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.0005690 | Defining the target and the effect of imatinib on the filarial c-Abl homologue | Previously we demonstrated the micro- and macrofilaricidal properties of imatinib in vitro. Here we use electron and multiphoton microscopy to define the target of imatinib in the adult and microfilarial stages of Brugia malayi and assess the effects of pharmacologically relevant levels of imatinib on the adult parasites.
After fixation of adult B. malayi males and females, sections were stained with polyclonal rabbit anti-c-Abl antibody (or isotype control) and imaged with multiphoton fluorescent microscopy. Microfilariae were fixed and labeled with rabbit anti-c-Abl IgG primary antibody followed by anti-rabbit gold conjugated secondary antibody and imaged using transmission electron microscopy (TEM; immunoEM). In addition, adult B. malayi males and females were exposed to 0 or 10μM of imatinib for 7 days following which they were prepared for transmission electron microscopy (TEM) to assess the drug’s effect on filarial ultrastructure.
Fluorescent localization of anti-c-Abl antibody demonstrated widespread uptake in the adult filariae, but the most intense signal was seen in the reproductive organs, muscle, and intestine of both male and female worms. Fluorescence was significantly more intense in the early microfilarial stage (i.e. early morula) compared with later development stages (i.e. pretzel). Anti-c-Abl antibody in the microfilariae localized to the nuclei. Based on TEM assessment following imatinib exposure, imatinib appeared to be detrimental to embryogenesis in the adult female B. malayi.
At pharmacologically achievable concentrations of imatinib, embryogenesis is impaired and possibly halted in adult filariae. Imatinib is likely a slow microfilaricide due to interference in intra-nuclear processes, which are slowly detrimental to the parasite and not immediately lethal, and thus may be used to lower the levels of L. loa microfilariae before they are treated within the context of conventional mass drug administration.
| While Loa loa, or the “African eye worm”, does not commonly cause clinical disease, infection with high blood levels of microfilariae from this helminth infection is problematic in those who receive mass drug administration (MDA) in the eradication efforts of lymphatic filariasis and onchocerciasis. Ivermectin, the drug of choice for both MDA programs, causes a rapid killing of Loa loa microfilariae and in those with high pre-treatment levels, a severe inflammatory reaction can result in encephalopathy, and rarely death. Using the filariae Brugia malayi as an in vitro surrogate model for any of the blood-borne filarial parasites (including Loa loa) we have previously shown that imatinib can act as a slow microfilaricide. Understanding imatinib’s targets in adult and microfilariae may predict the in vivo effects on Loa loa microfilarial loads, and anticipate potential side-effects for future clinical trials. In this study, we localized c-Abl, the target of imatinib, to the reproductive structures of adult B. malayi, and to the nuclei of the microfilariae. Pharmacologically achievable levels of imatinib most strikingly caused necrosis of developing microfilariae in adult female B. malayi. These data further support pursuing clinical trials in the safety and efficacy of imatinib for Loa. loa and other filarial co-infections.
| The World Health Organization has prioritized the elimination of lymphatic filariasis and onchocerciasis by 2020, and 2025 respectively through yearly or bi-annual administration of ivermectin or other anthelmintics in filarial-endemic regions of the world, a strategy aimed at interrupting transmission [1]. Despite the significant resources being devoted to this, challenges remain in meeting this goal. In particular, the presence of Loa loa, also known as “the African eyeworm”, that is co-endemic with other filariae (e.g. W. bancrofti or O. volvulus) in 10 Central African countries has complicated mass drug administration (MDA) programs because of the ivermectin-associated severe/serious adverse events (SAEs [e.g encephalopathy and coma]) [2–4]. While the pathophysiology of these post-ivermectin SAEs is not fully understood, it is believed that rapid killing of the microfilariae (MF) by ivermectin and the inflammatory response induced as a consequence underlies these SAEs [5–7].
Given that severe post-treatment reactions correlate with both the rapidity of MF killing (60–70% killing within 3 days) and the pre-treatment MF levels, [6] a potential alternative approach is to safely lower MF with another agent before treatment with ivermectin. We have recently demonstrated that the tyrosine kinase inhibitor imatinib at pharmacologically achievable concentrations is a slow microfilaricide of B. malayi in vitro [8].
As efforts are underway to test imatinib on microfilaremic patients we aimed to pursue further understanding in the mechanism of action by which imatinib may be harmful to both the microfilarial and adult stages of blood borne pathogenic filariae. We hypothesized that pharmacologically achievable concentrations of imatinib are able to damage the adult filarial worms as assessed by transmission electron microscopy following drug exposure. We also sought to localize the target of imatinib (c-Abl) in both adult and microfilarial B. malayi worms using multiphoton fluorescent microscopy and immuno-electron microscopy and to infer its mode of action by assessing the nature of the damage it induces.
Male and female adult B. malayi worms were obtained under contract from the University of Georgia (Athens, GA) and cleaned as previously described [9]. In vitro exposure of parasites to imatinib methanesulfonate salt was performed as previously described for 7 days [8].
B. malayi crude antigen (BMA) was prepared as previously described [10] from equal number of adult female and adult male worms.
BMA was mixed with Laemmli sample buffer (1610737 Bio-Rad) and 2-mercaptoethanol; 11 μg (lanes 3 and 9) 18.6 μg (lanes 4 and 10) of protein was loaded into a 4–15% precast gel (4561084 Bio-Rad) and transferred onto polyvinylidene difluoride (PVDF) membranes (Bio-Rad) using the Trans Blot Turbo Transfer System (1704150 Bio-Rad). Following blocking with 5% nonfat milk, the membrane was cut in half and incubated with either rabbit anti-c Abl antibody (Ab15130 Abcam) or an isotype control (011-000-003 Jackson Immunoresearch Labs), each diluted to a final concentration of 0.1 μg/mL and incubated overnight at 4°C. After washing with TBS, 0.1% Tween-20, the membrane was incubated with HRP-conjugated anti-rabbit IgG (7074 Cell Signaling Technology) at 1:5000 diluted in TBS, 0.1%Tween-20/5% milk at room temperature for 2 hrs. The membrane was washed again with TBS/0.1%Tween-20 and a chemiluminescent substrate was added (1705060 Bio-Rad) and exposed to x-ray film for various times prior to developing.
Approximately 100 adult males or adult females were fixed in 4% paraformaldehyde (15710 Electron Microscopy Sciences, 16% PFA diluted with 1X PBS) at 4°C overnight, then placed in 2% agarose in DMEM (1054 Gibco). Sections 200 μm thick were sliced with a Leica VT1000 S Vibrating Blade Microtome (Leica Microsystems) at speed 5, in ice-cold phosphate buffered saline (PBS), and placed in 1% or 2% bovine serum albumin (BSA) in PBS. Sections were permeabilized with 2.5% BSA, 1% Triton-X100 in PBS by agitating slowly at 4°C for 1 hour. Blocking was performed with 1% BSA, 10% goat serum and 0.1% Tween-20 in PBS agitating slowly for 1 hour at 4°C. Sections were washed with 1% BSA/PBS. Polyclonal rabbit antibody raised against human c-Abl (Ab15130 Abcam) or isotype control rabbit IgG (011-000-003 Jackson Immunoresearch Labs) at a concentration 0.05ug/mL was added to 1% BSA, 1% Triton-X100 in PBS and agitated slowly at 4°C for 72–96 hours. Sections were washed twice with 1% BSA/PBS, then counterstained with Alexa Fluor 594 goat anti-rabbit IgG (A31632 Life Technologies) at 1:2000 with 1% Triton-X100, 1%BSA and incubated overnight at 4°C agitating. Sections were then washed with 1%BSA/1% Triton/PBS for 16–24 hours agitating at 4°C. DAPI (R37606 Molecular probes) was then added. The experiment was repeated 6 times.
Immunostained sections were mounted in 14 mm microwell dishes (MatTek) and imaged using a Leica SP8 inverted 5 channel confocal microscope equipped with dual multiphoton (MP) lasers and a motorized stage. Microscope configuration was set up for three-dimensional analysis (x,y,z) of nematode sections. Mai Tai laser was tuned to 840 nm and InSight Deep See laser to 1150 nm excitation wavelengths. Second harmonic generation (SHG) signal was recorded at 420 nm wavelength. To collect tiled images of a whole section, z stacks consisting of 20 single planes (1 μm each, over a total tissue depth of 20 μm) were acquired and stitched automatically in LAS X (Leica Microsystems) post-acquisition. Images were processed using Imaris (Bitplane) software. Mean intensities of c-Abl fluorescence in various internal structures were analyzed using ImageJ (National Institutes of Health). Statistical analysis was performed using GraphPad Prism 7. Three sections per condition were analyzed.
Statistical analysis of fluorescence intensities was performed using GraphPad Prism 7. Sections from 3 independent experiments were analyzed. Paired Student's t test was used to compare mean fluorescence intensities between structures within one section. Values are presented as means ± SEM as indicated. *, P<0.05; **, P<0.01.
Transmission electron microscopy was performed on imatinib-treated and untreated male and female adult B. malayi. Day 7 following exposure to 0, 10, 25, or 50 μM of imatinib, worms were fixed in 2% glutaraldehyde and 2.5% paraformaldehyde buffered with 0.1 M sodium cacodylate and 1% sucrose and immediately cut into 3 sections, with the gonad-containing mid-portion further processed for imaging as described elsewhere [11].
For ultrastructure localization of the c-Abl protein in B. malayi MF, the worms were fixed and processed as described previously [12] using rabbit anti-c-Abl antibody (ABIN753613 Antibodies-Online.com) or rabbit isotype control at a concentration of 1:50. In brief, worms were fixed in 4% paraformaldehyde and 0.1% glutaraldehyde buffered with 0.1 M sodium cacodylate and 1% sucrose. Worms were embedded in LR White resin, sectioned using an ultramicrotome and grids containing the sections were incubated with primary antibody overnight at 4°C. The following day, the grids were washed with buffer, incubated with gold conjugated secondary antibody and contrasted with uranyl acetate before imaging with TEM.
To assess the ability of the purified mono-specific antibody raised against the human c-Abl protein to bind specifically to protein(s) in BMA, immunoblotting was performed. As seen (S18 Fig), there was binding of the antibody to a single band of ~ 50kD, whereas the isotype control run at the same dilution failed to react to BMA.
Using the same highly purified mono-specific antibody raised against the human c-Abl protein we were able to localize the filarial c-Abl-like tyrosine kinase in both adult male and female worms (Fig 1). As can be seen, there was specific staining throughout the hypodermis, including the lateral, ventral and dorsal hypodermal cords, somatic muscle, and reproductive tracts. Within the lateral cords, there was particularly high signal in the excretory-secretory canal (best seen in Fig 1B and 1C). There was variable, but specific staining throughout the intestine (Fig 1).
Quantitative analysis of mean fluorescence intensities detected in internal structures (S1 Fig) has confirmed significant differences in the amount of filarial c-Abl-like tyrosine kinase expression in various organs of adult worms (Fig 2).
In the females, it is notable that among the early stages of embryonic development, (Fig 1A and 1B) c-Abl expression is seen throughout the interstitium surrounding the developing embryo, as well as in the ovary and uterine lining. However, over the course of microfilarial development the expression of c-Abl appears to decrease in the interstitium surrounding the developing worm, and by later development (pretzel stage, Fig 1C), c-Abl expression surrounding the developing embryo is minimal. This signal change is further quantified in Fig 2, where the early morula stage has higher c-Abl expression compared with the pretzel stage (p = 0.0035).
When the anti-c-Abl primary antibody was used in immunoEM of B. malayi microfilariae the protein was localized solely to the MF nuclei (Fig 3B) and not to any other specific structures within the microfilariae. As expected, isotype control antibody exhibited very little staining of microfilariae (Fig 3A).
Having identified where in the parasite the c-Abl homologue is expressed, we next wanted to determine how imatinib exploits these anatomical niches to harm or kill the parasites. Thus, using TEM of sections prepared from adult B. malayi parasites exposed to 10 μM imatinib we were able to demonstrate significant damage to the reproductive apparatus in female worms (Fig 4). The most striking feature was the significant necrosis of developing microfilariae. In the reproductive tracts of females treated with imatinib, some structures of the MF, such as the endoplasmic reticulum, were clearly distorted (Fig 4B). However, most other structures in the developing microfilariae were completely unrecognizable, with only clusters of electron dense vacuoles found throughout the bodies of individual MF surrounding remnants of nuclei (Fig 4 panels A2 and B2-B4). Interestingly, there were no significant changes to Wolbachia at 10 μM (Fig 5A and 5B), nor at higher concentrations of imatinib (S7). At 10 μM, the hypodermis (Fig 5C and 5D), muscle (Fig 5C and 5D), nerve chords (Fig 6A and 6B), and intestine (Fig 6C and 6D) in treated females showed minimal or no morphologic difference compared to controls.
Adult males treated with 10 μM of imatinib showed minimal disorganization of the somatic muscle in some areas (Fig 7), however, overall there were no striking differences in features at this drug concentration compared with control males.
Doses higher than 10 μM were substantially more macrofilaricidal in vitro. As can be seen in Supplemental images (S5 Fig in comparison to S2–S4 Figs), at 25 μM all developing microfilarial structures are destroyed, which is similarly seen at higher doses (S8 and S9 Figs). At 50 μM, both males and females showed significant architecture distortion of the hypodermis, and the nerve contained within the hypodermal cords also appeared obviously damaged (S7, S11, S15 and S16 Figs).
Imatinib mesylate (Gleevec) is an orally administered small molecule c-Abl (and other tyrosine kinase) inhibitor with over 15 years of safety and pharmacokinetic data. Initially designed to target the constitutively active Bcr-Abl1 tyrosine kinase seen in chronic myelogenous leukemia (CML), the pharmacokinetics (PK) have been described in healthy volunteers, patients with CML as well as other malignancies (gastrointestinal stromal tumor, glioma)[13, 14]. Following oral administration, Cmax is achieved after 2–4 hours, with 98% bioavailability, and is highly protein bound[13–15]. The half-life of the imatinib parent compound is 18 hours, and 40 hours for its primary active metabolite[13, 14]. There is some inter-patient drug level variability, thought to be due to differences in drug metabolism by cytochrome-P450 isoform polymorphisms or efflux pumps,[14] although there has been conflicting data in this respect [16]. There is also evidence that blood concentrations of drug do not necessarily mirror intracellular levels [17]. No studies have assessed drug pharmacokinetics on the Central African population, although one study has evaluated genetic polymorphisms and the influence on imatinib blood levels in a West African population. It found that Nigerians were more likely to have genotypes that were associated with lower trough drug levels, however even between ethnic tribes allelic frequencies were significantly different [18]. Thus, it is clear that in areas endemic to Loa loa infection in Central Africa that more work still needs to be done in assessing imatinib pharmacokinetics in this population.
In general, imatinib is extremely well-tolerated. With chronic daily use, common side effects of imatinib include edema, gastrointestinal complaints, fatigue, rash, and headache[19]. Infrequently are cytopenias seen outside the setting of hematologic malignancy [19]. In a recent large series examining 10 years of safety data prospectively gathered on CML patients, only 6.9% of subjects discontinued treatment due to adverse events, and no new adverse events were discovered due to long term exposure of the drug [20]. The targeted mechanism of action of the drug, and the fact that it does not act by DNA alteration likely account for these findings. However, with a one-time dose of imatinib, which is what we would propose for filarial treatment, side effects most likely to be encountered would more accurately be reflected in the data from the pharmacokinetic studies evaluating a single dose of drug in healthy subjects. In these studies, a minority of subjects developed headaches or nausea, and none developed any other significant adverse events [21, 22].
We have previously shown that the filarial homologue of the human c-Abl protein is highly expressed in adult B. malayi¸ and that imatinib likely acts by inhibition of the filarial protein. Here we have shown that its inhibition in vitro in worms following exposure to 10 μM of imatinib causes physical damage to structures crucial for reproduction in adult female B. malayi. Following 600 mg of imatinib, a dose corresponding to that used in gastrointestinal stromal tumors, an average blood concentration of 13.2 μM (95% CI ± 6.4 μM) is achieved [23, 17].
We have used B. malayi given that these are the only human pathogenic filariae that can be obtained in large quantities from suitable small animal models. B. malayi, like the other pathogenic blood borne filariae including L. loa, has sheathed microfilariae, and each life stage has significant structural similarity [24–26]. We have previously reported the high degree of genetic identity in the c-Abl protein sequence between L. loa and B. malayi. Moreover, the predicted protein-drug interaction of the L. loa c-Abl homologue and imatinib, a small molecule inhibitor of c-Abl, [8] is largely identical to the site of interaction between the B. malayi c-Abl homologue and imatinib. Our previous study also demonstrated that at pharmacologically achievable concentrations (5 μM and 10 μM) imatinib acted as a slow (<50% killing by 4 days) microfilaricide in B. malayi in vitro. Slow killing is believed to be an important characteristic of a drug to be used as a microfilaricide in L. loa, as it is thought that rapid antigen release in the course of a high burden of microfilarial death is what causes adverse treatment reactions [6].
In B. malayi microfilariae, the c-Abl homologue was localized to the somatic nuclei (Fig 3). Over the course of development, these cells coalesce into syncytia called the hypodermis in adult filariae [27]. In nematodes, the hypodermis is responsible for cuticle integrity [28] and many metabolic processes, as it contains fat, glycogen, nerves, mitochondria, and endoplasmic reticulum and is a major site for carbohydrate metabolism [27, 29, 30]. The hypodermis separates the cuticle from the somatic muscle, and, in four places, it bulges into two lateral cords, one dorsal and one ventral cord (together called “hypodermal cords”). Each lateral cord contains an excretory secretory canal, the endosymbiont Wolbachia, nerve bundles, and hypodermal nuclei [27]. We have demonstrated that the protein expression of the c-Abl homologue is consistent in the anatomical structures associated with the hypodermis from microfilariae to adult stages.
Similar to the hypodermis, the female uterus is also highly metabolically active, and is also the site of a large amount of diverse protein transcription to support the developing embryos [31]. Direct evidence for the involvement of the filarial c-Abl homologue in reproduction is demonstrated by examining the impact of 10 μM imatinib on embryogenesis in the female. TEM images demonstrated necrotic MF (Fig 4A and 4B) compared with untreated controls. The localization of c-Abl to the uterine epithelium supports the idea that c-Abl is involved in processes necessary for MF development.
Filarial embryonic development proceeds linearly down two parallel uteri until mature sheathed and elongated microfilariae are expelled from the vagina. After multiple divisions of the fertilized zygote, the eggshell, which will ultimately become the microfilarial sheath, separates from outer layer of the embryo, soon followed by the eggshell becoming intricately folded (best seen in Fig 3A control) [32]. At this development stage within the uterus and distally to the vulva the uterus is lined with large apocrine cells [32]. Just as the composition of the uterine epithelium is different at various embryonic stages, c-Abl localization around the embryos appears to change and decrease as the developing microfilariae mature and eventually elongate. Early in development, fluorescence is measured throughout the fertilized ova (Figs 1A and 2). As the embryo matures, the specific staining consolidates around the outer surface, as can be seen in the morula stage (Fig 1B), however the overall fluorescence is similar to that seen in the fertilized ova (Fig 2). Finally, as can be seen in the young microfilarial pretzel stage, the staining is limited to only the uterine lining (Fig 1C), and the signal is significantly less than earlier stages (Fig 2). Interestingly, these changes mimic what was seen in a previous study examining the localization of the microfilarial sheath protein 2 [33]. Additionally, the effect of imatinib on developing microfilariae is similar to effects observed on the reproductive apparatus of Schistosoma mansoni [34], and in Echinococcus multilocularis [35]. One unifying observation made in the localization of c-Abl to the hypodermis and uterine lining and the destruction in these areas seen at very high imatinib concentrations (S7, S10, S11 and S14 Figs) in both of these areas is that significant glucose metabolism takes place in the hypodermis and uterine epithelium [36]. We therefore hypothesize that the c-Abl homologue may be involved in maintenance of the developing embryos’ outer eggshell and/or in glucose utilization related to supplying energy for this process. Interruption of this protective layer may ultimately lead to the microfilarial necrosis seen in utero in imatinib-treated females.
Given the genetic similarities between B. malayi and L. loa, the above offers evidence that a single oral dose of imatinib may not only affect circulating L. loa microfilariae but may also impair embryogenesis, and potentially future fecundity in adult L. loa filariae. In an effort to avoid the deleterious effects of ivermectin administration in Loa-endemic areas a strategy termed Test (and) Not Treat (TNT) has been suggested whereby screening the entire “at risk” population is performed [37] and treatment with ivermectin in only those with levels of L. loa below a certain threshold. However, left unaddressed is what to do with those people with levels of microfilariae above this “safe” threshold. Moreover, given that for some other blood-borne filariae (e.g. M. perstans) there is a positive correlation between the intensity of M. perstans and that of L. loa, [38] not treating those with the highest L. loa levels may have implications for other co-incident filarial infections as well. Thus, if the TNT strategy is to be adopted, then the dosing of imatinib in those excluded individuals may be a safe and already available treatment. Clinical trials of imatinib in Loa-endemic areas to assess its efficacy and safety in reducing blood microfilarial levels are of utmost importance.
|
10.1371/journal.pntd.0003395 | Impact of a Dengue Outbreak Experience in the Preventive Perceptions of the Community from a Temperate Region: Madeira Island, Portugal | The ability to effectively modify behaviours is increasingly relevant to attain and maintain a good health status. Current behaviour-change models and theories present two main approaches for (healthier) decision-making: one analytical/logical, and one experiential/emotional/intuitive. Therefore, to achieve an integral and dynamic understanding of the public perceptions both approaches should be considered: community surveys should measure cognitive understanding of health-risk contexts, and also explore how past experiences affect this understanding. In 2011, community perceptions regarding domestic source reduction were assessed in Madeira Island. After Madeira’s first dengue outbreak (2012) a unique opportunity to compare perceptions before and after the outbreak-experience occurred. This was the aim of this study, which constituted the first report on the effect of an outbreak experience on community perceptions regarding a specific vector-borne disease. A cross-sectional survey was performed within female residents at the most aegypti-infested areas. Perceptions regarding domestic source reduction were assessed according to the Essential Perception (EP)-analysis tool. A matching process paired individuals from studies performed before and after the outbreak, ensuring homogeneity in six determinant variables. After the outbreak, there were more female residents who assimilated the concepts considered to be essential to understand the proposed behaviour. Nevertheless, no significant difference was observed in the number of female residents who achieved the defined ‘minimal understanding’’. Moreover, most of the population (95.5%) still believed at least in one of the identified myths. After the outbreak some myths disappeared and others appeared. The present study quantified and explored how the experience of an outbreak influenced the perception regarding a dengue-preventive behaviour. The outbreak experience surprisingly led to the appearance of new myths within the population, apart from the expected increase of relevant concepts’ assimilation. Monitoring public perceptions is therefore crucial to make preventing dengue campaigns updated and worthy.
| Since there is no vaccine or treatment available for dengue fever, its prevention relies on community participation. Residents are asked to remove from their houses and gardens all receptacles where mosquitoes can breed. Exploring the public perception regarding dengue prevention is crucial for detecting obstacles to their participation in the proposed preventive activities. The authors explored and compared the community's perceptions before and after the first dengue outbreak in Madeira Island. For the first time it was possible to study the effect of a dengue outbreak in the public perceptions regarding its prevention. After the dengue outbreak, the authors found an improvement in the perception of the community. However, even after experiencing an outbreak, the majority of the residents still did not understand their role in the dengue prevention and, thus were not ready to adhere to it. Moreover, the authors also observed some new myths within the community after the outbreak (which were not present before the outbreak). The improvement of community perceptions was expected. However this search also revealed that this experience can surprisingly promote the emergence of new myths which may hamper the community engagement in the dengue prevention.
| Most of the 2011 worldwide major causes of death (MCD), rely on behaviour changes for their prevention [1]. Increasing physical activity, fruits/vegetables intake, hand-washing, use of condoms, and decreasing not only fat, salt and sugar intake but also smoking habits, are crucial in the control of heart disease (1st MCD), stroke (2nd MCD), chronic obstructive lung disease (4th MCD), diarrhoea (5th MCD), HIV (6th MCD), or diabetes (8th MCD). Behaviour changes are increasingly relevant to attain and maintain a good health status, especially when facing health threats for which there is no efficient or timely treatment. This is the case for dengue fever that such as other mosquito-borne diseases, requires a good compliance to certain preventive, protective or therapeutic actions. Moreover, since there is no vaccine or treatment for dengue fever, neither 100% effective insecticides, community behaviour has a huge impact on its prevention and control [2].
It is still not widely understood how to effectively promote behaviour changes [3]. During several decades, many behaviour impact campaigns were shown to be fruitless. In the past 50 years, literature extensively presented theoretical models which tried to clarify cognitive ways of (healthier) behaviour acquisition [4,5,6,7,8,9]. Recently, the concept of ‘past experiences’ was stated as being crucial in determining (healthier) decision-making. Many authors claimed that, due to the type of emotions and intuition that they produced, ‘past experiences’ could strongly encourage or discourage a particular action [6,7,8,9,10].
Altogether, these contributions seem to present two different approaches by which humans perceive decision-making and then make decisions: one analytical and one experiential [11]. In order to improve the efficacy of the behaviour-promoting messages, the authors firmly suggested that messages should be not only meaningful but also emotionally adequate for the targeted community. This way, the assessment of community’s cognitive and emotional perceptions, is hence useful in the guiding of effective health-seeking messages. However, few studies explored emotional experience-driven perceptions but rather frequently only focused on the assessment of the cognitive ones [12].
Some evidence suggested that experience can influence public perceptions and reactions in two ways [13]. In one aspect, it can over-estimate the risk perception [10,13] (i.e. alert-feeling, referred to as ‘availability bias’ [14]) and consequently promote protective/preventive actions. It can also underestimate the risk perception [12,15] (i.e. habituation effect also called ‘gambler’s fallacy’[14]) which can discourage protective/preventive actions. Only few studies have explored this issue in real situations. Besides the scientific interest of scrutinizing the complex process of (healthier) decision-making, the monitoring of public perceptions and behaviours contributes to the continuous and adequate update of the behaviour-promoting messages concerning their (rational and emotional) content. This is the case of any chronic and endemic disease, where the (health) risk is maintained during time such as dengue epidemic and endemic areas [12].
In 2005, a dengue vector species, Aedes aegypti was reported in Madeira archipelago. In 2012, the first dengue outbreak was recorded in the territory [16]. Community perception regarding preventive behaviours (domestic source reduction) was assessed and described in details by the current investigators, before the outbreak had been declared [17]. At the end of the outbreak, a unique opportunity to explore and compare community perception before and after the outbreak appeared. The aim of this study was thus to re-assess community perceptions regarding the same preventive behaviour (domestic source reduction) just after the dengue outbreak in order to compare how it has altered with the outbreak experience. This constitutes to our knowledge the first report of this kind describing the effect of an outbreak experience on community perceptions regarding a specific vector-borne disease.
As subsequently explained, methodology of the present survey (herein stated as POST-outbreak study) followed as much as possible the methodology used in the prior-to-the outbreak survey (herein mentioned as PRE-outbreak study) [17]. This ensured an accurate comparison between public perceptions before and after the dengue outbreak in Madeira Island.
Therefore, the tool used in the assessment of the community perceptions was maintained, i.e., the ‘Essential-Perception analysis’ (see sub-section of the same name). However, since the outbreak was not planed ahead nor predicted, in the POST-outbreak survey was not possible to reproduce exactly the same methodology used before the outbreak. Due to ethic, time and logistic constrains implicit in the preparation of this survey during the outbreak and in its implementation just after it, adjustments in the size of the studied sample and in the sampling methodology, were mandatory to make the POST-outbreak survey possible. The introduced alterations were sample size reduction (through rural and male residents’ exclusion) and intentional sample selection instead of the previous random one. These alterations are explained in detail in ‘Studied population’ sub-section. Finally a matching process was developed in order to overcome those constrains and guarantee an unbiased comparison between the two studies despite their differences in sampling methodology. For that, populations surveyed in both PRE/POST-outbreak studies and who had fully completed the questionnaires were scored according to the perceptions demonstrated (for EP-Score calculation) and marked according to the six socio-demographic characteristics (for the matching process). After this, populations were matched according to critical socio-demographic variables, as described in sub-section ‘Matching Process’ and EP-score was compared within matched pairs. Individuals who presented missing questions were excluded from the analysis.
Out of the several municipalities which were covered by the PRE-outbreak study area, only some were selected to be included in this POST-outbreak study (Fig. 1). In order to decrease the sample size, ‘Câmara de Lobos’ was excluded since it was the sole rural municipality. Facing the impossibility of including both urban and rural municipalities, the urban ones were preferred based on two main reasons: (i) they presented a dengue incidence rate greater than 200 during the outbreak (S1 Fig.); and (ii) they comprise the capital city of the archipelago, Funchal, and thus an important point of aegypti-dispersion. Part of the Funchal municipalities: ‘Sé’, ‘Santa Maria Maior’ and ‘Imaculado Coração de Maria’, were also included in the POST-outbreak study area besides those considered in the PRE-outbreak study (‘São Pedro’ and ‘Santa Luzia’). These extra-included area were also covered by the 2012’s most aegypti-infested area (presenting a density level of 31% or over along the year), thus ensuring a homogeneous level of natural exposure to the A. aegypti among the studied residents [18]. The geographic area covered in the present study will be mentioned as ‘Extended-AEGYPTI area’ and consists, thus, in part of five Funchal’s municipalities from the 2012 most aegypti-infested area.
Due to the mentioned unfeasibility to include a representative sample of the resident population of the study area, an intentional sample of exclusively female subjects who lived in the study area—‘Extended-AEGYPTI area’ aged 18 years old or over and who didn’t integrate the previous PRE-outbreak survey was selected from customers of central hairdressers and pharmacies, placed in the selected area. The women selection was based on three main reasons: (i) before the outbreak women were significantly less aware of domestic source reduction than men (S2 Fig. and S1 Table); (ii) women are the majority within the studied population [19]; (iii) women above 15 years-old were the age/gender-group more affected by the disease during the outbreak [20]; and (iv) culturally, in Madeira Island, women are more related to the main dengue-preventive behaviour proposed than men do (see details about the behaviour proposed in ‘Essential-Perception’ subsection). All women who entered in the establishment and who met the inclusion criteria were invited to participate.
The type of establishment were chosen in order to allow the study to cover the most possible heterogeneous women sample, in what concerns their age groups, education levels and socio-economic background. In order to stimulate participation of women from all the included municipalities, two central establishments of each service were chosen to participate in the study according to their convenient geographical location, being one placed in the east and the other in the west region of the studied area.
The study area has a population density which can be as high as 1433.5 habitants per square kilometers [21]. The sample size was calculated using Epitools’ sample size calculators (2014, AusVet Animal Health Services) in order to perform a comparison of two means using the t-test (2-tailed) where a 1 point is relevant in Essential-Perception Score difference (variation between −10, 10) [22]. The sample size calculation considered a 95% confidence level, a power of 80% and a pooled variance equal to 10 (S = √10 = 3.16). The obtained n was 157 (S2 Table). Finally, this sample size was inflated in 30% to account for incomplete interviews.
A cross-sectional survey was performed to assess residents’ perceptions through face-to-face interviews. Before data collection, establishments’ managers/participants gave their written/oral informed consent respectively. Previous to the beginning of this survey, the questionnaire was pre-tested in a non-selected establishment placed in the selected area. During the interview, a questionnaire comprising 21 questions was applied, covering dengue-preventive issues and personal-socio-demographic characteristics. In agreement with what was inquired in the PRE-outbreak study, questionnaire covered five main topics: ‘Medical Importance’ (two questions), ‘Local Context’ (two questions), ‘Domestic Attribute’ (three questions), ‘Mosquito Breeding’ (three questions) and ‘Control Measures’ (three questions) [17]. Besides the variables ‘gender’, ‘education level’, ‘age group’, and ‘geographical area’, two other variables were assessed: ‘travels to dengue endemic countries (DEC)’ which measures who had already been to any dengue endemic country and ‘admitted mosquito exposure (AME)’ which measures who recognized to had been bitten by mosquitoes. The survey was performed by trained personnel from the local health authority from 22nd of March until 16th of April, 2013. In each establishment (pharmacies/hairdressers), interviews were performed during a Monday-to-Saturday week, between 9am and 7pm (according to establishments’ opening hours) The study was approved by Instituto de Higiene e Medicina Tropical Ethics Committee, Instituto de Higiene e Medicina Tropical, Universidade Nova de Lisboa, Lisbon (reference: 09-2013-TD).
Populations studied in both PRE/POST-outbreak surveys were matched into pairs, ensuring homogeneity in six critical socio-demographic variables. Resulting matching population comprised thus pairs of individuals composed of an individual from the PRE-outbreak study and an individual from the POST-outbreak study with equivalent personal-socio-demographic characteristics. Matching pairs of individuals were equal in (or “blocked” on) gender, education level, age group, geographical area, travels to DEC and AME variables, already shown to be determinants to the individual perception [17]. This sampling methodology can also be called as randomized block design, and the latter variables as blocking factors [23].
For comparative purposes, the criteria ‘geographic area’ was applied in two different ways, generating two different matching approaches. In one matching approach, herein called ‘adjusted matching’, the ‘geographic area’ criteria distinguished only residents living in urban areas from residents living in rural areas. According to this criteria, ‘Câmara de Lobos’ (covered exclusively in PRE-outbreak study area) was the sole rural municipality. In the other matching approach, herein called ‘restricted matching’, the geographic criteria besides the previous distinction between urban and rural areas also differentiated urban municipalities covered in both PRE-outbreak and POST-outbreak studies (‘Santa Luzia’ and ‘São Pedro’) from the remaining urban ones which were exclusively included in the POST-outbreak study (‘Sé’, ‘Santa Maria Maior’ and ‘Imaculado Coração de Maria’). The other criteria (gender, education level, age group, travels to DEC and AME) were strictly applied in both matching approaches, i.e. only individuals who were equal in this variables were matched.
Both matching procedures were built in Excel (Microsoft Office, Windows 8), and guaranteed that individuals were randomly selected within those that were personal-socio-demographically equivalent. Moreover, matching procedures were optimized in order to re-include all the non-selected individuals in the subsequent matching rounds.
The assessment of the community perception was performed using the Essential-Perception analysis (EP-analysis), as described below.
Essential-Perception analysis assesses community perception regarding a particular behavioural proposal: the domestic aegypti’s source reduction, considered the most critical dengue-preventive practice by the World Health Organization [24]. This corresponds to the elimination (emptying, covering or removing) of water-containers present inside or around residential buildings. The EP-analysis’ considers ten essential concepts which assimilation by individuals was revealed to be determinant to the performance of the proposed behaviour (Table 1) [17]. Essential-Perception analysis allows the characterization and estimation of the community’s perceptions through four different approaches, all of them used here: (i) score of Essential-Perception, (ii) concept assimilation, (iii) topic understanding and (iv) myth identification and estimation. The first measures the number of concepts that were assimilated (out of those defined to be ‘essential’) and thus how far-off is the studied population from achieving the complete ‘Essential Perception’ (EP-Score = 10). The second describes how much those ‘essential’ concepts were assimilated or not-assimilated by the community. The third, organizes the ‘essential concepts’ in topics and describes how topics are/not being understood. Residents who have acknowledged both topic-related concepts are according to this tool considered to have ‘completely understood the topic’, the acknowledgement of only one out of the two topic-related concepts is considered as a ‘partial understanding of the topic’, and residents who did not perceive any of the two topic-related concepts are considered to have ‘not understood the topic’. Finally the fourth, by analysing the concept assimilation, identifies erroneous beliefs which may persist in the community (herein mentioned as ‘myths’) and estimates their putative frequency in the studied population (see example in S3 Fig.).
All collected information was introduced and records were double-checked. Statistical analysis was performed using Statistical Package for Social Sciences 19.0 (SPSS, Inc., Chicago, IL, USA). Answers obtained from the questionnaires were re-coded to obtain other categorical variables implicit in the EP-analysis. The personal-socio-demographic feature of the studied population presented in Table 2 was described in what concerns the gender, age group, education level, municipal division, travels to dengue endemic countries (DEC) and AME. The age groups were categorized in ten-year intervals and the education level was divided into five categories starting from ‘never studied’ until ‘upper graduation’. This categorization allow that groups were similar in number of individuals.
Comparison of the two urban municipalities covered in both PRE-outbreak and POST-outbreak studies (‘Santa Luzia’ and ‘São Pedro’) confirmed a priori the validity of the criteria ‘geographic area’ in the restricted matching. In this matching those municipalities were considered equivalent. S3 Table shows that despite the previous observed differences observed in their EP-score level, when “blocking” the education level there are no significant differences between the two municipalities.
Analysis of the demographic data of the extra-included areas compared the new added municipalities (‘Sé’, ‘Santa Maria Maior’ and ‘Imaculado Coração de Maria’) and the previously studied (‘Santa Luzia’ and ‘São Pedro’). Comparison is presented in S4 Table showing that there were no relevant differences between them in what concerns the two critical socio-demographic determinants: age group and education level supporting thus a priori the validity of the criteria ‘geographic area’ in the adjusted matching.
Comparisons of EP-score medians between populations from PRE/POST-outbreak studies were made using the non-parametric Wilcoxon Test (Table 3), after rejecting the normality of Essential-Perception score difference through Kolmogrov-Smirnov test. Additionally, the number of individuals who achieved an EP-score equal to or higher than seven (EP-score≥7) was calculated in both studies and differences were compared, using the McNemar Test (Table 4). This cut-off was chosen due to lack of subjects that achieved an EP-score equal to ten (EP-score = 10). In order to evaluate the methodology used during the matching process, comparisons between paired and non-paired samples were performed, according to their EP-score and their socio-demographic characteristics. In order to ensure that restricted and adjusted matching sample sizes (n = 47and n = 90) were satisfactory, the power associated to Wilcoxon test (the non-parametric alternative to t-test) was calculated a posteriori using free statistical power analysis program, G*Power 3.0 [25].
A total of 154 female Extended-AEGYPTI residents answered the complete questionnaire in the present POST-outbreak survey. A total of 90 pairs resulted from the adjusted matching between 154 female from the POST-outbreak survey and 1145 subjects who participated in the PRE-outbreak study. Each pair composed of an individual from the PRE-outbreak study and an individual from the POST-outbreak study with equivalent personal-socio-demographic characteristics. Nine individuals out of those surveyed had dengue and five were paired. The personal-socio-demographic feature of the studied sample populations is described in Table 2. When not mentioning subsequent paragraphs as well as the data presented in Figs. 2, 3, 4 and 5, and Tables 2, 3, 4 and 5 present results from the adjusted matching.
Statistical tests were performed in order to explore the differences between medians of the EP-score from studied populations in both PRE/POST-outbreak studies, confirming a significant increase in the EP-Score median after the outbreak (p<0.001, Table 3). An increase in the number of individuals who achieved an EP-score equal to or higher than seven (EP-score≥7) in the POST-study population was also statistically confirmed (p<0.001, Table 4).
In general, the community perception regarding preventive domestic practices improved within female residents of most aegypti-infested areas in Madeira Island after they experienced a dengue outbreak. By analysing how and how much assimilation of each 'Essential-concept’ has changed, crucial information can be retrieved regarding people´s perceptions about this experience and their future role in its prevention.
For many Madeira residents, the experience of this dengue outbreak was probably the first contact with a mosquito borne disease (it was the first in almost a hundred years in Europe, [26]. This can explain the observed increase in the assimilation of the idea that ‘mosquitoes can transmit diseases’ (MI1-concept). Moreover, before experiencing the outbreak, the community's worst incident with mosquitoes were allergic reactions, which could be considered as the sole health consequence of mosquito bites. After the outbreak, it was not surprising that the percentage of residents that were aware of ‘the kind of diseases that mosquitoes can transmit (such as dengue, yellow fever and malaria)’ (MI2-concept) almost doubled. Therefore, in the POST-outbreak study there were a higher percentage of people who rightly appraised the impact of mosquitoes in health. Since no fatal cases occurred during the dengue outbreak, some beliefs such as, ‘dengue disease does not kill’ and ‘dengue in Madeira is less aggressive’ may be present in the community. These questions should be clarified within the community due to the possibility of a different virus serotype reaches the Madeira territory, increasing the risk of dengue haemorrhagic cases to occur.
Even though assimilation of both ‘Local Context’ concepts increased after the outbreak, the majority of residents still ignored that ‘there is a high possibility for a (second) dengue outbreak in Madeira’ (LC2-concept). The acknowledgement of this concept was expected to increase after the outbreak, assuming that the previous identified myth which states that ‘Madeira were not at risk of have dengue’ would be opposed with the experience of a dengue outbreak. However, its assimilation merely increased 10%. Even though people had probably realized that Madeira was at risk and that several dengue cases occurred, two erroneous interpretations could explain this 10% result. Firstly, the false belief that the ‘dengue outbreak have ended due to the eradication of the disease or the mosquito’ (alleged myths 6a/6b, Table 5). Secondly, gambler’s fallacy, the invalid belief that when something happens more frequently than normal during a period of time, the probability of happening again in the future decreases (alleged myth 3, Table 5) [14]. People who believe in these alleged myths underestimate the probability of another dengue epidemics occur in Madeira Island.
Improvements in DA1-concept, DA2-concept, LC1-concept and MB1-concept can be attributed to the “boom” of educational information transmitted during the outbreak. This information was transmitted by the news, by official reports, and most importantly by the exhaustive door-to-door campaign that was rapidly implemented in the areas where most dengue cases were being reported during the outbreak period. In the latter, trained personnel of the health-authorities entered in residential buildings and supported the residents in performing correct and extensive elimination of mosquito breeding sites inside and in the surroundings of their houses (i.e. aegypti source reduction). This provided a useful opportunity for residents to realize ‘the existence of larval forms/mosquitoes in their own houses’ (DA 1-concept), to ‘recognize containers that were serving as breeding sites’ (MB1-concept), to emphasize the idea that ‘domestic control could be efficient in the A. aegypti control’ (DA2-concept), and finally to comprehend that their ‘residential area had (indeed) vector-mosquitoes’ (LC1-concept).
In contrast to the improvement in the above stated concepts, the percentage of people who believed in ‘false mosquito breeding inducers, such as, animals or food debris’ augmented after the outbreak and thus, MB2-concept was the sole concept of which assimilation had declined after the outbreak. Female residents may have ‘erroneously indorsed A. aegypti’s proliferation to dirty environments’ (with food debris or animals). This assumption could be interpreted as an intuitive explanation for the appearance/establishment of the A. aegypti and dengue disease in the Island. As stated in psychology in the attribution theory, humans need to “attribute” causes to events which are not understood [27]. Female residents, who agreed with latter belief, and believe to live in clean households, will not feel responsible to perform domestic source reduction.
Finally, almost all the female residents agreed with the efficacy of domestic source reduction in the control of mosquitoes (CM1-concept). However, the majority still erroneously consider ‘insecticide application or flyswatter usage’ as effective measures to control mosquito population (CM2-concept). In fact, these practices are protective (i.e. can, in some manner, avoid the mosquito bite) but are not preventive (i.e. are able to control the mosquito proliferation). This mistake is determinant because people that believe in it tend to focus their efforts on these easier but less efficient practices and to disfavour the truly efficient ones, which are more difficult to implement (such as, domestic source reduction). Moreover, previous studies had shown that the local A. aegypti population, present in Madeira Island, was resistant to the most common insecticides, which raised questions about the reasonability of its application, even when used with protective objectives [28].
Overall, there were only three Essential Concepts that were still not considered by the majority of the studied population (LC2-concept, MB2-concept and CM2-concept). Under the assumptions of the EP-analysis, the individual minimal understanding and putative subsequent compliance to the proposed behaviour, requires the assimilation of all the ten concepts defined as ‘essential'. Consequently, the weak integration of one of these concepts by the community can compromise the usefulness of the behaviour impact campaigns. It is worth pointing out that, even though concept assimilation had generally improved after the outbreak, only 4.5% of the studied population achieved the referred ‘minimal understanding’ (EP-Score equal to ten). Consequently, there were still very few residents that are ready to engage in the proposed behaviour.
Along with the observed improvement of essential concept assimilation, myths believed by the community also changed. Even though the community is now closer to the needed ‘minimal understanding’, the task of local authorities is still difficult since after the outbreak they have to cope with new/different beliefs, following ideas such as ‘Madeira is immune to suffer a second outbreak’ (alleged myth 3 and 6).
In reality, myths could subtly persist in the community, weakening the effect of strategies aimed at behaviour changes. Therefore, an adequate monitoring of public perceptions is undoubtedly crucial to (more quickly) detect them, allowing preventive campaigns to be planned accordingly. Apart from the here observed public erroneous interpretations (probably caused by their short contact with the vector and the disease) community can provide other enriching contribution such as technical hitches in implementing proposed behaviours, pointing out messages or expressions difficult to understand, and suggesting housewives-friendly solutions [29,30].
The similarity found between paired and non-paired samples, regarding their EP-score levels supported the validity of the criteria used in the adjusted matching approach. Moreover, the observed equivalent results between the adjusted and the restricted matching procedures corroborated the validity of the geographical adjustments. Furthermore, the calculated power value supported the strength of the results albeit the apparently small size of the sample. In fact, prior sample size estimations indicated a minimal amount of 157 subjects required to fulfil the objectives of this study (as mentioned in Methods section), assuming a minimal difference (1 point) between the EP-score levels from PRE/POST-outbreak studies. However, since a difference of 2 point was observed, only 40 pairs of subjects were needed to detect it fulfilling the same objectives (S2 Table). The studied sample size was thus higher than the required to the aimed analysis. Therefore, the power associated to Wilcoxon test is also high as described in Table 6.
It is worth noticing that considering the studied sample, only women from urban areas were covered, and therefore results may not be equivalent in male subjects, rural communities or in long-term dengue regions.
In conclusion, after experiencing a dengue outbreak in Madeira Island, female community perception towards the aimed preventive engagement improved in some aspects (as intuitively expected) but also deviated in other aspects, particularly by the emergence of new myths. The most frequent myths may be used in the future to outline appropriate priority messages. Subsequent health-messages tailored according to present findings could strengthen community engagement in dengue-preventive behaviours.
Monitoring public perceptions (before/after an intervention or an outbreak) revealed a great value, not only for public health professionals but also for researchers who may be interested in investigating the complex interplay between experiences, perceptions and decision-making. Thus, lessons taken from this work can be useful not only for local authorities but also for all professionals who are engaged in dengue preparedness in endemic or epidemic countries, as well as, to those interested in strengthening tools for other behaviour-based preventable diseases.
|
10.1371/journal.pcbi.1004900 | A Model for Direction Sensing in Dictyostelium discoideum: Ras Activity and Symmetry Breaking Driven by a Gβγ-Mediated, Gα2-Ric8 -- Dependent Signal Transduction Network | Chemotaxis is a dynamic cellular process, comprised of direction sensing, polarization and locomotion, that leads to the directed movement of eukaryotic cells along extracellular gradients. As a primary step in the response of an individual cell to a spatial stimulus, direction sensing has attracted numerous theoretical treatments aimed at explaining experimental observations in a variety of cell types. Here we propose a new model of direction sensing based on experiments using Dictyostelium discoideum (Dicty). The model is built around a reaction-diffusion-translocation system that involves three main component processes: a signal detection step based on G-protein-coupled receptors (GPCR) for cyclic AMP (cAMP), a transduction step based on a heterotrimetic G protein Gα2 βγ, and an activation step of a monomeric G-protein Ras. The model can predict the experimentally-observed response of cells treated with latrunculin A, which removes feedback from downstream processes, under a variety of stimulus protocols. We show that G α 2 β γ cycling modulated by Ric8, a nonreceptor guanine exchange factor for G α 2 in Dicty, drives multiple phases of Ras activation and leads to direction sensing and signal amplification in cAMP gradients. The model predicts that both G α 2 and Gβγ are essential for direction sensing, in that membrane-localized G α 2 *, the activated GTP-bearing form of G α 2, leads to asymmetrical recruitment of RasGEF and Ric8, while globally-diffusing Gβγ mediates their activation. We show that the predicted response at the level of Ras activation encodes sufficient ‘memory’ to eliminate the ‘back-of-the wave’ problem, and the effects of diffusion and cell shape on direction sensing are also investigated. In contrast with existing LEGI models of chemotaxis, the results do not require a disparity between the diffusion coefficients of the Ras activator GEF and the Ras inhibitor GAP. Since the signal pathways we study are highly conserved between Dicty and mammalian leukocytes, the model can serve as a generic one for direction sensing.
| Many eukaryotic cells, including Dictyostelium discoideum (Dicty), neutrophils and other cells of the immune system, can detect and reliably orient themselves in chemoattractant gradients. In Dicty, signal detection and transduction involves a G-protein-coupled receptor (GPCR) through which extracellular cAMP signals are transduced into Ras activation via an intermediate heterotrimeric G-protein (G α 2 β γ). Ras activation is the first polarized response to cAMP gradients in Dicty. Recent work has revealed mutiple new characteristics of Ras activation in Dicty, thereby providing new insights into direction sensing mechanisms and pointing to the need for new models of chemotaxis. Here we propose a novel reaction-diffusion model of Ras activation based on three major components: one involving the GPCR, one centered on G α 2 β γ, and one involving the monomeric G protein Ras. In contrast to existing local excitation, global inhibition (LEGI) models of direction sensing, in which a fast-responding but slowly-diffusing activator and a slow-acting rapidly diffusing inhibitor set up an internal gradient of activity, our model is based on equal diffusion coefficients for all cytosolic species, and the unbalanced local sequestration of some species leads to gradient sensing and amplification. We show that Ric8-modulated G α 2 β γ cycling between the cytosol and membrane can account for many of the observed responses in Dicty, including imperfect adaptation, multiple phases of Ras activity in a cAMP gradient, rectified directional sensing, and a solution to the back-of-the-wave problem.
| Many eukaryotic cells can detect both the magnitude and direction of extracellular signals using receptors embedded in the cell membrane. When the signal is spatially nonuniform they may respond by directed migration either up or down the gradient of the signal, a process called taxis. When the extracellular signal is an adhesion factor attached to the substrate or extracellular matrix, the response is haptotaxis [1], and when it is a diffusible molecule the process is called chemotaxis. Chemotaxis plays important and diverse roles in different organisms, including mediation of cell-cell communication [2], in organizing and re-organizing tissue during development and wound healing [3–5], in trafficking in the immune system [6], and in cancer metastasis [7].
Chemotaxis can be conceptually divided into three interdependent processes: direction sensing, polarization, and locomotion [8, 9]. In the absence of an external stimulus, cells can extend random pseudopodia and ‘diffuse’ locally, which is referred to as random motility [10]. Direction sensing refers to the molecular mechanism that detects the gradient and generates an internal amplified response, providing an internal compass for the cell [11]. Polarization involves the establishment of an asymmetric shape with a well-defined anterior and posterior, a semi-stable state that allows a cell to move in the same direction without an external stimulus. These three processes are linked through interconnected networks that govern (i) receptor-mediated transduction of an extracellular signal into a primary intracellular signal, (ii) translation of the primary signal into pathway-specific signals for one or more signalling pathways, and (iii) the actin cytoskeleton and auxiliary proteins that determine polarity of the cell. A single extracellular signal may activate numerous pathways, but our focus herein is on the first pathway, which involves transduction of an extracellular cAMP signal via a GPCR, and one specific pathway of the second type, the Ras pathway, which is involved in activating the appropriate downstream networks that govern chemotactic locomotion.
Dicty is an amoeboid eukaryotic cell that utilizes chemotaxis during various stages of its life cycle. In the vegetative phase, it locates a food source by migrating toward folic acid secreted by bacteria or yeast. When the food supply is depleted Dicty undergoes a transformation from the vegetative to the aggregation phase, in which cells sense and migrate toward locally-secreted 3’-5’ cyclic adenosine monophosphate (cAMP), which serves as a messenger for control of chemotaxis and other processes [9, 12]. Dicty has served as an excellent model for studying the interconnected signalling pathways governing chemotaxis due to its genetic and biochemical tractability [13–15]. The major components of the network topology for chemotaxis have been identified by analyzing the effects of gene knockouts and the response of cells to various spatio-temporal signalling protocols [8, 16, 17].
The first step of the chemotactic process involves signal transduction by GPCR’s that activate G-proteins, which is described in detail in the following section. Activated G-proteins can in turn activate numerous pathways, and the pathway we analyze here involves Ras, which is a monomeric G protein that functions as a molecular switch that activates downstream effectors such as PI3K in its activated GTP-bound state. Activation of Ras is the earliest measurable polarized signalling event downstream of G protein activation [14, 18]. A major question from both the experimental and the theoretical viewpoints is how the cell transduces a shallow spatial gradient of extracellular cAMP into a steeper internal gradient of activated Ras. Recent experiments show that Ras activity exhibits multiple temporal phases in cAMP gradients [19]. The first phase is transient activation of Ras that is essentially uniform over the entire cell boundary. In the second phase, symmetry is broken and Ras is reactivated exclusively at the up-gradient side of the cell. The third phase is confinement, in which the crescent of activated Ras localizes further to the region exposed to the highest cAMP. Other recent observations that are not incorporated in existing models are as follows. Firstly, the Ras symmetry breaking does not depend on the presence of the actin cytoskeleton—treatment of cells with latrunculin A (LatA), which leads to depolymerization of the network—does not destroy the symmetry-breaking [19]. Secondly, it was found that when two brief stimuli are applied to the same cell, the response to the second stimulus depends on the interval between the stimuli, which indicates that there is a refractory period [20]. Other experiments show that the adaptation of Ras activation is slightly imperfect, and Ras activity is suppressed when the chemoattractant concentration is decreasing in time, a phenomenon called rectification [21]. Finally, it was reported that there is a persistent memory of Ras activation, even when the cells are treated with LatA [22].
These new results are difficult to interpret in the framework of existing models, a number of which have been proposed [11, 20, 22–29]. Most current models are based on an activator and inhibitor mechanism called LEGI—local excitation, global inhibition—to explain both direction sensing and adaptation when the chemoattractant level is held constant [30]. While these models shed some light on direction sensing, their usefulness is limited due to the oversimplification of the signal transduction network—as will be elaborated later. In particular, none of the existing models incorporates sufficient mechanistic detail to satisfactorily explain the spectrum of observations described above, which provides the rationale for a more comprehensive model that enables us to test hypotheses and make predictions concerning the expected behavior of the signal transduction pathways.
The key components in the model we develop herein are the G-protein G α 2 β γ, RasGEF and RasGAP, which control rapid excitation and slower adaptation of Ras, and Ric8, a guanine nucleotide exchange factor that activates the G α 2-component of G α 2 β γ [31]. The model is developed for LatA-treated cells so as to remove the feedback effect from the actin cytoskeleton on Ras, and we show that it can replicate many of the observed characteristics of Ras activation in Dicty. It is known that activated Ras activates PI3K, which stimulates further downstream steps that affect actin polymerization, but we can restrict attention to the Ras dynamics and its upstream effectors because there is no known direct feedback to Ras from downstream steps between Ras and the actin cytoskeleton. We show that Gβγ mediates adaptation of Ras activity in a uniform stimulus and transient activation in a gradient. It is also shown that Gα2 contributes to the imperfect adaptation in a uniform stimulus, and that it is an essential element for front-to-back symmetry breaking in a gradient, highlighting the important roles of Gα2 and G α 2 β γ cycling between the bound and dissociated states. We also show that Ric8 contributes to the amplification of Ras activity by regulating Gα2 dynamics: the reactivation of Gα2 by Ric8 induces further asymmetry in G α 2 β γ dissociation, which in turn amplifies the Ras activity. Finally, we investigated the effects of diffusion and cell shapes on direction sensing, and the potential role of Ric8 in the establishment of persistent Ras activation, which provides a solution to the back-of-the-wave problem.
In light of the restriction to LatA-treated cells, the backbone of the chemotactic pathway activated in response to changes in extracellular cAMP is Δ cAMP →Δ GPCR occupation →Δ Gαβγ activation →Δ Ras activity. We describe this pathway in terms of three modules: the GPCR surface receptors cAR1-4, G α 2 β γ and Ras, as illustrated in Fig 1.
Next we investigate how cells respond to a linear cAMP gradient along the x-axis, which we define as follows.
C ( x , y , z ) = Δ C 10 · ( x - x r ) + C r
where C(x, y, z) is the cAMP concentration on the membrane at ( x , y , z ) ∈ S 5 2 (a sphere of radius 5), ΔC ≡ Cf − Cr, and subscripts f and r denote the
points (5,0,0) (the ‘front’) and (-5,0,0) (the ‘rear’).
In the context of Dicty aggregation, the ‘back-of-the-wave’ problem refers to the fact that cells do not turn to follow the cAMP gradient after the wave has passed, despite the fact that the spatial gradient reverses as the wave passes over a cell [15, 62]. This requires some level of persistence of ‘orientation’ of a cell, but there is as yet no agreed-upon mechanistic solution for this problem, since polarization and other factors may play a role [63]. Under uniform stimuli, cells are said to show rectification if there is an asymmetry in the amplitude and evolution of the response to a step increase in cAMP compared with the response following removal of the stimulus [21]. To test whether the proposed network exhibits rectification in this sense, we apply a uniform stimulus of various concentrations for 60 seconds and then remove it, as was done experimentally in fully aggregation-competent cells [21]. Fig 20 (left and center) show the simulation and the experimental results, resp. In both cases the concentration of cAMP is increased from 0 M to the concentrations indicated for 60 seconds (green shaded area), followed by a decrease to 0 M, and in both cases one sees a much larger and faster change in RBD following application of the stimulus than on removal. We also applied the same stimuli as used above to gα-null cells and ric8-null cells. Results given in the Supporting Information show that Ric8 plays a significant role in the rectification, as will also be seen later in the traveling wave analysis.
Some insight into this behavior can be gained from simple models of excitation and adaptation, such as the cartoon description defined by the system of equations
d y 1 d t = S ( t ) - ( y 1 + y 2 ) t e , d y 2 d t = S ( t ) - y 2 t a . (1)
Here S(t) represents the signal and the magnitudes of te and ta reflect the time scale for excitation and adaptation, resp., and one see that y1 adapts perfectly to a constant stimulus whereas y2 compensates for the stimulus. However, the temporal responses to increasing and decreasing stimuli are symmetric, and therefore such a simple model cannot explain the observed response. Nakajima et al. [21] suggest that a single-layered incoherent feedforward circuit with zero-order ultrasensitivity [64] is necessary to generate rectification, but our model does not include an ultrasensitive circuit. Instead, rectification is induced solely by the balanced regulation of RasGEF and RasGAP activity. The ratio of RasGEF* to RasGAP* increases 2–4 fold very rapidly in response to a step increase in the cAMP concentration, but when the stimulus is removed this ratio does not drop significantly, as shown in the right panel of Fig 20. Thus Ras activation persists because the ratio equilibrates rapidly while the absolute levels of the factors decrease more slowly.
To study how cells would respond in wave-like spatially-graded stimuli, we first generate a simple trianglular wave that approximates a natural cAMP wave. Let W(x, y, z, t) denote the cAMP concentration at (x, y, z) of the cell at time t, and specify it as
W ( x , y , z , t ) = 0 , 0 + 350 k ≤ t ≤ x + 5 v + 350 k 10 ( t - x + 5 v - 350 k ) , x + 5 v + 350 k < t ≤ x + 5 v + 100 + 350 k - 10 ( t - x + 5 v - 350 k ) + 2000 , x + 5 v + 100 + 350 k < t ≤ x + 5 v + 200 + 350 k 0 , x + 5 v + 200 + 350 k < t ≤ 350 ( 1 + k ) ,
where v is the wave speed and −5 ≤ x, y, z ≤ 5, k = 0, 1, ⋯. This wave resembles a natural wave when we choose the natural wave speed v = 5μm/s, as shown in Fig 21. The wave length is 1000μm, and at the natural speed any point on a cell is subject to an increasing stimulus for 100 sec on the upstroke of the wave and a decreasing stimulus for 100 sec on the downstroke.
As shown in Fig 22, Ras is activated everywhere as the wave passes, but Ras activation is delayed about 1 sec in the rear half (Fig 22 -right) for a wave traveling at the natural wave speed. Ras activation is higher at the front of the cell than at the rear throughout passage of the wave, thereby providing persistent directionality in Ras activation and the potential for persistent orientation as the wave passes. It should be emphasized that we are simulating the rounded LatA-treated cells that have no intrinsic polarity, which suggests that polarity is not necessary for the persistence of direction sensing at the natural wave speed, even at the level of Ras activity. By comparing Figs 20 and 22, one sees a similar pattern in Ras activation. In fact, due to the rectification characteristic observed in uniform stimuli, Ras* activity does not drop significantly in a wave, and therefore the front is able to maintain a higher Ras*. To determine whether the cell is able to respond after the first wave passes, we applied the same wave for three periods, and one sees in Fig 23 that the cell responses are almost identical for three successive passages of a wave.
It is also known that wave speeds affect the spatial pattern of Ras activity over a cell [21], in that Ras is activated uniformly for a fast wave, and activated at both the wavefront and waveback for slow waves. To test the effects of the wave speed, we apply a fast wave (50μm/s) and a slow wave (0.5μm/s) to the rounded LatA-treated cells. The results are shown in Fig 24. At a wave speed of 50μm/s, Ras activation is uniform along the cell periphery, as is observed in the experiments, but at 0.5μm/s we see a significant Ras reactivation at the rear of the cell and the Ras* distribution reverses at the back of the wave.
In order to demonstrate the effect of wave speed on rectification more clearly, we plot the time course of Ras activation at the front-most and rear-most points of the cell in Fig 25. At a wave speed of 0.1μm/s, Ras is reactivated at the rear of the cell when the back of the wave passes over the rear. As the wave speed increases, the reactivation at the rear becomes weaker, and at the normal wave speed of 5μm/s persistent directionality is well-preserved. Of course, when a fast wave passes over the cell, Ras activation is almost spatially uniform.
As was pointed out earlier, Ric8 plays an essential role in rectification under uniform stimuli, and to further emphasize that the back of the wave problem is closely connected with the disparity in the response to increasing vs. decreasing stimuli, we applied the same wave used previously to a ric8-null cell. The Ras* activity is shown in Fig 26, where one sees that the persistence of directional information is essentially lost. It is not surprising to see that Ras* at the front becomes smaller than the rear, which indicates a reversal in the Ras* distribution, further reinforcing the importance of the asymmetric response to increasing vs decreasing stimuli in solving the back of the wave problem.
Clearly there is a trade-off between the persistence of directionality in Ras activation and the ability of cells to respond to new gradients. To investigate whether the Ric8-induced rectification has an adverse effect on reorientation in response to a reversed gradient, we subject cells in a 0–100 nM gradient to reversals to increasingly weaker gradients. In each case we keep the mean concentration experienced by the cell fixed to eliminate the mean concentration effect (see. Fig 14). For an equally strong reverse gradient (100–0 nM), the directional persistence of Ras* is reversed within 100 seconds of gradient reversal, as shown in Fig 27. The spatial profile also indicates that Ras* distribution is strongly reversed after switching to equally strong reversed gradients, (Fig 27 –center and right). It is observed in Dicty that all cells (20/20) reversed their direction of migration under this protocol [22]. For intermediate gradients (75–25 nM), Ras* is slightly reversed (Fig 28 –left) in the same time window (0–200 s). The spatial plot of Ras* indicates spatial oscillations along the cell periphery at almost the end of the time window t = 180 s, (see Fig. M in S1 Text) suggesting spatio-temporal complexity in Ras* redistribution. Consistently, experiments show that a fraction of the cells (5/17) did not reverse their migration direction. For weak gradients (60–40 nM) a difference in Ras activation is still maintained at the end of the time window (t = 200 s) (Fig 28 (right)), consistent with the observation that that all cells continued moving in their original direction in this case [22]. These simulations suggest that Ric8-induced rectification does not harm cells’ reorientation in response to large amplitude reversals of the gradient, but it delays the reorientation in a weak reversed gradient.
Chemotaxis is a dynamic spatio-temporal process that involves direction sensing, polarization, and cell movement, and direction sensing is the first essential step in this process, becuase it defines the cell’s compass. A growing body of evidence suggests that Ras is an ideal candidate within the chemotactic signalling cascade to play an essential role in direction sensing [31, 68]. In this article, we developed a novel modular model of direction sensing at the level of Ras activation. The model incorporates biochemical interactions in Dicty and captures many aspects of its response. The model consists of the cAMP receptor, the G-protein G α 2 β γ, and a Ras GTPase module in which both adaptation and amplification occur. Utilizing a rounded cell pretreated by LatA as was done in experiments, we investigated Ras activation patterns in various cAMP stimuli. Simulations of this model give insights into how the signal transduction network determines Ras activation characteristics in wild type cells, how an altered network in mutant cells changes Ras activation, and how the spatial profile and persistence of Ras activation can lead to directional persistence.
We proposed an experimentally-based kinetic model of G α 2 β γ signaling in which the intact G α 2 β γ and the Gβγ subunit can cycle between the membrane and the cytosol, while the G α 2 subunit remains membrane-bound. Moreover,G α 2 can be reactivated by the only known (to date) GEF for G α 2, Ric8. The regulation of Ric8 is not well-defined, but we assume that it is also cycles between the cytosol and the membrane, and that its recruitment to the membrane is promoted by G α 2 *. The model replicates the persistent Gαβγ dissociation in the presence of cAMP, and also demonstrates that Gβγ and G α 2 * are produced in a dose-dependent manner. Interestingly, the model reveals that G α 2 exhibits dose-dependent kinetic diversities. The variety of G α 2 dynamics revealed here may have important implications in direction sensing because in neutrophils G α 2-GDP accumulates at the leading edge and is involved in regulating directionality [74], although it has not been demonstrated that Ric8 is involved there.
Adaptation of Ras activity is controlled by a balance between RasGEF and RasGAP, both of which can cycle between the membrane and the cytosol. This component of the network involves incoherent feed-forward, and becuase both can cycle betweenmembrane and cytosol, can give rise to spatial asymmetry in Ras activation. Both RasGEF and RasGAP are activated at the membrane by free Gβγ, but the translocation of RasGEF from the cytosol is enhanced by G α 2 *. The proposed translocation-activation topology is able to capture the dose-dependent Ras activation and various patterns such rectification and refractoriness under uniform stimuli. It also predicts that imperfect adaptation is inevitable in wild type cells due to the asymmetrical translocation of RasGEF. Takeda et al. [28] proposed an incoherent feedforward activation model to explain adaptation of Ras activity in which RasGEF is assumed to be confined to the membrane and RasGAP diffuses in the cytosol. In our model, both RasGEF and RasGAP can diffuse in the cytosol at equal rates, and both can be recruited to the membrane and activated by Gβγ.
Direction sensing, biphasic Ras activation and signal amplification are achieved by complex interactions between the modules. The incoherent-feedforward-activation by globally-diffusing Gβγ contributes to a transient activation along the entire cell perimeter. The activation at the front of the cell (facing the higher cAMP concentration) is initially faster and stronger due to the cAMP gradient, but it provides no symmetry breaking or signal amplification since diffusion eliminates the initial Gβγ concentration gradient. This means that Gβγ does not reflect the external stimulus gradient and provides no basis for direction sensing in LatA-treated cells, although it is essential for RasGEF and RasGAP activation. It is the Ric8 regulated, membrane-bound G α 2 * that determines the symmetry breaking and signal amplification. G α 2 * creates an asymmetrical recruitment of RasGEF in a cAMP gradient, which in turn induces asymmetrical RasGEF activation, providing a basis for symmetry breaking. More importantly, Ric8 recruitment to the membrane is elevated by G α 2 *, while activated Ric8 reactivates G α 2, forming a positive feedback loop. In addition, faster G α 2 β γ reassociation at the back of the cell due to less reactivation of G α 2 there induces faster G α 2 β γ cycling. Since G α 2 β γ diffuses in the cytosol, this provides a potential redistribution of G α 2 β γ from the back to the front, which in turn results in more G α 2 * at the front, thereby forming another positive feedback loop. These two positive feedback loops generate the symmetry breaking and signal amplification of Ras activation in a cAMP gradient.
We also studied cell responses to gα2 and ric8 mutations extensively. It is predicted in numerical simulations that in the presence of uniform stimuli, adaptation of Ras activity is perfect and the maximum cytosolic RBD depletion is reduced in gα2-null cells. In a cAMP gradient, gα2-null cells fail to sense directions and there is only an initial transient Ras activation. Adaptation of Ras activity is still imperfect in ric8-null cells, but the magnitude of imperfectness is reduced as compared with wild type cells. Moreover, simulations suggest that ric8-null cells fail to sense direction when they are exposed to a shallow gradient or a steep gradient with high mean concentration, highlight the importance of Ric8 in regulating Ras activation.
In contrast to LEGI-type models, the global diffusing Gβγ does not act as an inhibitor directly in our model—instead, it induces both activation and inhibition by activating RasGEF and RasGAP respectively. Gβγ also serves as a ‘global’ activator for the pool of RasGEF* and as a ‘global’ inhibitor by creating a uniform inhibition pool of RasGAP*. Asymmetry in their localization at the membrane arises from the fact that membrane-bound G α 2 * recruits RasGEF from the cytosol, thereby creating an asymmetrical pool of RasGEF*. Hence, our model can be regarded as a local-global transitions of both excitation and inhibition with a delayed local sequestrations of excitation model, in the sense that initially both activation and inhibition go through a local-global transition due to diffusion of Gβγ while a delayed localized translocation by G α 2 * contributes to a local excitation. Direction sensing is results from the Gβγ- mediated,G α 2-Ric8 dependent signal transduction network.
Although the model is based on cAMP induced Ras activation in Dicty, GPCR-mediated Ras activation is highly conserved between Dicty and mammalian leukocytes [8]. GEF translocation via interaction with an upstream GTP-bound G protein is a principle conserved in evolution [47] and Gα’s role in GPCR-mediated signalling has been emphasized in other systems [50, 75] and in drug discovery [76]. Therefore, our model could serve as a generic framework for GPCR mediated Ras activation in other systems and suggest new experiments in those systems.
We first formulate the reaction-diffusion system of signal transduction in general terms and then list the specific equations for the model.
Consider a bounded three dimensional domain Ω ⊂ R3 representing a cell, and denote ∂Ω as the plasma membrane. Then the reaction diffusion equation for a cytosolic species A is
∂ C ∂ t = ∇ · ( D ∇ C ) + ∑ i s i r i , (2)
in which C = C(t, x) represents the concentration of A at time t at x ∈ Ω and D is the diffusion coefficient of A. The summation is a reaction term indicating A participates in cytosolic reactions which either depletes it or produces it. The ith reaction produces si molecules of A, or consumes −si > 0 molecules of A with a reaction rate ri = ri(t, x). In the signal transduction network considered in this article, si = 0, 1.
The boundary conditions involve reactions on the boundary and binding and release of molecules at the membrane. We assume that the volume density C (the concentration in the cytosol) for A has the units μM and that the surface density (the concentration on the membrane), Cm, has the units #/μm2. We also assume that the binding reactions at the membrane take place within a layer of thickness δ(nm) at the membrane. Then the net flux to the boundary, which can be positive or negative, can be written as
- n → · D ∇ C = - D ∂ C ∂ n = k + · δ · C - k - · C m ≡ j + - j - , (3)
where n → is the exterior unit normal to ∂Ω, k± are the on and off rate of binding to the membrane, and κ = 602 relates the units of volume density and surface density scaled by Avogadro’s constant.
For the membrane form of species A we have the translocation-reaction-diffusion equation,
∂ C m ∂ t = ∇ · ( D m ∇ C ) + κ ( j + - j - ) + ∑ s m i r m i , (4)
where Cm = Cm(t, x) denotes the concentration on the membrane and Dm is the surface diffusion coefficient [77, 78]. The first term represents the diffusion on the membrane, which we ignore throughout, and the second represents transolcation between cytosol and membrane, which could be absent if A is confined on the membrane, such as Ras, Ras*.
There may also be conservation laws for certain substances. If the substances are confined to the membrane we write
∫ ∂ Ω ∑ i = 1 n A n d S = A t o t , (5)
where Ais are the concentrations of different forms and Atot represents the total amount in the cell. If the substances are present both in the cytosol and on the membrane, we write
∫ Ω ∑ i = 1 k A i c d x + ∫ ∂ Ω ∑ j = 1 n A j m d S = A t o t , (6)
where Ais are the concentrations of different forms in the cytosol and A i m s are the concentrations of different forms on the membrane.
We are now ready to assemble the system of equations that constitute the full kinetic model in a given geometry Ω. We have to account for 6 cytosolic species in the system Gαβγ, c, Gβγ, c, RasGEFc, RasGAPc, Ric8c and RBDc. The evolution can be described by a system of diffusion-translocation equations
∂ G α β γ , c ∂ t = ∇ ⋅ ( D G α β γ , c ∇ G α β γ ) ∂ G β γ , c ∂ t = ∇ ⋅ ( D G β γ , c ∇ G β γ ) ∂ R a s G E F c ∂ t = ∇ ⋅ ( D R a s G E F c ∇ R a s G E F c ) ∂ R a s G A P c ∂ t = ∇ ⋅ ( D R a s G A P c ∇ R a s G A P c ) ∂ R i c 8 c ∂ t = ∇ ⋅ ( D R i c 8 c ∇ R i c 8 c ) ∂ R B D c ∂ t = ∇ ⋅ ( D R B D c ∇ R B D c )
with the following conditions on ∂Ω,
D G α β γ , c ∂ G α β γ , c ∂ n = j 1 D G β γ , c ∂ G β γ , c ∂ n = j 2 D R a s G E F c ∂ R a s G E F c ∂ n = j 5 − j 6 D R a s G A P c ∂ R a s G A P c ∂ n = j 7 D R i c 8 c ∂ R i c 8 c ∂ n = j 3 − j 4 D R B D c ∂ R B D c ∂ n = j 8 − j 9
The species that evolve on the membrane are: R*, Gαβγ, m, Gβγ, m, G α *, Gα, Ric8m, Ric8*, RasGEFm, RasGAPm, RasGEF*, RasGAP*, Ras, Ras* and RBDm. The evolution equations for these are given by
∂ R * ∂ t = r 1 ∂ G α β γ , m ∂ t = − κ j 1 − r 2 + r 7 ∂ G β γ , m ∂ t = − κ j 2 + r 2 − r 7 ∂ G α * ∂ t = r 2 − r 3 + r 5 ∂ G α ∂ t = r 3 − r 5 − r 7 ∂ R i c 8 m ∂ t = − κ j 3 + κ j 4 − r 4 + r 6 ∂ R i c 8 * ∂ t = r 4 − r 6 ∂ R a s G E F m ∂ t = − κ j 5 + κ j 6 − r 8 + r 9 ∂ R a s G A P m ∂ t = − κ j 7 − r 10 + r 11 ∂ R a s G E F * ∂ t = r 8 − r 9 ∂ R a s G A P * ∂ t = r 10 − r 11 ∂ R a s * ∂ t = r 12 − r 13 + r 14 − r 15 ∂ R a s ∂ t = − r 12 + r 13 − r 14 − r 15 ∂ R B D m ∂ t = − κ j 8 + κ j 9
The following conservation laws are also imposed:
∫ ∂ Ω ( R + R * ) d s = R t , (7)
where Rt is the total amount of receptors.
∫ Ω G α , c + G β γ , c + G α β γ , c + G α * d x + ∫ ∂ Ω G α + G β γ , m + G α β γ , m d s = G α β γ t , (8)
where G α β γ t is the total amount of heterotrimetric G protein, indicating the cell does not produce additional heterotrimetric G protein.
∫ Ω R a s G E F c d x + ∫ ∂ Ω R a s G E F m + R a s G E F * d s = R a s G E F t . (9)
Similarly, for RasGAP
∫ Ω R a s G A P c d x + ∫ ∂ Ω R a s G A P m + R a s G A P * d s = R a s G A P t . (10)
For Ras, we have
∫ ∂ Ω R a s + R a s * d s = R a s t . (11)
The parameters involved in the Receptor module are taken from the literature. We estimated the parameters in the heterotrimeric G protein module from steady state analysis (SSA) of the spatially lumped model averaged from the spatially distributed model. The parameters in the Ras module are also estimated from SSA and time dynamics of Ras activation. The detailed estimation scheme is described in the supporting information (see section Parameter estimation in S1 text). We summarize the parameters in Table 3.
|
10.1371/journal.pntd.0005776 | Phylogeography of Angiostrongylus cantonensis (Nematoda: Angiostrongylidae) in southern China and some surrounding areas | Angiostrongylus cantonensis is of increasing public health importance as the main zoonotic pathogen causing eosinophilic meningitis or meningoencephalitis, which has been documented all over the world. However, there are very limited studies about its phylogeography and spread pattern. In the present study, the phylogeography of A. cantonensis in southern China (including Taiwan) and partial areas of Southeast Asia were studied based on the sequences of complete mitochondrial cytochrome b (Cytb) gene. A total of 520 individuals of A. cantonensis obtained from 13 localities were sequenced for the analyses and grouped into 42 defined haplotypes. The phylogenetic tree (NJ tree and BI tree) revealed a characteristic distribution pattern of the four main lineages, with detectable geographic structure. Genetic differentiation among populations was significant, but demographic expansion could not be detected by either neutrality tests or mismatch distribution analysis, which implied a low gene flow among the local populations in different regions where the samples were collected. Two unique lineages of the A. cantonensis population in Taiwan were detected, which suggests its multiple origin in the island. Populations in Hekou (China) and Laos showed the highest genetic diversities, which were supported by both genetic diversity indices and AMOVA. These results together infer that the area around Thailand or Hekou in Yunnan province, China are the most likely origins of Angiostrongylus cantonensis.
| Since it was described in 1935, more than 2800 cases of the disease have reportedly been caused by A. cantonensis worldwide, primarily in tropical and subtropical regions. Despite a relevant body of research on pathology, diagnosis and treatment, little is known about the phylogeography of A. cantonensis. Since southern China is one of the endemic regions, we performed this experiment to reveal the distribution pattern of A. cantonensis in southern China based on mitochondrial Cytb data. Our results revealed a unique pattern probably shaped by the biological features of its hosts and geographical barriers, simultaneously reflecting a low gene flow among populations. Nevertheless, the connective consanguinity between some locations (Taiwan and Southeast Asia) provides new evidence of the impact on its dispersal as influenced by human activities, indicating the emerging need of an effective strategy to control this helminth. In addition to the corresponding investigation on its hosts, more attention to the situation in southwest China and Southeast Asia is suggested to facilitate the understanding of the phylogeography of A. cantonensis.
| The nematode Angiostrongylus cantonensis (Chen, 1935) is one of 21 described species in its genus, and is the zoonotic parasite responsible for human eosinophilic meningitis or meningocephalitis [1, 2]. Originally discovered in the pulmonary arteries and hearts of rats Rattus rattus and R. norvegicus [1], A. cantonensis has been isolated from a number of intermediate hosts including terrestrial and aquatic snails species such as Achatina fulica and Pomacea canaliculata, from which it moves on to a number of species that serve as paratenic hosts, such as crustaceans, monitor lizards and various frogs species. The Angiostrongylus larvae develop to the third larval stage in snails or slugs with ingestion of infected rat-feces, and the main route of human infection of this nematode is through ingestion of raw or undercooked foods containing the third stage larvae, although human beings are not the normal definitive host [3]. Thousands of cases of human eosinophilic meningitis and meningoencephalitis caused by A. cantonensis have been reported, mainly from Southeast Asia and the Pacific. Due to the deliberate or unintentional introduction of its hosts, the epidemic areas of the angiostrongyliasis have expanded to novel countries and regions including Australia and Latin America [4, 5], in an apparent correlation with the dispersal of its intermediate host, especially Achatina fulica infected with A. cantonensis [6]. Globally, increasing interest in the diagnosis and treatment of angiostrongyliasis has followed its recent spread, however, diagnosis and treatments of this potentially fatal disease are difficult because of the unfamiliarity with the distinguishing biological features of this worm, and the lack of awareness of food security in some regions [7, 8].
A more comprehensive understanding of the phylogeography of A. cantonensis may provide insight into the spread of this zoonotic pathogen, as exemplified by a few recent molecular phylogeographic studies on this nematode. Four species of Angiostrongylus spp. including A. cantonensis were distinguished based on the sequence analysis of the mitochondrial cytochrome c oxidase subunit I (COI), among which China isolates and those sampled in Thailand formed two parallel subclades [9]. Another survey based on the same gene marker COI for A. cantonensis sampled from Japan, mainland China, Taiwan, and Thailand revealed that the current geographical distribution of A. cantonensis probably reflects multiple independent origins that are likely to have been influenced by human activities [10]. Similarly, a study by Monte [11] et al. on phylogenetic relationship of COI for A. cantonensis demonstrated that some haplotypes from Brazil clustered with isolates from Asia, while the rest formed distinctly divergent clades, implying multiple origins of A. cantonensis in Brazil. Besides, Alicata [12] proposed that A. cantonensis originated in East Africa and spread worldwide with its host Achatina fulica, while Drozdz [13] considered Southeast Asia as the original source of this parasite. Thus far, no widely-accepted explanation to the spread and distribution of A. cantonensis has emerged. Nevertheless, the results shown above suggest that Asia is one of multiple probable places of origin.
Although numerous studies on genetic differentiation of A. cantonensis have been carried out recently [11, 14–18], only two of these were based on analysis of the sequences of cytochrome b (Cytb) as a means of evaluating the genetic differentiation of this parasite [17, 18]. The complete mitochondrion genome has been reported for A. cantonensis from the Chinese mainland (GenBank accession number GQ398121—complete mitochondrial genome) and from Thailand (GenBank accession number KT186242—complete mitochondrial genome), which therefore allowed us to study the phylogeography of this parasite based on variable mitochondrial gene sequences in different areas of China in the geographic range where it was initially reported. This species has not been previously investigated for its molecular phylogeography in China.
The objective of the current study is to reveal the pattern and processes of geographic distribution of A. cantonensis in China. Our analysis focused on the Cytb gene of A. cantonensis from Achatina fulica mainly collected from southern China (including sampling locations in Taiwan) and the surrounding region, with some samples from Laos (Vientiane) close to northern Thailand included.
All the samples for the present study are isolated from snails which did not concern with ethnic issue. All the experimental manipulation accorded with animal safety and ethnic rule issued by the School of Life Sciences, Sun Yat-sen University.
Specimens of Achatina fulica, identified by morphological characteristics, were captured from 13 different locations in southern China and Laos. These samples were collected in Guangzhou (GZ, 23°7', 113°15'E), Zhuhai (ZH, 22°16'N, 113°34'E), Xiamen (XM, 24°29'N, 118°05'E), Fangchenggang (FCG, 21°41'N, 108°21'E), Baise (BS, 23°54'N, 106°37'E), Hekou (HK, 24°15'N, 106°17'E), Wenchang (WC, 30°33'N, 114°19'E; Hainan Island), Senya (SY, 18°15'N, 109°31'E; Hainan Island), New Taipei (NT, 25°03'N, 121°31'E), Taichung (TC, 24°09'N, 120°41'E), Kaohsiung (KH, 22°37'N, 120°17'E), Hualien (HL, 23°58'N, 120°36'E) and Laos (LA, Vientiane, 17°58'N, 102°37'E), as presented in Fig 1 for details. The snails captured were transported to field laboratory facilities for parasite examination in a ventilated, humid container. The sampling was conducted either between 20:00 and 24:00 or 6:00 and 8:00 between July 2013 and November 2015. To examine the third stage larvae (L3), snails were individually dissected after surface cleaning. The pulmonary membrane and part of mesenterium were obtained from each snail. After treatment in a bottle with 50 ml digestive solution (2g/L 1:3000 pepsin, 0.7% HCl) [19] for 15 min, the digested samples were stirred into pieces using a blender. Individual worms were obtained under microscopy and preserved in 70% ethanol for subsequent DNA extraction.
Genomic DNA was extracted from individual worms, strictly following the protocol provided by manufacturer (TIANamp Marine Animals DNA Kit). The final product extracted from each individual worm was preserved in 50 μl TE buffer solution (10 mM Tris-HCl, 1 mM EDTA) and stored at below -20°C.
The DNA product extracted from each nematode was used as the template to amplify the complete mitochondrial cytochrome b (Cytb). Amplification was conducted by nested PCR reaction in a volume of 25 μL containing 2.5 μL of 10 × Ex Taq buffer, 1.5 mM of MgCl2, 0.2 μM of each dNTP, 1U of Ex Taq polymerase (TaKaRa, Japan), 0.2 μL of extraction product for the primary amplification, and 0.2 μl of preliminary product for the subsequent nested reaction. Each reaction was carried out in a Biometra thermal cycler (TC-96/G, BIO-DL) using the following procedure: 3 min at 94°C for denaturation, followed by 35 cycles of 30 s at 94°C, 30 s at 50–53°C for annealing, 90 s at 72°C, and a final extension at 72°C for 3 min. The primers used for amplification are listed in supplementary S1 Table. Final products were analyzed by electrophoresis on 1.5% agarose gel and visualized under UV light staining with ethidium bromide.
PCR products were purified with the UNIQ-10 Spi Column PCR Product Purification Kit (Sangon, China) and subjected to automated DNA sequencing (BGI, China) with the same primers used for nested amplification.
Nucleotide sequences were compiled and aligned in MEGA 6.0, followed by visual inspection. Nucleotide sequences were then translated into acid sequences with the invertebrate mitochondrial code to ensure that no unclear pseudogenes were amplified.
Dnasp 5.0 [20] and Arlequin 3.5 [21] were used to evaluate molecular diversity through the number of haplotypes (H), polymorphic sites (S), haplotype diversity (h), nucleotide diversity (p) and the average numbers of pairwise nucleotide differences (k). Pairwise and overall distances among haplotype sequences were calculated in MEGA 6.0 [22].
The best-fit substitution model for Cytb gene data and the gamma distribution parameter for the rate of heterogeneity among sites were determined using Modeltest 3.07 [23] based on the Hierarchical Likelihood Ratio Tests (hLRTs). The TrN model [24] of evolution with the gamma shape parameter (TrN + G) was selected for the subsequent analysis of molecular variances (AMOVA) and phylogenetic analysis.
Neighbor-joining (NJ) trees were constructed using MEGA 6.0 with Angiostrongylus costaricensis (Genbank: GQ398122) as outgroup. The genetic distances were estimated under Tamura and Nei model of substitution suggested by Modeltest. Bootstrapping was conducted with 1,000 replicates [25] as the assessment support for the NJ tree. The calculated best-fit parameters were adopted to reconstruct the Bayesian tree in MrBayes 3.2.1 [26]. Four Markov Chain Monte Carlo (MCMC) were run for 100,000,000 generations sampled every 100 generations with a burnin value of 250,000. Maximum likelihood phylogenetic tree was generated using PhyML 3.1 [27] with BIONJ methods and bootstrap analysis (1,000 replicates). The haplotype network was constructed with Popart 1.7 [28] using the median joining network (MJN) approach [29].
The AMOVA was undertaken to describe the population structure, which was implemented in Arlequin 3.5 by F-statistics at three subdivided geographical hierarchical levels: the proportions of variations among regions (FCT′), among populations within region (FSC′) and within populations (FST′) with a 5,000-times permutation to assess the significance of the covariance components associated with the different possible levels of genetic structure. Both the calculation of fixation index (FST) with 10,000 permutations and statistical significance were used to evaluate the genetic differentiation between pairwise populations. The genetic distances between haplotypes was revised under the Tamura and Nei model of nucleotide substitution with a gamma shape parameter (G = 0.3017) suggested by Modeltest. A comparison between the observed distribution frequency and the expectations under panmixia [30] was conducted as the exact test of the differentiation of haplotypes among populations, which is aimed to test the null hypothesis of population panmixia. Probabilities were estimated by permutation analyses using 10,000 randomly permuted r (populations) × k (different haplotypes) contingency tables of haplotype frequencies. All statistics described above were performed in Arlequin 3.5.
Two different methods were adopted to the historical demographic analysis. First, the frequency distribution of pairwise differences among all haplotypes (mismatch distribution) was tested under the sudden expansion model of Rogers [31]. Deviations from the estimated demographic model were evaluated by the tests of Harpending’s raggedness index [32] and the sum of squared differences (SSD) with a parametric bootstrapping approach using 10,000 replicates. The mismatch distribution of samples drawn from populations at demographic equilibrium is usually multimodal while that for samples from populations with recent demographic and distributional expansions is usually unimodal [33]. Given that mismatch distributions could be very conservative occasionally [34], both Tajima’s D [35] and Fu’s Fs [36] tests based on neutral hypothesis were carried out under coalescent simulation algorithm in Arlequin 3.5. Tajima’s D test compares two estimators of the mutation parameter θ: Watterson’s estimator θs and Tajima’s estimator θπ; significant D values are typically generated as the result of factors such as population expansion, bottlenecks and selection. In Fu’s Fs test, the number of haplotypes observed is contrasted with that expected in a random sample under the assumption of an infinite-sites model without recombination. Additionally, Fs is sensitive to demographic expansion, which generally results in a negative Fs value.
Complete Cytb gene sequences of 1110 bp were obtained from 520 individuals of A. cantonensis in samples representing 13 populations. Among these sequences, a total of 42 haplotypes were identified with 229 polymorphic sites including 192 transitions, 44 transversions and 8 indels. Except for 5 haplotypes (H1, H5, H21, H23 and H25) shared by multiple localities, all others are private to their specific populations. H1 and H5 appear in southern China and Hainan Island, while the distributions of H21, H23 and H25 are restricted to Taiwan Island. H1 was the most prevalent haplotype (194 of 520) and occupied the widest range of localities (all sites excluding Taiwan and Laos). The H21 haplotype was the most commonly detected haplotype in Taiwan Island (Fig 2).
Genetic diversity indices of all regions are presented in Table 1 with an overall haplotype diversity (h) of 0.8253±0.0138 and nucleotide diversity (π) of 0.087096±0.041454, exhibiting a high level of haplotype diversity but low nucleotide diversity. Among regions, Laos and Hekou exhibited the highest haplotype diversities, 0.7385±0.0614 and 0.7357±0.0395 respectively, and the highest nucleotide diversities of 0.154297±0.074972 were also detected for samples from Laos. Samples from Xiamen and Hekou had considerably high values of nucleotide diversities, 0.029379±0.014567 and 0.022392±0.011127 respectively.
Tamura and Nei with the gamma shape parameter (TrN + G, G = 0.301) was selected as the best-fit substitution model suggested by Modeltest to reconstruct the phylogenetic tree for Cytb haplotypes data (Fig 2). Haplotypes from 13 regions were scattered in 4 distinct clades with high support (bootstrap supports >75%): Clade A comprises haplotypes from all localities, whereas Clade B includes samples from GZ, ZH, XM, HK and LA; Clade C includes samples from GZ, HK and LA; while Clade D only consists of samples from HL and LA. Clade A was further divided into two subclades, A1 consisting of the majority of haplotypes from southern China with few from two islands (Taiwan and Hainan) and A2 including samples from GZ, WC, KH and LA. In spite of the farraginous composition in clade A, the majority of haplotypes from southern China were clustered with samples from Hainan Island and accordingly separated from haplotypes from Taiwan Island (squares displayed in Fig 2, NJ bootstrap support / Bayesian posterior probabilities as 85% / 0.95), implying a genetic divergence of haplotypes between Taiwan Island and southern mainland China. In contrast with Clade A, no significant connection between haplotype composition and sampling localities was found in the other 3 clades. Nonetheless, the distribution of haplotypes among clades was tendentious and meaningful: haplotypes form HK interspersed among Clade A1, B and C; GZ interspersed among Clade A, B, and C; LA interspersed among Clade A2, B, C and D. Phylogeny of haplotypes was more distinctly and remarkably revealed in Bayesian analysis (Fig 3). The tree topology with branch length showed the same pattern consists of 5 clades with high posterior probabilities and bootstrap support. Noteworthy, neighbour-joining as well as Bayesian analysis revealed the strongly supported monophyletic Clade D.
The NJ tree for partial Cytb sequences with additional haplotypes from Thailand obtained from GenBank under accession number KP721442—KP721453 was constructed using the same parameters (TrN + G, G = 0.301). All the haplotypes from Thailand were interfused into Clade A2, forming a distinctive clade with haplotypes from Laos, Hainan and Taiwan regions, as shown in Fig 4. This strongly suggests the close relationship of haplotypes from Taiwan and Thailand.
The network analysis similarly revealed that the haplotypes nested into 5 clusters (Fig 5) corresponding to the clades obtained from phylogenetic tree analysis. Clade D was confirmed to be significantly distinguished with other clades by the branch length and mutational steps between them. Additionally, a cluster corresponding to Clade A2 appeared to be closer to Clade D than that to Clade A1 as displayed in the phylogenetic tree.
Genetic differentiation among populations (Fst) was summarized in Table 2. All except five pairwise comparisons of Fst revealed significant differences (P<0.05), implying the existence of significant population structure across the range investigated. These results were further confirmed by hierarchical AMOVA tests which attributed nearly 50% of the genetic variation to the variabilities among populations within groups (Table 3). Besides, the between-group differences for the two groupings accounted for less variability than those between-populations in groups, which was similarly revealed by exact test. Hence, the null hypothesis was not applicable, or in other words, these results taken together suggest that populations of A. cantonensis in southern China are not panmictic.
Despite a lack of significance of the goodness of fit test (HRI, P > 0.05) of the distribution from that expected under the expansion model (Table 4), the mismatch distribution of pairwise differences among regions exhibited irregular multimodal patterns rather than the unimodal pattern generally produced by populations that have experienced demographic expansion (Fig 6). The results of neutrality tests were consistent with the observed mismatch analyses. Neither Fs statistic of all regions nor Tajima’s D statistic of most regions revealed any significant difference (P > 0.05) from that under neutral assumption. These results lead us to conclude that populations of A. cantonensis in the range studied were inconsistent with stable demographic history, although no strong evidence emerged in support of episodes of expansion.
Because this parasitic species is associated with severe tropical diseases, extensive research has been conducted on A. cantonensis, particularly regarding clinical pathology, epidemiology and diagnosis [37–40], however, studies on its spread mechanism and genetic variation have obtained relatively little attention. Cytb nucleotide sequence has been adopted for the phylogenetic studies of A. cantonensis [17, 18]. This gene sequence has also been accepted as good marker to reveal phylogeographical patterns [41, 42].
From 520 Cytb sequences (1110 bp) of A. cantonensis sampled from 13 geographic regions, a total of 42 haplotypes were identified. The intraspecific diversity of haplotypes among all regions ranged from 0.1 to 14.0% in p-distance with an overall mean value of 6% (Tamura and Nei distance). High level values were generated between Clade D (H33, H36, H37, H38) and other clades (haplotypes) as 12.8 to 14.0%, while those of haplotypes within other three clades only ranged from 0.1 to 6.8% (S2 Table). This was corroborated by the phylogenetic tree in which these four distinct haplotypes (H33, H36, H37, H38) were grouped into a monophyletic Clade D. The complete mitochondrial genomic sequence of A. cantonensis collected from Thailand also demonstrated that A. cantonensis from Thailand is distinctly isolated from the population in China with p-distance of 11.6% [43]. Newly published results have revealed the existence of two dramatically divergent lineages of A. cantonensis in Thailand based on mitochondrial and nuclear sequences data with an average of 11% p-distance, which is in agreement with our results, since the sample site in Laos is adjacent to Thailand [44]. Such remarkable intraspecific divergence is worthy of further attention.
Numerous species of rodents serve as the definitive hosts of A. cantonensis, and these organisms are capable of highly promoting the spread and intraspecific transfer of this parasite. The genetic structure of populations from different regions was still different from those under random mating in some way. By contrast, owing to the high dispersal potential and lack of geographic barrier in the marine environment, marine organisms normally do not display evidence of genetic differentiation throughout wider geographic range and consequently their parasites typically exhibit a similar genetic pattern [45]. The phylogeography of Pseudokuhnia minor, a species of monogenean on the host of chub mackerel Scomber japonicus, exhibited no significant genetic structure along the coast of China, implying panmixia within the range of the population [46]. As the parasitic organisms depend on their hosts, the biological characteristics of their hosts can profoundly impact their population genetic structure, in addition to their own dispersal ability [47]. Although A. cantonensis has many definitive and intermediate hosts, these hosts have limitations in traversing habitat barriers, such as water for Achatina fulica and low temperature on the high mountains for both Achatina fulica and Pomacea canaliculata. Inevitably, the definitive host rodents have complex impacts on the population genetic structure of A. cantonensis because of their wide adaptation in terrestrial habitats, which is in need of further study.
Notably, haplotypes from Hualien (HL) and Laos (LA) generated Clade D which substantially diverges from other clades, implying the population from Hualien is a unique lineage in Taiwan. The exact test and low level of genetic diversity of population from Hualien also revealed the absence of gene flow between Hualien population and populations in other regions. Another survey on the genetic diversity of A. cantonensis in Taiwan based on partial COI sequence data also revealed the existence of a distinct strain of A. cantonensis in Hualien, which is distinguished from strains in other regions in Taiwan by not only the genetic distance, but also discrepancies in infectivity and pathogenicity [48]. Hence, we inferred that the separation between western and eastern Taiwan by the central mountain range presents a substantial geographical barrier; this, combined with the limited propagative capability of hosts simultaneously shaped the unique distribution pattern of A. cantonensis in Taiwan.
The unique lineage of the Hualien strain showed a closer relationship with lineages from Laos (p-distance = 6% - 7%) indicating that they might share a common origin, despite the thousands-miles separation by the Pacific Ocean between these two locations. This unusual pattern may be attributed to the worldwide dispersal of the hosts of this parasite as influenced by human activities. With regard to the spread of the widely introduced and invasive land snail species around the world, Achatina fulica, it was proposed that numerous impolitic introduction and activities of the Japanese army during World War II figured importantly in establishing its range [49, 50]. Phylogeographic study on the introduced Rattus rattus in the western Indian Ocean islands also revealed effects of human-mediated colonization [51]. The inference that A. cantonensis has established global-scale dispersal with various organisms/hosts or vectors that are largely influenced by human transportation has gained wide acceptance [10, 11, 17]. Besides the situation in Hualien, the association between A. cantonensis populations along the west coast of Taiwan Island with those in Thailand are also supported by the reconstructed phylogeographic tree (Figs 2 and 4). In conclusion, it could be inferred that Southeast Asia is a likely origin of A. cantonensis in Taiwan, but with variable and independent introductions.
In the study of invasion biology, we generally consider introduced populations as having low genetic variability [52, 53]. This could be observed in the survey on population genetics of invasive American bullfrog Lithobates catesbeianus that genetic diversity was greatly reduced in colonizing populations due to demographic bottlenecks [54]. Hekou and Laos are two sampling locations where the A. cantonensis populations have significant high levels of genetic diversity (Table 1). It can generally be deduced that A. cantonensis has higher genetic diversity in its origin area than the diversity seen among more recently established populations with regard to the fact that its host Achatina fulica is a notorious invasive snail species. Moreover, these two locations were highlighted since phylogeny reveals multiple relationships between populations from these two sites and other sites as displayed in Figs 2 and 3. These results therefore support our inference that Hekou and Laos are the most likely origin of A. cantonensis in our study region, and further speculation that Southeast Asia might be a potential site of origin in Asia. Although the A. cantonensis population in KH (Taiwan) was clustered together with those from Thailand and Laos (Fig 4), the populations along west coast of Taiwan have close association with those in southern China (Clade A1), the latter were also closely linked to A. cantonensis in Hekou (Clade B). Hence, it could be proposed that Hekou is more likely the site of the origin of A. cantonensis in southern mainland China. It is still unknown how the population of A. cantonensis in Hekou may have been genetically connected to populations in Laos or Thailand, an understanding that may require further insight into the potential interaction of A. cantonensis populations between these locations.
With regard to the notable high genetic diversities of A. cantonensis populations in Guangzhou (GZ) and Xiamen (XM), it is likely that these port cities have a high probability of accepting the A. cantonensis carrying organisms, as it has been observed that epidemics frequently break out initially in port areas [55, 56]. It is also potentially relevant that multiple introductions of A. cantonensis from diverse regions have increased the genetic diversity in these port cities, at least in part as a consequence of human activities that influence the distribution of A. cantonensis.
Many relevant studies based on mtDNA sequence could reveal patterns of demographic history or association with known historic events [57–61], however, the present study based on a single mtDNA locus can only partially reveal the demographic history of this nematode, since the intricate life cycle of this parasite involves many species of snails as intermediate hosts and numerous rodents as definitive hosts. Besides, the limited documented records about its ecology and distribution also restrict our capacity to retrace its origin and spread. For better understanding of its phylogeography, further surveys concentrating on the meta-population of A. cantonensis in different species of hosts and based on more molecular markers will be required.
The remarkable genetic differentiation between isolations indicated a low gene flow among the populations of A. cantonensis in different areas of Southern China. The dispersal via human activities across ocean, coupled with the natural spread of its hosts might have led to the establishment of several separated lineages in Taiwan Island. Additionally, no significant indication of demographic expansion was detected in the scope of survey, although significant genetic variation of A. cantonensis was found between southern China and Southeast Asia. The structured pattern of phylogenetic lineages has no precise correspondence to geographic distribution, underscoring the complicated nature of the task of tracing the population dynamics of this species. More samples from Southeast Asia and further survey of its host organisms together could provide more detailed evidence, leading to a higher degree of resolution in our understanding of the phylogeography of this parasite.
|
10.1371/journal.pgen.1005941 | Structural and Genetic Studies Demonstrate Neurologic Dysfunction in Triosephosphate Isomerase Deficiency Is Associated with Impaired Synaptic Vesicle Dynamics | Triosephosphate isomerase (TPI) deficiency is a poorly understood disease characterized by hemolytic anemia, cardiomyopathy, neurologic dysfunction, and early death. TPI deficiency is one of a group of diseases known as glycolytic enzymopathies, but is unique for its severe patient neuropathology and early mortality. The disease is caused by missense mutations and dysfunction in the glycolytic enzyme, TPI. Previous studies have detailed structural and catalytic changes elicited by disease-associated TPI substitutions, and samples of patient erythrocytes have yielded insight into patient hemolytic anemia; however, the neuropathophysiology of this disease remains a mystery. This study combines structural, biochemical, and genetic approaches to demonstrate that perturbations of the TPI dimer interface are sufficient to elicit TPI deficiency neuropathogenesis. The present study demonstrates that neurologic dysfunction resulting from TPI deficiency is characterized by synaptic vesicle dysfunction, and can be attenuated with catalytically inactive TPI. Collectively, our findings are the first to identify, to our knowledge, a functional synaptic defect in TPI deficiency derived from molecular changes in the TPI dimer interface.
| Glycolysis is the metabolic pathway that cells use to break down the sugar glucose, and mutations in the genes that control the glycolytic pathway elicit a collection of diseases known as glycolytic enzymopathies. Glycolytic enzymopathies are rare genetic diseases that lead to the degeneration of patient red blood cells. Triosephosphate isomerase is a gene that encodes a part of this glycolytic process, and patients with mutations in this gene experience the typical blood disorder, as well as severe neurologic dysfunction and often infant death. Until now, a molecular source of neurologic dysfunction in triosephosphate isomerase mutants was unknown. We have discovered that mutations that disrupt the self-association of the triosephosphate isomerase enzyme lead to neurologic dysfunction in fruit flies. This neurologic dysfunction is characterized by the abnormal cycling of neuronal vesicles that contain neurotransmitters. Given the evolutionary conservation of triosephosphate isomerase and neuronal synaptic function, we believe that these observations represent the dysfunction seen in human patients.
| Triosephosphate isomerase (TPI) is a glycolytic enzyme that converts dihydroxyacetone phosphate (DHAP) into glyceraldehyde-3 phosphate (GAP). TPI is a non-linear member of the glycolytic pathway, enhancing the efficiency of the catabolic process, and several missense mutations within TPI lead to a disease known as TPI deficiency [1]. TPI deficiency is one member of a group of disorders caused by mutations in glycolytic enzymes, collectively called glycolytic enzymopathies. Glycolytic enzymopathies are largely characterized as blood disorders, with all patients experiencing hemolytic anemia [2,3]. TPI deficiency is one of the few glycolytic diseases associated with patient neurologic dysfunction, and by far the most severe [1,4]. Clinical examinations of TPI deficiency patients have established that this disease is often characterized by episodic seizures, periodic dystonia, and progressive weakness and flaccidity in extremities [5–11]. Cellular studies have centered on erythrocytes and lymphocytes, leaving it unclear how TPI molecular dysfunction influences the nervous system. Further, the absence of neurologic dysfunction in many other glycolytic enzymopathies has made it unclear whether these symptoms are related to glycolysis or an as-yet-unidentified function of TPI.
Several reports have suggested that TPI deficiency is a disease caused by changes in protein conformation rather than metabolic defects [4,12–14]. A human protein structure and yeast genetic studies have asserted that defects in TPI dimerization are the primary determinants of pathology [15,16], revealing no catalytic defects in vitro or from cell lysate. However, not all TPI deficiency mutations lack catalytic defects. An erythrocyte study examining two Hungarian brothers with identical TPI alleles revealed equivalent reductions in TPI activity in both individuals [17], yet one exhibited severe neurologic dysfunction and the other was asymptomatic. Further, a recent structural study demonstrated that the hTPII170V substitution significantly altered enzyme kinetics and protein stability through a molecular alteration near the catalytic pocket [18]. Collectively, each of these studies failed to establish a causal relationship between TPI activity and disease.
Previously, a pathogenic substitution in Drosophila TPI (dTPIM80T) was identified, eliciting mechanical- and thermal-stress dependent paralysis [19,20]. These behavioral phenotypes have been independently established as hallmarks of neurologic dysfunction and each has been used in forward genetic screens to identify novel components of neuronal transmission [21–23]. The dTPIM80T allele was identified in such a screen [20,24], and to-date D. melanogaster is the only model organism to exhibit neurologic dysfunction caused by TPI deficiency. The dTPIM80T protein was found to be prematurely degraded with reduced catalytic activity [25,26]. This reduction in catalytic activity was shown to inhibit glycolytic flux as well as induce metabolic stress [19,25], yet did not change ATP levels in vivo [20]. Subsequent studies demonstrated that the dTPIM80T point mutation could be complemented by the addition of a catalytically inactive TPI without increasing lysate isomerase activity or alleviating metabolic stress [25], suggesting that dTPIM80T may elicit pathology through a change in protein conformation. Thus, we initially examined the molecular source of dTPIM80T pathology.
To determine whether Drosophila TPI deficiency was caused by changes in protein conformation, we purified and assessed the physical characteristics of hTPIM82T, the human equivalent of dTPIM80T, and revealed impaired TPI dimerization. These results were further supported when independent alleles bearing mutations at the TPI dimer interface phenocopied the Drosophila behavioral dysfunction seen in the dTPIM80T allele. These experiments provided novel insight into the pathogenesis of TPI deficiency leading to the conclusion that alterations of TPI dimerization are sufficient to elicit neuropathology.
Defining the molecular source of TPI neurologic dysfunction led to the generation of new alleles containing dimer-interface mutations. These novel TPI alleles were characterized by extreme behavioral defects, directing new investigations into the neuropathogenic mechanism of TPI deficiency. An examination of vesicle dynamics at the larval neuromuscular junction (NMJ) revealed a severe impairment that appears to be related to vesicle recycling. Further, complementation with a TPI allele encoding catalytically inactive TPI rescued both synaptic dysfunction and behavior, thereby characterizing a cellular mechanism of TPI deficiency neuropathology.
Collectively, the results of this study support the conclusion that an improperly formed dimer interface is sufficient to elicit TPI deficiency neuropathology. Further, our experiments establish that a functional synaptic defect occurs in our Drosophila model of TPI deficiency.
TPI is a homodimeric enzyme with catalytic sites in the C-terminal of a triose isomerase (TIM) barrel tertiary structural motif [27]. Each catalytic site is rigidified through dimerization to increase catalytic turnover, yet each active site works independently [28–30]. A dTPIM80T substitution was previously isolated and demonstrated to elicit pathology in a Drosophila model of TPI deficiency [19,20]. The dTPIM80T substitution is physically located in a solvent-exposed region of the protein near the dimer interface [25]. Numerous misfolding events could be hypothesized to occur as a function of the TPIM80T substitution, among them alterations of dimerization [15,16] and aggregation [31]. To examine the structural change elicited by M80T in vitro we purified Drosophila dTPIM80T.
Previous purification experiments had yielded Drosophila TPI enzyme, but these samples proved aggregation prone at high concentrations. Conversely, purified human TPI (hTPI) was well-behaved; therefore, in order to physically characterize TPI we studied the human protein in vitro. To validate the use of human protein in vivo we generated human WT (hTPIWT) and human M80T (hTPIM82T) alleles in the Drosophila TPI gene locus using an established genomic engineering (GE) system [25]; the hTPIM82T substitution is equivalent to dTPIM80T (Fig 1A). We found that hTPIM82T was able to recapitulate the disease phenotypes observed in dTPIM80T (S1 Fig). The phenotypes of hTPIM82T were remarkably similar but less severe than dTPIM80T, possibly due to subtle organism-specific changes in the dimer interface [32]. Confirmation that hTPIM82T pathologically phenocopied dTPIM80T indicated that any conformational change elicited by dTPIM80T was likely retained in the human protein.
We utilized dynamic light scattering (DLS) to examine potential conformational differences between hTPIWT and hTPIM82T. Analyses of 15 μM solutions of hTPIWT revealed a hydrodynamic radius of 4.3±0.08 nm, while hTPIM82T exhibited a significant reduction to 3.3±0.06 nm (Fig 1B); these results were consistent across two additional protein concentrations, 3.75 μM and 30 μM (Fig 1B). The linear slope generated by plotting the intensity correlation data suggested the hTPIWT sample was largely monodispersed, much like that of the 15 μM sample of bovine serum albumin (Fig 1C). Conversely, hTPIM82T samples exhibited a non-linear slope (Fig 1C), suggesting the possibility of a polydisperse protein population. Polydisperse protein populations indicated the sample was a mixed population in solution, and the observed reduction in TPI mean hydrodynamic radius suggested the sample could be a mixture of monomer and dimer TPI species.
We examined enzyme dimerization by assessing protein size via gel filtration chromatography. A standard curve was used to establish column resolution, 15 μM samples of hTPIWT and hTPIM82T were injected onto the gel filtration column, and their migration monitored by UV light at 280 nm. hTPIWT samples separated into two distinct peaks–one at ~24 min. and another at ~27 min. corresponding to ~50 kDa and ~28 kDa, respectively (Fig 1D). Dimeric and monomeric hTPI are 54 and 27 kDa, respectively. Integrating the peak areas revealed an ~80:20 split in dimer:monomer ratio of hTPIWT (Fig 1D inset). In contrast, the majority of the hTPIM82T sample eluted at 27 min., resulting in a ~5:95 dimer:monomer ratio (Fig 1D inset). These data led us to conclude that the hTPIM82T substitution elicited a dramatic conformational change in TPI resulting in a disruption of dimerization. Interestingly, the gel filtration results did not precisely reflect the monodisperse vs. polydisperse observations of the DLS experiments; we believe this could be due to dilution effects as the proteins migrated over the large gel filtration column.
Having established that the hTPIM82T mutation alters enzyme dimerization in vitro, we sought to assess whether other substitutions at the TPI dimer interface were sufficient to elicit neuropathology. Two novel TPI alleles (dTPIT73R and dTPIG74E) were generated using GE. Our hTPI dimer analyses (S2 Fig) and data from previous TPI studies [28] indicated these substitutions would result in dimer defective TPI. dTPIT73R and dTPIG74E dimer interface mutants elicited a more severe pathology than dTPIM80T, and stocks required maintenance over balancer chromosomes due to their poor viability. Test crosses of balanced stocks yielded significantly fewer homozygous animals than the Mendelian predicted 33%, and homozygous animals were extremely short-lived (Fig 2A), with median lifespans of 2 and 5 days for dTPIT73R and dTPIG74E, respectively.
Mechanical- and thermal stress-dependent behavioral defects were assessed at Day 1 and Day 2, respectively, as these phenotypes have been demonstrated to be hallmarks of Drosophila TPI deficiency [19,20,25,26,33,34]. dTPIM80T was previously described to exhibit a modest phenotype at early time points [19], and these data corroborate our analyses of the GE dTPIM80T allele (Fig 2B and 2C). Comparatively, the dimer interface mutants displayed a more severe degree of behavioral dysfunction than that seen in dTPIM80T (Fig 2B and 2C). These data support the hypothesis that mutations at the dimer interface are sufficient to induce neurologic dysfunction.
Lysate isomerase activity was compared between samples taken from animals homozygous for the dimer interface mutants. First, it was noted that all dimer interface mutants exhibited reductions in TPI activity (Fig 2D). However, a comparison of the dTPIM80T, dTPIT73R, and dTPIG74E lysates revealed a striking observation–the least phenotypically severe mutation (dTPIM80T) was characterized by the lowest isomerase activity (Fig 2D). These data support previous observations that TPI activity does not predict the presence or severity of TPI deficiency [25].
Many conformational diseases are elicited through changes in protein structure and stability leading to misfolding, then either sequestration and degradation, or aggregation [35]. First, we examined whether these new dimer interface alleles produced robust levels of TPI protein. We determined TPI levels in our dimer interface mutants as previously [34], and found that both dTPIT73R and dTPIG74E homozygotes exhibited reduced protein levels (Fig 3A and 3B).
It has previously been shown that TPI has the capacity to aggregate and thereby seed the aggregation of other proteins such as tau [31]. When measuring protein levels via SDS-PAGE, it is important to note that not all aggregate species are SDS soluble and a reduction in protein levels can indicate that the aggregates are not passing through the gel matrix. To determine whether the dTPIT73R and dTPIG74E proteins aggregate we used a dot-blot filter trap assay to assess retention differences between TPI mutant isoforms, as performed previously [36]. Lysates were collected from homozygous animals, and PC12 cell lysates expressing EGFP-huntingtin-Q97 (GFP htt-Q97) were used as a positive aggregation control. The results indicate that little TPI was trapped on the 200 nm filter, yet each sample showed a concentration-dependent increase in signal (Fig 3C). Importantly, no differences were observed in TPI signal between the WT and mutant alleles (Fig 3C). These data support similar findings established by sedimentation assays performed on dTPIM80T [34], and led us to conclude that although these dimer interface mutants display reduced protein levels via SDS-PAGE, this is not due to the insolubility of large aggregates.
To date, all but one study examining TPI deficiency in Drosophila have highlighted a reduction in TPI protein levels in disease-associated alleles [18–20,25,26]. To independently examine the importance of TPI protein levels in vivo, we employed the GAL4-UAS expression system to knock down wild type (WT) TPI using a UAS-RNAi line directed toward dTPI messenger RNA (mRNA) [37]. These lines were driven with actin-GAL4 + UAS-GAL4 (actin/UAS-GAL4) to obtain a dramatic reduction of TPI in all tissues.
Using UAS-RNAi in conjunction with actin/UAS-GAL4, we found that w;actin-GAL4,UAS-GAL4/+;UAS-RNAiTPI/+ animals exhibited a dramatic reduction in TPI protein levels similar to that seen in head and thorax tissue from w;;dTPIM80T homozygotes (S3A and S3B Fig). Head and body tissues were assessed separately to ensure equivalent knockdown in both tissues. Next, we examined animal behavior in these knockdown populations to determine whether depletion of cellular TPI was sufficient to elicit TPI deficiency behavioral abnormalities. Mechanical stress responses were used to quantify behavioral dysfunction. None of the knockdown genotypes exhibited abnormal mechanical-stress dependent responses (S3C Fig). No paralysis or seizure-like activity was observed in the knockdown genotypes at elevated temperatures, though hypoactivity was noted, with the knockdown animals consistently dwelling near the bottom of the vial relative to their TPI+ and UAS only controls. These observations suggested that a general depletion of TPI is not sufficient to elicit paralysis or seizure-like locomotor dysfunction, yet do not exclude the possibility that changes in protein conformation, localized subcellular depletions, or changes in protein stability may play a greater role in animal pathology.
Previous work demonstrated that a catalytically inactive allele of TPI (dTPIΔcat) complemented the behavior and longevity defects of the dTPIM80T allele [25], a mutation now established to disrupt enzyme dimerization. This previous study suggested that TPI deficiency is a loss-of-function disease caused by either i) the depletion of cellular TPI, or ii) a conformational change that could be rescued through the addition of a properly folded yet catalytically open/inactive isoform [25]. Having utilized knockdown strategies to examine the necessity of total TPI levels, we sought to confirm the capacity of dTPIΔcat (Lys-to-Met, position 11, Fig 1A) to complement additional dimer-interface mutations. To evaluate whether dTPIΔcat was sufficient to support normal behavior and longevity, dTPI+/dTPI+, dTPI+/dTPIT73R, dTPIT73R/dTPIT73R, dTPIT73R/dTPIΔcat, dTPI+/dTPIG74E, dTPIG74E/dTPIG74E, and dTPIG74E/dTPIΔcat animals were collected and tested as outlined above. These experiments demonstrated that the dTPIT73R allele was nearly fully complemented by dTPIΔcat (Fig 4A–4C), similar to the results found with dTPIM80T [25]. It should be noted that this complementation was not fully penetrant; 5 out of the 30 dTPIT73R/dTPIΔcat animals did paralyze after an extended thermal stress period (Fig 4B). The penetrance of the thermal stress complementation is reflected in an increased time to paralysis relative to the homozygous mutant animals (Fig 4B and 4E). dTPIG74E was also complemented by dTPIΔcat, although more modestly than was observed for dTPIT73R. Mechanical stress responses were unchanged in dTPIG74E/dTPIΔcat relative to dTPIG74E homozygotes, though the penetrance of thermal stress sensitivity was decreased to 20 out of 30 animals, and the median lifespan of the dTPIG74E mutants was extended from 5 to 21 days (Fig 4D–4F). Importantly, neither of the dimer interface mutants elicited dominant negative effects within the dTPI+ heterozygotes; to the contrary, dTPIT73R and dTPIG74E promoted a significant increase in animal health, extending the median 48 day dTPI+/dTPI+ lifespans to 77 and 71 days, respectively (Fig 4C and 4F). In toto, dTPIΔcat partially but significantly complemented each of the new TPI dimer alleles.
Our experiments with TPI dimer mutations demonstrated that alterations of the dimer interface were sufficient to elicit TPI neurologic dysfunction, and that these phenotypes were able to be complemented with dTPIΔcat. To address how the TPIΔcat substitution may influence its structure, we purified, crystallized, and determined the structure of hTPIΔcat at 1.7Å resolution, refining against native data to Rwork and Rfree values of 15.8%, and 19.6%, respectively (Table 1). These crystals grew in conditions that were nearly identical to conditions in which we have previously determined the structure of wild-type human TPI [18], minimizing the effects that changes in the crystallization condition or crystal packing might have on the resulting structure. While the overall fold of hTPIΔcat is highly similar to wild-type (r.m.s.d of 0.35 Å over all atoms) there are a number of important differences within the catalytic pocket and neighboring regions. First, the active site pocket of our previous hTPIWT structure contained a highly ordered phosphate and bromide ion located where the phosphate and triose groups of the natural substrate, DHAP, would be located [18]. In contrast, the active site pocket of hTPIΔcat was filled with solvent. At the site of the hTPIK13M substitution (hTPIΔcat), the M13 side chain adopts a different conformation than its lysine counterpart, shifting 4 Å away from the catalytic site and interacting with N11, G233 and L236 at the back of the pocket (Fig 5A). The sidechain positions of important active site residues S96 and E165 are also altered in hTPIΔcat, breaking critical solvent networks and shifting E165 2.7 Å away from the position it adopts in wild-type TPI [18] and substrate analog bound structures [38–40] (Fig 5A). Lastly, the lid moves as much as 7 Å away from the active site pocket, adopting an open conformation [40,41] (Fig 5A) and corroborating established kinetic data demonstrating that this enzyme is catalytically inactive [42]. These data are in agreement with a structure of yeast TPIK12M,G15A containing two mutations within the active site [43], but were an important control to isolate the structural impact of hTPIK13M. Importantly, examinations of the dimer interface of hTPIΔcat revealed that it is unchanged relative to hTPIWT (Fig 5B). The peptide backbone and side chains of Loop3 form the majority of the TPI dimer interface, and as shown previously, perturbations of this loop disrupt TPI dimer stability [28–30,44]. The new crystal structure revealed that the backbone of Loop3 along with important side chains M14, T75, G76, M82, and E104, are unaltered in hTPIΔcat (Fig 5B). These structural data indicated that TPIΔcat homodimers are catalytically inactive, with no observable alterations of the overall folding of the monomers or their dimeric assembly (Fig 5B).
Complementation of the TPI dimer mutant alleles with dTPIΔcat suggested a physical and/or functional interaction between the two enzymes. Given the dimeric nature of TPI, we sought to first examine physical interactions between TPI species. All of the dimer-interface substitutions used in this study disrupt homodimerization (Figs 1 and S2), though no experiments had yet addressed how these alterations may change heterodimerization with dTPIΔcat. These putative heterodimers could support or inhibit a critical function of TPI.
To examine heterodimer formation in vivo, we measured the capacity of the dimer-mutant TPI isoforms to co-precipitate using a C-terminal Cerulean cyan fluorescent protein (CFP) tagged variant of dTPIΔcat-CFP; this allele was previously confirmed to complement dTPIM80T [25]. anti-GFP was covalently conjugated to the AminoLink resin, and the CFP tag was immunoprecipitated (IPed) in dTPIWT/dTPIΔcat-CFP, dTPIM80T/dTPIΔcat-CFP, dTPIT73R/dTPIΔcat-CFP, and dTPIG74E/dTPIΔcat-CFP animal lysates and probed. Unconjugated resin was incubated with dTPIWT/dTPIΔcat-CFP lysate and used as a negative control (-) (Fig 6A, IP). Upon elution and SDS-PAGE separation, protein size was used to discriminate between the tagged and untagged TPI isoforms; the CFP tag roughly doubled the molecular weight of dTPI-CFP monomer (~50 kDa) relative to dTPI monomer (~25kD) (Fig 6A, Input).
Robust amounts of dTPIWT precipitated with dTPIΔcat-CFP, establishing substantial heterodimerization between the two species (Fig 6A and 6B) in agreement with the similarities between their respective dimer interfaces (Fig 5B). Conversely, dTPIM80T and dTPIT73R displayed markedly reduced associations with dTPIΔcat-CFP, corroborating their previously established dimerization deficiencies and reflecting their overall prevalence in the lysate (Fig 6A). Finally, it was surprising to see that dTPIG74E produced heterodimerization similar to that seen in dTPIWT (Fig 6A and 6B); it was predicted that the rotational flexibility of G74 was necessary for the appropriate positioning of loop 3 and establishment/rigidification of the dimer interface [28].
The coIP experiments suggested that the TPI species responsible for dTPIT73R phenotype suppression in the animals was not a dTPIΔcat heterodimer; the heterodimer was a very small fraction of the total TPI enzyme in lysate (Fig 6). Conversely, the dTPIΔcat-CFP::dTPIG74E heterodimer existed as a substantial fraction of the total TPI (Fig 6), yet exhibited modest complementation of the abnormal behavioral phenotypes (Fig 3). The substantial and unanticipated presence of the dTPIΔcat-CFP::dTPIG74E heterodimer could indicate an allele-specific dominant interaction. To examine whether we could enhance the capacity of dTPIΔcat to suppress dTPIG74E, we designed a double-mutant aiming to revert heterodimer formation.
An allele was generated bearing both substitutions, dTPIT73R,G74E, and animals homozygous for this allele displayed aggressive behavioral phenotypes and shortened lifespans (S4A, S4B and S4D Fig). Immunoprecipitation experiments found that the double substitution reduced dTPIT73R,G74E heterodimerization with dTPIΔcat-CFP relative to dTPIG74E (Fig 6). Finally, when paired with the dTPIΔcat allele, the addition of the T73R substitution to dTPIG74E enhanced the capacity for dTPIΔcat-behavioral complementation (S4A and S4B Fig). The mean time to recovery after mechanical stress was reduced from 204 sec. in dTPIT73R,G74E homozygotes to 50 sec. in dTPIT73R,G74E/dTPIΔcat animals with approximately 60% of the animals no longer responding to the stressor (defined as a recovery time ≤ 5 sec.) (S4A Fig); similar complementation was observed in the thermal stress assay (S4B Fig). Curiously, the longevity of the dTPIT73R,G74E/dTPIΔcat animals was unchanged relative to dTPIT73R,G74E homozygotes (S4D Fig); this is the second time that TPI deficiency behavioral abnormalities and longevity have not paralleled each other [18], suggesting the possibility of independent pathogenic mechanisms (see Discussion).
The inverse correlation between dTPIΔcat heterodimerization and behavioral complementation suggested that dTPIΔcat did not complement TPI deficiency behavioral phenotypes via heterodimer formation. Further, disease severity did not correlate with isomerase activity (S5B Fig); complementation of the dimer-mutant alleles with dTPIΔcat failed to increase isomerase activity and in all but one case significantly decreased activity (S4C and S5 Figs). These data led us to conclude that dTPIΔcat does not “suppress” TPI deficiency behavioral phenotypes through a general influence on TPI catalytic activity.
TPI deficiency complementation was not corroborated by an enhancement of TPI catalysis; however, the mean temperature-dependent time to paralysis of the dTPIT73R allele (27 sec) was a striking result (Fig 2B) and suggested a previously unknown role of TPI. Rapid (<60 sec.) temperature-dependent paralysis had only been identified in a handful of mutants in Drosophila and typically results from neural conductance or synaptic vesicle recycling defects [45]. To determine whether TPI was influencing vesicle dynamics, we first examined vesicle endocytosis at the synapse using the lipophilic dye, FM1-43. FM1-43 is a water soluble membrane dye that increases its fluorescence when bound to cellular membranes. During endocytosis, the dye will bind to the outer leaflet of the plasma membrane and become internalized within the synapse providing an optical measurement of endocytosis. Measuring vesicle dynamics in this context allowed us to assess two possibilities; i) a primary recycling defect due to impaired endocytosis, or ii) a secondary recycling defect due to aberrant exocytosis.
We dissected larvae homozygous for dTPIWT, dTPIT73R, and Shits1 as previously detailed [46]. The NMJ preparations were heated to 38°C over 3 min. and a loading curve was generated from a series of three different high [K+] + FM1-43 loading times– 15 sec., 30 sec., and 60 sec. as previously detailed [47]. dTPIWT displays a progressive increase in dye loading from 15 sec. to 60 sec. (Fig 7A), while the temperature sensitive dynamin mutant control Shits1 showed no signs of vesicle recycling at any heated time points (Fig 7D, data not quantified). Conversely, although dTPIT73R displayed similar loading to dTPIWT at 15 and 30 sec., dTPIT73R exhibited a striking 50% decrease in loading at 60 sec. (Fig 7A, 7B and 7D). This progressive decrease in endocytosis was stimulation and temperature dependent; loading experiments performed at room temperature did not exhibit an endocytic defect (Fig 7C). As previous experiments had highlighted the capacity of dTPIΔcat to complement the adult behavioral defects of dTPIT73R, we examined dTPIT73R/dTPIΔcat larvae to assess the relationship between vesicle endocytosis and animal behavior. The dTPIT73R/dTPIΔcat animals displayed a significant increase in vesicle endocytosis relative to dTPIT73R (Fig 7B and 7D). These results demonstrate that dTPIΔcat complements adult behavior and vesicle endocytosis defects. The utilization of chemical stimulation in these preparations demonstrates a synaptic defect arising from the severe dTPIT73R dimer mutation as this methodology bypasses conductance requirements.
A reduction in vesicle dye uptake could be derived from defects in endocytosis or exocytosis, and indeed, these activities are intimately linked [48]. To examine temperature-dependent changes in vesicle fusion, dTPIWT and dTPIT73R animals were i) loaded with dye at RT for 3 min., ii) washed with 0 mM Ca2+ HL-3, iii) imaged, iv) heated to 38°C, v) vesicle fusion initiated with 30 sec. of high [K+] HL-3, and vi) reimaged. Care was taken to ensure the same synapses were imaged at loading and unloading timepoints. Preliminary experiments demonstrated that 60 sec. of high [K+] stimulation completely unloaded the synapses in each genotype; therefore 30 sec. was analyzed to achieve a measurable dynamic range. Unloading experiments at elevated temperatures demonstrated no change in vesicle exocytosis between dTPIWT and dTPIT73R at 38°C (Fig 7E and 7F).
Finally, functional changes at the synapse can be the result of acute impairments in recycling machinery or more chronic developmental defects. Mutations in the E3- ubiquitin ligase Highwire or alterations in the trans-synaptic signaling proteins wingless and Glass-bottom boat have been shown to alter synaptic function through primary changes in development [49–51]. These changes in synaptic physiology are accompanied by dramatic alterations in synaptic morphology, a hallmark of neurodevelopmental defects. To examine whether aberrant neurodevelopment may contribute to this recycling deficit, we morphologically characterized the Drosophila NMJ from segment A2, muscle 6/7; this particular NMJ is highly elaborate and therefore sensitive to developmental perturbations. The dTPIM80T, dTPIT73R, and dTPIG74E alleles all exhibited early lethality if maintained at 25°C, therefore development was scored at RT. Third instar larva were dissected, and an assessment of bouton number and branches revealed no significant developmental differences in the thermal-stress sensitive mutants relative to dTPIWT (Fig 8). These results suggest that the synaptic defect is an acute disruption of function, and not likely a secondary defect caused by altered development. Collectively, these data demonstrate that TPI deficiency thermal-stress sensitivity is characterized by acute perturbation of synaptic vesicle dynamics.
We conclude that the impairment of TPI dimerization is sufficient to elicit neurologic dysfunction. Dimerization of TPI is critical to its conformation, stability, and catalytic properties; and disruption of these molecular properties impedes vesicle dynamics at the synapse. Impaired synaptic function is thermal-stress dependent, and both vesicular and behavioral abnormalities can be genetically complemented through a catalytically inactive TPI allele.
The pathogenic hTPIM80T substitution impairs TPI dimerization. These results were obtained from purified proteins and do not corroborate those from non-denaturing gel filtration experiments performed on animal lysates [26]. However, several in vitro studies have found that mutations that impair TPI dimerization severely destabilize the protein [28,30,44,52,53]. In vivo, unstable proteins are bound by chaperones and either refolded, targeted to the proteasome, or aggregate [54]. The results presented here suggest that dTPIM80T does not cause TPI to aggregate (Fig 3), while previous work extensively details the recruitment of Hsp70 and Hsp90 to dTPIM80T and its degradation through the proteasome [34]. Therefore, we hypothesize that TPI monomer may not have been detected previously in animal lysates due to its rapid sequestration and degradation. We believe these data, along with the previous inability to identify monomer in vivo, collectively suggest that TPI does not stably exist in vivo as a soluble monomer.
We utilized our GE system to generate two additional TPI alleles with point mutations at the dimer interface that have previously been shown to impair homodimerization [28]. These substitutions were located at the tip of the 3rd loop of TPI that extends into its dimer partner and stabilizes/rigidifies a network of hydrophobic interactions and hydrogen bonds which form the dimer interface [28,29,55]. The substitution of these dimer interface residues resulted in severely pathogenic TPI alleles, eliciting greater behavioral dysfunction and shorter lifespans than dTPIM80T (Fig 2).
A universal molecular mechanism of TPI deficiency pathogenesis is currently unclear. To date, two crystal structures of disease-associated TPI mutations have been reported [15,18]. These mutations were found in two distinct structural regions of the TPI homodimer. The first structure to be solved was from the most commonly diagnosed TPI deficiency substitution, hTPIE104D [15]. The hTPIE104D substitution is a conservative alteration of a charged residue at the dimer interface that results in reduced dimer stability, but unchanged catalytic activity [15,16]. The second disease-associated structure was an hTPII170V substitution, a conservative substitution found on the catalytic lid of the enzyme that enhances thermal stability and reduces catalytic activity [16,18]. The dimer substitutions used in this study share several molecular characteristics with the TPIE104D human mutation, and given the conservation of the TPI enzyme, we propose that these patients likely share similar molecular and cellular dysfunction as identified in this study. Conversely, the dimer substitutions are not predicted to share many molecular similarities with the hTPII170V mutation indicating that dimerization defects are sufficient but not necessary to elicit TPI deficiency. It is interesting to note that although dTPIT73R and hTPII170V both exhibit mechanical and thermal stress sensitivities in Drosophila, the behavioral dysfunction caused by the dimer substitutions is far more severe, and hTPII170V does not influence animal longevity [18]. Further, the capacity to attenuate behavioral dysfunction but not longevity in dTPIT73R,G74E/dTPIΔcat suggests an independent pathogenic mechanism (S4 Fig) that may not be determined by TPI dimerization.
TPI dimerization and protein stability in vivo will ultimately influence catalytic capacity, and many TPI activity measurements from animal models and patient tissue samples have identified a reduction in isomerase activity [7,14,17,18,25,56–58]. However, our measurements of isomerase activity from healthy and affected animal lysates argue that TPI dimer integrity is a stronger determinant of behavioral dysfunction. Still, this could indicate that reduced TPI activity is corollary to the disease or a contributing factor to an alternative pathogenic mechanism. Previous studies have demonstrated that the redox state is altered in TPI deficient cells and organisms [33,59,60]. The redox status in TPI deficiency is proposed to be altered by abnormal flux through the pentose-phosphate pathway as well as potential accumulation of advanced glycation end-products (AGEs) [16,20,33,59–61], and the accumulation of redox damage in the nervous system is strongly linked with several neurodegenerative diseases [62–64]. Interestingly, a study in yeast demonstrated differences in redox responses between the E104D and I170V mutations [16]; these results could imply different modes of pathogenesis that are dependent on the conformational and catalytic states of TPI.
One unresolved aspect of this study was the inability of dTPIΔcat to fully complement the behavioral defects of dTPIG74E. Co-IP experiments suggested that complementation correlated with an inability of dimer-interface mutants to form heterodimers with dTPIΔcat (Fig 6). These data would imply that dTPIG74E may be exhibiting a dominant negative effect as a heterodimer, but this conclusion was inconsistent with our genetic analyses of dTPIWT/dTPIG74E animals (Fig 4). To investigate whether TPIG74E may be interacting differently with TPIWT than with TPIΔcat, we purified, crystallized, and determined the structure of hTPIΔcat at 1.7Å resolution.
While the hTPIK13M substitution (hTPIΔcat) resulted in multiple rearrangements that mimic the open or non-catalytic TPI conformation, the dimeric interface remained essentially unchanged, including the peptide backbone of Loop3 and sidechain positions hT75, hG76, and hM82 (Fig 5). To address how hTPIT75R and hTPIG76E substitutions may influence the dimer interface, we generated models of heterodimers in which one subunit contained either hT75R, hG76E, or both hT75R and hG76E substitutions, while the other monomer remained unaltered. Models were made using either wild-type TPI hTPIWT (PDB: 4POC) or the hTPIΔcat (PDB: 4ZVJ) as the structural template, and subjected to analysis by RosettaBackrub [65]. Briefly, Rosetta scores are predictions of the most energetically stable conformations with higher scores indicating less favorable positioning of the model. The algorithm was run 50 times for each mutation to be modeled. Of these 50 simulations, the lowest scores of hTPIΔcat::hTPIT75R and hTPIΔcat::hTPIG76E were selected and shown (Fig 9A and 9B), while the modeled structures whose Rosetta scores fell within the best 10% of its respective ensemble were collected for analysis (Fig 9C).
Modeling the hTPIT75R and hTPIG76E substitutions as homodimers or heterodimers with the hTPIWT structure produced high Rosetta scores, predicting poor energetic favorability (Fig 9C) in agreement with our gel filtration experiments (S2 Fig). To examine TPIΔcat heterodimers we used the new hTPIΔcat structure to model hTPIΔcat::hTPIT75R and hTPIΔcat::hTPIG76E. These experiments predicted a high Rosetta score for hTPIΔcat::hTPIT75R and a very low one for hTPIΔcat::hTPIG76E, corroborating the results of our animal lysate coIP experiments and suggesting the simulated heterodimers may accurately represent the conformations of these molecules.
The hTPIΔcat::hTPIT75R heterodimer with the lowest Rosetta score suggests the hR75 residue may orient toward the catalytic pocket of hTPIΔcat, lining the floor of the substrate-binding pocket through the displacement of hE165 and hK13M (Fig 9A). This orientation of hR75 into the catalytic pocket is similar to that previously described by Wierenga and colleagues [28]. The hTPIΔcat::hTPIG76E heterodimer with the lowest Rosetta score suggests that hE76 finds a stable position in the dimer interface through the displacement of hE104 and hR98, possibly via coordination of the terminal amide of hN65 (Fig 9B). Interestingly, perturbation of hE104 has been shown to significantly alter the TPI dimer interface and elicits TPI deficiency in humans through a conservative hTPIE104D substitution [15]. These modeling predictions suggest that the character of the hTPIΔcat::hTPIG76E dimer interface is drastically altered relative to hTPIΔcat homodimers.
Drosophila TPI deficiency neurologic dysfunction is characterized by impaired vesicle dynamics at the neuronal synapse, a defect we believe is likely conserved in human patients. The key to deciphering this pathogenic mechanism was the behavioral severity of the newly generated TPI dimer interface mutants. The dTPIT73R allele exhibited temperature-dependent paralysis at a mean time of approximately 27 sec., an acute behavior that is rare and highly enriched for synaptic or conductance defective mutants. Only a handful of Drosophila mutant alleles have been identified with rapid temperature-dependent paralysis, including those of voltage-gated Na+, K+, and Ca2+ channels (para, sei, cac) [66–68], the sodium-potassium exchanging ATPase (ATPα) [69], and components of vesicle fusion and recycling (N-ethylmaleimide sensitive factor–dNSF1, dynamin–Shi) [70,71]; and after noting these phenotypic similarities we broadly examined synaptic function. Stimulation was conducted using high [K+] bath applications, thereby bypassing the participation of Na+ and K+ channels. The dTPIT73R mutants were characterized by normal endocytosis during acute stimulations (15 and 30 sec.) but exhibited a dramatic reduction after 60 sec. of stimulation (Fig 7A), suggesting a time/excitation dependent phenotype. Further, complementation with the dTPIΔcat allele significantly increased the temperature-dependent FM1-43 loading at these terminals (Fig 7B and 7D), similar to its complementation of adult thermal stress-induced paralysis. Finally, measurements of vesicle exocytosis at these elevated temperatures did not indicate an exocytic defect (Fig 7E and 7F). We were unable to detect terminal signal above background after 60 sec. of unloading, so we cannot unequivocally define the nature of the vesicular dysfunction, but the data suggest it is likely due to impaired endocytosis.
The observed synaptic defect provides insight into why the majority of human patients present with TPI mutations affecting the dimer interface. First, all substitutions that disrupt the dimer interface have been shown to destabilize the enzyme in vitro [28,29,55]. This destabilization is likely responsible for the reduced cellular TPI found in patient samples, and our work with the dTPIM80T, dTPIT73R, and dTPIG74E mutants provide additional evidence that dimer interface substitutions reduce TPI levels in vivo (Fig 3A and 3B). Secondly, the cellular depletion of dTPIM80T has been shown to be mediated by heat shock protein sequestration and proteasomal degradation [34]. If chaperones sequestered and degraded these misfolded or unstable proteins, this would likely prevent the distribution and maintenance of TPI at specific subcellular locales. Recent work has shown that the anterograde transport of globular/soluble proteins to the terminals is a slow process, moving at a rate of approximately 0.008–0.01μm/sec [72]. To put this in the context of the Drosophila nervous system, the length of the relatively short larval motor axon innervating muscle 4 of segment A3 has been measured to be ~220μm [73]. Based on these approximations, one could estimate that it would take ~6 hrs for TPI translated in the soma to be transported to the axonal terminal. In this way, substitutions that affect protein stability would likely result in improper localization or sequestration of TPI during distal transport, ultimately depleting TPI at the synapse. This proposal is also consistent with the inability of RNAi knockdown to recapitulate Drosophila TPI deficiency behavioral phenotypes, as RNAi alters mRNA transcript levels rather than enzyme conformation or stability. RNAi knockdown of TPIWT would reduce, but still allow the transport and stable accumulation of TPIWT at the synaptic terminal.
Two other Drosophila glycolytic mutants, aldolase and phosphoglycerate kinase, have been shown to exhibit temperature-dependent paralysis though with longer onsets [74,75]. The phosphoglycerate kinase mutant was shown to exhibit synaptic dysfunction, and the authors asserted that an inhibition of vesicle recycling was likely the cause of the functional defect [75]. In both cases, the animals were found to have depletions in lysate ATP [74,75]. It is attractive to speculate that all three of these glycolytic mutants may suggest a pivotal role for glycolysis within synaptic vesicular dynamics, and indeed, recent measurements of ATP consumption in the synapse suggest that glycolytic ATP is the primary substrate used to support synaptic function [76,77]. However, the role of glycolytic proteins and their putative energetic importance at the synapse is controversial. Many research groups assert the preeminent utilization of mitochondrial ATP at these sites [78–81], while the lactate shuttle hypothesis largely circumvents a role for neuronal glycolysis [82,83]. Further, the absence of a correlation between TPI catalytic activity and behavioral phenotypes suggests that the enzyme may be complexing with another molecule to facilitate synaptic vesicle cycling.
How the dimerization or integrity of the TPI dimer interface impacts synaptic vesicle dynamics remains a mystery, though one candidate for a molecular complex is the actin-regulatory protein cofilin. In Drosophila, cofilin (twinstar) and twinfilin mutants have been demonstrated to elicit functional and developmental neurologic defects [84–86], and actin-regulatory proteins such as cofilin, actin-depolymerizing protein (ADP), and twinfilin are known to influence synaptic vesicle dynamics [85,87,88]. Recently, cofilin was found to bind to TPI in both its inactive and active forms [89]. The precise binding site between cofilin and TPI is unknown, though with its mixture of charged and hydrophobic pockets, the TPI dimer interface may provide a suitable site for this interaction. Additional studies will be needed to specifically delineate the role of TPI in the synapse.
In conclusion, to our knowledge this work is the first to highlight a critical role for TPI in the cycling of vesicles at the synapse, with behavioral correlates similar to the inactivation of vesicle fusion/recycling proteins. These observations help clarify the neurologic symptoms seen in patients and will direct future therapeutic strategies. The findings of this study will guide future investigations regarding the contribution of TPI localization and function to synaptic vesicle dynamics, and ultimately how these properties are perturbed in TPI deficiency.
The Vienna Drosophila RNAi Center (VDRC) line used for knockdown experiments was stock #25644 [37]; experiments were also conducted with #25643 with similar results. The w;actin-GAL4,UAS-GAL4; animals were generated by recombining the second chromosomes of the Drosophila Genetic Resource Center (DGRC) stock y1 w1118; P(w+mC = UAS-Gal4.H)12B and Bloomington Stock Center stock y1w*; P(Act5C-GAL4)25FO1/CyO,y+; recombinants were screened molecularly and balanced in a w1118 background. The following TPI alleles in this study were generated using the GE system: dTPIWT, hTPIWT, dTPIWT-CFP, dTPIM80T, hTPIM82T, dTPIT73R, dTPIT73R-CFP, dTPIG74E, dTPIG74E-CFP, dTPIT73R,G74E, dTPIΔcat, and dTPIΔcat-CFP. The development of the GE system and the production of the dTPIWT, dTPIM80T, dTPIΔcat, and dTPIΔcat-CFP alleles were initially described elsewhere [25]. Briefly, GE involves the replacement of the TPI gene locus with a phiC31 integration site through homologous recombination. The phiC31 integration system allows nearly seamless integration of complementary vector constructs directly into the TPI gene locus to maintain endogenous spatial and temporal regulation. The TPInull allele used in S1 Fig was generated previously, formerly known as TPIJS10 [19]. RNAi experiments used the TPIM80T allele formerly known as TPIsgk [19], due to similarities with the UAS-RNAi and GAL4 genetic backgrounds.
This study uses the established nomenclature for TPI, assuming the start methionine is removed following translation [13]; all residue numbering in this study uses the same convention. An alignment is included for clarification (Fig 1A). All animal populations assessed were approximately equivalent mixtures of males and females.
Site directed mutagenesis was performed using the QuikChange Lightening Site-Directed Mutagenesis Kit (Agilent Technologies). Mutagenesis primers were generated (Integrated DNA Technologies) to introduce a Thr-to-Arg codon change at position 73, and a Gly-to-Glu change at position 74 –both separately and together for the purpose of creating the double-mutant. Mutagenesis was performed on the previously published pGE-attBTPI+ plasmid [25] and confirmed by sequencing. Once the constructs were generated, TPI GE was performed using previously published methods [25,90,91]. Briefly, the PGX-TPI founder animals were mated to vasa-phiC31ZH-2A animals expressing the integrase on the X chromosome and their progeny injected with pGE-attBTPI constructs. Integration events were identified via the w+ phenotype and verified molecularly.
Human TPI enzyme was purified as outlined previously [18].
DLS measurements were taken using a DynaPro Plate reader (Wyatt Technology) equipped with a temperature control unit. Purified BSA (Sigma Aldrich), hTPI+, and hTPIM80T were diluted to concentrations of 3.75 μM, 15 μM and 30 μM in 100 mM triethanolamine (TEA); pH 7.6. Three 75 μl aliquots were loaded onto a 384-well microplate and read at 37°C. Ten measurements were taken per sample and Dynamics V6 software (Wyatt Technology) was used to process the scattering data, generating autocorrelation functions. Autocorrelation functions were then analyzed to obtain the hydrodynamic radii. Student’s T test was used to compare samples. DLS experiments were performed three times.
Gel filtration was performed as outlined previously [26]. Purified TPI samples were diluted to 15 μM in mobile phase, 100 μl were injected in triplicate, and their elution monitored at 280 nm. Experiments were performed three times. Chromatography traces were collected and analyzed using EZStart 7.3 (Shimadzu) to quantify the relative monomer and dimer populations. Curve integration data were compared using Student’s T test.
Isomerase activity was determined using an NADH-linked assay as previously detailed [25,92]. Lysates were diluted to 0.1 μg/μl in 100 mM TEA pH 7.6 + inhibitors and enzyme activity was assessed. Reaction assays were performed in triplicate using 80 μl mixtures composed of 0.5 mM NADH, 0.752 mM GAP, 1 unit glycerol-3-phosphate dehydrogenase and 1 μg of lysate protein in 100 mM TEA; pH 7.6. Consumption of NADH was monitored at 340 nm and 25°C using a SpectraMax Plus 384 microplate reader (Molecular Devices). All reactions were performed at least three times. Reaction components were purchased from Sigma-Aldrich. Enzyme activity curves were normalized to reactions performed without GAP. A one-way ANOVA was performed to assess variance, and data sets were compared using Tukey’s post-hoc analysis.
Mechanical stress sensitivity was examined on Day 1 by vortexing the animals in a standard media vial for 20 seconds and measuring time to recovery, similar to [93]. Briefly, recovery is defined as two purposeful WT movements including righting, grooming, climbing, or walking. Thermal stress sensitivity was assessed on Day 2 by acutely shifting animals to 38°C and measuring time to paralysis, as previously described [24,69]. In these assays, the animals typically seized and either flipped over onto their backs or fell sideways with no successive coordinated movements, i.e. righting, climbing, walking, grooming. Behavioral responses were capped at 360 and 600 seconds where indicated and reported as 360 and 600 sec. Animal lifespan determinations were performed at 25°C as previously described [69]. All assays used approximately equivalent numbers of males and females. One-way ANOVAs were performed with Tukey's post-hoc analysis to compare behavior, and lifespans were assessed with Log-rank (Mantel–Cox) survival tests.
Animals were collected and aged 1–2 days at room temperature. Ten fly heads were obtained in triplicate from each genotype and processed as described previously [34]. Blots were incubated with anti-TPI (1:5000; rabbit polyclonal FL-249; Santa Cruz Biotechnology), anti-ATPalpha (1:10,000; mouse monoclonal alpha5; Developmental Studies Hybridoma Bank), or anti-β tubulin (1: 6,000; rabbit polyclonal H-235; Santa Cruz Biotechnology). Densitometric analyses of the scanned films were performed on unsaturated exposures using ImageJ software available from the National Institutes of Health. A one-way ANOVA was performed to assess variance of TPI levels and data sets were compared using Tukey's post-hoc analysis.
The filter-trap dot blot was modified from methods published previously [36]. Animals were aged 1–2 days, collected and homogenized in 1X PBS (2.7 mM KCl, 137 mM NaCl, 2 mM NaH2PO4, 10 mM Na2HPO4; pH 7.4) supplemented with cOmplete mini Protease Inhibitors (Roche Diagnostics), and diluted to 1 μg/μl; wells were loaded as indicated. Samples were diluted 1:2 in 1% SDS, 1X PBS, boiled for 5 min., and filtered through a cellulose acetate membrane (Whatman, 0.2μm pore) using a 96-well vacuum dot blot apparatus. Positive controls were collected from PC12 cells transfected with a construct expressing huntingtin “exon1” bearing a stretch of 97 glutamines and C-terminally tagged with EGFP. The membrane was washed four times with 1% SDS-PBS, blocked with Odyssey Blocking Buffer (LiCor), and primary antibodies applied in Odyssey Blocking Buffer. Blots were incubated with anti-TPI (1: 5000) and anti-GFP (1: 5000; rabbit polyclonal; FL; Santa Cruz Biotechnology). The membranes were then washed and incubated with the secondary antibody IRDye 800-conjugated goat anti-rabbit (LiCor) at 1: 20,000 in Odyssey Blocking Buffer. Direct-to-scanner detection and blot visualization were performed using a LiCor Odyssey scanner. Filter-trap experiments were performed twice.
Coimmunoprecipitation was performed using the Pierce Co-Immunoprecipitation Kit (Thermo Scientific) as per manufacturer’s instructions. Lysates were generated by mechanically homogenizing 50 animals in 0.5 ml of IP Lysis buffer (25 mM Tris, 150 mM NaCl, 1 mM EDTA, 1% NP-40, 5% glycerol; pH 7.4) supplemented with cOmplete mini Protease Inhibitors. After homogenization, lysates were frozen in liquid nitrogen, thawed, then centrifuged twice at 5,000 g to pellet exoskeletal debris. Supernatants were collected and diluted to 1 μg/μl, and 400 μg were loaded onto 25 μl of gel pre-coupled with 10 μg of anti-GFP. A negative control was performed using uncoupled gel and dTPI+/dTPIΔcat lysate. Samples were incubated overnight at 4°C and washed ten times with IP Lysis buffer at 4°C. Beads were eluted by boiling with 70 μl of 2X SDS–PAGE sample buffer, separated via SDS-PAGE, immunoblotted, and analyzed as outlined above. Coimmunoprecipitations were performed three times.
Images were taken with an Olympus BX51WI fluorescence microscope with Till Photonics Polychrome V monochromator excitation, and Hamamatsu C4742-95 digital camera. Heterozygous TPIT73R larvae were maintained over TM6B, and Tb+ 3rd-instar larvae selected for analysis. Dissection and preparation of larval NMJs were performed as described previously [46]. FM1-43FX dye [Molecular Probes, Invitrogen] loading was performed as previously detailed [47]. Briefly, animals were dissected in ice cold 0 mM Ca2+ HL-3 with 0.5 mM EGTA, motor neurons severed, and the preps heated to room temperature or 38°C over the course of 3 min. Bath temperature was monitored throughout the experiments with a microthermal probe to ensure consistency [Fisher Scientific]. Loading experiments were performed with room temperature or 38°C preheated 90 mM KCl, 1.5 mM CaCl2 HL-3 supplemented with 4 μM FM1-43FX, and preparations were washed quickly and thoroughly during the experiments to avoid Ca2+ chelation. After loading, preparations were washed with 15 ml of 0mM Ca2+ HL-3 with 0.5 mM EGTA at room temperature for 10 min. Unloading experiments were performed as follows: preparations were loaded for 3 min at room temperature; washed with 15 ml of 0mM Ca2+ HL-3 with 0.5 mM EGTA at room temperature; imaged; washed with 38°C 0 mM Ca2+ HL-3; heated to 38°C over 3 min.; unloaded using 38°C 90 mM KCl, 1.5 mM CaCl2 HL-3 for 30 sec.; and the same synapse was imaged again.
Preparations were imaged with a water immersion 60X objective, using 450 nm excitation and a 500 nm longpass filter [Chroma Technology]. Simple PCI imaging software was used for acquisition and ImageJ for analysis. Two NMJs from muscles 6/7 were assessed per animal, one from segment A2 and A3. Six animals were assessed per genotype per time point for a total of 12 NMJs per experimental condition. After acquisition, images were relabeled by an independent researcher and blinded analysis was performed on the raw images. Boutons were outlined and intensity measured, with background subtracted from adjacent tissue. Pair-wise analyses were performed using a two-tailed Student’s t test, while comparisons among multiple experimental conditions were performed using a one-way Analysis of Variance (ANOVA) with Tukey’s post-hoc analysis.
For NMJ morphological analyses, 3rd-instar larvae were collected and dissected as detailed above without transection of the descending motor neurons. Preparations were fixed in 3.5% paraformaldehyde HL-3, permeabilized with 0.1% Triton X-100 in 1X PBS (PBST), and blocked with 0.2% BSA in PBST (PBSTB) for 2 hrs at room temperature. Preps were washed and incubated with goat anti-HRP [Jackson Laboratories] at 1: 200 in PBSTB for 2 hrs at room temperature. Primary antibodies were removed, washed in PBSTB, and incubated with Cy3-labeled donkey anti-goat in PBSTB at 1: 400 for 1.5 hrs at room temperature. Preps were washed, mounted in VectaShield [Vector Laboratories], and imaged within three days. Images were acquired with an Olympus confocal FV1000 microscope, using a 559 nm excitation laser. Z stacks of segment A2 of muscle 6/7 were taken using 1 μm steps. The Z stacks were merged using Olympus FV1000 Fluoview Viewer, and morphology determined. Ten animals were assessed per genotype, one NMJ per animal, for a total of ten NMJs per experimental condition. Boutons were defined as varicosities at least 2 μm in diameter, and branches defined as extensions containing at least 2 boutons. Images were relabeled by an independent researcher for blinded analysis. Variance within the data set was examined using a one-way ANOVA, with comparisons made using Tukey’s post hoc test.
All image quantification was performed on raw image files acquired below saturation. Representative images were selected on the basis of raw image measurements, and post-acquisition processing was performed uniformly with grouped images in parallel using ImageJ; in agreement with published guidelines [94].
Recombinant hTPIΔcat containing the K13M substitution was expressed and purified as previously described [18] using affinity, anion exchange, and size exclusion chromatography. Purified protein was dialyzed into a buffer containing [20 mM Tris pH 8.8, 25 mM NaCl, 2.0% glycerol and 1 mM β–mercaptoethanol], and concentrated to 6 mg/ml prior to crystallization. Crystals of TPIΔcat were obtained using the vapor diffusion method with sitting drops containing 1 μl of protein and 2 μl of well solution [28–34% PEG2000 MME, 50 mM KBr]. Initial crystals grew within 3 days and were improved by successive rounds of microseeding. Crystals were cryoprotected in 40% PEG 2000MME, 20% glycerol, 50 mM KBr, prior to flash freezing in liquid nitrogen.
Data collection was performed at the National Synchrotron Light Source at beamline X25 and using a Pilutas 6M detector. Diffraction data was integrated, scaled, and merged using HKL2000 [95]. hTPIΔcat crystals belong to space group P212121 and contain a dimer in the asymmetric unit. Initial phases were estimated for hTPIΔcat via molecular replacement using a previously determined structure of wild-type as our search model [18]. Model bias was reduced through simulated annealing and the model was further improved by manual model building combined with positional and anisotropic B-factor refinement within Phenix [96]. Model quality was validated using MolProbity [97]. Figs were generated using PyMOL (PyMOL Molecular Graphics System, Schrödinger, LLC). Coordinates and structure factors for hTPIΔcat have been deposited within the Protein Databank under accession code 4ZVJ.
The effect of hT75R and hG76E substitutions were modeled onto hTPIΔcat or hTPIWT structural templates using the RosettaBackrub analysis as implemented within the RosettaBackrub server [98]. For all predictions, Rosetta version 3.1 was used as the algorithm with a backrub radius of 15 Å to ensure that perturbations extending away from the site of the substitution could be sampled. An ensemble of 50 structures was predicted for each TPI model, and their reported Rosetta scores normalized to the starting template. Structures whose Rosetta scores were in the most favorable 10% of the ensemble were used in Fig 9D.
Triosephosphate isomerase research has been split between those studying the enzymatic and structural properties of the protein, and those studying its role in disease. Researchers focusing on the pathology of TPI deficiency typically use the abbreviation “TPI”, whereas enzymologists and structural biologists used the abbreviation “TIM”. The aim of this study was to determine the molecular mechanism of a disease mutation, and as such we have used the abbreviation “TPI”.
|
10.1371/journal.ppat.1004174 | Selective Chemical Inhibition of agr Quorum Sensing in Staphylococcus aureus Promotes Host Defense with Minimal Impact on Resistance | Bacterial signaling systems are prime drug targets for combating the global health threat of antibiotic resistant bacterial infections including those caused by Staphylococcus aureus. S. aureus is the primary cause of acute bacterial skin and soft tissue infections (SSTIs) and the quorum sensing operon agr is causally associated with these. Whether efficacious chemical inhibitors of agr signaling can be developed that promote host defense against SSTIs while sparing the normal microbiota of the skin is unknown. In a high throughput screen, we identified a small molecule inhibitor (SMI), savirin (S. aureus virulence inhibitor) that disrupted agr-mediated quorum sensing in this pathogen but not in the important skin commensal Staphylococcus epidermidis. Mechanistic studies employing electrophoretic mobility shift assays and a novel AgrA activation reporter strain revealed the transcriptional regulator AgrA as the target of inhibition within the pathogen, preventing virulence gene upregulation. Consistent with its minimal impact on exponential phase growth, including skin microbiota members, savirin did not provoke stress responses or membrane dysfunction induced by conventional antibiotics as determined by transcriptional profiling and membrane potential and integrity studies. Importantly, savirin was efficacious in two murine skin infection models, abating tissue injury and selectively promoting clearance of agr+ but not Δagr bacteria when administered at the time of infection or delayed until maximal abscess development. The mechanism of enhanced host defense involved in part enhanced intracellular killing of agr+ but not Δagr in macrophages and by low pH. Notably, resistance or tolerance to savirin inhibition of agr was not observed after multiple passages either in vivo or in vitro where under the same conditions resistance to growth inhibition was induced after passage with conventional antibiotics. Therefore, chemical inhibitors can selectively target AgrA in S. aureus to promote host defense while sparing agr signaling in S. epidermidis and limiting resistance development.
| New approaches are needed to lessen the burden of antibiotic resistant bacterial infections. One strategy is to develop therapies that target virulence which rely on host defense elements to clear the bacteria rather than direct antimicrobial killing. Quorum sensing is a bacterial signaling mechanism that often regulates virulence in medically relevant bacterial pathogens. Therefore, drugs that inhibit quorum sensing can promote host defense by rendering the pathogenic bacteria avirulent and/or less fit for survival within the host. Our work addressed this strategy in the pathogen Staphylococcus aureus which is the major cause of acute bacterial skin and soft tissue infections. We conducted a high throughput screen to identify compounds that could inhibit signaling by the quorum sensing operon, agr. We found a compound that we termed savirin (S. aureus virulence inhibitor) that could inhibit signaling by this operon. The drug helped the innate immune system in animals to clear bacteria that express this operon without affecting clearance of bacteria that do not have this operon. We addressed the mechanism of action of this compound and whether resistance or tolerance to this compound would likely develop. Our data indicate for the first time that host defense against S. aureus skin infections can be enhanced by chemical inhibition of agr-mediated quorum sensing.
| The global health threat of antibiotic resistant bacterial infections mandates rethinking of how antibiotics are used, how targets for new antibiotics are identified, and how mechanisms for promoting host defense can be enhanced [1], [2]. In this regard, there is much interest in chemical inhibition of bacterial signaling systems, particularly quorum sensing, because of its regulation of virulence in many medically relevant pathogens where antibiotic resistance is problematic [3], [4]. While chemical inhibitors of quorum sensing (QSIs) have been described in vitro, few have demonstrated in vivo efficacy [5]. Moreover, concerns have been raised about the specificity and selectivity of these compounds [6] as well as the potential for resistance development to quorum sensing inhibition [7]. Therefore, the future of quorum sensing inhibition as a medical strategy to replace or augment conventional antibiotics is uncertain.
Of the quorum sensing systems in Gram positive pathogens being targeted for chemical inhibition, the agr operon of Staphylococcus aureus has received noteworthy attention [3], [8]. This interest derives from its significant medical burden [9], its known propensity for developing resistance to newly introduced antibiotics [10], and the failure of all vaccines to date to prevent infection [11]. While chemical inhibitors of agr have been identified [8], none have demonstrated efficacy in mammalian models of infection. Moreover, none have demonstrated selectivity towards agr signaling in the pathogen S. aureus while sparing agr signaling in the skin commensal Staphylococcus epidermidis, an important contributor to host defense against skin infection [12].
Approximately 90% of S. aureus infections involve skin and soft tissues (SSTIs) [9], [13] and agr is positively associated with human SSTIs [14], [15]. Moreover, competitive interference with agr signaling is sufficient to abrogate experimental skin abscesses [16], and we have shown that innate immunity against experimental S. aureus skin infection requires active suppression of agr signaling [17]–[19]. Therefore, we postulated that selective chemical inhibition of agr signaling in S. aureus could promote host defense against SSTIs, providing evidence for limiting conventional antibiotic use in the majority of S. aureus infections. Here we describe a QSI identified in a high throughput screen that selectively inhibited agr signaling in S. aureus, but not in S. epidermidis, by blocking the function of the transcriptional regulator of the operon, AgrA, preventing upregulation of the agr-regulated genes essential for skin infection. It was efficacious in murine models of agr-dependent skin infection without apparent induction of resistance or tolerance after passage in vivo. These data provide proof-of-principle that AgrA transcriptional function in S. aureus can be selectively inhibited to attenuate quorum sensing with minimal toxicity to the bacterium or induction of stress responses observed with conventional antibiotics. Thus, selective AgrA blockade could enhance agr-dependent host defense in the skin while potentially preserving the normal microbiota, limiting resistance induction, and sparing conventional antibiotics for treatment of invasive systemic infections.
The agr quorum sensing operon encodes two promoters [3], [20]; P2 that drives production of a two component sensor-regulator, AgrC and AgrA, and its autoinducing peptide pheromone ligand, and P3 that drives production of a regulatory molecule RNAIII that together with AgrA is responsible for transcriptional control of approximately 200 genes including multiple virulence factors and metabolic pathways involved in stationary phase growth [15]. P3 also drives P2 providing positive feedback to the production of the receptor (AgrC), the transcriptional regulator (AgrA), and the cyclic thiolactone peptide pheromone (AIP). Critically, the virulence factors most closely associated with human SSTIs, alpha hemolysin (hla), phenol soluble modulins (PSMs), and Panton-Valentine Leukocidin (PVL) are agr regulated [14], [15]. We screened 24,087 compounds selected for diversity for inhibition of AIP-induced agr::P3 activation using a reporter strain where P3 drives production of GFP (ALC1743) (http://pubchem.ncbi.nlm.nih.gov/assay/assay.cgi?aid=1206&loc=ea_ras). We pursued one compound where dose-response experiments using an additional reporter strain indicated that it had minimal impact on exponential phase growth during the 3 hr assay starting at a CFU of 2×107/ml and ending at ∼1×108/ml. It inhibited optimally at 5 µg ml−1 (13.5 µM) (Fig. S1). Termed savirin (Fig. 1A), for Staphylcoccus aureus virulence inhibitor, its molecular weight (368) and lipophilicity (XLogP3-3.5) meet standards for drug development [21].
S. aureus isolates belong to one of four agr alleles depending on variations in AIP (amino acid sequence and length) and the cognate receptor, AgrC [3], [20]. While agr I alleles predominate in human disease, all four can contribute to SSTIs [9]. Therefore, an optimal chemical for agr disruption should work against all agr alleles. Savirin (5 µg ml−1) inhibited agr::P3 activation in reporter strains of each agr type (Fig. S2). Therefore, we pursued its efficacy in vitro and in vivo using a strain (LAC) of the epidemic methicillin-resistant USA300 clone and the predominant agr group I [15], [18], [19], [22]. We demonstrated by qRT-PCR that savirin (5 µg ml−1) inhibited both AIP1-induced RNAIII (Fig. 1B) and RNAIII produced at a longer time point without addition of exogenous AIP1 (Fig. 1 C) with no effect on exponential phase growth (Fig. 1 D). Stationary phase growth was negatively affected by both the genetic deletion of agr (Δagr) and by savirin treatment (Fig. 1D) consistent with the known role of agr in regulating metabolic pathways of this growth phase in LAC [15]. Importantly, savirin did not significantly affect AIP1-induced RNAIII levels (Fig. 1E) or agr-dependent stationary phase growth (Fig. 1F) in the related Gram positive member of the skin microbiota, S. epidermidis. Nor did it affect growth of a Gram negative member of the skin microbiota, Pseudomonas aeruginosa [23] (data not shown). Because savirin could have different effects on growth in larger bulk cultures, we evaluated the effects of savirin on both exponential and stationary phase growth in 5 ml cultures diluted to measure OD600 under 0.8. The results were qualitatively similar (Fig. S3). In addition, savirin did not disrupt membrane integrity (Fig. S4A) or membrane potential (Fig. S4B), properties that are altered by antibiotic compounds that could affect agr signaling [24], [25] and that could be impaired by agr-independent, non-specific toxic effects [6].
To pursue the molecular mechanism by which savirin inhibits agr signaling in S. aureus but not in S. epidermidis, we examined the differences in histidine kinase function and transcriptional control between the two. Because residues within the histidine kinase domain of AgrC that are critical for agr activation are conserved between S. aureus and S. epidermidis [26], we pursued AgrA function as the molecular target of savirin. We used in silico docking of savirin to the C-terminal DNA binding domain (AgrAc) [27] of both S. aureus and S. epidermidis using the online server Swissdock [28]. Savirin docked to AgrAc of S. aureus between Tyr229, which is adjacent to a residue critical for AgrA folding [29] (Cys228), and Arg218 near the DNA binding interface with a calculated binding energy of −6.1 kcal/mol (Fig. 2A). Notably, mutation of Arg218 to His has been described in clinical isolates with defective agr function [30]. At this position, savirin is within hydrogen bonding distance of the backbone carbonyl of Glu217 and within π-stacking distance of Tyr229 (Fig. 2A, enlarged view). Importantly, this site differed in S. epidermidis where the key Tyr229 is a Phe and His227 is an Asn. Consistent with this, attempts to dock savirin to this site in S. epidermidis were unsuccessful, demonstrating that the DNA binding domain of AgrA is the likely target of savirin. We performed electrophoretic mobility shift assays to prove that savirin blocked the DNA binding function of AgrA. Incubation of purified AgrAc (2 µM) with the high affinity site in P2 and P3 (0.1 µM) (Fig. 2B) shifted electrophoretic mobility of the FAM labeled nucleotide and increasing concentrations of savirin (5–160 µg ml−1 or 13.5–432 µM) vs. vehicle inhibited this shift with an IC50 of 83 µM or 30.3 µg ml−1 (Fig. 2B). To prove that AgrA was the target within the pathogen, we constructed a novel reporter strain (AH3048) where plasmid-encoded AgrA constitutively produced without induction drives activation of agr::P3 lux in the absence of the rest of the agr operon, including agrB, agrC, and agrD [31]. As positive controls, we evaluated the ability of diflunisal and 4-phenoxyphenol, compounds published by others as inhibitors of AgrAc DNA binding ability [27], [32] and both inhibited dose-dependently (Fig. S5). Additionally, increasing concentrations of savirin (0.4–6.3 µM or 0.29–2.33 µg ml−1) suppressed constitutive luminescence without affecting viability where AIP2, an inhibitor of non-agrII AgrC signaling [16], [20], had no effect on luminescence demonstrating that savirin specifically suppressed AgrA-dependent activation of P3 within the microorganism (Fig. 2C). In comparison to the positive control compounds (Fig. S5), savirin inhibited luminescence at 6.3 µM equivalent to the inhibition of the controls at 100 µM. Moreover, the concentration of savirin required for optimal inhibition of the agrA reporter was equivalent to the concentration that optimally inhibited agr::P3 activation within the pathogen in strain LAC (1–5 µg ml−1)(Fig. 1). However, the concentration of savirin required to inhibit in the EMSA assay was much higher due to the excess of AgrAc required for the optimal shift in electrophoretic mobility of the labeled nucleotide. Together, these mechanistic studies indicate that AgrA within S. aureus is savirin's molecular target.
We investigated the transcriptional impact of savirin on agr virulence by microarray analysis [15], [33] and confirmed the results by qRT-PCR and direct measurement of virulence factor function in LAC and in multiple clinical isolates. All of these were performed with the same concentration of savirin, 5 µg ml−1. The effect of savirin vs. vehicle on AIP1 induced transcription in LAC was compared to the differences between LAC and Δagr LAC. Two hundred and five non-redundant transcripts were different and changed by greater than two fold between LAC and Δagr LAC (Table S1). Of these, savirin affected 122 or 60% of agr-regulated transcripts by a similar magnitude and direction including downregulation of agr secreted virulence factors (the majority of transcripts affected), transcriptional regulators, and metabolic pathways important for SSTIs [14], [15] (Fig. 3A). Of the remainder of the potentially agr-regulated transcripts not affected by savirin, the majority were hypothetical or involved in metabolism. In contrast, savirin affected only 5% of the non agr-regulated transcriptome (Table S1) demonstrating selectivity towards agr-dependent transcription. The transcripts upregulated by savirin in both LAC and Δagr LAC could reflect a stress response or be implicated in resistance or tolerance induced by savirin exposure. Of the 19 transcripts upregulated, only 5 with potential roles in drug efflux or resistance were significantly affected (Table S2). However, this did not include the two most closely implicated with antibiotic resistance, norA (SAUSA300 0680) or mecA (SAUSA300 0032). Importantly, transcripts were not affected for known stress response genes and the anti-inflammatory exotoxins induced by bactericidal agents [24], [34]–[36] or agr ablation [25], [37] (Table S1). We confirmed by qRT-PCR that savirin inhibited AIP1 induced transcripts for RNAIII and AgrA regulated genes including hla, psm alpha, pvl (lukS), agrA, and agrC, (Fig. 3B). We also confirmed by qRT-PCR that the anti-inflammatory exotoxin set7 was not affected by savirin (fold increase of vehicle 4.66±1.47 SEM vs 4.66±0.7 SEM for savirin, n = 3). Alpha hemolysin activity (Fig. 3C) and PMN lysis capacity (Fig. 3D) in savirin-treated bacterial supernatants were inhibited as well as lipase and protease activity (data not shown). Moreover, savirin inhibited psm alpha transcripts in clinical isolates of all four agr alleles (Fig. 4). Additionally, savirin reduced alpha hemolysin activity in supernatants from numerous MRSA and MSSA clinical isolates from multiple sites of infection (Fig. S6).
Given this SMI's selective effect on virulence factor production by multiple isolates, we pursued savirin's in vivo efficacy in two murine models of skin and soft tissue infection. To confirm that savirin inhibited agr signaling in vivo and that it did not affect infection with LAC Δagr, we used an airpouch skin infection model. Mice genetically deficient in the NADPH oxidase (Nox2−/−) lack control of agr::P3 activation in this model causing maximal in vivo quorum sensing [17]–[19]. The airpouch in the skin was infected with LAC expressing a fluorescent reporter of agr::P3 activation (AH1677) [19] and savirin (10 µg) was co-administered at the time of infection. Savirin treatment significantly inhibited agr::P3 activation in bacteria from a lavage of the pouch as well as consequential weight loss (as a measure of morbidity) and bacterial burden in the pouch lavage and systemically in the kidney (Fig. 5A). Moreover, when C57BL/6 mice were infected with LAC Δagr using the same model, savirin (10 µg) did not affect weight loss or bacterial burden in the pouch lavage or the kidney (Fig. 5B). These in vivo data are consistent with our in vitro data demonstrating that savirin selectively inhibits agr activation and that it has minimal impact on bacteria lacking agr.
In addition, we evaluated savirin in an established model of agr-dependent dermonecrotic skin infection in hairless immunocompetent mice [15], [38]. In this model, clearance of Δagr LAC vs. LAC was enhanced by day 7 (Fig. S7) demonstrating that agr contributes not only to early tissue injury [15], [38] but to persistence in the skin. Subcutaneous injection of savirin (5 µg) vs. vehicle at the time of infection abrogated abscesses and dermonecrosis (measured as area of ulceration) (Day 1–3) (Fig. 5C) similarly to the genetic deletion of agr (Fig. 5C, images) and prevented early morbidity (measured as weight loss). At day 3 the bacterial burden in the skin abscess was unaffected by savirin treatment (Fig. 5C), indicating that savirin inhibited toxin-induced tissue injury and not bacterial viability at this time point. In contrast, at day 7 savirin treatment promoted bacterial clearance from abscesses and systemically from the spleen (Fig. 5C), replicating the phenotype of agr deletion (Fig. S7). Because ongoing quorum sensing is likely as the pathogen reaches the required density in discrete locales to accumulate AIP and activate AgrC, we examined the effect of delayed delivery of savirin both in vitro and in vivo. Delayed delivery inhibited RNAIII production in vitro, dermonecrosis in vivo, and promoted bacterial clearance from the skin and systemically from the spleen at day 7 (Fig. 5D).
These data indicate that savirin promoted bacterial clearance not by inducing non-specific, agr-independent toxicity in the bacteria, because it did not lead to a reduction in CFU of Δagr at 24 hr (Fig. 5B) or of LAC agr+ at 3 days (Fig. 5C), but by rendering LAC less able to survive within the skin leading to clearance by skin host defense mechanisms during the resolution of the infection (Fig. 5 C, D) (Figs. S7). Skin host defense mechanisms are comprised in part of phagocytes, antimicrobial peptides, lytic lipids, and an acidic environment [39]–[41]. Given the time frame that clearance was enhanced, we postulated that savirin treatment of LAC (5 µg ml−1) but not Δagr LAC would augment killing of the bacteria in vitro by macrophages. As predicted, survival of vehicle treated LAC intracellularly from 1–5 hrs was significantly greater than savirin treated LAC (Fig. 6 A). In contrast, savirin had no effect on the intracellular survival of LAC Δagr (Fig. 6A) indicating that savirin's effect on intracellular viability was agr-specific. Because optimal killing of S. aureus within macrophage phagolysosomes requires acidification [42] and because agr regulates transcripts involved in acid resistance [43] (urease, kdpDE, Fig. 3A), we incubated savirin- and vehicle-treated LAC and LAC Δagr at pH 2.5 and evaluated viability. As with survival inside macrophages, savirin treatment promoted killing of agr+ but not Δagr bacteria (Fig. 6B). Of interest in both of these assays, the vehicle treated Δagr bacteria were more easily killed compared to the vehicle treated agr+ bacteria indicating that agr contributes to survival inside macrophages and to acid resistance (Fig. 6A,B). However, savirin treatment did not enhance killing by the antimicrobial peptide beta defensin 3, reactive oxidants, or lytic lipids (data not shown) indicating that savirin enhanced killing by some but not all skin defense mechanisms. These data suggest that enhanced killing by macrophages or the acidic environment of the skin may contribute in part to the ability of savirin to promote clearance of agr+ bacteria from the skin.
S. aureus has a remarkable propensity for developing resistance or tolerance to antibiotics [10] but whether it would become resistant to inhibition of quorum sensing, as has been postulated for Gram negative bacteria [7], is unknown. Resistance or tolerance to savirin suppression of quorum sensing could occur by either selecting for the survival of spontaneously arising agr dysfunctional mutants or by stimulating drug efflux necessitating higher concentrations of savirin for efficacy. To be clinically significant, resistance or tolerance induced by repeated exposure should occur in vivo. To address this, we serially passaged LAC with savirin (5 µg) vs. vehicle sequentially through the skin of ten individual mice 24 hrs after infection. We compared this to in vivo passage with sub-inhibitory concentrations of antibiotics known to induce resistance in USA300 strains, erythromycin and clindamycin, because of the genetic expression of ermC [44]. We chose clindamycin as a control because it is used clinically for the treatment of SSTI's and emergence of resistance to clindamycin is clinically important [44]. Passage in vivo with conventional antibiotics induced resistance to killing by clindamycin (Fig. 7A) but passage with savirin did not affect its ability to inhibit agr signaling in the savirin passaged bacteria, as measured by AIP1 induction of RNAIII by qRT-PCR, or the dose response of savirin optimal for inhibition of RNAIII production (1–5 µg ml−1) (Fig. 7B). Equivalent data were obtained with in vitro passage every day for ten days (Fig. 7C, D).
To address resistance at the colony level, we plated the in vivo passaged bacteria on milk agar plates where proteolysis is agr-dependent and contributes to colony growth (Fig. 7E). While passage of S. aureus in vitro leads to the production of agr dysfunctional colonies [45], whether this happens with in vivo passage is unknown. The passaged bacteria were diluted to give 15–20 colonies per plate, spread on plates containing either vehicle or 10 µg ml−1 savirin, and proteolytic and non-proteolytic colonies enumerated at 72 hr. Both antibiotic- and savirin- passaged bacteria plated on vehicle had equally large colonies with clear zones of proteolysis (≥1.0 mm) and neither had small non-proteolytic colonies indicative of agr dysfunction (Fig. 7E, F). In contrast, when both the savirin and antibiotic passaged bacteria were plated on savirin containing plates, the majority of the colonies converted to a non-proteolytic phenotype however a small number had zones of proteolysis ≥1.0 mm (Fig. 7E,F). These data demonstrated that plating on savirin was able to suppress agr-dependent protease production and that there was no difference between antibiotic- and savirin- in vivo passaged bacteria in their sensitivity to savirin inhibition. In total, these data indicate that under conditions where resistance to growth inhibition can be induced in vivo with a conventional antibiotic used for treatment of SSTI's, savirin exposure did not lead to loss of agr function or tolerance to savirin inhibition of agr function at both the population and colony level.
In this work we have used an SMI as a tool to address many of the concerns raised about the use of quorum sensing inhibitors as therapies or adjuncts for the prevention or treatment of antibiotic resistant bacterial infections [6], [7]. We identified an SMI in a high throughput screen that inhibited signaling of the agr quorum sensing operon in the medically significant pathogen, S. aureus (Fig. 8 model). We addressed the specificity of the inhibitor for agr signaling in this pathogen, its lack of generalized non-specific, agr-independent toxic effects on the bacterium, its molecular mechanism of action, its selective efficacy in vivo, and the potential for resistance development. If QSIs are to be efficacious for treating bacterial infections, they must work by enhancing host defense against the pathogen rendered either avirulent by the inhibitor or less fit for survival within the host. The evidence that our SMI works this way rather than by some non-specific, agr-independent toxicity on the bacterium in vivo includes: 1) its lack of effect on the number of Δagr bacteria 24 hr after infection, 2) its lack of effect on the number of agr+ bacteria early (day 3) at the site of skin infection, and 3) its lack of effect on macrophage or low pH killing of Δagr bacteria. Moreover, the reduction in CFU observed in our 2 models of SSTIs both at the site of infection and systemically (1.5–2.0 logs) was similar to that seen with conventional antibiotics tested in a murine surgical wound infection model [46] suggesting that if drugs were developed as QSIs with adequate bioavailability and pharmacokinetic properties that the eventual reduction in bacterial number could approach that seen with currently used antibiotics.
Because the majority of S. aureus infections involve skin and skin structures and are dependent on agr signaling in humans and animals [9], [14], [15], limiting antibiotic use in these infections could have a major impact on preserving conventional antibiotics for systemic, life-threatening infections [47]. In this regard, a clinical trial is ongoing which is testing whether treatment of uncomplicated skin abscesses could be limited to incision and drainage without systemic antibiotic use (NCT00730028, Uncomplicated Skin and Soft Tissue Infections Caused by Community-Associated Methicillin-Resistant Staphylococcus aureus). Our data are consistent with this approach and suggest that a QSI could either substitute for or be used as an adjunct to conventional antibiotics in this setting. Additionally, a QSI could be substituted for antibiotics used prophylactically to prevent wound infections until clinical signs of infection were apparent. Whether a QSI like savirin could be an adjunct with conventional antibiotics for treating systemic infections with or without a biofilm component is a matter of speculation and was not addressed by our studies. In fact, QSI's may have very different clinical utility in Gram negative and positive infections. Because even appropriate antibiotic use drives resistance [1], any strategy that spares conventional antibiotic use could positively impact resistance development. However, more work is needed in understanding the host defense status of patients presenting with acute bacterial skin infections because the effective use of a QSI is dependent on patients having adequate host defense systems to clear the QSI-treated, less virulent and/or less fit pathogen. Intriguingly, our compound was efficacious in mice lacking the Nox2 phagocyte oxidase, an important component of host defense against S. aureus in humans [17]–[19], suggesting that agr inhibitors may have efficacy in some patients with impaired host defense systems. More experimental work is required to determine which host defense elements are essential for agr inhibitor efficacy.
The potential for resistance development to QSI's has been addressed primarily in Gram negative bacteria (particularly Pseudomonas aeruginosa) where QS mutants (cheaters) arise during infection by taking advantage of the metabolic effort exerted by QS enabled bacteria for survival [7]. Whether this happens even experimentally in vivo with S. aureus infection is uncertain. Based on studies in Gram negative bacteria [7], the use of a QSI like savirin could give rise to mutants with a selective advantage over wild-type organisms. However, given the mechanism of action of savirin and its potential binding site in AgrAc, mutants resistant to savirin are most likely to be agr dysfunctional. Mutations in either agrA or agrC do arise in human infection [30] and savirin's potential binding site includes a known mutation in agrA (Arg 218 to His) [30]. However, elegant epidemiologic investigation has determined that these arise primarily from colonizing strains prior to the initiation of infection and not spontaneously from agr enabled bacteria during the course of infection [48]. Moreover, these mutants are less fit for transmission between patients [30], [48] suggesting that even if agr mutants arise with savirin exposure, they are unlikely to have a selective advantage over wild-type bacteria. Importantly, infection with agr mutants is primarily associated with bacteremia in hospitalized patients with impaired host defense systems and not with acute skin infection in immunocompetent individuals [14], [48]. This information along with our experimental data with in vivo passage in mice suggests that agr inhibitors may not drive the selection of agr mutants in skin infection. However, resistance or tolerance to agr inhibitors could arise by inducing a survival response in the bacteria that leads to upregulation of efflux mechanisms. Our microarray data suggest this as a possibility but neither in vivo nor in vitro passage with savirin resulted in resistance or tolerance to agr inhibition at either the population or colony level under the conditions we used. Currently, it is impossible to predict whether these issues would arise in human infection and whether our method for chronic exposure with in vivo passage in mice actually reflects how skin bacteria would be exposed to a QSI during human infection.
The mechanism of action of our SMI suggests that focusing on a site for targeted drug development within the DNA binding domain of the transcriptional regulator AgrA that is different between S. aureus and S. epidermidis, would be optimal for creating an agr inhibitor that spares the important contribution of S. epidermidis to host defense against skin infection [12], [23]. However, additional work is required to prove that savirin binds directly to the proposed site in AgrAc and to prove that savirin does not affect skin colonization by S. epidermidis. Other investigators have reported compounds that inhibit AgrA DNA binding but whether these compounds would also inhibit in S. epidermidis was not addressed [27], [32]. Our novel AgrA activation reporter assay could be duplicated using AgrA from S. epidermidis for dual screening of compound libraries for inhibitors of S. aureus but not S. epidermidis AgrA DNA binding function. Using this strategy a drug selective for agr inhibition in S. aureus could be developed with appropriate bioavailability and pharmacokinetic properties to enhance host defense against skin and soft tissue infections while minimizing the impact on normal microbiota and on antibiotic resistance.
All animal experiments were conducted at the AAALAAC accredited Veterinary Medical Unit of the New Mexico Veteran's Affairs Health Care Service in accordance with the applicable portions of the USA Animal Welfare Act as regulated by USDA, the Eighth Edition of The Guide for the Care and Use of Laboratory Animals, and the rules and regulations of the USA Department of Veterans Affairs governing experimental vertebrate animal use. These studies were approved by the NMVAHCS Institutional Animal Care and Use Committee (Protocol #10-HG-41). Human neutrophils were purchased from AllCells and the source of the neutrophils was anonymous.
AIPs1-4 were synthesized by Biopeptide Co., Inc and stored in DMSO at −80°C. Savirin (3-(4-propan-2-ylphenyl) sulfonyl-1H-triazolo [1,5-a] quinazolin-5-one, CID#3243271) was synthesized by ChemDiv, confirmed purified by HPLC, and stored in DMSO at −80°C.
The S. aureus strains used in this study were as follows: USA 300 strain LAC and its agr deletion mutant as described [15], [37]; ALC1743 (agr I [agr::P3-gfp]) and ALC3253 (Newman [agr::P3-gfp]) as described [17], [18]; AH1677 (agr I LAC [agr::P3-yfp]); AH430 (agrII 502a [agr::P3-yfp]), AH1747 (agr III MW2 [agr::P3-yfp]), AH1872 (agr IV MN TG [agr::P3-yfp]) as described [19]; and agr IV clinical isolates (NRS165 and NRS166) were obtained through the Network on Antimicrobial Resistance in Staphylcoccus aureus (NARSA) supported under NIAID, NIH contract No. HHSN272200700055C. MRSA and MSSA clinical isolates were provided by Dr. Larry Massie, Pathology Service, NMVAHCS and agr typed by PCR as described [19]. Staphylococcus epidermidis strain #12228 was obtained from ATCC and a Pseudomonas aeruginosa isolate was provided by Dr. Graham Timmins, College of Pharmacy, University of New Mexico. To generate early exponential phase, non-quorum sensing bacteria, frozen stocks were cultured in trypticase soy broth (TSB) (Becton Dickinson) as described [17]. CFU were determined after washing in PBS/0.1% Triton X-100 and sonication to disrupt clumps by plating serial dilutions on blood agar plates. Growth in TSB was measured at OD600 in 96 well plates using a plate reader (Molecular Devices) at 37°C with shaking, reading at 30 min intervals for 16 hr. The initial cultures were sufficiently diluted such that the maximal OD600 was confirmed to be within the linear range of the plate reader (<1.25 OD600). Additionally, growth was measured in 5 ml cultures in 50 ml sterile conical tubes with shaking and the OD600 determined on 1∶2 and 1∶4 dilutions of the bacterial cultures to ensure that maximal growth was adequately detected and the OD600 of the diluted samples was under 0.8 and clearly within the linear range of the spectrophotometer.
A fluorescence-based, high throughput assay was developed to screen 24,087 compounds selected for diversity from the Molecular Libraries Small Molecule Repository of the NIH Molecular Libraries Screening Center Network (summary available at http://pubchem.ncbi.nlm.nih.gov/assay/assay.cgi?aid=1206&loc=ea_ras). Using the Hypercyt flow cytometry sampling platform [49], a 384 well plate format was used that contained per well 2.5×107 early exponential phase ALC1743 containing agr::P3 driving expression of GFP. After incubation for 3 hrs with 100 nM synthetic AIP1, the induced fluorescence of the bacteria was compared between vehicle controls and compounds in 0.2% DMSO. Erythrosin B generated singlet oxygen was used as a positive control to inactivate AIP1 [17], [19]. Secondary assays included evaluation with a separate reporter strain ALC3253 in 1 ml assays and analysis of viability at 3 hr by CFU determination.
Early exponential phase non-fluorescent agr:: P3 reporter strains (2×107/ml TSB) were incubated (200 rpm at 37°C) in polystyrene tubes with broth, 50 nM synthetic AIP, or indicated concentrations of savirin for the indicated time. After incubation, bacteria were centrifuged (3000 rpm, 4 minutes, 4°C), supernatants decanted, and the pellet washed with PBS/0.1% Triton X-100, fixed with 1% paraformaldehyde containing 25 mM CaCl2, sonicated, and then evaluated for fluorescence by flow cytometry (Accuri C6, Accuri Cytometers, Inc., Ann Arbor, MI). Promoter activation was measured as induction of fluorescence.
Quantitative RT-PCR was carried out for transcripts of interest relative to 16S RNA using a probe-based assay as described with minor modifications [18], [19]. Early exponential phase S. aureus strains and clinical isolates were cultured as indicated in the figure legends. For S. epidermidis, overnight culture supernatant was used as a source of AIP. It was Millipore filtered and diluted 1∶2 with TSB. RNA was isolated and purified using the Qiagen RNA Protect Bacteria Reagent and RNeasy Mini Kit (Qiagen) using both mechanical and enzymatic disruption. RNA was purified with silica columns and subjected to DNase treatment to remove contaminating DNA. cDNA was generated using a high capacity cDNA RT kit with an RNAse inhibitor (Applied Biosystems) and a Bio-Rad thermocycler. Thermal cycling conditions were as follows: 10 minutes at 25°C, 120 minutes at 37°C, 5 minutes at 85°C, hold at 4°C. Quantitative PCR was performed using an ABI7500 Real-Time PCR system with Taqman Gene Expression master mix, ROX probe/quencher, and appropriate primer sequences (Applied Biosystems). Samples were assayed in triplicate. The data are represented as the fold increase of the transcript relative to 16S compared to the inoculum bacteria. The primer-probe sequences used were as follows: For S. aureus: RNAIII forward primer AATTAGCAAGTGAGTAACATTTGCTAGT, RNAIII reverse primer GATGTTGTTTACGATAGCTTACATGC, RNAIII probe FAM-AGTTAGTTTCCTTGGACTCAGT-GCTATGTATTTTTCTT-BHQ; psmα forward primer TAAG-CTTAATCGAACAATTC, psmα reverse primer CCCCTTCAAATA-AGATGTTCATATC, psmα probe FAM-AAAGACCTCCTTTGTTTGTTA-TGAAATCTTATTTACCAG-BHQ; hla forward primer ACAATTTTAGAGAGCCCAACTGAT, hla reverse primer TCCCCAATTTTGATTCACCAT, hla probe FAM-AAAAAGTAGGCTGGAAAGTGATA-BHQ; pvl-lukS forward primer CACAAAATGCCAGTGTTATCCA, pvl-lukS reverse primer TTTGCAGCGTTTTGTTTTCG, pvl-lukS probe FAM-AGGTAACTTCAATCCAGAATT-TATTGGTGTCCTATC-BHQ-2; 16S forward primer TGATCCTGGCTCAGGATGA, 16S reverse primer TTCGCTCGACTTGCATGTA, 16S probe FAM-CGCTGGCGGCGTGCCTA-BHQ; agrA forward primer CTACAAAGTTGCAGCGATGGA, agrA reverse primer TGGGCAATGAGTCTGTGAGA, agrA probe FAM-AGAAACTGCACATACACGCT-BHQ; agrC forward primer AAGATGACATGCCTGGCCTA, agrC reverse primer TGTGCACGTAAAATTTTCGCAG, agrC probe FAM- TGGTATCGAGAATCTTAAAGTACGTG-BHQ; and set7 forward primer ACGGAAAAACCAGTTCATGC, set7 reverse primer GCTTATCTTTGCCAATTAAAGCA, set7 probe FAM-CAGGTTATATCAGTTTCATTCAACCA-BHQ. For S. epidermidis: 16S forward primer TACACACCGCCCGTCACA, 16S reverse primer CTTCGACGGCTAGCTCCAAAT, 16S probe FAM-CACCCGAAGCCGGTGGAGTAACC-BHQ; and RNAIII forward primer ACTAAATCACCGATTGTAGAAATGATATCT, RNAIII reverse primer ATTTGCTTAATCTAGTCGAGTGAATGTTA, RNAIII probe FAM-ATTTGCTTAATCTAGTCGAGTGAATGTTA-BHQ.
Membrane integrity was measured as described using propidium iodide [25]. LAC was cultured overnight (18 hr) in RPMI supplemented with 1% casamino acids in the presence of savirin (5 µg ml−1) or vehicle control. The cultures were washed by centrifugation and the pellet resuspended in PBS supplemented with 1% BSA. The samples were set to an OD600 of 0.4 and an aliquot was heat killed (90°C for 10 minutes) to serve as a positive control. Samples (50 µl) were mixed with 1 ml PBS/1% BSA containing propidium iodide. Membrane damage was determined by measuring bacterial fluorescence by flow cytometry (Accuri C6).
Membrane potential was measured using the BacLight Membrane Potential Kit (Molecular Probes) following the manufacturer's recommendations. Membrane potential in this assay is based on the shift between the green fluorescence of DiOC2 to red in the cytosol of bacteria with higher membrane potential. The proton ionophore CCCP was used as a positive control for disrupting membrane potential. LAC was cultured with 50 nM AIP1 for 5 hr in TSB in the presence of savirin (5 µg ml−1) or vehicle control. After diluting into TSB, the bacteria were incubated with 30 µM DiOC2 in the dark for 16 min prior to analyzing by flow cytometry (Accuri, C6). Measurements from both the red and green channels were taken and data presented as a ratio of red channel divided by the green channel to reflect the shift to greater change in membrane potential.
Savirin (PubChem ID SMR000016143) was docked onto the C-terminal domain of AgrA of S. aureus AgrAc (residues 137–238 with an initiator methionine) deposited in the Protein Data Bank (PDB) accession number 3BS1 [50] using the online server SwissDock (http://www.swissdock.ch) [28]. The docking origin was set near Val235 with a search area of 10 Å in all directions and allowing for flexible side chains within 3 Å of the ligand. A model of the S. epidermidis AgrA DNA binding domain was prepared by threading the amino acid sequence (UniProt database accession number Q84FX9) onto the structural coordinates of the S. aureus protein (PDB 3BS1) using the I-TASSER server (http://zhanglab.ccmb.med.umich.edu/I-TASSER/). Savirin docking to S. epidermidis AgrAc was performed as described above for S. aureus AgrAc with the origin set to the Cα atom of Phe229. Structural images were generated using PyMOL (PyMOL Molecular Graphics System, v. 1.5.04, Schrödinger, LLC).
E. coli expressing the 6X-histidine tagged C-terminal DNA binding domain of AgrA (AgrAC) from S. aureus isolate Newman was provided by Dr. Chuan He (University of Chicago, Chicago, IL, USA). Expression and purification of AgrAC was carried out as previously described with minor modifications [29]. Briefly, AgrAC expressing E. coli were grown in Terrific broth to an OD600 of 0.6 and induced with 1 mM isopropyl β-D-1 thiogalactopyranoside overnight at room temperature. Harvested cells were flash frozen in liquid nitrogen, thawed and lysed using lysozyme and sonication. Soluble AgrAC was affinity purified using Talon Superflow Metal Affinity Resin (Clonetech) followed by gel filtration on a Superdex S200 column (GE Healthcare). Tris (2-carboxyethyl) phosphine (TCEP) at 1 mM was used as a reducing agent throughout purification. Protein was stored at −80°C in PBS, 20% glyercol, 5 mM DTT, 1 mM TCEP and 1 mM MgCl2. Electrophoretic mobility shift assays (EMSA) using purified AgrAC (2 µM) were performed as described [27] using a 16 base pair DNA duplex probe (0.1 µM) containing the high affinity LytTR binding site present in both agr P2 and P3 [27]. It was synthesized with a 3′ 6-fluorescein (FAM) to facilitate detection (Integrated DNA Technologies, USA). Samples including AgrAC, DNA probe, vehicle and/or savirin (5–160 µg ml−1 or 13.5–432 µM) were loaded in Tris-acetate-EDTA (TAE) buffer containing 10 mM dithiothreitol. Assays including the 16 bp probe were run with 10% native polyacrylamide gels.
An AgrA-dependent lux reporter strain, AH3048, was generated by transforming S. aureus Δagr strain ROJ48 [31] with pCM63 [51]. Plasmid pCM63 consists of the agrA gene cloned into plasmid pEPSA5, which placed transcription of agrA under the control of the xylose-inducible Tx5 promoter. To construct plasmid pCM63, the agrA gene was PCR amplified from AH1263 genomic DNA using primers CML609 (5′-GTTGTTGAATTCCCATAAGGATGTGAATG-3′) and CLM610 (5′-GTTGTTTCTAGACTTATTATATTTTTTTAACGTTTCTCACCG-3′), the PCR product was digested with EcoRI and XbaI, and ligated into similarly digested pEPSA5. Preliminary experiments demonstrated that light production by AH3048 increased in a xylose dose-dependent fashion, without impacting growth, up to a xylose concentration of 0.25%. For testing the impact of savirin on light production, AH3048 cultures were not induced with xylose because the constitutive level of agrA transcription from pCM63 was sufficient for luminescence induction. An overnight culture of AH3048 grown in TSB with 10 µg ml−1 chloramphenicol (for plasmid maintenance) was used to inoculate (at 1∶500 dilution) TSB containing antibiotic in 96-well microtiter plates (Costar 3603) at 200 µl per well. A 2-fold serial dilution series of savirin (0.4–6.3 µM or 0.29–2.33 µg ml−1) was used and the concentrations were tested in quadruplicate. Microtiter plates were incubated at 37°C with shaking (1000 rpm) in a Stuart SI505 incubator (Bibby Scientific, Burlington, NJ) with a humidified chamber. Luminescence and OD600 readings were recorded at 30 min increments using a Tecan Systems (San Jose, CA) Infinite M200 plate reader. Maximal light production occurred after 6 hrs of growth. As a specificity control, a 2-fold dilution series (0.5 nM to 1000 nM) of AIP-2 (Anaspec, Fremont, CA) was tested in quadruplicate, as well as 12 control wells containing vehicle (DMSO). As positive controls, two compounds demonstrated by others to inhibit AgrAc in EMSA assays, diflunisal and 4-phenoxyphenol (Sigma) [27], [32], were evaluated for luminescence inhibition in the same assay at concentrations from 1.56–100 µM.
To compare the transcript levels of LAC and the Δagr mutant in the presence or absence of savirin (5 µg ml−1), the bacteria were grown for 5 hr in TSB with 50 nM AIP1 or an equivalent amount of DMSO as the vehicle control and processed for microarray analysis as described [33]. The comparisons were LAC vehicle vs. LAC savirin, Δagr vehicle vs. Δagr savirin, and LAC vehicle vs. Δagr vehicle, n = 3. The bacterial RNA was purified as described [32]. Samples were hybridized to a custom Affymetrix GeneChip (RMLchip7) that contains all open reading frames of the USA300 genome. Samples were scanned using Affymetrix 7Gplus GeneChip scanner according to standard GeneChip protocols with the image files converted using GeneChip Operating Software (GCOS v1.4). The data were quantile-normalized and a 3-way ANOVA with multiple test correction using the false discovery rate (p<0.05) was performed using Partek Genomics Suite software (Partek, inc. v6.5). These data were combined with fold change values, signal confidence (above background), and call consistency (as a percent) as calculated using custom Excel templates to generate final gene lists for each comparison. The microarray data were deposited in the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/projects/geo/) under the accession number GSE52978. All microarray data are MIAME compliant.
Alpha hemolysin activity was measured in 0.45 µm filtered cultured supernatant standardized by OD600 after bacterial strains were grown overnight in TSB in the presence or absence of savirin (5 µg ml−1). The assay was performed using rabbit erythrocyte lysis as described [18]. One unit of hemolytic activity was defined as the amount of bacterial supernatant able to liberate half of the total hemoglobin from the erythrocytes and expressed as HA50.
The ability of secreted toxins to lyse human neutrophils was determined by LDH release. Overnight supernatant from MRSA agr group I clinical isolate #32 generated with either savirin (5 ug ml−1) or DMSO vehicle control was 0.45 µm filtered, stored at −80°C, and thawed on ice. Human neutrophils (AllCells) were washed twice in saline to remove EDTA, suspended in RPMI with 10 mM HEPES and 1% HSA, and assessed for viability by Trypan blue staining (>97%). The experiment was run in triplicate and each tube contained 3×106 neutrophils in 100 µl RPMI to which was added 100 µl of either RPMI, TSB diluted 1∶5 or 1∶10 in RPMI, or treated supernatants diluted 1∶5 or 1∶10 in RPMI. PBS with 0.1% Triton-X100 (100 µl) was used for 100% lysis. Tubes were incubated at 37°C in a 5% CO2 incubator for 1 and 2 hours. At each time, the tubes were centrifuged at 13,000 rpm, at 4°C, for 5 minutes. Cell free supernatant (100 µl) was transferred to a micotiter plate and immediately processed for LDH according to the Cytotoxicity Detection Kit (Roche). A blank was created for each plate with 10%TSB in RPMI. The data are depicted as the percentage of total lysis after correction for LDH release stimulated by media alone.
For all in vivo experiments, savirin was solubilized at 1 mg ml−1 in 0.5% hydroxypropyl methylcellulose (Sigma) in endotoxin-free sterile water made 3.0 mM in NaOH with cell culture tested 1 N NaOH (Sigma), and put through a 0.22 µM filter (Millex-GV). The vehicle control was the HPMC used to solubilize the savirin. Sample sizes were determined by preliminary experiments to determine the number of mice required to observe significance. Dermonecrosis model: SKH1 hairless immunocompetent mice (≈8–16 wk, ≈26–34 g, male) were obtained from Charles Rivers (Wilmington, MA). At Day 0, early exponential phase bacteria (4×107) washed in sterile normal saline were injected concurrently with savirin (5 µg) or vehicle in 50 µl subcutaneously into the flank using a 3/10 cc insulin syringe with a 28 ½ gauge needle (Becton Dickinson). For delayed delivery, 10 µg savirin was administered 24 and 48 hr after infection in 50 µl. The animals were divided into two groups to have equivalent mean abscess sizes prior to administering drug or vehicle. Abscess area (maximal on Day 1) and ulcer area (necrosis optimal on Days 3–4) were measured with calipers as described [15], [38] and recorded daily in addition to weight loss. The slightly raised abscess area (mm2) was calculated from the equation (π/2)[(length of the abscess)×(width of the abscess)]. The flat ulcer area (mm2) was calculated from the equation (length of the ulcer)×(width of the ulcer) or alternatively from digital images using Adobe Photoshop standardized to a micrometer with equivalent results. On Day 3 or Day 7, the mice were euthanized using isoflurane inhalation. The abscess/ulcer area was excised (1.5 cm2) and the spleens removed. Tissues and spleens were placed in 1 ml of HBSS/0.1% HSA in a bead-beating tube containing sterile 2.3 mm beads (Biospec) and were processed for bacterial CFU by homogenizing the spleens in a bead beater, diluting all samples 1∶10 in 1 ml PBS/0.1%Triton, sonicating, and plating serial dilutions on blood agar as described [17]–[19]. Airpouch model: age matched Nox2−/− male mice (Jackson Labs) or C57BL/6 male mice (Charles Rivers) were infected with either 2×107 early exponential phase non-fluorescent AH1677 bacteria (Nox2−/−) or 5×107 LAC Δagr (C57BL/6) into an air pouch generated by the injection of 5 ml of air subcutaneously as described [17]–[19]. Savirin (10 µg) vs. vehicle in 50 µl was injected into the pouch at time 0. After 24 hours, weight loss was determined, the air pouch was lavaged with HBSS/0.1% HSA and the kidneys removed. The bacteria in the lavage were analyzed by flow cytometry for promoter activation (AH1677) and both the lavage and kidneys processed as above for CFU determination.
Early exponential phase LAC+50 nM synthetic AIP1 or Δagr LAC (2×107/ml TSB) were incubated for 5 hr at 37°C with shaking (200 rpm) in the presence of savirin (5 µg ml−1) or vehicle control. Bacteria were opsonized (1×108/ml) with rabbit IgG anti-Staphylcoccus aureus (Accurate Antibody YVS6881) (100 µg/ml) in phenol red-free Dulbecco's Modified Eagle Media, DMEM, containing 4.5 g/L D-glucose/2%Hepes+1% FCS). The experiment was performed in triplicate. Murine macrophage RAW264 cells (5×106) in 250 µl of DMEM+2% FCS were combined with 5×106 opsonized bacteria in 250 µl of DMEM+1% FCS (MOI 1∶1) in sterile polystyrene 12×75 mm tubes, centrifuged briefly to initiate contact, and incubated for 1 hr at 37°C in 5% CO2. The infected cells were treated with lysostaphin (Sigma) (2 µg/ml for 15 min) to kill extracellular bacteria and then washed and suspended in fresh media. Half of the samples were incubated for an additional 4 hrs. To determine the intracellular CFU at 1 and 5 hr, the relevant cells were centrifuged, suspended in PBS/0.1% Triton-X-100 and sonicated to disrupt cells and dilutions plated on blood agar. The cell line was tested for Mycoplasma sp. contamination by PCR (Life Technologies).
Early exponential phase LAC+50 nM synthetic AIP1 or Δagr LAC (1×108/ml DMEM, 4.5 g/L D-glucose/2%Hepes) were incubated for 5 hr at 37°C with shaking (200 rpm) in the presence of savirin (5 µg ml−1) or vehicle control. Bacteria were centrifuged, washed, resuspended in DMEM/2%Hepes acidified with either HCl to pH 2.5 or 10 µg ml−1 linoleic acid (Sigma) and incubated for the indicated times. Dilutions were plated on sheep blood agar to determine the residual viability.
In vitro data were analyzed by the two-tailed Student's t-test or two way measures ANOVA as indicated in figure legends. In vivo data were analyzed by the two-tailed Mann-Whitney U test for non-parametrics. All evaluations were conducted using GraphPad Prism v. 5.o and results were considered significantly different with p<0.05.
|
10.1371/journal.ppat.1000547 | Expression and Processing of a Small Nucleolar RNA from the Epstein-Barr Virus Genome | Small nucleolar RNAs (snoRNAs) are localized within the nucleolus, a sub-nuclear compartment, in which they guide ribosomal or spliceosomal RNA modifications, respectively. Up until now, snoRNAs have only been identified in eukaryal and archaeal genomes, but are notably absent in bacteria. By screening B lymphocytes for expression of non-coding RNAs (ncRNAs) induced by the Epstein-Barr virus (EBV), we here report, for the first time, the identification of a snoRNA gene within a viral genome, designated as v-snoRNA1. This genetic element displays all hallmark sequence motifs of a canonical C/D box snoRNA, namely C/C′- as well as D/D′-boxes. The nucleolar localization of v-snoRNA1 was verified by in situ hybridisation of EBV-infected cells. We also confirmed binding of the three canonical snoRNA proteins, fibrillarin, Nop56 and Nop58, to v-snoRNA1. The C-box motif of v-snoRNA1 was shown to be crucial for the stability of the viral snoRNA; its selective deletion in the viral genome led to a complete down-regulation of v-snoRNA1 expression levels within EBV-infected B cells. We further provide evidence that v-snoRNA1 might serve as a miRNA-like precursor, which is processed into 24 nt sized RNA species, designated as v-snoRNA124pp. A potential target site of v-snoRNA124pp was identified within the 3′-UTR of BALF5 mRNA which encodes the viral DNA polymerase. V-snoRNA1 was found to be expressed in all investigated EBV-positive cell lines, including lymphoblastoid cell lines (LCL). Interestingly, induction of the lytic cycle markedly up-regulated expression levels of v-snoRNA1 up to 30-fold. By a computational approach, we identified a v-snoRNA1 homolog in the rhesus lymphocryptovirus genome. This evolutionary conservation suggests an important role of v-snoRNA1 during γ-herpesvirus infection.
| Epstein-Barr virus (EBV) infects about 90% of people worldwide and is associated with different types of cancer. So far, only two large virus-encoded non-coding RNAs (EBER1 and EBER2) and 25 microRNAs (miRNAs) have been identified in the EBV genome. In this study, we report identification of the first member of another abundant non-coding RNA class, a small nucleolar RNA (snoRNA), designated as v-snoRNA1. We show that v-snoRNA1 is located in the nucleolus and interacts with the same proteins as reported for canonical eukaryal snoRNAs. Its biological function is consistent with its high conservation in a distantly related simian herpesvirus genome. Interestingly, v-snoRNA1 might serve as a miRNA-like precursor, which is processed into a 24 nt sized RNA species, designated as v-snoRNA124pp. The viral DNA polymerase BALF5 was identified as a potential target for v-snoRNA124pp. Taken together, these experiments strengthen the crucial function of v-snoRNA1 in EBV infection.
| The Epstein-Barr virus (EBV), a member of the γ-herpesvirus subfamily, possesses a large (170 to 180 kb) double-stranded DNA genome. EBV infection is etiologically linked with various cancers of the lymphoid and epithelial lineages that include Burkitt's lymphoma (BL), Hodgkin's disease, nasopharyngeal carcinoma (NPC) and post-transplant lymphoproliferate disease (PTLD) [1]–[4]. In vitro and in vivo, EBV transforms normal B cells through establishment of a type III latency during which a restricted set of viral genes is expressed (eight Epstein-Barr nuclear antigens and two latent membrane proteins) [5]. More restricted expression patterns such as latency type II in NPC and latency type I in BL have also been characterized. In fact, recent work on Burkitt's lymphoma has shown that a subset of these tumours display a latency pattern intermediate between latency I and III showing that the boundaries between the latency types are not always sharply established as initially thought [6].
More then two decades ago, the group of J. Steitz discovered two highly abundant ∼170-nt long non-coding RNAs (ncRNAs) in the EBV genome, designated as Epstein-Barr encoded RNAs (EBER1 and EBER2) [7]. EBER RNAs have subsequently been shown to bind to human ribosomal protein L22. However, no unequivocal biological functions could be assigned to EBER transcripts, up till now [8]. The list of non-coding RNAs encoded by EBV has since rapidly expanded with the recent discovery of 25 microRNAs (miRNAs) [9]–[14].
In addition to miRNAs, numerous other ncRNAs have been discovered in all three domains of life, i.e. Archaea, Bacteria and Eukarya, as well as in various viruses [15],[16]. A large number of these ncRNA species were found to be involved in multiple regulatory functions including cellular differentiation and development, chromatin architecture, transcription and translation, alternative splicing, RNA editing, virulence and stress responses [17]–[20].
Small nucleolar RNAs (snoRNAs) consist of more than 200 stable ncRNA species in Eukarya of about 60 to 300 nt in size which are located in a sub-nuclear compartment, the nucleolus [21],[22]. SnoRNAs guide nucleotide modifications within ribosomal RNAs (rRNAs) or spliceosomal RNAs (snRNAs), i.e. 2′-O-ribose methylation or pseudouridylation, respectively. The snoRNA class has been identified in Archaea and Eukarya, but not in Bacteria, and is subdivided into box C/D and box H/ACA snoRNAs. In Eukarya, the majority of snoRNAs is located within introns of protein-coding genes and is processed by splicing followed by endo- and exonucleolytic cleavage [19],[23],[24].
Each member of the box C/D snoRNA family possesses characteristic sequence elements called box C (PuUGAUGA) and box D (CUGA), optional degenerate C′/D′ boxes and a short 5′-3′ terminal stem structure [24],[25]. 10–21 nt long sequence-specific antisense elements upstream of the boxes D/D′ guide the box C/D snoRNA core proteins fibrillarin, a RNA methyltransferase, Nop56, Nop58 and the 15.5 kD protein to the target RNA. 2′-O-methylation of the ribose at the fifth nucleotide upstream of the D/D′ box on the target RNA is carried out by the fibrillarin core protein [24]. Box H/ACA snoRNAs possess a distinctive common ACA sequence motif at their 3′-terminus and one to two stem-loop structures linked by a hinge (the so-called H-box motif: ANANNA, with N being any nucleotide), and guide the conversion of uridine to pseudouridine within the RNA target [26],[27]. The large number of conserved modifications in functionally conserved regions of rRNAs, such as the peptidyl-transferase centre, has suggested an important role for rRNA modifications in fine-tuning the structure and/or function of rRNAs [28]. It is important to note that a significant number of so-called “orphan” snoRNAs, lacking rRNA or snRNA targets, have been identified in Eukarya [29],[30]. However, the biological functions of orphan snoRNAs are still elusive.
In this study, we report, for the first time, the identification of a functional C/D box snoRNA within the EBV genome. We demonstrate that this viral snoRNA exhibits all bona fide box C/D snoRNA features with respect to its processing and expression, nucleolar localization as well as to canonical core protein binding partners. We also provide evidence that v-snoRNA1 is processed into a 24 nt long miRNA-like species which might target the 3′-UTR of the viral DNA polymerase mRNA.
We have established an experimental strategy, designated as SHORT, to identify viral-induced ncRNAs in cord blood lymphocytes (CBL) infected with the EBV strain B95.8 [31]. The SHORT method is based on subtractive hybridisation of ncRNA populations of virus-infected cells from non-infected cells. NcRNAs, selectively expressed in the infected cell population, were subsequently converted into cDNAs. Sequencing of a small number, i.e. about 500 cDNA clones, allowed identification of several ncRNAs from the human as well as from the EBV genome whose expression was up-regulated upon viral infection [32].
Deep-sequencing analysis of 40.000 cDNA clones from this subtracted cDNA library further extended the list of differentially expressed ncRNAs (Hutzinger et al., manuscript in preparation). Interestingly, one of these sequences was represented by 95 cDNAs and exhibited all defining features of canonical C/D box snoRNA sequence motifs, i.e. C, C′, D′ and D boxes [24],[25]. Crucially, this potentially novel snoRNA species mapped to the EBV genome and was therefore designated as v-snoRNA1 (Figure 1A, and see above; Accession number FN376861). It is noteworthy that the canonical terminal stem-structure, formed by the 5′ and 3′ ends of eukaryal snoRNAs, was absent in the viral snoRNA, a feature shared with snoRNAs identified from archaeal or fungal species [33],[34].
To assess expression of v-snoRNA1, northern blot analysis was performed employing RNA from EBV-positive cell lines (Rael, Raji, BL2-B95.8, BL41-B95.8 and a LCL generated in vitro with the B95.8 virus) or EBV-negative cell lines (BL2 and BL41; Figure 1B). As expected, v-snoRNA1 could only be detected in infected cells but not in the EBV-negative control cells. Comparison with an internal RNA marker showed that the hybridized RNA species was 65 nt in size, which fully matched the size suggested by the original sequence obtained by cDNA cloning (see above and Figure 1B). Repeated attempts to identify v-snoRNA1-precursor transcripts by northern blot analysis were unsuccessful (unpublished data), suggesting that they are subjected to rapid processing.
The v-snoRNA1 gene is located within the BamHI A rightward transcripts, known as BARTs, on the sense strand of the viral genome and maps about 100 nt downstream of the EBV mir-BART2 (Figures 2A and 2B). The BARTs represent abundant RNA species in EBV that are expressed in all latently infected EBV-B cell lines, in peripheral blood B cells of EBV-positive individuals and, at higher levels, in nasopharyngeal carcinoma [35],[36]. They do not encode for proteins but are processed into 22 different BART miRNAs (Figure 2A) [14]. Thereby, v-snoRNA1 as well as mir-BART2 arise from the same intron, which was found to be 4.9 kb in size in the AG876 strain (Accession number AJ507799) [35].
BART transcripts were previously shown to be predominantly transcribed from the P1 promoter [36]. However, P2 promoter-initiated BARTs were also detected in different B-cell lines with the exception of the EBV-positive BL cell line Raji. As shown in Figure 1B, v-snoRNA1 expression was verified in all tested EBV cell lines, including Raji cells, although expression levels varied considerably. In particular, v-snoRNA1 was expressed in Raji cells at barely detectable levels. Therefore, we infer that v-snoRNA1 transcription can be initiated at the P1 promoter but that the P2 promoter might be required to obtain full expression.
To determine the sub-cellular location of v-snoRNA1 within EBV-infected cells, we employed fluorescent in situ hybridization (FISH) with dye-labeled antisense oligonucleotides complementary to v-snoRNA1. As a control, we also investigated the localization of U3 snoRNA, which is known to be localized in the nucleolus [37],[38]. Examination of EBV-infected BL2 cells by confocal microscopy revealed that both v-snoRNA1 and U3 snoRNA in fact co-localized to the nucleolus (Figure 3A). In contrast, a v-snoRNA1 hybridization signal was absent in non-infected B cells.
Canonical C/D box snoRNAs have previously been shown to bind to four snoRNA core proteins: fibrillarin, Nop56, Nop58, and the 15.5 K protein, respectively. These proteins have previously been shown to be strictly required for RNA maturation, stabilization and function [22],[39]. The C/D box proteins assemble with snoRNAs thus forming ribonucleo-protein complexes (snoRNPs) that localize to the nucleolus. In order to assess whether v-snoRNA1 assembles into a canonical C/D box snoRNP, binding of v-snoRNA1 to three of these canonical snoRNA-binding proteins (fibrillarin, Nop56 and Nop58) was assessed by co-immunoprecipitation using specific antibodies. Immuno-precipitated samples were subsequently analyzed for the presence of v-snoRNA1 by northern blot analysis. These assays demonstrated that v-snoRNA1 and the canonical U81 snoRNA, used as a positive control, were both co-immunoprecipitated with similar efficiencies with antibodies against all three snoRNA-binding proteins (Figure 3B). In contrast, none of the snoRNAs was precipitated in controls without antibodies or employing an IgG-specific antibody. Hybridization with an oligonucleotide specific for 5.8S rRNA was used to test for the specificity of the employed antibodies. Thereby, a faint, unspecific signal was detected in all samples after antibody addition, except the control without an antibody. This is likely caused by the high expression levels of 5.8S rRNA in our samples. From these results we conclude that the newly identified 65 nt long viral RNA transcript displays all hallmark features of a genuine box C/D snoRNA.
A common trait shared by all herpesviruses is their ability to infect their target cells under several modes; cells can support lytic replication during which new virus progeny is replicated or instead induce virus latency. Viral proteins used in both modes are usually, but not always, distinct. We therefore assayed v-snoRNA1 expression in latently infected cells or in cells undergoing lytic replication. We took advantage of LCLs established with viruses that are devoid of the lytic immediate early gene BZLF1 (ΔBZLF1) and therefore cannot initiate lytic replication [40] and examined v-snoRNA1 expression in these cells by northern blot analysis (Figure 4). Northern blot signals were clearly visible in these cells thereby demonstrating that v-snoRNA1 is a latent transcript. We then performed the same experiment with replication-competent 293/EBV-wt cells lytically induced by transfection of the BZLF1 gene (Figure 4). Comparison with non-induced cells showed that the v-snoRNA1 expression levels were up-regulated up to 30-fold following induction (Figure 4). V-snoRNA1 is therefore especially part of the EBV lytic expression programme.
In an attempt to discover the function of v-snoRNA1 during the EBV life cycle, we constructed a recombinant virus that lacks a functional v-snoRNA1. To this aim, the C-box motif of v-snoRNA1 from the B95.8 strain was exchanged against the sequence of the kanamycin resistance gene flanked by two FLP recombinase recognition sites (Figure 5A). Excision of this cassette left an unrelated bacterial sequence containing a HindIII restriction site in place of the box C of v-snoRNA1 (Figure 5A and 5B, lane 2). DNA from the recombinant virus was stably transfected into 293 cells to generate a virus producer cell line, here referred to as 293/Δv-snoRNA1. Multiple clones were screened for their ability to support virus replication. One of the replication-competent clones was chosen at random for further experiments. Recombinant episomes purified from this producer cell line and transformed into E. coli cells were found to be intact as assessed by restriction analysis (Figure 5B, lane 3). Sequencing of the recombination site on these rescued episomes confirmed exchange of the Box C against unrelated DNA TTTCCCGCGCCAAGCTTCAAAAGCGCTCTGAAGTTCCTATACTTTCTAGAGAATAGGAACTTCGGAATAGGAACTTCCAACC (EBV DNA around the insertion is indicated in bold). A northern blot, performed on 293/Δv-snoRNA1 cells using a v-snoRNA1-specific probe, yielded negative results while signals could be clearly identified in the 293/EBV-wt positive control (Figure 5E, left panel). We therefore conclude that the Δv-snoRNA1 virus is devoid of the viral snoRNA and that destruction of the putative C box of v-snoRNA1 is sufficient to exert this effect.
We then conducted a series of experiments aiming at defining phenotypic traits of the mutant strain. We first assessed the ability of the 293/Δv-snoRNA1 to support viral replication. Viral titres were quantified either as packaged viral genome-equivalents (physical titres) or as green Raji units, i.e. as the concentration of viruses able to infect the Raji cell line determined by exposure to a limiting dilution of the viral supernatants (functional titres). Both assays revealed nearly identical titres for both the mutant and the wild type control (Figure 5D). The Δv-snoRNA1 viruses and producer cell line were then examined in electron microscopy; both displayed normal morphological features: encapsidation, primary and secondary egress were unchanged in the absence of the viral snoRNA (unpublished data). We further evaluated viral gene expression by western blot or immunostains (BZLF1, EA/D-BMRF1, gp350). Again, we could not discern any differences between the mutant and its wild type counterpart (unpublished data).
We then exposed various established cell lines or primary cells to the Δv-snoRNA1 mutant and monitored the efficiency of infection by counting the percentage of GFP-positive (293 cell line, primary epithelial cells) or EBNA2-positive (primary B cells) lymphocytes three days post-infection. The rate of infection was nearly identical in both wild type and mutant viruses (unpublished data). We finally investigated the transforming capacity of the mutant by performing infections of normal resting B cells from three different normal individuals at decreasing multiplicity of infections (Figure 5D). Wild type and mutant viruses both exhibited a transforming potential that resulted in a very similar number of outgrowing cell clones. We confirmed the identity of the viruses present in the growing LCLs by northern blot analysis; only LCLs generated by infection with wild type B95.8 virus expressed the snoRNA while those infected with Δv-snoRNA1 remained negative (Figure 5E, right panel).
The majority of snoRNAs have been found to target rRNAs or snRNAs by guiding ribose methylation or pseudouridinylation, respectively. In contrast, a number of snoRNAs lack telltale complementarities to canonical targets and hence are designated as “orphan” snoRNAs [19],[24],[30]. We therefore examined 18S and 28S rRNAs for putative v-snoRNA1 target sites using criteria established by Cavaille and Bachellerie [25]: the putative target sites were required to display at least a 7 nucleotides-long perfect complementarity with a region that ended within 3 nucleotides of the end of the snoRNA antisense boxes, and at most one nucleotide should be involved in a bulge or loop [25]. In particular we searched for putative target sites of the v-snoRNA1 box D antisense elements and for two potential alternative box D′ antisense elements (see Figure 6A). Using a program that was successfully used to predict targets of bacterial ncRNAs [41] we identified two putative ribose methylation site within the 18S rRNA and 23 sites within the 28S rRNA for box D′ (Table S1). However, none of the predicted target sites coincided with known methylated nucleotides within 18S and 28S rRNA. The same strategy applied to box D failed to reveal any putative ribose methylation sites within rRNAs. Nevertheless, we experimentally tested the ribose methylation status of the highest-scoring predictions for rRNA targets (Figure 6B) by primer extension analysis [42],[43]. However, no methylation at the predicted nucleotide positions C617 of human 18S rRNA and C3140 and C3152 of human 28S rRNA was observed in EBV-infected LCL B95.8 cells (data not shown), suggesting that v-snoRNA1 is a member of the still growing class of orphan snoRNAs.
In addition to full-length cDNA clones encoding v-snoRNA1, we also identified nine identical partial cDNA clones of 24 nt in size in our cDNA library derived from the very 3′-end of v-snoRNA1 (Figure 2B). Previously, two studies were able to demonstrate processing of specific snoRNA species into functional miRNAs [44],[45]. Attempts to verify expression of the 24 nt long v-snoRNA1-derived processing product, designated as v-snoRNA124pp, by northern blot analysis with conventional DNA oligonucleotide probes or by splinter ligation [44],[46] were unsuccessful (data not shown). In contrast, by applying a locked nucleic acid (LNA) probe, complementary to v-snoRNA124pp, we were able to verify its expression (Figure 7). An additional hybridization signal at 40 nt was also observed that might represent a processing intermediate. All hybridization signals, except for full length v-snoRNA1, were only detected in the 293/EBV-wt cells induced with BZLF1, likely due to the high expression level of v-snoRNA1 within this strain. Notably, v-snoRNA124pp was not detected in the snoRNA knock-out strain (Figure 7).
Since the 3′-UTR of the BALF5 mRNA exhibits full complementarity to v-snoRNA124pp (Figure 8) we investigated whether it might serve as a potential target site for cleavage by applying a 5′-RACE approach, as previously described [47],[48]. 5′-RACE products from the predicted 3′-UTR cleavage site were amplified by specific primers and sequenced (Figure 8). Indeed, we detected two clones corresponding exactly to a predicted cleavage site by v-snoRNA124pp 11 nt from its 5′-end in 293/EBV-wt cells induced with BZLF1 which exhibits highest expression levels of v-snoRNA1 (Figures 4 and 7). Remaining clones from this region exhibited shorter sequences likely due to exonucleolytic degradation of the BALF5 mRNA following initial cleavage by v-snoRNA124pp as described previously for plant miRNAs [47]. Notably, not a single sequence was observed that was longer than the expected size, which is indicative of a specific cleavage event triggered by v-snoRNA124pp and followed by exonucleolytic degradation. In contrast, no fragments cleaved within the 3′-UTR of BALF5 mRNA were observed in the snoRNA knock-out strain.
The identification of a snoRNA species in a viral genome raised two obvious questions: is v-snoRNA1 conserved among the different herpesvirus subfamilies or even among several EBV strains and do v-snoRNA1 homologs exist in other virus families? This prompted us to perform a BLAST alignment search using all available databases. This search showed that the v-snoRNA1 sequence is 100% conserved among the tested EBV strains (B95.8, AG876, M81, GD1, Raji). It further revealed that the distantly related rhesus lymphocryptovirus (rLCV) genome (exhibiting an overall sequence identity of 65% with the EBV genome; Accession number NC_006146) contains a 65 base pair sequence that shows 86% identity with v-snoRNA1 (Accession number FN376863). In particular, the canonical D, D′ and C, C′ boxes were universally conserved as well as antisense elements, preceding D or D′ boxes. This high degree of sequence identity did not extend to the v-snoRNA1 flanking regions; these showed only 69% sequence identity and were therefore clearly less conserved (Figure 9A). Northern blot analysis, employing an rLCV-specific antisense oligonucleotide, confirmed that the rLCV sequence homolog of v-snoRNA1 is actively transcribed and processed into an RNA species of 65 nt in simian B cells (Figure 9B). Despite the high degree of sequence identity between human and rLCV v-snoRNA1s, hybridization with the rLCV-specific probe did not detect its EBV counterpart. Altogether, these findings strongly indicate that rLCV also encodes a box C/D snoRNA homolog to v-snoRNA1.
Herpes virus genomes carry numerous cellular gene homologs [49]. Many of these genes encode house keeping proteins but others serve more specialized functions e.g. within the host immune system. This is particularly true of γ-herpesviruses whose genomes encode homologs of cytokines (e.g. CSF-1 and IL10 for EBV, IL6 for Kaposi's sarcoma-associated herpesvirus (KSHV) or of anti-apoptotic mediators (e.g. BCL2 in EBV and KHSV). These striking homologies between a virus and a cellular genome were reinforced by the discovery that herpesviruses encode multiple miRNA clusters. Here we report that herpesviruses and their host share yet another fundamental ncRNA species.
Deep-sequencing analysis of a subtracted cDNA library that was constructed to specifically identify transcripts expressed in EBV-infected B cells allowed discovery of a viral transcript that exhibited all defining features of a C/D box snoRNA. Indeed, v-snoRNA1 comprises canonical C/C′ as well as D/D′ boxes. It is of note that v-snoRNA1 is lacking the canonical terminal stem-structure usually encountered in eukaryal snoRNAs. In this respect, v-snoRNA1 appears to be closer to snoRNA species previously identified in fungi or in the domain of Archaea [33],[50]. In addition to the EBV-encoded v-snoRNA1, the genome of the Herpesvirus saimiri (HVS), a member of the γ-herpesvirus family, was recently reported to encode seven small nuclear RNAs [51],[52]. Thereby, in latently infected HVS-transformed T cells, the Herpesvirus saimiri U RNAs (HSURs) represent the most abundant viral transcripts. Similar to EBERs, HSURs are not essential for viral replication or transformation, but are involved in the activation of specific genes in virus-transformed T cells during latency [51].
V-snoRNA1 was found to be expressed in all samples of a panel of EBV-positive cell lines that included several BLs and in particular the latency I Rael cell line, LCLs and the 293/EBV-wt producer cell line (Figure 1). Detection of reduced levels of v-snoRNA1 in LCLs, generated with the BZLF1-null virus that therefore cannot undergo lytic replication, demonstrated that v-snoRNA1 is an integral part of the EBV latent transcription program (Figure 4). However, expression levels of v-snoRNA1 increased significantly up to 30-fold upon induction of the lytic replication cycle. This is consistent with a model that v-snoRNA1 serves, presumably different, functions in both the latent and the lytic mode of infection.
Three findings demonstrated that v-snoRNA1 is likely to represent a fully functional ncRNA species. V-snoRNA1 was found to co-localize with canonical snoRNA to the nucleolus (Figure 3). Furthermore, we could show that v-snoRNA1 assembles into a canonical snoRNP that at least includes the fibrillarin, Nop56 and Nop58 proteins. Finally, selective destruction of the C box resulted in a complete down-regulation of steady state levels of v-snoRNA1 (Figure 5E). This is consistent with previous work that ascribed an essential role in the regulation of the stability of snoRNA to this sequence motif [21],[53],[54].
V-snoRNA1 could be localized to the BARTs region which follows a complex splicing pattern and also encodes a cluster of non-coding miRNA genes (Figure 2). V-snoRNA1 was located outside the putative BARTs open reading frame and is therefore, as previously observed for canonical eukaryal snoRNAs, likely processed from an intron. The BARTs transcripts can be initiated from two promoters P1 and P2 [36]. Analysis of v-snoRNA1 expression levels showed a large degree of variation within the tested cell lines, as was also observed for EBV's miRNAs [55]. In principle, this could be related to the highly variable virus copy numbers among different EBV-positive cell lines. Alternatively, it may be related to the propensity of some of these cell lines to undergo lytic replication. The low expression levels of v-snoRNA1 in Raji are probably due to an inactive BART P2 promoter; this suggests that the P2 promoter initiates most of the v-snoRNA1 transcripts.
The discovery of a snoRNA in a Herpesvirus genome prompted us to search for homologs in other viral or cellular genomes. This search revealed that the v-snoRNA1 is strictly conserved across five distinct EBV strains. It further led to the identification of a transcript within the rLCV genome that displays a high degree of homology to v-snoRNA1. This genetic element comprises perfectly conserved canonical C/D and C′/D′ boxes and was expressed in a simian LCL which suggests that rLCV also encodes a snoRNA. Discovery of a v-snoRNA1 homolog in rLCV is not entirely unexpected; rLCV is the closest EBV relative as both genomes exhibit 65% sequence identity and, therefore, display more than 80% sequence identity for protein-coding genes and ncRNA genes. Indeed, seven rLCV miRNA were found to be closely related to their EBV counterparts [11]. The relatively crude approach (BLAST) we initially took failed to reveal further v-snoRNA1 relatives; we nevertheless consider that this question is still open and hope that our work will stimulate research in this direction.
The strict conservation of v-snoRNA1 domains within various EBV strains and among evolution strongly suggests that this element serves an essential role in the natural history of EBV infection. We therefore initiated a series of experiments that aimed at defining potential functions of v-snoRNA1. We thereby combined a computational with an experimental approach to determine putative ribosomal or spliceosomal RNA targets for v-snoRNA1 using previously identified criteria (see results section). However, both attempts failed to identify any obvious rRNA candidates. Hence, v-snoRNA1 can be assigned in all probability to the class of so-called “orphan” snoRNAs that lack rRNA or snRNA targets (see below).
Another strategy to discover the function of v-snoRNA1 consisted in constructing a v-snoRNA1-null mutant and defining its phenotypic traits using well-characterized in vitro assays. As of now, the Δv-snoRNA1 mutant remained indistinguishable from its wild type counterparts in terms of lytic replication, infection and B cell transformation (Figure 5). However, this does not exclude that v-snoRNA1 serves an important function during the virus life cycle; unraveling miRNAs contributions to EBV infection has also proven a difficult enterprise. Aside from a few notable exceptions such as miR-BART5 and miR-BART2 that respectively target the cellular gene PUMA [56] and the viral gene BALF5 [57] or the BART cluster 1 and BHRF1-2 that respectively modulate LMP1 expression and BHRF1 mRNA processing [58],[59], the essential functions served by these ncRNAs remain unclear. Indeed, the B95.8 strain that lacks a large number of miRNAs perfectly replicates and immortalizes primary B cells with high efficiency.
Recently, specific snoRNA species have been characterized as miRNA precursors, which are processed to mature miRNAs and assemble into a functional RNA induced silencing complex [60],[61]. Indeed, by deep-sequencing we identified nine identical cDNA clones of 24 nt in size, that mapped to the very 3′-end of v-snoRNA1. The expression of v-snoRNA124pp was verified by northern blot analysis employing a specific LNA oligonucleotide antisense probe (Figure 7). Thereby, the hybridization signal was especially apparent in 293/EBV cells induced by BZLF1, which results in a 30-fold up-regulation of v-snoRNA1 expression; the hybridization signal was absent, however, in non-induced wild type cells. This could be explained by lower v-snoRNA1 expression levels in non-induced 293/EBV cells, compared to BZLF1-induced cells (Figure 7), resulting in reduced processing of v-snoRNA124pp below the northern blot detection limit. Alternatively, this finding could result from preferential processing of v-snoRNA1 into v-snoRNA124pp during lytic replication.
Subsequently, by a 5′-RACE approach we also investigated a potential target for snoRNA124pp. Since the RNA species maps in antisense orientation to the 3′-UTR of the BALF5 mRNA, which encodes the viral DNA polymerase, BALF5 mRNA might represent a likely target site. As has been shown previously, the 3′-UTR of the BALF5 mRNA encodes in antisense orientation, in addition to v-snoRNA124pp, a bona fide EBV miRNA, designated as mir-BART2. Thereby, it has been reported that mir-BART2 down-regulates the mRNA levels by cleavage within the BALF5 3′-UTR [57]. According to the proposed model, mir-BART2 thereby inhibits the transition from latent to lytic viral replication. By 5′-RACE analysis, we provide evidence that v-snoRNA124pp might also target BALF5 mRNA for cleavage and subsequent degradation. In contrast to mir-BART2, however, expression of v-snoRNA124pp was only apparent upon induction of the viral lytic cycle by BZLF1 (Figure 7). Future experiments will focus on the function of v-snoRNA1 and v-snoRNA124pp especially in respect to its function in the latent and lytic cycles of EBV infection.
The cell lines BL2, BL2 B95.8, BL41, BL41 B95.8, CBL B95.8, LCL B95.8, Raji, Rael, HEK293 and LCL8664 were cultured in RPMI 1640 supplemented with 10% FCS, L-glutamine (2 mM ml−1) and antibiotics (100 U penicillin ml−1 and 100 µg streptomycin ml−1). BL2 and BL41 are EBV-negative Burkitt's Lymphoma cell lines, BL2 B95.8 and BL41 B95.8 are cell lines infected with EBV strain B95.8, Raji and Rael are EBV-positive Burkitt's lymphoma cell lines [62],[63]. CBL B95.8 and LCL B95.8 cells were obtained after in vitro transformation of cord blood lymphocytes (CBL) or primary human blood lymphocytes (LCLs) with the B95.8 strain of EBV. The EBV deletion strain B95.8, used in this study, lacks a 12 kb large portion of the genome [64]. LCL8664 is a rhesus LCV (cercopithicine herpesvirus 15)-infected B-cell line derived from a retro-orbital B-cell lymphoma in a rhesus monkey [65]. 293/EBV-wt contains the wild type EBV B95.8 genome in a recombinant form. Two LCL/EBV-wt and LCL/ΔBZLF1 pairs were established by immortalization of B cells from two different donors with either wild type or BZLF1-negative recombinant viruses [40].
Total RNA from EBV-infected and non-infected cells was isolated by using the Tri Reagent method according to the manufacturers protocol. Northern blot analysis was performed as described in Mrazek et. al [32] applying a mix of five oligonucleotides (F1: CCTCTCATCAGAATCTCAACC, F2: TCTCAACCGATTTCGTCAGC, F3: CGTCAGCCGCTTCAGACAG, F4: GACAGCCGCGGTTGTCATC, F5: GGTTGTCATCATCATCGGGAA) covering the whole v-snoRNA1 sequence. For the detection of the homologous rhesus lymphocryptovirus a rLCV-specific v-snoRNA1 oligonucleotide (5′-AATCTCAACCAATTTCCTCAGC-3′) was used. Detection of v-snoRNA124pp by an LNA oligonucleotide (5′-CATCAGAATCTCAACCGATTTCGT-3′, Exiqon) was performed according to the standard protocol, except 60 µg RNA was loaded and membrane was washed under stringent conditions. 5.8S rRNA antisense oligonucleotide 5′-TCCTGCAATTCACATTAATTCTCGCAGCTAGC-3′ was used as negative control in immunoprecipitations. Ethidium bromide-stained 5S rRNA were used as loading control for normalization after polyacrylamid gel electrophoresis. Northern blot signals were either put onto Kodak MS-1 film, using an intensifier screen or analyzed with a Molecular Dynamics Storm PhosphorImager (Image quant software version 5.0).
For the detection of the viral and U3 snoRNA the following amino-modified DNA oligonucleotides were used:AT*CTCAACCGATT*TCGTCAGCCGCT*TCAGACAGCCGCGGT*TGTCATCAT*CAT for v-snoRNA1 and GT*TCTCTCCCTCT*CACTCCCCAAT*ACGGAGAGAAGAACGAT*CATCAATGGCT*G for U3 (the amino-modified T nucleotides are marked with asterisks). The probes were labeled with CY3 (v-snoRNA1; Amersham Biosciences) or Oregon Green 488 (U3; Molecular Probes) according to the manufacturers protocol.
BL2 and BL2-B95.8 cells were washed in 1× PBS (PBS: 100 mM Na2HPO4, 20 mM KH2PO4, 137 mM NaCl, 27 mM KCl, pH 7.4) and diluted in 1× PBS to an appropriate concentration. The cell suspension was dropped onto glass slides and hybridized according to [21],. The slides were washed 3 times for 20 min after hybridization and mounted with 15 µl Mowiol containing 0.1 µg/ml DAPI. Slides were analyzed by confocal fluorescence microscopy (LSM 510 META, Carl Zeiss GmbH) using Zeiss LSM Software, version 3.2.
For the preparation of snoRNP extract, BL2-B95.8 cells were washed in 1× PBS, lysed in 5-fold amount of 1× RNP lysis buffer (25 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.2 mM EDTA pH 8.0, 0.2% TritonX-100), sonicated and incubated for 1 h on ice. After centrifugation steps at 18000×g, 4°C, 10 min and 30000×g, 4°C, 30 min the snoRNP extract was used for co-immunoprecipitation.
250 µl Protein A/G PLUS-Agarose (Santa Cruz Biotechnology Inc.) was washed three times in 1× PBS and resuspended with 1× RNP lysis buffer to receive a final volume of 125 µl. 50 µl of the suspension was added to the total cell lysate containing 500 µg protein extract for each approach and precleared for 1 h at 6°C during rotation. The precleared supernatant was equally distributed and incubated with specific antibodies for fibrillarin (ab5821; Abcam), NOP56, NOP58, and IgG (Santa Cruz Biotechnology Inc.) for 1 h at 6°C. After addition of 12 µl of washed beads to each approach and rotation for 4 h at 6°C, samples were centrifuged at 800×g at 4°C for 5 min. The supernatant was used as a control for unbound RNA and the remaining beads were washed four times with 1 ml 1× RNP lysis buffer and the co-immunoprecipitated RNA was eluted with 200 µl IP elution buffer (100 mM Tris-HCl pH 7.5, 150 mM NaCl, 12.5 mM EDTA pH 8.0, 20% SDS) after heating for 5 min at 95°C. The supernatant was used to perform phenol-chloroform-isoamylalcohol (Fluka) extraction, RNA was ethanol-precipitated over night and analyzed by northern blot analysis.
The wild type EBV recombinant plasmid (p2089) is cloned onto the prokaryotic F factor origin of replication and carries the green fluorescent protein (gfp), the chloramphenicol (cam) resistance gene and the hygromycin (hyg) resistance gene [67]. The EBV snoRNA mutant was constructed by replacing the C box sequence motif (B95.8 coordinates 153331–153341) with the kanamycin (kan) resistance gene using homologous recombination [68]. Composite primers were used whose internal parts (underlined) are specific for the kan resistance gene, and whose external parts (40 bp) are specific for the snoRNA gene (5′-ACGCTCCCCTGGGGGCTTCATGATCCCACCGCCTTTCfCCGCGCCAAGCTTCAAAAGCGCTC-3′; 5′-CTCAACCGATTTCGTCAGCCGCTTCAGACAGCCGCGGTTGGAAGTTCCTATTCCGAAGTTCC-3′).
These primers allowed PCR-mediated amplification of the kan resistance gene through their internal sequences and then homologous recombination of the amplified PCR product with the EBV wild type genome via their external sequences. PCR amplification products were incubated with the restriction enzyme DpnI to remove traces of the parental plasmid and introduced by electroporation (1000 V, 25 µF, 100 Ω) into E. coli DH10B cells carrying the recombinant virus p2089 and the temperature sensitive pKD46 helper plasmid encoding the phage lambda red recombinase to foster homologous recombination. Cells were grown in LB with cam (15 µg/ml) at 37°C for an hour and then plated onto LB agar plates containing cam (15 µg/ml) and kan (10 µg/ml). Incubation at 42°C induced the loss of the helper plasmid. DNA of positive clones was purified and analyzed with HindIII restriction enzyme to confirm correct recombination. The kan resistance gene was excised using the Flp recombinase cloned onto the temperature-sensitive plasmid pCP20 [69] which also carries the amp resistance gene. The bacterial clones that resulted from selection on cam/amp plates were further grown on cam plates at 42°C to induce the loss of the pCP20 plasmid. Resistant clones were then submitted to restriction analysis to confirm the expected restriction pattern. Sequencing further confirmed successful recombination and the intactness of the flanking regions.
HEK293 cells were transfected with the properly recombined mutant viral DNA (clone B 253) using Lipofectamine (Invitrogen) as described [70]. Selection of stable 293 cell clones carrying the EBV recombinant plasmid was performed by addition of hygromycin to the culture medium (100 µg/ml). Cell clones surviving selection were first assessed for GFP fluorescence and the positive clones were further expanded. Fifteen clones were assessed for their ability to support lytic replication by qPCR. Ten of those were found to produce virus at high levels, one of which was selected for further analysis. The cell clone used in this study is referred to as 293/Δv-snoRNA1. Viral episomes from this clone were transferred back in E.coli and submitted to restriction analysis and sequencing.
Circular plasmid DNA from 293/ΔsnoRNA and was extracted using a denaturation-renaturation method as described previously [71]. E.coli strain DH10B was transformed with the viral recombinant DNA by electroporation as described before [68] and clones were selected on LB plates containing cam (15 µg/ml). Single bacterial colonies were expanded and DNA plasmid preparation submitted to digestion with restriction enzyme HindIII.
Producer cell clones 293/EBV-wt (carrying p2089) and 293/Δv-snoRNA1 were transfected with a BZLF1 (Accession number NC_007605.1) expression plasmid (0.5 µg/well) to induce lytic cycle [72] using lipid micelles (Metafectene, Biontex) according to manufactures instructions. Virus supernatants were harvested four days post transfection, filtered through a 0.8 µm filter and stored at −80°C. Viral titers were determined by infecting 104 Raji cells with increasing dilutions of EBV-wt or Δv-snoRNA1 supernatants. Three days after infection, gfp-positive Raji cells were counted using a fluorescent microscope (Leica). For immortalization assays, primary B cells were mixed with infectious supernatants at various multiplicities of infections (MOI) and seeded into U-bottom 96-well plates coated with gamma-irradiated WI38 feeder cells [73] at a concentration of 102 cells per well. Wells containing outgrowing LCL clones were counted.
Detection of viral DNA and calculation of viral titers was carried out by quantitative real-time PCR (qPCR) using BALF5-specific primers and probe as described [74]. The DNA content was calculated using a serial dilution of Namalwa DNA, a human Burkitt's lymphoma cell line that contains two EBV genome copies per cell, as a standard curve.
We predicted putative rRNA target sites for the snoRNAs in this study as follows. We first downloaded from Genbank the sequences of the human 18S (Accession NR_003286) and 28S (Accession NR_003287) rRNAs. The sequences of the antisense D-box (TGACGAAATCGGTTGAGATT) and D′-box (TGACAACCGCGGCTGT) were used to search for subsequences with good complementarity to the rRNAs with the program described in Mandin P et al. [41]. As the study of Cavaille & Bachellerie [25] indicated that snoRNA-rRNA interactions involve regions of at least 7 nucleotides complementarity that are located at most 3 nucleotides from the end of the snoRNA antisense box, and that bulges and loops of more than 1 nucleotide are disfavored, we implemented these constraints in our programs. That is, we first used relatively large penalties for the introduction and extension of bulges and loops (a score penalty of 8), and we restricted the maximum size of loops and bulges to 1 nucleotide. The energy parameters of nucleotide-nucleotide interactions were kept with their default values coded in the program. We then extracted only hybrids that contained at least 7 nucleotide-nucleotide pairs, that ended within 3 nucleotides of the end of the antisense box, and that did not contain more than one bulge or loop.
2-OH ribose methylation of rRNA was assayed as follows. Oligonucleotides (0.6 pM) were 5′-end-labeled with 32P-γ-ATP and heat-denaturated after addition of 3 µg of total RNA (LCL B95.8) for 2 min at 96°C. Primer annealing was performed in presence of 30 mM KCl and 25 mM Tris-HCl pH 8.4 for 30 min at 42°C. Reverse transcription was carried out for 45 min at 42°C in buffer containing 100 mM Tris/HCl pH 8.4, 10 mM MgCl2, 15 mM KCl, 10 mM DTT, 0.5/0.02/0.005 mM dNTPs and 0.4 U AMV reverse transcriptase. Additionally, a final concentration of 0.0625 mM dideoxynucleotides was added to the sequencing reactions. The reactions were stopped by addition of twice volume of 4 M NH4Ac and 20 mM EDTA, cDNA products were precipitated, resolved on a 10% denaturating polyacrylamide gel and visualized by autoradiography.
Total RNA of 293/Δv-snoRNA1 and 293/EBV-wt induced by BZLF1 and was adaptor-ligated and reverse transcribed using a gene-specific primer (5′- TTCGCCCTTGCGTGTCCATTGT-3′) according to the FirstChoice RLM-RACE Kit (Ambion). cDNA was PCR-amplified with the non-specific 5′ RACE outer primer and the same reverse primer and further amplified by nested PCR using the 5′ RACE inner primer and a second gene-specific reverse primer (5′- GCAAGGAGCGATTTGGAGAAAATAAAC-3′). PCR DNA was gel purified, cloned (pGEM-T Easy Vector System I, Promega) and subjected to Sanger sequencing employing the ABI Prism 3100 capillary sequencer (Perkin Elmer).
v-snoRNA1: FN376861; BZLF1: NC_007605.1; 18S rRNA: NR_003286; 28S rRNA: NR_003287; Epstein-Barr-Virus genome, strain AG876: AJ507799; Rhesus lymphocryptovirus genome: NC_006146
|
10.1371/journal.pntd.0002197 | IRES-driven Expression of the Capsid Protein of the Venezuelan Equine Encephalitis Virus TC-83 Vaccine Strain Increases Its Attenuation and Safety | The live-attenuated TC-83 strain is the only licensed veterinary vaccine available to protect equids against Venezuelan equine encephalitis virus (VEEV) and to protect humans indirectly by preventing equine amplification. However, TC-83 is reactogenic due to its reliance on only two attenuating point mutations and has infected mosquitoes following equine vaccination. To increase its stability and safety, a recombinant TC-83 was previously engineered by placing the expression of the viral structural proteins under the control of the Internal Ribosome Entry Site (IRES) of encephalomyocarditis virus (EMCV), which drives translation inefficiently in insect cells. However, this vaccine candidate was poorly immunogenic. Here we describe a second generation of the recombinant TC-83 in which the subgenomic promoter is maintained and only the capsid protein gene is translated from the IRES. This VEEV/IRES/C vaccine candidate did not infect mosquitoes, was stable in its attenuation phenotype after serial murine passages, and was more attenuated in newborn mice but still as protective as TC-83 against VEEV challenge. Thus, by using the IRES to modulate TC-83 capsid protein expression, we generated a vaccine candidate that combines efficient immunogenicity and efficacy with lower virulence and a reduced potential for spread in nature.
| Venezuelan equine encephalitis virus (VEEV) is transmitted by mosquitoes and widely distributed in Central and South America, causing regular outbreaks in horses and humans. Often misdiagnosed as dengue, VEEV infection in humans can lead to lifelong neurological sequelae and is fatal in up to >80% of equine cases, representing a significant socio-economic burden and constant public health threats for developing countries of Latin America. The only available vaccine, the live-attenuated TC-83 strain, is restricted to veterinary use due to its high reactogenicity in humans and risk for reversion to virulence, which could initiate an epidemic. By using an attenuation approach that allows the modulation of the virus capsid protein expression, we generated a new version of TC-83 that is more attenuated but still induces a protective immune response in mice. Additionally, this new vaccine cannot infect mosquitoes, which prevents the risk of spreading in nature. The attenuation approach we describe can be applied to a lot of other alphaviruses to develop vaccines against diseases regularly emerging and threatening developing countries.
| Arboviruses (Arthropod-Borne viruses) comprise a group of viruses transmitted among vertebrates by hematophagous arthropods. They include members of a wide range of viral families, such as Rhabdoviridae, Bunyaviridae, Flaviviridae and Togaviridae, with a worldwide distribution. The presence of an arbovirus in a particular area depends on the availability of transmission-competent arthropods, as well as amplifying vertebrates (in particular birds or small mammals) susceptible to virus infection and producing sufficient viremia to maintain transmission cycles. Although mostly restricted to sylvatic, enzootic cycles between reservoir vertebrate hosts (mainly rodents and birds) and arthropod vectors, environmental alterations and continuous changes in human and animal demographics have created factors favorable to arboviral emergence from limited cycles, threatening domestic animals and humans [1]. Thus, arboviral epizootics in animals and/or epidemics in human populations are regularly reported. They have significant socio-economic impacts, and contribute to the maintenance of continuous public-health threats around the world.
Venezuelan equine encephalitis virus (VEEV), a positive-strand RNA arbovirus and member of the Alphavirus genus in the Togaviridae family, is one of the most pathogenic mosquito-borne viruses circulating in South and Central America [2]. In the VEE antigenic complex of alphaviruses that includes 6 subtypes (I to VI), all VEEV strains are found in antigenic subtype I. In this subtype, VEEV strains occur in 4 different antigenic varieties: IAB and IC strains are called “epizootic” or “epidemic” because they efficiently infect equids and produce sufficient viremia to allow oral infection of mosquitoes, thus facilitating high levels of transmission and amplification. These highly efficient equine-mosquito amplification cycles can generate widespread circulation in agricultural areas, usually resulting in spillovers into humans. Varieties ID and IE include enzootic strains, which are typically avirulent for equids and unable to induce high levels of viremia, although some recent IE strains from outbreaks in Mexico are neurovirulent [3], [4]. However, subtypes ID and IE can cause large numbers of human infections via spillover from their sylvatic cycles [2]. Phylogenetic studies indicate that IAB and IC strains derived from subtype ID progenitors [5]. Experimental studies have linked the emergence of VEEV IAB and IC strains to mutations in the E2 glycoprotein, allowing the virus to replicate more efficiently in equids, resulting in greater exposure and/or increased susceptibility to epizootic vectors [6], [7].
Human VEEV infection typically generates moderate to highly incapacitating flu-like symptoms, and is usually misdiagnosed as dengue, resulting in its neglect. Progression to severe encephalitis is observed in about 14% of cases and ultimately death occurs in less than 1%. Although the incidence of fatal disease is relatively low, the neurovirulence of some VEEV strains can lead to lifelong sequelae [8]. In horses, up to >80% of cases can be fatal [9]. Since the first documented outbreaks in the 1930s, several major epidemics have been reported in many countries in Latin America, including Venezuela, Colombia, Peru, Ecuador, Costa Rica, Nicaragua, Honduras, El Salvador, Guatemala, Panama, Mexico, involving hundreds-of-thousands of human and equine cases [2]. VEEV is also highly infectious by aerosol, and had been developed as a biological weapon [10]. Therefore, it represents a major target for which a vaccine is urgently needed to prevent amplification in equids and to protect against human disease.
Like other alphaviruses, VEEV has a positive-sense, single-strand RNA genome of ca. 11.5 kb [11]. The nonstructural protein genes are translated from genomic RNA via a cap-dependent mechanism but the structural genes are translated from a subgenomic message transcribed from negative strand replicative intermediates. The subgenomic RNA is produced in molar excess compared to the genomic RNA, allowing the production of large amounts of the capsid and envelope glycoproteins needed for virion formation [12].
To date, no VEEV vaccine has been licensed for use in humans. VEEV strain TC-83, a live-attenuated, licensed veterinary vaccine, is used to immunize horses in regions endemic for IAB and IC strains, as well as laboratory workers and military personnel. TC-83 was generated by 83 serial-passages of the Trinidad donkey (TrD) IAB strain in guinea pig heart cells [13], and its attenuation relies on only 2 point mutations 14,15. Because RNA viruses exhibit high mutation rates [16], [17], there is a concern that TC-83 may revert to a wild-type, virulent phenotype and cause potentially fatal disease in vaccinees. TC-83 can also infect mosquitoes, as occurred in 1971 during an equine vaccination campaign to prevent spread of an epidemic [18], and thus could initiate an outbreak. In addition, only 80% of human TC-83 vaccinees seroconvert, and reactogenicity is observed in nearly 40% of immunized individuals [19], [20], [21].
In an effort to improve TC-83 attenuation and safety, particularly regarding its potential to be transmitted by mosquitoes from vaccinated horses, a recombinant TC-83 virus, VEEV/mutSG/IRES, was engineered to eliminate the subgenomic promoter and place the expression of the viral structural proteins under the control of the Internal Ribosome Entry Site (IRES) of encephalomyocarditis virus (EMCV) [22], which functions inefficiently in arthropod cells [23], [24]. In this vaccine candidate, the viral subgenomic promoter was inactivated by the introduction of 13 synonymous mutations, and the EMCV IRES was placed upstream of the structural polyprotein gene open reading frame. The resulting recombinant virus, VEEV/mutSG/IRES/1, exhibited an attenuated phenotype in cell culture and in vivo in the mouse model, and was unable to replicate in mosquito cells or in live mosquitoes [22]. However, no neutralizing antibody response was detected in vaccinated NIH Swiss mice, and only partial protection against virulent VEEV challenge was achieved.
To improve the immunogenicity of VEEV/mutSG/IRES/1, we developed a new IRES-based variant of TC-83 in which only the capsid protein is placed under IRES translational control, leaving an intact subgenomic promoter driving the expression of the major antigens, the glycoproteins E1 and E2. This new vaccine candidate showed a similar, highly attenuated profile like the original VEE/mutSG/IRES/1 strain and was also unable to replicate in mosquitoes. However, this second generation of IRES-based vaccine candidate was more immunogenic and induced complete protection against lethal VEEV challenge.
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 protocol was approved by the Institutional Animal Care and Use Committees of the University of Texas Medical Branch or the University of Wisconsin.
Vero (African green monkey kidney) and baby hamster kidney (BHK-21) cells were obtained from the American Type Cell Culture (ATCC, Manassas, VA) and maintained at 37°C in Dulbecco's minimal essential medium (DMEM) supplemented with 5% fetal bovine serum (FBS), penicillin and streptomycin (PS). C6/36 Aedes albopictus cells (ATCC) were propagated at 29°C in DMEM containing 10% FBS, PS and supplemented with 1% tryptose phosphate broth.
Plasmid pVEEV/mutSG/IRES/1 was described previously [22]. It encodes the complete genome of VEEV strain TC-83, in which the subgenomic promoter is inactivated by 13 synonymous mutations and the structural protein genes are placed under the translational control of the EMCV IRES. The nsP2-coding sequence contains an adaptive mutation that increases replication efficiency. Our new pVEEV/IRES/C strain (Fig. 1) encodes the VEEV TC-83 genome with an active SG promoter. In this plasmid, the capsid gene, under control of the EMCV IRES, was positioned downstream of the E1 gene. This plasmid was constructed using standard PCR-based techniques and details are available from the authors.
After sequencing, a large-scale preparation of pVEEV/IRES/C was obtained using standard methods and purification on CsCl gradients. The plasmid was then linearized with MluI restrictase and subjected to RNA transcription using SP6 RNA polymerase (Ambion, Austin, TX) in the presence of a cap analogue. Each of these steps was analyzed by agarose gel electrophoresis. To rescue virus, in vitro-transcribed RNA was transfected into BHK-21 cells by electroporation as previously described [25], [26]. Briefly, one T150 flask of BHK-21 cells was trypsinized, the cells were washed 3 times in PBS, and finally resuspended in 400 µl. One µg of transcribed viral RNA was added to the cells and the mixture was subjected to five pulses at 680v for 99 µsec, at 200 msec intervals. Electroporated cells were resuspended in DMEM containing 10% FBS, seeded into one T75 flask, and incubated at 37°C. When cytopathic effects (CPE) were observed (18-to-24 h post-electroporation), supernatants containing infectious virus were harvested and titrated on Vero cells by plaque assay [27].
The procedure described above was used to electroporate 4 µg of transcribed RNA into BHK cells. One-fifth of electroporated cells were seeded into 35-mm dishes, and supernatants were harvested at designated time-points post-electroporation and replaced with fresh medium. Alternatively, Vero cells in T25 flasks were infected with a multiplicity of infection (MOI) of 0.1 PFU/cell. After a one-hour incubation at 37°C, cells were washed with PBS and covered with 4 ml of DMEM with 2% FBS. Cell supernatants were collected at different time-points post-infection and replaced with fresh medium. Viruses in harvested supernatants were titrated on Vero cells by plaque assay [27].
One-fifth of electroporated BHK cells were seeded into 35-mm dishes and incubated for 4.5 h at 37°C before the supernatant was replaced with 0.8 ml of DMEM supplemented with 1 µg/ml of Actinomycin D and 20 µCi/ml of [3H]-uridine. After 4 h of incubation, medium was removed and cells were harvested in 0.8 ml of Trizol (Invitrogen, Carlsbad, CA) for RNA extraction according to manufacturer's protocol. Purified RNA was analyzed by agarose gel electrophoresis after denaturation with glyoxal in dimethyl sulfoxide, as previously described [28]. The gel was then impregnated overnight with 2,5-dipheniloxazol (PPO) and dried. Kodak X-OMAT AR film (Sigma-Aldrich, Saint Louis MO) was exposed to dried gel at −80 °C and autoradiographed.
The VEEV/IRES/C virus was passaged 5 times on Vero cells to determine its genetic stability in vitro. Two parallel replicate series were performed at an MOI of 0.1 PFU/cell, and infectious supernatants were harvested 48 h post-infection, titrated by plaque assay and used for the next passage. For viral sequence analysis, RNA was extracted from passage 5 viruses using QIAamp Viral RNA mini kit (Qiagen, Valencia CA) and subjected to 2-step RT-PCR with Superscript III RT System (Invitrogen) and the Phusion DNA polymerase kit (New England BioLabs, Ipswich MA). The resultant 2000 bp amplicons were sequenced using an ABI 3500 Genetic Analyzer (Applied Biosystems, Carlsbad, CA) and alignments and analysis were performed using Sequencher 4.9 software (Ann Arbor, MI).
To assess mosquito cell infectivity, 5 blind serial passages were performed on C6/36 (Aedes albopictus) cells seeded in 35-mm dishes and infected at a starting MOI of 1 Vero cell PFU/mosquito cell. After 1 h incubation, inocula were removed, cells were washed 4 times with PBS and 2 ml of DMEM were added. Supernatants were collected 48 h post-infection and 0.4 ml were used to infect C6/36 cells for the next passage. After 5 passages, plaque assays were performed on supernatants from each passage to determine viral titers, as well as RNA extraction and RT-PCR to quantify viral genomes.
To evaluate replication competence in vivo, we used Aedes aegypti mosquitoes from a colony established with individuals collected in Galveston, TX. Five-to-six days post-emergence, mosquitoes were allowed to feed for one hour on an infectious artificial blood meal containing 33% (v/v) defibrinated sheep erythrocytes (Colorado Serum Company, Denver, Co), 33% (v/v) heat-inactivated fetal bovine serum (FBS) (Omega Scientific, Inc., Tarzana, CA) and 33% (v/v) of each individual virus in cell culture medium. The titer of each blood meal was of approximately 5×108 PFU/ml, the highest achievable with Vero cell-derived virus stocks. After feeding, mosquitoes were cold-anesthetized, and engorged individuals were incubated at 27°C with a relative humidity of 70–75% and 10% sucrose ad libitum for 10 days. Alternatively, Ae. aegypti mosquitoes were injected intrathoracically with ca. 1 µl of a 108 PFU/ml virus stock and incubated as described above.
After 10 days of incubation, mosquitoes were placed individually into 2 ml tubes containing 350 µl of MEM 10% FBS supplemented with 5 µg/ml of Fungizone (Invitrogen) and triturated for 4 min in a Tissue Lyser II (Qiagen, Venlo, Netherlands). Homogenized mosquito samples were centrifuged at 10,000×g for 5 min and 50 µl of supernatants were applied to Vero monolayers in 24-well plates. After incubation for 1 h at 37°C, cells were covered with 1 ml DMEM with 2% FBS and observed for 5 days to detect CPE as signs of infection.
To study virulence, six-day-old CD-1 mice (Charles Rivers, Wilmington, MA) were inoculated intracranially (IC) with 106 PFU of virus in a volume of 20 microliters (µl), or subcutaneously (SC) with 5×104 PFU in a volume of 50 µl. Animals were observed for 2 weeks with daily weight and survival recording. Mice that survived the SC injection were used for immunogenicity and protection studies. Six weeks following initial inoculation with recombinant viruses, blood was collected from the retro-orbital sinus for antibody screening by PRNT as previously described [27], using VEEV TC-83 virus for neutralization. Animals were challenged 3 weeks later with 104 PFU SC of virulent VEEV subtype IC strain 3908 and monitored twice daily for signs of illness, survival and weight loss.
In another experiment, 8-week-old CD-1 mice were vaccinated SC with VEEV strain TC-83 or the IRES-based vaccine candidates at a dose of 105 PFU/mouse, or PBS for unvaccinated controls. Six weeks post-vaccination, animals were challenged SC with 104 PFU of VEEV strain 3908, with daily monitoring for signs of illness, survival and weight loss. Blood samples were collected for 4 days post-vaccination and post-challenge for viremia detection, as well as 5 weeks post-vaccination for antibody measurement by PRNT.
To assess genetic and phenotypic stability of the new IRES-based vaccine candidate in vivo, VEEV/IRES/C was subjected to 10 serial, IC passages in six-day-old CD1 mice at a dose of ca. 5×104 PFU per animal. Two parallel passage series were performed (A and B). Animals were euthanized 48 h post-inoculation, and their brain harvested and triturated to determine viral titer by plaque assay. Homogenized brain samples containing the highest titers were used as the inoculum for the next passage in each series. Virulence of the mouse passage 10 (mp10) viruses was compared to the parental strain by inoculating 6-day-old CD1 mice SC with 5×104 PFU, as described above. Stability of the genomic sequences was assessed by RT-PCR on RNA extracted from mp10 viruses and sequencing, as described above. VEEV strains TC-83 and TC-83 mp10A and mp10B, previously described by Kenney et al. [29], were included as controls.
All statistical analyses were performed using Prism software (GraphPad version 4.0c, La Jolla, CA). Logrank tests were used to determine significance in survival differences between individual groups. One-way repeated measures ANOVA analyses were performed on the weights of mice following vaccination/challenge. Significance was determined at P<0.05 for all tests.
This study was designed to develop a VEEV vaccine candidate that would replicate at high titers in vertebrate cells but not in mosquitoes, and that would be immunogenic and protective against lethal VEEV challenge. To evaluate the performances of the new VEEV/IRES/C vaccine candidate compared to the previous IRES-based construct, VEEV/mutSG/IRES/1, we used the latter as a control in this study [22].
After SP6-driven in vitro RNA synthesis and electroporation into BHK-21 cells, production of viral RNAs (genomic and subgenomic) from the newly designed IRES-based vaccine candidate was confirmed in vitro (Fig. 2A). As previously shown, VEEV/mutSG/IRES/1 was incapable of producing subgenomic RNA due to the 13 point mutations introduced into the subgenomic promoter [22]. In VEEV/IRES/C, the subgenomic promoter was left intact, allowing efficient production of subgenomic RNA, which migrated more slowly than its TC-83 counterpart due to the introduction of the additional IRES sequence. However, it appeared that VEEV/IRES/C genomic RNA was produced at slightly lower levels compared to TC-83 and VEEV/mutSG/IRES/1. To assess the potential effect on viral replication, the production of infectious virus was monitored (Fig. 2B). As expected, both IRES constructs produced significantly less infectious virus than unmodified TC-83, with a difference of approximately 1.5 log10 at the peak of production; TC-83 titer reached 4.8×109 PFU/ml at 24 h post-electroporation. Despite the difference in genomic RNA production, both IRES constructs reached their peak titer 24 h post-electroporation, with similar titers of 9×107 PFU/ml for VEEV/mutSG/IRES/1 and 8×107 PFU/ml for VEEV/IRES/C. Thus, the lower amount of genomic RNA produced by VEEV/IRES/C did not severely impair viral replication compared to VEEV/mutSG/IRES/1. Moreover, VEEV/IRES/C viral production was consistently detected ca. 2 hr earlier than that of VEEV/mutSG/IRES/1, and the latter exhibited significantly lower titers of production during the first 24 h post-electroporation.
Viral replication following infection of Vero cells, an approved vaccine substrate, was also measured (Fig. 3A). Replication profiles were very similar to those obtained on BHK cells after electroporation, with an advantage of approximately 1 log for TC-83 compared to the IRES-modified strains, and a peak TC-83 titer of 6.2×109 PFU/ml at 24 h post-infection, compared to 5.2×108 and 5.3×108 PFU/ml for VEEV/mutSG/IRES/1 and VEEV/IRES/C, respectively. Additionally, plaques produced under 0.4% agarose on Vero cells were visible as early as 24 h post-infection for TC-83, whereas 48 h of incubation was necessary for IRES-based viruses to produce visible plaques. This slower replication level was also correlated to the size of the plaques produced by IRES-based viruses on Vero cells (Fig. 3B). At 48 h post-infection, TC-83 produced 3–6 mm plaques whereas VEEV/mutSG/IRES/1 and VEEV/IRES/C plaques were 2–3 mm and 1.5–3 mm, respectively.
VEEV/IRES/C was subjected to 5 serial passages in Vero cells or 10 serial passages in mouse brains. No discernible change was observed in plaque morphology after 5 serial passages in vitro or 10 passages in vivo (data not shown). Genetic stability was confirmed by full-genome sequencing of passaged viruses; no mutations were found in consensus sequences of Vero- or mouse-passaged viruses, aside from the deletion of one adenosine in a poly-A tract within the IRES itself, which appeared between passage 3 and 4 on Vero cells, and before passage 5 in mouse brains. No changes were detected in virulence for the mp10 VEEV/IRES/C compared to the parental strain (P = 0.95 for series A and P = 0.75 for series B) after SC injection of 6-day-old mice (Fig. 4), whereas a significant increase in virulence was observed for the mp10 TC-83 viruses compared to parental TC-83, as previously described [29].
To confirm its predicted inability to replicate in mosquito cells, VEEV/IRES/C was blind-passaged 5 times in C6/36 cells (along with TC-83 as a control). For each passage, supernatants were subjected to RT-PCR for viral RNA detection, and plaque assay for infectious virus. Virus and viral RNA were only detected in passages 1 and 2, presumably due to residual virions that were incompletely washed from the cells after the original inoculation (Fig. 5A and 5B). Indeed, the VEEV/IRES/C viral titer declined from 105 PFU/ml in passage 1 to 10 PFU/ml after passage 2, along with weakening of the RT-PCR signal. No infectious virus or viral RNA was detected after 2 passages. Meanwhile, TC-83 virus consistently produced ca. 1010 PFU/ml, confirmed by the detection of viral RNA in supernatants for all 5 passages.
The predicted VEEV/IRES/C inefficiency of replication was also confirmed in live mosquitoes, and compared to TC-83. Ae. aegypti were allowed to feed on infectious blood meals containing 3×108 PFU/ml of TC-83 or VEEV/IRES/C, and incubated for 10 days before being triturated and tested for the presence of infectious virus by detection of CPE on Vero cells. Fifty percent (24/48) homogenates from mosquitoes exposed to TC-83 produced detectable CPE, whereas none of the VEEV/IRES/C-exposed mosquitoes produced CPE after 10 days of incubation (Table 1). Because only 50% of the mosquitoes were found susceptible to TC-83, a second experiment was performed using a more permissive route of infection, intrathoracic injection. Using the highest dose achievable of 105 PFU per mosquito (ca. 1 µl of a 108 PFU/ml viral stock), 100% of mosquitoes injected with TC-83 became infected, whereas only 14/55 mosquitoes inoculated with VEEV/IRES/C produced CPE after incubation. Plaque assays performed on these homogenates revealed a mean titer of only 100 PFU/mosquito for the VEEV/IRES/C-infected mosquitoes, a titer incompatible with VEEV transmission by mosquitoes [30], [31], [32], [33], [34], [35], [36], [37], [38], [39]. In contrast, an average of 9×106 PFU/mosquito were recorded in the TC-83-infected group. To determine if the presence of VEEV/IRES/C in mosquitoes 10 days after inoculation could have represented residual inoculum without replication, replicates of a viral suspension containing 106 PFU/ml (VEEV TC-83 or VEEV/IRES/C) were incubated at 29°C for 10 days; these samples still contained on average 103 PFU/ml after these 10 days, indicating that the CPE-positive mosquitoes injected with VEEV/IRES/C most likely contained only residual virus from the inoculum rather than supported active viral replication.
Because TC-83 does not typically induce mortality in adult mice, an infant mouse model was used to compare virulence of the vaccine constructs. Cohorts of 6-day-old CD-1 mice were inoculated subcutaneously with 5×104 PFU of TC-83, VEEV/mutSG/IRES/1 or VEEV/IRES/C and monitored for signs of illness, weight and survival. Another cohort of mice was inoculated with PBS as a negative control. As shown in Fig. 6A, while TC-83 induced 100% mortality by day 9 post-inoculation, the IRES-based viruses were both markedly attenuated (Logrank test, P<0.0001), with no significant difference observed between VEEV/mutSG/IRES/1 (76% survival) and VEEV/IRES/C (50% survival) at 14 days post-inoculation (P = 0.28). However, the mean weights recorded throughout the experiment indicated that growth of mice inoculated with VEEV/IRES/C was delayed compared to those inoculated with VEEV/mutSG/IRES/1 or PBS (P = 0.019 and P = 0.004 respectively, Fig. 6B), suggesting lesser attenuation of VEEV/IRES/C vaccine candidate compared to the previous IRES-based virus. However, the delayed growth in the VEEV/IRES/C was temporary and animals recovered in a few days, whereas no recovery was observed in the TC-83 group.
To assess neurovirulence, six-day-old mice were inoculated intracranially with 1×106 PFU of virus. Similar to that observed after subcutaneous inoculation, there was no significant difference between VEEV/mutSG/IRES/1 and VEEV/IRES/C cohorts in mortality, with 60% and 67% of the animals surviving, respectively (Logrank test, P = 0.45), whereas 100% mortality was observed at 6 days post-inoculation in the TC-83 group (P<0.0001, Fig. 7A). The mortality observed in the VEEV/mutSG/IRES/1 and VEEV/IRES/C groups was also delayed compared to TC-83. Nevertheless, animals inoculated with VEEV/mutSG/IRES/1 showed more signs of illness than animals inoculated with VEEV/IRES/C, illustrated by the observation of more delayed growth compared to the PBS group (P = 0.01, Fig. 7B) and neurological signs such as ataxia, paralysis and lethargy in most mice infected with VEEV/mutSG/IRES/1. Thus, in this model VEEV/IRES/C appeared to be less virulent than VEEV/mutSG/IRES/1.
The ability of the new VEEV/IRES/C vaccine candidate to induce neutralizing antibodies and to protect against a lethal VEEV challenge was evaluated in neonatal and adult mouse models and compared to VEEV/mutSG/IRES/1 and TC-83. Animals that survived the single SC inoculation with VEEV/mutSG/IRES/1 and VEEV/IRES/C at 6 days of age were held for 6 weeks post-infection before sera were collected and tested by PRNT. Seroconversion was detected in 6 of 7 (85%) animals vaccinated with VEEV/IRES/C and in 6 of 10 (60%) animals vaccinated with VEEV/mutSG/IRES/1, with mean PRNT80 titers of 26±8 and 57±22, respectively (Table 2). Challenge was performed on these animals 3 weeks later with virulent VEEV strain 3908, a human isolate from the last major VEE epidemic [40], at a SC dose of 104 PFU (ca. 104 LD50). All sham-vaccinated animals died between days 6 and 8, whereas 30% mortality was recorded for the animals that received VEEV/mutSG/IRES/1, and all animals vaccinated with VEEV/IRES/C survived challenge (Fig. 8A). No weight loss was observed in the VEEV/IRES/C-vaccinated cohort after challenge, whereas the VEEV/mutSG/IRES/1- and sham-vaccinated animals lost an average of 6.5% and 19.4% of pre-challenge weight by day 6 post-challenge, respectively (Fig. 8B).
In a second experiment, adult mice were vaccinated SC with a single dose of 105 PFU of each vaccine strain. No viremia was detected in VEEV/mutSG/IRES/1- and VEEV/IRES/C-vaccinated groups at days 1 and 2 post-vaccination. In the TC-83-vaccinated group, 3 out of 5 animals were viremic on days 1 and 2 with mean titers of 2×103 and 2×102 PFU/ml, respectively. No significant weight changes were detected in any of the groups post-vaccination (data not shown). Animals were bled 2 months later and neutralizing antibody titers were determined. In the TC-83 vaccinated group, 100% of the animals seroconverted and PRNT titers all exceeded the endpoint of 1280. Although the titers in the IRES-recombinants vaccinated groups were lower than those in the TC-83 group, mean PRNT80 and PRNT50 titers were 2.5 times higher in the VEEV/IRES/C group (184±184 and 424±482, respectively) compared to VEEV/mutSG/IRES/1 group (74±98 and 160±195 respectively), with 80% seroconversion in VEEV/IRES/C-vaccinated animals and 70% in the VEEV/mutSG/IRES/1 cohort (Table 3). A challenge was performed 6 weeks post-vaccination with 104 PFU of wild-type VEEV strain 3908. All sham-vaccinated animals died between days 7 and 9 post-challenge, whereas all animals vaccinated with VEEV TC-83 or VEEV/IRES/C were protected. One VEEV/mutSG/IRES/1-vaccinated animal died on day 10 post-challenge (Fig. 9). All sham-vaccinated animals had detectable viremia up to 4 days post-challenge, reaching an average of 1.3×107 PFU/ml on day 3 (Table 4). In the VEEV/IRES/C-vaccinated group, viremia was recorded in 1, 3 and 1 animals out of 10 on days 1, 2 and 3 post-challenge, respectively, with average titers of 1×102 PFU/ml on days 1 and 3, and 1×103 PFU/ml on day 2. Challenge viremia was detected in 2 out of 10 animals vaccinated with VEEV/mutSG/IRES/1 on days 1 and 3, with average titers of 1×104 PFU/ml and 1×102 PFU/ml respectively. No virus was detected after challenge in animals vaccinated with TC-83 (Table 4). No significant difference was observed in weight change among the vaccinated groups (data not shown).
Vaccines remain the best tools to control viral infectious diseases, for which there are few treatments available. Because they induce robust and often life-long protective immune responses, live-attenuated vaccines have been developed and used extensively for decades against viral diseases with remarkable successes [41]. Traditionally, these vaccines were derived empirically from wild-type virus strains by serial passages in animals or cell cultures. However, this approach often yields unpredictable results and poses safety concerns, including the risk of reversion to a wild-type phenotype, especially when the attenuation relies on a limited number of point mutations. VEEV vaccine strain TC-83 exemplifies this safety issue, as only 2 point mutations are responsible for its attenuation [14]. Probably as a consequence, TC-83 is reactogenic in many human vaccinees, which has prevented its licensure [21], [42], [43]. However, TC-83 has been studied extensively and licensed in several countries for veterinary use, for which it is sufficiently attenuated and immunogenic [42], [44]. Thus, it represents a suitable backbone to develop a safer and more attenuated VEEV vaccine.
In a previous study, a recombinant TC-83 virus was developed by placing the expression of the viral structural proteins under the vertebrate-restricted translation control of the EMCV IRES, which does not efficiently drive protein expression in mosquito cells [22]. This strategy resulted in 2 critical improvements over unmodified TC-83: 1) the IRES-recombinant TC-83 was more attenuated and thus potentially less reactogenic, and; 2) it was incapable of replication in mosquitoes, which dramatically reduces the risk of initiating a mosquito-vertebrate amplification cycle from a vaccinated and viremic equid, and the subsequent potential for reversion to virulence. However, this first generation of IRES-based TC-83 vaccine did not induce detectable neutralizing antibodies in the NIH Swiss mice model and failed to protect 100% of challenged animals. As suggested previously [45], the low level of structural protein expression observed for the IRES-recombinant TC-83 virus could explain its poor immunogenicity, as critical B cell epitopes are located in the surface glycoproteins E1 and E2 [46], [47].
To retain the benefits of the first generation of IRES-based TC-83 vaccine while increasing the expression of the glycoproteins E1 and E2, we placed the capsid gene at the 3′ end of the structural protein open reading frame and under EMCV IRES control. Expression of the surface glycoproteins E1 and E2 was left under the control of the viral subgenomic promoter in a cap-dependent manner, as in the parental TC-83. As in the first TC-83 IRES-recombinant version, the deletion of the IRES sequence would make VEEV/IRES/C non-viable because the capsid gene could not be translated from the subgenomic RNA. VEEV/IRES/C was efficiently rescued and produced high titers on Vero cells, an acceptable substrate for vaccine production, making VEEV/IRES/C a vaccine candidate feasible to produce to large scale.
By comparing the new VEEV/IRES/C to the previous IRES-based TC-83 vaccine candidate, VEEV/mutSG/IRES/1, and the parental strain TC-83, we demonstrated that placing the capsid protein under IRES control while leaving the envelope glycoproteins under the subgenomic promoter control did not increase viral yields in vitro or greatly increase virulence in the mouse model. This could simply reflect an unbalanced ratio of capsid versus glycoproteins, which would not allow highly efficient encapsidation and release of viral particles. Overall, VEEV/IRES/C exhibited a similar attenuation profile compared to VEEV/mutSG/IRES/1 and markedly greater attenuation compared to VEEV TC-83. Additional studies in adult mice and eventually in non-human primates and horses will be necessary to link the increased attenuation of this virus to a decreased reactogenicity. Further investigating the pathogenesis of VEEV/IRES/C in terms of tissue tropism will also be needed to support its further development.
In terms of environmental safety, we also demonstrated that the consensus genome sequence of VEEV/IRES/C was stable after serial passages in vitro or in vivo, which translated to phenotypic stability in vivo with no significant change in virulence. In contrast, TC-83 underwent a rapid and significant increase in virulence after mouse passages, presumably reflecting its unstable attenuation based on only 2 point mutations [14]. Kenney et al. showed similar results and the increased TC-83 virulence was associated with a mixture of mutants, suggesting that a complex quasispecies population determined the virulence phenotype [29]. VEEV/IRES/C was also incapable of replicating in mosquito cells in vitro. Although we found small amounts of residual virus in a small proportion of IT-injected mosquitoes after 10 days of incubation, the low titers suggested residual viral inoculum rather than productive viral replication. In parallel, we showed that no mosquitoes were infected with VEEV/IRES/C after exposure to a large oral dose that far exceeded the 3 log10/ml detected in humans or 3.5 log10/ml detected in horses vaccinated with TC-83 [21], [42]. Thus, the inability of VEEV/IRES/C to replicate in mosquitoes offers a major advantage, even compared to another live-attenuated VEEV vaccine candidate, strain V3625, which is able to replicate to high titers in mosquitoes [48], [49].
Immunogenicity and efficacy were assessed after vaccination of infant and adult mice. In both models, VEEV/IRES/C appeared to be more potent at inducing a neutralizing antibody response compared to VEEV/mutSG/IRES/1. Although the PRNT titers were lower for VEEV/IRES/C compared to TC-83, all animals were protected from lethal VEEV challenge, whereas VEEV/mutSG/IRES/1 failed to do so. Moreover, animals that survived challenge after VEEV/mutSG/IRES/1 vaccination showed weight loss, where no signs of disease were observed in the VEEV/IRES/C-vaccinated group either after vaccination or challenge. Volkova et al. showed that adult NIH Swiss mice vaccinated with ca. 105 PFU of VEEV/mutSG/IRES/1 virus failed to develop detectable neutralizing antibodies and only 80% of the vaccinated animals were protected against a challenge with 104 PFU of wild-type VEEV strain 3908, versus 100% protection obtained with TC-83 [22]. In similar experiments with NIH Swiss mice, VEEV/IRES/C induced neutralizing antibody response and were fully protected against lethal challenge with VEEV 3908 (data not published). These results support the greater immunogenicity of VEEV/IRES/C compared to the first IRES-based TC-83 vaccine candidate.
These promising observations need to be confirmed by more extensive exploration of the immune response induced by VEEV/IRES/C, by testing different vaccine doses, and by evaluating the duration of immunity and protection against aerosol exposure. It would also be interesting to investigate the innate immune response induced, as it was previously shown that more type I interferon (IFN) was produced by cells infected with VEEV/mutSG/IRES/1 compared to TC-83 [22]. The capsid proteins of VEEV (and the closely related eastern equine encephalitis virus), involved indirectly in the antagonism of cellular antiviral responses through cellular transcription shutoff [50], [51], [52], remains under the control of the IRES in VEEV/IRES/C, which could imply a lower level of its expression and thus a reduced inhibition of the cellular antiviral response, including type I IFN. If this pattern is confirmed in the course of VEEV/IRES/C infection, it could potentially influence the nature and quality of the adaptive immune response, which is regulated by the innate immune response [53], [54], [55]. Moreover, neutralizing antibodies are not absolutely required for protection against VEEV challenge [22], [56], a finding supported by our data showing the survival of some challenged animals without detectable neutralizing antibodies. These observations suggest a significant role of the cellular adaptive immune compartments in protection against VEEV infection. Paessler et al. also demonstrated that T-cells alone protected against encephalitis following VEEV infection [56]. Thus, although the humoral response to VEEV/IRES/C appears to be lower than that induced by TC-83, the cellular compartment should also be evaluated.
In conclusion, we demonstrated that this novel, IRES-based TC-83 recombinant virus is superior to TC-83 in attenuation yet provides equivalent protection in a mouse model. Its inability to infect mosquitoes increases its safety by reducing the potential for natural spread after vaccination followed by reversion, which could lead to the initiation of an epidemic. Finally, this study also demonstrates that the IRES can be positioned alternatively to achieve the optimal balance between attenuation and immunogenicity, and along with other studies performed with chikungunya and eastern equine encephalitis viruses [28], [45], [57], further validate the IRES attenuation strategy as an effective and predictable approach for vaccine development against other alphaviruses constantly threatening developing countries.
|
10.1371/journal.pgen.1007788 | Single-cell RNA-sequencing reveals transcriptional dynamics of estrogen-induced dysplasia in the ovarian surface epithelium | Estrogen therapy increases the risk of ovarian cancer and exogenous estradiol accelerates the onset of ovarian cancer in mouse models. Both in vivo and in vitro, ovarian surface epithelial (OSE) cells exposed to estradiol develop a subpopulation that loses cell polarity, contact inhibition, and forms multi-layered foci of dysplastic cells with increased susceptibility to transformation. Here, we use single-cell RNA-sequencing to characterize this dysplastic subpopulation and identify the transcriptional dynamics involved in its emergence. Estradiol-treated cells were characterized by up-regulation of genes associated with proliferation, metabolism, and survival pathways. Pseudotemporal ordering revealed that OSE cells occupy a largely linear phenotypic spectrum that, in estradiol-treated cells, diverges towards cell state consistent with the dysplastic population. This divergence is characterized by the activation of various cancer-associated pathways including an increase in Greb1 which was validated in fallopian tube epithelium and human ovarian cancers. Taken together, this work reveals possible mechanisms by which estradiol increases epithelial cell susceptibility to tumour initiation.
| Women who take estrogen replacement therapy are at higher risk of developing ovarian cancer. When ovarian epithelial cells are exposed to estrogen, there is a heterogeneous cellular response, with some cells appearing unaffected, while others become disorganized and grow at accelerated rates consistent with pre-cancerous cells. This heterogeneity confounds traditional methods for surveying gene expression, which rely on averaging the signal across a population of cells. Here, we employ single cell RNA sequencing in order to measure gene expression profiles at single-cell resolution. This allowed us to distinguish between estrogen-responsive and unresponsive populations and identify defined expression signatures for each. Also, because cellular responses are asynchronous, we were able to use the snapshot of expression profiles to infer the transcriptional changes as cells respond to estrogen and become increasingly disorganized. These techniques revealed not only the processes that may contribute to the earliest stages in the formation of estrogen-driven pre-cancerous cells, but also identified biomarkers of that transition. We have confirmed that the protein GREB1 appears in the pre-cancerous cells and is present in the majority of human ovarian cancers.
| Ovarian cancer is the deadliest gynecological cancer with an overall 5-year survival rate of only 45%. However, if detected early, this survival rate increases to 92% [1]. These statistics demonstrate the importance of understanding the initiating events of ovarian cancer to better develop strategies for prevention and early detection. To date, there are many known risk factors for ovarian cancer but the molecular events leading to transformation remain unclear. Meta-analysis of 52 epidemiological studies investigating menopausal hormone use and ovarian cancer risk found that 55% of women who developed ovarian cancer had also used estrogen replacement therapy [2]. Using a mouse model of ovarian cancer, 17β-estradiol (E2) was shown to accelerate onset of ovarian cancer as E2 treated mice reached their end-point 30–40 days before microscopic tumours were observed in control mice [3]. Wild-type mice exposed to exogenous E2 over 60 days (modelling hormone replacement therapy) developed dysplastic lesions on the ovarian surface epithelium (OSE) and fallopian tube epithelium (FTE) [3,4]. On the ovary, these lesions are characterized by an increased incidence of both columnar epithelial cells and stratified hyperplastic cells. On the oviduct, there was significant increase in both stratified and hypertrophic FTE. Recent use of unbiased genomic and proteomics approaches demonstrate that both OSE and FTE are capable of giving rise to epithelial ovarian cancer [5,6]. Since our lab showed that prolonged E2 exposure impacts both cell types and genome-wide transcriptional regulation of estrogen receptor targets in mouse FTE was recently investigated [7], we present the transcription dynamics of estrogen-induced dysplasia in primary cultures of mouse OSE to further advance our understanding of how estrogen can accelerate transformation from either cells of origin.
The E2-induced OSE dysplasias are reproducible using primary cultures of OSE cells, where areas of columnar OSE and foci of multi-layered cells can both be observed after 15 days of exposure to E2. Immunohistochemical (IHC) staining of mouse ovaries and immunofluorescent (IF) staining of OSE cultures showed that these distinct phenotypes are associated with opposing gene expression patterns [4]. These co-existing morphologies highlight the heterogeneity of the responses to E2, confounding conventional methods used to measure transcript and protein expression changes. Here, we use single-cell RNA-Sequencing (scRNA-Seq) to tease apart this heterogeneity and to understand the phenotypic trajectory of OSE cells throughout the process of foci formation then extend our main finding to mouse FTE and human ovarian cancer tissue.
Substantial dysplasia was observed in primary cultures of OSE cells after prolonged E2 exposure, with disparate morphological changes present in the same dish (Fig 1A). To identify subpopulations of OSE and their responses to E2, we used scRNA-Seq to compare control cells and those that had been treated with E2 for 15 days. After processing and filtering out poor quality libraries (Figure in S1 Fig), the dataset comprised 329 control and 260 E2-treated cells, with a median of approximately 120,000 reads/cell and 4,000 genes/cell (Fig 1B).
Principal component analysis (PCA) revealed two distinct clusters of cells along the first principal component that did not segregate based on E2 treatment (Fig 1C; Figure in S2 Fig). PC2 and PC3, however, did correlate with the treatment. No known technical variable or typical biological source of variation (eg. cell cycle) correlated with these principal components, suggesting they represent biological variation of interest.
To determine the gene expression differences that drive structure in the PCA, we used SC3 [8]—a consensus-based k-means clustering algorithm—to cluster the data. Given the two distinct populations along PC1, we first set k = 2 to define two clusters, splitting the mixed populations of cells along the first principal component (Fig 2A and 2B). We then used monocle [9] to identify 1,132 differentially expressed genes between clusters (Table in S1 Table).
Since the cells of the rightmost cluster did not spread along PC2 according to E2 exposure, while the cells of the left cluster did, we predicted that these two clusters capture differences in the cells’ ability to respond to E2. None of the additional principal components captured E2-associated differences in the cells of the right cluster and further sub-clustering does not separate the E2-treated cells from the untreated cells of this population (Figure in S3 Fig), suggesting these cells are effectively indistinguishable, despite differences in their culture conditions. The estrogen receptors Esr1 and Esr2 were detected in 23.4% and 4.9% of all cells, respectively, and Gper1 (G protein-coupled estrogen receptor 1) was not detected in any cell (Figure in S4 Fig). Esr1 is the predominant mediator of the transcriptional effects of E2 signalling in the OSE [10] and was expressed significantly higher in the leftmost cluster (q = 0.02). Together, this suggests that the right cluster represents a population of cells that are E2-unresponsive, despite having some degree of Esr1 expression. To further support that the distribution of cells along the first principal component corresponds to estrogen responsiveness, we scored each cell based on their expression of genes comprising the “Early estrogen response” hallmark gene set from the Molecular Signatures Database [11]. All cells of the rightmost cluster, regardless of whether they had been exposed to E2, had lower gene set scores (t-test p = 9.16x10-160; Figure in S4 Fig).
Functional enrichment analysis of differentially expressed genes between the right and left clusters was performed to explore the biological differences between these two clusters (Table in S2 Table). The E2-responsive cluster seems more metabolically active than the unresponsive cluster, as demonstrated by the top ontology terms for this cluster (Fig 2C). The E2-unresponsive cluster was characterized by regulation of Actin cytoskeleton and, surprisingly, pathways in cancer (Fig 2C). However, the majority of genes in this term are involved in evading apoptosis (Xiap [12], Bmp4 [13]). Although some pro-proliferative signals were present (Kras, Hras, and Araf), there was also significant up-regulation of Pten, a tumour suppressor that is sufficient to inhibit the proliferative signals of KRAS in OSE [14]. Cells in this cluster also have high expression of Actin, Mmp2, and Rac1—key genes for cell motility, filopodia, and lamellipodia formation. Thus, cells in the E2-unresponsive cluster are predicted to be mesenchymal in morphology and more quiescent than E2-responsive cells.
The initial decision to produce two clusters was based on the observed PCA structure; however, we found that k = 3 produced optimal clusters, maximizing the cells’ average silhouette width. Interestingly, the E2-unresponsive cluster was retained (Cluster 1), while the E2-responsive cluster was split into two clusters of almost exclusively control (Cluster 2) or E2-treated (Cluster 3) cells (Fig 2D and 2E). A total of 189 genes were differentially expressed between the two condition-specific clusters (Clusters 2 and 3) (Table in S3 Table). Control cells of the responsive cluster (Cluster 2) had higher expression of genes involved in wound healing, aging, response to hypoxia, negative regulation of Notch signalling, and NOD-like receptor signalling. E2-treated cells (Cluster 3) had an up-regulation in genes involved in metabolic, PI3K-Akt, VEGF, and Rap1 signaling pathways (Fig 2F) (Table in S4 Table).
To confirm that the clustering patterns identified in the scRNA-Seq data represented distinct populations of cells in vitro, we defined marker genes for each cluster (Fig 3A; Table in S5 Table). Protein expression encoded by the top marker genes Ltbp2 (unresponsive cluster), Ptgis (control-specific cluster), and Igfbp5 (E2-specific cluster) was assessed using IF (Fig 3B; Figures in S5–S7 Figs). LTBP2, a secreted protein, was detected in the cytoplasm and extracellular matrix surrounding control and E2-treated cells that had a flat, mesenchymal appearance (in contrast to the cobblestone, epithelial appearance of most OSE). PTGIS is an endoplasmic reticulum membrane protein that contributes to the “cellular response to hypoxia” ontology term in control cells (Fig 2F). It was modestly expressed in all cells, with a typically perinuclear localization. Expression was highest in control OSE cells where PTGIS was detectable throughout the cell cytoplasm. IGFBP5 was mostly absent in control OSE, but was predominant in E2-treated OSE. These results support the existence of 3 populations of cells in the in vitro model of E2-induced dysplasia, as predicted by PCA, where high expression of PTGIS and IGFBP5 are observed in the majority of cells in control and E2-treated cultures, respectively. This also supports that both control and E2-treated cultures have a subpopulation of mesenchymal OSE cells that uniquely expresses LTBP2. Consistent with this population being more mesenchymal, we found that the hallmark gene set “TGFB signalling” was dramatically higher in this cluster (t-test p = 1.17x10-107; Figure in S8 Fig).
To determine if these subpopulations are found in vivo, ovaries from mice treated with exogenous E2 were evaluated by IHC with these same markers (Fig 4). E2-treated mice had significant areas of columnar and stratified OSE, as previously reported [3,4], and these cells had faint or absent LTBP2 and PTGIS expression, but strong punctate apical staining of IGFBP5. Placebo mice remain subject to endogenous estrus cycle activity and some areas of columnar OSE were observable, but the majority of OSE were squamous to cuboidal. PTGIS and LTBP2 were present at varying levels in the squamous to cuboidal cells, while IGFBP5 was consistently faint. These findings are largely consistent with the expression patterns predicted from the scRNA-Seq and IF data.
The formation of dysplastic regions is an asynchronous process: following E2 exposure, we observed both non-confluent monolayers and variably-sized foci of dysplastic cells. Therefore, we hypothesized that sampling cells for scRNA-Seq at a single time point would capture cells at various stages of foci formation, allowing us to use pseudotime analysis to reconstruct the trajectory of transcriptional changes that occur throughout the process. The continuous distribution of cells along the principal components, rather than discrete clustering, supports this hypothesis.
We used monocle [9] to construct a transcriptional trajectory of all the cells, naive to the condition that the cells came from. This trajectory was characterized by a branch common to cells from both conditions (representing the E2-unresponsive cluster) that diverges from a branch point into two condition-specific branches (Fig 5A). This can be interpreted as a linear trajectory for each condition, where depending on whether the cells had been exposed to E2, the trajectory diverges into a different state space. If the columnar and stratified states were separate “lineages” arising from a common precursor cell, we would expect a branched pattern in the trajectory of E2-treated cells, and if they arose from two distinct cell populations, we would expect two disconnected trajectories in E2-treated cells. Given that cells form a connected trajectory, and that each culture condition’s trajectory seems linear, this data supports the hypothesis that the columnar and stratified phenotypes represent different stages of OSE cell dysplasia along a common transcriptional path. Additionally, the trajectory connects the E2-unresponsive population with the responsive population, suggesting that cells can transition between these two states.
To characterize the divergent phenotypes in the trajectory, we used monocle [9] to pseudotemporally order the cells, ranking their distance along the trajectory from the cell of the E2-unresponsive branch farthest away from the branch point (Fig 5B). While pseudotime is often used to reconstruct temporal, albeit asynchronous processes, such as differentiation, we only use it here to define the relationship between cells along a phenotypic continuum. In this context, the defined root state is not meant to represent the starting point of a biological process, and unidirectionality is not assumed as cells transition through this continuum.
After assigning a pseudotime value to cells, we identified 693 genes with branch-dependent gene expression dynamics (Fig 5C; Table in S6 Table). Clustering of these genes based on their expression dynamics identified groups with common expression patterns (Fig 5D). We found that OSE cells cultured long-term in a steroid-free environment acquire more characteristics of cell stress as they diverge from the branch point. There is induction of genes associated with cell aging, inflammatory pathways, positive regulation of cell death, and other immunogenic signals for apoptosis and clearance (Fig 5E; Table in S7 Table). E2-treated cells show either no change or a decrease in these genes, but show increased expression of genes involved in metabolic activity, pathways in cancer, and proliferation via PI3K-Akt signaling (Fig 5E). Additionally, E2-treated cells show decreased expression of genes associated with regulating cell shape and cell-cell adhesion, which likely contributes to the development of dysplasia.
While the characteristics of the transcriptional trajectory are consistent with the observed biology of the culture system, we sought to validate that the structure of the trajectory relates to foci formation in the E2-treated cells and cell stress in the control cells. To do this, senescence-associated β-galactosidase (SA-βGal) activity was used as a marker for stressed cells [15–17], which we have previously shown identifies stressed OSE cells in culture [4]. Phase contrast images of the same field of view were overlaid to track cell morphology (Fig 6A). KI67 staining was used as a marker for proliferating cells [18], and actin staining was used to track cell morphology and approximate cell height (Fig 6B). E2-unresponsive cells occupying the common branch were SA-βGal-negative and sometimes KI67-positive in both control and E2-treated cultures. Also common between control and E2-treated cultures were areas of sub-confluent OSE where cells in small epithelial rafts or cells on the periphery of larger rafts were cuboidal in morphology, SA-βGal-negative, and ~50% KI67-positive. These areas of proliferative and cuboidal OSE cells, seen in both control and E2-treated cultures, are likely cells that exist shortly before phenotypic divergence. As the rafts become larger and more confluent, the divergence predicted by trajectory analysis becomes apparent. Confluent control cells remain similar in height to sub-confluent cells and remain SA-βGal-negative but become less KI67-positive. The final phenotype observed in control cultures, and presumably the last stage in the control-dominant branch of the trajectory, was SA-βGal-positive cells that remain mostly KI67-negative and have become enlarged and flattened in morphology, characteristic of senescent cells [19]. In contrast, E2-treated cells remain proliferative at confluence and either progressively assume a columnar morphology (as demonstrated with increased cell height) or become stratified. Consistent with the few E2-treated OSE cells existing in the control branch of the trajectory, some areas of SA-βGal-positive cells are seen in E2-treated cultures, but SA-βGal-negative columnar and stratified cells are the predominant phenotype.
When exploring the expression patterns of all genes from the “Early estrogen signalling” gene set, we noted that, of all genes, Greb1 had the most strikingly exclusive expression within the branch putatively associated E2-induced foci. GREB1 is a known E2-responsive gene that we have previously shown to be over-expressed in human ovarian cancer relative to healthy human OSE and is known to promote cancer progression [20,21]. In the trajectory, Greb1 was only present in E2-treated cells beyond the branch point and was detected in only 37% of E2-treated cells (Fig 7A). IF staining of OSE cultures demonstrate that all control cells are GREB1-negative and GREB1 is only present in E2-induced stratified OSE (Fig 7B). Together, these data suggest that the modelled trajectory represents the process of E2-induced foci formation where E2-treated cells first become columnar then progress into stratified OSE that are uniquely GREB1-positive. To further validate that the trajectory is representative of foci formation in vivo, we stained ovaries and oviducts of E2-treated mice and found increased expression of GREB1 in both stratified OSE and FTE (Fig 7C). If, in humans, similar dysplastic populations give rise to ovarian tumours, a similar gene expression pattern may be expected. We stained 40 human ovarian tumour samples and found detectable GREB1 in 75–85% of each of the various histological subtypes (Fig 7D).
Assessing a more general similarity to human disease, we also constructed a representative expression profile of human ovarian cancer cells by averaging the expression values from a recent scRNA-seq study of a human high-grade serous ovarian tumour [22] and correlated this profile with each cell in our study. This revealed that as cells form foci, their expression profile becomes more similar to human ovarian cancer cells (Fig 7E). Also, while E2-treated foci and confluent control cells both display a hypoxia response expression signature (Fig 7F), E2-treated cells fail to activate p53 signalling (Fig 7G). This may contribute to their sustained proliferation and could also drive genetic instability. Along with the activation of cancer-associated pathways as cells progress through this trajectory, these data support that the molecular profile of dysplastic regions may be consistent with pre-cancerous lesions in humans.
The OSE monolayer is well-documented as a heterogeneous population with multiple morphologies observable in both in vitro primary cultures and in vivo models [23]. Full appreciation of the heterogeneous nature of OSE cells was made possible using scRNA-Seq and, in doing so, we have resolved the asynchronous response to estrogen that OSE cells display, resulting in the formation of dysplastic lesions (Fig 8).
In the in vitro model of E2-induced OSE dysplasia, where control and E2-treated OSE cells are expanded in culture over 15 days, we propose that sub-confluent cells from both cultures are phenotypically similar where they are proliferative, metabolically active, and cuboidal. With increasing confluence, cell-cell adhesions form between control cells and proliferation is slowed due to contact inhibition mediated by regulators of polarity, such as Dab2 [4]. Without E2, control cells become stressed with increasing confluence and upregulate genes associated with hypoxia (Ptgis, Cryab [24]), cell death (Casp1, Fas [25]), and inflammatory responses involving immune cell recruitment through the TNF signalling pathway (Ccl2, Icam1) [26,27]. E2-treated cells appear to overcome hypoxia via up-regulation of E2-responsive Hif1a [28] and Connexin 43 (Gja1) [29,30]. Inflammatory immunogenic signals were suppressed in E2-treated cells, suggesting that E2 may contribute to an immune-restricted microenvironment, allowing pre-neoplastic lesions to develop and evade immune surveillance [31]. In addition to bypassing stress and apoptosis, E2-treated cells show up-regulation of growth factors (Fgf9, Kitl), mediators of MAPK signaling (Map2k1, Mapk14), Wnt signaling (Wnt5a), PI3K-Akt signaling, and Fos (a known E2-responsive oncogenic driver of proliferation and differentiation [32,33]). This increased proliferation is likely sustained by increased metabolism, where E2-treated OSE activate PPAR signaling to promote fatty acid oxidation and mitochondrial enzymes (Acat1, Acadm) that can utilize these fatty acids in the Krebs cycle [12].
Krt19 was consistently higher in E2-treated OSE cells in all analyses performed. While this suggests that E2 promotes epithelial differentiation, components necessary for proper monolayer formation such as integrin (Itga7, Itgav), tight junction (Tjp2), cell adhesion (Tln1), cytoskeleton (Actb) genes were all decreased in E2-treated cells. By confocal microscopy, E2-induced columnar cells observed in confluent areas consistently outnumber confluent control cells in a given field of view, suggesting that increased cell height maximally accommodates proliferative cells in an already confluent monolayer while still allowing attachment to the basement membrane. With increased proliferation associated with progressive loss of epithelial architecture, E2-treated cells also upregulate key genes involved in amoeboid migration (Rock2 [34], Nanog [35], Cdc42 [36]). This phenotype can be driven by Hif1a [37] and has been shown to be the preferred mode of migration of epithelial cancer cells to detach from the ECM and move through dense hypoxic conditions, providing a possible explanation for OSE stratification.
Given the lack of strategies for early detection of ovarian cancer and consequent poor patient prognosis, investigating genes upregulated throughout E2-induced dysplasia formation may inform development of diagnostic and prognostic markers for E2-driven and -responsive tumours. Igfbp5 [38–40], Enpp2 [41], Ctsh [42,43], Lpl [44,45], and Tacc1 [46–48] are the top five genes correlated with the E2 branch in pseudotime and all are currently under investigation as cancer biomarkers. Preliminary studies in various human pathologies have demonstrated that IGFBP5 [49], ENPP2 [50], ENPP2 metabolites [44] and LPL [44] are secreted factors detectable in serum. Given that they are up-regulated in pre-neoplastic cells, these are strong candidates to be investigated as markers for the early detection of ovarian cancer. Greb1 stands out because it is a known E2-responsive gene in breast and ovarian cancer that drives tumour progression [20], and it is currently being investigated as an alternative prognosis marker for tamoxifen treatment [51]. Greb1 has previously been shown to be E2-inducible in normal mouse FTE [7] and our present finding is the first to show E2 induction of GREB1 in OSE, where it is uniquely expressed in dysplastic cells after prolonged E2 exposure. Given that GREB1 expression is highest in dysplastic OSE and FTE, this suggests that GREB1 may play a role in the initiation of tumour formation in both cell types.
In this study, scRNA-Seq was used to provide a global mechanistic explanation for how prolonged E2-exposure can lead to OSE dysplasia and increased susceptibility to transformation, and to reveal potential biomarkers for early detection. Currently, the data remains largely correlative and additional biological assays are required to validate the model and delineate correlation from causation. Nevertheless, this study is the first in the field of ovarian cancer to explore the initiating events in the transformation of ovarian epithelial cells at single-cell resolution. It provides a solid foundation for supporting current hypotheses and a framework for developing new strategies for ovarian cancer prevention and early detection.
Experiments involving mice were performed according to the Canadian Council on Animal Care Guidelines for the Care and Use of Animals on a protocol approved by the University of Ottawa Animal Care Committee. Mice in this study were euthanized using CO2. Tissue microarray of human ovarian cancers was obtained from the Cooperative Human Tissue Network (University of Virginia, U.S.A.).
Exogenous E2 was delivered to FVB/N mice and their tissue was collected as previously described (n = 5/treatment) [4].
OSE cells were isolated from mouse ovaries, maintained, and treated with E2 as previously described [4,52]. OSE were maintained at 37°C in OSE media that consists of α-MEM media (Corning) with 5% fetal bovine serum, 2ng/mL epidermal growth factor (Sigma), and 0.01mg/mL insulin-transferrin-sodium selenite supplement (Roche). For E2 treatment of OSE, cells were seeded and allowed to normalize to hormone-free media consisting of 5% charcoal-stripped fetal bovine serum in phenol red-free DMEM-F12 media (Sigma) for 48 hours before treating with 100nM E2 (Sigma). An equivalent volume of 100% EtOH (vehicle) was added to control dishes for a final concentration of 0.0002% EtOH. Media was refreshed every 3–4 days and collected for single cell RNA-sequencing or immunofluorescent staining 15 days after E2 treatment. All control and E2-treated OSE cell experiments analyzed in this study are after 15 days in culture.
Cells were processed and sequencing libraries were prepared according to the Fluidigm C1 Single-Cell mRNA Seq HT IFC v2 protocol, using a medium (10–17μm) integrated fluidic circuit (IFC). For quality control, samples were stained with a LIVE/DEAD viability stain (ThermoFisher) and each capture site was visually inspected using an EVOS FL Cell Imaging System (ThermoFisher) to score the number and viability of cells in each capture site. Successful library tagmentation was assessed using the Advanced Analytical Fragment Analyzer and sample molarity was calculated using the average fragment size. Samples were pooled at equal molarity, spiked with 20% PhiX, and sequenced together in a single high-output 150 cycle NextSeq500 run with read 1 set to 26bp and read 2 set to 75bp.
Fastq files corresponding to the 20 cDNA libraries were demultiplexed into 800 fastq files, each representing one of the 800 capture sites on the Fluidigm C1 IFC, using Fluidigm’s API. Transcripts were quantified using kallisto [53], and gene-level read counts were imported into R using the tximport package [54]. The scater package [55] was used for quality control, normalization, and exploratory analysis of the data.
Only capture sites annotated to contain one live cell were kept. Cells with less than 20,000, or more than 250,000 reads were removed to eliminate poor quality libraries, and capture sites that may contain multiple cells, respectively. Cells with a high or low proportion of detected genes (>3 median absolute deviations—MADs—from mean) and those with a high proportion of mitochondrial reads (>3 MADs) were removed. Lastly, only genes that were detected in a minimum of 10 cells were retained. The final expression matrix included the expression of 14299 genes across 636 cells.
The R package scran [56] was used to calculate scale factors for each cell as described previously [57]. Each cell’s expression profile was then scaled by this factor. To restore structure in the data that may be lost as the result of drop-out characteristic of scRNA-Seq, we performed data imputation using the MAGIC algorithm [58] with the following parameters: n_pca_components = 20, t = 4, ka = 3, k = 9, epsilon = 1. These values were used for downstream clustering and pseudotime analysis.
Imputed expression values were used to define cell clusters. k-means-based consensus clustering was performed using SC3 [8] to segregate the cells into 2 or 3 clusters. The optimal number of clusters (k = 3) was determined by finding the value of k that maximized average silhouette width. k = 2 segregated the two large groupings of cells along the first principal component, so we also explored this clustering pattern in downstream analysis. Differential expression analysis was performed using monocle [9] to fit a generalized linear model to the expression values of each gene and filtering for genes with a q-value < 0.05 and a log2 fold change > 0.5.
The DAVID Gene Functional Classification Tool [59,60] was used to explore KEGG pathways and GO terms significantly associated with differentially expressed genes between clusters in PCA and clusters of genes with similar expression dynamics throughout pseudotime. The top 5 terms ranked by EASE p-value were determined by subjecting each cluster to identical filtering parameters. Terms were removed if there was intracluster term conflict, (eg. both positive and negative regulation of apoptosis appearing in the same cluster), if term is not unique to one cluster, if the biological process had 10 or more child terms (to eliminate broad and uninterpretable processes), or if they could be reasonably agreed upon as not applicable to the model system (eg. liver regeneration, Alzheimer's disease, etc.). KEGG pathways were given priority because they provide the most information for biological interpretation of data. Full list of unfiltered GO Terms are made available in Tables in S2, S4, and S7 Table.
Marker genes for each cluster were identified using SC3 [8], which constructs a binary classifier for each gene based on the mean expression of the gene in each cluster. The area under the receiver operator characteristic curve (AUC) is used to determine the accuracy of the gene as a marker for the cluster. The AUC values were ranked, and top-ranking genes for each cluster were chosen for validation by IF based on commercial availability of antibody and quality of stain (assessed by preliminary antibody optimization experiments).
A pseudotime trajectory of all cells was constructed using the DDRTree method implemented in monocle2 [61]. The root branch was defined as the branch comprising cells from each condition. A generalized linear model was fit to the expression values with an interaction term to model branch-specific expression dynamics as a function of pseudotime. Because the imputation and normalization eliminate a large amount of variation, the interaction coefficients for each gene were used to identify differentially expressed genes (>2 standard deviations from mean of all coefficients).
Hallmark gene sets were acquired from the Molecular Signatures Database [11]. Each cell was scored based on its expression of the genes within each gene set using the AddModuleScore function in the R package Seurat [62]. For each cell, this function determines the average relative expression of each gene of the gene set compared to groups of expression level-matched control genes. These relative expression values were then standardized using a Z-score transformation.
Processed scRNA-seq data was retrieved from Winterhoff et al. [22] Expression values were log-transformed and scaled before using the highly variable genes from the manuscript to cluster the cells into two groups using hierarchical clustering. Similar to the data presented in the original publication, this separated the population into a CD24+ cluster of cancer cells, and a CD24- cluster of stroma cells. The expression profiles of all cancer cells were averaged to produce a representative expression profile of human ovarian cancer cells. Using genes present in both the mouse and human data sets, this expression profile was correlated (Spearman’s rank correlation) with each cell of our data set.
All data are available at GSE121957 and analysis notebooks are hosted at https://github.com/dpcook/scRNASeq-Estrogen.
Cells were seeded onto glass coverslips and treated with E2 for 15 days. Cells were fixed, permeabilized, blocked, and probed according to antibody datasheet instructions, then mounted onto slides using ProLong Gold mountant with DAPI (ThermoFisher Scientific). See Supp.Table.8 for details on antibodies used for IF. All IF experiments have at least two independent experimental replicates and “n” refers to fields of view.
Confocal images were acquired using a Zeiss LSM 800 confocal microscope system using Plan-Apochromat 40X/1.3NA oil objective. Images were processed using the Imaris Image Analysis Software v8 (Bitplane). IF images are shown as z-stack maximum intensity projections of DAPI (blue) or Actin (orange) channels. Protein of interest (green) is represented as solid surface objects created based on the dynamic range of all control and E2 treated images acquired within the biological replicate. The surface object intensity threshold was set based on the maximal intensity value of the dynamic range. Control slides processed without primary antibody and original unmerged images are included as supplementary (Figures in S5–S7 Figs). %KI67+ cells was determined using ratio of KI67+nuclei/total nuclei (DAPI) (n = 6). Cell membrane co-staining was not ideal due to permeabilization required for IF antibodies so Actin was used as an alternative means to estimate cell height (n = 6). Cell height was determined by Actin surface rendering in Imaris. Due to mounting of coverslip, cell height was generally limited to <~20μm.
Cells were seeded onto glass coverslips and treated with E2 for 15 days. Cells were fixed with 4% paraformaldehyde (w/v, in PBS) for 10 mins, then incubated with SA-βGal staining solution [63] overnight at 37°C. Coverslips were mounted onto slides using ProLong Diamond antifade mountant (ThermoFisher Scientific) and allowed to cure for 24 hours. Slides were visualized using an EVOS 507 XL Core imaging system under phase-contrast and brightfield to best capture morphology and SA-βGal stain, respectively. Quantification of SA-βgal was performed using Fiji version 1.0. SA-βgal-positive (SA-βgal+) cell pixels were quantified using automated colour deconvolution (Methyl Green DAB vector, colour 1) on brightfield images. SA-βgal- pixels were quantified by subtracting bright-field pixels from the corresponding phase-contrast pixels using Image Calculator. %SA-βgal+ cells was determined using ratio of SA-βgal+ cell pixels/total cell pixels.
IHC was performed using a previously described protocol [3]. IHC staining was performed on ovaries from 3–5 mice per treatment group and on a tissue microarray of human ovarian cancers obtained from the Cooperative Human Tissue Network (University of Virginia, U.S.A.). See Table in S8 Table for details on antibodies used for IHC. No-primary control sections are included in Figure in S9 Fig. Images were acquired using ScanScope CS2 (Leica Biosystems, Concord, Canada).
|
10.1371/journal.pntd.0003941 | A Field Study in Benin to Investigate the Role of Mosquitoes and Other Flying Insects in the Ecology of Mycobacterium ulcerans | Buruli ulcer, the third mycobacterial disease after tuberculosis and leprosy, is caused by the environmental mycobacterium M. ulcerans. There is at present no clear understanding of the exact mode(s) of transmission of M. ulcerans. Populations affected by Buruli ulcer are those living close to humid and swampy zones. The disease is associated with the creation or the extension of swampy areas, such as construction of dams or lakes for the development of agriculture. Currently, it is supposed that insects (water bugs and mosquitoes) are host and vector of M. ulcerans. The role of water bugs was clearly demonstrated by several experimental and environmental studies. However, no definitive conclusion can yet be drawn concerning the precise importance of this route of transmission. Concerning the mosquitoes, DNA was detected only in mosquitoes collected in Australia, and their role as host/vector was never studied by experimental approaches. Surprisingly, no specific study was conducted in Africa. In this context, the objective of this study was to investigate the role of mosquitoes (larvae and adults) and other flying insects in ecology of M. ulcerans. This study was conducted in a highly endemic area of Benin.
Mosquitoes (adults and larvae) were collected over one year, in Buruli ulcer endemic in Benin. In parallel, to monitor the presence of M. ulcerans in environment, aquatic insects were sampled. QPCR was used to detected M. ulcerans DNA. DNA of M. ulcerans was detected in around 8.7% of aquatic insects but never in mosquitoes (larvae or adults) or in other flying insects.
This study suggested that the mosquitoes don't play a pivotal role in the ecology and transmission of M. ulcerans in the studied endemic areas. However, the role of mosquitoes cannot be excluded and, we can reasonably suppose that several routes of transmission of M. ulcerans are possible through the world.
| Buruli ulcer is a neglected tropical disease due to M. ulcerans, an environmental mycobacteria. Modes of transmission to human remain unclear and water bugs and mosquitoes had been incriminated with more or less experimental laboratory evidences and filed studies. In this context, we have investigated the presence of M. ulcerans DNA in mosquitoes and other flying insect in a highly endemic area of Buruli ulcer in Benin. No trace of the bacteria was found in mosquitoes and other flying insects, while 8,7% of aquatic insects, including water bugs, caught in the same area and in the same period were found positive to M. ulcerans DNA. Our results support the hypothesis that mosquitoes don’t play a major role in ecology of M. ulcerans in our research area and is in favor of a transmission from the aquatic environment.
| Buruli ulcer, which is caused by M. ulcerans, is a neglected tropical disease affecting mostly poor rural communities in West and Central Africa. In 2013, 75% of all new cases of Buruli ulcer worldwide were declared by Ivory Coast, Ghana and Benin. This skin disease, which mostly affects children, causes large ulcerative lesions often leading to permanent disabilities [1,2,3]. The cutaneous lesions are caused by a M. ulcerans toxin called mycolactone with cytotoxic, immunomodulatory and analgesic effects [4]. At early stages, Buruli ulcer can be treated with a combination of streptomycin and rifampin for eight weeks; at later stages, antibiotic therapy is associated with extensive surgery [5,6,7,8].
Buruli ulcer occurs mostly in low-lying swampy areas [9,10]. Epidemiological studies have shown that the aquatic environment is the main reservoir of M. ulcerans, with many aquatic vertebrates and macro-invertebrates harboring this bacillus. The exact ecological features and mode of transmission of M. ulcerans to humans remain to be identified. In recent decades, several studies have suggested that water bugs and mosquitoes may play a role in M. ulcerans transmission [11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]. Water bugs have been implicated as potential hosts and vectors of the bacillus in laboratory experiments and field ecology studies in Africa [26,27,28,29,30]. Outside the aquatic environment, adult mosquitoes tested positive for M. ulcerans DNA in an area of endemic Buruli ulcer in Australia, leading to the suggestion that these insects might transmit the bacterium to humans [26,28,29,30]. However, this hypothesis was not confirmed by laboratory experiments, and, surprisingly, no study has investigated the possible involvement of mosquitoes in M. ulcerans ecology in Africa, the continent with the highest level of endemicity for Buruli ulcer.
The objective of this study was to investigate the presence of M. ulcerans DNA in flying insects, including mosquitoes, in an area of Buruli ulcer endemicity in Benin. We monitored, in parallel, the levels of M. ulcerans in the aquatic environment, as a marker of the presence of the bacterium in the study area.
The study was carried out in the Oueme administrative area in South-East Benin, where Buruli ulcer has been endemic for several decades [31,32,33,34,35,36]. Sampling was carried out in three districts crossed by the Oueme River (Bonou, Adjohoun and Dangbo). The districts were selected for study because they are accessible throughout the year (including the rainy season) and because data were available for relevant epidemiological studies. Flying insects were sampled at four sites and aquatic sampling was carried out at nine sites (Fig 1).
The Oueme River originates in the Taneka hills in the Atacora Mountains and flows into the Atlantic Ocean close to Cotonou. The study area is characterized by a subequatorial climate with two rainy seasons. The first rainy season extends from April to July and the second extends from October to November. Mean annual precipitation is 1122 mm, and temperatures range from 22°C to 26°C. There are two main types of soil: alluvial soils, which are fertile but liable to flooding, and sandy soils, which are less fertile but suitable for growing coconut, palm, and other tropical trees. Most of the population in this area is engaged in farming (rice, maize, cassava, cowpeas, market garden crops, etc.), fishing and trade. The natural vegetation consists of grassy savannah and swampy mangrove forest.
This study focused on the adult stage of mosquitoes and other flying insects and the immature stages of mosquitoes. Insects were collected during four surveys in June, July, November, and December 2013, at four sites in the Bonou Centre, Kode, Gbada and Houeda areas (Fig 1). The collection periods correspond to the start, middle and end of the rainy season and the dry season, respectively. Flying insects were collected with Centers for Disease Control (CDC) light traps. A CDC light trap consists of a 150 mA incandescent light bulb and a fan, powered by 6 V batteries. At each survey, once consent had been received from the heads of household, insects were trapped from two selected houses in each village, over a period of two days. Traps were placed both indoors and outdoors at each house, from 6:00 pm to 6:00 am, corresponding to the period from dusk to dawn. The indoor traps were suspended from the ceiling, about 2m above the ground. The outdoor traps were hung on trees at about the same height. The insects collected were identified in the field in two steps. In the first step, mosquitoes were separated from the other insects. All mosquitoes were identified to species level under stereoscopic microscopes, according to morphological criteria in dichotomous keys [37,38,39]. They were counted and stored, in pooled groups of up to 15 individuals of the same species, in 70% ethanol for transport to the laboratory. In the second step, the remaining flying insects were identified to order level on the basis of their morphology under a stereoscopic microscope, with the appropriate keys [40,41]. They were stored in 70% ethanol, in pooled groups of up to 15 individuals from the same order, and were transported to the laboratory for PCR analysis (Fig 2).
During each survey, mosquito larvae were collected throughout the selected area by dipping with a 350 ml ladle. Samples were collected from various temporary and permanent bodies of water constituting potential habitats for the development of populations of mosquito larvae. All larvae were transported in clean water, in plastic containers, to the field laboratory. Larvae were identified to genus level with appropriate morphological keys [37,38,39]. The larvae of each genus were then separated into two groups. The larvae of the first group were preserved in 70% ethanol, in pools of 20 individuals for each genus. The larvae of the second group were reared to emergence. The resulting adults were then stored in 70% ethanol, in pools of up to 15 individuals. Exuviae were also preserved in 70% ethanol, in pools of 20, for laboratory analysis (Fig 2).
Samples were collected from the principal sources of water for domestic washing, bathing, fishing and recreation. The sampling sites were located in nine villages in the three districts: Bonou Centre, Agbonan, Agbomahan, Agonhoui, Gbame, Kode, Assigui, Houeda, and Mitro (Fig 1). Aquatic sampling was carried out with the same methods at each site, at least twice, between January 2013 and December 2013. Invertebrates and fish were captured with a square net (32 x 32 cm and 1 mm in mesh size), from the surface down to a depth of 0.2 to 1 m, over a distance of 1 m. A sample was considered to correspond to all the insects collected in 10 such sweeps with the net. All insects were preserved in 70% ethanol for laboratory identification. For the detection of M. ulcerans DNA, the insects were sorted into pooled groups, each including no more than 20 specimens from the same family. For each body of water, we collected plant samples from the predominant and the second most frequent types of living plant. Each of these plant samples consisted of one to five plant leaves, stems or roots, depending on the size of the plant sample. They were placed directly in a clean 100 ml bottle containing 70% ethanol (Fig 2).
Pooled insect bodies were ground and homogenized in 50 mM NaOH. Tissue homogenates were heated at 95°C for 20 min. The samples were neutralized with 100 mM Tris-HCl, pH 8.0. DNA was extracted from the homogenized insect tissues with the QIAquick PCR purification kit (Qiagen), according to the manufacturer’s recommendations. Negative extraction and purification controls were included in each series of manipulations. The homogenizers were decontaminated by incubation overnight in 1 M NaOH, to eliminate any traces of DNA. For each aquatic plant sample, the material was cut into small pieces with a scalpel and then ground in 50 mM NaOH. The extract was heated and neutralized and the DNA was purified with the Mobio purification kit, according to the manufacturer’s recommendations.
Oligonucleotide primer and TaqMan probe sequences were used for detection of the IS2404 sequence and the ketoreductase B (KR) domain of the mycolactone polyketide synthase (mls) gene from the plasmid pMUM001 [13,42,43]. PCR mixtures contained 5 μl of template DNA, 0.3 μM of each primer, 0.25 μM probe, and Brilliant QPCR Master Mix (Agilent Technologies) in a total volume of 25 μl. Amplification and detection were performed with a Thermocycler StepOne (Applied Biosystems), using the following program: heating at 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. DNA extracts were tested at least in duplicates, and negative controls were included in each assay. Quantitative readout assays were set up, based on an external standard curve generated with five tenfold serial dilutions of M. ulcerans (strain 1G897) DNA. Samples were considered positive only if both the IS2404 sequence and the gene sequence encoding the ketoreductase B domain (KR) were detected, with threshold cycle (Ct) values strictly < 35 cycles. An inhibition control was performed as previously described [44] and 10% negative controls (water alone) were included in each assay.
Mosquito abundance was compared between sites and between seasons in nonparametric Kruskal–Wallis tests.
We collected 7230 flying insects from nine orders: Coleoptera, Diptera, Heteroptera, Homoptera, Hymenoptera, Lepidoptera, Nevroptera, Odonate, Tricoptera. At all sites, Diptera was by far the most frequent order of flying insects caught, accounting for 84% of all insects trapped. Heteroptera was the least abundant order at each site and was not detected at Gbada and Houeda (Table 1).
The 6047 dipteran specimens collected during the four surveys included 4322 mosquitoes from 10 species. Mansonia africana (50%), Culex nebulosus (27%), and Culex quinquefasciatus (22%) were the most abundant species, accounting for 98% of all the mosquitoes trapped. The four least represented species were Anopheles pharoensis, Aedes vittatus, Culex decens, and Culex fatigans, with no more than four individuals each (S1 Table).
No significant differences in the abundance of the mosquitoes and other flying insects caught were found between sites (p>0.05). Flying insects were significantly more abundant (p<0.05) in the wet season than in the dry season, whereas no significant difference in mosquito densities was observed within seasons (p>0.05; S2 Table).
During the surveys, we collected a total of 5407 mosquito larvae. These larvae were identified as Culex spp., Anopheles spp. and Aedes spp. In total, 3146 mosquito larvae belonged to the genera Culex and Anopheles. Culex spp. were the most abundant, accounting for 66.35% of the mosquito larvae collected. For the adults emerging in the laboratory following the rearing of larvae collected in the field, 2261 individuals belonging to the genera Culex, Anopheles and Aedes were identified. Culex was the most abundant genus, accounting for 79.08% of the sample (S3 Table).
During the survey, we collected 3377 aquatic vertebrates and macro-invertebrates from various bodies of water in the Oueme administrative area (Table 2). Insecta accounted for 72% of the animals collected, with a majority of Hemiptera. The bodies of water studied were of various natures (flooded land, river and swamp) and were scattered around the Oueme, making it possible to sample diverse types of specimens from different ecological niches. In total, 95 plants were collected from the various bodies of water. They were identified as belonging to the Poaceae, Lemnaceae, Nymphaeaceae, Araceae and Potamogetonaceae families.
We tested flying insects, larvae, aquatic vertebrates and invertebrates, and plants collected in 2013 from various sites in Oueme for the presence of M. ulcerans DNA. We found that 942 pools of flying insects (corresponding to the 7230 captured flying insects and the 5407 collected larvae) tested negative for M. ulcerans DNA by PCR. Positive PCR results were obtained for 8.7% (28/322) of aquatic animal sample pools from the various bodies of water. No positive specimens were obtained at two sites, and 5.5 to 25% of the sample pools at the other seven sites tested positive (Table 2 and S4 Table). Decapoda was the invertebrate family with the highest level of mycobacterial contamination (26%). We performed 295 PCR analyses on the 95 plants collected. These analyses were carried out on leaves, stems and roots, and three samples tested positive for M. ulcerans DNA by PCR: a leaf pool and a stem pool from the same plant from a water body in Kode and a leaf pool from Mitro (S4 Table). Both plants concerned belonged to the Poaceae plant family.
The ecological characteristics and mode of transmission of M. ulcerans are not entirely understood, and several fundamental questions remain unanswered. One key concern relates to the routes by which M. ulcerans crosses the human skin barrier. There are currently two main hypotheses: (i) direct contact between an existing wound and water containing M. ulcerans; (ii) the inoculation of M. ulcerans into the skin [2,45]. Comparisons with the modes of transmission of other environmental mycobacteria in immunocompetent humans (e.g. M. fortuitum, M. chelonae, M. xenopy) and recent studies of M. ulcerans [46] have suggested that direct inoculation into the skin is the most likely mode of transmission. In this context, the two most likely scenarios for the inoculation with the bacterium are either inoculation by an active vector harboring M. ulcerans, as described for various microorganisms, including parasites (e.g. Leishmania sp. or Plasmodium sp.), arboviruses (e.g. the Dengue and Chikungunya viruses), and bacteria (e.g. Yersinia pestis and Borrelia sp.), or inoculation by a mechanical vector, such as aquatic plant thorns or sharp leaves, biting or sucking insects (bacilli present on the outside of the insects) [13,15,16,17,18,19,20,21,25,26,27,28,29,30,47,48,49,50,51,52]. M. ulcerans ecology is highly complex. It is therefore possible for these scenarios to co-exist, and their importance or significance is dependent on a number of different criteria (e.g. human behavior, including access to drinking water, rural or urban life and work, fauna and flora biodiversity, presence of permissive species, season).
Several experimental studies have explored the role of aquatic hemipterans as passive or active vectors of M. ulcerans. These approaches were supported by various environmental and epidemiological studies conducted in Africa. However, the importance (unique, major, or marginal) of this transmission route has yet to be established and other transmission routes should therefore be explored. For instance, it has been suggested that mosquitoes act as vectors of M. ulcerans in Australia, but, surprisingly, this possibility has never been explored in Africa. In this context, the aim of our study was to assess the role of mosquitoes in M. ulcerans ecology. We carried out an extensive field study in an endemic area in Benin, involving temporal and spatial monitoring of the presence of M. ulcerans in mosquitoes and other flying insects, used as a control for the distribution of M. ulcerans in aquatic flora and fauna.
M. ulcerans DNA was detected in various aquatic macro-invertebrates and vertebrates, and some aquatic plants. The global rate of detection was about 9%, consistent with the findings of other environmental studies [26,27,28,29,30]. M. ulcerans DNA was not detected in any of the flying insects collected in CDC light traps inside and around houses over the same period (including mosquito families in which M. ulcerans DNA was detected in Australia). As only one type of sampling method was used to collect flying insects (CDC light traps), it is possible that this introduced a bias in terms of species diversity. Nevertheless, in a recent study performed in the same area with three other types of sampling method for mosquito collection, the three most abundant mosquito species were the same as in our study, and eight of the 14 species identified were common to our study [53].
Our results suggest that mosquitoes and non-aquatic flying insects are not involved in the ecology and dissemination of M. ulcerans in an area of South-East Benin in which Buruli ulcer is highly endemic, and confirm that the aquatic environment is the main environmental reservoir of the bacillus. However, a role for mosquitoes in other areas, including Australia, cannot be definitively excluded.
The ecology and mode of transmission of micro-organisms may differ between geographic locations, with biological diversity affecting bacterial adaptation and human activities. This concept could be applied to M. leprae, a mycobacterium that also causes a dermatosis. Indeed, a recent study has suggested that the ecological features, reservoirs and transmission routes of M. leprae may differ between continents. It has been shown that, in North America, wild armadillos harbor the same strain of M. leprae as leprosy patients. Leprosy may thus be a zoonosis in this region [54]. This situation cannot be transposed to other continents in which leprosy is highly endemic such as Africa and Asia, where there are no armadillos and no other mammal is known to harbor the bacillus. A similar situation may apply to M. ulcerans. In Australia, mammals such as possums have been shown to be hosts of M. ulcerans and may play a key role in its dissemination, together with mosquitoes. However, there are no possums in Africa, and M. ulcerans has never been detected in the tissues of any mammal other than humans in Africa.
Based on the results of various studies performed in recent decades and aiming to decipher the ecological characteristics of M. ulcerans, it seems likely that M. ulcerans can be transmitted via several routes, potentially differing between locations in different parts of the world.
|
10.1371/journal.pbio.1002260 | BOLD Response Selective to Flow-Motion in Very Young Infants | In adults, motion perception is mediated by an extensive network of occipital, parietal, temporal, and insular cortical areas. Little is known about the neural substrate of visual motion in infants, although behavioural studies suggest that motion perception is rudimentary at birth and matures steadily over the first few years. Here, by measuring Blood Oxygenated Level Dependent (BOLD) responses to flow versus random-motion stimuli, we demonstrate that the major cortical areas serving motion processing in adults are operative by 7 wk of age. Resting-state correlations demonstrate adult-like functional connectivity between the motion-selective associative areas, but not between primary cortex and temporo-occipital and posterior-insular cortices. Taken together, the results suggest that the development of motion perception may be limited by slow maturation of the subcortical input and of the cortico-cortical connections. In addition they support the existence of independent input to primary (V1) and temporo-occipital (V5/MT+) cortices very early in life.
| While it is known that the visual brain is immature at birth, there is little firm information about the developmental timeline of the visual system in humans. Despite this, it is commonly assumed that the cortex matures slowly, with primary visual areas developing first, followed by higher associative regions. Here we use fMRI in very young infants to show that this isn’t the case. Adults are highly sensitive to moving objects, and to the spurious flow projected on their retinas while they move in the environment. Flow perception is mediated by an extensive network of areas involving primary and associative visual areas, but also vestibular associative cortices that mediate the perception of body motion (vection). Our data demonstrate that this complex network of higher associative areas is established and well developed by 7 wk of age, including the vestibular associative cortex. Interestingly, the maturation of the primary visual cortex lags behind the higher associative cortex; this suggests the existence of independent cortical inputs to the primary and the associative cortex at this stage of development, explaining why infants do not yet perceive motion with the same sensitivity as adults.
| The infant visual brain is immature at birth. To date there is no direct functional evidence from awake infants showing how the various cortical areas of human visual cortex develop. The available evidence suggests that infants are capable of discriminating motion-direction soon after birth [1–3], and that sensitivity to global-motion continues to mature slowly over the first 4–7 y in humans and 2–3 y in monkey [4–7]. The protracted development (after an early emergence) of global-motion sensitivity is attributed to late maturation of higher-level motion areas, such as temporo-occipital complex (V5/MT+) [2,8–12].
Previously it was hypothesized that the visual cortex develops in a hierarchical fashion with higher-order areas developing later, driven by feed-forward projections from previously developed lower-order cortical areas [9,13]. There is also evidence that some aspects of motion processing, such as the control of opto-kinetic eye movements (OKN), are limited by the development of subcortical relative to cortical function [14]. Brisk monocular OKN responses can be elicited in newborn infants, but only by motion in the temporal-to-nasal direction. Early development of subcortical mechanisms (probably of the Nucleus of the Optical Tectum) may mediate the eye-following response in this direction, while the directional sensitivity in the nasal-to-temporal direction, emerging later at about 10 wk, may be mediated by cortical mechanisms [11,15]. Even more peripheral factors, such as photoreceptor efficiency or myelination, may provide important constraints on some functional development. For example [16–18] chromatic and achromatic contrast sensitivities are limited by retinal immaturities, rather than by immaturity of cortical processing. This, together with evidence from infants who suffered from visual deprivation [19], suggests that the visual cortex may mediate vision very early after birth, provided that the incoming input is mature and transmits reliable visual information.
The subcortical input to associative visual cortex can undergo strong reorganization during development. In adult monkey and human, V5/MT+ input originates mainly from cortico-cortical connections and from independent konio-cellular LGN-Pulvinar projections that bypass V1 [20–22]. However, in the first few post-natal weeks, the major inputs to MT+ in marmoset monkey are a disynaptic connection from the Retino-Pulvinar projections (from the medial portion of the Inferior Pulvinar) [23]. This input may mediate the directional response of MT+ neurons observed very early postnatally. The present study examines whether the cortical mechanisms of motion processing are functional in very early infancy, and in particular aims to compare the selectivity of MT+ and primary visual cortex (V1) to flow motion to highlight a possible differential development.
To date there is no direct evidence about the functional development of the various cortical areas of human visual cortex or of their Blood Oxygenated Level Dependent (BOLD) response selectivity from awake infants, although a few studies have shown that it is feasible to record BOLD acoustic responses [24–26], or BOLD flash responses during deep anaesthesia or sleep [27–31]. Reliable BOLD activation to flashes in calcarine sulcus has been reported, but many studies found a reduction of the BOLD response to visual stimulation respect to no stimulation, particularly evident after the eighth week of age [28,30]. The origin of this negative, or more generally delayed BOLD response, is still under debate, and the issue is further complicated by the effect of the use of anaesthesia or sedation that have been shown to modulate BOLD response in human and animal models. In awake infants, studies using Near Infrared Spectroscopy (NIRS) [13,32–36], which measures signals of similar origin to the MRI-BOLD response, show evidence for positive responses in both young (around 8 wk of age) and older (4–6 mo of age) infants to visual stimulation. The discrepancy between MRI- and NIRS-BOLD responses is still unresolved and may reflect either the state of sedation, the sleepiness of infant or different types of visual stimulation. Interestingly, at 6 mo NIRS BOLD amplitude is reduced in response to homogenous flashes and increases in response to a structured high contrast pattern [34], suggesting that the type of visual stimulation matters.
Here, by measuring BOLD responses to coherent versus random flow-motion stimuli in cooperative infants, we demonstrate that the major circuits mediating motion perception are operative very early, by 7 wk of age. We found a selective response to coherent flow-motion in the temporo-occipital area, cuneus, posterior parietal and posterior associative insular cortex, with similar activation and localization as adults for visual motion [37–39] and vection perception [39–44]. Previous works have shown that resting-state connectivity networks are well segregated in newborn and even in pre-term infants [45–48] and mature rapidly in the first 2 y of age [48–50]. Here we demonstrate adult-like functional connectivity between many motion selective associative areas, but not between primary visual cortex and temporo-occipital (putative MT+) and posterior-insular cortices. The results localize for the first time cortical areas with high selectivity to visual stimuli in the first weeks of life, revealing an unexpected early maturation of the cortical system for motion processing.
We presented random-dot coherent flow-motion patterns, which cycled through radial, spiral, and contraction trajectories [37] at optimal conditions for infant vision: low temporal frequency, large dot-size, and high contrast (examples of the stimuli and the fixation of an infant are shown in the S1 Movie). We first located V1 using high contrast flow-motion versus blank, which elicited very strong responses in the occipital pole and in particular along the calcarine sulcus in both hemispheres of all infants, with a positive BOLD hemodynamic (see Table 1 and the single subject amplitude and average BOLD response in S1 Fig). This result contrasts with previous reports of negative BOLD activation to flashes in calcarine sulcus observed in sleeping or anaesthetized [27,28] infants, but is in agreement with those of Morita [30] who reported negative BOLD only in infants older than 60 d of age and with several NIRS studies [13,32–34] performed at similar ages.
Using this calcarine activation (labelled V1 seed), we generated a mask comprising all voxels that correlated positively or negatively with the labelled V1 activity (S2 Fig). To allow for variability in the hemodynamic time-constant demonstrated in previous studies [24,25,27,28,51], we computed the correlation using the delays between 0 and 3 s, after verifying that using additional delays would not change significantly the final labelled regions of interest (ROIs) (see Methods for details). Within this functional mask, we studied significant response modulation (positive and negative) to alternation of coherent- against random-motion stimuli, constructed to match locally the motion velocities of the coherent flow stimulus. The motion difference between the two stimuli is very subtle, but in adults it is able to elicit a consistent BOLD activity in a network of areas [37–39], similar to that we observed in infants. Fig 1A and 1B shows examples of the responses of two infants (CT and MG, p < 0.01 uncorrected) for a temporo-occipital cortical area that strongly preferred coherent to random flow motion, both in the right (light orange) and left (dark orange) hemisphere.
Both the anatomical localization (close to the inferior-temporal sulcus as assessed by an expert neonatal neuro-radiologist), and the time-course of the infant responses, were very similar to those of adults using similar statistical thresholds and procedure (see Fig 1A and 1B, subject MCM, p < 0.01). Table 1 reports the location and extension of these areas for all subjects. The adult V5/MT+ was clearly labelled in a location similar to previous studies [37]. Despite the great difference in atlas and morphometry between adult and infant brains, the coordinates of these areas were highly consistent between the two groups of subjects [37]. Given the similarity in anatomical localization (for congruency between localization in infants and in adults see multi-subject section) and the response selectivity to coherent flow-motion between adult and infants, we labelled this area as the putative infant MT complex (MT+). In all infants, the responses of MT+ were reliable at a threshold of p < 0.01 in 22 out of 24 ROIs. For only two MT+ ROIs, the statistical threshold had to be reduced to p < 0.05 before activity of similar cluster-size could be identified at the expected anatomical location. An independent measure of reliability is given by the Signal to Noise (S/N) ratio, which measures the relative power of a modulated signal at the fundamental frequency of the stimulus repetition respect to close-by frequencies. Scalar S/N values were similar to those measured in adults (Fig 1, compare C with D). Also the phases of the responses, that in principle could vary uniformly in the 0–360° range, were tightly clustered in infants and adults and the vectorial statistical analysis between the S/N response of infants and adults (Left and Right MT+ pooled together) showed that the BOLD responses were not significantly different (See S1 Table). Also the differences between Left and Right hemisphere BOLD responses averaged across subjects (see Fig 1C and 1D top panels) were not significant at either age. However, the variability across infants is somewhat larger than between adults, but this is to be expected given the higher level of measurement errors, including the size of stimulated visual field, eye movements and age range.
The network of regions analysing flow in adults is very extensive, comprising (besides V5/MT+) many dorsal associative visual areas (V3, V3A&B, TOS, LO) [37,52], as well as multimodal cortices such as V6 [40,41], Pre-Cuneus, cingulate sulcus and an associative vestibular cortex located in the posterior insular cortex, PIVC/PIC [39,41,43,44]. Responses to coherent versus random stimuli were observed in all adults, both in the dorsal visual and the associative multisensory cortices of this network (Fig 2 reports the anatomical location of the labelled regions in adult CS and Table 1 the Talairach coordinates of their foci, with the peak Z-score and the size of labelled ROIs at p < 0.05). Nearly all these areas, particularly V6 and the Cuneus/Pre-Cuneus areas, showed a strong positive preference to flow-motion, while the vestibular cortex PIVC/PIC and V1-seed (the ROI selected in response to coherent motion versus blank) showed a positive preference for random motion, consistent with previous reports in adults [40,52]. Not only are the anatomical locations of these areas very similar in adults and infants (see Table 1), but so also are the BOLD modulation (Fig 3A), the reliability and response delay (S/N and phase respectively, Fig 3B). Only minor differences emerge. The V1-seed ROI, selected by the response to flow-motion against blank, showed a significant negative response in adults to coherent versus random flow-motion (two tailed t-test, t = 2.70, p < 0.02), but not significantly different from zero in infants (two tailed t-test, t = 0.65 p = 0.47), indicating possible delayed development. While it was not possible to compare BOLD amplitude of the ROIs between infants and adults, given that the ROIs were defined using the same dataset (statistical circularity), it was possible to compare delays. The responses of V6 (ɸ(infant) = 82 ± 40 °; ɸ(adult) = 55 ± 31 °) and PIVC/PIC (ɸ(infant) = -146 ± 35 °; ɸ(adult) = -117 ± 38 °) regions had a slightly larger hemodynamic delay indicating a delayed BOLD hemodynamics in infants, in agreement with previous results in different cortical area [25,51] (S1 Table reports the significance tests calculating by vectorial statistics). Interestingly the infant delays with respect to adult delays vary between areas and between stimuli. For example, the response of the MT+ and V6 ROIs to flow versus blank is negative in some infants (see S3 Fig), suggesting a specificity of the Negative BOLD effect for the visual stimulus [34].
Besides the ROIs reported in Table 1, many other occipital areas were labelled in the individual infants. We did not analyse in detail these associative occipital cortices (yellow in Fig 2), given that these regions often blurred with each other (because of their close proximity in the small infant brain) and are very difficult to classify by anatomical landmarks.
The flow-selective area in the Cuneus/Pre-Cuneus was labelled in only seven infants, while it was always present in all except one adult; in another subject, the V6 and PIVC/PIC ROIs were not significant at α = 0.05. To assess the congruency of ROI localization, we built an anatomical template by brain segmentation using dedicated tissue probability maps (illustrated in S4 Fig) and we performed a fixed-effect GLM multi-subject analysis (see Methods). Fig 4 (top panels) shows the results mapped on the same anatomical template of S4 Fig for infants and in Talairach atlas for adults, at the same statistical threshold of q < 0.05 (false discovery rate [FDR] corrected and without mask). In adults, the occipital foci associated both with dorsal and with ventral pathways [53,54] showed a stronger response to coherent motion, with the exception of V1/V2 areas. A negative response was also labelled in the posterior insular cortex of the left hemisphere (slice at z = 8), while the right PIVC/PIC became labelled lowering the threshold at p < 0.05 (lower panels). This result is consistent both with the large variation in phase of the PIVC/PIC responses (see Fig 3 and S3 Fig) and with the large variation in the localization between subjects (see Table 1). The variation in phase was to be expected given that the BOLD amplitude of PIVC/PIC is modulated by the strength in vection perception [39–43]: subjects with stronger vection illusion might show stronger negative responses. Similarly, also the large scatter in position of the PIVC/PIC region has been already reported [43]. Both factors indicate that the multi-subject GLM may not be the most suitable technique to locate this area. Nevertheless it is reassuring that when decreasing the threshold to p < 0.05 (bottom panels) only one clear additional cluster in the posterior insular cortex became labelled, reinforcing the suggestion that PIVC/PIC localization can vary considerably across subjects. In infant multi-subject GLM, no activity was significantly labelled for V1/V2; all the ventral areas along the fusiform and the lingual gyri had negative responses, while the adult responses were positive, indicating a late development of direction selectivity for the ventral pathways. In contrast, the dorsal areas, like V3/V3A, LO, TOS, V6/V6A, and MT+, were clearly labelled, and showed a preference for the coherent flow motion. In the right hemisphere (slice at ζ = 16) a negative response corresponding to the location of the PIVC/PIC was labelled. As in adults, decreasing the threshold at p < 0.05 (bottom panels) also the PIVC/PIC in the left hemisphere became detectable, suggesting that, as in adults, for infants, the localization variability of this area is high. There were also other negative and positive clusters in the temporal lobe (see for example slice at ζ = 0) that can be also observed in adults, but whose function and circuitry are still unknown. The similarity in the localization pattern between infants and adults strongly suggests that the development of visual primary and dorsal associative cortex for motion processing is quite well advanced by 7 wk of age.
In nine out of twelve infants, we were also able to record resting-state activity during spontaneous sleep. Fig 5 shows the correlation of the resting-state activity between each pair of the ROIs of Fig 2, for all possible combinations in infants and adults. In most regions of infants and adults the correlation was strong and significantly non-zero (indicated by stars in the individual cells in Fig 5). Significant correlations reflecting functional connectivity were present in both adults and infants between homologue areas of the two hemispheres, including PIVC/PIC, V6 and MT+. The correlation matrixes were similar in adults and infants, corroborating the similarity of the BOLD responses with two main exceptions. In infants, the correlation between bilateral V1 activity and MT+ was significant (although just marginally for the right hemisphere), but it was positive, while in adults the correlation was strong and negative, as previously reported [55,56]. Similarly, the PIVC/PIC activity in infants showed a significant positive correlation, while in adults it was negative.
Our results show that visual motion cortical areas have a high degree of functional specialization in very young infants: direction selectivity is well established in many dorsal cortical regions at 7 wk of age. To distinguish between coherent and random flow-motion, the neuronal mechanisms must be selective to motion direction and also integrate local-motion signals along complex trajectories. Direction selectivity in adult primates occurs initially in V1, while large spatial integration along complex trajectories occurs in V6 and V5/MT+, and propagates to higher visual cortices [42]. Given that 1-mo-olds (or younger) infants show defensive motor responses such as blinking and avoidance head movements in response to large-field radial expansion patterns [1,3], and that motion direction selectivity has been demonstrated behaviourally [2], it is reasonable to assume that selectivity for motion direction emerges around 7 wk of age or earlier in many regions of the dorsal pathways, including MT+ and V6, both crucially important for motion perception. This suggests that cortical processing of motion is more mature than has been proposed at this age [8,11,15], with a relatively stronger response of MT+ compared with occipital areas in infants. Interestingly, V5/MT+ is situated ventro-laterally, just posterior to the meeting point of the ascending limb of the inferior temporal sulcus and the lateral occipital sulcus, and corresponds, on individual subjects, almost precisely with Flechsig's Field 16 [57], one of the areas that is myelinated at birth, corroborating the suggestion that V5/MT+ should be considered as a primary sensory area with an early maturation [58]. An earlier maturation of the V5/MT+ is also consistent with ERP evidence at 5 mo of age of a more lateralized selectivity to coherent motion in infants with respect to adults [59], indicating that the delayed maturation of more occipital areas (with respect to V5/MT+) is protracted over several months. Considering that in adults there is evidence that direction selectivity is computed in MT+ rather than being inherited from inputs of earlier cortical regions [22,60], it is feasible that the MT+ directional selectivity BOLD response at 7 wk of age is also computed locally, and it is independent from V1 input. The neuronal processing of the ventral and dorsal visual pathways are quite distinct and mediate different visual functions [53,54]. Ventral area, like V4 and fusiform gyrus, showed preference for coherent motion in adult, although these areas are not considered crucial for the perception of motion direction and probably inherit the direction selectivity from dorsal areas. Interestingly in infants, these areas prefer incoherent motion as illustrated in Fig 4. These results suggest that the early development of cortical specialization might be a prerogative of the dorsal pathway and visual motion analysis, corroborating the evidence of a later maturation of the ventral stream in children [61] sub-serving object and face recognition [62,63].
Overall, our data suggest that both primary and dorsal associative cortices develop direction selectivity at a similar stage. The motion selectivity of V6-, Pre-Cuneus and associative vestibular cortex has been related to the illusory percept of body-motion induced by large flow field (vection) [39–41,43]. The similar activity of these regions in infants and adults suggests that infants may have a sense of vection and hence a sense of body position. The continuous stimulation of the semi-circular canals in the womb may induce early development of the vestibular system, so the few weeks of visual experience may be sufficient to endorse the vestibular-visual integration necessary for vection perception. While this is consistent with early development of the vestibulo-ocular response in newborn infants [64], it is surprising that visuo-vestibular cortex is selective to motion so early, as optimal multisensory integration develops slowly in humans [65–67].
Our data show that the hemodynamic response of infants is delayed with respect to adults, similar to that reported at similar ages in response to acoustic [25] and motor/tactile stimulation [51]. The delay of the response is of the order of few seconds, consistent with the delay observed by Arichi [51], and cannot explain the phenomenon of Negative BOLD corresponding to 180 degree of phase shift and delay on the order of tens of seconds. It is worthwhile noting that other laboratories have observed positive BOLD visual responses in this age window, and negative BOLD responses for older infant [30], reinforcing the suggestion that the Negative BOLD phenomenon is related to the functional maturation of the cortex. However, the fact that NIRS studies in awake infants consistently find a positive BOLD responses at all ages [13,32–36] suggests that the MRI-BOLD negative response may not result from synaptogenesis or the cerebral metabolic rate for oxygen, but rather from the alertness state of the infant. In addition, the positive versus negative BOLD response could be biased by the use of flash stimuli versus black background, rather than a spatial pattern whose contrast is modulated over time. This is supported by the finding that the phase of V6 and MT+ became negative in response to the flow versus blank (S3 Fig), and blank stimuli engage usually little attention in infants. Interestingly, a recent NIRS study showed a negative BOLD response to flashes and a positive BOLD response to high contrast flickering patters in 6-mo-old infants [34]. It is well known that flash stimuli are strongly subject to suppression during blinks, and the suppression may be even stronger when the eyes are shut or during sleep. Repeating our study with older collaborative and awake infants could provide important clues to resolve the origin of negative MRI-BOLD response in infants, although the difficulty in recording stable signal increases rapidly with age because of the increased motility of older infants.
During post-natal development, there is a continuous refinement of anatomical connections in the mammalian visual system, which are initially diffuse and then progressively pruned to increase target selectivity. Between 0 and 4 mo, synaptic density increases rapidly [68,69], and myelination intensifies along the visual pathway. Diffusion Tensor Imaging (DTI) studies suggest that at birth all major fiber systems are in place, despite low anisotropy value [70] and incomplete myelination [71]. The functional connectivity observed in our data showed an adult-like correlation between inter-hemispheric ROIs and between associative cortices, in agreement with the anatomical DTI findings of well-established fiber connections. However, connectivity between V1 and MT+ and PIVC/PIC were different than in adults. In infants, the optic radiations are diffuse (see for example Fig 3 in [72]) and may project not only to V1 but also to other visual cortices. In addition, other transient projections not originating from LGN may be functional in young infants. These diffuse afferent projections to the cortex might mediate the strong BOLD response observed here in all areas, without implicating adult-like cortico-cortical connections, which at this age are highly immature [69]. This model would also explain the weak (although significant) correlation between MT+ and V1, given that these areas would receive independent V1 inputs to generate the strong BOLD response selective to motion. It would also reinforce the suggestion that, in infants, BOLD directional selectivity is computed in MT+ rather than being inherited from inputs of earlier cortical regions. However, some caution is required in comparing adult and infant functional connectivity given that the infant, but not the adult, data were acquired during sleep. Several studies show that functional connectivity during sleep and relaxation [73] may be different, but the differences are not significant for the sensory-visual resting state networks [74]. Resting-state functional connectivity appears to be invariant even under anaesthesia [75].Therefore, we do not foresee that different depths of state of alertness would affect the outcome of our studies. In addition, most of the functional connectivity strengths are equal in infants and adults, in agreement with previous results [48]: only V1-MT+ was very different. It would be rather strange for sleep to affect so selectively only this connectivity.
As in human, in cat and monkey, the putative corresponding area of human MT+ is well developed at birth, with a developmental time-course similar to primary visual cortex [57,58]. V5/MT+ neurons of newborn monkey show directional selectivity [8]. Interestingly, in monkey it has been suggested that the fast maturation of V5/MT+ is mediated by strong and direct retino-pulvinar-cortical projections, which are later pruned during development [23]. In adult humans the LGN-MT+ projections, which bypass V1, probably overtake the functional role of this direct retinal-pulvinar input [20–22], helping to explain several motion abilities retained after lesion of V1 (such as “blindsight”). The existence of a direct retino-pulvinar input to MT+ also in human infants would explain the weak functional connectivity between MT+ and V1. However, other explanations cannot be excluded, including delayed development of feedback projections, which may have an overall suppressive effect, explaining the negative correlation in adults. The suppression of V1 activity by feedback connections would also be consistent with the negative BOLD response to motion (against noise) of V1 in adults, and the smaller negative (not significant) modulation in infants, suggesting an overall delayed development of V1 circuitry respect to MT+.
Whatever the anatomical connections between the various areas in infants, the extensive, well-developed network of visual associative areas selective to motion at 7 wk, demonstrated by this study, suggests that direction selectivity, the most fundamental property for motion perception, emerges very rapidly, and simultaneously in many dorsal cortices. This developmental pattern may be consistent with the early suggestion [76] that direction selectivity development is limited by peripheral factors such as myelination and speed of neuronal signal transmission. Timing of arrival of spike input to cortex is crucial for motion perception. Jitter in temporal delay of the order of milliseconds is sufficient to disrupt or even invert motion [77]; disorganization of input spike trains or high levels of noise may be sufficient to impede the formation of direction selectivity at the cortical level. As input motion signals become more reliable and organized, the cortex may be capable of developing the complex neuronal circuitry for direction selectivity and trajectory integration only within a very limited time window. Similar fast cortical development has been observed previously for stereo acuity or chromatic contrast (for review see [76]), and also for these functions, the limiting factors may be peripheral. If the present suggestion of independent emergence of direction selectivity in MT+ is confirmed by additional evidence, we should revise the widely accepted idea of a slow, uniform, and progressive maturation of the various cortical hierarchy levels [9] and of cortical motion responses [59]. We propose an alternative model in which all dorsal cortical regions have equal potential for fast maturation and for developing appropriate circuitry, once the input begins to transmit reliable neural information.
The study was approved by Ethics Review Board of Fondazione Stella Maris and Regional Paediatric Committee (Meyer Paediatric Hospital). Written informed consent was obtained by all subjects or their parents before the experiment.
Sixteen (five females and eleven males) healthy, full-term, awake infants, mean age 7.7 ± 1.2 wk (range = 6.6–10.4w), were scanned by a 1.5T MR scanner (GE Healthcare, USA). Data from three subjects were not included in the analysis because of movement artifacts or deep sleep, and for one subject we obtained only the anatomical dataset, leaving twelve subjects with full data. All infants were assessed by an expert paediatric neurologist by means of the Hammersmith Neurological exam [78] and a battery of visual function tests, including fixation to white and black targets and the ocular following response. The infants were again assessed at 3 mo with a neurological follow-up that confirmed normal development. After the MR exam and before any data analysis, an expert child neuro-radiologist performed also an examination of the anatomical scans to reveal possible anomalies and all infant brains were referred as normal.
The MR protocol comprised acquisition of a three-dimensional (3-D) T1w FSPGR sequence (TR/TE = 12.28/5.14, isotropic voxel = 1x1x1mm3), and an fMRI session of three different series (GRE-EPI, TR/TE = 3000/50, FA = 90°, FOV = 240 X 240 mm, matrix = 96 X 96, slice thickness = 3 mm). fMRI experiments included two series of 84 time points (4’12” duration, block design, six periods of alternating coherent flow motion versus a blank or a random motion condition, each lasting 21 s). In nine subjects (age range = 6.6–8 wk) a resting state fMRI series (120 time points, 6’00” duration, no stimulus presentation) was successfully acquired during the spontaneous sleep. All functional series were preceded by four dummy time points to allow signal stabilization. Stimuli were generated in Matlab (TheMathWorks) and displayed on LCD goggles (Resonance Technology) positioned inside the head coil 10 to 15 cm from the infant eye, giving a visual field of about 27 X 20 degrees.
Infant eyes were refracted with retinoscopy at a distance of 87 cm (the virtual image of the goggles is greater than 1 m). All infants were in the normal range between 0 and +2D: given the viewing distance, no additional correction was introduced. Fixation of infant gaze was monitored by an infrared camera installed within the goggles (sample frequency 60 Hz, see final segments of S1 Movie for fixation example). Usually infants scanned the stimulus very attentively and this, together with the real-time movement signals, gave an indication of their level of wakefulness. Infants entered the magnet lying down on the scanner table, swaddled in a sheet by an expert neonatal nurse to calm the baby and to reduce movement. One operator (MCM, SAC, or LB) accompanied the child into the magnet, taking the sphinx position, surrounding the body of the child with her arms and wrapping his/her head with her hands in order to maintain a physical contact, to facilitate calming and to reduce motion. The operator was in constant communication with the staff at the acquisition terminal. Infant ears were protected by cotton wool padding at the auditus of the external auditory canal, and sound-attenuating headphones. In order to reduce the stress of the babies, we adopted specific strategies in accordance with the character and routines of each individual child. Some preferred to use a pacifier to relax and to sleep, especially for anatomical and resting state fMRI scanning. If the infant moved or changed position the operator could reposition the goggles by tilting them. In these cases, the recording was continued, but the series was cut offline (see details below). Importantly, the operator could not see the visual stimulus delivered, or if the goggles were switched off for the resting state scan. During the anatomical scan, the goggles were switched off and nine of the infants fell asleep, allowing us to acquire the resting-state scan.
The same MR protocol described for children was repeated in nine healthy adults (six females and three males, mean age 35 y). In adult subjects, resting state scans were acquired by asking to the subjects to relax and stay still with eyes closed.
The visual motion stimuli comprised 100 dots, half black and half white, of 1.3 degrees diameter, moving at constant speed (5 deg/s) with limited lifetime of 10 frames (at 60 Hz about 160 ms). For coherent motion the trajectory of the dots changed gradually from expansion, inward-spiral, rotation, outward-spiral, contraction, then repeating the cycle again (see S1 Movie, initial segments). The full cycle lasted 2 s. The random motion was constructed using the same dot velocities shuffled randomly over the dots, so it had matched local motion power. The dots covered all the visual field except the central 2 degrees. Dot density was kept constant in all displays, and collision between dots was not allowed. The mean luminance was 20 cd/m2 and contrast 0.85 (for further information, see [37]).
Three-dimensional T1-weighted images of each infants were visually inspected to select good quality data in order to create a specific infant template for this study. Only one infant’s data (subject DE) was discarded because of the poor grade of anatomical images; all the remaining twelve 3-D datasets were selected and analysed with the SPM8 package with the Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra (DARTEL) algorithm [79]. DARTEL uses diffeomorphic warping to obtain a study-specific template, with an optimal inter-subject realignment and an improved co-registration of small structures, such as those of infant brains. A pre-conditional step for the use of this algorithm is the segmentation of brain tissues using standard tissues probability maps. For infants, we used the segmented partial estimate volumes for grey matter, white matter, and cerebral spinal fluid derived by the 36 MRI dataset of 3-mo-old infants available at http://jerlab.psych.sc.edu/NeurodevelopmentalMRIDatabase/index.html as tissue probability maps [80]. The obtained atlas is shown in S4 Fig.
Data analyses were performed with BrainVoyager (BV, Brain Innovation). First, for each functional series, motion spikes or periods of heavy movement that could not be compensated for were eliminated, and the resulting separate intervals were considered as independent shorter samples. Infant motion was estimated by calculating the six rigid body parameters (three for translation and three for rotation) across time. Time series with head motion greater than 4 mm (translation) or 5° (rotation) were excluded. For each subject and stimulus, the mean and maxima values of the six rigid body parameters registered in the survived (and used) time courses are reported in S2 Table.
The recorded eye movements were analysed to verify fixation and alertness of the infant; intervals of sleep or jerk movements that could not be adequately compensated were discarded from analysis, typically for the duration of half period of stimulation. For all infants, we were able to select at least half duration of the complete recorded run (on average, 4.9 ± 0.8 periods for coherent versus random flow motion, 4.8 ± 0.9 periods for coherent flow motion versus blank, and 110 ± 10 data points for resting state stimulus) for further analysis, with the remaining periods discarded due to head motion or sleep. Data preprocessing included mean intensity adjustment to compensate for interscan intensity differences, temporal interpolation and re-sampled to compensate for slice dependent time differences (sinc function), 3-D motion correction (sinc interpolation), and high-pass temporal filtering (GLM-Fourier approach, two cycles).
Functional data were co-registered on the three-dimensional anatomical T1-weighted images by using an affine alignment with the standard BV nine parameters (three for translation, three for rotation, three for FOV scale). Infant anatomical datasets were in turn transformed into the AC-PC coordinate system by applying a rigid transformation (6 parameters; 3 for translation and 3 for rotation), whilst adult data were transformed into the standard Talairach space. For each subject, BOLD responses were analysed using a GLM modelling the regressor of interest (by convolving a box-car function for each stimulation block with a gamma variate function for the hemodynamic response) and six spurious movement-regressors (outputs of the 3-D motion correction procedure).
The first stimulus (flow-motion versus blank) was used to select a bilateral region with very strong response, located along the left and right calcarine sulci (p < 0.001 in infant and p < 10−10 in adults). Many other strong activities in the occipital pole were present but not analysed. The signal registered in this ROI (labelled V1-seed) was used to calculate correlation maps by cross-correlating the V1-seed ROI signal with all other brain voxels, using temporal delays in the range 0 to 21 s, given the unknown hemodynamic delay of infant BOLD responses. S2 Fig shows the maps of four different infants computed for positive and negative correlation at zero delay and at p < 0.05. The low conservative threshold of p < 0.05 marked about 40% of all voxels for delay = 0 s and additional 26% for delay = 3 s. Computing the maps for the remaining delays in the range of 3 to 21 s added only about 30% new voxels. Repeating the same analysis only on the selected ROIs (see below the selection procedure), we obtained that 80% of the voxels were labelled at delay = 0 s and 94% combining delay 0 s and 3 s. Given the high proportion of labelled voxels, in the analysis presented here we applied a mask obtained by the union of those computed at 0 and 3 s delays.
Both for infants and adults, the GLM analysis of the coherent versus random motion was performed within this mask, and all positive or negative foci at a threshold of p ≤ 0.05 were labelled (see Table 1 for statistical thresholds of the ROIs). These ROIs were located by a neuro-radiologist (with more than 20 y of experience with newborn infants) along the major infant sulci and gyri. In adults, we restricted the analysis to the major regions of interest selective to motion [37–39] that could be easily traced in infants.
Table 1 reported the maximum foci localization in millimetres in the AC-PC coordinate system for infants and in Talairach space for adults, and the respective Z-score for areas that were identified in all adults and in most infants. For multiple foci areas, the coordinates corresponded to the positions of the centre of mass of the aggregate ROIs. In particular, for the occipital activities we reported the union of the bilateral activations of V1 (V1-seed) (reporting the average coordinates of the two foci); an area located dorsally just posterior of the parietal-occipital sulcus, which in adults correspond to the union of V6 and V6A; an area positioned posteriorly in Pre-Cuneus/Cuneus, close to the parieto-occipital sulcus and anterior of the most extreme periphery of V1/V2. This area could be the human equivalent of the pro-striate [81], not to be confused with the Pre-Cuneus selective motion area reported by Cardin et al [39,82]. Many other occipital activation sites were labelled in response to coherent versus flow motion in the mask, but given the difficulties in distinguishing between them and locating them in all the sulci, we decided not to report or analyse them in detail. The area MT+ was the easiest to locate along the inferior temporal sulcus, both in adults and infants. In adults, this ROI includes MT and MST. An area showing a differential response to coherent versus incoherent motion was labelled in the fundus of the most apical portion of the posterior insula, corresponding to the PIC+ complex of the visual-vestibular network [39–41,43,44]. The complex comprises two different foci named PIVC and PIC respectively, that can be differentiated on the selectivity to vestibular stimulation. We labelled this area as PIVC/PIC given that we used only visual and not vestibular stimulation.
For each ROI, and for both the first (coherent flow-motion versus blank) and the second stimulus (coherent versus random flow-motion), we extracted the time-courses averaged across periods of repetition, and then across subjects. We also evaluated the Signal to Noise ratio (S/N) and the phase of the response, performing an FFT on the extracted time-course before averaging. S/N was defined as the amplitude of the ratio of the fundamental harmonic and the root-mean-squared amplitude of the two frequencies closest to the fundamental [83]. The phase of the response is a good measure of the hemodynamic delay. In adults, a standard hemodynamic model corresponds to a phase value of 64 deg in our plots. The relationship between the phases of the responses across different areas and single subjects is reported in S3 Fig, both for infants and adults. To evaluate statistical significance between the adult and infant phases of the responses for each ROIs, we calculated the distances of each vector data point from the resultant of the vectorial average and its standard deviation. Then assuming a normal spread, we calculated the variation in phase associated with the predicted 2-D Normal dispersion of data around the average. Results of average, SD and t-tests are reported in S1 Table.
In order to consider the infants as a homogenous group, inter-subjects alignment was performed. In particular, the obtained atlas and the volumetric dataset of each subject were co-registered (by affine transformation) to one single subject (MG) with optimal quality of the anatomical image. The same transformations, one for each subject, were in turn applied to the respective functional data. One subject (DE) was discarded from this kind of analysis for poor quality of three-dimensional anatomical dataset. The 11 co-registered functional datasets were used for a multi-subjects analysis of the infants group, by using a fixed–effect (FFX) GLM-based analysis and a statistical threshold corrected for False Discovery Rate (FDR) q < 0.05 and a minimum cluster size of 81 mm3, corresponding to three functional voxels (Fig 5, left panel). Similarly a fixed–effect (FFX) GLM-based analysis at the same statistical threshold of q(FDR) < 0.05 and a minimum cluster size of 135 mm3 was performed also on adult data, after co-registration in Talairach space (Fig 5, right panel). In both infant and adult multi-subjects GLM analysis, no mask was applied. In agreement with previous work that show a great variability of the PIC+ complex location and selectivity between subjects [43], the statistical threshold of the multi-subject GLM had to be reduced to p < 0.05 uncorrected before activity of similar cluster-size could be identified at the expected PIVC/PIC anatomical location in infants and in adults.
Resting state series were used to study the correlation between the ROIs described above. For each subject and for each pair of ROIs, the cross correlation of the signals extracted in the resting state was calculated and the value reported in the correlation matrix. Note that given the high-pass temporal filtering the functional connectivity maps could also show a negative correlation. The striking negative correlation between V1 and MT+ in adults that is often reported also depends on the state of the subject (fixation versus eyes-closed) [56]. Note that the resting state in infants, but not in adults, was acquired during sleep. Two mean correlation matrices were obtained averaging the single-subject matrices for infants and adults, respectively. In order to achieve a value of significance for each value in the mean correlation matrices, for each group of subjects, the signals of all single subjects were normalized and concatenated, obtaining the signal of a super-subject for each ROI. The calculation of cross correlation of these signals between each pair of ROIs corresponds to the average across subjects of the correlation matrices and gives the significance of the values in terms of p-value. Statistical analysis (two-tailed t-test) was performed between correlation matrices of the two groups of subjects.
|
10.1371/journal.pcbi.1003649 | An HMM-Based Comparative Genomic Framework for Detecting Introgression in Eukaryotes | One outcome of interspecific hybridization and subsequent effects of evolutionary forces is introgression, which is the integration of genetic material from one species into the genome of an individual in another species. The evolution of several groups of eukaryotic species has involved hybridization, and cases of adaptation through introgression have been already established. In this work, we report on PhyloNet-HMM—a new comparative genomic framework for detecting introgression in genomes. PhyloNet-HMM combines phylogenetic networks with hidden Markov models (HMMs) to simultaneously capture the (potentially reticulate) evolutionary history of the genomes and dependencies within genomes. A novel aspect of our work is that it also accounts for incomplete lineage sorting and dependence across loci. Application of our model to variation data from chromosome 7 in the mouse (Mus musculus domesticus) genome detected a recently reported adaptive introgression event involving the rodent poison resistance gene Vkorc1, in addition to other newly detected introgressed genomic regions. Based on our analysis, it is estimated that about 9% of all sites within chromosome 7 are of introgressive origin (these cover about 13 Mbp of chromosome 7, and over 300 genes). Further, our model detected no introgression in a negative control data set. We also found that our model accurately detected introgression and other evolutionary processes from synthetic data sets simulated under the coalescent model with recombination, isolation, and migration. Our work provides a powerful framework for systematic analysis of introgression while simultaneously accounting for dependence across sites, point mutations, recombination, and ancestral polymorphism.
| Hybridization is the mating between individuals from two different species. While hybridization introduces genetic material into a host genome, this genetic material may be transient and is purged from the population within a few generations after hybridization. However, in other cases, the introduced genetic material persists in the population—a process known as introgression—and can have significant evolutionary implications. In this paper, we introduce a novel method for detecting introgression in genomes using a comparative genomic approach. The method scans multiple aligned genomes for signatures of introgression by incorporating phylogenetic networks and hidden Markov models. The method allows for teasing apart true signatures of introgression from spurious ones that arise due to population effects and resemble those of introgression. Using the new method, we analyzed two sets of variation data from chromosome 7 in mouse genomes. The method detected previously reported introgressed regions as well as new ones in one of the data sets. In the other data set, which was selected as a negative control, the method detected no introgression. Furthermore, our method accurately detected introgression in simulated evolutionary scenarios and accurately inferred related population genetic quantities. Our method enables systematic comparative analyses of genomes where introgression is suspected, and can work with genome-wide data.
| Hybridization is the mating between species that can result in the transient or permanent transfer of genetic variants from one species to another. The latter outcome is referred to as introgression. Mallet [1] recently estimated that "at least 25% of plant species and 10% of animal species, mostly the youngest species, are involved in hybridization and potential introgression with other species." Introgression can be neutral and go unnoticed in terms of phenotypes but can also be adaptive and affect phenotypes. Recent examples of adaptation through hybridization include resistance to rodenticides in mice [2] and mimicry in butterflies [3]. Detecting regions with signatures of introgression in eukaryotic genomes is of great interest, given the consequences of introgression in evolutionary biology, speciation, biodiversity, and conservation [1]. With the increasing availability of genomic data, it is imperative to develop techniques that detect genomic regions of introgressive descent.
Let us consider an evolutionary scenario where two speciation events result in three extant species A, B, and C, with A and B sharing a most recent common ancestor. Further, some time after the splitting of A and B, hybridization occurs between B and C (that is, sexual reproduction of individuals from these two species). This scenario is depicted by the phylogenetic network in Fig. 1. Immediately upon hybridization, approximately half of the hybrid individual's genome comes from an individual in species B, whereas the remainder comes from an individual in species C. However, in homoploid hybridization, where the hybrid offspring has the same ploidy level as the two parental species, hybridization is often followed by back-crossing (further mating between the hybrid population and either of the two parental populations). Repeated back-crossing, followed by the effects of genetic drift and natural selection, results in genomes in the hybrid individuals that are mosaics of genomic material from the two parental species, yet not necessarily with a 50–50 composition. Thus, detecting introgressed regions requires scanning across the genome and looking for signals of introgression.
In a comparative framework, detecting introgressed regions can be achieved by evolutionary analysis of genomes from the parental species, as well as genomes from introgressed individuals. In such an analysis, a walk across the genomes is taken, and local genealogies are inspected; incongruence between two local genealogies can be taken as a signal of introgression [4]. (Here, we focus on topological incongruence; see [5] for a related discussion on local variation of coalescence times.) However, in reality, the analysis is more involved than this, owing to potentially confounding signal produced by several factors, a major one of which is incomplete lineage sorting (ILS). As recombination breaks linkage across loci in the genome, the result is that independent loci might have different genealogies by chance, which is known as ILS. ILS is common to several groups of eukaryotic taxa where species diverged with insufficient time for all genomic loci to completely sort, resulting in a scenario where introgression and ILS effects need to be distinguished [3], [6]–. Fig. 1 illustrates this issue, where local genealogies across recombination breakpoints differ due to ILS, but also differ inside vs. outside introgressed regions. While other factors, such as gene duplication and loss [10], could potentially add to the complexity of the phylogenetic and genomic patterns, we focus here on introgression and ILS.
Recently, new methods were proposed to detect introgression in the presence of ILS. Durand et al.'s statistic allows for a sliding-window analysis of three-taxon data sets, while accounting for introgression and ancestral polymorphism [11]. However, this statistic assumes an infinite-sites model and independence across loci. Yu et al. [5] proposed a new statistical model for the likelihood of a species phylogeny model, given a set of gene genealogies, accounting for both ILS and introgression. However, this model does not work directly from the sequences; rather, it assumes that gene genealogies have been estimated, and computations are based on these estimates. Further, the model assumes independence across loci. Of great relevance to our work here is an array of hidden Markov model (HMM) based techniques that were introduced recently for analyzing genomic data in the presence of recombination and ILS [12]–[14]; however, these methods do not account for introgression. A recent extension [15] was devised to investigate the effects of population structure and migration. Finally, Saguaro is a recent method that combines HMMs with artificial neural networks to annotate genomic regions into different classes based upon local phylogenetic incongruence [16]. The classes are meant to categorize local genealogies, but the method is not aimed at elucidating the cause of incongruence.
In this paper, we devise a novel model based on integrating phylogenetic networks with hidden Markov models (HMMs). The phylogenetic network component of our model captures the relatedness across genomes (including point mutation, recombination, ILS, and introgression), and the HMM component captures dependence across sites and loci within each genome. Using dynamic programming algorithms [17] paired with a multivariate optimization heuristic [18], the model can be trained on genomic data, and allows for the identification of genomic regions of introgressive descent. We applied our model to chromosome 7 genomic variation data from three mouse data sets. Our analysis recovered an introgression event involving the rodenticide resistance gene Vkorc1, which was recently reported in the literature [2]. Based on the analysis, 9% of sites within chromosome 7 are in fact of introgressive origin, which is a novel finding in that previously only a localized region (that included Vkorc1) had been identified, with no further regions scanned. When applied to the negative control data set, our model did not detect any introgression, further attesting to its robustness. Our software is publicly available as part of the open-source PhyloNet distribution [19]. The method and software will enable new analyses of eukaryotic data sets where introgression is suspected, and will further help shed light on the Tree of Life—or, Network of Life.
Let be a set of aligned genomes , and denote the site in the alignment (if we view the alignment as a matrix where the rows are the genomes and the columns are the sites, then is the column in the matrix). Since the genomes are aligned, every has evolved down a local genealogy, and since we assume that hybridization has occurred, each local genealogy has evolved within the branches of a parental tree. This is illustrated in Fig. 2.
It is important to note that for each , any tree could be the local genealogy. That is, if we denote by the set of rooted binary trees on leaves, then for each , it is the case that , for every tree along with its branch lengths . However, the set of parental species trees is always constrained by the actual evolutionary history of species. For example, in Fig. 2, only the two shown trees and are the possible parental species trees. Given a set of aligned genomes, each of length , and a set of parental species trees, we define a set of random variables each of which takes values in the set . We are now in position to define the problem for which we provide a solution:
(1)for every and .
Once this problem is solved and the method is run on a set of aligned genomes, we will be able to deduce the evolutionary history of every site, thus answering questions such as (1) which regions in the genomes are of introgressive descent (these would be the ones whose parental species tree, for the example in Fig. 2, is ; (2) is there recombination within introgressed regions (these would be indicated by switching among local genealogies in a region yet all genealogies evolved within ); and, (3) what is the distribution of lengths of introgressed regions.
Let us consider the scenario of Fig. 2, where only one individual is sampled per species. We propose a hidden Markov model (HMM) for modeling the evolution of the three genomes. The HMM for this simple case would consist of 7 states: a start state , and six additional states: (), corresponding to three possible local genealogies within parental tree , and (), corresponding to three possible local genealogies within parental tree . We denote by and the local genealogies to which states and correspond, respectively; see Fig. 3.
In this model, transition between two states or two states corresponds to switching across recombination breakpoints. The probabilities of such transitions have to do with population parameters (e.g., population size, recombination rates, etc.). Transition from a state to an state indicates entering a introgressed region, while transition from an state to a state indicates exiting an introgressed region. The probabilities of such transitions have to do, in addition, with introgression and evolutionary forces (back-crossing, selection, etc.). Each state emits a triplet of letters that corresponds to a column in the three-genome sequence alignment. The probability of emitting such a triplet can be computed using a standard phylogenetic substitution model [20].
Following the approaches of [12], [21], the transition probabilities in our model do not represent parameters in an explicit evolutionary model of recombination and introgression. Our choice was made to ease analytical representation and to permit tractable computational inference. We contrast our choice with alternative approaches: examples include (in order of increasing tractability of computational inference at the cost of more simplifying assumptions) methods incorporating the coalescent-with-recombination model [22], the sequentially Markovian coalescent-with-recombination model [14] (which adds the single assumption that coalescence cannot occur between two lineages that do not share ancestral genetic material), and the discretized sequentially Markovian coalescent-with-recombination model [23] (which additionally discretizes time).
Assuming that the probability of a site (or locus) in the genome of B being introgressed (in this case, inherited from C) is , we follow the model of [5] and use this parameter to constrain the transition probabilities. Furthermore, we capture topological changes in local genealogies due to recombination using parameters —the probability of switching from a local genealogy congruent with its containing parental tree to one that is incongruent—and —the probability of switching from a gene genealogy incongruent with its containing parental tree to one that is congruent. Finally, we model incomplete lineage sorting by allowing every local genealogy with the probability of observing it given its containing parental tree [24].
For example, assume a site is emitted by state and consider the next site. If the next site is in an introgressed region, the HMM should switch, with probability , to an state. If the next site is not in an introgressed region, then the HMM should stay in the states, with probability , and the next HMM state depends upon whether or not the two sites are separated by a recombination breakpoint that causes a change in local genealogy incongruence (with respect to the containing parental tree ): if they are, then the HMM should switch from state to a different state () with probability ; otherwise, the HMM should stay in state with probability . Thus, the transition probability from to any other () state is and to any () state is , where
is either or depending on whether or not the HMM transition corresponds to a change in local genealogy incongruence, is the probability of genealogy 's topology given the parental tree in , and is the probability of genealogy 's topology given the parental tree . The quantities are computed under the coalescent using the technique of [24].
If we denote by the set of (non-start) states, then a transition from the start state to a state occurs according to the the normalized gene tree probability
For such that and correspond to the same parental tree, let . Furthermore, for , let . Then, the full transition probability matrix, with rows labeled from top to bottom, and similarly for columns (from left to right), is
Given that
and for every pair of indices and , it follows that the entries in each row of the matrix add up to . Further, the HMM always starts in state ; that is the initial state probability distribution is given by for state and for every other state.
Once in a state , the HMM emits an observation , which is a vector in the genomic sequence alignment. Emissions occur according to a substitution model (we used the generalized time-reversible (GTR) model [25]), yielding the emission probability
where are the branch lengths of the gene tree associated with state . (It is straightforward to extend our model to other substitution models, including models nested within the GTR model and the GTR+ model, where is an additional parameter for rate variation across sites.)
Modeling a phylogenetic network in terms of a set of parental trees fails for most cases [26]. For example, if two individuals are sampled from species B in Fig. 1, then one allele of a certain locus in one individual may trace the left parent (to C), while another allele of the same locus but in the other individual may trace the right parent (to A). Neither of the two parental trees in Fig. 3 can capture this case. Similarly, if one individual is sampled per species, but multiple introgression events occur or divergence events follow the introgression, the concept of parental trees collapses [5].
To deal with the general case—where multiple introgressions could occur, multiple individuals could be sampled, and introgressed species might split and diverge (and even hybridize again later) —we propose the following approach that is based on MUL-trees [5].
The basic idea of the method is to convert the phylogenetic network into a MUL-tree and then make use of some existing techniques to complete the computation on instead of on . A MUL-tree [27] is a tree whose leaves are not uniquely labeled by a set of taxa. Therefore, alleles of individuals sampled from one species, say , can map to any of the leaves in the MUL-tree that are labeled by . For network on taxa , we denote by the set of alleles sampled from species (), and by the set of leaves in that are labeled by species . Then an allele mapping is a function such that if , and , then [5]. Fig. 4 shows an example of converting a phylogenetic network into a MUL-tree along with all allele mappings when a single allele is sampled per species. The branch lengths and inheritance probabilities are transferred from the phylogenetic network to the MUL-tree in a straightforward manner (see [5] for details).
Now, two changes to the PhyloNet-HMM given for the simple case above are required. While in the simple case above, we used two classes of states (the and states), in the general case, the PhyloNet-HMM will contain classes of states, where is the number of all possible allele mappings. As above, the transitions within a class of states corresponds to local phylogeny switching due to recombination and ILS, whereas transitioning between classes corresponds to introgression breakpoints. Second, the probability of observing a genealogy's topology given a containing parental tree is now computed using the method of [5], since the methods of [24], [28] are not applicable to MUL-trees.
We used a hill-climbing heuristic to infer model parameters that maximize the likelihood of the model . Here, the model consists of
Notice that the values are completely determined by the parental tree branch lengths and gene tree topology; hence, they are not free parameters in this model.
The standard forward and backward algorithms [17] were used to compute the model likelihood for fixed . We used Brent's method [18] as a univariate optimization heuristic during each iteration of the hill-climbing search heuristic. To reduce the possibility of overfitting during optimization, branch length parameters were optimized for each topologically distinct parental tree, and similarly for each topologically distinct unrooted gene genealogy (since we use a reversible substitution model). States therefore "shared'' branch length parameters based on topological equivalence of parental trees and gene genealogies.
To evaluate the effectiveness of our optimization heuristic, we utilized different starting points for the model inference phase. We found that our heuristics were robust to the choice of starting point since the searches all converged to the same solution (data not shown). We found that the choice of starting point only affected search time.
After model parameter values were inferred, Viterbi's algorithm [17] was used to compute optimal state paths and, thus, annotations of the genomes. More formally, using Viterbi's algorithm, we computed
Further, we used the forward and backward algorithms to conduct posterior decoding and assess confidence for the states on a path :
where is the probability of the observed sequence alignment up to and include column , requiring that (computable with the forward algorithm); is the probability of the last columns ( is the total number of columns in the alignment), requiring that (computable with the backward algorithm); and, is the probability of the alignment (computable with either the forward or backward algorithms).
In the Results section, we show results based on both the optimal path, , as well as posterior decoding, as the latter provides the probabilities in Eq. (1) in the problem formulation above.
To evaluate the performance of PhyloNet-HMM in scenarios where the true history of evolutionary events are known, we simulated data under the coalescent model [29] with recombination, isolation, and migration [22] using ms [30]. The specific model used for our simulation (Fig. 5) is based upon the consensus phylogeny for the species in our empirical study [31], to which we added migration processes. It is important to note that the model differs in one aspect compared to the one in the empirical study: the empirical data sets were sampled so that one Mus musculus sample had a very low chance of being introgressed, whereas both M samples in the simulation may be involved in introgression.
The simulation conditions were based upon consensus estimates from relevant prior literature (summarized in Table 1). We used a divergence time between in-group taxa of 1.5 Mya, generation time of 2 generations per year, and an effective population size of 50,000, which implies divergence time between the M and S populations. The outgroup population split from the ancestral population of A and B at time . We used a cross-over rate , corresponding to cM/Mb (compare with the cM/Mb reported for mice and the cM/Mb reported for humans [32]). We explored multiple migration scenarios hypothesizing either no migration () or migration at one of two different rates ( or ). For scenarios including migration, we utilized two different sets of relatively recent migration times (either between and or between and ) compared to the divergence time between A and B. Finally, substitutions occurred according to , corresponding to substitutions/site/year based on the estimate above (compared with substitutions/site/year reported by [33]).
A simulation condition consisted of a setting for each simulation parameter (in units, as required by ms [30]). For each condition, we repeated simulation to produce twenty replicate datasets per condition. The simulation of an individual dataset proceeded in two steps. First, ms was used to simulate local gene genealogies given the the coalescent model specified by the simulation condition. Then, using seq-gen [34], DNA sequence evolution was simulated on each local genealogy under the Jukes-Cantor model of substitution [35]. Sequences were simulated with total length of 100 kb distributed across the local genealogies.
Our study utilizes six mice that were either newly sampled or from previous publications. Details for the six mice are listed in Table 2.
Newly sampled mice were obtained as part of a tissue sharing agreement between Rice University and Stefan Endepols at Environmental Science, Bayer CropScience AG, D-40789 Monheim, Germany and Dania Richter and Franz-Rainer Matuschka at Division of Pathology, Department of Parasitology, Charité-Universitätsmedizin, D-10117 Berlin, Germany (reviewed and exempted by Rice University IACUC).
The M. m. domesticus data set was constructed as follows. We included a wild M. m. domesticus sample from Spain, part of the sympatry region (i.e., where the species co-occur geographically) between M. m. domesticus and M. spretus. To help maximize genetic differences as part of the design goals of our pipeline, we also selected a "baseline'' M. m. domesticus sample that originated from a region as far from the sympatry region as possible. Thus, we chose a mouse from the country of Georgia in Asia where M. spretus does not occur, and, presumably, M. m. domesticus there are ancestral to those M. m. domesticus that are part of derived populations in Western Europe, including Spain, and that encountered M. spretus during their westward dispersal. We utilized two M. spretus samples. The samples came from different parts of the sympatry region in Spain. The M. m. musculus control data set contained two wild M. m. musculus samples from China and the above two M. spretus samples.
The Mouse Diversity Array was used to obtain the empirical data used in our study [36]. Data for previously published samples were obtained from [31], [37], [38]. Since the probe sets in these studies differed slightly, we used the intersection of the probe sets in our study. A total of 535,988 probes were used.
We genotyped all raw reads using MouseDivGeno version 1.0.4 [38]. We utilized a threshold for genotyping confidence scores of 0.05. We phased all genotypes into haplotypes and imputed bases for missing data using fastPHASE [39]. Less than 15.1% of genotype calls were heterozygous or missing and thus affected by the fastPHASE analysis. The genotyping and phasing analyses were performed with a larger superset of samples. The additional samples consisted of the 362 samples used in [38] that were otherwise not used in our study. After genotyping and phasing was completed, we thereafter used only the samples listed in Table 2 in the Appendix.
Genomic coordinates and annotation in our study were based on the GRCm38.p2 reference genome (GenBank accession GCA_000001635.4). MouseDivGeno also makes use of data from the MGSCv37 reference genome (GenBank accession GCA_000001635.1).
To assess confidence in our method's detection of regions of introgressive origin, we used a modified version of the posterior decoding. In our simulations as well as biological data analyses, there are 15 states corresponding to the "introgressed" parental tree: . As we are interested in assessing confidence in whether a column in the alignment falls within an introgressed region, we computed for column the quantity (2)
We evaluated the performance of PhyloNet-HMM using simulated data sets. Here, we focus on results concerning inferred probabilities (computed using Eq. (2)) on simulations with different migration processes.
In Fig. 6, we plot the percentage of sites for which ( is computed using Eq. (2)) as a function of the migration rate. For the isolation-only model (), the method effectively infers no introgression for any of the sites. For the isolation-with-migration models (), the inferred percentages of introgressed sites were greater than zero and increased as a function of the migration rate . A potentially more informative comparison would be between the inferred percentages of introgressed sites and the percentages of sites in the simulation that involved migrant lineages. However, the simulation software that we used does not support annotating lineages in this way, nor is it a simple task to modify it to achieve this goal. (Furthermore, as noted above, we were unable to exactly simulate evolution under the evolutionary scenario in the empirical study since the simulation software did not permit us to constrain lineage evolution so that one of the samples from population A was not introgressed.)
On the other hand, for all simulated sites, the simulation software outputs the simulated gene genealogy under which the site evolved, along with branch lengths in coalescent units. This output from simulation can be used to obtain lower bounds on the true percentage of introgressed sites. Specifically, if a site evolved under a gene genealogy where one of the two A lineages and any subset of the B lineages are monophyletic and the lineages have a simulated coalescence time greater than and smaller than , then migration must have occurred for those lineages to coalesce in that time span, based on the model used for simulation (Fig. 5). As shown in Fig. 6, for all simulated model conditions, the introgression frequency reported by PhyloNet-HMM is greater than or equal to lower bounds on the true introgression frequency, obtained using this observation.
Clearly, when the duration of the migration period increases, the variation in the estimates of our method increases, which results in a pattern that seemingly does not change from migration rate to . However, it is important to note that the extent of variability in this case precludes making a conclusion on the lack of increase in the percentage of sites. Nonetheless, the important message here is that the estimates of our method start varying more as the duration of the migration period increases.
We also found that the probability of observing a gene genealogy conditional on a containing parental tree differed between the two parental trees (results not shown). Under all simulation conditions, the inferred gene tree distribution (conditional on the containing parental tree) had multiple genealogies with non-trivial posterior decoding probabilities, suggesting that within-row transitions were capturing switching in local genealogies due to ILS. That is, the simulated data sets clearly had evidence of incongruence due to both introgression and ILS.
Finally, Fig. 7 and Fig. 8 show that in training our PhyloNet-HMM model on the simulated data, base frequencies were accurately estimated at 0.25 (which are the base frequencies for all four nucleotides we used in our simulations) and substitution rates were estimated generally between and (we used in our simulations). Further, the results were robust to the migration rates and durations of migration periods.
We applied the PhyloNet-HMM framework to detect introgression in chromosome 7 in three sets of mice, as described above. Each data set consisted of two individuals from M. m. domesticus and two individuals from M. spretus. Thus the phylogenetic network is very simple, and has only two leaves, with a reticulation edge from M. spretus to M. m. domesticus; see Fig. 9(a). As we discussed above, the evolution of lineages within the species network can be equivalently captured by the set of parental trees in Fig. 9(b-c). Since in each data set we have four genomes, there are 15 possible rooted gene trees on four taxa. Therefore, for each data set, our model consisted of 15 states, 15 states, and one start state , for a total of 31 states.
We use our new model and inference method to analyze two types of empirical data sets. The first type includes individuals of known introgressed origin, and our model recovers the introgressed genomic region reported in [2] (Fig. 10). On the other hand, the second type consists of "control" individuals collected from geographically distant regions so as to minimize the chances of introgression (though, it is not possible to rule out that option completely). Our model detected no regions of introgressive descent in this dataset (Fig. 11).
We ran PhyloNet-HMM to analyze the M. m. domesticus data set, which consisted of samples from a putative hybrid zone between M. m. domesticus and M. spretus (Fig. 10). The data set covered all of chromosome 7, the chromosome containing the Vkorc1 gene. Vkorc1 is a gene implicated in the introgression event and the spread of rodenticide resistance in the wild [2].
Based on the pattern of recovered parental trees, the PhyloNet-HMM analysis detected introgression in the vicinity of the Vkorc1 gene from approximately 123.0 Mb to 130.8 Mb, reproducing the findings of [2]. The presence of the introgression in the M. m. domesticus sample from mainland Spain but not the one from the country of Georgia suggests that the putative introgression may be polymorphic; preliminary results on additional Spanish samples (not shown) support this hypothesis. The analysis also uncovered recombination and incomplete lineage sorting in the region, as evidenced by incongruence among the rooted gene genealogies that were ascribed to loci.
The PhyloNet-HMM analysis detected introgression in 8.9% of sites in chromosome 7, containing over 300 genes. Notably, the analysis located similar regions in other parts of chromosome 7 which were not investigated by prior studies such as [2]. Examples include the region from 107.7 Mb to 108.9 Mb and the region from 115.2 Mb to 117.6 Mb. It is worth mentioning that the method does detect ILS within introgressed regions and outside those regions as well; yet, it does not switch back and forth between these two cases repeatedly (which is an issue that plagues methods that assume independence across loci).
As described by our model above, if we sum the transition probabilities from any state to all states, we obtain a value for . We performed this computation for each state, and took the average of all estimates based on each of the 15 states. Our model estimates the value of as . This can be interpreted as the probability of switching due to introgression, and can shed light on introgression parameters.
The posterior decoding probabilities, based on Eq. (2), for all positions in chromosome 7, are shown in Fig. 10(a). Clearly, the introgressed regions indicated by green bars in Fig. 10(d) have very high support (close to 1), particularly the region around the Vkorc1 gene.
To further validate our approach, we repeated our scans on the M. m. musculus control data set (Fig. 11), which contained two sets of genomes of mice that are not known to hybridize. The first set of mice consisted of the M. spretus samples from the previous scan, and the second set of mice consisted of geographically and genetically distinct samples from M. m. musculus, which is not known to hybridize with M. spretus in the wild.
PhyloNet-HMM did not detect introgression on the control data set. The analysis recovered signatures of ILS, though, based on local incongruence among inferred rooted gene genealogies.
In this paper, we introduced a new framework, PhyloNet-HMM, for comparative genomic analyses aimed at detecting introgression. Our framework allows for modeling point mutations, recombination, and introgression, and can be trained to tease apart the effects of incomplete lineage sorting from those of introgression.
We implemented our model, along with standard HMM algorithms, and analyzed an empirical data set of chromosome 7 from mouse genomes where introgression was previously reported. Our analyses detected the reported introgression with high confidence, and detected other regions in the chromosome as well. Using the model, we estimated that about 9% of the sites in chromosome 7 of an M. m. domesticus genome are of introgressive descent. Further, we ran an empirical analysis on a negative control data set, and detected no introgression. On simulated data, we accurately detected introgression (or the lack thereof) and related statistics from data sets generated under both isolation-with-migration and isolation-only models.
We described above how to extend the model to general data sets with arbitrary hybridization and speciation events, by using a MUL-tree technique. However, as larger (in terms of number of genomes) data sets become available, we expect the problem to become more challenging, particularly in terms of computational requirements. Furthermore, while the discussion so far has assumed that the set of states is known (equivalently, that the phylogenetic network is known), this is not the case in practice. This is a very challenging problem that, if not dealt with carefully, can produce poor results. In this work, we explored a phylogenetic network corresponding to a hypothesis provided by a practitioner. In general, the model can be "wrapped" by a procedure that iterates over all possible phylogenetic network hypotheses, and for each one the model can be learned as above, and then using model selection tests, an optimal model can be selected. However, this is prohibitive except for data sets with very small numbers of taxa. As an alternative, the following heuristic could be adopted instead: first, sample loci across the genome that are distant enough to guarantee that they are unlinked; second, use trees built on these loci to search for a phylogenetic network topology using techniques described in [40]; third, conduct the analysis as above. Of course, the phylogenetic network identified by the search might be inaccurate, in which case the use of an ensemble of phylogenetic networks that are close to that one in terms of optimality may be beneficial.
|
10.1371/journal.pcbi.1004236 | Multiplex Eukaryotic Transcription (In)activation: Timing, Bursting and Cycling of a Ratchet Clock Mechanism | Activation of eukaryotic transcription is an intricate process that relies on a multitude of regulatory proteins forming complexes on chromatin. Chromatin modifications appear to play a guiding role in protein-complex assembly on chromatin. Together, these processes give rise to stochastic, often bursting, transcriptional activity. Here we present a model of eukaryotic transcription that aims to integrate those mechanisms. We use stochastic and ordinary-differential-equation modeling frameworks to examine various possible mechanisms of gene regulation by multiple transcription factors. We find that the assembly of large transcription factor complexes on chromatin via equilibrium-binding mechanisms is highly inefficient and insensitive to concentration changes of single regulatory proteins. An alternative model that lacks these limitations is a cyclic ratchet mechanism. In this mechanism, small protein complexes assemble sequentially on the promoter. Chromatin modifications mark the completion of a protein complex assembly, and sensitize the local chromatin for the assembly of the next protein complex. In this manner, a strict order of protein complex assemblies is attained. Even though the individual assembly steps are highly stochastic in duration, a sequence of them gives rise to a remarkable precision of the transcription cycle duration. This mechanism explains how transcription activation cycles, lasting for tens of minutes, derive from regulatory proteins residing on chromatin for only tens of seconds. Transcriptional bursts are an inherent feature of such transcription activation cycles. Bursting transcription can cause individual cells to remain in synchrony transiently, offering an explanation of transcriptional cycling as observed in cell populations, both on promoter chromatin status and mRNA levels.
| Transcription initiation is an important process that contributes to determining mRNA and eventually protein levels. In multicellular organisms transcription activity of a single gene is regulated by many different signals. This leads to multiple transcription factors binding to the same promoter. Here we study fundamental aspects of this regulation. We show that the formation of a single regulating protein complex, consisting of tens of proteins is a very inefficient mechanism: it is slow and hard to regulate. The formation of small complexes, each leaving a histone modification on the promoter solves these problems. Optimally complexes are assembled in a strict order and new histone modifications sensitize the chromatin for the formation of the next complex. This leads to a cyclic, ordered series of irreversible events that is fast and can be tightly regulated—a cyclic ratchet mechanism; like a mechanical ‘clock’. In the ratchet model, one particular chromatin state allows for RNA polymerase rebinding, which makes bursts in mRNA production a basic feature of the ratchet mechanism. Such bursting transcriptional activity has indeed been observed for eukaryotic genes. The duration of the entire transcription cycle can easily become tens of minutes, even though single proteins reside on the chromatin for only tens of seconds. Although the formation time of protein complexes on chromatin can be highly variable, due to invariable stochasticity in protein binding and disassociation, the duration of the entire transcription cycle can be very precise. As a consequence, cells that are activated at the same moment in time display synchronous transcriptional activity for several transcription cycles. This provides an explanation for transient transcription cycles observed at the level of cell populations.
| Eukaryotic transcription depends on dozens of proteins, including transcription factors (TFs), chromatin remodellers and RNA polymerase II components [1–7]. It is frequently more complex than prokaryotic transcription [8].
Even though we refer by ‘transcription regulation’ only to regulation of the transcription process itself and not to the regulation of gene expression through transcription, the complexity of eukaryotic transcription regulation is immense. It varies with gene function. ‘House-keeping’ genes maintain nucleosome-free regulatory regions permissive to transcription [9, 10]. Genes involved in developmental transitions switch between ‘OFF’ and ‘ON’ states under the control of multiple TFs [11] affected by the epigenetic state of the corresponding chromatin region [12]. Conditionally-active genes can respond to multiple TFs that activate RNA polymerase II complex assembly and modify the chromatin state of the gene’s regulatory region [13].
Regulation of gene activity involves protein complex formation on chromatin and changes in nucleosome states [14, 15]. Genome-wide analyses confirmed this for human transcription [16–18]. In the OFF state, nucleosomes often mask regulatory DNA sequences. Upon nucleosome remodelling, promoter regions become transiently accessible for TFs that, in turn, recruit RNA polymerase II to start transcription of selected genes in the context of chromatin [7]. Thus, the activity of a gene is determined by multiple stochastic events [19–21]. Protein complex assembly and DNA binding can be thermodynamically reversible, whereas other processes are Gibbs-energy-dependent, for instance requiring ATP hydrolysis, and irreversible, including covalent histone modifications, nucleosome eviction and transcription initiation. Irreversible events typically require enzymatic activity for their reversal. Irreversible events therefore lead to chromatin and nucleosome states that have a longer life-time than protein complex (dis-)assembly and protein residence times on chromatin. This may function as a molecular memory of the transcription cycle state [22].
In some cases transcription dynamics at the cell population level proceeds in an oscillatory fashion [23–28] at frequencies between 30 and 60 min [24, 27], i.e. much slower than the observed protein dwell times on chromatin of up to 1 min [29]. It is currently not understood how these fast molecular association and dissociation events cause such slow alterations in gene activity. Using a minimal model it was shown [30] that sequential assembly of protein complexes can in theory give rise to oscillatory gene activity. However, whether models with realistic kinetic parameters can also generate such slow dynamics remains unclear.
Despite many recent experimental studies, we lack a realistic, dynamic picture of transcription initiation that integrates time scales of nucleosome modifications, protein complex formation, RNA polymerase assembly and its escape from regulatory regions, mRNA elongation and mRNA-concentration dynamics. This hiatus precludes resolving a number of issues: do cyclic transcriptional mechanisms perform better than the reversible ones found in prokaryotes; and are they perhaps even essential? How well can an ordered transcription cycle be regulated by many factors in comparison to a random set of equilibrium binding events? Is irreversibility essential for transcription regulation by multiple TFs? Does such organization of transcription imply special regulatory and stochastic properties? Do the stochastic dynamics of the transcription cycle model agree with single-cell and population-level studies of transcription when realistic kinetic parameters are considered? Is the seconds-timescale of molecular events in agreement with transcriptional cycling of tens of minutes?
Using a variety of experimental data and observations, we here compare a number of transcription activation mechanisms in terms of their ability to be both fast enough and responsive to multiple factors. We show that fully reversible transcription activation mechanisms would not work for promoters regulated by many TFs, which is expected for conditionally regulated genes in mammals. Rather, a transcription activation clock model with irreversible ratchets does resolve diffusion limitations, the required multi-factorial regulation as well as the experimental observations on transcriptional cycling and bursting.
In the equilibrium-binding mechanism, TFs populate individual binding sites in a concentration-dependent manner. For convenience, we define every protein involved in transcription regulation as a TF, regardless of whether it acts merely as a scaffold, a modifier of a nucleosome or enzyme. For n TFs, the activity of the gene Φ(T), with T as the vector of TF concentrations, corresponds to the product of the saturation ϕi(Ti) of the individual, independent TF-binding sites, i.e. Φ(T)≡vk′=∏i=1nϕi(Ti), with v as the transcription rate and k′ as the apparent transcription rate constant. Here ϕ(Ti) denotes the saturation function of a single site, in the case of an activating TF it becomes TiTi+KD,i′ and when the TF is inhibiting it equals KD,i′KD,i′+Ti, with KD,i′ as the apparent dissociation constant of site i for TFi. Taking the product of the saturation functions of the individual, independent sites has the effect that saturating 10 activating sites by 50% leads to an activity of the gene of only 0.1% (Φ(T)=(12)10=11024). In order to achieve 75% of maximal gene activity, i.e. Φ(T) = 0.75, each individual site would have to be virtually saturated, e.g. ϕi(Ti) = 0.98, which is unrealistic. The tendency of requiring enhanced saturation of individual binding sites is shown in Fig 1A.
Gene regulation by many regulatory factors based on equilibrium binding is also limited in terms of responsiveness. The sensitivity of the transcription rate to a specific TF depends on the extent of saturation of the gene with that TF: ∂lnv∂lnTi=1−ϕi(Ti)=0.02 for ϕi(Ti) = 0.98. This analysis shows that increasing the single-site saturation, which is sufficient to produce significant activation of transcription, causes the gene to loose sensitivity to regulators. Thus, in case of the equilibrium binding mechanism, genes become progressively harder to regulate when the number of regulatory factors increases. This finding is highly relevant as the number of TF-binding sites actively regulating many eukaryotic genes is readily around ten or more. A genome-wide analysis of data from ENCODE project shows in Drosophila that the number of functional TF-binding sites per enhancer sequence is between 2 and 15 with the average around 8 [31]. Combined evidence from several studies of individual promoter studies supports similar conclusions [32–34]. Additionally, many transcription co-factors and histone modifiers are regulated by signaling and metabolic pathways [35], increasing the number of potential regulatory inputs.
We have also considered whether cooperative binding of the TFs can overcome the deficiencies of the equilibrium-binding model (see S1 Text). We find that for a high number of TFs unrealistically high levels of cooperativity between regulatory factors are required to solve the problem of sensitivity or saturation loss (S1 Fig). We will further discuss alternative mechanisms for eukaryotic gene regulation that do not suffer from this limitation.
The mean assembly time for a protein complex of n factors, with identical kinetics, depends on the reversibility of complex formation (KD′) and the first-order association rate constant (kF′) as: τ(n)=nkF′+∑i=1n−1ikF′(KD′)n−i. The irreversible assembly time equals n/kF′ for n identical factors; if factors have different association rate constants, this time is given by ∑i=1n1/kF,i′. In these equations, kF′ denotes the effective first order rate constant (unit: min-1) for the binding of a regulatory factor to a partially assembled complex. It equals the multiplication of a (diffusion-limited) second-order rate constant for binding (unit: nM-1 min-1) with the (nuclear) concentration of the regulatory factor (unit: nM).
Using these equations, one can estimate both the irreversible and the reversible assembly times of a protein complex composed out of 10 components. We take the cell nucleus volume to be 1.2 pL given that the measured cell volume is 3–4 pL [36] and that 30% of this volume is occupied by the nucleus [37]. The average copy number of sequence–specific TFs per cell, as found in whole-cell proteomic studies, is around 2500 (S2 Fig), which results in a concentration of 3.5 nM [38–41]. Similar average concentrations are found for the chromatin modifiers (S2 Fig). Protein-chromatin association rates are found to be about two orders of magnitude below the diffusion limit [42], which can be directly calculated to be around 0.01 nM-1s-1 (see S3 Table). With these numbers, kF′ is approximately 2 min-1, i.e. transcription activation could take place twice per minute if it would depend on just a single TF. For a complex of 10 TFs, assembling irreversibly in a defined sequence, assembly should then take 5 min. Ficz et al. [43] measured an off-rate constant (kB) of 2 min-1 for Polycomb group proteins from chromatin. This sets the dissociation equilibrium constant (KD′) close to 1, which is consistent with our analysis of the requirements for the promoter sensitivity to TFs. Assuming the dissociation rate constant, kB, to be equal to kF′ and substituting these numbers into the equation for τ(n), supplied above, gives for complexes of 2, 5, 7, 10, 12 and 24 TFs assembly times of 1.5, 7.5, 14, 27.5, 39 and 150 min, respectively, i.e. much longer than the times for irreversible assembly (39 versus 5 min for a complex of size 10). Notably, for the complex size of 10 the assembly time is more than twice as high as the assembly time of a complex of five proteins (7.5 min). While the exact timing of the complex assembly depends on the values of the protein copy number and the rate constants, which may vary, the overall conclusion that the reversible mechanism is considerably slower at high protein copy number is always correct for effective KD around 1 or higher. These numbers indicate that protein complex formation on chromatin can become potentially rate limiting for transcription rate. The delay in reversible assembly is due to ‘hesitation’, i.e. frequent disassembly of a partially assembled complex before full assembly. Division of the reversible assembly time, τ(n), by the time for irreversible assembly of the complex, τ(n)/(n/kF), indicates the influence of reversibility and complex size on assembly time (Fig 1B): large complexes have very long assembly times and this effect becomes dramatic if the processes are more than half reversible, i.e. when the effective dissociation constant favours dissociation. For an effective dissociation equilibrium constant of 2 and an on-rate constant of 2 min-1 the assembly for a decameric complex is delayed by a factor of 200; i.e. the process is hesitating all the time. Then transcription activation would take 17 h rather than 5 min. This clearly points out that, if a substantial number of TFs are involved in transcription activation, the reversible mechanism may well become too slow for appropriate timing of transcription. The assembly time also depends on the precise mechanism of assembly, as discussed in S2 and S3 Texts. In particular, we find that, in general, random assembly mechanisms are faster than the sequential ones (S3 Fig) and that direct assembly on the chromatin is faster compared to the pre-assembly in the nucleoplasm (S4 Fig). In summary, our analysis indicates that for many eukaryotic genes mechanism of reversible association of all TFs becomes too slow.
Fig 1B shows that for a KD′ of 2, the reversible assembly of 9 TFs is 100-times slower than the irreversible assembly, whilst for a complex size of 3 it is only about 3-times slower. This suggests that a sequence of two trimeric association processes could be faster than a single 6-proteins association process. Likewise, for an effective dissociation constant of 1, the reversible continuous formation of a complex composed of 24 proteins should be 5-times slower than the corresponding assembly of 6 complexes, each of which is composed of 4 in sequence, i.e. τ(24)/6τ(4) = 5 (for KD = 1 and 140 for KD = 1.33). These calculations indicate that the batch-wise assembly of multi-factor complexes can reduce their otherwise long assembly times. But for this, the partial complex established in the first phase should not dissociate whilst the protein complex is being formed.
This conclusion brings us to an important finding: to be fast enough, the process of sequential formation of small protein complexes requires irreversible marking of progress (which we shall refer to as ‘ticking’), for instance, by way of covalent histone modification of nucleosomes. The transcription activation process should therefore resemble a molecular ratchet: it can ‘hesitate’ during the reversible assembly of the small protein complex but cannot return to a state where the previous set of proteins were assembling due to the histone ticking. This batch-wise mechanism has been suggested before [22] but it has never been deduced as essential for eukaryotic transcription to be fast and regulative enough (see below).
One way to make the process of transcription activation revertible is to make it thermodynamically reversible. This corresponds to the equilibrium transcription activation mechanism discussed above. S5A Fig shows this mechanism for the case where 10 TFs reversibly bind to the chromatin, producing the activated transcription complex. In the S4 Text we detail why this type of mechanism and 4 others (S5 Fig), proceeding in reverse order through the transcription activation pathway, are either too slow or not revertible.
An alternative to reversing on a linear scheme is a combination of a forward linear pathway and a separate reverse linear pathway; together constituting a cycle. In Fig 2A, four TFs bind irreversibly to and (a number of steps later) dissociate irreversibly from their DNA-binding site, referred to as a response element. The covalent histone modifications are shown by the various symbols in the rectangles. In the simplest irreversible cyclic mechanism (Fig 2A) the regulatory factors dissociate according to a last-in, first-out principle. In the S4 Text we show that this mechanism corresponds to the one Fig 2B and that this mechanism is too linear to be effective. A mechanism in which TFs dissociate irreversibly on the basis of a first-in, first-out principle (Fig 2C) is able to attain much higher transcriptional activity than the equilibrium binding model, whilst transition times between active and inactive gene states are also much shorter. Although every step in Fig 2C is irreversible, the cycle could also operate batch-wise (leading to a considerable advantage, see above): one or a few regulatory factors could bind reversibly after which one would bind really strongly (nearly irreversibly; so, the bound state has a long lifetime), thereby fixing the information that all regulatory factors have bound (see Fig 2D). A mechanism in which chromatin is ‘ticked’ or ‘stamped’ upon encountering a TF, which subsequently dissociates (Fig 2E), is equally competent kinetically. We call this the ‘ticking mechanism’. We note that the most realistic mechanism for leaving marks on the chromatin, following the activity of a particular short-lived protein complex, is via histone modifications, as we doubt that many regulatory proteins can stay bound to chromatin for tens of minutes without covalent interactions.
Our analysis thus far has used criteria of regulatory sensitivity and the speed of regulation of transcription to show that a ‘ticking’ transcription cycle mechanism is an attractive mechanism for eukaryotic genes that are regulated by multiple TFs. There is considerable experimental evidence that shows that this mechanism is indeed operative (S5 Text).
The ticking mechanism described above is consistent with the wealth of experimental evidence on the role of chromatin covalent-modification and remodelling in regulation of eukaryotic gene transcription [3, 5–7, 9, 10, 22, 44]. There are many examples of histone modification recruiting new modifiers, and subsequent modifications recruiting new proteins actively involved in transcription induction or cessation. For instance, during promoter activation, H4R3 methylation, mediated by the methylase PRMT1, increases the affinity of histone acetyltransferase p300 for chromatin [7]. Some acetylation marks, e.g. H3K14Ac, H4K16Ac, increase binding of chromatin remodelling complexes in various experimental systems [45, 46] while others (H3K9Ac, H4K12Ac, H4K8Ac) attract basal TFs required for initiation [47, 48]. This could potentially create an orderly sequence of modifications of a chromatin site thus providing the ticking mechanism of the gene activation process [22].
There is also ample evidence for TFs forming complexes with each other and with co-factors that have “ticking activity”. One example is the PPARγ-RXRα heterodimer that forms a complex with the co-activator PGC1α and the histone acetyltransferases CBP and SRC1 (Fig 3A) [49]. Another example is the repressive complex Mi-2/NurD that includes the chromatin remodeller Mi-2, the histone deacetylases 1 and 2, and the methyl CpG-binding protein MBD, which can also bind repressive TFs [50]. Fig 3B shows the possible transitions’ sequence between ‘ON’ and ‘OFF’ phases of the promoter cycle based on literature data (see S5 Text for full description and references). The ‘ON’ state is attained through three successive ticks, or irreversible modifications: arginine methylation, lysine acetylation and chromatin remodelling that make the transcription site accessible to basal TFs. This is the state during which RNA polymerases are assembled into transcriptionally competent complexes, and start the elongation of an mRNA transcript. The RNA polymerase II-associated complexes then add a fourth mark—methylation of a lysine group(s). The inactivation of the TF-binding site follows a similar mechanism of four successive protein complex formations, now accompanied by the removal of ticks from the chromatin. The transcriptionally active state of chromatin can in principle persist during de-ticking, depending on the order of the deactivating events. Each transition in the cycle could be dependent on DNA-binding TFs, which would regulate the transcription rate by changing the duration of ‘ON’ or ‘OFF’ states.
In the proposed mechanism, the sequence of chromatin modifications proceeds in a defined order. When the activation and inactivation routes would cross however, as would happen in a random mechanism, this would obviously disturb the progress marking, causing the system to erroneously move backward or forward, and skip steps. These considerations may explain the preference for a unique order of histone mark addition/removal and specificity of each partially assembled complex to a chromatin state.
The mean waiting time for a single TF-binding event equals its standard deviation, making its timing very imprecise. The time to complete a defined sequence of n identical first-order reactions follows an Erlang distribution [30]. The variance in the duration of such a sequence, denoted by ⟨δ2τ⟩, equals n/k2. The noise in the duration of the sequence, defined as ⟨δ2τ⟩/⟨τ⟩2 equals 1/n. Therefore, the noise equals 1 for the single step mechanism but it is much less for a sequence of such steps: a sequence of reactions with identical duration has a more precise completion time than any of its component reactions. This conclusion holds true if a sufficient number of the steps in the cycle have different but comparable durations, but falters should one reaction be much slower than all the others, then the noise increases.
We modelled a sequence of nine chromatin-state transitions as approximation of an entire transcription cycle. Each of the transitions involved the reversible assembly of a complex of five proteins followed by irreversible histone modification (Fig 4A). In total 45 regulatory proteins were involved. We considered a batch process, i.e. a preferentially random-assembly mechanism on chromatin for protein complex formation, with a strictly ordered sequence of the chromatin modifications guiding the sequence of complex assemblies and sensitization of the chromatin for the assembly of the next protein complex. We chose realistic values for the rate constants and protein abundances (see S3 Text and S3–S5 Tables). Fig 4C shows cycling through several promoter states by four individual cells after transcription was activated in each of them at the same time. We observe that, for example, state 2 occurs in the first cycle in all cells almost simultaneously; and in subsequent cycles the occurrence of this state slowly loses synchrony across the four cells, as indicated by the increasing dispersion of the occurrence time. This means that the cells are progressing through the transcription cycle slowly (as compared to the TF binding time scale) and synchronously in the beginning, but desynchronize over time. This can be further illustrated by plotting the probability of observing each of the four states in a population as function of time (Fig 4B).
To illustrate how the stochasticity of assembly times leads to transient de-synchronization of a population of cells, we will consider what occurs with increasing number of cycles of n steps. The overall waiting time distribution to complete Nth cycle becomes narrower with an increasing number (N) of cycles: 〈δ2τN〉〈τN2〉=(n·N)−1. To quantify the effect of de-synchronization, we consider the noise in the timing of the end of the N-th cycle relative to the mean duration of a single cycle, i.e. using 〈τN〉=N〈τ〉:〈δ2τN〉〈τ〉2=Nn. This relation shows that cells will progressively become asynchronous with increasing number of completed transcription cycles, and that phase-noise in the Nth cycling time increases linearly with N. At the same time, it shows that more transitions per transcription cycle, n, tend to prolong the persistence of population level synchrony. These relations explain the behaviour calculated in Fig 4.
We simulated the stochastic dynamics for a population of 1,000 cells that had simultaneously started transcription activation of two gene copies. Fig 4D shows the fractions of cells that are in a given state at any moment in time (also shown for a single cell in Fig 4C).
Fig 4D shows that population level transcriptional cycles are observable for around 100 min and decrease in amplitude over time due to de-synchronization as expected. The transcriptional cycling time that we predict with this realistically parameterized model is approximately 54 min. The predicted waiting time distribution of the cycling time indeed peaks (Fig 5A), as expected for a multi-step sequential process.
The number of RNA polymerase molecules that initiate elongation per transcription cycle depends on the tendency of genes to engage in a phenomenon called re-initiation [20, 51–53]. Many experimental studies have shown that mRNA is produced in bursts [14, 54–56] of variable size [57, 58]. We incorporated RNA polymerase II binding and promoter escape during the permissive ON state into the model. We considered promoter state 5 to be transcriptionally permissive. The overall lifetime of the state is around 6 min (Fig 5A, orange line). However, as we assumed competition between polymerase and deactivating complex, the effective permissive state only occurs while the deactivating complex is not bound and its lifetime is close to 1 min (Fig 5A inset, red line). The polymerase concentration in the model was taken higher than the concentrations of other TFs, estimate supported by the experimental copy number data (S2 Fig). Therefore, polymerase binding and promoter escape time had an average of 0.1 min (Fig 5A inset, black line), a much shorter time than the lifetime of the effective ON state, resulting in multiple transcripts being initiated during most of the permissive periods. The average burst size should then be 10 mRNAs/cycle and the distribution of burst size can be shown to be geometric; which is indeed the case for calculated burst size distributions in Fig 5B. This finding is in good agreement with experimental observations that mRNA burst-size distributions for regulated genes across a cell population are often geometric [54, 55]. The contribution of the stochasticity of transcriptional bursts, as generated by the ratchet mechanism, to the total stochasticity of transcript concentrations in a cell depends also on factors, independent of the ratchet mechanism [59].
Whether bursting will occur in synchrony across cells depends on whether the cells were induced at the same time. Fig 5C and 5D show the prediction of mRNA trajectories in five individual cells for the 9-state ratchet cycle model that were induced at the same time (Fig 4A). The model predicted mRNA bursting: all five modelled cells fired mRNAs periodically and for the first 2 h they did this in fair synchrony. When mRNA decay was rapid (Fig 5C), most of the mRNA was degraded during the promoter OFF period. For multiple cells, this led to the prediction of population level transient oscillation in mRNA levels (Fig 5E). If no mRNA degradation occurred on the considered timescale the bursts led to step-wise accumulation of mRNA in individual cells (Fig 5D) as well as on the cell population level (Fig 5F).
The model developed in the previous sections explains a number of experimental observations dealing with cyclical changes in protein occupancy, nucleosome modifications, looping of eukaryotic regulatory regions, and transcriptional bursts. The transcriptional cycling periods that have been observed experimentally vary between 40 and 90 min [24, 25, 27, 28, 60], in line with the above predictions for the 9-state ratchet cycle model.
The most detailed dataset available is for the regulatory region of the estrogen-responsive trefoil factor 1 (TFF1) gene [27]. It exhibits the orderliness of the binding events that is a component of cyclic ratchet model: first, co-activator complexes containing histone acetyltransferases and histone methyltransferases bind, then the basal TFs and RNA polymerase II, and finally the de-activator complexes, containing remodelling/HDAC activity. This experimental data (Fig 6B) can be reproduced fairly well by a 9-state cycle model with realistic kinetic parameters (Fig 6A). The de-synchronization observed in the experiment was however slower than in the model simulations. This could be explained by an even higher number of proteins and chromatin modifications involved serially in the initiation process than currently known experimentally, by more peaked waiting time distributions for individual cycle transitions, which could again be due to multi-step serial processes, or by differences in kinetic parameters. A clear example where the de-synchronization predicted by our model is observed experimentally has been provided by the measurement of the TF occupancy on chromatin, using fluorescence-based methods in single cells [24].
Population level oscillations in mRNA concentrations as predicted by our cycle model (Fig 5E) have also been observed experimentally (Fig 6D). The yeast metallothionein (CUP1) gene, displayed 50-min transcriptional cycles of its main TF, Ace1p, upon activation and the CUP1 mRNA was oscillating at the same frequency [24]. Adjustment of the protein concentrations and the consideration of two promoter states (PR2 and PR3), which can each be transcriptionally permissive, allows for qualitative correspondence of the model simulations and experimental data.
Evidence for reversible looping between a response element and the TSS has been found in the regulatory region of the human cyclin-dependent kinase inhibitor 1A (CDKN1A) gene, which is regulated by several REs for the TF VDR [28]. Some of the co-regulators were shown to oscillate at the population level with a 60-min period. The experimental data (Fig 6F) can be reproduced by our model (Fig 6E) if the looping event between the distant RE and TSS in the model occurs during the chromatin state corresponding to RNA polymerase II binding.
In this study, we asked how transcription regulation of conditionally active genes in eukaryotes could be organized so that it is both fast enough, given diffusion limitations, and sensitive enough towards a large number of TFs. We showed that neither an equilibrium binding mechanism nor a more irreversible binding mechanism, reversing its steps upon inactivation, could be fast and sensitive enough. We then showed that the next simplest, competent mechanism is a batch-like multi-step process of transcription activation, followed by a separate batch-like multi-step inactivation process, together constituting a transcription (in)activation cycle. Each step would correspond to the reversible association of a limited number of TFs leading to irreversible marking of the transcription activation complex, after which the TFs dissociate. This transcription (in)activation cycle should not be confused with the more restricted use of the term transcription cycle when it refers to the process of an RNA polymerase producing an entire mRNA. We have a much more involved process of gene activation and transcription initiation in mind.
If one were to specify that histone modification is the process that marks the completion of a step in the multi-step process and sensitizes the local chromatin region for the assembly of the next protein complex, this model comes close to the accepted view of transcription activation in eukaryotes [7, 10, 22, 44]. The difference is that we derived this view as a requirement for delivering kinetic competence, revertibility and sensitivity, whereas the accepted view is based on experimental observations of the transcription activation process. All in all, our study suggests that we have found a plausible explanation as to why transcription (in)activation of many regulated genes in eukaryotes is organized the way it is, and why it is different from the activation of transcription that is regulated by a small number of TFs, such as in prokaryotes, or the house-keeping genes.
Metivier et al. [27] proposed a branched mechanism for the transcription cycle at the TFF1 promoter. Our model can straightforwardly be extended with such details. We considered such an extension in a minimalist manner when we incorporated the transcription re-initiation mechanism to allow for variable burst size of the transcription cycle.
In our model, a sequence of protein complex assemblies intermitted by covalent-modification of histones to mark the phase of the transcription cycle, leads to a fairly deterministic duration of the active transcription state. This has two consequences. Firstly, our model explains how the duration of the entire transcription cycle could be close to deterministic, i.e. clock-like. Secondly, in the transcription re-initiation formulation of the model the number of transcription (re-)initiations per cycle becomes more deterministic. This means that the eukaryotic transcription cycle can be both clock-like in terms of duration and quantal in terms of its activity. This is within the limit of many molecular events in series with similar reaction rates. To what extent real genes function within this limit is unclear. Recent experiments indicate that the ON and OFF durations of genes can have non-exponential waiting time distributions, which is in agreement with our predictions [58, 61, 62]. The clock-like nature of the model underlies the tendency of cells in a population to display transiently synchronous transcription activity upon simultaneous activation from the same initial state. The model does not, however, exclude a possibility of the transcription cycle times being less precise, due, for instance, to the presence of a very slow step in one or more of the cycle transitions, or a possibility of variable gene induction times due to heterogeneity of the initial promoter states in a cell population.
We have shown that our model explains precise durations of the entire transcription cycle of about 1 hour on the basis of molecular processes that are faster than 1 per minute. This is because of a sequence of several multi-step processes, each of which is consolidated by irreversible marking (ticking). This ticking corresponds to a ratchet mechanism, but the overall process corresponds to a clock mechanism as already proposed by Reid et al. [22]. A mechanical clock has the same type of ratchet mechanism and is similarly accurate at longer time scales: the timing of its individual ticks is not accurate, but the timing of large numbers (for example 60) of its ticks is. From a coarse-grained perspective, the transcription activation cycle model assumes that eukaryotic genes switch between transcription active ON states and ditto, inactive OFF states. The duration of these phases could be under the control of some regulator [56, 63, 64]. The same applies to burst size. From this perspective, transcription rates can be controlled in different ways. A suitable definition of transcription rate would be the mean burst size divided by the mean cycle time: 〈b〉〈τON〉+〈τOFF〉. Control of transcription rate can then be achieved via modulation of burst frequency (cycle duration; the FM mode) or burst size (the AM mode). Skupsky et al. [65] found evidence for regulation of burst size rather than frequency. The transcription cycle models we presented can accommodate both mechanisms.
Not all protein complexes involved at the various stages of the transcription cycle assemble on the chromatin, some may already be assembled in the nucleoplasm, such as mediator complex [66]. These pre-assembled complexes do not affect the total duration of the transcription cycle and its precision, as they bind in one step. However, even if all protein complexes would assemble in the nucleoplasm, which we know is not the case, histone marking would be still advantageous to make sure that they bind in the correct order.
In our model we propose that histone modification function as marks of the progress of the transcription cycle, because they have a longer lifetime than protein complexes—proteins reside on chromatin only for several tens of seconds. This does not mean that all histone marks should be remembered along the entire transcription cycle. We expect the minimal life of a histone mark to be related to the time that it takes to form the next-in-line protein complex. So, if protein complex i has been formed after which mark i is added, the lifetime of mark i should be long enough to make sure that the protein complex i+1 has had its time to form and leave mark i+1. Since marks can always be removed by accident by an enzyme, or fall off spontaneously, it makes sense that a short memory of previous marks should be present on chromatin, say in addition to mark i, mark i-1 and i-2, to make sure that if mark i gets removed by accident the transcription process does not erroneously reset to its resting state. Considering that the formation of a protein complex on chromatin takes several minutes, we would therefore expect that histone modifications that mark the progress of the transcription cycle stay for about 10 minutes or more on chromatin.
Transcription dynamics differ between prokaryotes and eukaryotes [14, 19–21, 54, 55, 67–69]. Eukaryotes tend to display transcription bursts, which in prokaryotes have only been found under conditions of leaky transcription repression [56, 64, 70], and occur infrequently across the majority of genes in Escherichia coli [71]; in contrast to what is found in Saccharomyces cerevisiae [72, 73]. RNA polymerase II assembly, and its escape from the regulatory region, only take place during a fraction of the entire transcription cycle, suggesting that many eukaryotic genes are prone to transcriptional bursting even under conditions of high transcriptional activity [24, 27, 54, 55]. The mechanisms for transcriptional bursts in bacteria [64, 74, 75] are likely very different in molecular detail from the mechanisms in eukaryotes, even though they can be coarse-grained to a similar mechanism giving rise to a variable number of RNA polymerases that initiate transcription during a single ‘ON’ state of the gene [59].
Transcriptional cycling differs from regular oscillations induced by a negative feedback loop, as found, for example, in NADH fluorescence and glycolytic activity in yeast [76], the dynamics of calcium concentrations [77], and the activities of NF-κB [78] or the kinase ERK [79]. In the latter examples the dynamics of metabolite or signal transduction factor pools are coupled through nonlinear kinetics. In some of these cases sustained oscillations have been observed, and for one of these the requisite active synchronization mechanism requiring dynamic communication between individual cells, has been elucidated [80]. The mechanism we propose for the transient oscillations at the population level is entirely cell-autonomous, i.e. no active communication between cells is involved. For transcriptional cycling, we propose that the transient cyclic dynamics at the population level are the consequence of a simultaneous start and accurate durations.
The mass balances, rate equations and parameters for protein-assembly and complete transcription initiation models were generated by using custom algorithms implemented in Mathematica 7.0. ODE system simulation were done using NDSolve function; stochastic simulations—by implementing direct-method of Gillespie algorithm [81]. For detailed description of the models’ structures, parameters and initial conditions see S3 Text, S1–S6 Tables and S6 Fig. All Mathematica files are available as part of Supporting Information (S1 Dataset).
|
10.1371/journal.pgen.1003164 | A Framework for the Establishment of a Cnidarian Gene Regulatory Network for “Endomesoderm” Specification: The Inputs of ß-Catenin/TCF Signaling | Understanding the functional relationship between intracellular factors and extracellular signals is required for reconstructing gene regulatory networks (GRN) involved in complex biological processes. One of the best-studied bilaterian GRNs describes endomesoderm specification and predicts that both mesoderm and endoderm arose from a common GRN early in animal evolution. Compelling molecular, genomic, developmental, and evolutionary evidence supports the hypothesis that the bifunctional gastrodermis of the cnidarian-bilaterian ancestor is derived from the same evolutionary precursor of both endodermal and mesodermal germ layers in all other triploblastic bilaterian animals. We have begun to establish the framework of a provisional cnidarian “endomesodermal” gene regulatory network in the sea anemone, Nematostella vectensis, by using a genome-wide microarray analysis on embryos in which the canonical Wnt/ß-catenin pathway was ectopically targeted for activation by two distinct pharmaceutical agents (lithium chloride and 1-azakenpaullone) to identify potential targets of endomesoderm specification. We characterized 51 endomesodermally expressed transcription factors and signaling molecule genes (including 18 newly identified) with fine-scale temporal (qPCR) and spatial (in situ) analysis to define distinct co-expression domains within the animal plate of the embryo and clustered genes based on their earliest zygotic expression. Finally, we determined the input of the canonical Wnt/ß-catenin pathway into the cnidarian endomesodermal GRN using morpholino and mRNA overexpression experiments to show that NvTcf/canonical Wnt signaling is required to pattern both the future endomesodermal and ectodermal domains prior to gastrulation, and that both BMP and FGF (but not Notch) pathways play important roles in germ layer specification in this animal. We show both evolutionary conserved as well as profound differences in endomesodermal GRN structure compared to bilaterians that may provide fundamental insight into how GRN subcircuits have been adopted, rewired, or co-opted in various animal lineages that give rise to specialized endomesodermal cell types.
| Cnidarians (anemones, corals, and “jellyfish”) are an animal group whose adults possess derivatives of only two germ layers: ectoderm and a bifunctional (absorptive and contractile) gastrodermal (gut) layer. Cnidarians are the closest living relatives to bilaterally symmetrical animals that possess all three germ layers (ecto, meso, and endoderm); and compelling molecular, genomic, developmental, and evolutionary evidence exists to demonstrate that the cnidarian gastrodermis is evolutionarily related to both endodermal and mesodermal germ layers in all other triploblastic bilaterian animals. Little is known about endomesoderm specification in cnidarians. In this study, we constructed the framework of a cnidarian endomesodermal gene regulatory network in the sea anemone, Nematostella vectensis, using a combination of experimental approaches. We identified and characterized by both qPCR and in situ hybridization 51 genes expressed in defined domains within the presumptive endomesoderm. In addition, we functionally demonstrate that Wnt/Tcf signaling is crucial for regionalized expression of a defined subset of these genes prior to gut formation and endomesoderm maintenance. Our results support the idea of an ancient gene regulatory network underlying endomesoderm specification that involves inputs from multiple signaling pathways (Wnt, FGF, BMP, but not Notch) early in development, that are temporarily uncoupled in bilaterian animals.
| During metazoan development one cell gives rise to thousands of daughter cells, each acquiring a particular fate depending on their temporal and spatial coordinates within the organism. The information required to assume a specific fate of a given cell is present in the genome of all cells, requiring a fine tuned mechanism for controlling and coordinating gene expression during development of the growing embryo. The fate of each cell is determined by its set of expressed genes and controlled by the action of transcriptional activators and/or repressors whose activity is governed by intracellular (e.g. localized cytoplasmic factors, RNA binding proteins), or extracellular signals (e.g. endocrine or exocrine signaling pathways). All together, these components form gene regulatory networks that underlie the formation of distinct cell types or germ layers. Understanding the relationship between intracellular factors and extracellular signals can provide key insight in how and when the molecular and morphological characters of each organism are built.
Triploblastic organisms, also called “bilaterians” due to their bilaterally symmetrical body (possessing an anterior-posterior axis and dorso-ventral polarity), constitute the vast majority of all metazoan animals. These animals are characterized by the formation of three distinct primary germ layers during embryogenesis called the endo-, meso- and the ectoderm, that subsequently differentiate into more specialized adult tissues. Ectoderm gives rise to skin and nervous system, endoderm gives rise to the derivatives of the digestive tract including the intestine and digestive glands, and mesodermal derivatives include muscle, connective tissue, blood, coelomic cavities, kidneys/nephridia, somatic portions of the gonad, and skeletal elements. Both classic descriptions as well as modern molecular analyses of germ layer formation in bilaterian organisms as diverse as nematodes, sea urchins, and vertebrates have indicated that these decisions are largely made in a two steps: ectodermal fates first separate from a bicompetent endomesodermal (also called mesendodermal) domain, and then endodermal fates become distinct from mesodermal tissues [1]–[3].
In 2002, the extensive amount of experimental data collected during the past decades by the sea urchin community was assembled into a provisional endomesodermal (EM) gene regulatory network representing interactions between signals/transcription factors (TF) and their downstream targets, which in turn activate/repress other signals/TF's required for endomesoderm formation in the sea urchin embryo [4]–[11]. To date, a very limited number of model organisms have been used to establish GRN's for endomesoderm specification and/or differentiation (for review see [12]). Endomesodermal GRNs have been established only for the nematode C. elegans [13], the sea urchin (S. purpuratus, P. lividus, L. variegatus) [6], [7], [10], [14]–[16], a sea star (A. miniata) [17], [18] and Xenopus [19]. Comparison of the sea star and sea urchin endomesoderm GRNs indicates that there is a set of highly conserved genes, thought to be part of the “kernel” of the endomesodermal circuit present in the echinoderm ancestor [18], [20]. In Drosophila, a well-established genetic model system, mesoderm and endoderm are created by fundamentally different regions of the animal [21]–[23], reviewed in [24]. Although some of the endomesodermal kernel genes appear to be involved in gut formation in insects, the differences in gut development in flies has so far made it difficult to compare with other endomesodermal GRNs from other bilaterian studied.
The origin of the mesodermal germ layer and all of its unique cell types (e.g. muscle, connective tissue, blood, kidney and somatic gonad) during metazoan evolution is a matter of intense debate and investigation (reviewed in [25]–[34]. The sister group to all triploblastic animals is a group of animals called cnidarians (sea anemones, corals, sea fans, and ‘jellyfish’). Cnidarians are diploblastic animals formed exclusively by an epidermis (ectoderm) and a gastrodermis (also historically called entoderm). There are no classical bilaterian muscle cells [35] or a mesodermal tissue layer in cnidarians, however, the cnidarian gastrodermis is a bifunctional tissue capable of both absorption and contractile functions via myoepithelial cells [29], [36]–[38]. The cnidarian gastrodermis also express a large number of both endodermal factors and genes historically associated with mesoderm formation such as otx, snail, twist [26], [39], [40] suggesting that the cnidarian gastrodermis has a bifunctional endomesodermal capacity that never segregates into two distinct tissues. It also suggests that it contains components of an ancestral triploblastic (bilaterian) endomesodermal gene regulatory network and that endodermal and mesodermal tissues in triploblastic organism may be derived from the bifunctional gastrodermis of the cnidarian/bilaterian ancestor. This provides us with the opportunity to gain insight in to the ancestral endomesodermal GRN in a living organism.
Recent studies have shown the favorable features and utility of the cnidarian Nematostella vectensis as a developmental and evolutionary model system [39], [41]–[46]. Importantly the whole genome has been recently sequenced by the Joint Genome Institute (JGI) and is publicly available [47]. As an anthozoan, it has a simple anatomy, an undetermined long life span, and a short life cycle of 10–14 weeks. The sexes are separate allowing in vitro fertilization and manipulating the light cycle can induce spawning of several hundreds of eggs/female. When raised at 17 degrees Celsius, a hollow blastula forms approximately 10–12 hours post fertilization (hpf) and the embryo begins to gastrulate around 24–28 hpf via invagination at the animal pole [48]–[50], the side of the animal that gives rise to the single oral opening and the gastrodermis (endomesoderm).
The canonical Wnt (cWnt) signaling pathway plays crucial roles during various bilaterian developmental processes such as axis specification and germ layer formation [51]–[58]. Recent studies have suggested that the cWnt/β-catenin pathway has an ancient role in axis and endomesoderm formation in N. vectensis [50], [59]. Treatments with lithium chloride (LiCl), perturbs nuclear ß-catenin (nß-catenin) distribution ectopically stabilizing nß-catenin in all blastomeres along the A/V axis and induces hyper-proliferation of endomesoderm. In addition, inhibition of the cWnt pathway by overexpressing either cadherin, a cell adhesion molecule that titrates the cytoplasmic pool of ß-catenin, or a β-catenin∶engrailed fusion (acting as transcriptional repressor) blocks gastrulation and endomesoderm formation [59]. Recently, Lee and colleagues have shown that Dsh is required for nuclearization of β-catenin and endomesoderm development by over expression of a dominant negative form of Dsh (NvDsh-DIX) that specifically stabilizes the canonical Wnt pathway [50]. Taken together, those results show that the cWnt/β-catenin pathway is required for proper endomesoderm formation in N. vectensis. Although the authors of these studies suggest that endoderm specification may be affected by cWnt inhibition, they only characterize endomesodermal gene expression by the analysis of a single gene at the late gastrula stage, a time point long after endomesoderm specification. Therefore, additional information is required to better understand early endomesoderm specification in N. vectensis.
Deciphering the cnidarian endomesodermal GRN is important for a number of reasons. It can become a useful resource to understand the basic developmental mechanisms of a “simple” animal, help understand germ layer formation in a diploblastic animal providing a framework for future developmental studies (predicting relationships with new identified genes, cis-regulatory analysis etc.), and comparative work may provide important information to understand how components of the GRN have been adopted, re-wired or co-opted that lead to the evolution of biological novelties (such as “true” mesoderm). Recent studies comparing echinoderm endomesodermal (EM) GRNs, revealed changes in GRN structure and offered the opportunity to present testable hypotheses for the molecular basis of body plan and cell type evolution across echinoderms [17].
In order to understand how and when the cnidarian endomesodermal GRN is deployed and to define the initial input of the cWnt pathway, we employed a set of complementary approaches (Figure S1). We re-analyzed previously published genes expressed in the pharynx or gastrodermis using a combination of fine scale qPCR for the first 48 hours of development coupled to whole mount in situ hybridization prior to the onset of gastrulation. In order to identify additional putative members of the cnidarian “endomesoderm” GRN, we performed genome wide microarrays on mRNA extracted from embryos in which the canonical Wnt pathway has been activated using two distinct reagents, Lithium chloride (LiCl) and 1-azakenpaullone (AZ). These two pharmaceutical drugs both induce ectopic nuclearization of ß-catenin, but intriguingly, cause significant differences at the molecular and morphological levels. Fine scale temporal and spatial gene expression analysis of newly identified genes in combination with re-evaluated expression data allowed us to draw a first blueprint of putative transcriptional interaction in the presumptive cnidarian endomesoderm (gastrodermis). Finally, using complementary knockdown experiments, we investigated the earliest input of the cWnt pathway into the first non-bilaterian endomesoderm GRN. While inhibition of cWnt blocks pharynx formation, affects endomesodermal gene transcription and is required for spatial restriction of gene expression domains within the animal hemisphere prior to gastrulation, our global analysis suggests that proper specification of endomesoderm in N. vectensis also requires activation of both FGF and BMP, but not Notch, signaling pathways.
Activation of the cWnt pathway can be induced by inhibition of Gsk3ß using pharmaceutical or chemical components. In order to compare the concentration dependent effects of two Gsk3ß inhibitors, lithium chloride (LiCl) and 1-azakenpaullone (AZ) we treated zygotes with increasing concentrations of LiCl and AZ and analyzed their effects on expression of NvfoxB (an oral/pharyngeal marker [42]) in the presumptive oral endomesoderm) and NvfgfA1 (an aboral pole marker [60], [61]) at 24 hpf, prior to the onset of gastrulation and the appearance of endomesoderm (Figure 1, Table 1).
With the exception of embryos treated with 100 mM LiCl that appeared developmentally delayed (Figure 1F, 1L), the general external morphology of the AZ and LiCl treated embryos (Figure 1B–1E, 1H–1K, 1N–1R, 1T–1X) resembled blastula control embryos (Figure 1A, 1G, 1M, 1S). Both treatments induced in a concentration dependent manner an extension of NvfoxB expression towards the vegetal hemisphere (Figure 1B–1E, 1N–1R) and a decrease in Nvfgfa1 expression (Figure 1H–1K, 1T–1X), compared to control embryos (Figure 1A, 1G, 1M, 1S). However, while Nv-fgfA1 expression was undetectable in AZ treated embryos at 10 µM and 30 µM (Figure 1W, 1X) its expression appeared only slightly reduced in LiCl treated embryos at the highest concentrations (Figure 1J, 1K). Based on the strong expansion of Nv-foxB expression in 30 mM LiCl and 10 µM AZ treatments (Figure 1D and 1Q, Table 1) we utilized these treatments for further developmental and molecular characterization.
To compare the effects of LiCl and AZ on ß-catenin nuclearization in N. vectensis, we injected mRNA encoding a GFP tagged form of Nvß-catenin (Nvßcat:GFP) [59] (Figure 2A), treated the injected uncleaved zygotes with either LiCl (30 mM, Figure 2E) or 1-azakenpaullone (10 µM, Figure 2I) and determined nuclear localization of ß-catenin at 24 hpf. As previously described [59], Nvßcat:GFP was uniformly expressed during early cleavage stages (data not shown), then progressively degraded in one hemisphere of the embryo and localized to the nuclei of cells in the presumptive endomesoderm (animal pole) prior to the onset of gastrulation (Figure 2A, Figure S2 [59]). In both treatments (Figure 2E, 2I), the domain of nuclear localization of Nvßcat:GFP was drastically expanded compared to control embryos. However, in LiCl treated embryos the nuclear localization of ß-catenin did not appear to extend all the way to the vegetal pole (aboral pole, Figure 2E), while in AZ treated blastula stages all cells of the embryo showed nuclear staining (Figure 2I).
Treatment of embryos with either LiCl or AZ did not cause any visible developmental perturbation for the first 48 hours post fertilization and the embryos gastrulated normally (Figure 2B, 2C, 2F, 2G, 2J, 2K). However after four days of development when control embryos reached the planula stage (Figure 2D), we distinguished two clear phenotypes resulting from the treatments. LiCl treated embryos became elongated with an increased amount of disorganized endomesodermal tissue and were devoid of any definite pharyngeal structure (Figure 2H, [59]). In contrast, AZ treated embryos displayed presumptive pharyngeal structures and endomesoderm everting from the oral pole, causing progressive exogastrulation after 10 days of development (Figure 2L, Figure S3). In AZ treated embryos the formation of endomesoderm increased at the expense of ectodermal tissue. The extension of Nv-foxB expression and nuclear ß-catenin localization towards the vegetal pole suggests a shift of the endomesoderm-ectoderm boundary and may involve changes in proliferation rates of endomesodermal cells (Figure 2L). Both of these treatments reinforce the idea that interfering with cWnt signaling affects endomesoderm formation in N. vectensis development. However, the distinct phenotypes suggested differences in either the efficacy or specificity of drug interaction.
Taken together these results support previous ideas of an ancestral role of Wnt/ß-catenin in endomesoderm specification and axial patterning in N. vectensis [50], [59] and suggest that AZ might be more effective than lithium in affecting the cWnt pathway.
In order to identify genes expressed in the presumptive endomesoderm of N. vectensis, and to analyze in more detail the similarities (and differences) in Gsk3ß inhibition using different reagents, we treated zygotes with either AZ or LiCl, extracted RNA prior to the onset of gastrulation (24 hpf) and screened an expression array designed to represent all protein coding genes in the N. vectensis genome. Out of 24,021 represented genes in our Nimblegen (Inc.) expression microarray, we selected genes with a significant 2-fold or greater change compared to the wild-type controls in the average of two biological replicates. Although the Pearson's correlation factors between biological replicates were low (0.53 and 0.42 for the AZ and LiCl arrays respectively), a total of 399 or 411 genes were significantly (P<0.05) upregulated in AZ or LiCl treated embryos, respectively, while 362 or 256 genes were significantly (P<0.05) down regulated in AZ or LiCl treated embryos, respectively (Table S1). To gain insight into the percentage of genes that are affected by either one of the cWnt activating treatments, we compared the two datasets to determine the degree of overlap of significantly up- or downregulated genes (Figure 3A, 3B).
Surprisingly, from the total of 731 unique significantly upregulated genes, only 79 genes (10.8%) were shared in both datasets. Of the remaining 652 genes, 303 genes (41,5%) were upregulated by AZ but not by LiCl and 349 genes (47.7%) were upregulated by LiCl but not by AZ (Figure 3A). Similarly, from a total of 538 genes that were significantly downregulated in both treatments, 132 genes (25.7%) were unique to LiCl, 282 genes (52.4%) were unique to AZ and only 124 genes (23%) were shared between the two treatments (Figure 3B).
Both compounds are supposed to target the ATP-binding pocket of Gsk3ß [62] and have been used in a wide range of organisms to study the role of cWnt signaling during early development [55], [63]–[66], regeneration [67] and cells in culture [68], [69]. Previous biochemical studies have described the difference in Gsk3ß affinity of AZ and LiCl [62] and shown that lithium chloride has additional targets such as inositol-phosphate phosphatases [70]. In order to gain insight into which Gsk3ß-inhibiting treatment in N. vectensis may be more specific to cWnt activation we over-expressed a stabilized form of Xenopus ß-catenin-GFP (Xßcat69:GFP, [50], [59] in which the GSK-3ß/CK-1 phosphorylation sites had been mutated to alanines and is resistant to proteolytic destruction [71].
In contrast to LiCl, but similar to AZ treatments, over-expression of Xßcat69:GFP mRNA induced ectopic localization of its protein in the nuclei of all cells along the oral-aboral axis (Figure 3C) and caused a strong exogastrulation phenotype after 4 days of development (Figure 3D–3F). In addition, expression of Nv-foxB in Xßcat69:GFP mRNA injected embryos was strongly expanded (Figure 3G), and Nv-fgfA1 expression downregulated (Figure 3H) similar to that seen in AZ treatments (Figure 1Q, 1W). These observations suggest that in N. vectensis the effects caused by AZ treatments may reflect a more specific activation of the cWnt pathway than LiCl, although a more thorough analysis perhaps including other commonly used Gsk3ß inhibitors such as alsterpaullone [72]–[75] is required to identify the best cWnt activator in this system.
cWnt signaling has previously been shown to be involved in endomesoderm formation in N. vectensis [50], [59] and ectopic activation of the pathway not only induces exogastrulation (Figure 2L, Figure 3F) but also the expansion of at least one endomesodermal transcription factor in the animal hemisphere prior to the onset of gastrulation (Figure 1D and 1Q, Figure 3G). To determine additional transcriptional differences between nß-catenin stabilized and control embryos with the goal of identifying putative genes that are required for specification and formation of endomesoderm in N. vectensis, we used gene profiling with a N. vectensis specific oligonucleotide based genome-wide microarray (Nimblegen, Inc). We chose to analyze differential expression in late blastula stages prior to the onset of gastrulation (24 hpf) of AZ and LiCl treated embryos. Transcription factors and signaling molecules build the basis of complex gene regulatory network that are deployed during embryogenesis [14], [76]. Therefore, we focused on the identification and characterization of genes that can be separated in the following classes: i) transcription factors, ii) signaling molecules (ligands and receptors) and iii) signaling pathway modulators (extracellular, membrane bound or cytoplasmic), that will constitute the main structure of the cnidarian endomesoderm GRN. Although the specificity of LiCl to activate the canonical Wnt pathway is questionable, at least one gene expressed in the presumptive endomesoderm, Nv-foxB, was visibly upregulated in embryos treated with that chemical (Figure 1D). For the purpose of identifying the largest possible set of new genes putatively playing a role in the gene regulatory network underlying endomesoderm formation in N. vectensis, we included microarray data from LiCl as well as AZ treatments that displayed at least a 2-fold upregulation from two biological replicates (Table S1).
Of the 731 genes identified as being upregulated by LiCl or AZ treatments, 104 unique genes belonging to distinct definitive/putative transcription factors or signaling molecules (Table 2) met our selection criteria for detailed characterization.
The majority of the selected genes (∼66%, 64/104) belonged to various families of transcription factors (Table 2), defined by their structure and DNA binding motifs, and involved in diverse developmental and biological processes. The largest group of transcription factors we selected belongs to the homeodomain containing molecules (28/64, e.g Nvevx, Nvhd050, NvhlxB9) that constitute an ancient class of regulatory genes with diverse roles in fungi, plants and animals [77]. Other transcription factors that were upregulated following Gsk3ß inhibitor treatment prior to gastrulation in N. vectensis belong to the Forkhead (e.g NvfoxQ1, NvfoxA, NvfoxB), T-box (e.g Nvtbx20-like, Nvbra), Ets (e.g NvelkA-like), Mad1 (e.g Nvsmad4-like, Nvnfix-like), HMG (e.g Nvtcf), zinc finger (e.g NvsnailA), bHLH (e.g Nvtwist, Nvhes3) or achaete-scute (e.g NvashB). These data indicate that a diverse set of transcription factor families may be involved in endomesoderm formation during cnidarian development (Table 2).
The Wnt, Hedgehog (Hh), RTK (Receptor Tyrosine Kinase, e.g. FGFR), Notch, Tgfß/Activin and Bmp signaling pathways are associated with diverse biological events during embryonic development in metazoan and have been previously described from N. vectensis [50], [56], [59], [78], [79]. With the exception of Notch signaling, putative ligands and/or receptors associated with all remaining pathways have been upregulated by ectopic canonical Wnt activation (Table 2). In particular, we identified 9 of the 13 described N. vectensis Wnt ligands [56], [80], Nvactivin [81], three Activin/TGFß Receptor-like genes, Nvbmp2/4 [81], Nvadmp-related, one Bmp Receptor-like gene, Nvfgf8A [61], two FGF-like, three Tyrosine Kinase Receptor-like genes, Nvhint3 [82] and one Patched-like receptor gene (Table 2). Interestingly, we also identified Nvfollistatin [81] a putative modulator of Activin [83], Nvsprouty3-like a putative modulator of FGF [84], as well as three modulators of Wnt signaling, Nvaxin-like, Nvnkd1-like (naked cuticle) and Nvporcupine-like [85]–[87], suggesting that these three signaling pathways (Activin, BMP and FGF), in addition to cWnt signaling, are deployed to specify and pattern the early N. vectensis embryo.
53 of the 104 genes identified above have been previously isolated, however only 23 have had their expression pattern characterized (e.g. Nvbrachyury, NvfoxA, Nvtcf/lef [26], [39], [50], [88]). All but two (Nvhint3 [82] and Nvhes3 [79] of the 23 previously characterized genes are expressed in endomesodermally related regions during development, demonstrating the effectiveness of the approach in N. vectensis.
Previous work in N. vectensis has shown that there appears to be at least two distinct complementary expression domains within the animal plate that give rise to endomesdoerm prior to gastrulation: i) the central domain, located at the animal pole of the embryo and characterized by NvsnailA expression and ii) the central ring expressing NvfoxA that surrounds the central ring [26], [39], [48]. To gain a basic understanding of when and where the transcription factors and signaling molecules with potential roles in endomesoderm formation are expressed in the developing embryo, we performed whole mount in situ hybridization (Figure 4, Figure 5).
We combined genomic sequence information with available EST data to design primers for the longest possible probes and were able to subclone and synthesize Dig-labeled antisense probes for 73 of the 104 identified genes. 49 of the 73 genes had never been characterized before in N. vectensis. In order to analyze their expression pattern and determine their putative implication in the N. vectensis endomesoderm GRN, we performed in situ hybridization focusing on the late blastula stage (24 hpf) (Figure 4). This embryonic stage is the same stage that was used to perform the initial microarray experiments that lead to the identification of the genes and corresponds to the timing in which the presumptive endomesoderm is specified.
We identified 18 new genes expressed in defined domains within the presumptive endomesoderm (Figure 4A–4R) that were upregulated by treatments described to affect cWnt signaling. Two genes (Nvhd043 and Nvngfr-like) were expressed in the gastrodermis at the late gastrula stage (http://www.kahikai.org/index.php?content=genes) but we were unable to detect differentially localized gene expression for the 29 remaining probes during the first 48 hours of development after fertilization. From the 20 genes that displayed localized expression, eleven were exclusively induced by AZ, five exclusively by LiCl and four by both treatments (Table 2).
Although it was difficult to identify sharp boundaries of expression for a few genes (e.g. Nv-smad4-like, Nv-unc4-like and Nv-foxQ1) at the blastula stage, detailed analysis of animal views of the expression patterns revealed that the newly identified genes could also be characterized as being expressed in one of these two domains (Figure 4A–4R insets) that may constitute distinct synexpression groups [89]. Fourteen genes (Nvtbx20-like, Nvadmp-related, Nvvasa-like, NvduxABC, Nvnk2-like, Nvsmad4-like, NvelkA-like, Nvhd050, Nvphtf1-like, Nvporcupine-like, Nvnk-like 13, Nvnfix-like, Nvunc4-like and NvfoxQ1 (Figure 4A–4N)) were expressed in the central domain, the transcripts of two genes (Nvhd147 and NvbicaudalC-like1 (Figure 4O, 4P) were detected in the central ring surrounding the central region, while Nvaxin-like and Nvnkd1-like appeared to be expressed in cells spanning both territories (Figure 4Q, 4R).
In order to establish the ground work for analyzing the gene regulatory network underlying endomesoderm specification/formation that includes the largest possible number of candidate genes, we re-analyzed spatial gene expression with longer probes at 24 hpf (blastula) of 51 formerly published genes (Table S2, highlighted in green). From all re-analyzed genes, we obtained clear expression patterns prior to gastrulation (Figure 5) for 33 genes: the transcription factors NvotxA, NvotxB, N-otxC, Nvsmad1/5, NvsnailA, NvsnailB, Nvgli, Nvgsc, NvhlxB9, NvashB, Nvevx, Nvbra, NvfoxA, NvfoxB, Nvtcf, Nvlmx, Nvlhx1, the signaling molecules and receptors, Nvfgf8A, Nvfz10, Nvbmp2/4, Nvwnt3, Nvwnt2, Nvwnt4, Nvwnt8, NvwntA, Nvstrabismus the modulators of FGF and BMP signaling, Nvsprouty, Nvtolloid, Nvchordin and putative germ line specific markers Nvpl10, Nvnanos2, Nvvasa1 and Nvvasa2. In addition, the genes Nvactivin, NvmoxD, Nvrepo, Nvwnt1, Nvwnt11, and NvWnt16 [80], [81], [90]–[92] show faint expression in the animal hemisphere but require additional analysis to confirm a localized expression at the blastula stage (data not shown).
Systematic analysis of animal views of the obtained expression patterns allowed us to extend the number of genes that belong to the above-mentioned co-expression groups within the animal hemisphere. Eighteen genes NvotxA, NvotxB, NvotxC, Nvpl10, Nvsmad1/5, Nvnanos2, NvsnailA, NvsnailB, Nvsprouty, Nvvasa1, Nvvasa2, Nvgli, Nvgsc, Nvfgf8A, Nvfz10, Nvtolloid, NvhlxB9 and Nvevx (Figure 5A–5R) are expressed in the central domain. The transcripts of nine genes Nvwnt3, Nvbmp2/4, Nvbra, NvfoxA, NvfoxB, Nvwnt8, NvwntA, Nvtcf, and Nvlmx (Figure 5S–5Za) are detected in the central ring surrounding the central domain, while NvashB, Nvstrabismus appeared to be expressed in cells spanning both territories (Figure 5Zb, 5Zc). The genes Nvwnt4, Nvwnt2, Nvlhx1 and Nvchordin are expressed in a third domain defining the animal hemisphere, the external ring (Figure 5Zd–5Zg).
While we confirmed localized expression at the blastula stage for NvotxB, Nvsmad1/5, NvsnailA, NvsnailB, Nvsprouty, NvfoxA, NvfoxB, Nvtcf, NvashB and Nvlhx1, (Figure 5B, 5E, 5G, 5H, 5I, 5V, 5W, 5Z, 5Zf) [26], [39], [40], [42], [48], [50], [61], [81], [93]–[95] we also detected an earlier onset of gene expression than previously reported for NvotxA, NvotxC, Nvpl10, Nvnanos2, Nvvasa1, Nvvasa2, Nvgli, Nvgsc, Nvfgf8A, Nvfz10, Nvtolloid, NvhlxB9, Nvevx, Nvwnt3, Nvbmp2/4, Nvbra, Nvwnt8, NvwntA, Nvlmx, Nvwnt4, Nvwnt2 and Nvchordin (Figure 5A, 5C, 5D, 5F, 5J, 5K, 5L, 5M, 5N, 5O, 5S, 5T, 5U, 5X, 5Y, 5Zb, 5Zc, 5Zc) [40], [56], [61], [80]–[82], [90]–[92], [96]–[100] (Table S2).
Taken together, our systematic gene expression analyses of 18 new and 33 previously identified genes (Figure 4, Figure 5) define at least four complementary expression domains (central domain, central ring, central domain+ring, external ring) within the animal hemisphere at the blastula stage (Figure 6A, 6B).
Because in situ hybridizations are not the most sensitive way to detect the onset of gene expression we used qPCR in order to gain a more precise idea about the temporal expression on cDNA made at embryonic stages sampled every two to four hours, up to 48 hpf. As a frame of reference, embryos at 8 hpf, 18 hpf and 24 hpf contain approximately 430, 2160 or 3480 nuclei respectively (Figure S4). Collected data were analyzed for the presence of maternal transcripts (Cp value>34.00) in unfertilized eggs and, if detectable, for their first zygotic expression inferred from positive changes in transcript levels (Figure 6C, Figure S5). Maternal transcripts were detected for 42.5% (31/73) of the analyzed genes, no significant zygotic upregulation observed for 8.2% (6/73) while only one maternally expressed gene, Nvtcf, appears to be zygotically expressed after the onset of gastrulation 32–40 hpf (Figure 6C). The remaining genes (89%, 65/73) are zygotically upregulated between 8 and 24 hpf, with NvashB, Nvbra, NvfoxB, NvduxABC (Figure 6C), Nvhd043, Nvhd032 and NvmoxC (Figure S6A) being the first upregulated genes 8–10 hours post fertilization. Zygotic expression of 29 genes (Nvbmp2/4, Nvfgf8A, Nvnfix-like, NvfoxA, Nvfz10, Nvhd050, Nvhd147, NvhlxB9, Nvlhx1, Nvlmx, Nvnkd1-like, NvsnailA, NvsnailB, Nvvasa-like, Nvvasa2, Nvwnt2, Nvwnt3, Nvwnt8, Nvsprouty (Figure 6C), Nvactivin, NvfoxA/B-like, Nvhes3, Nvtwist, Nvwnt1, Nvwnt11 and Nvwnt16 (Figure S6A) are detected only a couple of hours later, 10–12 hpf (Figure 6C, Figure S6A). An additional three waves of zygotic upregulation were observed at 14–16 hpf (Nvevx, Nvfoxq1, Nvgli, Nvnk2-like, NvotxB, Nvsmad1/5, Nvstrabismus, Nvtbx20-like, NvwntA (Figure 6C), Nvfollistatin-like, Nvhd017, NvmoxD, NvmsxB, and Nvrepo (Figure S6A), 16–18 hpf (Nvwnt4, NvbicaudalC-like1, Nvporcupine (Figure 6C) and Nvgata (Figure S6A)), and just prior the onset of gastrulation at 20–24 hpf (Nvchordin, NvelkA-like, Nvgsc, Nvnanos2, Nvnk-like13, NvotxA, NvotxC, Nvpl10, Nvtolloid, Nvunc4-like (Figure 6C), Nvfgf8/17-like and Nvtbx15-like (Figure S6A).
Transcripts of genes zygotically activated during the first 5 waves of expression (8–10, 10–12, 14–16, 16–18 hpf) are localized to one of the four animal hemisphere domains at 24 hpf (Figure 6C, Figure S6). With the exception of Nvchordin that is expressed in the external ring, 90% (9/10) of the genes zygotically upregulated at 20–24 hpf are expressed in the central domain, suggesting the beginning of segregation events that define distinct domains within the animal hemisphere at this time of embryonic development in N. vectensis.
A spatial and temporal co-expression map (Figure 7) summarizes our expression data analysis (in situ hybridization and qPCR) and provides a visual representation of the sequential deployment of the putative members of the cnidarian endomesoderm GRN. The distinction of three co-expression domains within the animal hemisphere has only been determined for the blastula stage at 24 hpf (Figure 4, Figure 5). We assume that genes we analyzed that were detected ubiquitously may also have a defined (not necessarily exclusive) role in the presumptive endomesoderm/animal hemisphere prior to gastrulation. We have organized the genes thought to be involved in endomesoderm formation by their maternal presence and zygotic upregulation in presumptive endomesoderm during the first 48 hours of development and by the co-expression group they belong to at 24 hpf (Figure 6A, 6B).
We have shown that treatments designed to ectopically activate the cWnt pathway can be used to identify genes expressed spatially and temporally consistent with involvement in a putative cnidarian endomesodermal GRN. In order to specifically analyze the effect of disrupting canonical Wnt signaling at the phenotypic and transcriptional level in N. vectensis and to determine provisional inputs of that pathway into the cnidarian endomesoderm GRN prior to the onset of gastrulation, we injected morpholino antisense oligonucleotides targeting the translation initiation site of the canonical Wnt effector NvTcf (MoTcf_trans) (Figure 8A). While control (Figure 8C–8E) and dextran injected embryos (not shown) gastrulate normally and form distinct pharyngeal structures (arrows in Figure 8E), MoTcf_trans injected embryos (Figure 8F–8H) gastrulate but fail to form a pharynx (Figure 8H). Previous reports using various approaches to inhibit cWnt signaling in N. vectensis have shown that the gastrodermis initially forms normally but later loses its epithelial organization [50], [92]. In contrast, in Nv-Tcf morphants, the body wall endomesoderm went ahead and formed a monolayer of epithelial cells (Figure 8H), suggesting only a partial effect of NvTcf knock down.
In order to verify the efficiency of the translational MoTcf_trans that targets a region spanning the 5′ UTR and the translational initiation site of Nvtcf, we performed a series of experiments (Figure 9). We made two constructs of NvTcf fused to the fluorescent protein Venus: i) NvTcf:Venus, lacking 15 nucleotides of the morpholino recognition site and ii) Nv-Tcf5′:Venus that contains the entire 5′UTR+ORF region targeted by MoTcf_trans (Figure 9A). When mRNA encoding Nvtcf:Venus (400 ng/µl) was injected alone or in presence of MoTcf_trans (1 mM), we observed nuclear localization of NvTcf:Venus in all the cells at the blastula stage (Figure 9B, 9C). In contrast, nuclear localized NvTcf5′:Venus (Figure 9D) was no longer detected when co-injected with MoTcf_trans (Figure 9E). These results show that MoTCF_trans effectively inhibits translation of a synthetic mRNA encoding Nvtcf (sequence based on genome prediction corroborated by EST data) and that Nv-tcf:Venus mRNA is not recognized by MoTcf_trans making this construct suitable for the following rescue experiments (Figure 9F).
When we injected Nvtcf:Venus (400 ng/µl) alone we observed no significant variation in expression of four genes putatively downstream of canonical Wnt signaling (Nvlmx, Nvbra, NvfoxA and Nvnkd1-like) by qPCR compared to dextran injected control embryos (Figure 9F). The only exception was Nvbra, which was slightly downregulated, reflecting the repressive capacity of Tcf in the absence of nß-catenin [101]. Microinjection of MoTcf_trans (1 mM) causes a downregulation of all four of these genes, while co-injection of Nvtcf:Venus together with MoTcf_trans restores similar expression levels compared to the injection of Nvtcf:Venus alone (Figure 9F). While NvotxA (a gene not affected by ectopic Wnt activation) is slightly upregulated in Nvtcf_Venus injections, it remains unaffected following knock-down or rescue conditions (Figure 9F). Taken together, these data support the idea that MoTcf_trans can effectively block translation of Nvtcf:Venus and that the observed effects on reduced gene expression in MoTcf_trans injected embryos are primarily caused by the inhibition of NvTcf function (Figure 9F).
Nvtcf transcripts are strongly detected in the egg and during early cleavage stages ([56], Figure 6C2) suggesting that the presence of maternally loaded Nv-Tcf protein may circumvent the translational morpholino approach we used to knock-down NvTcf function. In order to interfere with maternally presence of NvTcf, we injected mRNA encoding a dominant negative form of NvTcf fused to Venus (Figure 8A, Nvdntcf:Venus) lacking a 92 amino acid region of the N-terminus that contains the ß-catenin binding domain required for proper signal transduction of canonical Wnt signaling [102]. While injection of Nvdntcf:Venus into the egg clearly induced nuclear localization of Venus in all cells of the blastula stage (24 hpf, Figure 8B) no effect was observed on early invagination and gastrulation movements (Figure 8I, 8J). However, similar to MoTcf_trans injections, 4 day old Nvdntcf:Venus planula larvae (96 hpf) lacked an identified pharynx in over 90% (30/32) of the cases, with no mouth opening observed in appoximately 50% (15/32) of injected embryos (Figure 8K). Intriguingly, in 30% (11/32) of cases we observed various degrees of exogastrulation (Figure S8B, S8C), in addition to the lack of pharynx. When injected at slightly higher concentrations (450 ng/µl) the endomesoderm loses his epithelial organization (Figure S7D), similar to earlier observations of inhibition of cWnt [50], [92] that may eventually lead to apoptosis of the cells [103].
The morpholino (MoTcf) and dominant negative (NvdnTcf:Venus) based approaches we used to interfere with Nv-Tcf function did not perturb gastrulation movements but clearly affected pharynx formation. In Nvdntcf:Venus injected embryos we also observed the absence of a mouth opening in addition to a disorganized gastrodermis, supporting the idea that the dominant negative approach interferes with the maternal pool of NvTcf and is thus a more effective strategy to study the role of this gene during early N. vectensis development.
In this study we took advantage of the growing number of molecular and functional resources in the cnidarian sea anemone N. vectensis to establish the framework for the first provisional GRN underlying endomesoderm (EM) formation in a non-bilaterian metazoan. We used ectopic activation of cWnt signaling (using two different approaches) to identify new putative members of the GRN underlying endomesoderm specification in N. vectensis, combined high density temporal gene expression profiling by qPCR as well as detailed spatial expression analysis by in situ hybridization to build the framework for the EM GRN. Furthermore, we initiated a functional dissection of potential network components by using antisense oligonucleotide morpholino and mRNA (encoding a dominant negative form of NvTcf) injection to detect downstream targets of Wnt/ß-catenin signaling prior to the onset of gastrulation. The main observations from this study are: i) Gsk3ß inhibition using either AZ or LiCl treatments induces significantly different developmental endomesodermal phenotypes at the morphological and molecular levels, ii) within the animal hemisphere at the blastula stage, N. vectensis is already subdivided in at least four co-expression domains prior to the onset of gastrulation, iii) canonical Wnt activation in the animal hemisphere is essential (direct or indirect) for normal expression of some, but not all, genes belonging to all four co-expression groups, iv) cWnt activation appears essential for specifying cell types in the vegetal hemisphere as well as derivatives of the animal hemisphere, and v) that at least two other signaling pathways appear to be involved in particular components of endomesoderm specification.
It is currently too early to make assumptions about the evolutionary changes in network wiring, especially the network circuitry important for particular processes [104] leading to the formation of true mesoderm in bilaterians. Additional gene specific functional and epistasic studies in N. vectensis are required to obtain a better understanding of the genetic interactions of endomesodermal genes that will serve as a comparative basis. However, this current study already provides data to point out several conserved features as well as some differences from other endomesodermal GRNs.
The Gsk3ß/APC/Axin protein complex plays a crucial role in regulating the cytoplasmic pool of ß-catenin and inhibition of that complex by its naturally interacting protein, Dsh (disheveled). This complex is also the target of a variety of pharmaceutical drugs causing the activation of canonical Wnt signaling. Historically, lithium chloride (LiCl) was used to inhibit Gsk3 function, mimic Wnt signaling and interfere with sea urchin, zebrafish and Xenopus development [105], [106]. While currently more than 30 different pharmalogical Gsk3 inhibitors have been described and characterized biochemically [62] only a handful of reagents (lithium chloride (LiCl), 1-azakenpaullone (AZ), 1-alsterpaulllone (AP) and 6-Bromoindirubin-30-oxime (BIO) are commonly used in developmental and cellular [107] studies. The IC50 values (the half maximal (50%) inhibitory concentration (IC) of AZ, AP and BIO are comparable (0.004–0.0018 µM), while LiCl requires higher concentration for effective Gsk3 inhibition (∼2000 µM) [62]. Nonetheless, all four components are broadly used in a variety of animals and generally considered universal canonical Wnt activators [59], [72], [74], [105], [108], [109]. While direct comparisons of two or more Gsk3 inhibitors in a single organism are sparse, recent studies in Hydractinia primary polyps (hydrozoan cnidarian) [110], or acoel flatworms [111] have shown that AZ and LiCl or AZ and AP respectively induce similar phenotypes. These results as well as the fact that different Gsk3ß inhibitors are interchangeably used to ectopically activate canonical Wnt signaling in various animals, predict that AZ and LiCl cause comparable developmental perturbations and should affect a largely overlapping pool of downstream targets. Surprisingly, at the molecular level, the genes affected by these treatments in N. vectensis are largely non-overlapping and closer analysis of the morphological phenotype revealed clear differences. While AZ causes an exogastrulation (Figure 2L), LiCl treated embryos become elongated and the internal endomesodermal tissue disorganized (Figure 2H). Both treatments enhance Nv-foxB expression at the blastula stage at the working concentrations (Figure 1D, 1Q) but only AZ has drastic effects on Nv-fgfa1 at the vegetal pole (Figure 1J, 1W). A higher concentration of LiCl is needed to visibly reduce Nv-fgfa1 expression (Figure 1K). Our array data show that only approximately 11% of significantly upregulated genes or 25% of significantly downregulated genes are simultaneously affected by AZ and LiCl treatments (Figure 3A, 3B). One plausible explanation for this observation would be that the concentrations used for the treatments only cause a partial overlap of common targets. However, although only two biological replicates were performed, and the Pearson's correlation factors between biological replicates were low (0.53 and 0.42 for the AZ and LiCl arrays respectively), both our molecular and morphological observations of different phenotypes caused by LiCl or AZ treatment (Figure 1, Figure 2), suggest that these drugs might have radically different modes of action during N, vectensis development. A greater understanding of targets of LiCl action might also lend insight into additional inputs of endomesoderm specification acting in parallel to other signaling systems.
A recent study on N. vectensis suggests that continuous AP treatments for the first 48 hours after fertilization induces a phenotype that is similar to LiCl treated embryos [59]. While the duration of drug application by the authors was different from the continuous treatments of AZ or LiCl in our study, the described similarities between AP and LiCl add another level of confusion on what pharmaceutical drug to use to mimic ectopic canonical Wnt signaling. Interestingly, overexpression of a constitutively active form of ß-catenin, Xßcat69:GFP, causes exogastrulation (Figure 3F) similar to AZ treatments (Figure 2L). These data suggest that AZ may better mimic ectopic activation of ß-catenin than LiCl (and perhaps AP) in N. vectensis. The differences in morphological phenotypes and molecular targets revealed by our array experiments also highlight that these drugs may have additional non-canonical Wnt specific targets in addition to the effect on Gsk3. A broader comparative study that includes a wide range of different Gsk3 inhibitors would be beneficial to better understand which component actually mimics cWnt activation in vivo. Because AZ and LiCl treatment generate different phenotypes and molecular responses, it raises concerns about the interpretation of experiments made with pharmacological treatments, and underlines the importance of gene specific knock-down experiments for making concrete statements about gene function.
The observation that some genes upregulated by AZ/LiCl treatments were also upregulated by NvTcf inhibition (and not downregulated as expected, Figure 10, Figure 11Z, 11Za) further illustrates how misleading ectopic activation experiments that are not followed up by gene specific knock-down analysis can be.
For the sake of identifying putative downstream targets of the canonical Wnt pathway that may be part of the cnidarian endomesoderm GRN, we focused this study on genes that are upregulated by treatment of inhibitors of Gsk3ß and therefore could positively respond to canonical Wnt signaling. However, a total of 538 genes were significantly (2-fold or more) downregulated by ectopic activation of cWnt signaling (Figure 3B, data not shown). One gene that was downregulated in the array data obtained from AZ but remains unaffected in LiCl treatments is a gene expressed in the presumptive apical domain (vegetal pole), NvfgfA1 (Figure 1W, [61]), supporting the different phenotypes and molecular effects observed by these two treatments (Figure 1, Figure 2). A thorough analysis of genes negatively affected by AZ or LiCl treatments will be the focus of a subsequent paper.
A precise understanding of the timing of gene expression and their spatial distribution in the embryo is crucial in order to gain insight into the architecture of developmental GRNs. As our goal was to determine a large framework for future endomesoderm GRN studies in N. vectensis, we carefully analyzed spatial and temporal expression of previously published as well as newly identified genes by in situ hybridization and high-density qPCR (Figure 4, Figure 5, Figure 6).
In some bilaterian embryos, the initiation of the bulk of zygotic gene expression is called the MBT (mid-blastula transition, [112]. While the timing of the MBT seems controlled by the ratio of nuclei to cytoplasm [113]–[115], the pre-MBT embryo is defined by synchronous cell divisions [116], heterochromatically repressed genes [117] and the translation of the maternal pool of mRNA [118]. Interestingly, our systematic gene expression profiling analysis shows that in N. vectensis more than 40% of the endomesodermal genes analyzed are expressed maternally (Figure 6). In addition, of the 66 genes for which we detected zygotic upregulation, none were activated earlier than 8–10 hours post fertilization. While we could have simply not identified earlier zygotically controlled genes, these observations suggest that N. vectensis undergoes an MBT-like event approximately 10 hours post fertilization. Interestingly, the timing correlates with the previously described end of blastula oscillations and the associated shift from synchronous to asynchronous cell divisions in N. vectensis [49]. Additional experiments including a careful analysis of the early cleavage pattern and analysis of the heterochromatic state are however required to better understand the initial zygotic transcriptional control of N. vectensis.
To determine spatial expression patterns and potential clustering of putative endomesodermal co-expression groups we carried out whole mount in situ hybridization at the blastula stage. Figure 6 A, 6B summarizes the presence of at least five clear distinct co-expression groups present in the blastula in N. vectensis: Four in the animal hemisphere and one at the vegetal pole (the apical domain). In the animal hemisphere 32 genes are expressed in the central domain, 11 genes in the central ring, 4 genes in a territory that covers both the central domain and the central ring vegetal to the central ring, and 4 in an external ring (Figure 6A, 6B). The existence of co-expression groups in the animal hemisphere is not only of interest for establishing the endomesoderm GRN but also for our understanding of the putative “blastoporal organizer” in cnidarians. In fact, a recent work using ectopic grafting experiments has shown the potential of the N. vectensis blastoporal lip (a derivate of the central and external rings) to induce a secondary axis suggesting an expression of the same subset of signaling molecules in cnidarian and chordate blastoporal lips as axial “organizers” [119]. While our analysis allowed us to cluster gene expression patterns at the blastula stage to one of the co-expression groups, double in situ hybridization experiments are required to better understand the spatial relationship between genes on a cell-by-cell basis.
A previous study from N. vectensis has shown by double in situ hybridization that the expression domains of the Nvsnail (central domain) genes and NvfoxA (central ring) at the blastula/early gastrula stages do not overlap and proposed that their boundary can be viewed as the boundary between the endomesoderm and ectoderm [48]. In later stages (gastrula/early planula) NvsnailA and NvsnailB are expressed in body wall endomesoderm [26], [39] while NvfoxA is detected in ectodermal portions of the pharynx and the mesenteries [26], [39]. In order to verify the generality of this observation, we compared genes expressed at blastula stages in either the central domain or the central ring, to their expression at the late gastrula/early planula stage (if data available, Table S3). Of the 32 genes expressed in the central domain (including NvsnailA), 12 genes were detected in endomesodermal structures in later stages, 6 genes were expressed in ectoderm related tissue and two genes were associated with endo- as well as ectodermal territories. On the other hand, of the 11 genes expressed in the central ring (including NvfoxA) the majority (8/11) are detected in ectodermal structures and 3 in endomesodermal tissue. While clearly not all genes from this analysis follow a similar pattern to NvsnailA, NvsnailB and NvfoxA, it seems that the gastrodermis forms primarily from the central domain and pharyngeal/oral ectoderm from the central and external ring and support the idea that ectodermal versus endomesodermal structures are being specified prior to the onset of gastrulation. However, transcriptional control of gene expression is context dependent and can quickly change during embryonic development. In fact, NvashB is expressed in the central domain and central ring at 24 hpf (Figure 5Zb), is not detectable during gastrula stages but is re-expressed in the blastoporal ectoderm in planula stages, suggesting differential transcriptional control during embryogenesis [94]. Therefore using gene expression domains at 24 hpf does not provide a clear answer to the cellular fate of the central domain or ring, or their relationship to an ectodermal-endomesodermal boundary. Labeling of the cells belonging to either of the co-expression groups and following them over time is required to definitively address this question.
The comparison of gene expression domains in N. vectensis also reveals something subtler about regional patterning during early development relative to other systems studied. In echinoderms, the basic principle for the origin of the endomesoderm GRN follows four principal steps. Maternal factors activate (1) endomesoderm specific specification genes in the vegetal hemisphere, which after a signal that induces endo- and mesodermal segregation signal activate (2) two distinct sets of endo- or mesoderm specification genes that in turn inhibit (3) the reciprocal specification genes in a given tissue and activate (4) the germ layer specific differentiation genes [9], [120]. This would suggest that in sea urchins once the mesodermal germ layer is differentiated, its specification genes are either downregulated or maintained at basic levels while differentiation genes are upregulated. At the same time endoderm specification genes have to be strongly downregulated in the mesodermal germ layer so as not to interfere with its own specification program. Therefore, no specification genes are expressed in either one or the other germ layer after the segregation signal. The current version of the echinoderm endomesoderm GRN is in agreement with this idea (http://sugp.caltech.edu/endomes/). Our observations in N. vectensis suggest significant differences in the GRN architecture. Not only are endodermal and mesodermal genes expressed in the same gastrodermal precursors (e.g. not repressing each other) (Table S3) but genes of the presumptive endomesoderm (central domain) are later expressed in derivatives of the central ring (ectoderm) and vice versa. These data suggest that in N. vectensis the feedback loop mechanisms for segregation and subsequent specification of two distinct germ layers (endo- and mesoderm) are not operating as they are in triploblastic (bilaterian) animals.
Comparisons of the endomesoderm GRNs from sea urchins and sea stars suggested the existence of a network “kernel”: a conserved GRN subcircuit of five regulatory genes (blimp1, otx, bra, foxA and gataE) that are tightly linked by positive feedback loops. This kernel is required upstream of initial endomesoderm specification and if expression of any of the genes is perturbed, endomesoderm specification is disrupted [17]. In N. vectensis, no Nvblimp1 orthologue is expressed prior to the end of gastrulation (Ormestad & Martindale, unpublished) and Nvgata is not expressed in the animal plate at the blastula stage but only in individual cells of the presumptive ectoderm [26]. The temporal expression of Nvblimp-like after the initial specification of endomesoderm and the spatial expression of Nvgata suggests, that neither of these two transcriptional regulators are part of a putative ancestral kernel for endomesoderm formation. On the other hand, Nvotx (A,B and C), Nvbra and NvfoxA are all expressed in time and space suggesting that they may play a crucial role in specifying this germ layer in this cnidarian. Knock-down experiments analyzing the individual roles of these transcription factors in connecting the network and germ layer specification will shed light on the question about the existence of an endomesderm “kernel” that precedes the bilaterian split.
In order to functionally analyze the role of canonical Wnt signaling during early N. vectensis development, we specifically knocked down NvTcf function using an antisense oligonucleotide morpholino and a dominant negative approach. Overexpression of NvdnTcf:Venus shows that while canonical Wnt signaling has no effect on gastrulation movements (Figure 8), it is required for germ layer specification (Figure 10, Figure 11), proper pharynx and mouth formation (Figure 8H, 8K) and maintenance of endomesoderm (Figure 8, Figure S7 [50], [92]). The lack of oral structures (pharynx and mouth) is in agreement with the expression of Nvtcf in the pharyngeal and blastoporal endomesoderm in late gastrula/early planula stages [56]. One puzzling observation was the exogastrulation phenotype observed in 30% of NvdnTCF:Venus injected planula stages (Figure S7), suggesting that a normal pharynx is required for maintaining the developing endomesoderm inside the planula larvae. However, a properly patterned endomesoderm may also be a pre-requisite for the formation of a normal pharynx. Therefore, additional experiments are required to address the question about the relationships between pharyngeal structures and endomesoderm integrity.
In past studies, the role of cWnt signaling in N.vectenis has been analyzed by interfering with the function of the cytoplasmic/membrane-bound members of that pathway Disheveled (dsh) and Axin, as well as the over-expression of constructs designed to inhibit ß-catenin function (ß-catenin:engrailed fusion (Xßcat-Eng) or the cytoplasmic domain of Cadherin) [50], [59], [92]. With the exception of Cadherin (whose specificity to cWnt remains unclear, [92]) that blocks gastrulation movements and gut formation, over-expression of the other constructs has no significant effects on early gastrulation movements but clearly prevents maintenance of the gut epithelium. The NvdnTcf:Venus injection phenotypes observed in our study are in line with these results. Currently, we cannot rule out that the knock-down experiment from our study, as well as from previous studies [50], [59], [92] are incomplete which may explain the lack of gastrulation phenotype. NvdnTcf:Venus injected embryos show a weak downregulation of Nvstrabismus (Figure 10), a gene that has been shown to be required for gastrulation movements in N. vectensis [92]. However, the current data in N. vectensis [92] and work in another cnidarian [121], [122] suggests that the PCP/Wnt pathway is involved with the morphological aspects of epithelial folding/invagination in N. vectensis and that the cWnt pathway is required for activation of a partial subset of genes involved in endomesoderm specification.
In this study we combined predicted genome-wide microarray approaches (Figure S1, Figure 3, Table 2, Table S1), with precise temporal and spatial gene expression analysis (Figure 4, Figure 5, Figure 6, Figure 7) as well as NvTcf gene specific functional information (Figure 8, Figure 9, Figure 10, Figure 11) to propose the assembly of the framework for the first provisional cnidarian endomesoderm GRN (Figure 12). The current view of endomesoderm specification up to 24 hours post fertilization (Figure 12) allows to clearly distinguish four co-expression domains characterized in this study (Figure 4, Figure 5). No assumptions about direct or indirect interactions are made at this point, and detailed gene specific cis-regulatory analyses are needed to address this question in the future.
NvTcf function is required for normal expression of genes belonging to all four co-expression domains of the animal hemisphere (central domain, central ring, central domain+ring and external ring (Figure 6)). An interesting finding is that most of the genes affected by NvTcf inhibition are expressed in the central ring (e.g. Nvbra, NvfoxA, Nvbmp2/4 and Nvwnt8). This observation is consistent with NvTcf expression in that domain at the blastula stage ([56], Figure 5Za) and with the lack of pharynx formation in NvTcf depleted embryos (Figure 8H, 8K). In addition, NvTcf is crucial for regionalizing the animal hemisphere prior to gastrulation. In fact, analysis of the spatial expression of NvduxABC and Nvfgf8A by in situ hybridization shows that central domain expression of both genes is extended to the central ring (Figure 11T, 11U, 11Z, 11Za) suggesting that NvTcf function is required to restrict NvduxABC and Nvfgf8A expression to the central domain in wild-type embryos.
Endomesoderm GRNs have been proposed for only one protostome (C.elegans, [13], [123]) and three deuterostomes (sea urchin, sea star and Xenopus, [6], [7], [17], [18], [9], [10], [19]). However, for the sake of simplicity, and because early development between N. vectensis and echinoids is in certain aspects comparable [124], we will begin our discussion with echinoderms. However, it is obvious that the GRNs of a broad range of organisms including Xenopus and C.elegans will need to be included in the future.
In echinoderms, a maternal canonical Wnt pathway in the vegetal hemisphere plays a crucial role in patterning the animal - vegetal (A/V) axis and is required for endomesoderm specification and gastrulation [55], [58], [125], [126]. In N. vectensis, genes from all four animal expression domains are downregulated in NvTcf depleted embryos prior to the onset of gastrulation (Figure 10, Figure 11). However, gastrulation movements and invagination of the endomesodermal germ layer is initiated normally in NvTcf depleted embryos (Figure 8, [50], [59], [92]). One reason for normal endomesoderm formation may be that we did not efficiently block maternal NvTcf proteins or the existence of additional signals that specify the endomesoderm in cnidarians. However multiple functional approaches used to inhibit cWnt all failed to prevent gastrulation (Figure 8, [50], [59], [92]). Interestingly, putative molecules activating other signaling pathways are also expressed in the animal hemisphere prior to gastrulation. Nvfgf8A (a putative ligand for Fgf/MAPK signaling) and its putative modulator Nvsprouty are both expressed in the central domain ([61], Figure 5I, 5M). Nvbmp2/4 a putative ligand for Bmp signaling is also expressed in the central ring (Figure 5T) while the potential effector of this pathway Nvsmad1/5 is expressed in the central domain (Figure 5E).
In echinoderms MAPK and Fgf signaling are required to maintain initial cell-autonomous specification of the skeletogenic mesoderm (primary mesenchyme cells, PMCs), specification of a subset of non-skeletogenic secondary mesenchyme cells (SMCs), PMC ingression and differentiation of the larval skeleton [127]–[130]. In contrast, Bmp2/4 signaling is involved in dorso-ventral (oral-aboral) patterning of all three germ layers after the segregation of the mesoderm from endomesodermal precursors [131]–[134].
The role of Bmps has been recently analyzed in N. vectensis and shown a clear implication of NvBmp2/4 in patterning the directive axis (which is perpendicular to the oral/aboral axis) of the endomesoderm and oral ectoderm and patterning and differentiation of the endomesoderm at the late gastrula stages using morpholino approaches [135]. While no delay in gastrulation or morphological signs of a defective endomesoderm was reported from NvBmp2/4 morphants, all endomesodermal markers analyzed in this study were strongly downregulated [135]. This observation is similar to inhibition of cWnt signaling, in that morphogenetic movements of gastrulation and initial gastrodermis formation occurs normally, but endomesodermal markers are no longer detected at the end of gastrulation ([50], [59], [92], this study), suggesting that Bmp2/4 signaling may also be involved in endomesoderm specification prior to gastrulation in N. vectensis. As our experiments interfering with the cWnt pathway show, dominant negative approaches might revel additional roles for these other pathways in early endomesodermal patterning.
Unfortunately, little is known about the early role of Fgf/MAPK signaling in the animal hemisphere in cnidarians, it would be important to analyze the role of NvFgf8A signaling on endomesoderm specification in N. vectensis. Re-analyzing the role of NvBmp2/4 signaling prior to gastrulation and formation of the directive axis may also reveal whether NvBmp2/4 is involved in endomesoderm specification prior to its role in patterning the directive axis. This would considerably improve our basic understanding of the ancestral relationship between three main signaling pathways (Bmp2/4, Fgf/MAPK, and Wnt/Tcf) and underline their respective inputs into the endomesoderm GRN required to form a functional gut in N. vectensis. In the context of our study it appears likely that Bmp2/4 and FGF signaling are likely to be involved in specification of the central domain while Wnt/Tcf is more important for specifying the central ring and its derivatives (e.g. pharynx).
Another very important signaling pathway involved in endoderm and mesoderm segregation from an initial endomesodermal germ layer in echinoderms is the Notch signaling pathway. After initial endomesoderm specification by maternal cWnt, nß-catenin induces the expression of the Notch ligand, Delta, in the presumptive endoderm, which in turn activates the Notch signaling pathways in the neighboring cells (presumptive mesoderm) that actively inhibits cWnt signaling and induces the mesodermal specification program [136]–[138], [8] [16], [139]–[142]. Recently, gene expression of members of the Notch signaling pathway and its role during N. vectensis development have been reported [79]. Using pharmaceutical and gene specific approaches to knock-down Notch signaling this study has shown that this pathway is required for proper cnidocyte (cnidarian-specific neural sensory cells) development. While the endomesoderm in Notch inhibited embryos appeared disorganized during later development, expression of two markers (NvsnailA and NvotxA) was largely unaffected suggesting that initial endomesodermal patterning occurs normally in these animals. This study also suggests that, in contrast to echinoderms, the Notch signaling pathway does not seem to be involved in early germ layer segregation. However, a more detailed analysis of endomesodermal markers prior and during gastrulation after Notch inhibition might be required to fully exclude any important role of that pathway in specifying endomesodermal territories.
To summarize, we have used ectopic activation of canonical Wnt signaling to carry out a genome wide survey of putative members of the cnidarian endomesoderm GRN. In combination with previously described endomesodermal genes we systematically analyzed over 70 genes by in situ hybridization and real time qPCR to establish a set of potential components of an extensive gene expression network. Finally we have used functional NvTcf knock-down experiments to assemble the framework for the first provisional inputs into a complex cnidarian gene regulatory network underlying germ layer formation and show that canonical Wnt function is required to regionalize the animal pole into a central domain, central ring and an external ring at the blastula stage and to allow normal pharynx formation of the early planula. The current view of the network suggests that additional signaling pathways (Bmp2/4 and FGF) are tightly interwoven to correctly specify and pattern the endomesoderm of N. vectensis prior to the onset of gastrulation.
N. vectensis embryos were cultivated at the Kewalo Marine Laboratory/PBRC of the University of Hawaii. Males and females were kept in separate glass bowls (250 ml) in 1/3x seawater (salinity: 12pp) [41]. To keep the animals in a healthy reproductive state, they were kept at 17°C in dark and water was changed weekly. Animals were fed twice a week with oysters or brine shrimps. Manipulating the light cycle induced spawning and oocytes and sperm were collected separately [143]. The gelatinous mass around the eggs was removed with 2–4% L-Cystein in 1/3x seawater before fertilization and then washed 3 times with 1/3x seawater. For a simultaneous development of the embryos, all the oocytes were fertilized in glass dishes at the same time with 0.5 ml of sperm dilution. The fertilized eggs were kept in dark in filtered 1/3 seawater (12pp) at 17°C until the desired stage.
The canonical Wnt agonist 1-azakenpaullone (AZ, Sigma, #A3734) was dissolved at a stock concentration of 10 mM in DMSO and added at final concentrations as indicted (1–30 µM) in 1/3x-filtered seawater. Lithium chloride (LiCl) was dissolved in H2O and added at final concentrations as indicated (1–100 mM) [81]. Embryos were treated with 1-azakenpaullone or lithium chloride directly after fertilization and kept at 17°C. At 12 hours the 1-azakenpaullone and lithium chloride solution were replaced with fresh solutions to maintain activity of the Gsk3ß agonists. The described phenotypes were observed in more than 80% of the analyzed embryos in at least three individual experiments. Treatments were compared to DMSO (for AZ treatments) treated or untreated control embryos. Embryos were fixed for in situ hybridization and morphological analysis at indicated stages. mRNA of embryos was extracted at 24 h after fertilization (late blastula) from two distinct biological replicates for microarray analysis.
RNA for qPCR and microarray analysis was isolated with TriPure (Roche, # 11667157001) or TRIzol (Invitrogen, #15596-026) according to the manufacturer's instructions and genomic contamination removed using RNase-free DNase (Quiagen, #79254) for 15 minutes at 37°C. The total amount of RNA was quantified with a NanoDrop 2000 spectrophotometer (Thermo Scientific) and the quality analyzed with a Bioanalyzer 2100 (Agilent Technologies Inc.). 1 µg of total RNA was used to generate cDNA with the Advantage RT-PCR kit (Clontech, #639506) for qPCR analysis. For the fine scale temporal analysis (Figure 6, Figure S5, Figure S6) total RNA was extracted from the following stages (in hours post fertilization, hpf): 0,2,4,6, 8,10,12,14,16,18,20,24,28,32,40,48.
qPCR analysis using a LightCycler 480 (Roche) utilizing LightCycler 480 SYBR Green 1 Master mix (Roche, #04887352001) was carried out as described previously [94]. Efficiencies for each gene specific primer pair was determined using a five-fold serial dilution series and only primers with an efficiency ranging from 80% to 115% were used for further analysis (Table S4). The houskeeping genes Nvactin and/or Nvgadph were used to normalize relative fold changes between control and manipulated embryos and each qPCR analysis was repeated on independent biological replicates. 20 µg of total RNA was sent to NimbleGen, Iceland for further cDNA synthesis, labelling and array hybridization. The 4-plex microarray (72,000 features) is an oligonucleotide-based chip version, custom designed and produced by NimbleGen Systems (Roche). Gene expression levels were normalized in the Nimblescan software according to [144] and [145] and fold-changes calculated by comparing expression values from control and treated embryos.
Array results were screened based on the provided genome annotations assigned to each array spotID. If no clear blast hit or gene information was assigned to the prediction gene model from the Joint Genome Institute, we retrieved the genomic sequences (http://genome.jgi-psf.org/Nemve1/Nemve1.home.html) for the given gene and performed manually Blast (blastx) searches [146] against the NCBI database to determine the nature of the predicted gene product. All sequences from genes of interest have been used for Blast analysis to confirm their nature and to determine previously published genes.
To distinguish between previously published genes, and newly identified putative TFs and signaling molecules, we used the best Blast Hit identification, followed by “- like” to designate the newly identified gene sequences. In order to verify the potential accuracy of the “best blast hit” naming system, we used published phylogenetic reconstruction techniques to confirm the orthologies of Nv-admp-related, Nvfgf20-like, Nvfgf20-like as well as forkhead transcription factors (see Table 2 for references). Thus, while “Blast hit” approaches can be used to provide a general idea of the protein family, a detailed phylogenetic analysis is required to better resolve these gene orthologies, especially when paralogy issues or when multiple gene predictions are present for one gene family.
The constructs pC2+Nvßcat:GFP and pCs2+Xßcat69:GFP have been described previously [59], [71]. cDNA constructs encoding the wild type ORF (NvTcf), the wild type ORF including 16 nucleotides of the 5′UTR (NvTCF5′) and a dominant negative form (NvdnTcf) lacking 276 nucleotides of the 5′ coding sequence of NvTcf, were generated by PCR. The forward primers used were:
NvTcf_FWD (5′ CACCATGCCTCAGCTTCCTAGGAATTCC 3′)
NvTcf5′_FWD (5′ CACCACATGAGACGGTAGTATGCCTCAG 3′)
NvdnTcf_FWD (5′ CACCATGAACCAGCATGGTAGTGACAGTAAAC 3′)
The reverse primer (5′ GTGTCTGATGTTACTGGATTACTTG 3′) used was lacking the stop codon for fusion with a C-terminal Venus fluorescent tag.
NvTcf cDNA constructs were cloned into pENTR dTOPO vectors (Invitrogen) and subsequently recombined into a C-terminal Venus containing pDEST expression vector [147]. pDest expression vectors were linearized with the restriction enzyme ACC651 and transcribed using the Ambion mMessage mMachine T3 kit (Ambion, AM1348). pCs2+ expression vectors were linearized with the restriction enzyme Not1 and transcribed using the Ambion mMessage mMachine SP6 kit (Ambion, AM1340M). Synthetic mRNA was purified using Megaclear columns (Ambion, AM1908) followed by one phenol-chloroform extraction and isopropanol precipitation. Nvßcat:GFP, Xßcat69:GFP, NvTCF:Venus, NvTCF5′:Venus and NvdnTCF:Venus mRNAs were injected in zygotes at final concentrations of 0.3–0.5 µg/µl.
A morpholino antisense oligonucleotide (Gene Tools) was designed to target a region spanning the 5′UTR and tranlsation inition site of Nv-Tcf (MoTcf_trans: 5′ CTG AGG CAT ACT ACC GTC TCA TGT G 3′, Figure S7). The morpholino was used at 1 mM without noticeable toxicity. Absence of gene expression perturbation after injection of a control morpholino (5′ AGAGGAAGAATAACATACCCTGTCC 3′) at 1 mM has been reported previously [94]. All injections were compared to either rhodamine dextran injected or uninjected control embryos. Microinjections were performed using a PLI-90 Pico-Injector (Harvard Appartus). All embryos developed in 1/3x filtered-seawater at 17°C.
Previously described gene sequences were used to sub-clone into pGemT (Promega, #A3600) from mixed stage cDNA. All other sequences used in this study were isolated in the course of a microarray analysis. Genome predictions as well as EST sequence information were combined to design primers (Table S5) that allow the amplification and cloning of genes between 05.kb and 2 kb as described above. Accession numbers for all analyzed genes in this study can be found in Table 2.
Embryo fixation, probe synthesis and in situ hybridization were performed as previously described [26], [148]. 0.5 kb–2 kb digoxigenin-labelled (Roche, #11573152910) riboprobes were synthesized using the MegaScript Transcription Kit (Ambion). Hybridization of riboprobes (1 ng/µl) was carried out at 62°C in 50% formamide hybe buffer and visualization of the labeled probe was performed using NBT/BCIP as substrate for the alkaline phosphatase-conjugated anti-DIG antibody (Roche, #11093274910). To analyze embryonic and larval morphology, we used Biodipy FL Phallacidin (Molecular Probes/Invitrogen, #B607) and propidium iodide (Sigma, #81845) to stain f-actin and the cell nuclei respectively as described previously [48].
in situ hybridization images were taken on a Zeiss AxioScop 2 mounted with with an Axiocam camera triggered by Axiovision software (Carl Zeiss). All expression patterns described here have been submitted to Kahi Kai, a comparative invertebrate gene expression database [149] hosted at http://www.kahikai.org/index.php? content = genes. Scoring of treatment, overexpression and morphant phenotypes was performed on a Zeiss Z-1 Axio imager microscope and confocal imaging was conducted on a Zeiss LSM710 microscope running the LSM ZEN software (Carl Zeiss). Fluorescent images were false-colored, the fluorescent channels merged using ImageJ (http://rsbweb.nih.gov/ij/) and cropped to final size in Photoshop Cs4 (Adobe Inc.).
|
10.1371/journal.pbio.0050329 | GATA3-Driven Th2 Responses Inhibit TGF-β1–Induced FOXP3 Expression and the Formation of Regulatory T Cells | Transcription factors act in concert to induce lineage commitment towards Th1, Th2, or T regulatory (Treg) cells, and their counter-regulatory mechanisms were shown to be critical for polarization between Th1 and Th2 phenotypes. FOXP3 is an essential transcription factor for natural, thymus-derived (nTreg) and inducible Treg (iTreg) commitment; however, the mechanisms regulating its expression are as yet unknown. We describe a mechanism controlling iTreg polarization, which is overruled by the Th2 differentiation pathway. We demonstrated that interleukin 4 (IL-4) present at the time of T cell priming inhibits FOXP3. This inhibitory mechanism was also confirmed in Th2 cells and in T cells of transgenic mice overexpressing GATA-3 in T cells, which are shown to be deficient in transforming growth factor (TGF)-β–mediated FOXP3 induction. This inhibition is mediated by direct binding of GATA3 to the FOXP3 promoter, which represses its transactivation process. Therefore, this study provides a new understanding of tolerance development, controlled by a type 2 immune response. IL-4 treatment in mice reduces iTreg cell frequency, highlighting that therapeutic approaches that target IL-4 or GATA3 might provide new preventive strategies facilitating tolerance induction particularly in Th2-mediated diseases, such as allergy.
| Specific immune responses against foreign or autologous antigens are driven by specialized epitope-specific T cells, whose numbers expand upon recognition of antigen found on professional antigen-presenting cells. The subsequent maturation process involves the differentiation of certain T cell phenotypes such as pro-inflammatory cells (Th1, Th2, Th17) or regulatory T (Treg) cells, which serve to keep the immune response in check. The current study focuses on the role of two key transcription factors—FOXP3 and GATA3—in controlling the commitment of these cells. We demonstrate that the Th2 cytokine IL-4 inhibits the induction of FOXP3 and thus inhibits the generation of inducible Treg cells. We show that IL-4–induced GATA3 mediates FOXP3 inhibition by directly binding to a GATA element in the FOXP3 promoter. We hypothesize that therapeutic agents aimed at neutralizing IL-4 could be a novel strategy to facilitate inducible Treg cell generation and thus promotion of tolerance in allergies and other Th2-dominated diseases.
| Effective immune responses are characterized by T cell activation, which directs adaptive and innate immune responses to kill pathogens efficiently. Dependent on the pathogen, T cells differentiate into different subtypes, such as Th1 or Th2 cells, which are most efficient in defeating microbial or parasitic invaders respectively. A hallmark of Th1 and Th2 differentiation pathways is the exclusiveness of the individual phenotype leading to either Th1 or Th2, but not to mixed populations. The exclusiveness of this mechanism is provided by a polarization process, where Th2 differentiation inhibits Th1 commitment and vice versa. Specifically interleukin 4 (IL-4)–induced STAT6 and GATA3 inhibit differentiation into Th1 cells in the early phase of commitment [1,2]. GATA3 is sufficient to induce a Th2 phenotype [3] and acts not only through the induction of IL-4, IL-5 and IL-13, the Th2 cytokines, but also through the inhibition of Th1 cell-specific factors [3]. Recently, it was shown that T-bet directly modulates GATA3 function, suggesting that transcription factors compete in the early differentiation phase of T cells, potentially integrating environmental signals to finally imprint the T cell phenotype [4,5]. A GATA3-dominated immune response has been shown to be essential in airway hyperresonsiveness [6] and IL-4–dominated responses can break antigen-specific immune tolerance [7]. Overexpression of a dominant negative form of GATA3 [8] or treatment with antisense-mediated GATA3 blockade [9] decreased the severity of the allergic airway hyper-responsiveness.
The discovery of regulatory T (Treg) cells highlights another phenotype of T cells, which is essential for tolerance against self-antigens. Naturally occurring, thymus-derived Treg (nTreg) cells are generated in the thymus and are assumed to protect against the activity of autoreactive T cells in the periphery. These cells express the forkhead transcription factor FOXP3 and constitutively express CD25 on their surface, but they lack expression of Th1 or Th2 cytokines. Particularly interesting are those Treg cells that are generated in the periphery and thus are potential targets for therapeutic interventions. These induced Treg (iTreg) cells were reported to express FOXP3 [10]. The exact mechanisms of iTreg generation are unclear, but T cell receptor (TCR) triggering has been shown to induce FOXP3 expression and suppressive cells in human [11,12], however the phenotype appears to be of transient nature [13,14]. TGF-β has been demonstrated to be important for the persistent induction of these cells in vitro and in vivo, since animals lacking the TGF-βRII on T cells have fewer peripherally iTreg cells [15] and suffer from a T cell–dependent multiorgan inflammatory disease [16]. Although the effect of TGF-β on natural and inducible Treg cell induction has been demonstrated repeatedly [15,17], its molecular mechanisms remain to be identified.
The current study provides evidence that GATA3 and FOXP3 play a competitive role in iTreg cell commitment as T-bet and GATA3 for Th1 and Th2 differentiation, respectively. We show that GATA3 inhibits FOXP3 induction and that IL-4 limits FOXP3 expression in a GATA3-mediated way, both in vitro and in vivo. We also show that GATA3 directly binds to the FOXP3 promoter and thereby prevents the induction of this gene, demonstrating that Th2 differentiation overrules iTreg induction.
It is assumed that FOXP3 expression can be induced in nonregulatory T cells, which is an important step in iTreg cell differentiation. However, it is not known if all CD4+ T cells have the same capacity to express FOXP3. To investigate whether FOXP3 can be expressed by any T cell subset or if expression is restricted to a distinct lineage, FOXP3 mRNA expression was analyzed in freshly isolated T cells such as CD25-depleted CD4+ cells, CD45RA+ naive or CD45RO+ memory T cells (Figure 1A), as well as T cells driven in vitro toward Th1, Th2, or iTreg phenotypes (Figure 1B; phenotype on Figure S1). The CD4+CD25–, CD45RA+, CD45RO+, and CD4+CD45RO+CD25– were able to significantly induce FOXP3 mRNA up to 30-fold upon TCR activation and addition of TGF-β. Th1 cells showed only a 10-fold increase. In contrast, Th2 cells stimulated under the same conditions did not increase FOXP3 expression. The in vitro generated iTreg cells were unable to further up-regulate FOXP3, which was already at high levels under the resting conditions (Figure 1B, right panel).
Th2 cells are known to produce IL-4 upon activation, which may interact with TGF-β signaling and thus prevent FOXP3 induction. However, the neutralization of IL-4 with a blocking IL-4 antibody did not rescue FOXP3 expression in the differentiated Th2 cells (unpublished data). These data demonstrated that Th2 cells have a limited capacity to express FOXP3 (Figure 1B). The inability of Th2 cells to express FOXP3 was also documented at the single-cell level, confirming that Th2 cells lack FOXP3 expression (Figure 2A). Only iTreg cells expressed FOXP3 in resting conditions. Interestingly, we observed that resting iTreg cells express FOXP3 but show low CD25 surface expression. Repeated exposure to TGF-β did not further increase the FOXP3 expression in the iTreg lineage but transiently induced FOXP3 expression in Th1 cells. Naturally occurring Th2 cells such as CRTH2+ T cells, T cells isolated according to their IL-4 secretion, or an IL-4–producing T cell clone (BR8) were also lacking FOXP3 expression (Figure 2B). Furthermore TGF-β–mediated FOXP3 induction failed in these cells in contrast to the naive T cells (Figure 2B). Because IL-4 is the key Th2 cytokine, the expression of IL-4 and FOXP3 in freshly isolated CD4+ T cells was analyzed by fluorescence activated cell sorting (FACS). IL-4–expressing cells were most abundant among CD45RO+CD25– cells, which did not co-express FOXP3 (Figure 3A, left panel). In contrast, CD45RO+CD25+ cells abundantly expressed FOXP3, while lacking IL-4 (Figure 3A, right panel). As shown in Figure 3B, the frequency of the IL-4+ cells was always below 1% in the FOXP3+ cells close to the background. The IL-4+ cells were confined to the FOXP3– cells, as shown for the CD45RO+CD25–, CD45RO+CD25+, and CD45RO+CD25+high cells (Figure 3B). In addition, neither the Th2 clone (BR8) nor CRTH2 cells significantly expressed FOXP3 (Figure 3C). Cells enriched for their IL-4 secretion using the magnetic cell isolation technology contained some FOXP3-expressing cells, but importantly, the expression did not overlap. Taken together, these data indicate that FOXP3 was not expressed by Th2 cells and was not inducible in those cells.
The limited capacity of differentiated effector cells to induce FOXP3 expression suggests that iTreg induction has to occur before effector T cell differentiation occurs. Therefore, we analyzed the expression of FOXP3 and GATA3 during the differentiation of naive CD4 T cells into Th0 (neutral, anti-IL-4, anti-IFN-γ, and anti IL-12), Th2, and iTreg phenotypes. After initiation of the differentiation process, FOXP3 and GATA3 showed a similar expression kinetic within the first 3 d, which are considered to be critical in T cell commitment [18]. Under Th2 differentiation conditions, FOXP3 mRNA expression increased only marginally (Figure 4A, left panel). Thus, although GATA3 and FOXP3 showed similar kinetics, their expression polarizes at the end of the differentiation process when cells were cultured towards Th2 or iTreg cells, respectively (Figure 4A and 4B). Interestingly, the Th0 cells were expressing more FOXP3 than the Th2 cells, but expressed low levels of GATA3; however, the protein expression slightly differed from mRNA expression, suggesting also posttranslational regulation and degradation as potential additional mechanisms in the differentiation process. The phenotype of iTreg cells included an anergic phenotype upon anti-CD3 re-stimulation (Figure S1A), CD103, CTLA-4, GITR, and PD-1 surface expression (Figure S1B). On the single-cell level, it can be seen that cells progress through a transition phase, where GATA3 and FOXP3 expression coexist to some degree in the same cells, which is resolved in iTreg cells after 7 d (Figure 4B). Taken together, these data demonstrated that Th2 cells have lost their capacity to express FOXP3 and showed that Th2 and iTreg cells arise from two different differentiation pathways.
IL-4 induces differentiation of naive T cells, upon antigen encounter, into the Th2 cell lineage. We therefore asked whether IL-4 is able to inhibit TGF-β induction of FOXP3 during the priming of naive T cells. Human CD4+CD45RA+ T cells were activated with plate-bound anti-CD3/CD28 in the presence of TGF-β and/or IL-4 and harvested after 5 d. IL-4 efficiently repressed the TGF-β–mediated induction of FOXP3 expression (Figure 5A) in a concentration-dependent manner (Figure 5B). Low levels of GATA3 were induced also in the absence of IL-4, as it was previously observed [3]. However, at low concentration, IL-4 was able to marginally induce FOXP3 expression. Of note, GATA3 was also induced in the presence of TGF-β at high IL-4 concentration (Figure 5A and 5B). The IL-4–mediated prevention of FOXP3 expression was not caused by interferences of the receptor signaling, because the phosphorylation of SMAD2 or STAT6 was not affected by the addition of IL-4 and/or TGF-β, which demonstrates that IL-4 as well as TGF-β signaling were functional under these conditions (Figure 5C). Increasing amounts of IL-4 increase intracellular GATA3, whereas FOXP3 decreased, which is consistent with the mRNA analysis (Figure 5D). Furthermore, injection of IL-4 into wild-type B6 mice decreased the inducible or natural Treg number in vivo. A distinction of the Treg subsets is not possible, because recently activated iTreg cells also express surface CD25. We used complexes of recombinant mouse IL-4 (rmIL-4) plus anti-IL-4 monoclonal antibodies (mAbs), which have been shown to dramatically increase the potency of the cytokine in vivo [19]. In these mice, the percentage of CD4+CD25+ and FOXP3+ T cells dramatically decreased when the antibody-cytokine immune complexes were injected (Figure S2). Upon administration of rmIL-4 plus anti-IL-4 mAb complexes, the total number of CD4+CD25+ T cell, as well as the Foxp3+ T cells diminished by half (Figure S2G and S2H), confirming that the lower percentage was not due to an increase in the CD4+CD25– cells, but a real decrease of CD4+CD25+ T cells. In conclusion, IL-4 negatively regulates the natural or inducible Treg cell turnover not only in vitro but also in vivo. To study the effects of IL-4 on already-existing human natural or inducible Treg cells, we exposed sorted CD25+ T cells (nTreg cells) to IL-4 and analyzed FOXP3 expression and suppressive capacity. In already-existing Treg cells, IL-4 failed to inhibit FOXP3 expression (Figure S3A), and the suppressive capacity was not altered (Figure S3C). Similarly pre-existing iTreg cells did not decrease FOXP3 expression upon IL-4 exposure (Figure S3B).
Although TGF-β–reduced CD25, IL-4 expression, and CD25 expression (Figure 6B), IL-4 significantly inhibited TGF-β–mediated induction of FOXP3 in naive T cells driven toward FOXP3+ T cells, as shown by FACS analysis (Figure 6A). It is known that IL-4 is a potent growth factor and may therefore favor the proliferation of FOXP3– cells and thus decrease the relative percentage of FOXP3+ cells. However, analysis of cell division kinetics by CFSE-labeling demonstrated that IL-4 did not differentially promote cell growth of FOXP3+ over that of FOXP3–. In fact both populations showed similarly enhanced proliferation (Figure 6A). Furthermore the TGF-β–mediated induction of FOXP3 expression was not caused by overgrowth of a CD25–FOXP3+ minority, since the number of FOXP3+ cells was low/absent in the purified CD4+CD45RA+ T cells (between 0% and 1%), and the FOXP3+ cells were not confined to the highly divided cells. CD25 was down-regulated in TGF-β–treated cells compare to activated T cells, which was even more pronounced in cells cultured with TGF-β and IL-4.
The addition of IL-4 to iTreg-driving conditions decreased the number of FOXP3+ cells (Figure 6B). In line with the previous findings, the IL-4–producing cells and the FOXP3 expressing cells are nonoverlapping populations. Since FOXP3 is known to act as a repressor of cytokine expression [20], we therefore analyzed GATA3 and FOXP3 expression. The expression kinetic of naive T cells exposed to IL-4 and TGF-β demonstrated that GATA3 and FOXP3 are initially found in separate populations (day 2), but transiently co-express both factors (days 4–8), before establishing separate populations at the end of the differentiation process (day 10; Figure 6C), suggesting that GATA3 inhibits the development of iTreg cells by repressing FOXP3.
These results showed that IL-4 acts in vitro as an inhibitor of FOXP3 expression, without interfering with TGF-β signaling, probably acting at the level of transcription factors, and possibly by a GATA3-dependent mechanism.
FOXP3 expression decreased once GATA3 expression is high; therefore, we hypothesized a potential role for GATA3 in repressing FOXP3. Besides GATA3′s well-known positive effect on gene regulation, GATA3′s repressive capabilities were previously shown to restrict Th1 commitment by inhibiting STAT4 expression [2,21], and therefore GATA3 prevents differentiation into Th1 cells. To investigate whether GATA3 can directly inhibit FOXP3 induction, we transduced GATA3 or a truncated GATA3 lacking the DNA-binding domain in human primary CD4+CD45RA+ T cells using a TAT-fused, recombinantly expressed GATA3. After transduction, the cells were activated with soluble anti-CD3/CD28 in the presence or absence of TGF-β. TAT-GATA3 was successfully transduced in a homogeneous and dose-dependent manner into human CD4+ T cells (Figure 7A, upper panel). TAT-GATA3 reduced FOXP3 expression in a dose-dependent manner, whereas a DNA-binding domain truncated version (TAT-ΔDBD-GATA3) did not affect FOXP3 expression as compared with expression in untransduced cells (Figure 7A). In addition, we analyzed the inhibitory effect of GATA3 on FOXP3 in transgenic DO11.10 mice, constitutively overexpressing GATA3 under the control of the CD2 locus control region (DO11.10xCD2-GATA3). The thymic selection into the CD4 lineage is largely intact in DO11.10xCD2-GATA3 (RW Hendriks, unpublished data). These mice develop lymphomas at an older age, but signs of autoimmune disease were not described [22]. To investigate the effect of GATA3 on iTreg, CD4+CD62L+CD25– cells were isolated, activated with OVA in the presence or absence of TGF-β, and Foxp3 expression was analyzed after 4 d. The naive CD4+CD25– cells were Foxp3– (unpublished data). As described for the human cells, TGF-β dramatically up-regulated Foxp3 in the DO11.10 littermate control mice. In contrast, cells from the CD2-GATA3xDO11.10 mice showed dramatically reduced Foxp3 expression when activated with TGF-β and OVA (Figure 7B). All mice produced similar amounts of TGF-β; in addition, Smad7 was equally expressed [23] in T cells of both mice strains (Figure 7C), indicating intact TGF-β signaling.
Taken together, these results demonstrated a repressive role of IL-4-induced GATA3 transcription factor in the generation of iTreg cells.
To investigate the molecular mechanism of GATA3-mediated repression of human FOXP3, the human FOXP3 promoter was studied and a palindromic binding site for GATA3 was discovered. The GATA-binding site is located 303 bp upstream from the transcription start site (TSS) [24]. This site is highly conserved between humans, mice, and rats (Figure S4) and may therefore play an important role in FOXP3 regulation. The functional relevance of this site was studied using a FOXP3-promoter construct [24]. We transfected human primary CD4+ T cells, in vitro differentiated Th2 cells, and Jurkat cells (Jurkat cells are known to constitutively express GATA3 [25,26]), and we measured FOXP3 promoter activity. The promoter was not active in the GATA3-expressing cell line Jurkat or in the in vitro-differentiated Th2 cells, whereas the construct was active in the CD4 cells, which express a lower amount of GATA3 (Figure 8A). Overexpression of GATA3 in naive T cells diminished luciferase activity of the FOXP3 promoter compared with the control vector (Figure 8B). To further address the function of the GATA3 site, we inserted a site-specific mutation deleting the GATA3-binding site. This mutation increased luciferase activity by 3-fold in memory CD4+CD45RO+ T cells, whereas no difference was observed in naive (GATA3–) CD4+CD45RA+ T cells, revealing a repressor activity of GATA3 on the FOXP3 promoter (Figure 8C). Furthermore GATA3 binds directly to the FOXP3 promoter as investigated by pull-down assay. HEK cells were transiently transfected with GATA3 or a control vector, and increasing amounts of lysates were incubated with oligonucleotides containing the GATA3 site of the FOXP3 promoter or a control oligonucleotide with a mutated GATA3-binding site. After the pull-down, GATA3 binding was detected by Western blot. Similarly, GATA3-expressing Th2 cells and iTreg cells were subjected to this approach. Only HEK cells overexpressing GATA3 and Th2 cells showed GATA3-binding activity (Figure 8D and 8E). These experiments demonstrated that GATA3 binds the palindromic FOXP3 promoter. To gain insights into the in vivo situation, we performed a chromatin immunoprecipitation (ChIP) using an anti-GATA3 antibody and showed that GATA3 binds to the FOXP3 promoter region in Th2 cells, but not in iTreg cells (Figure 7F). Taken together these data demonstrate that the GATA3-binding to the FOXP3 promoter is repressing FOXP3 expression.
The current study reveals that FOXP3 induction, an important step in iTreg commitment, is inhibited by GATA3, which is the key regulator for polarization toward Th2 cells. After differentiation, the effector Th2 cells become refractory to conversion into a FOXP3+ phenotype.
In accordance with other studies, we found that CD4+CD25– cells were able to up-regulate FOXP3 [12,27]. Already-committed cells such as memory T cells and Th1 cells showed only moderate and transient FOXP3 induction, which is not sufficient to change the phenotype toward a regulatory T cell profile. In contrast, naive T cells could efficiently up-regulate FOXP3 when treated with TGF-β to induce iTreg cells [10,28–34], suggesting that FOXP3 plays an important role in the early differentiation process and may act in a way similar to that known for the Th1/Th2 decision factors T-bet and GATA3. This commitment is characterized by competitive expression of these factors [35,36], which we also observed in differentiating FOXP3+ iTreg cells including a phase of co-expression, which turns into nonoverlapping expression upon completed differentiation. In this competitive process TGF-β appeared to be mandatory for the induction of FOXP3, possibly by keeping the expression of GATA3 and T-bet low [37,38]. In contrast, differentiating naive T cells in the absence of polarization factors (Th0) such as IL-4, IL-12, or TGF-β showed only a transient FOXP3 expression and failed to generate a population of FOXP3-expressing cells, but GATA3 and T-bet were up-regulated (unpublished data). Interestingly, as we and others previously described, FOXP3-promoting factors, such as dexamethasone [39], CTLA-4 [40], and estrogens [41], are also known as inhibitors of GATA3 expression [42–45]. Therefore GATA3 not only induces differentiation into Th2 cells but also inhibits FOXP3 expression and commitment into iTreg cells.
The Th2 cytokine IL-4 but not IL-13 (unpublished data) was able to inhibit TGF-β–mediated FOXP3 induction and therefore prevented conversion into the regulatory phenotype. To prove the inhibitory effect of IL-4 on inducible or natural Treg commitment in vivo, we treated mice with IL-4 and anti-IL-4. This has been shown to increase the effect of the cytokine in vivo [19]. Only the IL-4/IL-4 mAb complex resulted in a decrease of the amount of natural or inducible Treg (CD25+ and Foxp3+) cells 7 d after treatment. Our results suggest that IL-4 is only interfering with the differentiation of naive T cells into iTreg cells. But since a distinction of nTreg and iTreg cells is currently not possible, because iTreg cells also transiently express CD25 after activation, we cannot exclude that IL-4 may also inhibit Foxp3 expression in nTreg cells in vivo or that additional effects may contribute to the observed drop in Foxp3 expression. IL-4 has already been shown to negatively regulate the development of naive T cells into Th1 or the IL-17–producing T cells (Th17) [46,47]. Similar effects have been recently described for IL-6, which, combined with TGF-β, inhibits the generation of iTreg cells and induces differentiation into the Th17 cells by an unknown mechanism [48,49]. Thus the polarization into iTreg cells is negatively regulated by the effector cytokines IL-4 and IL-6.
IL-4 has been previously shown to induce the generation of FOXP3+ Treg cells out of CD4+CD25– [50]. In those experiments, the concentrations of IL-4 used were low, and as we also observed, IL-4 at low concentration slightly enhanced FOXP3 expression. Importantly, these concentrations were not sufficient to induce GATA3 expression. IL-4 may favor proliferation of nTreg cells [47] or directly regulate FOXP3 expression in a STAT-dependent fashion [51].
Since IL-13 does not effectively reduce FOXP3 and fails to induce GATA3, we hypothesized that the IL-4–dependent inhibition of FOXP3 could be mediated by GATA3. In fact, GATA3-inducing IL-4 concentrations repressed TGF-β–mediated FOXP3 expression, whereas IL-4 as well as TGF-β signaling were intact. This result suggested a competitive mechanism between GATA3 and FOXP3 transcription factors in determining lineage commitment during the early phase of differentiation. Accordingly, we investigated naturally high GATA3-expressing cells and confirmed the absence of FOXP3. Protein transduction of GATA3 into naive T cells inhibited FOXP3 induction in human, differentiating, naive T cells. This inhibitory effect of GATA3 was further confirmed in BALB/c transgenic mice, expressing GATA3 in T cells (DO11.10:CD2-GATA3 transgenic mice). In line with the transient overexpression of GATA3 in human T cells, cells of these mice failed to induce FOXP3 expression upon exposure with antigen in the presence of TGF-β. Strikingly, the DO11.10:CD2-GATA3 mice do have peripheral FOXP3+ cells, which however displayed a 10%–25% lower frequency compared to wild-type DO11.10 mice. Thus GATA3 restrains the development of certain Treg subsets, presumably the inducible, peripheral population and not those of thymic origin. Thymic T cells undergo a different maturation process, which may explain the insensitivity of nTreg to GATA3 overexpression [52]. In contrast to the Th2-differentiating and iTreg-inhibiting function of GATA3 in peripheral T cells, GATA3 acts in the thymus together with other transcription factors such as the Repressor of GATA3 (ROG) in the differentiation process toward CD8 cells [53,54] or participates in complex transcriptional feedback network to regulate sympathoadrenal differentiation [55]. Therefore, the role of GATA3 appears to be tissue specific and cannot be generalized.
Our study demonstrates that GATA3 repressed FOXP3 expression directly by binding to the FOXP3 promoter region. A palindromic GATA-site is located 303 bp upstream of the TSS in a highly conserved region, which we have previously identified as the FOXP3 promoter [24]. Site-specific mutation of this site increased the activity of the promoter constructs, thus revealing the repressive nature of this GATA element in memory T cells, which naturally express GATA3, whereas no difference was seen in naive T cells, which do not express GATA3. This palindromic GATA element is bound by GATA3 protein as proven with pull-down experiments. Furthermore, it is shown by ChIP that GATA3 binds this element also in intact cells. It is known that GATA3 can induce transcription by chromatin remodeling [56], by directly transactivating promoters [36], or, as shown in the current study, acts as a repressor of gene expression [21,57–59]. Therefore keeping GATA3 expression low might be required to induce efficient FOXP3+ iTreg cell generation.
The molecular interactions enabling GATA3 to inhibit FOXP3 are not identified yet, but the GATA-binding site is located adjacent to positive, inducing sites, composed of AP-1-NFATc2 sites [24], and GATA3 may compete with the binding of AP-1/NFAT to the promoter (unpublished observations).
In summary, we demonstrated that FOXP3 is negatively regulated by cytokines such as IL-4. GATA3 acts as an inhibitor of FOXP3 expression in early T cell differentiation, as well as in differentiated Th2 cells by directly binding and repressing the FOXP3 promoter. We therefore describe a new mechanism of how Il-4 avoids tolerance induction by repressing FOXP3 expression. These findings will give new perspectives toward understanding molecular mechanisms of iTreg induction and thus pathways of peripheral tolerance induction, particularly in allergy and asthma.
Normal B6 mice were purchased from the Jackson Laboratories (Bar Harbor, Maine). Transgenic DO11.10 mice, expressing a T cell receptor for OVA323–339 peptide in the context of H-2d, were backcrossed with mice expressing GATA-3, driven by the human CD2 locus control region (CD2-GATA3) [22], resulting in DO11.10xCD2-GATA3 mice. Mice used for experiments were backcrossed on a BALB/c background for a minimum of eight generations and used at an age of 8–12 wk. Mice were housed under specific pathogen-free conditions and all animal studies were performed according to institutional and state guidelines.
CD4+ T cells were isolated from blood of healthy human volunteers using the anti-CD4 magnetic beads (Dynal, Hamburg, Germany) as previously described [60]. The purity of CD4+ T cells was initially tested by FACS and was ≥ 95%. Monoclonality of T cell clones was confirmed by TCR-chain mapping and was identified to be Vbeta8 positive. The clones were characterized by high IL-4 secretion.
The PCR primers and probes were designed based on the sequences reported in GenBank with the Primer Express software version 1.2 (Applied Biosystems) as follows: FOXP3 forward primer 5′-GAA ACA GCA CAT TCC CAG AGT TC-3′; FOXP3 reverse primer 5′-ATG GCC CAG CGG ATG AG-3′; EF-1α forward primer and reverse primer as described [61]; GATA3 forward primer 5′-GCG GGC TCT ATC ACA AAA TGA-3′ and rwd 5′-GCT CTC CTG GCT GCA GAC AGC-3′. The prepared cDNAs were amplified using SYBR-PCR mastermix (Biorad) according to the recommendations of the manufacturer in an ABI PRISM 7000 Sequence Detection System (Applied Biosystems).
Quantitative PCR of murine samples was performed with Brilliant SYBR Green QPCR master mix (Stratagene) and the following primers: Ubiquitin C, 5′-AGG TCA AAC AGG AAG ACA GAC GTA-3′ and 5′-TCACACCCAAGAACAAGCACA-3′; Smad-7, 5′-GAA ACC GGG GGA ACG AAT TAT-3′ and 5′-CGC GAG TCT TCT CCT CCC A-3′; TGF-ß1, 5′-TGA CGT CAC TGG AGT TGT ACG G-3′ and 5′-GGT TCA TGT CAT GGA TGG TGC-3′. Primer pairs were evaluated for integrity by analysis of the amplification plot, dissociation curves, and efficiency of PCR amplification. PCR conditions were 10 min at 95 °C, followed by 40 cycles of 15 s at 95 °C and 60 °C for 1 min using an 7300 real-time PCR system (Applied Biosystems). PCR amplification of the housekeeping gene encoding ubiquitin C was performed during each run for each sample to allow normalization between samples. Relative quantification and calculation of the range of confidence was performed using the comparative ΔΔCT method.
Naive CD4+ T cells (CD4+, CD62L+, and CD25–) were isolated from pooled lymph nodes and spleens by FACS (FACS Aria, BD Biosciences). 5 × 105 T cells were co-cultured with 2.5 × 104 bone marrow–derived dendritic cells [62] and 0.01 μg/ml OVA323–339 peptide (Ansynth) in the presence or absence of 20-ng/μl rhTGF-ß1 (Peprotech) in 48-well plates. After 4 d, cells were harvested and analyzed for intracellular FOXP3 expression by FACS or gene expression by quantitative RT-PCR.
CD4+ CD45RA+ magnetically-sorted (CD45RO depletion, MACS, according to the protocol of the manufacturer) cells were stimulated with immobilized plate-bound anti-CD3 (1 μg/ml, Okt3, IgG1) and anti-CD28 (2 μg/ml) in Th1 conditions: 25 ng/ml IL-12, 5 μg/ml anti-IL-4 (R&D systems); in Th2 conditions: 25 ng/ml IL-4, 5 μg/ml anti-IFN-γ, 5 μg/ml anti-IL-12 (R&D systems); or in Treg conditions: 10 ng/ml TGF-β, 5 μg/ml anti-IFN-γ, 5 μg/ml anti-IL-12, 5 μg/ml anti-IL-4. Proliferating cells were expanded in medium containing IL-2 (30 ng/ml).
The FOXP3 promoter was cloned into the pGL3 basic vector (Promega Biotech) to generate the pGL3 FOXP3 −511/+176 [24]. Site-directed mutagenesis in the FOXP3 promoter region were introduced using the QuickChange kit (Stratagene), according to the manufacturer's instructions. The following primer and its complementary strand were used: GTT TCT CAT GAG CCC TAT TAA GTC ATT CTT ACC TCT CAC CTC TGT GGT GA.
T cells were rested in serum-free AIM-V medium (Life Technologies) overnight. 3.5 μg of the FOXP3 promoter luciferase reporter vector and 0.5 μg phRL-TK were added to 3 × 106 CD4+ T cells resuspended in 100 μL of Nucleofector solution (Amaxa Biosystems) and electroporated using the U-15 program of the Nucleofector. After a 24-h culture in serum-free conditions and stimuli as indicated in the figures, luciferase activity was measured by the dual luciferase assay system (Promega Biotech) according to the manufacturer's instructions. Data were normalized by the activity of renilla luciferase.
RNA was isolated using the RNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. Reverse transcription of human samples was performed with TaqMan reverse-transcription reagents (Applied Biosystems) with random hexamers according to the manufacturer's protocol.
The cDNAs encoding GATA3 protein or the truncated GATA3 (lacking the two zinc fingers) were cloned in frame into an expression vector along with the TAT sequence as previously described [63]. Proteins were expressed in BL21 Star (DE3)pLysS (Invitrogen) and lysates were purified by Ni2+-chelate column chromatography. Both TAT-linked proteins were more than 95% pure, based on Coomassie blue staining of sodium disulfate acrylamide gels.
CD4+CD45RA+ cells were cultured in AIMV medium and transduced with 20 nM, 10 nM, or 500 nM of full-length or truncated GATA3 over the course of 4 h. After 4 h, the cells were washed and activated with soluble anti-CD3 and anti-CD28 and TGF-β (10 ng/ml). Each day, the TAT proteins were freshly added to the medium. FOXP3 expression was measured after 5 d by intracellular staining.
T cells were stimulated with 2 × 10−7 M PMA and 1 μg/ml of ionomycin (Sigma Chemicals) for 4 h. The following mAb was used: anti-IL-4-PE (8D4–8, BD). Matched isotype controls were used at the same protein concentration as the respective antibodies. Four-color FACS was performed using an EPICS XL-MCL (Beckman Coulter) using the software Expo32 version for data acquisition and evaluation.
For analysis of FOXP3 expression at the single-cell level, cells were first stained with the monoclonal antibody CD25 (Beckman Coulter), and after fixation and permeabilization, cells were incubated with PE-conjugated monoclonal antibody PCH101 (anti-human FOXP3; eBioscience) based on the manufacturer's recommendations and subjected to FACS (EPICS XL-MCL). For cell surface marker staining, cells were incubated for 20 min at 4°C in staining buffer with the following antibodies: anti–CD152-PE (CTLA-4; BD), anti–PD-1 (eBiosciences), anti-GITR (R & D Systems), anti-CD69 (Beckman Coulter), anti-CD103 (DakoCytomation), anti-CD62L (Beckman Coulter), or anti-HLA-DR (Beckman Coulter). The controls were FITC, PE, or ECD-conjugated mouse IgG1 or rat IgG2a. For staining of mouse cells, the following mAbs from BD Biosciences were used following standard techniques as described above: anti-CD3, anti-CD4, and anti-CD25. Anti-FcγRII/III antibody (2.4G2, ATCC) was included in all stainings to reduce nonspecific antibody binding. To isolate naive murine CD4 T cells from murine DO11.10 or DO11.10xCD2-GATA3 T cells, cells were stained with anti-CD25-FITC, anti-CD62L-PE, and anti-CD4-APC prior to sorting. Dead cells were excluded with 4′',6-Diamidino-2-phenylindole (DAPI). To analyze murine Foxp3 expression in inducible Treg cultures, cells were stained intracellularly with anti-Foxp3-PE according to manufacturer's instruction, in conjunction with anti-CD4-APC and LIVE/DEAD fixable dead cell stain kit (Invitrogen) to discriminate live cells. All monoclonal antibodies for murine cell stainings were purchased from eBioscience or BD Biosciences.
For FOXP3 analysis on the protein level, 1 × 106 CD4+CD25– cells were lysed and loaded next to a protein-mass ladder (Magicmark, Invitrogen) on a NuPAGE 4–12% bis-tris gel (Invitrogen). The proteins were electroblotted onto a PVDF membrane (Amersham Life Science). Unspecific binding was blocked with BSA, and the membranes were subsequently incubated with an 1:200 dilution of goat anti-FOXP3 in blocking buffer (Abcam) overnight at 4 °C. The blots were developed using an anti-goat HRP-labeled mAb (Amersham Biosciences) and visualized with a LAS 1000 camera (Fuji). Membranes were incubated in stripping buffer and re-blocked for 1 h. The membranes were re-probed using anti-GATA3 (HG3–31; Santa Cruz Biotechnology), anti-T-bet (4B10, Santa Cruz Biotechnology), anti-GAPDH (6C5, Ambion), anti-phospho-SMAD2 (138D4), anti-phospho-STAT6 (5A4), and anti-STAT6 (Cell Signaling Technology),
Age- and gender-matched normal B6 mice received every other day intraperitoneal (ip) injections of PBS, 1.5 μg rmIL-4, 50 μg anti-IL-4 mAb (11B11 or MAB404), or a mixture of 1.5 μg rmIL-4 plus 50 μg anti-IL-4 mAb (11B11 or MAB404) for 7 d. Thereafter, spleen and lymph node cells were analyzed by flow cytometry for CD3, CD4, and CD25 expression. The anti-mouse IL-4 mAb MAB404 was obtained from R&D Systems, the second anti-mouse IL-4 mAb 11B11 was purchased from eBioscience.
ChIP analysis was performed according to the manufacturer's protocol (Upstate Biotechnology) with the following modifications. iTreg and Th2 cells were fixed with 1% formaldehyde for 10 min at room temperature. The chromatin was sheared to 200-1000 bp of length by sonication with five pulses of 10 s at 30% power (Bandelin). The chromatin was pre-cleared for 2 h with normal mouse IgG beads and then incubated with anti-GATA3-agarose beads (HG3–31; Santa Cruz Biotechnology) for 2 h. Washing and elution buffers were used according to the protocol of Upstate Biotechnology. Crosslinks were reversed by incubation at 65 °C for 4 h in the presence of 0.2 M NaCl, and the DNA was purified by phenol/chloroform extraction. The amount of DNA was determined by conventional PCR. The PCR addressed for the FOXP3 promoter region −246 to −511 and was performed using the following primers: 5′-gtgccctttacgagt catctg-3′ and 5′-gtgccctttacgagtcatctg-3′. The PCR products were visualized using an ethidium bromide gel.
CD4+ T cells were stimulated with PMA and ionomycin for 2 h at 37°C. The cells were pelleted, resuspended in buffer C (20 mM HEPES [pH 7.9], 420 mM NaCl, 1.5 mM MgCl2, 0.2 mM EDTA, 1 mM DTT, protease inhibitors [Sigma]. and 0.1% NP-40) and lysed on ice for 15 min. Insoluble material was removed by centrifugation. The supernatant was diluted 1:3 with buffer D (as buffer C, but without NaCl). The lysates were incubated with 10 μg of poly(dI-dC) (Sigma) and 70 μl of streptavidin-agarose (Amersham Biosciences) carrying biotinylated oligonucleotides, for 3 h at 4 °C. The beads were washed twice with buffer C:D (1:3) and resuspended in DTT-containing loading buffer (NuPAGE; Invitrogen), heated to 70 °C for 10 min, and the eluants on a NuPAGE 4–12% bis-tris gel (Invitrogen). The proteins were electroblotted onto a PVDF membrane (Amersham Biosciences) and detected using an anti-GATA3 mAb (Santa Cruz Biotechnology). Accumulated signals were analyzed using AIDA software (Raytest).
|
10.1371/journal.pcbi.1005495 | Simultaneous inference of phylogenetic and transmission trees in infectious disease outbreaks | Whole-genome sequencing of pathogens from host samples becomes more and more routine during infectious disease outbreaks. These data provide information on possible transmission events which can be used for further epidemiologic analyses, such as identification of risk factors for infectivity and transmission. However, the relationship between transmission events and sequence data is obscured by uncertainty arising from four largely unobserved processes: transmission, case observation, within-host pathogen dynamics and mutation. To properly resolve transmission events, these processes need to be taken into account. Recent years have seen much progress in theory and method development, but existing applications make simplifying assumptions that often break up the dependency between the four processes, or are tailored to specific datasets with matching model assumptions and code. To obtain a method with wider applicability, we have developed a novel approach to reconstruct transmission trees with sequence data. Our approach combines elementary models for transmission, case observation, within-host pathogen dynamics, and mutation, under the assumption that the outbreak is over and all cases have been observed. We use Bayesian inference with MCMC for which we have designed novel proposal steps to efficiently traverse the posterior distribution, taking account of all unobserved processes at once. This allows for efficient sampling of transmission trees from the posterior distribution, and robust estimation of consensus transmission trees. We implemented the proposed method in a new R package phybreak. The method performs well in tests of both new and published simulated data. We apply the model to five datasets on densely sampled infectious disease outbreaks, covering a wide range of epidemiological settings. Using only sampling times and sequences as data, our analyses confirmed the original results or improved on them: the more realistic infection times place more confidence in the inferred transmission trees.
| It is becoming easier and cheaper to obtain (whole genome) sequences of pathogen samples during outbreaks of infectious diseases. If all hosts during an outbreak are sampled, and these samples are sequenced, the small differences between the sequences (single nucleotide polymorphisms, SNPs) give information on the transmission tree, i.e. who infected whom, and when. However, correctly inferring this tree is not straightforward, because SNPs arise from unobserved processes including infection events, as well as pathogen growth and mutation within the hosts. Several methods have been developed in recent years, but often for specific applications or with limiting assumptions, so that they are not easily applied to new settings and datasets. We have developed a new model and method to infer transmission trees without putting prior limiting constraints on the order of unobserved events. The method is easily accessible in an R package implementation. We show that the method performs well on new and previously published simulated data. We illustrate applicability to a wide range of infectious diseases and settings by analysing five published datasets on densely sampled infectious disease outbreaks, confirming or improving the original results.
| As sequencing technology becomes easier and cheaper, detailed outbreak investigation increasingly involves the use of pathogen sequences from host samples [1]. These sequences can be used for studies ranging from virulence or resistance related to particular genes [1, 2], to the interaction of epidemiological, immunological and evolutionary processes on the scale of populations [3, 4]. If most or all hosts in an outbreak are sampled, it is also possible to use differences in nucleotides, i.e. single-nucleotide polymorphisms (SNPs), to resolve transmission clusters, individual transmission events, or complete transmission trees. With that information it becomes possible to identify high risk contacts and superspreaders, as well as characteristics of hosts or contacts that are associated with infectiousness and transmission [5, 6]. Much progress has been made in recent years in theory and model development, but existing methods either include assumptions that do not take the full uncertainty in the evolutionary process into account [7, 8], are designed for specific datasets, with fit-for-purpose code for data analysis [9–11], or make limiting assumptions about the relation between sampling times and infectivity [12]. An easily accessible method without these restrictions and with the flexibility to cover a wide range of infections is currently lacking, and would bring analysis of outbreak sequence data within reach of a much broader community.
The interest in easily applicable methods for sequence data analysis in outbreak settings is demonstrated by the community’s widespread use of the Outbreaker package in R [8, 13–15]. However, the model in Outbreaker assumes that mutations occur at the time of transmission, which does not take the pathogen’s in-host population dynamics into account, nor the fact that mutations occur within hosts. The publications by Didelot et al [7] and Ypma et al [11] revealed that within-host evolution is crucial to relate sequence data to transmission trees, as is illustrated in Fig 1A: there are four unobserved processes, i.e. the time between subsequent infections, the time between infection and sampling, the pathogen dynamics within hosts, and mutation. The difference in sequences between host 2 and infector 1 result from all of these processes. As a result, a host’s sample can have different SNPs from his infector’s (Fig 1B: hosts 1 and 2); a host can even be sampled earlier than his infector with fewer SNPs (Fig 1B: hosts 1 and 3).
Several recently published methods do allow mutations to occur within the host, but make other assumptions not fully reflecting the above-described process, such as using a phenomenological model for pairwise genetic distances [16], presence of a single dominant strain in which mutations can accumulate [9, 17], or absence of a clearly defined infection time [18]. To take the complete process into account, Didelot et al [7] and Numminen et al [10] took a two-step approach: first, phylogenetic trees were built, and second, these trees were used to infer transmission trees. Didelot et al [7] used the software BEAST [19, 20] to make a timed phylogenetic tree, and used a Bayesian MCMC method to colour the branches such that changes in colour represent transmission events. Numminen et al [10] took the most parsimonious tree topology, and accounted for unobserved hosts by a sampling model (which is an additional complication). This two-step approach is likely to work better if the phylogenetic tree is properly resolved (unique sequences with many SNPs), but less so if there is uncertainty in the phylogenetic tree. However, also in that case construction of the phylogenetic tree is done without taking into account that only lineages in the same host can coalesce, and that these go through transmission bottlenecks during the whole outbreak. That is likely to result in incorrect branch lengths and consequently incorrect infection times.
Hall et al [12] implemented a method in BEAST for simultaneous inference of transmission and phylogenetic trees. BEAST allows for much flexibility when it comes to phylogeny and population dynamics reconstruction (for which it was originally developed [19, 20]), e.g. by allowing variation in mutation rates between sites in the genome, between lineages, and in time. However, datasets of fully observed outbreaks often do not contain sufficient information for reliable inference: they typically cover only a few months up to at most several years (as in Didelot et al [7], with tuberculosis) and do not contain many SNPs (usually of the same order of magnitude as the number of samples). A more important limitation is that the transmission model implemented in BEAST is rather specific: it allows for transmission only during an infectious period informed by positive and negative samples, during which infectiousness is assumed to be constant. This may put prior constraints on the topology and order of events in the transmission and phylogenetic trees, which is undesirable if the primary aim is to reconstruct the transmission tree with little or no prior information about when hosts were infectious.
Previously, Ypma et al [11] had also developed a method for simultaneous inference of transmission and phylogenetic trees, albeit with rather specific assumptions on the within-host pathogen dynamics and the time and order of transmission events, and with no available implementation. However, their view on the phylogenetic and transmission trees was quite different. Instead of a phylogenetic tree with transmission events, they regarded it as a hierarchical tree. The top level is the transmission tree, with hosts having infected other hosts according to an epidemiological transmission model. The lower level consists of phylogenetic “mini-trees” within each host. A mini-tree describes the within-host micro-evolution. It is rooted at the infection time and has tips at transmission and sampling events; in its simplest form it is only a single branch from infection to sampling. The complete phylogenetic tree then consists of all these mini-trees, connected through the transmission tree. That description allowed them to develop new MCMC updating steps, some changing the transmission tree, some the phylogenetic mini-trees.
We built further on that principle to reconstruct the transmission trees of outbreaks, in a new model and estimation method. The method requires data on pathogen sequences and sampling times. The model includes all four underlying stochastic processes (Fig 1A), each described in a flexible and generic way, such that we avoid putting unnecessary prior constraints on the order of unobserved events (Fig 1B). This allows for application of the method to a wide range of infectious diseases, including new emerging infections where we have little quantitative information on the infection cycle. The method is implemented in R, in a package called phybreak. We illustrate the performance of the method by applying it to both new and previously published simulated datasets. We demonstrate the range of applicability by applying the model to five datasets on densely sampled infectious disease outbreaks, covering a wide range of epidemiological settings.
The method infers infection times and infectors of all cases in an outbreak. The data consist of sampling times and sequences of all cases, where some of the sequences may be non-informative if no sequence is available. Using simple models for transmission, sampling, within-host dynamics and mutation, samples are taken from the posterior distributions of model parameters and transmission and phylogenetic trees, by a Markov-Chain Monte Carlo (MCMC) method. The main novelty of our method lies in the proposal steps for the phylogenetic and transmission trees that are used to generate the MCMC chain. It makes use of the hierarchical tree perspective, in which the phylogenetic tree is described as a collection of phylogenetic mini-trees (one for each host), connected through the transmission tree (see Methods for details).
The posterior samples are summarized by medians and credible intervals for parameters and infection times, and by consensus transmission trees. Consensus transmission trees are based on the posterior support for infectors of each host, defined as the proportion of posterior trees in which a particular infector infects a host. The Edmonds’ consensus tree takes for each host the infector with highest support, and uses Edmonds’ algorithm to resolve cycle and multiple index cases [21], whereas the Maximum Parent Credibility (MPC) tree is the one tree among the posterior trees with maximum product of supports [12].
The models and parameters used for inference are as follows:
We generated 25 new simulated datasets of 50 cases with the above model, which we modified by taking a population of 86 individuals and a basic reproduction number R0 = 1.5 (instead of an infinite population with R0 = 1). Parameters were aG = aS = 10, mG = mS = r = 1, μ = 10−4 and sequence length 104, resulting in 1 genome-wide mutation per mean generation interval of one year.
Table 1 shows some summary measures on performance of the method (see S1 Results for additional measures and results for more simulations). A 5,000 cycle burn-in followed by sampling a single chain of 25,000 MCMC cycles took about 30 minutes on a 2.6 GHz CPU (Linux). Four sets of results are shown, all with an uninformative prior for μ: one with all parameters other than μ fixed at their correct value, and three with uninformative priors for mG and r, and different levels of prior knowledge on mS: informative with correct mean, uninformative, and informative with incorrect mean. The top of the table shows effective sample sizes (ESSs) for all parameters and for the infection times to evaluate mixing of continuous parameter samples. The path-distance approximate topological ESS [23] was calculated to assess phylogenetic tree mixing. To evaluate mixing across and within chains of infectors per host, we tested for differences between the chains and for dependency within the chains by Fisher’s exact tests: the proportion of accepted tests (P > 0.05) is a measure of mixing. The MCMC mixing is generally good for tree inference and model parameters, as most ESSs are above 200 and an expected 95% of Fisher’s tests is accepted; the only exceptions being mS with an uninformative prior.
The bottom part of Table 1 shows the results on tree inference. Infection times (using all MCMC samples) are well recovered if the mean sampling interval does not have a strong incorrect prior: coverage of 95% credible intervals is good, and medians may only be slightly positively biased (later than true infection time) if uninformative priors are used. For this simulation scenario, consensus transmission trees contained almost 70% (35 out of 50) correct infectors, as determined by counting infectors and resolving multiple index cases and cycles in the tree (Edmonds’ method [21]) and slightly fewer when choosing the maximum parent credibility (MPC) tree [12] among the 50,000 posterior trees. Infectors with high support are more likely correct: 84% (28 out of 33) are correct if the support is above 50%, and 96% (15.2 out of 15.8) are correct if the support is above 80%. These numbers are similar in smaller outbreaks (S1 Results). If sampling and generation interval distributions are wider, the sampling times contain less information on the order of infection, which reduces the accuracy of transmission tree reconstruction (S1 Results). Using prior information on the mean sampling interval did not improve on this, but if prior information is incorrect, fewer hosts have a strongly supported infector, which makes inference more uncertain. In conclusion, the method is fast and efficient if applied to simulated data fitting the model. In that case, no informative priors are needed for transmission tree inference, though correct estimation of the infection time is aided by some information.
For comparison, we analysed the same datasets with the Outbreaker package in R [8], which uses the assumption of mutation at transmission, and with the TransPhylo package [7, 24], which requires input of a phylogenetic tree that we created in BEAST v2 [19] with a constant population coalescent model and Jukes-Cantor substitution model. Both Outbreaker and TransPhylo require input of a generation and sampling interval distribution, for which we supplied the distributions used to simulate the data. Thus, the results are best compared to the results of the reference scenario of our model (Table 1, left-most column). The numbers of correctly identified infectors (Edmonds’ consensus tree [21]) were smaller with both alternative methods: in the 25 outbreaks of Table 1 (50 cases, aG = aS = 10), Outbreaker identifies on average 27.5 out of 50 infectors correctly, TransPhylo 32.2, and phybreak 34.9. Also in smaller outbreaks or with different generation and sampling interval distributions, phybreak identified 8–22% more infectors correctly (S1 Results).
We also analysed the simulated results with 20% of the cases removed from the dataset, to assess performance if outbreaks are not completely observed. Table 2 shows the results with reference (parameters fixed and correct) and uninformative analyses, in comparison with the reference scenario and all data observed. With some of the cases removed, some of the remaining cases did not have their infector in the dataset anymore; these cases are referred to as orphans in Table 2. Infection time estimation was less accurate, with only 85% of credible interval containing the correct value, and more infection times estimated too early in the outbreak. Surprisingly, this was not only the case with orphans, for which this may have been expected with their infector not present in the data. It turns out that infectors are correctly identified about 20% less accurately, for all threshold levels of support. However, when correcting for presence of the infector in the data, infectors are identified with the same accuracy as in the complete dataset. We also checked how frequently the identified infector of orphans was in fact an earlier ancestor in the transmission tree, i.e. the infector’s infector in most cases. It turned out that ancestors were often identified as infector, but not as accurately as the true infector identification in complete datasets (Table 2).
We applied the method to previously published outbreak simulations [12]. Briefly, a spatial susceptible-exposed-infectious-recovered (SEIR) model was simulated in a population of 50 farms, with a latent period (exposed) of two days and a random infectious period with mean 10 days and standard deviation 1 day, at the end of which the farm was sampled. Two mutation rates were used with an HKY substitution model, either Slow Clock or Fast Clock, equivalent to 1 or 50 genome-wide mutations per generation interval of one week, respectively.
Table 3 shows performance of the method with naïve and informative prior information on the sampling interval distribution (see S1 Results for uninformative). Effective sample sizes of parameters and phylogenetic trees are a bit smaller than with analysis of the new simulations, but in most cases still good for infection times, whereas sampling of infectors was excellent. The low variance of the sampling interval distribution caused some problems in efficient sampling of mS because of its high correlation with the associated infection times, but it also caused problems in the burn-in phase if inference starts with parameter values far from their actual values (not shown). This was especially the case in the uninformative Slow Clock analyses, resulting in unreliable estimates of the mean sampling interval and infection times (S1 Results). With the Fast Clock analyses there were no such problems, as long as the full set of proposal paths in the MCMC chain was used (see Methods for details). Posterior median mutation rates are slightly higher than used for simulation, which could be due to different rates for transition and transversion in the simulation model [12].
Consensus trees with uninformative and informative settings were almost as good as in the original publication [12], in which spatial data were used and in which it was known that there was a latent period and that hosts could not transmit after sampling. In the Slow Clock simulations about 62% of infectors were correct, and in the Fast Clock simulations about 92%. Infection times were also well recovered in most cases, but not in the uninformative Slow Clock analysis (S1 Results). In the naïve analyses, the Slow Clock consensus trees were only slightly less good (but not the infection times), whereas the Fast Clock consensus trees became much worse, with only 65% of infectors correct. In conclusion, the method performs well if applied to data simulated with a very different model. Good prior information on the sampling interval can significantly improve both MCMC mixing and transmission tree inference, especially if the genetic data contain many SNPs.
We finally applied the method to five published datasets on outbreaks of Mycobacterium tuberculosis (Mtb, [7]), Methicillin-resistant Staphylococcus aureus (MRSA, [25]), Foot-and-mouth disease (FMD2001 and FMD2007, [9, 11, 26, 27]), and H7N7 avian influenza (H7N7, [12, 28–30]).
The results for the four smaller datasets are shown in Table 4, which shows that mixing of the MCMC chains was generally good. Fig 2 shows the Edmond’s consensus trees (full details in S1 Results), with each host’s estimated infection time and most likely infector, colour coded to indicate posterior support. Fig 3 shows one sampled tree for each dataset (from the posterior set of 50,000), matching the MPC consensus tree topology.
The Mtb data were analysed with naïve prior information, which resulted in a median sampling interval of 419 days (similar to estimated incubation times [31]), a median generation interval of 107 days, and a mutation rate equivalent to 0.3–1.3 mutations per genome per year, as estimated before [32, 33]. The Edmonds’ consensus transmission tree (Fig 2A) shows low support for most infectors, which is a reflection of the low number of SNPs, but also of the long sampling interval relative to the generation interval, which makes the sampling time less informative of the order of infection. However, the same index case K02 and three clusters as identified in Didelot et al [7] are distinguished: one starting with K22, one with K35, and the remaining cases starting with K16 or infected by the index case. The main difference compared to the original analysis lies in the shape of the phylogenetic tree and the estimated infection times (Fig 3A). Whereas the topology is very similar, the timing of the branching events is different: in the original tree, internal branches decrease in length when going from root to tips. That shape is consistent with a coalescent tree based on a single panmictic population but also reflects the fact that three mutations separate the two clades after the root node, whereas the posterior median genome-wide mutation rate is estimated at 0.48 per year (mutation rate × sequence length). By taking into account the fact that coalescent events take place within individual hosts, our analysis shows branch lengths that are more balanced in length across the tree. Importantly, this results in a more recent dating of root of the tree: early 2008 (Fig 3A) instead of early 2004 [7].
The MRSA data were analysed with an informative prior for the mean sampling interval mS and a shape parameter aS based on data on the intervals between hospitalisation and the first positive sample. The estimated mutation rate is similar to literature estimates [34, 35], but the posterior median mS of 31 days is considerably higher than the prior mean of 15 days (Table 4). This may be explained by the two health-care workers (HCW_A and HCW_B), which have very long posterior sampling intervals that were not part of the data informing the prior (Edmonds’ consensus tree, Fig 2B). In contrast with the original analysis, we now identify a transmission tree rather than only a phylogenetic tree, resulting in the observation that the two health-care workers may not have infected any patient in spite of their long infection-to-sampling interval. Almost all transmission events with low support (<20%) involved unsequenced hosts. Two of them were identified as possible infector (P5 and P7), in the initial stage of the outbreak, when only few samples were sequenced. This indicates that some unsequenced hosts may have played a role in transmission, but that it is not clear which. Finally, a major difference between our results and those in the original publication is the shape of the phylogenetic tree and the dating of the tree root: around 1st January (Fig 3B) instead of 1st September the year before [25].
Analysis of the FMD2001 and FMD2007 datasets resulted in posterior sampling intervals with means of 14 and 10 days, respectively, close to the 8.5 days estimated from epidemic data [36]. Generation intervals were about the same (Table 4). Both datasets contained more SNPs than the Mtb and MRSA data, with unique sequences for each host and higher mutation rates, similar to published rates in FMD outbreak clusters [37]. This resulted in equal Edmonds’ and MPC consensus transmission trees, and much higher support for most infectors (Figs 2C, 2D, 3C and 3D). Our transmission tree is almost identical to the one from Ypma et al [11], who used a closely related method but did not allow for transmission after sampling. When comparing to the analysis of these data by Morelli et al [9], the topologies of the phylogenetic trees (Fig 3C and 3D) match the topologies of the genetic networks (Fig S18 in [9]), but the transmission trees are quite different. The main differences are that in the FMD2001 outbreak, they identify farm B as the infector of C, E, K, L, O, and P; and in the FMD2007 outbreak, they have IP4b infecting IP3b, IP3c, IP6b, IP7, and IP8. Differences are likely the result of their use of the spatial data [9]. Use of additional data is expected to improve inference, although their estimates of infection-to-sampling intervals (about 30 days) were unrealistically long.
The H7N7 dataset was analysed with the sequences of the three genes HA, NA, and PB2 separately, and combined; with informative priors for both the mean sampling and mean generation intervals. Five parallel chains were run, and mixing was generally good (Table 5); it took about 7 hours on a 2.6GHz CPU to obtain 25,000 unthinned samples in a single chain. Analysis of the three genes combined resulted in a posterior median mS of 8.4 days, slightly longer than the 7 days on which the informative prior was based [38], and longer than in the separate analyses. The mean generation time was slightly shorter than the prior mean: 4.6 days with all genes. We also calculated the parsimony scores of the posterior sampled trees, defined as the minimum numbers of mutations on the trees that can explain the sequence data [39], and compared these with the numbers of SNPs in the data (Table 5). It appeared that with the genes separately analysed, parsimony scores were 6–13% higher than the numbers of SNPs, indicating some homoplasy in the phylogenetic trees (this was not seen with any of the other datasets). The analysis of all genes together resulted in parsimony scores of 18% higher than the number of SNPs. The estimated mutation rates are among the highest estimates for Avian Influenza Virus, as already noted before in earlier analyses of the same data [28, 40]. Fig 4 shows the Edmonds’ consensus tree in generations of infected premises, indicating locations and inferred infection days (full details in S1 Results). Without the use of location data, there is a large Limburg cluster, a Central cluster including two sampled Limburg cases, and a small Limburg cluster of three cases with an exceptionally long generation time (asterisk in Fig 4). A closer look at the sequences makes clear that the first of these cases (L24/34) has 3 SNPs different from assigned infector G4/11, and 4 SNPs different from cases in the large Limburg cluster. Using geographic data as in earlier analyses [12, 30] will probably place these cases within that cluster.
We developed a new method to reconstruct outbreaks of infectious diseases with pathogen sequence data from all cases in an outbreak. Our aim was to have an easily accessible and widely applicable method. For ease of access, we developed efficient MCMC updating steps which we implemented in a new R package, phybreak. We tested the method on newly simulated data, previously published simulated data, and published datasets. Our model is fast: 25,000 iterations took roughly 30 minutes with the Mtb and MRSA datasets of about 30 hosts, and 7 hours with the full three-genes H7N7 dataset in 241 hosts. Two chains with 50,000 posterior samples proved sufficient (measured by ESS) for tree inference (infectors and infection times) and most model parameters with most simulated and published data. The package contains functions to enter the data, to run the MCMC chain, and to analyse the output, e.g. by making consensus trees and plotting these (as in Figs 2 and 3).
Analysis of simulated datasets showed that the sampling times play an important role in transmission tree reconstruction. Firstly, the use of prior information on the sampling interval distribution (shape parameter as well as mean) greatly improves mixing of the MCMC chain (Tables 1 and 3). Secondly, the use of (correct) prior information on the sampling interval distribution can significantly improve infection time estimation as well as transmission tree reconstruction (Table 3). Thirdly, the extent to which sampling times are correlated with infection times determines how well the method is capable of reconstructing transmission trees, which appears from the fact that outbreaks are less well reconstructed with wider sampling interval distributions (Table 1 vs S1 Results) and the low support for the posterior infectors in the Mtb analysis, where sampling intervals were much longer than generation intervals. Therefore, it is advisable to use prior information on sampling intervals in the analysis (if available), and also to base conclusions not only on the summary transmission tree, but also on the posterior support of links in that tree.
We tested the method on five published datasets, with outbreaks of viral and bacterial infections, and in diverse settings of open and closed populations, in human and veterinary context. The method performed well on all datasets in terms of MCMC chain mixing and tree reconstruction. With naive priors on mean sampling intervals and mutation rates, we obtained estimates that were all very accurate when compared to literature, and the inferred transmission trees seemed as good, or even better when considering estimated infection times. With two datasets (MRSA and H7N7) we included some prior information on sampling and/or generation intervals, which mainly affected the inferred infection times, but not so much the transmission trees. It is possible that not all cases have been observed in these outbreaks, especially in the Mtb and MRSA outbreaks, an assumption nevertheless made by our model. If not too many cases are missing, the analyses of simulations show that this does not disturb identification of infector-host pairs that are in the data. It will only limitedly affect identification of transmission clusters, because if a host’s true infector is not in the data, the true infector’s infector is often selected as the most likely infector. Only some of the infection times may have been estimated too early.
For wide applicability, we kept the underlying model simple without putting prior constraints on the order of unobserved events such as infection and coalescence times. Four submodels with only one or two parameters each were used for sampling, transmission, within-host pathogen dynamics, and nucleotide substitution. The sampling model, a gamma distribution for the interval between infection and sampling, has a direct link to inferred infection times, and is the model for which it is most likely that prior information is available from epidemiological data in the same or other outbreaks. We used simulated data to study the effect of uninformative or incorrect prior information on shape parameter aS and mean mS. It appears that an incorrect aS or an incorrect informative prior for mS does reduce accuracy of inferred infection times. However, consensus trees are hardly affected, at least if the number of SNPs is in the order of the number of hosts as we saw in the actual datasets (Table 1 and Table 2 Slow Clock). Only the precision of consensus trees is reduced, i.e. there are fewer inferred infectors with high support. Results with the Fast Clock simulations did show a significant reduction in consensus tree accuracy. In that case, there are so many SNPs that the phylogenetic tree topology and times of coalescent nodes are almost fixed; then, too much variance in sampling intervals (low aS) results in many incorrect placements of infection events on that tree. Possibly, with so many SNPs it could be more efficient to first make the phylogenetic tree, and then use that tree to infer transmission events [7, 10], but it is questionable whether genome-wide mutation rates are ever so high that this becomes a real issue [41].
The submodel for transmission is relevant for transmission tree inference in describing the times between subsequent infection events. Transmission is modelled as a homogeneous branching process, implicitly assuming that there was a small outbreak in a large population, with a reproduction number (mean number of secondary cases per primary case) of 1. If all, or almost all, infectors are in the data, the generation interval distribution reflects the course of infectiousness, separating the cases in time along the tree. This interpretation may be obscured with many unobserved cases, as in the absence of the actual infector, the method often identifies an earlier ancestor in the transmission tree as infector (Table 2). Apart from not taking heterogeneity across hosts into account (an extension we wish to leave for future development, see below), the current model also neglects the possibility that susceptibles can have contact with several infecteds in a smaller population or more structured contact network. That could be modelled by a force of infection, which would more realistically describe contraction of the generation interval during the peak of the outbreak, and provide estimates for relevant quantities such as reproduction ratios [6]. However, it requires information about uninfected susceptibles in the same population and a more complicated transmission model, which is a significant disadvantage when it comes to general applicability, one of our primary aims. More importantly, for transmission tree inference it does not seem to be a problem: the analyses of the published simulations were almost as accurate as in the original publication [12], and these simulations were in very small populations with almost all hosts infected.
The role of the within-host model is to integrate over all possible phylogenetic mini-trees and mutation events within the hosts, and through that, to obtain a posterior distribution of all transmission trees consistent with the (genetic) data. For this, we used a coalescence model based on a linearly growing within-host population, combined with a Jukes-Cantor substitution model. These models contain each only one parameter, but we think that—as long as only few mutations occur in each host, as in our own simulations, the published Slow Clock simulations, and most datasets—for most applications more complex models are not needed for the following reasons. First, the gross structure of the phylogenetic tree topology and branch lengths result from transmission and sampling models, and only the finer within-host details are determined by the within-host model. With only few mutations within each host, precise mini-tree inference is not possible, and for our aim of inferring transmission trees, unnecessary. Second, and confirming this imprecise mini-tree inference, most tree proposal steps include simulation of the within-host phylogenetic mini-trees, resulting in good mixing of transmission and phylogenetic tree topologies. The fact that proposing from the prior distribution works so well indicates that the sequence data do not contain much information on within-host branch lengths. Third, if there are few SNPs, the posterior contains almost only phylogenetic trees with fewest mutations (maximum parsimony). It is therefore not likely that tree inference will improve with more general substitution models. Fourth, inference of transmission trees and infection times appears not to be biased if the underlying simulation model was more realistic (Table 2). If data do contain many SNPs, as in the Fast Clock simulations, more detailed and realistic models for within-host pathogen growth and nucleotide substitution do probably improve inference, especially on the phylogenetic tree. Nonetheless, even then our method was still capable of correctly inferring the infection times and transmission trees with almost the same accuracy as in the original publication.
With two exceptions, the parsimony scores of posterior tree samples were always equal to the number of SNPs in the datasets (the minimum possible). The first exception is the set of Fast Clock published simulations, which had so many SNPs that many of the same mutations had occurred in parallel. The second exception is the H7N7 dataset. In that case, the analyses of the three genes separately resulted in parsimony scores with 6–12 (6%-13%) more mutations than the number of SNPs, whereas the analysis of all genes together resulted in a parsimony score of 313 (median) to explain only 257 SNPs, a surplus of 56 mutations (18%). The results for separate genes could indicate positive selection, confirming the analysis by Bataille et al [28], who even identified specific mutations that had occurred multiple times. The even higher discrepancy for the combined analysis is suggestive of reassortment events, also recognised by Bataille et al [28].
The proposed method and implementation opens perspectives for further extending the methodology to reconstruct phylogenetic and transmission trees from pathogen sequence data. One possible set of extensions arises from changes to the models embedded in our method, to include additional aspects of outbreak dynamics. For instance, the generation time distribution (infectiousness curve) could be made dependent on the sampling interval, which may be relevant for the MRSA outbreak analysis in which the two health-care workers may have transmitted the bacterium until late after infection. This dependence is implicit in methods in which transmission is modelled more mechanistically (e.g. [11, 12, 16]), but we chose not to do that to keep the model more generic. Another important extension would be to relax the assumption of a complete bottleneck at transmission; the bottleneck may be larger in reality [42, 43] and it has previously been relaxed by looking at transmission pairs [44] or modelling it as separate transmission events [18], but not yet in a timed transmission tree. In our model, this would mean that a host can carry multiple phylogenetic mini-trees, rooted at the same infection time to the same infector. A third extension would be to include the possibility of reassortment of genes within a host, primarily motivated by the results of the H7N7 analysis. This may be done by modelling the coalescent process within hosts, the phylogenetic mini-trees, differently for different genes, but constrained by a single transmission tree. Finally, it would be possible to allow for multiple index cases, which may play a role in open populations with possible re-introductions (as in the MRSA setting), or when only a subset of a large epidemic is analysed (the FMD2001 dataset). This is implemented in models using genetic models based on pairwise genetic distances [8, 16] and with a model assuming coalescence at transmission [45], but is considered a major challenge with a within-host coalescent model [46]. Multiple index cases could also reflect unobserved hosts in the outbreak itself, recently addressed by Didelot et al [24] in their two-step approach of first inferring a phylogenetic and then a transmission tree.
A second type of extension would stem from incorporating additional data. An example is the use of data that make particular transmissions more or less likely, such as contact tracing data, or censoring times for infection times per host or transmission times between sets of hosts, motivated by the MRSA dataset in which admission and discharge days are known for each patient. Sampling of infection times and infectors could be constrained by these additional data (as in [12, 30]) and could then become less dependent on the sampling times and sampling interval distribution, as in the current implementation. Another example is the use of spatial data in combination with a spatial transmission kernel, so that the likelihood of infectors includes a distance-dependency, a possible extension motivated by the FMD and H7N7 analyses (as in [9, 30]). A third example is the use of host characteristics to model infectivity as a function of covariates. With the MRSA data, it would then be possible to test for increased infectivity of the health-care workers, or to test for differences in transmissibility in the three wards. In general, the use of additional host data would make dealing with hosts for which a sequence is not available less problematic: the method currently can include these hosts, but without additional data their role is unclear and they are often placed at the end of transmission chains in consensus trees (Fig 2B, Fig 3).
We developed our model for fully observed outbreaks of size n hosts. Data consist of the sampling times S and DNA sequences G, which means that for each host i we know the time of sampling or diagnosis Si and the sequence Gi associated with the sampling time. Hosts without available sequence are given a sequence with noninformative nucleotides (only ‘n’).
We illustrate the method with the following five datasets from earlier publications (all in S1 Data):
The model describes the spread of an infectious pathogen in a population through contact transmission, the dynamics of the pathogen within the infected hosts, and mutation in the DNA or RNA of that pathogen. Furthermore, it describes how these dynamics are observed through sampling of pathogens in infected hosts. We infer the transmission tree and parameters describing the relevant processes by a Bayesian analysis, using Markov-Chain Monte Carlo (MCMC) to obtain samples from the posterior distributions of model parameters and transmission trees (infectors and infection times of all cases). We first introduce the models and likelihood functions; then we explain how we update the transmission trees and parameters in the MCMC chain.
The posterior distribution is given by
Pr(I,M,P,θ|S,G)∝Pr(S,G|I,M,P,θ)⋅Pr(I,M,P,θ).
(1)
Eq (1) is the probability for the unobserved infection times I, infectors M, phylogenetic tree P, and model parameters θ, given the data (sampling times and sequences). The posterior probability can be split up into separate likelihood terms representing the four processes, times a prior probability for the parameters (see S1 Methods):
Pr(I,M,P,θ|S,G)∝Pr(G|P,θ)⋅Pr(P|S,I,M,θ)⋅Pr(S|I,θ)⋅Pr(I,M|θ)⋅Pr(θ).
(2)
We now introduce the four models, the associated likelihoods, and prior distributions for associated parameters.
We use Bayesian statistics to infer transmission trees and estimate the model parameters from the data, and MCMC methods to obtain samples from the posterior distribution. The procedure is implemented as a package in R (phybreak), which can be downloaded from GitHub (www.github.com/donkeyshot/phybreak) and is available on CRAN (cran.r-project.org/web/packages/phybreak/index.html). The package also contains functions to simulate data, and to summarize the MCMC output.
The main novelty of our method lies in the proposal steps for the phylogenetic and transmission trees, used to generate the MCMC chain. It makes use of the hierarchical tree perspective, in which the phylogenetic tree is described as a collection of phylogenetic mini-trees (one for each host), connected through the transmission tree. Most proposals are done by taking one host, changing its position in the transmission tree, and simulating the phylogenetic mini-trees in the hosts involved in that change. In a second type of proposal, the transmission tree is changed while keeping the phylogenetic tree intact. A third type of proposal only resimulates the within-host mini-tree topology, keeping the transmission tree and coalescent times intact.
Initialization of the MCMC chain requires initial values for the six model parameters (aG, mG, aS, mS, r, and μ). The transmission tree is initialized by generating an infection time for each host (sampling day minus random sampling interval). The first infected host is the index case, and for the remaining hosts an infector is randomly chosen, weighed by the density of the generation time distribution. Finally, the phylogenetic mini-trees in each host are simulated according to the coalescent model and combined with one another to create a complete phylogenetic tree.
Each MCMC iteration cycle starts with updates of the transmission and phylogenetic trees, followed by updates of the model parameters. To start with the latter, the parameters mS and mG are directly sampled from their posterior distribution given the current infection times and transmission tree (Gibbs update). This is done by sampling the rate parameters bS and bG, which were given conjugate prior distributions (see above). If TS=∑Si−Ii is the sum of n sampling intervals in the tree, a0,S and b0,S are the shape and rate of the prior distribution for bS, then a new posterior value is drawn as
bS∼Γ(shape=a0,S+aSn,rate=b0,S+TS),
(9)
from which mS is calculated as aS/bS. Posterior values for mG are drawn from a similar distribution, with TG=∑Ii−IMi the sum of n– 1 generation intervals. The parameters r and μ are updated by Metropolis-Hastings sampling; proposals r’ and μ’ are generated from lognormal distributions LN(r,σr) and LN(μ,σμ), i.e. with current values as mean. The standard deviations are calculated based on the expected variance of the target distributions, given the outbreak size for σr, and number of SNPs for σμ (see S1 Methods).
In a single proposal path K, only a new topology of the phylogenetic mini-tree of focal host i is proposed; the coalescent times are kept unchanged.
We took three approaches to evaluate our method: analysis of newly simulated data, analysis of published simulated data [12], and analysis of published observed data. When not specified, the following parameter settings and priors were used: shape parameters for sampling and generation interval distributions aS = aG = 3, uninformative priors for mean sampling and generation intervals with μS = μG = 1 and σS = σG = ∞, and a prior for within-host growth parameter r with ar = br = 3. The prior for log(μ) (mutation rate) is always uniform.
Analyses were done by two MCMC chains, in each taking 25,000 samples (25,000 MCMC cycles). Burn-ins were different: 5000 MCMC cycles for the newly simulated data, 25,000 for the published simulated data [12], and 5000 for the observed data. With the H7N7 data, five MCMC chains were run, with a burn-in of 5000 samples, followed by 25,000 samples. Burn-in lengths of simulated data were based on visual inspection of convergence for two datasets, and then choosing a burn-in period at least 10 times longer than necessary for all the other simulations, followed by comparing ESS and infector sampling in the parallel chains. The Slow clock published simulations had not all converged in the uninformative analysis (S1 Results). For the published data, all chains were inspected visually to confirm convergence.
|
10.1371/journal.pcbi.1002068 | Mutation D816V Alters the Internal Structure and Dynamics of c-KIT Receptor Cytoplasmic Region: Implications for Dimerization and Activation Mechanisms | The type III receptor tyrosine kinase (RTK) KIT plays a crucial role in the transmission of cellular signals through phosphorylation events that are associated with a switching of the protein conformation between inactive and active states. D816V KIT mutation is associated with various pathologies including mastocytosis and cancers. D816V-mutated KIT is constitutively active, and resistant to treatment with the anti-cancer drug Imatinib. To elucidate the activating molecular mechanism of this mutation, we applied a multi-approach procedure combining molecular dynamics (MD) simulations, normal modes analysis (NMA) and binding site prediction. Multiple 50-ns MD simulations of wild-type KIT and its mutant D816V were recorded using the inactive auto-inhibited structure of the protein, characteristic of type III RTKs. Computed free energy differences enabled us to quantify the impact of D816V on protein stability in the inactive state. We evidenced a local structural alteration of the activation loop (A-loop) upon mutation, and a long-range structural re-organization of the juxta-membrane region (JMR) followed by a weakening of the interaction network with the kinase domain. A thorough normal mode analysis of several MD conformations led to a plausible molecular rationale to propose that JMR is able to depart its auto-inhibitory position more easily in the mutant than in wild-type KIT and is thus able to promote kinase mutant dimerization without the need for extra-cellular ligand binding. Pocket detection at the surface of NMA-displaced conformations finally revealed that detachment of JMR from the kinase domain in the mutant was sufficient to open an access to the catalytic and substrate binding sites.
| Protein kinases are involved in a huge amount of cellular processes through phosphorylation, a crucial mechanism in cell signaling, and their misregulation often results in disease. The deactivation of protein tyrosine kinases (PTKs) or their oncogenic activation arises from mutations which affect the protein primary structure and the configuration of the enzymatic site apparently by stabilizing the activation loop (A-loop) extended conformation. Particularly, mutation D816V of receptor tyrosine kinase (RTK) KIT, found in patients with pediatric mastocytosis, acute leukemia or germ cell tumors, can be considered as the archetype of mutation inducing a displacement of the population equilibrium toward the active conformation. We present a comprehensive computational study of the activating mechanism(s) of this mutation. Our multi-approach in silico procedure evidenced a local alteration of the A-loop structure, and a long-range structural re-organization of the juxta-membrane region (JMR) followed by a weakening of the interaction network with the kinase domain. Our results provided a plausible conception of how the observed departure of JMR from kinase domain in the mutant promotes kinase mutant dimerization without requiring extra-cellular ligand binding. The pocket profiles we obtained suggested putative allosteric binding sites that could be targeted by ligands/modulators that trap the mutated enzyme.
| Regulation of physiological functions in the cell is mostly governed by phosphorylation – a crucial mechanism in cell signaling – catalyzed by protein kinases [1]–[5]. Stem cell factor (SCF) receptor or CD117, also known as human receptor tyrosine kinase (RTK) KIT (according to the nomenclature defined in [6]), belongs to the type III RTK family [7]–[10]. Type III RTKs consist of a glycosylated extra-cellular ligand-binding domain (ectodomain) connected to a cytoplasmic region by means of a single transmembrane helix. The cytoplasmic region of KIT is composed of an auto-inhibitory juxta-membrane region (JMR) and a protein tyrosine kinase (PTK) that is subdivided into proximal and distal lobes separated by an insert sequence of variable length (70–100 amino acids). In human KIT, the 77-amino acid kinase insert domain (KID) possesses phosphorylation sites and provides an interface for the recognition of pivotal signal transduction proteins [11]–[13].
Binding of SCF to KIT leads to receptor dimerization [14], [15], intermolecular auto-phosphorylation of specific tyrosine residues [16] and PTK activation [8], [17], [18]. The activation process involves a large rearrangement of the activation loop (A-loop, ∼20–25 residues) situated in the C-lobe of PTK (Figures 1a,b). Conformational switch of A-loop from an inactive packed position (Figure 1a) to an active extended form (Figure 1b) releases access for Mg2+-ATP and protein substrate(s) to the kinase catalytic site [19], [20]. The inactive form of A-loop is maintained by JMR, which inserts directly in the domain interface between the N- and C-lobes of PTK (Figure 1a). JMR is composed of four fragments, namely JM-Proximal at the N-extremity (residues 547–552), the most buried JM-Binder (residues 553–559), JM-Switch (residues 560–570) and JM-Zipper (residues 571–581) [11], [21]. Phosphorylation of its primary sites Y568 and Y570 lifts the auto-inhibition (Figure 1b). Active KIT binds to intra-cellular substrates and phosphorylates them, thereby switching on multiple signaling pathways by interacting with enzymes and adaptor proteins [11], [22], [23]. For instance, the SCF-KIT interaction is essential for the development of melanocytes, erythrocytes, germ cells, mast cells and interstitial cells of Cajal (ICCs) [24]–[27].
The deactivation of tyrosine kinases or their oncogenic activation relates with mutations (point mutations as well as deletions and gene fusions) which affect the primary structure of the protein [28], [29]. A variety of mutations in the gene encoding the proto-oncogene KIT were found in different types of human cancer, in gastrointestinal stromal tumors (GISTs) [30], acute myeloid leukemia (AML) [31], mast cell leukemia (MCL) [32] and human germ cell tumors [33], among others. Mutations inducing tumorigenic effects were identified in the membrane-proximal Ig-like domain D5, in the auto-inhibitory juxtamembrane region and in the protein tyrosine kinase [34], [35]. Longley et al. [36] early proposed a classification of KIT gain-of-function mutations according to their structural and functional locations. The JMR mutations, frequently found in GISTs, are considered regulatory as they disrupt the auto-inhibitory mechanism which negatively regulates the activity of the protein [29], [37], [38]. The PTK mutations are considered catalytic as they directly affect the configuration of the enzymatic site probably by stabilizing the A-loop extended conformation [39]–[41]. To this category belongs the mutation of D in position 816 (indicated as a black sphere on Figure 1 a, b), most frequently substituted by V, found in most patients with mastocytosis, leukemia and germ cell tumors [42]. D816V is also resistant to Imatinib (Gleevec™) treatment [35], which has motivated to study its role in KIT activation mechanisms. Biochemical studies of KIT gave insights into the molecular mechanism of the ligand-independent activation of the D816V mutant receptor [43] but whether dimerization is required remains unclear [15], [44].
Crucial for kinase regulation is the orientation of the highly conserved Asp-Phe-Gly (D810-F811-G812) motif positioned at the N-extremity of A-loop, within the active site (Figures 1a,b). In the inactive state, the DFG triad adopts a “DFG-out” orientation for D810 points out of the ATP-binding pocket, while F811 is oriented toward the site and JMR is bound to PTK; in the active state, a canonical “DFG-in” conformation positions the catalytic D810 in the back of the site for chelation of magnesium while F811 is buried away and A-loop extends toward a completely solvent-exposed JMR. A-loop conformational switch is part of a global movement involving a tilt of the N-lobe towards the C-lobe and the rearrangement of several regions of the receptor such as the glycine-rich P-loop (residues 596–601, in yellow) and C-helix (residues 631–647, in lime) placed in the N-lobe (Figure 1). The link between the conformational changes of the DFG motif and a set of distinctive structural elements of the kinase core has recently been assessed through evolutionary analysis of PKs [45]. In addition, a surface comparison of PKs crystal structures has highlighted the role of F-helix (residues 766 to 786, labeled HF on Figure 1) in the C-lobe as a central scaffold for the dynamic assembly of the active kinase form [46].
The structural properties of PTKs were characterized mainly by X-ray analysis. Although crystallographic data yield valuable insights into such structural rearrangements, they represent only average conformation for a given set of crystallization conditions. Alternative experimental techniques, such as NMR spectroscopy, and computational approaches, such as molecular dynamics (MD) and normal mode analysis (NMA), provide a way to better understand the structure-dynamics-function relationships at the atomic level and further characterize the alteration of protein structure and internal dynamics induced by cancer mutations [47]. These theoretical methods also enable to describe intermediate conformational states, that can be used to guide the design of specific inhibitors acting as modulators of the enzymatic function by targeting putative allosteric sites [48], [49].
Recent (classical or advanced) molecular dynamics studies have begun to elucidate the molecular mechanisms of conformational transitions of PTKs [50]–[54] and to investigate the thermodynamic and mechanistic catalysts of kinase activation by cancer mutations [55]–[57]. Regarding drug-design oriented perspective, the inclusion of normal mode-based descriptions of protein flexibility was shown to improve the prediction of small molecules binding mode to PKs [58]–[61]. A combination of both methods, MD simulations and NMA (elastic network), was employed very recently to elucidate the inactive-to-active state transition of protein kinase B [62].
Early studies, including from our laboratory, shed light on the molecular mechanism by which the D816V mutation destabilizes A-loop inactive conformation, corresponding to a constitutive phosphotransferase activity [40], [63], [64]. This effect was then described in the context of Imatinib-induced resistance [65], [66]. The structure of KIT D816V mutant has not yet been determined. Nevertheless, recent crystallographic data [67] have suggested that the JMR auto-inhibitory conformation is destabilized in the D816H mutant, advocating a regulatory impact of this catalytic mutation. The mechanistic role of JMR in the initiation of the activation process was described by Zou et al. using classical and targeted MD [68]. The authors found that JMR is likely to detach from PTK before the A-loop conformational switch, due to electrostatic repulsion between the C-lobe and phosphorylated tyrosines in JMR.
In this study, we have carried out a detailed analysis of KIT receptor cytoplasmic region structural and dynamic changes related to D816V mutation through extensive description of the protein motions combining MD simulations and NMA. We first applied bioinformatics structural-based tools to accurately assign the secondary structure elements of JMR and A-loop and to characterize the hydrogen bonds stabilizing the active and inactive forms. We then employed homology modeling, MD simulations and free energy calculations to further evaluate the impact of the mutation on the stability of KIT cytoplasmic region auto-inhibited state. We observed both a local structural alteration and long-range structural and recognition effects on A-loop and JMR respectively, induced by the mutation. We then further explored the accessible motions of JMR relative to PTK with NMA and found that JMR is allowed larger amplitude motions in the mutant, promoting its triggering role for the inactive-to-active state transition. Displacements of MD conformations along chosen normal modes combined with pocket detection at the surface of the protein revealed that motions of JMR away from PTK in the mutant lead to the opening of an access to the catalytic and substrate binding sites. Consequently, we reveal D816V-induced alterations of KIT juxta-membrane region structure, dynamics and thermodynamics that were not previously described at an atomic level. We believe that our results may bridge experimental evidence of a regulatory activating role of the mutation and an alternative molecular recognition pattern determining the mutant dimerization.
The crystallographic structures of KIT auto-inhibited inactive and active enzymatic forms (PDB codes: 1T45 [69] and 1PKG [40]) were carefully analyzed to correctly assign their secondary structure and characterize their stabilizing interactions (Figures 1c,d). In the inactive state (Figure 1c), JMR adopts a twisted hairpin (V-shaped) conformation, with the well-structured elements on the external part. The main predicted structural element is an anti-parallel β-sheet (β1–β2). The backbone of β1 (residues 558–561) interacts with the backbone of β2 (residues 569–571) and the backbone of β6 (residues 788–789) from the C-lobe of PTK, through strong and multiple H-bonds. A-loop inactive conformation shows a mixed structure composed of two 310-helices (residues 812–814 and 817–819) adhered by a short coil and an anti-parallel β-sheet (β9–β10) stabilized by backbone-backbone H-bonds. In the active state (Figure 1d), JMR forms an extended coil that is fully solvent-exposed. A-loop is also positioned differently compared to the inactive state and displays a distinct secondary structure. Indeed the 810–820 sequence, featuring two helical motifs in the packed state, forms a β9' sheet (residues 815–816) separated from β10' (residues 823–824) by a turn in the extended conformation. The anti-parallel β-sheet β9'–β10' interacts with β6.
This structural-based analysis reveals that JMR and A-loop, two segments of KIT receptor cytoplasmic region capable of large conformational changes, exhibit distinct structural elements in the inactive and active forms of the enzyme, which are stabilized by peculiar hydrogen bonds. These elements are organized in such a way that residues from JMR and A-loop take the part of each other in the two structures interaction networks. The inactive packed conformation of A-loop is stabilized by intra-loop H-bond binding, whereas the active extended form is stabilized by interactions with the other regions of the receptor.
Noticeably, JMR and A-loop are the preferred regions of gain-of-function point mutations [11] (residues encircled in Figures 1c,d). The majority of these mutational hot spots participates in the stabilization of either the inactive (1T45) (Figure 1c) or active (1PKG) conformation (Figure 1d). For example, in the inactive form, V559 and V560 of β1 establish H-bonds with I571 of β2 and N787 of β6, respectively (Figure 1c). D816 serves as a negative capping for the 817–819 helix in 1T45 whereas it is a non-interacting residue of β9' in 1PKG, representing the active conformation (Figure 1d). D820 and N822 interact through their side-chains and participate in the formation of the 820–823 β-turn in the inactive state (Figure 1c). Y823 is involved in intra-loop interactions that stabilize turns in both states. Furthermore, this tyrosine is positioned in the catalytic site and H-bonded either to R787 or to the catalytic residue D792 in 1T45 (Figure 1c), whereas it bridges the anti-parallel β-sheet β10'–β11 by interacting with Y846 in 1PKG (Figure 1d). Its aromatic side chain was shown to be essential for the stabilization of the inactive conformation [70]. Mutation of any of these residues, except for V560, also confers resistance to Imatinib [35]. This analysis thus highlights the polymorphous structural properties of JMR and A-loop, tolerating mutations which provoke the deregulation of the kinase activity without altering the integrity of its structure.
Four 50-ns MD simulations of full-length KIT receptor cytoplasmic region (CR), in wild-type (WT547-935, crystallographic) and D816V-mutated (MU547-935, modeled) forms, were run to explore and compare the protein internal dynamics and energetics. Two additional 50-ns simulations were carried out for KIT domains (WT567-935 and MU567-935), where residues 547–566 of JMR – disordered in the active structure 1PKG – were removed. Further we shall refer to these different forms of KIT CR as full-length (547–935) and truncated (567–935) and we shall identify the two MD simulations of WT547-935 by indices 1 and 2, and the same for MU547-935.
To analyze the global behavior of the studied systems, the root mean square deviations (RMSDs) of the nitrogen and carbon atoms of protein backbone with respect to the initial frame were plotted versus simulation time (Figure 2a). All four trajectories of full-length KIT CR, WT547-935 and MU547-935, display comparable backbone conformational drifts with RMSD mean values in the range 2.35–2.77±0.33–0.54 Å. Corresponding values for the truncated forms tend to be larger and vary much more, mean values of 3.39±0.74 Å and 3.59±1.05 Å for WT567-935 and MU567-935, respectively (Figure S1a). The evolution of RMSD during the course of the MD trajectories indicates a reasonable stability of the systems after a 2-ns relaxation period. Hence, the last 48 ns of each trajectory were considered as productive simulation time for further analyses. To find out which parts of the protein deviate most from the initial template, the RMSDs of backbone atoms were monitored for the N-lobe, C-lobe, A-loop and JMR, separately. The RMSDs of the N-lobe remained stable along all simulations except for simulation 1 of WT547-935 where it increased by 1.5 Å after 16 ns (Figure 2b). Also in this simulation, the RMSD of the C-lobe reached a stable level earlier than in the other runs (Figure 2c). The RMSD curves for N- and C-lobes of the truncated forms show comparable profiles to those of the full-length CRs (Figures S1, b–c). A-loop displays a rather small deviation in both simulations of WT547-935 and in simulation 1 of MU547-935, with mean values in the range 1.26–1.43±0.25–0.35 Å, but the drift increased after 27 ns in simulation 2 of MU547-935 to reach a maximum of 4.58 Å (Figure 2d). The deviations of JMR are significantly larger than those of the other regions, with mean values in the range 5.26–5.90±1.07–1.65 Å, and their fluctuation profiles are unstable (Figure 2e).
The conformational stability of the studied systems was investigated through a convergence analysis of the trajectories [71]. Briefly, a set of reference structures are picked up randomly among the MD conformational ensemble and reference groups are formed, composed of conformations from the two halves of the trajectory (see Materials and Methods). A good convergence quality can be assessed when each reference structure is more or less equally represented in both halves of the trajectory. One defines a lone reference structure as a reference structure that is not represented in one half of the trajectory (one empty reference group) (insert in Table 1). To ensure the robustness of the method, the analysis was run with five different random seeds for the reference structure picking up (Table 1) similarly as we performed early [72]. Lone reference structures were found in almost all runs for the full-length forms, indicating a mean convergence quality for these simulations. The results were much improved for the truncated forms, especially regarding the wild-type. KIT wild-type cytoplasmic domain appears thus less stable in its full-length form than when the JMR is cleaved. The use of a 2.5 Å RMSD cutoff led to the identification of more reference structures in WT547-935 trajectories, indicative of a greater conformational diversity, than in MU547-935 trajectories (Table 1). By contrast, the optimal cutoff for WT567-935 (r = 3 Å) was smaller than that of MU567-935 (r = 3.5 Å). Consequently the D816V mutation seems to reduce the conformational variability of the full-length form but enhance that of the truncated form.
The thermodynamic effect of the activating D816V mutation could be quantified by combining the equilibrium MD simulations with the Molecular Mechanics Generalized Born Surface Area (MM-GBSA) analysis [73] of KIT stability changes. Free energies were averaged over 2,400 conformations taken at 20-ps time interval along simulations 1 and 2 of WT547-935 and MU547-935 (Figure 3a). Errors on estimates were calculated using the method of Straatsma [74], useful for evaluating the uncertainty of finite correlated series [75], [76]. Predicted errors for the entropy components (TStrans, TSrot, TSvib, TS) were small in all simulations (up to 1.87 kcal/mol), reflecting very good convergence properties. Predicted errors for the enthalpic components (Eele, Evdw, Eint, Egas, Gsa, Ggb, H) were larger (up to 24.84 kcal/mol). Nevertheless, error compensation occurred between large quantities and the auto-correlation functions suggested good convergence properties for the total enthalpy contribution H in simulation 2 of WT547-935 and simulation 1 of MU547-935. These two time series were thus retained to illustrate KIT enthalpic, entropic and total energy changes upon mutation on Figure 3a, although one should keep in mind that energy changes between WT547-935 and MU547-935 show the same trend whatever simulations considered.
We observed that the D816V mutation induced a significant decrease in the thermodynamic stability of KIT receptor cytoplasmic region autoinhibited inactive state (Figure 3a, on the right). This detrimental energetic effect was mainly due to loss of electrostatic interactions (Eele), whereas favorable van der Waals contributions were gained (Evdw) (Figure 3a, on the left). A slight reduction in conformational entropy (TStot) was observed.
To qualitatively estimate the energetic impact of D816V mutation on KIT active versus inactive states, free energies were computed on the equilibrated inactive and active conformations of KIT truncated CR in wild-type and mutant forms (Figure S2, and Materials and Methods). We observed that the D816V mutation has a deleterious effect on both inactive and active conformations. However, the decrease in thermodynamic stability is smaller for the active form, so that the free energy difference between inactive and active conformations is reduced in the mutant compared to the wild-type.
Our free energy calculations thus indicate a deleterious impact of the D816V mutation on the stability of KIT receptor autoinhibited inactive state and enable to quantitatively relate the associated energy changes. They further suggest that the mutation may modify the energetic balance between inactive and active conformations. This finding is in excellent agreement with the data supporting a regulatory role for the D816V mutation [67] and correlates with similar results obtained for other kinases [55], [56].
We analyzed the MD conformations in details to investigate the mutational effects of D816V on the internal structure and dynamics of KIT cytoplasmic region and understand what changes induced by this mutation promote the increased exchange rate between inactive and active states experimentally evidenced in [67]. MD snapshots taken at regular time intervals show that the A-loop position is systematically shifted between wild-type and mutated proteins (Figure 4a). For instance the small 817–819 helix, identified in our structural-based analysis of the inactive state is preserved in WT547-935 (Figure 4a, see in particular 38-ns snapshots) whereas it is unfolded in MU547-935. Secondary structure assignments averaged on MD trajectories confirm a 8% loss of helical structure (42% total structure loss) between residues 817 and 819 induced by the mutation (Figure 4b, right panel). This local destabilization results from the replacement of the negative capping D816 by a hydrophobic valine. Indeed H-bonds were recorded between the backbone and side-chain oxygen atoms of D816 and the backbone nitrogen atoms of K818, N819 and D820 for 22, 25 and 13% of WT547-935 total 96-ns productive simulation time whereas no H-bond was observed between V816 and residues 815–820 in MU547-935 simulations. Consistently, the computed solvent accessible surface area (SASA) of R815, mutated V816 and I817 were increased by 53, 31 and 84% in MU547-935 compared to WT547-935. The 817–819 helix unfolding was also assessed in the truncated form MU567-935. In addition, the helical contribution for residues 812–814 following the DFG motif is significantly larger (by 37%) in MU547-935 (Figure 4a, see in particular 26-ns snapshots) compared to WT547-935. This structural gain is counter-balanced by an equivalent decrease in the contribution of partially organized structure (turn) (Figure 4b, right panel). As a result, the proportion of helical/turn structures in this region are nearly identical in MU547-935, whereas it is displaced to the turn structure in WT547-935. The remaining part of A-loop (residues 820–835) shows a conserved structure between wild-type and mutated KIT. Consequently, we evidenced that the mutation D816V provokes an important alteration of the local structural organization of A-loop adjacent sequence regions.
Apart from this local effect, we noted that the distantly positioned JMR rapidly adopted a well-shaped anti-parallel β-sheet structure in MU547-935, moving from a position packed to the C-lobe toward an axial position, whereas it retained its non well-ordered structure in WT547-935 (Figure 4a). Effective structural re-organization of residues 568–572 upon mutation can be assessed based on secondary structure assignment performed on the MD trajectories (Figure 4b, left panel). To clarify this observed long-range structural effect, the interaction network between JMR and PTK was characterized by recording H-bonds and hydrophobic contacts in WT547-935 (Figure 5, upper panel) and MU547-935 (Figure 5, lower panel). To describe in details JMR interaction network, we considered its fragments separately, JM-Proximal (JM-P), JM-Buried (JM-B), JM-Switch (JM-S) and JM-Zipper (JM-Z). To visualize the established contacts occupancy, we used a color gradient from red to blue for strong to weak interactions, respectively. Upon mutation, both types of interactions, H-bonds and hydrophobic, vanish between JM-P/JM-B (residues 550–553) and the N-extremity of C-helix (residues 632–633) (Figure 5). H-bonds between JM-Z (residues 573–576) and C-helix (residues 640–642) (Figure 5a) and hydrophobic contacts between JM-P/JM-B (residues 551–553) and both P-loop and A-loop (Figure 5b) are weakened in the mutant. Noticeably, in the mutant JM-S also interacts less strongly with the C-lobe, including with β6. The two primary phosphorylation sites (JM-S) apparently swap their role in the interaction network between wild-type and mutated forms. Indeed Y846 in the C-lobe establishes a H-bond either with Y570 (46%) in WT547-935 or with Y568 (27%) in MU547-935 (Figure 5a). Moreover a hydrophobic contact between Y570 and I789 (β6) persistently exists in WT547-935 (55%), while it is not observed in MU547-935. Consequently, the long-range effect of the D816V mutation leading to noticeable structural re-organization and shifted position of JMR, is accompanied by an alteration of the interaction network between JMR and PTK.
The Cα atomic fluctuations depicted by ellipsoids on the averaged MD conformations (Figure 5) show two highly flexible clusters, with fluctuations between 2.4 and 4.5 Å corresponding to JM-S and A-loop in both wild-type (residues 563–567 and 827–829) and mutant (residues 563–567, 817, and 825–830). By contrast, JM-Z (residues 581–582), the loop preceding C-helix (residues 630–632) and the loop preceding G-helix in the C-lobe (residues 871–873 and 877–878) display high flexibility with values above 2.4 Å up to 3.6 Å in WT547-935 (Figure 5, upper panel), whereas their fluctuations are much reduced in MU547-935 (Figure 5, lower panel).
To further characterize KIT receptor cytoplasmic region motions in the inactive state, principal component analysis (PCA) of the MD trajectories was performed. 26 and 28 PCA modes are sufficient to describe 90% of the total backbone fluctuations of WT547-935 and MU547-935, respectively. The first three PCA modes cumulative contribution is 56% for the wild-type and 53% for the mutant (Figure 6a). Computed scalar products between the first ten PCA modes from the two proteins indicate that the correspondence is not straightforward between the two ensembles. However, the second principal modes share a high degree of 66% similarity (Figure 6b). Noticeably, mode 2 of WT547-935 bears a significantly larger contribution (20%) than mode 2 of MU547-935 (14%) (Figure 6a), and displays a two-fold higher degree of collectivity k (see Materials and Methods). Indeed, it illustrates atomic motions of JMR coupled to deformations of PTK in the N-lobe – interface with JM-Z and residues 627–633 preceding C-helix, and in the C-lobe – A-loop and residues 868–886 including G-helix (Figure 6c, left panel), whereas mode 2 of MU547-935 describes atomic motions of JMR independent from PTK (Figure 6c, right panel).
Consequently, the D816V mutation alters the global dynamics of KIT receptor cytoplasmic region, in particular regarding the participation of JMR in the main motions of the protein. The coupling between JMR and the N-lobe revealed by the PCA in wild-type KIT can be related to the large atomic fluctuations observed in the N-lobe of this form. By contrast in the mutant, no coupling is observed and fluctuations in the N-lobe are much smaller. These findings are also in agreement with the greater conformational variability of the wild-type over the mutant evidenced by the convergence analysis.
PCA applied on the MD trajectories of the truncated proteins (WT567-935 and MU567-935) reveals that independent motions of the residues 567–581 of the JMR portion (JM-Z and part of JM-S) are dominant in the total backbone fluctuations of the protein, contributing up to 52% and 74%, respectively. Indeed, the reduced JMR is highly solvent-exposed and flexible. It displays larger fluctuations in MU567-935 than in WT567-935 (Figure S3) and thus appears to be responsible for the greater conformational variability of the cleaved mutant over the cleaved wild-type evidenced by the convergence analysis. Noticeably, the A-loop is not further destabilized in these cleaved forms and its positions at the end of the simulations superimpose well between WT547-935 and WT567-935 on the one hand, MU547-935 and MU567-935 on the other hand (Figure S3). This observation is consistent with a recent biochemical characterization of KIT cytoplasmic domain showing that the cleavage of JMR does not automatically promote inactive-to-active transition of the A-loop [70].
To elucidate the thermodynamic origin of the structural and dynamics changes induced by the mutation in the remote JMR, we evaluated binding free energy changes between the equilibrated conformations of wild-type and mutated KIT (Figures 3c), where no structural re-organization of JMR was yet observed. Single-point MM-GBSA calculations were performed following the thermodynamic cycle shown on top of Figure 3c to estimate the relative attachment of JMR to PTK. We found a global binding free energy change (ΔΔG) of −42.68 kcal/mol, indicating that JMR is more tightly attached to PTK in the wild-type than in the mutant, due to a largely more favorable (mainly caused by the vibrational component) entropy (Figure 3c, table at the bottom). The greater conformational variability displayed by WT547-935 compared to MU547-935 in the simulations may enlighten this entropic effect. Indeed, the penalty endorsed by JMR and PTK upon binding due to reduction of their intrinsic vibrational entropy may be balanced in the wild-type by an emerging cooperativity between the two domains, as suggested by the PCA. By contrast in the mutant, the absence of cooperativity may lead to a larger entropic penalty upon binding. Binding free energy changes computed for the different JMR fragments show a dominant entropic penalty for JM-B and JM-S binding to MU547-935 compared to WT547-935. The largest energy change is observed for JM-Z – covalently bound to PTK, due to both large enthalpic and entropic penalties in mutated KIT. The smallest energy change is obtained for the solvent-exposed extremity JM-P. Overall, these calculations reveal the thermodynamic determinants responsible for the alteration of JMR structure and dynamics upon mutation and they enable to formulate the hypothesis that JMR is less tightly attached to PTK in the mutant than in the wild-type.
To further explore the motions accessible to JMR, all-atom Normal Mode Analysis (NMA) was conducted on representative MD conformations. Based on our convergence quality assessment (Table 1), we retained simulations 1 of WT547-935 and MU547-935 as they displayed the smallest number of lone reference structures. Two sets of four MD conformations – extracted through clustering analysis (see Materials and Methods), were considered, taken at: 4217, 34238, 42356, 49260 ps for WT547-935 and 2531, 19157, 30160, 36987 ps for MU547-935. These sets enabled to get the best convergence quality and thus were the most representative of the MD conformational ensemble. For comparison, NMA was also performed on the static crystallographic structure 1T45 [69]. The 97 non-zero lowest-frequency modes (ω<20 cm−1) obtained from each NMA were considered, leading to a total of 97 values for the X-ray structure and 388 values for the wild-type and the mutant respectively. On this ensemble, the degrees of collectivity of JMR atomic motions, , were computed, with values ranging from (only one atom among the total n involved in the motion) to 1 (high collectivity). The mean value is 0.57 for WT547-935 and 0.60 for MU547-935, with 48% and 54% of the values above 0.6 respectively (Figure S4). This indicates that JMR atomic motions illustrated by the normal mode ensemble are overall more collective in the mutant than in the wild-type. For comparison, a significantly lower mean value = 0.51 is found for the X-ray structure, underlining the relaxation of the protein in the MD simulations.
All calculated normal modes represent a limited ensemble of motions, as several modes significantly overlap. We describe qualitatively some of the modes which exhibit motions of JMR relatively to PTK. Three modes from WT547-935 and three modes from MU547-935 were picked up for their displaying of JM-Z and/or JM-S large displacements (Figure 7 and Table 2). In WT547-935, mode 18{42356-ps} shows a collective motion of JMR of especially large amplitude for JM-Z (Table 2), coupled to a rigid-body motion of C-helix (Figure 7, upper left panel). In MU547-935, mode 7{30180-ps} displays an even larger resultant displacement of JM-Z but a rather low (Table 2), indicative of disparities in the mode atomic components (Figure 7, lower left panel). Large displacements of JM-S are displayed in modes 21{49260-ps} of WT547-935 and 17{2531-ps} of MU547-935 (Table 2). JM-S concerted atomic motions are rather coupled to JM-P and the loop preceding C-helix in mode 21{49260-ps} of the wild-type (Figure 7, upper right panel) and to JM-P and G-helix in mode 17{2531-ps} of the mutant (Figure 7, lower right panel). Modes 29{34238-ps} of WT547-935 and 16{30180-ps} of MU547-935 illustrate combined displacements of JM-Z and JM-S with above 0.6 (Table 2). Concerted JMR atomic motions are oriented toward the back of PTK in mode 29{34238-ps} of the wild-type (Figure 7, upper middle panel) while the arrows representing JMR atomic motions in mode 16{30180-ps} of the mutant point away from PTK and are more numerous (Figure 7, lower middle panel), corresponding to a very high degree of collectivity (Table 2).
The selected normal modes reveal differences in the motions accessible to JMR in wild-type and mutated KIT. Consistent with the PCA results, the JMR atomic displacements in the wild-type are coupled to deformations in PTK, in particular to motions of C-helix and its preceding loop; in the mutant, the JMR atomic motions are more independent from PTK. In particular, mode 16{30180-ps} represents a possible way-out of JMR from PTK through a highly collective motion.
The wild-type and mutated structures were displaced up to 4 Å with a step of 0.1 Å along each selected normal mode in both positive and negative directions (Figure S5). The extreme conformations obtained from the 4-Å displacements of JMR away from PTK enable to produce JMR atomic motions (Figures 8c–h). Based on visual inspection of the modes components, C-helix appears more attached to JMR in wild-type conformations, resulting in deformations of PTK, than in mutated conformations (Figure 8). Furthermore, some extreme conformations show a coil structure of A-loop in the wild-type and mutant (Figure 8d, h).
A close inspection of the surface of wild-type and mutated proteins permitted to identify pockets on the NMA-displaced conformations and on the crystallographic structures 1T45 and 1PKG (Figure 8a,b). According to X-ray data, in the inactive state the extended catalytic region shows two adjacent pockets (Figure 8a, the areas colored in red and orange); while in the active state, we observed the occurrence of three adjacent pockets form the ATP-binding site and the substrate binding site (Figure 8b, the areas colored in red, green and purple, respectively). These three pockets are also detected in some extreme conformations obtained from the 4-Å displacement in mutant (Figure 8g and h). A more fragmented pocket profile is observed in conformations (f) of the mutant and (e) of the wild-type, with an additional small pocket (in olive). Consequently, pocket search applied to conformations obtained from NMA reveals that displacing JMR relative to PTK leads to the opening of a “path” to the catalytic site, even though A-loop remains in its inactive state. The access to the catalytic site is particularly facilitated in mutated KIT, whose structure (g) displaced along mode 16{30180-ps} displays a “path-of-pockets” very similar to that observed in X-ray structure 1PKG, in terms of volumes and shapes. Consistently, pocket search performed at the surface of the 50-ns MD conformations of WT567-935 and MU567-935 revealed a similar “path-of-pockets” in the mutant truncated CR (Figure S6). Overall these thorough normal mode analyses suggest that the D816V mutation may promote spontaneous detachment of JMR from PTK and concomitant access release to the substrate and ATP-binding sites, hence reinforcing the triggering role of JMR in the inactive-to-active state transition of the protein.
This study represents a detailed description at the atomic level of the impact of the D816V mutation on KIT cytoplasmic region structure, internal dynamics and thermodynamic stability and contributes to the basic concepts of the activating/deactivating mechanisms of RTKs. Unlike for many other kinases, the activation of type III RTKs such as KIT does not require the phosphorylation of the activation loop [70]. Instead, the primary phosphorylation sites are located in the juxta-membrane region, whose detachment from its auto-inhibitory position is likely to trigger the inactive-to-active state transition, due to repulsive negative charges [68]. Our structural-based bioinformatics analysis of KIT receptor auto-inhibited inactive and active states highlighted the strong polymorphous character of both A-loop and JMR and their crucial stabilizing roles for either conformation. We also found that mutational hot spots located in these two elements of the receptor play important role in the secondary structure stabilization and H-bond patterns of the protein. In particular, residue D816 in A-loop serves as a negative capping for a helical motif in the inactive form whereas it stands within a β-sheet in the active form.
Inspired by these preliminary observations, we were interested in characterizing the effect of the mutation on the auto-inhibition mechanism of KIT receptor cytoplasmic region. Hence we chose to explore the conformational space around KIT auto-inhibited inactive state, in wild-type and mutated forms, using multiple 50-ns MD simulations. The striking similarity between wild-type KIT (1T45 [69]) and the mutant D816H (3G0F [67]) suggested that the folding of the protein would be similar in the context of the mutant D816V.
In our simulations, A-loop demonstrated high flexibility, consistently with NMR studies [77]. We also evidenced a local structural alteration induced by D816V on A-loop inactive conformation. In particular, we observed unfolding of the small 817–819 310 helix in the mutant. As we pointed out in the introduction, this local effect has been previously characterized [40], [63], [64], [67]. By contrast, the behavior of the juxta-membrane region had not yet been explored in the context of the mutant. Our simulations revealed a long-range structural re-organization of JMR and a conformational drift of JM-Switch segment, from a position packed to the C-lobe to an axial arrangement, in the mutated form. Comparing our data with those obtained during the first third of a targeted MD simulation of the inactive-to-active state transition of wild-type KIT cytoplasmic region [68]), we propose that the drift we observed could be the first step towards the ligand-independent activation of D816V mutant.
Our recording of the hydrogen bonds and hydrophobic contacts gave a possible justification for this long-range effect, through the weakening of the interaction network between JMR and both N-lobe and C-lobe of PTK upon mutation. Furthermore, quasi-harmonic analysis (PCA) of the trajectories and computed free energy changes revealed that the mutation has a deleterious impact on the thermodynamic stability of the inactive state and on the coupling between JMR and catalytic domain. In the literature, differences between homology models of active wild-type KIT receptor kinase domain and the mutant D816V early suggested a strong influence of JMR on the folding of wild-type PTK but not on that of D816V-mutated PTK [64]. Moreover, it was reported recently that mutation in position 816 shifts the conformational equilibrium of the kinase away from the auto-inhibited state toward JMR being released to solvent and disordered [67]. These in silico and in vitro results are in good agreement with the coupling/decoupling balance put in light here between the wild-type and mutated proteins.
In an attempt to go beyond crystallographic structures or homology models static view and to get a qualitative insight into the modification of the protein energetic landscape upon mutation, we have conducted normal mode analysis on representative conformations sampled in our MD simulations. This method is inscribed in the same philosophy as consensus normal modes theoretical framework [78]. Including the well equilibrated first hydration shell from MD simulations permitted to obtain modes with good accuracy regarding JMR, which is located at the surface of the protein. Statistics computed on the NMA ensembles revealed more collective motions of JMR in the mutant. On one side, the overlap between different starting points reflects conformation population equilibrium; on the other side, the identification of particular modes for particular conformations relates to the search for a discrete transition path between inactive and active states. In this regard, several modes were chosen that described possible ways-out for JMR to depart from catalytic domain. A pocket search at the surface of the conformations displaced along these modes revealed that JMR accessible displacements relative to mutant PTK were sufficient to open a path of adjacent pockets to the substrate-binding sites. This observation was also confirmed by the MD simulations of the truncated proteins, where JMR was cleaved.
Noticeably, proto-oncogenic mutations located downstream of the DFG motif in the A-loop, as is the case of D816V, were identified in at least seven other kinases, including BRAF (V600) and EGFR (L858/L861) [79]. MD studies of EGFR evidenced a deleterious impact of mutation L858R on the thermodynamic stability of the protein inactive state [55], [79], the effect described here for KIT mutation D816V. It was also revealed that conformational changes in L858R and L861Q EGFR mutant, taking place in the A-loop and C-helix, may facilitate the inactive-to-active state transition. Regarding BRAF, Xie et al. proposed a mechanism by which mutation V600E mimics the effect of phosphorylation events in the A-loop, thus disrupting the kinase inactive conformation toward the active state [80]. These results match with our observations, according to which KIT mutation D816V favors departure of JMR from PTK, a process that is normally induced by phosphorylation events in the wild-type protein.
Another topic can be considered in the discussion: the question of whether dimerization is mandatory for D816V mutant activity. Kanakura et al. proposed that the D814V mutant of murine KIT (equivalent to D816V mutant of human KIT) acts as a dimer, the dimerization interface is not located in the ectodomain and the last exerts only negative regulation on the ligand-independent dimerization process [44]. Later, structural studies of KIT ectodomain illustrated this supposed negative regulation by revealing a large conformational change between monomeric and dimeric forms [14], [15]. Monomeric ectodomains encounter electrostatic repulsion through their domain D4, maintaining receptors at a minimum distance from each other. Upon SCF binding, domains D4 and D5 twist and form a contacting interface (Figure 9a). Recently, Bougherara et al. showed that D816V mutant was able to induce downstream oncogenic signaling without the need to reach the cell surface [43]. Our results shed new light on these experimental data. Our description of the molecular mechanism by which the activating D816V mutation promotes spontaneous detachment of JMR from PTK, the triggering first step of the enzyme inactive-to-active state transition, may reconcile the views of a functioning dimeric mutant, yet a mutant activated without the need for extra-cellular ligand binding. Indeed, the greater freedom of movement of JMR in the mutant implies an increased longest dimension of KIT cytoplasmic region, suggesting that JMR could act as an arm able to extend from PTK toward other interacting partners such as another KIT kinase monomer (Figure 9b). Under such hypothesis, dimerization and transphosphorylation of KIT kinase would still hold as the activation mechanism of the mutated enzyme. Recent biochemical and structural characterization of RTK dimers has shown a great variety of interfaces [81]. It was suggested that FGFR1 – another receptor tyrosine kinase – ectodomain dimer formation imposes steric constraints that reduce the number of possible interaction modes between kinase domains [82]. As a consequence, loss-of-function mutation located in the interface prevents in vivo activation of the receptor. In a reciprocal manner in KIT, according to the model we propose, the gain-of-function mutation D816V could permit the kinase domains to bypass the repulsion between ectodomains in the absence of ligand, allowing for receptor activation.
This study proposes atomic level description of the regulatory impact of the D816V mutation, through local and remote structural/dynamic changes. The conformational exploration of the kinases – particularly the receptor tyrosine kinase KIT - specific inactive states presents an obvious therapeutic interest. Understanding of the regulation/deregulation of kinase activation contributes to the design of novel generation of inhibitors targeting KIT and other structurally or functionally related kinases. As an illustration, the efficiency of Gleevec™ to treat chronic myelogenous leukemia (CML) and gastrointestinal stromal tumors (GIST) is a consequence of its capacity to bind to and stabilize the inactive form of KIT and its binding preferences are governed by conformational selection [83]. The multi-approach procedure we applied on KIT inactive structure enabled us to postulate a model where the mutated kinase is able to dimerize without the enlisting of the extra-cellular domain. Very recent unpublished data (Schlessinger, personal communication, 2010) suggest that two populations of wild-type KIT dimers coexist in the cell, both showing a symmetric arrangement of the extra-cellular domain but an asymmetric arrangement of the kinase domain. Those data could be used in the future to construct reliable models for the homo (mutant/mutant) or hetero (wild-type/mutant) dimers of KIT receptor. The extensive NMA study of both wild-type and mutant KIT we conducted and the careful selection of a set of relevant modes could be further exploited to determine a plausible conformational transition pathway between the inactive and active states. The pocket profiles we obtained also encourages us to search for putative allosteric binding sites that could be targeted by small molecules that would trap the enzyme in an active conformation. Compared to orthosteric sites, allosteric sites are less-well conserved and thus ligands acting at allosteric sites have a greater potential to achieve receptor selectivity.
The crystallographic structures representing the auto-inhibited inactive and active states of KIT cytoplasmic region (PDB entries: 1T45 [69] and 1PKG [40]) were analyzed using the bioinformatics tools DSSP [84] and Stride [85]. (i) The secondary structure elements were assigned with the two algorithms, based either on the inter-molecular H-bonds (DSSP) or on both backbone geometry and inter-molecular H-bonds (Stride). (ii) Hydrogen bond networks (interactions D–H•••A, where D is the H-donor atom, A is the H-acceptor atom) were characterized with HBPLUS 3.2 [86] and visualized with PyMOL 1.2 [87] and Maestro (Schrödinger LLC, New York NY). The H-bond assignment was made using geometrical criteria [88], [89].
Free energies were evaluated using the Molecular Mechanism Generalized Born Surface Area (MMGBSA) method – implemented in AMBER 10 [73], [103]–[105] which proposes to express the total free energy of the protein as a sum of contributions(1)where H is the enthalpy and TS is the configurational entropy of the solute. Egas is the molecular mechanics energy of the solute, obtained by summing the internal energy Eint, the electrostatics interactions Eele and the van-der-Waals contacts Evdw; Ggb is the polar solvation term whose evaluation is based on the continuum generalized Born solvent model; Gsa is the non-polar solvation term, proportional to the solvent accessible surface area (SASA), and was evaluated using the Linear Combinations of Pairwise Overlaps (LCPO) method. The translational TStrans, rotational TSrot and vibrational TSvib entropies, were evaluated through normal mode calculations with the NMODE module, using a dielectric constant of , where is the distance between atoms i and j.
In principle, considering a two-state model, the calculation of the free energy difference between the wild-type and the mutant inactive-folded and unfolded states would be required to evaluate the protein stability changes. Following the assumption that, in the unfolded state, individual residues may not interact and hence the contributions of all the residues except the one under mutation (D816V) are the same in the wild-type and the mutant, we considered that this difference between one-residue contributions should be small compared to the difference between the total energies of the wild-type and mutated folded states. We thus evaluated the thermodynamic stability difference for the inactive state directly from our MD simulations of WT547-935 and MU547-935, as:(2)
The quantities and were averaged over 2400 snapshots selected at 20-ps intervals along the four 50-ns MD simulations and the protein stability change was then approximated based on the free energy difference . Statistical errors on estimates were calculated from their variance and auto-correlation function using the method of Straatsma [74]:(3)where var(G) is the variance of the estimate, is the correlation length from the relaxation of the autocorrelation, and is the total length of the time series.
The relative free energies of active versus inactive equilibrated conformations were also evaluated for wild-type and mutant KIT truncated CR (single-point calculations).
The free energy of binding a ligand to a receptor is defined as:(4)
Here we computed the binding free energy of JMR and its fragments, the remaining parts of the protein being considered as the receptor, to get estimates of the different energetic contributions involved in the attachment of JMR to PTK. The binding free energies were evaluated on the equilibrated conformations of WT547-935 and MU547-935 (single-point calculations). The change in JMR (or its fragments) relative stability within the protein was then approximated based on the difference between the binding energies and .(5)
Normal mode analyses (NMA) were conducted using the DIMB method [106] of the VIBRAN module of CHARMM 35b3 [107], [108] on (i) the crystallographic structure 1T45 [69], (ii) MD conformations from WT547-935 taken at 4217, 34238, 42356, 49260 ps, (iii) MD conformations from MU547-935 taken at 2531, 18157, 30180, 36987 ps. The selected MD conformations were found to be the most representative of the trajectories, according to the convergence analysis. The first hydration shell (<5 Å, 2200 water molecules) around the MD conformations was kept to help prevent the solvent-exposed regions of the protein from collapsing during the minimization procedure [78]. During initial steepest descent energy minimization of the system, mass-weighted harmonic constraints (250 kcal/mol/A2) were applied to the starting structure and reduced by a factor of 2 every 1000 minimization steps until they fell below a threshold value of 5 kcal/mol/A2. The constraints were then removed and the system was minimized by conjugate gradient and adopted-basis Newton-Raphson steps until the RMS energy gradient fell below 10−5 kcal/mol/A. The Cα RMS deviations of the minimized conformations from their initial position were limited to less than 1 Å (Table S2). Normal modes were computed by diagonalizing the mass-weighted Hessian matrix of the energy-minimized conformations and the 97 non-zero lowest-frequency modes were analyzed.
The degree of collectivity of the JMR motions in a given mode l was calculated as [109], [110]:(6)where n = 596 is the number of atoms belonging to JMR. The quantity is defined as:(7)where , and are the components of mode l that correspond to the three degrees of freedom of atom i and such that . The degree of collectivity is comprised between 0 and 1. A value of indicates that only one atom is involved in the motion while a value close to 1 indicates high collectivity.
The resultant displacement, i.e. the norm of the resultant displacement vector, of any fragment of the protein was calculated as:(8)over the ensemble M of the m atoms belonging to the fragment – 172 for JM-Switch and 181 for JM-Zipper.
Displacements along selected normal modes, in both positive and negative directions, were performed with the VMOD facility in CHARMM 35b3. The range of displacement was set from −4 Å to 4 Å with steps of 0.1 Å with respect to the initial conformations extracted from WT547-935 and MU547-935 MD trajectories. Intermediate conformations were obtained using a restraint potential added to the internal standard potential [111], [112].
Pockets were detected at the surface of the crystallographic structures 1T45 and 1PKG and the 4-Å displaced conformations using fpocket [113], with the default parameters. This geometry-based algorithm was found to perform best on accurate binding site prediction in a recent large-scale comparison study [114]. Pockets were selected by visual inspection with PyMOL 1.2.
|
10.1371/journal.pntd.0000653 | A Systems-Based Analysis of Plasmodium vivax Lifecycle Transcription from Human to Mosquito | Up to 40% of the world's population is at risk for Plasmodium vivax malaria, a disease that imposes a major public health and economic burden on endemic countries. Because P. vivax produces latent liver forms, eradication of P. vivax malaria is more challenging than it is for P. falciparum. Genetic analysis of P. vivax is exceptionally difficult due to limitations of in vitro culture. To overcome the barriers to traditional molecular biology in P. vivax, we examined parasite transcriptional changes in samples from infected patients and mosquitoes in order to characterize gene function, define regulatory sequences and reveal new potential vaccine candidate genes.
We observed dramatic changes in transcript levels for various genes at different lifecycle stages, indicating that development is partially regulated through modulation of mRNA levels. Our data show that genes involved in common biological processes or molecular machinery are co-expressed. We identified DNA sequence motifs upstream of co-expressed genes that are conserved across Plasmodium species that are likely binding sites of proteins that regulate stage-specific transcription. Despite their capacity to form hypnozoites we found that P. vivax sporozoites show stage-specific expression of the same genes needed for hepatocyte invasion and liver stage development in other Plasmodium species. We show that many of the predicted exported proteins and members of multigene families show highly coordinated transcription as well.
We conclude that high-quality gene expression data can be readily obtained directly from patient samples and that many of the same uncharacterized genes that are upregulated in different P. vivax lifecycle stages are also upregulated in similar stages in other Plasmodium species. We also provide numerous examples of how systems biology is a powerful method for determining the likely function of genes in pathogens that are neglected due to experimental intractability.
| Most of the 250 million malaria cases outside of Africa are caused by the parasite Plasmodium vivax. Although drugs can be used to treat P. vivax malaria, drug resistance is spreading and there is no available vaccine. Because this species cannot be readily grown in the laboratory there are added challenges to understanding the function of the many hypothetical genes in the genome. We isolated transcriptional messages from parasites growing in human blood and in mosquitoes, labeled the messages and measured how their levels for different parasite growth conditions. The data for 5,419 parasite genes shows extensive changes as the parasite moves between human and mosquito and reveals highly expressed genes whose proteins might represent new therapeutic targets for experimental vaccines. We discover sets of genes that are likely to play a role in the earliest stages of hepatocyte infection. We find intriguing differences in the expression patterns of different blood stage parasites that may be related to host-response status.
| Renewed efforts to combat malaria have focused on the goal of total eradication. While most attention is on the more deadly Plasmodium falciparum malaria, Plasmodium vivax is the most geographically widespread human malaria parasite causing an estimated 80–250 million cases of vivax malaria each year [1]. P. vivax malaria has traditionally been found outside of tropical areas and was endemic throughout North America and Europe until the introduction of DDT. Despite the large burden of disease caused by P. vivax, it is overlooked and left in the shadow of the enormous public health burden caused by P. falciparum in sub-Saharan Africa. The widely held misperception of P. vivax as being relatively infrequent, benign, and easily treated explains it's nearly complete neglected across the range of biological and clinical research. In fact, P. vivax malaria seriously threatens more people than has been historically appreciated. Recent reports [2]–[4] provide abundant evidence that challenge the paradigm that P. vivax infection causes benign disease: P. vivax malaria may result in severe symptoms similar to P. falciparum. Selective pressure for resistance to malaria has had a great influence on the human genome and most Africans are immune to P. vivax malaria due to mutations in the Duffy receptor that the parasites use to invade the red cell.
A fundamental difference between P. vivax and P. falciparum is the formation of dormant liver stage parasites called hypnozoites that are resistant to schizonticidal drugs that kill erythrocytic stage parasites. Despite schizonticidal drug therapy, the patient may experience multiple relapses months or years following the primary infection. Attempts to eradicate malaria will depend on having effective and non-toxic drugs that target P. vivax liver stage hypnozoites. Hypnozoite biology is poorly understood but is likely related to persistence of the parasite in locales where mosquito populations vary seasonally. Little is known about what triggers a relapse but some strains form different numbers of hypnozoites and have different relapse frequencies, which may be correlated with latitude[5]. There is controversy about whether hypnozoites represent a different lifecycle stage or merely represent an arrested early exo-erythrocytic phase. Almost nothing is known about metabolic activity in the hypnozoite, thwarting all efforts at rationale drug design. The mechanism of the only drug known to eradicate hypnozoites, the 8-aminoquinoline drug, primaquine, is still not completely understood.
While rodent models have furthered our understanding of Plasmodium liver stage development, they do not form hypnozoites. The only other available model for studying hypnozoite biology is the closely related P. cynomolgi species that must be studied in the rhesus macaque. Studies of P. vivax biology still depend on obtaining erythrocyte stage parasites or sporozoites from infected humans or non-human primates or their mosquito vectors, respectively. Drug sensitivity testing remains difficult. In an earlier era compounds were tested for activity against hypnozoites in prisoners who had been given malaria [6]. Drug testing in vitro is limited by current difficulties in culture techniques. Only a few genetic manipulations of P. vivax have been successful [7]. These impediments have discouraged many researchers from working on P. vivax, and limited the potential for molecular genetic analysis of this elusive liver stage.
Given that forward and reverse genetic methods, which have been powerful in P. falciparum, are not readily available for investigating the genome of P. vivax, comparative genomics and virtual genetic methods offer the best opportunities to elucidate P. vivax gene function. Previous gene expression profiling of P. falciparum [8], [9] P. yoelii [10] throughout many developmental life cycle stages in the mammalian host and insect vector, and in P. vivax blood stages [11], [12] has provided fundamental insight into Plasmodium biology and illustrated how gene function can be predicted based on gene expression patterns throughout development. The power of this method is to assure that there is sufficient diversity in different stage parasites to allow for the delineation of distinct patterns of gene expression.
Here we used a systems biology approach to characterize the P. vivax transcriptome. Using a custom high-density tiling microarray, we obtained a diverse set of gene expression data from human and mosquito stages including sporozoites, gametes, zygotes and ookinetes, and in vivo asexual blood stages obtained from infected patients in the Peruvian Amazon. These data are combined with published short term in vitro culture data [11]. Using a guilt-by-association approach we create hypotheses about the function of many uncharacterized genes. Comparison to datasets of P. falciparum and P. yoelii reveals conserved and species-specific patterns. This analysis provides new insights into the metabolic state of parasites growing within humans. It shows that many of the orthologs of P. falciparum transcripts needed for exo-erythrocytic development are present in the P. vivax sporozoites suggesting that slight modifications in exo-erythrocytic development may allow hypnozoite formation.
The protocol used to collect human blood samples for this work was approved by the Human Subjects Protection Program of The Scripps Research Institute, and the University of California, San Diego and by the Ethical Committees of Universidad Peruana Cayetano Heredia and Asociacion Benefica PRISMA, Iquitos, Peru. Written informed consent was obtained from each subject or the parent, in the case of minors. The consent form states in English and Spanish that samples may be used for any scientific purpose involving this or any other project, now or in the future and that the samples may be shared with other researchers.
Patients who presented to local health clinics in the Peruvian Amazon region of Iquitos with typical signs and symptoms of malaria were evaluated by light microscopy examination of Giemsa-stained blood smears to have P. vivax parasitemia. After informed consent, 20 ml of blood was drawn into heparinized vaccutainer tubes, placed in a portable incubator maintained between 37 and 39°C, and transported to laboratory facilities within 30 minutes. Blood was centrifuged at 900×g for 5 minutes and plasma was removed. Red cells were resuspended in 2 volumes of suspended animation (SA) solution (10 mM Tris, 170 mM NaCl, and 10 mM glucose pH 7.4) and passed through a cellulose column (CF-11 powder, Whatman Ltd) placed in a 37°C incubator, to remove white blood cells. Asexual parasites were enriched by gradient purification. Following filtration, cells were washed twice in SA solution at 900 g for 5 minutes and then resuspended in a 1∶3 v/v ratio in SA solution. For the production of gametes/zygotes and ookinetes in vitro, The red cells containing parasites were resuspended in exflagellation medium (10 mM Tris, 170 mM NaCl, 10 mM glucose, 25 mM NaCO3, 10% AB+ human serum, 50 mM xanthurenic acid) to induce parasite emergence from gametocytes, exflagellation and fertilization. The cell suspension was then layered over a discontinuous gradient of 38%, 42% and 50% Percoll (Sigma, USA) in RPMI 1640 medium (Invitrogen USA) and centrifuged at 600 g for 15 minutes. Female gametocytes at the 38–42% interface were removed from the gradient. Red cells containing asexual parasites which sediment below the gradient were collected and washed separately in SA solution at 900 g for 10 minutes. Zygotes and ookinetes and untransformed gametes at the 11%–16% interface were collected, washed in PBS and resuspended in Trizol and stored at −70°C. Microscopic analysis of cellular morphology confirmed the enrichment of sexual stages. Small aliquots of purified cells were stained and 60 fields were examined to determine the relative percentage of asexual and gametocyte cells (Table 1). Red cells from asexual enrichments were subjected to lysis by a 0.1% Saponin solution in PBS for 15 minutes at 4°C. Parasites were pelleted at 1200 g for 5 minutes and washed three times in PBS before resuspension in Trizol and storage at −70°C for shipment to the Scripps Research Institute.
We designed a Affymetrix custom P. vivax whole-genome tiling microarray with 4.2 million 25-bp probes covering both strands at six base pair spacing, based on the genome assembly of 2809 contigs (PlasmoDB Ver. 5.4). This microarray includes 1.7 and 2.3 million probes uniquely mapped to coding regions and non-coding regions, respectively. Altogether 5419 P. vivax genes are represented on the array, 4676 of which have P. falciparum orthologs. This array will be made available for purchase from Affymetrix, part number PvivaxLi520507.
RNA was purified by Chloroform extraction and isopropanol precipitation and purified by RNeasy Mini Kit (Qiagen) following the manufacturer's instructions. RNA was quantified by spectrophotometer and qualitatively analyzed by BioRad Experion RNA StdSens Analysis kit (BioRad). 1 ug of total RNA from asexual blood stage parasites was used to produce cDNA in the Two-cycle cDNA synthesis kit (Affymetrix) and amplified to produce labeled cRNA in the IVT Labeling kit (Affymetrix), and purified using the Genechip Sample Cleanup Module (Affymetrix), according to manufacturer's instructions. For the sporozoite sample, 100 ng of total RNA was used to produce cDNA in the Two-cycle cDNA synthesis kit (Affymetrix) and amplified to produce labeled cRNA in the IVT T7 MEGAScript kit (Affymetrix). Following the first IVT reaction, the cRNA was split into two reactions, containing approximately 500 ng each for the second round of cDNA synthesis and amplified to produce labeled cRNA in the IVT Labeling kit (Affymetrix), and purified using the Genechip Sample Cleanup Module (Affymetrix), according to manufacturer's instructions. 20 ug of amplified cRNA was hybridized to the P. vivax tiling microarray for 14 hours. The genechips were washed on Affymetrix Wash Station using standard Affymetrix protocol FlexGE-WS450_00001 and scanned on the Affymetrix scanner. The Affymetrix CEL files microarray data are available for download from our companion website (http://carrier.gnf.org/publications/Pv). Gene expression data are also visible in PlasmoDB on the P. vivax gene info pages. For more information on gene expression interpretation from our custom whole genome tiling array please see the website.
All gene expression values were subjected to a probe-level two-way ANOVA test to determine variability of expression across all samples [13]. A total of 4,326 genes were identified as differentially expressed with pANOVA <0.05 and FC >2 (Table S1) and were subjected to further OPI clustering analysis. All data can be downloaded from the companion web site.
All Plasmodium gene descriptions and name aliases were downloaded from PlasmoDB (version 6.1). Function annotation data for P. falciparum, P. yoelii, P. vivax, P. berghei, P. chaubaudi and P. knowlesi were obtained from PlasmoDB, then merged with P. falciparum function annotation data from Gene Ontology (November 2009 release). We had previously collected gene co-citation data through Google Scholar and NCBI [10], this dataset is appended with co-citation data found in “gene notes” section of the PlasmoDB annotation files. We also considered previously published data such as P. falciparum protein complexes [14], P. falciparum cell cycle k-means clusters [9], and our own literature-based annotations [10]. For each gene list, we further recruited additional gene members that are homologs or orthologs among Plasmodium species according to the latest OrthoMCL database (version 3). The final annotation database contains 3,732 gene lists that contains at least one P. vivax gene, including 2,538 GO groups, 1,066 literatures, 84 custom GNF lists, 29 complexes and 15 cell cycle clusters. From these 2,002 lists contains at least two differentially expressed P. vivax genes, therefore were used to create clusters of co-regulated genes that share gene ontology annotation using the ontology-based pattern identification (OPI) algorithm described previously [15]-[17]. To weight samples correctly, replicate samples were weighted 50% each and effected counted as one sample. When our dataset were merged with Bodzech data, the overall weighting of samples in each data set was adjusted so that two data sets contributed equally in the clustering analysis.
Visualization of the OPI expression patterns in Figure 1 used the OPI query pattern data as described previously [17]. Briefly, the query pattern is the best representative expression pattern of a given cluster, and was typically derived from common pattern shared by most of the known GO members of the cluster. Each query profile is normalized by subtracting the mean and divided by the standard deviation, a standard practice used for heat map plotting. The query patterns representing the 192 gene functions were hierarchically clustered so that related biological processes are close to each other.
For each of the 192 statistically significant P. vivax OPI clusters, we identified the P. falciparum and P. yoelii orthologs of its cluster members. We then calculated the average pair-wise Pearson correlation coefficient of the P. falciparum orthologs in a previously published cell cycle dataset of 2,235 genes. The correlation coefficients are almost all positive with a median at 0.45. We also calculated using a similar approach the average correlation coefficients of the orthologs in a combined P. yoelii and P. falciparum life cycle dataset. All results are available in Table S2.
For each piece of P. vivax gene function prediction, we checked if its P. falciparum or P. yoelii orthologs were also predicted to be in the same function category based on the previous P. yoelii and P. falciparum life cycle dataset. Both previously published OPI clusters [10] and new OPI clusters based on the latest gene annotation database were used in this cross validation analyses. Notice here an orthlog was considered to be associated with a GO group as long as it appeared in the group or one of its descendant groups. Support was found for 89 of the 192 OPI clusters.
We first projected the previously published P. falciparum yeast two-hybrid protein interaction database into P. vivax by introducing interactions to all the corresponding P. vivax ortholog pairs using OrthoMCL version 3. We then constructed a protein network for each OPI cluster and recorded the size of the network. To evaluate the statistical significance of the resultant network, we replaced the cluster members through random sampling and repeated the network construction process 1000 times. The p-value of the network was estimated based on the probability of an equal or more complex network to occur by chance. The whole process was carried out by either only considering direct protein-protein interactions or including indirect protein-protein interactions and whichever led to a better p-value was selected for presentation. An indirect protein-protein interaction refers to two proteins interact via another protein. A total of 63 networks were determined to have p-values <0.05.
It was previously shown that genes co-cited in a literature tend to be more correlated in their expression profile compared to gene members of the same GO group, probably due to the careful review process involved in the publication [10]. Therefore, it could be worthwhile introducing virtual protein-protein interactions among genes co-cited. With these virtual interactions appended to the yeast two-hybrid dataset, we repeated the above network evaluation processes for non-literature derived OPI clusters and identified 11 additional networks.
Protein networks were visualized using Cytoscape (Version 6.1), where nodes can be color coded according to the level of confidence in their function predictions, and edges can be color coded to reflect the data sources.
Statistical enrichment of motifs in upstream regions was determined using GeneSpring 7.3 software, using the “search for regulatory sequences” with parameters of 0 to 1000 bases, motif sizes of 5 to 9 bps and allowing for up to two N's in the central region. Only cutoffs with a corrected p-value of less than 0.05 are reported except for very small clusters.
P. vivax sporozoites were obtained from Sanaria, Inc. from mosquitoes fed on P. vivax infected chimpanzees infected with India VII strain P. vivax [18]. Sporozoites were dissected from mosquito salivary glands and purified. Approximately 150,000 sporozoite cells were used for RNA preparation. Similarly purified sporozoites of P. falciparum strain 3D7 isolated from mosquitoes infected with in vitro cultured parasites were also obtained from Sanaria, Inc.
P. falciparum salivary gland sporozoites were obtained from Sanaria, Inc. P. falciparum sporozoite RNA was isolated and amplified using Affymetrix kits as described for P. vivax sporozoite samples. P. falciparum 3D7 strain RNA from in vitro synchronized trophozoite stage parasites was isolated and amplified as described for P. vivax samples. Amplified cRNA was hybridized to the Pftiling array described previously [19].
To validate the expression comparison of genes that are differentially expressed in sporozoites of P. vivax and P. falciparum, we performed quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) on 22 genes with orthologs in both species, and two P. vivax specific genes. Primers used are listed below. All primer sets were optimized using genomic DNA from 3D7 strain P. falciparum and Salvador I strain P. vivax at three dilutions of 10 ng/ul, 1 ng/ul and 0.1 ng/ul to ensure that the amplification threshold values accurately reflected the difference in DNA template concentration. Primer sets for the two different species produced similar threshold Ct values (what does Ct stands for) (+/− 1.5 Ct) for all primer sets for both species. An additional aliquot of 150,000 sporozoites for both P. falciparum and P. vivax from Sanaria, Inc, were used to isolate total RNA using Trizol as described previously. This total RNA sample was split into equally into three reactions to produce single stranded cDNA using reverse transcriptase and a T7-Oligo dT primer from the cDNA synthesis kit (Affymetrix) according to manufacturer's instructions. The single stranded cDNA was used as template for QRT-PCR reactions using the primer sets presented. To account for variability in the input cDNA between different cDNA reactions from the same species, we normalized the threshold Ct values by the average difference between reactions across all genes. We cannot know the identity of any one gene, which is expressed at the exact same level in both species that can be used to control for variation. However, when comparing P. vivax and P. falciparum threshold Ct values, we found that the highest expressed gene (CSP) and lowest expressed gene (Pv117045 zinc finger) in both species showed very similar threshold Ct values within 1.5 Ct cycles, which was the within the error observed by DNA optimization. Our gene expression data from multiples species, including P. yoelii, as well as proteomic data and conventional experiments show CSP is often one of the most abundant proteins in plasmodium sporozoites. It seems safe to use this in normalization. qRT-PCR reactions were prepared using SYBR GREEN PCR Master Mix (Applied Biosystems) according to manufacturer's instructions, and were run on Applied Biosystems TaqMan machine using SDS 2.2.1 software. Threshold Ct values were determined using default settings and automatic threshold determination. All amplification results were manually inspected to ensure that threshold levels were determined within the logarithmic amplification phase of the reaction for accurate determination of Ct values. Fold difference between P. falciparum and P. vivax qRT-PCR determined expression values are equal to 2 raised to the power of the difference in Ct values between the two species. qRT-PCR results and primers used are listed in Supplemental Table S2.
The list of P. vivax sporozoite-specific genes used to seed the OPI cluster includes: S13, MAC/Perforin (PVX_000810, PFD0430c); SIAP-1 (PVX_000815, PFD0425w); pf52 protein (PVX_001015, PVX_001020, PFD0215c); ECP1, cysteine protease (PVX_003790, PFB0325c); asparagine-rich antigen Pfa35-2 (PVX_081485, PFA0280w); S24, hypothetical protein (PVX_081555, PFA0205w); TRSP (PVX_081560, PFA0200w); TRAP (PVX_082735, PF13_0201); S14, hypothetical protein (PVX_084410, PFL0370w); S25, kinesin-related protein (PVX_084580, PFL0545w); MAEBL (PVX_092975, PF11_0486); S1, hypothetical protein (PVX_094625, PF10_0083); kinesin-related protein (PVX_094710, PFL0545w); conserved hypothetical protein (PVX_097795, PFE0230w); hypothetical protein (PVX_118360, PF14_0404); circumsporozoite (CS) protein (PVX_119355, PFC0210c); early transcribed membrane protein 13, ETRAMP13 (PVX_121950, PF13_0012); S23, conserved hypothetical protein (PVX_123155, PF08_0088); S4, conserved hypothetical protein (PVX_123510, PFL0800c); conserved hypothetical protein (PVX_123750, PFL1075w).
To infer gene expression of un-annotated genes in P. vivax, we performed a BLAST search of all P. falciparum and P. knowlesi annotated genes against the P. vivax genome to identify all putative orthologous genes that may not be annotated in P. vivax. The BLAST similarity coordinates were used to define the coding region in P. vivax. We have not validated the coding sequence for proper gene translation nor have we defined intron-exon boundaries for these genes. These gene boundary definitions were used to pick probes to evaluate the level of gene expression from these regions in the same way as all other annotated P. vivax genes. These genes were originally named using the GeneID numbers of their P. falciparum and P. knowlesi orthologs. We have included these gene expression values for these putative genes in Table S1. We also performed an analysis of all P. vivax RNA microarray hybridization data to identify highly transcribed regions of 50 bp that do not overlap with existing gene annotations. We found a few of these regions, but they appeared to correspond to additional exons, intronic regions, or 5′ or 3′ untranslated regions of existing genes. One additional gene identified by this method is the Pv_PF11_0140 gene. We provide putative P. vivax Gene ID numbers for these genes based on their position relative to existing flanking genes. We provide a list of these new putative gene coordinates in Supplemental Table S3.
Blood samples were collected from eight different P. vivax-infected patients with uncomplicated malaria in Iquitos, Peru. Human leukocytes were removed by filtration and P. vivax gametocytes were separated from asexual stage parasites by gradient centrifugation. There was only a small proportion of gametocytes (0–15%) in the final sample (Table 1), so we hereafter refer to these samples as asexual profiles. By microscopy the asexual cells in the patient blood samples appeared to be rings and early trophozoite stages, with no late trophozoite or schizont stages, likely as the result of natural synchronization in the Peruvian patients. For one sample, we put the isolated gametocyte stages into in vitro culture and induced sexual stage development to obtain a mixed gamete/zygote stage sample and a mostly pure ookinete stage sample from the patient isolate (Table 1). Additionally, we isolated salivary gland sporozoites from dissected mosquitoes fed on an experimentally infected chimpanzee.
We assayed gene expression using a custom P. vivax whole genome tiling microarray with 4.2 million 25-base pair probes covering both strands at six base pair spacing. A semi-quantitative estimate of transcript abundance for each gene could be obtained with this microarray design because the 5,419 P. vivax genes were probed by hundreds of independent oligonucleotides. While this array can be used to find noncoding RNAs, including antisense RNAs, the labeled cRNA for hybridization was prepared using a polyA reverse transcriptase priming method that would not give accurate descriptions of noncoding RNAs and thus we limited our analysis to predicted coding regions. In order to calculate an approximate gene expression level, E, we used the MOID algorithm, which ranks the probe intensity values for the 20 probes at the 3′ end of the transcript having similar GC values. It then assigns the expression value, E, to difference between the background (computed from thousands of probes with a similar GC content not predicted to be in the P. vivax or human genome) and the probe at the 70th percentile (Table S1). Because the varying GC content of different P. vivax genes could give rise to spurious apparent expression levels, extensive optimization of probe selection was undertaken to ensure robust measurements of gene expression across all samples. We show that MOID expression values are not changed by selection of the independent 13th or 14th of 20 ranked probes (Methods S1 Figure IIIA). Additional optimization for different GC contents (Methods S1) shows that we can select probes from a whole-genome tiling array to accurately detect gene expression. To verify reproducibility of our analysis, we analyzed three samples, two asexual and one sporozoite in two technical replicates each, and obtained Pearson correlation coefficients of 0.986 to 0.996, indicating excellent reproducibility, whereas lower correlation is observed between samples from divergent asexual groups (Methods S1 Figure IIIB-D). Results were also confirmed by qRT-PCR, and compared to expressed sequence tags (ESTs), as described in Methods S1.
We first sought to address our hypothesis that genes involved in similar processes would be co-expressed. We identified 4,326 differentially expressed genes using the cutoff of ANOVA p-value <0.05 and fold change (FC) >2 using the individual probe intensity values for each gene using just our data. Differentially expressed genes were clustered using ontology-based pattern identification (OPI), an algorithm previously used with P. falciparum and P. yoelii expression datasets [10] (Figure 1A). This algorithm begins with 2,002 lists of genes sharing a common gene ontology (GO) annotation, literature co-citation, or other annotated parasite-specific process, e.g., there are 38 genes known to be involved in DNA replication. For each group a representative expression profile vector is computed using E values from all conditions for all genes in the group. Then all of the 4,326 differentially expressed genes are ranked by the correlation coefficient calculated between the gene's expression vector and the representative expression profile vector. The algorithm then uses a correlation coefficient optimization routine and creates expression clusters that contain the largest number of genes with common annotation and high correlation. In the case of the GO process “DNA replication” a group of 60 genes is created which contains 11 of 38 annotated DNA replication genes. The probability of this distribution occurring by chance is less than 10−13. Similarly, there is less than 10−8 probability of identifying 5 of 15 genes with an annotation of glycolysis in a cluster of 21 genes. In addition to using gene ontologies we also used other groupings of genes. In a previous analysis of P. falciparum lifecycle stages [9] we had identified 106 genes upregulated in sporozoites, which corresponds to 66 P. vivax orthologs in our dataset (GO:CCYCL01). Of these 35 were found in a group of 366 genes with a probability of enrichment by chance of 10−20. Overall these data showed that patterns of gene expression were nonrandom and that genes with similar functions showed much greater cohesion than would be expected by chance. Altogether, 121 clusters of highly correlated genes with shared annotation were identified (see companion web site: http://carrier.gnf.org/publications/Pv).
Previous analysis of P. vivax blood stage gene expression for parasites taken into synchronous short term culture has been performed [11] and we compared our results to these. Because this study involved two color microarrays and did not produce expression levels, direct comparisons between expression levels are not possible. However, we could still apply the same OPI clustering algorithm to the Bozdech data. Clustering of the Bozdech data (see companion web site) gave less information about sporozoites and sexual stages but revealed highly significant functional enrichments, especially within the area of protein biosynthesis and ribosome function, which is expected because of the higher sampling throughout the erythrocytic cycle. For example, 48 or the 58 annotated genes with a predicted role in cytosolic ribosome (GO:0022626) were found in a cluster of 126 genes, with a probability of enrichment by chances of 10−63. The data showed that in many cases the same genes that cluster with “cytosolic ribosome” in the Bozdech data also cluster with “small ribosomal subunit” in our data. The gene, PVX_084645, co-clusters with ribosomal genes in both cases and is listed as hypothetical but its P. falciparum ortholog, PF14_0360, is listed as an eukaryotic translation initiation factor 2A protein and thus its association with ribosomes is not surprising. PVX_101135, a hypothetical, clusters with ribosomal proteins in both cases. BLASTP (p = 1.3×10−35) shows a strong match to the yeast protein YOR091W, a protein of unknown function that associates with ribosomes that interacts with GTPase Rbg1p [20].
We also co-clustered Bozdech data with our data to generate more accurate predictions of gene function creating a set of 192 different clusters containing between 2 and 493 genes and p-values between 10−3 and 10−52 (Figure 1, Table S2). Many of our functional predictions can be cross-validated with previously published data sets. In particular we checked if the same function prediction can be made based on combined P. falciparum and P. yoelii data set, using either previously published OPI clusters [10] or an updated cluster set using the latest gene annotations. For each P. vivax cluster we ran permutation tests to see their P. falciparum orthologs form denser protein networks than what would be expected by chance using both published two hybrid data [21] and literature co-citation data [10]. In total, 75 of the 192 OPI clusters led to protein networks with a p-value less than 0.05 based on 1000 permutation simulations. For example, PVX_123920, a putative ubiquitin-activating enzyme e1 clusters with genes involved in the proteosome regulatory particle in both P. vivax and in P. falciparum and has two-hybrid support as well [21]. While there are numerous examples that can be derived from well-studied processes, the greatest value of this data is in supporting predictions for genes that may not be found in other model organisms. PVX_092415 and PVX_113830 cluster with genes involved in merozoite development in P. falciparum and in P. vivax (GO:GNF0218) and furthermore, are supported by two-hybrid interaction studies from P. falciparum (Figure 2). Likewise, PVX_000945 shows a similar pattern. The Toxoplasma gondii homolog of this protein has been isolated from rhoptries [22] as has, the Toxoplasma ortholog of PVX_113800, which also clusters with genes involved in merozoite development in P. falciparum. There are numerous examples from pre-erythrocytic stages as well. Of course, some caution must be used in evaluating the data because genes involved in two different processes may be co-expressed (e.g. DNA replication is occurring during gamete production) and yet be involved in relatively different processes. Nevertheless this clustering exercise gives functional predictions for the many uncharacterized genes found in an OPI cluster.
Having established the overall quality of the data we examined the groups of genes that were differentially expressed in any one stage. While our initial expectation was that patient-derived blood samples would be relatively homogenous, unexpectedly, we observed large differences in the expression profiles of asexual samples with the most pronounced differences observed in genes involved in glycolysis. While we found little or no differential expression of the first three enzymes of the pathway: hexokinase (PVX_114315), glucose-6-phosphate isomerase (PVX_084735) and 6-phophofructokinase (PVX_099200), the eight remaining glycolytic enzymes: fructose 1,6-bisphosphate aldolase (PVX_118255), triosephosphate isomerase (PVX_118495), glyceraldehyde-3-phosphate dehydrogenase (PVX_117321), phosphoglycerate kinase (PVX_099535), phosphoglycerate mutase (PVX_091640), enolase (PVX_095015), lactate dehydrogenase (PVX_116630) and pyruvate kinase (PVX_114445) all showed very strong differential expression between samples with some genes showing up to a 100-fold higher expression in some asexual samples relative to others (Figure 3). These last eight enzymes are among the top 1% when ranked by transcript abundance in vitro trophozoites of P. falciparum and are among the top 5% of P. vivax genes in some samples but not others (Figure 3). These differences are unlikely to occur by chance (p = 1.9×10−7 comparing P. vivax group 1 and group 2 defined in Figure 3 with a paired t-test). While the high glycolysis samples were the minority (two of eight), they were more similar to that described in Cui et al. Here 22,236 ESTs from P. vivax-infected Thai blood samples [12] were sequenced. Our data showed good Spearman rank correlation with the Thai EST numbers and showed that those samples with high glycolysis gene E values were most similar (r = 0.53) (Table 1 and Methods S1) to the Thai strain. While it may be that these expression differences are due to contamination with gametocytes [23] the presence of 0–10% contaminating gametocytes cannot mathematically explain why lactate dehydrogenase is present at ∼3,300 units in two asexual samples and as low as 75 (almost indistinguishable from background) in others [24]. Furthermore, genes typically associated with gametocytogenesis (e.g. Pvs25, PVX_111175) are higher in the high glycolysis samples (CM12 and CM13). An alternative is that although morphologies looked similar, a substantially different proportion of early and late cell cycle stages were contained in high and low glycolysis samples. Genes typically associated with schizogony and invasion such as myosin motor proteins and reticulocyte binding proteins, transcribed later in the Bozdech erythrocytic cycle data, were expressed at higher levels in the low glycolysis samples (Figure 1). Despite this, glycolysis transcripts generally do not show 50–100 fold changes in expression levels throughout the in vitro erythrocytic cycle in P. falciparum cultured in vitro [8], [9] nor are such large fold changes observed with P. vivax cultured in vitro [11]. In addition, such a hypothesis would require the patient samples to have been tightly synchronized. While more samples will be need to be examined, the data raise an intriguing possibility that there may be differential regulation of metabolism in a subset of patient-derived samples as observed in P. falciparum patient–derived samples (Figure 3) [24].
The molecular determinants regulating hypnozoite formation and relapse are unknown. One hypothesis is that the hypnozoite is an early stage exo-erythrocytic form (EEF) that is arrested in development. If this hypothesis is correct, we would expect the transcriptional profile of P. vivax sporozoites to be similar to species that do not form hypnozoites. Thus, we compared our P. vivax sporozoite expression profile to a P. falciparum sporozoite sample and three P. yoelii sporozoite samples analyzed previously [25], [26]. Sporozoite transcriptome comparisons showed that P. vivax sporozoites are generally similar to P. yoelii and P. falciparum sporozoites with positively correlated expression of highly expressed genes (r = 0.5). Despite some species-specific differences, we observed similar expression for many P. vivax genes whose P. falciparum and P. yoelii orthologs were previously shown to be upregulated in sporozoites, including some of the most highly expressed genes in the P. vivax sample (Table 2): In addition to genes known to be upregulated in sporozoites such as the initiation factor UIS1, and the serine threonine phosphatase UIS2, the list contains a number of known candidates for pre-erythrocytic subunit vaccines including the apical membrane antigen 1 (AMA1) gene and the circumsporozoite protein (CSP). In addition there are genes whose disruption has led to genetically attenuated sporozoites that may eventually be used in whole organism vaccines, including the ortholog of P. falciparum etramp10.3 or UIS4 in P. yoelii [27], [28], UIS3 [29] and P52 [30]. From the set of genes upregulated in multiple species (Table S4) we identified a set of sporozoite conserved orthologous transcripts (SCOT) that were upregulated in multiple species but not yet annotated that may provide a list of possible candidates antigens for P. vivax pre-erythrocytic vaccines, or which when disrupted may yield genetically attenuated sporozoites.
There are also, of course, genes which are not found in P. falciparum that are upregulated in P. vivax sporozoites and may play some unique role in the biology of P. vivax or other hypnozoite-forming parasites. Examples from the sporozoite specific combined OPI cluster (GO:GNF0006) include up to 63 genes without P. falciparum orthologs. There are also some genes previously shown to be upregulated in P. falciparum sporozoites, but whose orthologs were downregulated or barely increased in P. vivax sporozoites or vice versa (Table S4). One ApiAP2 gene (PVX_090110) and two zinc finger proteins (PVX_099045, PVX_099045, PVX_081725), which often function as transcription factors, were downregulated as were several RNA binding proteins (PVX_098995, PVX_098995, PVX_100715) and several kinases including two serine/threonine protein kinases (PVX_081395, PVX_081395, PVX_002805), a FIKK calcium-dependent protein kinase (PVX_091755) and a MAP kinase (PVX_084965). Although some of these differences could be due to strain specific or sporozoite collection variation we were able to confirm these species-specific differences in gene expression for 19 genes using qRT-PCR (Table S5).
The process of sexual development and meiosis are poorly characterized in many species. An expression profiles from the mix of macrogametes and zygotes (Table 1) derived from in vitro cultivation of fertilized patient-derived gametocytes showed upregulation of only a few characterized genes include those with roles in DNA replication, meiosis and chromatin structure (Table 3). The majority of strongly differentially expressed genes are uncharacterized, although many had been shown to be upregulated during induction of in vitro gametocytogenesis in P. falciparum[15]. As predicted, high expression of the P. vivax ortholog of Pvs25 (PVX_111175), the ookinete surface protein, is found in this stage. Another membrane protein gene, the ortholog of the Toxoplasma gondii PhIL1 (photosensitized INA-labeled protein 1) (PVX_081335), was among the most highly upregulated (>20-fold) in P. vivax gametes/zygotes. This protein associates with the cytoskeleton and is localized to the apical end of the plasma membrane [31] where it may function as an ookinete-specific surface protein for midgut invasion. The transcription factor high mobility group protein, HMGB2 (PVX_089520) is found among the top 20 genes expressed in this sample (See Table S1). It was shown to be a critical regulator of oocyst development and appears to activate genes that are most highly transcribed in gametocytes, but then stored and translated in ookinetes [32]. Interestingly, some of these uncharacterized proteins are also upregulated during meiosis in humans, such as the human orthologs of PVX_117890 [33]. An exceptionally large proportion of the genes upregulated in zygotes (Table 3) did not fall into clusters with enrichments of known genes, highlighting the problem of using “guilt by association” when the existing knowledge base is sparse.
Gene ontology-based OPI clustering revealed few annotated gene functions to be upregulated in ookinetes. This finding likely reflects the difficulty of obtaining sufficient quantities of this developmental stage for in vitro study. We found high expression of genes involved in chromatin (e.g. histones) and translation, potentially reflecting the fact that once the parasite reaches the midgut and forms an oocyst, thousands of rounds of DNA replication will likely commence. Genes specifically upregulated in oocysts were mostly uncharacterized. Ookinete surface protein genes such as the P. vivax ortholog of the CSP and TRAP-related protein (PVX_095475), previously shown to be essential for ookinete invasion [34], [35], the transmission-blocking target antigen Pfs230 (PVX_003905), a sexual stage antigen s48/45 domain containing protein (PVX_003900), the transmission blocking target antigen precursor Pfs48/45 (PVX_083235), showed modest levels of upregulation (<3-fold) in ookinetes relative to any other stage. In P. falciparum, transcripts for many of these peak in early stage gametocytogenesis [15], and thus high expression would not necessarily be expected in P. vivax ookinetes, even if protein is detected in present.
On the other hand there were a number of uncharacterized genes that showed substantial upregulation in ookinetes (>3-fold). These may encode the proteins that are needed for early oocyst formation. Some examples (see Table S1 for details) include a possible calcium dependent kinase, (PVX_083525), and 5-aminolevulinic acid synthase (PVX_101195), a key enzyme in the porphyrin synthesis pathway that leads to heme synthesis. Proteases and other degradative enzymes may be needed to exit the blood meal, penetrate the peritrophic matrix or form the oocyst on the mosquito midgut. A number of proteases were unregulated in ookinetes, including Ulp1 (ubiquitin-like protein-specific protease, PVX_100650), which is specifically required for cell cycle progression in other species.
While in many cases drug resistance in malaria parasites is conferred by single nucleotide polymorphisms [36]–[38], in some cases transcript amplification events in a target or in a pump [39] confer greater tolerance to drugs. We therefore compared levels of drug resistance gene transcripts to determine if anything interesting might be found. Most genes involved in drug resistance such as pvcrt (chloroquine resistance transporter, PVX_087980), dhps (dihydropteroate synthase, PVX_123230), gtpch (GTP cyclohydrolase, PVX_123830) or mdr1 (multidrug resistance gene 1, PVX_080100) showed similar patterns in P. vivax and P. falciparum. An exception was dihydrofolate reductase (dhfr, PVX_089950), the target of the widely used antimalarial antifolate drug, pyrimethamine. Remarkably, dhfr expression was 50-fold higher in the India VII P. vivax sporozoites (confirmed by qRT-PCR, see Table S6) and 10-fold to 30-fold higher in P. vivax asexual stages compared to P. falciparum, where it was near background in all stages. Interestingly, while P. falciparum is sensitive to pyrimethamine, P. vivax is considered intrinsically resistant [40], although in vitro drug sensitivity data is not available for most P. vivax strains. Both dhfr mutations [41], [42] and amplifications [43] have been shown to confer resistance in P. falciparum. Thus, the higher dhfr expression in P. vivax relative to P. falciparum may confer higher tolerance to pyrimethamine.
While much of our data shows that patterns of gene expression are conserved across Plasmodium species the data also reveals possible roles for information about the many uncharacterized genes which are specific to P. vivax including members of multigene families. The largest paralogous gene family of P. vivax is the diverse superfamily of variant surface protein genes (vir) found in subtelomeric regions [44] and may be related to the rif genes of P. falciparum and yir genes of P. yoelii [45]. The highly variable surface protein var gene family in P. falciparum has no orthologs in P. vivax. Expression levels for most of the 274 vir genes on the array were lower than for other genes. The average for vir genes was 210 units (approximately the 35th percentile) using the highest value in any one stage (MaxExp in Supplemental Table 1), versus 572 units (approximately the 80th percentile) on average for other genes. In addition 39 of the 109 genes in the genome, which were not detected as differentially expressed in any sample, were vir genes. Only 27 of the 274 probed vir genes showed strong differential regulation (>5-fold change, pANOVA<0.001) versus 1,044 of the remaining 5,420 genes. Interestingly, four of the highest expressed vir genes (PVX_086860, PVX_096970, PVX_096980, PVX_096985) were among the highest expressed genes in the two high glycolysis asexual samples, indicating possible alternate mechanisms of expression control. Three of these highly expressed vir genes are adjacent to one another suggesting coordinated regulation of the entire region. Cui et al. also found these three vir genes and another gene in the same region (PVX_096975) had the highest numbers of ESTs in their patient samples [12].
Malaria parasites are known to be able to decorate the surface of infected erythrocytes with proteins that play roles in sequestration and immune evasion. Many of the exported proteins contain sequences, called PEXEL or VPS motifs that direct them out of the parasitophorous vacuole and to the surface [46], [47] and many are members of multigene families, presumably because enhanced levels of recombination between members or epigenetic transcriptional switching between members allows the parasite to evade the host immune responses to these exposed proteins. Many of the exported genes in P. falciparum are transcribed at specific, mid-trophozoite stages of the parasite cell cycle [47]. The number of predicted exported proteins in P. vivax is not as large as in P. falciparum most likely due to problems in recognizing PEXEL/VPS motifs in this species. For example, only 160 of the 346 vir proteins contain predicted export motifs [48]. In addition to vir genes the 20 members of the P. vivax PHIST (Plasmodia helical interspersed subtelomeric) exported gene family (Pv-fam-b)[49] on the array are expected to be exported as well as some members of the Pv-fam-h and Pv-fam-e families.
Remarkably many of the members of the multigene families as well as most predicted exported proteins that show strong differential transcription show peak expression in just one of our blood stage samples, CM115 (Figure 4). Five of the eight exported PHIST genes, six of the ten of Pv-fam-e family of RAD GTPases (some exported), nine of the 11, Pv-fam-d, and 9 or the 11 Pv-fam-a genes that show strong (>5X) differential expression again peak mostly in CM115. While vir genes show lower levels of expression overall, many of those that are differentially expressed also peak in CM115. Notably, 17 of the 38 expressed above 300 units show peak transcript levels in sample CM115. Many of the genes showing dramatic upregulation (up to 100 fold) in sample CM115 are abundantly transcribed. The Pv-fam-d family with 16 genes of unknown function has 2 genes in the top 1% of all genes ranked by maximum expression as does the Pv-fam-a family of tryptophan rich antigens (PvTRAg, average max expression in any one sample = 1,969 units). One P. vivax tryptophan-rich antigen, PvTRAg (PVX_090265), has shown a very high seropositivity rate for the presence of antibodies in P. vivax malaria patients [50]. Another highly immunogenic antigen in this multigene family is PvATRAg74 (PVX_101510), recombinant versions of which showed erythrocyte binding activity and were recognized by all P. vivax patient sera tested [51]. This gene also ranks in the top 1% of transcripts in CM115. Many of the exported genes in P. falciparum are transcribed at specific, mid-trophozoite stages of the parasite cell cycle [47] and it seems likely that most of the parasites in asexual sample CM115 are at this export permissive stage. Thus many of the genes that are specifically upregulated in CM115 may play a role in immune evasion. While genes involved in DNA replication are also upregulated in CM115, these same genes are also upregulated in zygotes, while those encoding exported proteins may not be. These data nicely illustrate how the random collections of gene expression data that may be obtained from a neglected parasite can be used to create high quality predictions if enough random and yet diverse data is available. The data also illustrate that an advantage of using P. vivax patient samples is that a high level of synchrony may exist, that is confirmed by the Bozdech data (Figure 4).
Of course members of multigene families that share sequence similarity may have different cellular roles depending on their expression in the parasite lifecycle. An example is the group of cysteine protease genes referred to as serine repeat antigens (SERAs). In P. falciparum, PFB0325c, one of the seven SERA genes is expressed in sporozoites, while the others are expressed in blood stages. Disruption of this SERA ortholog, called ECP1 in P. berghei, results in parasites that are unable to migrate out of the oocyst[52]. Although P. vivax contains 13 SERA cysteine protease paralogs, only the ECP1 ortholog (PVX_003790) was dramatically upregulated in sporozoites (105X). Two others showed substantial upregulation in sample CM12. None of the SERAs show maximal expression in sample CM115, nor do any members of the Pv-fam-c, homologous to the SURFIN gene family in P. falciparum. These are not appreciably expressed in any of our samples and may be functioning in other life stages.
In many organisms, including Plasmodium species, genes that are co-expressed share common sequence motifs in the regions upstream of their translational start sites [16]. To determine whether this would be true in P. vivax we looked for enriched motifs in various OPI groupings created with our data or the combined larger dataset. Here all possible 6–9 mers are counted within a group of co-expressed genes and these numbers are compared to the number found in all upstream regions. A corrected p-value that accounts for the multiple testing hypothesis is then computed. In general we find that transcriptional control is conserved and that many of the same motifs, which appear to control gene expression in P. falciparum also appear to control gene expression in P. vivax. For example an unbiased search for overrepresented sequences upstream of the 70 sporozoite-specific genes with promoter sequences (GO:GNF0006) relative to upstream regions in the whole genome found a motif similar to one associated with P. falciparum sporozoite genes. This putative regulatory sequence, TGCATG, is found upstream of 44 of the 70 P. vivax sporozoite-specific OPI cluster genes. The corrected probability of enrichment by chance in this set relative to the rest of the upstream regions in the genome is 0.02. Additionally, 48 of 65 genes in the list of SCOT genes contained the TGCATG motif (Table S6, p = 9.16×10−7). This is similar to the PfM24.1 sporozoite-specific regulatory motif CATGCAG identified in P. falciparum [16], sharing the core CATGC sequence and identical to the sequence (Table S6) bound by the P. falciparum ApiAP2 transcription factor PF14_0633 [53]. The P. yoelii ortholog of this transcription factor (PY00247) is among the highest expressed genes in midgut sporozoites [10], supporting the hypothesis that this protein functions as a specific activator of sporozoite transcription [25].
A rich variety of other promoter motifs could be found by searching the various other OPI clusters as well. The sequence TGTAnnTACA was found enriched in the 1000 bases upstream of 383 genes from an OPI cluster containing 12 of the 19 genes mentioned in a an analysis of P. falciparum sexual development (GO:PM16005087, 92/364 genes, p = 1.69×10−26). This motif (Figure 5) is identical to the motif that is found in 65 of the 246 genes upregulated during sexual development in P. falciparum [15]. A cluster of genes enriched for ones with a role in gliding motility gave the motif TGTnnACA (86 or 155 genes, p = 0.00224). A search of a cluster with many genes predicted to be involved in merozoite development (GO:GNF0218) produced the motif GTGCA in 392 of 443 genes with a probability of enrichment by chance of 4.31×10−13. Protein binding microarrays have shown that the P. falciparum AP2 transcription factor, PFF0200c binds this motif [53]. It is also found upstream of P. falciparum genes transcribed during schizogony [16]. As with P. falciparum [54] the sequence CACAC was enriched in a cluster of genes containing an abundance of DNA replication genes (GO:0030894, 192 of 228 promoter regions, p = 4.7×10−6). Novel motifs were also found. A cluster enriched for genes identified in a male gametes (GO:PM16115694 [55]) yielded the motif CGTACA in 35 of 66 genes (p = 0.0645) and the sequence GCTATGC was found upstream of 36 of the 105 genes with promoter sequences in the cluster containing many of the structural constituents of the ribosome (GO:0003735, p = 0.007). This motif is similar to binding site for the AP2-O transcription factor (TAGCTA) that functions as a positive regulator of ookinete gene expression [56]. While these motifs may not be the same, expression of ribosomal proteins is upregulated in ookinete stages and one must keep in mind that transcription factor binding is likely to be combinatorial and a given gene may have multiple regulatory sites contained within its upstream region.
Analysis of the transcriptome of blood stages and sporozoites of P. vivax shows that the general mechanisms of growth, development, metabolism, and host-parasite interactions are shared by all Plasmodium species. Some of the expression differences that exist between species may provide insight into the molecular and genetic basis of biological differences that distinguish the species. The most significant differences include the formation of hypnozoites by P. vivax, pathogenic processes of sequestration and antigenic variation, and the wider geographical distribution of P. vivax into temperate regions. However, examination of a much larger number of different strains will probably be need to determine which differences of the differences we find are likely to be due to speciation and which are due to strain or experiment variability.
Because of the difficulties of working with P. vivax, transcriptional and proteomic studies represent one of the most effective ways to find candidate genes for vaccines or other processes. Many of the genes that cluster with proven antigens for pre-erythrocytic vaccines such as CSP may be worth investigating especially if sera drawn from individuals living in P. vivax endemic regions shows cross reactivity to their cognate protein [57]. A high-throughput analysis P. falciparum proteomic data [58] revealed one exceptionally abundant sporozoite protein named Ag2 in P. falciparum [59] or CelTos in P. berghei [60], which was more immunogenic than previously identified antigens such as CSP. In P. falciparum this is the third most abundant transcript in sporozoites. Ag2 one of the few antigens that is able to provide cross-species immunity and indeed we also find it highly expressed in our sporozoite sample. Given that many of the SCOT genes show strong expression only in sporozoite stages, their disruption may lead to genetically attenuated sporozoites that cannot develop in the liver, but which nevertheless provide immunity.
The gamete/zygote and ookinete expression data provides insight into mosquito midgut biology of a second human-infecting malaria parasite, and confirms common stage-specific gene expression shared by multiple Plasmodium species. Processes of sexual development, meiosis and DNA replication were evident in the gamete/zygote transcriptome. Some ookinete surface protein genes show highest expression in the gamete/zygote stage. Ookinetes produce stage-specific surface proteins, secreted proteases and invasion-related gliding motility proteins. While we observe upregulation of many cell-surface genes involved in invasion in our gametes/zygotes and ookinetes relative to asexual stages, this conclusion should be considered in light of the fact that our asexual stage gene expression represented ring and trophozoite stages, but no schizont and merozoite stage gene expression, thus biasing the analysis against the invasive blood stage parasites.
The function of the vir gene family is still unknown, but published data seems to indicate its function is fundamentally different from var genes. While P. falciparum var genes display mutually exclusive expression only during the mature stages [61], different vir [11], yir [62] and rif [63] genes are expressed in different intra-erythrocytic stages and gametocytes [64]. Previous studies showed that vir expression is not clonal, and multiple vir subfamilies are expressed in individual parasites from infected patients [65]. This is supported by our analysis of vir expression in vivo, with high-level expression of a small subset of vir genes in the high glycolysis samples along with other genes such as the PvTRAG genes that are likely involved in antigenic variation and immune evasion. However, examination of our data will also show that expression levels for many vir genes was very low or not detectable, indicated that they may be silenced. It is formally possibly that none of the parasites that we collected were at a stage where vir genes are being actively transcribed, potentially because of sequestration. The overall low levels could also be attributed to having mixed stage parasites in our samples, however many other cell-cycle regulated genes such as histones showed strong (>50) fold differential expression in some of our blood stage samples. Finally, vir gene genetic differences between the genome reference strain, Salvador I, and the Peruvian samples [66] could contribute to reduced expression levels, although high expression was certainly found for other members of variable gene families. Excepting these caveats, our data are consistent with a model in which some vir genes are transcriptionally-silenced.
The OPI analysis here provides functional predictions for a large numbers of genes. However, its limitation is that it uses existing knowledge and it is likely that interesting clusters of genes may be of mixed function, which can be estimated by examining the false positive and true positive rates in each cluster in Supplemental Table 2. In some cases the functional enrichment may be misleading and thus caution should be used in interpreting the labels. Many of the exported protein and blood stage antigens are found in an OPI cluster named “ribonuclease activity” reflecting the fact that this is the only well-annotated cellular process going on at this time. Finally, 501 of the differentially expressed genes were not contained in any of the OPI clusters. This may be because they are playing a role as ookinete, oocyst or hypnozoite function. Traditional hierarchical or k-means clustering is likely the best way to find out the function of these genes (see companion web site). Finally, 296 were not considered differentially expressed. Some of these may be upregulated in liver or oocyst sporozoite stages or they may be silenced.
One of the more remarkable things about the gene expression analysis is its robustness and insensitivity to sample contamination or admixture. Our group contained few samples focused on early sexual stage parasites and only a single zygote sample and yet we were able to extract groups of genes from which we could extract a sexual development transcription factor motif. Of course, the fact that we had no oocyst salivary gland sporozoite or liver stages means that this dataset is not comprehensive and more work will need to be done. While P. vivax is by no means a model organisms the gene expression data described here may be more useful for discovering motifs involved in regulating transcription as the lower AT content may be less problematic. It should also be noted that successes were not dependent on having the larger structured Bozdech dataset as many of the motifs could be extracted from the OPI clusters created with our data alone. A group of genes enriched for those with roles in sexual development (GO:PM16005087) could be extracted from our data when clustered independently (7 of the 19 genes in a group of 172, p = 10−5) and could be used in motif finding. However no groups of genes enriched for ones with roles in sexual development were found when the Bozdech data was analyzed independently (See Supplemental Table 2). However, when the datasets were combined, the quality of the cluster improved (12 of the 19 genes in a group of 246, p = 10−8) even though the Bozdech samples were not predicted to contain sexual stage parasites. Thus, more data and diversity is better.
These gene expression data for in vivo patient infections, gametes/zygotes, ookinetes and sporozoites of the P. vivax parasite provide an important foundation and reference for future studies. Numerous differences in gene expression suggest many hypotheses to be tested by researchers in the laboratory and in the field, and may be used to guide drug treatment and vaccine development for P. vivax. Further studies may provide correlations between in vivo parasite gene expression variability between patient samples and phenotypes of disease severity and drug resistance. In combination with accurate drug treatment outcomes and patient data, we will begin to identify the key determinants of the host-parasite interactions in this important pathogen.
|
10.1371/journal.ppat.1000072 | Genetic Compatibility and Virulence of Reassortants Derived from Contemporary Avian H5N1 and Human H3N2 Influenza A Viruses | The segmented structure of the influenza virus genome plays a pivotal role in its adaptation to new hosts and the emergence of pandemics. Despite concerns about the pandemic threat posed by highly pathogenic avian influenza H5N1 viruses, little is known about the biological properties of H5N1 viruses that may emerge following reassortment with contemporary human influenza viruses. In this study, we used reverse genetics to generate the 63 possible virus reassortants derived from H5N1 and H3N2 viruses, containing the H5N1 surface protein genes, and analyzed their viability, replication efficiency, and mouse virulence. Specific constellations of avian–human viral genes proved deleterious for viral replication in cell culture, possibly due to disruption of molecular interaction networks. In particular, striking phenotypes were noted with heterologous polymerase subunits, as well as NP and M, or NS. However, nearly one-half of the reassortants replicated with high efficiency in vitro, revealing a high degree of compatibility between avian and human virus genes. Thirteen reassortants displayed virulent phenotypes in mice and may pose the greatest threat for mammalian hosts. Interestingly, one of the most pathogenic reassortants contained avian PB1, resembling the 1957 and 1968 pandemic viruses. Our results reveal the broad spectrum of phenotypes associated with H5N1/H3N2 reassortment and a possible role for the avian PB1 in the emergence of pandemic influenza. These observations have important implications for risk assessment of H5N1 reassortant viruses detected in surveillance programs.
| The influenza pandemics of 1957 and 1968 were caused by hybrid viruses consisting of a mixture of human and avian influenza genes. The introduction of avian genes resulted in a sudden change of the virus surface antigens, allowing its worldwide spread due to lack of immunity in the population. The highly pathogenic avian influenza H5N1 virus has continued its spread in domestic and wild birds in Asia, Europe, and Africa. Although H5N1 infection in humans is rare and person-to-person transmission is very inefficient, the steady accumulation of human cases has raised concern over the possible reassortment between H5N1 and human seasonal influenza resulting in a virus with new surface antigens and pandemic potential. In this study, we used recombinant DNA technology to generate a systematic collection of hybrid viruses (with genes from human and avian viruses) bearing H5N1 surface antigens and analyzed their properties in cell culture and in mice. The H5N1 hybrid viruses revealed a broad range of viability and multiplication capacity in cell cultures. In addition, several H5N1 hybrid viruses were highly virulent in mice. Results from this systematic analysis provide important insight to support risk assessment of reassortant H5N1 avian influenza viruses.
| The emergence of an influenza virus that will cause a pandemic is inevitable and therefore preparedness is mandatory. The new pandemic influenza virus is likely to carry a hemagglutinin (HA) gene other than the currently circulating H1 and H3 lineages in order to escape immunity in the human population. However, we cannot predict the mechanism by which the pandemic influenza virus will emerge. One possibility is the transfer of an avian influenza virus from birds to humans, made possible by adaptive mutations, as postulated for the 1918 pandemic [1],[2]. Another possible scenario would follow the paradigm of the H2N2 and H3N2 influenza pandemics of 1957 and 1968 in which avian virus genes were incorporated into circulating human influenza viruses by reassortment [3], giving rise to viruses with novel surface antigens; i.e. antigenic shift. The segmented structure of the viral genome facilitates exchange of gene segments between two viruses co-infecting a single host cell. Dual infection with avian and human influenza viruses and subsequent reassortment may occur in hosts that are susceptible to both kinds of viruses and serve as mixing vessels that generate novel reassortants [4],[5].
Wild aquatic birds are the natural reservoirs for influenza A viruses and have been found to harbor each of the 16 known HA subtypes [6]. Highly pathogenic avian influenza (HPAI) H5N1 viruses are now enzootic among wild birds and poultry in three continents (http://www.who.int). Since 1997, when HPAI H5N1 viruses first emerged in Hong Kong to cause human respiratory illness and death, over 360 laboratory-confirmed human infections have been reported. Most human infections are caused by contact with infected poultry and to date H5N1 viruses have not yet acquired the ability to transmit efficiently among humans.
A major obstacle to transmission of the H5N1 virus among humans is thought to be the preferred receptor specificity of the H5 HA towards sialic acid (SA) with α2,3 linkage to galactose (the so-called avian receptor) [7],[8]. A switch of receptor specificity towards α2,6 linked SA (the human receptor) is considered to be a pre-requisite for sustained human to human transmission [9],[10]. However, it is not known whether other genes from H5N1 viruses would confer virulence and transmissibility in humans. It has been shown that a reassortant virus with the HA and NA from an H3N2 human virus and the PB2, PB1, PA, NP, M, and NS (so-called internal genes) of an H5N1 virus did not transmit efficiently in a ferret model [11]. (In this report, the term “internal genes” refers to the gene constellation comprising PB2, PB1, PA, NP, M, and NS, although the M gene encodes for the M2 protein, which is surface-exposed in virions.) The internal genes from this avian H5N1 virus were therefore postulated to lack at least one essential functional attribute to initiate a human pandemic. These critical function(s) might be acquired through a reassortment event between the H5N1 virus with a circulating human H3N2 influenza virus that generates the appropriate gene constellation.
In theory, a single reassortment event between two influenza A viruses can yield up to 254 (28 minus two parental viruses) hybrid genotypes. However, the few available reports suggest that the number of natural or experimental reassortants is likely to be smaller [5],[12],[13],[14],[15],[16]. Reliable estimates of the expected frequency of hybrid genotypes resulting from dual infections are not possible in the absence of systematic studies on human-avian influenza reassortment. Comprehensive in vivo co-infection studies and in vitro evaluations of all the reassortant genotypes derived from a human influenza virus and an HPAI virus would help bridge this gap of knowledge. In this report, we analyze the repertoire of reassortants between contemporary avian H5N1 and human H3N2 viruses by evaluating the phenotypes of 63 (26-1) viral reassortants with HA and NA genes from an avian H5N1 virus and the six internal genes from either parental virus, assigned higher priority because only viruses with novel surface antigens may cause a pandemic. We used reverse genetics to derive the reassortant virus panel, and subsequently examined their replication characteristics in cell culture and their virulence in a mammalian system. Our in vitro and in vivo analyses revealed a high frequency of viable reassortants with a wide spectrum of virulence for mice, providing insight into their potential for future emergence in nature.
To generate the collection of human-avian reassortant viruses for this study, we first developed plasmid-based reverse genetics (rg) systems for the two parental viruses; A/Wyoming/3/2003 (subtype H3N2) (WY03) and A/Thailand/16/2004 (H5N1) (TH04) [17]. The parental WY03 virus showed α2,6 linked SA receptor specificity [18], replicated to high titers in MDCK cell culture, and was avirulent in mice (data not shown). The TH04 virus showed α2,3 receptor specificity [8], replicated efficiently in MDCK cells and was highly virulent for mice [17]. Virus recovery from plasmid DNA transfections was evaluated by quantitative plaque analysis at 72 hours (h) post-transfection; herein referred to as rescue efficiency. Cell cultures transfected with WY03 and TH04 rg plasmids yielded >107 plaque-forming units (pfu)/ml of progeny virus, termed rgWY03 and rgTH04, which formed 4–5 mm diameter plaques, comparable to those of parental wildtype (wt) viruses (Figure 1). A wide range of virus yields and plaque diameters were obtained for each of the 63 H5N1 human-avian reassortant (rH5N1) plasmid transfections. In order to categorize the in vitro properties of each reassortant, rH5N1 genotypes were segregated into 4 phenotypic groups, according to their rescue efficiencies (Figures 1 and 2): (1) rH5N1 genotypes with wt or near-wt replication efficiency. Twenty-eight rH5N1 viruses (cell culture phenotype group 1) consistently yielded ≥106 pfu/ml in the transfected cell cultures (Figure 1), which represented rescue efficiencies similar to those of rgWY03 and rgTH04. Most of the cell culture group 1 viruses formed ∼2–4 mm plaques in diameter (Figure 1). The efficient in vitro growth phenotypes of nearly one-half of the rH5N1 viruses in the group revealed a high frequency of functional compatibility between avian and human virus genes. (2) rH5N1 genotypes with moderate cell culture replication impairment. Fourteen rH5N1 viruses (22% of the rH5N1 genotypes) had rescue efficiencies between 104 and 106 pfu/ml (cell culture phenotype group 2), and most of these viruses formed 1–3 mm plaques (Figure 2). (3) rH5N1 genotypes with severe cell culture replication impairment. Eight reassortants (13% of the rH5N1 genotypes) yielded ∼102–104 pfu/ml from transfected cell cultures, with plaque size ranging from 0.5–2 mm (cell culture phenotype group 3 in Figure 2). (4) Non-viable or marginally viable rH5N1 genotypes. Thirteen rH5N1 viruses (∼20% of rH5N1 genotypes) yielded <100 pfu/ml from transfected cell cultures (cell culture phenotype group 4 in Figure 2). Their rescue efficiencies were 5 log10 pfu/ml lower than their rg parent viruses. The severe replication defects of viruses in this group may reflect structural or functional incompatibilities in avian-human viral RNA and/or protein complexes. Collectively, these categories guided our rationale for excluding reassortants with severe replication defects from further in vivo studies.
Notably, the severely impaired rH5N1 viruses in group 4 (Figure 2) were all characterized by the association of the nucleoprotein (NP) gene from WY03 virus with matrix (M) and/or nonstructural (NS) genes derived from TH04 virus. For example, the single gene reassortant r5 (group 4), which carried the NP from WY03 and the five other internal genes from TH04 had a rescue efficiency of <102 pfu/ml. However, replacement of the TH04 NS gene with the WY03 NS in the same background increased rescue efficiency to ∼104 pfu/ml (r5/8 virus, group 2, Figure 2), which was significantly higher than r5 (P≤0.0001). Further introduction of the WY03 M segment into this gene constellation restored the rescue efficiency and plaque size of the reassortant virus (r5/7/8, group 1, Figure 1) to nearly wt level (P≤0.0001). In contrast, introduction of polymerase complex genes did not improve replication (r5 replication is similar to r1/5, r2/5, r3/5, and r1/2/3/5; P>0.9) (Figure 2). Conversely, only 6 out of 28 rH5N1 viruses (group 1) that replicated efficiently had NP of human origin, in every case along with human NS (Figure 1). These observations suggest that the NP gene of WY03 origin preferentially interacts with M and NS genes of the same origin for optimal replication. In contrast, the NP gene of the TH04 avian virus appears to be more compatible with the M or NS of heterologous origin (e.g., r1/2/3/7/8 virus replication was similar to r1/2/3/7 or r1/2/3/8; P = 1.0). Although not all viruses with human NP and avian M or NS were severely impaired, they generally displayed significantly reduced replication, suggesting that avian M and/or NS may not be incorporated into seasonal human H3N2 viruses in the absence of avian NP.
Another remarkable gene incompatibility was noted with the r2/3/5/7/8 virus, bearing TH04 PB2 and the remaining five internal genes from WY03 virus (Figure 2, group 3). This reassortant virus formed tiny plaques (0.5 mm diameter) and had a very low rescue efficiency. This defect was repaired by providing a PA gene of avian virus origin; i.e., the rescue efficiency of r2/3/5/7/8 was significantly lower than r2/5/7/8 (Figure 1, group 1) (P<0.0001), suggesting a functional interaction of TH04 PB2 with the cognate avian PA gene. This finding suggests that reassortment of avian PB2 genes with human viruses may be linked to co-incorporation of the avian PA gene.
A set of 38 rH5N1 viruses with cell culture replication efficiencies comparable to those of parental viruses, or with only modest reductions, were chosen for study in a BALB/c mouse model to assess their virulence in a mammalian host. The plasmid-derived rgTH04 virus was highly virulent for mice, as indicated by a very low intranasal 50% mouse infectious dose (MID50 = 101.5 pfu) and lethal dose (LD50 = 101.8 pfu) (Figure 3). This virus replicated to high titers (>107 pfu/ml) in lungs by day 4 following an intranasal inoculation of 104 pfu and caused >19% body weight loss. Viral replication was also detected at systemic sites, such as brain and spleen, recapitulating the outcome of infection with the wt TH04 isolate [17]. In contrast, replication of the rgWY03 virus in mice was very inefficient as evidenced by an MID50 of 106 pfu and an LD50 of >106 pfu (determination of MID50 for rgWY/03 required additional doses of 105 and 106 pfu to detect virus in tissues; data not shown). However, the reassortant virus bearing HA and NA from TH04 and the remaining genes from WY03 virus (r1/2/3/5/7/8) replicated efficiently (MID50 of 101.8 pfu and titer of 107 pfu/ml in the lung), suggesting that the HA and/or NA from WY03 lack appropriate interaction with receptors or other host factors in the mouse respiratory tract [19]. Most importantly, the internal genes from WY03 mediated efficient viral replication of r1/2/3/5/7/8 virus in mice validating the BALB/c mouse as a useful model to evaluate the influence of human/avian internal gene combinations on the virulence phenotypes of rH5N1 viruses.
Three rH5N1 viruses were highly virulent for mice, with an LD50<103 pfu (Figure 3, group A1). Each of these rH5N1 had an MID50 of ≤101.5 pfu, replicated to high titers in the lung (≥106.5 pfu/ml), and caused >17% weight loss by 6–7 days post-infection (dpi) on average. The virulence of these viruses was comparable to that of wt TH04. In addition, the high frequency of virus detection in the spleen and brain of mice indicated systemic spread of these viruses, resembling infection with wt TH04. The ten viruses in group A2 were moderately virulent, with a mean LD50 value that was significantly different from that of the highly virulent group A1 viruses (P<0.001). The remaining 25 rH5N1 viruses in groups B1 and B2 exhibited low virulence phenotypes in mice with LD50 values >104 pfu. However, five rH5N1viruses (Figure 3, virulence group B1) caused significant transient weight loss (>16%), clinical signs, such as ruffled haircoat and lethargy, and three viruses (r1/3/7/8, r1/2/3/8, r1/5/7/8) each caused mortality in a single mouse infected at 104 pfu, suggesting potential for increased virulence at higher virus inoculums (data not shown). The other 20 rH5N1 viruses (Figure 3, virulence group B2) caused subclinical infections in mice, with minor weight loss (<15%). These viruses spread to the spleen and /or brain sporadically and their pulmonary replication capacity ranged from substantially efficient to nil. Although many rH5N1 viruses with high rescue efficiencies and large plaque phenotypes also displayed highly virulent phenotypes in mice, several rH5N1 viruses belonging to virulence group B2 (i.e., r2/7/8 and r1/3/8) had high rescue efficiencies but did not replicate well in mice. This finding highlights the limitations of inferring in vivo virulence properties solely from efficient in vitro replication characteristics.
Interestingly, r1/3/5/7/8, one of the most virulent rH5N1 among the 38 reassortants inoculated into mice had a gene constellation resembling that of the pandemic viruses from 1957 and 1968. In 1957, HA, NA and PB1 genes from an avian H2N2 virus were introduced into the circulating human H1N1 virus and caused the so-called “Asian flu” pandemic, whereas the 1968 “Hong Kong” pandemic virus incorporated the HA and PB1 genes from an avian donor [3]. In this study, a virus carrying HA, NA and PB1 of avian origin and the remaining genes from a human virus, namely r1/3/5/7/8, was highly virulent for mice (LD50 = 102.5 pfu). In contrast, a reassortant virus (r1/2/3/5/7/8) with all the internal genes from WY03 virus, including PB1, caused minimal mortality and had a significantly different LD50 (1.3 log10 pfu increase; P<0.001), suggesting that the PB1 of contemporary H5N1 viruses can reassort into circulating H3N2 viruses and increase their virulence for mice.
Efficient viral replication at the lower temperature of the upper respiratory tract is thought to be essential for droplet transmission of influenza virus between humans. Avian influenza viruses with a PB2 polymerase bearing glutamic acid at position 627 instead of lysine have decreased replication at 33°C in mammalian cells [20],[21],[22]. Although both WY03 and TH04 viruses have lysine at position 627 in PB2, it is not known whether new avian and human gene constellations would compromise viral replication at lower temperature. To address this question, we determined reassortant viral titers in the nasal turbinates collected at 4 dpi. We found that in general, rH5N1 viruses replicated less efficiently in nasal turbinates than in lungs, as reported previously [21]. Interestingly, some reassortants (i.e., r2/7, r2/7/8, and r3/7/8) showed extremely low replication in nasal turbinates despite considerable titers in lungs (Figure 3). These reassortants would be expected to lack efficient transmissibility by generation of nasal secretion droplets.
Although mice are regarded as a useful mammalian model for studying the replication of HPAI viruses, the species differences between humans and mice mandate studies with models from the target species to complement the data. The epithelial cells of the respiratory tract are the primary targets of influenza infection. Therefore, we used in vitro differentiated HTBE cultures to evaluate the replication potential of the rH5N1 viruses in humans [23]. HTBE cells were infected with rgWY03 and rgTH04 viruses, or each of 38 rH5N1 viruses that were previously analyzed for virulence in mice. We quantified virus progeny released into the apical side of the pseudostratified epithelium because budding of HPAI H5N1 virus in the HTBE model remains polarized (data now shown). As shown in Figure 4A, both rgTH04 and rgWY03 parental viruses replicated efficiently in the HTBE cells. The rapid rise of WY03 virus titers to 108 pfu/ml at 32 h post-infection was consistent with the efficient spread of human viruses in HTBE cells, as described previously [23]. The plateau in WY03 virus production may be caused by virus-induced cell death, first noted at 40 h post-infection. In contrast, HTBE cells infected with rgTH04 virus showed no cytopathology and virus progeny increased steadily throughout the 56 h infection. The majority of rH5N1 viruses produced ≥104 pfu/ml by 24 h post-infection, and the growth kinetics were similar to parental rgTH04 or slightly delayed (e.g., r1/2/3/5/7/8 in Figure 4B and data not shown for others). In comparison, four rH5N1 viruses, r/3, r2/3, r3/8, r3/7/8, replicated substantially less efficiently in the HTBE cells (Figure 4, C and D). Interestingly, these four viruses also replicated poorly in mice; had MID50 values of ≥4 log10 pfu and caused minimal weight loss (Figure 3). These results supported the virulence data provided by the mouse model.
To study the mechanisms underlying the differences in the replication phenotypes of certain rH5N1, we exploited a mini-genome reporter assay which dissects the function of the viral ribonucleoprotein (RNP) complex from the rest of the viral gene products [24],[25]. The 16 possible RNP combinations of PB2, PB1, PA and NP from either the TH04 or WY03 viruses were studied at 33°C or 37°C, to recapitulate the temperatures of the upper and lower respiratory tract, as reported previously [20]. Another panel of RNP combinations with A/Vietnam/1203/2004 (VN04) viral genes replacing TH04 genes was also analyzed in parallel to extend the results for other H5N1 viruses. The human RNP was almost equally active at 33°C and 37°C, whereas the avian RNP activity was substantially reduced at 33°C despite the presence of lysine at position 627 of PB2, in both TH04 and VN04 backgrounds (Figure 5). The RNP constituted by PB1 and PA from WY03 virus and PB2 from TH04 (or VN04) virus resulted in extremely low polymerase activities at 33°C and 37°C (Figure 5A and B, RNP denoted by asterisks). Although the RNP complexes carrying PB2 and PB1 from TH04 and PA from WY03 virus showed partially reduced polymerase activity, a similar combination derived from VN04 and WY03 viruses showed a more pronounced loss of replication activity (Figure 5A and B, RNP denoted by arrows). The reduced polymerase activities of these gene constellations were consistent with the low viral titer from lungs and nasal turbinates of mice infected with reassortant viruses r3/7/8, r3/8, r3 and r2/3 (Figure 3). Interestingly, the polymerase activity of the RNP with PB1 from TH04 and the other proteins from WY03 was comparable to that of the wt WY03 RNP. These findings indicated that the increased mouse virulence attributed to avian PB1 in the WY03 genetic background (r1/3/5/7/8) may not directly result from stimulation of the polymerase activities of the RNP. Alternative hypotheses to reconcile these observations would include an in vivo role for PB1 in RNP function, a possible modulation of host cell function by PB1-F2, or unknown interactions of PB1 with the remaining 4 genes absent in this assay: HA, NA, M and NS.
Although H5N2 subtype viruses have been isolated from poultry in North America and Asia for many years [26], it is not clear whether the N2 derived from contemporary human H3N2 virus can support the efficient replication of a reassortant virus bearing the HA from circulating H5N1 virus. Balanced HA and NA activities are critical for efficient influenza virus infection and replication in various hosts. The HA of contemporary H5N1 viruses has retained a strong preference for α2,3 linked SA [7],[8]. In contrast, the NA derived from H3N2 human seasonal isolates has adapted over time to acquire α2,6 SA specificity [27],[28]. To evaluate the H5N2 reassortant, the NA of TH04 was replaced with the N2 from WY03 virus by reverse genetics. This virus, termed r6 (H5N2), was virtually identical to wt TH04 in rescue efficiency and replication in HTBE cells (Figure 1 and Figure 4B). In addition, this reassortant was highly virulent in mice, with an LD50 comparable to the parental TH04 virus (Figure 3). These results suggested that the NA activity from circulating H3N2 viruses can functionally support the activity of the H5 HA and promote H5N2 virus spread in the mammalian host.
The influenza pandemics of the past century were caused by viruses carrying at least one internal gene of avian origin and a novel HA subtype that acquired α2,6 SA receptor binding specificity [3]. While many studies have focused on adaptive mutations in the avian HAs required for acquisition of human receptor specificity, little is known about the importance of the avian virus internal genes in pandemic emergence [7],[29],[30]. In this report, we used reverse genetics to systematically study reassortants with each of the 63 possible combinations of internal genes from contemporary avian and human viruses; all with H5N1 surface protein genes.
Collectively, these studies revealed that certain genes, such as human NP and avian PB2, often caused severe replicative impairment in cell culture when transferred alone to the heterologous virus background, whereas transfer of other genes, such as PB1, was less detrimental. The incompatibility of the human NP with a full complement of avian influenza genes was noted in early studies with Fowl Plague virus [13]. This is significant because the NP gene of influenza virus plays an important role in host range specificity [5],[31],[32]. In this report, we provide evidence suggesting that reassortants with NP of avian origin in a human virus background can replicate efficiently in mammalian cell cultures. This phenotype does not require the presence of other avian virus internal genes, whereas the NP gene of human origin depends on cognate NS and M genes for expression of the efficient replication phenotype. The best characterized event of the viral infectious cycle involving NP, M, and NS gene products is the nuclear export of viral RNP. In the nucleus, the influenza nuclear export protein (NEP; encoded by the NS gene) interacts with the M1 protein, which binds to the newly assembled viral RNP. NEP also interacts with host protein CRM1, thereby directing the nuclear export of the viral RNP complex [33],[34],[35]. Although a direct interaction between NEP and NP proteins has not been shown, the severely defective growth of reassortants possessing heterologous M and NS relative to NP suggests an unidentified crosstalk between these viral proteins, with the possible involvement of a host protein(s).
Striking viral phenotypes were also evident in rH5N1 viruses with heterologous polymerase subunits. The PB1 protein interacts with PA and PB2 forming transcriptionally active heterotrimers [36],[37],[38]. Although a direct interaction between PB2 and PA has never been reported, our genetic analyses pointed towards a specific interdependence between PB2 and cognate PA genes of avian origin, either through direct protein-protein interaction or concerted interaction with other viral or host factor(s). Interestingly, natural avian-mammalian reassortant viruses isolated from humans and swine possess PB2 and PA of the same host origin and sometimes carry a PB1 derived from a virus adapted to a third host species [39],[40]. Thus, linkage between avian PB2 and PA would be expected in the event of reassortment between an H5N1 virus and a seasonal H3N2 virus from humans.
The role of the avian PB1 genes in the emergence of reassortant viruses that caused the 1957 and 1968 influenza pandemics has remained enigmatic. This study shows that incorporation of an avian PB1 gene into a background of human virus internal genes significantly increased mouse virulence. We postulate that acquisition of the avian PB1 gene, as was seen in the 1957 and 1968 pandemic influenza strains may be a critical factor in the early stages of a pandemic, allowing the emerging reassortant to overcome competition with seasonal influenza viruses by enhancing its replication or virulence. Our results, therefore, have implications for assessing the potential virulence of novel reassortant viruses possessing human virus internal genes and PB1 from currently circulating H5N1 viruses.
Although reassortment between two different viruses could yield 254 possible new genotypes, this study characterized the subset of 63 genotypes with H5N1 surface antigens, of highest public health significance. In addition, these studies show that a reassortant virus with NA from a contemporary human H3N2 virus and the remaining 7 genes from TH04 replicated efficiently and was as lethal as wt H5N1 virus in mice, indicating that the current human N2 is compatible with the receptor binding function of the H5 HA. Although we did not analyze all the 63 additional genotypes carrying H5N2 surface genes, we anticipate that their virulence would be similar to their rH5N1 counterparts. However, these data should be interpreted in a broader context of human and avian influenza virus replication and evolution. The genotype of the rH5N1 that would emerge from natural co-infection is dictated by many factors besides the replication competency of a given reassortant. Dual infection of a single cell with human and avian influenza viruses involves co-replication of two genomes that may complement, interfere, and compete with each other. These events and the subsequent expansion of the reassortan will be further conditioned by the host species and tissue tropisms of the parental viruses and resulting reassortants. Ultimately, while use of reverse genetics technology to generate reassortants provides an experimental platform free of these many variables, natural reassortment between two viral genomes, and the consequences therein, are more complex.
In summary, we report a strikingly high level of compatibility between avian and human virus genes. Because few studies have described naturally occurring or experimentally derived avian-human reassortants, our results were surprising in that almost half of the rH5N1 viruses tested showed a high frequency of functional compatibility between avian and human virus genes. In addition, approximately 1 in 5 of all possible H5N1 reassortants was lethal for mice at doses below 104 pfu. The highly virulent reassortant genotypes identified in this study suggest that introduction of certain H5N1 viral segments into circulating human H3N2 viruses may increase their virulence for mice and perhaps other mammalian species. In addition, the moderately virulent reassortant viruses could circulate in a mammalian host, evolve by compensatory and/or adaptive mutations, and become more virulent for humans. The results of this study, therefore, underscore the necessity for enhanced viral surveillance strategies, which monitor reassortment events in nature to reduce the public health threat posed by H5N1 HPAI viruses currently circulating in three continents.
A/Thailand/16/2004 (TH04) and A/Vietnam/1203/2004 (VN04) H5N1 viruses and A/Wyoming/3/2003 (WY03) H3N2 virus obtained from the WHO Global Influenza Surveillance Network were provided by Alexander Klimov (CDC, Atlanta, USA). Madin-Darby canine kidney (MDCK) and human lung carcinoma (A549) cells were obtained from the American Type Culture Collection and propagated in Dulbecco's Modification of Eagle's Medium with 10% fetal bovine serum. Viral infectivity was determined by plaque assay on MDCK cells as described [41]. Reassortant viruses containing any segment derived from the H5N1 virus were generated in compliance with the Institutional Biosafety Committee and NIH Guidelines for Research Involving Recombinant DNA Molecules. Viruses were handled in biosafety level 3 containment, including enhancements required by the U.S. Department of Agriculture and the Select Agents program http://www.cdc.gov/od/ohs/biosfty/bmbl5/bmbl5toc.htm.
RT-PCR amplicons of the eight viral genes from WY03 and TH04 viruses were cloned into a dual-promoter plasmid for influenza A reverse genetics [42]. Virus rescue was performed by plasmid DNA transfection into co-cultured 293T/MDCK cells [42]. Culture medium from the transfected cells was harvested at 72 h and analyzed by plaque assay on MDCK monolayers. The plaque count and diameter were recorded as a measure of the virus rescue efficiency from plasmid DNA. DNA transfection of each genotype was performed at least three times independently. WY03 and TH04 rg plasmid sets were included as controls during each reassortant rescue to evaluate experimental variation. Viruses with H5 HA were propagated in 10–11 days old embryonated chicken eggs. The H3N2 virus was propagated in MDCK cells in the presence of 1 μg/ml TPCK-treated trypsin. Following propagation, the full genomes of reassortant viruses were sequenced to confirm presence of parental virus sequence.
A549 cells cultured in 24-well tissue culture plates were co-transfected with pPol1-NS-Renilla (100 ng) encoding a reporter mini-genome under transcriptional control of the human RNA polymerase I, pSV-Luc (200 ng) encoding firefly luciferase under SV40 virus RNA polymerase II promoter control, and four plasmids expressing viral PB2, PB1, PA, NP (50 ng each) from the strain of interest. Twenty-four hours after the transfection, the cell lysates were harvested and further diluted to perform the dual luciferase assay according to the manufacturer's protocol (Promega). The influenza polymerase catalytic activity derived from the Renilla luciferase plasmid (pPol1-NS-Renilla) was corrected to account for well-to-well differences in transfection efficiency using the firefly luciferase activity values from pSV-Luc plasmid.
Groups of 6-8 week old female BALB/c mice (Jackson Laboratories, Bar Harbor, ME) were placed under light anesthesia and inoculated intranasally with 50 μl of serial 10-fold dilutions of infectious virus in PBS. For reassortant viruses tested, 104 pfu of virus was the highest dose used to infect mice; for WY03 virus, 106 pfu of virus was tested. Three mice from each group were euthanized at 4 days post-infection (dpi) and nasal turbinates, lungs, spleens, and brains were harvested, immediately frozen on dry ice, and stored at −80°C until further processing. Whole tissues were thawed, homogenized in 1 ml of cold PBS, and clarified by centrifugation (2,200×g) at 4°C. Virus titers of homogenates were determined by plaque assay in MDCK cells. Five additional mice in each group were monitored daily for clinical signs for 14 dpi. Mice that lost more than 25% of their body weight were euthanized humanely. The fifty percent mouse infectious dose (MID50) and fifty percent lethal dose (LD50) were calculated and expressed as the pfu value corresponding to 1 MID50 or LD50. Animal studies were conducted per approved Institutional Animal Care and Use Committee protocols.
Statistically significant differences of rescue efficiencies of avian-human reassortants in cell culture were determined by F-test adjusted for multiple comparisons. LD50 and MID50 values were calculated using the method of Reed and Muench [43]. Statistically significant differences between LD50 values of viruses in virulence group A1 and A2 were determined by comparing groups A1 and A2 to TH04 WT and group A1 to group A2 using an analysis of variance performed by an F test for multiple comparisons.
Growth and differentiation of primary human tracheobronchial epithelial cells were performed as described previously [23],[30],[44]. Briefly, primary cells (passage level 3) were seeded in porous membrane inserts (Corning, 4.5 μm, 12 mm diameter) at the density of 5×104 cell/cm2. Three days after seeding the cells, the medium from the apical side was removed and the confluent monolayers were cultured at an air-liquid interface. The medium from the basal compartment was replaced daily, and the in vitro differentiation of primary cells was achieved after 4–6 weeks. Differentiated cells with trans-epithelial electrical resistance of ≥300Ω cm2 were used in our study. Kinetic analysis of reassortant virus growth was performed after infection of the monolayer at a multiplicity of infection (moi) of 0.02 pfu/cell as described [23],[30]; apically released virus was harvested at the appropriate times and analyzed by plaque assay.
The GenBank (http://www.ncbi.nlm.nih.gov/sites/entrez) accession numbers for the genes described in this paper are: EU268216 (A/Thailand/16/2004, PB2 gene), EU268217 (A/Thailand/16/2004, PB1 gene), EU268218 (A/Thailand/16/2004, PA gene), EU268219 (A/Thailand/16/2004, HA gene), EU268220 (A/Thailand/16/2004, NP gene), EU268221 (A/Thailand/16/2004, NA gene), EU268222 (A/Thailand/16/2004, M gene), EU268223 (A/Thailand/16/2004, NS gene), EU268224 (A/Wyoming/03/2003, PB2 gene), EU268225 (A/Wyoming/03/2003, PB1 gene), EU268226 (A/Wyoming/03/2003, PA gene), EU268227 (A/Wyoming/03/2003, HA gene), EU268228 (A/Wyoming/03/2003, NP gene), EU268229 (A/Wyoming/03/2003, NA gene), EU268230 (A/Wyoming/03/2003, M gene), EU268231 (A/Wyoming/03/2003, NS gene). |
10.1371/journal.pbio.1001222 | Coupled Motions Direct Electrons along Human Microsomal P450 Chains | Protein domain motion is often implicated in biological electron transfer, but the general significance of motion is not clear. Motion has been implicated in the transfer of electrons from human cytochrome P450 reductase (CPR) to all microsomal cytochrome P450s (CYPs). Our hypothesis is that tight coupling of motion with enzyme chemistry can signal “ready and waiting” states for electron transfer from CPR to downstream CYPs and support vectorial electron transfer across complex redox chains. We developed a novel approach to study the time-dependence of dynamical change during catalysis that reports on the changing conformational states of CPR. FRET was linked to stopped-flow studies of electron transfer in CPR that contains donor-acceptor fluorophores on the enzyme surface. Open and closed states of CPR were correlated with key steps in the catalytic cycle which demonstrated how redox chemistry and NADPH binding drive successive opening and closing of the enzyme. Specifically, we provide evidence that reduction of the flavin moieties in CPR induces CPR opening, whereas ligand binding induces CPR closing. A dynamic reaction cycle was created in which CPR optimizes internal electron transfer between flavin cofactors by adopting closed states and signals “ready and waiting” conformations to partner CYP enzymes by adopting more open states. This complex, temporal control of enzyme motion is used to catalyze directional electron transfer from NADPH→FAD→FMN→heme, thereby facilitating all microsomal P450-catalysed reactions. Motions critical to the broader biological functions of CPR are tightly coupled to enzyme chemistry in the human NADPH-CPR-CYP redox chain. That redox chemistry alone is sufficient to drive functionally necessary, large-scale conformational change is remarkable. Rather than relying on stochastic conformational sampling, our study highlights a need for tight coupling of motion to enzyme chemistry to give vectorial electron transfer along complex redox chains.
| Enzymes are proteins that catalyze a large array of chemical reactions, often in partnership with other enzymes. We understand in detail the chemical mechanisms of many of these reactions; however, the importance of the physical movements of enzymes during catalysis (or protein dynamics) is, increasingly, becoming apparent. In this study, we have placed fluorescent markers on an enzyme called cytochrome P450 reductase (CPR) to probe the dynamic changes in the physical conformation of the protein as the reaction chemistry proceeds. CPR catalyses the transfer of electrons from a small molecule donor (called NADPH), ultimately passing them to their partner enzymes called CYPs. We were able to correlate specific conformational changes with distinct chemical steps in CPR. We found that the chemical transformation itself induces the enzyme to adopt conformations that are required for its efficient interaction with CYPs. These findings have allowed us to develop a model of CPR activity in which electron transfer along the pathway from NADPH through CPR to CYP is tightly integrated with physical conformational control of the enzyme.
| The relationship between dynamics and the function of proteins is important. Proteins undergo a wide range of motions in terms of time (10−12 to >1 s) and distance (10−2 to >10 Å) scales and any of these may be significant catalytically and related directly to function [1]–[5]. Proteins exist in an equilibrium of conformational states that define a multi-dimensional free energy landscape, enabling proteins to explore high energy states [6]. Mutagenesis can induce altered landscapes leading to energy traps with consequent effects on catalytic efficiency [7],[8]. It is in the nature of catalysis that high energy states are populated transiently during the course of an enzyme-catalyzed reaction. The ability to study these states experimentally, and to assess their impact on biological function, is a major challenge. Evidence points to a range of spatial and temporal dynamical contributions to substrate binding, product release, and chemical catalysis [9]–[11].
There is evidence supporting a role for domain motion in catalysis in the important family of diflavin oxidoreductases typified by human cytochrome P450 reductase (CPR) and human methionine synthase reductase (MSR) [12],[13]. Pulsed Electron Electron Double Resonance (PELDOR) studies of both CPR and MSR indicate landscape remodeling induced by ligand binding. Domain motion in this enzyme family has also been inferred from structural studies (crystallographic [14]–[17] and solution state [18]) and from pressure-dependent kinetic studies of electron transfer in CPR [12]. CPR is a membrane-bound NADPH-dependent oxidoreductase that contains FAD and FMN cofactors housed in discrete redox domains separated by a flexible hinge region [15]. CPR catalyzes electron transfer from NADPH to cytochrome P450 (CYP) enzymes in the endoplasmic reticulum. The relative orientation of the two flavin redox domains is variable, giving rise to “open” and “closed” conformations of the enzyme as seen in crystallographic analysis of homologous wild-type and mutant forms [16],[19]. NMR and small angle X-ray scattering studies suggest that CPR adopts a more closed conformation on coenzyme binding [18], similar to the conformation of crystallized rat CPR in which the dimethylbenzene moieties of the FAD and FMN cofactors are juxtaposed [15]. This closed conformation is optimal for interflavin electron transfer since the short interflavin distance enhances electronic coupling. Despite this close approach, interflavin electron transfer is slow (∼50 s−1) as measured by temperature jump [20],[21] and flash photolysis [22] time-resolved spectroscopies. These studies imply adiabatic control of electron transfer through conformational sampling [23]. This is consistent with temperature [24], pressure [12], and viscosity dependence [20] analysis of electron transfer kinetics, and with the multiple conformational states of human CPR seen in PELDOR studies [12]. Whilst the closed state of CPR is optimal for interflavin electron transfer, interaction with CYP enzymes requires a more open state. FMN domain residues that interact with CYP enzymes are occluded in the closed state [25]. A sequential opening and closing of CPR during the catalytic cycle is therefore proposed to facilitate internal electron transfer and subsequent transfer of electrons to CYP enzymes [16],[26]. This proposed cycling between open and closed conformations is consistent with impaired CYP reduction by CPR containing a non-native disulphide bond that links the FAD and FMN domains and the rescue of activity following reduction of this bond [27].
Evidence for conformational cycling during CPR catalysis is largely circumstantial. A direct means of analyzing conformational variations during enzyme catalysis is required to link the kinetics (and energy barriers) of conformational change to the chemical (redox) changes that result from hydride transfer (NADPH→FAD) and electron transfer (FAD→FMN). There are major problems to be addressed, including (i) identification of the “drivers” that open and close CPR; (ii) discrimination between electron transfer mechanisms that rely on conformational change coupled to chemical or binding events, or stochastic sampling of conformational space (i.e., conformational sampling mechanisms of electron transfer [28],[29]); (iii) whether the timescales for opening and closure support directional electron transfer from NADPH to CYP enzymes. With these key questions in mind our strategy has been to develop a direct method for analyzing the spatial and temporal properties of domain motion in human CPR using time-resolved Fluorescence Resonance Energy Transfer (FRET) during catalytic turnover. Our approach employs extrinsic fluorophores (Alexa 488 and Cy 5) attached at different positions on the solvent exposed surface of CPR to enable spatial (range ∼20–80 Å) and temporal (range ms to s) mapping of conformational variation during stopped-flow studies of flavin reduction by NADPH. In this way, we have been able to correlate the time dependence and extent of conformational change with individual rate constants for hydride and electron transfer in CPR. Using this direct approach, we have elucidated how motions link to enzyme chemistry, identified the “drivers” of these motions, and gained important new insight into how these motions facilitate directional transfer of electrons along human microsomal P450 chains.
We generated homology models for several closed and open structures of human CPR based on X-ray crystal structures of the homologous rat CPR (94% sequence identity) [15],[16]. The fully closed structure is shown in Figure 1. Based on these models we reasoned that CPR can bind at least two mole equivalents of an extrinsic fluorophore through a thiol linkage to cysteine residues. Figure S1 shows the absorbance spectra for the donor (Alexa 488 (D)) and acceptor (Cy 5 (A)) fluorophores attached to cysteine residues, in fluorophore-labeled CPR (CPR-DA). The fluorophore and protein concentrations determined from this spectrum indicate stoichiometric attachment of the two fluorophores, giving a total fluorophore∶CPR ratio of 2∶1. Mass spectral analysis indicates that three cysteines are labeled using our protocol, namely C228, C472, and C566 (Figure S2), suggesting fractional labeling of each cysteine (see Text S1 for detailed discussion). C228 is located in the FMN domain and C472/C566 in the FAD domain, as shown in Figure 1. We have not attempted to remove the multiple cysteines in the FAD domain as we wish to study the wild-type enzyme, particularly since mutagenesis may have unknown effects on the protein dynamics. We note that there is only one labeled residue in the FMN domain, C288. From our homology models, opening of CPR results in a decreased distance between C228 in the FMN domain and C472/C566 in the FAD domain.
Figure 2A shows the emission spectra of both donor labeled CPR (CPR-D) and CPR-DA, where the donor is excited at 495 nm. For CPR-DA, there is significant emission arising from the acceptor (∼670 nm) with a corresponding decrease in the emission arising from the donor (∼520 nm) compared to CPR-D. This indicates that there is efficient FRET from donor to acceptor when bound to CPR. We observed a small emission peak at ∼670 nm for the Cy 5 labeled CPR (CPR-A) when excited at 495 nm (Figure S3), but the relative emission is far smaller (<∼3%) than that attributed to FRET. If the FMN domain moves significantly relative to the rest of CPR (as is proposed to occur following flavin reduction), the FRET efficiency is expected to change, manifesting as a change in the ratio of acceptor to donor emission (A∶D). We were able to determine contributions to the FRET signal from inter-protein FRET as well as direct physical interaction of the extrinsic fluorophores with the flavin cofactors. A description of these control studies is given in Supporting Information (Text S1; Figure S4). We found no evidence to indicate that either of these processes contributes to the observed emission that we attribute to intraprotein FRET. We are therefore confident that the experimental setup reports on conformational change.
Several studies have suggested that CPR undergoes a conformational change associated with coenzyme binding [12],[18],[20],[30]. Specifically, PELDOR spectroscopy of di-semiquinoid CPR (containing FAD semiquinone and FMN semiquinone) has revealed that binding of NADP+ leads to formation of a more closed distribution of CPR structures compared to ligand-free di-semiquinoid enzyme [12]. It is possible to form an enzyme-coenzyme complex by incubating oxidized CPR with NADP+. Should binding of NADP+ induce CPR closure, the distance between C228 and the cysteines in the FAD domain will increase (as discussed above), resulting in poorer FRET efficiency between donor and acceptor (i.e., a decrease in the A∶D emission ratio). Figure 2B shows the resulting A∶D ratio for the emission of the CPR-DA fluorophores excited at 495 nm when titrated against NADP+. The individual donor and acceptor emission titrations are shown in Figure S5, normalized for the corresponding changes in fluorescence of CPR-D and CPR-A as described in Materials and Methods. This removes effects such as quenching by aromatic residues/NADP+ and FRET involving the flavin cofactors, leaving only changes attributable to FRET between the extrinsic fluorophores. From Figure 2B, the A∶D ratio decreases with increasing NADP+ concentration and saturates with a constant, KS = 1.6±0.5 mM. These data indicate that coenzyme binding induces formation of a more closed form of CPR and demonstrate that our experimental system can detect relative domain movements in CPR.
By monitoring the change in fluorescence emission of the fluorophores in stopped-flow studies of flavin reduction by NADPH, we have been able to correlate the kinetics of conformational change with enzyme chemistry. We assessed the degree of photo-bleaching of the fluorophores in oxidized CPR-D (ex 495 nm) and CPR-A (ex 655 nm). Example traces are given in Figure S6A. In each case there is a small decrease in fluorescence emission of ∼1% over 500 s. This small amount of photo-bleaching is not used to correct subsequent traces as the magnitude of the quenching is relatively small. Next, we determined if binding of NADP+ in stopped-flow studies causes a measurable change in protein conformation as demonstrated also in NADP+ titration experiments (Figure 2B). The change in FRET (CPR-DA excited at 495 nm) was monitored on mixing oxidized and 2-electron reduced CPR-DA with a saturating concentration (5 mM) of NADP+. Example traces are given in Figure S6B–C. The observed changes in emission following mixing with NADP+ are similar to those recorded for photo-bleaching. However, we observed small shifts in the absolute magnitude of fluorescence at t = 0 for enzyme versus NADP+ mixes compared to enzyme versus buffer control mixes. This indicates a loss in the fluorescence signal of a magnitude similar to the titration study (Figure 2B) in the dead time of the stopped-flow instrument, consistent with fast (<5 ms) conformational closure of CPR. Since coenzyme-induced closure of CPR is fast, we infer that any fluorescence changes observed beyond the instrument dead time in reactions of CPR with NADPH would be related to conformational change accompanying chemical (redox) change in the enzyme catalytic cycle.
We extracted time-resolved changes in FRET between the D and A fluorophores as flavin reduction proceeds. In this way we assessed relative conformational change associated with flavin reduction in CPR. The time-resolved FRET response is deconvoluted from other contributions to the emission response such as quenching by aromatic residues or FRET involving the flavin moieties in a similar manner as our NADP+ titration study (Figure 2B). This is achieved by subtracting the traces for CPR that contained only a single fluorophore (species CPR-D and CPR-A; fluorescence traces shown in Figure S7A and B, trace (i)) from the fluorescence traces for the corresponding fluorophore in a FRET pair (Figure S7B, trace (ii)). The resulting difference traces (Figure 3A) then show the fluorescence emission due to FRET between the extrinsic fluorophores alone. Opposition of the D and A traces was not observed (as expected for FRET data) despite deconvolution of the FRET response, suggesting we recover an approximation of the pure FRET signal. Consequently, we have not calculated detailed distance information from the FRET data, but simply used the FRET signal qualitatively to follow changes in the distribution of CPR conformations in a time-resolved manner. To extract rate constants for the observed changes in the conformational distribution, we simultaneously fit both the donor and acceptor traces in Figure 3A to a multi-exponential expression (Text S1) with linked rate constants for each kinetic phase. This method is robust as the shifting sign of the amplitude for each kinetic phase facilitates good resolution of potentially similar rate constants and small amplitudes. Data fitting is described in detail in Supporting Information (Text S1). The extracted observed rate constants and amplitudes are given in Table S1. These data can be minimally fit to a four exponential function, suggesting there are at least four conformational transitions that occur during flavin reduction by NADPH. The extracted rate constants are essentially the same for each kinetic phase for any of the traces shown in Figure S7 (Table S1). This is consistent with our assertion that the observed changes in fluorescence emission of the fluorophores are due to conformational changes in the enzyme only. That is, the traces give the same rate constants, despite different mechanisms (quenching, FRET, etc.), since the changes in fluorescence emission are caused by the same conformational change as flavin reduction proceeds. Moreover, changes in the FRET signal report specifically on distance changes between C228 and C472/C556 as shown by control experiments in which negligible changes in A∶D ratio were seen with a variant form of CPR containing the C228S mutation (see Text S1 and Figures S8 and S9).
Absorption studies of flavin reduction by NADPH in CPR using stopped-flow methods have previously been reported [12],[24],[31] and can be used to dissect flavin reduction in CPR in detail. These studies indicate that NADPH binds to the FAD domain where it transfers a hydride to the N5 of FAD followed by electron transfer from FAD to FMN to yield a distribution of 2-electron reduced species (FADH• FMNH•, FADH2 FMN and FAD FMNH2). In the absence of an electron acceptor (such as CYP) a second equivalent of NADPH binds to the FAD domain and transfers a hydride to FAD, driving the equilibrium distribution of enzyme states towards the fully (4-electron) reduced species (FADH2 FMNH2). The observed rate constants for formation of 2-electron (FMNH• FADH•) and 4-electron (FMNH2 FADH2) reduced CPR can be monitored by following the formation and decay of the di-semiquinoid (FMNH• FADH•) 2-electron reduced species at 600 nm on mixing with a saturating concentration of NADPH in a stopped-flow instrument [31]. The two exponential phases extracted from these reaction traces correspond broadly to the observed rate constants for 2-electron and 4-electron reduction (termed k1 and k2, respectively). CPR-DA reacts with NADPH in a similar way, and the kinetics of absorption change at 600 nm for the two exponential phases are identified as “flavin reduction” (black trace) in Figure 3B. Further, there is a very slow phase (increase in 600 nm absorbance; represented as “EQ” in Figure 3B) observed after 4-electron reduction. In wild-type CPR, this phase has been attributed previously to the formation of an internal equilibrium between redox states in the absence of an electron acceptor [31]. This slow adjustment to the final equilibrium of redox states is retained in CPR-DA. In the present study, we focus on the chemical steps, k1 and k2. Figure 3B shows a typical reaction trace for flavin reduction by NADPH in CPR monitored at 600 nm. The data are fit to a four exponential function (Text S1, Equation S1) accounting for all the observed absorption changes discussed above. The rate constants for k1 and k2 at 25°C are given in Table 1, extracted as kobs1 = 16.2±0.2 s−1 and kobs2 = 4.0±0.1 s−1. Table 1 also shows the observed rate constants for the first two kinetic phases extracted from the fluorescence data that represent conformational change (Figure 3A). The rate constants extracted from the fluorescence data (kobs1 = 24.5±1.0 s−1 and kobs2 = 4.2±0.1 s−1) are similar to the observed rate constants for flavin reduction, suggesting that flavin reduction and conformational change are linked in CPR.
We now correlate the observed changes in flavin redox state with the conformational changes extracted from our fluorescence data. Figure 3B shows the ratio of acceptor to donor (A∶D) emission extracted from the stopped-flow traces shown in Figure 3A, effectively describing the trend in FRET efficiency during flavin reduction by NADPH. These data clearly show that the FRET signal is increased (i.e., more “open” conformations are populated) as CPR is sequentially reduced to the 2-electron and then 4-electron levels (indicated by the time-resolved absorption measurements at 600 nm; Figure 3B). Further, following flavin reduction we observed a gradual closing of CPR (reduced A∶D emission ratio) over prolonged time periods (10 to 1,000 s) as CPR relaxes to the final equilibrium position. These FRET data indicate, therefore, that conformational closure is a key part of this long-time base equilibration of the reduced enzyme species and that the more open state is a metastable form of reduced CPR. The fluorescence and absorption changes shown in Figure 3B occur over very similar timescales (0–10 s), suggesting that domain motion is linked to redox chemistry. The potential for direct coupling of redox change with conformational opening of CPR is addressed below.
Strong evidence for the coupling of conformational change with enzyme chemistry would arise if the energetic barriers for electron transfer and motion are shown to be equivalent. This was addressed experimentally by monitoring the temperature dependence of the rate of flavin reduction (absorption change at 600 nm, reporting on k1 and k2) and associated conformational change (from time-resolved fluorescence data). The temperature dependence of the rate constants k1 and k2 for structural change and flavin reduction is shown in Figure 4. The Eyring plots for the fluorescence data are linear (Figure 4), suggesting that the first two exponential phases each report on rate constants for a single process (i.e., structural change). The values of ΔH‡ for flavin reduction and conformational change are given in Table 1 and are similar for both kinetic phases (k1 and k2). That ΔH‡ is very similar for both flavin reduction and structural change for both exponential phases (2-electron and 4-electron reduction) suggests a tight coupling of the reaction chemistry with the observed structural transitions.
The kinetics and energetics of flavin reduction and conformational change are consistent with the two processes being tightly coupled. These data might suggest that flavin reduction is responsible for conformational opening of CPR or that conformational change induces electron transfer associated with flavin reduction. The analysis, however, is complicated by the opposing effects of coenzyme binding, which in the absence of redox chemistry is known to effect closure of CPR (Figure 2B). We therefore conducted stopped-flow measurements in which CPR-DA was mixed with the chemical reductant, sodium dithionite, to investigate the effects of redox change in the absence of coenzyme binding. A typical trace from these experiments is shown in Figure 5A. As with NADPH reduction, when dithionite is used to reduce the flavin centers the appearance and subsequent disappearance of the flavin semiquinones is observed at 600 nm corresponding to formation of the 2-electron and 4-electron reduced species of CPR. The individual exponential components corresponding to 2-electron and 4-electron reduction are less well defined, which complicates data analysis, but approximate rate constants can be obtained. The observed rate constant for flavin reduction is far slower, with dithionite being ∼0.05 s−1 and ∼0.04 s−1, compared to NADPH, ∼19 s−1 and ∼2.5 s−1, at 20°C for 2- and 4-electron reduction, respectively. The change in FRET efficiency for CPR-DA was also monitored as flavin reduction proceeds (Figure 5A). Individual fluorescence traces used to calculate the FRET response are shown in Figure S11A,B. The change in FRET efficiency can be adequately fit to a two-exponential function (Text S1, Equation S1) with observed rate constants of ∼0.05 s−1 and ∼0.03 s−1 for the first and second kinetic phases, respectively. In general, there is a large increase in A∶D (CPR opening) as flavin reduction proceeds. We note that the first exponential phase shows a slight decrease in A∶D, though this is on a faster timescale than either 2- or 4-electron reduction (Figure 5A). Therefore, as seen with NADPH, the rate of flavin reduction by dithionite correlates with the observed rate of conformational change (Figure 5) and reduction to the 2-electron and 4-electron levels is accompanied by an opening of CPR. These data indicate that reduction of the flavin cofactors alone is sufficient to induce conformational opening of CPR. Further, these data are consistent with our temperature dependence data from which we inferred a tight coupling of the conformational transition with the flavin redox state.
We also examined the effects of coenzyme binding on the rate of flavin reduction and conformational opening during enzyme reduction with dithionite. This is achieved by mixing CPR-DA that had been pre-incubated with a saturating concentration of NADP+ (5 mM) with dithionite. There was no evidence for reduction of NADP+ to NADPH (monitored by absorption changes at 340 nm) by dithionite over the timescale of the study. In the presence of NADP+, the rate constant for flavin reduction by dithionite (2-electron reduction) increases approximately 2-fold (∼0.01 s−1; Figure 5B) compared with reactions performed in the absence of NADP+. The conversion of 2-electron reduced CPR to the 4-electron reduced species is not well-resolved, due likely to overlap with the slow kinetic phase(s) involved in establishing the final equilibrium of redox states (EQ) (Figure 5B). The corresponding change in FRET efficiency is shown in Figure 5B with the individual fluorescence trace shown in Figure S11C and S11D. As with dithionite alone, there is a significant increase in the A∶D ratio corresponding to CPR opening and this occurs on a similar timescale to flavin reduction (approximate rate constant ∼0.08 s−1).
The effect of NADP+ is therefore to accelerate (approximately 2-fold) flavin reduction and the associated conformational opening of CPR with dithionite as reductant. Further, after the opening of CPR with NADP+ bound there is a subsequent decrease in A∶D reflecting CPR closure as the reduced enzyme relaxes to the final equilibrium (EQ) state (Figure 5B). On the timescale of our measurements we do not observe the establishment of this equilibrium in dithionite studies performed in the absence of NADP+ (Figure 5A). We find therefore that NADP+ not only increases the observed rate of flavin reduction, but also increases the observed rate of EQ formation. This is consistent with previous t-jump studies of interflavin electron transfer in di-semiquinoid human CPR, where the observed rate of electron transfer (55 s−1) is increased 5-fold on adding NADP+ compared to reactions performed in the absence of nicotinamide coenzyme [20]. The precise reasons for accelerated flavin reduction in the presence of NADP+ are unclear. However, cofactor binding likely induces a shift in the equilibrium distribution of enzyme forms towards a more closed conformation (Figure 2B). We suggest that internal electron exchange between the flavin cofactors is enhanced due to a minimized cofactor separation induced by NADP+ binding. Clearly, once CPR is reduced by dithionite the distribution then adjusts first to the metastable open conformations (0–50 s) and then relaxes to the more closed EQ conformations (>50 s) (Figure 5B).
We have monitored two separate conformational transitions in CPR, namely opening and closing, which correspond to increased and decreased separation of the FMN and FAD cofactors, respectively. Opening is driven by reduction of the flavin cofactors and closure by coenzyme binding (Figure 6A). We now develop an integrated model for CPR action that incorporates these conformational transitions (Figure 6). In this model, domain motions driven by flavin reduction are crucial in mediating electron transfer to CYPs. The open conformations expose residues required for CYP interaction that are occluded in the closed conformation [25]. The open state signals that CPR is “ready and waiting” to transfer electrons to CYP partners in the microsomal membranes. Should productive interaction with CYP partners not occur, subsequent closure of CPR (formation of EQ) may offer some protection of the reducing equivalents in the flavin cofactors by suppressing their adventitious transfer (e.g., to molecular oxygen). The open conformation of CPR is not appropriate for “electron loading” from the reducing coenzyme NADPH. Rapid equilibration of electrons across the flavin centers is required to generate 2- and 4-electron reduced CPR from the oxidized form. Therefore, “closure” of the oxidized form of CPR is induced on nicotinamide coenzyme binding to facilitate efficient loading with reducing equivalents prior to redox-driven opening of the structure. We note that there is evidence for both 2, 4 and 1, 3 electron cycling in CPR [32],[33], and we propose a similar mechanism of opening/closing driven by the flavin redox state can occur in either case. However, in vivo the need for a second hydride transfer from NADPH may be less important (Figure 6B). The redox and ligand-bound forms of CPR therefore drive the re-distribution of CPR conformations across the associated energy landscape to more open or closed forms of CPR. These different conformations direct downstream interaction of CPR with CYP partners and facilitate directional transfer of reducing equivalents for CYP-mediated catalysis. Such motion therefore drives the vectorial transfer of electrons from NADPH to CYP to catalyze the wide range of mono-oxygenation reactions in the endoplasmic reticulum.
It is important to distinguish between the relatively large-scale redox-coupled and ligand-coupled motions discussed above and other stochastic motions that can limit the rate of electron transfer. Our model therefore also recognizes that smaller-scale motions can also limit electron transfer, either between flavin cofactors (in the closed state) or to CYP enzymes (in the open state). Localized searches for productive reaction geometries are common in biological electron transfer reactions and these are often responsible for the slower observed rates of electron transfer compared to those predicted for “pure” (nonadiabatic) electron transfers on the basis of distance criteria alone [23],[34]. Indeed, our temperature-dependence studies indicate that the reaction cannot be modeled using the Marcus nonadiabatic formalism for electron transfer (see Text S1 for details; Figure S10 and Table S2). As such, Figure 6 should not be taken to imply that single discrete open and closed conformations of CPR exist under a defined ligand-bound or redox form of the enzyme. Rather, an ensemble of conformations exist (as indicated by PELDOR studies of different liganded forms of di-semiquinoid CPR [12]) and that redox change and ligand binding drive the equilibrium distribution towards more open or closed states, respectively. We suggest that localized searches for reactive electron transfer geometries could be rate limiting for interflavin electron transfer in CPR, consistent with the slow observed rates of electron transfer in t-jump [20] and laser flash photolysis experiments [20],[35].
The use of time-resolved FRET in conjunction with stopped-flow absorption analysis of the CPR catalytic cycle has enabled us to present a dynamic model of catalysis in which redox change and ligand binding drive large-scale redox domain motion. By coupling conformational change to redox change and ligand binding, CPR optimizes internal electron movements between the flavin cofactors and signals “open and ready” conformations to partner CYP P450 enzymes. This linking of motion with enzyme chemistry enables fine control of vectorial electron transfer along the NADPH→FAD→FMN→heme (CYP) chain that supports all P450-mediated catalysis in the microsome. Given the structural similarity of CPR with other major mammalian diflavin oxidoreductases, including the isoforms of nitric oxide synthase, we anticipate CPR will be a prototype for similar coupling of reaction chemistry, ligand binding, and motions in biology.
The human CPR structures were modeled using SwissModel from the rat CPR crystal structures. For the closed and intermediate forms, the sequence was simply fit to the crystal structure (1AMO_A and 3ES9_C, respectively, with 94% and 93% sequence identity). Loops not present in the crystal structures were modeled using the built-in loop database (closed form: residue 235–241 and 499–505; intermediate form: 235–238 and 499–504). For the most open form, a significant portion of the FMN domain is missing in the crystal structure (3ES9_B). This portion was modeled by aligning the FMN domain from the intermediate form onto the existing coordinates of the crystal structure with the loop connecting the two domains modeled using the built-in loop database.
Human CPR was expressed and purified essentially as described previously [31]. Labeling of CPR with extrinsic fluorophores was achieved by incubating CPR in 50 mM potassium phosphate, pH 7 at <20°C in an anaerobic glove box (Belle Technology) with either Alexa 488 C5-maleimide (Molecular Probes) or Cy 5 mono-maleimide (GE Healthcare). To achieve a 1∶1 ratio of Alexa 488 and Cy 5 bound to CPR, incubation was with 1 mM and 5 mM of the fluorophores, respectively. Non-reacted fluorophore was separated from the sample by running through a desalting column equilibrated with 50 mM potassium phosphate, pH 7. Details of mass spectral analysis are given in Supporting Information (Text S1). Unless otherwise stated, CPR was fully oxidized prior to all experiments by adding a few grains of potassium ferricyanide and elution through a desalting column as above. 2-electron reduced CPR was formed by reaction with an equimolar concentration of NADPH (Melford), and elution through a desalting column under anaerobic conditions.
Fluorescence emission spectra were monitored on a Varian Cary Eclipse fluorescence spectrophotometer (Varian Inc., Palo Alto, CA, USA). Multiple wavelength absorbance spectra were monitored on a Varian Cary 50 Bio UV/Vis spectrophotometer. Specific experimental conditions are given in the main text. All experiments were performed in 50 mM potassium phosphate, pH 7. The saturation constant, KS, was extracted by fitting concentration dependence data to a weak binding function (Equation 1):(1)
To prevent oxidase activity of CPR, all kinetic studies were performed under strict anaerobic conditions within a glove box (Belle Technology; <5 ppm O2) using a Hi-Tech Scientific (TgK Scientific, Bradford on Avon, UK) stopped-flow spectrophotometer housed inside the glove box. Spectral changes accompanying flavin reduction and flavin di-semiquinone formation/decay were monitored at 456 nm and 600 nm, respectively. For reduction with sodium dithionite, the same solution of dithionite was used for all experiments within 6 h. Fluorescence emission changes associated with donor emission were monitored using a 550 nm short wave pass optical filter. Fluorescence emission changes associated with acceptor emission were monitored using a 650 nm long wave pass optical filter. For the given reaction conditions, no bleed through of fluorescence was observed. Data were collected over a log timebase (15 decades, 3,000 data points total). Typically 3–5 measurements were taken for each reaction condition. Fitting of reaction traces is described in detail in Supporting Information (Text S1).
|
10.1371/journal.pcbi.1001025 | Identifying a High Fraction of the Human Genome to be under Selective Constraint Using GERP++ | Computational efforts to identify functional elements within genomes leverage comparative sequence information by looking for regions that exhibit evidence of selective constraint. One way of detecting constrained elements is to follow a bottom-up approach by computing constraint scores for individual positions of a multiple alignment and then defining constrained elements as segments of contiguous, highly scoring nucleotide positions. Here we present GERP++, a new tool that uses maximum likelihood evolutionary rate estimation for position-specific scoring and, in contrast to previous bottom-up methods, a novel dynamic programming approach to subsequently define constrained elements. GERP++ evaluates a richer set of candidate element breakpoints and ranks them based on statistical significance, eliminating the need for biased heuristic extension techniques. Using GERP++ we identify over 1.3 million constrained elements spanning over 7% of the human genome. We predict a higher fraction than earlier estimates largely due to the annotation of longer constrained elements, which improves one to one correspondence between predicted elements with known functional sequences. GERP++ is an efficient and effective tool to provide both nucleotide- and element-level constraint scores within deep multiple sequence alignments.
| There are millions of sequences in the human genome that perform essential functions, such as protein-coding exons, noncoding RNAs, and regulatory sequences that control the transcription of genes. However, these functional sequences are embedded in a background of DNA that serves no discernible function. Thus, a major challenge in the field of genomics is the accurate identification of functional sequences in the human genome. One approach to identify functional sequences is to align the genome sequences of many divergent species and search for sequences whose similarity has been maintained during evolution. We have developed GERP++, a software tool that utilizes this “comparative genomics” approach to identify putatively functional sequences. Given a multiple sequence alignment, GERP++ identifies sites under evolutionary constraint, i.e., sites that show fewer substitutions than would be expected to occur during neutral evolution. GERP++ then aggregates these sites into longer, potentially functional sequences called constrained elements. Using GERP++ results in improved resolution of functional sequence elements in the human genome and reveals that a higher proportion of the human genome is under evolutionary constraint (∼7%) than was previously estimated.
| The identification and annotation of all functional elements in the human genome is one of the main goals of contemporary genetics in general, and the ENCODE project in particular [1], [2], [3]. Comparative sequence analysis, enabled by multiple sequence alignments of the human genome to dozens of mammalian species, has become a powerful tool in the pursuit of this goal, as sequence conservation due to negative selection is often a strong signal of biological function. After constructing a multiple sequence alignment, one can quantify evolutionary rates at the level of individual positions and identify segments of the alignment that show significantly elevated levels of conservation.
Several computational methods for constrained element (CE) detection have been developed, with most falling into one of two broad categories: generative model-based approaches, which attempt to explicitly model the quantity and distribution of constraint within an alignment, and bottom-up approaches, which first estimate constraint at individual positions and then look for clusters of highly constrained positions. A widely used generative approach, phastCons [4], uses a phylo-Hidden Markov Model (HMM) to find the most likely parse of the alignment into constrained and neutral hidden states. While HMMs are widely used in modeling biological sequences, they have known drawbacks: transition probabilities imply a specific geometric state duration distribution, which in the context of phastCons means predicted constrained and neutral segment length. This may bias the resulting estimates of element length and total genomic fraction under constraint.
One of the leading bottom-up approaches is GERP [5], which quantifies position-specific constraint in terms of rejected substitutions (RS), the difference between the neutral rate of substitution and the observed rate as estimated by maximum likelihood, and heuristically extends contiguous segments of constrained positions (RS>0) in a BLAST-like [6] manner. However, GERP is computationally slow because its maximum likelihood computation uses the Expectation Maximization algorithm [7] to estimate a new set of branch lengths for each position of the alignment; this step is also undesirable methodologically because it involves estimating k real-valued parameters from k nucleotides of data. Furthermore, the extension heuristic used by GERP (and other bottom-up methods [8]) may induce biases in the length of predicted CEs.
In this work we present GERP++, a novel bottom-up method for constrained element detection that like GERP uses rejected substitutions as a metric of constraint. GERP++ uses a significantly faster and more statistically robust maximum likelihood estimation procedure to compute expected rates of evolution that results in a more than 100-fold reduction in computation time. In addition, we introduce a novel criterion of grouping constrained positions into constrained elements using statistical significance as a guide and assigning p-values to our predictions. We apply a dynamic programming approach to globally predict a set of constrained elements ranked by their p-values and a concomitant false positive rate estimate. Using GERP++ we analyzed an alignment of the human genome and 33 other mammalian species, identifying over 1.3 million constrained elements spanning over 7% of the human genome with high confidence. Compared to previous methods, we predict a larger fraction of the human genome to be contained in constrained elements due to the annotation of many fewer but longer elements, with a very low false positive rate.
Like other bottom-up approaches, the GERP++ algorithm consists of two components: calculation of position-specific constraint scores for each column of a multiple alignment, and subsequent aggregation of neighboring columns into segments that score significantly higher than expected by chance (Fig 1; see Methods for more detailed description). These are largely independent procedures: the GERP++ score for a specific position depends entirely on the nucleotides at that position and not on any global element predictions, while identification of statistically significant high-scoring segments depends only on the additivity of individual position scores and can potentially be used in conjunction with other position-specific scoring metrics.
Constraint intensity at individual alignment positions is quantified in terms of “rejected substitutions” (RS), defined as the number of substitutions expected under neutrality minus the number of substitutions “observed” at the position [5]. Thus, positive scores represent a substitution deficit (which would be expected for sites under selective constraint), while negative scores represent a substitution surplus. To estimate this quantity at each aligned position, GERP++ begins with a pre-defined neutral tree relating the genomes present within the alignment that supplies both the total neutral rate across the entire tree and the relative length of each individual branch. For each alignment column, we estimate a scaling factor, applied uniformly to all branches of the tree, that maximizes the probability of the observed nucleotides in the alignment column. The product of the scaling factor and the neutral rate defines the ‘observed’ rate of evolution at each position.
Then, in the element-finding step, GERP++ uses the position-specific RS scores to generate a set of candidate elements. For each putative element it computes a p-value based on the element's length and score (defined as the sum of RS scores for each position within the element) that represents the probability of observing such an element in the null model. These p-values are used to rank CEs in order of significance and report a set of non-overlapping predictions, starting with the lowest (best) p-value. Rather than applying a fixed cutoff, GERP++ estimates the false positive rate by randomly permuting the input RS-scores and treating any prediction within the shuffled sequence as a false positive, similar to the first version of GERP [1], [5].
We used GERP++ to analyze the TBA alignment of the human genome to 33 other mammalian species (the most distant mammalian species is Platypus) spanning over 3 billion positions with a phylogenetic scope of 5.83 substitutions per neutral site. We identified 1,354,034 constrained elements covering 214,749,502 nucleotides, or approximately 7% of the human genome, with an estimated false positive rate of 0.86% at the nucleotide level (see Methods for details). Compared to a slightly negative background average of −0.125 RS, GERP++ predictions and certain known functional elements display an elevated level of constraint, in excess of 1.7 RS. GERP++ elements range in size from 4 to nearly 2000 bases, with mean length of 158.6 nucleotides. The minimum (4 bases) and maximum lengths (2000 bases) are parameters of the algorithm, and the tail of the length distribution (Fig S2A) suggests that with a more permissive upper bound even longer elements could be identified.
We observe significant variation among entire chromosomes of both average RS score and fraction of positions predicted to belong to constrained elements (Fig 2). The mean constraint level varied from −0.3 to −0.05 RS with the exception of chromosome X, which was the only chromosome with a positive average RS score, just under 0.1 RS. This result is consistent with earlier work [9], which suggested that the X chromosome in rodents has a reduced mutation rate. We also observe substantial fluctuation in the fraction of each chromosome predicted to be inside constrained elements, which varied from 1% of the Y chromosome to 4–9% for other chromosomes. We expect this metric to be low for the Y chromosome because a large portion of the alignments for the Y chromosome are too shallow to perform a rate estimation, but even when adjusting for “effective” chromosome size much of the fluctuation remains (Fig 2B). Surprisingly, despite a low fraction of the Y chromosome being within constrained elements, it does not have a particularly low average RS score, while the X chromosome does not exhibit a high CE fraction despite the positive average RS. In fact, there appears to be at best weak correlation between these two metrics of constraint: since the null model is derived from the actual distribution of RS scores for a given region, any (additive) difference in RS score applied uniformly to every position in the region would not change the p-value of any candidate element (although in practice this would alter the exact boundaries, resulting in a slightly different candidate set). The chromosomal fraction within predicted constrained elements ultimately depends more on the distribution and variance of the scores rather than the mean. Unfortunately, this is impossible to quantify exactly due to confounding factors such as differences in alignment quality and depth.
The only major parameter for GERP++ is a false positive rate cutoff that determines at what point the algorithm should stop generating predictions in order to avoid too many false discoveries. Throughout its execution GERP++ keeps track of the constrained elements predicted so far, as well as estimates of the number and total size of false positive predictions for the specified cutoff level. Examining how these quantities grow as the cutoff parameter increases permits us to estimate the amount of total constraint that can be detected using this methodology and give an approximate upper bound on the amount of constraint within the human genome.
Let B(c) be the number of bases within constrained elements predicted at false positive cutoff c, and let B*(c) = B(c)−F(c) be the same quantity adjusted for false positive predictions by subtracting the estimated number of false positive bases (as found in shuffled alignments) at cutoff c. Fig 3 shows B and B* as a function of c from 0 to 50%: while B continues to increase, B* starts to level off right as B begins to grow linearly. This suggests that maxc B*(c) can be used to estimate the total number of bases in constrained elements that can be annotated using this method in any given region or the entire genome. Approximately 225 megabases, or nearly 7.3% of the human genome can be detected as contained in CEs using GERP++ at the mammalian phylogenetic scope. If we adjust for the portions of the genome where rate estimation was not performed (but with a deeper alignment might be in the future), we estimate that up to 8% of the human genome consists of CEs detectable using this kind of methodology. Combined with the observation that about 190 megabases, or 6.2% can be detected at a false positive cutoff of 0 (Fig 3), we obtain a fairly narrow estimate of 6–8% of the human genome under detectable evolutionary constraint, in the mammalian scope. We note that this estimate depends on alignment quality, since we may fail to pinpoint constrained elements not only due to method-intrinsic limitations but also because an appropriate signal may be absent in a given multiple alignment.
We next examine the relationship between evolutionary constraint and several classes of biologically important regions. Overall, coding exons exhibit by far the strongest levels of constraint, as quantified both by the average RS score within functional elements (Fig 4A), and by fraction of bases that overlap the predicted CEs (see Table 1). Both 5′ and 3′ UTR regions show weaker but noticeable constraint levels and, somewhat surprisingly, introns on average have slightly lower RS scores than the overall genomic baseline. However, a nontrivial fraction of introns does exhibit evidence of constraint, as nearly 7% of intron positions overlap predicted elements (Table 1), and these positions make up a large fraction of constrained element bases (see Fig 4B).
Over 94% of the coding exons in the human genome overlap at least one predicted CE; conversely, only about 16% of constrained elements overlap a coding exon. CEs that overlap exons are on average ∼60 nucleotides or 40% longer, and consequently have more than two-fold higher scores, than elements that do not overlap exons (both t-tests significant at p-value<2.2·10−16). While overall these results are consistent with what was observed using the previous version of GERP [5] on much more limited alignments, the length difference between exon-associated and non-overlapping CEs is somewhat smaller than what was previously found. This is partially explained by the differences in the pattern of constraint between coding exons and other regions. Because the previous GERP by default only merges blocks of contiguous constrained positions if they are separated by at most one unconstrained position [5], it is far more likely to generate longer elements in exonic regions where most unconstrained bases correspond to 3rd positions of a codon and are usually flanked by constrained positions. In noncoding regions where unconstrained positions are distributed more irregularly and often occur consecutively, the previous GERP algorithm [5] ends up fragmenting longer constrained regions and generating shorter elements. Because GERP++ does not base merging decisions on any such fixed threshold it is able to better annotate longer noncoding CEs.
To further test this hypothesis, and to investigate a potentially useful signal for detecting coding exons, we introduce a metric that rigorously quantifies this pattern of constraint for any region. For any given segment, we define the 3-periodicity bias as the maximum over the 3 possible reading frames of the mean RS score at positions 1 and 2 minus the mean RS score at position 3. This metric quantifies a periodic bias in constraint and effectively deals with unknown reading frame location and lack of a reading frame altogether, since the maximum is taken over all 3 possibilities. As Fig 5 shows, the 3-periodicity bias is a strong signal characteristic of coding exons (mean 2.96) compared to other regions such as UTRs, introns, and ncRNAs (mean 0.13–0.38, difference significant at p-value<2.2·10−16). We partitioned the constrained elements predicted by GERP++ according to exon overlap, and found that CEs overlapping coding exons have a much greater mean 3-periodicity bias (Table 2). However, the difference between CEs that did not overlap any annotated exons, and known nonexonic regions such as introns was still significant, suggesting some of these CEs intersect unannotated exonic regions. To test this hypothesis, we checked the constrained elements that did not overlap any known coding exons against exon predictions made by the computational gene prediction tool CONTRAST [10]. We found 16,881 CEs (making up 1.5% of all CEs that did not overlap known genes) that overlapped CONTRAST predictions, and these CEs had a significantly higher 3-periodicity bias (1.33) than those that did not overlap CONTRAST predictions (0.54). As this latter figure is still higher than the average 3-periodicity of clearly non-exonic elements, it is possible that a fraction of these elements overlap unannotated exons or pseudogenes with recently lost function. It is interesting to note that the difference between 3-periodicity bias of GERP++ CEs that overlap known exons (2.46) and CEs that overlap CONTRAST predictions (1.33) is also significant. This is likely a combination of two factors: false positives (or errors in identifying the exact boundary) in CONTRAST predictions, and selection bias that manifests as exons with higher 3-periodicity being more conserved and/or easier to identify, and thus annotated in the UCSC Known Genes set.
We compared the GERP++ constrained element predictions in placental mammals (see Methods) to phastCons [4], the leading generative model-based tool. Not surprisingly, we found significant overlap between GERP++ and phastCons predictions: 80% of GERP++ predictions overlapped at least one phastCons prediction, and vice versa. However, aside from both algorithms detecting clearly constrained areas, there are substantial differences: GERP++ predicts significantly fewer elements, which are much longer on average (see Fig S2B for distribution of phastCons element lengths) and cover a substantially larger portion of the human genome - almost twice as much as the 4% predicted by phastCons (Fig 6A). As a result, on a nucleotide level GERP++ overlaps 90% of phastCons predictions while only half of GERP++ CE positions are covered by phastCons.
Part of the reason for these differences is that often phastCons predicts multiple elements where GERP++ makes one longer prediction. PhastCons thus skips intermediate positions which may be under weaker constraint yet still part of one large functional element, as the example in Fig 6E shows. In order to demonstrate that this is not an isolated occurrence and to quantify fragmentation of known functional elements, we computed the number of distinct predicted constrained elements overlapping each annotated coding exon. While the total number of exons that overlap at least one constrained element prediction is approximately the same between the two methods, GERP++ is significantly more effective at identifying entire exons as a single predicted CE, rather than fragmented between two or more CEs like phastCons (Fig 6C & 6D). This phenomenon is not limited to coding exons, as we observed similar behavior for experimentally identified RNA Polymerase II (PolII) binding sites (see Methods), which correspond to poised or active promoters. GERP++ overlaps a larger fraction of nucleotides within 50 base pairs of a PolII binding site (26% vs 19% for phastCons), and exhibits similarly reduced fragmentation as with coding exons (Fig 7).
Due in part to its ability to annotate larger elements in one piece, GERP++ is more effective at predicting constraint within several types of known functional regions. At the nucleotide level GERP++ elements cover a substantially larger fraction of several major types of functional elements, especially coding exons and UTRs (Fig 6B). The improved resolution in detection of known functional elements suggests GERP++ may also be more effective at predicting unannotated regions that are not only constrained but also functional.
One of the main challenges in constrained element detection is the lack of a clear gold standard for evaluating the quality of predictions. Human functional elements are sometimes unconstrained at the mammalian scope or missed at the assembly or alignment stages, and CE predictions that do not correspond to any known annotations may have unknown function, and cannot be definitively considered false positives. Given these limitations, we have shown that GERP++ offers several advantages over its predecessor GERP and makes fewer assumptions about the shape of conservation than previous approaches such as PhastCons. Previous bottom-up approaches have been limited largely by the simple heuristics used to merge constrained positions into longer elements; these heuristics may introduce biases in element length due to patterned constraint such as the 3-periodicity in coding exons. With GERP++ we evaluate a much richer set of candidate elements, selecting and ranking final predictions according to statistically meaningful p-values.
Despite the added computational cost at this stage, GERP++ overall is more than 100 times faster than GERP due to the speedup in rate estimation. Because GERP++ estimates a single parameter that directly translates into evolutionary rate, rather than an independent parameter for each branch of the tree, the computation is not only faster but also results in more statistically robust estimates as alignment depth increases. GERP++ takes a few days on a typical machine or a few hours on a small cluster to complete an analysis of the human genome aligned to 33 mammalian species, and can scale to virtually any reasonable genome size and alignment depth.
Our understanding of the evolutionary forces constraining sequence variation is still limited, especially in noncoding regions. This presents a challenge for generative model-based approaches, which model implicitly or explicitly the distribution of length and intensity of constrained elements and the total genomic fraction under constraint. In contrast, rate estimation and element prediction in GERP++ are largely independent procedures, and while GERP's rejected substitution metric [5] accurately quantifies constraint intensity at individual positions, any additive position-specific scoring scheme could potentially be used instead. For example, in future implementations of the GERP++ package more elaborate or context-dependent models of nucleotide evolution could be easily incorporated in order to improve position-specific evolutionary rate estimation without drastically changing the overall algorithm.
One drawback of GERP++ and other similar approaches is sensitivity to variation in and erroneous estimates of the neutral rate of substitution. Neutral rate estimates are often subject to some uncertainty and can vary depending on the methodology, alignment quality, and genomic region. To test the ability of GERP++ to tolerate a reasonable amount of error in neutral rate estimates, we repeated our analysis with the neutral tree scaled up or down by 5 or 10%. Not surprisingly, overestimating the neutral rate leads to overprediction of constraint, and vice versa. For a fixed false positive cutoff, we observed a linear relationship between the input neutral rate and the amount of constrained element bases predicted; a 5/10% change in neutral rate leads to approximately 8/15% change in the number of predicted constrained bases.
It is important to note that our false positive rates and p-values are computed based on the implicit assumption that the score distribution is homogeneous within a region and all sites are independent. While this assumption has been present in previous approaches that also relied in permuted alignments for false positive rate estimation, it is central to the GERP++ p-value computation. Finally, the greedy manner of resolving candidate element overlap conflicts by smallest p-value presents another potential limitation, as for elements with equal average constraint this will break ties in favor of the longer element. This may or may not be biologically meaningful, especially if complicated conservation patterns are involved or two strongly conserved functional elements are very close together (and the segment between them is at least somewhat constrained). These hypothetical effects are likely mitigated by GERP++'s position-specific scores, which enable higher resolution analysis within individual CEs, and which ultimately may be the criterion upon which to decide whether any particular long element may better be regarded as two shorter ones.
GERP++ recapitulates known biology, at both the nucleotide level and on the scale of entire functional elements and even chromosomes. GERP++ scores are accurate enough to obtain a strong signal of synonymous substitution in coding exons, and the elevated average RS score for chromosome X (Fig 2A) agrees with earlier findings [2], [3]. Compared to phastCons, GERP++ predictions overlap a larger fraction of known functional elements (Fig 4B) and have greater 1∶1 correspondence to constrained coding exons (Fig 6C & 6D) and promoters (Fig 7). Our analysis has also yielded interesting biological insights, including the likely presence of unannotated coding exons among our predicted constrained elements. We detect around 7% of the human genome to be contained in CEs in the mammalian scope, a slightly larger amount than previous predictions, yet with a lower estimated false positive rate. While this estimate is inexact, our analysis suggests 6% and 8% as reasonable lower and upper bounds, a somewhat tighter range than earlier estimates [1], [2].
Computationally, GERP++ is efficient enough to perform whole-genome analysis of deep mammalian alignments within a few cpu-days, making it suitable for high-throughput analysis of the ever increasing amounts of genomic data. We hope GERP++ will prove to be a useful tool in analyzing, quantifying, and annotating constraint and discovering novel functional elements in the human and other genomes for which sufficient comparative data exist.
GERP++ is available at http://mendel.stanford.edu/SidowLab/downloads/gerp/index.html
Given a multiple sequence alignment and a phylogenetic tree with branch lengths representing the neutral rate between the species within that alignment, GERP++ quantifies constraint intensity at each individual position in terms of rejected substitutions [5], the difference between the neutral rate and the estimated evolutionary rate at the position. For our analysis the alignment was compressed to remove gaps in the reference sequence (human), although the RS score computation algorithm does not assume any specific reference sequence. In order to estimate the evolutionary rate we model nucleotide evolution as a continuous-time Markov process, which specifies for each pair of nucleotides a and b and duration t the probability of a transforming into b over time t, designated by pab(t). Many such evolutionary models have been developed [11], [12], each with its own set of simplifying assumptions. GERP++ implements the HKY85 model [13], but any time-reversible model (where papab(t) = pbpba(t) for all pairs of nucleotides a and b) that permits efficient computation of pab(t) can be used instead.
For each individual alignment column GERP++ labels the leaves of the phylogenetic tree with the corresponding nucleotides c1, …, ck; gapped species are projected out. Although this is not necessarily ideal and sometimes leads to information loss, it avoids some of the common difficulties and potentially serious biases that accompany modeling gaps in alignments: aligner errors and artifacts that result from simplified gap penalties and incorrect handling of duplications and rearrangements, assembly mistakes, and missing sequence data. Furthermore, this treatment of gaps avoids explicitly penalizing constrained elements that have undergone lineage-specific deletion [5].
Once the gapped species are removed, the site-specific neutral rate is computed as the sum of the branch lengths in the trimmed tree. When there are fewer than 3 species remaining no rate estimation is performed for that position, as there are not enough species to even form a valid tree. We estimate by maximum likelihood a homogeneous scaling factor of the neutral tree at each position; similar but independently developed methods were used for rate estimation in [14], [15]. Specifically, we introduce a scaling parameter r that represents the site's rate of evolution relative to neutrality. When r<1 the quantity (1−r) can be naturally interpreted as the fraction of neutral substitutions “rejected” by evolutionary selection. GERP++ estimates r by maximum likelihood, where the likelihood is given by L(r) = Pr(c1, …, ck | Tr), where Tr is the neutral tree T scaled by r. For any given r, and therefore fixed tree Tr, this function can be computed efficiently using a dynamic programming algorithm due to Felsenstein [16]. If n is an internal node with children n1 and n2, and {c1, …, ck}n represents the subset of the leaves corresponding to the subtree rooted at n, thenwhere Tr(x,y) is the branch lengths in Tr between two neighboring nodes x and y.
Since the leaf nucleotides are observed, this equation can be used to compute the subtree probability for all internal nodes, starting at the bottom and reaching the root, where we can compute L(r) = Pr(c1, …, ck | Tr) = Σa Pr({c1, …, ck}n | root = a) pa. Assuming a fixed alphabet and an evolutionary model where the probabilities pab(t) are computable in constant time, this algorithm runs in time O(k) where k is the number of species in the phylogenetic tree.
Using this algorithm as a subroutine to calculate L(r), GERP++ computes the maximum likelihood value of r using Brent's method [17], [18], a numerical optimization technique that tends to require relatively few computations of the function being optimized. The evolutionary rate for a site with neutral rate n is estimated to be rn, and the final RS score is computed as n−rn = n(1−r). As maximum likelihood may estimate very large or even infinite values of r, we impose a cap of r = 3 on GERP++ rate estimates, yielding RS scores that range between −2n and +n. These scores are then used as the basis for prediction of constrained elements within the region.
Given position-specific constraint scores, GERP++ generates a list of elements that exhibit evidence of evolutionary constraint beyond what is likely to occur by chance. For each element, we compute a p-value that represents the probability of a random neutral segment of equal length having an equal or higher RS score. In addition to being used to select final predictions from the set of candidate elements, these p-values in conjunction with position-specific scores provide useful information for biological analysis.
Every segment of contiguous multiple alignment columns is a candidate element. Because considering all possible segments within the alignment is computationally infeasible, GERP++ generates a list of candidate elements using several simple biological heuristics to prune the possibilities. First, we impose a user-specified minimum and maximum on candidate element length; while real functional elements vary in length, very few extend beyond several thousand bases, and even these will not be missed entirely as GERP++ will identify their most constrained parts. Second, since positive RS scores indicate constraint, GERP++ allows only candidate elements that start and end at positions with RS≥0 and cannot be extended further in either direction; this rule has the additional benefit of imposing sensible boundary conditions on predicted elements. Finally, we only consider candidate elements with score above a certain value, which is a function of the element length and the median neutral rate of the region. This allows pruning of candidate elements that have low scores relative to their lengths, and since they will end up with poor p-values anyway ignoring them early reduces the memory requirements considerably.
Using neutrality as the null hypothesis, we can now define p-values for candidate and predicted elements on the basis of score and length. If the probability of a single neutral position having RS score x is given by P(x), then for an element of length L and score S the p-value is the probability of having score at least S in exactly L positions, and is given by:The RS score distribution is irregular (Fig S3) and therefore cannot be easily modeled by common statistical distributions; however, the p-values can be computed using dynamic programming, for L = 1, …, Lmax, provided the distribution P(x) can be computed and the space of possible scores x is not too large. The latter is assured by discretizing to within a specified tolerance t; since individual scores range from −2n to +n, there are 3n/t possible discretized scores. We now build a histogram of these discrete scores from the alignment, with two exceptions. First, we exclude long consecutive runs of “shallow” positions (default at least 10), i.e. positions with neutral rate below specified cutoff (default 0.5 substitutions per site), as there are many such primate-specific regions and they tend to skew the score distribution. Additionally, remaining shallow positions are given a small penalty to discourage GERP++ from predicting CEs consisting mostly of shallow positions. Second, we exclude positions that belong to clearly constrained regions, which are identified using a preliminary pass of the algorithm (with false positive cutoff set to 0). All other scores are used to build a score histogram for each region. In order to eliminate artifacts caused by zero probabilities, we add a small uniform prior to the histogram to ensure every discretized score appears at least once.
Once all candidate elements have been assigned p-values, GERP++ selects elements in a greedy manner, from smallest to highest p-value, discarding any elements that overlap previously reported elements. As the p-value increases so does the expected false positive rate of our predictions; when this reaches a user-specified threshold the algorithm terminates. While it would be ideal to compute this directly from the p-values, the multiple hypothesis correction in this case is non-trivial because GERP++ reports a non-overlapping set of predictions. Therefore, we adopt the approach of Cooper et al [2], [5] and estimate the false positive rate by generating several independent permuted alignments. These alignments are obtained by randomly shuffling columns of the original multiple alignments, excluding long stretches of shallow positions.
TBA [19] alignments of the human genome (hg18) to 43 other vertebrate species were obtained from the UCSC genome browser [20], [21] together with a phylogenetic tree with the generally accepted topology (Fig S1) and neutral branch lengths estimated from 4-fold degenerate sites. Both the tree and alignments were projected to the 34 mammalian species. The alignment was compressed to remove gaps in the human sequence, and GERP++ scores were computed for every position with at least 3 ungapped species present, or approximately 88.9% of the 3.08 billion positions on the 22 autosomes and X/Y chromosomes. We used the HKY85 [13] model of evolution with the transition/transversion ratio set to 2.0 and nucleotide frequencies estimated from the multiple alignment.
To limit memory requirements and allow parallelization of the constrained element computation, each chromosome was broken up into regions of approximately 2 megabases, with long segments where no RS score was computed chosen as boundaries. These boundary segments contain no information usable by GERP++ and because the algorithm never annotates constrained elements spanning them, excluding such segments did not sacrifice any predictive ability. These boundary regions made up approximately 6.8% of the human genome, including a 30.2 megabase region that made up more than half of chromosome Y. Constrained element predictions were generated using default parameters and a 5% false positive cutoff measured in terms of number of predictions; the estimated nucleotide-level false positive rate was under 1%. As additional validation, we computed overlap between our predictions and a set of ancestral repeats (L2) annotated by RepeatMasker. We found the overlap to be in line with what we expected given our estimated false positive rates: about 5% of the repeats overlap a predicted CE, with around 1.6% nucleotide-level overlap.
Gene, noncoding RNA, and PhastCons conserved element annotations were obtained from the UCSC genome browser's [20], [21] Known Genes [22], RNA Genes, and Conservation [4] tracks respectively. To avoid skewed statistics due to alternative splicing, gene annotations were resolved to a consistent nonoverlapping set where any segment belonging to multiple conflicting annotations was assigned a single annotation in the following order of priority: coding exon, 5′ UTR, 3′ UTR, intron. For meaningful comparison against phastCons, separate GERP++ scores and constrained elements were generated according to the same procedure as above but using only placental mammal data (ignoring platypus and opossum in the alignment and projecting them out of the phylogenetic tree).
PolII binding regions were defined as 50 bp upstream and downstream of PolII binding ‘peaks’ as identified from ChIP-seq experiments performed by the ENCODE Consortium [3]. A 100 bp window allows capture of the likely PolII binding site and its flanking sequence. We obtained data from nine ChIP-seq experiments conducted in two labs (the Snyder lab at Yale and the Myers lab at Hudson Alpha) on six cell types. Data was downloaded through the DCC at UCSC (ftp://encodeftp.cse.ucsc.edu). All data have passed publication embargo periods. Overlap statistics were calculated as described above for other annotation sets and averaged across all nine experiments.
|
10.1371/journal.ppat.1007357 | Gut and blood differ in constitutive blocks to HIV transcription, suggesting tissue-specific differences in the mechanisms that govern HIV latency | Latently-infected CD4+ T cells are widely considered to be the major barrier to a cure for HIV. Much of our understanding of HIV latency comes from latency models and blood cells, but most HIV-infected cells reside in lymphoid tissues such as the gut. We hypothesized that tissue-specific environments may impact the mechanisms that govern HIV expression. To assess the degree to which different mechanisms inhibit HIV transcription in the gut and blood, we quantified HIV transcripts suggestive of transcriptional interference (U3-U5; "Read-through"), initiation (TAR), 5' elongation (R-U5-pre-Gag; "Long LTR"), distal transcription (Nef), completion (U3-polyA; "PolyA"), and multiple splicing (Tat-Rev) in matched peripheral blood mononuclear cells (PBMCs) and rectal biopsies, and matched FACS-sorted CD4+ T cells from blood and rectum, from two cohorts of ART-suppressed individuals. Like the PBMCs, rectal biopsies showed low levels of read-through transcripts (median = 23 copies/106 cells) and a gradient of total (679)>elongated(75)>Nef(16)>polyadenylated (11)>multiply-spliced HIV RNAs(<1) [p<0.05 for all], demonstrating blocks to HIV transcriptional elongation, completion, and splicing. Rectal CD4+ T cells showed a similar gradient of total>polyadenylated>multiply-spliced transcripts, but the ratio of total to elongated transcripts was 6-fold lower than in blood CD4+ T cells (P = 0.016), suggesting less of a block to HIV transcriptional elongation in rectal CD4+ T cells. Levels of total transcripts per provirus were significantly lower in rectal biopsies compared to PBMCs (median 3.5 vs. 15.4; P = 0.008) and in sorted CD4+ T cells from rectum compared to blood (median 2.7 vs. 31.8; P = 0.016). The lower levels of HIV transcriptional initiation and of most HIV transcripts per provirus in the rectum suggest that this site may be enriched for latently-infected cells, cells in which latency is maintained by different mechanisms, or cells in a "deeper" state of latency. These are important considerations for designing therapies that aim to disrupt HIV latency in all tissue compartments.
| Available antiretroviral drugs significantly prolong life expectancy and reduce morbidity in people living with HIV. However, HIV can escape host immune responses and drug treatment by establishing a reversibly silent ("latent") infection in CD4+ T cells. This latent infection represents the major barrier to a cure. While much of the research to date has highlighted the importance of peripheral CD4+ T cells as reservoirs for latent HIV, it is becoming increasingly apparent that the gut may play an integral role as a major tissue reservoir for HIV. In this study, we show that the transcriptional blocks that underlie HIV latency in CD4+ T cells differ in the blood and gut. In HIV-infected people on effective treatment, the major blocks to HIV transcription in blood cells occur at transcriptional elongation, distal transcription/polyadenylation (completion), and splicing. In the gut, the major block to HIV transcription occurs at transcriptional initiation, suggesting that HIV latency is maintained by different mechanisms in the gut, which may be enriched for latently-infected cells and/or cells in a "deeper" state of latency. These differences in the blocks to HIV transcription are important to consider in designing therapies that aim to cure HIV.
| The major barrier to a cure for HIV is thought to be latently-infected cells that do not produce HIV constitutively but can be induced to produce infectious virus upon activation [1–3]. The latent HIV reservoir cannot be eliminated using currently available antiretroviral drugs, and due to their long half-lives and ability to proliferate [4], latently-infected cells can persist for many years [5–8]. While an extensive body of research has underscored the importance of peripheral CD4+ T cells as reservoirs for latent HIV, it is becoming increasingly apparent that the gut may play an integral role as a major tissue reservoir for HIV [9]. First, a large proportion of all lymphocytes reside in lymphoid tissue, of which the gut accounts for up to 85 per cent [10]. Second, CD4+ T cells of the gut are likely to be more vulnerable to infection than their peripheral blood counterparts [10]. This increased permissivity to HIV [11, 12] may be due to factors such as elevated levels of activation or CCR5 expression [13–15]. Consequently, the depletion of CD4+ T cells in the gut during acute HIV [16] and SIV [17–21] infection is both more rapid and severe than peripheral blood. Furthermore, this depletion occurs prior to and is more profound than that in the blood or lymph nodes [17, 22]. The disproportionate effect of HIV infection on the gut may result in an increased HIV burden in gastrointestinal tissue. Both HIV DNA and RNA are found to be concentrated in the gut [23, 24] and replication-competent HIV has been recovered from the rectal mucosa [25], suggesting that a proportion of gut CD4+ T cells harbor replication-competent proviruses. Prior data also suggest differences between blood and gut in infected cell types, levels of T cell activation, HIV DNA levels, relationship to activation, and levels of HIV RNA per cell [23, 26], suggesting these tissues differ in the mechanisms that govern HIV transcription and latency.
Using a novel panel of reverse transcription droplet digital polymerase chain reaction (RT-ddPCR) assays that can simultaneously quantify multiple different blocks to HIV transcription, we recently showed that the major reversible blocks to HIV transcription in peripheral CD4+ T cells from ART-suppressed patients are blocks to proximal elongation, distal transcription/polyadenylation (completion), and splicing [27]. We hypothesized that the mechanisms and degrees of HIV transcriptional blocks underlying HIV latency differ between gut and peripheral blood. In this study, we applied our "transcriptional profiling" assays to two cohorts of ART-suppressed individuals to simultaneously assess the mechanisms that govern HIV transcription in the gut and blood. We quantified the levels of different HIV RNAs in PBMCs and intact rectal biopsies (n = 9), as well as sorted CD4+ T cells from peripheral blood and dissociated rectal biopsies (n = 7). The relative levels of the different HIV RNAs suggested blocks to distal HIV transcription, completion, and splicing in all samples, and these observations were not explained by mutations in the corresponding HIV DNA primer/probe sequences or differential RNA stabilities. However, in contrast to our findings in peripheral CD4+ T cells [27], we found a much greater block to HIV transcriptional initiation in the rectum (both biopsies and sorted cells) compared to the blood. These differences in HIV transcriptional blocks, which could reflect tissue-specific differences in viral or cellular factors, are important to consider in designing therapies that aim to eliminate or silence HIV-infected cells.
We used a novel panel of HIV “transcription profiling” assays (Fig 1) to quantify HIV transcripts suggestive of transcriptional interference (U3-U5; "Read-through"), initiation (TAR [Trans-activation Response region]), 5' elongation (R-U5-pre-Gag; "Long LTR” [Long Terminal Repeat]), distal transcription (Nef), completion (U3-polyA; "PolyA"), and multiple splicing (Tat-Rev) in PBMCs and intact rectal biopsies from nine ART-suppressed individuals. In PBMCs, these assays revealed a reproducible gradient in the relative abundance of HIV transcripts (normalized to cell equivalents by ddPCR for Telomere Reverse Transcriptase [TERT]) where total (TAR) > elongated (Long LTR) > distally elongated (Nef) > polyadenylated (PolyA) > multiply-spliced Tat-Rev transcripts (medians: 7289, 420, 108, 44, and 2 copies/106 cells, respectively; Fig 2A). Read-through transcripts (U3-U5), suggestive of transcriptional interference, were also detected in every individual (median 155 copies/106 cells), but were 30-fold lower than total (TAR) transcripts (median [Read-through/TAR ratio] = 0.033). The median level of 5' elongated (Long-LTR) transcripts was 17-fold lower than that of total transcripts (median [Long LTR/TAR] = 0.06), suggesting a block to proximal elongation. The median level of Nef (3') was almost 4-fold lower than that of 5' elongated transcripts (median [Nef/Long LTR] = 0.26), suggesting a block to distal transcription. The median level of multiply-spliced transcripts was 23-fold lower than levels of polyadenylated transcripts (median [MS Tat-Rev/PolyA] = 0.04), in accord with prior data suggesting a reversible block to multiple splicing [27].
A similar trend in relative transcript levels (albeit at lower overall levels) was observed in the rectal biopsies, wherein the relative abundance of HIV transcripts was also: total > elongated > Nef > polyadenylated > multiply-spliced Tat-Rev (medians: 679, 75, 16, 11, and <1 copies/106 cells, respectively; Fig 2B). Just as in the PBMCs, Read-through transcripts were detected in all individuals (median 23 copies/106 cells), but were much lower than that of total transcripts (47-fold lower; median [Read-through/TAR] = 0.021). The median level of elongated transcripts was 9-fold lower than that of total transcripts (median [Long LTR/TAR] = 0.11). The median level of distally-elongated (Nef) transcripts was 4-fold lower than that of 5’ elongated transcripts (median [Nef/Long LTR] = 0.21), and the level of polyadenylated transcripts was nearly 2-fold lower than that of distally-elongated transcripts (median [PolyA/Nef] = 0.66]). A block to multiple-splicing is also likely (PolyA > MS Tat-Rev), given that polyadenylated HIV transcripts were detected in rectal biopsies from 6 of 9 individuals, while MS Tat-Rev transcripts were detected in biopsies from only 2 of 9 individuals (vs. 5 of 9 from PBMCs; Fig 2A and 2B). These data suggest that in both PBMCs and rectal biopsies, HIV transcription is blocked at the stages of elongation, distal transcription/polyadenylation (completion), and splicing.
To address the possibility that proviral deletions or hypermutations in primer/probe regions could account for the varying levels of HIV transcripts, we quantified the levels of U3-U5 ("Read-through"), TAR, Long LTR and Nef in DNA extracted in parallel with the RNA from the same PBMCs and rectal biopsies (n = 9 individuals) using the same primers/probes and ddPCR conditions used to measure each HIV RNA (S1 Fig). Comparisons between DNA from the same tissue revealed no differences in the levels of TAR and Read-through regions (both present at 2 copies in an intact provirus) and these levels were ≥2-fold greater than long LTR DNA (1 copy per intact provirus) in rectal biopsies and PBMCs (median TAR/Long LTR = 1.99 and 3.49, respectively; S1 Fig). Levels of Nef DNA (1 copy per intact provirus) were similar to Long LTR DNA and tended to be lower than both Read-through and TAR DNA for both tissues (P<0.05 for all comparisons).
Next, we measured the ratio of each HIV RNA to the corresponding HIV DNA sequence region quantified using the same ddPCR assay (Fig 2C) and normalized to 106 cells in the same manner. This measure expresses the average level of transcription per provirus and is independent of normalization to cell numbers. For both PBMCs and intact biopsies, the gradient pattern in the levels of successive HIV transcripts was preserved after normalization of each HIV RNA region to the corresponding HIV DNA region, suggesting that the differences are unlikely to be due to proviral deletions or hypermutations in primer/probe regions.
The average level per provirus of each HIV transcript was quantified using two approaches. First, we expressed the ratio of each HIV transcript to HIV DNA measured using the same primers/probe (Fig 2C), which revealed lower levels of Read-through, TAR and Nef transcripts per provirus in intact rectal biopsies compared to PBMCs (P<0.05 for all). As an orthogonal method, we expressed the ratio of each HIV transcript to DNA measured using the Long LTR assay alone, which is present in only one copy per intact provirus. This analysis revealed the same trend for all transcripts (Fig 2D; P<0.05 for all). Together, these data strongly suggest lower levels of HIV transcriptional initiation and distal transcription, in addition to lower levels of transcriptional interference, in the rectal biopsies relative to PBMCs.
To determine whether sequence-specific differences in RNA stability/degradation contribute to the divergent levels of HIV transcripts detected, we measured the RNA decay rate of each transcript in peripheral CD4+ T cells isolated from an ART-suppressed patient in the presence or absence of the RNA Pol II inhibitors Triptolide (Fig 3A) or Actinomycin D (Fig 3B). In the absence of RNA Pol II inhibitors, levels of each HIV transcript over 16 hours (h) remained relatively stable (S2 Fig). In contrast, the presence of either Triptolide (100nM) or Actinomycin D (5 mg/mL) resulted in decay of all HIV transcripts over 16h, irrespective of normalization (Fig 3, S1 Table). The half-lives of TAR- and Long LTR-containing transcripts were similar (in the order of ~3–5 hours) irrespective of treatment (Triptolide or Actinomycin D) and generally similar to the half-lives of read-through, Nef, and PolyA transcripts in the presence of Triptolide (4.58, 2.80 and 2.29h, respectively), although the latter three transcripts had shorter half lives in the presence of Actinomycin D (2.54, 1.66, and 2.34h, respectively). The half-life of MS Tat-Rev transcripts was longer with Triptolide (6.25h) and Actinomycin D (5.49h) treatments. The relatively short, similar half-lives of most HIV transcripts (S1 Table) suggest rapid decay in vivo, and any differences do not readily account for the measured differences in levels of the various HIV RNAs.
Since PBMCs and gut biopsies contain different mixtures of T and non-T cells, we also compared CD4+ T cells (defined as CD3+CD8-) sorted from blood and dissociated rectal biopsies from a different cohort of seven ART-suppressed patients. HIV DNA (as measured by the Long LTR assay) was higher in CD4+ T cells from the rectum (10,736 copies/106 CD4+ T cells) than blood (3,841 copies/106 CD4+ T cells; P = 0.016; Fig 4A). HIV DNA can exist in non-integrated or episomal forms, such as 2-LTR circles, which have been interpreted as either labile markers of recent infection or stable forms that decrease only with cell division [28, 29]. Therefore, we also measured the levels of 2-LTR circles using a new assay (S3 Fig), as well as the levels of 2-LTR circles relative to total HIV DNA, in the blood and rectal CD4+ T cells by ddPCR. 2-LTR circles were detected in both blood and rectal CD4+ T cells from the same 5 of 7 individuals, and in 4 of these individuals, levels of 2-LTR circles/106 CD4+ T cells (normalized by TERT) were higher in the rectum (Fig 4A). However, the ratio of 2-LTR HIV DNA to total HIV DNA did not appear to differ in CD4+ T cells from rectum and blood.
In order to measure blocks to HIV transcription in the sorted CD4+ T cells, levels of read-through, total, elongated, completed, and multiply-spliced Tat-Rev transcripts were measured by RT-ddPCR (S4 Fig). Since HIV DNA levels differed in the CD4+ T cells from blood and rectum, levels of each HIV RNA were divided by the HIV DNA (measured using the Long LTR assay) to express the average transcription per provirus (Fig 4B). In CD4+ T cells from both sites, we observed low average levels per provirus of read-though transcripts compared to total transcripts, suggesting little transcriptional interference, and a gradient where elongated > polyadenylated > multiply-spliced transcripts, suggesting blocks to distal transcription and splicing. For the four individuals for whom ileal CD4+ T cells were available, we observed a similar trend (S5 Fig). As in the rectal biopsies, levels of total (TAR) transcripts per provirus were much lower (median 12-fold) in rectal CD4+ T cells compared to peripheral CD4+ T cells (Fig 4B; P = 0.016), suggesting less initiation of HIV transcription in the rectum. Unlike the intact rectal biopsies, PBMCs, or blood CD4+ T cells, we observed no difference between levels of elongated (Long LTR) and total (TAR) transcripts in the rectal CD4+ T cells (Fig 4B). With the caveat that isolation of gut CD4+ T cells requires additional processing, these data suggest little or no block to elongation in the sorted rectal CD4+ T cells.
Ratios of one HIV transcript to another are independent of normalization to cell numbers and can be used to measure the presence and degree of different HIV transcriptional blocks [27]. We did not find a difference between PBMCs and rectal biopsies in the proportion of HIV transcripts that are read-through (read-through/total) or in the proportion of HIV transcripts that proceed through blocks to elongation (elongated/total), completion (polyadenylated/elongated), or multiple splicing (MS Tat-Rev/polyadenylated). In contrast, the sorted CD4+ T cells from blood and rectum showed a 6-fold difference in the proportion of HIV transcripts blocked at the stage of elongation (P = 0.016; Fig 4C), with little block to elongation in the rectal CD4+ T cells (median total/elongated = 1.31) and a strong block to elongation in CD4+ T cells from the blood (median total/elongated = 7.50). These data suggest that a block to HIV transcriptional initiation plays a greater role in inhibiting virus expression in the rectal CD4+ T cells, whereas a block to elongation likely plays a bigger role in CD4+ T cells from the blood.
Blocks to HIV transcriptional initiation could be due to transcriptional interference caused by transcription from neighboring cellular genes that perturbs assembly of preinitiation complexes at the 5’ HIV LTR [30]. To assess the likely contribution of transcriptional interference to the decreased HIV transcriptional initiation in rectal CD4+ T cells, we measured levels of read-through transcripts in relation to total and elongated HIV transcripts. A trend toward lower read-through/elongated transcripts was observed in sorted CD4+ T cells from rectum compared to blood (0.03 vs. 0.08, respectively, P = 0.078; Fig 4D), suggesting less transcriptional interference in the rectal CD4+ T cells. Furthermore, the levels of read-through transcripts tended to be low compared to total transcripts in all samples from both tissues. These findings suggest that transcriptional interference plays a relatively modest role in inhibiting HIV transcription in both sites.
It is conceivable that the process of dissociating gut tissue using collagenase may degrade HIV RNA, which could contribute to lower levels of HIV RNAs in the sorted gut CD4+ T cells but not the intact biopsies. Alternatively, the tissue processing could induce HIV transcriptional elongation, which could contribute to the lower block to elongation in sorted gut CD4+ T cells. To address these concerns, we treated PBMCs from an HIV-infected individual using the same protocol employed for dissociation of gut biopsies. HIV RNA transcripts were measured in PBMCs that were untreated, FACS-stained only, or collagenase treated and FACS-stained (S6 Fig). Interestingly, we found that the FACS-staining procedure necessary to sort live cells of interest may itself cause increases in HIV transcription. In PBMCs that were not treated with collagenase but were FACS-stained, an increase was observed in all HIV transcripts in relation to untreated PBMCs (S6 Fig). FACS-staining alone resulted in a 6.6- and 4.4-fold increase in MS Tat-Rev relative to untreated cells when normalized by RNA mass and TERT, respectively. In contrast, the combination of collagenase treatment and FACS-staining resulted only in an ~2.5 fold increase in MS Tat-Rev relative to untreated cells. The effect of collagenase without FACS staining was also determined in PBMCs from a second HIV-infected individual (S6 Fig). Collagenase treatment alone increased levels of all HIV RNA transcripts, although the change was less than two-fold for each transcript irrespective of normalization (S6 Fig). Collagenase treatment did not alter the pattern of differences between various HIV transcripts. Given that the flash-frozen biopsies that were not subject to collagenase treatment also demonstrated lower levels of HIV transcription, these data support our findings that RNA stability or processing disparities do not account for differences in HIV transcription between tissues.
The reversible lack of virus expression in latently-infected cells is widely considered to be the main barrier to cure of HIV, but it is unclear what mechanisms inhibit constitutive virus expression in lymphoid tissues such as the gut, where most infected cells reside, or whether these mechanisms differ from HIV-infected cells in the blood. In this study, we employed our "transcriptional profiling" [27, 31] approach to measure the degree to which different mechanisms inhibit HIV transcription (and hence virus expression) in gut and blood within HIV+ individuals on suppressive ART.
We measured levels of six HIV transcripts in blood and rectum from two different cohorts of ART-suppressed patients. Distinct types of samples were chosen to minimize sample processing that could affect transcription (PBMCs and flash frozen biopsies) or to facilitate comparison between the same cell type (sorted CD4+ T cells from blood and rectum). In both blood samples and the rectal biopsies, we observed a reproducible pattern in the abundance of HIV transcripts that declined in successive increments (TAR>Long LTR> Nef> PolyA > MS Tat-Rev). These findings, which accord with our previous findings in peripheral CD4+ T cells from ART-suppressed individuals [27], suggest constitutive blocks to HIV transcriptional elongation (TAR > long LTR), distal HIV transcription (Long LTR > Nef), completion (Long LTR > PolyA), and multiple splicing (PolyA > MS Tat-Rev). In contrast to these results from the blood and even the biopsies, we found little evidence for a block to elongation in the rectal CD4+ T cells. Moreover, the average levels of total (TAR) transcripts per provirus were much lower in both types of rectal samples than the corresponding blood samples, suggesting a much greater block to HIV transcriptional initiation in the rectum.
As with any study that attempts to quantify levels of RNA or DNA, differing assay efficiencies can greatly influence the levels of RNA or DNA detected and thus interpretations of these data. Our transcription profiling approach utilizes special methods to minimize bias towards any one sequence region, and we have previously measured the performance characteristics of each assay [27, 31]. Given that all our HIV assays demonstrate similar efficiencies [27], it is unlikely that differing assay characteristics account for the marked differences between levels of the various HIV RNAs.
Internal deletions and hypermutations are present in a substantial proportion of proviral sequences [32, 33] and could cause sequence mismatches with primers or probes, which could impair detection of some HIV transcripts [33]. To assess the impact of proviral sequence, we measured the levels of each sequence region (except Tat-Rev) in the DNA from the PBMCs and rectal biopsies, and we also normalized levels of each RNA to levels of the corresponding DNA measured using the same assays employed for the HIV RNA and normalized to 106 cells using the same method (Fig 2C). The gradient in levels of different HIV RNAs was preserved even after normalizing to the corresponding DNA, and was similar when all RNAs were normalized to the same Long LTR DNA region (Fig 2D), underscoring that these differences in RNA levels cannot be attributed solely to mutations in the corresponding DNA sequence regions. We were unable to quantify HIV DNA regions other than the Long LTR region in the sorted CD4+ T cells, for which the detection of 2-LTR circles consumed a large proportion of the DNA, but it seems unlikely that these would differ much from the results in PBMCs and gut biopsies.
The steady state level of each HIV RNA likely reflects a balance between production (transcription) and destruction (degradation). To determine whether sequence-specific differences in RNA stability could contribute to differences in levels of the HIV RNAs, we measured the decay of Read-through, TAR, Long LTR, Nef, PolyA, and MS Tat-Rev transcripts in CD4+ T cells from an ART-suppressed individual using two RNA Pol II inhibitors, Triptolide and Actinomycin D. In Triptolide-treated cells, the half-lives of Nef (2.80h), PolyA (2.29h), and Read-through (4.58h) did not vary considerably from TAR and Long LTR (3.13h and 2.86h, respectively). MS Tat-Rev seemed to be more stable than the other HIV transcripts after treatment with Triptolide (6.25h) and Actinomycin D (5.49h). However, the very low levels of MS Tat-Rev cause considerable imprecision in calculating its half-life, and even if MS Tat-Rev transcripts were more stable, this would not explain the very low levels of MS Tat-Rev relative to other HIV RNAs.
The half-lives determined from decay of HIV transcripts in cells treated with Triptolide tended to be higher than those determined from cells treated with Actinomycin D. Triptolide acts by inducing proteasome-dependent degradation of RNA Pol II [34], whereas Actinomycin D is thought to intercalate into DNA to sterically-inhibit RNA Pol II [35]. Potential discrepancies in the half lives measured with Triptolide and Actinomycin could be due to the different mechanisms by which these agents inhibit RNA Pol II, incomplete arrest of de novo transcription, or imprecision due to the limited number of time points. For both Actinomycin and Triptolide, the most striking observation is that the half-lives of less abundant HIV transcripts, such as Read-through, Nef, PolyA, and MS Tat-Rev, are comparable to those transcripts that are detected at much higher levels (TAR and Long LTR). These data strongly argue that differences in RNA stability alone do not explain the differential abundance of these HIV transcripts. Although we did not have sufficient numbers of rectal CD4+ T cells to assess the stability of these transcripts in the rectum, and cellular factors may contribute to differences between cell or tissue types, it seems unlikely that there are major sequence-dependent differences between the various HIV transcripts.
Surprisingly, these data also demonstrate that the half-lives of all HIV transcripts tested were relatively short (<7 hours for all except MS Tat-Rev). The short half-life of the TAR transcripts and the average levels of TAR RNA per provirus (>2) suggest a dynamic transcriptional environment with multiple rounds of HIV transcription initiation per day per provirus in the PBMCs, in contrast to the prevailing model of transcriptionally-silent proviruses in ART-suppressed individuals. Ultimately, however, HIV transcripts may be maintained at low levels, despite their active transcription, because of rapid RNA turnover rates, particularly in activated T cells [36]. Furthermore, blocks to elongation and splicing could lead to low expression of HIV Rev, which is critical in circumventing the degradation of unspliced viral transcripts containing introns and AU-rich sequences that contribute to instability [37, 38]. The presence of instability elements within gag also contribute to HIV-1 unspliced mRNA instability [39]. The expression of Gag, Pol, Vif, Vpr, Vpu, and Env proteins from unspliced and partially spliced human immunodeficiency virus type 1 (HIV-1) mRNAs depends on Rev protein, and intron-containing HIV-1 transcripts undergo nuclear downregulation as they are further spliced or degraded in the absence of Rev [39, 40]. This finding may contribute to the short half-lives of unspliced or partially spliced transcripts ex vivo, even those that have been polyadenylated.
Our data from collagenase-treated PBMCs provide evidence that collagenase treatment of gut cells is unlikely to promote HIV transcript degradation or selectively alter levels of a particular transcript. Collectively, our RNA stability and collagenase treatment data suggest that the differential expression of HIV transcripts in the gut and blood are not explained by either intrinsic or treatment-mediated changes to HIV RNA stabilities.
Although the PBMCs and rectal biopsies differ in cell composition, direct comparison between HIV transcription in these samples was possible using the ratio of HIV RNA to HIV DNA, which yields a measure of average transcription per provirus. In intact rectal biopsies, the average levels per provirus of Read-through, total, elongated, Nef, polyadenylated, and multiply-spliced Tat-Rev transcripts were all lower compared to PBMCs, supporting previous work that reported lower levels of HIV transcription in the rectum [23, 26]. The sorted CD4+ T cells showed largely congruent results, where the average levels per provirus of Read-through, total, and multiply-spliced Tat-Rev transcripts were all lower in the rectal CD4+ T cells than the blood CD4+ T cells. One notable exception was the average level per provirus of elongated transcripts, which did not appear to differ between rectal and peripheral CD4+ T cells. Moreover, the TAR/Long LTR ratios suggest a considerable block to elongation in the peripheral CD4+ T cells (TAR/Long LTR = 7.5) but little block to elongation in the rectal CD4+ T cells (TAR/Long LTR = 1.31).
This lack of a block to elongation in the rectal CD4+ T cells seems to disagree with the results from the intact biopsies, where we observed a larger block to elongation. Aside from differences in the patient cohorts, it is likely that the rectal biopsies contain a different mix of infected cell types, including non-T cells. Alternatively, it is possible that the tissue processing (collagenase digestion and shearing) used to dissociate the rectal biopsies into single cells could change cellular or viral transcription in ways that selectively induce elongation. The control experiments in PBMCs suggest that collagenase and FACS staining may cause a small increase in all HIV transcripts, likely reflecting an increase in initiation rather than a specific effect on elongation, although it is possible that these effects of tissue processing differ in adherent and nonadherent cells. However, other findings in the sorted gut CD4+ T cells accord with those in the flash-frozen gut biopsies (which were not subject to such processing) and suggest that blocks to HIV transcriptional initiation (low TAR RNA per provirus) and blocks to later stages of HIV transcription (distal transcription, completion, and splicing) may be important for HIV latency in the gut.
Our major finding that HIV transcription initiation in the rectal CD4+ T cells is 12-fold lower than that observed in blood CD4+ T cells (Fig 4B) provides evidence that differing transcriptional blocks operate in different tissues (Fig 5). These findings are particularly striking given that the gut harbors a much higher proportion of activated CD4+ T cells, and that T cell activation usually stimulates HIV transcription and reverses latency [10]. It is not clear what mediates the 12-fold greater block to initiation of HIV transcription in the rectum. Blocks to HIV transcriptional initiation have been attributed to integration into heterochromatin, epigenetic modification, transcriptional interference, lack of host transcription initiation factors, or insufficient activity of the viral transcription factor Tat [41–44]. It seems very unlikely that the greater block to HIV transcriptional initiation in the gut is driven by higher levels of transcriptional interference from neighboring cellular genes, since levels of Read-through transcripts per provirus tended to be lower in the gut than blood, the ratio of Read-through to elongated transcripts tended to be lower in rectal CD4+ T cells than the blood CD4+ T cells, and T cell activation is supposed to reverse transcriptional interference.
The differences between the blood and gut in the blocks to HIV transcription could be attributable to either the characteristics of the proviral sequences found in each site or the prevailing host cellular conditions and/or cellular environment. It is possible that proviruses in the rectum are more likely to be integrated into transcriptionally-silent regions, although most studies from the blood suggest that HIV is more likely to be integrated in actively-transcribed genes [45, 46], and the higher levels of T cell activation in the gut might also correlate with more actively transcribed genes. It is also possible that proviruses in the gut have more or different mutations that could affect transcription, such as those in the LTRs, Tat, or Rev [47–49]. Compartmentalization of HIV could occur in anatomic sites or tissues where viral trafficking may be impaired or restricted, such as the brain, central nervous system and genital tract [50–52], but prior studies disagree on whether there is compartmentalization of HIV in the gut [47–49]. The HIV DNA levels from the rectal biopsies suggest that many proviruses in the rectum have 2 full LTRs (containing U3-U5 and TAR) for every R-U5/pre-Gag and Nef region, but future studies will be needed to determine whether HIV integration sites or full-length proviral sequences differ between blood and gut.
Epigenetic modifications, host cell factors, and extracellular milieu at either site may also dictate the basal transcriptional activity and the consequences for silencing integrated HIV [30, 45, 46, 53–57]. Previous reports have shown that CD4+ T cell tolerance and anergy can be mediated by epigenetic modification [58–60]. Given that epigenetic modification (for example, of the LTR) can also inhibit HIV transcription [56], it is possible that the unique environment of the gut favors both induction of CD4+ T cell tolerance and HIV latency through epigenetic modification. The gut and blood also show differences in the phenotypes of HIV-infected and uninfected cells, which likely differ in the cellular factors that govern HIV transcription. Naïve and central memory T helper (TCM) cells account for the largest proportion of T lymphocytes in peripheral blood, while effector memory (TEM) and transitional memory (TTM) constitute the predominant populations in the gut [10]. Most HIV DNA in the blood is found in central and transitional memory CD4+ T cells [4], whereas in the gut, most HIV DNA and RNA are found in effector memory CD4+ T cells [26]. The expansive surface area of the GI mucosa is under constant exposure to diverse microbial and food antigens, resulting in sustained immune activation [10]. A higher proportion of gut CD4+ T cells express HIV coreceptors and markers of T cell activation, which may enhance their susceptibility to infection or depletion during acute infection [10]. The gut is also enriched for tissue resident memory cells and different subsets of helper T cells, such TFH [61–64], TH17, TH1/TH17 and TH22, which may serve as reservoirs for latent HIV [10]. In addition, a higher proportion of gut CD4+ T cells express immune checkpoint blockers such as PD-1 and CTLA-4 [65, 66], which have been associated with HIV latency [66]. Non-T cells such as macrophages or dendritic cells might also constitute a larger portion of the infected cells in the gut [26]. These differences in infected cell types likely contribute to the discordant progression through various transcriptional blocks observed in these two tissues.
It is not clear which cellular or viral factors might explain the differences between HIV transcriptional blocks in the blood and gut. A multitude of cellular factors reportedly interact with the HIV LTR to control transcription, and a regulatory feedback mechanism mediated by Tat and Rev drives HIV transcription through its distinct phases [67, 68]. MS Tat-Rev was more frequently detected in the blood than the gut, resulting in a lower median level per provirus in the gut, but MS Tat-Rev may have been harder to detect in the gut because of the lower frequency of CD4+ T cells in the rectal biopsies and the lower yield of sorted CD4+ T cells from the rectum. If the lower block to elongation in the rectal CD4+ T cells is not the result of induction of elongation during collagenase digestion, it could reflect higher levels of P-TEFb or lower levels of NELF in the gut CD4+ T cells, which are more likely to be activated. At the same time, activation should also increase cellular factors that increase HIV transcriptional initiation, such as NFAT and NF-κB, but HIV transcriptional initiation was lower in the rectum, even in the flash frozen biopsies. It is possible that "activation" markers have different meanings or that T cell activation results in different changes in the gut, where immune cells are exposed to more microbial products but mechanisms are needed to maintain tolerance to the normal flora.
In both blood and rectum, we found evidence suggesting blocks to HIV transcriptional completion and multiple splicing. The block to completion could represent incomplete processivity of the RNA polymerase II, which could be modulated by protein levels or post-translational modifications of the enzyme itself, cellular co-factors that affect transcriptional processivity, or secondary structure in the HIV DNA or RNA. In addition, the block to completion could reflect viral and cellular factors involved in end processing of HIV transcripts, including Vpr [69], CDK11 [70, 71], and members of the polyadenylation complex. We have previously shown that the block to multiple splicing in blood CD4+ T cells is partially reversed by T cell activation, which could change levels of cellular factors involved in splicing, such as spliceosome components, SR proteins, MATR3, and PSF [67]. However, a difference between polyA and Tat-Rev remained even after activation, which could reflect intrinsic sequence-dependent factors that inhibit HIV splicing (including multiple inefficient splice donor and acceptor sites, intronic and exonic splice silencers, secondary structure, etc. [67]) as well as cell-specific differences in activation, proviral mutations affecting Tat-Rev or splice sites [32, 33], and the effect of Rev to export unspliced HIV RNA [67].
The blocks to splicing found in both gut and blood suggest that latency could be governed, in part, by post-transcriptional mechanisms. This assertion is supported by the observation that the various latency-reversing agents tested to date fail to completely eradicate infected cells despite inducing viral transcription, albeit to varying degrees [27, 72]. Post-transcriptional blocks to HIV expression, such as blocks to nuclear export [73], RNA interference [74–76], and inefficient translation [77], have been observed in latency models and patient cells. In a primary cell model of latency, levels of intracellular Gag protein were found to be markedly low despite high levels of gag RNA [77], and in resting CD4+ T cells from ART-suppressed individuals, both partially- and fully-spliced HIV transcripts were retained in the nucleus [73], alluding to a block to nuclear export of HIV RNAs. Polyadenylation is important for nuclear export [38] as well as translation, so the block to completion could contribute to lack of export or translation. Because Rev protein facilitates export of unspliced and incompletely-spliced HIV transcripts from the nucleus [67], the block to multiple splicing could also contribute to very low levels of Rev and therefore blocks to nuclear export of unspliced and incompletely-spliced HIV transcripts. Our methods do not allow us to detect other post-transcriptional blocks, which are possible but would have to be in addition to the blocks described here.
Additional studies are needed to determine what cellular factors govern the different blocks to HIV transcription and how cellular gene expression differs between infected and uninfected cells in the blood and gut. Although a limited number of gut biopsies and cells can be obtained from endoscopic procedures, gut cells from HIV-infected patients could be tested ex vivo for their responses to T cell activation or drugs known to act on distinct cellular factors, and gut biopsies could be obtained during clinical trials with interventions designed to disrupt latency. Other techniques of gut cell isolation, such as those that require either no or alternative collagenases [78, 79], could be explored in an effort to minimize tissue processing and potentially increase dissociated cell recovery. Even if cell recovery is limited, single cell transcriptomic and/or proteomic studies could also be used to investigate the cellular factors that are associated with different HIV transcripts in blood and gut cells from patients. Moreover, larger cell numbers could be obtained from unused tissue left over from surgical resections in either HIV+ or HIV- patients. Future mechanistic studies could also employ primary cell latency models derived from lymphoid tissue, such as ‘human lymphocyte aggregate culture’ [80] or ‘lamina propria aggregate culture’ [81] systems, which might achieve higher numbers of HIV-infected cells.
In addition to the points addressed above, other limitations of this study should be noted. One limitation is the relatively small number of ART-suppressed individuals in each of the two cohorts from whom we had samples from blood and gut. Nonetheless, given the statistically significant findings from two different cohorts with two distinct types of gut samples, and findings that largely concur with previous work [23, 27], it is likely that there are differences between blood and gut in the molecular mechanisms that constitutively block HIV transcription. A second limitation is that we did not address the degree to which these blocks are reversible after T cell activation. However, we have previously demonstrated the reversibility of the blocks to HIV transcriptional elongation, completion, and splicing in blood CD4+ T cells [27], so we also expect some reversibility in gut cells, although the magnitude may differ. Finally, although our transcriptional profiling technique enables us to simultaneously assess multiple blocks to transcription, these assays are unable to distinguish which of these transcripts originate from replication-competent proviruses. Given the many cellular factors that can influence HIV transcription independently of viral fitness, the large proportion of the HIV genome in which mutations could eliminate infectivity without affecting transcription (or, conversely, could prevent any transcription), and prior evidence suggesting these blocks to HIV transcription operate in most infected CD4+ T cells from the blood, it seems likely that the same mechanisms operate in many cells with infectious proviruses. However, this question is extremely important and should be addressed in future studies, although it is an exceedingly challenging question to answer using patient cells and will require novel single cell approaches.
Findings from this study have important implications. The large disparities between levels of different HIV RNAs in both blood and gut highlight the importance of critically evaluating the regions targeted when quantifying HIV RNA in different tissue compartments. This fact is particularly important when designing studies to evaluate the effectiveness of interventions designed to reverse latency, since putative "latency reversing" agents (LRAs) exert differential effects on the various transcriptional blocks in blood CD4+ T cells [27], and the same is likely true in other tissues. For such studies, a multi-target approach, such as the transcriptional profiling technique applied in this study, might yield greater insight than quantifying unspliced HIV RNA alone.
The lower levels of HIV transcriptional initiation and of most HIV transcripts per provirus in the rectum suggest that this site may be enriched for latently-infected cells, cells in a "deeper" state of latency, or cells in which latency is maintained by different mechanisms. Future studies are needed to determine whether the same finding holds true in other gut regions or tissues, and whether these tissues differ from blood in the proportion of cells that contain infectious proviruses or can be induced by activation to produce infectious viruses. Since infected cells in the rectum make less HIV RNA (and likely less HIV protein) [23], they may be less likely to trigger cell-intrinsic defense mechanisms or extrinsic immune responses that are designed to recognize and kill infected cells. This may facilitate survival of infected cells in the rectum and might contribute to the higher levels of HIV DNA per million CD4+ T cells in the gut. For the same reason, HIV-infected cells in the rectum might be less susceptible to killing by immune-based therapies that require HIV protein or antigen expression, such as broadly-neutralizing antibodies to Env, immunomodulators, CAR T cells, and vaccines designed to elicit B or T cell responses.
Given the stronger block to HIV transcriptional initiation, as well as persistent blocks to later stages of HIV transcription (completion, splicing) despite higher levels of T cell activation, infected cells in the rectum may also be less susceptible to agents designed to reverse latency or may require different LRAs or combinations. Direct evidence for this idea comes from a recent multi-dose trial of vorinostat in HIV-infected patients, which showed that the median increase in cell-associated HIV RNA in the rectum was 5-fold less than in the blood [82]. Future studies should investigate how gut cells differ from blood in the response to different LRAs, and whenever feasible, the gut should probably be sampled in clinical trials designed to evaluate new therapies aimed at HIV cure. Additional studies should investigate how cellular gene expression differs between CD4+ T cells in the gut and blood, since differentially-expressed genes might suggest new proteins or pathways involved in suppressing HIV transcription. As an alternative to latency reversal, some recent studies have investigated therapies designed to create a "deeper" latency and prevent reactivation from latency [83–85]. Infected gut cells that are activated in vivo but do not transcribe RNA could serve as a model for therapies designed to silence HIV-infected cells. In these ways, an improved understanding of the mechanisms that govern HIV transcription/latency in the gut and blood could help inform new therapies aimed at HIV cure, functional cure, or reducing HIV-associated immune activation and organ damage.
The study was approved by the Committee on Human Research (CHR), the Institutional Review Board for the University of California, San Francisco (approval #11–07551). All study participants provided written informed consent.
The study participants were HIV-infected adults on suppressive ART from two cohorts (median age = 51; median CD4 count = 611 cells/mm3; median years of suppression = 5). Matched rectosigmoid biopsies and cryopreserved PBMCs were obtained from study participants (Table 1) in the Reservoirs and Drug Levels (RADL) study. This study was a prospective, randomized study designed to measure HIV levels and antiretroviral drug (ARV) levels after suppression of plasma viremia with ART regimens containing 2 nucleotide inhibitors of reverse transcriptase and either an integrase inhibitor or a protease inhibitor. Inclusion criteria included: confirmed HIV-1 infection, ART suppression for ≥12 months on initial ART regimen, plasma HIV-1 RNA <40 copies/mL, and a willingness to undergo rectal biopsy. Matched blood and gut biopsies were obtained and aliquoted for parallel measurement of ARV levels and HIV levels. Six rectosigmoid biopsies (flash frozen) and 107 cryopreserved PBMCs were available from 9 study participants for measurement of HIV levels. Ileal biopsies were available for three of the 9 study participants.
While these samples offer the advantages of an inclusive mix of cell types and less processing that could affect HIV RNA or drug levels, the PBMCs and gut biopsies also differ in composition of infected and uninfected cells. For more direct comparison of CD4+ T cells in the two tissues, we also sorted CD4+ T cells (defined as CD3+CD8-) from fresh gut biopsies (rectosigmoid +/- ileum) and blood from a different group of 7 ART-suppressed study participants recruited prospectively and sequentially from the SCOPE cohort (Table 1).
For flash frozen rectosigmoid tissue, 1 mL TRI Reagent with 2.5 μL polyacryl carrier (both from Molecular Research Center, Cincinnati, OH) was added to the pooled rectal biopsies (six rectal biopsies per individual), which were homogenized using a Mini Beadbeater (Biospec Products, Bartlesville, OK). PBMCs from the same participants were thawed quickly and pelleted by centrifugation (300g for 5 min at 4°C). Following centrifugation, cryopreservation medium was removed and 1 mL TRI Reagent with 2.5 μL polyacryl carrier was added to homogenize cells. Total cellular RNA and DNA were subsequently extracted per the TRI Reagent protocol with back extraction for DNA[27].
RNA and DNA concentrations and quality were measured using the Nanodrop 1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA). Up to 1μg of total RNA was used for a polyadenylation-reverse transcription-ddPCR assay for the TAR region, which is found in all HIV transcripts. This assay employs an initial polyadenylation step, since efficient reverse transcription (RT) of short, prematurely-terminated TAR transcripts requires RT from a linker molecule [31]. Following an RT reaction employing a combination of oligo (dT) and random hexamers, replicate 5μL aliquots of the cDNA were used in a ddPCR reaction containing TAR-specific primers and probe. Up to 5μg of RNA was used for a separate 50μL common RT reaction, from which aliquots of cDNA (5μL/well) were used in ddPCR assays for other sequence regions, including U3-U5 ("Read-through"), R-U5/pre-Gag ("Long LTR"), Nef, U3-R-polyA ("Poly A"), and multiply-spliced Tat-Rev (MS Tat-Rev) regions [27].
Since prior studies suggest that the average level per cell of "housekeeping" transcripts and the average RNA content per cell differ between cell types or even between corresponding cell types in different tissues or conditions (such as T cell activation [27, 86, 87]), we did not use housekeeping transcripts or total cellular RNA to determine the cell equivalents in the RNA from the PBMCs or rectal biopsies, which contain different mixtures of cell types from different tissues. Instead, we determined the total cell equivalents in the DNA extracted from the same samples as the RNA by measuring the absolute copy numbers of a nonduplicated cellular gene, Telomere Reverse Transcriptase (TERT), using ddPCR [88]. Using the assumption that the extraction efficiency is similar for both DNA and RNA, which we have verified in prior TRI reagent extractions from PBMCs, we normalized the absolute copy numbers of cDNA in each well to copies per million cells by using the total cell equivalents in the DNA and the fraction of the total RNA used for each RT and ddPCR [27]. It should be noted that the same method of normalization was used for each HIV RNA transcript, and therefore would not explain any difference between levels of various HIV RNAs in the same sample. As an additional method of normalization, we also normalized levels of each HIV RNA to the total cell equivalents in the DNA as determined by DNA mass, using the concentration measured by NanoDrop and the total resuspension volume. The findings were essentially identical.
We also quantified the levels of the U3-U5 ("Read-through"), TAR, Long LTR, and Nef regions in DNA extracted in parallel with the RNA from the same PBMCs and rectal biopsies (from n = 9 individuals) using the same primers/probes and ddPCR conditions used to measure each HIV RNA [27]. Since the PolyA assay does not detect HIV DNA, polyadenylated transcripts were normalized to HIV DNA levels of the U3-U5 (Read-through) assay, which contains the U3-R region of the PolyA assay and shares the same forward primer and probe. Note that we excluded MS Tat-Rev from this analysis, since this assay also does not detect HIV DNA and spans the two exons of tat and rev, so there is no DNA equivalent. Cellular DNA was fragmented by passage through a QIAshredder column [88]. Levels of each HIV DNA sequence region were measured by ddPCR in replicate aliquots (at least 2) of 500ng DNA/well and expressed as copies per million cells by levels of TERT (measured in duplicate) in other aliquots from the same DNA.
To further assess the effect of proviral sequences, we calculated the ratio of each HIV RNA to the corresponding HIV DNA sequence region quantified using the same ddPCR assay and normalized to 106 cells in the same fashion. This measure expresses the average level of transcription per provirus and is independent of normalization to cell numbers. While this measure may be best to assess the impact of proviral mutations in each sequence region, it should be noted that some sequence regions (TAR, "Read-through") are present in 2 copies per intact HIV DNA (one in each LTR), while the others are present in only one copy. As an alternative method to express the average transcription per provirus in the PBMCs and biopsies, we also calculated the ratio of each HIV RNA to HIV DNA as measured using the R-U5/pre-Gag (Long LTR) assay, which is present in only one copy per intact provirus. This method also allows comparison of Tat-Rev to the other transcripts. Finally, we calculated the ratios of one HIV transcript to another, which are independent of normalization to cell numbers and can be used to measure the presence and degree of different HIV transcriptional blocks [27].
To determine whether sequence-specific differences in RNA stability could contribute to differences in levels of the HIV RNAs, we measured the decay of Read-through, TAR, Long LTR, Nef, PolyA, and MS Tat-Rev transcripts in CD4+ T cells from an ART-suppressed individual using RNA Pol II inhibitors, Triptolide and Actinomycin D. CD4+ T cells were isolated from blood from an ART-suppressed individual using the Dynabeads Untouched Human CD4 T cells kit (Thermo Fisher, Waltham, MA). Replicate aliquots of CD4+ T cells (6x106 cells/well) were seeded into 6-well tissue culture plates (Corning Inc., Corning, NY) at a concentration of 1x106 cells/ml in complete RPMI with either DMSO (negative control), 100nM Triptolide (Sigma, St Louis, MO), or 5mg/ml Actinomycin D (Sigma, St Louis, MO). Cells were harvested at the following time points: DMSO: 0, 1 and 16h; Triptolide: 0, 1, 3, 6 and 16h; Actinomycin D: 0, 1, 3 and 16h. HIV transcripts (Read-through, TAR, Long LTR, Nef, Poly A, and MS Tat-Rev) were quantified using RT-ddPCR as described above. Levels of each HIV RNA were quantified by RT-ddPCR, normalized by alternative measures (cell counts, DNA mass, RNA mass), and expressed as a fraction of the value at time zero. The half-life for each transcript was determined using an exponential one-phase decay model.
For more direct comparison of CD4+ T cells in the blood and gut, we also analyzed CD4+ T cells isolated from blood and rectosigmoid (+/- ileum) from a different group of 7 ART-suppressed individuals. Fresh gut biopsies (15–20) were obtained by colonoscopy, placed immediately in RPMI (supplemented with L-Glutamine, penicillin, streptomycin and 15% fetal calf serum), washed, and dissociated into total gut cells using collagenase with DNase and mechanical shearing [26]. Blood was obtained immediately before colonoscopy and PBMCs were isolated using Ficoll as previously described [26]. PBMCs and total gut cells were counted, stained with LIVE/DEAD stain and fluorescently-conjugated antibodies, and sorted for live, single, CD45+CD3+CD8- cells as previously described [89]. Sort yields from the rectum ranged from 83,474 to 976,738 (median 360,279; S2 Table). Cells were sorted into FACS buffer, centrifuged to pellet cells, and immediately frozen.
Total cellular RNA and DNA were isolated from the sorted CD3+CD8- cells using TRI Reagent, as described previously [26]. DNA was resuspended in 20μL (for rectal T cells) to 75μL (for blood T cells with higher cell counts) of QIAGEN buffer EB. 2.2μL were used to measure the DNA cell equivalents by ddPCR for TERT in duplicate (1μL/well). At least one (and up to 3) aliquots (5μL/well) were used to measure HIV DNA by ddPCR for the "Long LTR" region. At least two (and up to 4) aliquots (5μL/well) were used to measure levels of HIV DNA 2-LTR circles using a new ddPCR assay (see below). Levels of HIV DNA (Long LTR) and 2-LTR circles were normalized to copies/106 cells using the absolute levels of TERT (2 copies/cell) in the DNA and the volumes used for each assay.
Total cellular RNA was resuspended in 20μL of RNase-free water. 5–10μL (no more than 1μg) was used for the polyadenylation-RT-ddPCR assay for TAR, while the remainder was used for a common RT reaction from which aliquots were used in replicate ddPCR assays for Read-through, Long LTR, PolyA, and MS Tat-Rev transcripts (S4 Fig) [27]. Since the average level per cell of "housekeeping" transcripts may differ in CD4+ T cells from blood and gut (which consist of different mixtures of TH subtypes with differing proportions of activated cells), and we wanted to reserve as much of the RNA as possible for measurement of the 5 different HIV transcripts, we did not attempt to measure levels of housekeeping transcripts. Instead, HIV RNA levels were normalized to copies per million cells using the total cell equivalents in the DNA, as measured using ddPCR for TERT. As an alternative method to normalize the HIV RNA levels to cell numbers, we also used total cell counts from the sorts; findings were the same.
To express the average level of each transcription per provirus in the sorted CD4+ T cells, we calculated the ratio of each HIV RNA to HIV DNA (both expressed as copies/106 cells using normalization to TERT levels) as measured using the R-U5/pre-Gag (Long LTR) assay, which is present in only one copy per intact provirus. Finally, we calculated the ratios of one HIV transcript to another, which are independent of normalization to cell numbers and can be used to measure the presence and degree of different HIV transcriptional blocks [27].
HIV-infected, ART-suppressed individuals were recruited from the SCOPE cohort (individual 2475) and the VA Healthcare System (individual 129). Following a blood draw, PBMCs were isolated using Ficoll as previously described [26]. For individual 2475, a proportion of PBMCs (40x106 cells) were untreated. The remaining cells were split for two treatment protocols: 1) 40x106 PBMCs were treated with collagenase as described previously [26] and stained using the same antibodies employed for sorting CD4+ T cells from blood and gut (CD45, CD3, CD8, and LIVE/DEAD); 2) 40x106 PBMCs were not treated with collagenase but stained using CD45, CD3, CD8, and LIVE/DEAD. A similar procedure was followed for individual 129, where 40x106 PBMCs were untreated and 40x106 cells were treated with collagenase, except these cells were not stained with fluorescently-conjugated antibodies. Following treatments, cells were counted and aliquoted into 10x106 cells prior to centrifugation to pellet. After removal of supernatant, cells were directly lysed in TRI reagent and stored at -80°C until RNA and DNA extraction.
Total cellular RNA and DNA were extracted using TRI Reagent (Molecular Research Center, Inc., Cincinnati, OH) as per manufacturer’s instructions, with the following modifications: polyacryl carrier (Molecular Research Center, Inc., Cincinnati, OH) was added to TRI reagent prior to lysis, RNA was resuspended in RNase free-water, DNA was extracted using back extraction buffer (4M guanidine thiocyanate, 50mM sodium citrate, 1M Tris), polyacryl carrier was added to the aqueous phase containing the DNA, and DNA was resuspended in QIAGEN buffer EB.
A common RT reaction was used to generate cDNA for all ddPCR assays except TAR [27]. Each 50μL RT contained cellular RNA, 5μL of 10x Superscript III buffer (Invitrogen, Carlsbad, CA), 5μL of 50mM MgCl2, 2.5μL of 50ng/μl random hexamers (Invitrogen), 2.5μL of 50μM dT15, 2.5μL of 10mM dNTPs, 1.25μL of 40U/μL RNaseOUT (Invitrogen), and 2.5μL of 200U/μL Superscript III RT (Invitrogen). Control RT reactions were performed in parallel with participant samples. A ‘6-assay’ synthetic HIV standard was utilized as a positive control [27]. HIV- donor PBMCs and water that was subjected to nucleic extraction by TRI Reagent served as negative controls for each transcript. The thermocycling conditions were as follows: 25.0°C for 10min, 50.0°C for 50min, followed by an inactivation step at 85.0°C for 5min.
Reverse transcription from a linker molecule (which we achieve using polyA polymerase to attach a polyA tail) is necessary for efficient reverse transcription of short, prematurely-terminated HIV transcripts limited to the TAR loop [31]. Therefore, a polyadenylation step was employed prior to reverse transcription and ddPCR for the TAR region [27, 31]. Each polyadenylation reaction comprised cellular RNA with 3μL of 10x Superscript III buffer (Invitrogen), 3μL of 50mM MgCl2, 1μL of 10mM ATP (Epicentre), 2μL of 4U/μL poly-A polymerase (Epicentre), and 1μL of 40U/μL RNaseOUT (Invitrogen) in a 20μL reaction. The reaction was incubated at 37μC for 45min prior to addition of RT reaction components, including 1.5μL of 10mM dNTPs (Invitrogen), 1.5μL of 50ng/μL random hexamers (Invitrogen), 1.5μL of 50μM oligo dT15, and 1μL of 200U/μL Superscript III reverse transcriptase (Invitrogen). Reverse transcription was performed on the final 30μL reaction at 25.0°C for 10 min, 50.0°C for 50 min, followed by an inactivation step at 85.0°C for 5 min.
Droplet digital PCR was employed because it enables absolute quantification, circumvents the requirement for external HIV standards, and is more forgiving of differences in PCR efficiency due to sequence mismatches [90]. These assays have been validated previously [27, 31]. cDNA from each sample was assayed in duplicate wells for Read-through, TAR, Long LTR, PolyA, and MS Tat-Rev regions (all samples) and (for rectal biopsies and PBMCs) one replicate for Nef. Total cellular DNA was used for the following ddPCR assays: 1) TERT (all samples); 2) Long LTR DNA (all samples); 3) U3-U5 ("Read-through"), TAR, and Nef DNA (PBMCs and rectal biopsies); and 4) 2-LTR circles (sorted CD4+ T cells). Each reaction consisted of 20μL containing cDNA (5μL) or DNA, 10μL of ddPCR Supermix for Probes (no dUTP) (Bio-Rad, Hercules, CA), 900 nM of primers, and 250 nM of probe. Following production of droplet emulsions using the QX100 Droplet Generator (Bio-Rad), the samples were amplified under the following thermocycling conditions: 10 minutes at 95°C, 45 cycles of 30 seconds at 95°C and 59°C for 60 seconds, and a final droplet cure step of 10 minutes at 98°C, using a 7900 Thermal Cycler (Life Technologies, Carlsbad, CA). Droplets were quantified using the QX100 Droplet Reader (Bio-Rad Laboratories Inc., Hercules, CA) and analyzed using the QuantaSoft software (version 1.6.6, Bio-Rad Laboratories Inc., Hercules, CA) in the “Absolute” quantification mode. Gates were set above the negative controls. False positives were rare, generally limited to a single droplet, and usually identifiable by abnormally high fluorescence in both channels.
A new ddPCR assay was used to measure levels of 2-LTR circles. Primers were "F Kumar" (5’ GCCTCAATAAAGCTTGCCTTGA 3’; HXB2 522–543) and "R Butler mod 2-LTR" (5' YCCACAGATCAAGGATMTCTTGTC 3’; 51–28). The probe, "P Kumar" (5’ CCAGAGTCACACAACAGACGGGCACA 3’; 559–84, was dual labelled with FAM (5') and Black Hole Quencher (BHQ; 3'). Each reaction consisted of 20μL containing DNA, 10μL of ddPCR Supermix for Probes (Bio-Rad, Hercules, CA), 900 nM of primers, and 250 nM of probe. Thermocycling conditions and analysis were as described above. A 2-LTR junction standard was created to determine the performance characteristics of the assay. The 2-LTR junction region was amplified from HIV NL4-3 infected PBMCs using the primers "F Buzon mod 2-LTR" (5’ CTARCTAGGGAACCCACTGCT 3’; HXB-2 498–518) and "R Buzon 2-LTR" (5’ GTAGTTCTGCCAATCAGGGAAG 3’; 92–71), then cloned and sequenced. The copy numbers in the standard were determined using the calculated molecular weight and the DNA concentration as determined by NanoDrop. Replicate dilutions of the standard were used in replicate experiments to determine the detection limit, efficiency, and linearity (S3 Fig). Detection frequencies were 2/6 at 0.9 copy, 3/6 at 1.8 copies, and 6/6 at 3.6 copies or above. No apparent inhibition was observed with addition of 1 or 2μg of QIAshredded cellular DNA.
The Wilcoxon signed rank test was performed to assess differences between levels of different HIV RNA or DNA sequence regions. Wells with no positive droplets (most common for MS Tat Rev) were assigned a value of zero for purposes of calculating the median and p values in Figs 2 and 4, and S1 Fig. The Wilcoxon signed rank tests were also repeated with no value (blank) for the undetectables, which did not change the major findings. GraphPad Prism (Version 5.0) was used for the Wilcoxon tests and exponential one phase decay modeling. As an additional method to account for undetectable samples, a negative binomial regression analysis was performed in STATA using the cell equivalents used in each ddPCR well and the number of replicates. The major findings did not change.
|
10.1371/journal.pcbi.1002842 | Phosphorylation Variation during the Cell Cycle Scales with Structural Propensities of Proteins | Phosphorylation at specific residues can activate a protein, lead to its localization to particular compartments, be a trigger for protein degradation and fulfill many other biological functions. Protein phosphorylation is increasingly being studied at a large scale and in a quantitative manner that includes a temporal dimension. By contrast, structural properties of identified phosphorylation sites have so far been investigated in a static, non-quantitative way. Here we combine for the first time dynamic properties of the phosphoproteome with protein structural features. At six time points of the cell division cycle we investigate how the variation of the amount of phosphorylation correlates with the protein structure in the vicinity of the modified site. We find two distinct phosphorylation site groups: intrinsically disordered regions tend to contain sites with dynamically varying levels, whereas regions with predominantly regular secondary structures retain more constant phosphorylation levels. The two groups show preferences for different amino acids in their kinase recognition motifs - proline and other disorder-associated residues are enriched in the former group and charged residues in the latter. Furthermore, these preferences scale with the degree of disorderedness, from regular to irregular and to disordered structures. Our results suggest that the structural organization of the region in which a phosphorylation site resides may serve as an additional control mechanism. They also imply that phosphorylation sites are associated with different time scales that serve different functional needs.
| Cells employ protein phosphorylation – the addition of a phosphate group to serine, threonine or tyrosine residues – as a key regulatory mechanism for modulating protein function. Proteomics technologies can now quantify thousands of phosphorylation sites to reveal the dynamics of phosphorylation at each site in response to a biological process. It is known that phosphorylation does not occur randomly with regard to a protein's structure, but so far the relationship between the dynamics of phosphorylation and these structural properties has not been investigated. Here we relate the relative levels of phosphorylation for more than 5,000 sites through the cell cycle to the predicted structural features of the vicinity of the sites. We find that dynamic phosphorylation tends to occur in disordered regions, whereas phosphorylation sites that did not vary as much over the cell cycle are often located in defined secondary structure elements. Kinases that prefer charged amino acids in their substrate motives are more often associated with unchanging sites whereas proline-directed protein kinases phosphorylate cell cycle regulated sites in disordered regions more frequently. The structural organization of the region in which a phosphorylation site resides may therefore serve as an additional control mechanism in kinase mediated regulation.
| Phosphorylation is a ubiquitous post-translational modification that is known to be important for the regulation of a myriad of cellular processes, among which are cell growth, apoptosis, differentiation, signal transduction and transport [1]. Rapidly evolving mass spectrometry (MS)-based technologies, innovative labeling techniques and advances in computational proteomics provide powerful means for overcoming the low abundance problem of this modification and are making it possible to obtain large-scale, high-resolution quantitative data. With these advances, not only can single protein phosphorylation experiments be done with high accuracy, but also whole-phosphoproteome studies are becoming increasingly feasible [2], [3].
Given the availability of these data, much research has been devoted to analyzing and understanding the structural features of phospho-sites. This includes creation of online resources containing structural information [4], combining data on linear motifs and structural properties [5], and development of software tools that use three-dimensional data for the prediction of phosphorylation sites (DISPHOS [6], Phos3D [7]). Large-scale studies of the structural characteristics of phosphorylation sites have focused on solvent exposure, local and global structure, amino acid context of the spatial surrounding, and structural motifs [7]–[9]. The mechanism of modification suggests that serine, threonine and tyrosine residues should be located on the protein surface where they are accessible for the modifying kinase [7].
The main challenge in studying structural properties of phospho-sites from experimental data is their preference for unstructured regions [6] for which electron density is often missing in X-ray structures. Disorder is strongly associated with protein-protein interactions [10]. Modified residues found within disordered regions can act as on/off switches, either promoting or inhibiting an interaction. Due to the specific structural organization of some protein kinases, in which the catalytic loop resides within a small cleft between two lobes, flexible regions within the substrate's interaction surface are well suited for binding to the kinase. However, a recent systematic study suggested that kinase preference for disordered regions is only marginal [8]. Furthermore, a computational study of kinase specificity reported that approximately 60% of the sites modified by protein kinase A lie within α-helical regions [11]. These considerations raise an interesting question: can a distinction between kinases be made with respect to the level of structural organization of their substrates.
Since phosphorylation events both depend on the structural environment and influence its properties, protein structure and phosphorylation should be considered interrelated and mutually dependent. On one hand, disorder facilitates the access of a kinase to the residue to be modified. On the other hand, the addition of a phosphate moiety may lead to structural changes. Both order-to-disorder and disorder-to-order transitions upon phosphorylation have been observed in nature or studied via molecular dynamics simulations [12]–[15]. The major driving forces of conformational changes observed upon phosphorylation are the electrostatic interactions between the negatively charged phosphate group and the surrounding charged residues. The functional roles of charged residues range from stabilization to correct substrate identification and facilitation of conformational changes.
Although numerous previous studies have focused on structural properties of phosphorylation sites [6]–[8] no systematic analysis has been performed combining large-scale quantitative data with structural features. To bridge this gap we here build on data from a recent study by Olsen et al., which elucidated phosphorylation site occupancy during mitosis [16]. Quantitative data were measured at six time points, corresponding to major phases of the cell division cycle. The additional temporal dimension of these data makes it possible to examine how various phosphorylation sites are dynamically regulated. Olsen et al. clustered sites according to their distinct phosphorylation patterns and similarities in regulation with the aim to infer each site's functional importance. Here, in contrast, we focus on structural properties of the phosphorylation sites and, for the first time, distinguish between two groups of sites with respect to the overall variation of phosphorylation over time.
We find that sites that lie within regular secondary structures exhibit less variable phosphorylation fold changes during the cell cycle than sites that are found in disordered regions. Analysis of the amino acid composition of the flanking regions of these two groups of sites revealed enrichment of positively charged residues and depletion of disorder-related residues such as proline, serine and threonine in the former group.
Using the data from the Olsen et al. investigation [16], we here computed the overall variation of the phosphorylation ratios during six time points of the cell cycle and investigated the differences between the sites with small variation as opposed to the sites with large variation. The original data set comprised 6,027 proteins with 20,443 unique phosphorylation sites. We retained only those sites that had quantitative information for all six time points available (1,059 proteins with 5,173 sites). The phospho-site variability is calculated as the standard deviation of the phosphorylation ratios over the six time points measured during the cell cycle.
We sought to investigate a possible relation between the structural organization of the environment in which a modified residue is found and the experimentally measured changes in phosphorylation during the cell cycle. To do so, we compared the phosphorylation variation of two groups of sites. These two groups were composed of sites that reside in ordered regions and sites that lie within disordered regions as predicted with DISOPRED [17]. In agreement with previously observed tendency we found over 90% of the modified residues to lie within disordered regions (4,675 sites versus 498 sites). Figure 1 shows three examples from our large-scale dataset, illustrating a non-variable site on a regular secondary structure (α-helix), a slightly variable site on a short loop and a variable site in a disordered region.
Our results revealed notable differences in the distributions of phosphorylation variations of the two sets (Kolmogorov-Smirnov test p-value 6.6E-13). The sites associated with structurally characterized regions were found to exhibit smaller changes in phosphorylation during the cell cycle (median 1.77) as compared to sites located in disordered regions (median 2.22, Figure 2A).
Having investigated the difference between ordered and disordered regions on a global scale, we next predicted protein secondary structure in more detail using PsiPred [18]. First we classified sites into regular structures (92 in α-helices or 53 in β-sheets) and sites with irregular structures (5,028 in loops, turns and coils). Phosphorylation in regular secondary structures showed smaller variation over the six time points of the cell cycle. This effect was small but statistically significant (ANOVA p-value 1.8E-04).
Although there is a large intersection between ordered structures and regular secondary structures, and the terms are often used interchangeably, the two sets are not identical. We observed that a large number of regions predicted as ‘coil’ by PsiPred are predicted as ‘ordered’ by DISOPRED. This reflects a distinction between ordered and disordered coils. A major difference between these two groups of coils is the length distribution of their elements (p-value 4.12E-114): ‘ordered coils’ are much shorter on average as they mainly correspond to turns and short loops connecting regular secondary structures. By contrast, ‘disordered regions’ are longer and represent large protein regions lacking defined structure (see Text S1 for details).
In order to take this distinction into account, we redefined the structural environments into three categories: regular structures (predicted as helix/sheet and ordered), irregular structures (predicted as coil and ordered), and disordered regions (predicted as coil and disordered) (Figure 2B). We found significant differences in the variation of the phosphorylation ratios between these distinct structural groups (ANOVA p-value 3.02E-09). Sites within ordered structural environments appeared to be subjected to the lowest level of regulation during the cell cycle (median 1.65). Interestingly, a distinction emerged between coils (median 1.83) and disordered regions (median 2.22), signifying that the latter exhibited the largest variation in phosphorylation changes. We speculate that the increased variation of phosphorylation in longer, disordered coils correlates with their higher solvent exposure, which makes them more easily accessible for both kinases and phosphatases. Overall, our data shows that the phosphorylation variation of a site clearly scales with the level of order of its structural context (i.e. the tendency of a site to be found within a regular, irregular or disordered region).
We wanted to investigate if sites with distinct phosphorylation patterns over the cell cycle differ not only according to structural context, but also with respect to the amino acid content in their local sequence environment. A two-sample logo [19] was computed to contrast the two data sets, using the highly variable sites as a negative set (Figure 3). For each position and each possible amino acid, a two sample t-test was used to evaluate the null-hypothesis that the vectors of residues at a given position in both the positive and negative data sets (i.e. low and high variation) come from the same distribution. We found statistically significant enrichment of charged amino acids and depletion of proline, serine and threonine in the surrounding of sites with small phosphorylation variability (p-value<0.05). Additional comparisons of the amino acid distributions of the two sets against a background distribution accounting for structural differences using the composition profiling technique [20] revealed similar trends (see Text S1 for the detailed analysis and results).
The enrichment of serine and threonine residues in the vicinity of the detected phosphorylation sites could correlate with additional modification events. To check this hypothesis, we determined if multiple phosphorylation sites are found with higher preference in disordered regions. Phosphorylation sites that had at least one neighboring phosphorylation site in both ordered and disordered regions were compared. A ‘neighbor’ was defined as any phosphorylated residue that lies within +/−1,2,3,4, or 5 residue-long flanking region of a given modification site. Regardless of which of these five cut-offs was chosen, multiple phosphorylation sites were always highly significantly enriched in disordered regions (Table 1).
Next, we were interested in potential differences in evolutionary constraints on the phospho-sites in structured and disordered regions. When analyzing conservation it is important to take into account the different evolutionary rates of disordered and ordered regions. We therefore compared conservation scores between phosphorylated serines, threonines and tyrosines with ‘control’ serine, threonine and tyrosine residues with a similar structural background. We define the set of ‘control’ residues, as all potential phosphorylation sites that were not found to be phosphorylated in the study of Olsen et al.
As expected, phospho-sites that were predicted to lie in regular regions appeared significantly more conserved than phospho-sites in disordered regions (p-value 3.23E-120), due to the more conserved structural background of the former (Figure 4). In agreement with a previous study [21] modified residues in regions that lack defined structure were more conserved than the control serine, threonine and tyrosine residues with the same surrounding environment (Mann-Whitney Wilcoxon test p-value 3.4E-03). The same holds true for phospho-serine, phospho-threonine and phospho-tyrosine in ordered regions as compared to their equivalent control sets (p-value 2.24E-16). Despite the small size of the effect (groups' means −0.38, −0.28, 0.14 and 0.22 for pS/pT/pY ordered, S/T/Y ordered, pS/pT/pY disordered and S/T/Y disordered respectively) the higher evolutionary pressure on phosphorylated residues suggests functional importance of these sites in a broad range of species.
We next asked if different groups of kinases would exhibit preferences for less variable or highly variable phosphorylation sites. To identify kinase recognition motifs that show similar behavior with respect to two variables – protein disorder and phosphorylation variation, we used the recently described 2D Annotation Enrichment technique (see Methods and [22]). It employs a two dimensional generalization of the nonparametric two-sample test to detect preferences of a certain group of elements for two numerical attributes simultaneously relative to all other elements. The motifs separation is plotted in Figure 5 (the complete data are available in Table 2). The general trend between disorder and phosphorylation variation is reflected in the plot as sites with more disordered background show also higher variability. For individual kinases a very clear separation reflecting their preference for specific amino acids in their consensus motifs becomes apparent. Overall, four classes can be distinguished: (i) tyrosine kinases (black squares), (ii) proline-directed kinases (red circles), (iii) non-proline directed kinases with charged residues in their substrate recognition motif (green and blue triangles) and (iv) proline-oriented kinases, which contain a proline residue in their motif (red triangles and pentagons).
The class of tyrosine kinases shows a strong preference for low phosphorylation variability and structured regions, whereas the other three classes favor more disordered regions, but span a wider range of phosphorylation variation values. Also among the latter three groups higher quantitative variability is clearly associated with higher disorder. As seen in Figure 5, basophilic and acidophilic kinases occupy the regions on the graph corresponding to low phosphorylation variation, whereas proline-directed kinases are located on the right part of the graph, demonstrating their preference for more variable sites. The motif corresponding to the highest variability is the consensus motif for the proline-directed CDK5 kinase, which is in agreement with the important regulatory role of this enzyme during the cell cycle.
The proline-related class represented by two acidophilic, one basophilic and one atypical protein kinases shows preferences for intermediate level of disorderedness and phosphorylation variation properties. The Casein kinase II is characterized by various substrate recognition motifs [23], but the main differences are related to presence or absence of a proline residue preceding the phosphorylated site. These two motifs show distinct structural and variation preferences – the former type being more similar to proline-directed kinases and the latter – to non-proline ones. The occurrence of the G protein-coupled receptor kinase 1 (GRK1) near the proline-directed kinases (red triangle) can be explained analogously by the presence of a proline residue in the consensus sequence for that kinase. Interestingly, the reported consensus motifs of the MAPKAPK2 kinase (blue triangle) do not contain proline residues, however, it is still grouped together with the more variable kinase motifs. After careful examination of the amino acid composition of substrates of the MAPKAPK2 kinase in our data set we found multiple examples that contained proline within +/−6 residue window around the phospho-site. Together with our structural analysis, this suggests that this residue may play an important role in the substrate recognition. Overall, the group of proline-oriented kinases has similar preferences for disorder and phosphorylation variability as the proline-directed group. This observation also extends to the functional relevance of the member kinases to the regulation of the cell cycle. For example, the DNA-dependent protein kinase (DNA-PK) is involved in stress response and DNA repair and is known to play a role in the progression of the cell cycle [24]. Furthermore, the MAPKAPK2 kinase is involved in DNA repair processes and thus can provide an alternative to checkpoints activation [25].
Proline-directed kinases such as PLK1 are known to actively regulate the progression of the cell division cycle, thus implying that disordered regions (which are enriched for prolines) are subjected to regulation and therefore to variable phosphorylation patterns. We therefore checked if the tendency of phosphorylation variability to scale with the level of disorder persists if we control for proline-directed kinases and excluded all sites modified by such from the data set. Indeed, the effect of lower phosphorylation variability being associated with ordered regions and higher - with disordered regions remained the same.
Previous studies had already found a preference of phosphorylation to occur in loops or disordered regions [6],[7]. However, those studies generally did not have access to the dynamics of phosphorylation and they therefore based their analysis on the absence or presence of phosphorylation sites alone. Here, we instead made use of a large-scale quantitative phosphorylation data set to investigate a possible relation between the structural features of phosphorylation sites with their degree of regulation. This allowed us to contrast the behavior of less variable sites to those that were dynamically regulated. Our data clearly demonstrate that the propensity of phosphorylation sites to be regulated during the cell division cycle is related to the level of structural organization of the environment in which these sites reside. Furthermore, we discovered that this effect occurs in a graded manner: regions with regular structure are least likely to harbor regulated phosphorylation sites, followed by irregular regions (short loops or random coils). Note that over 90% of the sites were found within disordered structures and their high phosphorylation variability relates them to regulated phosphorylation events.
Interestingly, the sets of sites within ordered loops and disordered structures showed significant differences. It has been shown before that different flavors of disordered regions exist with regards to their lengths, amino acid composition, and the conformational transitions that they undergo upon binding [26], [27]. Liu et al. defined regions with no regular secondary structures (NORs) as one specific category of disordered regions. They demonstrated that NORs differ significantly from regular structured loops and argued that these might have different functional implications, a hypothesis which finds support in our study.
Functional analysis of the highly variable set of sites revealed enrichment of cell cycle-related, biosynthesis and cellular organization and localization processes (Table S1). Some examples are RNA, DNA and mRNA processing, localization and transport, regulation of gene expression and biosynthesis. Cell cycle-associated processes such as regulation of the different phases of the cycle, DNA replication and repair, telomere organization and maintenance and chromatin assembly were also strongly over-represented in the variable set of sites.
Phosphorylation is an important mechanism for regulation of a myriad of intricate processes during cell division. A detailed study of the cell cycle regulation through phosphorylation focused on functional analysis of protein groups that are up or down regulated at specific time points [2]. These were the proteins that contained sites that reached phosphorylation peaks at S or M phases. As expected, proteins involved in mitotic and cell cycle processes were shown to be maximally phosphorylated at mitosis.
Interestingly, Olsen et al. found proteins that regulate metabolic processes to be weakly phosphorylated during S phase and highly phosphorylated at mitosis. An explanation to this discovery is the possibly inhibitory character of phosphorylation on proteins that regulate metabolic processes, as protein synthesis and related functions tend to shut down during mitosis. Furthermore, DNA replication takes place during S phase, which rationalizes the up-regulation through phosphorylation of various proteins involved in DNA replication repair. High phosphorylation of cytokinesis-related proteins in S phase appears to play an important role in the control of the correct segregation of the two daughter cells.
The tendency of modification sites in regular structures to be less variable may be facilitated by proximal charged residues acting as stabilizers of the phosphate group. Charged flanking regions offer a suitable environment for hosting a phosphate group and allow for favorable interactions that potentially result in phosphorylation acting on a longer time scale. For instance, these favorable interactions could reduce the efficiency of phosphatases in removing a phosphate group, thereby contributing to the tendency for smaller variation in the phosphorylation level that we observe in our data. In contrast, negatively charged residues could lead to repulsion-driven conformational changes and polarization of the entire protein surface by creating clusters of negatively charged residues.
Several mechanisms that are known from literature furthermore contribute to the observed tendency for structural rather than regulatory phosphorylation sites to be present in ordered regions. Specific structural changes due to phosphorylation include stabilization of the N-termini of α-helices via favorable interactions of the added phosphate group with the helix backbone [28]. This is effected by the interaction of the phospho group with the helix dipole moment. Yet, the same modification introduced at the C-terminus would have the opposite effect [29]. The optimal stabilizing position for the phosphate group was estimated as −2 relative to the N-cap of a helix. Additionally, favorable electrostatic interactions between proximal positively charged residues (e.g. at a helix cap) and the phosphate group can enhance helix formation. The stabilizing effect of salt bridges formed between a phosphate group and a lysine side-chain has been recognized as one of the strongest possible α-helix inducers [30]. In contrast, the phosphate-guanidinium interaction leads to disruption of the local regular structure [31]. Phosphorylation has also been reported to cause conformational changes in β-sheets and disruption of β-hairpins. In those cases repulsive interactions with an aromatic tryptophan residue in the spatial vicinity of the phospho-site are observed [32]. A related question that arises from our investigation is to what extent the phosphorylation variability of a site is connected to a role in the overall structural re-arrangements of a protein. A phosphorylation event can alter the energy that is required for a conformational change [15], and thus hinder or facilitate it. Further experiments including 3D structural information or computational models are needed to increase our understanding of the interplay between structure and phosphorylation.
Multiple experimental studies show the regulatory role of modification sites that show variation in their phosphorylation patterns and lie within intrinsically disordered regions For example, the cyclin-dependent kinase inhibitor 1B (p27) is an intrinsically unstructured protein, which is multiply phosphorylated and regulates the cell cycle by inhibition of cyclin-dependent kinases (CDKs) [33]. The disorderedness of p27 plays an important role in keeping the complex formed between CDK and p27 flexible. Due to this flexibility the segment, which blocks the ATP binding site becomes exposed. This allows a tyrosine residue to become accessible for phosphorylation, upon which the space previously occupied by the inhibitor becomes available for ATP binding. Then the partially reactivated CDK phosphorylates p27 at another residue, which leads to its degradation and allows CDK to regain full activity and guide the progression through the cell cycle [34].
In another example, multiple phosphorylation sites on the transcription regulator Retinoblastoma protein (Rb) influence its ability to interact with transcription factors and other regulatory proteins. A detailed structural study reports that the different phospho-sites found within disordered regions induce distinct conformational changes and also serve different functional roles [35]. For instance, one of the modified residues decreases the affinity of Rb for binding the transcription factor E2F by reordering the pocket domain. At the same time another modified site at a loop in the pocket domain induces complete blocking of E2F binding.
We found that the set of sites with varying phosphorylation patterns was enriched in amino acids associated with disorder, specifically Pro, Gly and Ser. Interestingly the same sites were more likely to have additional modified residues in their vicinity. Phosphorylation of a protein often occurs at several distinct residues and it has been reported that modification sites tend to cluster and function in a cooperative manner [36]. Mathematical models suggest that this phenomenon leads to an increase in the sensitivity and robustness of the cellular response [37] and may promote a switch-like behavior [38]. In such a case, the exact position of a modification site in a cluster would not be a determining factor on its own, but would rather contribute to a cumulative effect. It would be worth studying how different levels of phosphorylation variability in regions with different structural organization may be implicated in the cellular regulation of the cell cycle. Multiple phosphorylation sites with highly dynamic phosphorylation patterns may be suitable for both rapid and robust response. In contrast, the robustness of the response of sites within regular regions might be achieved on a longer time scale and be related to longer lasting effects of phosphorylation.
We showed that phosphorylated residues tend to be more conserved than their equivalent non-modified residues. Conservation of phosphorylated residues has been a broadly debated issue [21], [39], but the general consensus appears to be that the overall conservation of phospho-sites is low. Even though statistically it is significantly stronger than that of the equivalent non-modified sites, the effect size is relatively small. Possible explanations include (i) loss and gain of phospho-sites at different positions in disordered regions, likely due to clusters of sites acting as functional units regardless of the exact sequence position [40] and (ii) potential silent phosphorylation events [39].
The idea that it is the cluster of phosphorylation sites that plays a functional role is becoming increasingly accepted [36], [37]. The functional roles of multiple phosphorylated residues span a wide range: (i) targeting for sub-cellular localization, (ii) targeting for degradation, (iii) control of protein-protein and protein-nucleic acid interactions (often through electrostatic effects) and (iv) enhancement of a robust and rapid response to a stimulus [1]. Furthermore, mechanisms of ‘priming’ phosphorylation are also well-known [41].
Here we showed that disordered regions harbor variable sites, which tend to be surrounded by additional phosphorylated sites. This raises the possibility that the variability of these sites is related to some of the above-described phenomena. It is known that disordered regions can facilitate a large number of interaction partners, and that multiple sites can control their association and dissociation. Given the wide range of functions of multiple phosphorylated sites in disordered regions, a larger variability in their phosphorylation patterns may provide an adequate functional mechanism to effect the desired regulation. In contrast, structural regularity imposes certain constrains on the less variable sites. The necessity of evolutionary conservation of the structure tends to prevent the accumulation of disorder-associated serine and threonine residues and a consequent change of their positions. Furthermore, the more rigid structure implies a more limited number of interactions partners. Therefore, we reason that the requirement for regulation for these sites in structured regions can be smaller.
Our data allowed us to investigate the kinase preferences of phosphorylation sites with high vs. low levels of regulation. Tyrosine kinases and kinases that require charged residues in their substrate recognition motives clearly preferred sites with smaller phosphorylation variation, whereas proline-directed kinases were clearly associated with sites that were dynamically regulated. Proline is known to be a helix and sheet breaker, due to the planarity of its side-chain. Proline lacks an NH backbone donor to form a hydrogen bond and thus disrupts the formation of regular hydrogen bond patterns, which are the basis of regular structure formation. Due to its unique stereochemistry the proline residue can adopt two different conformational states – cis and trans – and a large number of folded proteins contain both states of the residue. The intrinsic conformational changes resulting from the proline isomerization play an important role in determining the function, ligand recognition and interactions of the protein [42]. For instance, certain kinases, such as MAPKs and CDK2 preferentially modify substrates with the trans isomer [43]. Proline isomerization in a S/TP motif, where S/T is phosphorylated, can also control the opposite step – dephosphorylation, as some phosphatases appear to be conformation-specific and prefer the trans state [44]. Therefore, the preference of proline-directed kinases for sites with higher variation illustrates a connection between dynamic regulation and disordered regions.
Our results highlight the central role of proline, as a disorder-promoting residue that is also part of regulatory motifs [45]. The directing role of proline together with the multiple functions associated with disorder explain the more variable character of phosphorylation of sites with these properties that we observe in our study. Furthermore, our statistical analysis of the interplay between structure and phosphorylation variation in relation to specific kinase recognition motifs presents a new approach of describing and classifying protein kinases. We showed that the combination of both properties can be used to gain conceptual and specific insights into regulation. We were able to reproduce known relations and to identify new links between kinases, which may reflect functional dependencies emerging from common regulatory behavior and structural preferences.
In conclusion, we have related the tendency of phosphorylation sites to be dynamically regulated throughout the cell cycle with the structural features of the sites. While we have found clear relations between phosphorylation dynamics and protein structure, we are only scratching the surface of what we believe could be an exciting new area at the interface of proteomics and structural biology.
In the data set underlying our analysis [16], human HeLa S3 cells were labeled with SILAC [46], [47] to produce three different isotopic forms of lysine and arginine (light, medium and heavy). The light and heavy isotopes were synchronized in six different stages of the cell cycle, while the medium one was kept non-synchronized as a reference. Relative quantification of protein abundances (protein ratios) and/or phosphorylation (phospho-peptide ratios) were computed by taking the ratio between two cell states at each time point (i.e. synchronized heavy-labeled cells in S phase and non-synchronized medium-labeled cells). In order to account for the possible influence of protein abundance, changes in the phosphorylation ratios between the reference and the stimulated cells were normalized by the protein change. We mainly focused on the phosphorylation ratios as they were available for a larger number of sites compared to the absolute occupancy values.
The data set contained information about the UniProt id of the phosphorylated protein, sequence positions of phosphorylated residues, and quantitative measures of phosphorylation (normalized phosphorylation ratio) at 6 time points (i.e. cell cycle phases: G1, G1S, Early S, Late S, G2 and M). In total 1,059 proteins and 5,173 phosphorylation sites with measured phospho-ratios for each of the six time points of the cell cycle were used in the analysis.
In order to assure that the observed phenomena are not due to the properties of the chosen subset of sites, we repeated the analysis of phosphorylation variation between different structural groups with data sets containing five (5,254 sites), four (8,537 sites), and three (8,731 sites) time points only (see Text S1 for details). Although slight fluctuations were observed, the main tendencies remained stable and the conclusions did not change. Therefore, no bias in the reduced data set (i.e. the one containing information about all six time points) was found.
Phosphorylation variation value for each modified site was computed as the standard deviation of the phosphorylation ratios over the six time points. High variation corresponds to sites with temporal variation of phosphorylation ratios (e.g. a peak is observed in S phase), while low variation describes those sites that retain constant or slightly variable phosphorylation fold change during the cell cycle.
The secondary structure of each site was predicted with PsiPred [18]. Each site was assigned one of three possible states: ‘H’ for α-helix (92 sites) ‘E’ for β-sheet (53 sites), and ‘C’ for random coil, turn or loop region (5,028 sites).
An intrinsic disordered state was also predicted for each site using DISOPRED [17] with standard settings. We found 498 sites to be in the ‘order’ state while the remaining 4,675 were predicted to be in the “disorder” state.
Based on a combination of secondary structure and disorder predictions, we defined three distinct structural categories for each phosphorylated site: (i) regular regions (helices and sheets in ordered regions, 145 sites), (ii) irregular regions (coils in ordered regions, 353 sites), and (iii) disordered regions (coils in disordered regions, 4,675 sites).
Statistical analyses were performed within the R environment [48] and using the in-house statistics work frame Perseus. The lattice package was utilized for comparing distributions of phosphorylation variation in different structural categories. Differences between distributions were assessed with the standard non-parametric Kolmogorov-Smirnov test. In the case of three structural categories, analysis of variance of the phosphorylation fold change was performed using the structural category as an independent variable. Data on phosphorylation site variation and structure predictions are available in the Supporting material (Table S2). Enrichment of functional Gene Ontology (GO) categories was performed with the GOrilla tool [49].
We performed conservation analysis on phosphorylated residues in ordered and disordered regions. The proteins from our data set were mapped to pre-computed EggNOG groups of orthologs [50]. We used the maximum likelihood-based rate4site algorithm to build phylogenetic trees from the EggNOG clusters and to compute residue-based evolutionary rates [51]. Lower evolutionary scores correspond to stronger conservation.
The ‘control’ sets of sites were defined as all serine, threonine and tyrosine residues from the phospho-proteins that were not measured to be phosphorylated in our data set with equivalent structural background (i.e. disordered and ordered as predicted by PsiPred [18]).
We tested if disordered regions are enriched in multi-phosphorylation sites, as compared to ordered regions. We considered phosphorylation sites with at least one modified neighbor as multi phospho-sites. A neighbor residue is defined as a phosphorylated serine, threonine or tyrosine located within +/−1,2,3,4 or 5 residue-long flanking regions of a central phospho-site. For each cut-off length, we built a contingency table. Each contingency table contained the number of sites with and without neighboring phospho-sites for both ordered and disordered regions. The significance of the enrichment was estimated with the Fisher's Exact Test.
The 2D Annotation Enrichment technique [22] enables analysis of the preference of a certain group of elements (i.e. phosphorylation sites, characterized by the same consensus motif) for two numerical attributes simultaneously relative to all other elements (in our case all other phosphorylation sites). It employs a two dimensional generalization of the nonparametric two-sample test and uses the Benjamini-Hochberg method to correct for multiple hypotheses testing. We used the default settings to distinguish the statistically significant groups, corresponding to false discovery rate <0.01. We used the Human Protein Reference Database motif definitions in this analysis [23].
The difference between the amino acid content of the flanking regions of the sites with low and the sites with high phosphorylation variation was computed, assessed and visualized with the help of the Two Sample Logo method [19]. The highly variable set was used as the negative set. Residues significantly enriched in a certain position are shown above the horizontal line in the logo.
|
10.1371/journal.pntd.0002745 | Serological Screening of the Schistosoma mansoni Adult Worm Proteome | New interventions tools are a priority for schistosomiasis control and elimination, as the disease is still highly prevalent. The identification of proteins associated with active infection and protective immune response may constitute the basis for the development of a successful vaccine and could also indicate new diagnostic candidates. In this context, post-genomic technologies have been progressing, resulting in a more rational discovery of new biomarkers of resistance and antigens for diagnosis.
Two-dimensional electrophoresed Schistosoma mansoni adult worm protein extracts were probed with pooled sera of infected and non-infected (naturally resistant) individuals from a S. mansoni endemic area. A total of 47 different immunoreactive proteins were identified by mass spectrometry. Although the different pooled sera shared most of the immunoreactive protein spots, nine protein spots reacted exclusively with the serum pool of infected individuals, which correspond to annexin, major egg antigen, troponin T, filamin, disulphide-isomerase ER-60 precursor, actin and reticulocalbin. One protein spot, corresponding to eukaryotic translation elongation factor, reacted exclusively with the pooled sera of non-infected individuals living in the endemic area. Western blotting of two selected recombinant proteins, major egg antigen and hemoglobinase, showed a similar recognition pattern of that of the native protein.
Using a serological proteome analysis, a group of antigens related to the different infection status of the endemic area residents was identified and may be related to susceptibility or resistance to infection.
| Despite intensive efforts towards disease control, schistosomiasis is still highly prevalent in most endemic countries. Although effective treatment is available and widely used, it does not prevent reinfection, as it could be achieved with the use of a vaccine. Efforts to control and eradicate schistosomiasis rely on praziquantel, the only drug available for treatment. Therefore, the identification of antigens that can induce protective immunity is highly desirable, as well as the need for more sensitive assays, useful to detect low intensity infections and treatment follow-up. The occurrence of natural resistance in schistosome endemic areas suggests that there is protective immunity. However, the mechanisms involved in protection, or the proteins that induce this protective immunity, are not yet known. These proteins, once identified, may constitute the basis for a successful vaccine. In this study, we compared the profile of reactive proteins to the serum antibodies of infected and non-infected individuals residing in a schistosomiasis endemic area using two-dimensional western blotting. The association of proteomic and serological screening methodologies enabled the identification of immunogenic proteins of the parasite, which could be an informative source for the development of vaccines and new diagnostic assays. In this manuscript we describe the discovery of potential candidate proteins for subsequent testing as protective or diagnostic antigens.
| Schistosomiasis is one of the most important parasitic diseases, being prevalent in 76 countries [1]. Despite many control efforts, mainly after the introduction of a chemotherapeutic treatment in 1980s, the disease is still highly prevalent [2]. The control of the main medically important species Schistosoma mansoni, Schistosoma japonicum and Schistosoma haematobium is based on the use of praziquantel, the only drug available for chemotherapy [3]. The use of the chemotherapy has a clear effect on morbidity [4], [5]. However, repeated mass drug administration has exerted selective pressure on parasite population and resistance to praziquantel is being described by different investigators [6].
The development of long-term protection based on vaccination would be of significant benefit for disease control [7]. Despite a large body of research in this area and one ongoing clinical trial [8], there is no effective vaccine against schistosomiasis. Together with the fact that mass drug administration has been applied widely and the increasing drug pressure on the parasite population, it becomes more evident the need to find alternative methods of schistosomiasis control/elimination. In this context development of an effective vaccine is a plausible alternative.
The lack of understanding of the protective immunological mechanisms, and the difficulty in identifying antigens which stimulate such a response, remain the major barriers towards the development of anti-schistosome vaccines [9]. Many single antigens with potential use as a vaccine have been proposed, but most have showed disappointing results even with different immunization schemes and experimental models [10], [11]. Nevertheless, distinct observations in animals and humans indicate that it is feasible to achieve protection against infection. Significant levels of protection were obtained in experiments with irradiated cercariae [12] and with some recombinant antigens [13]–[16]. Furthermore, several reports from our group and others have suggested that resistance to infection is acquired naturally or drug induced [17]–[21]. In our studies specifically, we have shown that resistance may develop naturally in endemic areas, describing a group of individuals, that live in areas where transmission is active but do not get infected, called Endemic Normals [22]. These individuals were defined using specific criteria such as being S. mansoni egg-negative over 5 years despite continuous exposure to contaminated water, no previous treatment with anthelmintic drugs and having vigorous cellular and humoral immune response to crude schistosome antigen preparations [23], [24]. The immune response of individuals with natural resistance to schistosomiasis differs significantly from that of post-treatment resistant and infected individuals [17].
The immunological mechanisms that prevent the infection in drug-induced and naturally resistant individuals living in endemic areas for schistosomiasis may constitute the basis for the development of a successful vaccine [7], [25], [26]. Therefore, we believe that using the most recent technology to identify antigens reactive to antibodies from resistant individuals, both natural and drug induced, we will be able to screen at a much faster pace for putative protective antigens.
Schistosomiasis control will benefit from a vaccine, but a new generation of diagnostic tools is as much a part of any control and eradication strategy. Available tools, especially fecal exams, encounter limitations in low parasitic load and low infection rates settings and in the follow-up to treatment [27], [28]. The next generation of assays needs to be simple, inexpensive, fast, sensitive, specific and capable of distinguishing active from prior infection [28], [29].
The empirical science used in the last decades is strikingly changing with the use of high throughput global approaches to a less biased, and more encompassing development for the proposition of new biomarkers to the discovery of vaccine candidates, drug and diagnostic targets [24], [25], [30]. Significant progress has been made in S. mansoni and S. japonicum proteomic studies, mainly with the description of proteins differentially expressed in the different life cycle stages of the parasite [31]–[33], between male and female worms [34] and irradiated and normal cercariae [35]. Furthermore, proteomic studies have concentrated mainly on the studies of proteins exposed on the parasite surface and readily accessible to the host, i.e. identification of tegumental proteins [36]–[42] or secreted/excreted proteins [38], [43]–[49].
A combination of proteomic and serological analyses has been used as a promising experimental approach for screening new biomarkers candidates to different diseases, identifying proteins useful for diagnosis, therapy and vaccine design [50]–[54]. A limited number of studies have performed serological-proteomic analysis using schistosome proteins. Serum of experimentally infected animals was used to screen antigens of S. japonicum in two-dimensional electrophoresis (2-DE) in an attempt to identify suitable antigens for diagnostic purposes, identifying four proteins out of 30 immunoreactive protein spots [30]. Comparative proteomic and immunological analysis of S. haematobium, S. bovis and Echinostoma caproni revealed some common cross-species antigens and species-specific targets [55]. Additional studies using S. mansoni immunoprecipitated proteins from protective and non-protective rat serum analyzed by 2-DE showed four spots specifically reactive with the protective rat serum [56]. High and low worm burden serum from infected Rhesus macaques were used to probe S. mansoni gut secretions and tegument surface proteins in 2-DE. The study identified gut digestive enzymes, tegument surface hydrolases and antioxidant enzymes as IgG targets of the IgG high titer serum of low burden animals [57]. The use of infected human serum was conducted only for S. haematobium, where a total of 71 immunoreactive protein spots were identified as 26 different proteins [58]–[60].
Although important observations have been made in relation to schistosome antigen identification, a human schistosomiasis mansoni coherent screening for new antigens is still necessary. In the present study we used, for the first time, serum antibodies of infected and naturally resistant individuals from a S. mansoni endemic area to compare the recognition profiles of adult worm antigens by these serum antibodies using two-dimensional Western blotting. We identified a total of 47 S. mansoni antigenic proteins. We also observed that some of the antigens were differentially recognized by antibodies of infected and naturally resistant individuals. This panel of antigens may constitute an informative source for the improvement of diagnostic tools and vaccine development to schistosomiasis.
This research was approved by the Ethics Committee for Human Research of CPqRR – FIOCRUZ (CAAE: 1.0.245.000-08). Written informed consent was obtained from all participants at the time of the stool collection.
All procedures involving animals were conducted in compliance with the Manual for the Use of Animals/FIOCRUZ and approved by the Ethics Committee on the Use of Experimental Animal (CEUA – FIOCRUZ) license number LW-17/09.
BALB/c mice were infected by the subcutaneous route with 100 S. mansoni cercariae of the LE strain. After 45 days, adult worms were recovered by perfusion of the portal mesenteric system, as described by Pellegrino and Siqueira [61]. The adult worms were washed three times in RPMI medium (SIGMA), snap frozen in liquid nitrogen and stored at −70°C until use.
S. mansoni adult worm total protein extract (AW-TOT) was obtained from direct lysis of the parasites in lysis buffer [8 M Urea, 2 M Thiourea, 4% 3-3-Cholamidopropyl-dimethylammonio-propane-sulfonate (CHAPS), 50 mM dithiothreitol (DTT), 20 mM Tris and Complete Mini Protease Inhibitor Cocktail Tablets (Roche)]. After homogenization under continuous agitation for 2 hours at room temperature, followed by 10 repeated passages through a 30-gauge hypodermic needle, the homogenate was centrifuged at 20,000×g for 30 min at 25°C and the supernatant was collected and stored at −70°C until use.
S. mansoni adult worm tegument protein extract (AW-TEG) was obtained by freeze/thaw/vortex method in Tris Buffered Saline (TBS) supplemented with Complete Mini Protease Inhibitor Cocktail Tablets, according to Roberts and co-workers [62] with some modifications. Briefly, after thawing on ice, the outer tegumental membrane complex was removed by ten 1 second vortex pulses at maximum speed. All the supernatant content, obtained after decanting the stripped worms, was passed through a 30-gauge hypodermic needle 10 times and then concentrated using a 3 kDa cutoff centrifuge filter (Millipore). Next, acetone precipitation was performed and the pellet was solubilized in SBI buffer [7 M Urea, 2 M Thiourea, 15 mM 1,2-diheptanoyl-sn-glycero-3-phosphatidylcholine (DHPC), 0.5% Triton X-100, 20 mM DTT and Complete Mini Protease Inhibitor Cocktail Tablets], as described by Babu and co-workers [63], and stored at −70°C until use.
Protein concentration of both protein extracts was measured by the Bradford method [64] and the quality of the extracts was verified by SDS-PAGE 12% [65].
The human serum samples were obtained from a rural population of the Virgem das Graças village (VDG). This is a hyperendemic area for schistosomiasis located in the Jequitinhonha Valley in northern Minas Gerais State, Brazil. In VDG there is no treated water or basic sanitation and water contact was determined by direct observation and translated to Total Body Minutes (TBM). Stool samples were processed by the Kato-Katz method to detect eggs of S. mansoni and other intestinal helminthes. Two slides from each stool sample collected in three consecutive days were used for quantification of number of eggs/gram of stool [66].
In the present study, the serum samples were chosen according to the following criteria: individuals not infected by other helminthes (Ascaris lumbricoides, Trichuris trichiura and Ancylostoma), between 20–50 years of age, man or non-pregnant women. Individuals positive for S. mansoni eggs at the start of the study in 2001 were called Infected Individuals (INF), and those who were egg negative in the three years of the study (2001, 2002 and 2006) were identified as Non-Infected Individuals from Endemic Area (NE). All the serum samples used in this study were obtained at the initiation of the study in January 2001, before mass chemotherapy with praziquantel was administered. Serum samples of 13 INF, who had 8 to 304 eggs/gram of stool, and 9 NE were used (Table 1). Sera of Non-Infected volunteers from non-endemic sites (NI) were also used in this study, 7 from USA or 2 from UK sites.
For each two-dimensional-polyacrylamide-gel-electrophoresis (2D-PAGE), 100 µg of proteins from AW-TOT or AW-TEG extracts were used. The AW-TOT proteins were solubilized in IEF rehydration buffer [8 M Urea, 2 M Thiourea, 4% CHAPS, 0.0025% bromophenol blue, 65 mM DTT and 1% BioLyte 3–10 buffer 100× (Bio-Rad)] and the AW-TEG proteins in SBI buffer supplemented with 0.0025% bromophenol blue and 0.4% BioLyte 3–10 buffer 100× (Bio-Rad) [63], both to 125 µl final volume. After homogenization under continuous agitation for 1 hour at room temperature, the samples were centrifuged at 16,000×g for 30 min. The supernatants were loaded onto 7 cm IPG strip 3–10, 3–10NL or 5–8 pH ranges (Bio-Rad) by in-gel sample rehydration. Isoelectric focusing was carried out in a Protean IEF Cell (Bio-Rad) at 20°C and 50 µA/strip. Passive rehydration was performed for 4 hours, followed by active rehydration at 50 V for 12 hours, and focalization at 500 V for 30 min, followed by 1,000 V for 30 min, 4,000 V for 1 hour and 4,000 V up to 16,000 V/h. The IPG strips were equilibrated in reducing buffer (6 M Urea, 30% glycerol, 2% SDS, 50 mM Tris-HCl pH 8.8, 0.001% bromophenol blue and 130 mM DTT) for 10 min, and in alkylating buffer containing 135 mM iodoacetamide for a further 10 min. The IPG strips and molecular weight standard were placed on top of 12% SDS-PAGE gels and sealed with 0.5% agarose. The second dimension electrophoretic protein separation was carried out using a Mini-Protean III (Bio-Rad) under 60 V constant voltage for 10 min, and then under 100 V until the dye front reached the bottom of the gel.
For each 2-DE experiment, at least two 2D-PAGEs were performed in parallel, one to be used in a Western blotting experiment and another corresponding 2D-PAGE to be stained by Colloidal Coomassie Blue G-250 for spot excision and protein identification.
The 2D-PAGEs were immediately transferred to a PVDF-based membrane (Immuno-blot 0.2 µm, Bio-Rad) using a Trans-Blot Electrophoretic Transfer Cell (Bio-Rad) at 100 V (2–3 mA cm2) for 120 min with transfer buffer (25 mM Tris-Base, 192 mM glycine, 20% methanol). The membranes were washed in water and air-dried. Before proceeding with Western blotting, the membranes were re-activated in 100% methanol and blocked for 16 hours in TBS (20 mM Tris-HCl, 500 mM NaCl, pH 7.5) containing 0.05% Tween-20 and 3% BSA (TBS-T/3% BSA) at room temperature. Each membrane was incubated separately for 2 hours with each pool of INF, NE or NI (from USA volunteers) sera diluted 1∶500 in TBS-T/1% BSA. After 2×30 min washes in TBS-T/1% BSA, the membranes were incubated with goat anti-human Ig's polyvalent antibody, HRPO conjugated (Caltag Laboratories), diluted 1∶100,000 in TBS-T/1% BSA. After 2×30 min washes in TBS-T and 1×15 min wash in TBS, the immunoreactive proteins were developed using ECL Plus Western Blotting Detection System (GE Healthcare) and the membranes were exposed for 16 hours to X-Ray film. All the 2D-WB experiments were performed in triplicate.
The X-Ray films and its corresponding Colloidal Coomassie Blue stained 2D-PAGE were overlapped. The antigenic protein spots were manually and individually excised from the corresponding 2D-PAGE for mass spectrometry identification. First, spots were washed in Milli-Q water, and then destained 2×15 min in 50% acetronitrile (ACN)/25 mM ammonium bicarbonate (AB) pH 8.0 until clear of blue stain. The gel fragments were dried in 100% ACN for 5 min, followed by rehydration in 100 mM AB for 5 min and addition of same volume of 100% ACN. The solution was removed and 100% ACN was added again. After removing the ACN, the spots were completely dried in a Speed Vac Concentrator Plus (Eppendorf) for 20 min. The final dried spots were re-swollen in 10 µl of 20 µg/ml Sequencing Grade Modified Trypsin (Promega) in 25 mM AB for 10 min and then, additional 10 µl of 25 mM AB were added. Protein digestion was conducted at 37°C for 16 hours. After the incubation, the supernatant was transferred to a clean tube and 30 µl of 5% formic acid (FA)/60% ACN were added to gel spots for the extraction of the tryptic peptides. This procedure was performed 2×30 min under constant agitation. The supernatant was pooled to the respective tube containing the initial peptide solution. This solution was dried in a Speed Vac and the peptides were resuspended in 8 µl of 0.1% FA. The peptides were desalted in reverse phase micro-columns Zip Tip C18 (Millipore), according to manufacture instructions. Peptides were dried again and resuspended in 1 µl of 50% ACN/0.1% trifluoracetic acid (TFA) solution.
MALDI-ToF-ToF analysis was performed on the 4700 Proteomics Analyzer (Applied Biosystems). Briefly, 0.5 µl of the micro-column eluate was mixed with 0.2 µl of alpha-cyano-4-hydroxycinnamic acid matrix (20 mg/ml in 30% ACN/0.3% TFA). Samples were spotted onto the ABI 192-targed MALDI plate by co-crystallization and mass spectrometry data were acquired in positive and reflectron mode, mass range 900–4,000 Da, using a neodymium-doped yttrium aluminum garnet (Nd: YAG) laser with a 200-Hz repetition rate. Typically, the analyses were conducted using 2,000 shots of MS and 4,000 shots of MS/MS to the 10 most abundant ions. External calibration was performed using a mixture of four peptides: des-Arg1-bradykinin (m/z = 904.47), angiotensin I (m/z = 1296.69), Glu1-fibrinopeptide B (m/z = 1570.68) and adrenocorticotropic hormone (18–39) (m/z = 2465.20) (mass standards kit for the 4700 Proteomics Analyzer). The list of peptide and fragment mass values generated by the mass spectrometer for each spot were submitted to a MS/MS ion search using MASCOT (Matrix Science, Boston, MA) to search in the NCBInr database. The parameters used were: allowance of two tryptic miss cleavages, peptide error tolerance of ±0.6 Da, MS/MS error tolerance of ±0.2 Da, peptide charge +1 and variable modifications of methionine (oxidation), cysteine (carbamidomethylation and propionamidation). To avoid random matches, only ions with individual score above of the indicated by the MASCOT to identity or extensive homology (p<0.05) were considered for protein identification. To those protein matches obtained from NCBInr search that did not retrieve a Smp number, a blastp search was conducted using the SchistoDB database Version 2.0 (www.schistodb.net) [67].
Hierarchical clustering was performed with the identified immunoreactive proteins according to similarities in recognition profile with each pool of serum used in this study, in three two-dimensional Western blotting assays. Recognition profile similarity was measured by the Euclidean distance, with complete-linkage among samples, using the Hierarchical Clustering (HC) algorithm available at the GenePattern Platform [68].
The S. mansoni hemoglobinase precursor (Smp_075800) and major egg antigen (Smp_049300.3) were expressed using a wheat germ cell-free expression system (TNT SP6 High-Yield Wheat Germ Protein Expression System, Promega) by coupled in vitro transcription-translation. The coding region of the corresponding genes were obtained by PCR amplification using a previously constructed S. mansoni adult worm cDNA library as template. Primers were designed using the Flexi Vector Primer Designer Tool (Promega), adding an extra C-terminal histidine tag. The primers used to amplify the coding region of major egg antigen and hemoglobinase genes were as follow: forward (5′-GGCTGCGATCGCCATGTCTGGTGGGAAACAACATA-3′) and reverse (5′-TGATGTTTAAACGTGGTGGTGGTGGTGGTGAGTAATTGCATGTTGCTT-3′), and, forward (5′-CCTGGCGATCGCCATGGTATCCGATGAAACTGTTAGTGA-3′) and reverse (5′- TGATGTTTAAACGTGGTGGTGGTGGTGGTGACCGCAAATTTTTATGATTGCT-3′), respectively. PCR reactions were composed of 25 µl JumpStart REDTaq ReadyMix Reaction Mix (Sigma-Aldrich), 2.5 µl of each gene specific forward and reverse primer (10 µM), 2 µl of S. mansoni adult worm cDNA library in 50 µl final volume. The purified PCR products were inserted into pF3A WG (BYDV) Flexi Vector (Promega) using the Flexi Vector System (Promega) and the plasmids were transformed into electrocompetent DH5α Escherichia coli cells by electroporation. Colonies were selected and grown in LB-Amp broth. Plasmids were purified using QIAprep Spin Miniprep Kit (Qiagen) and verified by DNA sequencing using the SP6 and T7 terminator primers (Source Bioscience, Nottingham, UK). Protein synthesis was initiated by adding the DNA plasmids as template according to the instructions described in the TNT SP6 High-Yield Wheat Germ Protein Expression System protocol. Protein expression was analyzed by the incorporation of labeled lysine residues (FluoroTect GreenLys, Promega) loading 5 µl of the reaction in 4–20% Mini-PROTEAN TGX Precast Gel (Bio-Rad) and detecting with a laser-based fluorescent gel scanner (Fujifilm LAS-4000 Imaging System).
For Western blotting analyzes, 3 µl of the protein synthesis reaction were blotted onto 0.45 µm nitrocellulose membranes after electrophoresis. The membranes were blocked for 16 hours in TBS-T/3% nonfat dry milk and then incubated separately for 2 hours with each pool of INF, NE or NI (from UK volunteers) sera diluted 1∶500 in a pre-adsorbed solution. These serum pools were pre-adsorbed for 16 hours in TBS-T/3% nonfat dry milk and 5% wheat germ protein extract. After 2×30 min washes in TBS-T/1% nonfat dry milk, membranes were incubated with rabbit anti-human IgG antibody (Sigma) diluted 1∶10,000 in TBS-T/1% nonfat dry milk for 1 hour and with anti-rabbit IgG HRP conjugated (ECL Plus Western Blotting Reagent Pack, GE Healthcare) in the same conditions, with 2×30 min washes between antibodies incubations. The membranes were revealed using ECL Plus-Western Blotting Detection System (GE Healthcare) and the proteins were visualized by chemiluminescence detection using a Fujifilm LAS-4000 Imaging System.
S. mansoni adult worm total and tegumental protein extracts were used initially in 2D-WB in order to evaluate the antigenicity of proteins using total immunoglobulins in pooled sera of S. mansoni infected individuals.
Firstly, Colloidal Coomassie Blue stained 2D-PAGEs were conducted using different IPG strip pH ranges (3–10, 3–10NL and 5–8) in order to visualize the separation pattern of the AW-TOT protein extract. All the different IPG strip pH ranges used showed good resolution of the spots and minimal streaking. Protein spots were reproducibly resolved in a broad pH range and molecular weight (Figure 1A).
Corresponding 2D-WB using each of the IPG strip pH ranges were performed to determine which pH range would better separate the immunoreactive protein spots when using pool of INF serum against AW-TOT protein extract. The use of IPG strip 3–10, 3–10NL and 5–8 pH ranges showed that distinct antigenic spots were evidenced when resolved by different IPG strip pH ranges (Figure 1B). Although most of the antigenic proteins were common among all the IPG strip pH range used, some proteins were exclusively identified in specific IPG strip pH range, contributing to increase the total number of identified immunoreactive proteins (Table 2).
AW-TEG protein extract was also used in an attempt to enrich the analysis with immunologically exposed parasite proteins. The AW-TEG protein extract was separated in pH 3–10 IPG strip and it was observed a distinct 2D-PAGE and 2D-WB pattern to the AW-TOT protein extract, although there are some common immunoreactive protein spots in both extracts (Figure 1).
When the 2D-WB X-Ray films and the corresponding Colloidal Coomassie Blue stained 2D-PAGEs were overlapped it was observed that there was no direct correlation between the amount of protein in the AW-TOT and AW-TEG protein extracts and its antigenicity level. Although the most of immunoreactive spots recognized by the INF serum were visible in its corresponding 2D-PAGE, there were some strongly stained protein spots that showed weak or no immunoreactivity (dashed circles in Figures 1A and 1B). Conversely, there were some barely visible protein spots in the 2D-PAGE that were highly immunoreactive (circles in Figures 1A and 1B). Most of the immunoreactive protein spots show a pI above 6.5 and molecular weight above 25 kDa.
Immunoreactive spots visualized in each of the pH ranges were excised from the corresponding Colloidal Coomassie stained 2D-PAGE for proteins identification by mass spectrometry (MS/MS). A total of 37 immunoreactive spots were excised from the 2D-PAGE using AW-TOT protein extract and pH 3–10 IPG strip. Additional 37 and 39 spots were excised from the 2D-PAGE using the pH 3–10NL IPG strip and the pH 5–8 IPG strip, respectively. From the 2D-PAGE using AW-TEG protein extract, 36 immunoreactive spots were excised. Some of the protein spots could not be identified by MS/MS. Using AW-TOT protein extract and pH 3–10 and pH 5–8 IPG strips, 7 and 14 immunoreactive spots were not MS/MS identified, respectively. Two immunoreactive spots from the 2D-PAGE using AW-TEG protein extract were also not identified (gray circles in Figure 1C).
A total of 47 different S. mansoni immunoreactive proteins were identified. Using AW-TOT protein extract 22, 29 and 18 proteins were identified from 2D-PAGE of pH 3–10, pH 3–10NL and pH 5–8, respectively. AW-TEG protein extract yielded 25 proteins identified from 2D-PAGE of pH 3–10. Most proteins were identified in more than one pH range, but others were exclusive. Additionally, 9 immunoreactive proteins were identified only in AW-TEG protein extract (Figure 2). All proteins identified by mass spectrometry in at least one 2D-WB experiment were included (Table 2). In some cases, as the result of post-translational modifications, splice variants or paralogue genes, for example, the same protein description was identified for different immunoreactive spots, but a representative gene ID was used.
In order to identify antigens differentially recognized by antibodies from pooled sera of S. mansoni infected (INF) and non-infected (NE) individuals from the endemic area, 2D-WB experiments were performed using both serum pools. A serum pool of non-infected individuals from a non-endemic area (NI) was also used. AW-TOT and AW-TEG protein extracts were focused using pH 3–10 IPG strips. Four 2D-PAGEs were electrophoresed simultaneously. Three of them were used in the 2D-WB with the three serum pools separately (INF, NE, NI). The fourth gel was stained with Colloidal Coomassie for spot excision and MS/MS identification.
Quantitative variations on the reaction intensity of the immunoreactive spots were observed among the 2D-WB using the same dilution of the three pooled sera. However, a similar overall pattern of reactive spots was observed, but with higher signal intensity when using the INF serum pool when compared to the NE, which in turn showed higher signal intensity than NI serum pool (Figure 3).
Qualitative variations of the immunoreactive spots were also visualized among the 2D-WB. Using AW-TOT and AW-TEG protein extracts it was observed that 9 spots were detectable exclusively with the INF serum pool and a single spot exclusively with the NE serum pool (Table 3). From those that reacted only with the INF serum pool, some strong immunoreactive spots of 40 kDa with approximately pI 7.0 were observed, as indicated by a circle in Figure 3. Interestingly, the corresponding spots in the 2D-PAGE were weakly stained (Figure 1A). The spots excised from this region in the 2D-PAGE were identified as major egg antigen, annexin and troponin T proteins. Two spots corresponding to major egg antigen were identified in all the triplicate assays of both AW protein extracts (spots 14 and 15 in Figure 4 and Table 3). One extra spot of the major egg antigen was identified in all triplicate assays of AW-TEG protein extract (spot 42 in Figure 4 and Table 3). The presence of major egg antigen was confirmed in all of the IPG strip pH ranges using the INF serum pool (Table 2). Annexin was identified in all triplicate assays of the AW-TEG protein extract, but major egg antigen was co-extracted in the same spot as they co-migrate (spot 41 in Figure 4 and Table 3). Two spots corresponding to troponin T were identified in the AW-TOT protein extract (spots 31 and 32 in Figure 4 and Table 3). However, they were not present in all 2D-WB triplicate experiments, two experiments showed the spot 32 and one, the spot 31 (Figure 4 and Table 3). Other two spots, also immunoreactive only with the INF serum pool, were localized above 60 kDa, within a pH range 6.0 and 7.0 in the triplicate of AW-TEG protein extract, as pointed by arrows in the Figure 3. The corresponding proteins were identified as disulphide isomerase ER-60 precursor and filamin (spots 44 and 45, respectively, in Figure 4 and Table 3). From spot 21 indicated by the arrowhead in the Figure 3, two proteins were co-excised: actin and reticulocalbin (Figure 4 and Table 3), with low pI and at approximately 45 kDa. This spot was immunoreactive in the triplicate of AW-TEG protein extract and in one of the triplicate 2D-WB experiment of the AW-TOT protein extract when probed with INF serum pool (Figure 3). The transketolase (spot 24 in Figure 4 and Table 3) was identified as an immunoreactive spot exclusively when probed with INF serum pool in one 2D-WB experiment of the AW-TOT triplicate. However, in two 2D-WB experiments of the AW-TEG triplicate this protein was immunoreactive to both INF and NE serum pools.
Only one immunogenic spot reacted exclusively with NE serum pool in two 2D-WB experiments of the AW-TEG triplicate. This spot of approximately 33 kDa and pI 8.0, as indicated by a dotted circle in Figure 3, corresponds to eukaryotic translation elongation factor (spot 40 in Figure 4 and Table 3). Although highly immunoreactive, this protein showed a low expression levels in AW-TEG protein extract (Figures 3 and 4).
All the immunoreactive proteins were clustered by Euclidean distance with complete-linkage method according to reactivity pattern against INF, NE and NI serum pools in triplicate assays with AW-TOT (Figure 5A) and AW-TEG (Figure 5B). Using the heat map representation, the proteins which reacted exclusively with antibodies from the INF serum pool were clustered, and it was highlighted that the protein recognition pattern by NE and NI serum pools is closer than the NE and INF serum pools (Figures 5A and 5B).
Two proteins identified in the 2D-WB experiments were selected for further validation of the immunoreactive pattern to the serum pools as recombinant protein expressed in a cell free in vitro system. The major egg antigen (MjE) was selected since it was identified in several spots immunoreactive only to the INF serum pool in the 2D-WB experiments (spots 14, 15, 41 and 42 in Table 3) using AW-TOT and AW-TEG protein extracts. The hemoglobinase precursor (Hem) was also selected since it was identified using all serum pools in the 2D-WB (spot 11 in Table 3).
The coding region of the genes encoding these proteins was successfully amplified by PCR using S. mansoni adult worm cDNA library as template. The amplified fragments were inserted into pF3A WG (BYDV) Flexi Vector and the sequences confirmed by DNA sequencing. The in vitro coupled transcription-translation using the wheat germ system showed to be a suitable approach for expression of these schistosome proteins. The expressed proteins were visualized in the SDS-PAGE with the expected theoretical molecular mass, 40.3 kDa for MjE and 47.7 kDa for Hem, including the 6×His-tag (Figure 6A, lanes 1 and 2, respectively). In the negative control reaction, without DNA template, two protein bands of approximately 17 kDa were observed, as well as in the reactions using the plasmids containing the coding region of MjE and Hem. These bands correspond to newly synthesized biotinylated translation products of the TNT SP6 High-Yield Master Mix, which incorporate the labeled amino acid FluoroTect GreenLys and are seen as background (Figure 6A, lane 3). In the reaction using no DNA plasmid template and no FluoroTect GreenLys no fluorescent translation products were observed (Figure 6A, lane 4).
Western blotting analysis of recombinant proteins expressed by in vitro wheat germ expression system was performed using serum pool of S. mansoni infected (INF) and non-infected (NE) individuals from endemic area and also of non-infected volunteers (NI). The recombinant proteins maintained the recognition pattern of INF and NE serum pools used in the 2D-WB experiments, confirming their correct identification and the maintenance of the antigenic epitopes in the in vitro expressed proteins. MjE remained strongly reactive when tested against the INF serum pool, and was non- or weakly reactive with NE serum pool. The Hem recombinant protein reacted with both INF and NE serum pools, however, with greater reaction intensity against the INF serum pool. Neither MjE nor Hem was recognized by antibodies present in NI serum pool. Interestingly, the NI serum pool from UK volunteers reacted with a higher molecular weight protein from wheat germ extract, even in the negative reaction control using no DNA plasmid template. This reactive band was not observed using serum pools of individuals from schistosomiasis endemic area (Figure 6B).
S. mansoni adult worms can survive for decades in the hepatic portal system of the vertebrate host in spite of the parasites being constantly exposed to the host immune system and must, therefore, display effective strategies to evade the host immune response [69]. On the other hand, it has been well demonstrated that the development of protective immunity against schistosomes depends on both humoral and cell-mediated immunity [70]. In this study, we aimed to identify parasite antigens applying an approach that would enable the screening of a large number of antibody targets using serum of individuals residing in a schistosomiasis endemic area. This approach allowed the identification of antigens associated to the disease infection or resistance status.
A number of Schistosoma serological-proteomic studies have already been performed. However, this is the first which was conducted with S. mansoni using human sera, including sera of resistant and susceptible infection individuals living in an endemic area, allowing a more rational screening for new tangible human biomarker discovery. Although these studies have searched for schistosome vaccine candidates, most of them were based on animal models, with the caveat that they may not be directly translatable to humans [26]. In addition, similar studies with S. haematobium may not correlate with S. mansoni, as there is considerable variability in immune responses to crude antigens from both parasites [71], [72].
In this study we were able to identify 47 different antigenic proteins, a slightly larger number when compared to previous S. mansoni serological-proteomic studies. This can be attributed to the use of different IPG strip pH ranges and to the fact that all experiments in this study were performed in triplicate, increasing our chances of identifying new antigens. Although previous serological-proteomic studies were performed using protein extracts from other Schistosoma species and serum sources, some proteins were commonly identified such as HSP-70, enolase, GAPDH, triose phosphate isomerase, fructose-bisphosphate aldolase, glutathione S-transferase 28 kDa, 14-3-3 protein [30], [59], [73].
Among the immunoreactive proteins several are related to housekeeping metabolic pathways, such as glycolysis, gluconeogenesis and citric acid cycle; protein synthesis and proteolysis; transport; response to stress; detoxification process; cytoskeleton organization. In addition, some proteins that have already been tested as vaccine candidates, such as triose phosphate isomerase [74], glutathione S-transferase 28 kDa [75], fatty acid-binding protein (Sm14) [13] and superoxide dismutase [76] were also identified in this study.
For this work we decided to detect total immunoglobulins from sera of the different groups of individuals, since there is no clear mechanism of immunity in human schistosomiasis that defines a class of immunoglobulin as key for an effective immune response. Although some studies indicate IgE as an important immunoglobulin in schistosome post-treatment resistance, a vaccine trial for other helminth with an antigen that induces high IgE response showed significant adverse events [17], [77]. Therefore, we performed an analysis of all immunoreactive proteins regardless of immunoglobulin isotype using polyvalent anti-human Ig. Mutapi and co-workers [60] showed that there are qualitative and quantitative differences in S. haematobium antigen recognition profiles by human antibody isotypes (IgA, IgE, IgG1 and IgG4) although the majority of the adult worm antigens were recognized by all of these four isotypes. In earlier experiments, Delgado and McLaren [78] showed that IgG1 and IgG3 were involved in protective immunity against S. mansoni infection in mice. To address this possibility, we also preliminarily assayed for reactivity using anti-IgG1 and IgG3 in our serological screening. However we could not observe any significant differences in the antigenic pattern of INF, NE and NI sera recognition (data not shown).
Tegumental protein extract of S. mansoni adult worm was used to enrich our analysis with proteins that are directly exposed to host immune response. Although a large number of proteins were identified in both AW-TOT and AW-TEG protein extracts, there are still some differences between these extracts to be explored. Using 1D SDS-PAGE and LC-MS/MS, van Balkom and co-workers [40] identified 429 proteins, from which only 43 were specific to the tegument. All the tegumental specific proteins identified in our study were also identified by these investigators, and the proteins filamin and hydroxyacylglutathione hydrolase were exclusively identified in the tegumental protein fraction in both studies. Although we have used detergent buffer to solubilize integral membrane proteins for 2-DE in the AW-TEG, we were not able to identify immunoreactive proteins with transmembrane domains, as indicated in the SchistoDB database [67]. Despite extensive research, the large-scale analysis of membrane proteins by 2-DE remains a difficult task [79]. It is critical that other methods for membrane protein extraction that allow separation by 2-DE are developed.
Serum of non-infected volunteers from non-endemic schistosomiasis area showed immunoreactivity to some protein spots of AW-TOT and AW-TEG. It has been previously shown by Losada and co-workers [80] that the sharing of molecules among organisms is an expected finding since there are several functional molecules that are conserved during the process of evolution. These molecules may elicit immune responses between different species of various genera and is responsible for antigenic cross-reactivity. According to these findings, Escherichia coli and Saccharomyces antigens induce cross-reactivity with S. mansoni crude antigens, sharing T- and B-lymphocyte epitopes [81], [82]. Nevertheless, in our study 12 schistosome specific protein spots were detected only when sera of INF and NE were used.
Although most immunoreactive spots were visible in its corresponding 2D-PAGE, some highly immunoreactive protein spots were barely visible in the 2D-PAGE. Our analysis was conducted using only adult worm protein extracts, and there is a different level of protein expression during the life cycle of the parasite [31], [83]. Major egg antigen protein spots showed to be highly immunoreactive in the 2D-WB to the INF serum pool, although poorly expressed in AW-TOT and AW-TEG protein extracts. When a comparative analysis of the S. mansoni proteome among the life cycle stages was described by Curwen and co-workers [31], the major egg antigen was among the top 40 proteins expressed in egg protein extract (SEA). Therefore, the high level of major egg antigen immunoreactivity with the INF serum is probably due to the host immune response to this highly expressed protein in the parasite eggs that is also present in other life cycle stages.
As described by Wilson and Coulson [69] a single “magic bullet” has been shown not to be an efficient target for the development of a schistosomiasis vaccine. An antigen cocktail is suggested as a way to acquire protection. In line with this principle, in the current study we were able to cluster proteins with similar immunoreactivity pattern when using serum pools of infected or non-infected individuals. We observed that the major egg antigen, annexin, troponin T, filamin, disulphide-isomerase ER-60 precursor, actin and reticulocalbin proteins reacted exclusively with serum antibodies of the infected individuals and the eukaryotic translation elongation factor with antibodies present in the serum of the non-infected individuals from endemic area.
The major egg antigen, or Smp40, has been described as highly immunogenic in humans [84]. The cytokine profile obtained by PBMC from S. mansoni infected patients stimulated with purified Smp40 was associated with a reduction of granuloma formation and anti-pathology vaccine [85]. This protein was also previously suggested as a potential antigen to be used in an immunodiagnostic test, since it was effectively immunoprecipitated by S. mansoni infected human and chronically infected mouse serum [86], [87]. As for the Smp40 protein expression across the parasite life cycle, Nene and co-workers [86] showed that Smp40 could be easily detected by Western blotting assays in adults, cercariae, schistosomulum and egg stages, when probed with serum raised against a p40 fusion protein. Furthermore, van Balkom and co-workers [40] in their S. mansoni tegumental proteomics study have identified the Smp40 in both, tegumental and stripped worms protein fractions.
Filamin proteins are mechanical linkers for actin filaments and are also involved in signal transduction and transcription [88], [89]. Filamin was previously identified in an immunoscreening of cercarial cDNA library using IgG fraction of rabbit antiserum raised against immature female worms. A polyclonal antiserum specific to recombinant S. mansoni filamin revealed a tegument associated fluorescence in adult worms that reacted mainly with a band of 84 kDa, instead of the 280 kDa as indicated by the RNA sequence [90]. The filamin protein spot that was identified in this study using S. mansoni adult worm tegument protein extract was approximately 80 kDa, suggesting that they are the same protein. Previous studies using IgG specific antibodies to the recombinant filamin showed 36.6% killing of schistosomula in vitro and in a DNA vaccine immunization filamin induced a mean of 50% protection in mouse following challenge with adult worms by surgical transfer [90], [91].
Troponin T is one of the three protein subunits of the troponin complex that mediate Ca2+-regulation that governs the actin-activated myosin motor function in striated muscle contraction [92]. Troponin T was initially suggested to be a good candidate for use in the diagnostic test for Taenia solium. However, ELISA tests using pooled sera from cysticercosis-positive and negative patients showed disappointing results [93]. Troponin T was identified by van Balkom and co-workers [40] in a non-tegumental fraction of S. mansoni adult worm protein extract. Our study is in agreement with their study since we also identified it only in total protein extract, but not in tegumental protein extract. Another muscle protein identified in our study was actin. Actin was also previously observed to be reactive to prepatent infected mouse serum [94].
Reticulocalbin is a Ca2+-binding protein. It is localized in the endoplasmic reticulum, being involved in the secretory pathway, although its detailed role remains unknown. Overexpression of reticulocalbin may also play a role in tumorigenesis, tumor invasion, and drug resistance [95]. This protein has not been previously indicated as a diagnostic test or vaccine candidate for schistosomiasis.
Immunolocalization experiments and proteomic analysis of tegumental membrane preparations confirmed that annexin is a protein localized mainly in the tegument of S. mansoni schistosomula and adult worms [36], [96]. In this study, annexin was again identified exclusively in tegumental protein extract. Schistosoma bovis recombinant annexin was shown to be biologically active in vitro, with fibrinolytic and anticoagulant properties [97].
Significant attention is being given to the excretory system of schistosomes, since accumulating evidence suggests that it plays an important role in the host-parasite interaction. The S. mansoni cysteine protease ER-60 is one of four members of the parasite disulfide-isomerase protein family. It is expressed in adult worms and larvae excretory organs, suggesting a role for ER-60 during the host-parasite interaction [98], [99]. In this study, a disulphide-isomerase ER-60 precursor was recognized by INF serum in tegumental protein extract.
The S. mansoni eukaryotic translation elongation factor was the only protein identified in this study that reacted exclusively with antibodies present in NE serum pool. In addition to its canonical function in polypeptide chain elongation, the isoform eEF1A has been associated to viral propagation, apoptosis in metazoans, cytoskeleton organization and unfolded protein degradation [100]. Moreover, elongation factor 1b/d (31 kDa) has been shown to be immunoreactive to the sera of Echinococcus granulosus infected patients [101]. In our study, the immunoreactive spot corresponding to S. mansoni eukaryotic translation elongation factor was also approximately 31 kDa.
Although we have identified only one protein that was recognized exclusively by serum of the natural resistant individuals, which represents our major candidate for a vaccine development, all other 46 proteins identified also are candidates, once they were recognized by antibodies present in serum of infected individuals. The next step would be to assess the level of protection induced by these proteins in animal models.
Expression of recombinant proteins was important to confirm the antibody reactivity pattern, mainly for the proteins that were weakly stained in the 2D-PAGE, as it was performed with the major egg antigen. In vitro expressed S. mansoni major egg antigen was strongly recognized by the INF serum pool, maintaining a similar serum recognition pattern of the native protein which indicates the correct spot excision. Anti-human IgG were used in this Western blotting experiment, which means that major egg antigen is recognized by one of the IgG subclasses. However, in 2D-WB using anti-human IgG1 and IgG3, the spots corresponding to major egg antigen were no longer detectable using INF serum pool (data not shown). Additional studies with post-treatment sera and from individuals low stool egg counts will be important to test its potential, as well as of other promising candidates, as antigens for monitoring the cure and schistosomiasis infection. This will be particularly important for those individuals living in endemic areas for the detection of low infection rates. The antigens identified in this study may also be used as an immunodiagnostic test based, for example, on a qualitative predictive model able to distinguish the clinical status of the schistosomiasis endemic area residents.
S. mansoni hemoglobinase was also in vitro expressed in wheat germ cell-free system to confirm the serum recognition pattern of the 2D-WB experiment. While hemoglobinase was found to be immunoreactive to all the pooled sera used in 2D-WB, the recombinant protein showed to be associated with the INF and NE pooled sera, but not with the NI. It may be due to the use of anti-human IgG in the Western blotting instead of anti-human Ig's in the 2D-WB experiments, suggesting a specific reaction of other immunoglobulin class in 2D-WB experiments.
Serological-proteomics is a demanding methodology with a number of arduous steps, consequently the reproducibility in 2D-WB experiments is a difficult task [102], causing missed immunoreactive spots in the triplicate experiments and among the immunoblots. Furthermore, matching the immunoreactive spots of 2D-WB to stained spots in 2D-PAGE gels used to excise the spots to MS/MS analyses is not a trivial task. To precisely locate immunoreactive spots in the 2D-PAGE gels, the molecular weight, pI and the distribution pattern of spots neighboring the spot of interest had to be taken into account. The same difficulty occurs when matching the spots among the immunoblots. It must also be taken into account that spots do not represent proteins but rather protein species with post-translational modifications, partly degraded polypeptides, or may be splicing variants or paralogues genes. Additionally, spots often contain several different proteins or protein species [102].
Despite these difficulties, serological-proteomics seems to be a good approach to characterizing the host immune response profile to parasite antigens in a large-scale analysis, overcoming the case-by-case and empirical science used in the past and providing prominent antigens for the development of new schistosomiasis diagnostics and vaccines. The manageable repertoire of S. mansoni antigens identified in this study warrants further investigation by profiling the antibody response of a larger panel of individual sera using different immunoglobulin classes/subclasses. Understanding the human immune response associated with the infection/protection profile to these antigens represents a huge step towards the improvement of diagnostic tools and development of vaccine against schistosomiasis, using not only one but multiple antigens.
|
10.1371/journal.pcbi.1004556 | FireProt: Energy- and Evolution-Based Computational Design of Thermostable Multiple-Point Mutants | There is great interest in increasing proteins’ stability to enhance their utility as biocatalysts, therapeutics, diagnostics and nanomaterials. Directed evolution is a powerful, but experimentally strenuous approach. Computational methods offer attractive alternatives. However, due to the limited reliability of predictions and potentially antagonistic effects of substitutions, only single-point mutations are usually predicted in silico, experimentally verified and then recombined in multiple-point mutants. Thus, substantial screening is still required. Here we present FireProt, a robust computational strategy for predicting highly stable multiple-point mutants that combines energy- and evolution-based approaches with smart filtering to identify additive stabilizing mutations. FireProt’s reliability and applicability was demonstrated by validating its predictions against 656 mutations from the ProTherm database. We demonstrate that thermostability of the model enzymes haloalkane dehalogenase DhaA and γ-hexachlorocyclohexane dehydrochlorinase LinA can be substantially increased (ΔTm = 24°C and 21°C) by constructing and characterizing only a handful of multiple-point mutants. FireProt can be applied to any protein for which a tertiary structure and homologous sequences are available, and will facilitate the rapid development of robust proteins for biomedical and biotechnological applications.
| Proteins are increasingly used in numerous biotechnological applications. A key property determining proteins’ applicability is their stability under operating conditions. Natural proteins can be stabilized by modification of their structure. Methods of molecular biology allow introduction of modifications–mutations–to the protein structure at will, but it is not straightforward where to mutate and which amino acid to introduce for better stability. Computational methods can be used for prediction of stabilizing mutations using computers. Current computational methods predict libraries of single-point mutations, which need to be constructed individually, tested and recombined, resulting in non-trivial experimental effort. Here we present a robust computational strategy for predicting multiple-point mutants, providing extremely stabilized proteins with a minimal experimental effort.
| Proteins are increasingly used in biotechnological applications as therapeutics [1], diagnostics [2], nanomaterials [1] and biocatalysts [3]. Despite numerous advantages, the utility of proteins is frequently restricted by their limited stability under practical conditions, such as high temperatures, extreme pH, or the presence of organic solvents or proteases. Their thermostability is usually positively correlated with stability and performance in the presence of denaturing agents [4], expression yield [5], serum survival time [6] and shelf-life [7]. Thus, it is a key determinant of proteins’ applicability in biotechnological processes. High temperatures may also be required to prevent bacterial contamination during enzymatic food processing [8]. Moreover, thermostable proteins can tolerate much larger numbers of mutations than mesophilic variants and show enhanced evolvability in protein engineering projects [9].
Protein engineering is frequently applied to obtain more stable proteins. If successful, such efforts typically enhance the melting temperature (Tm) of engineered proteins by 2 to 15°C [7, 10]. Extremely stabilized proteins with even greater increases in melting temperature (ΔTm) have been engineered by incorporating multiple mutations, and several outstanding increases of up to 35°C have been achieved using directed evolution methods [8]. However, these methods generally require extensive experiments, including screening up to 108 colonies of organisms expressing mutant variants to identify stable constructs, and appropriate high-throughput screening assays must be available [11]. A currently popular strategy is saturated mutagenesis of hotspots identified by (semi-)rational approaches [7, 8, 12], such as the most flexible residues [10]; tunnel-forming residues [13]; or residues at multimeric interfaces [14]. The selected hotspots are then subjected to site-saturation mutagenesis (while leaving the rest of the protein unchanged) to create smaller smart libraries, markedly reducing the required screening to thousands of colonies.
A long-sought alternative to screening-based approaches is reliable in silico design of stability-enhancing mutations. Numerous stable proteins have been computationally engineered via diverse approaches (singly or in combination), e.g., identification of back-to-consensus or ancestral mutations, calculation of changes in folding free energies upon mutation, introduction of disulfide bridges and elimination of highly flexible regions [7, 8, 12]. However, mutants generated using computational methods have rarely surpassed the 15°C ΔTm threshold of outstanding stabilization as a result of neutral, destabilizing or function-corrupting mutations that were predicted as stabilizing due to moderate accuracy of these methods [15, 16]. To overcome this obstacle and provide substantial stabilization, predicted mutations are usually introduced by site-directed mutagenesis and tested individually. The most viable mutations are then recombined in multiple-point mutants assuming they have additive effects, but this is often invalid due to antagonistic epistatic effects of individual mutations [17]. For all those reasons, no computational method capable of directly designing highly stable multiple-point mutants has been previously published.
Here we introduce a strategy, FireProt, for computationally designing multiple-point mutants, enabling significant improvements of protein stability with minimal experimental effort. We demonstrate its power by stabilizing the model proteins haloalkane dehalogenase (HLD) DhaA and γ-hexachlorocyclohexane dehydrochlorinase LinA. The method’s general applicability was further verified by validation against information from the ProTherm database [18], demonstrating that it can be used to identify stabilizing mutations in diverse proteins with known tertiary structures and homologous sequences allowing phylogenetic analysis, and thus should have broad utility in protein stabilization projects.
The FireProt strategy for protein stabilization is based on combining the best multiple-point mutants obtained from predictions of ΔΔG following mutation from a set of crystal structures and evolutionary information derived from multiple sequence alignment (Fig 1). Additional pre- and post-processing filters are applied in both approaches to improve prediction reliability and reduce the required computational effort.
DhaA enzyme was selected as the first model protein due to the wealth of knowledge available on mutants engineered towards higher thermostability, prolonged half-life and stability in organic co-solvents that enables quantitative comparison of their performance with the mutants designed by FireProt [13, 28].
γ-hexachlorocyclohexane dehydrochlorinase LinA enzyme was selected as the second model protein to illustrate broader applicability of FireProt strategy to other proteins of very different characteristics: (i) LinA is natively homotrimer (DhaA is monomer), (ii) LinA monomers form α+β barrel fold (DhaA possesses α/β-hydrolase fold), (iii) LinA is mainly composed of β-sheets (α-helices and β-sheets and equally represented in DhaA) and (iv) LinA is with 156 amino acids two-times shorter (DhaA has 294 residues).
Visual inspection of mutant structures coupled with detailed analysis of their individual energy terms calculated by Rosetta provided indications of the possible structural basis of protein stabilization by mutations of DhaA115 and LinA01 (S11 Table). These mutations were introduced to various locations in the protein structure with different types of secondary structures.
The last decade has seen significant advances towards more rational approaches to reduce the experimental effort required to engineer highly stable proteins (Fig 4 and S12 Table). As a contribution to these efforts we have developed a hybrid strategy integrating energy-based and evolution-based approaches, with smart filtering of mutations that are destabilizing or may impair enzymes’ functions, enabling the identification of additively stabilizing substitutions in multiple-point mutants. It is essential to correctly configure all of the tools used in both the energy- and evolution-based approaches of the FireProt workflow in order to achieve robust and reliable predictions. Therefore, individual steps of the workflow were verified using a dataset featuring diverse proteins from the ProTherm database. The predictions carried out for 656 mutations confirmed the FireProt's precision: the energy- and evolution-based approaches identified stabilizing mutations with success rates of 100% and 80%, respectively. Strikingly, only one stabilizing mutation that exceeded our thresholds was identified by both approaches, suggesting that they are highly complementary. The potential downside of the stringent conditions imposed to avoid false positives was that 92% of the available stabilizing mutations were discarded. However, the remaining correctly identified stabilizing mutations should be more than sufficient to construct highly stable catalysts (S3 Table).
When the energy-based approach was applied to DhaA and LinA enzymes, the removal of conserved and correlated positions from analysis helped to avoid modification of structurally and functionally important residues, thereby greatly reducing the number of possible mutations requiring evaluation by computationally intensive free energy calculation. Since FoldX computation is about an order of magnitude faster than Rosetta, it was applied as a pre-filter, further reducing numbers of mutations to be analyzed by Rosetta. Regarding the prediction of multiple-point mutants, simple recombination of the most promising mutants could weaken stabilization, since strong antagonistic effects were detected even at the level of double-point mutants. The thermostability enhancement for the eight- and four-point mutants predicted by this approach, DhaA112 (ΔTm 16°C) and LinA01 (ΔTm 21°C), both exceeded the threshold for outstanding stabilization, although none of the introduced mutations optimized either hydrogen bonds or charge-charge interactions. This may be due to sampling limited rotamer libraries during the calculations and the requirement for both FoldX and Rosetta to unambiguously evaluate selected mutations as stabilizing. FoldX and Rosetta employ simplified scoring functions and despite using three protein structures for analysis, only limited protein flexibility is allowed, implying that it should be possible to supplement mutations proposed by free energy calculations with beneficial substitutions identified using different principles.
To this end, additional mutations were selected by the evolution-based approach. The mutations predicted by the back-to-consensus method were filtered by FoldX to discard mutations proposed due to function-related evolutionary constraints rather than structural stabilization. This filtering step proved to be very important as over half of the mutations were discarded as potentially destabilizing. Interestingly, all five multiple-point mutants DhaA100-DhaA103 and LinA02 were predicted as destabilizing by Rosetta and had to be tested experimentally. While this prediction was accurate for three of them (DhaA100, DhaA102 and LinA02), the other two mutants (DhaA101 and DhaA103) were clearly more stable than the wild-type. This result suggests that some underlying principles important for stability detected by the back-to-consensus method are not captured by the applied Rosetta protocol. We speculate that these may include larger backbone rearrangements, interactions with ions present in the solvent, or other entropic contributions that are not well accounted for in the current protocols. Experimental characterization of these mutants by microcalorimetry, temperature-jump stopped-flow and protein crystallography is currently on-going in our laboratory. Despite its lower reliability, the evolution-based approach should still be considered as useful supplement to the energy-based approach, potentially enabling further improvement in the stability of designed proteins. The final 11-point mutant DhaA115 arising from this hybrid prediction strategy is one of the most stable HLD protein known to date (ΔTm > 24°C) [13, 30].
We have compared our strategy against several methods providing exceptional protein stabilization (S12 Table). The experimentally intensive protocols of directed evolution and hot-spot predictions can provide engineered enzymes with comparable enhancement. However, since their success rate is generally below 1%, stable proteins can only be obtained after extensive screening. Notably, two of these studies also focused on improving stability of the enzyme DhaA. In one, an eight-point mutant DhaA was obtained with a ΔTm of 18°C after screening all 121,000 possible variants [28]. We have obtained a clearly superior enzyme after experimental evaluation of as few as six mutants, highlighting the importance of removing mutations with antagonistic and uncertain stabilizing effects. In the other study performed with DhaA, four hotspots in an access tunnel were experimentally randomized, requiring experimental screening of 5,000 mutations [13], and the ΔTm for the best four-point mutant was 19°C.
Highly stable proteins have been obtained by in silico prediction of stabilizing effects of single-point mutations in four recently published studies [15, 31, 32, 33]. In one, 67 variants of epoxide hydrolase with mutations identified as potentially stabilizing by the FRESCO method were experimentally tested, 24 were reportedly more stable than the parent protein, and the variant with the best permutation of mutations had remarkably enhanced thermostability (ΔTm 36°C) [15]. Much of this enhancement arose from disulfide bridges at the dimer interface, making this approach particularly suitable for multimeric proteins. In another of the studies, four out of six engineered methionine aminopeptidases designed by the RosettaVIP method were found to be stabilizing and a combined five-point mutant reportedly had a ΔTm of 18°C [31]. The authors noted that their final construct is still less stable than the most thermostable native aminopeptidases and that the method is particularly effective for mutagenesis of buried residues around internal cavities. In the other study, a 12-point mutant of Tobacco 5-epi-aristolochene synthase was generated using the SCADS method with an impressive ΔTm (45°C), but at the expense of 98% of catalytic activity at the optimal temperature [32]. In comparison to the methods applied in these and other relevant studies (S12 Table), FireProt affords a reduction of experimental screening effort due to robust identification of stabilizing mutations and ensuring their additivity. In addition, it has promising applicability to diverse proteins, potentially all proteins with known tertiary structure and homologous sequences, due to the diverse locations of introduced mutations and universal applicability of underlying principles.
In summary, the presented hybrid strategy FireProt affords rapid design of stable proteins. Consideration of the additivity of identified potentially beneficial mutations enables prediction of multiple-point mutants with significantly enhanced stability. Despite a dramatic reduction in experimental effort, the workflow provided two proteins with outstanding stability. One of them a HLD with greater thermostability than all known HLD enzymes, either obtained from thermophilic organisms or engineered using extensive combinatorial screening. Furthermore, owing to the smart filtering, this strategy is affordable by users with limited access to powerful computer facilities. In addition, implementation of the FireProt strategy in the web-based protein engineering tool Hotspot Wizard [34] is currently on-going in our laboratory to ensure user convenience.
|
10.1371/journal.pgen.1000945 | Gene Transposition Causing Natural Variation for Growth in Arabidopsis thaliana | A major challenge in biology is to identify molecular polymorphisms responsible for variation in complex traits of evolutionary and agricultural interest. Using the advantages of Arabidopsis thaliana as a model species, we sought to identify new genes and genetic mechanisms underlying natural variation for shoot growth using quantitative genetic strategies. More quantitative trait loci (QTL) still need be resolved to draw a general picture as to how and where in the pathways adaptation is shaping natural variation and the type of molecular variation involved. Phenotypic variation for shoot growth in the Bur-0 × Col-0 recombinant inbred line set was decomposed into several QTLs. Nearly-isogenic lines generated from the residual heterozygosity segregating among lines revealed an even more complex picture, with major variation controlled by opposite linked loci and masked by the segregation bias due to the defective phenotype of SG3 (Shoot Growth-3), as well as epistasis with SG3i (SG3-interactor). Using principally a fine-mapping strategy, we have identified the underlying gene causing phenotypic variation at SG3: At4g30720 codes for a new chloroplast-located protein essential to ensure a correct electron flow through the photosynthetic chain and, hence, photosynthesis efficiency and normal growth. The SG3/SG3i interaction is the result of a structural polymorphism originating from the duplication of the gene followed by divergent paralogue's loss between parental accessions. Species-wide, our results illustrate the very dynamic rate of duplication/transposition, even over short periods of time, resulting in several divergent—but still functional—combinations of alleles fixed in different backgrounds. In predominantly selfing species like Arabidopsis, this variation remains hidden in wild populations but is potentially revealed when divergent individuals outcross. This work highlights the need for improved tools and algorithms to resolve structural variation polymorphisms using high-throughput sequencing, because it remains challenging to distinguish allelic from paralogous variation at this scale.
| Plant growth is a very complex character impacted by almost any aspect of plant biology and showing continuous variation among natural populations of a single species like Arabidopsis thaliana. Although difficult, it is important to reveal the precise genetic architecture of such a trait's variation to improve our understanding of the mechanisms and evolutionary significance of phenotypic variation. By using recombinant inbred lines derived from a cross between the reference strain ‘Col-0’ and the Irish strain ‘Bur-0’, we have localized several regions of the genome impacting plant growth. When attempting to confirm one of this region's effect, we revealed an even more complex genetic architecture where a first locus (which had remained undetected initially) has a major effect on growth only when a specific genotype was present at a second locus. We have shown here that the reason for this epistatic interaction between the two loci is that the functional allele for a gene important for photosynthesis efficiency and, consequently, growth, had been transposed from one locus to the other in Bur-0 compared to Col-0. This type of structural polymorphism seems to be frequent among strains and, although more difficult to detect, is likely to be of significant evolutionary importance.
| Natural phenotypic variation observed among different genotypes (accessions, varieties, populations, etc) is partly explained by alterations of the genetic material. Identifying the molecular basis of phenotypic variation provides candidates to test how and where in the pathway adaptation is shaping natural variation [1]. Of particular interest is the type of sequence variation from which intra-specific diversity originates. In Arabidopsis for example, there is reason to suspect that along with single nucleotide polymorphisms and short indels, structural variants in the genome as well may be an important source of natural variation [2]–[5]. Structural submicroscopic variants (smaller than those recognized microscopically) are defined as genomic alterations (insertions, deletions, inversions, duplications and transpositions) involving segments of DNA ranging from the kb to the Mb scale [6]. They can occur in genomes after large segmental duplications and subsequent gene loss, or as the result of unequal or illegitimate recombination (tandem duplications/insertions, deletions/gene loss) [7], DNA segment inversions [8] or transposable elements activity (dispersed gene duplication) [9].
Little is known about the prevalence of this phenomenon in plants and its phenotypic consequences, but it was recently found to be widespread in yeast [10] and in humans [11] where large structural variants (>50 kb) are confirmed to affect ∼4,000 genomic loci among healthy individuals [12]. Two individuals are also estimated to differ at approximately 1,000 copy-number variations (CNVs) alone [13]. In humans, structural variation may have a more significant impact on phenotypic variation than SNPs and they were implicated in gene expression variation, female fertility, susceptibility to HIV infection, systemic autoimmunity, genomic disorders… [13], [14].
Structural variation may also contribute to postzygotic isolation through the production of genetically deficient hybrids [15], as recently demonstrated among Arabidopsis strains in a study describing how the transposition of the functional copy of an essential gene (balanced structural polymorphism) results in recessive embryonic lethality in an intraspecific cross [16]. Complex structural variation at the GS-Elong locus, including tandem MAM-gene duplication, gene loss, gene conversion and CNV was shown to cause natural variation for insect herbivore resistance [17]. Another example of natural variation for disease resistance in Arabidopsis was explained by the reciprocal loss of R-gene paralogues located in an ancient segmental duplication, which resulted in certain combinations lacking either functional copy [18].
Dispersed kind of structural variation will result in epistasis in intraspecific crosses (if the functional structural variation has phenotypic consequences) and could therefore be detected in mapping populations. Although experimental mapping populations of the type of recombinant inbred lines (RILs) allow detection of genetic interactions between loci, the number of RILs observed as well as possible segregation distortions caused by lethality or reduced fitness of particular genotypes may hamper the power to detect interacting QTL [19]–[21]. In this context, the residual heterozygosity existing in some RIL sets is a plus, since deleterious genetic combinations can be restored and studied from their ‘latent’ heterozygous states.
Currently, not much interest has been manifested for the detection and consequences of structural polymorphisms in plants, probably because it is even less convenient to detect and complement than ‘simple’ coding sequence changes in a gene, for example. Although the global impact of structural variation is unknown, it might have dramatic consequences on phenotypic diversity [22]. Unfortunately, array-based re-sequencing projects are limited to this respect as they can only easily detect deletions relative to the reference sequence [5]. For A. thaliana, the short sequence reads produced by deep-sequencing on Illumina proved to support, in addition to SNPs, the detection of short to medium-size indels, and the discovery of new sequences (absent from the reference genome) but not their sequence context [2]. Therefore, nothing is obvious about the frequency of structural variation and its association with phenotype as no ideal methodologies yet allow direct detection of structural variants in plants. Here we give an example of a functional structural polymorphism resulting from the divergent evolution of duplicate genes among A. thaliana accessions.
We have used genome-wide molecular quantitative genetics to investigate natural genetic variation for shoot growth as a complex trait. Since the parental accessions were showing phenotypic differences with regard to shoot growth in our conditions, a subset of 164 Bur-0 × Col-0 Recombinant Inbred Lines optimized for QTL mapping [23] was grown and phenotyped in vitro in standard conditions in order to map loci affecting early stage shoot growth. Transgressive segregation of the shoot phenotypes observed among RILs (Figure S1A) indicates that the genetic potential for the study of shoot growth exists in this set. Indeed, four significant QTLs with LOD scores greater than 2.5 were mapped in this cross (Figure S1B).
In this work, we are now focusing on allelic variation in the genomic region underlying the QTL predicted between 14 and 15 Mb on chromosome 4. Confirmation of the phenotypic effect related to this locus was performed using specific NILs differing only for a small genomic region spanning a few cM around the QTL. NILs for this QTL were obtained by producing Heterogeneous Inbred Families (HIFs) which are easily generated taking advantage of the residual heterozygosity still segregating in F6 RILs [24], [25]. Initially, four candidate RILs (# 067, 081, 212 and 332), heterozygous only around the chromosome 4 locus, were used. In all HIFs except HIF332, the comparison of plants that were fixed for each parental allele (Col or Bur) at the QTL region revealed a highly significant difference in shoot growth (P<0.001). The HIFs[Bur] had marked phenotypic features (Figure 1) strongly reducing shoot size, as estimated either at a young stage in vitro (−25%) or at a later stage on soil (−70%), and chlorophyll content (−40% in greenhouse conditions). Then, bolting in plants carrying the Bur allele was delayed by approximately two weeks, plants were less robust with shorter inflorescence stems and yielded significantly fewer seeds than the HIFs[Col] plants, though flowers were normally developed and fully fertile. We named this QTL SG3 (Shoot Growth-3). Yet, the observed allelic effect was surprising: the Bur allele at the QTL was negatively affecting shoot size, while the QTL mapping had predicted that an opposite effect was segregating with this region (Figure S1). Moreover, HIF332 was not segregating for SG3 although its heterozygous region seemed to fully cover the QTL locus (and other positive HIFs' heterozygous regions), suggesting that the QTL is most probably involved in an epistatic interaction. The analysis of several (12) other independent RILs genotypically segregating at the locus revealed that SG3 was in complete interaction with a region at the top of chromosome 4, which we called SG3i (SG3-interactor; Figure 1). The phenotypic segregation of SG3 is conditioned by the presence of a Col allele at SG3i. Hence, with only two RILs (among our 164 subset) fixed for the combination of alleles giving rise to the defective growth phenotype, we could not have mapped this locus with the RILs and SG3 is a distinct locus than mapped initially in this region. Indeed, further analysis of HIFs genotypically segregating for the bottom of chromosome 4 region, but that do not segregate for SG3 (because harbouring a Bur allele at SG3i), showed that the locus mapped in the QTL analysis is real but distinct from SG3 and of much smaller–opposite–effect (HIF332; Figure S2). It is very likely that plants with the Col allele at SG3i and the Bur allele at SG3 (SG3i[Col]/SG3[Bur] combination) were unintentionally counter-selected during the single-seed descent process used for RIL fixation (despite the care to avoid any obvious bias) because of their retarded and small stature early after germination.
We used HIF212 to further perform the fine-mapping of SG3 by recurrent (genotypic) selection and (phenotypic) analysis of recombinants within the target interval (rHIF, see Materials and Methods), as described earlier [26]. A first set of nine recombinants were identified in the initial heterozygous region of ∼5 Mb, when fixing HIF212. Each recombinant was tested for the segregation of the phenotype and the region of interest was fast reduced to ∼1 Mb. A following screen of 600 plants resulted in the isolation of 35 recombination events in the remaining interval. Recombinants with homozygous Bur-genotype at the QTL were easily phenotyped during the selection process (the Col allele is dominant at SG3) and seven other interesting recombinants were analysed by progeny-testing. Reliable phenotyping results allowed us to further narrow down the candidate interval to ∼100 kb. Finally, a last screen of 5,000 plants, descending from one positive rHIF, provided 34 additional recombinants in the 100 kb interval. The analysis of 12 interesting recombinants delineated the region of interest to a 9 kb-interval (Figure 2), precisely 9,325 bp between bordering polymorphisms internal to the recombination events. Based on the most informative recombinants (rHIF212.60 and rHIF212.77), an ‘advanced rHIF cross’ (arHIF; see Materials and Methods and [26]) was designed in order to obtain a line (arHIF212.97) that was segregating solely for the final candidate interval. This confirmed the presence of SG3 within the 9 kb-interval when comparing arHIF lines fixed for each allele (arHIF[Col] vs arHIF[Bur]). Three predicted genes (At4g30720, At4g30730 and At4g30740) are present in this interval.
The 9 kb-interval was sequenced in both parental accessions to identify putative functional polymorphisms. The presence of “heterozygous” (double) peaks in the sequence chromatograms when amplifying from Bur-0 DNA with primers within At4g30720, was taken as predictive of a duplication of (at least part of) the candidate region. SNP information was isolated from the “heterozygous” peaks, converted into CAPS markers and the duplication was mapped thanks to the original RIL set genetic map [23] between two markers at physical positions 5,629 and 6,923 kb, i.e. in the region of SG3i.
Some major polymorphisms were identified in the SG3 candidate region when comparing Bur-0 to Col-0 (Figure 2). A ∼2 kb region including At4g30740 is absent from the Bur-0 accession. Conversely, a large (1,130 bp) insertion is present in Bur-0, 180 bp upstream of At4g30720 (it includes LTR and ADN/MuDR transposon traces and a putative target site duplication TGATG/TGATG). Also, a 1 bp-deletion in exon four of At4g30720 results in a frame shift, predicting a premature stop codon which terminates the ORF after 5 amino acids. Finally, one non-synonymous SNP is present in the coding sequence of At4g30730, changing glutamic acid (E)#66 into aspartic acid (D).
The predicted At4g30720 is encoding a 707aa-putative oxidoreductase/electron carrier detected in the chloroplast stroma [27]. It has a predicted FAD-dependent oxidoreductase domain (IPR006076), adrenodoxin reductase domain (IPR000759) and N-terminal domain (www.arabidopsis.org; www.ncbi.nlm.nih.gov). Both At4g30730 and At4g30740 putatively encode very small proteins of unknown functions, if not pseudogenes (both are partially matching neighbouring genes, respectively At4g30750 and At4g30710, and they are not supported by ESTs and expression data).
T-DNA insertions in At4g30730 (SALK_049026-promoter) and At4g30740 (SALK_057859-3′UTR) had wild-type phenotypes for shoot growth. Instead, a homozygous mutant line (SALK_059716) with a T-DNA insertion in the fourth exon of At4g30720 in a Col-0 background, phenotypically resembled plants with SG3i[Col]/SG3[Bur] allelic combination (Figure 3). In addition, RT-PCR assays using primers annealing to the exon sequences flanking the T-DNA insertion site failed to amplify any full-length transcript, indicating that this mutant is very likely a knock-out.
Quantitative complementation tests showed that the Col allele at SG3 was able to rescue the mutant allele phenotype, while, in an identical F1 genetic background, the Bur allele failed to complement the T-DNA allele (Figure 3). This indicates that the phenotype observed in arHIF[Bur] is probably due to a defect in At4g30720 and that (in the absence of any observed expression-level differences) the 1 bp-deletion found in At4g30720[Bur] is likely to be the causal polymorphism.
A Bur-0 BAC genomic library was used to sequence the duplicated copy of At4g30720 at SG3i. Although a non-synonymous SNP is causing an amino-acid change compared to At4g30720[Col], converting alanine #43 into threonine, it seems that this copy must be encoding a functional protein and explains why the Bur-0 accession itself does not have a small, pale green phenotype. Our sequencing results on the BAC identify a 10 kb-region (including At4g30720' paralogue) at SG3i clearly corresponding to the SG3 region surrounding At4g30720. Efforts have been made but we did not manage to obtain any good sequence (corresponding to any known stretch of DNA that would correspond to our reference sequence, Col-0) when sequencing from the 10 kb duplicated region toward the outside. Direct sequencing of the BAC was then impossible due to double priming which suggests the presence of repetitive DNA around the duplicated region. In addition, attempts to amplify specific loci (especially genes neighbouring the 10 kb duplicated region from SG3, like At4g30710, At4g30750, …) from the SG3i BAC gave no results, so we consider likely that the paralogues arose from a small-scale dispersed duplication event.
Overall, our results support the following: At4g30720 is responsible for the observed phenotype; two copies are present in the Bur-0 genome, one of which is not functional (at SG3); only one (functional) copy is present in the Col-0 genome (at SG3). Thus, plants that are homozygous for the Col allele at the SG3i locus and at the same time homozygous for the Bur allele at SG3, have no functional copy of At4g30720, resulting in the defective growth phenotype (Figure 4). This theory is supported by the mutant line SALK_059716 which represents an equivalent combination of alleles and by the observation of the segregation of the phenotype in other independent crosses (see below). We suggest that the transposition of the functional copy of the gene in Bur-0 (compared to Col-0) is the result of divergent evolution of the paralogues after an ancestral (dispersed) single gene duplication. This type of structural variation could have been previously underestimated as it has no phenotypic impact on the parental accessions themselves but seems to evolve quickly and could therefore represent a major source of intraspecific diversity and constraints [16].
We sequenced ∼3,400 bp of At4g30720 from 52 different accessions (essentially our core-collection of 48 [28]; Table S1) to investigate the species-wide patterns of diversity for this duplication event. The SG3i amplicons were distinguished from SG3 by a specific 7 bp-indel located 125 bp upstream of the start codon. The vast majority of accessions analysed bear two copies, while only 5 have a single copy, like Col-0. Most of the two copies-accessions seem to have both copies functional (no obvious functional polymorphisms in the coding sequence). Bur-0, Mc-0 and Fab-4 share the same polymorphism resulting in a premature stop codon in At4g30720, while Sue-0 presents a distinct 1 bp-indel causing a premature stop codon in the copy present at SG3i (Figure 4). RT-PCR analyses were performed for a subset of 26 accessions. In all two copies-accessions tested, transcripts were detected from both paralogues, with no exceptions.
In addition, crosses between Bur-like and Col-like accessions were designed in order to independently validate the candidate genetic mechanism and the functional group in which accessions belong (Figure 4). The F2 progeny of a cross between the single copy-accession Cvi-0 and Bur-0 is clearly segregating for the small, pale green phenotype in accordance with segregation of alleles at SG3/SG3i loci, indicating that Cvi-0 has a functional copy of At4g30720 at SG3 despite bearing a haplotype much closer to Bur-0 than to Col-0 at this locus (see below). Also, the F2 progeny of reciprocal crosses between Col-0 and either Mc-0 or Fab-4 (both Bur-like accessions) confirmed that a similar allelic combination is present in these Bur-like accessions. As expected, segregation of the phenotype was also observed in the reciprocal crosses between Sue-0 and Bur-0, in agreement with our sequencing results. In each case the causality link between SG3/SG3i allelic combination and the phenotype was confirmed by genotyping at these two loci. As negative controls, no phenotype was observed in the Sue-0 × Col-0, Cvi-0 × Col-0 reciprocal crosses' progeny, as well as in crosses between Bur-0 and other Bur-like accessions. The clear link observed between the segregation of diverse functional alleles at SG3/SG3i and the phenotype in distinct unrelated genetic backgrounds is another strong confirmation of the identity of the QTL.
Phylogenetic analysis, using A. lyrata as outgroup, essentially clustered the two paralogues in distinct haplotype groups. The SG3 cluster, contains most of the SG3 copies except for those from Bur-0, Fab-4, Mc-0, Sue-0 and Cvi-0 which branch together as a divergent group with a distinct haplotype (Figure 5). Overall, each of the three clusters have distinctive polymorphisms and we see no obvious genealogy between the two main clusters and either the minor cluster or A. lyrata (Table S2). It seems that the minor cluster evolved, in some way, away from the rest of the SG3 copies. Ectopic recombination and gene conversion (the extents of which remain to be shown in plants) could also participate in generating new haplotypes during duplicated gene evolution [29]. Very likely, the At4g30720 alleles found in the minor cluster derive from a common ancestor. It is nevertheless surprising to see the very divergent path followed by the five accessions bearing this haplotype, with at least one belonging to each of the first three functional groups depicted on Figure 4. The Bur-0, Fab-4 and Mc-0 SG3 copies lost their function (premature stop codon) probably very recently; while the Cvi-0 and Sue-0 copies maintained undifferentiated functions (as confirmed in the crosses) and are found associated with, respectively, a deleted and a non-functional paralogue at SG3i. Hybridization between accessions bearing different alleles could explain the diversity of allelic combinations detected at these unlinked loci.
As found previously [16], we saw no clear correlation between our haplotype network and the population structure described on a very similar accession sample [30].
Phenotypically, at the vegetative stage, SG3[Bur] plants are always pale green, with fewer leaves and rosettes about 70% smaller than the SG3[Col] plants (Figure 6A). SG3[Bur] plants contain less chlorophyll (a and b) and the chlorophyll a/chlorophyll b ratio is significantly modified (Table S4).
Transient expression assays of the GFP-fused (N-terminal) target peptide in Arabidopsis cotyledons confirmed the exclusive localization of the protein into the chloroplast (Figure 6B).
Chlorophyll fluorescence studies were undertaken to find evidence of the physiological role of At4g30720. Light-induced absorption changes in the absence/presence of DCMU and hydroxylamine (PSII inhibitors) at 520 nm measured 100 µs after a single turnover flash was employed to quantify the PSII/PSI stoichiometry in arHIFs. The PSII/PSI ratio seems not to be altered in the arHIF[Bur] compared to the arHIF[Col]. However, the arHIF[Bur] showed a lower Fmax/F0 ratio suggesting a higher F0 and, hence, a significant amount of PSII antenna not being connected to the PSII photochemical trap (Figure 6C). Together with the unaltered PSII/PSI stoichiometry, these data suggest that the overall PSII + PSI content is likely to be lower in arHIF[Bur].
In line with this hypothesis, the fluorescence changes observed during an illumination of a few minutes show that the line with a Bur allele at SG3 is marked by strongly quenched Fmax' and that the recovery of this quenching is severely delayed (Figure 6D). This may reflect the delayed activation of the Benson-Calvin cycle which would stem from the decrease in the electron flow through the photosynthetic chain.
We tend to think that it is not a specific effect on one of the photosynthetic complexes because the PSII/PSI stoichiometry is not affected. Rather, by one way or another, the main consequence on the photosynthetic chain is a decrease in the number of complexes involved in electron transfer, resulting in an overall impaired CO2 assimilation. The resulting defect in carbon metabolism could justify the delay in growth observed in the arHIF[Bur] and T-DNA mutant.
Our work highlights the very dynamic rate of evolution of duplicate genes in Arabidopsis where multiple divergent–but still functional–combinations of alleles can be fixed in different backgrounds over a limited period of time. Even genes essential for fundamental processes, like photosynthesis here, can be affected. In the present work, just as in previously described examples from Arabidopsis [16], [18], the structural variation has its origin in a duplication event followed by paralogue's extinction, a loss that can occur early in the process of duplicate gene evolution [31]. Due to the numerous large-scale segmental duplications and dispersed small-scale gene duplications, we can expect a high prevalence of this phenomenon in plants, potentially significantly impacting and constraining phenotypic variation generated at the intraspecific level.
The real extent of structural variation remains to be evaluated and sequencing projects at the species level are willing to consider this [22]. However, to enable a more comprehensive detection of structural variation contributing to intraspecific genetic diversity, significantly longer reads and paired-end sequencing–at the least–are necessary [11], [43] as well as new algorithms to analyse this data [44]–[46]. This should reveal at least CNV polymorphisms, but an extreme case of structural variation is balanced gene transposition which remains challenging to solve because it is then difficult to distinguish allelic from paralogous variation [10]. In Arabidopsis species-wide sequencing studies, one should expect to commonly face new DNA sequences, for which we have no reference and/or no idea of the insertion context, as it is clear that most Arabidopsis accessions have genome sizes 5 to 10% larger than the reference Col-0 genome [3]. A true de novo assembly of high-quality complete Arabidopsis genomes should elucidate all types of polymorphisms that can dictate natural variation.
A subset of 164 Bur-0 × Col-0 RILs (http://dbsgap.versailles.inra.fr/vnat/; [23]) optimized for QTL mapping was grown in vitro on standard media and phenotyped to map QTLs affecting early-stage shoot growth (see below). RILs 067, 081, 212, 332 that still segregated only for a limited region around 15 Mb on chromosome 4 were used to generate HIFs [24], which enabled the comparison of lines containing either of the parental alleles at the locus of interest in an otherwise identical background. The progeny of RIL212 was genotypically screened to find recombinants used in the fine-mapping process (see below). All (22) lines in the complete RIL population that were still heterozygous around 15 Mb were analysed by progeny-testing to identify and roughly map an interactor controlling SG3 phenotypic segregation. T-DNA insertion lines at At4g30720 (SALK_059716), At4g30730 (SALK_049026) and At4g30740 (SALK_057859) were ordered from NASC and grown under greenhouse conditions. The 52 accessions used to explore species-wide diversity at SG3/SG3i are listed and described in Table S1. They represent most of the core-collection of 48 accessions from the Versailles Center for Biological Resources ([28]; http://dbsgap.versailles.inra.fr/vnat/), plus a few more references and specifically selected accessions.
Seeds were surface-sterilized by soaking 10 minutes in 70% EtOH, 0.1% TritonX-100, followed by one wash with 95% EtOH for another 10 minutes. Under sterile conditions, the seeds were suspended in a 0.1% agar solution and stratified in the dark at 4°C for 4 days. Then, seeds were sown on square Petri dishes (120 mm) containing classical Arabidopsis media [47], with 9 RILs per plate and 9 seeds representing each RIL. Plants were grown for 11 days in a culture room (21°C, 16 hours light/8 hours dark cycle) where plates were rotated daily. At 10 DAG (days after germination) plantlets were carefully flattened onto the surface of the media and scanned on a flatbed scanner. Projected shoot area was measured with Optimas 6.5 as an estimate of shoot growth.
To find genetic loci that affect trait variation, average shoot areas were used as quantitative values to carry out a simple interval mapping using the WebQTL tool (www.genenetwork.org). A 1000 permutations-test was used to estimate a significance threshold at 2.5 LOD. Scanning of the genome with 2 loci mapped simultaneously including their interaction (pair-scan) was performed to search for complex epistasis between pairs of loci that could explain the trait variation.
Phenotyping for the confirmation of the QTL segregation during the fine-mapping process was performed as described above, except that seeds were not surface-sterilized and plants were grown in the greenhouse on soil. For each RIL, 48 plants were grown and genotyped to isolate individuals fixed for the parental alleles in the interval (HIF) as well as possible recombinants within the heterozygous region (rHIF). For the chosen HIF (HIF212), more recombinants were searched in two successive screens of respectively 600 and 5,000 plants. Genotyping during screens involved microsatellite or indel markers to identify recombination events within the candidate region. Once recombinants had been identified, CAPS markers and direct sequencing were used to refine and localize recombination breakpoints to smaller intervals when needed. Interesting (informative) rHIFs were then tested for the segregation of the defective growth/pale green phenotype by progeny-testing.
Advanced rHIF line 212.97 that segregate solely for the candidate region (and, hence, for the phenotype) was obtained from crosses between two different rHIFs lines with adequate genotypes (rHIFs recombined immediately to the north or immediately to the south of the SG3 final interval and with adequate genotype elsewhere), following a strategy initiated earlier [21] and as described by Loudet et al. [26].
F1 plants used for the quantitative complementation assay were generated by reciprocally crossing heterozygous arHIF212.97 to the heterozygous T-DNA insertion line, SALK_059716. F1 plants were phenotyped and genotyped to detect which allelic combination at SG3 was restoring the wild-type phenotype or not.
Bur-0 SG3 and SG3i copies were sequenced using a BAC library (Amplicon Express). Corresponding BACs were selected by PCR using primers for IND414975 amplifying a fragment of ∼1,300 bp at SG3 and respectively ∼100 bp at SG3i. BAC DNA was then extracted with the NucleoBond BAC 100 kit (Macherey-Nagel) and sent for sequencing.
For the sequencing of the 52 accessions, a specific primer (INDEL'75R) based on an indel polymorphism was used to specifically amplify a 4.5 kb-fragment covering the SG3i copy of the gene. The obtained PCR product was used as template to further PCR amplify and sequence five overlapping fragments. We inferred the SG3 sequence from an unspecific sequencing reaction by manually subtracting the SG3i polymorphisms (according to the specific SG3i sequence above), assuming none of our accessions had residual heterozygosity at these loci (which is extremely likely after the numerous SSD cycles they have been through). Primers used for sequencing are listed in Table S5. A. lyrata sequence was obtained from the Joint Genome Institute sequencing project data (http://genomeportal.jgi-psf.org/Araly1/Araly1.home.html).
At4g30720 expression in the transgenic plants was analyzed by RT-PCR. RNA was extracted with the RNeasy Plant Mini Kit (Qiagen) and the reverse transcription performed with the RevertAid First Strand cDNA Synthesis Kit (Fermentas) to yield single-strand cDNA. Transcription was tested with the following primers, forward primer 5′-CGTTTTCAACACCGCTAGAAC-3′ and reverse primer 5′- TGGTTTGTTTGCTGCTCTTG-3′, flanking the T-DNA insertion site. The adenine phosphoribosyl-transferase gene was used as control in the RT-PCR reaction.
Accessions were grown in vitro for 2 weeks then shoots were frozen and ground in liquid nitrogen. RNA was extracted and reverse-transcribed as above. PCR was carried out with the following primers, forward primer 5′- TGGTTTGTTTGCTGCTCTTG -3′ and reverse primer 5′- AACAACACGAGAATCCTCTACCA -3′, complementary to both SG3 and SG3i copies in all accessions. Amplicons generated (410 bp) were digested with BclI (Fermentas). Due to a SNP in the sequence shared by all SG3i copies among accessions tested, SG3i amplicons are digested while SG3 amplicons remain undigested, allowing to distinguish each paralogue's expression.
All accessions' sequences were aligned using CodonCode Aligner v3.01 and manually checked for all polymorphisms. The haplotype network was generated using Splitstree4 V4.10 on all sequenced accessions [48]. Population genetics analyses were generated on a subset of 41 accessions with both SG3 and SG3i functional copies, and A. lyrata ortholog was used as the outgroup in the analyses. Estimates of nucleotide diversity θπ, nucleotide divergence (Ks), Tajima's D and Fu et Li D* were performed using DnaSP5 [49]. The multilocus HKA test was performed on silent sites using a program kindly provided by J. Hey (Rutgers University, Piscataway, NJ).
The fluorescence changes were measured as described previously [50]. The fluorescence changes were induced by a continuous green light (100 µE.m2.s−1) allowing the measurement of the fluorescence yield of the dark-adapted sample (F0) and the fluorescence yield reached under quasi steady-state conditions (Fstat). The maximum fluorescence yield was measured 50 µs after applying an intense (5,000 µE.m2.s−1) light-pulse of 200 ms duration.
|
10.1371/journal.pntd.0005553 | Mycobacterium ulcerans low infectious dose and mechanical transmission support insect bites and puncturing injuries in the spread of Buruli ulcer | Addressing the transmission enigma of the neglected disease Buruli ulcer (BU) is a World Health Organization priority. In Australia, we have observed an association between mosquitoes harboring the causative agent, Mycobacterium ulcerans, and BU. Here we tested a contaminated skin model of BU transmission by dipping the tails from healthy mice in cultures of the causative agent, Mycobacterium ulcerans. Tails were exposed to mosquito (Aedes notoscriptus and Aedes aegypti) blood feeding or punctured with sterile needles. Two of 12 of mice with M. ulcerans contaminated tails exposed to feeding A. notoscriptus mosquitoes developed BU. There were no mice exposed to A. aegypti that developed BU. Eighty-eight percent of mice (21/24) subjected to contaminated tail needle puncture developed BU. Mouse tails coated only in bacteria did not develop disease. A median incubation time of 12 weeks, consistent with data from human infections, was noted. We then specifically tested the M. ulcerans infectious dose-50 (ID50) in this contaminated skin surface infection model with needle puncture and observed an ID50 of 2.6 colony-forming units. We have uncovered a biologically plausible mechanical transmission mode of BU via natural or anthropogenic skin punctures.
| Buruli ulcer is a neglected tropical disease caused by infection with Mycobacterium ulcerans. Unfortunately, how people contract this disease is not well understood. Here we show for the first time using experimental infections in mice that a very low dose of M. ulcerans delivered beneath the skin by a minor injury caused by a blood-feeding insect (mosquito) or an experimental needle puncture is sufficient to cause Buruli ulcer. This research provides important laboratory evidence to advance our understanding of Buruli ulcer disease transmission.
| Among the 17 neglected tropical diseases the World Health Organization (WHO) has targeted for control and elimination, only Leprosy and Buruli ulcer (BU) have unknown modes of transmission [1]. The search to understand how humans contract BU spans more than 70 years since the causative agent, Mycobacterium ulcerans, was first identified [2]. There are persistent and emerging foci of BU cases across the world, in particular Africa and Australia [3]. BU is characterized by necrotizing skin lesions, caused by localized proliferation of M. ulcerans in subcutaneous tissue. BU is rarely fatal, but untreated infections leave patients with significant disfigurement and disability, with damaging personal and economic consequences [4, 5]. Researchers have long been struck by the characteristic epidemiology of BU, with cases occurring in highly geographically circumscribed regions (sometimes less than a few square kilometres) and risk factors for infection that include gardening, insect bites and proximity to (but not necessarily contact with) lacustrine/riverine regions [6–14]. Human-to-human spread is considered unlikely [14]. Disease transmission is thought to occur by contact with an environment contaminated with Mycobacterium ulcerans but exactly where the pathogen resides and why it appears so geographically restricted have yet to be determined. [15].
M. ulcerans is very slow growing (doubling time >48 hrs) and this poses a problem for source tracking efforts as it is difficult to isolate the bacteria in pure culture from complex environmental specimens [16]. M. ulcerans has only once been isolated from a non-clinical source, an aquatic water bug (Gerridae) from Benin [16]. Quantitative PCR targeting M. ulcerans-specific DNA is the most frequently used technique in surveys of environmental specimens. A comprehensive review of the many field and lab studies that have examined transmission of BU has highlighted the range of organisms from aquatic plants, snails, insects, fish, amphibia, and in Australia certain native marsupials that can serve as potential reservoirs for M. ulcerans [15, 17–20]. Since the first observation that biting aquatic insects can harbour M. ulcerans [21], studies of BU transmission have largely focused on the potential for insects to biologically vector M. ulcerans implying that M. ulcerans undergoes a propagative or reproductive mode of development in an insect [22–26]. Several case-control studies, including from both Australia and Africa have suggested insects may play a role in transmission [10, 11]. For example, in southeastern Australia, we noted Buruli lesions on exposed areas likely to attract biting insects, some patients with every brief exposure times to endemic areas [27, 28] and in 2004 we began a study that identified M. ulcerans DNA associated with mosquitoes captured in endemic areas [22]. However, there is no compelling experimental evidence for single-mode biological transmission of M. ulcerans via insect vectors.
The animal ethics committee (AEC) of the University of Melbourne approved all animal experiments under approval number AEC: 1312775.2, in accordance with the National Health and Medical Research Council Australian code for the care and use of animals for scientific purposes 8th edition (2013).
M. ulcerans strain JKD8049 and bioluminescent M. ulcerans JKD8049 (harbouring plasmid pMV306 hsp:luxG13) [29, 30] were cultured in 7H9 broth or Middlebrook 7H10 agar, containing 10% oleic-albumin-dextrose-catalase growth supplement (Middlebrook, Becton Dickinson, Sparks, MD, USA) and 0.5% glycerol (v/v) at 30°C. Colony counts from bacterial cultures or tissue specimens were performed using spot plating. Five x 3μl volumes of serial 10-fold dilutions (10−1 to 10−5) of a culture or tissue preparation were spotted onto 7H10 agar plates with a 5x5 grid marked. The spots were allowed to dry, the plates loosely wrapped in plastic bags and then incubated as above for 10 weeks before counting colonies. Data analysis was performed using GraphPad Prism v 7.0a. All culture extracts were screened by LC-MS for the presence of mycolactones as previously described to ensure bacteria used in transmission experiments remained fully virulent [31].
BALB/c mice were purchased from ARC (Canning Vale, Australia) and housed in individual ventilated cages. Upon arrival, animals were acclimatizing for 5 days. Food and water were given ad libitum.
Wild caught mosquitoes were sourced from around Cairns, Queensland, Australia. A. notoscriptus and A. aegypti colonies were reared in a Physical Containment Level 2 (PC2) laboratory environment at 26°C using previously described methods, with the addition of brown paper used as the oviposition substrate for A. notoscriptus [32]. Females were aged for at least one week prior to blood feeding ensure maturity, during which they were mated (seminal accessory fluid is required to stimulate searching and feeding behavior in females). The adults were provided with access to a 10% sucrose solution, which was withdrawn 24h prior to a blood feeding experiment.
Six-week old female BALB/c mice were anaesthetized and their tails coated in a thin film of M. ulcerans wild type strain JKD8049 by dipping the tails in a Petri dish containing 20mL of bacterial culture (concentration ~106 CFU/mL) (Table 1). Tails were allowed to air-dry for 5 minutes after dipping. The tail only was then exposed to a 200mm x 200mm x 200mm cage of 20 adult female mosquitoes for a period of 15 minutes. Twenty mosquitoes were used per feeding bout to minimize stress on the mice. The number of insects biting each mouse was recorded over the exposure period by continuous observation. Tails were not wiped or disinfected post-biting. Mice were then observed weekly for up to six months for signs of tail lesions. Sterile needle stick (25G or 30G needle) and no-trauma were used as controls. An additional control consisted of tails dipped in sterile culture broth only and subjected to mosquito biting or sterile needle stick (See Table 1 for experiment design). The concentration of bacteria used was calculated by dilution plating as described above. Contingency tables and Fisher’s exact test were used to test if instances of mosquito-associated or needle stick-associated disease transmission were significantly different to no trauma (GraphPad Prism v 7.0a).
To estimate the infectious dose, we first measured the surface area of five dissected naïve mouse-tails to obtain an average surface area (493.3 ± 41.1 mm2). Using ten naïve mouse-tails and a precision balance, we then calculated the average volume of M. ulcerans 7H9 Middlebrook culture adhering to the tail surface (32.4 ± 4.2 μL), the concentration of bacteria in the cultures used, and the surface area of the tips of 25G and 30G needles used to deliver the puncture wounds (0.207 mm2 and 0.056 mm2 respectively). These parameters were then used to calculate the infectious dose, assuming the bacteria were evenly distributed over the tail surface (Fig 1B). A standard protocol to calculate an ID50 was followed. Four-week old female BALB/c mice were anaesthetized and their tails coated as described above for the transmission experiments with 10-fold dilutions of bioluminescent M. ulcerans JKD8049 from 106 to 10 CFU/mL in each Petri dish, using five mice per dilution, followed by a single needle stick puncture with a 25G needle. The mice were monitored using a Lumina XRMS Series III In Vitro Imaging System (IVIS) (Perkin Elmer). Images were captured with the following settings: binning of 4 (medium), Field of View (FOV) of 24, Relative aperture at f1.2 and exposure time of 180s. Bioluminescence was calculated using Living Image software V4.1. The number of mice per dilution developing BU up to six-months was recorded. Mice were observed up to six-months. The data were plotted and a standard curve was fitted using least squares non-linear regression. The ID50 and 95% confidence intervals were interpolated with reference to the standard curve using GraphPad Prism v 7.0a.
For each mosquito that blood-fed DNA was individually extracted from the dissected head, abdomen and legs of each insect using the MoBio Powersoil DNA extraction kit following manufacturer’s instructions (MoBio Laboratories Inc., Carlsbad CA USA). The insects were held by the wings and each body portion (legs, abdomen, head) individually and separately removed with sterile, fine forceps to avoid cross contamination during dissections. Body parts were stored individually (legs were pooled per insect) in sterile 1.5mL tubes at -20°C until DNA extraction. DNA was similarly extracted from mouse tissue. Procedural extraction control blanks (sterile water) were included at a frequency of 10% to monitor potential PCR contamination, in addition to no-template negative controls. IS2404 quantitative PCR (qPCR) was performed as described [33]. IS2404 cycle threshold (Ct) values were converted to genome equivalents (GE) to estimate bacterial load within a sample by reference to a standard curve (r2 = 0.9312, y = [-3.000Ln(x)+39.33]*Z, where y = Ct and x = amount of DNA [fg] and Z = the dilution factor]), calculated using dilutions of genomic DNA from M. ulcerans strain JKD8049, quantified using a fluorimeter (Qubit, Invitrogen) [33].
At the end of the experimental period or when a clinical end-point was reached mice were humanely killed. The region of a mouse-tail spanning a likely lesion was cut into three equal sections for histology, qPCR and CFU counts. Individual tail pieces for CFU counts were weighed and placed into sterile 2ml screw capped tube containing 0.5g of large glass beads and 600μl of sterile 1x PBS. Tissues were homogenized using four rounds of 2x 30second pulses in a high-speed tissue-disruptor at 6500 rpm, with tubes placed on ice for 5 minutes between each round. A 300μl volume of this homogenate was decontaminated with 300μl of 2% NaOH (v/v) and incubated at room temperature for 15 minutes. The preparation was neutralized drop-wise with a 10% solution of orthophosphoric acid (v/v) with added bromophenol blue until the solution changed from blue to clear. The mixtures were diluted in PBS and CFUs determined by spot plating as described above.
Sections of mouse-tails were fixed in 10% (w/v) neutral-buffered-formalin and imbedded in paraffin. Each mouse-tail was sectioned transversely (4μm thick) and subjected to Ziehl-Neelson and hematoxylin/eosin staining. The fixed and stained tissue sections were examined by light microscopy.
In experiment one, we established a murine model of M. ulcerans transmission that represented a skin surface contaminated with the bacteria and then subjected to a minor penetrating trauma, via either a mosquito bite or needle stick puncture. For this experiment, only Aedes notoscriptus mosquitoes were used, a local species previously associated with BU in south east Australia [22]. We first coated the tails of 12 mice with M. ulcerans then exposed only the tails to A. notoscriptus. Six of the 12 mice exposed to mosquitoes were bitten once each, and of these six mice, two developed BU lesions (Table 1, Fig 1B, Fig 2A). Histology of these lesions confirmed a subcutaneous focus of AFB, within a zone of necrotic tissue. There was also characteristic epithelial hyperplasia adjacent to the site of infection (Fig 2B and 2C). Material extracted from the lesions was IS2404 qPCR-positive and culture positive for M. ulcerans (S1 Table). Mice bitten by mosquitoes but with tails coated only with sterile culture media did not develop lesions (Table 1). In the same experiment, we also subjected five mice to a single needle stick puncture. Each mouse had their tail coated with M. ulcerans as for the mosquito biting. Four of these five mice developed M. ulcerans positive lesions (Table 1, Fig 2D), with subcutaneous foci of infection and viable bacteria (Fig 2F). The histology of these lesions was the same as the mice subjected to mosquito blood feeding, however bacterial burden was higher following needle stick puncture (Fig 2C compared with Fig 2F). Six mice with their tails coated with M. ulcerans but not subjected to a puncturing injury did not develop lesions and remained healthy until the completion of the experiment at six months. This experiment suggested that minor penetrating skin trauma (defined here as a puncture <0.5mm diameter and <2mm deep) to a skin surface contaminated with M. ulcerans is sufficient to cause infection. It also revealed a means by which mosquitoes could act as mechanical vectors of M. ulcerans.
In experiment two, using approximately the same dose of bacteria to coat the mouse-tails, we repeated experiment-1 but with Aedes aegypti, because of the close association of this mosquito to humans world-wide and their potential to vector pathogens. Despite multiple mosquito bites per mouse in the second experiment compared to the first, none of the five Aedes aegypti-exposed mice developed lesions (Table 1). However, as for experiment 1, four of five mice subjected to single, needle stick puncture developed M. ulcerans positive tail lesions (Table 1). A third needle stick puncture experiment was then conducted, this time using a smaller diameter, 30-gauge needle, to assess the impact of a smaller injury. There were 13/14 mice that developed BU when subjected to a single needle stick puncture through a contaminated skin surface, while eight mice with contaminated skin but no injury did not progress to disease (Table 1). Thus, across the three experiments there were 21/24 mice (88%) with needle stick puncture that developed BU, suggesting that this is an efficient mode of disease transmission (Table 1). Either A. notoscriptus mosquito bite or needle stick trauma significantly increased the risk of developing BU in our mouse skin surface contamination model (Tables 2 & 3).
We assessed the likely burden of M. ulcerans by individual IS2404 qPCR of the head, abdomen and legs for each mosquito that blood fed (Fig 3). A summary of these results is shown in Fig 3A. We noted that the bacterial load (expressed as genome equivalents [GE]) was significantly higher in the mosquito heads associated with mice that developed lesions (p<0.05) (Fig 3B, S1 Table). These data point to a threshold, above which some mosquitoes may become competent mechanical vectors for M. ulcerans transmission.
Based on the time until a tail lesion was first observed, and when using the highest concentration of bacteria (dose range 9–55 CFU, Table 1) we estimated a median incubation period (IP) of 12 weeks (Fig 4A). This result overlaps with the IP in humans for BU, estimated in different epidemiological studies from 4–10 weeks in Uganda during the 1960s [14] and 4–37 weeks in south east Australia [28]. We then estimated the infectious dose-50 (ID50). We used six different concentrations of M. ulcerans to coat the tails of mice (n = 5 mice/dilution), subjected each mouse-tail to a single needle stick puncture, and then observed the number of mice for each dilution that developed Buruli ulcer, allowing an ID50 estimate of 2.6 CFU (95% CI 1.6–3.6 CFU) (Fig 4B). To our knowledge this is the first estimate of an M. ulcerans infectious dose and indicates that like Mycobacterium tuberculosis and Mycobacterium leprae, a small quantity of this slow growing mycobacterium is sufficient to cause disease.
Pathogen transmission by arthropods is generally characterized by either biological transmission, such as malaria [34] or mechanical transmission, where replication or biological transformation of the pathogen within the vector is not necessary for disease spread [35, 36]. Here, we show for the first time an efficient mechanical mode of transmission of Mycobacterium ulcerans to a mammalian host that implicates both puncturing injuries and arthropods. In our study, the uninfected host is externally contaminated with M. ulcerans that is certainly plausible in many areas of the world. We propose that a micro-puncture wound of any sort whether it is by natural means e.g., a thorn, arthropod bite or artificially induced via a human-mediated puncture has the potential to inject M. ulcerans and generate an ulcer. This research was designed around established frameworks for implicating vectors in disease transmission and provides the necessary causational evidence to help resolve the 80-year mystery on how M. ulcerans is spread to people [15, 37]. The efficient establishment of BU we have shown here via minor penetrating trauma such as a needle puncture through a contaminated skin surface helps fulfil one of four Barnett Criteria [37]. In vector ecology, mechanical transmission is defined as a non-circulative process involving accidental transport of the pathogen [36]. That is, the pathogen, in some fashion, nonspecifically associates or contaminates the mouthparts (stylet) of an arthropod vector. Insect mechanical transmission of BU implies that if M. ulcerans were ingested and then egested via regurgitation or salivation, the mechanism would act more like a syringe than a needle [38]. Such a mode of M. ulcerans disease transmission is supported by previous laboratory studies in which Naucoris and Belostmatid water bugs were contaminated via feeding on maggot prey that had been injected with M. ulcerans or fed naturally on dietary contaminated larval mosquito prey [26, 39].
Our demonstration in the current study of mechanical transmission suggests there are potentially multiple or parallel pathways of M. ulcerans infection [37]. Examples of bacterial diseases with multiple transmission modes include tularemia, plague and trachoma [40, 41]. Support for our mechanical transmission model also comes from the many field reports over the decades of M. ulcerans infection following trauma to the skin. Case reports have noted BU following a suite of penetrating injuries ranging from insect bites (ants, scorpions), snake bite, human bite, splinters, gunshot, hypodermic injections of medication and vaccinations [42–44]. Epidemiologists in Uganda during the 1960s and 70s suggested sharp-edged grasses might introduce the bacteria [45]. However, a recent laboratory study established that abrasions of the skin in Guinea pig models and subsequent application of M. ulcerans was not enough to cause an ulcer, however, this same study established that a subcutaneous injection would cause an ulcer [46]. As a sequel to this study in Guinea pigs, we raised the question of how likely it was that mammalian skin could be sufficiently coated in M. ulcerans that an injury from natural or anthropogenic sources could lead to infection. Other explanations for the transmission of M. ulcerans include linkages with human behavior that increase direct contact with human skin and contaminated water [15]. A recent study from Cameroon recorded the persistence of M. ulcerans over a 24-month period in a waterhole used by villagers (including BU patients) for bathing [47]. A similar study in Ghana documented a 90% positivity rate for MU for water bodies frequented by community members for bathing and washing purposes [48]. Hence, it is reasonable to envisage a scenario where a villager’s skin surface becomes contaminated after bathing in such a water body and is primed for infection if (i) the concentration of bacteria is sufficiently high, and (ii) an inoculating event occurs. Whereas, in Australia, earlier studies have shown that M. ulcerans contamination of possum feces in and around the gardens of BU patients might present a similar skin surface contamination model in this region [17, 49]. Future experiments will address the possibility that insect vectors may be able to move M. ulcerans from one source and inject it into an animal or human.
Our focus on mechanical mosquito transmission arose from previous surveys in southeastern Australia where a strong association between M. ulcerans positive mosquitoes and human cases of BU has shown that M. ulcerans has not only been found on adult mosquitoes from both lab and field studies but also a biological gradient, where maximum likelihood estimates (MLE) of the proportion of M. ulcerans-positive mosquitoes increased as the number of cases of BU increased [22, 39, 50–53]. However, a recent study in Benin, West Africa found no evidence of M. ulcerans in association with adult mosquitoes [54]. The authors concluded that the mode of transmission might differ between southeastern Australia and Africa. Although, laboratory and fieldwork in West Africa suggest that aquatic insects, including mosquito larvae, play a role as reservoirs in nature for M. ulcerans that may be indirectly tied to transmission by serving as dispersal mechanisms [21, 26, 55]. Epidemiological studies have shown that direct contact with water is not a universal risk factor for BU [8, 11]. Prior exposure to insect bites and gardening are also independent risk factors for developing BU, while use of insect repellent is protective [11, 56].
Laboratory support to show mosquitoes can be competent vectors to spread BU is important additional evidence required to satisfy accepted vector ecology criteria for implicating insects in disease transmission [15, 37]. We found that infection was established following very minor penetrating trauma. Mosquitoes, in general, feed by insertion of a stylet, sheathed within the proboscis, beneath the skin of a host. The stylet has an approximate diameter 10 μM tapering to 1 μM at its tip and extending 1–2 mm below the skin surface. We estimated the density of M. ulcerans on the mouse-tails surface was 100–200 CFU/mm2. Thus, the number of bacteria potentially injected during mosquito feeding through this contaminated surface is likely to be low, but this is consistent with our infectious dose estimates from needle-stick punctures, indicating an ID50 of only 2.6 CFU (Fig 4B). Aedes notoscriptus mosquitoes are approximately twice as large as Aedes aegypti. Larger size may imply a longer stylet length, longer blood feeding time with a deeper penetration to a depth that may initiate M. ulcerans infection more frequently than that from Aedes aegypti mosquitoes that are smaller and do not blood feed as long. Such subtle differences in mosquito morphology and behavior may indicate why mechanical transmission in any form may be uncommon. There are strong parallels here with Mycobacterium leprae, the agent of leprosy. Like BU, the mode of transmission of the leprosy bacillus is unclear, but the infective dose is known to be very low (10 bacteria) and epidemiological evidence suggests multiple transmission pathways, including entry of the bacteria after skin trauma [57, 58]. Our infective dose estimate for M. ulcerans is consistent with observations that pathogens producing locally acting molecules to cause disease (e.g. the polyketide toxin mycolactone of M. ulcerans) have lower infective doses [59].
In summary, we have uncovered a highly efficient (88% rate) for needle stick skin punctures and a lower rate for mosquito punctures, suggesting a plausible mechanical transmission mode of M. ulcerans infection via anthropogenic or natural skin-puncturing microtrauma. We conclude from these experiments that reduction of exposure to insect bites, access to clean water for bathing, and prompt treatment of wounds and existing BU are concrete measures likely to interrupt BU transmission.
|
10.1371/journal.pntd.0002717 | Memantine, an Antagonist of the NMDA Glutamate Receptor, Affects Cell Proliferation, Differentiation and the Intracellular Cycle and Induces Apoptosis in Trypanosoma cruzi | Chagas' disease is caused by the protozoan parasite Trypanosoma cruzi and affects approximately 10 million people in endemic areas of Mexico and Central and South America. Currently available chemotherapies are limited to two compounds: Nifurtimox and Benznidazole. Both drugs reduce the symptoms of the disease and mortality among infected individuals when used during the acute phase, but their efficacy during the chronic phase (during which the majority of cases are diagnosed) remains controversial. Moreover, these drugs have several side effects. The aim of this study was to evaluate the effect of Memantine, an antagonist of the glutamate receptor in the CNS of mammals, on the life cycle of T. cruzi. Memantine exhibited a trypanocidal effect, inhibiting the proliferation of epimastigotes (IC50 172.6 µM). Furthermore, this compound interfered with metacyclogenesis (approximately 30% reduction) and affected the energy metabolism of the parasite. In addition, Memantine triggered mechanisms that led to the apoptosis-like cell death of epimastigotes, with extracellular exposure of phosphatidylserine, increased production of reactive oxygen species, decreased ATP levels, increased intracellular Ca2+ and morphological changes. Moreover, Memantine interfered with the intracellular cycle of the parasite, specifically the amastigote stage (IC50 31 µM). Interestingly, the stages of the parasite life cycle that require more energy (epimastigote and amastigote) were more affected as were the processes of differentiation and cell invasion.
| Trypanosoma cruzi is a parasite transmitted to mammal hosts by insect vectors known as kissing bugs. This species can result pathogenic for humans, causing Chagas' disease in the Americas. Its treatment relies on two drugs discovered more than 40 years ago. Besides their toxicity, a main drawback of these drugs is the fact that they are highly efficient only during the acute phase of the infection. But due to the lack of specific symptoms, the acute phase of the infection is largely not diagnosed. In fact, most of patients are diagnosed in the chronic phase, where the treatments are not satisfactory. In view of that, it is urgent to look for new drugs with low toxicity and able to kill the parasite in chronic patients. On the basis of previous finding, we looked for drugs against glutamate recognizing surface molecules, keeping special attention on those that are already in use in humans for other purposes (this strategy is called drug repositioning, and allow to save time and money in clinical trials: several parameters such as toxicity, pharmacokinetics, side effects in humans are already known). Here we report that Memantine, a NMDA glutamate receptors antagonist already in use to treat Alzheimer's disease, presents interesting perspectives as a trypanocidal drug.
| Trypanosoma cruzi is the etiological agent of Chagas' disease, which affects approximately 10 million people living in endemic areas of Mexico and Central and South America, with 28 million people at risk of infection [1]. T. cruzi has a complex life cycle that alternates between a reduviid insect vector and mammalian hosts (humans among them). During its biological cycle, the parasite differentiates several times between infective, non-dividing forms and dividing forms that inefficiently or are unable to infect mammalian cells. Epimastigotes, the replicative form in the insect vector, colonize the digestive tract and differentiate into metacyclic trypomastigotes, the insect-derived infective form, in the terminal midgut. During a blood meal on a mammalian host, the insects defecate and deposit these forms with the feces, which are internalized by the mammalian host and invade cells where they differentiate into the replicative amastigote stage in the cytoplasm. Amastigotes replicate by binary fission until differentiating into mammal-derived trypomastigotes, passing through a transient epimastigote-like stage [2], [3]. These trypomastigotes induce the lysis of the host cells, bursting into the extracellular milieu where they invade new cells or reach the bloodstream. The parasites disseminate throughout the infected mammal through the blood and can eventually be taken up by a new reduviid insect during a blood meal. In the midgut, the ingested trypomastigotes differentiate into epimastigotes, which replicate, thereby colonizing a new insect vector [3].
The clinical evolution of Chagas' disease in humans can be divided into two phases: acute and chronic. The acute phase is usually asymptomatic with patent parasitemia and non-specific symptoms. The chronic phase is characterized by infrequent tissue parasitism and subpatent parasitemia that persists for the life of the host. Most patients in the chronic phase (60–70%) will never develop clinically apparent disease. However, approximately 30–40% of chronic patients will develop important physiological alterations: the heart is affected, with hypertrophy and dilatation, and furthermore, the digestive tract, mainly the esophagus and large intestine, are affected, with dilatation and the appearance of megaviscera [4]–[6] as reviewed in reference [7].
Chemotherapy relies on two drugs that were discovered approximately 40 years ago: Nifurtimox and Benznidazole. Both drugs are effective for treating the acute phase of the disease. However, their efficacy in treating the chronic phase, when most patients are diagnosed, is controversial [7]. Moreover, drawbacks for both drugs have been reported, such as serious toxic side effects and more recently, the emergence of drug-resistant parasites. These facts underscore the urgent need to intensify the search for new drugs against T. cruzi [7], [8].
Our group has been exploring drug repositioning strategies, which are being widely employed for the discovery of novel therapeutic strategies to treat tropical diseases [9], [10]. This strategy seeks new uses for drugs that are already approved for the treatment of diseases in humans. Paveto and colleagues have suggested that T. cruzi epimastigotes have an N-methyl-D-aspartate (NMDA)-type L-glutamate receptor that is involved in the control of cytosolic Ca2+ levels, functionally analogous to that reported in neural cells [11]. Moreover, our group characterized a glutamate transporter [12] which is able to bind NMDA, behaving as a glutamate receptor (unpublished data). In addition, analogs of amantadine and Memantine (1,2,3,5,6,7-hexahydro-1,5:3,7-dimethano-4-benzoxonin-3-yl)amines with NMDA receptor antagonist activity were also demonstrated to have significant trypanocidal activity against Trypanosoma brucei [13]. These data led us to hypothesize that trypanocidal activities are present in compounds directed against mammalian glutamate receptors. In the present work, we tested the anti-T. cruzi activity of three compounds that have antagonistic effects on NMDA receptors: Amantadine and Memantine, tricyclic amines with low-to-moderate affinity for the NMDA receptor and used for the treatment of Alzheimer's disease [14], and MK-801, which is currently being tested in preclinical studies [15]. Memantine, an uncompetitive blocker of continuously overactivated NMDA receptors in neurons, exhibited the highest antiproliferative activity on epimastigotes and a relevant trypanocidal effect against infective forms of T. cruzi. Our experiments show that Memantine mobilizes intracellular Ca2+ and induces apoptosis, which supports the presence of a receptor with similar activity to glutamate NMDA receptors that can be used as drug targets against this parasite.
Memantine was purchased from TOCRIS; MK-801, MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltretazolium bromide) and a kit for bioluminescent somatic cells were purchased from Sigma-Aldrich (St. Louis, MO, USA). Amplex red, horseradish peroxidase, Fluo-4 AM and annexin V-FITC were purchased from Invitrogen (Eugene, Oregon, USA). Culture medium and fetal calf serum (FCS) were purchased from Cultilab (Campinas, SP, Brazil).
The Chinese Hamster Ovary cell line (CHO-K1) was cultivated in RPMI medium supplemented with 10% heat-inactivated FCS, 0.15% (w/v) NaCO3, 100 units mL−1 penicillin and 100 µg mL−1 streptomycin at 37°C in a humidified atmosphere containing 5% CO2. T. cruzi CL strain clone 14 epimastigotes [16] were maintained in the exponential growth phase by subculturing every 48 h in Liver Infusion Tryptose (LIT) medium supplemented with 10% FCS at 28°C. Trypomastigotes were obtained by infection of CHO-K1 cells with trypomastigotes as described previously [17]. Trypomastigotes were collected from the extracellular medium five or six days after infection.
T. cruzi epimastigotes in the exponential growth phase (5.0–6.0×107 cells mL−1) were cultured in fresh LIT medium. The cells were treated with different concentrations of drugs or not treated (negative control). A combination of Rotenone (60 µM) and Antimycin (0.5 µM) was used as a positive control for inhibition as previously described [5]. The cells (2.5×106 mL−1) were transferred to 96-well culture plates and incubated at 28°C. Cell proliferation was quantified by reading the optical density (OD) at 620 nm for eight days. The OD was converted to cell density values (cells per mL) using a linear regression equation previously obtained under the same conditions. The concentration of compounds that inhibited 50% of parasite proliferation (IC50) was determined during the exponential growth phase (five days) by fitting the data to a typical sigmoidal dose-response curve using OriginPro8. The compounds were evaluated in quadruplicate in each experiment. Except where otherwise indicated, for experimental purposes epimastigotes (1.0×106 cells mL−1) were cultured in LIT and treated with a concentration corresponding to the IC50 (172.6 µM) Memantine or not treated (control). Before conducting the experiments, epimastigotes were washed twice in PBS and resuspended in 50 µl of binding buffer (10 mM HEPES, 140 mM NaCl and 2.5 mM CaCl2, pH 7.4). The results shown here correspond to three independent experiments.
Parasites were treated with Memantine for four days or not treated (control). Annexin V-FITC and propidium iodide were added to the final concentration indicated by the manufacturer. The cells were analyzed by flow cytometry on a Guava cytometer (General Electric).
Epimastigote were treated with Memantine for 24 hours, washed and resuspended in PBS (5 mM succinate). The cells were incubated with 12 µM amplex red and 0.05 U mL−1 horseradish peroxidase. Fluorescence was monitored at a λexcitation of 563 nm and a λemission of 587 nm on a Spectra Max M3 fluorometer (Molecular Devices). Calibration was performed using hydrogen peroxide as a standard.
Parasites (1.0×108 cells) treated Memantine for four days were incubated with 5 µM Fluo-4 AM (Invitrogen) for one hour at 28°C. After this period, the cells were washed twice with HEPES-glucose (50 mM HEPES, 116 mM NaCl, 5.4 mM KCl, 0.8 mM MgSO4, 5.5 mM glucose and 2 mM CaCl2, pH 7.4), resuspended in the same buffer and distributed into 96-well plates (2.5×107 per well) in triplicate. Readings were performed on a Spectra Max M3 fluorometer using a λexcitation of 490 nm and a λemission of 518 nm [18].
Intracellular ATP levels were measured in treated (or not) epimastigote forms. To assess the effect of Memantine on the levels of intracellular ATP, a kit for bioluminescent somatic cells purchased from Sigma-Aldrich was used according to the manufacturer's instructions. Briefly, 50 µl of PBS was added to 100 µl of cellular ATP-releasing reagent and added to 50 µl of parasite suspension containing 5.0×106 treated or untreated (control) cells mL−1. The concentration of ATP was determined using a standard curve of different concentrations of ATP. Luminescence was obtained by the reaction between luciferase and the ATP that was released after cell lysis. Light emission levels were measured on a Lumat LB 9507 luminometer at 570 nm.
Epimastigotes (5.0×106 cells mL−1) were grown in LIT medium, transferred to Grace's medium [19] and treated or not treated (control) with 172.6 µM Memantine (IC50 value). On the sixth day, after transfer, the parasites were counted in a Neubauer chamber, and the percentage of metacyclic forms was determined.
CHO-K1 cells (5.0×105 cells mL−1) were seeded in 24-well plates in RPMI medium supplemented with FCS (10%) with different concentrations of drugs or not treated (control). Cell viability was evaluated 48 h after the initiation of treatment using the MTT assay [20]. The IC50 was determined by fitting the data to a typical sigmoidal dose-response curve using OriginPro8.
CHO-K1 cells (5.0×104 per well) were maintained in 24-well plates in RPMI medium supplemented with 10% FBS and maintained at 37°C. After 24 h, the cells were infected with trypomastigote forms (2.5×106 per well) and treated with different concentrations of Memantine (50–300 µM) for four hours. After this period, free parasites and the Memantine were removed. The infected cells were washed twice with PBS. The RPMI medium was replaced, and the plates were incubated at 33°C. Trypomastigotes were collected from the extracellular medium on the fifth day and counted in a Neubauer chamber.
CHO-K1 cells (5.0×104 per well) were maintained in 24-well plates in RPMI medium supplemented with 10% FBS and maintained at 37°C. After 24 h, the cells were infected with trypomastigote forms (2.5×106 per well) for four hours. After this period, free parasites were removed. The infected cells were washed twice with PBS, the RPMI medium was replaced, and the cells were kept in culture in the presence of different concentrations of Memantine (5–300 µM). The plates were then incubated at 33°C. Trypomastigotes were collected from the extracellular medium on the fifth day and counted in a Neubauer chamber.
CHO-K1 cells (5.0×104 per well) were maintained in 24-well plates in RPMI medium supplemented with 10% FBS and incubated at 37°C. After 24 h, the cells were infected with trypomastigote forms (2.5×106 per well) for four hours. The infected cells were washed twice with PBS, the RPMI medium was replaced, and the cells were treated at different times during invasion, after 24 h (amastigote stage) and after 60 h (epimastigote-like stage) with 31 µM Memantine (corresponding to the IC50 value obtained for the treatment of infected cells). The plates were incubated at 33°C. Trypomastigotes were collected from the extracellular medium on the fifth day and counted in a Neubauer chamber.
One-way ANOVA followed by the Tukey post-test was used for statistical analysis. The T test was used to analyze differences between groups. P<0.05 was considered statistically significant.
To investigate the possible presence of targets for mammalian NMDA glutamate receptor inhibitors, leading to a trypanocidal activity, Amantadine, Memantine and MK-801 were evaluated. A preliminary screening for the ability of these compounds to inhibit epimastigote growth was performed. T. cruzi epimastigotes were cultured in LIT medium with different concentrations of the selected drugs or no drug (control). Amantadine, Memantine and MK-801 produced a dose-dependent inhibition of epimastigote growth at 28°C and pH 7.5, the optimal growth conditions for these cells. The observed growth differences between the treated cells and the control were statistically significant (p<0.05), and the IC50 was determined to be 172.6 µM for Memantine, 300 µM for MK-801 and 451.2 µM for Amantadine (Figure 1A, 1B and 1C, respectively). In spite of being a relatively high IC50 when compared to that obtained herein for Benznidazole, (which resulted to be 7 µM, see Figure S1), the fact that Memantine is considered a safe drug for humans (few side effects have been reported) at relatively high doses (up to 20 mg/kg day), together with the facts that is commercially available and is inexpensive, led us to choice it for further study by investigating its effects on the biological processes of T. cruzi.
Programmed cell death is characterized by morphological and biochemical changes. A major change observed in cells undergoing PCD is exposure of phosphatidylserine on the extracellular face of the cytoplasmic membrane. Treated parasites were incubated with annexin V-FITC to assess external exposure of phosphatidylserine (feature of PCD) and propidium iodide to assess the possible rupture of the parasite membrane (feature of necrosis), and were further evaluated by flow cytometry. As shown (Figure 2A), untreated parasites (control) showed 10% positivity for phosphatidylserine exposure, whereas the parasites treated with Memantine showed 42% positivity (Figure 2B). Another type of necrotic process was excluded because the maintenance of parasite membrane integrity was confirmed by the absence of propidium iodide staining (Figure 2C). To confirm that Memantine induces apoptosis in epimastigotes, hallmarks for this process in trypanosomatids, such as an increase in reactive oxygen species (ROS), decreased ATP levels, increased intracellular Ca2+ levels and cell shrinkage [21]–[23], were explored. To evaluate the production of H2O2 in parasites treated with Memantine, epimastigote forms were treated with 172.6 µM Memantine (concentration corresponding to the IC50 value). After treatment for 24 hours, the parasites were incubated with amplex red and horseradish peroxidase. As observed, treated parasites produced a slightly increased amount of H2O2 than untreated parasites (Figure 3A). To determine intracellular concentrations of Ca2+, epimastigote forms were incubated with Memantine (172.6 µM) or no drug (control) for four days. After treatment, the parasites were incubated with Fluo-4 and analyzed by fluorometry. Treated parasites exhibited higher intracellular Ca2+ concentrations compared with untreated parasites (Figure 3B). The levels of intracellular ATP in treated and untreated cells were also determined using a bioluminescence assay. Intracellular ATP levels decreased in the treated parasites compared with the control (untreated parasites) (Figure 3C), indicating that the energy metabolism of the parasite is affected by the drug. Finally, we evaluated potential morphological changes in treated parasites compared with the control (Figure 4). Epimastigotes treated with Memantine exhibit dramatic changes in morphology (Figure 4C–D), presenting a characteristic rounded shape corresponding to shrinkage, a feature that is also described for apoptotic cells including trypanosomatids [23], [24]. These changes were reflected by changes on the values obtained for the forward and side light scattering (Table 1).
Because Memantine produced apoptotic activity in epimastigotes, we evaluated whether the drug could interfere with parasite differentiation. Metacyclogenesis is a well-characterized process in T. cruzi that involves transient modulation of Ca2+ levels and is dependent upon the parasite's metabolic status [25], both of which were affected by Memantine. On this basis, we evaluated the effect of Memantine on metacyclogenesis. Memantine-treated parasites sustained a 30% decrease in the number of metacyclic forms compared with the control (parasites without treatment) (Figure 5).
To evaluate the effect of treatment on the intracellular forms of the parasites, we first evaluated the toxicity of Memantine for mammalian CHO-K1 cells by MTT assay to avoid cytotoxic doses. Memantine was well tolerated by CHO-K1 cells, with an IC50 of 624.5±46 µM (Figure 6A). Based on this result, we evaluated the effect of Memantine on parasite infection using concentrations up to 0.4 mM (below the IC50 for CHO-K1 cells). To verify the effect of the drug on trypomastigote infectivity, CHO-K1 cells were infected and treated with different concentrations of Memantine (ranging from 50 to 400 µM) or not treated (control). The parasites were treated for four hours (the interval corresponding to the process of cell invasion). At all concentrations, a significant decrease in the number of trypomastigotes released from the lysed treated cells on the 5th day after infection was observed compared with the control, indicating that Memantine interferes with the infection process, and the IC50 under these conditions was determined to be 206.3 µM (Figure 6B). We also evaluated the effect of treatment after invasion of the mammalian cells by T. cruzi. All treatments produced a significant reduction in trypomastigote bursting on the 5th day after infection compared with the control (Figure 6C). This result suggests that Memantine also interferes with processes involved in the intracellular cycle. Under these conditions, the IC50 value was 31 µM, less than 20 times the IC50 for CHO-K1 cells (selectivity index: 20.13).
Given the effects of treatment of infected cells throughout the entire infection cycle, we determined which stages of the intracellular cycle (trypomastigote, amastigote or epimastigote-like) are more susceptible to treatment with Memantine. To explore this question, we took advantage of the fact that the CL14 strain produces a synchronic infection in CHO-K1 cells as previously reported [17]. In this experiment, 31 µM Memantine (concentration corresponding to the IC50 value when applied throughout the infection) was added to the infected cultures at different times: period of infection (four hours), between 24 and 60 hours post-infection (when the parasites are in the host-cells cytoplasm, as amastigotes) and between 60 and 96 hours post-infection (when most of the intracellular parasite population is at the epimastigote-like stage and differentiating into trypomastigotes). The stage most susceptible to treatment was the amastigote stage (Figure 6D), with a 35% decrease in the number of egressed trypomastigotes compared with the control.
The discovery of novel drugs for neglected diseases is a necessity for the development of more efficient chemotherapies. However, some alternative strategies should be followed in parallel to accelerate the process of optimizing the treatment of these diseases. In this sense, the search for new therapeutic uses (in this case, against T. cruzi) of well-known drugs already in use for humans (such as Memantine) may help to reduce time- and resource-consuming steps because parameters for their application in humans (such as pharmacokinetics, toxicity, maximum tolerable doses and interactions with other drugs) are already well characterized [10], [26]. Drug repositioning was the main objective of the present work.
The uncompetitive NMDA receptor antagonists Amantadine, Memantine, and MK-801, which are described in the pharmacopeia as antagonists of NMDA glutamate receptors, exhibited trypanocidal activity. These receptors have not been described in T. cruzi at the molecular level, although evidence of their existence in T. cruzi has been reported [11].
All three evaluated drugs produced a dose-dependent inhibition of proliferation and death in T. cruzi epimastigotes. Interestingly, Amantadine and Memantine, which share their basic structure consisting in a tricyclic amine (Figure 7), were the less and the more effective antagonists, respectively. The presence of two methyl groups in Memantine, which are absent in Amantadine, diminished the IC50 of the first with respect to the second by a factor of 2.5, showing that little modifications on the leader structure can result in an optimized drug. To investigate the mechanism of death, several parameters were evaluated. First, the integrity of the cytoplasmic membrane and the exposure of phosphatidylserine on the extracellular face were evaluated and strongly suggested PCD with the characteristics of apoptosis. This type of PCD has been described for unicellular protists, including T. cruzi, Leishmania and Plasmodium [27]–[30]. Similar to metazoans, apoptosis is triggered by changes in mitochondrial function. The role of mitochondria in different PCD processes including apoptosis is well characterized [21]–[23]. The production of ROS together with diminished intracellular ATP levels suggest this organelle as a main actor in gating this process. Second, increased intracellular Ca2+ levels and morphological changes were consistent with this cell death mechanism. Taken together, these results demonstrate that Memantine triggers PCD with characteristics of apoptosis.
Given that Memantine alters epimastigote physiology, we were interested in determining whether in addition to PCD, this drug may also interfere with differentiation into metacyclic trypomastigotes (metacyclogenesis). This process normally occurs in the terminal midgut of the insect vector. It is worth noting that differentiation requires initial metabolic stress conditions and is mainly energetically supported by amino acids present in reduviid urine and feces, such as proline, aspartate and glutamate [31]. These amino acids allow the parasite to reestablish the intracellular ATP levels required to energize metacyclogenesis [32]. Because Memantine reduces parasite ATP levels, we propose that the inhibition of metacyclogenesis occurs as a result of low ATP levels.
To evaluate Memantine as a trypanocidal of interest for developing new treatments against T. cruzi infection, its effect throughout the parasite life cycle in mammalian cells was evaluated. Memantine affected the infectivity of trypomastigote forms, which resulted in a reduced number of trypomastigotes bursted from infected host cells. In addition, the amastigote stage was shown to be the most sensitive stage among those infecting the mammalian cells. This is particularly interesting because amastigotes are the forms involved in maintenance of the chronic phase of infection.
Taken together, these results reveal promising prospects for a new use for Memantine, a drug that is already approved for use in humans, as an anti-T. cruzi drug. Preclinical studies are underway to support this proposal.
|
10.1371/journal.pntd.0004196 | Multiple Origins of Mutations in the mdr1 Gene—A Putative Marker of Chloroquine Resistance in P. vivax | Chloroquine combined with primaquine has been the recommended antimalarial treatment of Plasmodium vivax malaria infections for six decades but the efficacy of this treatment regimen is threatened by chloroquine resistance (CQR). Single nucleotide polymorphisms (SNPs) in the multidrug resistance gene, Pvmdr1 are putative determinants of CQR but the extent of their emergence at population level remains to be explored.
In this study we describe the prevalence of SNPs in the Pvmdr1 among samples collected in seven P. vivax endemic countries and we looked for molecular evidence of drug selection by characterising polymorphism at microsatellite (MS) loci flanking the Pvmdr1 gene.
We examined the prevalence of SNPs in the Pvmdr1 gene among 267 samples collected from Pakistan, Afghanistan, Sri Lanka, Nepal, Sudan, São Tomé and Ecuador. We measured and diversity in four microsatellite (MS) markers flanking the Pvmdr1 gene to look evidence of selection on mutant alleles.
SNP polymorphism in the Pvmdr1 gene was largely confined to codons T958M, Y976F and F1076L. Only 2.4% of samples were wildtype at all three codons (TYF, n = 5), 13.3% (n = 28) of the samples were single mutant MYF, 63.0% of samples (n = 133) were double mutant MYL, and 21.3% (n = 45) were triple mutant MFL. Clear geographic differences in the prevalence of these Pvmdr mutation combinations were observed.
Significant linkage disequilibrium (LD) between Pvmdr1 and MS alleles was found in populations sampled in Ecuador, Nepal and Sri Lanka, while significant LD between Pvmdr1 and the combined 4 MS locus haplotype was only seen in Ecuador and Sri Lanka. When combining the 5 loci, high level diversity, measured as expected heterozygosity (He), was seen in the complete sample set (He = 0.99), while He estimates for individual loci ranged from 0.00–0.93. Although Pvmdr1 haplotypes were not consistently associated with specific flanking MS alleles, there was significant differentiation between geographic sites which could indicate directional selection through local drug pressure.
Our observations suggest that Pvmdr1 mutations emerged independently on multiple occasions even within the same population. In Sri Lanka population analysis at multiple sites showed evidence of local selection and geographical dispersal of Pvmdr1 mutations between sites.
| Chloroquine combined with primaquine has been the recommended antimalarial treatment for Plasmodium vivax malaria infections for sixty years but the efficacy of this treatment regimen is threatened by chloroquine resistance. In this study we describe the prevalence of mutations in the P. vivax gene, Pvmdr1 among samples collected in seven endemic countries. The mutations are thought to be associated with chloroquine resistance and here we looked for evidence of drug selection by characterising polymorphism in DNA repeat regions (microsatellite (MS) loci) flanking the Pvmdr1 gene. Mutations in the Pvmdr1 gene were mainly identified at codons T958M, Y976F and F1076L. Just 2.4% of samples were wildtype at all three codons, while 63% were single mutants (MYF). Clear geographic differences in the prevalence of these Pvmdr mutation combinations were observed. At the flanking MS loci, we found high levels of diversity, and significant differentiation between geographic sites. This pattern of variation could indicate directional selection through local drug pressure. In summary, our observations suggest that Pvmdr1 mutations and thus, chloroquine resistance has emerged independently on multiple occasions even within the same population.
| Malaria is one of the world’s leading causes of mortality and morbidity. Since late in the 1940s the antimalarial drug chloroquine (CQ) has been the primary chemotherapeutic for prophylaxis and treatment of malaria because of its good safety profile, low cost and high efficacy against the blood stages of CQ sensitive (CQS) Plasmodium parasites, causing the disease. Since the 1950`s, CQ treatment of P. vivax infections has been combined with the hypnozoitocidal drug primaquine (PQ) for clearance of the latent P. vivax liver stages, responsible for later relapses of the disease [1–3]. Compared to P. falciparum, development of CQ resistance (CQR) in P. vivax has been relatively slow with the first reports emerging in 1989 in Papua New Guinea (PNG) [4]. Since then, CQR has spread and today it is considered to be present in vivax-malaria endemic countries all over the world (reviewed in [5] and more recently in [6]). Development of CQR has been slower in P. vivax than P. falciparum and this is sometimes attributed to the use of combined treatment (CQ with PQ), where PQ acts synergistically with CQ against CQR parasites [7]. It is also proposed that CQR in P. vivax has a different CQR mechanism than P. falciparum [8].
Knowledge of the mode of action of CQR in P. vivax is limited. In P. falciparum, reduced CQ sensitivity is strongly associated with single nucleotide polymorphisms (SNPs) in the chloroquine resistance transporter-gene, Pfcrt [9;10]. However, studies of the Pfcrt orthologue in P. vivax, Pvcg10, have not been able to find an association to CQR [8;11]. In P. falciparum is the P-glycoprotein–like molecule Pgh-1 encoded by Pfmdr1, is also associated with CQR though it may only modulate the effects of the Pfcrt gene [12;13]. In 2005, Brega et al. characterized the mdr-like gene Pvmdr1 in P. vivax isolates [14], and evidence suggests that SNPs in the Pvmdr1 gene are a possible genetic determinant of CQR [11;14;15]. Cross-species comparisons led the focus in P. vivax to be primarily on the mdr-codons orthologous to codons implicated in P. falciparum CQR namely 86, 184, 1034, 1042 and 1246 [14;16]. However codons 91 and 189 which are homologous to codons 86 and 184 in P. falciparum [14;17] and codons 1071 and 1079 which are homologous to codons 1034 and 1042 in P. falciparum [14] are rarely polymorphic in Pvmdr1 Instead SNPs at codons, 976 and 1076, have been detected multiple times [11;14;15;18;19]. Suwanarusk et al. observed an association between the Y976F mutation and increased CQ IC50 in samples from Thailand and Papua province of Indonesia, and stated that the Y976F mutation is a useful tool to indicate foci of chloroquine resistance [11]. Others detect the mutations, but doubt their association with CQR [15;18;20–23].
Studies of P. falciparum drug resistance loci have used flanking microsatellite (MS) variation to describe selective sweeps around Pfdhfr [24] and Pfdhps [25] and Pfcrt [26]. This approach has revealed lineages of drug resistance mutant alleles which are derived from a single emergence event. Notably, in P. falciparum some resistance lineages were found to have spread across vast geographical distances [26;27]. When the same approach was repeated antifolate drug targets in P. vivax Pvdhfr [28] and Pvdhps [29] contrasting results were found. In Pvdhfr and Pvdhps there was evidence of multiple independent mutation events with little selective sweep around mutant alleles at those loci. This result may reflect the limited antifolate drug selection pressure that has historically been applied to P. vivax, or it may point to differences in transmission and selection dynamics in the two species. In this study we looked for evidence of drug selection on the CQR candidate Pvmdr using MS flanking the Pvmdr1 gene.
We analysed samples collected from Pakistan, Afghanistan, Sri Lanka, Nepal, Sudan, São Tomé and Ecuador. In Pakistan approximately 83% of the malaria cases are caused by P. vivax and in Afghanistan, 95% are P. vivax. In both countries P. vivax infections are still being treated with CQ + PQ [30]. Around 83% of reported malaria infections in Sri Lanka were caused by P. vivax and CQ with PQ were still efficient and recommended treatment of P. vivax infections on the island until autochthonous cases of both P. falciparum and P. vivax in Sri Lanka fell to zero [30]. In Nepal 88% of the malaria cases are caused by P. vivax and treated with CQ + PQ [30]. In Ecuador, P. vivax accounts for 86% of all malaria infections and is treated with CQ + PQ [30]. To the best of our knowledge no cases of CQR have been reported from either Nepal or Ecuador. Only, 5% of the malaria infections in Sudan are caused by P. vivax and these are treated with artemether-lumefantrine + PQ [30]. No reports of P. vivax CQR have been published from Sudan. Falciparum malaria is the main cause of malaria in São Tomé and no recommendations are provided regarding treatment of P. vivax [30].
The objectives of the present study were to 1) Determine the diversity of SNPs in the Pvmdr1 gene, a putative marker of CQR, in P. vivax samples collected from Pakistan, Afghanistan, Sri Lanka, Nepal, Ecuador, Sudan and São Tomé and 2) Characterise flanking MS variation and use this to explore the evolutionary origin of Pvmdr mutations.
The total number of P. vivax samples analysed in this study was 267. The samples originated from 7 countries: Pakistan (n = 36), Afghanistan (n = 13), Nepal (n = 55), Sri Lanka (n = 136), Ecuador (n = 17), São Tomé (n = 4) and Sudan (n = 6). The samples were all (with the exception of Sao Tomé, and Sudan) derived from larger sets of PCR positive samples and the subset selected by computer randomisation. The DNA from a total of 263 P. vivax samples selected were already extracted as part of a P. vivax microsatellite study described previously [31].
Pakistan and Afghanistan: Forty-nine samples from a cluster of neighbouring sites in Pakistan and Afghanistan were analysed. Thirty-six were from Pakistan (n = 36); Ashaghroo refugee camp in Adizai from 2003 (n = 10), Adizai Refugee Village in Peshawar from 2004–2005 (n = 3) sampled as part of another study [32] and lastly Adizai, Baghicha and Khagan villages located near Peshawar 2005–2006 (n = 23) described in [33]. Thirteen samples were from Afghanistan collected at the Malaria Reference Center in Jalalabad in 2004–2005 [32]. The samples from all these sites were grouped together because of similar study designs and close geographical distance between the sites.
Sri Lanka: The samples from Sri Lanka were collected in 9 malaria endemic districts during 2002–2007, see [34]. For this study, the samples were divided into 9 groups of districts, where after randomised computerisation was used to select samples from each district (N = 136).
Nepal: Samples from two separate studies in Nepal were grouped together. Thirty-eight samples collected in 2009–2010 from the districts of Jhapa (N = 34) and Banke, Chitwan and Dang (N = 5) have been previously described in [35]. The other study collected samples in the districts of Kanchanpur (N = 5) and Jhapa (N = 12) in 2005–2006 as a part of a cross-sectional prevalence survey estimating the malaria burden and risk behaviour in two endemic districts of Nepal (S. Hewitt, personal communication). The Kanchanpur samples were grouped with Banke, Chitwan and Dang.
Ecuador: Twenty-one P. vivax samples were collected from 2007–2009 in the Province of Sucumbíos through the network of laboratories of the Ministry of Health.
Sudan: Six P. vivax samples from Sudan were collected in the village of Asar in Gedaref state in 2006 as a part of an artemether-lumefantrine efficacy trial community based survey [36]. The amount of available P. vivax DNA was small, and only limited analysis was possible.
São Tomé: The island São Tomé is a part of the Democratic Republic of São Tomé and Príncipe in the western equatorial coast of Central Africa. Four samples were available. As with the samples from Sudan, limited analysis was possible because only a small amount of extracted DNA-solution was available.
A fragment spanning nucleotide 2596 and 3532 (amino acids 865–1177) of the Pvmdr1 gene and was amplified using semi-nested primers Pvmdr1-4F [11], Pvmdr1-AS and Pvmdr1-S [14]. Primer sequences are shown in Table 1. Cycling conditions were as follows: 94°C for 15 min, followed by 30 cycles of 94°C for 30 s, 55°C for 1 min, and 72°C for 1 min, and finally 72°C for 10 min. The amplified Pvmdr1 fragments were sequenced on an ABI Prism 377 (Perkin-Elmer) using the Big Dye terminator reaction mix (Perkin-Elmer). After sequencing, the individual haplotypes were aligned and analysed by use of the DNASTAR-Lasergene software.
The Pvmdr1 gene is located on chromosome 10, and sequences flanking the gene were screened for suitable microsatellite marker loci in the Salvador-1 (Sal-1) reference strain (accessed through the European Bioinformatics Institute homepage (www.ncbi.nlm.nih.gov/). Multiple repeats were identified using the software Tandem Repeats Finder [37] and semi-nested primers designed (Table 1). The primary reaction comprised of 1μl template, 0.5 unit Taq polymerase, 1.1μl Thermopol Reaction buffer (New England Biolabs Inc, Glostrup, Denmark), and 0.4μM dNTPs, 0.1μM of forward (F) and reverse primers (R) with cycling conditions as follows; 2 min at 94°C and then 25 repeated cycles of 30 s at 94°C, 30 s at 42°C, 30 s at 40°C and 40 s at 65°C followed by 2 min at 65°C and a minimum of 10 min at 15°C. In the secondary PCR, the same concentrations of reagents were added, but with 0.15μM of reverse primers (R) and fluorescent-labelled inverse primers (I). The cycling conditions were initiated with 2 min at 94°C followed by 25 repeated cycles of 20 s at 94°C, 20 s at 45°C, 30 s at 65°C, and finished with 2 min at 65°C and 10 min at 15°C.
A panel of four MS were selected for further analysis on the basis of their successful amplification and were named according to their distance to the Pvmdr1 gene: m9.5 (9,489 bp downstream of Pvmdr1), m10.1 (10,120 bp downstream), m10.4 (10,420 upstream) and m43.1 (43,168 bp upstream), (Fig 1). PCR amplified fragments were run with LIZ-500 size standard on an ABI 3730XL genetic analyzer (Applied Biosystems), and analysed using the GeneMapper software (Applied Biosystems). Samples that were negative by PCR were repeat amplified with 2μl template in the first PCR.
The number and length of the repeats in each of the four MS is shown in Table 1. In the cases where multiple (≥ 2) microsatellite alleles were detected in a single sample the major/predominant allele chosen, (‘predominant’ is defined by the electropherogram peak height which had to be twice that of the minor allele).
Linkage disequilibrium (LD) was calculated to test for a non-random association of Pvmdr1 allele and the MS alleles. Only the non-mixed samples were used in the calculations. LD was measured by the formula D' = D/Dmax, where D equals derivation of random association between alleles at different loci, and D’ measures D standardized by the maximum value (Dmax), given the observed allele frequencies. LD values range from -1 to +1, where the value +1 refers to a complete non-random association between the alleles. Values of gene diversity were calculated by expected heterozygosity by the formula He = (n/[n-1])(1-∑pi2), by use of the Arlequin software [38], where He is expected heterozygosity, n the number of samples, and pi the frequency of the i-th allele in the sample set.
Clearance for analysis of Plasmodium genes were approved by London School of Hygiene Tropical Medicine and Hygiene Ethics Board, locally by Bioethics Committee, Pakistan Medical Research Council and Directorate of Public Health, Jalalabad, Nangahar, Comitee de Bioetica Universidad San Francisco de Quito, Committee on Research and Ethical Review at the Faculty of Medicine, Peradeniya, Kandy and the Nepal Health Research Council. All data analysed were anonymized.
The analysis of Pvmdr1 in 39 samples from Nepal was previously published [35]. Of the remaining 228 samples, sequencing was successful in 173 (75.9%); Pakistan (n = 24), Nepal (n = 4), Sri Lanka (n = 120), São Tomé (n = 3), Sudan (n = 4) and Ecuador (n = 17). Combined with the 39 samples from Nepal, 212 Pvmdr1 sequence fragments were available for further analysis (Table 2, Fig 2).
SNP variation in fragment 2 was largely confined to three codons; T958M (ACG→ATG), Y976F (TAC→TTC) and F1076L (TTT→CTT), (Fig 1). Three novel SNPs were detected in two samples from the Jhapa district in Nepal. These were sequenced twice to confirm the results. One of the samples carried a SNP at c1080 (S1080N, AGT→AAT), while the other sample possessed SNPs at c979 (F979S, TTT→TCT) and c980 (M980V, ATG→GTG) (Table 3). These two samples and the SNPs were described by Ranjitkar et al. [35].
The substitutions at codons 958,976, and1076 were found in various configurations. TYF (the wild-type) was found only in Ecuador in 5 of 17 samples (Fig 2, S1 Table). All the remaining samples carried one of three mutant haplotypes, MYF (single mutant, T958M), MYL (double mutant, T958M and F1076L) and MFL (triple mutant T958M, Y976F and F1076L). The double mutant MYL was present in 63.0% of the samples (n = 133), the triple mutant MFL in 21.3% (n = 45), and the single mutant MYF in 13.3% (n = 28). Their relative abundance at the different sites is shown in Fig 2. In Pakistan only the MYL double mutant haplotype was detected, while the Ecuador samples (apart from the wild-type TYF) possessed the single mutant MYF (n = 12). The Sudan P. vivax samples a mix of MYL (n = 2) and MFL (n = 2) haplotypes were found, while the three samples from São Tomé all possessed the MFL haplotype (Fig 2).
The heterozygosity of Pvmdr1 measured as He is shown in Table 2. Measured over all samples He was 0.54. When divided by collection site, Sri Lanka was the most diverse (He = 0.54), and Pakistan the least diverse (He = 0) with only one allele-the MYL haplotype. Although the He value for Sudan was high (0.67), the sample size was small (n = 4), and the broad variation in sample size between sites precluded further in-depth analysis of difference between the sites.
District level analysis was possible for Sri Lanka (Fig 3). In Sri Lanka, samples were collected in 9 districts, and despite the small sample size per district, the distribution of alleles was similar at district level when compared to the pooled sample set, with a dominating MYL haplotype, followed by MFL and MYF. The exception was Kurunegala district where the MFL haplotype was most common (n = 15), followed by the MYL haplotype (n = 5).
Four MS from Pvmdr1 flanking genomic regions were genotyped in the 267 samples although these were amplified with varying successes; for m9.5 (n = 190), m10.1 (n = 196), m10.4 (n = 229) and m43.1 (n = 181) alleles (Table 2). The m10.1 locus was the most polymorphic with 26 alleles identified among 196 samples (He = 0.90), while m9.5 had 4 alleles, m10.4 had 13 and m43.1 had 3 different alleles (Table 2, and S1 Table). The number of mixed samples (those containing more than one allele) detected using each locus differed; only 1% were mixed at the m9.5 locus, while 12% were mixed at the m43.3 locus. When combining all 5 loci, 26% were mixed among 125 samples (Table 2).
The m10.1 locus had a mono-A-repeat as well as an AT dinucleotide repeat, and this was reflected in the high number of different alleles (He = 0.90). Ecuador was an exception, only possessing 4 different m10.1 alleles among the 16 positive samples (He = 0.68). Allele size variation is shown in S1 Table, the highest number of observed MS alleles were generally of intermediate size.
The combination of all 5 loci resulted in 57 different 5-loci haplotypes among the monoclonal samples (Table 3) and 78 when majority alleles in the mixed genotype samples were included. All 5-locus haplotypes differed from the CQS wild-type Sal-1 haplotype. The most commonly observed 5-locus haplotype was detected in Ecuador (n = 6, haplotype number 8 in Table 3). The other frequently observed 5-loci haplotypes were all from Sri Lanka. The distribution of the Pvmdr1 5- locus haplotypes among the sub-populations sampled in Sri Lanka is shown in Fig 4. A large number of haplotypes occurred only once and these are indicated by grey segments in the pie charts. Haplotypes which occurred multiple times are indicated each by a different colour. The sample collections with the greatest degree of haplotype sharing were from Trincomalee and Anuradhapura (6 haplotypes) and Anuradhapura and Polonaruwa (5 haplotypes of which only 1 was found in Trincomalee). Other sites appear more isolated, for instance all the haplotypes found in the district of Batticaloa were unique.
Unfortunately, the small amount of DNA-solution available from Sudan and São Tomé prevented reanalysis of these samples, and no 5-loci haplotype from Sao Tomé could be created.
We tested for LD, between Pvmdr1 alleles and flanking MS alleles. Significant associations are shown in S2 Table. Strong linkage associations were seen in Ecuador where MS alleles occurred in association with the MYF single mutant allele and also with the wildtype allele TYF. Other population level LD associations were observed in Sri Lanka, and Nepal. In each case different MS alleles were associated with the Pvmdr1 allele. When the 4 MS loci were combined, significant LD between certain 4-loci haplotypes and either the TYF, MYF or MFL haplotypes was found in Ecuador and Sri Lanka, (S2 Table). When samples from all sites were pooled the LD analysis found significant associations of MS alleles with TYF and MFL which are likely attributable to admixture.
The aim of this study was to characterise SNPs in the Pvmdr1 gene, to examine whether the putative CQR mutations have one, few or many origins, and to determine whether there has been geographical spread of certain Pvmdr1 haplotypes. SNPs were found at three codons, T958M, Y976F and F1076L among 267 samples from Pakistan, Nepal, Sri Lanka, Ecuador, Sudan and São Tomé. Polymorphism in the last two codons has been described in multiple studies [11;14;15;18;19;22;23], but the high prevalence of 958M which we observed (206/211, 97.6%) was surprising since this SNP has only been mentioned in two earlier studies; in Madagascar (with a 100% fixation of the 958M) [18] and in a few samples from Indonesia and Brazil [39]. Besides these studies, all others either report the presence of the wild-type T958 allele, or do not mention the locus [11;20–23]. Since the 958M mutation was present in countries with both high and low level of reported CQR over a wide time span, we hypothesize that the T958M is an allelic variant of the wildtype and most likely not associated with CQR. In the present study the T958 wild-type was only detected in Ecuador. It is also seen in the Sal-1 reference sample which originates from El Salvador, so it is possible this allele might be a characteristic of American samples while the 958M is characteristic of Asia and Africa.
Rare mutations, F979S (TTT→TCT), M980V (ATG→GTG) and S1080N (AGT→AAT) were found in two samples from the Jhapa district of Nepal, both possessing the MYL double mutant Pvmdr1 haplotype (Table 3); One of these samples was mutated at codon 1080 while the other was mutated at codon 979 and codon 980. These results have been previously published by Ranjitkar in 2011 [35]. Thus, in total only five Pvmdr1 SNPs were detected which was an unexpected result. Orjuela-Sanchez et al. (2009) [23] reported up to 24 Pvmdr1 mutations in a study of only 7 samples from Brazil, while Barnadas et al. (2008) [18] reported 21 mutations among 105 samples from Madagascar. However, both studies amplified longer fragments of the Pvmdr1 gene than the present study, which might be a part of the explanation.
Genotyping of microsatellites flanking the Pvmdr1 gene revealed high levels of diversity around single, double and triple mutant alleles. There were too few wildtype TYF alleles to meaningfully compare the MS heterozygosity surrounding wildtype and mutant alleles for evidence of selective sweeps on the mutant alleles. However our finding that all 3 wildtype TYF alleles were flanked by an identical microsatellite haplotype would not support the view that reduced diversity among microsatellite haplotypes is attributable to selective sweeps, but rather suggests a tendency to clonal population structure in some populations.
The evidence for association of MS alleles with particular mutations was patchy. TYF and MYF Pvmdr1 haplotypes occurred together with the “201” m9.5 allele, while the “204” allele at m9.5 was more commonly seen with MFL and MYL. No obvious pattern of distribution was seen for the other 3 MS, suggesting this association was caused by greater representation of certain mutant alleles in particular geographic localities rather than a selective sweep.
The combination of the 5 Pvmdr1 loci into a 5-loci haplotype revealed 57 different haplotypes among the 125 samples positive at all loci, many of them unique. Country-wise, Nepal was the most diverse, when analysed by locus and for the combined 5-loci haplotype, whereas Ecuador was more conserved. The diminished diversity within the Ecuadorian samples is consistent with the general finding of little diversity amongst P. vivax samples from the Americas, although this is not a hard and fast rule [40]. Just 10 samples of African origin were available for this study but both double and triple Pvmdr1 mutant alleles were present, and their microsatellite fingerprint was distinct from that associated with the same alleles in Asia. Likewise, our South American sample from Ecuador (n = 17) was distinct from the other populations being less diverse, and unique in having the wild-type Pvmdr1 allele.
Studies of P. falciparum have revealed a contrasting pattern of resistance evolution in which relatively few resistance mutants emerge but then become globally disseminated. The pattern is consistent for both CQR [26;41] and high levels of resistance to SP [24;27;42–44]. Hawkins et al. [28;29] analysed the origin and dissemination of SPR in P. vivax by analysing SNPs in and surrounding the Pvdhfr and Pvdhps genes. They concluded that the genes are considerably more diverse than seen in P. falciparum [28;29] and that highly pyrimethamine-resistant Pvdhfr alleles arose three times in Thailand, Indonesia and PNG/Vanuatu, and that sulfadoxine resistance associated SNPs had evolved independently on multiple occasions. This is consistent with our findings on Pvmdr and may be explained by comparisons of total genomic diversity among P. vivax and P. falciparum isolates. A study by Neafsey et al. [45] reported twice as much SNP diversity, significantly higher MS diversity and a far deeper divergence among P. vivax geographic isolates than among a comparable set of P. falciparum isolates. The higher level of diversity in P. vivax can explain the multiple origin pattern of resistance emergence in of Pvmdr1, Pvdhfr and Pvdhps.
Our findings can indicate three things. First, Pvmdr1 mutant alleles have developed on multiple haplotype backgrounds by convergent evolution in Pakistan, Nepal, Sri Lanka, Ecuador, Sao Tomé and Sudan. Second, assuming that Pvmdr1 is a reliable CQR marker, there is little evidence that the variation around mutant haplotypes has been subject to a selective sweep, (the result of positive natural selection causing diminished diversity in sequences flanking the selected marker). Third, the historical pattern of drug resistance emergence in P. falciparum is not repeated in P. vivax. The time-delay of around 30 years between initial reports of CQR in the two malaria species, and the two different treatment regimens (mono-in P. falciparum and usually combined CQ/primaquine treatment in P. vivax) might explain this, and may suggest that it is just a matter of time before the effect of selection of the markers can be seen since high grade treatment failure has not yet been reported in any of the sites sampled in this study. P. vivax is generally a chronic disease with low parasitaemia causing mild symptoms compared to P. falciparum, therefore there is less selective drug pressure on the parasite. Furthermore, the fast gametocytogenesis in P. vivax enables uptake of gametes by vectors before clinical symptoms arise and antimalarial treatment is initiated which balances the spread of sensitive and resistant P. vivax parasites.
Equally, these differences may lie with the biology and transmission dynamics of the two species. Our district level analysis of Pvmdr and linked MS variation in Sri Lanka found evidence of exchange of genotypes between districts which may or may not be linked to their resistance phenotype. Latent P. vivax infections cause relapses of the disease, which may increase the possibility of migration of parasites within the human host while the broader temperature tolerance by P. vivax parasites compared to P. falciparum might increase the likelihood of gene flow between sites with varying temperature or microclimate.
Finally CQR may be a complex trait including other genes in addition to Pvmdr1 and future research will hopefully illuminate the genomic-level change underpinning changes in CQ sensitivity. Meanwhile, monitoring and research of CQR is of highest importance for the public health in the afflicted areas.
|
10.1371/journal.ppat.1005786 | The 17D-204 Vaccine Strain-Induced Protection against Virulent Yellow Fever Virus Is Mediated by Humoral Immunity and CD4+ but not CD8+ T Cells | A gold standard of antiviral vaccination has been the safe and effective live-attenuated 17D-based yellow fever virus (YFV) vaccines. Among more than 500 million vaccinees, only a handful of cases have been reported in which vaccinees developed a virulent wild type YFV infection. This efficacy is presumed to be the result of both neutralizing antibodies and a robust T cell response. However, the particular immune components required for protection against YFV have never been evaluated. An understanding of the immune mechanisms that underlie 17D-based vaccine efficacy is critical to the development of next-generation vaccines against flaviviruses and other pathogens. Here we have addressed this question for the first time using a murine model of disease. Similar to humans, vaccination elicited long-term protection against challenge, characterized by high neutralizing antibody titers and a robust T cell response that formed long-lived memory. Both CD4+ and CD8+ T cells were polyfunctional and cytolytic. Adoptive transfer of immune sera or CD4+ T cells provided partial protection against YFV, but complete protection was achieved by transfer of both immune sera and CD4+ T cells. Thus, robust CD4+ T cell activity may be a critical contributor to protective immunity elicited by highly effective live attenuated vaccines.
| The 17D line yellow fever virus (YFV) vaccines are some of the safest and most effective live-attenuated virus vaccines ever produced, protecting recipients for life against deadly yellow fever (YF). As a testament to this safety and efficacy, the 17D line of live-attenuated vaccines has become an important model for the design of future vaccines. However, we still lack a fundamental understanding of the protective immunity elicited against the virulent YFV, a knowledge gap that must be overcome to inform the design of future live-attenuated and subunit vaccines. Humans develop robust antibody and T cell responses following vaccination, leading some to suggest that vaccine-elicited CD8+ T cells are important for protection against virulent YFV. Since this can never be tested in humans, we have used mice to model immunity to the 17D-204 vaccine strain. We found that CD4+ T cells elicited by 17D-204 contributed to protection against virulent YFV, but CD8+ T cells had no effect on the outcomes of survival or disease. Our study is the first to demonstrate that vaccine-elicited CD4+ T cells can protect against YFV infection. These results suggest that vaccine developers should consider the importance of CD4+ T cells when designing vaccines against viruses similar to YFV.
| Live-attenuated vaccines (LAV) generally provide the highest level of protection against infectious diseases. The most effective LAVs duplicate the pathogen-specific conditions of natural infection but have their replication curtailed by the innate and adaptive immune responses prior to the onset of clinical disease. A well-balanced combination of authentic antigen expression and control can induce a prolific adaptive immune response and the formation of long-lived memory. The development of LAVs is generally a results-driven empirical process controlling first for attenuation and subsequently for protection. Although the broad immunological response to these vaccines is often times examined exquisitely, the immunity that directly contributes to protection is more difficult to define. Exploring the protective immunity elicited by LAVs would require the use of human subjects, which is often not appropriate, or animal model systems which may not accurately represent immunity or disease. However, understanding the immune properties that are required for protection is crucial to the rational design of vaccines against pathogens for which empirical production of a LAV has failed or for which usage of a LAV is prevented by current vaccine standards.
One of the most successful lines of LAVs uses the 17D-based vaccine strains of yellow fever virus (YFV). Since its introduction in the 1930s [1] the 17D-based vaccines (substrains 17D-204 and 17DD) have proven themselves to be amongst the most successful and efficacious vaccines created [2]. Centuries prior to the introduction of the 17D line of YFV vaccines, yellow fever (YF) was one of the most feared and widespread epidemic diseases in Africa, Europe, and the Americas. More than 500 million people have been vaccinated with the 17D-based vaccines and only 12 known cases of vaccine failure have resulted in YF [3]. The 17D line of vaccines also has an excellent safety record resulting in extremely rare severe adverse events [4]. Immunization with the 17D line of vaccines remains a mainstay in the YF endemic zones of Central/South America and Sub-Saharan Africa where sylvatic YFV reservoirs still seed endemic disease and outbreaks, offering the only protection to over 900 million people world-wide [5]. The strong and enduring response from a single vaccination has led to the recommendation that only a single dose is required for life-long immunity [6,7]. As such, the human response to immunization with the 17D line of vaccines has been used to identify gene signatures that correlate with desirable vaccine traits like immunogenicity [8], with the prospect of improving the designs of other vaccines.
Neutralizing antibodies elicited by the 17D-based vaccines have long been considered a correlate of protection against YFV. Titers are detected in humans as early as six days post vaccination and have been recovered after forty years [9,10]. Antisera protects non-human primates (NHPs; [11]) and intracranially (i.c.) inoculated mice [11,12] against challenge. Neutralizing antibodies rise to maximum levels around 2 weeks post-vaccination [13]. There is robust activation of CD4+ and CD8+ T cell responses that appear to transition into stable, polyfunctional memory [14–17]. Virus-specific CD4+ (conventional and T regulatory) and CD8+ T cells peak between 10 [17] and 11–30 [16,18,19] days post-vaccination, respectively. The magnitude of the CD8+ T cell response is influenced by viral load, peaking shortly after virus in the blood falls below detectable levels [19]. CD8+ T cells degranulate and produce IFNγ, IL-2, and TNFα [16,17]. CD4+ T cells predominantly produce IFNγ and express CXCR3, indicating a largely TH1 polarization [20]. Circumstantially, T cells appear to be capable of killing virus-infected cells [17]. Although it is widely speculated that YFV-specific T cells contribute to the superior efficacy of the 17D line of vaccines [21], this remains unproven as such questions cannot be addressed in human subjects.
Progress toward understanding, optimizing and exploiting 17D-based virus-elicited immunity has been hindered by the lack of a tractable small animal that accurately models severe YF disease and 17D-based vaccination. Few studies of naturally acquired wild type infections have examined immunologic parameters [22–24], and experimentally controlled studies in humans are virtually impossible. Therefore, human studies of the 17D line of vaccines have been limited to the evaluation of blood circulating factors [14–18,20,25,26]. NHP models have proven advantageous for studies modeling human disease and protection [11,27,28], but the cost of these studies is prohibitive. The resources and flexibility for immunologic studies in inbred rodent models are superior to NHPs. Unfortunately, normal and acquired immune-deficient rodents are refractory to infection and disease with both the 17D-based viruses and wild type YFV (wtYFV), with the exception of i.c. infection where both viruses are equally virulent [29,30].
Previously, we determined that subcutaneous (s.c.) infections with wtYFV were strongly restricted by the type I IFN response in mice [30], whereas in severe human or NHP infections, the virus appears relatively unhindered by the type I IFN system. Although the 17D-based viruses can elicit an immune response in immunocompetent mice [31,32], they are also restricted by type I IFNs making studies of immune development in these animals less relevant to humans where they actively replicate and spread from the site of inoculation. In mice lacking the receptor to type I interferon (IFNAR-/-), the 17D-204 vaccine strain replicates and disseminates from the site of s.c. inoculation, which functionally mimics human vaccination. 17D-204 is cleared in all mice following only minor clinical signs of infection. In contrast to 17D-204, infection with the wild type Asibi virus reproduces remarkably human-like disease [30], accompanied by a substantial cytokine response with high levels of IL-6 and MCP-I. This coincides with the viscerotropic dissemination of virus and severe liver and splenic pathologies similar to those found in post-mortem studies of YF patients [22–24,33]. Furthermore, wtYFV is uniformly lethal in these mice, reproducibly modeling the most severe aspects of YF disease.
Here, we demonstrate that vaccine substrain 17D-204 vaccinated C57BL/6 IFNAR-/- (AB6) mice are completely protected against an otherwise lethal infection with wtYFV from 3 weeks to at least 1 year post-vaccination. Using this model, we have evaluated the formation of adaptive immunity to the 17D-204 virus in a context that closely resembles human vaccination, in that it is a self-limiting attenuated infection. Similar to human studies, 17D-204 infection of AB6 mice resulted in the induction of neutralizing antibodies and a large polyfunctional 17D-204-specific CD4+ and CD8+ T cell response. Passive and/or adoptive transfer of immunity into naïve mice implicated both humoral and cellular immunity for conferring protection against infection with wtYFV. 17D-204-specific CD4+ T cells conferred partial protection against challenge with wtYFV whereas complete protection was achieved only when anti-sera and CD4+ T cells were provided together. Virus-specific CD8+ T cells produced cytokines and were cytolytic but surprisingly offered no protective effect upon challenge. Our data provide evidence, for the first time, that some T cell subsets elicited by 17D-based vaccination can play a role in protection against wtYFV infection and disease, suggesting that T cells may contribute to the unparalleled efficacy of the 17D line of vaccines.
Previously [30], we demonstrated that s.c. footpad infection of IFNAR-/- mice with wtYFV resulted in severe disease resembling human YF and requiring euthanasia. However, infection with the 17D-204 vaccine strain of YFV remained attenuated and resulted in clearance of the virus and only minor swelling at the site of injection. The clearance of 17D-204 suggested that infection was inducing an adaptive immune response capable of controlling virus infection. We therefore sought to define the cellular and humoral responses to 17D-204 as well as the protective efficacy against challenge with a virulent wtYFV.
The global success of the 17D line of vaccines resides in its ability to protect against challenge with diverse genotypes of wtYFV. Thus, we reasoned that a stringent test of protective immunity against wtYFV would require us to challenge with a genetically divergent strain of wtYFV. The 1971 isolate of YFV from Angola (Ang71, [34]) is a highly divergent phylogenetically distinct YFV virus of the Angola genotype [34–36]. Ang71 displays a 6.8 percent amino acid divergence from the 17D-204 parent strain, Asibi [36], making it an excellent wtYFV strain to assess the protective immunity elicited by the 17D-204 virus. Previously, we determined that Ang71 is virulent in IFNAR-/- mice [30].
Our previous work was conducted in IFNAR-/- mice from the 129 genetic background (A129) [30]. Virulence of the Asibi virus and Ang71 in A129 mice displayed an age-dependent attenuation that would make it impossible to complete studies of immune memory and protection when challenged with a virulent virus. We found that 17D-204 and Ang71 did not display the same age-dependent virulence in IFNAR-/- mice on the C57BL/6 background (AB6). 17D-204 remained attenuated following infection of 5–6 week old AB6 mice with 1x104 PFU per footpad, and resulted in only minor swelling at the site of injection. Similar to our previous observations in A129 animals [30], 17D-204 plaque forming units (PFU) remained undetectable in C57BL/6 mice whereas in AB6 mice, virus was detected in the serum, lymph nodes, spleen and bone marrow (Fig 1). Virus was never detected in the liver or the brain. 17D-204 was not detectable on d12 post infection (p.i.), suggesting that by this time the virus had been cleared from the AB6 mice. Importantly, subcutaneous injection of 2x104 PFU per footpad of the Ang71 was found to be virulent in AB6 mice independent of age, including mice aged 5–6 weeks (Fig 2A), 8 weeks (Fig 2B), and one year (Fig 2C). Disease included progressive weight loss beginning on d4 p.i., swelling of the limbs surrounding the injection site, lethargy, piloerection, and hunched posture as described previously [30]. Ang71 was uniformly lethal in AB6 animals.
We hypothesized that immunity generated against 17D-204 in AB6 mice would be protective against infection with the virulent Ang71 strain of YFV. To test this, naive 5–6 week old AB6 mice were vaccinated s.c. with 1x104 PFU of 17D-204 or virus diluent (mock) in both rear footpads. Mice were then challenged 21 days following vaccination by s.c. injections in both rear footpads with 2x104 PFU of Ang71 (Fig 2B). The challenge with Ang71 in mice having received a mock vaccination resulted in disease and required 100% of animals to be euthanized by 9d p.i. In contrast, vaccination of mice with 17D-204 immunized them against challenge with Ang71 as evidenced by continued weight gain and the absence of clinical signs of disease. All 17D-204 immunized animals survived challenge (Fig 2B) with no clinical signs of disease for up to three months prior to being euthanized. Furthermore, AB6 mice having received a single 17D-204 immunization at five to six weeks of age were completely protected against clinical signs of infection when challenged one year later (Fig 2C). Thus, immunization with the live-attenuated 17D-204 YFV vaccine strain resulted in long-term protective immunity against challenge with a highly virulent, genetically divergent isolate of wtYFV.
We previously reported that YFV infection of mice resulted in rapid and transient induction of high levels of the proinflammatory cytokines IL-6 and MCP-1 in serum [30], consistent with human studies of YF [22], YF vaccine-associated viscerotropic disease [37] and NHP infection with the wild type Asibi virus [IL-6 only, [38]]. We screened 18 additional serum cytokines in addition to IL-6 and MCP-1 (Fig 3) in mice following Ang71 infection and found that numerous cytokines with adaptive immunomodulatory roles were elevated including: proinflammatory and effector cytokines IFNγ and TNFα [39,40], IL-12—a proinflammatory T cell and IFNγ stimulating protein [41], IL-2 –a T cell growth factor [42], and the B-cell differentiation and growth factor IL-5[43]. All reached peak levels around 4d p.i., declining by d5. IL-12p70 levels were high on d2 through d4 p.i. before declining by d5. The immune cell chemoattractant proteins IP-10 and MIG peaked on d4 p.i. and remained at similarly high levels through 5 d p.i. Several additional cytokines were found to be largely non-responsive to Ang71 infection in non-immunized animals (S1 Fig). The peak of the cytokine storm coincided with the peak of viral titers in most tissues (Fig 4A and 4B), the onset of weight loss (Fig 2B) and the rapid progression of disease, suggesting that a broad cytokine response may be a biomarker for and a contributing factor to disease. Importantly, cytokine levels were not elevated in 17D-204-immunized animals, remaining similar to those recorded for mock-infected mice at d4 post challenge (Fig 3).
The lack of disease observed in 17D-204-immunized mice when challenged with a lethal dose of Ang71 suggested the induction of a rapid and long-lived adaptive immune response that was efficient at controlling infection and/or dissemination of Ang71. To determine the degree to which replication and dissemination of Ang71 was restricted in 17D-204 immunized mice, we infected mice with Ang71 and harvested tissues at 2, 4 and 5 d post-challenge. In mock-vaccinated mice, infectious Ang71 was detected in the popliteal lymph node draining the inoculation site (DLN), serum and spleen within 2d (Fig 4A). DLN titers reached a sustained 2-5x103 PFU/LN. Serum viremia peaked on d2 p.i. at ~2 x 103 PFU/ml and fell below the LD in all but one animal on d4 and then in all animals by d5. Virus was detected in the liver in all but one mouse by 2d p.i. and in all mice by d4 and d5 p.i. Two mice had detectable virus in the kidney at 2d p.i. and all mice had virus in the kidney on d4 and d5 p.i. By d4 and d5 p.i. the heart and brain in all mice had detectable virus (Fig 4B). Generally, viral titers declined in the visceral tissues between d4 and d5. In the brain, the titer rose by approximately 20-fold. Despite rising levels of virus in the brain, as previously reported [30], the mice did not present signs of neurologic disease.
No tissues harvested from mice immunized with 17D-204 and then challenged with Ang71 contained any detectable infectious virus above the limit of detection (Fig 4C and 4D). For greater sensitivity, we compared viral GE by RT-qPCR in mock-vaccinated mice (Fig 4E and 4F) versus 17D-204-immunized mice (Fig 4G and 4H) after Ang71 challenge. RNA genomes for Ang71 were readily detected in unvaccinated AB6 mice. GE in the DLN and serum peaked on d4 p.i. and by d5 p.i. had decreased (Fig 4E). GE in the spleen increased through d4 p.i. and remained at similar levels through d5 p.i. GE in the liver, kidney, and heart remained similar from d4 to d5, whereas an increase in GE was observed in the brain from d4 to d5 (Fig 4F). In 17D-204-immunized mice, however, Ang71 RNA was consistently detected only in the DLN (Fig 4G) but at more than 100-fold lower levels compared to mock-vaccinated mice (Fig 4E) at d2 p.i. and approximately 6 x104-fold lower by 4d p.i. In a subset of mice, the spleen [33%], liver [17%], kidney [17%] and brain [33%] showed low GE levels peaking at d3 p.i. and falling below the LOQ at 4d p.i. One mouse had low levels of detectable GE in the heart at 4d p.i. (Fig 4H). The variation in GE to PFU ratio observed across tissues and time points likely reflects the cellular dynamics and formation of adaptive immune responses, primarily neutralizing antibodies, against Ang71 that obscure the detection of PFU but not GE.
Sera collected from 17D-204-immunized mice were evaluated for 17D-204 neutralizing antibodies by plaque reduction neutralization test (PRNT; Fig 5A). Consistent with the high seroconversion rate in human vaccinees of between 90% and 100% [44,45], all mice tested between 19 days post immunization and 1.5 years post immunization demonstrated PRNT80 neutralizing antibody titers against both 17D-204 and Ang71 (Fig 5A). We found that serum neutralizing titers were reduced by approximately nine-fold on average against Ang71 when compared with 17D-204 virus. The reduced titers were most likely explained by the amino acid divergence displayed between 17D-204 and Ang71, as discussed above.
To test whether 17D-204 antisera would protect against challenge, sera and magnetically enriched CD19+ B-cells - those responsible for antibody production - (S2 Fig) were harvested from 17D-204 immunized mice on d21 after immunization and transferred into naive AB6 mice 24 hours before challenge with Ang71. To serve as positive and negative controls for protection, total splenocytes and sera from 17D-204-immunized or mock-vaccinated AB6 mice were transferred into naïve AB6 mice. Similar to the disease described in Fig 2, all mice [6/6] receiving total splenocytes and sera from mock vaccinated animals experienced weight loss (Fig 5B and 5C) and had to be euthanized (Fig 5D and Table 1). On average, animals receiving total splenocytes and sera from 17D-204-immunized mice experienced mild disease indicated by significantly reduced weight loss following challenge (Fig 5B and 5C) and 100 percent survival [5/5] (Fig 5D and Table 1). On average, mice receiving only serum from 17D-204 immunized animals also experienced mild disease as illustrated by significantly reduced weight loss compared to animals receiving mock-vaccinated splenocytes and serum (Fig 5B and 5C). Additionally, seventy-five percent [6/8] of mice receiving only serum survived challenge with Ang71, a significant increase compared to mice receiving mock splenocytes and serum (Fig 5D and Table 1). Mice receiving CD19+ cells from 17D-204 immunized mice experienced weight loss similar to mice receiving mock splenocytes and sera (Fig 5B and 5C). However, 37.5% [3/8] of mice receiving CD19+ cells from 17D-204 immunized mice survived challenge with Ang71, a significant increase compared to mice receiving mock serum and splenocytes (Fig 5D and Table 1). These data suggest that convalescent serum may be more effective than CD19+ cells at preventing severe disease. However, both serum and CD19+ B-cell recall responses can lead to survival of mice following challenge with Ang71.
Acute viral infections, including the 17D-based vaccines, result in the proliferation of virus-specific CD4+ and CD8+ T cells responding to diverse epitopes. Once T cells become activated, they upregulate CD44 and downregulate CD62L (CD44hiCD62Llo) and represent circulating T cells. Although CD44 and CD62L comprise a conventional means of assessing T cell activation, CD11a is also upregulated on activated T cells in mice [46,47] and in humans [16]. Enhanced expression of CD11a on the surface of responding T cells is indicative of T cells responding to antigen-specific stimulus resulting from TCR cross-linking rather than bystander activation that can result from exposure to an inflammatory environment [46]. CD11a upregulation may constitute a more accurate means to assess total virus-specific T cell responses in systems where all possible T cell epitopes have not been defined [46]. Additionally, CD11a expression increases on YFV-specific CD8+ T cells in humans [16] and has been used to measure the broad T cell response to bacterial infection [46], malaria [48,49], and viruses [46,47] in mice. CD11a expression remains high for the life of a T cell following activation.
To evaluate T cell immunity to 17D-204, we harvested popliteal DLNs and spleens from 5–6 week old immunized mice at d7 p.i. At this time, 17D-204 was not detectable in the lymphoid organs by plaque assay following the peak of replication on d4 p.i. (Fig 1). CD4+ (Fig 6A and 6B) and CD8+ (Fig 6F and 6G) T cells in 17D-204-immunized mice demonstrated an expansion of the CD44hiCD62Llo subset commensurate with activation and proliferation. In addition, we evaluated T cells responding to 17D-204 by measuring CD11a upregulation. S3 Fig demonstrates the gating strategy for CD11ahi cells and that the majority of CD11ahi cells are CD44hiCD62Llo. Comparison of CD11a expression (Fig 6C, 6D, 6H and 6I) also demonstrated a significant increase in the percent of total CD4+ and CD8+ T cells over mock in the DLN and spleen. The increase in the percentage of activated cells corresponded to an increase in total numbers of CD11ahi CD4+ and CD8+ cells indicating a true expansion of activated T cells responding to immunization with 17D-204 (Fig 6E and 6J).
Following an acute T cell response to virus infection, T cells contract into memory, which can persist for the life of the animal. Memory cells can be broadly characterized as central (TCM, CD62Lhi) and effector (TEM, CD62Llo) memory [50]. TCM are characteristically found in the lymphoid compartments and are associated with the long-lived rapid recall responses that are most associated with adaptive immune memory [51–53]. TEM are relatively short-lived cells that remain in circulation and retain effector functions similar to those seen in the acute response [51–53]. Compared to CD4+ T cells, differentiation of CD8+ effector T cell subsets have been more thoroughly defined by monitoring the expression of the inhibitory C-type lectin, KLRG1 and the IL-7 receptor-α, CD127 [54,55]. Four subgroups of effector cells have been described that predict the formation of memory (Fig 7D): early effector cells (KLRG1loCD127lo, EEC) which probably represent an early state of transition into the other subtypes; short lived effector cells (KLRG1hiCD127lo, SLEC) which have a limited duration; memory precursor effector cells (KLRGloCD127hi, MPEC) which give rise to long-lived memory and may be analogous to TCM; and double positive effector cells (KLRG1hiCD127hi, DPEC) which may be similar to TEM.
On d7 after 17D-204 immunization (acute T cell response) CD4+ T cells were predominantly CD44hiCD62Llo (Fig 7A). Following contraction, at d27 after immunization, TCM-CD44hiCD62Lhi cells predominated in the DLN and the spleen (Fig 7B). As a proportion of total CD4+ T cells, TCM-CD44hiCD62Lhi cells were enriched from d7 to d27 post immunization. In contrast, the percent of total CD4+ T cells that were TEM-CD44hiCD62Llo was decreased (Fig 7C). The acute CD8+ T cell response on d7 post immunization was predominated by EEC followed by SLEC (Fig 7D and 7E). A population of MPEC was present at d7 following immunization, suggesting that a subset of T cells was beginning to transition to long-lived memory. By d27 post immunization (Fig 7F), contraction of all cell types had taken place as indicated by the decrease in total numbers of all cell populations. The contraction corresponded to a decrease in the proportion of total CD8+ T cells of the SLEC and DPEC phenotypes in the DLN (Fig 7G). The EEC phenotype was enriched in the DLN on d27 post immunization, whereas the proportion of the MPEC phenotype remained unchanged. In contrast to the DLN, the contraction observed on d27 post immunization in the spleen corresponded to decreases in the proportions of SLEC and EEC. An increase in the proportions of the DPEC and MPEC phenotypes were observed in the spleen. The enrichment of MPEC cells in the spleen but not the DLN suggests that the formation of long-lived CD8+ T cell memory to 17D-204 may be dependent on the environments of specific lymphoid organs.
To verify that 17D-204 elicited a specific T cell response and to determine whether those cells were functional, we evaluated DLN and spleen cells from 17D-204-immunized mice by intracellular cytokine staining (ICS). Cells were stimulated with the YFV MHC-II (YFII-E; I-Ab) and two MHC-I (YFI-NS3; H2-Kb and YFI-E; H2-Db) restricted determinants [31]. Stimulation with the YFII-E peptide resulted in CD4+ T cells producing IFNγ (Fig 8A). Since polyfunctional 17D-based vaccine-specific T cells have been observed in humans, we evaluated the CD4+ T cells' ability to produce multiple cytokines (IFNγ, IL-2 and IL-4) and grouped them by how many cytokines they produced (single, double, triple). The majority of CD4+ T cells at d7 and d27 post immunization produced only one cytokine, dominated by IFNγ (Fig 8B). The remaining cells produced IFNγ and IL-2 with no IL-4 detected above background levels.
Stimulation of T cells with YFI-E or YFI-NS3 resulted in the majority of CD8+ T cell responders producing both IFNγ and CD107a (LAMP-1) (Fig 8C). CD107a is deposited at the cell surface during T cell degranulation and is linked with a T cell's ability to lyse targets [56], suggesting that CD8+T cells were largely functional and capable of killing target cells. When we assessed CD8+ T cells, we found that nearly all cells responding with cytokines also produced CD107a. We gated on CD107a+ cells and grouped them according to their ability to produce combinations (none, single, double, triple) of IFNγ, TNFα or IL-2 in response to peptide stimulation (Fig 8D). On d7 post immunization, CD8+ T cells responding to both YFV peptides demonstrated polyfunctional behavior with double and triple positive cells most predominant in the DLN while the spleen was comprised of mostly single and double cytokine producers. At d27 post immunization, cells from both tissues produced fewer cytokines. This effect was most dramatic from CD8+ T cells in the DLN and specifically those cells responding to the subdominant YFI-E epitope. The trend that T cell polyfunctionality becomes less diverse over time is consistent with the literature studying human vaccination with the 17D line of vaccines [17].
We tested whether 17D-204-specific T cells were cytolytic by examining in-vivo cytotoxicity against 17D-204 determinants. Mice were immunized with 17D-204 on d7, d27 or 1.5 years prior to intravenous transfer of fluorescently labeled naive splenocytes loaded with control peptides from ovalbumin (OVA-I; H2-Kb), SV40 large T-antigen (SV40 TAg; SV40T-I; H2-Db), hepatitis B core (HBV-Core; I-Ab) or YFV peptides YFI-NS3, YFI-E or YFII-E. Sixteen hours after transfer, spleens or popliteal draining lymph nodes were harvested and analyzed by flow cytometry to determine specific cytolytic activity (Fig 9A). As a positive control for specific cytotoxicity, mice were immunized with SV40 virus-transformed cells and cytolytic activity was confirmed against the SV40T-I determinant of SV40 TAg [57] (Fig 9A and 9B). During both the acute response (d7 post immunization) and the early memory response (d27 post immunization), strong cytolytic activity was detected against all CD8+ T cell determinants with a significant decrease in activity between d7 and d27 post immunization (Fig 9B). Although weaker than CD8+ T cell activity, cytotoxicity was detected against the CD4+ T cell determinant YFII-E during the same time frame. The activity was relatively strong in the DLN on d7, showing a significantly higher signal than in the spleen. By d27 post immunization, the signal had increased in the spleen (Fig 9B). Cytolytic activity was detected as early as d3 p.i. in the DLN for all determinants. In both the DLN and spleen, cytolytic activity was detected out to one and a half years following immunization (Fig 9C and 9D). Activity against the YFI-NS3 and the YFII-E peptides showed no drop from peak levels during this time whereas activity against the subdominant YFI-E peptide decreased over time in both organs.
To determine whether T cells could contribute to protection against challenge with Ang71, we isolated serum and magnetically enriched CD4+, CD8+ or CD19+ cells (S2 Fig) on d21 post immunization and adoptively transferred combinations of the enriched populations into naïve AB6 mice. Mice were challenged with Ang71 and monitored for weight loss (Fig 10A, 10B and 10D and Table 1) and survival (Fig 10C and 10E and Table 1). Compared to mice receiving splenocytes and serum from mock vaccinated animals, mice receiving CD4+ T cells alone or in combination with CD8+ T cells experienced similar disease as indicated by weight loss. However, survival of both groups (CD4, [5/8]; CD4, CD8, [4/8]) was increased compared to mice receiving splenocytes and serum from mock animals. Surprisingly, mice receiving only CD8+ T cells from 17D-204 immunized mice experienced similar disease and mortality as indicated by weight loss (Fig 10A and 10B) and survival [0/8] (Fig 10C and Table 1) as animals receiving mock splenocytes and serum. Interestingly, the subset of animals that had to be euthanized after receiving CD4+ T cells from 17D-204 immunized mice experienced a significantly reduced AST, by approximately one day, compared to mice receiving mock vaccinated total splenocytes and serum (Table 1). Although overall the severity of clinical disease in this subset of mice was similar when comparing weight loss (Fig 10A and 10B) to what was observed in mice receiving mock-vaccinated splenocytes, the onset was earlier. This observation suggests that under as yet undefined conditions, CD4+ T cells may contribute to immunopathology following challenge with a wtYFV. Overall, these results suggest that in AB6 mice, 17D-204-elicited CD4+ T cells can impart protection against challenge with a wtYFV whereas CD8+ T cells are not protective.
Since neither 17D-204 sera (Fig 5B–5D) nor T cells (Fig 10A–10C) alone were capable of completely protecting mice against Ang71 (Table 1), we adoptively transferred combinations of T cells, B cells and sera into naïve mice prior to challenge. When serum from 17D-204-immunized mice was transferred in addition to both CD8+ and CD4+ T cells, 100% of mice were protected against challenge with Ang71 compared to mice receiving total splenocytes and serum from mock vaccinated animals. These groups (CD8, CD4, Serum and CD4, CD8, CD19, Serum) displayed reduced weight loss (Fig 10A and 10D) and increased survival ([8/8] and [7/7]) compared to animals receiving splenocytes and serum from mock vaccinated mice. Since CD8+ T cells did not contribute to protection (Fig 10A–10C), these data suggest that CD4+T cells and sera from 17D-204 immunized mice can act together to confer protection against challenge with a wtYFV.
The success of the live-attenuated 17D line of YFV vaccines stems from the induction of long-lived, probably life-long, immunity against wtYFV. Due to the wide use of the 17D-based vaccines, volunteer vaccinees have in recent years contributed immensely to our understanding of vaccine-induced immunity in humans. These studies have shown that a single vaccination results in long-lived neutralizing antibodies [9,10] as well as long-lived [14,15] functional memory [16,17] following a robust CD4+ and CD8+ T cell response. Additionally, the use of systems biology approaches to track genomic signatures following immunization paints a picture that more thoroughly decodes the superior immunogenicity of the 17D line of vaccines and may lay groundwork for predicting the efficacy of other vaccines [8]. Despite these advances, studies linking humoral and T cell responses to protection against wtYFV have been lacking due to the absence of a cost-effective model that accurately recapitulates both vaccination and disease. In this study, we used an AB6 murine model to evaluate the immunity induced by 17D-204 immunization and its corresponding protection against challenge with a virulent wtYFV. Our results suggest that both neutralizing antibodies and CD4+ T cells elicited by 17D-204 can be protective against challenge with a virulent wtYFV.
Infection of AB6 mice with wtYFV elicits a broad elevation in the cytokine profile reminiscent of natural wtYFV infection in humans or rare vaccine-associated viscerotropic disease (YFV-AVD) [22,37,58]. Elevation of MCP-1, IL-6 ([30] and this study), TNF-a and IP-10 in the AB6 model is consistent with studied cases of patients that develop a fatal case of YF [22] or YFV-AVD [37,58]. The role of these and other elevated cytokines following wtYFV infection and YFV-AVD is unknown. However, pathological cytokine responses including IFNγ, IL-6, IL-10, TNFα, and IP-10 [59], all of which are elevated in AB6 mice during fatal Ang71 infection, have been implicated in inducing vascular permeability with the onset of hemorrhagic fever after infection with the related Dengue virus (DENV). Incidentally, serum cytokine concentrations become elevated on d4 p.i. ([30] and this study), marking the onset of clinical disease in AB6 mice. 17D-204-immunized AB6 mice show no signs of cytokine elevation following challenge with Ang71 possibly due to limited viral replication, which is mostly restricted to the DLN. We suggest that the cytokine response associated with high viral replication in unimmunized mice may contribute to disease. Future studies are warranted to assess whether suppressing these responses after Ang71 challenge of unvaccinated mice could prevent the onset of disease and perhaps lead to treatments for human YF.
The adaptive immunity induced by the 17D line of vaccines in humans has been partially characterized in recent years [14–18,26,60]. However, little has been studied concerning the specific components of this response that are required for protection against wtYFV. Antibodies are a correlate of protection against wtYFV and are considered by the WHO as the primary measure of vaccine efficacy [61]. The importance of antibodies for protection against wtYFV is supported by reports that lower levels of YFV-specific antibodies correlate with severe and fatal cases of YF in contrast to mild non-fatal cases [22]. Additionally, lethal infection in rhesus monkeys produces necrotic B cell germinal centers [28]. Our serum transfer results support the importance of antiserum for protection against wtYFV and suggest that it may be required for protection from disease. However, our results indicate that even in fully 17D-204-immune mice, Ang71 was not immediately sterilized upon challenge since virus was detected in the DLNs of all mice. It remains unknown whether 17D-based vaccines produce sterilizing immunity in humans against challenge with a wtYFV. Many experimental studies that have assessed the importance of antisera for protection have challenged with the Asibi virus, the parent virus to the 17D vaccine line. Our use of the divergent Ang71 virus, which is less efficiently neutralized in 17D-204-immunized mice, may explain why antiserum is not entirely protective against challenge. Long-term studies are needed to determine whether neutralizing antibodies alone are sufficient to clear virus or whether neutralizing antibodies simply limit the spread of virus while T cells or other immune cells eliminate virus reservoirs. Our findings support a critical role for neutralizing antibodies for protection against wtYFV.
As in humans, 17D-204 immunization of AB6 mice results in a robust expansion of polyfunctional CD8+ T cells that form functional memory. Our analysis shows that 17D-204-specific T cells produce IFNγ and a subset also produce combinations of TNFα and IL-2. Additionally, CD8+ T cells express surface CD107a following specific stimulation and efficiently lyse target cells in-vivo at early and late (memory) time points following vaccination. These observations led us to hypothesize that in this model the cytolytic capacity of CD8+ T cells would be important for controlling Ang71 in adoptive transfer experiments. A similar mechanism of control is seen in a related mouse model following vaccination against DENV [62] and West Nile virus (WNV) [63,64]. Instead, CD8+ T cells played no detectable role in protection against Ang71, as determined by the severity of disease or AST. Additionally, the fact that clinical disease and AST were not exacerbated also suggests that CD8+ T cells were not contributing to disease through immunopathologic mechanisms. The importance of CD8+ T cells for vaccine-related protection or challenge immunopathology in humans has never been tested.
We found that 17D-204-specific CD4+ T cells were functional during the early and late time points post immunization. 17D-204-specific CD4+ T cells displayed a strong TH1 polarization during the acute response, and similar cytokine profiles were maintained following the formation of memory. In-vivo cytolysis of targets displaying 17D-204 derived MHC-II restricted peptides was detected with similar efficiencies during both the acute and long-term memory response. Importantly, adoptively transferred CD4+ T cells contributed to survival following Ang71 challenge, resulting in the recovery of a majority of infected mice. These results suggest that CD4+ T cells may functionally contribute to the superior efficacy of the 17D line of vaccines and as such constitute a critical component for effective vaccination against wtYFV. The mechanism by which CD4+ T cells exert their effect remains unknown. However, the MHC-II restricted in-vivo cytotoxicity suggests that direct cytolysis by YFV-specific CD4+ T cells may be an important mechanism. Perforin and Fas/FasL mediated mechanisms, as measured by in-vivo cytotoxicity are involved in the control of WNV [65]. The same mechanisms that contribute to the protective efficacy of CD4+ T cells may also be involved in the CD4-mediated immunopathology that we observed in a subset of mice. Bystander cytotoxicity was observed from CD4+ T cells responding to DENV [66]. Determining the specific mechanisms by which CD4+ T cells promote their effects is outside the scope of this study but remains an important question for future studies. In order to understand the implicit requirements for control of wtYFV, it will be important to study the mechanisms that lead to CD4+ T cell-mediated protection and the apparent inability of CD8+ T cells to limit disease.
To conduct these studies, we have used a murine system that lacks type I interferon signaling. Type I interferon can act as a third signal [67] of T cell activation and is a pro-survival and proliferative cytokine [68] possibly affecting the responses in AB6 mice. Type I interferon plays a minimal role in T cell responses for some pathogens due to compensation by other third signal cytokines like IL-12 or as yet undefined pathogen-specific environments [68,69]. For example, Lymphocytic Choriomeningitis Virus (LCMV)-specific T cells lacking IFNAR have substantially inhibited proliferation whereas during Listeria monocytogenes (LM) infection, IL-12 acts as the predominant third signal [69–72]. The CD4+ and CD8+ T cell responses induced following 17D-204 vaccination were robust even in the absence of type I interferon signals, and were comparable in size to those seen against LCMV and LM in B6 mice [46,47]. Importantly, the size of the dominant CD4+ and CD8+ T cell responses in the IFNAR-/- mouse model accurately reflects the response to individual dominant epitopes in humans [16,17]. These observations suggest that the magnitude of activated T cell responses in 17D-204 vaccinated IFNAR-/- mice may accurately represent human responses.
The influence of type I interferon on the adaptive immune response in humans to vaccinations with a 17D-based vaccine is unknown. Numerous publications indicate that human cells produce type I interferon in response to a 17D-based virus infection [73–78]. It is unknown what influence type I interferon has on human T cells during vaccination with the 17D line of vaccines, thus we cannot rule out the existence of deficiencies in the T cell phenotypes elicited by IFNAR-/- mice. To more carefully consider the role of type I interferon in humans, we assessed whether the 17D vaccine line induces a substantial type I interferon response in humans by independently analyzing gene expression data from PBMCs of human vaccinees published by Querec et al. [8]. This analysis indicated that immunization of humans with the 17D vaccine line does not result in significant increases in IFNα/β gene transcription in PBMCs (S4 Fig). Our analysis does not rule out more localized production of type I interferons (e.g. in the lymph nodes) or in a small subset of PBMCs, and the original study did not measure levels of protein in the sera. However, should these transcription data be representative of the levels of type I interferon following 17D-based vaccination, it suggests that type I interferons may play a limited role physiologically for 17D based vaccine-specific T cells. Under the same scenario, the proliferation of 17D-based vaccine-specific T cells in humans may be driven by an undetermined third signal cytokine, or T cell expansion may be driven by abundant antigen from relatively high titers of virus in the lymphoid compartments [30].
In summary, we have used a murine model of YFV infection and disease to characterize the immune response to 17D-204 and determine empirically which immune constituents contribute to protection against wtYFV. The 17D-based vaccine strains constitute one of the world's most successful vaccines, demonstrating superior safety, efficacy and longevity compared to most subunit vaccines and LAVs. The characteristics of immune development and function seen following 17D-204 immunization in small animal models can serve as benchmarks for the development of other vaccines that aim to mimic its superior immunogenicity and long-lasting protection. Since we can now assess the role that cellular immunity plays in conferring protection, we can form a more complete picture of what contributes to vaccine efficacy in general. More specifically, 17D-204-elicited protection against wtYFV, defined by neutralizing antibodies and CD4+ T cell immunity, may be directly applied towards the development of vaccines against other flaviviruses like DENV and WNV. Continued study of 17D-204-elicited immunity may facilitate vaccine research by uncovering novel immune mechanisms of action, discovering innate immune correlates of protection, developing immunotherapies, or providing a platform for screening vaccine candidates.
Animals were maintained and procedures were performed in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Research Council. Protocols 1004668, 1103456, and 14033545 were approved by the University of Pittsburgh's IACUC committee. Approved euthanasia criteria were based on weight loss and morbidity.
Mice deficient in receptors for type I interferons, (IFNAR-/-, AB6) were bred under specific pathogen-free conditions. At 5–8 weeks of age, randomized male and female mice were transferred to the ABSL-2 or ABSL-3 facility for infection. Vaccination was administered subcutaneously (s.c.) in both rear footpads to 5–6 week old AB6 mice with 104 PFU of 17D-204 in a volume of 10ul in an ABSL-2 facility. Mock-vaccinated animals received 10uL of virus diluent (PBS[with Mg+, Ca+] containing 1% donor calf serum). Mice were observed for clinical disease and weighed daily. At 8–9 weeks of age, mice were transferred to an ABSL-3 facility and challenged s.c. in both rear footpads with 2x104 PFU Ang71 in a volume of 10uL. Mice were monitored daily for clinical disease and weight loss. Animals were euthanized based on morbidity or weight loss of 20%. For virus titers and qPCR, serum was separated from whole blood by centrifugation using microtainer tubes (BD). Mice were perfused with 10mL of virus diluent prior to tissues being collected and frozen at -80C in virus diluent or Tri Reagent-LS (MRC) until the time of processing. 50uL of serum was frozen undiluted for titer or placed in Tri Reagent-LS for qPCR.
Vero (ATCC-CCL-81), Huh7 (Charles M. Rice, The Rockefeller University) and B6/WT-19 (Todd D. Schell, The Pennsylvania State University: College of Medicine) cells were maintained in Dulbecco's modified Eagle's medium (DMEM), supplemented with 10% fetal bovine serum (FBS), 0.29 mg/mL L-glutamine, 100 U/mL penicillin and 0.05 mg/mL streptomycin (37C; 5% CO2). B6/WT-19 cells are transformed by SV40 virus [79,80].
Stocks of YFV 17D-204 were produced from infectious clone [81] by electroporation of in-vitro transcribed (IVT) viral genomic RNA. 1ug of infectious clone DNA was linearized by restriction digest with Xho1. Linear DNA was purified and used as a template for IVT (mMESSAGE mMACHINE SP6, Ambion). 20ug of IVT RNA was electroporated twice into vero cells during exponential growth phase harvested from three 50% confluent T175 tissue culture flasks using the following settings: 220V, 1uF, exponential decay. Electroporated cells were seeded into a single T175 flask in 15mL of media with HEPES (0.02M) and sodium bicarbonate (0.15%) and incubated for 7 days at 37C + 5% CO2. Supernatant was harvested by centrifugation at 4000 RPM for 30 minutes and stored at -80C. The Ang71 (14FA [34,35]) virus was amplified on vero cells as described previously [30]. Infectious virus titers were determined by a plaque assay on Huh7 cells, expressed as plaque forming units (PFU)/mL.
RNA was isolated first by crushing tissue in Tri Reagent-LS and following the protocol provided by the manufacturer. Polyacryl carrier was added for serum isolation only. Reverse transcription of 100ng of purified RNA was performed for +-strand viral RNA using primer T7-YFV-antisense (GCGTAATACGACTCACTATATACCATATTGACGCCCAGGGTTTT) targeting a region of the 5'-non-coding region that is conserved between Ang71 and 17D-204 and 18s-antisense (CGAACCTCCGACTTTCGTTCT) using TaqMan reverse transcription reagents (AB). Synthesis of cDNA consisted of 25C, 10 m followed by extension at 48C, 30 m and concluding with inactivation of RT at 95C, 5 m. Quantitative determination of YFV and 18s cDNA was performed using separate reactions in Maxima Probe qPCR Master Mix (Thermo) and read on a 7900HT Real-Time PCR System (AB). Primers for YFV: T7 (GCGTAATACGACTCACTATA), YFV-sense (AATCGAGTTGCTAGGCAATAAACAC), and YFV-Probe (CAGTTCTCTGCTAATCGCTCAACGAACG). Primers for 18s: 18s-antisense, 18s-sense (CGCCGCTAGAGGTGAAATTCT) and 18s-Probe (CAAGACGGACCAGAGCGAAAGCATTTG). Cycling conditions consisted of: denaturing and polymerase activation at 95C, 10 m; followed by 40 cycles of denaturing at 95C, 15 s then extension at 60C, 1 m. Fluorescence intensity data was collected during the extension step. The YFV GE standard curve was based on 10 fold dilutions of 17D-204 IVT RNA. The LOQ was set to the GE represented by the greatest dilution of standard remaining on the logarithmic curve. We found that expression of 18s in 100ng of RNA varied among tissues, thus mean 18s Ct values were calculated for each tissue type (e.g. spleen, brain, etc.) and these values were used to correct for sample loading error.
A mouse cytokine 20-plex panel kit was purchased from Invitrogen (LMC0006). According to the manufacturer's instructions, serum was diluted 1:1 with diluent and analyzed using a Luminex 100/200 instrument.
Serial dilutions of control antibody or serum were incubated for 1 hour at 37C + 5% CO2 with approximately 100 PFU of 17D-204 or Ang71. PFU of non-neutralized virus was determined using a standard plaque assay on Huh7 cells. Plaques remaining at all dilutions were counted and expressed as a percent of plaques remaining with mock serum samples. A best fit non-linear curve constrained to maximum 100 percent of mock and minimum zero percent of mock was used to calculate the 80 percent reduction in PFU.
Peptides were ordered from GenScript at > 90% purity and resuspended in PBS + 5% DMSO at 1mM concentration. Peptides designated as YFI or YFII are respectively MHC-I or MHC-II restricted peptides originating in 17D-204. The designations E or NS3 indicate the 17D-204 protein in which the peptide originates. Peptides: YFI-E (4–12) IGITDRDFI; YFI-NS3 (268–275), ATLTYRML; YFII-E (231–245), LVEFEPPHAATIRVL [31]; SV40T-I (206–215), SAINNYAQKL [82]; OVA-I (255–262), SIINFEKL; HBV Core (128–140), TPPAYRPPNAPIL. All YFV peptides are conserved between 17D-204 and Ang71.
ICS and cell surface staining was performed on single cell suspensions from spleens or lymph nodes created by pushing the tissues through a 70uM nylon mesh (Fisher). For ICS, up to 4x106 total cells were incubated in 200uL of T cell growth media (RPMI 1640 containing 10% FCS, 100 U/mL penicillin, 20uM 2-ME, 1mM sodium pyruvate and 10mM HEPES) supplemented with 1uM of peptide, monensin (BD GolgiStop) according to the manufacturer's instructions and 1:50 anti-CD107a-PE. Cultures were incubated for 5 hours at 37C and 5% CO2.
Staining for ICS: Cell were stained for 30 minutes on ice with 1:1000 dilution of Invitrogen Blue LIVE/DEAD stain in PBS. Cells were washed 1x in FACS buffer (PBS supplemented with 2% FBS and 0.1% sodium azide). Cells were then incubated in a 1:200 dilution of rat anti-mouse CD16/CD32 on ice for 20 min then washed 1x with FACS buffer. Fluorophore conjugated antibodies (1:200 dilution) were added to the cells and incubated for 30 minutes on ice than washed 3x with FACS buffer. Cells were then fixed and permeabilized using Cytofix/Cytoperm solution (BD) on ice for 20 minutes then washed 3x in Perm/Wash (BD). Staining was completed with 1:200 dilutions of fluorophore conjugated antibodies in Perm/Wash for 15 min at room temperature then washed 3x with Perm/Wash. Cells were fixed in 4% PFA in PBS and stored at 4C O/N prior to analysis.
Surface staining: Cells were stained for 15 minutes at 1:1000 dilution of Invitrogen Blue LIVE/DEAD stain in PBS. Cells were washed 1x in FACS buffer then incubated in a 1:200 dilution of rat anti-mouse CD16/CD32 for 15 minutes then washed 1x with FACS buffer. Fluorophore conjugated antibodies (1:200 dilution) were added to the cells and incubated for 15 minutes at room temperature than washed 3x with FACS buffer prior to fixation in 4% PFA. FACS analysis was performed the following day.
All data were collected using a LSR II or Fortessa flow cytometer (BD) administered by the University of Pittsburgh’s Unified Flow Core. If available, a minimum of 300,000 live events were collected based on FSC-A/SSC-A and live/dead gating. Data were analyzed using FlowJo software (Tree Star). The following flow cytometry antibodies were purchased from eBiosciences or Tonbo biosciences: CD8α (53–6.7), CD4 (GK1.5), CD19 (eBio 1D3), CD44 (IM7), CD62L (MEL-14), CD11a (M17/4), IFNγ (XMG1.2), TNFa (MP6-XT22), IL-2 (JES6-5H4), IL-4 (11B11), KLRG1 (2F1), CD127 (A7R34), CD107a-PE (eBio1D4B).
Splenocytes from naive AB6 mice were incubated for 1 hour at 37°C in complete media with 1uM of the indicated peptides. Free peptide was removed by washing three times in PBS. Each splenocyte population was fluorescently marked by staining with 1uM eFluor 450 or eFluor 670 Cell Proliferation Dye (eBioscience) and 5, 0.5 or 0.05 uM CFSE (eBioscience) in PBS for 10 minutes at 37°C. Staining was halted by addition of T cell growth media then cells were washed 3 times in PBS, combined in equal ratios then transferred to infected mice. 16 hours following transfer, spleens and/or popliteal lymph nodes were harvested and stained with anti-CD19 prior to evaluation by flow cytometry. As a positive control for cytolytic T cell activity, a group of mice was immunized i.p. with 2x107 B6/WT-19 cells and evaluated on d7 following immunization for specific lysis against the SV40 TAg site I determinant (SV40T-I). The number of splenocytes remaining in the MHC-I restricted peptide populations (YFI-NS3, YFI-E, OVA-I and SV40T-I) was calculated from total splenocytes. The number of cells remaining in the MHC-II restricted peptide populations (YFII-E and HBV-Core) was calculated using total CD19+ splenocytes. Specific lysis was calculated by the following formula for the YFI-NS3 peptide: ([Peptide in mock mice/OVA-I in mock mice]–[Peptide in immunized mice/OVA-I in immunized mice]) / [Peptide in mock mice/OVA-I in mock mice]. Specific lysis was calculated by the following formula for the YFI-E peptide: ([Peptide in mock mice/SV40T-I in mock mice]–[Peptide in immunized mice/SV40T-I in immunized mice]) / [Peptide in mock mice/SV40T-I in mock mice]. Specific lysis for the YFII-E peptide was calculated using: ([Peptide in mock mice/HBV-Core in mock mice]–[Peptide in immunized mice/HBV-Core in immunized mice]) / [Peptide in mock mice/HBV-Core in mock mice].
Splenocytes and lymphocytes from 17D-204 immunized mice, on d21 post immunization, were pooled and stained with MicroBeads (Miltenyi) as per the manufacturer’s recommendations. Stained cells were subjected to positive selection using the AutoMACS automated sorting system. First the CD19 positive fraction was collected. Then the CD19 negative fraction was stained with CD8 MicroBeads and the positive fraction was collected. Finally, the CD8 negative fraction was stained with CD4 MicroBeads and the positive fraction was retained. Positive fractions were evaluated for purity (S2 Fig) and combined as indicated (Figs 5 and 10), washed two times with PBS and adoptively transferred into naive mice by intravenous injection. The number of each cell type transferred into each mouse was maximized resulting in each mouse from each experiment receiving approximately: CD4+, 1.5x106 cells; CD8+, 3.5x106 cells; CD19+, 5.0x106 cells. In addition, a mix of serum pooled from all 17D-204 immune mice was injected i.p. into naive mice in a volume of 170ul. Twenty-four hours following transfer, mice were challenged with Ang71.
See each table or figure legends for statistical analysis.
|
10.1371/journal.pntd.0003473 | Detection of Rickettsia spp in Ticks by MALDI-TOF MS | Matrix Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry (MALDI-TOF MS) has been shown to be an effective tool for the rapid identification of arthropods, including tick vectors of human diseases.
The objective of the present study was to evaluate the use of MALDI-TOF MS to identify tick species, and to determine the presence of rickettsia pathogens in the infected Ticks. Rhipicephalus sanguineus and Dermacentor marginatus Ticks infected or not by R. conorii conorii or R. slovaca, respectively, were used as experimental models. The MS profiles generated from protein extracts prepared from tick legs exhibited mass peaks that distinguished the infected and uninfected Ticks, and successfully discriminated the Rickettsia spp. A blind test was performed using Ticks that were laboratory-reared, collected in the field or removed from patients and infected or not by Rickettsia spp. A query against our in-lab arthropod MS reference database revealed that the species and infection status of all Ticks were correctly identified at the species and infection status levels.
Taken together, the present work demonstrates the utility of MALDI-TOF MS for a dual identification of tick species and intracellular bacteria. Therefore, MALDI-TOF MS is a relevant tool for the accurate detection of Rickettsia spp in Ticks for both field monitoring and entomological diagnosis. The present work offers new perspectives for the monitoring of other vector borne diseases that present public health concerns.
| Tick-borne rickettsioses include mild to life-threatening diseases in humans worldwide. When removing an attached tick from the human body, patients and physicians may have two questions: 1) is the tick a known vector of a human infectious disease, and 2) is the tick infected by a pathogenic agent that could have been transmitted during the attachment period? The morphological identification of Ticks is difficult, and requires expertise and specific documentation. The use of Matrix Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry (MALDI-TOF MS) has recently emerged as an effective, rapid and inexpensive tool to identify arthropods including Ticks. Here, we show the utility of MALDI-TOF MS for the dual identification of tick species and the rapid detection of Rickettsia spp in Ticks. Such results can be used to guide decisions related to specific patient monitoring or the administration of preventive treatment. Additionally, the low consumable costs, the minimum time required for sample preparation and the rapid availability of the results of MALDI-TOF MS could be useful for epidemiological studies and tick-borne disease monitoring via the dual identification of vectors and the pathogens they carry in one step. These results present new opportunities for the management of other vector-borne diseases that are of importance to public health.
| Ticks are obligate hematophagous arthropods that parasitize vertebrates in almost all regions of the world and are currently considered to be the second-most important vectors of human infectious diseases worldwide, after mosquitoes [1]. Tick-borne rickettsioses are caused by obligate intracellular bacteria belonging to the spotted fever group of the genus Rickettsia. These zoonoses are among the oldest known vector-borne diseases, and include Mediterranean spotted fever, which is caused by Rickettsia conorii conorii and transmitted by the brown dog tick Rhipicephalus sanguineus. Additionally they include most of the emerging tick-borne diseases such as the infection caused by R. slovaca which is transmitted by Dermacentor spp [1, 2].
When removing an attached tick from the human body, patients and physicians may have two questions: 1) is the tick a known vector of human infectious disease, and 2) is the tick infected by a pathogenic agent? Identifying the species of the tick may alert the physician to the diseases that may appear, and knowledge of the infectious status of the tick is a key to evaluating the risk of disease transmission. Both pieces of information, if obtained quickly may be clinically helpful, particularly with regard to decisions about the use of antibiotic prophylactic treatment to prevent tick-borne diseases.
The routine method of identifying Ticks has traditionally been morphological identification using taxonomic keys, entomological expertise and specific documentation [1]. In the past decade, molecular tools have been developed to identify Ticks but these techniques also have their limitations including the selection of ideal primers, the requirement for technically time-consuming and expensive of PCR assays, and the availability of gene sequences in GenBank [1, 3]. More recently, we implemented the use of Matrix Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry (MALDI-TOF MS) in our laboratory as an effective tool to rapidly identify arthropods including Ticks [4–7]. Furthermore, with the creation of a database of reference spectra MALDI-TOF MS profiling of tick leg protein extracts will allow the rapid, cost-effective and accurate identification of Ticks.
For the detection and identification of Rickettsia species in infected Ticks, the most widely available tools remain molecular methods [1], and several Rickettsia DNA sequences can be detected and precisely identified in Ticks by different PCR methods [1]. However, to date, no system allows for the rapid and accurate identification of both the tick species and the Rickettsia spp that the Ticks harbor. Although the MALDI-TOF MS approach has emerged as a routine method for the identification and classification of bacteria for clinical diagnostics [8], no reference spectrum is available for the identification of intra-cellular Rickettsia in the commercial reference spectra database.
The aim of the present study was to determine whether it is possible, to simultaneously identify the tick species and the presence of an associated intra-cellular pathogen in a single assay. To test this, Rh. sanguineus and D. marginatus Ticks that were infected or not, by R. c. conorii or R. slovaca, respectively, were used as experimental models.
Adult laboratory-reared Rh. sanguineus (n = 15) and D. marginatus (n = 20) were used, including rickettsia free specimens and specimens infected by R. c. conorii and R. slovaca respectively. Rh sanguineus were collected in France and Algeria and maintained at the URMITE laboratory. The Rh. sanguineus infected by R. c. conorii were obtained from specimens collected in the field, which were initially infected naturally by R. c. conorii. The vertical transmission of the Rickettsia in these Ticks during their laboratory rearing maintained the presence of R. c. conorii in this colony from generations to generation [9]. The presence of R. c. conorii was regularly confirmed by molecular biological analyses. The laboratory specimens were reared in an environmental incubator (19°C for D. marginatus and 25°C for Rh sanguineus with a relative humidity of 80–90%) and successive generations were obtained by allowing the Ticks to feed on rabbits as previously described [10]. The Ticks infected by Rickettsia spp were maintained in a biosafety level 3 laboratory (BSL-3). D. marginatus Ticks were also collected on dead wild boars killed by hunters in Southern France in order to obtain specimens infected by R. slovaca (see below). They D. marginatus Ticks were morphologically characterized using standard taxonomic keys [11]. For further analysis, each specimen was placed in 1.5 mL microcentrifuge tubes and immobilized or anesthetized at -20°C for 30 min. Whole Ticks were rinsed once with 70% ethanol for 2 min followed by 2 washes with distillated water. After air-drying, all of the legs were removed and two- to four-legs were used either for DNA extraction or sample preparation for MALDI-TOF MS analysis. Additionally, infected Ticks removed from patients including 2 specimens of Rh. sanguineus infected with R. c. conorii, 1 specimen of Rh. sanguineus infected with R. massiliae and 1 specimen of D. marginatus infected with R. slovaca were used. The presence of Rickettsia spp was previously confirmed by qPCR [4].
All processing of infectious Rickettsia spp was carried out in a BSL 3 laboratory. R. c. conorii (ATCC N° VR613) and R. slovaca (CSUR N° R154) were grown into the cell line L929 (ATCC N° CCL-1) for approximately 7 days (+/- 2 days) at 32°C as previously described [12].]. To purify each Rickettsia strain, the infected L929 cells were centrifuged at 11650x g for 10 min. The pellets were rinsed twice in 30 mL of phosphate-buffered saline (PBS) (BIOMERIEUX/France) and centrifuged again at 11650x g for 10 min. The pellets were harvested in 18 mL of sterile PBS, vortexed, diluted in 12 mL of 2.5% concentrated Trypsin (Gibco®) and incubated at 37°C for 60 min. The suspensions were vortexed every 15mn and centrifuged at 11650x g for 10 min. This washing step was repeated three times using sterile PBS; the final suspensions were centrifuged and the pellets were collected in 1 mL of PBS. To eliminate the last cellular debris, two filtrations were performed using 5 μm and 0.8 μm filters (Millipore/France). The purity level and the quantification of the Rickettsia strains was evaluated by Gimenez staining [13] to detect residual cellular debris and to determine bacteria concentration. After purification, serial dilutions of each purified strain was performed in PBS and 10μL of each Rickettsia sample was applied to a 18 Well microscope slide (THERMO Cel-Line Diagnostic 6mm well), fixed by heat during 15min at 100°C, and stained by the Gimenez method [13]. Whole cells or cell debris were stained green and bacteria stained red. The purification rate was determined visually based on the absence of green labelling and the presence of red staining reflecting the individual purified bacteria. Bacteria concentration was estimated by counting all the bacteria in 5 different fields by well at two dilutions under microscopy.
After purification Rickettsia counting was also performed using flow cytometry (BD Accuri C6). The combination of side scatter (SSC) and forward (FSC) correlates with the cell size and the density of the particles of the sample analyzed. In this manner, a bacterial population can be distinguished according to the differences of its size and density without any fluorescent staining. In addition, flow cytometry allowed us to control for the purity of the bacterial based on the absence of whole cells or cell debris.
Serial dilutions of each purified Rickettsia bacteria strains in PBS buffer were performed to determine the optimal concentration for MALDI-TOF MS analysis. The rickettsial strain suspensions were then either immediately used for MALDI-TOF MS analysis or stored overnight at 4°C before MS analysis.
DNA extractions were performed with one or two legs of each tick specimen included in the present study (laboratory and field specimens) using the EZ1 DNA Tissue kit (Qiagen, Hilden, Germany). Rickettsial DNA detection was performed by quantitative PCR using a CFX 96 Real Time System (BIO-RAD, Singapore) and the Eurogentec MasterMix Probe PCR kit (Qiagen, Hilden, Germany) following the manufacturer’s instructions. The presence of R. c. conorii and R. slovaca was determined using the primers R_conorii_6967 and R.slo_7128-R, respectively, which target tRNA intergenic spacers as previously described [14, 15]. A negative control (sterile water containing DNA extracted from uninfected Ticks maintained in laboratory colonies) and a positive control using DNA from R. c. conorii or R. slovaca strains were included in each respective test.
Ticks. Two to four legs of Rickettsia-infected and uninfected Ticks were homogenized manually in 40 μL of 70% formic acid (Sigma, Lyon, France) and 40 μL of 100% acetonitrile (VWR Prolabo) using pellet pestles (Fischer Scientific). All homogenates were centrifuged at 6700 x g for 20 sec and 1 μL of each supernatant was spotted onto a steel target plate (Bruker Daltonics) in quadruplicate. Then, 1 μL of matrix suspension composed of saturated α-Cyano-4-hydroxy-cinnamic acid (CHCA) (Sigma), 50% acetonitrile, 10% trifluoroacetic acid (Sigma) and HPLC water was directly spotted onto each sample on the target plate. Following the drying of the matrix at room temperature, the target plate was immediately introduced into the MALDI-TOF MS instrument for analysis.
Rickettsia species. For protein extraction from each Rickettsia species, a suspension of 500 μL of purified bacteria was centrifuged for 5 min at 14,000 x g. The supernatant was discarded and the pellet was washed twice in 500 μL of pure water, vortexed and centrifuged for 5 min at 14,000 x g. The pellet was then homogenized with 7.5 μL of 70% formic acid and 7.5 μL acetonitrile; after centrifugation at 14,000 x g for 5 min, 1 μL of supernatant was deposited on the target plate in quadruplicate and overlaid with 1 μL of CHCA matrix buffer.
L929 cell line. Uninfected cells were treated with 0.05% trypsin (1X), counted with Kova-Slide and washed twice in 10 mL of PBS; the cells were then centrifuged for 10 min at 262 x g. The pellet was homogenized in 1 mL of buffer to obtain a final concentration of 107cells/mL. After a centrifugation at 14,000 x g for 5 min, 1 μL of the supernatant was deposited on the target plate in quadruplicate and overlaid with 1 μL of CHCA matrix buffer, as described above. The mass spectrometer was calibrated using the Bruker Bacterial Test Standard in the mass range of 2–20 kDa.
Protein mass profiles were acquired using a Microflex LT spectrometer (Bruker Daltonics) with Flex Control software (Bruker Daltonics). The spectra were recorded in a linear, positive ion mode with an acceleration voltage of 20 kV, within a mass range of 2,000–20,000 Da. Each spectrum corresponds to an accumulation of 240 laser shots from the same spot in six different positions. To control the loading on the steel target, the matrix quality and the MALDI-TOF apparatus performance, the matrix solution was loaded in duplicate onto each MALDI-TOF plate with or without Bacterial Test Standard (Bruker Protein Calibration Standard I). The spectrum profiles obtained were visualized with Flex analysis v.3.3 software and exported to ClinProTools version v.2.2 and MALDI-Biotyper v.3.0 (Bruker Daltonics, Germany).
MALDI-TOF MS spectra from the leg protein extracts of 9 D. marginatus infected or not by R. slovaca, and 10 Rh. sanguineus infected or not by R. c. conorii were imported into ClinProTools v.2.2 (Bruker Daltonics, Germany) to identify the specific peaks related to the infection status of the tick. The parameters for ClinProTools software analysis were similar to those previously described [4]. An average spectrum was generated for each condition (i.e., tick species infected or not by Rickettsia spp), using the algorithm “average peak list calculation” tool within the range of 2–20 kDa. The detection of discriminating peak masses was performed by comparison of the average spectrum generated between two classes. The Genetic Algorithm (GA) model of the ClinProTools software was then used to automatically display a list of discriminating peak masses. Based on the selected peak masses, the values of Recognition Capability (RC) and Cross Validation (CV) were determined [16, 17]. The presence or absence of each discriminating peak masses generated by the model was verified by the comparison of each peak mass contained in the peak report created for each species, with the total average spectrum created from all the replicates between two classes (i.e., Rickettsia-infected and uninfected) for each tick species. Additionally the peak mass lists of each Rickettsia strain were retrieved from the Flex analysis v.3.3 software.
The accuracy of MALDI-TOF MS for the detection both of the Ticks and pathogens was assessed in a validation step involving a blind test using other tick specimens that were infected or not by Rickettsia spp, including Ticks collected in the field or removed from patients. MALDI-TOF MS spectra from the leg protein extracts of 3 uninfected D. marginatus, 3 D. marginatus infected by R. slovaca, 2 uninfected Rh. sanguineus and 4 Rh. sanguineus infected with R. c. conorii, were used for a blind test (Blind test 1) with 1 to 4 new specimens per species against our laboratory’s database of reference spectra for (Database 1). This database includes the leg protein spectra of 6 rickettsia free tick species (Amblyomma variegatum infected by R. africae, Rh. sanguineus, Hyalomma marginatum rufipes, Ixodes ricinus, D. marginatus and D. reticulatus), 30 mosquito species (Anopheles gambiae molecular form M and An. gambiae molecular form S, An. funestus, An. ziemanni, An. arabiensis, An. wellcomei, An. rufipes, An. pharoensis, An. coustani, An. claviger, An. hyrcanus, An. maculipennis, Culex quinquefasciatus, Cx. pipiens, Cx. modestus, Cx. insignis, Cx. neavei, Ae. albopictus, Aedes excrucians, Ae vexans, Ae. rusticus, Ae. dufouri, Ae. cinereus, Ae. fowleri, Ae. aegypti, Ae. caspius, Mansonia uniformis, Orthopodomyia reunionensis, Coquillettidia richiardii and Lutzia tigripes,), and other arthropods including louse (Pediculus humanus corporis), triatomine (Triatoma infestans) and bedbugs (Cimex lectularius), as well as the spectra obtained from the bodies (without the abdomens) of 5 flea species (Ctenocephalides felis, Ct. canis, Archaeopsylla erinacei, Xenopsylla cheopis and Stenoponia tripectinata) [4–7]. Then, MALDI-TOF MS spectra from uninfected D. marginatus (n = 4), D. marginatus infected by R. slovaca (n = 4), uninfected Rh. sanguineus (n = 4) and Rh. sanguineus infected with R. c. conorii (n = 5) were added to our database; this upgraded database is referred to as Database 2. The same specimens of D. marginatus, D. marginatus infected by R. slovaca, uninfected Rh. sanguineus and Rh. sanguineus infected with R. c. conorii, were tested in a blind test against Database 2 (Blind test 2). Additionally, the spectra from the leg protein extracts of 3 Ticks removed from 3 patients were also tested against Database 2. The presence of Rickettsia spp was previously confirmed by qPCR including 1 specimen of Rh. sanguineus infected with R. c. conorii (Ct = 22), 1 specimen of Rh. sanguineus infected with R. massiliae (Ct = 24), and 1 specimen of D. marginatus infected with R. slovaca (Ct = 19) (Table 1) [4].
The reliability of the identification was estimated based on the Log Score values (LSVs) exhibited by the MALDI-Biotyper software, between 0 and 3. These LSVs correspond to the degree of homology between the query mass spectra and the reference spectra. An LSV was obtained for each spectrum of the samples tested.
The maintenance of laboratory colony of Rhipicephalus sanguineus and Dermacentor marginatus Ticks [18] has been approved by the Institutional Animal Care and Use Committee of the Faculty of Medicine at Aix-Marseille University, France. The collection of Dermacentor marginatus Ticks in the field did not involve privately owned, wildlife, national park or other protected areas and endangered or protected species.
When the legs of 15 Rh. sanguineus specimens including 8 specimens presumably infected with R. c. conorii and 7 Rickettsia-free specimens from the laboratory colony were tested by qPCR, R. c. conorii DNA was detected in 100% (8/8) of the Rh sanguineus legs predicted to be infected by this bacterium, with a mean Ct ± SD value of 28.76 ±3.27 (Table 1). As expected, R. c. conorii DNA was not detectable in the Rh. sanguineus Rickettsia-free specimens. When the legs of 12 D. marginatus collected in the field were tested by qPCR, 58% (7/12) of the tick legs tested positive for the presence of R. slovaca with a mean Ct ± SD value of 23.93 ± 5.62 (Table 1). Additionally, the absence of R. slovaca from the laboratory reared D. marginatus colony was confirmed by quantitative PCR.
Gimenez straining was performed to determine the purity and concentration of each Rickettsia strain (S1A and S1B Fig.). The absence of green labelling indicated that the purified bacteria samples were free of cells and cell debris. The purity of the samples was confirmed by flow cytometry (BD ACCURI C6 instrument) to detect a homogeneous population of bacteria. Serial dilution of the purified bacteria samples was performed to determine the Rickettsia concentration. Flow cytometry and direct counting on slides by Gimenez labelling led to similar results (S1C and S1D Fig.). The concentration of each purified strain was of 1.6 x107 bacteria /mL and 1.35 × 107 bacteria /mL for R. c. conorii and for R. slovaca, respectively (S1E Fig.) for the MALDI-TOF MS analysis.
Legs from a total of 19 Rickettsia-infected and 13 uninfected specimens belonging to Rh. sanguineus (n = 17) and D. marginatus (n = 15) were subjected to MALDI-TOF MS analysis (Table 1). Although one leg of adult tick was sufficient to generate an accurate MS spectra, to increase the rate of identification, at least two adult tick legs should be included in the preparation for mass spectra analyses (Yssouf et al 2013). Similar MALDI-TOF MS spectra profiles from the leg protein extracts were obtained for each tick species and infectious status. Representative MS profiles with high intensities peaks in the range of 2–20 kDa are presented in Fig. 1. Using Flex analysis software, the alignment of the leg MALDI-TOF MS spectra of 2 uninfected specimens of R. sanguineus and 2 specimens of Rh. sanguineus infected by R. c. conorii, confirmed the reproducibility of the spectra and also revealed changes in the MS pattern according to the infectious status. Comparable results were obtained from MS spectra of D. marginatus specimens infected or not by R. slovaca. Although several protein peaks were conserved in the spectra from specimens belonging to the same species, modifications of the MS patterns were detectable in Rickettsia-infected specimens compared to uninfected specimens (Fig. 2). Technical and biological replicates yielded reproducible spectra (Fig. 1). The spectra of at least 4 specimens of each species (infected and uninfected) were added to our arthropod database (Database 1) in MALDI-Biotyper 3.0, which was designated as Database 2. In parallel, MALDI-TOF MS spectra of each Rickettsia strains were compared to that of the L929 cell line. The alignment of the spectrum profiles of the strains with the cell line using Flex analysis software revealed the absence of peaks with identical masse-to-charge ratios, supporting the conclusion that Rickettsia strains were not contaminated by L929 cell proteins and that the MS spectra corresponded to the Rickettsia strains.
To determine whether the mass spectra data were suitable for the identification of discriminating peaks (m/z-values) according to the Rickettsia-infectious status, 16 to 20 MS spectra per group were selected for further analysis and loaded into the ClinProTools software. Among the Rh. sanguineus and D. marginatus Ticks that were infected or not, by R. c. conorii or R. slovaca, respectively, 76 spectra from 19 specimens that were selected for the MALDI-Biotyper database were imported into the ClinProTools software. The Genetic Algorithm model displayed the peak masses that discriminate between the Ticks that were infected or not by Rickettsia spp with RC and CV values of 100% for both comparisons. After verification of the peak report in the averaged spectrum of the Rh. sanguineus species, 30 biomarker masses were identified that could distinguish Rh. sanguineus specimens that were infected or not by R. c. conorii (Table 2). Among them, 22 peak masses were observed uniquely in the R. conorii-infected specimens and 8 peak masses were associated with the uninfected Rh. sanguineus specimens (Table 2). To confirm the specificity of several of these discriminant biomarker masses, a comparison of the MSP between Rh. sanguineus infected by R. c. conorii and the purified R. c. conorii strain was performed (Table 2). Twelve peak masses were common to both samples, and they were localized in the spectra of Rh. sanguineus infected by R. c. conorii using Flex analysis software (Fig. 3A). Using a comparable strategy for D. marginatus specimens, 35 discriminating peak masses were identified, among which 21 peak masses were specific to spectra from D. marginatus infected by R. slovaca (Table 3). Moreover, among these 21 specific peak masses, 4 were shared between D. marginatus infected by R. slovaca and the purified R. slovaca strain. These 4 peak masses were localized on the spectra profiles of infected D. marginatus using the Flex analysis software (Fig. 3B).
A total 15 specimens, including uninfected and Rickettsia-infected Ticks, were queried successively against the MS reference Database 1 and Database 2 (i.e., Database 2 = Database 1 plus the spectra from Rickettsia-infected Ticks). Using Database 1, the blind test yielded 100% correct identification at the species level for the specimens tested irrespective of their infectious status and their origin of collection (i.e., Ticks that were laboratory-reared, collected in the field or removed from patients). The LSVs of the first top-ranking hits against Database 1 varied from 1.756 to 2.449 (Table 1). Interestingly, the tick specimens infected by Rickettsia spp had lower LSVs than the uninfected specimens. The same specimens were then tested against Database 2, and 100% of the specimens tested possessing a corresponding reference spectrum in Database 2 were correctly identified at the levels of tick species and infectious status (Table 1). Moreover, with the exception of the Rh. sanguineus specimen infected by R. massiliae, only the LSVs from Rickettsia-infected Ticks were increased, and all of these specimens had an LSV larger than 1.85. Interestingly, no association was observed between the cycle threshold value of qPCR and the LSVs. Although no reference spectrum was included in the database for the Rh. sanguineus specimen infected by R. massiliae, it was correctly identified at the level of the tick species as an uninfected Rh. sanguineus specimen, with an LSV greater than 2.
After the demonstration that MALDI-TOF MS profiling is an accurate tool to identify arthropods [19–23], including vectors of infectious diseases such as Ticks [4, 24], the possibility of identifying the presence of microorganisms inside the vectors became evident.
Recently, we showed that the MALDI-TOF MS approach could successfully detect and screen Borrelia spp in their soft tick vectors [25]; the legs of Ticks were used for the dual identification of tick species and the detection of Borrelia relapsing fever [25]. It has also been shown that MALDI-TOF-MS could be employed for the rapid screening of pathogens in tick vectors within the same experiment used for tick identification.
Here, we assess the application of MALDI-TOF MS for the detection of intracellular Rickettsia bacteria and the identification of their respective tick vectors. The present study revealed that the MALDI-TOF MS spectra obtained from two to four tick leg protein extracts were sufficient to accurately identify both the arthropod species and its infectious status. The advantage of performing both of these identifications using only legs is that allows the remaining body parts to be utilized for other analyses. In our study, the infection of Ticks by Rickettsia spp was confirmed by molecular approaches using DNA extracted from the remaining tick legs. In addition to validation of the tick infectious status, Rickettsia specific quantitative PCR confirmed the dissemination of these bacteria in the tick body including the legs.
To evaluate the consequences of Rickettsia infection on the MS profiles of Ticks, we compared the spectra produced by Rh. sanguineus and D. marginatus Ticks that were, infected or not by R. c. conorii and R. slovaca, respectively. The alignment of the MS profiles from Rh. sanguineus Ticks that were uninfected or infected by R. c. conorii led to incomplete superimposable protein profiles. Similar results were obtained for D. marginatus specimens that were infected or not by R. slovaca. The uniqueness of the MS profiles according to the tick species and infectious status suggests that the detected variations could be attributed to the presence of Rickettsia spp. The analysis of the spectra with ClinProTools revealed the existence of specific discriminating peak masses between infected and uninfected specimens. In total 30 and 35 biomarker mass sets distinguished the uninfected specimens of Rickettsia spp from the infected specimens of D. marginatus and Rh. sanguineus species, respectively. Interestingly, although the majority of the discriminating peaks appeared in the protein profiles of the infected Ticks, some were not maintained. This loss of some peak masses could be detrimental to the level of significant identification (i.e., LSVs) of Ticks at the species level. Effectively, our blind test experiments indicated that the LSVs of the infected specimens were lower than those of the uninfected Ticks when compared with Database 1, which included only uninfected specimens. In the future, it will be necessary to test the infectious status of a specimen of new species prior to including the results in the reference database. Moreover, the addition of MS spectra from specimens infected with pathogens will improve the identification of arthropod species and the pathogens that they carry.
In addition, among the discriminating peak mass sets found in the infected Ticks, few of them were shared with their respective purified Rickettsia strains. These masses could correspond to Rickettsia-specific proteins. Moreover, some discriminating peaks detected uniquely in the Rickettsia-infected Ticks were not present in the spectra peaks of the bacteria strains. These differential peak masses could be attributed either to Rickettsia strains (i.e., variations between laboratory and field strains) [26] or to a response of the Ticks to infection [27]. Complementary experiments are needed to test these hypotheses.
The validity of the databases was established by blind tests using infected and uninfected specimens. A query against Database 2 demonstrated that all the specimens possessing reference spectra in the database were correctly identified at the level of the tick species and the Rickettsia-infectious status. Moreover, 86% (n = 12/14) of these spectra presented LSVs greater than 1.9, which is considered to be reliable score for bacterial species identification [28]. Thus, the spectral variations that are detected following Rickettsia infection are sufficient to avoid cross-recognition between uninfected and infected Ticks. Moreover, the presence of Rickettsia in the Ticks did not mask the protein profiles for unambiguous identification at the species level (e.g., querying the MS spectra against Database 1). These results are in agreement with a previous study showing that these variations do not interfere with species determination [24]. However, the absence of corresponding reference spectra in Database 2 for Rh. sanguineus infected by R. massiliae resulted in an incorrect identification of this sample. It is necessary to complete this database with additional tick species infected by Rickettsia strains.
The present study shows that MALDI-TOF MS can be used to reliably identify tick species infected or not by Rickettsia spp without the use of a molecular method requiring DNA sequence information. It is important to note that no Rickettsia spp spectrum is available in the Bruker reference database and that this is the first analysis of Rickettsia strain by MALDI-TOF MS. This work also demonstrated that MALDI-TOF MS could be applied for the rapid detection of Rickettsia spp in Ticks removed from patients. The rapid determination of a tick’s identity and it infectious status should guide decisions related to specific patient monitoring or the administration of preventive treatment. Additionally, the low consumable costs, minimal time required for sample preparation and rapid availability of the results of MALDI-TOF MS could be useful for epidemiological studies and the monitoring of tick-borne diseases via the dual identification of vectors and their borne pathogen in one step. The main obstacle to the use of the MALDI-TOF MS approach is the cost of acquiring the machine, but its use is cost effective thereafter [29]. These results also open new doors for the monitoring and management of other vector-borne diseases that are of importance for public health in human and veterinary medicine. For example, it would be advantageous to test whether MALDI-TOF MS, which has been shown to be a relevant tool for the identification of mosquito species [5, 7, 29, 30], could be useful for detecting the Plasmodium-infectious status of mosquito malaria vectors.
|
10.1371/journal.ppat.1001337 | Clonal Structure of Rapid-Onset MDV-Driven CD4+ Lymphomas and Responding CD8+ T Cells | Lymphoid oncogenesis is a life threatening complication associated with a number of persistent viral infections (e.g. EBV and HTLV-1 in humans). With many of these infections it is difficult to study their natural history and the dynamics of tumor formation. Marek's Disease Virus (MDV) is a prevalent α-herpesvirus of poultry, inducing CD4+ TCRαβ+ T cell tumors in susceptible hosts. The high penetrance and temporal predictability of tumor induction raises issues related to the clonal structure of these lymphomas. Similarly, the clonality of responding CD8 T cells that infiltrate the tumor sites is unknown. Using TCRβ repertoire analysis tools, we demonstrated that MDV driven CD4+ T cell tumors were dominated by one to three large clones within an oligoclonal framework of smaller clones of CD4+ T cells. Individual birds had multiple tumor sites, some the result of metastasis (i.e. shared dominant clones) and others derived from distinct clones of transformed cells. The smaller oligoclonal CD4+ cells may represent an anti-tumor response, although on one occasion a low frequency clone was transformed and expanded after culture. Metastatic tumor clones were detected in the blood early during infection and dominated the circulating T cell repertoire, leading to MDV associated immune suppression. We also demonstrated that the tumor-infiltrating CD8+ T cell response was dominated by large oligoclonal expansions containing both “public” and “private” CDR3 sequences. The frequency of CD8+ T cell CDR3 sequences suggests initial stimulation during the early phases of infection. Collectively, our results indicate that MDV driven tumors are dominated by a highly restricted number of CD4+ clones. Moreover, the responding CD8+ T cell infiltrate is oligoclonal indicating recognition of a limited number of MDV antigens. These studies improve our understanding of the biology of MDV, an important poultry pathogen and a natural infection model of virus-induced tumor formation.
| Many viral infections target the immune system, making use of the long lived, highly proliferative lymphocytes to propagate and survive within the host. This characteristic has led to an association between some viruses such as Epstein Barr Virus (EBV), Human T cell Lymphotrophic Virus-1 (HTLV-1) and Mareks Disease Virus (MDV) and lymphoid tumors. We employed methods for identifying the T cell receptor repertoire as a molecular bar-code to study the biology of MDV-induced tumors and the anti-tumor response. Each individual contained a small number of large (high frequency) tumor clones alongside some smaller (lower frequency) clones in the CD4+ T cell population. The tumor infiltrating CD8+ T cell response was highly focused with a small number of large clones, with one representing a public CDR3 sequence. This data is consistent with the recognition of a small number of dominant antigens and understanding the relationship between these and protective immunity is important to improve development of new vaccination strategies. Collectively, our results provide insights into the clonal structure of MDV driven tumors and in the responding CD8+ T cell compartment. These studies advance our understanding of MDV biology, an important poultry disease and a natural infection model of virus-induced tumor formation.
| Virus driven lymphoid oncogenesis is a serious consequence of infection with a wide range of herpes and retroviral pathogens in a variety of hosts. Major lymphoma-associated infections of humans include Epstein Barr virus (EBV) and Human T cell lymphotropic virus (HTLV) [1], [2]. With both EBV and HTLV tumor progression is a relatively rare event considering the prevalence of infection and the persistent nature of the virus [2], [3]. In contrast, Marek's Disease Virus (MDV) is a widespread, oncogenic α-herpesvirus infection of chickens which readily causes lymphoid tumors and has immense impact on the poultry industry [4]. The oncogenicity of MDV, combined with the ability to vaccinate against tumor formation make the MDV-chicken system an excellent natural infection model for understanding the biology and treatment of viral induced lymphomas [1], [5]–[7].
The spread of MDV occurs through the inhalation of infectious particles in dust. After a brief lytic phase in B lymphocytes (∼2 to 7 days post infection [dpi]), MDV establishes a life-long latent infection in CD4+ T lymphocytes [8]. The life-cycle is completed by transfer of the MDV to the feather follicle epithelium [8]. In susceptible birds, MDV infection leads to a high incidence of CD4+ T cell tumors (up to 100%) in a wide range of organs including heart, liver, ovary, testes, lungs and skin [9]–[14]. These CD4+ tumors express high levels of CD30, a tumor necrosis factor receptor II family member, also over-expressed on human lymphomas with diverse etiologies [5]. MDV latency and tumor formation is dependent upon viral encoded genes such as EcoRI-Q (meq), a c-Jun related molecule [15]–[17].
The penetrance (up to 100%) and temporal reproducibility of tumor appearance after infection (within 3 to 4 weeks) in susceptible lines of bird raises important questions regarding tumor clonality. These include the clonality of transformed cells in individual sites and between sites where multiple discrete solid tumors are evident in a single individual. The MDV genome readily integrates into the host cell genome particularly at telomeric or sub-telomeric locations [18], [19]. The profile of MDV integration within the tumor host cell suggested restricted clonality of most Marek's Disease-derived cell lines and cells taken from tumor sites [18], [19]. Between two and twelve independent integration sites were detected in each sample and the pattern of integration was stable over time in culture. In contrast, analysis of T cell receptor (TCR) Vβ family usage in CD30hi cells from primary lymphomas led to the conclusion that MD tumors were polyclonal [10]. During the analysis of MDV integration patterns, samples obtained from a single chicken contained at least two major distinct patterns [19] suggesting at least two independent transformation events. These data coupled with the possibility of favoured sites for MDV integration [e.g. telomeric or sub-telomeric preference, [19]] suggest that a non-viral integration site dependent analysis of clonality would be appropriate. Since the tumors are derived from CD4+ T cells, the clonally expressed T cell receptor (TCR) would be an appropriate target for the molecular definition of tumor clonality.
The development of successful anti-tumor vaccines against MDV has been critical in poultry production and led to the proposal of utility for MDV as a model for developing vaccines against other lymphoma-inducing viral infections reviewed in [1], [6]. The vaccines are highly effective at preventing tumor formation but fail to eliminate infection or block transmission over prolonged periods [20]. Periodically, circulating strains of MDV develop enhanced pathogenicity and vaccine break has necessitated the development of different generations of vaccines over the past 50 years [Reviewed in [21], [22]]. The success of vaccination indicates acquisition of protective adaptive immunity and both antibody and T cell responses are readily detected [23], [24]. Other evidence for immune protection includes the association of genetic resistance with the MHC (B locus) haplotype [25]–[29]. Similarly, natural infection induces measurable natural killer cell, antibody, T cell and cytokine and interferon responses [30]–[34]. The highly cell associated nature of MDV supports the notion that cell mediated responses may predominate in protective immunity (reviewed by [23], [35], [36] with the CD8+ T cell mediated cytotoxic killing demonstrated in several studies [37]–[39]. The cytotoxic activity in MHC B19 and B21 homozygous chickens was focussed on the MDV-encoded pp38, meq and gB antigens [38]. Importantly, transient depletion of CD8+ T cells rendered chickens more susceptible to infection with MDV [40]. The response to persistent viral infections in humans is often characterised by cytotoxic T cells specific to latency-associated antigens. Indeed, large clones of T cells are readily detected during infection with CMV [41], [42] and EBV [2], [43], [44]. This type of clonal structure within CD8+ T cells is indicative of a response focussed on very few antigens.
The issue of tumor clonality and the nature of the CD8+ T cell response during MDV infection prompted application of the T cell receptor repertoire analysis tools we have recently developed for the chicken [45]. The chicken TCRβ locus in chickens is much simpler than in mammals containing 13 Variable (V), 1 Diversity (D), 4 Joining (J) segments and 1 C segment [45]–[49]. The Vβ segments group into two families, which simplifies global analysis of the chicken TCR repertoire. We applied a combination of CDR3 length analysis (spectratyping) and sequencing of the VDJ-junction (also known as the complementary determining region 3 [CDR3]) to define the clonality of MDV cell lines and different populations of cells from tumors or other sites within MDV infected birds. These approaches revealed clonal structure within MDV tumors (but not always monoclonal) and a pattern of shared and distinct clonal origin in different sites within a single individual. Analysis of the tumor infiltrating and splenic CD8+ T cells allowed identification of large T cell clones within an oligoclonal framework of responding CD8+ T cells.
Inbred line P (MHC, B19/19) white leghorn chickens were reared pathogen free at the Poultry Production Unit of the Institute for Animal Health. One-day-old birds were infected with of MDV strain RB-1B [50] by intra-abdominal injection of ∼1000 pfu cell associated virus and observed for the development of MD using methods described previously [51], [52]. Two of the birds (15 and 16) were sentinel birds and infected by exposure to experimentally infected birds. Birds were reared with ad libitum access to water and vegetable-based diet (Special Diet services, Witham, UK) and wing-banded to allow identification of individuals.
This study was carried out according to the guidance and regulations of the UK Home Office with appropriate personal and project licences (licence number 30/2621). As part of this process the work has undergone scrutiny and approval by the ethics committee at the Institute for Animal Health.
Single-cell suspensions of lymphocytes were prepared from spleen, blood and tumor tissues by Histopaque-1083 (Sigma-Aldrich, Steinheim, Germany) density-gradient centrifugation. CD4+ and CD8+ T cell populations were isolated by positive magnetic cell sorting (AutoMACS Pro Separator, Miltenyi Biotec, Bergisch Gladbach, Germany) according to manufacturer's instructions using FITC conjugated mouse anti-chicken CD4, clone CT-4 and anti-chicken CD8β antibodies, clone EP42 [[53]; SouthernBiotech, Birmingham, Alabama, USA)] and goat anti-mouse IgG microbeads (Miltenyi Biotec). After each antibody treatment, cells were washed three times with PBS containing 0.5% bovine serum albumin with centrifugation at 450 xg for 10 min. The purity of sorted cells was >99% by flow cytometry.
Established lymphoma cell lines derived from MDV-1-induced tumors included MSB1[54], HP8 [55] and HP18 [56], RPL-1 [57]. Four additional MDV cell lines were established from four line P birds infected with pRB-1B5 [51], from testes (T), ovary (O) and spleen (S) tumors according to standard methods [56]. These have been given the following identifiers 4523(T), 4525(O), 4590(S) and 760(O). The Reticuloendotheliosis virus T (REV-T strain)-transformed CD4+ T-cell line AVOL-1 [58], [59] was included as a MDV-negative transformed cell line. Cell lines were grown at 38.5°C in 5% CO2 in RPMI 1640 medium containing 10% fetal calf serum, 10% tryptose phosphate broth and 1% sodium pyruvate.
Tissue samples were stored in RNAlater (QIAGEN Ltd. Crawley, United Kingdom) at −20°C before disruption by homogenization (Mini-bead beater; Biospec Products, Bartlesville, Okla.). Isolated cell subsets or cultured cells were disrupted by resuspension in RLT buffer (QIAGEN) and stored at −20°C. RNA was extracted with the RNeasy Mini kit (QIAGEN) according to the manufacturer's instructions. Contaminating DNA was digested on column with RNase-free DNase 1 (QIAGEN) for 15 min at room temperature. The RNA was eluted with 50 µl RNase-free water (QIAGEN) and stored at −80°C.
Reverse transcription reactions were performed using the iScript Reverse Transcription system (iScript Select cDNA synthesis Kit, Bio-Rad, USA) according to manufacturer's instructions, using 2 µg of isolated RNA from each sample and oligo(dT) primers. Twenty µl of cDNA was obtained for each sample and stored at -20°C.
PCR were performed according to standard protocols. Briefly, cDNA (2 µl) was incubated with 200 µm dNTP, 1.5 mM MgCl2, 1x reaction buffer [50 mM KCl, 20 mM Tris–HCl (pH 8.4)], 2 units Platinum Taq DNA polymerase (Invitrogen), 1 µl of each primer at 10 µM working concentration, in a 50 µl final reaction volume. The forward primer used for Vβ1 and Vβ2 was 5′ACAGGTCGACCTGGGAGACTCTCTGA CTCTGAACTG-3′ and 5′-CACGGTCGACGATGAGAACGCTACCCTGAGATGC-3′ respectively with a common Cβ reverse primer 5′ACAGGTCGACGTACCAAA GCATCATCCCCATCACAA-3′ [60]. The TCRβ locus lies on chromosome 1 with Vβ and Cβ primer design based upon genomic sequence (version 82; http://www.ensembl.org/Gallus_gallus) as described previously (45). The use of primers that lie in conserved regions of the TCR segments minimises any bias associated with PCR amplification. Sequence analysis of samples derived from uninfected birds reveals a polyclonal population of amplified TCR CDR3 with no evidence of PCR bias (45 and our unpublished data).
PCR conditions were as follows, one cycle, 94°C for 2 min, followed by 35 cycles of 94°C for 30 s, 50°C for 40 s and 72°C for 1 min, followed by one cycle at 72°C for 10 min using a G-storm thermocycler (Gene Technologies, Essex, UK) or Eppendorf mastercycler (Eppendorf, Hamburg, Germany). The amplified products were analysed by electrophoresis through 1% agarose (Sigma-Aldrich Ltd, Poole, UK) gels in 1x Tris-borate-EDTA buffer at 50 mA for 1 hr, and products visualized by staining with ethidium bromide (Bio-Rad, Ltd) or GelRed nucleic acid stain (Biotium).
PCR products were purified using QIAquick PCR purification kit (Qiagen Ltd) according to manufacturer's instructions. DNA was eluted in 50 µl nuclease free water and stored at −20°C.
To determine the sequence of the expressed Vβ-chain, PCR products were cloned directly into the pCR4-TOPO vector (Invitrogen) and used to transform competent E. coli, TOP10 (Invitrogen) according to the manufacturer's instructions. After incubation on selective LB agar plates containing 100 µg/ml Ampicillin (Sigma), single bacterial colonies were picked and screened for insert of correct size by PCR followed by agarose gel electrophoresis. Positive colonies were processed using the Qiagen Miniprep kit (Qiagen Ltd) and subsequently sequenced with plasmid-specific (M13 Forward; 5′-GTAAAACGACGGCCAG-3′or M13 reverse; 5′-CAGGAAACAGCTATGAC-3′) or Cβ specific reverse primer (5′-TGTGGCCTTCTTCTTCTCTTG-3′). Alternatively, the plasmid insert amplified by PCR was purified using QIAquick PCR purification kit (Qiagen Ltd) according to manufacturer's instructions and sequenced directly using a nested Cβ specific reverse primer (above). Sequencing was carried out by capillary electrophoresis on the CEQ 8000 sequencer according to the manufacturer's instructions (Beckman Coulter, Fullerton, CA).
Up to 22 (usually ∼15) independent sequences were obtained with each sample. The sample size (n) was chosen with reference to the coefficient of variation of the binomial distribution, which is proportional to 1/√n. This means that the increased precision obtained by raising sample size above ∼n = 15 rapidly reaches a point of diminishing return. Appropriate confidence limits for the repeated sequence frequencies were calculated using the Adjusted-Wald method for binomial proportinos [61]. All sequence data was considered with reference to data generated by spectratype analysis of the CDR3 length profile generated from the total population of cells examined.
To determine the CDR3 lengths of the amplified PCR products by spectratype analysis, a run-off reaction was performed as follows. Five µl of purified PCR product was incubated with 200 µm dNTP, 1 mM MgCl2, 1x reaction buffer [50 mM KCl, 20 mM Tris–HCl (pH 8.4)], 0.5 units Taq DNA polymerase (Invitrogen), 1 µl of a WellRED dye D4 (Sigma) labelled nested Cβ specific reverse primer (5′-TCA TCT GTC CCC ACT CCT TC-3′) at 4 µM working concentration in a 20 µl final reaction volume.
The reaction conditions were as follows, one cycle 95°C for 2 min, followed by 4 cycles of 57°C for 2 min and 72°C for 20 min using a G-storm thermocycler (Gene Technologies, Essex, UK) or Eppendorf mastercycler (Eppendorf, Hamburg, Germany). The run-off reaction products were diluted 5x with nuclease free water and 1 µl of the diluted product was mixed with 40 µl sample loading dye (Beckman Coulter, Fullerton, CA) containing 0.25 µl DNA size standard kit-600 (Beckman Coulter, Fullerton, CA). Samples were transferred into a 96 well plate, overlaid with mineral oil and immediately loaded into a capillary sequencer (CEQ8000 Genetic Analysis System, Beckman Coulter) for fragment analysis. For optimal results, samples were analysed using a modified fragment analysis program (Frag-4) by increasing separation time to 75 min. The data was compiled in CEQ8000 analysis module and for each sample the range of base pair lengths of products was identified and displayed as spectratype profiles. Peak size data was extracted from the fragment analysis software and transferred into Microsoft Excel. Chi-squared tests were used to test whether each CDR3 length distribution differed significantly from that obtained with uninfected birds (TCRVβ1 and TCRVβ2 from unsorted cells or those positively sorted for expression of CD4 or CD8β). The spectratype profiles derived from uninfected birds (n = 3 for each population) exhibited consistently broad CDR3 length distributions that were not statistically different to each other. Reference CDR3 length distributions were constructed for each population by calculating the mean proportion of signal obtained at each CDR3 length from uninfected samples.
In the first instance we selected eight MDV-transformed cell lines [[54], [56], [57], [62] and our unpublished data] and subjected these to TCR repertoire analysis. The REV-T-transformed CD4+ T-cell line AVOL-1 [58], [59] was included for comparison. All of the MD tumor cell lines expressed either Vβ1 or Vβ2 exclusively, whereas the REV-T transformed AVOL-1 cell line expressed both TCR Vβ1 and Vβ2 (Figure 1A). The majority of the randomly selected cell lines (6/7) expressed Vβ1 suggesting a bias in tumor formation between the two avian TCRβ families. The spectratype-derived CDR3 length profiles for each MD cell line comprised a single spectral peak, whereas AVOL-1 contained multiple spectral peaks (Figure 1B). PCR products were cloned into the pCR4-TOPO vector and the inserts sequenced from single colonies of transformed E. coli. For each MD cell line, all inserts contained identical TCRβ CDR3 sequences whereas three sequences were obtained for Vβ1 in AVOL-1 (Figure 1C and S1). Taken together, these data indicate the clonal nature of MD cell lines compared with an oligoclonal structure in the REV-T transformed AVOL-1 cell line.
A fresh ovarian tumor was obtained from one pRB-1B5 MDV-infected bird (designated Bird1) at post mortem (90 DPI). Spectratype analysis revealed a restricted TCRβ repertoire (Figure 2A) with a single spectral peak for Vβ1. The Vβ2 spectratype profile of the ovarian tumor had two main peaks and 3 or 4 minor peaks. With Vβ1 all CDR3 sequences were identical (Figure 2B) corresponding in size to the CDR3 length observed by spectratyping, a profile similar to the tumor-derived cell lines. In contrast, with Vβ2 two repeated CDR3 sequences were detected, one which coded for the amino acid (aa) sequence ‘GIDSD’ at a frequency of 9/21sequences which translates to an estimate of 43% (95%CI 24-63%) of the population and the second, ‘DRG’ at 7/21 (33%, 95%CI of 17–54% of the population). The remaining 5 sequences were singlets. The expanded Vβ2 clones may indicate presence of additional tumor clones, latently-infected T cells or a focussed T cell response infiltrating the tumor. These data demonstrate that MD tumor may consist of a monoclonal Vβ1 and an oligoclonal Vβ2 population. Application of spectratype and CDR3 sequence analysis to T cell populations from uninfected Line P birds revealed polyclonal repertoire profiles with no duplicated CDR3 sequence identified in any sample (data not shown).
Since MDV transforms CD4+ cells [9]–[12], [14] we compared the CDR3 length distribution within unsorted and CD4+ populations of cells derived from tumors. Spectratype analysis of the liver and kidney tumors (32 DPI) from two additional individuals (designated Bird 2 and 3) revealed dramatic restriction in Vβ1 CDR3 length in unsorted cells (Figure 3, left column). These profiles were mirrored by the spectratypes of the CD4+ cell populations in all four tumor samples (Figure 3, middle column). Flow cytometry analysis showed that CD4 + cells represented between 88–98% of the cells derived from whole tumor (data not shown). Cell lines were established from three tumors, two of which had spectratype profiles identical to those detected within isolated CD4+ cells (Figure 3, right column). With the kidney tumor of Bird 3, the CDR3 spectra of cultured cells included a dominant peak of identical length to that in CD4+ T cells but also included a second slightly shorter peak. Sequence analysis revealed dominant sequences that were enriched by sorting for CD4+ cells and by ex vivo culture with the majority being derived from monoclonal expansions (Figure 4). The second spectral peak in the cultured cells of Bird 3 represented a second sequence detected once in the sorted CD4+ cells. Moreover, as a result of analysing two tumors from different organs from each individual, this data set also demonstrated that different tumor clones were present in different sites, with each site dominated by a single Vβ1 clone (e.g. CDR3 aa sequences EWDRGTY and VGGDRLS for Bird 2).
In contrast to Vβ1, the Vβ2 spectratype profiles of the 4 tumors (Figure S2) and corresponding sequences (Figure S3) indicate a wider repertoire although relatively large CD4 + T cell clones were detected in Bird 2 liver and kidney. However, none of these clones could be generated into transformed T-cell lines and may represent non-culturable tumors or a focussed T cell response. To identify the frequency of profiles consistent with metastatic tumor clones (shared clones in multiple sites) and those with independent origin (different clones), we carried out the spectratype analysis of multiple tumor sites from further seven birds (Bird 4 to 10). The profiles obtained for both Vβ1 and Vβ2 are shown in Figure S4 (A for Vβ1 and B for Vβ2). Dominant spectral peaks shared between multiple sites were found in 6 of 7 birds but there were also site-specific over-represented spectral peaks in most individuals, for example with the kidney Vβ1 of bird 7. Overall, the data indicate large bias in the profile of CDR3 length in all tumor sites (p<0.001) and the shared peaks between sites will often be due to a common CDR3 sequence. However, as seen with Bird 2 sometimes the sequence will be distinct despite shared CDR3 length (Figure 4). Interestingly, the dominant spectral peak seen in multiple tumor sites was often evident in spleen and/or blood samples supporting an interpretation of metastatic spread for some tumor clones.
Further spectratype and sequencing analyses were performed to identify the nature of the CD8+ response (see below), where cells from multiple tumor sites were sorted into CD4 and CD8 fractions. The spectratype profiles for whole tumor or sorted CD4+ cells from tumor sites in Birds 11 to 14 were similar to those seen with Bird 1 to 10, with dominant spectral peaks in tumor sites (Figure S5). Some of the dominant peaks were shared between tumor sites within a single bird whilst others were specific for particular sites. The Vβ1 and Vβ2 products were sequenced for all tumor sites in Birds 11 and 12 (Figures S6 to S9). In the absence of culturable T-cell lines generated from these tumors, we tentatively defined tumor-like clones as CD4-enriched and representing greater than 30% of the sequences in any one site (most were much higher frequency than 30%). Specifically, the sequence data for Vβ1 in CD4+ cells from Bird 11 (Figure S6) identifies three large tumor-like clones, “LDGTGGY” (liver only), “RRLTGD” (kidney and as a singlet in ovary) and “LDTGGS” (liver, kidney and ovary). The sequence for Vβ2 in CD4+ cells of Bird 11 (Figure S7) revealed one highly over-represented sequence in all sites (ILRDRGW) that may represent a metastatic tumor and a second in the spleen (IRLGTGGY). For Bird 12 (Figure S8) no Vβ1 CDR3 were represented at over 30% of CD4+ T cell derived sequences but one Vβ2 sequence (Figure S9) with the CDR3 motif “QG” was dominant in the kidney (18/19 CD4+ sequences) and detected in ovary and spleen. A second CD4+, Vβ2 CDR3 sequence “FVMRGID” was dominant in the ovary but not detected elsewhere.
In most individuals the sequencing approach revealed smaller clones of CD4+ cells (repeated but <30% of sequences in any site) including Vβ1 with Birds 2, 11 and 12 and in Vβ2 with Birds 3, 11 and 12 (Figures 4, S3, S6 to S). These sequences may also represent small tumor clones or responding cells but the expansion of one of these sequences in cultured cells from Bird 3 kidney indicates that the “small tumor clone” explanation is valid. Global attribution of the smaller clones of CD4 T cells to a response or tumor phenotype is not possible with the current data sets. Nonetheless, our data clearly demonstrated that culturable tumors were usually dominated by a single T cell clone but that different sites within the same individual can contain independent tumor clones.
The detection of tumor clones in the blood, at post-mortem raised the possibility of identifying tumor clones prior to the occurrence of overt disease. Initial analysis with samples of blood collected ∼2 weeks before the birds exhibited clinical signs supported the notion that the TCR spectratype would be useful to detect tumor clones circulating in the blood. The results of Vβ1 analysis of peripheral blood leukocyte (PBL) samples for two birds (Bird 15 and 16) are given in Figure 5. The samples from liver, kidney, muscle, heart and spleen taken at 49 DPI from Bird 15 revealed a dominant spectral peak that could also be detected in the blood at 42 and 35 DPI (leading to a significant bias in the spectral profile; p<0.001). Similarly, Bird 16 shared the same spectral peak in liver, kidney and ovary with an overrepresented peak and a biased CDR3 profile in the blood at 35 DPI (p<0.001), one week prior to the onset of clinical disease. In Bird 16, there was also a second spectral peak in the ovary and a non-shared spectral peak in the muscle that were not detected in the blood.
A further two birds (17 and 18) were blood sampled serially (twice a week) throughout infection for more precise detection of the tumor clones in the blood, and the results for Vβ1 and Vβ2 spectratypes are depicted in Figure 6. The tumor profile for Bird 17 at post-mortem (33 DPI) indicated a shared spectral profile for Vβ1 in kidney, testes and spleen (and in CD4+ cells isolated from kidney and spleen) and a second site-restricted tumor in the kidney comprising CD4+ Vβ2+ cells. The multi-site tumor CDR3 spectral length was readily detected in the PBL from 16 DPI (p<0.01 and at later time points p<0.001) whereas earlier PBL samples exhibited a “normal” distribution of CDR3 lengths that were not significantly different to the spectral profiles obtained from uninfected birds. In contrast, the site specific Vβ2 tumor was not detected as a spectratype bias in the PBL at any time. The tumor profiles of Bird 18 revealed one shared site (ovary and spleen) Vβ1 tumor, one single site Vβ1 tumor (liver) and one shared site Vβ2 tumor in all three sites (although the more complex ovarian tumor spectratype suggest it may be less highly represented). The multi-site Vβ1 tumor was detected as spectral bias in the PBL between 16 and 19 DPI (p<0.001) although the overall bias was less dramatic than seen with Bird 17.
The spectral profiles of PBL from MDV infected birds indicate that multi site tumor clones can be readily detected in the blood over two weeks prior to clinical symptoms. Unlike the multi site tumors, those restricted to a single site were not detected in the blood. The appearance of tumor clones in the blood affected the repertoire of the overall PBL population especially within the TCRVβ family that comprise the tumor (e.g. for Bird 17, the blood Vβ1 profile was completely dominated by the tumor). Moreover, the disturbance caused by a large CD4+ T cell tumor clone in Vβ1 also affected the repertoire profile of Vβ2 (compare pre- and post- 12 DPI spectratype profiles) with significantly altered CDR3-length profiles in the PBL of Bird 17 at 16 DPI (p<0.005), 29 DPI and 33 DPI (both p<0.001).
Although the nature of the tumor complicates identification of CD4+ T cell responses the CD8+ TCRαβ+ T cells clearly represent a responding T cell population capable of specific recognition, cytokine production and anti-MDV capability [38], [39], [63]. Moreover, in humans infected with persistent viruses (e.g. EBV, CMV and HTLV) the responding CD8+ T cells develop a highly focussed repertoire [2], [41]–[43], [64], [65]. Hence, to define the repertoire of the CD8+ response in MDV infected birds, we isolated CD8+ T cell populations from a range of tumor sites and subjected them to spectratype and sequence based repertoire analysis (simultaneous analysis of CD4+ populations was used to determine the nature of the tumor profiles in these individuals, Figures S5, S6, S7, S8, and S9).
Spectratype profiles obtained for Vβ repertoire analysis of CD8+ cells isolated from multiple tumor sites in four birds (11 to 14) are presented in Figure 7. CD8+ cells represented a minority cell population within the tumor, ranging between 0.4 and 5% by flow cytometric analysis (data not shown). Highly purified CD8+ cells (>99%) exhibited a restricted Vβ1 CDR3 length spectral profile (p<0.001; Figure 7). Within birds, the spectral profiles taken from different sites often included shared peaks detected in multiple samples. The Vβ2 spectral profiles were more variable but were also characteristic of biased populations (p<0.01 to p<0.001) with large over-represented peaks in some samples. The Vβ1 products were sub-cloned and sequenced from all sites in two birds (11 and 12) (Figure 8) allowing identification of clonal expansion by the presence of repeated sequences. These sequences included the CDR3 aa motif “GGS” present in both Bird 11 and 12 as a large, multi-site, overrepresented “public” CDR3 sequence. Considering this clone was the only sequence at this length in either Bird 11 or 12 it is intriguing that this spectral peak was also over-represented in the CD8+ T cells from Bird 13 and 14. Other repeated CDR3 sequences in CD8+ T cells included “RDRGIY” (in liver kidney and spleen), “SRTGGS” (ovary and spleen) and “IFGIY” (spleen) of Bird 11 and “GGSI” in the spleen of Bird 12. Further candidate CD8+ CDR3 sequences were identified as present in unsorted populations and not present in CD4+ sorted populations. These included those revealed by the Vβ2 sequencing efforts; two from Bird 2 (ETGGVY and FAFIDRGI), one from Bird 3 (TIERVD), two from Bird 11 (EVGEILY and TTPQGDRSQ) and one from Bird 12 (RGGYQPA).
Collectively, these results indicate a highly focussed CD8+ T cell response with some clones present at high frequencies in multiple tumor sites and the spleen. The tumor profile of Bird 11 (5 tumor-like clones, with two metastatic) and Bird 12 (2 tumor-like clones with one metastatic and one ovary-restricted) may relate to the identity of the CD8+ T cell expansions seen in different sites. For example the public GSS CDR3 sequence was detected at most tumor sites, whereas some other CD8+ clones were more restricted in their distribution to certain locations.
Based upon an assumption of similar TCR mRNA levels in all cells and the known numbers of Vβ1+ and Vβ2+ CD8+ cells in the tumor and spleen we can estimate the size of the CD8+ clones in the tumor site and spleen. For example, within Bird 11, the splenic population of the three CD8+ Vβ1+ clones “GGS” “RDRGIY” and “IFGIY” each represented 12.5% of the CDR3 which translate into populations of ∼25 million cells (95%CI 4–74×106).. In Bird 12, the public CDR3 aa sequence “GGS” was present in 6/22 (27%) CDR3 sequences from CD8+ T cells representing ∼54 million cells (95%CI 24–96×106) and the private “GGSI” represented in ∼27.2 million cells (95%CI 8–72×106).
For comparative purposes we have displayed the aa identity of all over-represented CDR3 sequences identified in this study and grouped these according to frequency in different T cell subsets (Figure 9). All cell lines contained monoclonal CDR3 sequences except for one short-term cultured cell line, which was biclonal. Within CD4+ T cells derived from tumor sites, fourteen high frequency CDR3 (>50%) were identified with ten represented at greater than 70% of the sequences obtained. Of the 21 “high frequency” CDR3 (established cell lines, ex vivo cultured cells and tumor sites), these were distributed in Vβ1 and Vβ2 based CDR3 (13 and 8 respectively). All four Jβ segments were represented. Other CD4+ CDR3 were present at 10 to 30% with a small number of low frequency (<10%) repeated sequences. Within positively sorted CD8 or non-CD4 (presumably CD8+) populations some large clones were detected, most of which represented private CDR3 but one represented a public CDR3 sequence detected in multiple birds. Samples of T cells from uninfected birds were polyclonal (no repeated CDR3) and none of the CDR3 seen in MDV infected birds was detected (data not shown).
Virus driven transformation of lymphoid cells is a major clinical consequence of infection with persistent infections such as EBV and HTLV in humans. Progress in understanding these human diseases is hindered by the lack of suitable model systems. MDV represents a natural α-herpesvirus of galliform birds capable of inducing rapid onset of tumors in susceptible birds. Losses caused by this group of viruses also represent a substantial problem in their own right; without MDV vaccination the poultry industry would be unsustainable. Indeed the ability to vaccinate against MDV tumor formation has implications for control of medically relevant tumors [1], [6]. Within this framework, we addressed the issue of T cell clonality during infection and tumor formation, dissecting the tumor, spleen and blood to identify repertoire changes in the transformed CD4+ cells and the responding CD8+ cells. With MD almost all cell lines and in vivo tumors have been characterised as CD4+, [9]–[11], [13], [14]. In one study using the intraperitoneal infection route one of twelve cell lines was CD4- CD8α+ but this lacked expression of CD8β [12]. All of the CD8+ samples in this study were prepared using anti-CD8β to avoid isolation of non-classical CD8αα T cells. We chose to examine the Vβ profiles as a measure of clonality since this receptor is clonally expressed with a single in-frame sequence present in each clone of T cells due to the process of allelic exclusion that takes place during T cell development in the thymus [66]–[68].
Tumor clonality is a fundamental issue in MD pathogenesis. The infectious cycle involves transfer of the virus from the lungs to initiate a cytolytic infection in B cells. This is followed by spread and lytic cycling infection largely within CD4 TCRαβ T cell population, before development of latent infection and transfer of MDV into the feather follicle epithelium from where the infectious virus is shed [8]. All infected birds experience a persistent, latent infection, and susceptible birds develop tumors usually within 4 to 5 weeks. Herein resides the problem; if the transformation event is rare, how to explain the high penetrance and temporal reproducibility of the tumor phenotype, unless the “tumors” are induced as a result of polyclonal transformation. Previous studies have addressed this issue in relation to the pattern of MDV genomic integration within the host cell genome [18], [19] or by cell surface staining with CD30 as a tumor associated marker [10]. These two studies reached opposing conclusions, with the restricted MDV integration profiles used to propose clonal tumors (with metastasis), contrasted with the high expression of CD30 in both TCRαβ families within a single tumor being used to propose polyclonality. Our studies using TCRVβ repertoire analysis techniques [45] as a viral integration independent clonal “bar-code” to identify the repertoire of CD4+ TCRαβ+ T cells in tumor-derived cell lines and with in vivo derived tumor samples revealed a characteristic of clonal dominance within an oligoclonal framework of tumor-capable CD4+ T cells.
All of the established tumor-derived cell lines were monoclonal (each expressing a single TCRβ CDR3 sequence), although one short-term line developed during the course of these studies was biclonal at second passage. In contrast the REV-T-transformed AVOL-1 cell line was oligoclonal after over 37 passages expressing at least three TCRVβ1 and one Vβ2 TCR CDR3 sequences. The spectratype profiles obtained with all cell lines were diagnostic in terms of the clonality of the CDR3 as defined by sequence analysis. The clonal structure of the cell lines was not influenced by the length of time in culture which suggests that monoclonality is not an artefact of in vitro selection as a result of multiple passages. It is therefore likely that selection for dominant transformed clones had already occurred in vivo and is retained in MDV cell lines as suggested previously [19]. Furthermore, cell lines generated in this study from fresh tumors expressed a TCR identity shared with the source tumor in vivo. Where cell lines were established most (∼90%) expressed the Vβ1 family of T cell receptors with only one expressing Vβ2, a ratio consistent with the 84% bias previously reported [13].
The profile of most primary tumors was dominated by a single clone of transformed T cells, although biclonal dominance in individual tumor sites was not uncommon. However, sequence analysis revealed smaller secondary clones of expanded CD4+ T cells in most tumors (∼10% of the CDR3 sequences) and the outgrowth of one of these during ex vivo culture indicates the tumor potential of sub-dominant CD4+ clones. Some of the very large clonal populations also failed to establish as tumor cells lines ex vivo, perhaps indicating a phenotypic variability in transformation state. Indeed, considering the very large TCR clones (40 to 100% of CD4+ CDR3 in one site) these were evenly distributed between TCRVβ1 (9 sequences) and TCRVβ2 (8 sequences) (Figure 9). The bias in TCRVβ usage within cell lines may represent a cultivation artefact or reflect the biology of cells expressing different TCR family members. Nonetheless, the multi-step analysis of dominant CDR3 in the primary tumor, in sorted CD4+ cells and after establishment of lymphoblastoid cell lines ex vivo are important in confirming the capacity of the identified large clones to express a tumor capability. All four Jβ segments were present in the CDR3 of both TCRVβ1 and TCRVβ2 expressing large tumor-like clones or in cultured tumor cell lines.
Our data resolves many of the issues surrounding MD tumor clonality. Essentially, we demonstrate clonal dominance within MD tumors (broadly similar to that reported by Delacluse et al., [19]) although our integration-site independent analysis using the T cell receptor CDR3 region revealed a more complex clonal framework within, and between, tumor sites in vivo. Different tumor sites within a single individual may be dominated by shared or distinct clones, hence a single individual may experience multiple transformation events giving rise to tumors that have very different characteristics. On most occasions the dominant in vivo clone present at a particular site was the only clone represented in ex vivo cultured cell lines grown under tumor culture conditions. However, on one occasion one of the lower frequency clones exhibited tumor-like growth patterns ex vivo (alongside the dominant clone in the original site) indicating that some of the smaller clones exist in a transformation capable state. The fact that many individuals harbour both metastatic and single-site tumor clones indicates a complex interplay between transformation and clonal competition. Indeed, with most individuals the overall tumor burden was the result of a small number of independent transformation events (i.e. more than one but fewer than 3 or 4). In contrast, with some individuals the multi-site tumors were the result of metastasis from a single tumor clone. The relationship between these “successful” tumor clones and the infected cell population deserves attention.
In a broader context, the monoclonal origin of adult T-cell leukaemia/lymphoma (ATLL) induced by the human T-lymphotropic virus type -1 (HTLV-1) associated malignancy is well documented [2], [69], [70]. This profile is probably related to the rarity of ATLL even among HTLV-1 seropositive individuals [3] reflecting the acquisition of secondary genomic lesions in persistently infected T cells. Nonetheless, the rapid onset MD tumors with clonal dominance in the context of a more complex framework of oligoclonal expansion may also reflect a circumstance common to other tumor associated persistent viruses of lymphocytes including HTLV-1. Perhaps the main differences may lie in the vigor of MDV-induced T cell replication leading to a compressed time-frame compared with other lymphotropic, tumor associated viruses.
Biological differences were also detected amongst the very large clonal CD4+ “tumors”, with some clones found in multiple sites including the blood and spleen whereas others were located in a single site, indicating phenotypic diversity based upon metastatic capability. The identification of metastatic tumor clones in the blood allowed serial analysis of blood samples from infected birds to determine the dynamics of the appearance of the tumor clone, in relation to the time of infection and onset of clinical signs. The spectratype analysis of blood samples prior to infection and in the first 10–14 days revealed a profile consistent with a polyclonal population of circulating cells. However in some cases, the ‘tumor-specific’ spectratype signature could be detected in blood 12 to 16 dpi, more than two weeks before appearance of clinical signs. The appearance of the tumor clone at detectable levels in the blood supports the proposal of an early transformation event. The level of tumor clone expansion in the blood compartment at the onset of clinical disease was extreme, and in some individuals, these were the only T cell clones detectable (e.g. within TCRVβ1 for Bird15 and 17) represented the tumor (Figure 5 and 6). There was also evidence for disturbance within the polyclonal repertoire in TCRVβ2 expressing cells (Figure 6) suggesting that the blood niche for T cells was being filled by the tumor. Hence, with a circulating TCR profile dominated by a single clone, it is of little surprise that MDV-infected birds develop immune deficiency [reviewed in [71]]. These dramatic repertoire changes would have greater impact than the reported changes in cytokine production [72] and would be immunologically catastrophic. Infiltration of the skin with CD4+ T cells is a consequence of MDV infection [73], [74] [75] and the high frequency tumor clones in the blood are likely to represent the relocation of MDV to the site of transmission.
In mammals, many persistent viral infections including EBV, CMV and HTLV stimulate highly focussed repertoire expansion in responding CD8+ T cells [2], [76], [77]. Although the MDV tumors were populated by relatively small numbers of CD8+ T cells, their repertoire was highly structured and oligoclonal in nature. The CD8+ T cell clone sizes of around 25 to 50 million cells are similar to those reported during persistent viral infections in humans [78]. However in the case of MD, these are developed over a much shorter period of time than considered with mammalian infections. For example, taking a conservative estimate of prolonged T cell division of 12 hours/division [79] and assuming no cell death (unlikely), the latest time point for initial stimulation of the CD8+ T cell would be ∼15 days prior to sampling. This calculation would place the initiation of these clones of specific CD8+ T cells at ∼15 DPI, probably earlier, around the time at which latent infection was initiated. The rapid focussing and clonal expansion of the MDV-specific repertoire suggests restriction to a small selection of MDV antigens. Indeed, Omar and Schat [38] examined the cytolytic response of infected birds against a panel of cell lines expressing individual genes from MDV found that in MHC B19 homozygote Line P2a birds, the cytolytic activity was restricted to meq, gB and pp38 antigens, while the genetically-resistant line N2a (B21) birds also detected the ICP4 antigens. In our studies, tumor-infiltrating CD8+ T cells produce greater levels of IFNγ mRNA than CD8+ T cells derived from the spleen of uninfected birds (unpublished data, Mwangi, Peroval et al.,). The CD8+ T cell response of susceptible birds is insufficient to prevent tumor progression; our data provides a framework for comparisons with resistant or vaccinated birds which do not develop tumors. Our sequence analysis clearly detected large CD8+ T cell clones and allowed an approximation of the clone size, the application of higher throughput sequencing technologies may be useful in the future to identify smaller clonal expansions and provide more accurate estimations of clone sizes. Understanding the nature of the TCR repertoire to specific antigens after infection and vaccination can be used to improve vaccine approaches in the future. The rapid nature of focussing within the CD8+ population may reflect a combination of the minimal MHC configuration where each haplotype is dominated by presentation through a single MHC class I gene [80] and the minimal TCRVβ locus with 13 Vβ segments in just two families [45].
The high frequency CD8+ T cell clones were found in both tumor sites and in the spleen of infected individuals, either restricted to one tumor site or present in multiple tumor sites. One of the largest CD8+ clones has a CDR3 sequence (“GSS”) of note, in that identical sequences were detected in different individuals. This type of CDR3 is known as a “public” TCR rearrangement and, although previously reported with mammals, is relatively rare [81]. Upon closer examination, it was clear that the public GSS amino acid sequence for the CDR3 also represented shared nucleotide sequence in different individuals. Interestingly, the GSS sequence represents retention of a fragment of the D segment, after deletion of six nucleotides in the D and three nucleotides in the Vβ1 segment. Although not noted previously, it is clear that a CDR3 constructed by deletion (with no retained nucleotide addition) is much more likely to occur in multiple individuals than one generated by addition of nucleotides. We propose that public CDR3 sequences in other contexts (e.g. in humans) may also conform to this arrangement, representing a deletion-based junctional modification. This feature might be useful and exploitable in diverse scenarios to improve “public” responses to vaccines. The remaining CDR3 sequences positively identified as clonal expansions in CD8+ cells (or as not in CD4+ cells) all represented “private” CDR3 identities (Figure 9).
In this report, we have documented the TCR Vβ repertoire changes associated with infection, tumor development and anti-tumor response that characterise MDV pathogenesis. Upon consideration of our data in the context of previous reports, we propose that the MD tumors are dominated by clonal expansion in an oligoclonal framework of minor clones of pre-cancerous cells. We propose that this type of population structure explains the penetrance and narrow temporal window that characterise MD in susceptible birds. The CDR3 analysis identified that all established MDV-transformed cell lines tested were clonal (with one bi-clonal short term culture), and that these clones represent dominant clones detected in vivo. Within birds harbouring multiple tumors there was a mixture of metastatic and site-specific tumor clones. Overall, we examined 50 tumors derived from 21 individuals, and all tumors were dominated by one or two clones with some birds harbouring a single metastatic tumor clone and others with different clones in different sites. The TCR repertoire analysis system has allowed examination of diverse areas of MD lymphoma biology and the CD8+ response against the infection. We consider that this type of approach can be used to further define MD pathogenesis and the response generated against infection and/or tumors. These types of study also have the potential to impact much more broadly, identifying strategies to vaccinate against or otherwise control viral driven lymphomas in medical and veterinary fields.
|
10.1371/journal.ppat.1004628 | Within-host Competition Does Not Select for Virulence in Malaria Parasites; Studies with Plasmodium yoelii | In endemic areas with high transmission intensities, malaria infections are very often composed of multiple genetically distinct strains of malaria parasites. It has been hypothesised that this leads to intra-host competition, in which parasite strains compete for resources such as space and nutrients. This competition may have repercussions for the host, the parasite, and the vector in terms of disease severity, vector fitness, and parasite transmission potential and fitness. It has also been argued that within-host competition could lead to selection for more virulent parasites. Here we use the rodent malaria parasite Plasmodium yoelii to assess the consequences of mixed strain infections on disease severity and parasite fitness. Three isogenic strains with dramatically different growth rates (and hence virulence) were maintained in mice in single infections or in mixed strain infections with a genetically distinct strain. We compared the virulence (defined as harm to the mammalian host) of mixed strain infections with that of single infections, and assessed whether competition impacted on parasite fitness, assessed by transmission potential. We found that mixed infections were associated with a higher degree of disease severity and a prolonged infection time. In the mixed infections, the strain with the slower growth rate was often responsible for the competitive exclusion of the faster growing strain, presumably through host immune-mediated mechanisms. Importantly, and in contrast to previous work conducted with Plasmodium chabaudi, we found no correlation between parasite virulence and transmission potential to mosquitoes, suggesting that within-host competition would not drive the evolution of parasite virulence in P. yoelii.
| Malaria infections are very often composed of multiple strains of malaria parasites. It is thought that these strains may compete for resources such as space and nutrients within the host. Here we show that such “within-host competition” has repercussions for the virulence of the malaria infection, so that infections composed of multiple strains are more virulent in terms of disease severity, than single strain infections containing the constituent parasites. Following from this, it has been proposed that as such competition would favour those parasites with faster growth rates, then these parasites would be selected in nature when within-host competition is common. We show, however, that this is not necessarily the case, as parasites with faster growth rates in the mammalian host were no more successful at transmitting to mosquitoes than parasites with slower growth rates. These results show that a reassessment of our current understanding of the role of within-host competition in the selection of virulence in malaria parasites is required.
| Malaria is caused by a diverse group of parasites composed of at least six species of the genus Plasmodium. Genetic diversity within these species is high, with multiple strains often co-infecting the same host, and is driven and maintained by mutation and through recombination between strains. Concomitant infection of hosts, both man and mosquitoes, with multiple species and/or strains is a common occurrence in endemic areas [1–3]. Such infections may result from the bites of multiply infected mosquitoes or from the bites of multiple mosquitoes harbouring different species or strains. Co-infecting species or strains interact during their life cycles and such interactions may lead to intra-host competition with repercussions for the host, the parasite, and the vector in terms of disease severity, vector fitness, and parasite transmission potential and fitness.
Within-host interactions between different parasite genotypes have been observed in both empirical human [4, 5] and rodent [6] malaria studies, and these have often been observed to result in modulations of parameters such as infection dynamics (suppression or enhancement of a particular strain or species in a mixed infection) and virulence (harm caused to the host). A series of experiments performed exclusively with strains of the rodent malaria parasite Plasmodium chabaudi, suggested that faster growing strains gained a competitive advantage over slower growing strains [7], in that they often dominated mixed strain infections in terms of proportional numbers of parasites, and sometimes competitively excluded the slower growing strain at some point during the infection. This has been interpreted as suggesting that within-host competition could lead to the selection of virulence within a parasite population.
The effects of competition between parasites on disease pathology is of particular relevance in malaria, as understanding the links between parasite genetic and disease severity will allow an understanding of how interventions, such as drugs and vaccines, that reduce parasite diversity will impact human health. There are conflicting theories as to how the presence of multiple strains (and/or species) of malaria parasites in an infection impact on disease. Reports from malaria endemic regions suggest that it is possible that disease pathology may be exacerbated by within-host competition [8] but also that it can result in decreased parasite burden and may protect against some clinical outcomes of disease [9–14].
Interactions between different Plasmodium strains or species concurrently infecting the same host (vertebrate or vector) may also influence the transmission dynamics of each species or strain, affecting their fitness, and driving the selection of those parasites that are good at competing. This has been observed in both human [3, 15–18] and rodent [19, 20] malaria infections. Laboratory experiments, field studies and mathematical modelling have been employed to describe the mechanisms driving the evolution of various phenotypes, including virulence. Virulence (defined as “harm to the host”) is often, but not always, linked to replication rate, with more virulent strains growing and replicating faster in the host than less virulent strains. Virulence differences can occur as a result of inherent genetic differences between strains, and/or through the influence of environmental factors, the most relevant of which to a parasite is the condition of the host [21]. It has been proposed that the evolution of virulence is driven by within-host competition between strains of malaria parasites in mixed infections [7, 22]. This theory is based on the idea that faster multiplying parasites out-compete others in a mixed infection and therefore transmit more successfully to the vector.
The consequences of within-host competition during the mosquito stage development of malaria parasites for both the vector and the parasite are very poorly understood. This is partially due to the fact that there is little understanding of how malaria parasite infection influences mosquito fitness. Some reports associate malaria parasite infection with decreased survival and reproduction of mosquitoes [23–29], others find no effect [30] while there is also evidence to suggest that that malaria infection increases longevity in mosquitoes due to a trade-off with decreasing reproduction [31]. These contradictions notwithstanding, many lab-based and field studies have convincingly established that Plasmodium causes pathological changes not only in their vertebrate hosts, but also in insect vectors. This is particularly evident during earlier stages of Plasmodium infection in the mosquito whereby ookinetes penetrate mosquito midgut epithelium (physical damage) and provoke physiological stress [32]. As a result, the mosquito launches an immune response to curb the infection, which is an energetically demanding process [33], hence reduced fitness. In addition, meta-analyses suggest that malaria parasites reduce mosquito fitness and survival [34]. Overall, Plasmodium infection is generally considered to be harmful to their mosquito hosts. What effect competition between parasite species and/or strains has on the pathology of the mosquito stages of malaria infections is largely unknown.
Here we use the rodent malaria parasite Plasmodium yoelii to explore the consequences of within-host competition on disease severity and parasite fitness (including transmission potential) between isogenic parasite strains with varying degrees of virulence.
Infection parameters for all single and mixed infections are summarised in Table 1. Of the three strains, only the virulent strain (17XL) causes death of mice in single strain infections. In mice infected with this strain, death occurs early in the infection, by day 5 post-infection. Neither of the avirulent strains (17XNL and CU) or the intermediately virulent strain (17X1.1pp) cause host mortality at any time, and infections self-clear within 30 days.
Peak parasitaemia; Mean (±SEM) of highest parasitaemia, Cumulative parasitaemia; Mean of area under curve of the parasitaemia curve in single and mixed infections. Maximum Weight loss is the Mean (±SEM) of maximum weight lost by the infected mice. Minimum RBC count is the Mean (±SEM) of least Red Blood Cell number per mL of blood. Days post infections are given in parentheses. Data were generated from groups of 4 mice, and are representative of two independent repeat experiments.
To analyse time course data we fitted general linear mixed models with Treatment and Time and their interaction as fixed factors, and Mouse as a random factor nested within treatment. To account for possible autocorrelation of errors through time, we compared the fit of models with and without an autocorrelation term [35]. Models were fitted using REML, and compared using Likelihood ratio tests. In cases where the autocorrelation term significantly improved the fit of the model it was retained for subsequent analysis of the fixed effects. The significance of fixed effects was then determined by fitting models with and without the term of interest using Maximum Likelihood, and comparing the fit of these models with a Likelihood ratio test. Parasitaemia values were log transformed prior to analysis to meet assumptions of homogeneity of variance whilst other response variables were analysed on the measured scale. All analyses were carried out using the LME function in R [36]. Time courses for infections involving the virulent strain were only analysed up to day 5, after which point all mice had died.
Both avirulent/avirulent (CU+17XNL) and avirulent/intermediately virulent (CU+17X1.1pp) strain mixed infections were characterised by protracted parasitaemia and prolonged chronicity of disease compared to their constituent strains growing in single strain infections. The mixed strain infections resulted in higher parasitaemia than either of the constituent strains in single infections late on in the infection (treatment by time interaction term, L = 306.28, df = 50, P < 0.0001, and L = 579.48, df = 50, P < 0.0001, Fig. 1, Panels A and B). For the avirulent/virulent (CU+17XL) mixed strain infections, the course of infection followed that of the most virulent constituent strain (Fig. 1 Panel C, treatment by time interaction term, L = 111.69, DF = 8, P < 0.001). Strikingly, infection with the avirulent/intermediately virulent mixture resulted in 75% host mortality infection late in the infection (Log-rank test, χ2 = 8.165, df = 2, p = 0.0169, Fig. 1 Panel E), and this was associated with the inability to clear parasites from the blood.
In order to compare the virulence (here defined as pathological harm to the host) of mixed strain infections to that of single strain infections, we measured the red blood cell (RBC) density, and weight of mice daily throughout the course of the infections (Fig. 1, panels G-I and J-L, respectively), sampling ceasing when parasites were no longer visible in the blood by microscopy, or when mice succumbed to infection.
Mixed infections of avirulent/avirulent (CU + 17XNL) resulted in lower RBC density and greater weight loss compared to either of the constituent strains in single infections during the latter stages of the infection (Fig. 1, Panels G, L = 97.90, DF = 86, P = 0.0004 and J, L = 131.38, DF = 86, P < 0.0001), a phenomenon consistent with the prolonged chronicity of the mixed infection parasitaemia. In the avirulent/intermediately virulent (CU + 17X1.1pp) mixed infection, this effect was more pronounced, with dramatic and significantly lower RBC count (Fig. 1, panel H, L = 228.75 DF = 86, P < 0.0001) and significantly greater weight loss (Fig. 1, panel K, L = 171.70 DF = 86, P < 0.0001) compared to single strain infections, This weight loss and reduction in RBC density occurred in the latter part of the infection, reflecting the significantly higher parasitaemia of the mixed infection group during this period. Single infections of the most virulent strain (17XL) result in acute anaemia and dramatic weight loss early in the infection, and this pattern is also observed in the mixed strain infection containing this strain plus an avirulent strain (CU + 17XL) (Fig. 1, Panels I and L).
In order to determine whether mixed strain infections infect more mosquitoes and result in higher oocyst burdens than single strain infections, we allowed A. stephensi mosquitoes to feed on mice with single or mixed strain infections. As transmissibility varies dramatically throughout an infection, we allowed mosquitoes to feed at two separate time points; on day 3 post-infection, when the oocyst conversion rate (OCR, the number of oocysts produced per gametocyte) is at its highest in P. yoelii infections (R. Culleton, unpublished observations), and on day 4 post-inoculation when OCR is lower.
We observed no significant differences in the numbers of gametocytes produced in single strain infections compared to mixed strain infections, with the exception of the mixed infection composed of the avirulent CU and the intermediately virulent 17X1.1pp on day 3 pi, which contained significantly fewer gametocytes than the single 17X1.1pp infection (S1 Fig.).
All infections, regardless of virulence or whether mixed or single, resulted in a lower percentage of mosquitoes infected and lower oocyst burdens following feeding on day 4 pi compared to day 3 pi (Fig. 2). For the single infections, the greatest percentages of infected mosquitoes (mean, n = 4 mice) were found in those that had fed on CU (100%) or 17X1.1pp (100%), followed by those fed on 17XNL (70.5%) and 17XL (56.2%). The highest mean day 3 pi oocyst burdens were recorded for 17X1.1pp (105 oocysts per infected mosquito, p.i.m), followed by CU (96 oocysts p.i.m), 17XNL (17 oocysts p.i.m), and 17XL (10 oocysts p.i.m). This pattern was similar on day 4 pi, when the highest percentage of infected mosquitoes was achieved by 17X1.1pp (74.5%), followed by CU (65.6%), 17XNL (19.5%) and 17XL (8.3%), with associated mean oocyst burdens of 13, 8, 3, and 2 oocysts p.i.m, respectively. There was, therefore, no positive correlation between virulence and transmission potential.
To test for differences in transmission potential (defined as the percentage of mosquitoes carrying one or more oocyst following blood feeding on infected mice) between mixed and single infections, we fitted a generalised linear mixed model, with treatment and day and their interaction as fixed factors, and mouse as a random factor to account for the repeated measures on each mouse. Since infection is a binary trait, we fitted a binomial error structure. Models were fitted using the glmer function in R with Likelihood ratio tests used to compare models with different fixed effects.
In mixed infections containing two avirulent (CU + 17XNL), There was no significant interaction between day and treatment, (L = 2.231, df = 2, P > 0.05), whilst treatment affected the proportion infected on both days (L = 12.369, df = 2, P = 0.002, with 17XNL infecting fewer mosquitos than either CU or the mixture (Fig. 2, Panel D).
For the mixed infection containing one avirulent and one intermediately virulent strain (CU + 17X1.1pp), again, there was no indication of an interaction (L = 4.75, df = 2, P > 0.05). Infection was lower on day 4, than on day 3 (L = 30.545, df = 1, P < 0.001), but there was no effect of treatment (L = 0.2656, df = 2, p > 0.05, Fig. 2, Panel E).
When mosquitoes fed on mice infected with the avirulent/virulent (CU + 17XL) mixed infection, the effect of treatment on infection depended on day (L = 6.681, df = 2, P < 0.05), with CU having higher infectivity than both other treatments on day 4, and than 17XL on day 3, (Fig. 2, Panel F).
In summary, mixed infections did not infect significantly different percentages of mosquitoes than one of the constituent strains (CU, the most successful transmitter), with the exception of the CU+17XL infection when mosquitoes fed on day 4 of the infection. In this case, there were significantly fewer mosquitoes infected than in the CU single infection.
To analyse oocyst number, we used a general linear mixed model with day and treatment as fixed factors, and mouse as a random factor. Data was logged prior to analysis to meet the assumptions of homogeneity of variance.
In mixed infections containing two avirulent (CU + 17XNL), there was a significant effect of treatment on oocyst numbers (treatment main effect, F2,9 = 13.71, p = 0.002), with CU producing more oocysts that the mixture and 17XNL producing less. There were fewer oocysts on day 4 than day 3 (F1,9 = 78.01, p < 0.001), but the effect of treatment was consistent on both days (F2,9 = 0.781, p = 0.4868). Post-hoc tests revealed that the mixed infection produced significantly more oocysts than 17XNL (P = 0.0031), but oocyst numbers did not differ significantly from CU in a single infection (Fig. 2, Panel A).
In the mixed infection composed of an avirulent and an intermediately virulent parasite (CU+17X1.1pp), there were no significant differences in the numbers of oocysts produced by the mixed infection compared to the constituent strain single infections (Fig. 2, Panel B).
When an avirulent and a virulent parasite strain co-infect a host (CU + 17XL), thee numbers of oocysts produced in mosquitoes fed on single strain and mixed infections is significantly different (treatment main effect, F2,9 = 18.28, p < 0.0001), with CU producing more oocysts than the mixture and 17XL producing less. There were fewer oocysts on day 4 than day 3 (F1,9 = 52.14, p < 0.001), but the effect of treatment was consistent on both days (F2,9 = 3.42, p = 0.0787). Post-hoc tests revealed that CU differs significantly from the mixture (P = 0.003), but for 17XL the difference is marginally non-significant (Fig. 2, Panel C).
We measured the relative proportions of each of the strains within mixed infections by strain-specific qPCR every day throughout the course of the infection. In the mixed infection composed of the two avirulent strains CU and 17XNL, the proportions of the two strains fluctuate between 65% and 35% during the first 16 days of the infection, with 17XNL dominating for the majority of this period. From day 16 pi onwards, however, the proportion of 17XNL relative to CU drops daily until finally, at day 22 pi, there is complete competitive exclusion of 17XNL by CU (Fig. 3, panel A). In the mixed infection containing avirulent (CU) and intermediately virulent (17X1.1pp) parasites, the intermediately virulent strain completely dominates the infection from day 4 pi until day 14 pi, during which period no avirulent parasites could be detected. However, this situation is dramatically reversed from day 16 pi, when the avirulent parasite re-emerges, and completely dominates the infection from day 20 until the infection is cleared, completely competitively excluding the intermediately virulent strain (Fig. 3, panel B). In the case of the mixed infection composed of a virulent (17XL) and an avirulent (CU) parasite, the virulent parasite completely dominates the infection from day 4 pi, competitively excluding the avirulent parasite, as all mice die due to the infection on day 5 pi (Fig. 3, panel C).
We next compared the numbers of parasites produced by each strain (in terms of parasite density, defined as the number of parasites per mL of blood at a particular time-point) throughout the course of single infections, with the numbers produced when in competition with another strain. The avirulent strain CU is competitively supressed by all strains during the first 20 days of the infection, with competitive suppression strongest when in competition with the virulent strain (17XL) (mice die at day 5 pi), followed by the intermediate strain (17X1.1pp) and suppression mildest when in competition with the avirulent strain (17XNL). When in competition with the avirulent (17XNL) and the intermediately virulent (17X1.1pp) strains, competitive suppression ceases at day 20 pi, and competitive release occurs, with CU parasite densities reaching higher levels than in single infections (Fig. 3, Panel D). Thus, over the course of an infection in which CU is mixed with a strain of the same or higher virulence, both competitive suppression and facilitation occur.
The avirulent parasite 17XNL is not supressed in competition with CU compared to growth in single infections during the first 16 days of the infection, after which time it suffers competitive suppression (Fig. 3, panel E). A similar trend is seen with the intermediately virulent 17X1.1pp strain, which is completely unaffected by the presence of CU in mixed infections up to day 16, from which point on it suffers from competitive suppression (Fig. 3, panel F). The virulent 17XL strain is completely unaffected by the presence of an avirulent competitor (CU) in a mixed infection throughout the 5 days during which mice survive (Fig. 3, panel G).
We measured the relative proportion of each of the strains in mixed infections in mosquitoes using qPCR on DNA extracted from mosquitoes with known numbers of oocysts, and compared the adjusted number of oocysts per strain (total number of oocysts multiplied by the frequency of the strain measured by qPCR) to the numbers produced in single infections. This analysis was performed on oocyst DNA extracted from mosquitoes fed on mice at days 3 and 4 pi.
We fitted a general linear mixed model, with infection treatment and day fitted as fixed factors, and mouse fitted as a random factor nested within treatment. Adjusted oocyst number was log transformed to meet homogeneity of variance assumptions. For the mixed infection containing the virulent parasite 17XL, as there was very little transmission on day 4, we were unable to use a general linear model, and t-tests were used in its place.
The avirulent clone CU was significantly less successful at transmitting to mosquitoes in mixed infections with all competing strains on day 3 pi (F3,12 = 7.78, P = 0.0038), when transmission capacity is at its peak in the P. yoelii / Mus musculus / Anopheles stephensi malaria system. The effect of infection treatment differed for the two days (Treatment*Day interaction, F3,12 = 3.68, P < 0.0433) with CU not producing significantly differing numbers of oocysts in single compared to mixed infections on day 4 (Fig. 4, panel A). Importantly, the virulence of the competing clone had no effect on the degree of competitive suppression of transmission potential.
The transmission of the avirulent parasite 17XNL was severely and significantly reduced in mixed infections compared to single infections on both days 3 and 4 (F1,9 = 30.48, P = 0.0004, Fig. 4, panel B). The intermediately virulent strain 17X1.1pp suffered a less drastic and not statistically significant reduction in transmission success through competition with the avirulent strain CU on day 3 pi and there was no difference on day 4 (Fig. 4, panel C). Finally, the transmission of virulent strain 17XL to mosquitoes was severely and significantly reduced in the presence of an avirulent competitor on day 3 pi (Student’s two-tailed t-test, t = 3.282, df = 64, P = 0.0017), and completely abrogated on day 4 pi (Fig. 4, panel D).
Using the oocyst data and relative proportions of each strain described above, we determined a fitness coefficient, reflecting the relative contribution of each strain to the products of fertilization in the mosquito midgut (i.e. oocysts) for each parasite strain either in single infections (Fig. 5, panel A), or in competition with the other strains (Fig. 5, panel B). The CU strain exhibits the highest fitness in single infections, followed by the intermediately virulent 17X1.1pp, the avirulent 17XNL, and finally the virulent 17XL. All strains are negatively affected by competition, with 17XL and 17XNL particularly severely compromised when in competition with CU. The CU strain is least affected by the presence of 17XL followed by 17X1.1pp and 17XNL.
In order to test whether infections with avirulent parasite are infectious to mosquitoes during the latter stages of the infection, we allowed mosquitoes to feed on mice infected with 17XNL on day 18. We found that these mosquitoes were infected with oocysts following feeding, and so transmission is possible during the latter stages of infection, at least with this strain.
Finally, we assessed whether the proportion of strains measured in oocysts was representative of the proportion of strains inoculated into mice during mosquito feeding. We found a good correlation between the proportions of strains in oocysts, and the proportions in the blood of mice in infections resulting from inoculation of sporozoites from tested mosquitoes (S2 Fig.).
The consequences of within-host competition on disease severity in the mammalian host Mixed strain malaria parasite infections, in which multiple genetically distinct parasites co-infect the same host, are common in nature, and are probably more common than single-clone infections, certainly in regions with relatively high transmission rates [4, 10, 37, 38, 39, 40, 41, 42, 43]. It is becoming increasing clear that malaria parasites co-infecting the same host interact with each other [43–45]. The consequences of these interactions for disease severity have been the subject of a limited number of field studies, some of which report that disease severity increases with increasing parasite genetic diversity within infections [46–48]. The general consensus from these studies is that co-infections result in competition between strains, and this competition can lead to increased virulence of co-infections compared with single clone infections [22].
Our results indicate that simultaneous inoculation with two genetically distinct strains of P. yoelii result in infections that are more virulent than single-clone infections composed of either of the co-infecting strains. Virulence in this case was measured by assessing host weight loss, RBC density and parasitaemia. This was true when the co-infections were composed of two avirulent strains or one avirulent and one intermediately virulent strain. When co-infections contained a very virulent strain and an avirulent strain, then the virulence of the co-infection was not significantly different from the virulence of the infection caused by the virulent strain. In this case, however, the virulent strain causes very severe disease, and host death within five days, making measurement of any possible increased virulence in a co-infection very difficult. Co-infections always resulted in infections that were either more virulent than each of the constituent strains, or which were as virulent as the most virulent of the constituent strains; we did not observe any protective effects of co-infection on host disease severity.
Our results show that in experiments with the rodent malaria parasite P. yoelii, mixed infections cause more harm to the host than single infections, a phenomenon also observed in natural human infections with P. falciparum [46–48]. If we extrapolate from these experiments and assume that mixed strain infections are, in general, more harmful than single clone infections, then a case may be made for the implementation of malaria control measures that aim to reduce parasite genetic diversity. Of course, most existing interventions such as drug treatment and anti-mosquito measures do exactly this; reduce parasite genetic diversity by reducing parasite prevalence rates. Extrapolation from rodent malaria parasites comes, however, with the usual caveats, and it should be mentioned that many field reports do not find any correlation between parasite genetic diversity (most commonly measured by the “multiplicity of infection” (MOI) index), and disease severity [49–54], and some find a negative correlation [55, 56]. It may also be the case that many field-based studies consider chronic malaria infections, whereas acute infections are considered in rodent malaria studies. There are also, of course, problems and difficulties with the interpretation of field-based surveys that do not apply to laboratory based experiments in which conditions can be carefully controlled and confounding factors minimised.
It has been argued, on the basis of numerous experiments performed exclusively with P. chabaudi, that within-host competition leads to the selection of virulent malaria parasites [7, 22, 57–60]. This theory relies on the assumption that the “virulent” parasite (typically, the one with the fastest growth rate) outcompetes the less virulent parasites in the mammalian host, and then, crucially, is more successful at transmitting to mosquitoes, and subsequently into another mammalian host, than the less virulent parasites. Our results do not support this assumption. Firstly, the most virulent strain in a mixed infection does not always out-compete the least virulent. In the case of a mixed infection between an intermediately virulent and an avirulent parasite, it was ultimately the avirulent parasite that was responsible for the competitive exclusion of the intermediate virulence parasite. We hypothesise that the strain that dominates the infection during the acute phase (the first 7 days of the infection), is subsequently targeted by a stronger strain-specific immune response than the competing strain, leading to the competitive release of the less virulent clone later on in the infection. Crucially, we found that P. y. yoelii infections are infectious to mosquitoes during the latter stages of such infections, following the competitive exclusion of the most virulent strain. In the case of a mixed infection with a highly virulent strain, the avirulent strain was competitively excluded by the fifth day of the infection, at which point the death of the host occurred, effectively restricting the transmission of both the virulent and avirulent parasites to the first five days of the infection.
Secondly, we found no correlation between the virulence of a parasite and its transmission ability in single infections, with the avirulent strain (CU) resulting in the highest proportion of mosquitoes infected, and the highest number of oocysts per infected mosquitoes, than any of the other strains. Based on the transmission success of the clones in single infections, a fitness co-infection was derived which reflects the transmissibility of the strains on days three and four post-infection when transmissibility is at its highest in P. y. yoelii. This revealed that CU (avirulent) had the highest fitness in single infections, followed by 17X1.1pp (intermediately virulent), 17XNL (avirulent), and finally, the highly virulent 17XL. Furthermore, these relative finesses were calculated for only days three and four of the infections, and, as we found that transmission to mosquitoes was successful during the chronic phase of infection on the day on which it was tested, it is likely that the true relative fitness of the virulent 17XL is much lower than our estimates, as it kills the host on the fifth day of infection.
Thirdly, there was no correlation between a strain’s virulence and the relative fitness cost of competition with another strain. For example, the relative fitness of the highly virulent strain 17XL in mixed infection with the avirulent strain CU was ~20% of its fitness in a single infection, whilst the relative fitness of the intermediately virulent 17X1.1pp in a mixed infection with CU was ~70% of its fitness in a single infection. Of all the strains, the avirulent 17XNL suffered the largest cost of competition with the avirulent CU strain, followed by the virulent 17XL, and the intermediately virulent 17X1.1pp. The avirulent strain CU was least affected by competition with the virulent XL, and most adversely affected by competition with the avirulent 17XNL.
From these results we can infer that virulence is linked neither to competitive ability nor ‘fitness’ as measured through the ability of strains to transmit through mosquitoes in this species of malaria parasite, directly contradicting previous studies with P. chabaudi [7, 22], calling into question the validity of extrapolating general principles of the importance of within-host competition as a driver of the evolution of virulence from one parasite species to another. Malaria parasite species differ hugely in many important phenotypes, some of which, such as the timing of gametocytogenesis, will affect the evolutionary repercussions of within-host competition. Considering, for example, the cases of P. falciparum and P. vivax, the two most prevalent of the malaria parasites that infect humans, it may be reasonable to postulate that the evolution of P. vivax strains might be less influenced by within-host competition than P. falciparum, due to the former species’ propensity for producing gametocytes early on in infections [61], before the influence of inter-strain within-host competition would manifest.
In summary, previous experiments with P. chabaudi have appeared to show that within-host competition would drive the evolution of virulence; our results with P. yoelii contradict this, and this discrepancy is probably best explained by phenotypic differences between the species with respect to the timing of gametocyte production. We urge caution, therefore, when extrapolating the results of experiments dependent on variable phenotypic traits with one species of malaria parasite to any other.
The fact that our experiments with P. yoelii yield contrasting results to those performed with P. chabaudi highlights the importance of parasite biology when considering informative models for the evolution of various traits, including virulence. Extrapolation from one species to another is problematic when parasite biology varies greatly between species. Furthermore, as host-parasite interactions are of crucial importance in these types of studies, it should be emphasized that the rodent malaria parasites are not, naturally, parasites of Mus musculus, but rather of Grammomys surdaster and Thamnomys rutilans (Reviewed in [62]), and that the typical pathological outcomes of malaria parasite infections in these natural hosts is very different from that observed in laboratory mice. This point is illustrated further by studies showing that the outcome of within-host competition can be significantly different depending on the laboratory mouse strain used [59].
In summary, mixed strain infections of P. yoelii were found to cause more severe disease in mice than single infections of the constituent strains. There was no apparent increase in the infectivity of mixed infections to mosquitoes, and mixed infections did not result in greater oocyst burdens per infected mosquito. Within-host competition generally led to a reduction in parasite fitness, the degree of which varied between strains. Importantly, we found no evidence that virulent strains were more competitive than less virulent strains, and conclude that, in the case of P. y. yoelii, within-host competition would not lead to the selection of virulent strains.
Laboratory animal experimentation was performed in strict accordance with the Japanese Humane Treatment and Management of Animals Law (Law No. 105 dated 19 October 1973 modified on 2 June 2006), and the Regulation on Animal Experimentation at Nagasaki University, Japan. The protocol was approved by the Institutional Animal Research Committee of Nagasaki University (permit: 1207261005–2).
We used four strains of rodent malaria parasite Plasmodium yoelii yoelii, three of which are phenotypically distinct lines that are isogenic except for polymorphisms at those loci that confer virulence [63]. These are Plasmodium yoelii yoelii 17XNL (wild-type, non-virulent) [64], P. y. yoelii 17XL (virulent) [64], and P. y. yoelii 17X1.1pp (intermediate virulence) [65], and a genetically unrelated strain P. y. yoelii CU, which is of wild-type, non-virulent phenotype [65]. Eight-week old female CBA mice (SLC Inc., Shizuoka, Japan) were housed at 26°C and fed on maintenance diet with 0.05% para-aminobenzoic acid (PABA)-supplemented water to assist with parasite growth. Anopheles stephensi mosquitoes, used in the transmission experiments, were housed in a temperature and humidity controlled insectary at 24°C and 70% humidity, adult flies being maintained on 10% glucose solution supplemented with 0.05% PABA.
To address the question of whether within-host competition leads to increased virulence, we infected groups of mice with either of the strains on their own or together with a competitor strain. Densities of each strain in mixed infection were monitored using strain-specific real-time quantitative PCR [59, 66], replication rates were measured by asexual parasitaemia and virulence was quantified through monitoring anaemia, live-weight loss [67] and host mortality.
Seven experimental groups of four mice each were set up to understand the effects of interactions between different parasite strains on the host and to compare the fitness of a strain in single versus mixed strain infections. Four of these groups were each singly-infected by i.v inoculation with CU, 17XNL, 17X1.1pp, or 17XL parasites (1 × 106 parasitised erythrocytes in 0.1mL). The remaining three groups each received a total of 2 × 106 parasites comprising a mixture of equal numbers of CU + 17XNL, CU + 17X1.1pp, and CU + P. yoelii 17XL. Inocula were prepared by taking blood from the tail vein of the donor mouse and diluting it in medium suitable for parasite maintenance (50% heat-inactivated foetal calf serum, 50% Ringer’s solution [27 mM KCl, 27 mM CaCI2, 0.15 M NaCI], with 20 units of heparin/ml mouse blood) to the appropriate concentration for the inoculum size. The requisite volume of blood was calculated from the blood cell density and parasitaemia in donor mice counted immediately before experimental sub-inoculations.
To accurately quantify the proportion of each strain used in the mixed strain infections, DNA was later extracted from a sample of each inoculum for real-time quantitative PCR (qPCR) analysis. Mouse red blood cell (RBC) densities and live-body weights were monitored as indicators of virulence. RBC densities were measured using a Coulter Counter (Beckman Coulter, Florida) from a 1:40,000 dilution of 2 μl sample of tail blood in Isoton (Beckman Coulter, Florida) solution. Parasite replication rate was assessed for 30 days by counting the proportion of RBCs infected by asexual parasites (parasitaemia) on Giemsa’s solution-stained thin blood smears from tail vein blood. Densities of gametocytes, the blood stage parasites that are transmissible to mosquitoes, were obtained by counting the number of RBCs containing mature gametocytes (distinguishable from asexual parasites by their morphology and presence of pigment as detected by polarized light) in the same thin blood smears used for counting asexual parasites. Asexual parasite density and gametocyte density were calculated from the product of RBC density and parasitaemia or gametocytaemia. Parasite densities in mixed strain infections were measured at specific time-points from day 1 to day 30 p.i using strain-specific qPCR from DNA prepared from 10-µl tail blood samples which were collected daily into physiological citrate saline solution, spun down and the pellet stored at −80°C prior to DNA extraction. The experiment was repeated once.
To assess whether the outcome of the within-host competition was a determinant of transmission success, we fed mosquitoes on mice infected with either single or mixed infections. Transmission was measured by density of sexual forms, gametocytes, in the blood, the proportion of mosquitoes infected after taking a blood-meal from the mouse, and the numbers of oocysts present on infected mosquito midguts. The number of oocysts produced and the proportions of mosquitoes that were infected were also used as indicators of vector fitness.
Twenty female Anopeheles stephensi mosquitoes (seven- to eleven-days post-emergence) were allowed to take blood meal from anaesthetised and immobilised mice in both single and mixed infections groups on day 3 p.i and the same was repeated with a fresh group of 20 mosquitoes day 4 p.i. Groups of mosquitoes that had fed on the same mouse were housed in individual pots. Seven to eight days post-feed, mosquitoes were immobilised and their midguts dissected to determine the number of oocysts and the percentage of oocyst-infected mosquitoes. To quantify the proportion of each strain in mixed-strain infections, mosquito midguts from the mixed-infection groups were suspended in PBS, spun down and the pellet stored at −80°C before DNA extraction and subsequent qPCR.
The proportion of co-infecting strains in mixed infections was determined by using qPCR measurement of the copy number of parasite’s MSP-1 gene. The MSP-1 gene, located on chromosome 8 [68] contains regions of high sequence polymorphism between clones that facilitate the design of allele-specific primers that can act as clone-specific genetic markers. DNA was extracted from infected mouse blood and infected mosquito midguts using EZ1 DNA investigator kit (Qiagen) according to manufacturer’s instruction. The extracted DNA was used for qPCR using the Power SYBR Green PCR kit (Applied Biosystems, UK) on a 7500 Real Time PCR system (Applied Biosystems, UK). Copy numbers of parasite msp-1 were quantified with reference to a standard curve generated from known numbers of plasmids containing the same gene sequence. Plasmodium yoelii CU msp1 and P. yoelii 17X msp1 were amplified as previously described [69]. Description of the use of quantitative microsatellite markers to measure the proportions of parasites carrying markers linked to the putative genetic driver of virulence in mice and mosquitoes is given in S1 Text.
Fitness coefficients were determined for the four strains based on the numbers of oocysts produced per mosquito that fed on mice infected with each of the strains on days 3 and 4 post-inoculation. The mean number of oocysts observed on the mid-guts of mosquitoes fed on mice with infections of the various strains averaged between days 3 and 4 were taken as an infectivity index. These were then standardized against the strain with the highest infectivity (CU), so that CU had a fitness coefficient of 1. In mixed infections, the proportion of each strain was determined by qPCR, and the “adjusted number of oocysts” calculated for each strain (number of oocysts multiplied by the strain frequency):
Fitness coefficient = mean number of oocysts per mosquito fed on strain X (day 3 + 4) / mean number of oocysts per mosquito fed on strain CU (day 3 + 4)
All graphs were generated using GraphPad Prism (GraphPad Software Inc, USA). All statistical analyses were performed using R [36]. Detailed explanations of the statistical treatments used for each analysis are given in the relevant results section. All experiments were subject to full independent repeats with the exception of the experiment in which infectious mosquitoes were allowed to feed on naïve mice in order to measure whether parasite strain proportions present in mosquito oocysts were indicative of the proportions observed in mice following transmission, which were performed once.
|
10.1371/journal.pcbi.1000724 | Quantifying Aggregated Uncertainty in Plasmodium falciparum Malaria Prevalence and Populations at Risk via Efficient Space-Time Geostatistical Joint Simulation | Risk maps estimating the spatial distribution of infectious diseases are required to guide public health policy from local to global scales. The advent of model-based geostatistics (MBG) has allowed these maps to be generated in a formal statistical framework, providing robust metrics of map uncertainty that enhances their utility for decision-makers. In many settings, decision-makers require spatially aggregated measures over large regions such as the mean prevalence within a country or administrative region, or national populations living under different levels of risk. Existing MBG mapping approaches provide suitable metrics of local uncertainty—the fidelity of predictions at each mapped pixel—but have not been adapted for measuring uncertainty over large areas, due largely to a series of fundamental computational constraints. Here the authors present a new efficient approximating algorithm that can generate for the first time the necessary joint simulation of prevalence values across the very large prediction spaces needed for global scale mapping. This new approach is implemented in conjunction with an established model for P. falciparum allowing robust estimates of mean prevalence at any specified level of spatial aggregation. The model is used to provide estimates of national populations at risk under three policy-relevant prevalence thresholds, along with accompanying model-based measures of uncertainty. By overcoming previously unchallenged computational barriers, this study illustrates how MBG approaches, already at the forefront of infectious disease mapping, can be extended to provide large-scale aggregate measures appropriate for decision-makers.
| Reliable disease maps can support rational decision making. These maps are often made by interpolation: taking disease data from field studies and predicting values for the gaps between the data to make a complete map. Such maps always contain uncertainty, however, and measuring this uncertainty is vital so that the reliability of risk maps can be determined. A modern approach called model-based geostatistics (MBG) has led to increasingly sophisticated ways of mapping disease and measuring spatial uncertainty. Many health management decisions are made for administrative areas (e.g., districts, provinces, countries) and disease maps can support these decisions by averaging their values over the regions of interest. Carrying out this aggregation in conjunction with MBG techniques has not previously been possible for very large maps, however, due largely to the computational constraints involved. This study has addressed this problem by developing a new algorithm and allows aggregation of a global MBG disease map over very large areas. It is used to estimate Plasmodium falciparum malaria prevalence and corresponding populations at risk worldwide, aggregated across regions of different sizes. These estimates are a cornerstone for disease burden estimation and are provided in full to facilitate that process.
| Risk maps estimating the spatial distribution of infectious diseases in relation to underlying populations are required to support public health decision-making at local to global scales [1]–[3]. The advancement of theory, increasing availability of computation and growing recognition of the importance of robust handling of uncertainty have all contributed to the emergence in recent years of a new paradigm in the mapping of disease: the use of a special family of generalised linear models known as model-based geostatistics (MBG), generally implemented in a Bayesian framework [4],[5].
MBG models take point observations of disease prevalence from dispersed survey locations and generate continuous maps by interpolating prevalence at unsampled locations across raster grid surfaces. The most striking advantage of MBG in disease mapping is its handling of uncertainty. Interpolating sparse, often imperfectly sampled, survey data to predict disease prevalence across wide regions results in inherently uncertain risk maps, with the level of uncertainty varying spatially as a function of the density, quality, and sample size of available survey data, and moderated by the underlying spatial variability of the disease in question. MBG approaches allow these sources of uncertainty to be propagated to the final mapped output, predicting a probability distribution (known formally as a posterior predictive distribution) for the prevalence at each location of interest. Where predictions are made with small uncertainty, these distributions will be tightly concentrated around a central value; where uncertainty is large they will be more dispersed. These techniques have been used to generate robust and informative risk maps for malaria [6]–[12], as well as a range of other infectious diseases [13]–[19], at scales varying from national to global. Some studies have extended the handling of variation through space to also include the temporal dimension, allowing disease risk to be modelled and quantified over time as well as space [6],[20].
Implementation of MBG models over even relatively small areas is extremely computationally expensive. Not only are the matrix algebra operations required to generate predictions at each individual pixel costly compared to simpler interpolation methods [21],[22], but this cost must be multiplied many times because prediction uncertainty is evaluated by generating many, equally probable, “realisations” of prevalence at each pixel. Implementations of MBG disease models over large areas therefore tend to be via “per-pixel” computation whereby complete maps are built up by generating predictive realisations for each pixel independently. This allows the computational task to be broken down into many small, more easily manageable, operations. Such an approach yields appropriate measures of “local” uncertainty: the set of realisations for each pixel represents a posterior predictive distribution of prevalence from which summary statistics such as the mean, inter-quartile range or 95% credible intervals can be readily extracted, providing the user with valid uncertainty information for each individual location considered in isolation.
There is often a need to evaluate disease prevalence aggregated across spatial regions, temporal periods, or combinations of both [23],[24]. This may be to quantify and compare mean prevalence between countries or administrative units, for example, or to measure a shift in mean prevalence between the start and end of an intervention period or policy change. In addition, MBG prevalence models can be used to estimate derived quantities such as population totals living in regions at different levels of risk, or the burden of disease cases expected within individual countries or continents as a function of underlying prevalence [25], quantities that by definition exist only over aggregated space-time units. It is not possible, however, to construct posterior distributions for these aggregate quantities using a per-pixel approach. To estimate the mean of a region made up of multiple pixels, and the uncertainty around this estimate, the correlation between all the pixels in the region must be known. In a per-pixel approach, each pixel is modelled as independent of its neighbours, ignoring any spatial or temporal correlation. Failing to account for correlation between pixels leads to gross underestimates of the uncertainty in the aggregated quantity, especially over large regions [26].
The solution to the problem outlined above is to replace per-pixel simulation of prevalence realisations with the simultaneous or ‘joint’ simulation of all pixels to be aggregated, recreating appropriate spatial and temporal correlation between them [26]. Crucially, the set of pixel values can then be aggregated in any way, or used as input in derived aggregated quantities, and realisations of these aggregations will have the appropriate posterior predictive distributions. Whilst conceptually simple, the extension from local to regional simulation induces a fundamental computational constraint in that the necessary calculations can no longer be disaggregated into separate tasks for each pixel. This constraint has thus far prevented any use of MBG in disease mapping for the evaluation of aggregate quantities over very large areas, despite the profound public health importance of such measures. Where examples of joint simulation in MBG disease mapping exist, they tend either to be over very small spatial regions [8] or are achieved by simply breaking larger regions down manually into smaller more manageable tiles [17].
In this paper we use a new approximate algorithm for joint simulation to quantify, for the first time, aggregated uncertainty over space and time in a global scale MBG disease model for Plasmodium falciparum malaria prevalence [6]. We exemplify how this approach allows uncertainty in prevalence predictions to be enumerated at the continental, national, and sub-national scales at which public-health decisions are usually made. We then extend the model architecture to estimate a second quantity of particular epidemiological interest: national populations at risk (PAR) under different policy-relevant strata of P. falciparum transmission intensity.
PAR estimates form a fundamental metric for malaria decision-makers at national and international levels [24],[27] and have also been used to assess equity in donor funding distributions [28], chart the changing exposure of human populations to the disease [29] and provide baselines for predicted changes in exposure under climate change scenarios [30]. A range of techniques have been used to estimate PAR, including the use of MBG and other prevalence models to delineate risk strata in relation to underlying population distributions [6], [29], [31]–[33]. None of these studies have incorporated the inherent uncertainty in prevalence estimates, however, and the resulting PAR estimates are presented as point values with no uncertainty metrics. Here we use the joint simulation framework to generate posterior predictive distributions of PAR living under conditions of low, medium, and high stable transmission within each malaria endemic country, allowing the uncertainty inherent in these estimates to be quantified in a formal statistical framework. These PAR estimates are presented in full with this paper, making them available to any interested parties to support theoretical and applied epidemiological and public health applications.
In the remainder of this introductory section we outline the computational challenges of large scale joint simulation and review existing approaches to overcoming them. In the methods section we present our algorithm for efficient joint simulation over very large grids, detail its implementation and testing with the global P. falciparum model, and its extension to estimating PAR. The results section provides the outcome of the testing and validation procedures and examples of jointly simulated realisations of continental, national, and locally aggregated estimates of P. falciparum prevalence in 2007. We present our national level estimates of PAR and exemplify how the accompanying uncertainty metrics can be communicated effectively to enhance their utility to decision-makers. We conclude by discussing the strengths and weaknesses of our modelling architecture, the implications for the future of disease mapping, and useful directions for further research.
A general form for MBG models can be defined as follows:(1)such that in a disease survey of individuals at a given location, the number observed to be infected is modelled as binomially distributed with probability of infections given by , the underlying prevalence of the disease in question, which is modelled as a transformation via an inverse link-function of an unknown Gaussian process (GP) [34],[35]. A Gaussian process in the context of disease mapping is a convenient probability distribution for 2-d surfaces (or 3-d cubes if considering time), describing probabilities associated with different forms of the surface (or cube). Using Bayesian inference, the Gaussian process can be updated to take account of the input data, providing a refined description of these probabilities. Possible surfaces can then be drawn from this updated Gaussian process which, after passing through the inverse link-function, provide realisations of the target disease surface. The Gaussian process can take a wide range of forms: the central tendency at any location is governed by the underlying mean function , whilst textural properties (the roughness of the surface, and its tendency to fluctuate across space) are governed by the covariance function . The symbol denotes a set of parameters that define the form of the covariance and mean, which can include covariate coefficients.
In MBG, the aim is to estimate the joint posterior distribution of the model parameters and the values of evaluated at all locations and times for which a prediction is required - generally across the nodes of a regular raster grid. Computationally, this task can be split into two distinct phases. Firstly, Markov chain Monte Carlo (MCMC) can be used to generate realisations from the joint posterior of and at only the space-time locations where data exist, denoted . This is intuitive because it is only at these locations that the fit of the Gaussian process is evaluated, and this means the MCMC must only consider a multivariate normal distribution of dimension , which is generally several orders of magnitude smaller than if all prediction locations across the raster grid were considered. A realisation of and provides a ‘skeleton’ from which the Gaussian process can be evaluated at all prediction locations across a raster grid in a second computational stage. Conditional on these ‘skeleton’ realisations, the value of at each prediction location and time can be sampled from its posterior predictive distribution:(2)where the posterior predictive mean and covariance parameters are given by the standard conditioning formulas for multivariate normal variables [[36] (p.367)]:(3)(4)By carrying out this two-step procedure over many realisations, samples are built up from the target posterior predictive distribution .
In a per-pixel implementation, the predictive distributions , , at all prediction locations in the output raster are realised independently to generate local models of uncertainty. In this case, the largest single computational component is the population and factorisation (via a procedure known as the Cholesky decomposition [37]) of the data-to-data covariance matrix which, in typical disease prevalence data sets where is in the hundreds or thousands, is a relatively minor task that could generally be achieved on a standard desktop computer. The subsequent sampling from the posterior predictive distribution (as in Eq. 2) is trivial: the posterior predictive mean and covariance refer to a single prediction location and sampling therefore amounts to drawing from a univariate normal distribution. Total computation for each pixel is therefore modest, and the cost of generating the maps grows simply in proportion to the number of pixels involved, .
Switching from a per-pixel implementation to a joint simulation over many prediction locations increases profoundly the computational challenge. The efficiency of a per-pixel approach arises from the effective reduction of to one, as each pixel is considered in isolation. Joint simulation requires that is preserved as the total number of prediction points, which can be many millions if large areas are considered at reasonably fine spatial resolution. In addition to the × data-to-data covariance matrix, the × prediction-to-prediction and × data-to-prediction covariance matrices must be populated. More importantly, in the subsequent sampling from the posterior predictive multivariate normal distribution, the prediction-to-prediction covariance matrix must be factorised [37]. The computational cost of this operation is proportional to the cube of . To put this non-linear scaling in context, if a direct joint simulation of a 100×100 raster grid could be computed in one minute, a 1000×1000 grid would take approximately 6×107 seconds (around 694 days). In practice these scaling factors along with those of memory and storage requirements mean direct joint simulation using the equations outlined above is generally limited to predictions at a maximum of around 10,000 points [17],[38], at least two orders of magnitude too few for global scale mapping at sub-10 km resolution, even at a single time period.
In response to the strict computational limits of direct joint simulation outlined above, a wide range of algorithmic and mathematical tools have been developed that increase substantially the maximum number of prediction locations that can be feasibly handled.
The most widely used family of joint simulation algorithms in geostatistics is known as sequential simulation [39]–[41]. Instead of simulating the joint distribution across all prediction locations simultaneously, sequential simulation evaluates each prediction location in turn. The properties of the multivariate normal model are preserved by conditioning each prediction location not only on the input data, but also on the values simulated at previously evaluated prediction locations, which are effectively treated as conditioning data in subsequent simulations. This approach means the data-to-data covariance matrix gains an additional row and column after each simulation, ultimately approaching elements, which becomes prohibitively large as approaches around 10,000 points. In response, sequential simulation algorithms generally limit the conditioning data to a small neighbourhood of points around each prediction location, specified either by number or by spatial proximity. This computational shortcut is justified by the declining influence of more distant data, which means a simulation conditioned on data approximates asymptotically one conditioned on data [39]. Whilst allowing potentially very large prediction grids to be evaluated, the restriction of conditioning data to local neighbourhoods necessitates that, for each prediction location, these data are identified via a search algorithm [40],[42], and bespoke linear algebra systems are evaluated and solved (Eqs. 3 and 4). The cost of the latter for each pixel is proportional to , meaning that as the number of prediction locations grows, the size of that can be feasibly computed reduces sharply.
In a disease mapping context, the goal is to generate joint simulations conditioned by observed prevalence or incidence data. This precludes the direct use of a wider class of algorithms developed for unconditional joint simulation [5], where the goal is simply to realise random fields with correct mean and covariance properties, unconstrained by any observations. Conditional simulation can, however, be split into an unconditional joint simulation and a per-pixel prediction task [[42] (p.494),[43]], as follows:(5)where is the target conditioned field. In practical terms, this decomposition allows generation of the conditioned field in two stages: unconditional simulation is used to generate the unconditioned field (the first term above, ), which is then combined with the ‘skeleton’ of the conditioned field at the data points in a standard per-pixel prediction (the second right-hand term). The sum of these terms yields a conditioned field with an identical distribution to one generated directly via conditional simulation. The advantage in working with unconditional simulation algorithms arises because, in the absence of irregularly located data, all computations relate to locations arranged in a regular grid, a geometric convenience that can be exploited in a variety of ways [44]–[47]. An elegant and widely used family of techniques for grid-based unconditional simulation is based on spectral decompositions, principally the fast Fourier transform, of which the ‘circulant embedding’ algorithm is particularly popular [48]–[50]. These techniques offer extremely efficient exact simulations but become infeasible for more than around one million prediction locations [38], due in part to memory requirements resulting from the necessary replication of large covariance matrices. Furthermore, such algorithms require the prediction grid to be regular, that is, for pixels to be arranged in rows and columns of equal spacing, which is not the case with global grids defined using spherical coordinates.
A new global map of P. falciparum endemicity in 2007 has recently been published [6], the first such enumeration of global malaria risk in 40 years. This map was generated from an assembly of 7,953 community parasite surveys collated from 78 countries between 1985 and 2007 used with a Bayesian space-time MBG model to predict urban-adjusted P. falciparum parasite rate in the epidemiologically informative 2 up to 10 yr age range, PfPR2–10, across a regular spherical grid within the limits of stable transmission [31]. The model form is described in full elsewhere [6]. The original implementation of this model used an MCMC inference stage to generate 500 samples from the joint posterior distribution of the space-time Gaussian process at the 7,953 locations for which input parasite rate survey data existed , and of a 13 element parameter vector, . A per-pixel approach was then used to evaluate, for each realisation, values of the Gaussian process at all desired prediction locations , which were then combined with an independently sampled Gaussian random noise component, and subjected to an inverse logit transform and multiplication with an age-correction factor to yield the target quantity PfPR2–10. The set of realisations of PfPR2–10 for each pixel provided an appropriate measure of local uncertainty, with which the precision of PfPR2–10 predictions could be assessed at all individual pixel locations worldwide.
The aim of the current study was to implement the predictions described above via joint simulation, allowing quantification of uncertainty in predicted PfPR2–10 over spatially and temporally aggregated regions. This presented an unprecedented challenge in geostatistical disease modelling for a number of reasons. Firstly, the target prediction space was exceptionally large: a grid of resolution equivalent to 5×5 km at the equator spanning the extent of stable P. falciparum transmission in Africa (the largest contiguous region of interest), evaluated temporally for each of the 276 months of 1985–2007 (the study period of interest, corresponding to the temporal span of the collated PfPR survey assembly) constituted approximately 623 million individual prediction locations, several orders of magnitude larger than any other MBG disease model extent in the published literature. Secondly, the model had a relatively complex form, particularly in the covariance function [51] which was spatiotemporal (covariances were modelled between locations spread across time as well as space), spatially anisotropic (covariance between spatial locations was influenced by direction as well as separation distance), and included a periodic component in the temporal axis (to address observed seasonality). Finally, spatial locations of data and predictions were represented on a sphere, with their separations evaluated using great-circle distance, a geometric complication that was necessary to avoid the distorting effects of map projections when dealing with global scale phenomena. Together, these factors precluded the use of the existing approaches to joint simulation described above. Spectral decomposition-based algorithms for unconditional simulation would have required the data-to-data covariance matrix to be reflected along three axes, exceeding memory limits of currently available computers. More fundamentally, the incorporation of the curvature of the earth in the arrangement of prediction locations meant the matrix could not be considered to be in block Toeplitz form [38],[44],[48]. Whilst a standard sequential simulation could, in principle, have been achieved within available memory constraints, the very large number of prediction locations would have meant limiting conditioning data to insufficiently small prediction neighbourhoods in space and time in order to achieve computation in a feasible timescale. Instead, a novel approximate algorithm for joint simulation was developed that overcame these constraints, and this is presented in the next section.
Sequential simulation algorithms maintain feasible memory and computation requirements over large grids by limiting conditioning data to small local neighbourhoods, but the repeated identification of local data, evaluation of local covariance matrices, and subsequent linear algebra calculations are prohibitively inefficient for very large numbers of prediction locations. The extremely efficient algorithms developed for unconditional joint simulation over regular grids, such as circulant embedding, also reach memory limits for very large prediction tasks, and are not suited to sphere-based grids. In this study a new algorithm was developed that adopted and extended the principle of traditional sequential simulation - that joint simulation over very large areas can be broken down into many small simulations conditioned on nearby values - but incorporated some of the efficiencies exploited by unconditional algorithms operating on a grid whilst overcoming the complications of sphere-based grid systems.
Firstly, the decomposition of a conditional joint simulation into an unconditional joint simulation and a per-pixel conditioning stage was exploited (Eq. 5). The bulk of the computational challenge therefore lay in generating unconditioned realisations of the zero-mean Gaussian process across the nodes of the 3-d space-time prediction grid given only realisations of the scalar parameter vector , where = 6.23×108 was the largest individual prediction task (for the Africa region, with 1718 columns, 1315 rows and 276 months). Each grid pixel was 0.04165 decimal degrees in height and width, corresponding to approximately 5×5 km at the equator. Defining a regular grid in terms of spherical coordinates meant that the width of pixels varied with latitude. A second stage was then required to condition the field given realisations of the field at the data locations These two stages are now discussed in more detail.
Having implemented the algorithm described above, the jointly-simulated conditioned field was combined with per-pixel samples from an uncorrelated Gaussian noise component, subject to an inverse-logit transform, and multiplied by an age correction factor to yield realisations of PfPR2–10. Crucially, in contrast to the original per-pixel implementation [6], each realisation represented a joint simulation of prevalence, so the back-transformed and age-corrected space-time cube could be aggregated into any arbitrary spatial, temporal, or space-time unit, with realisations of the aggregated quantity representing samples from the posterior predictive distribution. This was exemplified by generating realisations of mean PfPR2–10 across the 12 months of 2007 for three scales of spatial aggregation: continental, national, and at the first sub-national administrative unit level, quantities that span the spectrum of information scales required by malaria public-health decision-makers.
Previous approaches to estimating PAR have used modelled surfaces of P. falciparum prevalence to delineate the boundaries of various risk strata, and combine these mapped boundaries with population maps to calculate the population living in each strata [6]. Because the prevalence modelling in this earlier work was set in a per-pixel framework, spatial uncertainty in the prevalence predictions could not be propagated into the PAR estimates since the latter is a spatially aggregated quantity. This limitation was removed in the current study since prevalence was modelled using a joint simulation framework. Population data [54] were obtained and adjusted to form a 1×1 km grid surface for 2007, and a previously defined stratification [31] was used to delineate areas in which stable transmission of P. falciparum malaria was likely to occur (defined as areas where incidence is likely to exceed 0.1 case per 1000 per annum). These inputs are explained further in Protocol S1. Within these limits of stable transmission, each jointly simulated realization of PfPR2–10 was converted into a categorical map identifying pixels where prevalence was predicted as either low stable (PfPR2–10≤5%), medium stable (PfPR2–10>5%≤40%) or high stable (PfPR2–10>40%) transmission. These prevalence classes have been proposed previously as of particular relevance to decision-makers when developing optimal strategies for intervention and control [55],[56]. Each realized endemicity class map was downscaled to a 1×1 km grid and combined with the population grid and an additional grid identifying national boundaries to allow calculation of a realization of PAR in each of the three endemicity classes in each country. Repeating this procedure across all 500 realizations allowed posterior predictive distributions to be constructed, from which the posterior mean was extracted as a point estimate for each country-class and the inter-quartile range was extracted as an accompanying uncertainty metric.
Figure 3 provides examples of five of the 500 jointly simulated realisations of PfPR2–10 within the global limits of stable transmission, aggregated temporally across the 12 months of 2007. None of these maps, taken individually, are intended to represent the true pattern of global prevalence. Each is driven by the underlying data but represents a random draw from a universe of possible maps given the model specification, the information in the data, and the resultant modelled uncertainty. Whilst the large-scale regional patterns of endemicity are similar in each realisation, small scale heterogeneity exists between each, and this variation across the 500 realisations defined the form of the posterior predictive distribution of the global surface. The validation procedures explained above provided evidence of the suitability of these surfaces to be aggregated spatially or temporally to provide appropriate posterior predictive distributions of mean PfPR2–10 within spatiotemporal units of different sizes. Figure 4 provides examples of this functionality: posterior predictive distributions are shown for mean PfPR2–10 across the entire African continent, across three individual countries (Ghana, Democratic Republic of Congo (DRC) and Kenya), and across a first level administrative unit in each of these countries (Ashanti Region, Ghana; Kinshasa Province, DRC; Nyanza Province, Kenya). As would be expected, the dispersion of the distributions, which can be interpreted directly as the modelled uncertainty in the predicted mean PfPR2–10, tended to decrease as aggregated predictions were made over progressively larger regions, such that the continent-wide mean was predicted with lower uncertainty than were national-level means which, in turn, were less uncertain than first-level administrative unit means. Dispersion was also moderated, however, by predictive uncertainty influenced by the availability of input survey data in different regions. This explains why the posterior predictive distribution for DRC, a country with very few available survey data, is substantially more dispersed than that for Kenya, for which many survey points exist, despite constituting a much larger spatial unit of aggregation. These example plots also illustrate in general terms why joint-simulation is necessary when predicting aggregated prevalence. Under a standard per-pixel implementation with all locations simulated independently, the variance of the aggregated mean PfPR2–10 would decline in proportion to the square-root of the number of pixels in the aggregated unit. At even the first administrative unit level, this would result in artificially small variances for the posterior predictive distributions. At national and continental levels the predicted uncertainty would effectively be zero. Under the joint simulation approach presented here, the space-time variance structure is preserved and this resulted in even the continent-wide prediction retaining a non-negligible level of uncertainty.
Estimated 2007 populations living under low, medium, and high stable transmission risk are presented by country in Protocol S1 along with accompanying posterior inter-quartile ranges. Figure 5(A) provides an example of mapped national PAR estimates for the high stable transmission risk class (PfPR2–10>40%). Figure 5(B) shows a ranking of each country in terms of model-based uncertainty, quantified by comparing the width of the posterior inter quartile range associated with each PAR estimate. Figure 5(C) shows an equivalent uncertainty ranking for relative PAR (percentages of each country's population living under high stable conditions). Further details on the methods and interpretation of the presented results can be found in Protocol S1.
Numerous algorithms exist that seek to increase the efficiency of joint simulation, including the widely-used family of sequential simulation algorithms, and those based on spectral decompositions operating on regular lattices. Whilst these elegant algorithms expand considerably the magnitude of joint simulation tasks that can be achieved relative to a direct calculation, none of them could produce simulations on arbitrary input grids on the scale required for global-scale disease maps as addressed in this study. We have overcome these limitations using a practical elaboration of standard sequential simulation that is empirically highly efficient but does not put any special requirements on the input grid or covariance function. The approach represents an important increase in the feasibility of aggregated uncertainty assessment over very large prediction spaces, expanding the scope of geostatistical models in global scale epidemiology.
The current study builds on a modelling framework for the global mapping of P. falciparum prevalence defined in an earlier study [6]. Like many large-scale MBG disease mapping studies published to date, this earlier work presented prevalence maps with per-pixel uncertainty metrics that could not be used to define uncertainty around aggregated prevalence predictions. Similarly, this per-pixel approach did not support the evaluation of uncertainty around important derived aggregate quantities such as populations at risk. By setting this earlier model in a joint simulation framework, the current study allows the formal prediction of aggregated P. falciparum prevalence and national populations under different prevalence strata with appropriate measures of uncertainty. Figure 5 shows these national PAR estimates for populations living under conditions of high stable transmission. A prevalence threshold of PfPR2–10>40% has been proposed as separating lower transmission settings, where universal coverage of insecticide treated bed-nets could interrupt transmission [56], from higher transmission settings where this coverage alone would be insufficient and scale-up of additional interventions would be required to achieve elimination [55],[56]. Under these recommendations, the quantification of PAR in this high stable transmission class has direct implications for national resource requirements as very large numbers of people in these nations will require more than universal coverage of bed-nets to interrupt transmission.
The direct practical utility to decision-makers of accompanying uncertainty metrics is less well established since they have not been available previously. Uptake by decision-makers will be aided by the packaging of uncertainty measures into easily understood information and the ranked uncertainty maps presented in Figure 5 (B and C) highlight one such approach. Recognising those countries where high-risk populations can be identified with least certainty provides a basis for rational deployment of global surveying, monitoring, and evaluation efforts for populations that will carry the largest burden of global malaria morbidity. Figure 5(B) illustrates that the least certain countries are, as would be expected, those which include areas of high transmission risk in conjunction with very large human populations, such as India, Myanmar, DRC and Nigeria. Figure 5(C) considers population proportions at risk and therefore standardises for absolute population size. This removes, amongst others, India, Nigeria and DRC from the set of least certain countries and adds some smaller high risk nations such as Togo, Liberia and Sierra Leone.
In this study we have presented the extension of a jointly simulated prediction framework for P. falciparum prevalence to estimates of PAR. The framework could readily be extended to the prediction of related aggregate quantities of substantial public health significance. An important example is the prediction of P. falciparum clinical case incidence which can be estimated empirically as a function of prevalence [25] and the underlying population density [54]. By basing these estimates on the jointly simulated prevalence predictions presented here, incidence estimates can be summed to provide national or continental-level estimates with appropriate credible intervals. A second important measure is the basic reproductive number, R0, which provides biological insight into the intensity of malaria transmission and is particularly useful when assessing the effect of current or future interventions [57]–[59]. Again, R0 can be estimated as a function of prevalence [59],[60] and the jointly simulated surfaces presented here can be incorporated with these models to provide country level estimates of R0 useful for strategic planning.
The approach presented in this study can be applied readily to any large-scale MBG prediction of infectious disease prevalence and corresponding populations at risk. An important caveat for further applications, however, is that the algorithm cannot be treated as a black-box that will generate appropriate output without user supervision. The algorithm relies on a key assumption: that the use of a relatively small proportion of conditioning data proximate to each target prediction column generates predicted values that are sufficiently similar to a theoretical (although infeasible) direct joint simulation based on all locations simultaneously. In reality, the footprint-based predictions will approach the theoretical “true” values asymptotically, such that the use of progressively more conditioning data in the footprint will result in progressively smaller increases in convergence between the two sets of values. This leads to a delicate trade-off between feasible computational demand and appropriate predictive precision. Suitable resolutions of this trade-off cannot be prescribed a priori since they rely on factors that will vary between settings. On one hand, the sparsest permissible footprint will be determined by the space-time covariance structure of the disease measure under study. On the other, the computational demand will scale non-linearly with the size of the study area and the spatial and temporal resolution at which it is to be modelled. In this study, appropriate footprint configurations were identified by systematically evaluating the empirical covariance functions of realisations generated under progressively sparser footprint configurations, and the necessary diagnostic scripts are freely available from the authors. In principle, appropriate configurations could be approximated in advance using the target covariance structure parameters, although the implications of the latter on the appropriateness of different configurations is likely to be complex and non-linear. Users of the algorithm should recognise that failure to consider these factors appropriately could lead to misleading or erroneous results.
Automatic optimization of the footprint would be a useful area for future research. Approaches to this problem, based on evaluation of the Markov properties of 2-d fields, have been proposed [61]–[63] although as yet these have not been extended to the 3-d setting necessary for space-time simulation. A number of recent advances in computational infrastructure have emerged that also warrant further investigation in the context of efficient joint simulation over very large grids. In particular the re-purposing of graphic processing units (GPUs) to support extremely efficient parallel processing, and their application to matrix calculations, offers potential decreases in processing time of several orders of magnitude [64] for the current rate-limiting steps of the algorithms presented here: the population and factorisation of large covariance matrices.
The expansion of MBG in epidemiology has been rapid and led to major advances in the handling of uncertainty in disease risk maps. To date, fundamental computational constraints have precluded the use of such models for predictions of aggregated prevalence and populations at risk required by decision-makers across national and continental spatial scales. In this study we have designed, implemented and tested a new algorithm that overcomes the prohibitive computational barriers of large scale joint simulation allowing, for the first time, appropriate handling of aggregated uncertainty in global scale disease maps. This epidemiological insight has been extended to defining national populations at risk with appropriate confidence intervals which are released here in the public domain to support informed efforts in disease burden estimation.
|
10.1371/journal.pntd.0001996 | Cell Death and Serum Markers of Collagen Metabolism during Cardiac Remodeling in Cavia porcellus Experimentally Infected with Trypanosoma cruzi | We studied cell death by apoptosis and necrosis in cardiac remodeling produced by Trypanosoma cruzi infection. In addition, we evaluated collagen I, III, IV (CI, CIII and CIV) deposition in cardiac tissue, and their relationship with serum levels of procollagen type I carboxy-terminal propeptide (PICP) and procollagen type III amino-terminal propeptide (PIIINP). Eight infected and two uninfected guinea pigs were necropsied at seven time points up to one year post-infection. Cell death by necrosis and apoptosis was determined by histopathological observation and terminal deoxynucleotidyl transferase dUTP nick end labeling, respectively. Deposition of cardiac collagen types was determined by immunohistochemistry and serum levels of PICP, PIIINP, and anti-T. cruzi IgG1 and IgG2 by ELISA. IgG2 (Th1 response) predominated throughout the course of infection; IgG1 (Th2 response) was detected during the chronic phase. Cardiac cell death by necrosis predominated over apoptosis during the acute phase; during the chronic phase, both apoptosis and necrosis were observed in cardiac cells. Apoptosis was also observed in lymphocytes, endothelial cells and epicardial adipose tissue, especially in the chronic phase. Cardiac levels of CI, CIII, CIV increased progressively, but the highest levels were seen in the chronic phase and were primarily due to increase in CIII and CIV. High serum levels of PICP and PIIINP were observed throughout the infection, and increased levels of both biomarkers were associated with cardiac fibrosis (p = 0.002 and p = 0.038, respectively). These results confirm the role of apoptosis in cell loss mainly during the chronic phase and the utility of PICP and PIIINP as biomarkers of fibrosis in cardiac remodeling during T. cruzi infection.
| Chronic Chagas heart disease (CHHD) caused by the infection with the parasite Trypanosoma cruzi is the most important infectious heart disease in the world. The typical manifestations are dilated cardiomyopathy and congestive heart failure; they result from death of cardiomyocytes and their replacement by collagen. Knowing the mechanisms of cardiomyocyte death is important for the development of therapies that prevent them. The contribution of apoptosis in cardiomyocyte death was evaluated in the guinea pig model of T. cruzi infection, and the detection of serum levels of collagen precursors were evaluated as biomarkers of cardiac fibrosis. We observed apoptosis of lymphocytes, cardiomyocytes, endothelial cells and epicardial adipose tissue in cardiac tissue of infected guinea pigs. The increase of serum levels of collagen precursors PICP and PIIINP were associated with cardiac fibrosis. Areas of inflammation and apoptosis of epicardial adipose tissue were associated with cardiac pathology, which suggests the importance of epicardial adipose tissue in CCHD. These results show that apoptosis is an important characteristic of cardiac cell death during CCHD and serum levels of PICP and PIIINP could be used as biomarkers of cardiac fibrosis.
| Chagas disease, a parasitic infection caused by Trypanosoma cruzi, remains a major public health problem in Central and South America with 8 to 10 million people infected [1]. Chronic Chagas heart disease (CCHD), the major clinical consequence, is the most important infectious heart disease in the world, with an estimated 50,000 attributable deaths per year [2].
Many factors have been implicated in the pathogenesis of CCHD, including parasite persistence, cardiac denervation, inflammation, autoimmunity, and microcirculatory changes [3]–[8]. The acute phase of Chagas disease is characterized by the presence of focal necrosis, severe inflammation and abundant amastigote nests in the heart [5]. In the early chronic phase, most infected individuals are asymptomatic, but over a period of years to decades, 30% of patients develop CCHD [3]. The earliest signs of CCHD are usually conduction system abnormalities such as right bundle branch block, with or without multifocal ventricular extrasystoles. Over time, higher-grade conduction deficits and ventricular arrhythmias may occur. The late stage of the disease typically features dilated cardiomyopathy and congestive heart failure. Sudden death may result from arrhythmias, heart block or emboli; patients with advanced cardiomyopathy may die of intractable heart failure [4]. These clinical manifestations result from sequential changes that occur at the cellular level, including lengthening of cardiomyocytes, rearrangement within the myocardial matrix, and death of cardiomyocytes and their replacement by connective tissue; all these changes are part of so-called cardiac remodeling [9].
Different types of cardiomyocyte death have been described, including necrosis, apoptosis, and most recently autophagy, oncosis and programmed necrosis [10]. Necrosis is a well-recognized mechanism of myocardial cell loss during CCHD [3]–[5]. Apoptosis tends to be a chronic process with subtle but harmful impact in different types of cardiac disease [10]. The contribution of apoptosis to cardiomyocyte death in CCHD remains uncertain. A study of chronic chagasic patients with heart failure demonstrated apoptosis in cardiomyocytes and inflammatory cells [11]. However, in another study of patients with chronic chagasic myocarditis, apoptosis was observed in inflammatory cells but not in cardiomyocytes [12]. Apoptosis of endothelial cells, cardiomyocytes and inflammatory cells was observed in the canine model of acute chagasic myocarditis, but dogs were not examined in the chronic phase [13]. In the cardiac tissue of the mouse model, as well as in humans with CCHD, up-regulation of apoptosis related genes such as caspase 12 [14],[15], apoptosis-related Fas antigen, IRF1 and IRF2 [16] has been reported; however, in these studies, the type of cells implicated were not identified [14]–[16].
The prevention of cell death is important to maintain the number of cardiomyocytes and cardiac function. Therapies that regenerate cardiomyocytes using stem cells or progenitor cells show promising results, but clinical trial data are still lacking [17]. Determining the role of apoptosis in myocardial cell loss in CCHD is critical to predicting the impact that interventions at the cellular level might provide in the treatment of heart disease [18].
During cardiac remodeling, cardiomyocyte death leads to replacement by connective tissue, mainly collagen. Collagen determines cardiac structure, function and mechanical properties. Cardiac remodeling during CCHD is characterized by altered collagen turnover and subsequent fibrosis [3], mainly due to an increase of collagen III (CIII) and collagen IV (CIV) [19]. The increase of collagen in cardiac tissue during CCHD is demonstrable in biopsies [20], but noninvasive methods to detect collagen turnover would be useful for prognosis, and in evaluation of therapies and their effect on cardiac remodeling [21]. Many potential biomarkers of fibrosis detect collagen synthesis, such as propeptides of collagen types I and III, procollagen type III amino-terminal propeptide (PIIINP), procollagen type I amino-terminal propeptide (PINP), and procollagen type I carboxy-terminal propeptide (PICP) [21]. Serum levels of PIIINP [21]–[23] and PICP [24] are increased in patients with dilated cardiomyopathy; however, the utility of these biomarkers in CCHD has not previously been evaluated.
We have demonstrated that the guinea pig infection model provides an accurate reflection of the cardiac pathology of Chagas disease in humans [25]. In the current study, we evaluated the role of apoptosis in cell death in the cardiac tissue of guinea pigs during the acute and chronic phases of T. cruzi infection. In addition, we evaluated collagen I, III, IV (CI, CIII and CIV) deposition and fibrosis in cardiac tissue, and their relationship with serum levels of PIIINP and PICP during the course of infection.
The protocol was approved by the San Marcos University Animal Use and Welfare Committee. All experiments adhered to the Guidelines for Animal Experimentation of the Universidad Nacional Mayor de San Marcos.
Trypomastigotes of T. cruzi Y strain were donated by Dr. E. Umezawa, Instituto de Medicina Tropical, Universidade de São Paulo, São Paulo, Brazil. The strain was maintained in an in vitro culture using LLC-MK2 cells following published procedures [26].
We used 70 female Andean guinea pigs, weighing 600–700 g (two months old). The animals were sourced from the Pachacamac region of Lima, an area without vector-borne transmission of T. cruzi. Prior to parasite inoculation, blood samples were taken from each animal and tested for the presence of anti-T. cruzi antibodies and T. cruzi DNA; all the animals were negative for both tests. The animals were fed with special food for guinea pigs (cuyina, Purina), alfalfa and water ad libitum.
Fifty six guinea pigs (experimental group, EG) were inoculated intradermally with 10000 parasites in 100 µl RPMI 1640 medium and 14 guinea pigs (control group, CG) were inoculated with medium alone.
At 20, 25, 40, 55, 115, 165 and 365 days post inoculation (dpi), eight guinea pigs from EG and two from CG were selected at random, sacrificed by intraperitoneal inoculation of sodium pentobarbital (20 mg/kg) and necropsied. The time points for necropsy were chosen based on previous observations of parasitemia and specific IgM levels in serum [21]. Each time point corresponds to an infection phase as follows: acute phase (20–25 dpi), late acute phase (40–55 dpi), early chronic phase (115–165 dpi) and late chronic phase (365 dpi). Blood samples were collected before sacrifice by cardiac puncture. The cardiac tissue was removed and fixed in absolute ethanol and 10% formalin in PBS. Blood and serum samples were stored at −20°C until use.
Levels of IgG1 and IgG2 against T. cruzi were measured by an in-house EAE-ELISA, as described previously [27]. Briefly, ELISA 96 well plates (Immulon 2, Thermo Lab systems, MA) were coated with 3.5 µg/ml (for IgG1 detection) or 2.5 µg/ml (for IgG2 detection) of epimastigote alkaline extract (EAE) antigen and incubated with guinea pig serum 1∶200 dilution. HRP-conjugate was used at a dilution of 1∶ 5000 of goat anti-guinea pig IgG1 or anti-guinea pig IgG2 (ABD Serotec, USA). The plates were incubated with 0.1 mg/well o-phenylene diamine dihydrochloride (Sigma-Aldrich, USA) and 0.05% H2O2. The OD was determined at 492 nm using a Versa Maxmicroplate spectrophotometer (Molecular Devices Corporation, USA). Each serum sample was analyzed in duplicate.
DNA was purified by proteinase K digestion (Invitrogen, Carlsbad, CA) and phenol-chloroform extraction as previously described [25]. A PCR targeting the kinetoplast DNA of T. cruzi was performed using primers 121/122 yielding 330 bp products [28],[29]. Internal control primers specific to guinea pigs (SINEs) were included [30].
Cardiac tissue samples fixed in 10% formalin-PBS were processed and embedded in paraffin. Four 3 µm sections were prepared for each animal: two were stained with hematoxylin-eosin stain and two with Masson's Trichrome stain. All sections were of approximately equal in size. Inflammatory infiltrate was classified as described previously [25]: Absent; focal or mild myocarditis (lymphocytes seen in 2–15% of the entire section); moderate (20–60%); or severe myocarditis (>70%). Mild myocarditis was focal; moderate and severe inflammation was either multi-focal or diffuse. Degrees of necrosis were classified as follows: absent; mild (necrosis in 5%–15% of cardiomyocytes in the entire section); moderate (16–45%); or severe (>45%). The quantification of parasites was based on the mean number of amastigotes seen in the entirety of the two examined sections: absent (no parasites seen), rare (one amastigote nest), moderate (2–10 nests), or abundant (more than 10 nests).
The semiquantification of collagen I, collagen III and collagen IV were performed by IHC. Cardiac tissue samples fixed in formalin-PBS were embedded in paraffin and serial sections of 3 µm were cut. After deparaffinization and hydration steps, the sections were digested with 15 ug/ml proteinase K (Invitrogen, USA) at 37°C for 15 min. Blocking steps were performed using 3% hydrogen peroxide and 5% fetal bovine serum. Rabbit polyclonal antibodies (Cosmo Bio Co, Ltd-USA) against collagens type I, III and IV were used at a dilution of 1/100. An ABC staining kit (Dako, USA) was used for the subsequent procedures. The sections were incubated with diaminobenzidine-hydrogen peroxide kit (Dako, USA) and then counterstained with hematoxylin. The determination of levels of collagen was performed with an optical microscope at 1000×; four quadrants were drawn on cardiac tissue samples and 15 microscopic fields were read in each quadrant. Collagen levels deposition was classified as: normal; mild (focal increase in collagen in 5–25% of fields); moderate (focal increase in 26%–40% of fields); or severe (diffuse increase in >40% of fields).
Levels of carboxy terminal propeptide of type I pro-collagen and N-terminal peptide of type III pro-collagen were measured by ELISA (MyBiosource, USA). Limit of detection was 0.1 ng/ml.
Quantification of apoptotic cells was performed by Terminal Deoxynucleotidyl Transferase dUTP nick end labeling (TUNEL) (Roche Laboratories, USA) according to manufacturer's instructions with some modifications. Briefly, sections of 3 µm were obtained from cardiac tissue stored in ethanol and embedded in paraffin. After deparaffinization and hydration steps, sections were permeabilized with 0.1% triton X-100, blocked with 3% H2O2, and 5% skim milk (Nestle, Peru) and 2% BSA (Sigma, USA). TUNEL reaction was used at a 1/18 dilution. The nucleus were also stained using 5 µg propidium iodide (Biovision, USA) and 300 nM DAPI (Invitrogen, USA). Two adjacent sections of block were used for detection of apoptosis; one section was stained for detection of apoptotic cells using TUNEL as described above and the other was stained for H&E staining to aid in differentiation of the cell types implicated in apoptosis. The sections were examined by fluorescent and optical microscopy at 1000×. Thirty microscopic fields were read in each tissue. The number of apoptotic cells and non-apoptotic cells was counted in each field to determine the percentage of apoptotic cells.
The associations between categorical variables were assessed by chi-square test. The differences in mean levels of apoptotic cardiomyocytes, PICP and PIIINP between guinea pigs with cardiac fibrosis and without cardiac fibrosis were examined by T-Test student with unequal variances. All the calculations were made using the software STATA version 10 (Stata Corp., College Station, TX, USA), p values less than 0.05 were considered significant.
Amastigote nests in cardiac tissue were observed in specimens collected from 20 dpi until 365 dpi. Parasites in cardiac tissue were judged as abundant (mean ± SD: 41±23.6, amastigote nests in the two entire sections) at 25 dpi, but mild or moderate at other time points. Parasites were seen between cardiac fibers and were surrounded with lymphocytes, but most inflammatory cells were distant from the parasites.
The percentage of animals with amastigote nests and positive kDNA PCR results varied by phase: acute phase (62.5% with nests, 100% positive by PCR), early chronic phase (87.5%, 75%), chronic phase (62.5%, 75%). In the chronic phase (365 dpi), the presence of kDNA in cardiac tissue (6/8) was significantly associated with cardiac fibrosis (5/8) (p = 0.035).
IgG1 anti-T. cruzi levels (an indicator of Th2 immune response) were detected from 55 dpi until 365 dpi with a peak at 165 dpi. IgG2 anti- T. cruzi levels (an indicator of Th1 immune response) were detected starting at 20 dpi and peaked at 55 dpi. Levels of IgG2 anti-T. cruzi were always higher (by 1.5–7.5-fold) than levels of IgG1 anti-T. cruzi.
Necrosis of cardiomyocytes was observed at each time point over the course of infection in higher degree than apoptosis. Severe (more than 45% of cardiomyocytes) and moderate (16%–45% of cardiomyocytes) necrosis were seen more often in the acute phase; in the chronic phase, most necrosis was moderate (17%) or mild (5%–15% of cardiomyocytes) (Figure 1A).
Apoptosis was rare in uninfected animals (0–1.9% of cells apoptotic). By contrast, substantial levels of apoptosis were observed in cardiomyocytes (8%–28.4% cells apoptotic), endothelial cells (5.2%–14.6%), epicardial adipose cells (3.3%–14.3%) and lymphocytes (27.2%–54.7%) in cardiac tissue throughout the course of T. cruzi infection (Figure 1B). The highest percentage of apoptotic cells were seen among lymphocytes (Figure 1B and 2a). Apoptosis of cardiomyocytes, endothelial cells and epicardial adipocytes was most frequent during the chronic phase (28.4%, 14.6% and 14.3% of cells apoptotic, respectively). Generally, apoptotic cardiomyocytes, endothelial cells and epicardial adipocytes were observed near inflammatory infiltrate, but distant from amastigote nests (Figure 2b, 2c and 2d). Apoptosis-like death was observed in all amastigote nests, both large and small, throughout the acute and chronic phases of infection (Figure 2e and 2f).
During the chronic phase, areas of adipose tissue between cardiac fibers, often with neighboring inflammatory cells were seen in tissue from 33% (2/6) of infected guinea pigs. Areas of mild to moderate inflammation were observed in epicardial adipose tissue (EAT) and in the cardiac tissue adjacent to it in 55% (22/40) of infected guinea pigs throughout the course of the infection, with highest frequency in the chronic phase (66.7%) (Figure 3), these areas of inflammation were always surrounded by large areas of fibrosis. No parasites were observed in epicardial adipose tissue of guinea pigs.
Isotypes of CI, CIII and CIV were detected in cardiac tissue by immunohistochemistry. From the acute to the early chronic phase (115–165 days pi), there was a mild to moderate increase in the levels of all three types of collagen. A moderate to severe increase in levels of CI, CIII and CIV was observed during the chronic phase (365 days pi), with CIII being present in highest levels (Figure 4a, 4b and 4c). CI and CIII were detected in interstitial and perivascular spaces (Figure 4d, 4e and 4f). The deposits of collagen were near inflammatory cells such as lymphocytes (Figure 4f). Deposits of CIII were also observed close to EAT, usually surrounded by inflammatory cells (Figure 4g). CIV was detected in some interstitial forms and on the basement membrane of cardiomyocytes (Figure 4h and 4i).
Serum levels of PICP and PIIINP were slightly increased during the acute phase. Levels of both PICP and PIIINP were increased in infected animals compared to non-infected animals during the chronic phase. High levels of PICP and PIIINP were associated with the presence of cardiac fibrosis during the course of the infection (p<0.05) (Figure 5).
Knowledge of the mechanisms involved in T. cruzi-induced cardiac pathology is an essential step toward improved understanding of pathogenesis at the cellular level and identification of useful biomarkers of cardiac progression. Serum biomarkers would provide valuable new tools for prognosis and to evaluate new drug candidates aimed at halting progression of CCHD. In the present study, the guinea pig model of T. cruzi infection reflected the known pathology of Chagas cardiomyopathy, including parasite persistence associated with cardiac fibrosis, apoptotic cell death, activation of Th1/Th2 immune response, and increased collagen I, III and IV in the chronic phase. Serum levels of PICP and PIIINP correlated well with fibrosis and could serve as potential biomarkers for cardiac disease progression in patients. Furthermore, our data suggest that epicardial fat may play a role in the pathology of CCHD.
The high levels of anti-T.cruzi IgG2 in the earliest stages of infection suggest that the guinea pig activates an early Th1 response [31] that succeeds in limiting parasite replication, leading to the transition to the chronic phase. These findings demonstrate a process similar to that observed in the chronic Chagas disease model using C57BL/6 mice, a mouse strain that activates a Th1 response and exercises efficient control of the acute infection [32]. However, in the chronic phase of the guinea pig model, the Th2 response (IgG1) was also high. This response may contribute to cardiac fibrosis; the Th2-type antibody response in BALB/c mice is associated with intraventricular conduction abnormalities, sinus bradycardia and widening of the interfibrillar space [33].
The results of this study showed that although necrosis is the main mechanism of cardiomyocyte death throughout T. cruzi infection, apoptosis is a significant contributor to cardiomyocyte loss in the chronic phase. Apoptosis was observed in cardiomyocytes, endothelial cells and adipocytes close to inflammatory cells but remote from parasites, suggesting that release of pro-inflammatory cytokines such as IL1β, INFγ and TNFα may be involved in triggering apoptosis in this setting. The role of cardiomyocyte death by apoptosis in CCHD is still uncertain, but apoptosis has been demonstrated in other dilated cardiomyopathies [34]–[35]. Diverse mechanisms have been described, including activation of protein kinase B signaling, cytochrome c and β-adrenergic receptor pathways, and increased Fas receptor signaling by up-regulation of TNF-α and degradation of inhibitors of Fas receptor signaling [36]. However, because TUNEL staining is not limited to the detection of apoptotic cells [37], further studies are needed to define the impact of apoptosis in cell loss during CCHD.
Apoptosis-like death has been reported in T. cruzi amastigote nests and trypomastigotes, and hypothesized to be the result of control of parasite burden regulated by the parasite itself or by the host, parasite evasion of the host immune response and clonal selection [13],[38]–[39]. However, these studies were conducted during the acute phase [13],[38], whereas we observed apoptosis-like death in nests during both the acute and chronic phases. We have observed apoptosis in small and large amastigote nests, providing evidence against the hypothesis that contact between amastigotes triggers apoptosis as a mechanism to limit intracellular population size [40]. This finding is more consistent with apoptosis in T. cruzi being viewed as a developmentally regulated process, with the activation of different apoptotic mechanisms to allow altruistic survival of the parasite population [39]. The fact that non-apoptotic cells were observed near the parasites even in the presence of inflammatory cells suggests that molecules released by the parasite such transialidases and/or cruzipain could be involved in inhibition of apoptosis in the host cell [41]–[43]. Apoptosis of parasites has been proposed as an important T. cruzi virulence factor, in which stimulation by apoptotic amastigotes leads to anti-inflammatory cytokine production favoring parasite persistence [44].
Although we did not observe parasites in epicardial adipose tissue (EAT), our results demonstrate that inflammation and apoptosis of EAT is an important characteristic of T. cruzi infection in the guinea pig model. EAT is thought to affect cardiac function through paracrine and endocrine regulation [45]. In metabolic and cardiovascular disease states, EAT reduces the production of cardioprotective adipocytokines, such as adiponectin, and induces the production of detrimental pro-inflammatory adipocytokines such as leptin, resistin, IL-6, tumor necrosis factor-α, and IL-17. The resulting inflammatory state alters the balance between vascular nitric oxide, endothelin-1, and superoxide production, promoting vasoconstriction [46],[47]. However, the role of EAT in the pathogenesis of CCHD is still unknown, in part because the more widely used animal models, such as mice and rats, have little or no EAT [46]. The presence of T. cruzi has been demonstrated in subcutaneous adipose tissue in chronic chagasic patients [48] and in brown and white adipose tissue of mouse [49],[50]. The parasite is also known to induce an inflammatory response in adipocyte culture [51], and in the murine model, infection leads to increased levels of inflammatory adipocytokines and decreased levels of anti-inflammatory adipocytokines [49],[50]. The presence of this tissue and the demonstration of T. cruzi-induced pathology make the guinea pig an attractive alternative model to study this feature of Chagas disease.
We also found replacement of cardiac fibers by adipose tissue close to areas of collagen accumulation, similar to the cardiac histopathology observed in dogs with chronic T. cruzi infection [52]. This myocardial adipose tissue may contribute to hyperactivation of β-oxidation of fatty acids leading to excess formation of reactive oxygen species and modulation of the sarco/endoplasmic reticulum Ca2+-ATPase, an early contributor to diastolic dysfunction in myocardial fibrosis and hypertrophy [53].
We observed an increased collagen deposition (CI, CIII and CIV) in cardiac tissue throughout the course of the infection, with CI and CIII in interstitial and perivascular spaces, and CIV in interstitial spaces and on the basement membrane of cardiomyocytes. These results resemble those of the murine model of chronic Chagas disease [19]. We performed all examinations with blinded observers and in duplicate to minimize observer bias. Nevertheless, the use of automated image analysis may provide more accurate quantification of collagen isotypes.
Serum levels of PICP are used as an indicator of collagen I synthesis, based on the release of one molecule of PICP during the synthesis of each molecule of collagen I [54]. Serum levels of PICP have been used as an indicator of the severity of cardiac fibrosis in hypertensive patients [55]. In our data, collagen deposition in heart tissue was significantly correlated with elevated levels of PICP from the late acute phase through the late chronic phase. In the early acute phase, mild levels of collagen I deposition (5–15% of microscopic fields with collagen deposition), but serum PICP levels were not significantly elevated. Some authors have observed that PICP may reflect the rate rather than the quantity of collagen I deposition [54],5[6]. Other potential explanations include suboptimal sensitivity of the serum PICP ELISA [54],[55] or the prolonged time over which the collagen is synthesized; because it has been demonstrated that large quantities of collagen may be deposited in the absence of elevated circulating markers of collagen synthesis [56]. PICP rose significantly in the late acute phase, suggesting an increase in collagen I synthesis following the severe inflammation and necrotic cell death in the early acute phase [21],[57]. Subsequently, activation of anti-inflammatory mechanisms (indicated by high levels of IgG1), induction of cell death by apoptosis and necrosis, and the persistence of the parasite may lead to activation of growth factors such as TGFβ1 which stimulate collagen synthesis resulting in intensified deposition of collagen in the chronic phase [56],[57]
PIIINP is reported to be a useful marker of collagen III synthesis in patients with dilated cardiomyopathy [58], hypertrophic cardiomyopathy [59] and heart failure [60]. In one study, high circulating PIIINP was shown to predict mortality in patients with dilated cardiomyopathy [53]. In contrast to PICP, the relationship between the number of molecules of PIIINP released to the circulation and the number of collagen III molecules synthesized is thought to be variable [54]. PIIINP is not completed eliminated from procollagen during the synthesis of collagen III and the remaining PIIINP may be eliminated from the collagen fiber during degradation. Thus serum levels of PIIINP reflect both collagen III synthesis and degradation [54]. In the late chronic phase we observed a significant increase in serum PIIINP together with severe collagen III deposition by microscopy, suggesting that in this phase collagen III synthesis predominates over degradation.
In summary, our data indicate that PICP and PIIINP are promising biomarkers of cardiac collagen metabolism during the development of Chagas heart disease, and that there may be a contribution to cardiac cell death through apoptosis as well as necrosis. Future studies designed to relate the levels of PICP and PIIINP with cardiac function are needed to determine its potential value for CCHD prognosis and measurement of response to therapy. Further studies could be useful to evaluate whether inhibition of apoptosis could slow the development of cardiomyopathy in patients with chronic Chagas disease.
|
10.1371/journal.pgen.1005622 | Disproportionate Contributions of Select Genomic Compartments and Cell Types to Genetic Risk for Coronary Artery Disease | Large genome-wide association studies (GWAS) have identified many genetic loci associated with risk for myocardial infarction (MI) and coronary artery disease (CAD). Concurrently, efforts such as the National Institutes of Health (NIH) Roadmap Epigenomics Project and the Encyclopedia of DNA Elements (ENCODE) Consortium have provided unprecedented data on functional elements of the human genome. In the present study, we systematically investigate the biological link between genetic variants associated with this complex disease and their impacts on gene function. First, we examined the heritability of MI/CAD according to genomic compartments. We observed that single nucleotide polymorphisms (SNPs) residing within nearby regulatory regions show significant polygenicity and contribute between 59–71% of the heritability for MI/CAD. Second, we showed that the polygenicity and heritability explained by these SNPs are enriched in histone modification marks in specific cell types. Third, we found that a statistically higher number of 45 MI/CAD-associated SNPs that have been identified from large-scale GWAS studies reside within certain functional elements of the genome, particularly in active enhancer and promoter regions. Finally, we observed significant heterogeneity of this signal across cell types, with strong signals observed within adipose nuclei, as well as brain and spleen cell types. These results suggest that the genetic etiology of MI/CAD is largely explained by tissue-specific regulatory perturbation within the human genome.
| Coronary artery disease (CAD) and its subcomponent, myocardial infarction (MI), are the leading causes of infirmity and death worldwide. Large-scale genetic association studies have identified many genetic markers associated with CAD and MI. However, it has been difficult to determine the precise functional effects of these markers. Furthermore, it is unknown which cell types are biologically important in the development of MI/CAD. By intersecting findings from large-scale genetic association studies with functional genomic annotations, we show that genetic markers located in genomic regions that regulate expression of genes make up a large proportion of the genetic risk of MI/CAD. Furthermore, we show that this effect is particularly strong in certain tissues, including adipose, brain and spleen tissue. These results highlight the role of tissue-specific regulatory mechanisms in the genetic etiology of MI/CAD.
| Coronary artery disease (CAD) and myocardial infarction (MI) remain among the leading causes of infirmity and death worldwide despite advances in and widespread adoption of medical therapies treating this disease. Studies have shown a large genetic component for CAD, with the heritability estimated to be 30–60% [1]. Large-scale genome-wide association studies (GWAS) have identified common single nucleotide polymorphisms (SNPs) at 45 loci associated with MI/CAD risk [2–8]. Although these newly discovered loci have led to important new biological insights for MI/CAD [9,10], the proportion of the heritability explained by these loci represents approximately 15% of the estimated heritability [8]. Therefore, a large proportion of the genetic effects are apparently not explained by known loci. This phenomenon has been similarly observed with GWAS for other complex diseases [11].
Previously, we modeled the genetic architecture of MI and CAD using GWAS data [12]. Using simulated genetic models, we inferred that a polygenic model comprised of thousands of associated common variants with small effects explains the majority of the heritability (proportion of total liability-scale variance explained is 0.48, 76% of family-study h2) for MI/CAD [12]. Furthermore, recent work has partitioned heritability of complex diseases into broad categories of the genome [13]. Pooling 11 common diseases, including CAD, Gusev et al. have shown that heritability is disproportionally represented in regulatory elements, specifically in DNase I hypersensitivity sites (DHS) (h2g = 79% for SNPs in DHS regions). However, a specific analysis of CAD yielded a wide interval of the true enrichment of DHS, with genotyped estimates of h2g = 10.9 to 71.3% (95% confidence intervals) for SNPs in DHS regions. Hence, other than being broadly distributed throughout the genome, the molecular consequences of MI/CAD-associated common variants largely remain undefined. Furthermore, it is unclear which cell or tissue types are influenced the most by MI/CAD-associated SNPs.
We addressed these unresolved issues by leveraging data from the National Institutes of Health (NIH) Roadmap Epigenomics Project [14,15] and the Encyclopedia of DNA Elements (ENCODE) Consortium [16]. These projects have comprehensively catalogued biochemical functional regions such as those critical for transcription, transcription factor binding, chromatin structure and histone modification in different cell types, providing unique opportunities to scrutinize links between non-protein-coding DNA sequence variants and gene function. Studies have shown that SNPs in these functional DNA elements can regulate gene expression [17] and common disease-associated loci [18,19].
Using these data, we partitioned the genetic risk of MI/CAD into different categories, to discern drivers in specific cell types that may biologically influence MI/CAD. First, we investigated components of polygenicity and heritability in distinct genomic compartments. Second, we tested for differences in polygenicity, enrichment measures and heritability across diverse cell types within three histone modification marks. Third, we examined clusters of 45 loci discovered from recent GWAS meta-analyses [8] for MI/CAD mapping to the three histone marks in different cell types. Finally, we investigated whether specific biological networks were expressed differently in certain cell types.
We imputed two GWAS datasets, the Myocardial Infarction Genetics Consortium (MIGen) and the Wellcome Trust Case Control Consortium (WTCCC) CAD, using reference haplotypes from the 1000 Genomes Project [20]. We imputed ~7 million SNPs with a high imputation quality metric (>0.5) in both datasets. First, we investigated the relative contributions of different genomic compartments to MI/CAD risk. We partitioned the human genome into three distinct variant sets: “genic noncoding”, “genic coding” and “intergenic” (S1 Table) (see Materials and Methods). For each variant set, we performed two different analyses: 1) polygenic risk score analysis, where we test association of a genetic score comprised of multiple SNPs and 2) SNP-heritability analysis, where we estimate the variance in liability to MI/CAD [21] (see Materials and Methods).
In the polygenic risk score analysis, we observed that the ‘genic’ variant set, defined as genomic regions within 10 kilobases (kb) upstream and downstream of a gene, showed a substantially stronger signal than SNPs in intergenic regions (defined as genomic regions outside of 10 kb of a gene) (Fig 1 and S1 Fig). Similar results were observed amongst genomic regions with window sizes of 20 kb and 50 kb (S2 and S3 Figs). Among this set of variants, both association (Fig 1) and the phenotypic variance explained by the polygenic risk scores before and after normalization by the number of SNPs (S1–S3 Figs) were the most significant in regions adjacent to the protein-coding DNA regions, called “genic noncoding” regions. We observed that polygenic risk scores were strongly associated with MI/CAD in “genic noncoding” regions (P<10−10), explaining 1 to 1.5% of the phenotypic variance. By comparison, polygenic risk scores in regions within protein-coding DNA regions, called “genic coding” regions, were less strongly associated (P<10−5) and explained approximately 0.5% of the variability. Similar patterns were observed after normalizing by the number of SNPs in the polygenic risk score. These patterns were particularly evident for P value thresholds <10−5 (see Materials and Methods). The association signal remained among the variant sets with discovery P value thresholds greater than or equal to 0.05 after excluding regions within ±1 megabase of the 45 known SNPs associated with MI/CAD risk [8] (S4 Fig).
We further examined the role of different genomic compartments on the heritability for MI/CAD risk by testing whether SNPs in the three compartments make up a large portion of the heritability for MI/CAD (Table 1 and S2 Table). Consistent with findings from our polygenic risk score analysis, we observed that most of the heritability resides within the “genic” regions. In a meta-analysis of the MIGen and WTCCC CAD studies, SNPs in “genic noncoding” regions explained approximately 58.9% (variance in liability = 0.25, P = 1×10−9) of the total heritability, resulting in a fold enrichment of variance of 1.2. In contrast, the heritability of MI/CAD explained by SNPs residing within “genic coding” regions was estimated to be only 10% of the total (variance in liability = 0.042, P = 0.07). SNPs residing within “genic coding” regions accounted for only 0.5% of the total number of variants, resulting in a high fold enrichment of variance of 19.1. Despite the enrichment of variance estimates for both “genic coding” and “genic noncoding”, neither category statistically deviated beyond expectation (P = 0.088 and P = 0.23 respectively). On the other hand, a statistically significant depletion of variance in liability was observed in the “intergenic” regions compared to expectation (observed variance in liability = 0.13, expected variance in liability = 0.22 [0.52 fraction of SNPs of total × 0.42 total variance in liability], difference in observed and expected variance in liability P = 0.0089). Similar results were observed amongst genomic regions with window sizes of 20 kb and 50 kb (S3 and S4 Tables). Heritability estimates with a prevalence of 3% of early-onset MI/CAD showed reduced heritability (variance in liability = 0.35) explained by the genomic compartment whereas P values and enrichment of variance remained the same (S5 Table).
Given the high polygenicity for MI/CAD explained by SNPs in noncoding regions surrounding protein-coding regions, we further examined whether polygenicity is stronger within specific regulatory elements. Using data from the NIH Roadmap Epigenomics Project (see URL), we specifically examined three histone modification marks that are indicative of active promoters (H3K4me3/H3K9ac) or enhancers (H3K27ac). A polygenic association signal comprised of SNPs with association P<0.05 for MI/CAD was stronger in the histone marks, beyond what we expect by chance after randomly sampling “genic noncoding” regions outside of the marks (Mann-Whitney test P = 1.1×10−95) (S5 Fig).
We next tested for differences in polygenicity, enrichment and heritability estimates between different cell or tissue types within the three histone modification marks (Fig 2). We observed heterogeneity on the polygenicity of MI/CAD between cell types (Fig 2A). For example, SNPs in the H3K27ac and H3K9ac in bone marrow derived mesenchymal stem cell cultured cells and SNPs in the H3K4me3 regions in mesenchymal stem cell derived adipocyte cultured cells were amongst the strongest signals. Meanwhile, SNPs in any of the three histone marks in hematopoietic CD3, Treg, CD4, CD25, CD45RA primary cells were amongst the weakest signals. Enrichment analyses showed strong excesses for highly associated SNPs (P<10−5), compared with matched random SNPs, for mesenchymal stem cells, heart tissues such as fetal heart and ventricle, and intestinal mucosa including rectal and colonic mucosa, among others (Fig 2B). Cell-type specific effects were also generally consistent for heritability estimates (Fig 2C). The cell-type specific enrichment signals were strongest among strongly associated SNPs (Fig 2B).
When tested for ENCODE data, statistically significant polygenicity and high heritability estimates were observed for variants residing in specific active chromatin regions, including enhancers and weak transcription regions (about 3 and 11% of reference genome, respectively) across several cell lines, including skeletal muscle and vascular cell lines (S6 Fig). Polygenicity and heritability estimates were substantially weaker for variants in inactive chromatin states than those in active states (S7 Fig), with the notable exception of the quiescent state. Heritability estimates for variants in a quiescent state were particularly high although the variants in this state accounted for >60% of the reference epigenome (S7 Fig). However, the enrichment of variance (% variance of total divided by % SNPs of total) in the quiescent state was low (on average 0.76 [0.50−0.92] in different cell lines) compared with those for enhancers (on average 6.5 [3.2−10.3]) and weak transcription (1.9 [1.1–2.8]). In general, we note that ENCODE cell line chromHMM state results were largely driven by the sizes of these genomic compartments (i.e. % SNPs), which would be expected if alignment of MI/CAD genetic effects and ChromHMM stats were random.
These observations have implications for fine-mapping loci discovered from large-scale GWAS. In particular, the results suggest that causal variants within GWAS loci are overrepresented in regulatory elements adjacent to protein-coding regions of the genome. Therefore, we investigated 45 independent, lead SNPs discovered from recent GWAS meta-analyses [8] for MI/CAD and tested whether a disproportionate number of these SNPs overlapped histone marks across diverse cell lines and tissues. We generated 10,000 random sets of SNPs that were selected to match the query GWAS SNPs based on similar minor allele frequency (±0.05 frequency), number of SNPs in linkage disequilibrium (LD) with the query SNP (±10% of number of SNPs in LD with query SNP using r2>0.5), distance to nearest gene (±10% of distance of nearest gene from query SNPs) and gene density (±10% of number of genes in loci around the query SNPs) [25]. We excluded two SNPs (rs3798220 and rs12205331) out of the 45 GWAS SNPs because we were unable to find appropriate matching null SNPs. We observed that 25 (58.1%) out of 43 GWAS SNPs overlapped one of three histone marks. From the 10,000 random null sets, a median of 15 (34.9%) out of the 43 random SNPs (1st quartile of 13 and 3rd quartile of 17) overlapped histone marks (S8 Fig). Only a very small fraction of random sets showed a higher number of GWAS SNPs overlapping with histone marks than what we observed (4 out of 10,000 random null sets) (P = 4×10−4).
Because histone modification differed between cell types, we next examined whether the 45 MI/CAD GWAS SNPs yielded differential gene regulation across various tissues. We mapped the 45 GWAS SNPs, along with SNPs in high LD (r2≥0.8), that reside within each of the three histone marks in different cell types. We observed distinct patterns between the different GWAS loci and cell types (Fig 3 and S9 and S10 Figs). For example, for the histone mark H3K27ac, 12 of the 45 loci were expressed in more than 80% of the cell types, whereas 13 of the 45 loci were expressed in less than 20%. Using HaploReg v2 [26], specific cell lines displayed enrichment with the 45 loci in strong enhancer regions. In particular, cell types related to adipose nuclei, spleen and brain tissue were amongst the strongest enrichment signals for inferred strong enhancer chromatin states (Table 2). We observed consistent results in these cell lines when also considering SNPs with less stringent significance levels using polygenic association analysis (Fig 2). HaploReg analysis for available ENCODE cell lines showed 24-fold enrichment with the 45 loci in enhancer regions in H1 cell lines (P = 3×10−5).
Finally, to investigate whether specific biological networks are expressed in specific cell types, we examined connectivity of protein-protein interaction (PPI) networks among the 45 GWAS loci (45 GWAS SNPs, as well as SNPs in r2≥0.8) in different cell lines. Consistent with the HaploReg v2 and polygenic association analysis, we observed high direct connectivity in a PPI network involving known lipid genes, particularly apolipoprotein E (APOE), apolipoprotein C3 (APOC3) and low-density lipoprotein receptor (LDLR) in adipose nuclei (P = 0.002) and mesenchymal stem cell line derived adipocyte cultured cells (P = 0.001), and apolipoprotein B (APOB) and Sortilin 1 (SORT1) in adult liver (P = 0.009) (S11 Fig). Specific effects were observed in PPI networks in other cell types as well, including adult liver (S6 Table).
We utilized several complementary human genetic approaches to partition the genetic risk of MI/CAD into different genomic categories and cell types. Three principal findings emerged: (1) genetic variants residing in noncoding regions flanking protein-coding genes make up a large proportion of the heritability for MI/CAD; (2) association signals are enriched in histone modification marks; and (3) clear cell-type specific effects emerged with genetic effects of MI/CAD-associated SNPs being enhanced in adipocyte cell lines.
We highlight an important role for genetic variants residing within specific regulatory elements including promoters and enhancer regions, suggesting that the genetic risk for MI/CAD is largely driven by variants in key regulatory elements. Because we did not adjust for traditional risk factors of MI/CAD, our heritability estimates may include the fraction derived from these risk factors. Furthermore, our heritability estimate of 0.43 for the total genome is consistent with previous family-based studies with estimates of 0.3–0.63 [1,28,29]. Our work is consistent with previous studies that showed that a significantly higher portion of GWAS SNPs overlap regulatory elements such as transcription factor binding regions and/or DNase I hypersensitive sites [18,30]. These findings have important implications for interpretation of GWAS signals, particularly for identifying causal variants. Our findings particularly highlight an important role for promoter and/or enhancer regions in fine-mapping GWAS signals.
Our analyses further highlighted the genetic effects of MI/CAD-associated SNPs occurring in specific cell types, including mesenchymal and adipocyte cell lines. Adipocytes may affect cardiovascular homeostasis by regulating diverse peptides and nonpeptide compounds, which have been implicated in cardiovascular disease pathogenicity [31]. Furthermore, obesity is a casual risk factor for MI/CAD [32]. Here, we demonstrate that alteration of chromatin-mediated gene regulation by DNA sequence variation at key genomic regions within adipocytes is an important determinant of MI/CAD risk. Our results are consistent with studies that have shown that adipocytes play an important role in the pathogenesis of obesity, with adverse effects on inflammation, hemodynamics, and cardiovascular function [33]. We propose that the adipose-cardiovascular axis is not only an acquired mediator of MI/CAD risk but also a critical genetic determinant of MI/CAD risk in the general population. Furthermore, we highlight other loci that are not known to be involved in cardio-metabolism but appear to map to histone marks in the adipocyte cell lines. For example, SNPs in the MIA3 and MRAS loci are highly associated with CAD [2–4] but have not been found to be associated with any cardio-metabolic intermediate trait [34]. Notably, we observe enrichment of MI/CAD-associated SNPs disrupting regulation in brain and spleen tissue raising new hypotheses in the pathogenesis of human atherosclerosis. Hypothalamic-pituitary regulation of the autonomic nervous system has a diverse array of impacts on cardiovascular disease determinants including blood pressure, heart rate, sodium regulation, metabolism, and physiologic responses to stress [35]. Furthermore, proinflammatory mediators within the spleen may have an important role in MI [36]. Because the spleen consists of a multitude of cell types, including immune B/T cells, macrophages, monocytes, endothelial cells, smooth muscle cells from larger arterioles, chromatin data of specific cell types in the spleen may be useful to identify specific functional roles related to MI. The mechanism by which these tissues contribute to a higher enrichment to MI/CAD associated SNPs may be through tissue-specific regulatory perturbation in functional regulatory regions (histone marks and/or chromatin enhancer states) adjacent to protein-coding regions, specifically in adipose, heart, brain and spleen tissues. Thus we propose that genetic determinants of early-onset myocardial infarction have biologic roles within distinct tissue types and these observations should prioritize experimental modeling strategies.
We note the following limitations. We note that the case samples in MIGen and WTCCC CAD were ascertained based on early-onset MI/CAD that was not fatal (male cases < 50 years old and female cases < 60 years old). Therefore, the genetic architecture and some of the tissue types highlighted in this study may be less relevant to the more general, broader CAD phenotype. Furthermore, we observed strong enrichment signals in some tissues (such as heart and intestinal mucosa) that may serve as a proxy for tissues that are biologically relevant to MI/CAD (such as smooth muscle in arterial walls). This may be due to tissues with similar cell types (i.e. smooth muscle cells in different tissues) having similar open chromatin structures. We note that we observed relatively weak signals for immune-related tissues and cell types despite a recent study suggesting that there was enrichment of tissue-specific eSNPs associated with CAD in pathways related to the immune system[37]. We have not examined our results in the context of biological pathways relevant to specific cell types (for example, immune/inflammation pathways in immune cells), and further investigation on this is warranted.
In summary, we have shown that disease-causing variants for MI/CAD are enriched in promoters and enhancer regions flanking protein-coding genes. Functional data from the NIH Roadmap Project and ENCODE provide key links to specific cell types such as mesenchymal stem cell derived adipocyte cultured cells, heart, brain and spleen cells with risk variants for MI/CAD. Our results highlight the importance of investigating the noncoding regions of the genome in genetic studies and suggest a key role of tissue-specific regulatory mechanisms on the etiology of MI/CAD. Identifying critical nodes that are significant drivers of a substantial portion of human disease can both inform biological investigation and prioritize therapeutic efforts.
Details regarding material and methods for primary analyses presented in main figures and tables are provided below and in corresponding figure and table legends. Details about methods for supplemental figures and tables are provided in the legends in the Supporting Information.
We obtained GWAS data for 5,903 samples from MIGen [5] (2,905 cases and 2,998 controls) and 4,837 samples from WTCCC [38] (1,914 cases and 2,923 controls). For quality control, samples with extreme heterozygosity, gender mismatch or sample call rate <95% and SNPs with Hardy-Weinberg equilibrium P<10−6, minor allele frequency <1% or SNP call rate <98% were excluded prior to imputation.
We prephased MIGen GWAS data using the MaCH software and imputed into the 1000 Genomes Phase I Integrated Release Version 3 Haplotypes panel using minimac [39]. We imputed WTCCC CAD data using IMPUTE2 [39–41]. To control for imputation quality, we removed low-frequency SNPs with minor allele frequency <0.5% and used SNPs with high imputation quality metric (minimac rsq or IMPUTE2 info) >0.5. We tested for association with additive tests for imputed dosage data using the SNPTEST [40,42] and PLINK [43] software. Gender, age and principal components for population structure [44] were used for covariates in the association analysis.
A categorization of SNPs was adopted in order to compare their relative aggregate properties (S1 Table). “Genic” regions were defined as ±10 kb of the 3′ or 5′ untranslated regions (UTR) of a gene. “Intergenic” regions were those that were beyond ±10 kb of the 3′ or 5′ UTR of a gene. “Genic coding” variants were those that code amino acid sequence (nonsense, missense, synonymous). “Genic noncoding” variants were those that were resided outside of the coding region but within the “genic” window size. This includes the region beyond the 3’ or 5’ UTR, as well as the introns. 19.5% of variants in the “genic noncoding” region are also observed in the DHS regions as defined by Gusev et al [45]. Genomic compartments for the “genic coding”, “genic noncoding”, and “intergenic” regions were defined to be non-overlapping. Other window sizes for “genic” regions of ±20 kb and ±50 kb were also tested.
To examine polygenic effects of specific chromatin marks, we obtained histone modification marks (histone H3 lysine 4 trimethylation or H3K4me3, histone H3 lysine 9 acetylation or H3K9ac, histone H3 lysine 27 acetylation or H3K27ac) in diverse cell types or tissues that are indicative of active enhancers or promoters from the NIH Roadmap Epigenomics Project [14,15] data repository (see URL). We identified histone mark using the MACS v1.4 software [46], with a P value cutoff of 10−5 and a false discovery rate cutoff of 0.01. We ran the histone mark test files with control files matched by individual if available in order to increase specificity. We also obtained chromatin core 15-state data for 16 ENCODE cell lines (Epigenome ID E114-E119) (see URL). We examined eight active chromatin states (active transcription start site [TSS] proximal promoter states [active TSS, flanking active TSS], a transcribed state at the 5′ and 3′ end of genes showing both promoter and enhancer signatures [transcription at gene 5′ and 3′], actively transcribed states [strong transcription, weak transcription], enhancer states [enhancers, genic enhancers], and a state associated with zinc finger protein genes [ZNF and repeats]) and seven inactive states (constitutive heterochromatin, bivalent regulatory states [bivalent TSS, flanking bivalent TSS/enhancers, bivalent enhancers], repressed PolyComb states [repressed PolyComb, weak repressed PolyComb], and a quiescent state) inferred by ChromHMM [15].
Polygenic risk score (PRSi) for individual i for a given variant set was calculated as PRSi = ∑j∈SNPsβj×dij, where βj >0 is the natural log of odds ratio for the risk allele of SNP j from the association result in the MIGen discovery set and dij is the dosage (0−2) for that same allele of individual i from the WTCCC CAD validation set, as previously described [12]. Association of polygenic risk scores with disease status was tested using the Wald test and variance explained was estimated by Nagelkerke’s R2 from logistic regression [47]. Variance explained after normalization was calculated using Nagelkerke’s R2 divided by the number of SNPs. We used different discovery P value thresholds in MIGen (association P<5×10−8, 5×10−7, 5×10−6, 5×10−5, 5×10−4, 0.005, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 1) to define various polygenic risk scores (Fig 1).
The proportion of phenotypic variance explained by each variant set was estimated using the restricted maximum likelihood method [48] implemented in the Genome-wide Complex Trait Analysis (GCTA) software, transformed to the underlying liability scale assuming a prevalence of 6% for CAD [22,23]. We removed one of each pair of individuals with estimated relatedness larger than 0.05 (grm-cutoff 0.05 in the GCTA software). Given a previous observation that LD does not substantially influence polygenic risk score analysis [49] or heritability explained by genotyped SNPs [13], and that imputed SNPs do not produce biased heritability estimates compared to genotyped SNPs [13], we included all SNPs in our variant sets (Table 1). We performed heritability estimates independently in the MIGen and WTCCC CAD studies, and then a meta-analysis across both studies using as weights the inverse variance (Tables 1 and S2–S4). We also estimated heritability with a prevalence of 3% for early-onset MI/CAD (S5 Table).
We performed analyses on SNPs within three histone marks (H3K27ac, H3K4me3, H3K9ac) in different cell types. We performed polygenic risk score analysis as described above. We used a discovery P value threshold of <0.05 in MIGen to form the polygenic risk score and then validation in WTCCC CAD (Fig 2A). We performed enrichment analysis by comparing the proportion of significant variants passing a specific association P threshold of a variant set with that of a baseline set. We tested different association P thresholds P<5×10−7, 5×10−6, 5×10−5. For the baseline set, we examined variants that resided in the intergenic regions, were not conserved (Genomic Evolutionary Rate Profiling score [50] <5) and did not overlap with any of the studied regulatory elements (Fig 2B). We performed heritability analysis as described above. We restricted heritability analysis to variants within histone marks for the indicated cell line using the restricted maximum likelihood method [48] implemented in the GCTA software (Fig 2C) [22,23]. For all approaches, we performed analyses restricting to SNPs residing in the three histone marks (H3K27ac, H3K4me3, H3K9ac) that were present in the different cell types
For enrichment analyses, we selected 45 independent SNPs that were shown to be significantly associated with CAD in a large-scale GWAS meta-analysis (Table 2) [8]. We also included tag SNPs in strong LD (r2≥0.8 in 379 individuals from European populations [85 CEU, 98 TSI, 93 FIN, 89 GBR, 14 IBS] from the 1000 Genomes Project [20]) with the 45 GWAS SNPs. We utilized the NIH Roadmap and ENCODE data and two mammalian conservation algorithms, GERP and SiPhy-omega, implemented in HaploReg v2 [26] (see URL) to examine if the 45 GWAS loci are enriched in regions of inferred strong enhancer chromatin states [27] in specific cell types. The background set for enhancer enrichment analysis was “All SNPs in 1000 Genomes phase I data”.
For hierarchical clustering analysis, we mapped all 45 MI/CAD GWAS SNPs, along with SNPs in high LD (r2≥0.8) (same SNP set in enrichment analysis), to each of the three histone marks (H3K27ac, H3K4me3 and H3K9ac), which are associated with active regulatory regions, in different cell types (Fig 3). Hierarchical clustering was performed using the heatmap function in R (R Project for Statistical Computing). Reordering of the rows and columns to produce the dendrogram was based on the presence or absence of a SNP residing in a histone mark in different cell types.
We tested for direct connectivity of genes in GWAS loci in specific cell types. We tested 45 MI/CAD GWAS SNPs, in addition to SNPs in high LD (r2≥0.8) (same SNP set in enrichment analysis), that overlapped with the three histone marks (H3K4me3, H3K9ac, H3K27ac) in a specific cell type. DAPPLE [51] was utilized to test for direct connectivity in PPI networks. Gene regulatory regions were defined as within 110 kb upstream of transcription start site and 40 kb downstream of transcription end site of each of the 45 lead SNPs or tag SNPs were included in the analysis. We performed 1,000 permutations to obtain empirical significance for the observed connectivity compared with the expected connectivity.
NIH Roadmap Epigenomics Project, http://www.roadmapepigenomics.org ENCODE Chromatin Core 15-State data, http://egg2.wustl.edu/roadmap/data/byFileType/chromhmmSegmentations/ChmmModels/coreMarks/jointModel/final/ HaploReg v2, http://www.broadinstitute.org/mammals/haploreg/haploreg.php
|
10.1371/journal.pbio.2005380 | The actin remodeling protein cofilin is crucial for thymic αβ but not γδ T-cell development | Cofilin is an essential actin remodeling protein promoting depolymerization and severing of actin filaments. To address the relevance of cofilin for the development and function of T cells in vivo, we generated knock-in mice in which T-cell–specific nonfunctional (nf) cofilin was expressed instead of wild-type (WT) cofilin. Nf cofilin mice lacked peripheral αβ T cells and showed a severe thymus atrophy. This was caused by an early developmental arrest of thymocytes at the double negative (DN) stage. Importantly, even though DN thymocytes expressed the TCRβ chain intracellularly, they completely lacked TCRβ surface expression. In contrast, nf cofilin mice possessed normal numbers of γδ T cells. Their functionality was confirmed in the γδ T-cell–driven, imiquimod (IMQ)-induced, psoriasis-like murine model. Overall, this study not only highlights the importance of cofilin for early αβ T-cell development but also shows for the first time that an actin-binding protein is differentially involved in αβ versus γδ T-cell development.
| T cells are produced in the thymus and are critical to fighting infections and combating cancer. To move through the body and to fulfill their specific functions, T cells need to dynamically reshape their cell body. This requires remodeling the actin cytoskeleton using a plethora of actin-binding proteins, including cofilin. Whereas the majority of T cells use one type of cell membrane protein called αβ T cell receptor (TCR) to recognize their target, a minor population of T cells use a different type of receptors that are called γδTCR. The decision on whether thymocytes, the precursors of T cells, develop into αβTCR or γδTCR-bearing T cells takes place within the thymus. By replacing the cofilin gene with a nf copy, we identified an important role for cofilin in T-cell development. These mutant mice exhibited a severe thymus atrophy. Importantly, αβ T-cell development was severely affected, but the γδ T cells were unaffected in number and functional, as they were capable of responding to activation signal both in culture and inside the body. Overall, our study reveals the importance of cofilin in early αβ T-cell development and shows for the first time that an actin-binding protein is differentially involved in αβ versus γδ T-cell development.
| One requirement for T-cell–mediated immune surveillance is the permanent reshaping of the cell body. Here, a functional remodeling of the actin cytoskeleton is important for changes of the cell shape during migration or immune synapse (IS) formation with antigen presenting cells (APCs) or target cells [1–5]. One protein that drives these actin dynamics is cofilin. Cofilin is a 19-kDa actin-binding protein that belongs to the actin depolymerizing factor (ADF)/cofilin family. In humans and mice, three different highly conserved isoforms are expressed [6,7]: nonmuscle cofilin (n-cofilin or cofilin-1 [Cfl1]) [8], muscle cofilin (m-cofilin or cofilin-2) [9], and destrin or ADF [10]. This study focused on Cfl1, which is highly expressed in T cells [11]. Cofilin has a dual function for actin dynamics, as it is both depolymerizing and severing actin filaments [12]. In resting human peripheral blood T cells (PBTs), cytoplasmic cofilin is constitutively phosphorylated at its serine 3 residue and thus inactive. Cofilin phosphorylation (inactivation) is mediated by LIM or testis-specific kinases (reviewed by Mizuno and colleagues [13]). Upon costimulation of resting T cells but not by TCR triggering alone, cofilin is dephosphorylated and thereby transmitted to its active state [11,14,15]. Once active, cofilin exerts its actin remodeling function which is crucial for proper IS formation and T-cell activation [16,17]. Dephosphorylated cofilin can also translocate to the nucleus where it may have anti-apoptotic functions and may enhance transcription [11,18]. It can furthermore serve as nuclear shuttle for actin [11,19], which is also involved in different nuclear mechanisms (reviewed by Falahzadeh and colleagues [20]). Besides T-cell costimulation, chemokine receptor triggering (e.g., by SDF-1α) can also lead to the dephosphorylation of cofilin [21]. In this regard, it was also shown that an active mitogen-activated protein kinase kinase (MEK) cofilin module is needed for T-cell movement [21], known to be driven by constant actin flow, i.e. migration in 3D environments [22–25]. The activity of cofilin is not only inhibited by phosphorylation but also by binding to phosphatidylinositol 4,5-bisphosphate (PIP2) near the plasma membrane and by a pro-oxidative microenvironment. Cofilin is inactivated by oxidation provoking T-cell hyporesponsiveness or in a long-term perspective necrotic-like programmed cell death [26,27]. In a reducing environment, however, even PIP2-bound cofilin becomes active, leading to enhanced actin dynamics in the vicinity of the plasma membrane [28].
Even though the essential role of cofilin for T-cell activation and migration was proven in in vitro studies of human PBTs, there is nothing known about the importance of cofilin for T-cell development in vivo. Thus, we created a mouse line in which T-cell–specific a nf form of cofilin was expressed instead of endogenous cofilin. The decision to use a cofilin knock-in rather than a knock-out mouse was due to the observation that knocking out a protein can result in an elevated expression of other proteins, which could in turn compensate for the lack of the protein of interest [29,30]. With the generated mice, we could show that cofilin is crucial for early αβ but not γδ T-cell development.
To overcome potential disadvantages of fusion proteins such as alterations of protein activity or subcellular localization, coexpression of fluorescent dyes together with the protein of interest is widely used to monitor protein expression and/or promoter activities. With the help of the viral 2A consensus motif, two proteins can be coexpressed from a single mRNA by a mechanism called “ribosome skipping” [31–33]. Upon cotranslational cleavage, most of the 2A sequence remains attached to the C-terminus of the upstream protein, whereas only a single proline stays attached to the N-terminus of the downstream protein. Cofilin is reported to undergo cotranslational processing at its N-terminus and its activity is post-translationally regulated by (de)phosphorylation at its serine 3 residue. We wondered whether or not addition of proline to cofilin’s N-terminus would lead to its inactivation. Therefore, we created a plasmid in which an enhanced green fluorescent protein (eGFP)-2A-Cfl1 expression cassette was cloned under a cytomegalovirus (CMV) promoter. To test expression and functionality of cofilin derived from the eGFP-2A-Cfl1 expression cassette, the plasmid was transfected into Jurkat leukemia cells in which the endogenous cofilin was knocked down via siRNA. A vector in which the C-terminus of cofilin was fused to eGFP served as positive control. Transfection efficiency and successful expression of the eGFP-2A-Cfl1 cassette was visible by eGFP analysis (Fig 1A). Cofilin protein expression was further confirmed by western blot analysis of total Jurkat cell lysates (Fig 1B).
We first examined whether the addition of proline to cofilin’s N-terminus influences cofilin phosphorylation. Comparing phosphorylation of endogenous cofilin (Fig 1C, lane 1, pCfl1) and the cofilin from the eGFP-2A-Cfl1 cassette (Fig 1C, lane 5, pCfl1) revealed much less phosphorylation of the latter. Note that cofilin expressed from the positive control vector (Fig 1C, lane 4, pCfl1-eGFP) showed no alteration in the phosphorylation state.
To test the functionality of nonphosphorylated cofilin encoded by eGFP-2A-Cfl1, the F-actin content of transfected Jurkat cells was determined by analysis of phalloidin binding via flow cytometry (Fig 1D). As expected, Jurkat cells with a successful cofilin knock-down showed an increase in total F-actin. Cotransfection with a positive control vector, which expressed eGFP-tagged WT cofilin, rescued F-actin depolymerization. However, Jurkat cells with cofilin knock-down that expressed eGFP-2A-Cfl1 harbored a similar high F-actin content as cells transfected with siRNA only. Thus, even though cofilin from the eGFP-2A-Cfl1 plasmid was expressed and less phosphorylated, it was not functional pointing towards defective regulation by phosphorylation.
Overall, cofilin expressed from the eGFP-2A-Cfl1 vector showed a defect in both phosphorylatability and actin remodeling function.
Having observed the functional inactivity of cofilin obtained from the eGFP-2A-Cfl1 expression cassette in vitro, we wondered about the consequences of cofilin dysfunction in T cells in vivo. Therefore, we generated mice expressing an eGFP-2A-Cfl1 expression cassette instead of endogenous cofilin specifically in T cells. Thus, the nf form of cofilin should be expressed only in T cells. The targeting strategy used for generation of knock-in mice is shown in S1A Fig. In short, an eGFP-2A-Cfl1 expression cassette was inserted into the intronic region between exon 1 and 2 of the mouse cofilin gene. To prevent transcription of the cassette, a floxed stop cassette was included in front. Another locus of X (cross)-over in P1 (loxP) site was cloned into the noncoding region of exon 1. T-cell–specific knock-out of endogenous cofilin and knock-in of the eGFP-2A-Cfl1 expression cassette was achieved by crossing mice carrying the construct with lymphocyte-specific protein-tyrosine kinase (Lck)-Cre mice that express Cre recombinase under the proximal p56lck (Lck) promoter [34]. WT mice (Cfl1+/+) could be discriminated from heterozygous (Cfl1+/nf) and homozygous knock-in mice (Cfl1nf/nf) by PCR (S1B Fig). All mice were born with an expected Mendelian ratio and developed without apparent signs of abnormality. Rarely, Cfl1nf/nf mice showed inflamed cheeks or intestinal prolapses. Successful T-cell–specific knock-in of the expression cassette was confirmed by flow cytometry (via eGFP expression; S1C Fig). Please note that eGFP positive cells were already detected in heterozygous DN1 thymocytes (S1D Fig) but not in common lymphoid progenitors in the bone marrow. This is also in line with earlier studies investigating the activity of the Lck proximal promoter [35].
To further characterize the nf cofilin mutant, cofilin obtained from T cells of Cfl1+/nf mice (expressing both WT and nf cofilin) and control B6 mice was subjected to mass spectrometry (S1E Fig). Besides its post-translational regulation by phosphorylation, cofilin was reported to undergo N-terminal excision of the initiator methionine followed by acetylation of alanine (Uniprot; P18760). Accordingly, mass spectrometry analysis of cofilin from B6 T cells revealed the presence of peptides starting with acetylated alanine (S1E Fig). Thereby, peptides were either phosphorylated at serine 3 (Ac+Ph) or dephosphorylated (Ac) (S1E Fig, left). These two peptide species (Ac and Ac+Ph) were also identified in MS/MS analysis of cofilin obtained from T cells of Cfl1+/nf mice, which express both wt and nf cofilin (S1E Fig, right). Additionally, a N-terminal cofilin peptide starting with proline, followed by methionine and alanine was found only in Cfl1+/nf mice. In this peptide, no serine phosphorylation and, due to the N-terminal proline-methionine, also no alanine acetylation could be detected. Thus, in the generated knock-in mice, the single remaining proline residue hinders co- and post-translational processing of cofilin in T cells.
Having established homozygous mice expressing nf cofilin in T-cells, we next characterized their immune cells. Nf cofilin knock-in mice (Cfl1nf/nf) had similar numbers of total splenocytes as wt B6 animals (S2A Fig). However, their lymph node (LN) cell numbers were significantly diminished (S2B Fig). Further analysis of leukocyte cell populations revealed that Cfl1nf/nf mice show a massive decrease in T-cell populations both in percentage and numbers in spleen (Fig 2A, S2A Fig). The almost complete lack of T cells in the periphery was accompanied by an absolute increase in other leukocyte cell populations such as splenic B-cells, natural killer (NK) cells, and dendritic cells (DCs) as well as eosinophils and neutrophils (S2C Fig). This finding explains why total cell numbers in the spleen were normal despite the nearly complete loss of T cells in Cfl1nf/nf mice. Note that mice carrying the nf cofilin construct homozygously without Cre-mediated knock-in and also mice carrying the nf cofilin construct heterozygously with Cre-mediated knock-in had similar T-cell populations as B6 mice (S2D Fig).
In regard to the small T-cell numbers in the periphery, we next investigated the thymic development of nf cofilin knock-in mice. Here, Cfl1nf/nf mice showed a severe thymus atrophy, which was characterized by a decrease in the thymic cell number of more than 99% (Fig 2B). Flow cytometric characterization demonstrated that thymocytes were almost exclusively found within the CD4- CD8- double negative (DN) stage, suggesting a very early block in T-cell development (Fig 2C, left bar chart). The DN stage can be further discriminated into 4 developmental steps by differential expression of surface CD25 and CD44 [36]. Within the DN stage, thymocytes were mainly detected in the DN2 (CD44+ CD25+) and DN3 (CD44- CD25+) stage, with most cells accumulating at the DN3 stage. Furthermore, a loss of cells in the DN4 stage (CD44- CD25-) was observed (Fig 2C, right bar chart).
T-cell development is not determined solely by T-cell progenitors themselves but is also influenced by the thymic stroma. To test whether the reason for impaired thymocyte development is T-cell intrinsic, mixed bone marrow chimeras were created (Fig 2D). To this end, irradiated B6 mice were injected with a 1:1 ratio of bone marrow (BM) tester cells (derived from Cfl1nf/nf knock-in mice; CD45.2+) and control competitor cells (B6; CD45.1+). Once successful reconstitution was verified in peripheral blood of the recipient mice, their thymus was taken out and cells were analyzed by flow cytometry. Thymocytes that originated from BM of Cfl1nf/nf accumulated in the DN stage (mainly in DN3), whereas control competitor cells derived from B6 mice developed completely normally. B-cells that originated from BM of Cfl1nf/nf mice developed to a normal extent (S2E Fig). This indicates that the disturbed T-cell development in nf cofilin knock-in mice is caused by T-cell intrinsic factors. Further, the number of CD45.2+ cells which were found in the thymi of reconstituted B6 mice was much smaller than the one of CD45.1+ control cells (3% versus 95%), implying not only a developmental but also a proliferative disadvantage of cells which originated from BM of homozygous knock-in mice. Overall, the severe thymus atrophy seems to be caused by a lack of DN thymocyte expansion.
Despite the enormous thymus atrophy and reduction in peripheral T-cell numbers, there were few CD3+ cells detected in secondary lymphoid organs of Cfl1nf/nf mice. Hence, we wondered if the remaining peripheral T-cells are of a specific subtype. Analysis of CD4 and CD8 expression in T-cells from the spleen revealed a strong accumulation of CD4- CD8- cells in nf cofilin knock-in mice (S3A Fig). We next checked splenic T-cells for TCRβ and TCRγδ surface expression. In B6 mice >95% of T-cells are of the αβ subtype and only a minor fraction of γδ T-cells are found (~ 2%) (Fig 3A). In contrast, Cfl1nf/nf mice do not harbor substantial amounts of αβ T-cells but possess normal absolute numbers of γδ T-cells (Fig 3A, bars on the right). Note that also the distinct CD4-CD8+ population of splenic T-cells isolated from Cfl1nf/nf mice expressed exclusively TCRγδ but not TCRβ (S3B Fig).
γδ T cells’ survival is not due to a missing Lck-Cre mediated knock-in of nf cofilin, as γδ T-cells do express Lck [37]. Moreover, successful recombination of the nf cofilin construct was confirmed by PCR and the expression of the cofilin protein in γδ T-cells was confirmed by flow cytometry (S3C and S3D Fig). Thus, cofilin appears to be essential only for αβ but not γδ T-cells.
As shown above, mice with a T-cell–specific knock-in of nf cofilin almost completely lacked peripheral αβ T-cells and showed a severe thymus atrophy. Residual thymocytes, which were found, accumulated in the DN stage. Thus, we next addressed why thymocytes arrested particularly at this stage of T-cell development and why expression of nf cofilin is critical for αβ but not γδ T-cell development. In regard to the actin depolymerizing function of cofilin, we first checked the cellular F-actin content of DN thymocytes as well as of thymic γδ T-cells. The F-actin content in DN thymocytes was highly increased in cells obtained from Cfl1nf/nf mice in comparison to those derived from B6 mice (Fig 3B, left bar chart). Interestingly, also thymic γδ T cells from Cfl1nf/nf mice accumulated more F-actin than control γδ T cells (Fig 3B, right bar chart).
Besides cofilin, destrin is another actin depolymerizing factor that can be expressed in mammalian cells. Flow cytometric studies revealed that destrin is expressed in DN and γδ thymocytes and its expression is not impaired in Cfl1nf/nf mice (S4A Fig). Thus, the presence of destrin could not compensate the effects of nf cofilin.
One important process during early T-cell development, which may be influenced by altered actin dynamics, is the outward migration of DN thymocytes from the entry site at the corticomedullary junction (CMJ) to the outer cortex. To determine the migratory capacity of DN thymocytes and γδ T cells that express nf cofilin, we employed a transwell assay, in which SDF-1α, the natural ligand of CXCR4, was used as chemotactic stimulus. DN thymocytes from Cfl1nf/nf mice showed both a decreased random migration (Fig 3C, left bar chart; none) and a diminished directed migration (Fig 3C, left bar chart; +SDF-1α). A similar reduction in the migratory potential was observed for thymic γδ T cells (Fig 3C, right bar chart).
Note that the decreased migratory capacity of nf cofilin–expressing cells was not due to a lack of CXCR4, which was expressed intracellularly and extracellularly to a similar extent as in control cells (S4B Fig).
A second process during T-cell development, which requires actin dynamics, is the redistribution of receptors to the cell surface, as e.g. the TCR. While normal surface expression of TCRγδ was observed in DN thymocytes of Cfl1nf/nf mice, they completely lacked TCRβ surface expression (Fig 3D, upper panel). However, TCRβ was detected inside nf cofilin expressing DN thymocytes (Fig 3D, lower panel). Although the number of icTCRβ+ DN thymocytes was decreased in Cfl1nf/nf compared to B6 mice, the mean fluorescence intensity (MFI) of TCRβ in icTCR+ cells was similar between Cfl1nf/nf and B6 mice (Fig 3E).
To test whether the surface expression of TCRβ in Cfl1nf/nf mice can be rescued by the disruption of actin filaments (e.g., cortical actin), we treated DN thymocytes with cytochalasin D (cytoD). Although TCRβ surface expression on thymocytes of B6 mice was slightly but not significantly enhanced after cytoD treatment, thymocytes of Cfl1nf/nf mice showed still no TCRβ on their surface (Fig 3F).
Our data demonstrate that DN thymocytes as well as thymic γδ T-cells from Cfl1nf/nf mice showed a strong accumulation of F-actin and a decreased migration capacity. However, only TCRβ but not TCRγδ surface expression was abolished in thymocytes of nf cofilin knock-in mice.
Since the heterogeneous γδ T-cell compartment consists of different subpopulations, the influence of nf cofilin on specific γδ T-cell populations was tested. First, the surface expression of Vγ1, Vγ2, and Vγ3 chains was analyzed on thymic or peripheral (skin, lung, spleen) γδ T cells (Fig 4A). As expected, these Vγ chains were tissue specifically expressed. However, comparing the tissue specific Vγ chain expression of γδ T-cells from B6 and Cfl1nf/nf revealed no differences at all. In line with these results, distinct γδ T-cell populations, which are characterized by the expression of different surface markers on γδ thymocytes (CD24, CD27, and CD44), were similar in Cfl1nf/nf and B6 mice (Fig 4B).
Since knock-in mice had no αβ T cells but normal numbers of γδ T cells in the periphery, we wondered whether γδ T cells are still functional. To evaluate their functionality, purified splenic γδ T cells were in vitro stimulated by anti-CD3 and anti-CD28 antibodies for 24 h. Stimulated γδ T cells of both control and Cfl1nf/nf mice showed increased expression of T-cell activation markers (CD25, CD69) compared to unstimulated cells (Fig 4C).
To further investigate the functionality of γδ T cells in Cfl1nf/nf mice under in vivo conditions, we chose the IMQ-induced psoriasis-like murine model. In this, the loss of γδ T cells was already linked to diminished psoriasis-like symptoms [38]. By applying IMQ containing Aldara crème topically on the shaved back skin of either WT (Cfl+/+) or Cfl1nf/nf knock-in mice for 6 days, the psoriasis-like phenotype was assessed (Fig 4D). After 3 days, both groups started to show signs of scaling, skin thickening, and erythema. Cfl1nf/nf mice exhibited slightly decreased erythema at day 4 and day 5 and diminished scaling and skin thickening at day 6 compared to WT mice. Also, the cumulative psoriasis area severity index (PASI) score was partially reduced in Cfl1nf/nf mice at day 5 and 6. Nevertheless, Cfl1nf/nf mice developed strong psoriasis-like symptoms over the course of the experiment and also the severity of inflammation increased up to the end of the experiment.
IL-17A–producing and RORγt-positive γδ T cells are crucial for proper development of psoriasis [39,40]. Therefore, we tested the IL-17A production and RORγt expression in skin-draining LN γδ T cells. Ex vivo restimulation of LN cells from control vehicle crème (Sham) or IMQ-treated mice revealed a slightly but nonsignificantly reduced percentage of IL-17A, producing as well as RORγt-expressing γδ T cells in Cfl1nf/nf mice compared to WT mice. Nonetheless, both groups showed significantly increased percentages of these cells after IMQ application compared to the mice treated with control crème only (Fig 4E). Together, these experiments demonstrated that γδ T cells of Cfl1nf/nf mice are functional and able to induce a psoriasis-like skin inflammation in the absence of αβ T cells.
Using knock-in mice in which T-cell-specific nf cofilin was expressed instead of endogenous cofilin, we demonstrate that cofilin is essential for αβ but not γδ T-cell development. Cfl1nf/nf mice lacked peripheral αβ T cells and showed a severe thymus atrophy, which was caused by an early developmental arrest at the DN stage. DN thymocytes exhibited impaired actin dynamics, a defective migratory capacity, and a lack of TCRβ surface expression (Fig 3). Even though γδ thymocytes were also impaired in actin dynamics and cell motility, nf cofilin knock-in mice harbored normal γδ T-cell numbers in the periphery. Those γδ T cells showed normal expression of Vγ chains and also the different subpopulations (discriminated via CD24, CD27, and CD44) were similar to those of γδ T cells from B6 mice. Thus, nf cofilin does not interfere with the development of different subsets of γδ T cells.
The functionality of peripheral γδ T cells from nf cofilin knock-in mice was proven in in vitro and in vivo experiments. First, in vitro CD3/CD28 stimulation confirmed the ability of splenic γδ T cells to be activated, although nf cofilin is expressed instead of WT cofilin. Second, we analyzed these mice via the γδ T cell–driven, IMQ-induced psoriasis-like murine model. Even though the cumulative PASI score was decreased in IMQ-treated Cfl1nf/nf mice compared to control mice, they developed clear psoriasis-like symptoms, and also, the severity of psoriasis-like skin inflammation increased within the course of the experiment. In line with this, nf cofilin knock-in mice exhibited a strong induction of IL-17A+ RORγt+ γδ T cells in skin-draining LNs of IMQ-treated animals.
In this study, we avoided knocking out cofilin completely and rather expressed a nf cofilin mutant. This was accomplished solely by the addition of a single proline to cofilin’s N-terminus by making use of the viral 2A sequence. The additional proline inhibits removal of the initiator methionine and as a consequence also N-terminal alanine acetylation. Moreover, no phosphorylation on serine 3 was detected. In the nf cofilin mutant, the lack of cofilin phosphorylation is most likely due to impaired cotranslational processing, which could render cofilin less accessible for kinases. Thereby, less phosphorylation does not automatically mean more activity. Similar findings were obtained with oxidized cofilin, which is a poor target for LIM kinase and was found to be less phosphorylated than untreated cofilin even though it is not able to remodel the actin cytoskeleton [27].
One consequence of the expression of nf cofilin was the drastically increased F-actin content of DN thymocytes. Functionally, a massive accumulation of actin filaments could cause a stiffening of the respective cells and could render them less dynamic. Indeed, early thymocytes of nf cofilin knock-in mice showed a decreased migratory capacity towards SDF-1α. Also, their spontaneous undirected migration in the absence of chemokines was diminished. In the postnatal thymus, DN1 cells are mainly found in the inner cortex (close to the CMJ) before they start an outward migration during their transition to DN2 and DN3 stage [41]. Thymocytes that are not able to migrate outward from the CMJ to the cortex due to deficiency of CXCR4, the chemokine receptor of SDF-1α, are not developing beyond the DN stage [42]. Another independent study of thymocytes derived from CXCR4-deficient progenitor cells also revealed that their development is already drastically altered before they develop into DP thymocytes [43]. Thus, the diminished migratory capacity of early thymocytes of nf cofilin knock-in mice—possibly due to stiffening of the actin cytoskeleton—may at least in part play a role for their developmental arrest. Moreover, in human PBTs, cofilin was shown to be dephosphorylated and thereby activated in lamellipodia upon triggering of cells with SDF-1α [21]. Thus, it is likely that also during thymocyte development, cofilin is one of the effector molecules, which get activated by chemokines secreted by thymic epithelial cells and are involved in the directed migration of DN cells to the outer cortex. Although normal numbers of γδ T cells were found in nf cofilin knock-in mice, γδ thymocytes also showed an accumulation of F-actin as well as a defective migratory capacity as observed for DN thymocytes. These findings imply that cofilin function and high cell motility are more crucial for αβ than for γδ T-cell development.
Besides the essential role of actin dynamics for cell movement, the dynamic rearrangement of the actin cytoskeleton is also important for clustering and (re)distribution of surface receptors, e.g., during immune synapse formation [44–46]. In our study, knock-in of a dysfunctional cofilin had detrimental effects on TCRβ but not on TCRγδ surface expression of DN thymocytes. Even though the TCRβ chain is rearranged and expressed intracellularly, it is not detectable on the the cell surface. Interestingly, the surface translocation of other proteins, e.g. CD25, CD44, or CXCR4, was not influenced, indicating a selective effect of nf cofilin on TCRβ surface translocation rather than an interference with the general surface transport of membrane proteins.
Alternatively to altered TCRβ transport to the cell surface, the diminished TCRβ expression among DN thymocytes could theoretically also be due to the lack of NKT cells, since the Lck promoter is also active in these cells [35]. However, NKT cells were still present in nf cofilin mice. Interestingly, these NK1.1+ DN thymocytes did also not express TCRβ on their surface, emphasizing the importance of cofilin for TCRβ translocation.
TCRβ surface expression is essential for pre-TCR signaling and the transition through the so-called β-checkpoint. Interestingly, thymocytes of nf cofilin knock-in mice accumulated in the DN3 stage, a phenotype which is indeed characteristic for impaired pre-TCR signaling (e.g., caused by knocking out components of the pre-TCR [47,48]). Whereas functional pre-TCR signaling and passage through the β-checkpoint induces extensive proliferation of DN thymocytes and development into DP thymocytes, those cells which are not able to express a functional pre-TCR get eradicated by apoptosis [49,50]. This proliferative burst is one of the key functions of pre-TCR signaling, and thus the dramatic thymic atrophy in nf cofilin knock-in mice is at least in part due to the lack of proliferation and/or induction of apoptosis. However, to finally conclude that nf cofilin is interfering with preTCR signaling, a comparison between TCRa-/- and Cfl1nf/nf mice would be necessary.
In the thymus, Tcrb, Tcrg, and Tcrd are all rearranged at the DN2/3 stage of development. It is at the DN3 stage in which final fate determination of αβ and γδ lineages takes place. If cells have rearranged the TCRβ chain, the β-selection process starts. In contrast to the αβ lineage cells, those DN3 cells that have rearranged functional γ and δ chains undergo γδ selection remain negative for both T-cell co-receptors and develop into γδ T cells (for details about the αβ versus γδ lineage decision, see the publication of Zarin and colleagues [51]). As αβ and γδ T cells undergo different developmental processes, they may also have varying requirements, e.g., in regard to up-regulation of receptors. This study shows that cofilin-driven cellular processes are essential for cell surface expression of TCRβ but appear to be less important for γδ TCR up-regulation. One possible explanation for this could be that other actin-remodeling proteins can partially compensate for the lack of cofilin function in γδ T cells. We investigated destrin, another closely related actin-depolymerizing factor. It was equally expressed in DN and γδ thymocytes of Cfl1nf/nf mice. This shows first that destrin could not compensate the effects of nf cofilin (massive increase of the F-actin amount in nf cofilin knock-in cells) and second that normal developmental and function of γδ thymocytes was not due to a higher expression of destrin. So far, we have no information about the expression and function of other actin-depolymerizing proteins. Furthermore, we did not find any information in the literature about the role of actin remodeling proteins in γδ T-cells. Additionally, there is no other study—at least to our knowledge—reporting about a differential role of an actin remodeling protein for αβ versus γδ T-cell development.
Previous studies in which cytoskeletal proteins were targeted in mice revealed that they are of major importance for the emigration of mature SP thymocytes from the thymus to secondary lymphoid organs. However, these proteins play only a minor role for early thymocyte development. This holds true for mice deficient in mDia (actin-nucleating-polymerizing protein) [52] as well as for L-plastin (actin bundling protein) [53] and Coronin1A (inhibits nucleation-promoting Arp2/3 complex) [54,55]. Closest to the phenotype observed for nf cofilin knock-in mice—albeit being more modest—was the phenotype of mice with a knock-out of WASP (Wiskott–Aldrich Syndrome protein). These mice exhibited a reduction of thymic cellularity and a relative increase in DN3 cells among the DN cell compartment [56]. However, in another study in which WASP was targeted, there was no effect on thymocyte development [57]. A study of Zhang and colleagues, in which T cells expressed WASP with a deleted VCA domain on the WASP knock-out background confirmed the importance of WASP for T-cell development [58]. DN cells from those mice do express pre-TCRα and TCRβ. However, in contrast to nf cofilin knock-in mice, they develop DP cells. Yet DP cells also do not express TCRβ on the cell surface and show a surface phenotype resembling the one of immature thymocytes from the DN population. Thus, even though WASP seems to be important for pre-TCR signaling and thymocyte development, it most likely plays only a partial role in this process, as there are still thymocytes which develop beyond the DN stage in WASP knock-out mice, and also, mature T-cells are present in their periphery.
Together, our data demonstrate the unique role of cofilin activity for proper development of αβ but not γδ T cells. Probably, cofilin and related signaling cascades are valuable starting points to decipher differences in developmental checkpoints for αβ versus γδ T-cell lineage decision. Besides this, usage of the Cre/lox system also allows us to knock-in the functionally inactive cofilin in other cell types. Our strategy not only allows the expression of a nf form of cofilin but also the coexpression of eGFP, which further enables to monitor knock-in cells and cofilin promoter activity. This makes the generated mice to a valuable tool for studying the relevance of cofilin in different cell types.
All mouse experiments were carried out in accordance with the relevant guidelines and regulations by the federal state Baden-Wuerttemberg and Rhineland-Palatinate, Germany. Psoriasis experiments were approved by Landesuntersuchungsamt Rheinland-Pfalz (TVA # G13-1-099). To dissect lymphoid organs (e.g., LNs, spleen, or thymus), mice were euthanized by cervical dislocation or lethal dose of CO2.
Strain details as well as procedure to generate nf cofilin knock-in mice are provided in S1 Text. All mice were bred and maintained at the central animal facility of the University of Heidelberg under specific pathogen-free conditions. Mice used in experiments were sex- and age-matched and were generally 6–12 weeks (or, for thymic experiments, 4–5 weeks) old.
For knock-down of endogenous cofilin in Jurkat cells and expression of eGFP-2A-Cfl1, Jurkat cells were transfected with the Bio-Rad GenePulser II. To this end, each 10 Mio of cells were mixed with 2 μg cofilin siRNA (CFL1 ON-TARGETplus siRNA; Dharmacon) and/or 15 μg plasmid DNA and electroporated at 230 V and 950 mF. Cells were cultured in RPMI1640 medium containing 10% FCS at 37 °C and 5% CO2. Transfection efficacy as well as successful down-modulation of endogenous cofilin was examined by western blot (see S1 Text).
For determination of the cellular F-actin content of Jurkat cells, 1 Mio of cells were fixed with 1.5% PFA in PBS for 10 min at 37 °C. Afterwards, cells were permeabilized in FACS buffer (PBS with 0.5% BSA) containing 0.1% saponine for 10 min at RT. Cells were stained in the same buffer containing Phalloidin-AF647 (Life technologies) for 20 min at RT. For determination of the F-actin content of DN thymocytes or γδ thymocytes, cells were stained with 500 nM SiR-actin (Cytoskeleton, Inc.) for 3 h at 37 °C. The higher the MFI of Phalloidin-AF647 or SiR-actin, the more filamentous actin is present inside the cell.
Chemotaxis of thymocytes was tested with 5-μm pore size Transwell plates (Corning). To this end, 50,000 cells in medium were plated in the upper compartment of the transwell insert and medium +/− 200 ng/ml SDF-1α (Peprotech) was added into the lower compartment. Migration was carried out for 3 h at 37 °C. The number of transmigrated thymocytes was determined via flow cytometry by making use of an internal bead standard (BD Biosciences).
To disrupt the cortical actin of thymocytes, 2 Mio cells were treated with 20 μM cytochalasin D (Sigma Aldrich) for 1 h at 37 °C. After cytoD treatment, cells were stained for TCRβ and surface expression was analyzed by flow cytometry.
Splenic γδ T cells were MACS isolated using “TCRg/d T-cell isolation kit, mouse” (Miltenyi Biotec). Cells numbering 300,000 were stimulated by plate-bound αnti-CD3 (10 μg/mL, BD Bioscience) and anti-CD28 antibodies (2 μg/mL, BD Bioscience) for 24 h at 37 °C. After stimulation, cells were stained for surface markers and analyzed by flow cytometry.
Age- and sex-matched Cfl1nf/nf and Cfl1+/+ mice at 7 weeks of age were used for IMQ-induced psoriasis-like model.
Mice received a daily topical dose of 50 mg of IMQ crème (5%) (Aldara, Meda Pharma) or 50 mg of a control vehicle crème (Sham crème, University medicine Mainz) over 6 days. Prior to topical application scores of individual parameters such as back skin thickness, scaling, and erythema formation were measured and the accumulated PASI was calculated as described previously [59]. At day 6, mice were killed humanely and LN cells were isolated for intracellular cytokine staining.
To induce cytokine production, single cell suspensions of isolated LNs were stimulated with 50 ng/ml phorbol 12-myristate 13-acetate (PMA, Sigma Aldrich) and 500 ng/ml ionomycin (Sigma Aldrich) in the presence of 1 μg/ml Brefeldin A for 4 h at 37 °C. After stimulation, cells were stained for surface markers and intracellular cytokines and analyzed by flow cytometry.
Statistical analysis was performed with Prism 6 software. Values are expressed as mean ± SEM. Unpaired two-tailed Student t test was used to test significant numerical differences between groups. Differences of p ≤ 0.05 were considered to be statistically significant (* p ≤ 0.05; ** p ≤ 0.01; *** p ≤ 0.001; **** p < 0.0001).
|
10.1371/journal.pgen.1006678 | Rare variants in fox-1 homolog A (RBFOX1) are associated with lower blood pressure | Many large genome-wide association studies (GWAS) have identified common blood pressure (BP) variants. However, most of the identified BP variants do not overlap with the linkage evidence observed from family studies. We thus hypothesize that multiple rare variants contribute to the observed linkage evidence. We performed linkage analysis using 517 individuals in 130 European families from the Cleveland Family Study (CFS) who have been genotyped on the Illumina OmniExpress Exome array. The largest linkage peak was observed on chromosome 16p13 (MLOD = 2.81) for systolic blood pressure (SBP). Follow-up conditional linkage and association analyses in the linkage region identified multiple rare, coding variants in RBFOX1 associated with reduced SBP. In a 17-member CFS family, carriers of the missense variant rs149974858 are normotensive despite being obese (average BMI = 60 kg/m2). Gene-based association test of rare variants using SKAT-O showed significant association with SBP (p-value = 0.00403) and DBP (p-value = 0.0258) in the CFS participants and the association was replicated in large independent replication studies (N = 57,234, p-value = 0.013 for SBP, 0.0023 for PP). RBFOX1 is expressed in brain tissues, the atrial appendage and left ventricle in the heart, and in skeletal muscle tissues, organs/tissues which are potentially related to blood pressure. Our study showed that associations of rare variants could be efficiently detected using family information.
| Hypertension is a risk factor for cardiovascular disease and the most important risk factor for stroke. Family studies suggest that hypertension related traits are heritable. Previous genome-wide association studies (GWAS) have identified multiple common blood pressure (BP) variants but these variants do not overlap with the linkage evidence observed from family studies. Rare variants have been suggested to play a substantial role and contribute to missing heritability of BP. In this study, linkage analysis identified 16p13 linked to SBP in a cohort of 517 white individuals in 130 families from the Cleveland Family Study (CFS). By combining linkage and association analyses, we searched for rare, coding variants that can explain the linkage evidence. Rare, coding variants within RBFOX1 were associated with lower systolic (p-value = 0.00403) and diastolic (p-value = 0.0258) blood pressures, and explained significant amount of observed linkage evidence. We replicated the identified variants in four independent cohorts (with total sample size N = 57, 234) and further observed consistent evidence that rare RBFOX1 variants are protectively associated with blood pressure traits. Our study clearly shows that family-based designs are powerful for identifying rare, coding variants underlying complex diseases.
| High blood pressure (BP) is a common condition associated with multiple health outcomes, including heart, brain, and kidney diseases [1, 2]. Previous studies have shown that BP is a genetically determined trait with estimated heritability of 30% to 60% [3, 4]. Multiple large genome-wide association studies (GWAS) meta-analysis and admixture mapping studies have identified over 190 genetic variants that explained only a small variation in BP [5–21].
For complex traits such as BP, rare variants are suggested to play a greater role in heritability than anticipated in the common disease-common variant hypothesis [22]. A Framingham Heart Study reported rare mutations in three renal salt handling genes causing large reductions in blood pressure and estimated that the overall prevalence of hypertension is reduced by about 1% because of the effects [23]. Linkage studies of family data can be used to uncover missing heritability and identify genetic markers linked to BP [24, 25]. However, the identified linkage regions from well-designed linkage studies such as the US Family Blood Pressure Program (FBPP) and the UK Medical Research Council British Genetics of Hypertension (BRIGHT) study [26–29] do not overlap with many BP loci identified by large BP GWAS of mostly unrelated individuals. In general, GWAS have good power to detect common variants of modest effect with attainable sample sizes, but less power for detecting rare variants with intermediate effect. In contrast, linkage analysis can have good power to detect multiple rare or lower frequency BP variants in a gene or region with relatively larger effect sizes [25]. Thus, we hypothesize that a linkage region observed in a family study, if not overlapping with the BP loci in reported GWAS, may harbor multiple rare or lower frequency BP variants.
Recently, many statistical approaches for rare variant association analyses have been developed for unrelated samples [30–34] and family data [35–38]. It has been suggested that rare or lower frequency variants can be enriched in families [35, 37], and therefore improving the statistical power for their identification. However, the existing rare variant association methods have not incorporated linkage evidence. In this study, we performed variance-component linkage analysis with BP traits, including systolic blood pressure (SBP), diastolic blood pressure (DBP), and pulse pressure (PP) in the Cleveland Family Study (CFS). We searched the published GWAS to examine whether there are reported BP variants in the linkage regions. Using the combined linkage and association analyses, we searched for potential functional variants that can explain linkage evidence and replicated the variants in independent cohorts.
Table 1 presents the characteristics of white participants in the CFS data. S1 Fig presents the distributions of the residuals of SBP, DBP, and PP, which are all approximately normally distributed. Linkage analysis identified a peak (LOD = 2.81) on chromosome 16p13 linked to SBP (S2 Fig, Materials and Methods). Linkage analysis of further pruned markers using a R2 threshold of 0.1 or modeling marker-marker linkage disequilibrium resulted in a slight decrease of LOD score in the same region (LOD = 2.30–2.42, S3 Fig). We selected a candidate region of 20cM with 2-LOD score drop for association analysis (Fig 1). This region did not overlap with published GWAS of BP variants. Therefore, we tested the hypothesis that the observed linkage evidence for SBP is due to the presence of multiple rare, coding variants in a gene(s) within the region.
The CFS was genotyped by an exome array, with most of the variants being coding variants. Within the linkage region, there are 454 exonic variants defined by the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium that are genotyped on the exome array [39]. We identified 13 exonic variants (S1 Table) that satisfy the following filtering criteria: 1) either have a SBP association p-value ≤ 0.1 or absolute regression coefficient beta ≥ 5; and 2) present at least twice in at least one family with a family-specific LOD score ≥ 0.1. A risk score based on the 13 identified SNPs has an effect size of 0.948 ± 0.135 for association with SBP in CFS. After adding this risk score as a covariate in linkage analysis, the MLOD score dropped from 2.809 to 1.055, suggesting that these 13 variants were able to account for most of the observed linkage evidence. To further assess the significance of this LOD drop, we sampled 1,000 independent SNPs from chromosomes other than chromosome 16. Hence, these SNPs should not contribute to the LOD score observed on chromosome 16. We performed linkage analysis with each of these 1,000 SNPs as a covariate in the linkage analysis. We calculated the differences between the original MLOD score and the MLOD scores of the 1,000 linkage analyses with a SNP as a covariate. The largest LOD score drop in these 1,000 linkage analyses was 0.347, suggesting the observed LOD score drop on the risk scores of 13 selected variants is statistically significant (p-value<0.001). Among these 13 exonic variants, two variants (rs149974858 and rs145873257) are present in RBFOX1 and the remaining 11 variants are each in separate genes. When adjusting for the risk scores of rs149974858 and rs145873257, we also observed a drop in LOD score (LOD = 1.97), which suggests that rs149974858 and rs145873257 account for a portion of the observed linkage evidence.
The variant rs149974858 shows association evidence with SBP (p-value = 0.0016) in CFS. The minor allele frequency of rs149974858 is 0.0036 and only segregates within a 17-member family with family-specific LOD of 0.697 (Fig 2). This missense variant (c.112C>G) results in a proline to alanine substitution (p.Pro38Ala). Five members from this family carrying this rare missense mutation had on average lower SBP (carrier average = 117 mmHg, noncarrier average = 125 mmHg) but higher BMI than other family members (carrier average = 60 kg/m2, noncarrier average = 31 kg/m2). A single SNP association test revealed that rs149974858 is also significantly associated with BMI in CFS (beta = -26.8±4.35, asymptotic p-value = 7.28E-10). The exonic variant rs145873257 segregates within a different family with family-specific LOD score of 0.215. A c.1072G>A base change resulted in a glycine to serine substitution (p.Gly358Ser). Six members of this family are heterozygous for the AT genotype. The estimated effect of this variant is protective in CFS, although it is not statistically significant (S1 Table).
Since both variants rs149974858 and rs148751394 consistently show a protective effect in two large families, we examined the other coding and rare variants in RBFOX1 genotyped on the exome array. Two exonic variants, rs151214012 and rs145873257, and one rare, intronic variant rs2345080 are available in the exome array. These three variants show protective effect despite not satisfying the filtering criteria (S2 Table). Single SNP associations and annotations for all exome array variants of RBFOX1 are provided in the S3 Table. Applying either the family-based burden or SKAT analysis, these five variants are significantly associated with SBP and DBP (Table 2).
We next sought the replication of the rare variants in RBFOX1 in four large independent cohorts (ARIC, WHI, BioUV, and HRS) of whites for the traits SBP, DBP and PP. We specifically looked at the five rare variants of RBFOX1 found in CFS and their associations with BP traits. Within the ARIC data, 4 of the 5 variants were present and all the 4 variants showed a consistent protective effect. For WHI, all 5 variants were present but only 1 variant (rs145873257) had a protective effect size. BioUV contained 4 out of 5 variants found in CFS; 2 variants were protective for SBP and 3 variants were protective for DBP. HRS contained all 5 variants found in CFS; 4 variants were protective for SBP and 3 variants were protective for DBP. Among the total 23 tests (CFS: 5, ARIC: 4, BioUV: 4, WHI: 5, HRS: 5) conducted across all cohorts, 16 of them were protective (p-value = 0.0173 based on binomial distribution), suggesting a consistent protective effect.
We next conducted a gene-based association analyses for RBFOX1 and BP traits using exome array data from the CFS with weights Beta (1, 25). Burden, SKAT, and SKAT-O tests were performed using the 5 rare variants of RBFOX1 found in CFS (Table 2). In CFS, the association between RBFOX1 and SBP was found to be significant by the burden test (p-value = 0.00214) and SKAT-O (p-value = 0.00403), but not by SKAT (p-value = 0.0702). All three gene-based tests for DBP were statistically significant (burden test p-value = 0.0148, SKAT p-value = 0.0494, SKAT-O p-value = 0.0258). When we conducted gene-based analyses using only the 4 coding variants (rs149974858, rs148751394, rs151214012, rs145873257) identified in CFS, the associations are also significant for SBP and DBP (p-value<0.037).
The same gene-based analysis for rare variants was done for all four replication cohorts separately and the results were meta-analyzed (Table 2). Individually, the ARIC cohort had significant gene-based associations for SBP (burden test p-value = 0.00572, SKAT p-value = 0.00259, SKAT-O p-value = 0.00356) and PP (burden test p-value = 0.00140, SKAT p-value = 0.000273, SKAT-O p-value = 0.000392). After meta-analyzing the results for ARIC, WHI, BioVU, and HRS, we found significant associations for SBP (burden test p-value = 0.0172, SKAT p-value = 0.00635, SKAT-O p-value = 0.0126) and PP (burden test p-value = 0.00377, SKAT p-value = 0.00266, SKAT-O p-value = 0.00234).
We observed that the variant rs149974858 co-segregated with BMI in the 17-member CFS family. Subsequently, we performed linkage analysis for BMI in CFS on chromosome 16, after adjusting for gender, age, age2, and PC1. We did not observe linkage evidence in this region (LOD = 0.721). The gene-based analysis of BMI using the same set of variants was only significant in CFS but not in any of the replication cohorts (Table 2).
We performed a linkage analysis of BP traits using the families collected in CFS. The largest linkage peak identified is on 16p13 linked to SBP. The 16p13 region has been reported of linkage evidence with BP in two studies: the Victorian Family Heart Study [40], the extreme-sib-pair study in Chinese adults [41]. In addition, the longitudinal change of BP in Mexican Americans [42], and the Hypertension Genetic Epidemiology Network Study in whites [43] reported linkage regions partially overlapped with the current study. The reported linkage evidence from multiple ethnic populations strongly suggests the linkage evidence on 16p13 is real. Linkage analysis using microsatellite markers was performed with 363 sib-pairs of CFS whites for a hypertensive status, adjusted for age, age2, sex, BMI, and BMI2. This analysis did not find linkage evidence in the 16p13 region. This is unsurprising because the power of using a binary hypertensive status is lower than that of quantitative phenotypes, such as SBP, DBP, or PP. In addition, hypertensive status was defined as either SBP ≥ 140, DBP ≥ 90, or taking antihypertensive medications and the sample size in the sib-pair analysis was smaller than the current study, all of which contribute to the lack of linkage evidence observed. Our study demonstrates that high-density SNP genotyping arrays are informative for detecting linkage signals.
We searched among published large GWAS studies of BP traits [5, 7–10] and did not identify any BP variants previously reported on 16p13, suggesting that multiple low frequency or rare variants with relatively large effect sizes are possibly contributing to the observed linkage evidence. We further assumed that variants with relatively large effect sizes are more likely to be coding and rare variants. Thus, we limited our search to only the coding and rare variants under the linkage peak genotyped on the exome array. By examining the associated variants that are able to account for the observed linkage evidence on 16p13, we were able to identify 13 exonic variants. Among these 13 variants, 2 of them fall in RBFOX1, which encodes for the ataxin-2 binding protein 1 (also known as A2BP1), and show a consistent protective effect for SBP in CFS. Gene-based analysis of the four available exonic variants and one rare intronic variant in RBFOX1 are significantly associated with SBP and DBP (p-value = 0.00403, 0.0258, respectively) using SKAT-O. Replication analysis of the rare variants at the gene level (but not at the variant level) is also significant for SBP and PP in the meta-analysis of four large cohorts of whites with a total replication sample size N = 57,234.
This study also provides evidence that rare variants within RBFOX1 are protective for BP traits among obese individuals. Among individuals of European ancestry within the CFS, 5 individuals within the same family carried the minor allele for rs149974858, the variant showing significantly protective effect with SBP by single SNP association test. All of the 5 individuals are morbidly obese (with average BMI of 60). However, their SBP (mean = 117) and DBP (mean = 78) are within the normotensive range. Ma et al. conducted a GWAS of BMI in Pima Indians using Affymetrix 100K array and identified two common variants in RBFOX1 associated with BMI [44]. The same two variants could be replicated in non-overlapped Pima Indians but not in French Caucasians, Amish Caucasians, German Caucasians, or Native Americans [44]. In our analysis, we identified four exonic and one intronic rare variants in RBFOX1 that are significantly associated with BMI but the association evidence could not be replicated (Table 2). Therefore, it is inconclusive whether RBFOX1 is an obesity gene. In all our analysis, either linkage or association analysis, BMI is included as a covariate. Furthermore, no linkage evidence was found for BMI on chromosome 16, after adjusting for gender, age, age2, and PC1. Our result indicates the RBFOX1 contributes to BP variation independent of obesity, although we are unclear whether RBFOX1 has a pleiotropic effect on both BP and obesity.
We also observed that the effect direction of single variant replication analysis in the four cohorts is not always consistent with that of CFS. However, 16 of the 23 tests were protective (p-value = 0.017), suggesting a consistent protective effect. Assuming a causal rare variant with an effect size equal to one quarter of the BP standard deviation, we estimate the probability of observing an opposite direction in a study to be 40.1%, which is consistent with 7 opposite directions among 18 single SNP replication tests in 4 replication cohorts.
It is interesting that all the four exonic variants in RBFOX1 are either monomorphic or extremely rare in African ancestry populations (S4 Table). Furthermore, the BP admixture mapping analysis by Zhu et al. also suggest local ancestry in this region is associated with SBP and DBP; however, the evidence is not genome-wide significant [11]. Thus, our result suggests that the rare exonic variants in RBFOX1 may contribute to a protective effect for hypertension and further work will be needed to establish whether the lack of these protective variants contribute to the disparity in hypertension occurrence and early age of onset between African Americans and whites.
To our knowledge, only one GWAS study so far has reported on the association of RBFOX1 variants with blood pressure using human genotyping data. Wang et al. reported an association between rs1507023, a candidate SNP in RBFOX1 involved in vitamin D metabolism and signaling, and SBP, DBP, and PP. Its association with blood pressure was significant before, but not after correction for multiple testing [45].
Under the linkage region of 16p13, there are 11 additional variants that either have an association test p-value less than 0.1 or an effect size larger than 5 mmHg in CFS. When we used the risk scores of these 11 variants as a covariate in linkage analysis, the MLOD dropped to 1.932, suggesting that there may be additional variants that contribute to linkage evidence in this region. However, the current exome array data is limited for further dissection of genes or variants contributing the linkage evidence. Whole genome sequencing data, including the sequencing data from the Trans-Omics for Precision Medicine (TOPMed) Program (https://www.nhlbi.nih.gov/research/resources/nhlbi-precision-medicine-initiative/topmed), will be necessary to identify the genes contributing blood pressure variation in this region.
Gene expression data and previous studies have demonstrated that RNA splicing factor RBFOX1 is important for heart and skeletal muscle development and function [46–48]. RBFOX1 expression has been associated with cardiac hypertrophy and heart failure in mice models [49]. Gao et al. found that RBFOX1 expression was significantly diminished in both mouse and human failing hearts [49]. We searched the GTEx database and RBFOX1 is highly expressed in multiple human brain tissues, atrial appendage and left ventricle of the heart, as well as muscle skeletal tissues (Fig 3; http://www.gtexportal.org/home/gene/RBFOX1). Further biological studies are needed to establish the direct role of RBFOX1 in regulating blood pressure.
Our study suggests that family-based linkage evidence can be extremely successful in searching for rare variants contributing to complex traits. In summary, we identified rare, exonic variants in RBFOX1 that have a protective effect on BP traits, which can be important in searching new drugs for cardiovascular disease. However, it should be pointed out that association analysis was performed using variants available in the exome array of this study. The variants identified in RBFOX1 may still be reflecting in LD with the causal variants to BP. While RBFOX1 is expressed in multiple tissues that may relate to blood pressure, the mechanism underlying how this gene contributes to BP variation needs to be further studied. The identification of these rare coding variants will facilitate precision medicine in treating cardiovascular disease.
The CFS is a family-based longitudinal study comprised of patients with laboratory-diagnosed sleep apnea, their family members, and neighborhood control families, as described before [50]. The data were analyzed anonymously at Case Western Reserve University. The CFS study was approved by Partners Human Research Committee with the proposal number 2011D001860. The analytical sample includes 517 white participants in 130 families who were genotyped with the Illumina OmniExpress Exome array, which includes both GWAS and exome variants (Table 1). Standard quality controls were performed, including checking Hardy-Weinberg equilibrium, Mendelian inconsistences, and verifying pedigree structure using the genetic markers by the software PLINK [51]. BP traits, including SBP and DBP were each determined following standardized guidelines using a calibrated sphygmomanometer [52]. Height and weight were directly measured and antihypertensive medications were ascertained by questionnaire. Data for this analysis were from the last available examination for each participant.
The samples used for replication analysis include five independent cohorts. We included 10,864 unrelated white participants from the Atherosclerosis Risk in Communities (ARIC) Study. The ARIC study is a prospective epidemiologic study designed to investigate the natural history and etiology of atherosclerosis (https://www2.cscc.unc.edu/aric/). There were 18,050 unrelated white participants from the Women’s Health Initiative (WHI), a study of postmenopausal women focused on strategies for preventing heart disease, breast and colorectal cancer, and osteoporotic fractures [53, 54]. From the Vanderbilt University Biobank (BioVU), we included 18,977 unrelated white individuals. BioVU uses leftover blood samples collected during routine clinical care from consented individuals who visit the Vanderbilt University Medical Center [55]. Lastly, we included 9,343 unrelated white participants from the Health and Retirement Study (HRS). This is a longitudinal survey of a representative sample of Americans over the age of 50 [56–58].
SBP and DBP for an individual taking antihypertensive medication were imputed using the standard approach in literature, by adding 15 mmHg and 10 mmHg, respectively. Pulse pressure (PP) was calculated as the difference between imputed SBP and DBP [5].
We calculated residuals of SBP, DBP, and PP after adjusting for gender, age, age2, body mass index (BMI; kg/m2) and the first principal component of genotype values in CFS separately. The residuals of these regressions were used for linkage analysis using the software MERLIN [59]. The principal components (PCs) were calculated using the software FamCC, which can be applied to family data [60]. Since the results were essentially the same for including the first PC or the first 10 PCs, we reported the linkage results including the first PC. We used the pairwise linkage disequilibrium (LD) pruning approach with a window size of 50 kb, step size of 5 variants, and R2 threshold of 0.2. We also required a minor allele frequency (MAF) ≥ 0.2. This resulted in 56,992 autosomal SNPs using PLINK [51]. Because marker-marker LD may result in biased linkage calculations, we performed linkage analysis by further reducing the R2 threshold to 0.1 and by modeling the marker-marker LD using MERLIN [59]. Linkage analysis using MERLIN decomposes phenotypic variance into three parts: the variance contributes to the quantitative trait locus (σQTL2), the variance contributes to the polygenetic effect (σG2), and the variance contributes to the random effect (σE2). It also tests the null hypothesis of no linkage H0:σQTL2=0 vs. HA:σQTL2>0.
We examined exonic variants genotyped in the exome array in the region of 2-LOD score drop from the linkage peak. We performed the family-based association analysis for the exonic variants only in the 2-LOD drop region using the ASSOC package in S.A.G.E [61]. The family-based association analysis was conducted using a linear mixed model y = β0 + β1g + δ + ε, where g is a genotype value vector, β0 is the intercept, β1 is the regression coefficient, δ ~ N(0, 2ΦσG2) where Φ is the kinship coefficient matrix and σG2 is the polygenic variance, and ε ~ N(0, Iσε2) where σε2 is the random error. ASSOC applies the likelihood ratio test to test the null hypothesis of H0: β1 = 0. For each of the variants, we first performed an association analysis with a BP trait using ASSOC and identified variants with either p-value ≤ 0.1 (marginal effect) or absolute regression coefficient beta ≥ 5 (large effect). We next estimated family-specific LOD scores and identified families with LOD score ≥ 0.1. We kept the variants with association p-value ≤ 0.1 or absolute regression coefficient ≥ 5, and that were present at least twice in at least one family with family-specific LOD score ≥ 0.1.
We defined the risk score as ri=xiTβ, where β is the regression coefficients of the SNPs, and xi is a vector of the number of risk alleles carried by individual i for these SNPs. Linkage analysis was further performed conditional on the risk scores.
We performed family-based burden and SKAT tests for CFS using the software famSKAT and for the replication cohorts using the R package SKAT [30, 32, 62]. The weight was set to wj=Beta(MAFj,1,25) as suggested to increase the weight of rare variants.
|
10.1371/journal.ppat.0030002 | Effective Post-Exposure Treatment of Ebola Infection | Ebola viruses are highly lethal human pathogens that have received considerable attention in recent years due to an increasing re-emergence in Central Africa and a potential for use as a biological weapon. There is no vaccine or treatment licensed for human use. In the past, however, important advances have been made in developing preventive vaccines that are protective in animal models. In this regard, we showed that a single injection of a live-attenuated recombinant vesicular stomatitis virus vector expressing the Ebola virus glycoprotein completely protected rodents and nonhuman primates from lethal Ebola challenge. In contrast, progress in developing therapeutic interventions against Ebola virus infections has been much slower and there is clearly an urgent need to develop effective post-exposure strategies to respond to future outbreaks and acts of bioterrorism, as well as to treat laboratory exposures. Here we tested the efficacy of the vesicular stomatitis virus-based Ebola vaccine vector in post-exposure treatment in three relevant animal models. In the guinea pig and mouse models it was possible to protect 50% and 100% of the animals, respectively, following treatment as late as 24 h after lethal challenge. More important, four out of eight rhesus macaques were protected if treated 20 to 30 min following an otherwise uniformly lethal infection. Currently, this approach provides the most effective post-exposure treatment strategy for Ebola infections and is particularly suited for use in accidentally exposed individuals and in the control of secondary transmission during naturally occurring outbreaks or deliberate release.
| Being highly pathogenic for humans and monkeys and the subject of former weapons programs makes Ebola virus one of the most feared pathogens worldwide today. Due to a lack of licensed pre- and post-exposure intervention, our current response depends on rapid diagnostics, proper isolation procedures, and supportive care of case patients. Consequently, the development of more specific countermeasures is of high priority for the preparedness of many nations. In this study, we investigated an attenuated vesicular stomatitis virus expressing the Ebola virus surface glycoprotein, which had previously demonstrated convincing efficacy as a vaccine against Ebola infections in rodents and monkeys, for its potential use in the treatment of an Ebola virus infection. Surprisingly, treatment of guinea pigs and mice as late as 24 h after lethal Ebola virus infection resulted in 50% and 100% survival, respectively. More important, 50% of rhesus macaques (4/8) were protected if treated 20 to 30 min after Ebola virus infection. Currently, this approach provides the most effective treatment strategy for Ebola infections and seems particularly suited for the use in accidental exposures and the control of human-to-human transmission during outbreaks.
| Editor's Note: The potential efficacy of pre- and post-exposure prophylaxis against Ebola virus infection, as well as the fundamentally important question of whether neutralizing antibodies are important for Ebola virus resistance, is addressed by a related manuscript in this issue of PLoS Pathogens. Please see doi:10.1371/journal.ppat.0030009 by Oswald et al.
Infection with the filoviruses, in particular Zaire ebolavirus (ZEBOV), Sudan ebolavirus, or Marburg virus (MARV), causes a severe haemorrhagic fever (HF) in humans and nonhuman primates that is often fatal [1–3]. In addition to the sporadic outbreaks that have occurred in humans in Central Africa since 1976 and caused more than 1,800 human infections with a lethality rate ranging from 53% to 90%, Ebola virus (EBOV) has also decimated populations of wild apes in this same region [4]. At this time, there is no preventive vaccine or post-exposure treatment option available for human use.
Much remains to be learned about these highly virulent viruses; however, important advances have been made over the last decade in understanding how filoviruses cause disease and in developing preventive vaccines that are protective in nonhuman primates [1,5]. For example, a recombinant replication-defective adenovirus vaccine completely protected nonhuman primates from uniformly lethal ZEBOV infection [6,7]. More recently, we generated live-attenuated recombinant vesicular stomatitis viruses (VSV) expressing the transmembrane glycoproteins (GP) of ZEBOV (VSVΔG/ZEBOVGP) and MARV (VSVΔG/MARVGP) and the glycoprotein precursor of Lassa virus (VSVΔG/LASVGPC) [8] and showed that these completely protected cynomolgus macaques against lethal challenge with the corresponding filoviruses and arenavirus [9,10]. Progress in developing therapeutic interventions against the filoviruses has been much slower [5]. Limited success was achieved in using an anticoagulant to treat EBOV infections [11], and very recently the VSV-based MARV vaccine platform (VSVΔG/MARVGP) demonstrated astonishing efficacy in post-exposure treatment of MARV-infected macaques [12]. Other than that, no post-exposure modality has been able to protect nonhuman primates against lethal filovirus infections [5,13,14].
There is clearly an urgent need to develop filovirus-specific effective post-exposure strategies to respond to future outbreaks in Central Africa, to counter acts of bioterrorism, and to treat laboratory exposures such as the recent EBOV exposures that occurred in the United States and Russian laboratories [15,16]. Post-exposure vaccine treatment is successful in preventing or modifying viral diseases such as rabies [17,18], hepatitis B [19], and smallpox [20,21] in humans, as well as MARV HF in nonhuman primates [12]. However, the faster disease course and higher lethality of ZEBOV in human and nonhuman primates may limit the success of a similar approach for EBOV HF. Here, we show remarkable efficacy of the VSV-based EBOV vaccine platform in the post-exposure treatment of rodents and nonhuman primates infected with ZEBOV. Currently, this is the most promising post-exposure treatment strategy for EBOV HF and is particularly suited for use in accidentally exposed individuals and in the control of transmission in the event of natural or deliberate outbreaks.
The recombinant VSV expressing the GPs of ZEBOV (strain Mayinga), MARV (strain Musoke), or Lassa virus (strain Josiah) were generated as described recently using the infectious clone for the VSV, Indiana serotype (kindly provided by J. Rose) [8]. Briefly, the appropriate open reading frames for the GPs (ZEBOV, Mayinga, MARV, Musoke) were generated by PCR, cloned into the VSV genomic vectors lacking the VSV G gene, sequenced, and originally rescued using the method described earlier [8,22]. ZEBOV (strain Kikwit) was isolated from a patient of the EBOV outbreak in Kikwit in 1995 [23]. The mouse- and guinea pig-adapted ZEBOV strains (MA-ZEBOV and GA-ZEBOV, respectively) were generated by serial passages in the different rodent species until uniformly lethal [24,25].
Total white blood cell counts, lymphocyte counts, red blood cell counts, platelet counts, haematocrit values, total haemoglobin, mean cell volume, mean corpuscular volume, and mean corpuscular haemoglobin concentration were determined from nonhuman primate blood samples collected in tubes containing EDTA, by using a laser-based haematology analyzer (Beckman Coulter, http://www.beckmancoulter.com). The white blood cell differentials were performed manually on Wright-stained blood smears.
RNA was isolated from nonhuman primate whole blood and swabs using appropriate RNA isolation kits (Qiagen, http://www1.qiagen.com). ZEBOV RNA was detected using primer pairs targeting the L genes [ZEBOV: RT-PCR, nt position 13344–13622; nested PCR, nt position 13397–13590]. The sensitivity of the ZEBOV-specific RT-PCR is approximately 0.1 pfu/ml. ZEBOV titration was performed by plaque assay on Vero E6 cells from all blood and selected organ (adrenal, ovary, lymph nodes, liver, spleen, pancreas, lung, heart, brain) and swab samples [23]. Briefly, increasing 10-fold dilutions of the samples were adsorbed to Vero E6 monolayers in duplicate wells (0.2 ml per well); thus, the limit for detection was 25 pfu/ml.
IgG and IgM antibodies against ZEBOV were detected with an enzyme-linked immunosorbent assay (ELISA) using purified virus particles as an antigen source [6]. Neutralization assays were performed by measuring plaque reduction in a constant virus:serum dilution format as previously described [9,26]. Briefly, a standard amount of ZEBOV (∼100 pfu) was incubated with serial 2-fold dilutions of the serum sample for 60 min. The mixture was used to inoculate Vero E6 cells for 60 min. Cells were overlayed with an agar medium, incubated for 8 d, and plaques were counted 48 h after neutral red staining. End point titres were determined by the dilution of serum, which neutralized 50% of the plaques (PRNT50).
Peripheral blood mononuclear cells were isolated from rhesus macaque whole blood samples by separation over a Ficoll gradient. Approximately 1 × 106 cells were stained for cell surface markers, granzyme B, and viral antigen using monoclonal antibodies. Staining procedures were performed as previously described [27].
To test the concept that the VSVΔG/ZEBOVGP vaccine may have utility as a post-exposure treatment for EBOV HF, we investigated its efficacy in two rodent models, mouse [25] and guinea pig [24], and a rhesus macaque model [11]. Initially, we treated groups of five BALB/c mice with i.p. injections of 2 × 105 pfu of the VSVΔG/ZEBOVGP vaccine 24 h prior to challenge or 30 min or 24 h post i.p. challenge with a 1,000 LD50 of the mouse-adapted ZEBOV (MA-ZEBOV) [25]. The immunization dose chosen was relatively high considering that as little as 2 × 100 pfu still conferred complete protection against the same challenge dose (unpublished data). Animals were weighed every day and scored for clinical symptoms (see Methods). Untreated control animals (naïve controls) rapidly lost weight, developed severe clinical symptoms, and died on day 6 post-challenge (Figures 1A and S1A). Surprisingly, all treated mice survived independent of the time of treatment (Figure 1A). Those animals treated 24 h prior to challenge did not show any clinical symptoms, whereas animals treated post-challenge developed mild clinical symptoms. With all protected groups, mild weight loss was observed during the first day post-challenge (Figure S1) indicating virus replication prior to clearance and survival.
Next, we treated three groups of guinea pigs (Hartley strain; six animals per group) with i.p. injection of 2 ×105 pfu of the VSVΔG/ZEBOVGP either 24 h before challenge or 1 or 24 h after challenge with 1,000 LD50 of the guinea pig-adapted ZEBOV (GA-ZEBOV) [24]. Disease progression was followed and measured as described for the mice. Untreated guinea pigs (naïve controls) showed weight loss at day 5 post-challenge progressing to death on days 7 to 9 (Figures 1B and S1B). Unlike the mice, the treatment groups were not fully protected (Figures 1 and S1). Two animals (33%) died from the group treated 24 h prior to challenge; one (17%) and three (50%) animals died from the groups treated 1 and 24 h post-challenge, respectively (Figures 1B and S1B). In all cases, the development of clinical symptoms, weight loss and time to death, were significantly delayed. All surviving animals lost weight and became sick with a degree of severity that correlated very well with disease outcome. The final survival rates were 66% for the pre-treatment group (24 h prior to challenge) and 83% and 50% in the 1- and 24-h post-treatment groups, respectively (Figures 1B and S1B).
Encouraged by the success in the rodent models, we treated eight rhesus monkeys (subjects 1 to 8) with i.m. injections of the VSVΔG/ZEBOVGP vaccine (2 × 107 pfu), and two rhesus monkeys (subjects c1 and c2) with VSV control vaccines (2 × 107 pfu) (see Methods) 20 to 30 min after challenge with 1,000 pfu of ZEBOV. The immunization and challenge doses were equivalent to what had been used in previous successful pre-exposure vaccine studies [6,9]. All animals became febrile by day 6 and haematology data indicated evidence of illness by day 6, usually manifested as lymphopenia, in most of these animals (Table 1). Surprisingly, 50% of the VSVΔG/ZEBOVGP-treated animals (subjects 1, 2, 5, and 7) survived the lethal ZEBOV challenge (Figure 2A; Table 1) without showing signs of severe disease, while three VSVΔG/ZEBOVGP-treated macaques (subjects 3, 4, and 8) developed characteristic ZEBOV HF including fever, perturbations in clinical chemistry values, and macular rashes (Figure S2); these animals died on days 9 (subject 3) and 10 (subjects 4 and 8) (Figure 2A; Table 1). Notably, all VSVΔG/ZEBOVGP-treated animals that succumbed to the ZEBOV challenge (subjects 3, 4, and 8) developed plasma viraemia on day 6 between 1 × 104 and 1 × 106 pfu/ml, whereas plasma viraemia was transient in the animals that survived (subjects 1, 2, 5, and 7) and did not exceed 1 × 102 pfu/ml on day 6 (Figure 2B). The final VSVΔG/ZEBOVGP-treated macaque (subject 6) died on day 18 (Figure 2A; Table 1). This animal had a transient low-level ZEBOV viraemia on day 6 and had cleared the ZEBOV infection by day 10 (Figure 2B). Furthermore, the animal never developed clinical symptoms consistent with severe ZEBOV HF, and organ infectivity titration showed no evidence of infectious ZEBOV in any of the tissues surveyed at post-mortem. Pathology results showed that this macaque died from disseminated septicaemia and peritonitis caused by Streptococcus pneumoniae as demonstrated by immunohistochemistry (unpublished data). The source of the bacterial infection is unknown. Both monkeys treated with the VSV control vectors (subjects c1 and c2) developed severe symptoms over the disease course with plasma viraemia titres in excess of 1 × 106 pfu/ml on day 6, macular rash (Figure S2) evident by day 7, and death on day 8 after ZEBOV challenge (Figure 2A; Table 1) with peak viraemia titre of >1 × 108 pfu/ml (Figure 2B). In addition, all animals were also tested for VSV viraemia using RT-PCR (unpublished data). In accordance with our previous results [9,10], VSV RNA was detected in most immunized animals only at day 3 post-immunization indicating transient viraemia of the vaccine vector. There was no correlation between VSV viraemia and survival.
All four animals that survived the ZEBOV challenge (subjects 1, 2, 5, and 7), and the animal that survived until day 18 (subject 6), developed ZEBOV-specific humoral immune responses with low titre IgM antibodies detected on days 6–14 (subjects 1, 5, and 7) (Figure 3A) and moderate IgG antibody titres detected on days 10–22 (subjects 1, 2, 5, 6, and 7) (Figure 3B). Neutralizing antibody titres to ZEBOV (1:80) were detected on days 14–37 after challenge in all four animals that survived the ZEBOV challenge (subjects 1, 2, 5, and 7) and the animal that survived until day 18 (subject 6) (Figure 3C). Humoral immune responses could not be detected in any of the non-survivors although these animals lived until day 9 and 10 post-challenge, which was sufficient to mount detectable IgM and IgG responses in the surviving animals.
We also evaluated changes in populations of peripheral blood mononuclear cells during the course of the study to identify any differences between the rhesus monkeys treated with the VSVΔG/ZEBOVGP vector and the controls. A rapid loss of CD4+ lymphocytes, CD8+ lymphocytes, and NK cells has been reported during ZEBOV infection of nonhuman primates [28]. In this study, we also detected a decline in the circulating CD4+ and CD8+ (2%–10% decrease) lymphocyte populations on day 6 in most of the animals regardless of treatment or outcome with a 7%–22% decrease and 2%–10% decrease in cell numbers observed, respectively (Table 1). However, the percentage of NK cells did not drop in any of the animals treated with VSVΔG/ZEBOVGP vector on day 6, but markedly increased. Interestingly, a sharp decline in NK cell number (10% decrease) was observed on day 10 in one of the animals treated with the VSVΔG/ZEBOVGP vector. Similarly, a marked increase in B cells was noted for all animals regardless of treatment or outcome on day 6, followed by a decline in B cell number on day 10.
Although no EBOV vaccine is currently licensed for human use, recent advances have been made and efficacy studies in nonhuman primates with several platforms have been encouraging [6,7,9]. Far less progress has been made in developing treatment interventions for EBOV infections [5,13,14]. Thus, there is clearly a need to develop effective strategies to respond to future EBOV outbreaks in Africa and to counter acts of bioterrorism using EBOV. Additionally, the potential EBOV exposure involving a researcher at a United States Army laboratory [16] and the unfortunate death of a Russian scientist after an accidental exposure to EBOV [15], underscore the need for medical countermeasures for post-exposure prophylaxis. Recently, a post-exposure strategy to mitigate the coagulation disorders that typify filoviral infections improved survival from 0% to 33% in the rhesus macaque model of ZEBOV HF [11]. Here, we show a significant advance in treating EBOV infections.
Our data clearly demonstrate the efficacy of the VSV-based EBOV vaccine vector in post-exposure treatment in three relevant animal models. In the mouse model it was possible to protect all animals following challenge with treatment as late as 24 h post infection. It is known from previous data that treatments and vaccines given to mice are more effective than seen in guinea pigs and nonhuman primates [1,2,13]. However, in this case it was possible to protect over 50% of guinea pigs and 50% of nonhuman primates from uniformly lethal ZEBOV challenge. It should be noted that mice received about 10 or 100 times more vaccine per weight than guinea pigs and nonhuman primates, respectively. Thus, it is possible that further optimization of dosing strategies could improve the results.
The rhesus macaques that survived infection all controlled the virus within the first 6 d of infection. The data clearly show that moderate or high-level viraemia on day 6 invariably resulted in a fatal outcome (Figure 2). In the current study, we can conclude that neutralizing antibodies were not essential for infection control (Figure 3) since they were not detected until after the animals had cleared the EBOV infection. Circulating CD4+ and CD8+ T cells were reduced in number in all animals regardless of treatment (Table 1); this indicates that the initial control of infection may not require classical T-cell responses. The time course for EBOV HF in rhesus macaques is very short (∼8 d) and therefore, CD8+ cytotoxic T-cell responses are very unlikely to be involved in the control of the infection because the cell numbers of specific responding cells could not have peaked until after the infection was controlled. The primary immune correlate of protection seems to be the rapid development of non-neutralizing antibody that was only seen in the protected animals (Figure 3). This, coupled with the NK-cell increase in the VSVΔG/ZEBOVGP-treated animals, may have resulted in significantly enhanced killing of virus-infected primary target cells and, consequently, elimination of the ZEBOV infection. An important role of NK cells for protection has also been described for immunization with virus-like particles [29].
Clearly, the adaptive response is essential to promote survival as animals immunized with the control VSV-based vaccines succumbed to the ZEBOV challenge (Figure 2, Table 1). Both control animals died on day 8, which is the historical mean for rhesus monkeys infected by the same route and dose with this seed stock (historical n = 23). However, other mechanisms probably contribute as well. Recently, Noble and colleagues described a new paradigm for an interfering vaccine in which one of the antiviral mechanisms of action is intracellular interference with the replication of the lethal wild-type virus [30]. In the current study, the VSV vectors exploit the EBOV GP, which largely determines host cell tropism and mediates viral entry [31]. We have demonstrated that the VSV vectors expressing the ZEBOV GP will infect the same cells as wild-type ZEBOV in vitro [8]. Also, the VSVΔG/ZEBOVGP vectors replicate significantly faster than wild-type ZEBOV [8]. Therefore, it is possible that these vectors compete with ZEBOV through viral interference. Clearly, even mild to moderate inhibition of ZEBOV replication may delay the course of infection and tip the balance in the favor of the host.
VSV has been shown to be a potent inducer of the innate and adaptive immune system [32–34]. In contrast, EBOV has acquired mechanisms to counteract the innate immune responses of the host at different levels [1,2,35]. The virion protein (VP) 35 of ZEBOV functions as an inhibitor of type I interferon production by blocking the activation of IRF-3 [36–38]. In addition to VP35, the ZEBOV protein VP24 functions as an inhibitor of type I interferon signaling by blocking nuclear accumulation of activated STAT-1 [35,39]. Recently, it was suggested that VP24 blocks the downstream signaling cascades activated by type I interferon by inhibiting the phosphorylation of p38 [40]. Therefore, treatment with the VSV vectors might induce or boost the innate immune response in the host, and thus, counteract the immune inhibitory effect of EBOV. In this case, the host will mount a nonspecific innate immune response allowing for time to develop a specific adaptive response that can overcome the EBOV infection and again tip the balance in favor of survival of the host.
In a historical context, it is important to note that the mechanism for post-exposure protection of humans against smallpox and rabies are also not fully understood. For post-exposure treatment of rabies, levels of neutralizing antibodies have been used as a measure of protection. However, several studies of HIV-infected patients with likely or proven exposure to rabies showed that these patients failed to develop neutralizing antibodies after post-exposure rabies vaccination, yet there were no reports of death of these patients attributed to rabies [41,42]. Moreover, studies in mice suggest that cell-mediated immunity may play an essential role in post-exposure protection [43]. In the case of smallpox, post-exposure protection is presumed to be due in part to differences in the route of exposure and growth kinetics of the wild-type variola virus versus the vaccinia vaccine [20]. Briefly, infection with variola usually starts in the upper and lower respiratory tract with subsequent spread to lymphoid tissues. Thus, the natural variola infection proceeds much slower than post-exposure i.m. vaccinia vaccination, which bypasses the respiratory tract infection. In addition, it appears that vaccinia has a shorter incubation period than variola virus resulting in a more rapid development of cell-mediated immunity and neutralizing antibody. However, a recent study using monkeypox in the macaque model demonstrated better results with antiviral therapy than post-exposure vaccination [44].
Post-exposure treatment with the VSV-based MARV vaccine vector against MARV challenge was more potent and resulted in complete survival, no disease, and undetectable viraemia [12]. The development of symptoms and viraemia in MARV-infected rhesus monkeys is delayed compared with ZEBOV [12,45], which may explain the difference in efficacy in post-exposure treatment with the VSV-based vectors. The efficacy of the VSV-based EBOV vector in post-exposure treatment might be increased by a higher treatment dose or multiple treatments over a longer period of time as is being done in post-exposure treatment of rabies [46]. Alternatively, combination therapy should be considered to increase therapeutic efficacy. In the case of EBOV, post-exposure treatment with the VSV-based EBOV vector could be combined with the previously published post-exposure strategy to mitigate the coagulation disorders [11].
Nevertheless, the VSV-based ZEBOV vaccine currently provides the most effective and promising single treatment strategy for EBOV HF. It is likely that the mechanism of protection by the VSV-based vaccine is multifactorial; while NK cells and antibody responses appear to be important to survival, viral interference and innate immune response are almost certainly essential in delaying the progression of the ZEBOV infection and extending the window for the adaptive response to become functional.
Post-exposure treatment is particularly suited for use in accidentally exposed individuals and in the control of secondary transmission during naturally occurring outbreaks or deliberate releases. Our results also suggest that this VSV platform might be even more beneficial as a fast-acting single-shot preventive vaccine. Finally, this system also provides an excellent opportunity to study the fundamental mechanisms that lead to such devastating disease following infection with ZEBOV.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession number for the ZEBOV Mayinga strain is AF272001; the accession number for the MARV Musoke strain is Z12132. |
10.1371/journal.pntd.0003467 | Bartonella spp. Bacteremia in Blood Donors from Campinas, Brazil | Bartonella species are blood-borne, re-emerging organisms, capable of causing prolonged infection with diverse disease manifestations, from asymptomatic bacteremia to chronic debilitating disease and death. This pathogen can survive for over a month in stored blood. However, its prevalence among blood donors is unknown, and screening of blood supplies for this pathogen is not routinely performed. We investigated Bartonella spp. prevalence in 500 blood donors from Campinas, Brazil, based on a cross-sectional design. Blood samples were inoculated into an enrichment liquid growth medium and sub-inoculated onto blood agar. Liquid culture samples and Gram-negative isolates were tested using a genus specific ITS PCR with amplicons sequenced for species identification. Bartonella henselae and Bartonella quintana antibodies were assayed by indirect immunofluorescence. B. henselae was isolated from six donors (1.2%). Sixteen donors (3.2%) were Bartonella-PCR positive after culture in liquid or on solid media, with 15 donors infected with B. henselae and one donor infected with Bartonella clarridgeiae. Antibodies against B. henselae or B. quintana were found in 16% and 32% of 500 blood donors, respectively. Serology was not associated with infection, with only three of 16 Bartonella-infected subjects seropositive for B. henselae or B. quintana. Bartonella DNA was present in the bloodstream of approximately one out of 30 donors from a major blood bank in South America. Negative serology does not rule out Bartonella spp. infection in healthy subjects. Using a combination of liquid and solid cultures, PCR, and DNA sequencing, this study documents for the first time that Bartonella spp. bacteremia occurs in asymptomatic blood donors. Our findings support further evaluation of Bartonella spp. transmission which can occur through blood transfusions.
| Bartonella is a genus of small bacteria with worldwide distribution, transmitted by blood-sucking insects, and is capable of causing disease in humans and animals. Some of the clinical presentations of Bartonella spp., such as cat scratch disease, trench fever, and bacillary angiomatosis are well documented; however, novel presentations have been described in the last two decades, ranging from cyclic flu-like syndrome to neurologic disease and life-threatening endocarditis. Asymptomatic human infection is possible and accidental blood transmission has been reported. Bacterium isolation is very difficult because they grow slowly and require special culture media and procedures. Serology testing poorly predicts active Bartonella infection, except in infection of cardiac valves. Therefore, diagnosis is generally challenging. However, when molecular detection techniques are coupled with special culture protocols, enhanced sensitivity and specificity can be achieved. We investigated Bartonella spp. infection prevalence in a large blood donor population and confirmed bacteremia in 1.2% of the subjects. Bloodstream infection was detected with at least three different molecular methods in 3.2% of donors. These results indicate that Bartonella is a genus of importance for transfusion medicine.
| Bartonella, a genus of fastidious bacteria with worldwide distribution, is responsible for persistent infections in animals and humans [1]. Bartonella spp. are considered neglected zoonotic pathogens, presumed to be transmitted to humans by a variety of arthropod vectors including sandflies, body lice, fleas, ticks, and keds [1,2]. During the past several years, the spectrum of clinical manifestations associated with bartonellosis, a term that now encompasses infection with any Bartonella spp., has widened substantially [3]. In humans, Bartonella spp. are known causative agents of Peruvian bartonellosis, cat scratch disease, trench fever, and bacillary angiomatosis [1]. However, more recent studies have documented bloodstream infections in patients with cardiovascular, neurological, and rheumatologic disease manifestations [4,5]. With the exception of localized lymphadenopathy or blood-culture-negative endocarditis, physicians rarely consider Bartonella sp. infection among differential diagnoses [6].
Bartonella spp. are able to infect and survive inside erythrocytes [7], resulting in a long-lasting intraerythrocytic and presumably intraendothelial infection, which can be associated with a relapsing pattern of bacteremia [8]. In vitro, these bacteria have been shown to invade, multiply within, and persist for the lifetime of the infected host cell [9,10]. Prolonged bacteremia allows greater opportunity for arthropod vector and other modes of transmission to occur between hosts. Although at least fifteen Bartonella spp. have been associated with human infections, B. henselae is the most frequent species identified from humans, as well as from companion animals such as cats and dogs [1,9]. There is no single gold standard methodology to diagnose bartonellosis and multi-step platforms are necessary to decrease false-negative test results [1]. Culture in liquid and solid media, multiple PCR reactions and serology have been used together to improve the diagnostic sensitivity [8,11].
Previous studies by our group using transmission electron microscopy and culture isolation have documented the ability of B. henselae to survive in stored blood for 35 days, suggesting the potential for transfusion-associated transmission [12]. We also documented B. henselae adhered to human erythrocytes 10 hours after inoculation of the bacteria into blood and intraerythrocytic infection after 72 hours [13]. These results suggested a potentially important role for Bartonella sp. in transfusion medicine, particularly as blood transfusion infection has been documented in cats [14] and needle stick transmission of Bartonella sp. has also been reported in two veterinarians [15,16]. Since the presence of selected Bartonella spp. was previously documented in blood samples of asymptomatic subjects [17–20], we hypothesized that bloodstream infection with Bartonella sp. occurs in blood donors at the time of donation. The objective of this study was to determine the seroprevalence and frequency of bacteremia caused by Bartonella spp. in a large asymptomatic blood donor population in Campinas, São Paulo State, Brazil.
A cross-sectional study was conducted at the UNICAMP Blood Bank (HEMOCENTRO), which serves a geographic region with an estimated population of 6.4 million people in the São Paulo state, Brazil. Healthy blood donors were randomly recruited from November 19th to December 23rd 2009, at the time of their voluntary blood donation. Sample size was estimated in 473 subjects to allow for estimation of at least 5% prevalence of bloodstream infection with Bartonella spp. in blood donors, with a desired precision of 5% given a 95% confidence limit and a design effect of 1. Therefore, with possible attrition, we enrolled 500 donors. This study was approved by the Research Ethics Committee of the University of Campinas (UNICAMP), Brazil (CEP122/2005). An informed written consent, approved by the UNICAMP Research Ethics Committee, was obtained from each participant. Donor selection, blood collection, and infectious disease screening were performed in accordance with current international standards [21,22]. Following aseptic preparation of the venipuncture site and immediately after the collection of a blood unit, an additional 5 mL of whole blood was collected into a tube with ethylenediaminetetraacetic acid (EDTA) and another 5 mL was collected into a serum separator tube via the accessory port. Samples were stored at −20°C until analysis at the University of Campinas and subsequently at Western University of Health Sciences. An overview of the diagnostic procedures performed in this study is presented in the Fig. 1.
Culture procedures used in this study followed the previous description by Duncan et al. [11] and Maggi et al. [8], with modifications. Two milliliters of whole EDTA blood from each subject were thawed and added into 8 mL of liquid Bartonella alpha-Proteobacteria growth medium (BAPGM) [23], incubated at 37°C in 5% CO2, water-saturated atmosphere and maintained with a constant shaking motion for 14 days. Blood donor sample cultures were manipulated in batches of 75 flasks. A negative control flask containing only BAPGM medium was added to each batch of samples tested and subjected to the same laboratory procedures and culture conditions. Subsequently, 1 mL of the blood-inoculated liquid culture was sub-inoculated onto agar enriched slant tubes containing 30% Bartonella spp.-negative sheep blood (confirmed by PCR and culture methods) [24] for additional 42 days. BAPGM-negative controls were also subcultured onto blood-agar. The agar slant tubes were inspected weekly for evidence of bacterial growth. Because blood cultures in BAPGM may yield other species of alpha-Proteobacteria [25], an initial screening process was performed as demonstrated in Fig. 1. Colonies were Gram-stained and those isolates with suggestive morphology were suspended and frozen in Brain and Heart Infusion (BHI) for future identification by DNA amplification and sequencing. All Bartonella spp. culture methods were carried out in a class 2 biosafety cabinet in order to minimize the possibility of specimen contamination and to protect laboratory personnel.
Molecular techniques were performed in four separate rooms to avoid DNA contamination. A uni-directional workflow was strictly enforced between pre-PCR areas (sample handling, PCR set up, DNA extraction) and post-PCR areas (DNA amplification, gel analysis, and amplicon purification). Dedicated sets of equipment, pipettes, and supplies were used in each of these locations. Strict laboratory procedures were implemented in order to avoid potential contamination of reagents and samples with amplicons. In order to prevent PCR contamination, different negative controls testing the different stages in the PCR process was included in every experiment as described in the S1 Appendix.
After a 14-day incubation period, a 1 mL aliquot of liquid culture medium was centrifuged and the pellet was subjected to DNA extraction using the QIAamp® DNA Mini Kit (QIAGEN Inc., Valencia, CA). The average of DNA yield obtained was 6.19 ng/ul (SD: 32.8 ng/ul), and the average 260/280 ratio was 1.47 (SD: 2.03). Screening of the 500 liquid culture samples was performed using Bartonella genus-specific single tube PCR. This assay was manually designed to target a hypervariable region of the 16S-23S rRNA gene intergenic transcribed spacer (ITS) of Bartonella species using primers and conditions described in the S1 Appendix. For all PCR reactions three controls were incorporated: negative BAPGM control (BAPGM medium with no blood inoculate and incubated simultaneously with each batch of liquid cultures), Mastermix reagent control, and a positive control (DNA extracted from a Houston 1-ITS strain of B. henselae; ATCC 49882). These screening tests were performed at the Multidisciplinary Center of Biological Investigation (CEMIB), UNICAMP, Brazil.
DNA obtained from Bartonella sp. positive liquid cultures and DNA from suspected subculture isolates were retested at Western University of Health Sciences, USA, for species identification. These DNA samples were tested by a Real-Time ITS PCR using primers manually designed to amplify a fragment of the 16S-23S rRNA ITS region of all Bartonella species. PCR primers and conditions are described in the S1 Appendix.
DNA samples from six isolates that were confirmed to contain Bartonella sp. DNA were further characterized using previously published conventional PCR assays for the ITS region [26], for the citrate synthase gene (gltA) [27], and the heme-binding phage-associated protein (pap31) gene [26]. In addition, each sample was also tested using a specific conventional PCR for Bartonella koehlerae [28]. DNA samples from liquid cultures were not tested by these assays due to insufficient genomic material. Amplicons generated from any one of the five PCR assays used during the confirmatory steps (Fig. 1) were sequenced for bacterial species identification (S1 Appendix).
Using B. henselae and B. quintana antigens supplied by the Centers for Disease Control and Prevention (CDC-Atlanta, USA), serum samples were analyzed for IgG antibodies to B. henselae and B. quintana antigens by an indirect immunofluorescence assay (IFA) as described in the S1 Appendix. Sera samples were tested at a 1:64 dilution. A positive test was warranted if brightly stained bacteria could be detected by fluorescence microscopy at 400× magnification. Previously IFA negative serum samples were used as negative controls.
Subjects with positive ITS PCR results from isolates were considered bacteremic. Subjects positive from liquid culture were considered with Bartonella sp. bloodstream infection. Molecular and serologic prevalence of Bartonella spp. were described as absolute frequencies, percentages, and 95% confidence intervals (computed using score method) using JMP Pro 10 for Windows (SAS Institute Inc., Cary, NC).
From the 500 blood donors tested, 16 (3.2%, 95% Confidence Interval [CI]: 2.0%–5.1%) subjects were infected with a Bartonella spp. based on culture in liquid and/or isolation on solid media followed by ITS Real-Time and/or conventional PCR in two different laboratories. ITS amplicon sequence analysis revealed B. henselae in 15 of the 16 cultures (3%, 95% CI: 1.8%–4.9%), and B. clarridgeiae in one culture (0.2%, 95% CI: 0%–1.1%) (Table 1). Among the 16 Bartonella sp. bloodstream-infected donors, 11 were confirmed by liquid culture followed by PCR amplifications and DNA sequencing, whereas six bacteremic individuals were confirmed after subculture onto blood slant tubes followed by PCR and DNA sequence confirmation (Table 1). Only one subject was PCR positive in both liquid and solid subcultures.
When subsequently amplified and sequenced, four of these six isolates contained a larger ITS region fragment (559 bp in size) that was 100% similar to B. henselae sequences deposited in GenBank (accession number NC_005956.1). From one liquid culture sample, a 190 bp ITS DNA sequence was obtained, being 100% similar to B. clarridgeiae (accession number NC_014932.1). Similarly, the presence of B. henselae DNA was confirmed in five isolates by also amplifying and sequencing the gltA gene (338 bp, 100% similarity to accession number BX897699.1), and from three by amplification and sequencing of the pap31 gene (501 bp, 100% similarity to accession number DQ529248.1). By testing the isolates, no blood donor was B. koehlerae PCR positive using species-specific ITS primers. The methodologies applied in our laboratory allowed the isolation of fastidious bacteria morphologically similar to Bartonella by Gram stain from 34 blood donors, with six isolates confirmed as Bartonella spp. by DNA sequencing. Some of the possible genera of the other isolates obtained in this study include Arthrobacter, Bacillus, Dermabacter, Methylobacterium, Propionibacterium, Pseudomonas, Sphingomonas, Staphylococcus and unknown “non-cultured” bacteria, as previously reported by Cadenas et al. using the same liquid enrichment culture method [25].
Of the 500 blood donors tested, antibodies against B. henselae or B. quintana were detected in 16.2% (81/500, 95% CI: 13.2%–19.7%) and 32% (160/500, 95% CI: 28.0%–36.2%), respectively. Seropositivity for both antigens was detected in 13.4% of blood donors (67/500, 95% CI: 10.6%–16.7%). B. quintana DNA was not detected in any blood donor in this study. Only two of the B. henselae seroreactive blood donors were confirmed by liquid culture/PCR as infected with B. henselae. However, two B. quintana seroreactive donors had confirmed B. henselae bloodstream infection, one of whom was also seroreactive to B. henselae. The B. clarridgeiae bacteremic subject was seronegative to both Bartonella spp. antigens (Table 1).
Using a combination of liquid and solid cultures, PCR, DNA sequencing, and serology, this study documented the presence of Bartonella spp. bloodstream infections in 16 (3.2%) of 500 healthy blood donors presented to a major blood bank in Southeastern Brazil. Despite the fact that exposure of blood donors to Bartonella spp. has been previously documented by serology methods [29,30], no previous study has confirmed the presence of Bartonella spp. in blood donors using similar culture and molecular diagnostic methods. A total of six B. henselae isolates were obtained in this study, with other ten blood donors having blood infection with B. henselae or B. clarridgeiae identified by liquid culture enrichment coupled with DNA amplification by PCR at two different laboratories. These results indicate, for the first time, that asymptomatic blood donors can be infected with Bartonella spp. at the time of blood donation.
Bartonella spp. are re-emerging infectious agents that can induce asymptomatic and intraerythrocytic infection in preferred and accidental hosts. These bacteria have been isolated from both immunocompetent and immunodeficient human patients [8,9,31]. Any asymptomatic infection with a blood-borne pathogen has the potential for transfusion transmission. The survival of B. henselae in collected human blood units [12] associated with the bacterium’s ability to cause infection by the intravenous route as described in animal models [14,32], and in humans after needle stick accidents [15,16], supports a potential risk for transmission through blood transfusions. Although well-documented human cases of blood transfusion transmission have not yet been published, our data further support the hypothesis that this can occur.
Transmission via blood transfusion is of great relevance if the transfused agent subsequently causes short or long-term disease manifestation in the donor recipient. In humans, cat-scratch fever, bacillary angiomatosis and endocarditis are the most common historically recognized disease entities caused by Bartonella sp. infection [1]. However, non-specific symptoms have been described, including fever of unknown origin, local or generalized lymphadenopathy, severe or recurrent anemia, chronic uveitis, optic neuritis, frequent headaches, fatigue, intermittent paresthesias, and granulomatous or angioproliferative reactions of undefined etiologies [15,16,33,34]. Similar non-specific clinical signs have been reported in Bartonella sp. bacteremic cats and dogs [35–37].
Diagnosis of bartonellosis remains a microbiological challenge because of the difficulty of culturing and isolating the bacterium from patient specimens. Bartonella sp. is highly fastidious and isolation by direct plating of blood or other clinical samples is not always successful. In order to improve the chances of isolation of Bartonella sp. from blood, samples should be cultured in liquid enrichment media to support growth [11], similar to culture techniques used for Haemophilus influenza and other bacteria species [38]. In our study, Bartonella spp. DNA was identified more frequently from BAPGM liquid culture samples than from subculture isolates, with only one subject PCR positive at both culture methods (Table 1). Similar discrepancies between results from liquid and solid culture methods were previously documented. Among 122 Bartonella-infected patients in the USA, 48 (39%) had BAPGM liquid culture samples positive for Bartonella DNA, but isolates were obtained only from five subjects (4%), including from two patients with PCR-negative liquid culture samples [4]. Causes for these differences are not completely elucidated and may be associated with low levels of bacteremia in asymptomatic humans [1]. In this study, the amount of viable Bartonella spp. in the bloodstream of these infected blood donors was not determined. Low bacterial burden in the bloodstream may limit the transmission of Bartonella spp. to a blood recipient or the development of infection. In addition, the discrepancy between the number of Bartonella sp. isolates obtained and the number of BAPGM samples positive by PCR could indicate that PCR-positive BAPGM samples contained non-viable bacteria. However, the two culture systems used in this study were designed to reproduce the vector environment (insect-based liquid culture medium BAPGM) and the host environment (blood-enriched agar medium), and each culture system may have facilitated the growth and detection of specific wild types of Bartonella. Consequently, the use of a combined diagnostic platform for testing the presence of Bartonella spp. DNA in BAPGM liquid culture and in subculture isolates in this study provided enhanced sensitivity, as previously demonstrated [39].
Evidence of Bartonella sp. infection may be confirmed by microbiological isolation, molecular techniques, and histopathologic visualization of Bartonella sp. antigens from tissue samples. Serology can be used to document exposure, but does not confirm infection. In this study, antibodies against B. quintana or B. henselae were detected in 32% (136/500) and 16.2% (78/500), respectively. A previous Brazilian study involving 437 healthy subjects from a rural area of Minas Gerais state documented B. quintana and B. henselae seroprevalences of 12.8% and 13.7%, respectively [40]. In Sweden, overall Bartonella spp. seroprevalence among 498 blood donors was 16.1%, but only 1.2% of those subjects were seropositive for B. henselae [29]. In New Zealand, 5% of 140 blood donors were seropositive for B. henselae [30]. In our study, antibodies against Bartonella sp. correlated poorly with infection as detected by PCR amplification followed by DNA sequencing or with the successful isolation of Bartonella organisms. Similar findings have been previously demonstrated in animals [14,41] and human patients [4,8], even in individuals with symptomatic disease [8,42]. Difficulties in Bartonella serodiagnosis are exemplified in the study by Vermeulen et al. [42]. It is suggested that Bartonella spp. manipulate the host immune system on a systemic scale to achieve a state of immunological attenuation, including stimulation of IL-10 secretion, which suppresses the capabilities of various immune cells, including T helper cells, monocytes/macrophages, and dendritic cells, thus interfering with both innate and adaptive immune responses [9,10]. Therefore, our results indicate that the predictive value of serology to detect Bartonella spp. infection in asymptomatic donors is low, supporting the recommendation that antibody status should not be used as a sole diagnostic modality to determine Bartonella spp. infection in blood donors. Moreover, negative results in one or more currently available diagnostic tests cannot exclude infection and, whenever possible, a combination of diagnostic tests is encouraged [42,43].
Another factor related to the apparent emergence of Bartonella spp. is the development of diagnostic techniques with improved sensitivity [4]. In recent years, the development of more sensitive and specific PCR methods, coupled with enrichment growth in specific culture media, has increased the detection of this pathogen in animal and human patient samples [4,8,11,26]. It is of clinical and epidemiological relevance that failure to amplify Bartonella sp. gene targets, following extraction of DNA from patient blood samples, does not rule out this bloodstream infection. It is estimated that Bartonella bacteremia in asymptomatic donors is approximately 10 CFU/mL of blood [1], which may be below the detection limit of most conventional or Real- Time PCR assays. Another reason for false-negative PCR or culture results is that Bartonella spp. typically cause a cyclic bacteremia [9]. It has been recently demonstrated that the detection of Bartonella sp. infection in humans is improved when three sequential blood samples are tested during a one-week period [39]. In that study, only 3 of 12 patients with Bartonella sp. bloodstream infection were documented as positive on more than one sample test date and no patient was positive on liquid culture/PCR for all three specimen test dates. Therefore, the number of bacteremic subjects may be underestimated in our study because only one blood sample from each blood donor was tested. We hypothesize that the actual number of blood donors infected with a Bartonella spp. may be higher in healthy humans than our current findings have documented. The low bacterial levels and the cyclic feature of Bartonella sp. bloodstream infection reinforce this hypothesis [9,10].
Based on recommendations by the Ethics Committee from UNICAMP Medical School, Bartonella sp. positive blood donors in this study were considered inappropriate for further blood donations. To the authors’ knowledge, there are no specific guidelines in the USA or other countries designed to prevent transfusion of human blood when donors are suspected to be infected with a Bartonella species. In Veterinary Medicine, at least two major medical boards have issued recommendations regarding blood donors. In 2005, the American College of Veterinary Internal Medicine (ACVIM) conditionally recommended the screening of canine and feline blood donors in order to obtain a Bartonella-free donor pool, especially for cats due to the high frequency of bacteremia in this host [44]. This recommendation for cats has been recently ratified by the European Advisory Board on Cat Diseases (ABCD) [45]. The results of our study indicate that guidelines for human blood transfusions should be designed, with special attention to the selection of Bartonella-free blood products for transfusion to immune suppressed subjects, which would include pediatric and geriatric patients.
The results of this study indicate that human exposure to Bartonella spp. frequently occurs in the Southeast region of Brazil, and that Bartonella sp. bacteremia occurs in asymptomatic blood donors. There is a risk of blood supply contamination with these pathogens from asymptomatic bacteremic donors. The impact of transmission of Bartonella spp. through blood transfusions recipients should be evaluated, as well as the use of advanced diagnostic techniques for the screening of Bartonella sp. infection among blood donors.
|
10.1371/journal.pntd.0002744 | Conjunctival Scarring in Trachoma Is Associated with the HLA-C Ligand of KIR and Is Exacerbated by Heterozygosity at KIR2DL2/KIR2DL3 | Chlamydia trachomatis is globally the predominant infectious cause of blindness and one of the most common bacterial causes of sexually transmitted infection. Infections of the conjunctiva cause the blinding disease trachoma, an immuno-pathological disease that is characterised by chronic conjunctival inflammation and fibrosis. The polymorphic Killer-cell Immunoglobulin-like Receptors (KIR) are found on Natural Killer cells and have co-evolved with the Human Leucocyte Antigen (HLA) class I system. Certain genetic constellations of KIR and HLA class I polymorphisms are associated with a number of diseases in which modulation of the innate responses to viral and intracellular bacterial pathogens is central.
A sample of 134 Gambian pedigrees selected to contain at least one individual with conjunctival scarring in the F1 generation was used. Individuals (n = 830) were genotyped for HLA class I and KIR gene families. Family Based Association Tests and Case Pseudo-control tests were used to extend tests for transmission disequilibrium to take full advantage of the family design, genetic model and phenotype.
We found that the odds of trachomatous scarring increased with the number of genome copies of HLA-C2 (C1/C2 OR = 2.29 BHP-value = 0.006; C2/C2 OR = 3.97 BHP-value = 0.0004) and further increased when both KIR2DL2 and KIR2DL3 (C2/C2 OR = 5.95 BHP-value = 0.006) were present.
To explain the observations in the context of chlamydial infection and trachoma we propose a two-stage model of response and disease that balances the cytolytic response of KIR expressing NK cells with the ability to secrete interferon gamma, a combination that may cause pathology. The data presented indicate that HLA-C genotypes are important determinants of conjunctival scarring in trachoma and that KIR2DL2/KIR2DL3 heterozygosity further increases risk of conjunctival scarring in individuals carrying HLA-C2.
| Chlamydia trachomatis is a pathogen that causes sexually transmitted infections (STIs) and the blinding disease trachoma. Natural Killer (NK) cells are part of the host immune system's first line of defence against infection. NK cell functions are genetically encoded and differences between individuals mean that some people are better able to respond to infections than others. We found that in certain combinations, specific variants of the gene HLA-C (Human Leucocyte Antigen, C) and of a complex set of genes called the Killer-cell Immunoglobulin-like Receptors (KIR) were associated with a six-fold increase in the relative risk of scarring tissue damage resulting from ocular C. trachomatis infection (trachoma). This combination of genetic variants may reduce the host's ability to effectively resolve infections and result in a harmful immune response that ultimately leads to tissue damage and scarring. KIR+ NK cells are potential cellular mediators of the damaging immune response. Previous studies have identified that the same HLA-KIR genetic constellation that associates with trachoma is actually protective against infectious diseases such as malaria and tuberculosis. The high frequency of the trachoma-associated constellation in African populations may therefore be explained by the evolutionary benefits of protection from the complications of severe disease.
| Chlamydia trachomatis (Ct) is an obligate intracellular bacterium [1] which causes significant morbidity as the causative factor of around 106 million new sexually transmitted infections per annum [2]. As the cause of trachoma, the same bacterium is the most common infectious cause of blindness [3]. Ct serovars exhibit highly specific tissue tropism, with serovars A–C being limited to the mucosal epithelium of the ocular conjunctiva. The remaining serovars are sexually transmitted, but whilst serovars D–K are limited to the mucosal epithelia of the genitourinary tract and rectum, the strains L1–L3 are able to invade other tissues including the lymph nodes. Ocular infection in trachoma is spread among young persons through exposure to secretions from the infected eye via direct physical contact, on fomites or by eye-seeking flies [4]. Repeated and prolonged cycles of infection and inflammation have been identified as the main factors that lead to the progressive formation of fibrotic scars on the tarsal conjunctiva, which ultimately becomes deformed. This can cause entropion and trachomatous trichiasis (TT), a condition where the eyelashes turn inwards and irreversibly damage the cornea by scratching the globe of the eye. If left unchecked, TT causes corneal opacity, visual impairment and blindness.
Active trachoma is frequently found in the absence of detectable Ct infection and both tissue damage and scarring are thought to be the result of a chronic immuno-pathological reaction [5]. Human conjunctival transcriptome studies in trachoma suggest that in addition to T cell and innate responses of epithelial cells, the activation and cytotoxic responses of natural killer (NK) cells is an important determinant of the severity of active trachoma [6], [7]. NK cells are a rich source of multiple chemokines and cytokines, including interferon gamma (IFNγ), a cytokine that is central to the control of chlamydial intracellular development and growth. IFNγ also has anti-fibrotic properties that can counteract the effects of TGF-β and inhibit fibroblast proliferation and collagen synthesis [8], but when inappropriately expressed may cause immunopathology. NK cells in mucosal-associated lymphoid tissues are known to be important in the maintenance of epithelial cell integrity via production of the cytokine IL-22 [9]. NK cells therefore have the potential to fulfil multiple roles that encompass tissue homeostasis, tissue re-modelling and immunity.
Early studies in murine chlamydial model infections found that NK cell depletion exacerbated disease, delayed clearance and limited the development of specific T cell responses [10], [11]. Subsequent studies have confirmed that in response to chlamydial stimulation, NK cells are promoters of T cell immunity and a major source of IFNγ [10], [12] but their role as lytic effector cells is less clear. Although Ct infected cell lines are lysed in vitro, NK cells purified from the peripheral blood of individuals with current chlamydial infection had diminished lytic activity (and reduced IFNγ) compared with uninfected controls [13]. Population diversity in the highly polymorphic genes that encode the variable NK receptors and their ligands [14] along with functional heterogeneity in the NK cell repertoire may account for these findings [15].
Trachoma is a complex inflammatory fibrotic disease in which host polymorphism in immune response genes plays a significant role [16]–[18]. The conjunctival epithelial surface is compromised in trachoma [5] as a result of the host response to the causative bacterium, which occupies an intracellular niche. Therefore the mechanisms used by NK cells in the control of other intracellular infections such as Hepatitis B [19], Hepatitis C [20] and HIV [21]–[23] might also be effective against intracellular Ct.
NK cells become activated when they are released from inhibition that is normally bound by interaction of specific HLA class I ligands with inhibitory Killer-cell Immunoglobulin-like Receptors (KIRs) [24]. The ligands of several inhibitory KIR have been described including HLA-A3 and HLA-A11 alleles, which are ligands of the KIR3DL2 receptor [25], [26] and the HLA-Bw4 public epitope which is the ligand of KIR3DL1 [27], [28]. HLA-C alleles can be classified (according to a functional dimorphism at amino acid position 80) as carrying one of two KIR binding epitopes, which are known as HLA-C1 and HLA-C2 [29]. The HLA-C2 group of alleles (HLA-C*02/04/05/06…) are ligands of the inhibitory receptor KIR2DL1 [30]–[32] and its activating counterpart KIR2DS1 [33]. The HLA-C1 group alleles (HLA-Cw*01/03/07/08…) are ligands of both KIR2DL2 and KIR2DL3 [30]–[32], however, the latter KIR are both able to cross-react (with differing avidities) with a small number of HLA-C2 and HLA-B allotypes [34]. Although germ-line encoded, the KIR gene system is highly polymorphic and exhibits extensive diversity both between individuals and between populations [35]–[38]. KIRs exhibit haplotype diversity such that different individuals possess variable gene contents. Since KIR and HLA are also found on different chromosomes, individuals can possess a KIR for which they have no cognate ligand, or vice versa. The extensive polymorphism in the KIR system culminates in a repertoire of NK cells within an individual that is more or less sensitive to release from inhibition under appropriate physiological conditions [39]. The strength of the signals mediated by interactions between specific HLA and KIR alleles is also highly variable [29], [40], [41] and this further limits overall NK cell responsiveness [42]. In part the responsiveness might be predicted by the presence of type ‘A’ and type ‘B’ KIR haplotypes. Type A haplotypes carry genes encoding predominantly inhibitory KIRs. B haplotypes contain some or all of the same genes found on A haplotypes, but additionally may carry the inhibitory KIR2DL2 and KIR2DL5 genes and numerous activating KIRs [36]. KIR haplotypes can be separated in to two variable regions, defined by their orientation towards the centromeric (Cen) or telomeric (Tel) regions of the chromosome [43]. The KIR A and B haplotypes are present in all populations studied to date and are thought to be maintained by the balancing selection pressures of infection, immunopathology and healthy reproduction [44]–[47]. In recent human history, a wide range of infectious diseases may have reduced the balancing effects in African populations, leading to more directional selection and a unique pattern of HLA and KIR diversity in this region [38], [47]. We therefore assessed the extent to which host genotypes at the HLA and KIR loci were associated with trachomatous scarring in a trachoma endemic population from The Gambia.
The study was conducted in accordance with the tenets of the Declaration of Helsinki. The Ethics Committee of the Gambian Government/Medical Research Council Unit, and the ethics committee of the London School of Hygiene and Tropical Medicine approved the study (MRC SCC1177). Individual written informed consent was obtained from all adult participants. Written informed consent was obtained from a parent/guardian on behalf of those subjects aged <18 years who wished to take part in the study. All samples were anonymised.
We selected a family study design and identified probands at a relatively early age for clinical signs of conjunctival scarring. This maximised statistical power whilst controlling for population stratification through the use of related control samples. The study population came from multiple regions of The Gambia and included multiplex families of mixed ethnic background. Families were ascertained through the identification of probands in which there were signs of trachomatous scarring at an early age (age ≤30 years). This approach maximised the extent to which genetic rather than environmental factors could be expected to have contributed significantly to the probands' phenotypes. We recruited first-degree relations of the probands. In most cases this meant that we sampled both biological parents of the probands and all their (self-described) full siblings. Samples for DNA analysis were collected from buccal mucosae using sterile cyto-brushes (Part Number F-440151, SLS, Nottingham, UK). After collection, brushes were returned to their original packaging and stored dry at room temperature for up to 6 months [48] before DNA extraction was performed using a salting out procedure.
An average of 4 offspring per family was assumed with a population prevalence of scarring in those <30 years of age in The Gambia of ∼2% [49]. The Pedigree Based Association Test (PBAT) v3.6 program [50] was used to calculate the power of the study to detect with 95% confidence (α<0.05) a genetic association with odds ratios 1.5, 2 and 3 when the hypothetical disease allele had a frequency between 0.01 and 0.50. Figure S1 shows the estimated power of this study to detect genetic associations with trachomatous scarring at a range of allele frequencies and effect magnitudes, given the sample size. We had >90% power to detect an effect size greater than an odds ratio (OR) = 3 when the allele frequency was ≥0.05 and similar power to detect an effect size of OR = 2 when the allele frequency was ≥0.19.
Trachoma was graded in the field using the WHO simplified grading system by field supervisors certified for trachoma grading with regular performance checks as described by the PRET clinical trial manual of operations [51]. Left and right tarsal conjunctivae of all subjects were photographed as described by Derrick et al. [52]. Photographs were subsequently reviewed by two ophthalmologists with experience of grading trachoma and a final grade agreed. Subjects were assigned to the ‘scarred’ group if there were any signs of trachomatous scarring, in either eye. Individuals where phenotypes could not be confirmed for reasons of poor quality photography (n = 5) did not contribute to the statistical tests of association.
KIR genotyping for the presence or absence of 17 KIR genes was performed by PCR using the set of sequence specific primers described by Vilches et. al. [53]. The genotyping method was validated by participation in the UCLA Immunogenetics Center KIR exchange programme (http://www.hla.ucla.edu/cellDna.htm). Medium resolution HLA-A, -B and -C genotyping was performed using LABtype sequence specific oligonucleotide probes (OneLambda, Canoga Park, CA. USA) on a Luminex platform (Luminexcorp, Austin, TX. USA). Medium resolution HLA typing data generates strings of possible allele combinations. Information from the HLA genotypes of family members was used to reduce the length of the strings of possible allele pairs and to eliminate alleles that were not compatible with Mendelian inheritance within a given pedigree. Strings were further shortened where possible to include only common and well-defined alleles [54]. In order to maximise statistical power, highly sequence similar HLA alleles were combined in to groups (table S1) before FBAT. KIR ligands of HLA (HLA-A*03/11/Bw4, HLA-B-Bw4, HLA-C1/C2) were inferred from the full HLA genotypes of individual specimens rather than the reduced strings. The HLA-C*16:01 (HLA-C1) and HLA-C*16:02 (HLA-C2) alleles were frequently ambiguous and where this was the case alleles were assigned to HLA-C*16:01 because the HLA-C*16:02 allele has not been observed in other West African populations whilst HLA-C*16:01 is very common (data from allelefrequencies.net). HLA types were used to identify cases of parental mis-assignment and inconsistent parent-offspring genotypes. KIR phenotypes (presence/absence) were tested for Mendelian inconsistencies. KIR2DL5, KIR2DS3 and KIR2DS5 were not included in the association tests as they can segregate to both Cen-B and Tel-B regions and confound haplotype assignments.
KIR gene frequencies were compared to those of other world populations using data from allelefrequencies.net and PCA using R. Family based tests of HLA association were carried out using FBAT v.2.0.3 [55] performing a series of bi-allelic tests (i.e. association of an index allele against all other alleles) under an additive genetic model and the null hypothesis of no linkage and no association of any factor of the HLA system with trachomatous scarring. This approach is robust to effects of population structure [55], [56] and is applicable to a data set with samples originating in mixed ethnic backgrounds. We tested for associations between scarring and all HLA alleles with a sample frequency greater than 0.05 with an offset value of 0.02 (population prevalence of scarring in persons ≤30 years of age) to allow the unaffected siblings to contribute to the test statistic. All FBAT p-values were adjusted using a conservative Bonferroni correction. Significant associations were tested again using a case/pseudo-control conditional logistic regression (CLR) [57], which generated estimates of odds ratios and associated p-values. To test for independence between the disease-associated alleles, we included all alleles that had a corrected p<0.05 in a multivariate CLR model. To establish whether significant HLA associations were restricted to F1 subjects with specific KIR genotypes we tested the full data set under a genotype model [58], [59], using CLR, in different subsets of the F1 data where the population was limited by the KIR genotype. Because of the high linkage disequilibrium between factors of the KIR system, these tests were not considered to be independent and test statistics were corrected using the Benjamini-Hochberg method.
We sampled 830 individuals from 134 pedigrees and 146 nuclear families in which scarring trachoma had been identified in the first filial (F1) generation. The self-described ethnic background of the parental (P0) population (n = 260) was approximately 40% Mandinka, 23% Fula, 15% Jola, 15% Wolof, 5% Bambara and 2% other minority ethnic groups. There were 570 persons in the F1 generation, where the gender distribution was 52% (n = 296) male and 48% (n = 274) female. The median number of offspring per pedigree was 4 (range 1–11). Eight families had one missing parent. There were 180 (∼32%) cases of trachomatous scarring in the F1 generation and of these, 72 (40%) were female and 108 (60%) were male. Three hundred and eighty six (∼67.8%) F1 individuals were unaffected and phenotypic status could not be confirmed for 4 (<1%). Table 1 gives a detailed description of the phenotype distribution in the families. Detailed examination of photographs revealed that 12 probands did not have sufficient signs of trachomatous scarring. One proband could not be graded. In all the families where there was no photography confirmed scarred proband, at least one sibling was identified who was under 30 years of age and had signs of scarring. HLA genotyping identified paternal misassignment in 63 F1 individuals (11%) who were reassigned to an unknown father but were otherwise retained for analysis.
Table 2 shows the Family Based Association Test (FBAT) estimates of the HLA allele and KIR epitope frequencies in the sample population. Figure 1 describes the 64 unique KIR genotypes that were observed in the P0 generation. Thirty-eight additional KIR genotypes were revealed by re-assortment of the parental haplotypes in the F1 generation (Figure 2). All observed genotypes were assigned as either the ‘AA’ or ‘Bx’ genotypes (where Bx includes both AB and BB genotypes) for the full KIR region and where possible, for each of the Cen and Tel regions. A number of unusual genotypes were identified in this population, most notably, 10.4% of P0 individuals (n = 27/260) possessed KIR2DL2 but not KIR2DS2.
Pairwise linkage disequilibrium data (LD) for the KIR genes were calculated (figure S2). Contrary to data from other studied human populations [58], [60], [61] and consistent with other findings within Africa [47], we observed reduced LD between KIR genes. We did not identify any pairs of KIR genes that were in perfect LD (r2 = 1 : only two of the four possible haplotypes observed), although a number of KIR genes were found to be in complete LD (D′ = 1 : only three of the four possible haplotypes observed). The extent of LD was insufficient for high confidence imputation of missing KIR genotypes for use in FBAT [58].
Any HLA alleles and KIR epitopes with estimated frequencies above 0.05 were included in the FBAT. Three sets of HLA alleles were significantly associated with trachomatous scarring (Table 2). These were HLA-B*08:01 (Z = −3.548, p = 0.0004, corrected p = 0.01), HLA-C*03:04 (Z = −3.201, p = 0.0014, corrected p = 0.04) and the KIR epitope HLA-C1/C2 (Z = 3.622, p = 0.0003, corrected p = 0.008). Only HLA-C1/C2 remained significant (HLA-C2, OR = 1.684 p = 0.0033) in a multivariate case/pseudo-control, additive model that included all three factors (Table 3), indicating that the HLA-C1/C2 epitope was the only significant independent factor of the HLA system that was associated with trachomatous scarring. In line with previous study designs and analyses we divided the data into several subsets [58], [59]. We identified that in the majority of subsets, as with the unselected sample, the relative risk of scarring increased with the number of genomic copies of the HLA-C2 epitope in an additive manner (Table 4). The association of the HLA-C2 homozygote genotype with trachomatous scarring was restricted to the subsets of offspring who were KIR2DL2+ and KIR2DL3+ (Cen-AB) (OR = 5.95, p = 0.0025, BH corrected p = 0.006) and to those who were KIR3DL1+ KIR3DS1− and KIR2DS1− (Tel-AA) (HLA-C2 homozygote OR = 4.89, p = 0.00006, BH corrected p 0.0004). Elevated odds ratios were observed in sensitivity analyses (Table 4) in F1 samples where the case definition was restricted to those with moderate or severe (WHO FPC grade C2 or C3) rather than evidence of any (C1, C2 or C3) scarring.
We used Principle Components Analysis (PCA) to compare the KIR gene frequencies observed in the P0 generation of the Gambian trachoma families to those observed in other populations where data was available (allelefrequencies.net database, (Figure 3)). The proportions of the total variance explained by the first three principle components were 0.42 (σ = 2.05), 0.28 (σ = 1.69) and 0.11 (σ = 1.03). The P0 specimens clustered with other populations of African descent, which could be recognised by the observation of high frequencies of the genes defining the Cen-B (KIR2DS2, KIR2DL2) and Tel-A (KIR3DL1 and KIR2DS4) haplotypes.
We identified three factors of the HLA system (HLA-C1/C2, HLA-B*08:01 and HLA-C*03:04) that associated with trachomatous scarring. However, the protective associations of HLA-B*08:01 and HLA-C*03:04 observed by univariate analysis were not independent of the HLA-C1/C2 association under a multivariate model. This dependence can be explained by the presence of a common HLA-B*08∼C*03 haplotype, which we estimate to have a frequency of around 5.7% in the Gambian trachoma families. In Senegalese Mandinka, this haplotype has an estimated population frequency of 5.5% [62] whilst in African-Americans the frequency is estimated to be as high as 22% [63]. The associations that were observed between HLA-B*08 and HLA-C*03 (an HLA-C1 group allele) appear to be proxies for the association of the HLA-C1/C2 epitope, which reached statistical significance in univariate analysis because of their high population frequencies and the relatively large contribution they therefore made to the HLA-C epitope data.
We observed patterns of transmission disequilibrium in our sample of families that suggest HLA and KIR genotypes associate with high magnitude increases in the relative risk of scarring in trachomatous disease. Through sensitivity analysis (Table 4) we demonstrated the robustness of the analysis and consistency of the results when the cases were defined by more stringently defined phenotypes relating to severity of scarring.
The chlamydial protease, CPAF has been reported to interfere with the surface presentation of HLA class I molecules [14], [64]–[66], but recently this has been called in to question by Chen et al. [67]. Kägebein et al. [68] then demonstrated that Ct infection does not lead to alteration in normal MHC Class-I expression, maturation or surface presentation. This implies that Ct infected cells are unlikely to be targets for missing-self reactions mediated by NK cells which selectively monitor down-regulation or loss of self-type MHC class I on target cells. Instead it is more likely that cytotoxic NK responses in chlamydial infections are controlled by dynamic changes in the expression levels of activating NK receptors. These changes may occur as a result of infection and other environmental triggers [69], [70] and might overwhelm the inhibitory effects of the more strictly expression-regulated [69] NK inhibitory pathways.
HLA-C1:KIR2DL3 inhibited NK cells have weaker inhibitory signals than other HLA-C inhibited cells [29] and may have a lower threshold for activation. Khakoo et al. [40] reported that the HLA-C1/C1 KIR2DL3/2DL3 genotype constellation increased probability of clearance of early stage Hepatitis C Virus (HCV) infections. Ahlenstiel et al. [42] provided evidence that HLA-C1 homozygotes might be better able to challenge early infections by showing that the proportion of the total NK cell repertoire that is educated and inhibited by HLA-C is ∼50% greater in this group than that in HLA-C2 homozygotes [42]. The same study showed that HLA-C1 inhibited NK cells are better able to mount rapid, intense responses to infection through degranulation and IFNγ secretion [42]. In Ct infections, HLA-C1/C1 individuals may be able to limit chronicity by controlling the early stages of Ct infections with an NK response that is easily activated, and involves a more substantial component of the NK repertoire than in HLA-C2/C2 individuals. This may also be true of HLA-C2+ individuals who possess only weakly responsive KIR2DL1 alleles, such as those alleles that are found on the commonest B haplotypes in Caucasian populations [41]. However, in the Ga-Adangbe population of Ghana, there was a great diversity B haplotypes, none of which were found at high frequency and many of which carried non-attentuated KIR2DL1 alleles [38]. Any assumption about how the presence of Cen-B might indicate reduced cellular inhibition in Gambians should therefore be made with some caution.
The role of KIR in mediating NK cytotoxic responses is well studied, but it is now clear that KIR expressing NK cells are also a major source of IFNγ [71]. The ability of NK cells to produce IFNγ in response to microbial stimuli is related to the density of NCAM-1 (CD56) expressed on their surface, their KIR genotype and the degree of stimulus by accessory cells. An indication of the strength of regulation imposed by the KIR genotype can be estimated as a ratio, known as the ‘DIM factor’, between the response of the CD56dim (KIR-HLA dependent) and CD56bright (KIR-HLA independent) IFNγ responding populations [71]. The majority of human NK cells in the periphery are CD56dim, express KIR and are susceptible to inhibition through KIR-HLA interaction. KIR genotype directly influences the DIM factor, but the exact genotypic conformation that defines this has yet to be elucidated. It has been proposed that the NK cell IFNγ response will be higher in individuals with more KIR educated NK cells, a situation found when there is a greater diversity of within-person inhibitory KIR genes. Experimentally, IFNγ production in CD56dim NK cells showed least inhibition (and the highest DIM factor) in KIR AB heterozygotes [71]. In HLA-C2 homozygotes, we observed a significant KIR2DL2/L3 heterozygote (Cen-AB) disadvantage (Table 4) and an increased relative risk in those with the Tel-AA genotype. The number of persons with Tel-B genotypes was very low in this study, which reflects the low diversity in the Tel region that was reported in another West African population [38]. The high phenotypic frequency of KIR3DL1 (Tel-A) in this Gambian population (∼99.6%) indicates that most individuals with the Cen-AB genotype possess at least one Tel-A haplotype. The Cen-AB, Tel-A+ genotype represents a full complement of the known MHC specific inhibitory KIRs (KIR2DL1, KIR2DL2, KIR2DL3, KIR3DL1 and KIR3DL2) and this genotype might define a high DIM factor [71]. NK cell clones with a Cen-AB genotype would therefore be relatively resistant to inhibition (DIM factor >1) and would retain the potential for high IFNγ production.
The KIR system exhibits extensive diversity in African populations [47], [72], [73] possibly driven by a high burden of life threatening infectious diseases, that have exerted strong (diversifying) selective pressures on each population [46], [47], [74]. The high prevalence of Ct STIs in some African populations has been implicated as a contributory factor to the high incidences of infection related infertility that are observed in Africa [75]. It is therefore surprising that Ct disease associated KIR and HLA genotypes are enriched in Africa. One explanation is that opposing selection pressures from other infectious diseases negate selection by Ct. Our sample was selected based on disease phenotype and we found KIR gene frequencies similar to other African populations (Figure 3). The Gambian samples are clearly separated from those in other geographical regions by high frequencies of the genes that define the Cen-B and Tel-A haplotypes (Figure 3) [72]. The frequency of HLA-C2 epitopes is reported to be higher in African populations than in other populations [46], [76], [77] and the HLA-C epitope frequencies that we observed are similar to those previously described [77]. Yindom et al. [78] reported that the proportion of persons with the constellation HLA-C1 and KIR2DL2/KIR2DS2 (Cen-B) is higher in cases of malaria than in population matched, cord-blood controls [78]. In a study of a South-East Asian population, Hirayasu et al. [79] reported that natural selection may have reduced the frequency of the HLA-C1 and KIR2DL3 (Cen-A) because this genotype associates with cases of cerebral malaria. Both studies identify HLA-C1 in association with malarial disease, but they implicate different KIR Cen haplotypes. In the Gambian trachoma families, we observed that many Cen-B haplotypes lacked KIR2DS2, whilst maintaining KIR2DL2. This genotype has previously been identified in an African population [73] and its presence could be explained if KIR2DS2, rather than KIR2DL2, were mediating the Cen-B risk effect. The combined evidence of several TB studies shows that KIR Cen-A [80]–[82] and Tel-B [82], [83] haplotypes associate with TB cases. We therefore suggest that the HLA-C2 homozygous, Cen-AB, Tel-A+ population are more resistant to the complications of both malaria and TB, but more susceptible to trachomatous scarring and that trachomatous scarring (and possibly reduced fertility) is the penalty of increased survival.
We identified KIR-HLA interactions as an important contributory factor to risk of scarring. The HLA-C2 homozygous, KIR2DL2+, KIR2DL3+ genotype associates with high relative risk of scarring. We suggest a model that may explain the data in which HLA-C2 may favour chronic infection, whilst KIR2DL2/L3 heterozygosity favours chronic inflammation (Figure 4). In some aspects this is similar to the model put forward by Hollenbach et al. to explain the observation of HLA-C1/C1, KIR2DL2/L3 heterozygote risk in Crohn's disease [84], which like trachoma is characterised by chronic inflammation and fibrotic immunopathology. It is possible that the high burden of trachomatous scarring, TT and infection related infertility in observed in Africa can be explained in part by unusually high frequencies of HLA-C2 and KIR2DL2/L3 heterozygosity and the effects of NK cell responsiveness. The therapeutic consequences of such a theory would impact on vaccine immune-therapies and we would expect that current efforts in the development of chlamydial vaccines, adjuvants and immunisation schedules would additionally monitor the boosting or modulating effects on the NK cell compartment. As early as the 1990 it was shown that vaccination with influenza virus was able to elicit NK cell responses [85]. More recent work has demonstrated that many vaccination regimes against viruses boost not only adaptive T and B cell response but also lead to repetitive expansions of NK cells [86]–[88]. Some immunologists have termed these “memory-like NK cell responses” and have now begun to consider the role of these responses in vaccine induced immunity [89]. The effectiveness of NK cells as targets of vaccine immuno-therapy has been described in oncology [90]. Efforts are now required to investigate the role of NK cells in immunity following vaccination with a wider spectrum of bacterial vectors and in natural immunity to infectious diseases such as trachoma.
|
10.1371/journal.pntd.0006874 | A MALDI-TOF MS database with broad genus coverage for species-level identification of Brucella | Brucella are highly infectious bacterial pathogens responsible for a severely debilitating zoonosis called brucellosis. Half of the human population worldwide is considered to live at risk of exposure, mostly in the poorest rural areas of the world. Prompt diagnosis of brucellosis is essential to prevent complications and to control epidemiology outbreaks, but identification of Brucella isolates may be hampered by the lack of rapid and cost-effective methods. Nowadays, many clinical microbiology laboratories use Matrix-Assisted Laser Desorption Ionization–Time Of Flight mass spectrometry (MALDI-TOF MS) for routine identification. However, lack of reference spectra in the currently commercialized databases does not allow the identification of Brucella isolates. In this work, we constructed a Brucella MALDI-TOF MS reference database using VITEK MS. We generated 590 spectra from 84 different strains (including rare or atypical isolates) to cover this bacterial genus. We then applied a novel biomathematical approach to discriminate different species. This allowed accurate identification of Brucella isolates at the genus level with no misidentifications, in particular as the closely related and less pathogenic Ochrobactrum genus. The main zoonotic species (B. melitensis, B. abortus and B. suis) could also be identified at the species level with an accuracy of 100%, 92.9% and 100%, respectively. This MALDI-TOF reference database will be the first Brucella database validated for diagnostic and accessible to all VITEK MS users in routine. This will improve the diagnosis and control of brucellosis by allowing a rapid identification of these pathogens.
| Brucella are bacteria that mainly infect animals. They can also be transmitted to humans and cause a serious disease called brucellosis. Half the world's population is considered exposed, especially in the poorest rural areas. Experts agree that prompt identification of Brucella isolates is essential to provide appropriate treatment to patients and to control epidemiological outbreaks. Mis-identification of these highly infectious pathogens may lead to delays in diagnosis, but also to increased risks of accidental exposure for laboratory workers. MALDI-TOF mass spectrometry is now the first line of bacterial identification in many routine diagnostic laboratories. However, not all clinical mass spectrometers can identify Brucella. In this work, we updated a database with Brucella spectra to improve the performance of MALDI-TOF mass spectrometers. These instruments will now be able to identify accurately Brucella isolates. This will greatly improve the diagnosis of brucellosis.
| Brucella are important pathogens in medical and veterinary context. These Gram-negative bacteria can be transmitted from their animal reservoir to humans, usually by ingestion of contaminated milk products or direct contact, causing brucellosis. This zoonosis causes a severely debilitating illness characterized by intermittent fever, chills, sweats, weakness, myalgia, osteoarticular or obstetrical complications and endocarditis.
This disease is largely unreported and the true incidence of human brucellosis is thus unknown [1]. According to the World Health Organization (WHO), half a million new cases are reported each year, most of them in the poorest rural areas of the world [2]. Indeed, while the disease has been successfully prevented in most industrialized countries, it remains a significant burden in the Mediterranean region, all over Asia, sub-Saharan Africa, and certain areas in Latin America. Approximately half of the human population worldwide is considered to live at risk of exposure [3]. Moreover, due to the low dose required to cause infection (10–100 colony-forming units) and the potential for aerosol dissemination, Brucella was considered a potential bioterrorism agent early in the 20th century [1] and its possession and use is still strictly regulated in many countries.
Currently, the Brucella genus consists of eleven recognized species plus several isolates that have not yet been officially designated. The major zoonotic species are B. melitensis, B. abortus and B. suis which are subdivided into biovars by a set of phenotypic characteristics including lipopolysaccharide (LPS) epitopes, phage sensitivity, dye sensitivity and a battery of biochemical tests. These three species are also the most common in domestic livestock. B. melitensis is responsible for the majority of human cases in the Mediterranean basin, the Arab peninsula, Latin America countries and Asia, while B. abortus is more prevalent in the United States, Northern Europe and Africa [4]. B. suis and B. canis infections are more sporadic in humans. Very rare human infections have also been reported with B. inopinata [5,6], B. ceti [7,8] and B. neotomae [9,10].
Clinical microbiology laboratories play a key role in the diagnosis and management of human brucellosis and should be able to provide a rapid and exact identification of Brucella spp. Currently, the most suitable tool for identification of bacteria is Matrix Assisted Laser Desorption/Ionization-Time of Flight Mass Spectrometry (MALDI-TOF MS). This method provides rapid, sensitive and cost-effective identification and is currently replacing phenotypic microbial identification. Its accuracy however largely depends on the coverage of the database of the commercially available MALDI-TOF MS systems. With regards to Brucella, identification was not possible because this genus was not represented in the databases of the two main MALDI-TOF MS system manufacturers (i.e. bioMérieux and Bruker) [11–13]. Only the Bruker Security Relevant (SR) database, or custom databases developed in some laboratories, can identify these highly pathogenic bacteria, but access to these databases is not possible in some countries due to export restriction regulations [13–15]. Moreover, only B. melitensis is included in the SR database.
The bacterial strains used for the construction of the database are listed in Tables 1 and S1. Each of these strains was cultivated on several different media (S1 Table). The bacterial isolates used for the external evaluation and their culture conditions are listed in Tables 2 and S2. All strains used in this study were previously characterized using an established workflow (phenotypic assays, Multiple-Locus Variable number tandem repeat Analysis -or MLVA-, whole-genome sequencing) [16].
Samples used to build the spectra database were prepared according to a previously established inactivation protocol [19] consisting in resuspending two full loops of bacteria (i.e. multiple colonies) in 200 μL of solvent mix, vortexing (10 sec), centrifuging (10,000 g, 2 min) at room temperature, removing 190 μL and resuspending in the 10 μL of solvent left in the tube. For the external evaluation study, this protocol was simplified by suspending only one loop of bacteria in 100 μL of solvent mix, vortexing (10 sec) and incubating at room temperature (20–25°C, 3 minutes). Bacteria were efficiently inactivated by this method and the biomass concentration of the samples allowed identification by MALDI-TOF MS, demonstrating that the centrifugation step in the original protocol was not required.
One μL of each sample was applied to a single well of a disposable, barcode-labeled target slide (VITEK MS-DS, bioMérieux), overlaid with 1 μL of a saturated solution of alpha-cyano-4-hydroxycinnamic acid matrix in 50% acetonitrile and 2.5% trifluoroacetic acid (VITEK MSCHCA, bioMérieux) then air dried. For the database construction, several independent measurements were recorded for each strain (see S1 Table for the different culture conditions).
For instrument calibration, an Escherichia coli reference strain (ATCC 8739) was directly transferred to designated spots on the target slide using the procedure recommended by the manufacturer.
Mass spectra were acquired using a VITEK MS Plus (bioMérieux, Marcy l’Etoile) and the Launchpad v2.8 software program (Kratos, Shimadzu group Compagny, Manchester, UK). Dendrograms showing taxonomic relationships between strains were constructed using the SARAMIS software (bioMérieux, Marcy l’Etoile, France).
The database was built as previously described [20]. Briefly, peak lists were binned by assigning each peak within the mass range of 3.000–17.000 Da to one of 1,300 bins. A predictive model was then established for each species using the Advanced Spectra Classifier (ASC) algorithm developed by bioMérieux (La Balme les Grottes, France). The outcome of this procedure provided an assignment of a dimensionless weight for each bin and for each species. As a result, a specific pattern of weights for the 1,300 bins was obtained and combined for all species in a weighted bin matrix.
For optimization, the spectral data were partitioned into 5 complementary subsets. One round of cross-validation involved a learning phase on 4 subsets (“training set”) and a validation of the predictive model on the remaining subset (“testing set”). Five rounds of cross-validation were performed by permutation, and the results from the five rounds combined.
To assess the accuracy of the database and calculate its performance in cross-validation, individual spectra were re-used as template for identification. The ASC algorithm compares the acquired spectrum to the specific pattern of each organism/organism group in the database and calculates a percent probability, or confidence value (%ID), which represents the similarity in terms of presence/absence of specific peaks between spectra. A perfect match provides a %ID of 99.9%. %ID >60 to 99.8% are considered as good. Scores <60% are considered to have no valid identification. The VITEK MS system renders the following types of identification results: “Single Choice”, when the spectrum acquired presents a high level of similarity (%ID >60 to 99.9%) with only one specific pattern in the database; “Low discrimination”, when the spectrum acquired presents a high level of similarity with 2 to 4 specific patterns in the database; or “No Identification”, when the spectrum acquired either does not match with any pattern in the database, or presents a high level of similarity to more than 4 specific patterns. During cross-validation, identification was considered as correct when the result was consistent with the reference identification. Low discrimination results were considered as correct if the expected identification was included in the matches. A misidentification was defined as discordant organism identification between the cross-validation result and the reference identification.
External spectra were generated from bacteria cultivated with different growth conditions (media, incubation time, etc) to mimic possible inter-laboratory variations. To reflect clinical laboratory practice, inactivated samples were spotted in duplicate, and analyzed with the updated database. If only one of the two spectra allowed a correct identification, the isolate was considered correctly identified. The cut-off for identification confidence was as described above.
To update the MALDI-TOF MS VITEK database, we used 84 Brucella strains, either reference strains or well characterized clinical/veterinary isolates (Tables 1 and S1), to generate independent spectra covering the Brucella genus. After initial selection based on quality criteria such as peak resolution, signal to noise ratio, number of peaks, absolute signal intensity, and intra-specific similarity, 590 spectra were retained and submitted for biomathematical analyses using an iterative system (bioMérieux patented ASC algorithm).
Using an optimization process, we next evaluated the possibility to discriminate between different Brucella species and biovars. Discrimination between the different species was obtained, with the exception of B. ceti and B. pinnipedialis. These two species could not be clearly separated, as illustrated by the intertwining of their spectra on a dendrogram (Fig 1). Distinguishing the different biovars of B. melitensis and B. abortus was not possible. Discrimination between several of B. suis biovars was obtained (S1 Fig), but biovars 1 and 4 gave cross-identifications.
Classes representing the different Brucella species were thus created by grouping together the different biovars of B. melitensis, of B. abortus and of B. suis, and the two species B. ceti and B. pinnipedialis. The eight species represented in the MALDI-TOF database are thus: B. melitensis (biovar 1, 2 or 3), B. abortus (biovar 1, 2, 3, 4, 5, 6 or 9), B. suis (biovar 1, 2, 3, 4 or 5), B. canis, B. ovis, B. ceti/B. pinnipedialis, B. inopinata and B. papionis.
After optimization, cross validation was performed to evaluate the performance of the updated database, which contains 37,902 spectra covering 1,095 bacterial species including Brucella. This mathematical method is used to assess how accurately the database can perform. Correct identification at the genus level was obtained in 97.29% of cases (Table 3). Importantly, the remaining 2.71% of results corresponded to “no ID”, but never to an incorrect identification. At the species level, the performance varied between the different classes. For the three main zoonotic species (B. melitensis, B. abortus and B. suis), correct identification was obtained with 96.06%, 100% or 89.34% of spectra, respectively.
Finally, as an external validation, the database was challenged with the MALDI-TOF spectra from 48 independent Brucella isolates, and 2 strains of Ochrobactrum, which are “near neighbors” of the Brucella genus (Tables 2 and S2).
The implemented database allowed correct identification at the genus level in 88.4% of cases, all the other results being “No-identification” but never misidentification as another genus (Tables 4 and 5). At the species level, the performances varied. For B. melitensis, B. abortus, and B. suis, correct identification was obtained for 100%, 92.3% or 100% of strains, respectively. It should be noted however that only one extra B. suis isolate was available to be tested in the external validation.
Interestingly, the rare clinical isolate 02/611, described as B. ceti-like after molecular characterization [7], was indeed identified within the B.ceti/B. pinnipedialis class. Also, both the Bullfrog (B13-0095) and the Australian rodent (NF2637) isolates were identified as B. inopinata, in agreement with previous work showing that these belong to the atypical Brucella clade of this genus [27,28]. The two isolates belonging to “B. abortus biovar 7”, a rare biovar of this species, were identified as B. abortus. Finally, the recombinant 16M strain overexpressing the green fluorescent protein (GFP) was correctly identified as B. melitensis using this database. Moreover, using different culture conditions for 16M did not affect its identification by MALDI-TOF MS (S3 Table).
A major asset of this MALDI-TOF MS database is its ability to identify Brucella isolates at the species level, which is essential for following epidemiological outbreaks. Obtaining such a resolution was very challenging for this genus, as highlighted in previous studies [29], because of the high similarity between species at the genetic level [30]. Discrimination between species was made possible using a patented approach to differentiate closely related species using internal calibration and a two-step algorithm. This was not sufficient to distinguish the two species of Brucella from marine mammals (B. ceti and B. pinnipedialis). This is in agreement with a recent Multi-Locus Sequence Analysis (MLSA) showing that the taxonomy is inconsistent with the phylogeny of these two species, and that taxonomic rearrangement should be envisaged [31]. This MALDI-TOF MS database is however able to discriminate eight different Brucella species, which include the most common in human or animal disease.
The updated database allowed correct identification of Brucella isolates at the genus level in 88.4% of cases. It is important to mention that none of them was identified as Ochrobactrum spp., a misidentification that is common with other standard identification methods [32–34] and recently reported using the VITEK MS database currently available [35]. Analysis at the species level gave only one discordant result, corresponding to cross identification between two Brucella species. Such result would have no consequence for human medicine, as identification at the genus level is sufficient to prescribe the appropriate treatment. As for all MALDI-TOF databases, the limitation of this system is its inability to identify non-clinically validated species or species not included in the database. However, the large coverage of the Brucella genus (in particular the most common species) in this database makes this risk is very minor.
Diminution of the performance at the genus and/or species level was due to “no ID” results for some rare and/or atypical Brucella spp. (B. neotomae strain 5K33, B. microti strain CCM4915, and the rodent isolate NF2653), several strains from marine mammals, and the vaccine strain B. abortus RB51. These results were not due to the quality of MALDl-TOF spectra, which was good (based on the number of spectral peaks, Table 5). In the spectra for RB51, we found that several masses characteristics of the B. abortus class were less frequently present, in particular the masses of 5,920.63, 6,040.32 and 7,467.89 Da were present in only 14.3% of spectra (vs. in 75–95% of the spectra of other B. abortus isolates, with a tolerance of 800 ppm). The only discordant result in our assay was obtained with B. canis Mex51, which was identified as B. suis. This was due to the presence in its spectra of additional masses that are common with the B. suis class in addition to the major peaks characteristics of the B. canis class. This finding is consistent with an exhaustive MLSA showing that B. canis strains are very close to B. suis biovars 3 and 4 [31].
Importantly, the MALDI-TOF database allowed the correct identification as Brucella of several recently discovered “atypical” isolates [5,6,28,36]. These strains represent a serious problem for diagnosis laboratories, as they are not identified as Brucella using classical phenotypic tests. It is possible that similar strains have been isolated in the past but misidentified. Very little is known concerning the ability of these new species to cause disease in humans or livestock. The possibility to identify these isolates as Brucella will thus be important for both human and animal health.
Overexpression of an exogenous protein (GFP) did not affect the identification of B. melitensis 16M. This is important since recombinant Brucella strains are common tools in research laboratories and could potentially infect lab workers. Moreover, the use of such Brucella strains as vaccines was proposed, since the presence anti-GFP antibodies would allow distinguishing vaccinated animals from naturally infected ones [37].
In conclusion, this updated MALDI-TOF MS database is a new diagnostic tool that allows the identification of Brucella. It combines precision of identification (broad coverage of the Brucella genus together with species-level identification) and widespread availability. After integration in the VITEK MS (v3.2), this will be the first Brucella database validated for diagnostic with CE accreditation and accessible to all users in routine. This will allow accurate diagnosis and timely treatment in brucellosis. These highly infectious pathogens also causing one of the most frequent laboratory-acquired infection [38], their rapid identification by MALDI-TOF MS will decrease the risk of accidental infection of laboratory workers. A paradox of global health however is that the countries where brucellosis is endemic may not have access to MALDI-TOF MS. This could be circumvented by the use of the in-tube inactivation method described earlier [19], which will allow the shipment of erstwhile infectious samples to mass spectrometry platforms.
|
10.1371/journal.pcbi.1002962 | Computational Predictions Provide Insights into the Biology of TAL Effector Target Sites | Transcription activator-like (TAL) effectors are injected into host plant cells by Xanthomonas bacteria to function as transcriptional activators for the benefit of the pathogen. The DNA binding domain of TAL effectors is composed of conserved amino acid repeat structures containing repeat-variable diresidues (RVDs) that determine DNA binding specificity. In this paper, we present TALgetter, a new approach for predicting TAL effector target sites based on a statistical model. In contrast to previous approaches, the parameters of TALgetter are estimated from training data computationally. We demonstrate that TALgetter successfully predicts known TAL effector target sites and often yields a greater number of predictions that are consistent with up-regulation in gene expression microarrays than an existing approach, Target Finder of the TALE-NT suite. We study the binding specificities estimated by TALgetter and approve that different RVDs are differently important for transcriptional activation. In subsequent studies, the predictions of TALgetter indicate a previously unreported positional preference of TAL effector target sites relative to the transcription start site. In addition, several TAL effectors are predicted to bind to the TATA-box, which might constitute one general mode of transcriptional activation by TAL effectors. Scrutinizing the predicted target sites of TALgetter, we propose several novel TAL effector virulence targets in rice and sweet orange. TAL-mediated induction of the candidates is supported by gene expression microarrays. Validity of these targets is also supported by functional analogy to known TAL effector targets, by an over-representation of TAL effector targets with similar function, or by a biological function related to pathogen infection. Hence, these predicted TAL effector virulence targets are promising candidates for studying the virulence function of TAL effectors. TALgetter is implemented as part of the open-source Java library Jstacs, and is freely available as a web-application and a command line program.
| While it had already been discovered that transcription activator-like (TAL) effectors from Xanthomonas pathogens act as transcription factors in the host plant, deciphering the modular code of DNA binding specificity of TAL effectors in 2009 fascinated the scientific community. This modular code opens the possibility to identify virulence targets of natural TAL effectors in host plants including valuable crops. Knowing these targets deepens our understanding of the role of TAL effectors in virulence. At the same time, it is an opportunity to create resistant plants by destroying TAL effector target sites, indispensable for the pathogen, in plant genomes. However, computational methods are needed to effectively scan full genomes or promoteromes for putative target sites. Hence, we propose TALgetter, a new approach for predicting TAL effector target sites. Using TALgetter, we predict target sites of Xanthomonas TAL effectors in the important crop plants rice and sweet orange. Besides novel putative virulence targets of several TAL effectors, we also gain new insights into the biology of TAL effector targeting. The predictions of TALgetter reveal that target sites are preferentially located in the vicinity of the transcription start and that many TAL effectors bind to the TATA-box in the promoters of target genes.
| The DNA-binding domain of transcription activator-like (TAL) effectors is unique in its modular DNA-specificity. Natural TAL effectors are potent virulence proteins from plant-pathogenic Xanthomonas bacteria that are injected into eukaryotic host cells where they function as transcription factors [1]. Specific DNA-binding of TAL effectors is mediated by highly conserved tandem repeats composed of usually 34 amino acids. Each repeat recognizes one base pair in a contiguous, non-overlapping fashion. DNA-specificity is determined by two amino acids per repeat at position 12 and 13, termed repeat-variable diresidues (RVDs) [2], [3]. Structures of TAL effector-DNA complexes showed that amino acid 13 interacts with the sense strand DNA base whereas amino acid 12 stabilizes the repeat arrangement [4], [5]. Individual RVDs have specificities for individual DNA bases or combinations thereof [2], [3]. Different RVDs contribute differently to the transcriptional activation by TAL effectors [6]. Typically, natural TAL effector target sites are directly preceded by the nucleotide T, while some target sites also have a C or an A at that position [2], [3], [7]–[10]. The modular repeat architecture allows a rearrangement of TAL effector repeats to easily generate any desired DNA-specificity. Accordingly, TAL effectors were adopted as a preferred biotechnology tool for targeted DNA binding [11]–[13]. Fusion of the TAL effector repeat domain with nuclease, activator, and repressor domains yielded highly specific mutagens, gene switches, and repressors, respectively [11]–[21]. Decoding the DNA specificity of TAL effectors also opens the possibility to identify virulence targets of natural TAL effectors in host plants including valuable crops. The computational prediction of TAL effector target sites in host genomes is a key step to provide candidates for subsequent experimental validation.
The recognition of signals in nucleic acid sequences, such as transcription factor binding sites, splice sites, or translation initiation sites, is one of the major fields of computational biology since the seminal work of Berg and von Hippel [22]. Berg and von Hippel propose a statistical-mechanical model where each base pair contributes independently to the total binding affinity of a DNA-binding protein. The same independence assumption is imposed by Stormo et al. [23], who use a scoring matrix learned by the perceptron algorithm for predicting translation initiation sites, and Staden [24], who estimates the entries of a position weight matrix as relative frequencies of nucleotides in a training data set.
Berg and von Hippel already note that the independence assumptions of position weight matrices are most likely not satisfied. First order Markov models or weight array matrix models [25], [26] address this issue and additionally model dependencies between neighboring positions of binding sites. Dependencies between neighboring positions are also taken into account by a special profile hidden Markov model proposed by Salama and Stekel [27] for predicting transcription factor binding sites. Higher order Markov models, which capture dependencies on a larger number of adjacent positions, are employed by Grau et al. [28] for the prediction of transcription factor binding sites and by Yakhnenko et al. [29] for predicting subcellular localization signals.
Dependencies between non-adjacent binding site positions are represented by Bayesian networks [30], Bayesian trees [31], [32], permuted Markov models [33], and Markov random fields [34]. All models capturing dependencies to other binding site positions share the disadvantage that the number of parameters increases exponentially with the number of positions considered. This problem is addressed by variable order Bayesian networks [35], variable length permuted Markov models [36], and hybrid-order models [37], which locally adapt the number of positions considered.
In principle, all of these models could also be employed for the prediction of TAL effector target sites, where a direct application would require to learn distinct parameters for the target sites of each TAL effector. However, the number of validated target sites of individual TAL effector is currently not sufficient to reliably estimate the parameters of any of these models. More importantly, such an approach would render the target site prediction for TAL effectors with currently unknown targets impossible. The ab-initio prediction of zinc finger transcription factor binding sites poses similar problems, which are addressed by an approach Kaplan et al. [38] specifically designed for that class of transcription factors. Regarding TAL effectors, this issue is addressed by several approaches specifically designed for the prediction of TAL effector target sites, which are outlined in the following.
We give an overview of current tools for the prediction of TAL effector target sites and TAL effector nuclease target sites in Table 1. Target Finder of the TALE-NT 2.0 suite [18], [39] predicts target sites of a TAL effector based on its RVD sequence. To this end, Target Finder represents RVD-dependent binding specificities as probabilities for the individual nucleotides, which are hard-wired into the code. These probabilities are combined as columns of a TAL effector-specific position weight matrix (PWM) model, which is used to scan user-supplied input sequences, promoteromes, or genomes for putative target sites. At the 5′ end of the target site, the user may either choose to restrict predictions to those with a preceding T, or to allow nucleotide C. Target Finder is available as a web-server and a stand-alone command line application, where the latter is published under an open-source license.
Storyteller and TALVEZ are provided as a web-server and stand-alone application as well. However, the methods behind both approaches are not published, yet, and are accessible only on e-mail request. For these reasons, we do not consider Storyteller and TALVEZ in the remainder of this paper. Paired Target Finder and idTALE use RVD-dependent binding specificities to predict target sites of TAL effector nucleases, which function as homo- or hetero-dimers to specifically cut genomic DNA. While Paired Target Finder is available as a web-server and command line application, idTALE is only available as a web-server and can only be applied to pre-defined input data sets. Both approaches are applicable to TAL effector nucleases but not to TAL effectors.
In this paper, we propose a new statistical model for the prediction of TAL effector target sites, which represents importance of RVDs and their binding specificity independently. The concept of importance is related to the efficiency of RVDs reported by Streubel et al. [6]. However, while efficiency denotes the positive contribution of specific RVDs to the transcriptional activation by TAL effectors, importance additionally affects the penalty for non-matching nucleotides in a target site. We model the importance of RVDs by a binary hidden variable that represents interaction or non-interaction of an RVD with the corresponding nucleotide. Important RVDs are assumed to interact with the DNA in the majority of cases and, hence, should obtain a high probability of interaction. In case of interaction of RVD and nucleotide, binding specificities are represented by probabilities for the interacting nucleotides that depend on the corresponding RVD. If an RVD does not interact with the DNA, the probabilities of nucleotides are determined by the genomic context. In the proposed model, neither the importance nor binding specificity of an RVD depends on the position of the repeat or on other RVDs in the TAL effector. These assumptions allow for a model with an acceptable number of parameters, which is independent of the number of repeats in a TAL effector. In contrast to previous approaches, the parameters of the proposed model are computationally estimated from known pairs of TAL effectors and target sites. This allows for a rapid and automatic adaption of the model parameters as new target sites are validated. We call the tool using this new approach TALgetter – TAL effector target site finder. TALgetter is implemented within the open-source Java library Jstacs [40], and will be part of the next public release. A web-application of TALgetter is available at http://galaxy.informatik.uni-halle.de, and can also be installed in a local Galaxy [41]–[43] server. In addition, we provide a command line version of TALgetter at http://jstacs.de/index.php/TALgetter. Both, the web-application and the command line application, also allow a user to estimate new model parameters from custom training data. Hence, users can adapt the parameters of the TALgetter model to improved sets of validated TAL effector target sites, which are to be expected in the near future.
The mechanism of transcriptional activation by TAL effectors is still not fully understood. However, there are indications that the presence of a suitable TAL effector target site in a promoter is not always sufficient to induce transcription of the downstream gene [3], [9], [44], [45]. Likely, other factors, e.g., promoter elements surrounding the target site, are required for efficient transcriptional activation, too. Since these factors are yet unknown, they cannot be incorporated into a computational model. Hence, we propose to currently assist the search for functional target sites by experimental approaches to measure activation. In this paper, we use gene expression microarray data of Oryza sativa (rice) and Citrus sinensis (sweet orange) measured after infection with different Xanthomonas strains, and gene expression microarray data of transgenic Arabidopsis thaliana lines endogenously expressing a TAL effector for this purpose.
The remainder of this paper is structured as follows. In the section Materials and Methods, we define the proposed statistical model, and introduce public and in-house gene expression microarray data sets, as well as sequence data used in our studies. The Results section is split in several parts. First, we investigate the capability of our approach to predict known target sites of different TAL effectors in rice. In a second part, we compare the prediction accuracy of TALgetter to the Target Finder of TALE-NT based on gene expression microarray data. In a third part, we scrutinize the RVD binding specificities and importances estimated for the proposed model. We then investigate properties of TAL effector target sites, namely positional preference and the relationship to core promoter elements, revealing novel insights into the biology of TAL effector target sites. Finally, we predict several new putative TAL effector target sites in Oryza sativa and Citrus sinensis, which are supported by gene expression data.
In this section, we define the statistical model used by TALgetter, we describe how the parameters of this statistical model are estimated from training data, and we explain how the trained model is used to scan genomes, promoteromes or other input sequences for putative target sites. Subsequently, we describe the gene expression data obtained from microarray experiments and sequence data used in the studies of this paper.
The statistical model employed by TALgetter is defined by its likelihood, which is derived in the following. Let be an input DNA sequence of length , where represents the nucleotide at position of the sequence, and . Let be the alphabet of known RVDs, and let with be the RVD sequence of the TAL effector of interest. For each position , we model the potential interaction of nucleotide and RVD , while nucleotide directly preceding the interacting positions is modelled independently of the RVD sequence.
We can decompose the likelihood of input sequence given the sequence of RVDs y and model parameters as(1)where is the probability of nucleotide given all previous nucleotides, the complete RVD sequence , and all model parameters .
Since a strong preference for nucleotide T at position directly preceding the target site at the 5′ end has been observed [2], while nucleotides C and A are accepted in natural targets as well [9], [46], this position is included in the model. We assume that the nucleotide preference at position does not depend on the RVD sequence or the specific TAL effector and, hence, define(2)where denotes the parameters of the model at position .
As motivated in the introduction, we model binding specificity and importance of an RVD independently. In addition, we impose several independence assumptions:
These assumptions may be formulated as a local mixture model for each position reflecting interaction vs. non-interaction of RVD and DNA by a binary hidden variable with values and , respectively. Let be the probability that RVD interacts with the DNA, and let be the converse probability that no interaction occurs. Finally, let be the probability of nucleotide given an interaction of RVD with the DNA at the corresponding position, and let , be the probability of nucleotide given its context under the condition that no interaction occurs. The context does not extend beyond the 5′ end of the sequence , and, hence, the context considered may be shortened. With these definitions, we have(3)where denotes the parameters of the mixture probabilities, denotes the parameters of the RVD-dependent component, and denotes the parameters of the RVD-independent component, and . For all subsequent studies, the order is fixed to .
In this section, we examine known TAL effector target sites in O. sativa, and analyze if these target sites are recovered by TALgetter. To this end, we consider two settings. In the first setting, we consider each TAL effector in turn and exclude all TAL effectors with RVD sequences identical to the TAL effector considered from the training set in a cross validation-like manner. For testing, we scan the standard region of 1 kb upstream of the start codon of all rice genes and rank the predictions of TALgetter according to the corresponding likelihood. We refer to this setting as TALgetter CV.
In the second setting, we use the final version of TALgetter, where we use the complete training data. This version is available as a web-application and command line program. For testing, we scan regions from 300 bp upstream the transcription start site (TSS) to 200 bp downstream the TSS or the start codon, whichever comes first. This choice will be motivated in the section Positional preference of target sites. We refer to this setting as TALgetter final.
The ranks of the known TAL effector target sites achieved in these two settings are listed in Table 4. For TALgetter CV, we find the known target sites of Tal1c [3], PthoXo6 [2], PthXo7 [2], and TalC [9] among the top 10 promoterome-wide predictions of TALgetter, while the known target site of PthXo1 [2], [3] is predicted at rank 20. For TALgetter final, we find the known target sites of all these TAL effectors at rank 1 or 2.
For the known target sites of PthXo3 and AvrXa7, the achieved ranks in both settings are considerably worse. Interestingly, these two TAL effectors contain atypical, long repeats, which might influence the overall binding of the TAL effector. Since such atypical repeats are not specifically modelled by TALgetter, this might explain the high ranks of the true target sites of PthXo3 and AvrXa7. However, once the impact of long repeats on TAL effector binding are understood, these could be implemented in the modular structure of TALgetter. Notably, the same effect can also be observed for Target Finder, where the known target sites of PthXo3 and AvrXa7 obtain ranks 558 and 1543, respectively. We compare the prediction accuracy of TALgetter and Target Finder in more detail in the next section.
In the first part of this section, we consider public gene expression microarray data of O. sativa studying the effects of Xanthomonas infections on the transcriptome. Using these gene expression data, we determine sets of genes that are up-regulated upon Xanthomonas infection. Under the assumption that a considerable subset of these genes is directly up-regulated by TAL effectors, we compare the number of predicted target sites of TALgetter and Target Finder that are consistent with the observed up-regulation. Besides virulence targets, these up-regulated genes presumably also include collaterally induced genes. When predicting target sites with TALgetter, we use the TALgetter CV setting described in the previous section and exclude all TAL effectors with RVD sequences identical to the current one from the training set. For Target Finder, we use the publicly available version (https://boglab.plp.iastate.edu/node/add/talef-off, https://github.com/njbooher/boglab_talesf) having fixed binding specificities, which might include knowledge from known target sites of a TAL effector considered. In the second part, we repeat this analysis for genes that are up-regulated in A. thaliana plants that endogenously express the TAL effector Hax2.
We visualize the binding specificities and importances of the different RVDs in Figure 5. Considering the nucleotide preferences at position 0 shown in panel (A) of Figure 5, we find that the most frequent nucleotide with a probability of 0.829 is T, followed by C with a probability of 0.100, A with a probability of 0.049, and G with a probability of , which is in accordance with previous findings [2], [3], [7], [8].
Turning to the binding specificities of RVDs, we find the highest specificities for HD (C), NG (T), NH (G), NI (A), and NK (G). This is in accordance to the experimentally determined DNA-specificities of these RVDs [2], [6], [13], [21], [57].
For HD, NG, NH, and NI, we also find a high importance. Hence, these RVDs are highly specific and mismatches according to the binding specificities are hardly tolerated. Other RVDs with a notably high importance are HN, NN, NP, NS, and NT, although these RVDs are less specific than the other high-importance RVDs.
As noted in the introduction, the concept of importance of RVDs is related but not identical to the efficiencies of RVDs as proposed by Streubel et al. [6], which classifies RVDs as strong, intermediate, and weak, respectively. TAL effectors with exclusively weak RVDs can not activate transcription of downstream genes, even if all binding specificities are matched [6]. Inclusion of three or more strong RVDs renders the TAL effectors fully functional. In contrast, more intermediate RVDs (e.g. six) are needed for full activity. The different efficiencies of RVDs likely reflect different DNA-binding strength and thereby affect overall TAL effector affinity to DNA.
The two RVDs classified as strong, namely HD and NN, also receive the highest importance in the TALgetter model. The intermediate RVDs NS and NH are assigned a fairly high importance as well, whereas the remaining intermediate RVDs, namely NP, HN, and NT, receive a lower importance. The RVDs NG and NI are assigned an importance comparable to that of the intermediate RVD NP, although these RVDs are classified as weak according to their efficiency. This result may be an effect of the related but different concepts of efficiency and importance, which we discuss in the following. An RVD with a low efficiency might prevent transcriptional activation in general, whereas a low importance has the effect that the binding specificities modeled by TALgetter for this RVD have a reduced influence on the overall score. An RVD with high efficiency has a strong positive influence on the transcriptional activation, whereas the contribution of an RVD with a high importance to the total score highly depends on the specificity. Hence, importance affects the penalty that is imposed if the binding specificity is not fulfilled, i.e., a nucleotide with a low probability for a specific RVD is present in a target site.
For some RVDs (HN, NN, NS, NT, N*), we observe a preference for more than one nucleotide, where we recognize gradually decreasing specificities. The most prominent example of this class of RVDs is N*, where we find a preference for C with a probability of 0.693, followed by T 0.272, and very low probabilities for A and G.
The RVD N* has experimentally been determined to specify for T and C with preference for T [6]. A preference for T is also expected, because RVDs are directly followed by a conserved glycine in the repeat sequence and, hence, N* might exhibit a binding preference similar to NG [5]. However, N* recognizes T and C in known TAL effector target sites instead [3], which differs from the RVD NG. In contrast, H* shows a binding preference that is similar to HG.
Figure 5 demonstrates that TALgetter correctly estimates the known specificities of RVDs [2], [3], [6], [13], [57]. Only amino acid 13 of each TAL effector repeat, i.e., the second amino acid of the RVD, interacts with the DNA base and should therefore be responsible for the RVD specificity [4], [5]. Accordingly, the parameters of TALgetter reflect that RVDs containing the same amino acid 13 have comparable specificities. Notable exceptions from this rule are the estimated binding preferences of HA and NA, and HI and NI, which can be explained by HA, NA, and HI being underrepresented in the training data set as shown in panel (C) of Figure 5.
For RVDs NC, YG, HH, IS, SS, NV, and S*, a uniform preference was estimated by TALgetter, since these RVDs are neither present in the training data nor do we have prior knowledge about their binding preference. However, under the assumption that only amino acid 13 of a repeat defines binding preference, we can overcome this issue by estimating a common binding preference for all RVDs with the same 13th amino acid. In this case, the probability (cf. section Materials and Methods) does not depend on the full RVD , but only on its second amino acid. Due to the modular structure of the TALgetter model, we may still estimate an individual importance for each RVD. We refer to this variant of TALgetter as TALgetter13.
The parameters estimated with these modifications are visualized in Figure 6. The binding preferences of the most prominent RVDs, namely HD, HG, NG, NI, NN, NS, and N* [5], remain highly similar to those estimated conditional on the complete RVDs (cf. Figure 5). By specification, the differences between HA and NA, and HI and NI are resolved, and the estimated binding preference is dominated by the more prominent RVD. For YG, HH, IS, SS, and S* that were assigned a uniform binding preference before, we gain binding preferences that are based on the preferences of the other RVDs with same amino acid 13. The importance of individual RVDs is only marginally affected by the modified binding preferences, with a slight decrease of binding importance for HG and H* and a slight increase for NG, NI, NS, and N*.
To investigate if the modified binding preferences influence the prediction accuracy of TALgetter13 compared to TALgetter, we repeat the assessment of prediction accuracy in complete analogy to the previous comparison to Target Finder. The results of this comparison are presented in supplementary Figure S4. We find that for 9 of the 32 combinations of rank cutoff and data set, TALgetter13 yields an improved prediction accuracy compared to TALgetter, whereas for 8 combinations the opposite is the case. For the remaining 15 combinations, both variants of TALgetter achieve an identical number of recovered target genes. Since we do not find an improved prediction accuracy for TALgetter13, we use TALgetter throughout the subsequent studies. However, TALgetter13 might be of value if we search for target sites of TAL effectors containing many rare RVDs. Hence, we include it as an option into the web-application and the command line program.
The variable importance of RVDs according to the parameters learned by TALgetter strengthens the observation that RVDs can differ in their efficiency and contribution to overall TAL effector function [6]. Our findings also suggest that the penalty of mismatching RVD-base combinations to overall TAL-binding differs for each individual RVD-base combinations, a concept that is novel.
In the following, we investigate if TAL effector target sites are located in a preferred distance to either the start codon or the transcription start site (TSS) of target genes. To this end, we scan broader regions of upstream sequences, which span from 1 kb upstream of the transcription start site to the start codon as described in section Data, and collect the positions of the top 200 predicted target sites for each TAL effector studied. As in the previous comparison to Target Finder, we define as positives all genes that achieve a log-fold change greater than in the corresponding experiment. We additionally create a set of negatives by extracting all genes with an absolute log-fold change of less than 0.5. The sets of positive and negative genes are then combined with the target site positions collected from the predictions of TALgetter to obtain positive and negative sets of target site positions. In the following, we consider the unions of these sets across all microarray experiments for O. sativa.
We analyze the collected target site positions by conducting a kernel density estimation with a box kernel and a bandwidth of 100. In Figure 7, we plot the density estimates against the relative position to the start codon (left) and against the relative position to the TSS (right), where the density estimates for the positive and negative set are plotted as a green and red curve, respectively.
Considering the relative position to the start codon, we find a clear enrichment of positive target sites compared to negative target sites in a region reaching from the start codon approximately 300 bp upstream. At regions farther than 400 bp upstream of the start codon, the density of false positive predictions according to the microarray experiments is consistently greater than the density of the true positives.
We also find a pronounced positional preference relative to the transcription start site as can be observed from the right panel of Figure 7. A substantial fraction of true positive predictions is located in a region extending approximately from 300 bp upstream to 200 bp downstream of the TSS. Again, we find a greater density of negatives than positives at positions farther than 400 bp upstream of the TSS. For many genes, the distance between TSS and start codon is at most 200 bp, i.e., many genes have 5′ untranslated regions of at most 200 bp. Hence, for these genes positions farther than 200 bp downstream of the TSS are not considered in the predictions and we generally find a low number of predicted target sites at distances greater than 200 bp.
We repeat this analysis for rank cutoffs of 100 and 500 with highly similar results (data not shown).
In the following, we investigate whether this strong positional preference may be exploited to reduce the number of false positive predictions. To this end, we use TALgetter to predict target sites in two modified sets of upstream sequences. First, we predict target sites in the 300 bp upstream sequences relative to the start codon of all O. sativa genes. Second, we extract for these genes the sequences from 300 bp upstream of the TSS to 200 bp downstream of the TSS or the start codon, whichever comes first. As for the benchmark study of section Comparison to Target Finder, we consider rank cutoffs of 10, 20, 50, and 100 on the predictions for a single TAL effector and join the sets of predictions according to the set of TAL effectors expressed by the Xanthomonas strain studied in the corresponding microarray experiment.
In supplementary Figure S6, we present the results of this analysis in complete analogy to Figure 1. We find that the restriction of predictions to 300 bp upstream the start codon does not improve the overall prediction performance. We even observe a decrease of prediction performance for 11 of the 32 combinations of data set and rank cutoff, while an improvement can only be found in 10 cases. We find an improvement if we restrict predictions to the [−300,200]-region around the TSS. Here, we observe an improvement of prediction performance for 17 of the 32 combinations, whereas performance decreases only in 5 cases. Hence, we might conclude that true target sites of TAL effectors are preferentially located at most 300 bp upstream and at most 200 bp downstream of the TSS, and that exploiting this positional preference for predictions by TALgetter increases the number of true positive predictions for relevant rank cutoffs. In supplementary Figure S7, we present and discuss models to explain this positional preference. Since we expect that the discovered positional preference is a general characteristic of functional TAL effector target sites, limiting the search region to the [−300,200]-region around the TSS should reduce the number of false-positives of any approach for TAL effector target site prediction.
In Figure 7, we additionally recognize that the peak of true positive predictions is not centered at the TSS but approximately 50 bp upstream. Since this peak is located in close vicinity to the preferred location of core promoter elements like the TATA-box or the TC-box [58], we scrutinize the relationship of TAL effector target sites and core promoter elements in the next section.
As a first core promoter element, we consider the canonical TATA-box with consensus TATAWA [58], [59]. Approximately 14% of the O. sativa genes contain a canonical TATA-box in a preferred distance of 39 to 26 bp upstream of the TSS. Genes containing a canonical TATA-box often belong to the group of highly expressed genes [58], [60].
We find a canonical TATA-box in the promoters of 1445 of the 10903 unique predicted target genes among the top 200 predictions for all TAL effectors considered. This is in well accordance to the rate of 14% reported by Bernard et al. [58]. Splitting TATA-related and TATA-less predicted target sites into positives and negatives as described in the previous section, we find 142 TATA-related predictions among the positives (38.4%), whereas 288 positive predictions belong to TATA-less genes. For the set of negative predictions, we find 1303 TATA-related (12.4%) and 9170 TATA-less predictions. Hence, the TATA-related predictions are considerably enriched in the set of positive compared to the negative predictions (odds ratio 3.5). This result is highly significant yielding a p-value below in a one-sided Fisher's exact test, which is the smallest possible p-value due to computational precision.
However, TATA-containing genes might be generally enriched in the set of up-regulated genes regardless of TAL effector target sites. Since genes containing a canonical TATA-box are often highly expressed, this could be an effect of the required log-fold change of 1 for experiments 24 hpi. Indeed, the enrichment of TATA-containing genes among all up-regulated genes is highly significant () with an odds ratio of 3.6. Hence, the enrichment of TATA-containing genes in the set of positive predictions is not greater than the enrichment of TATA-containing genes in all up-regulated genes.
Nonetheless, there could be a functional relationship between transcriptional activation by a subset of TAL effectors and the presence of a TATA-box. For instance, some TAL effectors might substitute the TATA binding protein and acquire the transcriptional machinery independently. The latter explanation is supported by known TAL effector target sites that overlap a TATA-box including the known target sites of AvrBs3 [2], PthXo3 [49], AvrXa7 [49], and PthXo6 [7]. The TAL effector AvrBs3 shifts the transcription start site of target genes and it has thus been speculated that TAL effectors might functionally mimic the TATA-box binding protein [10], [46], [61]. In contrast, several TAL effectors which recognize adjacent DNA boxes in an artificial target promoter primarily directed gene expression from the same original start site [44]. Therefore, it is likely that additional plant promoter elements contribute to TAL effector-mediated gene induction. In the following, we investigate if the predictions of TALgetter support a direct binding of TAL effectors to the TATA-box.
In the subset of 142 positive target sites in TATA-containing genes, we find 40 target sites (28%) that overlap the putative TATA-box. In contrast, only 134 of the 1303 (10%) TATA-related negative predictions overlap with the TATA-box. This enrichment is significant, yielding a p-value of in a one-sided Fisher's exact test. Only 33 of the 142 positive targets sites (23%) and 308 of the 1303 negative target sites (24%) are predictions for one of the 13 TAL effectors having TATAWA in their binding consensus. Hence, the enrichment of target sites overlapping the TATA-box can not be explained by an enrichment of TAL effectors with TATAWA in their binding consensus among the positive predictions.
Interestingly, 26 of the 40 TATA-overlapping target sites directly start with the TATA-box, and 12 overlap the TATA-box with an offset of 2, i.e., start with nucleotides 3 to 6 of the TATAWA consensus. Of the remaining two predicted target sites, one overlaps the TATA-box with an offset of 4, and one contains the TATA-box in the middle of the target site. The large number of TAL effector target sites overlapping TATA-boxes might entail an evolutionary advantage for Xanthomonas strains, since mutations in the TATA-box would lead to a change of the transcriptional behavior of the downstream gene and, hence, are disadvantageous for the host plant [49]. In addition, it might help to correctly position the TAL effector with respect to other promoter elements neighboring the TATA-box (cf. supplementary Figure S7).
In addition to the canonical TATA-box, we also examine the enrichment of TATA-variants [58], which are often found in the promoter sequences of housekeeping genes. Interestingly, we do not find an enrichment of genes containing TATA-variants (not including the canonical TATA-box) in the set of positives, where the corresponding p-value of 0.48 in Fisher's exact test is far from significant. Finally, we consider the TC-box [58], i.e., TTCTTC and variants, located in a similar distance to the TSS as the TATA-box. In this case, the enrichment of genes containing a TC-box in their promoters among the positives is not clearly significant (p = 0.031).
We might suspect that the relationship to core promoter elements, especially the observed overlap of predicted target sites with the canonical TATA-box, is the only reason for the positional preference of TAL effector target sites described in the previous section. However, if we remove all genes that contain a canonical TATA-box, a TATA-variant, or a TC-box from the sets of positive and negative genes, and repeat the kernel density estimation for the remaining sets, the overall picture remains unchanged (cf. supplementary Figure S5).
In Figure 8, we finally investigate the relative position of target sites to the core promoter element in TATA-box containing (left) and TC-box containing (right) promoters. Considering the set of genes containing a canonical TATA-box, we find a sharp cluster of target sites in close vicinity to and often overlapping the TATA-box, which can be recognized from the individual positions plotted as green points in the lower part of the plots. This again reflects that direct binding to the TATA-box might constitute one potential mode of TAL effector function. The majority of the remaining target sites is located upstream of the TATA-box. For the TC-box containing genes, we observe a broader cluster of target sites around the positions of the TC-box. Similar to the TATA case – and the set of all target sites – we find the remainder of target sites preferentially located upstream of the TSS.
In summary, we find an enrichment of genes with a promoter containing a canonical TATA-box among the predicted TAL effector targets, but a similar enrichment can be found for all up-regulated genes. Within the subset of TATA-containing genes, the number of target sites that overlap the canonical TATA-box is significantly enriched. The most conclusive explanation of this observation is a functional relationship between transcriptional activation by TAL effectors and the TATA-box.
In the following, we present and discuss a selection of putative target sites of TAL effectors in rice (O. sativa ssp. japonica) and sweet orange (C. sinensis). For O. sativa, we use the refined search region from 300 bp upstream to 200 bp downstream the TSS or the start codon, whichever comes first (cf. Positional preference of target sites). To limit false positives, we only study TAL effectors where gene expression data are available. Promising targets show a low rank in the TALgetter predictions and a significantly induced gene expression in microarray studies. Different TAL effectors are known to target the same or related plant genes which indicates that these host genes constitute major virulence targets, and that the pathogen has evolved different TAL effectors to target them [62]. Therefore, we consider it meaningful, if we predict novel targets that are either related to known virulence targets or a common target of different TAL effectors. We list a selection of predicted target sites in Tables 5 and 6, while a complete list is available as supplementary Table S2.
The first group of targets we consider belongs to the family of nodulin MtN3 genes, of which several members from A. thaliana and O. sativa encode functional sucrose/glucose transporters (SWEETs) [63], [64]. Some SWEET genes have been identified as susceptibility (S) genes whose induction is essential for a successful bacterial infection [65]. TAL-mediated SWEET gene induction and the resulting elevated export of sugars from plant cells is believed to support bacterial proliferation. In pepper, the MtN3-homolog UPA16 is induced by the TAL effector AvrBs3 [10]. The O. sativa SWEET gene Os8N3 (Os08g42350) is a known target of PthXo1 [65], which is predicted by TALgetter on rank 1 and shows log-fold changes above 1 in three of the microarray experiments. Similarly, the best prediction for the TAL effector TalC is the SWEET gene Os11N3 (Os11g31190), a known target as well [9]. Os11N3 is also a known target of PthXo3 and AvrXa7 [49], however their target sites are predicted on a much higher rank (cf. section Recovering known target sites).
In addition to these known target sites, we find a novel SWEET as putative virulence target of Tal7b and Tal8b, which are identical TAL effectors encoded by a duplicated gene in the genome of Xoo [56]. The predicted target gene of Tal7b/Tal8b is Os02g30910, which yields a log-fold change of 4.3 in the PXO99 experiment. The corresponding target site sequence is shown in the fourth column of Table 5. Notably, the nucleotide at position 0 of this predicted target site is G instead of the usually required T. A few exceptions to the invariable initial T have been reported in natural TAL effector target sites (C for TalC in Os11N3 [9]; A for AvrBs3 in UPA25 [10]) and the potential Tal7b/Tal8b target site might be another exception. Like Os8N3 and Os11N3, Os02g30910 belongs to the clade III of the SWEET family, which is the only known sub-family that encodes sucrose transporters, indicating that this specific function is important for Xoo [64,64]. However, a gene with unknown function (Os01g40290, no. 25) is predicted by TALgetter for Tal7b/Tal8b on rank 1. Since this gene yields a log-fold change above 1 in two of the microarray experiments and has been predicted as a target of Tal7b/Tal8b before [3], it is an alternative target candidate.
The second group of targets also addresses the nutrient supply of the pathogen. We identify several putative TAL effector targets that are related to phosphate metabolism. Xoo mutants impaired in the utilization of phytic acid, a storage form of phosphate, are impaired in virulence [66] indicating that the supply of phosphate is important for a successful Xoo infection. The phosphate transporter Os06g29790 is predicted as a putative target of three TAL effectors, namely AvrXa27 (RVDs identical to XOO1134_MAFF), Tal6a, and XOO2158_MAFF with the latter two TAL effectors differing by only one RVD. Os06g29790 is predicted by TALgetter on rank 1 for Tal6a and XOO2158_MAFF, and rank 21 for AvrXa27/XOO134_MAFF, and is supported by two and three microarray experiments, respectively. The predicted target sites of Tal6a and AvrXa27 do not overlap, and the common target is thus not due to general similarities between these TAL effectors. AvrXa27 can also trigger resistance due to induced expression of the resistance gene Xa27, but only in O. sativa ssp. indica and not in ssp. japonica [67]. Tal9d and XOO1132_MAFF differ in one RVD and are predicted to target the promoter of Os10g25310 which encodes OsSPX3, an SPX domain containing negative regulator involved in tolerance to phosphate starvation [68]. High induction of OsSPX3 after Xoo infection is supported by two microarray experiments. An alternative target of Tal9d and XOO1132_MAFF with slightly lower over-expression but rank 1 among the TALgetter predictions is Os08g05910 (no. 26), a peptide transporter.
The third group of target genes contains HEN1, a small RNA pathway component that contributes to plant immune responses [69]. HEN1 methylates the 3′ terminal nucleotide of all classes of small RNA (sRNA) duplexes thereby promoting sRNA stability [70]. The HEN1 gene is a known target of XOCORF_0460 (Tal1c) and a previously predicted target of Tal9a [3]. In addition, we identified HEN1 as a target of XOO1138_MAFF, a TAL effector which differs from Tal9a in 5 of 20 RVDs. HEN1 yields rank 1 among the TALgetter predictions for all three TAL effectors and is up-regulated in 3 different microarray experiments with Xanthomonas strains that express either Tal9a or XOO1138_MAFF. This demonstrates that TAL effectors interfere with the sRNA homeostasis of the host cell. This example also demonstrates that Xoo and Xoc TAL effectors can have common targets and thus common infection strategies despite their different modes of infection as vascular and leaf mesophyll pathogens, respectively.
The fourth group of targets is related to signal transmission. The TAL effectors Tal7a/Tal8a and XOO1998_MAFF have overlapping predicted target sites in the promoter of Os08g07760 with rank 4 and 1, respectively, and two different supporting microarray experiments. This target gene encodes a putative brassinosteroid-insensitive1 associated receptor kinase (BAK1) ortholog from rice. BAK1 is a leucine-rich repeat receptor-like kinase that is involved in both, brassinosteroid and pathogen signal perception [71]. In addition, XOO2127_MAFF is predicted to target and strongly induce Os01g50370, a MAPKKK protein kinase. Elevated expression of BAK1 or MAP kinase pathway components might interfere with signal transmission and cellular responses.
The fifth and largest group of predicted targets contains genes that are related to transcriptional regulation. AvrBs3 induces the pepper basic helix-loop-helix regulator UPA20 to trigger plant cell enlargement and a hypertrophy phenotype [46] demonstrating that TAL effectors can control complex plant responses via induction of regulatory genes. TALgetter predicts the known targets of PthXo6 and PthXo7, the transcription factor TFX1 and the transcription initiation factor TFIIA subunit, at rank 2. In addition, three TAL effectors are predicted to target different genes encoding zinc finger proteins (targeted by Tal9b and XOO1136_MAFF differing in 3 RVDs, and Tal2a) or a gene encoding a helix-loop-helix domain containing protein (targeted by Tal9e/XOO2001_MAFF). The predicted targets of XOO2865_MAFF (Os07g48450), Tal5a (Os04g43560), and XOO1996_MAFF (Os04g52810) encode no apical meristem proteins (NAC proteins), a large family of plant transcriptional regulators that are involved in diverse developmental and abiotic/biotic stress response processes, including drought tolerance. TALgetter also predicts NAC gene targets for Tal9d (Os05g34830) and XOO2127_MAFF (Os05g10620) for which we have described alternative targets above. Several NAC encoding genes are induced after pathogen infection and some repress defense-related gene expression, rendering them good candidates for TAL effector virulence targets [72], [73].
The sixth group of predicted targets comprises members with diverse function. AvrXa10 is predicted to induce two target genes (Os08g09040, Os08g09010) and XOO2127_MAFF one target gene (Os08g13440) that encode cupin domain-containing proteins. The cupin superfamily includes functionally diverse proteins that can be involved in transcriptional regulation, seed storage, enzymatic reactions to protect plants from oxidative stresses, and pathogen infection [74]. For AvrPth3 TALgetter predicts a monocopper oxidase (Os06g46500). It has been reported before that Xanthomonas influences the defense of rice plants by manipulating copper transport [75]. Tal4 and XOO2129_MAFF, which differ in one RVD, are predicted to induce a gene encoding a putative ATPase with unknown function.
We also use TALgetter to predict TAL effector target sites for Xanthomonas axonopodis pv. citri (Xac 306) in Citrus sinensis. A selection of predicted target sites of the four TAL effectors of Xac 306, namely PthA1, PthA2, PthA3, and PthA4, is presented in Tables 7 and 8, while the complete list of predicted target sites is available as supplementary Table S3. The predicted target of PthA1 is a late embryogenesis-abundant (LEA) protein (orange1.1g027210m). This family of proteins is often related to drought [76]. However, members of this family have also been reported to be metal-binding [77], which might indicate a role in copper transport [75]. For PthA2, TALgetter predicts a target gene from the Tetratricopeptide repeat (TPR)-like superfamily. Interestingly, it has been reported that PthA2 and PthA3 interact with TPX, which contains a TPR domain as well and is related to protein folding and activation, in Citrus [78]. Hence, PthA2 might play a role in supplying such an interactor to PthA2 and PthA3. The predicted target of PthA3 is a RAN GTPase, which might play a role in signal transduction. A LOB domain-containing protein (orange1.1g026556m) is the best target predicted by TALgetter for PthA4. According to gene expression data, this gene is highly up-regulated and achieves a log-fold change of 5.7. For these two reasons, we consider this the most promising candidate of the four Xac 306 TAL effectors. LOB domain containing proteins have been shown to act as transcription factors [79].
We aim to experimentally test, if the target sites predicted by TALgetter are valid targets for the corresponding TAL effector. For this, we analyzed the TAL effector AvrXa10, for which the target specificity has been experimentally verified [2], but targets have been unknown, so far. Seven putative target rice genes are both, in the top 100 TALgetter predictions and up-regulated in gene expression studies (cf. Tables 5 and 6, supplementary Table S2). Four of these target sites with predictions ranking at position 6, 38, 41, and 98, respectively, are cloned upstream of a minimal promoter and a promoterless reporter gene in a reporter vector and tested for gene activation in a transient reporter assay in planta (cf. [2], details in supplementary Text S1), and we present the results of this experiment in Figure 9. Three of the four predicted target sites trigger an AvrXa10-dependent transcriptional activation. Both cupin domain-containing genes which we propose as interesting targets for AvrXa10 (Tables 5 and 6) corresponding to ranks 38 and 41, respectively, therefore contain functional AvrXa10 target boxes. In contrast, the target site with rank 98 has more deviations to the optimal AvrXa10 target site than the other three targets, and this reporter is accordingly not expressed by AvrXa10. Our experimental approach demonstrates that TALgetter can indeed predict functional target sites and that the prediction rank gives an indication for the ability to be recognized by a TAL effector.
In this paper, we present TALgetter, a new tool for the prediction of TAL effector target sites. TALgetter uses a local mixture model that models binding specificity and importance of RVDs independently. In contrast to previous approaches, the parameters of this model are estimated from training data and, hence, allow for an easy adaptation to new validated target sites.
We demonstrate that TALgetter is able to identify known TAL effector target sites in rice and we show that TALgetter predicts a greater number of TAL effector targets that are consistent with up-regulation after Xanthomonas infection than Target Finder in a benchmark study using public and in-house gene expression data. In the benchmark study, a substantial fraction of target sites is uniquely predicted by TALgetter and, hence, these potential virulence targets would have been missed using previous approaches.
Scrutinizing the binding specificities learned by TALgetter, we find that for many RVDs, binding specificities are estimated in accordance to the literature. In addition, we observe gradually decreasing binding specificities for some RVDs, which have also been reported by recent experimental studies. Regarding the concept of RVD importance, we find substantially different parameters for the individual RVDs, which gives indication that different RVDs indeed contribute differently to transcriptional activation by TAL effectors.
In subsequent studies using target sites predicted by TALgetter, we discover a strong positional preference of target sites towards the transcription start site. Most true positive target sites are located within a window from 300 bp upstream to 200 bp downstream the TSS. We demonstrate that exploiting this positional preference for predicting TAL effector target sites further improves the overall prediction performance of TALgetter. This finding is of general value for the computational prediction of TAL effector target sites, since it may also help to reduce the number of false-positive predictions of other approaches.
We also study the relationship of TAL effector target sites to core promoter elements. We show that a considerable number of target sites overlaps with the TATA-box, which indicates that TAL effector binding to the TATA-box – and possibly substituting the TATA binding protein – might constitute one mode of transcriptional activation by TAL effectors. These two findings, positional preference and binding to the TATA-box, reveal new insights into the biology of TAL effector target sites that may aid the understanding of transcriptional activation by TAL effectors. For models to explain this observation, see supplementary Figure S7.
Against this background, we discuss predictions of TALgetter in Oryza sativa (rice) and Citrus sinensis (sweet orange). Besides several known target sites, TALgetter also predicts promising targets for many TAL effectors with currently unknown targets. We experimentally demonstrate that TALgetter predicts target sites that are functional in planta.
We make TALgetter available as a web-application at http://galaxy.informatik.uni-halle.de, which can be used without registration. For confidential analyses, this web-application can also be installed in a local Galaxy server. At http://jstacs.de/index.php/TALgetter, we additionally provide a command line program, which can be easily scripted. Web-application and command line program allow for estimating new model parameters from custom training data to account for the rapid emergence of new TAL effector target sites. Since TALgetter is implemented in the open-source Java library Jstacs, it can be easily extended or modified.
|
10.1371/journal.pntd.0005991 | Phylogenetic analysis of simian Plasmodium spp. infecting Anopheles balabacensis Baisas in Sabah, Malaysia | Anopheles balabacensis of the Leucospyrus group has been confirmed as the primary knowlesi malaria vector in Sabah, Malaysian Borneo for some time now. Presently, knowlesi malaria is the only zoonotic simian malaria in Malaysia with a high prevalence recorded in the states of Sabah and Sarawak.
Anopheles spp. were sampled using human landing catch (HLC) method at Paradason village in Kudat district of Sabah. The collected Anopheles were identified morphologically and then subjected to total DNA extraction and polymerase chain reaction (PCR) to detect Plasmodium parasites in the mosquitoes. Identification of Plasmodium spp. was confirmed by sequencing the SSU rRNA gene with species specific primers. MEGA4 software was then used to analyse the SSU rRNA sequences and bulid the phylogenetic tree for inferring the relationship between simian malaria parasites in Sabah.
PCR results showed that only 1.61% (23/1,425) of the screened An. balabacensis were infected with one or two of the five simian Plasmodium spp. found in Sabah, viz. Plasmodium coatneyi, P. inui, P. fieldi, P. cynomolgi and P. knowlesi. Sequence analysis of SSU rRNA of Plasmodium isolates showed high percentage of identity within the same Plasmodium sp. group. The phylogenetic tree based on the consensus sequences of P. knowlesi showed 99.7%–100.0% nucleotide identity among the isolates from An. balabacensis, human patients and a long-tailed macaque from the same locality.
This is the first study showing high molecular identity between the P. knowlesi isolates from An. balabacensis, human patients and a long-tailed macaque in Sabah. The other common simian Plasmodium spp. found in long-tailed macaques and also detected in An. balabacensis were P. coatneyi, P. inui, P. fieldi and P. cynomolgi. The high percentage identity of nucleotide sequences between the P. knowlesi isolates from the long-tailed macaque, An. balabacensis and human patients suggests a close genetic relationship between the parasites from these hosts.
| Anopheles balabacensis has been incriminated as the primary vector of zoonotic simian malaria, P. knowlesi in Malaysian Borneo with a high prevalence recorded in the states of Sabah and Sarawak. In this study, Anopheles spp. were sampled using human landing catch (HLC) method at Paradason village in Kudat district of Sabah. Total DNA was extracted from these specimens, followed by sequencing the SSU rRNA gene of Plasmodium using polymerase chain reaction (PCR) for the detection and identification of Plasmodium. PCR results showed that only 1.61% (23/1,425) of the screened An. balabacensis had either single or double Plasmodium spp infections. The simian malaria parasites isolated from An. balabacensis were P. coatneyi, P. inui, P. fieldi, P. cynomologi and P. knowlesi. Sequence analysis of these Plasmodium isolates showed high percentage of identity within the same Plasmodium sp. group. Consensus sequences phylogenetic tree of P. knowlesi isolates from An. balabacensis, human patients and a long-tailed macaque from the same locality had 99.7%–100.0% nucleotide identity. This study suggests a close genetic relationship between the parasites isolated from these hosts.
| Anopheles species of the Leucosphyrus group have been identified as medically important vectors in Southeast Asia region [1,2]. The Leucosphyrus group has three main subgroups; Hackeri, Leucosphyrus and Riparis subgroups [3], with the Leucosphyrus subgroup further divided into Dirus complex and Leucosphyrus complex [2,4]. In Peninsular Malaysia, three species of the Leucosphyrus group namely An. hackeri, An. cracens and An. introlatus had been incriminated as primary vectors for P. knowlesi [5–7]. However, in East Malaysia, An. latens in Sarawak and An. balabacensis in Sabah had been confirmed as primary vectors for P. knowlesi [8,9].
A study in Cambodia in 1962 has shown that An. balabacensis (identified as An. dirus later [10]) preferred biting human compared to monkeys placed at the ground level, but preferred monkeys at canopy level to monkeys on the ground [11]. A study in Sabah comparing human landing catch (HLC) and monkey baited trap (MBT) at ground level showed that more An. balabacensis were caught using HLC than MBT [12]. Recent studies showed that this species is more active during the early night with a peak biting time between 7 pm to 8 pm [9,13], and also prefers to bite outdoors than indoors [13]. Such biting behaviors coupled with an abundant source of simian malaria parasites in the reservoir long-tailed macaques (Macaca fascicularis) contribute to An. balabacensis becoming an effective vector for transmitting P. knowlesi malaria in Sabah.
Previous studies in Malaysia have shown that the long-tailed macaques harbor at least five species of simian Plasmodium [14,15], all of which have also been detected in An. balabacensis [9,16]. In Sabah, besides P. knowlesi, other simian malaria parasites recorded in An. balabacensis are P. coatneyi, P. inui, P. fieldi and P. cynomolgi [9,13]. Apart from recording these parasites in the mosquitoes, there is limited study on the phylogenetic relationship among these simian malaria parasites found in An. balabacensis, macaques and human.
In this study, we compare the partial nucleotide sequences of SSU rRNA of simian malaria parasites isolated from An. balabacensis caught in Kudat district of Sabah, from macaques as well as human patients with other published sequences of human and simian malaria parasites available in the GeneBank database. Building a phylogenetic tree of these malaria parasites will give us a clearer picture about their genetic relationship especially for P. knowlesi isolated from long-tailed macaque, An. balabacensis and human.
Kudat district, located at the northern tip of Borneo under the Kudat Division, is about 153 kilometers from Kota Kinabalu, the state capital of Sabah. Paradason village where the study was conducted is located in Kudat District and about 50 kilometers from Kudat town (Fig 1). Most of the villagers belong to the Rungus ethnic group who are dependent on small-scale farming (paddy), oil palm and rubber plantations as their primary source of income.
Anopheles mosquitoes were sampled monthly from October, 2013 to December, 2014 using human landing catch (HLC) method. A total 70 nights of sampling were performed starting from 1800 to 0600 hours (12 hours). Two pairs of volunteers were assigned working in shifts at a randomly selected habitat during each night of sampling. Anopheles was lured by the volunteers exposing their legs. The mosquitoes landing on the legs were caught by the volunteers using plastic specimen tubes (2 cm diameter X 6 cm) aided by a flashlight.
The next morning, the Anopheles mosquitoes were killed by keeping them in the freezer (-20°C) for a few minutes, then gently pinned onto Nu poly strip using ultra-thin micro-headless pins. Species identification was done under a compound microscope using published keys [2,17,18]. After identification, each individual specimen was stored separately in a new microfuge tube and transported to Faculty of Medicine & Health Sciences, Universiti Malaysia Sabah for further processing.
Each individual Anopheles specimen was placed separately inside a sterilized mortar and the tissue homogenized using a sterile pestle. The total DNA was extracted from the tissues using DTAB-CTAB method [19] with some modifications (for example: incubation time was reduce to 30 minute instead of overnight and at the final step of precipitation before adding TE buffer, DNA pellet was incubated at 45°C to completely evaporate any residue of ethanol).
First, 600 μl of DTAB solution was added into the mortar and the tissue was ground using pestle until homogenized. Then, the homogenized tissue was transferred into a clean 1.5 ml microfuge tube and incubated at 68°C for 30 min. Subsequently, 600 μl of chloroform was added into the microfuge tube which was inverted ten times to mix the contents and centrifuged at 13,000 rpm for 5 min. Then, 400 μl of the upper aqueous layer was carefully transferred into a new clean 1.5 ml microfuge tube and mixed with 900 μl sterile dH2O and 100 μl CTAB solution by gently inverting the microfuge tube for several times and allowed it to sit at room temperature for 5 min. The mixture was then spun at 13,000 rpm for 10 min. The supernatant was discarded and the DNA pellet was re-suspended in 300 μl of 1.2 M NaCl solution. Total DNA was precipitated by adding 750 μl of absolute ethanol and centrifuged at 13,000 rpm for 5 min. The supernatant was discarded, the DNA pellet washed with 500 μl of 70% ethanol and centrifuged at 13,000 rpm for 2 min. The DNA pellet was incubated at 45°C for 10 min and re-suspended in 30 μl Tris-EDTA (pH8.0) buffer and stored at -30°C.
Presence of malaria parasites in the mosquitoes was detected using nested PCR by targeting the small subunit ribosomal RNA (SSU rRNA) gene of Plasmodium. A PCR primer pair, rPLU1 and rPLU5, was used in first PCR reaction, while another pair (rPLU3 and rPLU4) was used in the second PCR reaction [20]. For internal control, another set of nested PCR was performed separately to amplify the cytochrome c oxidase subunit II (COII) gene of Anopheles [12]. When a mosquito was confirmed positive for malaria parasites, the Plasmodium species was determined using species specific primers. Both PCR reactions were performed with 25.0 μl final volume.
The reaction components were prepared by mixing 5.0 μl of 5X PCR buffer (Promega), 0.5 μl of (10 mM) dNTPs (Promega), 3.0 μl of (25 mM) MgCl2, 1.0 μl of (10 μM) forward and reverse primers, 0.3 μl of (5.0 U/μl) Taq DNA polymerase (Promega), 2.0 μl of DNA template and sterile dH2O to make up to 25.0 μl final volume. After completion of the first PCR, 2.0 μl of the PCR product was used as DNA template in the second PCR. The reaction was carried out using a thermal cycler (T100 Thermal Cycler, BioRad) with an initial denaturation at 95°C for 5 min followed by 35 cycles of denaturation at 94°C for 1 min, annealing for 1 min and extension at 72°C for 1 min and one final extension step at 72°C for 5 min. The annealing temperature was set at optimal temperature for each set of primers (see S1 Table). The PCR products were analyzed on 1.5% agarose gel electrophoresis stained with RedSafe nucleic acid staining solution (iNtRON Biotechnology), and visualized with an UV transilluminator.
The SSU rRNA gene of the five simian malaria parasite species extracted from An. balabacensis caught in Paradason were cloned and sequenced. In addition, we included in the study blood samples from two P. knowlesi patients and two long tail macaques, one infected with P. knowlesi while the other with P. inui. To make the data set larger, we included simian malaria parasites obtained from mosquitoes caught in three other villages (Tomohon, Mambatu Laut and Narandang) in Kudat district from another study.
A new universal forward primer (UMSF) combined with species-specific primers were used to amplify the SSU rRNA gene of Plasmodium. Details of the primers are provided in S2 Table. Preparation of the reaction mixture and the PCR conditions programmed are as described above. After the PCR was completed, the PCR products were purified to remove impurity and excess reaction mixture using MEGA quick-spin PCR & Agarose Gel DNA Extraction System (iNtRON Biotechnology, Korea) according to manufacturer’s procedure.
Cloning the SSU rRNA gene was done using pGEM-TEasy vectors (Promega, USA) and the plasmids were extracted from the transformed E. coli (JM109) using DNA-spin Plasmid DNA Purification Kit (iNtRON Biotechnology, Korea), all according to the manufacturer’s protocol. The extracted plasmid vectors were restricted using EcoRI restriction enzyme (Promega, USA) and sent to AITBIOTECH, Singapore for sequencing. Sequencing was carried in both directions using forward and reverse M13 primers.
The nucleotide sequences of SSU rRNA of 21 Plasmodium isolates in this study were aligned and compared with other SSU rRNA sequences available at the GeneBank database to determine the percentage identity using Basic Local Alignment Search Tool (BLAST) available online at https://blast.ncbi.nlm.nih.gov/Blast.
The SSU rRNA sequences were standardized to a fixed region for analysis based on the UMSF and UNR primers binding sites. Further analysis was performed using MEGA software, version 4.1 [21]. The nucleotide sequences were multi-aligned using ClustalW method [22] incorporated in the software and the number of variable nucleotides within each of the five Plasmodium spp. determined.
Phylogenetic tree was constructed using neighbor-joining method [23] and the evolutionary distances computed using maximum composite likelihood model with a bootstrap test of 1000 replicates [24] and pairwise deletion option. This method was adopted as it takes into account the different rates of evolution or substitution between nucleotides. The selected region for constructing the phylogenetic tree was nucleotides numbered nt81 to nt1041, based on the published P. knowlesi sequence (AY327551) isolated in Kapit Sarawak where there was a large focus of infected people [25]. This region includes the binding sites for universal forward (UMSF, used in this study) and reverse primers (UNR, [26]) of SSU rRNA. In constructing the phylogenetic tree, Theileria spp. (AF162432) was used as the outgroup. Details of the other 66 nucleotide sequences that were used in constructing the phylogenetic tree are given in S3 Table. Both Plasmodium simium (AY579415) and P. brasilianum (AF130735, KT266778) were not included in the sequence analysis as the selected sequence used in this study was not available in GeneBank database.
A second phylogenetic tree was constructed using the consensus sequences of five Plasmodium species found in Sabah to show the relationship between Plasmodium isolates found in the macaque, An. balabacensis and human.
This project was approved by the National Medical Ethics Committee (NMRR), Ministry of Health Malaysia (Ref. NMRR-12-786-13048). All volunteers who carried out mosquito collections signed informed consent forms and were provided with antimalarial prophylaxis during the study period. Blood spots on Whatman filter paper were collected from adult patients by Kudat hospital personnel, after they had signed informed consent forms. This human blood sample collection was also approved by the NMRR (Ref. NMRR–11–4539471). Blood spots on filter paper were collected by wild life department personel from ten wild macaques captured for relocation purposes and kept in cages following the guidelines in the Animals (Scientific Procedures) Act 1986 Code of Practice for the Housing and Care of Animals Used in Scientific Procedures (UK), with the approval from the London School of Hygiene and Tropical Medicine Animal Welfare and Ethical Review Body (AWER, Ref.2012/8N). Fecal samples were not used then as the protocol for storing the samples had not yet been established by primatology group of the research team.
A total of 1,599 Anopheles individuals belonging to ten species were caught during 14 months of sampling (Table 1). Anopheles balabacensis was the dominant species in Paradason village comprising 89.87% of the total catch, followed by An. barbumbrosus (5.75%), An. maculatus (1.38%) and An. donaldi (1.19%).
A total of 1,586 Anopheles mosquitoes (of which 1,425 were An. balabacensis) were tested for presence of malaria parasites using the PCR method. Only 23 An. balabacensis (1.61%) were found to have malaria parasites in them, being infected with one (78.3%) or two simian Plasmodium spp. (Table 2). The single infection was mostly by P. inui (n = 11).
BLAST analysis of 21 SSU rRNA sequences of Plasmodium spp. isolated from An. balabacensis, human and long tail macaques (3 samples of P. coatneyi, 1027–1029 bp; 4 samples of P. cynomolgi, 1015 bp; 3 of P. fieldi, 1039 bp; 6 of P. inui, 1039 bp and 5 of P. knowlesi, 1050 bp) showed high percentage of identity with the simian Plasmodium nucleotide sequences published in the GeneBank database.
The Plasmodium species in Sabah show a high percentage identity within the same species groups (98.4%–99.6%) but less between different species groups. The highest percentage identity (99.6%) was observed between the P. cynomolgi samples isolated from Tomohon, Membatu Laut and Paradason villages, while the least was for P. coatneyi isolates (98.4%) obtained from Narandang and Paradason villages.
The SSU rRNA sequences of Plasmodium spp. from Sabah also show high percentage identity with the same species from other Asian regions. Plasmodium coatneyi sequences showed 99% identity with P. coatneyi isolated from M. fascicularis in Kapit, Sarawak (FJ619094), as well as with CDC (AB265790) and Hackeri (CP016248) strains. Plasmodium cynomolgi sequences showed 99%–100% identity with P. cynomolgi isolated from M. fascicularis in Kapit, Sarawak (FJ619084), and from other macaque species viz. M. radiata (AB287290) of southern India and M. nemestrina (AB287289) from unspecified South-east Asian nation. Similarly, P. fieldi has high percentage identity with P. fieldi isolated from M. fascicularis in Kapit, Sarawak (KC662444). Of interest is P. inui, which not only has high identity (99%–100%) with those isolated in Kapit (FJ619074) but also with P. inui isolated from M. fascicularis from South China (HM032051), Southern Thailand (EU400388) and strain Taiwan II isolate from M. cyclopis (FN430725).
The P. knowlesi samples of Sabah showed 99% identity with P. knowlesi isolated from both human (AY327551) and M. fascicularis (FJ619089) in Kapit, Sarawak, as well as with that from a Swedish traveler who was infected during his visit to Sarawak (EU807923) [27].
The number of nucleotides in the analyzed region for the various Plasmodium spp. are: P. knowlesi 961 bp, P. inui 946, P. coatneyi 942, P. cynomolgi 935 and P. fieldi 934 respectively. Sequence alignment indicated that P. coatneyi has the highest number of variable nucleotides among the isolates (n = 3 isolates; 15 variable nucleotides) followed by P. knowlesi (n = 5; 9), P. inui (n = 6; 7), P. fieldi (n = 3; 5) and P. cynomolgi (n = 4; 4).
Further analysis of the P. knowlesi group using consensus sequences showed that there were three variable nucleotides between P. knowlesi isolated from the long-tailed macaque and human, two between long-tailed macaque and An. balabacensis isolates but none between An. balabacensis and human isolates (Fig 2).
In the phylogenetic tree generated for 13 Plasmodium species infecting monkeys and humans (Fig 3), all the 21 Plasmodium isolates obtained in the study were placed in the correct species group. P. knowlesi group was positioned below P. coatneyi group whereas P. inui, P. fieldi and P. cynomolgi were placed at the upper branches.
In the phylogenetic tree depicting relationship between the five Plasmodium species found in Sabah using consensus sequences, a similar tree topology was also observed (Fig 4). All Plasmodium group except for P. knowlesi group has two branches, each representing the host from which Plasmodium was isolated. However, P. knowlesi group has three branches with the isolates from both An. balabacensis and macaque closer to each other than to the isolates from humans.
In this study, we analyzed 21 nucleotide sequences of partial SSU rRNA of five Plasmodium spp. isolated from An. balabacensis collected in Kudat district of Sabah, infected humans and a long-tailed macaque together with other nucleotide sequences downloaded from the GeneBank database. The results suggest that in Sabah, there is a close genetic relationship between the P. knowlesi specimens in the long-tailed macaques, An. balabacensis and human.
Plasmodium inui appears to be a common simian malaria parasite found in 61% (14/23) of the infected An. balabacensis specimens. This was also the case in other investigations [9,28]. Hitherto, this simian malaria has not become zoonotic to humans yet although it has been proven experimentally to be infective to monkey through the bites of An. dirus [29]. The infection rate of P. knowlesi in An. balabacensis is low (0.14%, 2/1,425) with only two mosquitoes being infected along with other Plasmodium species. Nevertheless P. knowlesi is the dominant Plasmodium species recorded among the human cases in Sabah [30]. These cases were recorded mainly in the rural areas near to forests and also among the workers in the agricultural sector viz. in oil palm estates and vegetable farms [13,31].
Sequence data of the SSU rRNA of Plasmodium confirm that the five species of simian Plasmodium commonly harbored by the wild macaques in Malaysia are also found in An. balabacensis. BLAST results of Sabah’s Plasmodium sequences showed high identity with other simian Plasmodium sequences published in the GeneBank database, especially with the simian malaria parasites in long-tailed macaques in Kapit, Sarawak (FJ619069 and FJ619089). This could suggest that a similar or closely related cluster of simian Plasmodium is circulating among the monkey populations and Anopheles mosquitoes in both Sabah and Sarawak. This is highly plausible as these two states share a common boundary, and there is a continual movement of humans between these two states.
The total number of nucleotides in the analyzed region was different for the five simian Plasmodium spp. in Sabah, with P. knowlesi having a higher number. The differences in total number of nucleotides in the SSU rRNA gene confer a unique signature to each Plasmodium species. Furthermore the presence of conserved and variable sequences in the gene makes it suitable for species identification and phylogenetic study [32,33].
The percentage of identity between consensus sequences of SSU rRNA of P. knowlesi isolates from the monkey, mosquito and man was high (Fig 2). For example, 100% identity was observed between P. knowlesi isolates from An. balabacensis and human, 99.8% between An. balabacensis and the long-tailed macaque, and 99.7% between long-tailed macaque and human. This indicates a great genetic similarity in P. knowlesi found in the long-tailed macaque, An. balabacensis and human populations. However, it is not certain if this would indicate the same cluster of P. knowlesi is circulating between these hosts, since we did not dissect the mosquitoes’ salivary glands to detect for sporozoites, or carry out RT-PCR targeting the specific mRNA transcripts of the sporozoite stage. Thus further study is needed to determine this, using more P. knowlesi positive Anopheles balabacensis and analyzing other polymorphic markers or microsatellite loci of the parasite. Different P. knowlesi haplotypes have been observed in the macaque and human populations in Kapit Sarawak [14] as well as in the human population in Thailand [34].
Overall, the 13 Plasmodium species in the phylogenetic tree can be grouped into two main clusters, one containing the P. vivax/simian malaria parasites while the other human malaria parasites (Fig 3). Although P. simium (AY579415) and P. brasilianum (AF130735, KT266778) were not included in our analysis as their nucleotide sequences in the GeneBank database do not contain the same analyzed region, P. simium is closely related to P. vivax [32]and can be placed in the first cluster, while P. brasilianum is closely related to P. malariae and can be placed in the second cluster. It may be noted that P. cynomolgi, P. fieldi and P. simiovale were not clearly resolved as some of the isolates were grouped in different branches. This could be due to the high percentage of nucleotide identity (99.6%) among these three species.
The consensus tree (Fig 4) of Plasmodium species found in Sabah showed a very close relationship between the Plasmodium isolates from monkey as the reservoir, An. balabacensis as the vector, and human as the case. This is supported by P. knowlesi isolates from these three organisms having high nucleotide identity (99.7–100%).
Currently in Sabah, An. balabacensis is the only species found to carry P. knowlesi. The phylogenetic analysis here indicates that the vector picks up the malaria parasites from monkeys and transmits them to humans when it feeds on them. However, there is a lot more about the transmission dynamics of P. knowlesi that is still unknown and needs to be unpacked. A clearer picture on the interrelationship of simian malaria parasites found in An. balabacensis will help us to understand more about Plasmodium itself. Future research may focus more on the host-vector relationship that requires longer nucleotide sequence analysis so that new informed alternatives for malaria elimination strategy targeting on P. knowlesi as well as other simian malaria parasites may be formulated.
|
10.1371/journal.pbio.1000252 | SUMO Modification Regulates BLM and RAD51 Interaction at Damaged Replication Forks | The gene mutated in Bloom's syndrome, BLM, is important in the repair of damaged replication forks, and it has both pro- and anti-recombinogenic roles in homologous recombination (HR). At damaged forks, BLM interacts with RAD51 recombinase, the essential enzyme in HR that catalyzes homology-dependent strand invasion. We have previously shown that defects in BLM modification by the small ubiquitin-related modifier (SUMO) cause increased γ-H2AX foci. Because the increased γ-H2AX could result from defective repair of spontaneous DNA damage, we hypothesized that SUMO modification regulates BLM's function in HR repair at damaged forks. To test this hypothesis, we treated cells that stably expressed a normal BLM (BLM+) or a SUMO-mutant BLM (SM-BLM) with hydroxyurea (HU) and examined the effects of stalled replication forks on RAD51 and its DNA repair functions. HU treatment generated excess γ-H2AX in SM-BLM compared to BLM+ cells, consistent with a defect in replication-fork repair. SM-BLM cells accumulated increased numbers of DNA breaks and were hypersensitive to DNA damage. Importantly, HU treatment failed to induce sister-chromatid exchanges in SM-BLM cells compared to BLM+ cells, indicating a specific defect in HR repair and suggesting that RAD51 function could be compromised. Consistent with this hypothesis, RAD51 localization to HU-induced repair foci was impaired in SM-BLM cells. These data suggested that RAD51 might interact noncovalently with SUMO. We found that in vitro RAD51 interacts noncovalently with SUMO and that it interacts more efficiently with SUMO-modified BLM compared to unmodified BLM. These data suggest that SUMOylation controls the switch between BLM's pro- and anti-recombinogenic roles in HR. In the absence of BLM SUMOylation, BLM perturbs RAD51 localization at damaged replication forks and inhibits fork repair by HR. Conversely, BLM SUMOylation relieves its inhibitory effects on HR, and it promotes RAD51 function.
| Replication is the process in which cellular DNA is duplicated. DNA damage incurred during replication is detrimental to the cell. Homologous recombination, in which DNA sequences are exchanged between two similar or identical strands of DNA, plays a pivotal role in correcting replication processes that have failed due to DNA breakage and is tightly regulated, because deficient or excess recombination results in genomic instability. Previous studies have implicated the DNA-processing enzyme BLM in the regulation of homologous recombination; BLM is defective in Bloom's syndrome, which is characterized by excess recombination and cancer susceptibility. Here, we show that modification of BLM by the small protein SUMO controls BLM's function in regulating homologous recombination at sites where DNA replication failed. We showed that cells expressing a SUMO-deficient mutant of BLM accumulated more DNA damage and displayed defects in repair by homologous recombination. An enzyme involved in homologous recombination, RAD51, displayed a defect in localization to sites where DNA replication failed. Our data support a model in which SUMO modification regulates BLM's function in homologous recombination by controlling the localization of RAD51 to failed replication sites.
| Homologous recombination (HR) is a high-fidelity DNA repair mechanism that functions to rejoin double-strand breaks (DSBs) and restart broken replication forks. A major outcome of the repair of replication fork damage by HR is the generation of sister-chromatid exchanges (SCEs), which result from resolution of Holliday junctions during HR repair [1],[2]. Predictably, a large number of agents that cause DNA damage increase the frequencies of SCEs [3]–[5]. Bloom's syndrome (BS) is the only clinical entity in which increased levels of SCE are a prominent cellular feature [6]. It is an autosomal recessive disorder, which is characterized by proportional dwarfism, photosensitivity, immunodeficiency, hypogonadism, and predisposition to a wide range of different types of cancer [7]. BS is caused by biallelic null mutations of the BLM gene [8]. The BLM gene encodes a DNA helicase of the RecQ family, which is an evolutionarily conserved group of enzymes that operates at the interface of DNA replication, HR, and DNA repair [9].
The RecQ helicases are DNA-dependent ATPases that can translocate on single-stranded DNA (ssDNA) with 3′ to 5′ directionality [10]. In vitro, they preferentially unwind DNA substrates that resemble recombination intermediates, including G4 tetrahelical DNA, Holliday junctions, double Holliday junctions, and D-loops. A complex consisting of BLM, topoisomerase IIIα, BLAP75, and BLAP18 (BLAPs are BLM-associated proteins) can “dissolve” a substrate representing a double Holliday junction—a late intermediate in HR-mediated repair of DSBs—in such a way that crossing over would not occur between DNA strands [11]–[13]. This activity could provide an explanation for the increased SCEs in BS cells; however, recent genetic and biochemical studies have shown that BLM also has activities in upstream parts of the HR pathway. Because BLM is recruited to damaged replication forks early in the repair process [14]–[16], it could suppress the formation of aberrant recombination intermediates at the replication fork. Such a mechanism has been proposed for Sgs1, the yeast homolog of BLM [17],[18]. BLM interacts directly with the RAD51 recombinase, which is the enzyme that catalyzes homology-dependent strand invasion [19],[20], and in vitro it can displace RAD51 from ssDNA and unwind the invading DNA strand of a D-loop formed by RAD51 [21],[22], suggesting that BLM regulates the formation of D loops. Finally, BLM and Sgs1 each collaborate with exonucleases that process DSBs to generate ssDNA with a 3′ tail, which is the substrate for RAD51 [23]–[26]. Collectively, these data show that BLM has both pro- and anti-recombinogenic functions in HR. A key question that emerges from these studies is how are these different functions of BLM in HR regulated?
Modification by the small ubiquitin-related modifier (SUMO) has emerged as an important regulator of HR [27]. In response to replication fork damage in the budding yeast Saccharomyces cerevisiae, the polymerase processivity factor PCNA (proliferating cell nuclear antigen) is SUMOylated, and PCNA SUMOylation recruits the DNA helicase Srs2 to the fork, which functions to prevent aberrant recombination events between sister chromatids [28]–[31]. Mutants of the SUMO-specific E3 ligase gene MMS21 accumulate RAD51-dependent cruciform structures at damaged replication forks [32], which are aberrant structures that also accumulate in sgs1 deletion mutants [33]. These studies indicate that SUMO modification can play important roles in response to damaged forks; however, the role of SUMO in regulation of HR is not fully understood, and these mechanisms have not been studied in mammalian cells.
We have previously shown that BLM is SUMOylated and that failure to SUMOylate BLM results in changes in BLM's nuclear distribution [34]. Expression in cells of SUMO-mutant BLM, containing lysine to arginine mutations at residues 317 and 331 that prevent SUMOylation, induces excess γ-H2AX foci—a marker for DNA damage and repair—in the absence of exogenously induced DNA damage [34]. Despite the presence of excess γ-H2AX foci and micronuclei in cells that expressed SUMO-mutant BLM, there was insufficient evidence to conclude that SUMO-mutant BLM generated excess DNA damage. Because SUMOylation is known to regulate the localization of proteins in the nucleus [35], we hypothesized that the accumulation of BLM in γ-H2AX foci could result from a kinetic defect in recruitment of BLM back to the promyelocytic leukemia (PML) nuclear bodies (PML-NBs).
In the present study, we aimed to determine how SUMOylation regulates BLM's function in maintaining genomic integrity. We hypothesized that cells that express SUMO-mutant BLM have a DNA repair defect. Characterization of cells that expressed SUMO-mutant BLM revealed that SUMOylation of BLM regulates its association with RAD51 and its function in HR-mediated repair of damaged replication forks. Our data support a model in which SUMOylation of BLM acts as a switch to regulate its effects on recombination.
Because BLM functions at damaged replication forks and γ-H2AX is a marker for DNA damage, we hypothesized that SUMO-mutant BLM is defective in repair of damaged replication forks. To gain insight into this question, we introduced GFP-BLM expression constructs into the BS cell line GM08505, isolated clones that stably expressed either normal BLM (BLM+ cells) or SUMO-mutant BLM (SM-BLM cells), and studied these clones for responses to replication fork damage. We treated SM-BLM and BLM+ cells with 0.5 mM hydroxyurea (HU) for 24 h, which stalls replication forks, and quantified the production of γ-H2AX by immunofluorescence, Western blot, and flow cytometry analyses (Figure 1). The 24-h HU treatment blocks approximately 80% of the cells in S phase, simultaneously providing a primary synchronization of the cells and stressing replication forks through nucleotide deprivation.
As we reported previously, untreated SM-BLM cells exhibited higher levels of γ-H2AX foci compared to untreated BLM+ cells (35.9 vs. 16.4, respectively; Figure 1A). Treatment of SM-BLM and BLM+ cells with HU resulted in a larger increase in γ-H2AX foci per cell in SM-BLM compared to BLM+ cells (a gain of 56.8 vs. 42.7 foci). Particularly notable were the presence of SM-BLM cells that stained brightly with γ-H2AX (Figure 1B). HU induced twice the numbers of γ-H2AX-bright nuclei in SM-BLM cells than in BLM+ cells (Figure 1C). Consistent with the immunofluorescence analysis, by Western blot and flow cytometry analyses, the levels of γ-H2AX in HU-treated SM-BLM cells were higher than in HU-treated BLM+ cells (Figure 1D and 1E).
Because the results could have been influenced by cell-cycle effects, we analyzed nuclear DNA content and measured BrdU incorporation at different times after release from the HU block by flow cytometry. After treatment with HU, in both SM-BLM and BLM+ clones, the majority of cells were blocked in early S phase, and they progressed to mid-S phase by 6 h after release from the HU block (Figure S1). These data indicated that differences in position in the cell cycle or irreversibility of the S phase block did not explain the differences in the accumulation of γ-H2AX after HU treatment of SM-BLM and BLM+ cells.
In summary, these experiments showed that SM-BLM cells exhibited excess phosphorylated H2AX in both untreated and HU-treated conditions. The presence of increased levels of spontaneous and HU-induced γ-H2AX strongly suggests the presence of excess DNA damage.
In order to obtain direct evidence for the presence of DNA damage, we analyzed HU-treated and untreated cells for DSBs by pulsed-field gel electrophoresis (PFGE) (Figure 2). In the absence of treatments, SM-BLM cells exhibited over 1.5 times more DSBs compared to BLM+ cells (Figure 2B). This result confirmed the presence of increased numbers of DSBs consistent with the higher numbers of γ-H2AX foci. After a 24-h treatment with HU, SM-BLM cells again exhibited 1.5 times more DSBs compared to BLM+ cells. Because 80% of the cells are blocked with stalled forks, the DSBs detected in HU treatment conditions likely originate from fork breakage. After release from the HU block, DSBs accumulated over time, with the total number of DSBs observed in SM-BLM cells being greater at each time point than the total number in BLM+ cells (Figure 2B). A 24-h HU treatment induced more DSBs in BS cells compared to no treatment, and BS cells also accumulated more DSBs after release from the HU block compared to either SM-BLM or BLM+ cells (Figure S2). Consistent with the PFGE analysis, the total number of HU-induced micronuclei was greater in SM-BLM compared to BLM+ cells (Figure 2D).
We also analyzed the numbers of DSBs in cells treated with camptothecin (CPT), which generates replication-associated DSBs [36],[37]. Treatment with different concentrations of CPT for 3 h generated two times more DSBs in SM-BLM compared to BLM+ cells, showing that breaks accumulate at an accelerated rate in SM-BLM cells (Figure 2C). Altogether, these data were consistent with the hypothesis that SM-BLM cells have a defect in the repair of replication-associated DSBs.
Because SM-BLM cells exhibited higher levels of DSBs induced by replication damage, we expected SM-BLM cells to be hypersensitive to DNA damage encountered during S phase. To test this hypothesis, we compared the levels of cell death in cells exposed to replication damage, using a standard flow cytometry assay (Figure 3). In the absence of HU or etoposide treatment, BLM+ and SM-BLM cells exhibited similar levels of cell death. A 24-h treatment with HU induced a 3% increase in cell death in BLM+ cells as compared to a 10.6% increase in SM-BLM cells (p = 0.01), demonstrating that SM-BLM cells have increased sensitivity to HU treatment alone. Similarly, a 24-h treatment with etoposide induced an 11.3% increase in cell death in BLM+ cells compared to a 20.7% increase in SM-BLM cells (p = 0.003), demonstrating that SM-BLM cells are also hypersensitive to etoposide treatment compared to BLM+ cells. After HU pretreatment, etoposide induced a 13.5% increase in cell death in BLM+ cells, which was similar to the level of cell death observed without HU pretreatment (11.3%) (p = 0.84), whereas after HU pretreatment, etoposide induced a 43.1% increase in cell death in SM-BLM cells, which was 2-fold greater compared to the level observed without HU pretreatment (20.7%) (p<0.001). As expected, BS cells that lack BLM protein are also hypersensitive to DNA damage encountered during S phase. In corroboration of these results, in colony survival assays, we also observed increased sensitivity of SM-BLM cells to CPT compared to BLM+ cells (Figure S3). These data indicated that SM-BLM cells are more sensitive than BLM+ cells to DNA damage generated during S phase, again consistent with a defect in the repair of replication-associated DNA damage.
Replication-associated DSBs are repaired by HR, which generates increased numbers of SCEs [1]. Because SM-BLM cells exhibited excess HU-induced DSBs, we hypothesized that HR is impaired in SM-BLM cells. Therefore, we tested whether replication stalling induced fewer SCEs in SM-BLM compared to BLM+ cells (Figure 4). Untreated BLM+ and SM-BLM cells showed similar numbers of SCEs (17.4 vs. 16.7 SCEs/46 chromosomes, respectively). However, whereas HU treatment induced a 2-fold increase in SCEs in BLM+ cells (from 17.4 to 29.6 SCEs/46 chromosomes; p<0.001), HU treatment had almost no effect on the levels of SCEs in SM-BLM cells (from 16.7 to 18.5 SCEs/46 chromosomes; p = 0.32). The numbers of HU-induced SCEs in BLM+ compared to SM-BLM cells was significantly different (p<0.001). In contrast to SM-BLM cells, HU induced a large increase in SCEs in BS cells (Figure S4A). These data suggest that HR repair is not engaged normally at damaged replication forks, leading to the excess DSBs that are observed in HU-treated SM-BLM cells.
It is worth noting that in our earlier report on BLM SUMOylation [34], we found that the mean number of SCEs in untreated SM-BLM cells was greater than the mean number in BLM+ cells. This difference was caused by the presence in the SM-BLM cultures of 5% of cells with SCE levels equal to levels typically observed in BS cells, whereas we had detected no cells of this type in the BLM+ cultures. In the present study, we did not detect cells with high SCEs in the SM-BLM cultures; consequently, we suggest that these high-SCE cells were produced by extinction of SM-BLM expression in a small fraction of SM-BLM cells.
RAD51 is a key enzyme in HR repair, and normally it interacts with BLM at damaged replication forks [19],[20]. Because SM-BLM cells exhibited a defect in HR-mediated repair after replication stalling, we examined whether RAD51 and BLM colocalize normally at γ-H2AX–marked damage in SM-BLM cells (Figure 5). In untreated cells, SM-BLM cells contained more RAD51 foci than BLM+ cells (26.9 vs. 18.1 foci/cell). Similarly, as previously noted [34], untreated SM-BLM cells contained more BLM foci (19.0 vs. 11.0 foci/cell) (Figure 5B). However, after treatment with 0.5 mM HU for 24 h, whereas BLM+ cells exhibited a large increase in RAD51 foci from 18.1 to 51.7 foci/cell, SM-BLM cells exhibited only a modest increase from 26.9 to 34 foci/cell (Figure 5B). Consistent with these observations, whereas HU treatment induced substantial increases in RAD51-γ-H2AX and RAD51-BLM colocalization in BLM+ cells, HU treatment induced only a modest increase in these colocalizations in SM-BLM cells (Figure 5C and Figure S5). In HU-treatment conditions, whereas 66% of the γ-H2AX foci contained RAD51 in BLM+ cells (vs. 58% in untreated cells), only 25% of γ-H2AX foci contained RAD51 in SM-BLM cells (vs. 43% in untreated cells—a decrease). As a positive control, we found that RAD51 showed much higher levels of colocalization with γ-H2AX in HU-treated BS cells compared to BLM+ cells (Figure S4B), as previously reported [19].
To distinguish whether the RAD51 localization defect was an early or late effect of replication stalling, we treated BLM+ and SM-BLM cells with 10 mM HU for 1 h and stained cells with antibodies to RAD51 and PCNA (to identify cells in S phase). Whereas BLM+ cells exhibited a 2.3-fold increase in RAD51 foci from 14.4 to 33.6 foci/cell in PCNA-positive cells, SM-BLM cells exhibited no increase in total RAD51 foci from 19.7 to 21.0 foci/cell (Figure 5D). Synchronization experiments with mimosine followed by treatment with 10 mM HU for 1 h corroborated these data (Figure S6). It is worth noting that both BLM+ and SM-BLM proteins localized efficiently with PCNA foci in HU-treated cells, indicating that SUMOylation is not required for normal trafficking of BLM to stalled forks (Figure S7).
Altogether, these data demonstrated that there is a dramatic defect in RAD51's recruitment to and/or retention in repair foci induced by replication stalling. The RAD51 localization defect could explain both the impairment of HR after replication stalling and the excess DSBs observed in SM-BLM cells.
To investigate the mechanism that might explain the RAD51 localization defect, we considered the possibility that RAD51 interacts with SUMO noncovalently, which would facilitate interaction between RAD51 and SUMOylated BLM. To test this hypothesis, we assayed for possible noncovalent interactions between RAD51 and SUMO and also for possible effects of covalent SUMOylation of BLM on its interactions with RAD51 (Figure 6). In an in vitro binding assay, more RAD51 bound to SUMO-coated beads than to controls beads (Figure 6A), showing that RAD51 binds equally well to both SUMO-1 and SUMO-2. To test whether SUMO modification of BLM affects its interaction with RAD51, we incubated RAD51-coated beads with either unmodified BLM or with a mixture of SUMO-2–modified and unmodified BLM and analyzed bound proteins by Western blot with anti-BLM antibodies. Consistent with previous findings [20], unmodified BLM bound specifically to RAD51-coated beads, confirming that BLM and RAD51 interact directly (Figure 6B, left panel). SUMO-2–modified BLM also bound specifically to RAD51-coated beads (Figure 6B, right panel). To evaluate the effect of SUMO-2 modification on BLM's interaction with RAD51, the results of the binding assays were analyzed quantitatively. This analysis revealed that, whereas the ratio of SUMO-2–modified to unmodified BLM in input fractions and fractions bound to control beads was approximately 1.1∶1 and 1.7∶1, respectively, the ratio present in fractions retained on RAD51-coated beads was approximately 5.1∶1 (Figure 6C). These binding ratios reveal that SUMO-2 modification of BLM has a strong, positive effect on its binding to RAD51.
Altogether, these data demonstrated that RAD51 contains a SUMO binding site(s) and that SUMOylation of BLM can affect its interactions with RAD51. The data support the hypothesis that SUMOylation of BLM facilitates repair of damaged replication forks by HR by modulating RAD51's recruitment and/or retention at repair sites.
The data presented here demonstrate the importance of BLM SUMOylation in the repair of damaged replication forks. Failure to SUMOylate BLM resulted in excess damage-induced repair foci, DSBs, and hypersensitivity to DNA damage. Importantly, in SM-BLM cells, replication stalling by HU did not stimulate HR as measured by SCEs, suggesting a defect in RAD51 function. Consistent with these data, RAD51 failed to accumulate at stalled forks. SUMO-mutant BLM exerts a dominant effect, because excess γ-H2AX foci were induced by expression of SUMO-mutant BLM in HeLa cells, which express endogenous normal BLM [34]. Moreover, we showed here that several SM-BLM phenotypes differed from BS phenotypes, such as the presence of excess H2AX phosphorylation in untreated cells, the RAD51 localization defect, and the lack of HU-induced SCEs. We found that RAD51 is a SUMO-binding protein, implicating BLM SUMOylation in recruitment of RAD51 to repair sites through a mechanism involving noncovalent SUMO interactions. Our findings demonstrate that BLM SUMOylation regulates the recruitment and/or retention of RAD51 to damaged replication forks, and it is important in HR-mediated repair.
The steady-state levels of DSBs in cells are a function of the rate at which DNA damage accumulates and the rate at which it is repaired. We observed that SM-BLM cells exhibited greater numbers of DSBs than BLM+ cells under a variety of conditions (Figure 2). For example, SM-BLM cells exhibited more HU-induced DSBs, which arise due to breakage of stalled forks, and more CPT-induced DSBs, which arise due to replication runoff at sites where topoisomerase-cleavage complexes are bound to the DNA [38]. Formally, the presence of increased levels of DSBs in SM-BLM cells could result from an increase in the rate at which DNA damage accumulates (due to excess numbers of replication forks, increased numbers of topoisomerase cleavage complexes, or a failure to process aberrant replication intermediates) or from a decrease in the rate of DNA repair (due to a failure to recruit RAD51). We noted that the rate of DSB accumulation after release from the HU block was the same in both SM-BLM and BLM+ cells, indicating that the rate of breakage exceeds the rate of repair under these conditions in both types of cells. Consistent with these observations, the numbers of γ-H2AX foci in both BLM+ and SM-BLM cells increase 6 h after release from the HU block (unpublished data).
Recent work has shown that BLM is present on a class of ultrafine anaphase bridges [39], and it acts to separate interlinked DNA strands especially at loci with intrinsic replication difficulties [40]. BS cells consequently have a defect in separation of sister chromatids, resulting in more anaphase bridges; some fraction of the increased DSBs that arise in BS cells no doubt traces to breakage at sites of underreplicated DNA. BS cells exhibit an inadequate response to replication stress [16],[41]–[43], in which additional forks are initiated apparently to compensate for forks that have collapsed [44],[45]. One possible explanation for the excessive numbers of γ-H2AX foci and DSBs in treated and untreated SM-BLM cells is that these cells have increased replication difficulties, as BS cells do, and they compensate by activating additional replication forks, which are concomitantly more likely to break after replication damage. According to this view, BLM SUMOylation helps prevent the collapse of replication forks in regions with hard-to-replicate DNA, perhaps by stimulation of BLM's activity in subverting aberrant recombination intermediates at stalled replication forks.
Alternatively, BLM SUMOylation could promote HR-mediated repair of broken forks through the recruitment and/or retention of RAD51 at damaged forks. RAD51 is the DNA recombinase essential for HR-mediated DNA repair, and previous studies have demonstrated that RAD51 and BLM interact specifically in DNA-damaged cells [19],[20]. On the basis of immunolocalization studies, we found that the recruitment and/or retention of RAD51 at sites of stalled DNA replication forks is impaired in SM-BLM cells. Whereas BLM+ cells exhibited a 4-fold increase in the number of colocalized γ-H2AX-RAD51 foci upon HU treatment, SM-BLM cells exhibited a <1.5-fold increase in these foci. On the basis of this finding, and the finding that HR-mediated DNA repair is defective in SM-BLM cells, we propose that BLM SUMOylation mediates the recruitment and/or retention of RAD51 to sites of DNA damage and thereby facilitates HR-mediated repair processes.
Previous cell and biochemical studies have led to the view that BLM has both pro- and anti-recombinogenic functions. Most notably, BLM is important in stabilizing damaged replication forks [14]–[16] and repressing aberrant recombination events, as evidenced by the dramatic increase in levels of SCEs and loss of heterozygosity in BLM null cells [6],[46]. In contrast, BLM is also predicted to promote HR by facilitating exonucleolytic resection of DSBs [23]–[26], by stimulating synthesis-dependent strand annealing [21],[47], and by promoting noncrossover resolution of Holliday junctions [12]. Both pro- and anti-recombinogenic functions have likewise been proposed for Escherichia coli RecQ helicase [48]. Our finding that cells expressing SUMO-mutant BLM have a defect in the recruitment and/or retention of RAD51 to sites of DNA damage, and that they are defective in HR-mediated repair, supports a model in which SUMOylation of BLM acts as a switch to regulate its effects on recombination (Figure 7). In the absence of SUMOylation, we propose that BLM binds to stalled replication forks and suppresses aberrant HR by inhibiting excessive accumulation of RAD51 at repair sites. In the event that a stalled replication fork progresses to a DSB, stimulation of HR-mediated repair would be triggered by BLM SUMOylation and, consequently, more efficient recruitment and/or retention of RAD51 at the repair site.
SUMOylation of BLM could regulate the recruitment and/or the retention of RAD51 at sites of DNA damage through several different mechanisms. SUMOylation could dissociate BLM from broken DNA ends, where it might otherwise inhibit RAD51 binding and function by displacing RAD51 from ssDNA or by unwinding D-loops [21],[22]. SUMOylation could limit the binding of BLM to sites of DNA damage by altering its affinity to ssDNA or possibly by triggering its ubiquitin-dependent degradation. SUMO-mediated degradation has been described for PML and other proteins [49]–[51]. In the model that we currently favor, BLM SUMOylation could function to promote RAD51 localization at repair sites by stabilizing its interactions with BLM through a mechanism involving noncovalent SUMO binding. Consistent with this model, we found that RAD51 is a SUMO-binding protein and that it interacts more efficiently with SUMOylated BLM compared to unmodified BLM. RAD51 is recruited to damaged replication forks in BS cells [19],[20] and to a limited number of sites of DNA damage in SM-BLM–expressing cells. Thus, multiple factors appear to control RAD51 recruitment to sites of DNA damage. Nonetheless, our findings support the hypothesis that in BLM-expressing cells, BLM SUMOylation promotes RAD51 recruitment and/or retention at sites of DNA damage and thereby facilitates HR-mediated DNA repair.
SUMOylation of BLM is likely to have multiple roles. In addition to regulating its activity at sites of DNA damage as revealed in the current study, BLM SUMOylation may also be important in mediating the localization of BLM to PML-NBs in undamaged cells. SUMO modification can function to retain proteins in the PML-NBs [35], and fluorescence recovery after photobleaching studies have shown that BLM rapidly associates and dissociates from the PML-NBs [52]. This on–off process may be mediated primarily through BLM's SUMO interaction motif, which is required for BLM localization to the PML-NBs and for BLM SUMOylation [53],[54]. The presence of a SUMO binding site(s) in RAD51 suggests that its association with the PML-NBs may also be regulated through the SUMO pathway.
Our results have broad implications for understanding not only how the integrity of replication forks are maintained under stress but also how SUMO modification regulates its substrates, because many proteins in the DNA repair and signaling pathways are SUMO substrates. Although the role of SUMO in HR function is not yet understood, it is clear that SUMOylation plays multiple roles in regulating the HR pathway through modifications of various HR factors, including Sgs1 [32], Rad52 [55]–[57], PCNA [58], and other recombination-associated factors. sgs1 mutants and mutants of the SUMO-specific E3 ligase gene mms21 accumulate aberrant cruciform structures at damaged replication forks [32],[33]. This genetic evidence suggests that SUMOylation is important in the regulation of HR, but there has been no direct evidence that SUMOylation occurs at the repair site. Because SUMO-mutant BLM accumulates at HU-induced replication fork damage, the present results indicate that BLM SUMOylation occurs at the sites of damaged replication forks, where it affects stabilization of stalled forks, trafficking of RAD51 to repair sites, and HR repair of damaged forks. Further experiments are now needed to characterize the spatial and temporal regulation of SUMOylation of the different repair factors in HR. In particular, we need to determine what signals activate BLM SUMOylation and how BLM SUMOylation is regulated at damaged forks.
For BLM Western analysis, rabbit polyclonal anti-BLM antibodies raised against the first 431 amino acids of human BLM [59] or commercially available antibodies (A300-110A, Bethyl Laboratories) were used. Anti-SUMO antibodies were used as described [54]. For indirect immunofluorescence, we used mouse monoclonal anti–γ-H2AX antibody (Upstate), rabbit polyclonal anti-RAD51 antibodies PC130 (Calbiochem), mouse monoclonal anti-PCNA antibody sc-56 (Santa Cruz Biotechnology), Cy-5–labeled donkey anti-rabbit antibodies (Jackson Labs), Alexa Fluor 594–labeled goat anti-mouse, Alexa Fluor 594–labeled goat anti-rabbit, and Alexa Fluor 647–labeled goat anti-mouse antibodies (Invitrogen). Rat monoclonal anti-Hsc70 antibodies (Assay Design) were used as a loading control in Western analyses.
The full-length BLM cDNA was cloned into the EGFP-C1 vector (Clontech), which produced a GFP-BLM fusion protein with GFP at the N-terminus of BLM, as described previously [53]. The GFP-BLM construct was used as a template for the construction of BLMs that contain SUMO acceptor-site mutations, by substituting arginine for lysine at amino acid residues 317 and 331 using standard polymerase chain reaction–based methods [34]. The construct used in the experiments reported here contained mutations at both 317 and 331. We stably expressed the normal BLM and SUMO-mutant BLM constructs in the SV40-transformed fibroblast cell line GM08505 (BS cells) and isolated multiple clones expressing each construct, as described previously [34],[60]. To measure the levels of GFP-BLM expression, we prepared cell lysates in Laemli sample buffer, fractionated proteins by sodium dodecyl sulphate-polyacrylamide gel electrophoresis, and transferred the proteins to nitrocellulose membranes (Bio-Rad). The membranes were then processed for Western blot analysis and probed with anti-BLM antibodies as described earlier [53]. Varying levels of BLM expression were detected in BLM+ and SM-BLM clones. We chose BLM+ and SM-BLM clones that had comparable levels of transgene expression.
To measure DNA content and percentage of cells in each phase of the cell cycle, cells were harvested by trypsinization and fixed in 70% ethanol for >3 h at −20°C. After fixation, cells were pelleted and resuspended in a solution of 1× phosphate buffered saline (PBS; Gibco) containing propidium iodide (10 µg/ml) and RNase A (0.1 mg/ml). The fluorescence intensities of the propidium iodide–stained cells were measured using a FACScalibur (Becton-Dickinson), and data were analyzed with CellQuest (Becton-Dickinson) and WinMDI (Joe Trotter; http://facs.scripps.edu) software. To examine the percentage of cells in S phase, cells were treated with HU, released, and then analyzed at times after release using the BrdU Flow Kit (BD Biosciences) according to the manufacturer's instructions. BLM+ and SM-BLM cell clones in the logarithmic phase of cell proliferation had comparable proliferation rates, with cell-doubling times equal to ∼30 h for each clone examined.
BLM+ and SM-BLM cells were seeded on coverslips and then treated with 10 mM or 0.5 mM HU in culture medium for 1 or 24 h, respectively. To achieve cell synchronization through an independent mechanism, BLM+ and SM-BLM cells were seeded onto coverslips and treated with 0.5 mM mimosine for 24 h. The cells were released into normal medium for 5 h to allow entry into S phase, then treated or not with 10 mM HU for 1 h. For indirect immunofluorescence, cells were washed and fixed at the end of HU treatment. They were then stained with anti-RAD51 and with anti–γ-H2AX or anti-PCNA antibodies, and counterstained with secondary antibodies labeled with Alexa Fluor (Invitrogen). Fixation and staining was performed as described previously [34]. Coverslips were mounted with Prolong Gold antifade reagent containing 4′,6-diamidino-2-phenylindole (DAPI) (Invitrogen). Images were captured on a spinning disk confocal microscope (Carl Zeiss, LSM-510), and data were collected using Slidebook 4.1 software. Z-stacks were captured using a 100× oil immersion objective, and the optical slice thickness was 0.2 µm.
A focus was defined as a defined area of the nucleus greater than the minimum area of optical resolution (>0.125 µm2) in at least one Z-stack in which the fluorescence intensity was greater than the background fluorescence intensity of the nucleoplasm. Colocalization was defined as an area of overlap between two foci of different fluorophores. The maximum number of foci that could be counted in these cells was 150. For purposes of quantification of γ-H2AX foci in the γ-H2AX-bright cells, the γ-H2AX-bright cells were assigned 151 foci, which was one focus more than the maximum countable number. A typical immunofluorescence experiment consisted of assessment of 30–50 cells per condition. The data presented are from two to three independent experiments performed on two or three clones of each type. Immunofluorescence images for figures were created using Image J and Metamorph software (Molecular Devices).
Cells were untreated or treated with 0.5 mM HU for 24 h. Immediately after treatment, cells were harvested, washed twice with PBS, resuspended in fixation solution (2.3% paraformaldehyde, 0.6% methanol) at a density of approximately 1×106 cells/ml, and incubated on ice for 20 min. Following the fixation process, cells were washed twice with PBS to remove the fixative and were resuspended in permeabilization solution (0.25% saponin, 10 mM HEPES [pH 7.4], 140 mM NaCl, 2.5 mM CaCl2) at a concentration of approximately 5×106/ml. Cells were incubated overnight at 4°C in Alexa Fluor 647–conjugated anti-H2A.X (BioLegend). After incubation, the cells were washed twice with 0.1% saponin and 5% fetal bovine serum in PBS and were resuspended in 5% fetal bovine serum in PBS for analysis. Cells were analyzed on a FACScaliber as described above. Data were collected from a minimum of two experiments on five BLM+ clones and six SM-BLM clones. Median fluorescence intensity data was log normalized, and the differences in median intensity were calculated. Log-normalized median intensity differences were tested by Student's t-test.
Cells were untreated or treated with 0.5 mM HU for 24 h and subsequently released into fresh medium for an additional 0, 12, or 24 h, or they were treated with different concentrations of CPT for 3 h. For each damage condition, 4×105 cells were formed into individual 1% agarose plugs (Cleancut agarose, Bio-Rad). The plugs were then incubated in 100 mM EDTA (pH 8.0), 0.2% sodium deoxycholate, 1% sodium lauryl sarcosine, and 1 mg/ml proteinase K at 50°C for 24 h. The plugs were washed four times in 10 mM Tris-HCl (pH 8.0) and 1 mM EDTA for 30 min at room temperature with gentle agitation. Plugs were loaded onto a 0.8% agarose gel (Pulsed Field Certified Agarose, Bio-Rad), and PFGE was preformed on a CHEF DR III (96°, 100°, 106° angle ramp, 1,200–1,800 s switch time, 2 V/cm; Bio-Rad) for 72 h. The gel was stained with SYBR Gold, visualized under UV light, and analyzed using Quantity One and ImageJ software after contrast adjustment. Each lane on the gel was divided into seven areas, and the intensity in each area was analyzed and weighed according to fragment size. The amount of breakage in each lane was further normalized against total DNA content. Values are given relative to the level of DNA breakage in the untreated control. At least two independent experiments were performed on two clones of each type.
For SCE analyses, cells were cultured with 10 µM BrdU (Sigma-Aldrich). After 60 h, the cells were incubated with 0.02 µg/ml colcemid (Invitrogen) for up to 2 h, harvested and processed as described earlier [60]. The slides were examined under the microscope at 100×, and SCEs were counted from metaphases with an acceptable quality of sister-chromatid discrimination. For measurements of HU-induced SCEs, cells were cultured in 10 µM BrdU for 30 h, washed one time with 1× PBS, and treated with 0.5 mM HU for 24 h. Next, the cells were released into medium containing 10 µM BrdU for an additional 20 h. Metaphases were collected in colcemid and processed as described above. Two independent experiments were performed on two clones of each type.
To quantify the cytogenetic effects of replication-associated DSBs, cells were treated with HU, and micronuclei formation was assessed using the cytokinesis-block micronucleus assay [61],[62]. Cells were plated on chamber slides (Lab-Tek) for approximately 48 h. Next, cells were treated with 0.5 mM HU for 24 h, after which the cells were washed thoroughly with PBS and then incubated in culture medium containing 8.7 µM cytochalasin-B (Sigma-Aldrich). After 28 h of cytochalasin-B treatment, cells were fixed on the slides with a 9∶1 methanol:acetic acid solution and then stained with Diff-Quik (Baxter) according to the manufacturer's instructions. Using a blinded analysis, we examined cells under a light microscope at 100×. A minimum of 500 binucleated cells were assessed under each condition and categorized as follows: cells with no micronucleus, one micronucleus, more than one micronucleus, and nucleoplasmic bridges. Three independent experiments were performed on three clones of each type.
To measure cell viability under different DNA damage conditions, 2×105 cells were seeded onto six-well dishes. Cells were treated or not treated with 0.5 mM HU for 24 h and then treated or not treated with 50 µM of etoposide for an additional 24 h. At the end of the second treatment, the cell-growth medium was retained, and the floating cells were combined with adherent cells harvested by trypsinization. We added the ViaCount reagent (Guava Technologies) according to the manufacturer's instructions. Live and dead cells were counted with the Guava Cell Analyzer. A minimum of three independent experiments with three replicates for each condition were performed on three clones of each type.
Human RAD51 protein was purified as described previously [63]. The purified hRAD51 was either cleaved from the streptavidin-conjugated agarose beads (Ultralink, Pierce) using tobacco etch virus protease or left on the beads and directly used for the binding reactions. SUMO was biotinylated by Sulfo-NHS-Biotin (EZ-link, Pierce) according to the manufacturer's instructions. Equal amounts of unbiotinylated and biotinylated SUMO-1 and SUMO-2 proteins were incubated with RAD51 in binding buffer (PBS with 0.1% Triton X-100) for 2 h at 4°C. Streptavidin beads were then added to each reaction and incubated for another 1 h at 4°C. After five washes with binding buffer, proteins were eluted by SDS-PAGE buffer and analyzed by Western blotting with RAD51 antibodies.
BLM N-terminal fragment (1–431) was modified by SUMO-2 in vitro as described [54]. The SUMO-modified BLM was then aliquoted equally into two tubes with either RAD51-coated streptavidin beads or biotin-coated beads, and rotated in the binding buffer in the presence of 2% BSA at 4°C for 2 h. Unmodified BLM was used as a control. The eluted protein was analyzed by SDS-PAGE and Western blotting with anti-BLM antibodies. Fujifilm Multi Gauge image analysis software (Fujifilm Corp.) was used to determine relative Western blot band intensities and ratios of SUMO-2–modified to unmodified BLM.
Because observations within each clone may be correlated, we used mixed effects linear models to test the data for statistical significance. In the mixed effects models, each clone was treated as a random effect, and the experimental variables were treated as fixed effects. Because the foci and micronuclei data were not normally distributed, we first applied a square root transformation to stabilize the variance and normalize the data (results in the figures were still presented in the original scale). For testing changes in the number of foci per cell and the number of colocalized foci per cell, cell type (BLM or SM-BLM), treatment (with and without HU), and interaction terms for cell type by treatment were treated as fixed effects. Similarly, for testing changes in the number of micronuclei per cell and the number of SCEs per 46 chromosomes, cell type, treatment with HU, and their interaction were treated as fixed effects. Finally, for testing changes in percentage of cell death, cell type, treatment with HU, treatment with etoposide, and their interaction terms were included as fixed effects. If the random effects (i.e., the clonal variation) were found to be nonsignificant based on likelihood ratio test, the mixed effect models reduced to traditional analysis of variance.
|
10.1371/journal.pntd.0000567 | Chagas Disease, Migration and Community Settlement Patterns in Arequipa, Peru | Chagas disease is one of the most important neglected tropical diseases in the Americas. Vectorborne transmission of Chagas disease has been historically rare in urban settings. However, in marginal communities near the city of Arequipa, Peru, urban transmission cycles have become established. We examined the history of migration and settlement patterns in these communities, and their connections to Chagas disease transmission.
This was a qualitative study that employed focus group discussions and in-depth interviews. Five focus groups and 50 in-depth interviews were carried out with 94 community members from three shantytowns and two traditional towns near Arequipa, Peru. Focus groups utilized participatory methodologies to explore the community's mobility patterns and the historical and current presence of triatomine vectors. In-depth interviews based on event history calendars explored participants' migration patterns and experience with Chagas disease and vectors. Focus group data were analyzed using participatory analysis methodologies, and interview data were coded and analyzed using a grounded theory approach. Entomologic data were provided by an ongoing vector control campaign. We found that migrants to shantytowns in Arequipa were unlikely to have brought triatomines to the city upon arrival. Frequent seasonal moves, however, took shantytown residents to valleys surrounding Arequipa where vectors are prevalent. In addition, the pattern of settlement of shantytowns and the practice of raising domestic animals by residents creates a favorable environment for vector proliferation and dispersal. Finally, we uncovered a phenomenon of population loss and replacement by low-income migrants in one traditional town, which created the human settlement pattern of a new shantytown within this traditional community.
The pattern of human migration is therefore an important underlying determinant of Chagas disease risk in and around Arequipa. Frequent seasonal migration by residents of peri-urban shantytowns provides a path of entry of vectors into these communities. Changing demographic dynamics of traditional towns are also leading to favorable conditions for Chagas disease transmission. Control programs must include surveillance for infestation in communities assumed to be free of vectors.
| Chagas disease affects 8–10 million people in the Americas. Although transmission was previously limited to the rural poor, Chagas increasingly affects urban populations, especially near the city of Arequipa, Peru. We interviewed residents of five communities to learn about why and when they migrated to the city and how their movements may link to Chagas vectors and to explore the settlement patterns of shantytowns and traditional towns. We found that migrants to shantytowns were unlikely to introduce Chagas vectors to the city upon first arrival. Frequent seasonal moves, however, took shantytown residents to valleys surrounding Arequipa where vectors are prevalent. In addition, the settlement pattern of shantytowns and the practice of raising domestic animals create a favorable environment for vectors. Finally, population loss and replacement by low-income migrants in one traditional town has created the human settlement pattern of a shantytown. This study exposes potential links between population dynamics and Chagas vector infestation. Suggested methods for improving vector control include focusing future vector surveillance in areas with mobile populations, creating educational campaigns for migrant workers to Chagas-endemic areas, and fomenting collaboration between the Arequipa Ministries of Health and Housing to ensure the inclusion of new shantytowns in vector surveillance.
| Chagas disease, caused by infection with protozoan parasite Trypanosoma cruzi, causes more morbidity and mortality than any other parasitic disease in the Western Hemisphere [1]. T. cruzi is carried by numerous species of triatomine insects. Humans and other mammals usually become infected when the triatomine vector defecates during its blood meal, and fecal material containing the parasite is inoculated through the bite wound or mucous membranes [2]. Vector-borne transmission only occurs in the Americas, where 8–10 million people, including an estimated 192,000 Peruvians, are currently infected with T. cruzi [3],[4].
The member countries of the Southern Cone Initiative (INCOSUR) have worked since 1991 to eliminate household infestation with Triatoma infestans, the most important Chagas disease vector in the southern half of South America, through large-scale residual application of pyrethroid insecticides [5],[6]. Despite remarkable successes, major challenges remain to vector control, among them the increasing urbanization of the disease [7]. Chagas disease is traditionally associated with rural villages with adobe houses hospitable to T. infestans and other domestic vectors [8] and vector-borne transmission appears to be rare in urban settings [9]–[11]. However, in marginal communities of the city of Arequipa (pop. 750,000) in southern Peru, urban T. cruzi transmission cycles have become established [12],[13], and a vector control campaign has been in place in the city of Arequipa since 2002 [13].
The settlement and migration patterns in and around cities therefore may be important to understanding the dynamics that make certain communities more susceptible to Chagas disease vectors [14]. Latin America has experienced an overwhelming phenomenon of urbanization due in most part to in-migration, and Peru is no exception [15],[16]. Few studies have directly examined migration and settlement patterns, and their connections to Chagas disease transmission. Here we use qualitative methods to explore the migration and settlement patterns, and their links with vector infestation, in different communities around the city of Arequipa.
The research protocol was approved by the ethical review committees of the Asociación Benéfica PRISMA and the Johns Hopkins Bloomberg School of Public Health. All participants provided written informed consent prior to data collection, including consent for audio-recording.
Arequipa is the second largest city in Peru, located in an arid zone 2,300 m above sea level [15]. The outskirts of the city contain hundreds of peri-urban pueblos jóvenes (young towns or shantytowns) and pueblos tradicionales (traditional towns). Pueblos jóvenes are low-income hillside squatter settlements founded over the past 60 years [17],[18]. Pueblos tradicionales tend to be in lower-lying flat areas, are inhabited by higher-income landowners, and date back to the late 19th or early 20th century. (See Figure 1 for photos of the two types of communities.) Because preliminary data from our research group indicated that T. infestans prevalence differed between these two types of towns [12], we compared migration and settlement patterns in 3 pueblos jóvenes and 2 pueblos tradicionales.
The research team worked with 2–3 community leader “gatekeepers” in each community to ensure acceptance and to recruit people who could provide detailed information about their personal history of migration and settlement (for interviews) or community history (for focus groups). A total of 94 female and male participants were enrolled in the study.
This was a qualitative study that employed focus group discussions and in-depth interviews. Focus group sessions were carried out with 8–10 participants in each community at central, well-known locations (health establishments, community centers) selected by the gatekeepers. We used participatory methodologies to explore the community's demographic characteristics and mobility patterns, and historical and current presence of triatomine vectors. Participants created community maps [19] which formed the basis for discussions of the history and characteristics of communities. Participants then created a timeline [19] of important community events dating back approximately 40 years, first exploring general events and then focusing on events related to Chagas disease and vector infestation. All sessions were audio-recorded; participatory activities were recorded on large sheets of paper that were hung on the wall and visible to all participants.
In-depth interviews utilized an event history calendar (EHC), a highly structured but flexible interview style that facilitates recall by using the individual's own experiences as cues [20]. Interviews were carried out with 10 participants per community, and explored migration history, experience with Chagas disease, the presence of its vectors, and customs of raising animals in each place of residence, starting from birth. Each interview, held at a location selected by the participant (their home, the focus group location), was audio-recorded and lasted 45 to 60 minutes. All interviews and focus groups were conducted by two of the authors (AB and GH).
Presence of vectors was examined by asking participants when they had seen triatomine bugs. Because T. infestans is the sole insect vector for Chagas disease in southern Peru, we only asked participants to recall the presence of this species. All of the study communities had been involved in insecticide application programs run by the Ministry of Health (MOH) and as a result were broadly familiar with T. infestans, which they refer to as a “chirimacha.” In this context, there is no other bug that goes by the same name. We also showed images of T. infestans to participants to aid their recall. During interviews, participants were asked if they recalled seeing triatomine bugs in their house or in their community for the specific years they lived in each place of residence. During focus groups, participants were asked to reach a consensus regarding the year that triatomine bugs first appeared and the areas most infested.
Data on the number of households, estimated population, and year of insecticide application were collected by our research group in collaboration with the Arequipa Regional Office of the Ministry of Health (MOH). The domiciliary infestation index (DII) is a community level variable equivalent to the number of infested houses divided by the total number of houses surveyed. We used ArcView 9.1 (ESRI) and existing maps to estimate the distance of each community from Arequipa and its surface area (to calculate population density).
Detailed notes were taken from focus group audio-recordings and the large sheets of paper, and synthesized into matrices in Excel by theme and by community to carry out further analysis. Audio-recordings of interviews were transcribed verbatim from digital audio recorders to word processing programs and analyzed using the grounded-theory approach. Grounded theory refers to theory that is developed inductively from a body of data, in contrast to theory that is derived deductively from grand theory and not necessarily based on data. The grounded theory approach is applied by reading textual data, in this case transcripts and field notes, in order to discover the main concepts or themes that were mentioned during data collection, allowing the data to reveal its message (or theory) instead of looking for confirmation of a previously-developed hypothesis [21],[22],[23],[24]. Further information about grounded theory can be accessed via the web links listed in the references [25],[26]. Two authors (AB and GH) created a code set based on the main themes that emerged in the interviews after an initial reading of the transcripts. The Atlas-ti software (Scientific Software Development GmbH, 2005) was used to apply the codes to each interview transcript. The event history calendars themselves were also analyzed by entering the information into a single timeline in Excel, such that each decade from 1900 to the present year detailed the places of residence of participants and the presence of triatomine vectors in those places. The superimposition of the EHCs on a timeline allowed us to visualize patterns in migration and vector presence over time. Each coded interview was analyzed together with the interviewee's EHC and all focus group and interview data were analyzed by community and then across communities.
During the EHC interviews, all moves were recorded, including a change of abode within the same community. For the purposes of this analysis, only a change of resident community for one month or more, whether temporary or permanent, was considered as a movement.
Key demographic and T. infestans infestation data from the 5 study communities are listed in Table 1, and their geographic location in reference to the urban center is shown in Figure 2.
Fifty people provided in-depth interviews and 44 participated in focus groups. Interview participants had a median age of 52 and 47 years for males and females, respectively, with a range of 20 to 80 years. Focus group participants had a median age of 67 for males and 49 for females, with a range of 20 to 79 years (Table 2). Older community members were purposely recruited for focus groups, to ensure knowledge of historical events.
Males moved more frequently than females, with a median of 3.0 (Interquartile range (IQR) 1.0–4.0) lifetime moves for females and 4.0 (IQR 2.5–5.5) for males. There was also more migration among residents of pueblos jóvenes: 80% of participants from pueblos jóvenes were in-migrants, while 60% of participants from pueblos tradicionales were non-migrants (Table 3). Typical patterns of migration are shown in Table 4. Residents of pueblos jóvenes typically moved from a rural birth community to the Arequipa area early in life due to economic stress. Later moves took them to a pueblo jóven near Arequipa in search of cheap housing and then multiple, short-term moves continued throughout life in search of work. Some residents of pueblos tradicionales also moved from a rural area to Arequipa during childhood or adolescence, usually for schooling. Later moves were few since these residents tended to settle in the pueblo tradicional to focus on building agricultural enterprise and raising a family. The majority of participants who migrated from rural areas recalled raising farm animals in their birth places, and many continued to raise animals, especially guinea pigs, rabbits and poultry, in peri-urban Arequipa.
Study participants indicated that the pueblos jóvenes of Arequipa's urban periphery were mostly settled by people from greater Arequipa and the northern Andean regions of Cusco and Puno, and in lesser numbers by people from the southern coastal/Andean region of Moquegua (see map insert in Figure 2). The formal founding dates of the pueblos jóvenes in this study ranged from 1970 to 1981, although some residents had lived in these settlements since the 1960s. Migration from rural birthplaces to urban Arequipa was motivated by acute economic stress, with a few families sending their children away to work as early as age six. This early move was followed by migration later in life to a pueblo joven to acquire land and housing. Each quote presented in the Results section is followed by the participant's sex, age, type of current community, and the years corresponding to the movement. Two examples of interviewees who moved alone as children follow:
Other participants moved with their entire family for work:
Migration to pueblos jóvenes enabled early settlers to acquire land at little cost as squatters invading land. The pueblo tradicional of Tío Chico is located just below several hillside pueblos jóvenes, including Guadalupe:
By contrast, residents who arrived after the original invasion bought an existing house or land parcel, rented a house or room, or lived with family members. Pueblos jóvenes are still expanding in geographic area and population, and currently are composed of multiple generations, including the original migrants, their children and grandchildren, and newly-arrived migrants.
Pueblos tradicionales are made up primarily of people who have lived in the community since birth and whose families have lived there for several generations. As in pueblos jóvenes, the founding migrants arrived in search of land and housing. However they usually purchased this higher-cost land:
Later migrants tended to move to pueblos tradicionales motivated by a return to family roots or the search for a calmer environment:
Current migration dynamics are causing important changes to the population of pueblos tradicionales. Many people, especially young people, are moving out of pueblos tradicionales due to a lack of opportunities. In the two pueblos tradicionales in this study, out-migration has resulted in diminished populations and a shortage of agricultural workers. To compensate, land owners hire temporary workers who are often migrants to the peri-urban areas of Arequipa. In Tío Chico, these workers do not live in the town itself since there are few available houses and they can live more affordably in the surrounding pueblos jóvenes. In contrast, Quequeña, being further from the city, lacks housing options other than the town itself. Rental is therefore common since the property owners have moved and need people to watch over their properties and work their land. As a result, Quequeña is experiencing relatively recent in-migration, while Tío Chico is not.
A constant across most participants' adult lives was migration to live with a partner or spouse and children:
Across communities, the search for work was a constant that often spanned generations, with migration in early life by parents and in later life by participants. These participants described the almost constant movement of their fathers in search of work:
Male participants described their own search for work opportunities and female participants discussed similar searches by their partners:
Forty-six (92%) of 50 interviewees had seen triatomine vectors (locally known as chirimachas) during their lifetimes in some place of residence; the four who did not report seeing chirimachas lived in the pueblo tradicional of Tío Chico (Figure 3). Participants reported seeing no triatomines in the highland Andes regions of Cusco and Puno, where many migrants were born. In addition, the urban center of Arequipa (often the first place of residence for low-income migrants to Arequipa) was described as vector-free.
The earliest sighting of triatomines occurred in Moquegua in 1942, a sending area for some migrants to urban Arequipa that is located south of the study communities (Figure 2). Other early memories came from the rural areas of Valle de Vitor and La Joya, both valleys west of Arequipa. The study community of Villa La Joya is a pueblo joven founded near the more rural and longer established community of La Joya. Villa La Joya had its first residents in the early 1960s, but chirimachas were not memorable until the town was much more populated in the late 1970s and 1980s.
Closer to the urban center, reported vector presence showed a similar pattern of infestation following settlement, but much more recently. The first memories of urban vector presence from study participants were from the pueblo joven of Jacobo D. Hunter in 1968, a peri-urban district settled in early squatter invasions. Years later, in 2002, Hunter was the setting of a highly publicized child death due to acute Chagas disease [27]. Peri-urban pueblos jóvenes increased in number and size during the 1960s and 1970s, and vector presence was reported in the urban pueblos jóvenes of our study roughly 20 years following original settlement. Participants from Guadalupe (founded in 1970) and Nueva Alborada (founded in 1981) noted the first widespread appearance of triatomines in 1988 and 2002, respectively.
In pueblos tradicionales, vector reports followed a different pattern than in the pueblos jóvenes. Although their settlement dates back to the mid-1800s, focus group participants recalled first seeing vectors in Quequeña around 1978. They noted that chirimachas were especially concentrated in the area of town around a communal stable and zones of relatively newer settlement by migrants. Tío Chico, in contrast, had very few reports of infestation, and much later (from 2000). The following quotes contrast the degree of infestation between Tío Chico and Guadalupe, the hillside pueblo joven located above Tío Chico. Participants specifically associated infestation with the presence of domestic animals.
Residents reported continuous presence of chirimachas in all study communities except Tío Chico until insecticide spraying occurred (Figure 3).
According to MOH vector control data, the pueblos tradicionales in our study had considerably lower infestation rates than pueblos jóvenes (Table 1). The high infestation rates in pueblos jóvenes parallel higher population densities of 10,300–15,000 inhabitants/km2, compared to 1,700–3,300 inhabitants/km2 in pueblos tradicionales (Table 1).
The prevalence of Chagas disease varies widely among communities around Arequipa. High T. cruzi infection rates in children, reflecting recent transmission, were documented in several recently-formed pueblos jóvenes, while long-established pueblos tradicionales have almost no infection in children [12]. The proximate cause of infection heterogeneity is the difference in triatomine infestation rates [12],[28]. We show here how migration and associated activities contribute to conditions promoting infestation in pueblos jóvenes. Human migration patterns thus constitute an important underlying determinant of Chagas disease risk in and around Arequipa.
Most migrants to pueblos jóvenes came from areas without infestation and are unlikely to have brought the vector with them. However, community members later made multiple short- to medium-term moves, often to the valleys west of Arequipa for seasonal agricultural labor. These valleys are also the best-known historical foci of T. cruzi transmission in the region [29]. Migrants may have become infected while living temporarily in the valleys, or may have carried vectors back to their long-term communities in their belongings.
An alternative, not mutually exclusive hypothesis is that vectors were always present in pueblos tradicionales prior to the construction of pueblos jóvenes, but not in large enough numbers to be a memorable event for study participants. When pueblos jóvenes quickly took over peri-urban hillsides, existing vector populations may have exploded. Rural migrants from highland areas brought their domestic animal husbandry practices to the city with them, raising small animals such as guinea pigs and rabbits for sale or for personal consumption, and keeping them in small yards close to human dwellings. At the time of vector control spraying in 2004, the pueblo joven of Guadalupe contained a total of 5,006 domestic animals (predominantly guinea pigs, rabbits, poultry and dogs, but also sheep and cows); the presence of guinea pigs, in particular, was associated with an increased risk of infestation both in the animal enclosure and in the adjacent human house [13]. The high density of animals provides an abundance of blood meal sources to support large vector populations, and potentially contributes to Chagas disease transmission, since all except poultry are susceptible to T. cruzi infection.
Interestingly, one pueblo joven, Nueva Alborada, was highly infested with vectors, but none of the 1,460 insects examined during the Ministry insecticide application campaign were carrying T. cruzi [30]. Focus group participants in Nueva Alborada reported chirimachas in the community only back to 2000, while focus groups in the other two pueblos jóvenes recalled insects in their communities for a much longer period. It is possible that, given the short history of vector infestation in Nueva Alborada, the parasite has yet to be successfully introduced into this community. In contrast, in Guadalupe and Villa La Joya, the longer history of infestation may have led to single or multiple introductions of the parasite in these communities. The relationship between time since introduction of the vector and presence of T.cruzi merits further epidemiological investigation.
Two final considerations include the possible transformation of long-standing pueblos tradicionales by an influx of low-income migrants and the urbanization of rural areas. Although many of the original residents had emigrated from both traditional towns, we observed increased population density and levels of in-migration that were similar to those in pueblos jóvenes only in Quequeña, where low-income migrants are filling the population void. In contrast, no new migrants had moved to Tío Chico, because low-income farm workers who cultivate fields surrounding Tío Chico can live more cheaply in a nearby pueblo joven. The case of Quequeña draws attention to a previously undescribed phenomenon of population loss and replacement by low-income migrants, creating the human settlement pattern of a pueblo jóven within a pueblo tradicional, and providing insight into why T. infestans and T. cruzi are prevalent in Quequeña.
These complex patterns of human migration and triatomine infestation demonstrate the conditions underlying the urbanization of Chagas disease in Arequipa. Factors common to the infested pueblos jóvenes include recent, rapid settlement, high human and animal density, and frequent temporary migration to Chagas disease-endemic areas. However, some pueblos tradicionales are also undergoing similar processes. As peri-urban areas in South America continue to grow, vector control programs must remain vigilant against reinfestation, as well as infestation of communities not previously recognized to be at risk, such as the pueblos tradicionales of Arequipa. In addition, our data point to at least three potential interventions for improving vector control in Arequipa: 1) Intensifying vector surveillance efforts in areas with highly mobile populations; 2) Creating educational campaigns for migrant workers to Chagas-endemic areas; and 3) Fomenting collaboration between the Arequipa Region's Ministry of Health and Ministry of Housing to monitor the emergence of new pueblos jóvenes for their inclusion in the vector surveillance system.
|
10.1371/journal.ppat.1002796 | Effect of Biodiversity Changes in Disease Risk: Exploring Disease Emergence in a Plant-Virus System | The effect of biodiversity on the ability of parasites to infect their host and cause disease (i.e. disease risk) is a major question in pathology, which is central to understand the emergence of infectious diseases, and to develop strategies for their management. Two hypotheses, which can be considered as extremes of a continuum, relate biodiversity to disease risk: One states that biodiversity is positively correlated with disease risk (Amplification Effect), and the second predicts a negative correlation between biodiversity and disease risk (Dilution Effect). Which of them applies better to different host-parasite systems is still a source of debate, due to limited experimental or empirical data. This is especially the case for viral diseases of plants. To address this subject, we have monitored for three years the prevalence of several viruses, and virus-associated symptoms, in populations of wild pepper (chiltepin) under different levels of human management. For each population, we also measured the habitat species diversity, host plant genetic diversity and host plant density. Results indicate that disease and infection risk increased with the level of human management, which was associated with decreased species diversity and host genetic diversity, and with increased host plant density. Importantly, species diversity of the habitat was the primary predictor of disease risk for wild chiltepin populations. This changed in managed populations where host genetic diversity was the primary predictor. Host density was generally a poorer predictor of disease and infection risk. These results support the dilution effect hypothesis, and underline the relevance of different ecological factors in determining disease/infection risk in host plant populations under different levels of anthropic influence. These results are relevant for managing plant diseases and for establishing conservation policies for endangered plant species.
| Biodiversity has been proposed as a major ecological factor determining disease prevalence. However, the relationship between biodiversity and disease risk remains underexplored. Few studies focus on host-virus systems and, particularly on plant viruses. To address this subject the prevalence of virus infection and disease symptoms was monitored in wild-pepper (chiltepin) populations under different levels of human management. For these populations, species diversity, host genetic diversity and host plant density were determined. Higher levels of human management resulted in increased disease and virus infection risk, which was associated with decreased habitat species diversity and host genetic diversity, and with increased host plant density. More specifically, for wild chiltepin populations, species diversity of the habitat was the primary predictor of disease risk; and host genetic diversity was the primary predictor in managed populations, with host density being generally a poorer predictor of disease risk. These results support a dilution effect of biodiversity on disease risk, and underline the relevance of different ecological factors in determining disease risk in wild and in human-managed habitats.
| Understanding the relationship between the risk of infectious diseases and host ecology is a long-standing goal of biological research, central for the management of current infectious diseases and for preventing the emergence of new ones. Indeed, changes in host ecology are among the most frequently identified causes of disease emergence (i.e. the increase of disease incidence following its appearance in a new, or previously existing, host population) [1]–[3]. Because infectious diseases involve interactions between at least two species, it has been proposed for a long time that ecosystem biodiversity will play a key role in disease risk. Current declines in biodiversity have been proposed to be linked with the emergence of infectious diseases, which have fueled a renewed interest on this subject [4]. Two major hypotheses with different predictions relate biodiversity to disease risk. The “Amplification Effect” hypothesis predicts that diversity will be positively correlated with disease risk, as it will result in increased abundance of inoculum sources for a focal host. The “Dilution Effect” hypothesis predicts a negative correlation between biodiversity and disease risk, as a reduction in diversity could result in an increased abundance of the focal host species facilitating disease transmission [5]. These two hypotheses can be considered to represent extremes of a continuum, as the effects of diversity on disease risk would be related to the host range of the pathogen: an Amplification Effect would require a generalist pathogen, while the more restricted the host range of the pathogen, or the higher the differences between shared hosts in their ability to amplify or transmit the pathogen, the higher the Dilution Effect. Increasing evidence derived from pathogens with broadly different life-styles indicates that biodiversity reductions most often result in increased disease risk [4].
The idea linking biodiversity with disease risk is not new in animal or plant pathology. Two classical hypotheses in plant pathology state that the high impact of plant diseases in crops is associated with: i) the reduced species diversity, and higher host density, of agroecosystems as compared to wild ecosystems [6]; ii) the reduced genetic diversity of crops as compared to their wild ancestors or relatives [7]. However, despite that a number of recent studies on the ecology of plant diseases have been added to those dating from the 1980s, support for these hypotheses is still often circumstantial [8]. Attention has focused on analyses of foliar diseases caused by fungi, which mostly indicate that increased biodiversity reduces disease risk [9]–[14]. Remarkably, there are fewer reports referring to viral diseases, which represent a large fraction of emergent plant pathogens [15], and may differ from fungal ones in their relationship to biodiversity. While most plant pathogenic fungi are directly transmitted specialists [16], most plant-infecting viruses are vector transmitted, and are host generalists but often vector specialists [17]. Most studies with plant viral diseases have focused on generalist viruses infecting grasses, generally finding an amplification effect [18]–[21]. Interestingly, work on plant diseases largely failed to assess the role of various possible mechanisms by which reduced biodiversity may affect disease risk (but see [10], [11], [22]). Particularly, it is often difficult to differentiate the effects of increased host density and of reduced species diversity [4]. Hence, there is a need of research aimed at analyzing the effects of biodiversity on plant disease risk and, specifically, at disentangling the role of the various factors associated to ecosystem diversity. This is the goal of the present work.
The focal host in this study is the wild pepper Capsicum annuum var. glabriusculum (Dunal) Heiser and Pickersgill [23], also known as “chiltepin”. Chiltepin is found in Mexico in a variety of habitats from the Yucatan peninsula and the Gulf of Mexico to the Sonoran desert [24], [25]. Chiltepin is a deciduous, perennial bush that grows for 5–8 years and vegetates and reproduces during the rainy season. Birds disperse the seeds from its red pungent fruits [24]. Human harvesting of fruits from wild chiltepin plants is a common practice in central and northern Mexico [26], [27]. A second level of human exploitation involves tolerance or favoring the growth of spontaneously dispersed chiltepin plants in anthropic habitats, such as pastures and living fences (i.e., let-standing plants, sensu [28]). Last, chiltepin cultivation in home gardens or in small traditional plots has started in the recent past [25]. Cultivation has not yet lead to domestication, and cultivated chiltepin populations, which are managed as annual crops, do not show obvious phenotypic differences with wild ones [25]. Wild chiltepin populations show a large genetic variation and a strong spatial structure associated with the biogeographical province of origin, and human management results in a significant loss of both spatial structure and genetic diversity [25]. This habitat diversity makes chiltepin a uniquely good system to analyze the relationship between biodiversity and disease risk.
We focused on two contrasting pepper-infecting virus groups. The first involves two species of the genus Begomovirus (Geminiviridae): Pepper golden mosaic virus (PepGMV), and Pepper huasteco yellow vein virus (PHYVV), here treated collectively as “begomoviruses”. These species have a two-segmented single-stranded (ss) DNA genome; narrow host ranges limited in nature mostly to species of the genera Capsicum, Solanum and Datura (Solanaceae), and are transmitted in a persistent manner by the whitefly Bemisia tabaci Gennadius (Homoptera, Aleyrodidae) [29]–[31]. The B biotype of B. tabaci, characterized by a broad plant host range, a high reproductive potential, and a high efficiency as a vector for begomoviruses, is prevalent in Central and North America [32]. The second virus is Cucumber mosaic virus (CMV), genus Cucumovirus (Bromoviridae), with a tripartite, ssRNA genome, and a typical generalist, infecting more than 1000 species of both mono- and dicotyledonous plant families. CMV is transmitted in a non-persistent manner by more than 80 species of aphids, thus being also a vector generalist [33].
Utilizing these host-pathogen systems we specifically addressed if: i) modification of chiltepin habitat associated with different levels of human management resulted in changes in disease or infection risk, ii) reduction of species diversity increases disease or infection risk, iii) decreased host genetic diversity had an effect on disease or infection risk, iv) increased host plant density resulted in increased disease or infection risk and v) the above effects were different for viruses with different life-histories.
To answer these questions, we visited over three years neighboring wild, and human managed (i.e., of let standing and cultivated plants) chiltepin populations from different biogeographic provinces in Mexico. For each population, species diversity, host density and the prevalence of plants showing symptoms of virus infection were quantified in the field as an estimate of disease risk. Plants were collected at each population and their status (infected/non-infected by several viruses) was determined in the laboratory in order to estimate infection risk. Results indicate that disease and begomovirus infection risks, but not CMV infection risk, decrease with increasing biodiversity. We propose that observed differences between begomovirus and CMV infection risk can be due to different transmission modes.
Chiltepin populations were visited during the summers of 2007 to 2009 at different sites over the species distribution range in Mexico (Figure 1 and Table 1). A total of 26 populations were localized in different habitats representing three levels of human management: i) ten wild populations (W) in which fruit gathering by local people may occur; ii) six populations of let-standing plants (here from called “let-standing populations”), in anthropic habitats, either pastures (LSP) or live fences (LSF), in which chiltepin plants are tolerated or favored, and iii) ten cultivated populations (C) either at home gardens (CHG) or at small monocultures (CMC). Population sites were assigned to 6 biogeographical provinces: Yucatan (YUC), Eastern side of the Sierra Madre Oriental (SMO), Altiplano Zacatecano-Potosino (AZP), Costa del Pacífico (CPA), Costa del Pacífico Sur (CPS) and Sonora (SON) [34]. A total of 14 populations, located in YUC, SMO, AZP, CPA and CPS, were visited during 2007 and 2008. The 2009 survey was extended to other populations of these five biogeographical provinces and SON, to a total of 26 populations (Table 1). Populations were visited between the 15 of July and the 30 of August, in an attempt to homogenize plant phenology among locations at the stage of flowering and beginning of fruit setting. Due to the highly unpredictable rain regime at some regions, or to extinction, not all populations could be surveyed for the three years.
At each location, the following information was collected: 1) The census of the chiltepin population. 2) The status of each censused plant: asymptomatic or showing symptoms commonly related to virus infection (i.e., mosaic, leaf curl, leaf lamina reduction, and/or stunting). 3) The area (m2) occupied by the chiltepin population. 4) The inventory of the non-herbaceous vegetation, determined as the number of individuals of each bushy or arboreal species, in the same area of the chiltepin population, to estimate species richness, and evenness according to the Shannon index [35]. Populations BER-w, PEL-w, MOC-w and MAU-w (Table 1) were too large – i.e., more than 200 plants – to census all plants, and both chiltepin censuses and biodiversity inventories were limited to a fixed transect. In this case, the area occupied by the chiltepin population was calculated by prospecting a width of 4 m along the fixed itinerary.
At each population and visit, plants were systematically sampled for laboratory analyses. Plants were sampled regardless of their showing or not symptoms: One plant out of every x plants was sampled along fixed itineraries, with itinerary length and x (0<x≤4) depending on population size, 1–3 young branches with fresh leaves were collected per plant.
Infection by CMV and by Potyvirus species was analyzed by DAS-ELISA, using commercial antisera against CMV or a monoclonal antibody against a highly conserved motif in the coat protein of potyviruses (Agdia Biofords), according to the manufacturer's instructions. Infection by Chiltepin yellow mottle virus (ChYMV, Tymoviridae) was analyzed by molecular hybridization using a 32P-labeled RNA probe complementary to nucleotides 5365–5777 of ChYMV genomic RNA (Accession No. FN563124) [36]. Infection by species of the genus Begomovirus was detected by PCR using degenerate primers designed on the alignment of DNA-A sequences of 43 begomovirus species from the New World: BAOPsp (5′-GCGCCCTGCAGGGGCCYATGTAYAGGAAGCC-3′) and BAONsp (5′-GCGCGCGGCCGCGANGCATGNGTACATGCCAT-3′), which amplify a region in the coat protein gene located between nucleotide positions 392 and 884 in the genome of PepGMV (Accession No. AY928512).
Molecular hybridizations and PCR were performed on total nucleic acid preparations from chiltepin leaves extracted by grinding 200 mg of fresh leaf tissues in three volumes of 200 mM Tris-HCl pH 9, 25 mM EDTA, 1% SDS, 400 mM LiCl, followed by phenol-chloroform extraction [25]. Plants genotyped using the set of 9 nuclear microsatellites markers described in [25] were used to estimate genetic diversity of the 26 chiltepin populations.
Generalized linear mixed models (GLMM) were used to analyze the difference in the prevalence of virus infection (Begomovirus and CMV), and in the frequency of symptomatic plants, according to chiltepin population, biogeographical province and level of human management of the population, considering these factors as fixed effects. The rationale for considering population as a fixed effect is that all the chiltepin populations that we were able to find were included in the analyses, rather than using a random representation of them. The symptom and virus prevalence values determined for each population in the different years were considered as dependent measures; thus, they were treated as repeated measures in the GLMM. This seems the correct approach for wild and let-standing populations, in which at least a subset of the plants sampled over the years were the same, since chiltepin plants live for several years. This might not be so for cultivated populations, in which plants are managed as an annual crop and may change plots over the years. However, we considered that plots from different years were close enough to be spatially correlated, and therefore repeated measures are warranted. In addition, results did not differ when data from cultivated populations were analyzed as independent measures (not shown). To determine whether values of analyzed traits were significantly different among classes within each factor, Bonferroni analyses were employed in all cases using the GLMM marginal means calculated for each class [37]. GLMM accommodates missing data, so that the 26 chiltepin populations sampled could be included in the analysis. Parallel analyses using only the 8 populations for which data on the 3 years of sampling were available yielded comparable results (data not shown).
The contributions of each ecological factor to the variation in virus and symptom prevalence were estimated using a Principal Component Analysis (PCA). Host plant density (d), host plant genetic diversity estimated as expected heterozygosity (He), species diversity estimated as species richness (number of species, SR) and Shannon index (Sh), of 24 chiltepin populations were scaled to zero mean and unit variance, inserted in a regression matrix and rotated to obtain the principal components (PCs). Importantly, species diversity was not measured in TLA-w and HER-cmc, so that these populations were excluded from the analysis. Significance thresholds for the load of each ecological factor on a PC were determined using a broken-stick model [38]. Bivariate analyses, considering both linear and non-linear models, of begomoviruses, CMV and symptom prevalence onto the ecological factors and their corresponding PCs, yielded the proportion of the variance in each of these variables explained by each factor and each principal component (R2), and the significance of these correlations. For these bivariate analyses, we utilized the GLMM marginal means of begomoviruses, CMV and symptom prevalence calculated for each population. Statistical analyses were performed using the statistical software package SPSS 17.0 (SPSS Inc., Chicago, USA).
Information theory was used to determine the relative importance of the ecological factors in the variation of symptom and virus prevalence [39]. This approach was chosen because it allows making inferences across a set of causal model structures for symptom and virus prevalence [39]. To do so, a set of models that included a global model, which contained all ecological factors (species richness, Shannon index, expected heterozygosity and host plant density), and nested models, which contained different combinations of the predictor variables was fitted. Since species richness and Shannon index always loaded in the same PC, different selection model analyses in which the nested models considered SR, or Sh, or both variables together were performed. The three approaches gave similar results. For simplicity, only results considering species richness are shown. We ranked the models according to second order Akaike's Information Criteria (AICc) to account for small sample size (R library: AICcmodavg) [39]. The model with the lowest AICc score was selected as the best-ranked model. We calculated AICc Delta (Δi), as the difference between the AICc of a given model and that of the best-ranked model. Delta quantifies how strongly models compete (Δi = 0 for best-ranked model; Δi = 1–2 indicates substantial empirical support; Δi = 4–7 indicates considerable less support; and Δi>10 indicates no support [39]). Finally, the Akaike relative weight (ωi) of each model was calculated following the expression: ωi = exp(Δi)/Σexp(Δi).
The status of a total of 1820 censused plants was recorded during the summers of 2007–2009. The prevalence of plants showing symptoms of virus infection (symptomatic plants) (Table 1) marginally varied among year (χ2 = 5.86, P = 0.060), ranging between 16.2% and 21.6% of the census.
A subset of 1081 plants, either symptomatic or asymptomatic, was analyzed for infection by ChYMV (a chiltepin-infecting tymovirus, see [36]), CMV, begomoviruses or potyviruses. Low prevalence of potyvirus infection (2.87%) precluded further analyses. ChYMV infection was limited to locations around Tula, AZP (Table 1) where its prevalence was high (42.86%). Infection by CMV and by begomoviruses was detected during the three years of the study in all biogeographical provinces and under different levels of human management (Figure 1 and Table 2). PepGMV and PHYVV were the only begomovirus species detected infecting chiltepin, and their relative prevalence did not depend on the level of human management of the chiltepin population (data not shown). Therefore, from here on these two species will be considered together and referred to as “begomoviruses”. CMV prevalence remained stable (≈7%) among years (χ2 = 0.06, P = 0.970), while begomovirus prevalence was about 3–5 times higher (19–36%) and varied largely according to year (χ2 = 58.25, P<1×10−5) (Table 2).
Begomoviruses, CMV or ChYMV infection explained the symptoms of 212/281 (78.7%, of these 81% being infected by begomoviruses) laboratory-analyzed symptomatic plants from all populations and years. This fraction did not differ according to the level of human management of the population (59/76 analyzed symptomatic plants for wild populations; 50/70 for let-standing populations, and 103/135 for cultivated populations) (χ2 = 0.86, P = 0.651). The fraction of infected plants showing symptoms (212/369, i.e., 57.4% in total) was lower in wild populations (59/133, 44.4%) than in cultivated (103/153, 67.3%) or in let-standing populations (50/83, 60.2%) (χ2≥4.54, P≤0.033). This fraction did not differ between the later two levels of human management (χ2 = 0.89, P = 0.345).
The effect of geography (biogeographical province and chiltepin population), and level of human management in the prevalence of symptomatic plants, begomoviruses or CMV, was analyzed. GLMM analyses using biogeographical province as a fixed effect showed that neither the prevalence of symptomatic plants, begomoviruses or CMV did depend on this factor (F5,47>0.797, P<0.557). Similarly, the prevalence of CMV infection did not vary among chiltepin populations (F25,47 = 1.512, P = 0.108). However, population was a factor determining the prevalence of symptomatic plants and begomovirus infection (F25,47>4.369, P<1×10−4). Bonferroni-corrected multiple comparisons showed that this was solely due to the higher prevalence in populations HUJ-chg and LIB-cmc (P<0.046 in 21/25 populations in both cases), and when these populations were removed from the analysis, population was no longer a factor in the prevalence of symptomatic plants and begomovirus infection (F23,45 = 1.984, P = 0.183). Populations HUJ-chg and LIB-cmc were not excluded from further analyses in order to consider as much of the variability in the analyzed factors as possible. These results show that the biogeographical factors analyzed largely do not affect viral and symptom prevalence. Consequently, the populations corresponding to each level of human management could be analyzed together.
The level of human management was a factor in determining the prevalence of symptomatic plants, (F2,47 = 7.619, P<1×10−4), which was significantly lower in wild than in cultivated populations (P<1×10−3), with let-standing populations showing an intermediate value (P≥0.276). Similarly, human management also affected the prevalence of begomovirus infection (F2,47 = 5.774, P = 6×10−4). Values were higher in cultivated populations than in wild (P = 6×10−4) and intermediate in let-standing populations (P≥0.076). However, CMV prevalence did not vary depending on this factor (F2,47 = 1.459, P = 0.243). Thus, increased levels of human management are associated with higher prevalence of symptomatic plants and begomoviruses, but not to prevalence of CMV. We therefore explored which ecological factors varying between populations with different levels of human management were linked to these differences in disease and virus infection risk.
The relative importance of focal host plant density (d), host genetic diversity (expected heterozygosity, He) and species diversity of the habitat expressed either as species richness (SR) or considering also species evenness (Shannon index, Sh), on chiltepin populations was analyzed by including these variables in a PCA. Relevant statistical parameters of these ecological factors are provided in Table S1 of the Supporting Information. Parallel PCAs were performed considering all populations together and individually for each level of human management (Table 3). Importantly, SR and Sh loaded in the same PC in all cases. Since both variables represent the same ecological factor, we performed separate PCAs considering either SR or Sh, but the choice of index did not alter the results (data not shown).
The PCA using the data set that included all the populations (All) yielded three main PCs collectively explaining 95.7 percent of the total variance. Species diversity (SR and Sh) was highly associated with PC1, d with PC2, and He with PC3 (Squared loadings>81.9) (Table 3, All column). The PCA restricted to wild populations, largely mirrored the results obtained with the All data set. However, the fraction of total variance explained by PC1 was higher, and that explained by PC2 and PC3 lower than in the All analysis (Table 3, Wild column).
PCAs considering either let-standing or cultivated populations separately yielded PCs explaining similar percentages of the variance than in the All data set. However, variables loading in each PC differed from All and wild data sets. For let-standing populations, He was now associated with PC1, SR and Sh with PC2, and d with PC3 (Squared loadings>82.7) (Table 3, Let-standing column). Similarly, in cultivated populations He was associated with PC1, SR/Sh with PC2, and d with PC3 (Squared loadings>85.0) (Table 3, Cultivated column). Importantly, SR, Sh and d loaded positively into their respective PCs, but the loading of He was always negative (not shown). The results above indicate that the relative importance of the ecological factors considered in this study vary depending on the level of human management of the chiltepin population.
To determine how human management affects species and genetic diversity, and plant density, we performed GLMM analyses on each PC obtained with the All data set using level of human management as a factor. The three major PCs significantly differed depending on the level of human management (F2,24≥4.995, P≤0.015). Values of PC1 (species diversity) were significantly higher in wild than in cultivated populations (P = 0.045), with intermediate values in let-standing populations. The opposite trend was observed for PC2 (plant density). For PC3 (host genetic diversity), values for cultivated populations were lower than in let-standing and wild populations (P≤0.015), the later two types not differing (P = 0.952).
To further explore the association between the considered ecological factors and disease risk, the influence of each ecological factor on symptom, begomovirus and CMV prevalence was studied using model selection analyses. For the All data set, symptom, begomovirus and CMV prevalence were chiefly determined by species diversity, either measured as SR (Table 4) or Sh (not shown). The model including only species richness was unambiguously the best (ω>0.79) (Table 4). In wild populations, symptom prevalence was mainly associated with species richness and host density (ω = 0.45 and 0.33, respectively), and begomovirus prevalence was chiefly associated with host density (ω = 0.40). However the best-ranked model included also host genetic diversity (ω = 0.55). Although the best-ranked model explaining CMV prevalence included all the ecological factors, the single factor that best explained the variable was host genetic diversity (ω = 0.12) (Table 4). Model selection analyses were largely similar for let-standing and cultivated populations. In both types of populations, host genetic diversity best explained symptom and begomovirus prevalence (ω = 0.51 in both cases), with the model including all the ecological factors showing slightly lower weight. Finally, host density chiefly determined CMV prevalence (ω = 0.57 and 0.35), but the model considering all factors showed slightly higher weight in cultivated populations (Table 4). Thus, these analyses are in agreement with the PCA.
The effect of ecological factors in symptom and virus prevalence was analyzed by bivariate analyses of each factor onto the prevalence of symptomatic plants, begomoviruses and CMV. For the All data set, SR was negatively associated with symptom and begomovirus prevalence (P<0.050), explaining 31.1% and 20.2% of the variance in these variables, respectively (Figure 2 and Table S2). Therefore, species diversity was the primary predictor of symptom and begomovirus prevalence. In wild populations, SR explained 20.2% of the variance being negatively associated with symptom prevalence (P = 0.026) (Figure 2 and Table S2), and d explained 27.7% (P = 0.039) of the variance in symptom prevalence, and 44.4% (P = 0.038) of the variance in begomovirus prevalence. Thus, in wild populations species diversity had also a principal role in determining symptom prevalence, with a lesser effect of plant density in symptom and begomovirus prevalence. In contrast, in let-standing populations He was negatively associated with symptom and begomovirus prevalence (P≤0.025), explaining 85.7% and 65.1% of the variance in these two traits, respectively. In addition, a negative correlation between SR and CMV prevalence was found (P = 0.041, 45.8% of the variance explained), and d showed a positive association with CMV prevalence (P = 0.048, 31.1% of the variance explained) (Figure 2 and Table S2). Finally, He in cultivated populations was also negatively associated with symptom prevalence (P≤0.015, 55.2% of the variance explained), and d explained 28.5% of the variance in CMV prevalence (P = 0.022) (Figure 2 and Table S2). Parallel bivariate analyses using the PCs associated with species richness, Shannon index, expected heterozygosity and host density of each PCA, instead of the original variables yielded similar results (Table 3 and Figure S1).
We have analyzed the prevalence of virus disease and virus infection in populations of a wild plant to test whether increased host density and decreased host genetic diversity in agroecosystems, as compared with wild ecosystems, favors disease risk. These two classical hypotheses of plant pathology [6], [7] are particular cases of a more general one, which is receiving much attention recently, stating that habitat biodiversity is a determinant of disease risk [4] and may be at the root of disease emergence [3], [4]. The wild pepper or chiltepin was the focal host for this study, taking advantage of some unique characteristics of this species. First, wild populations of chiltepin are found in a large variety of habitats in different biogeographical provinces of Mexico [24], [25], which anticipated large differences in species diversity among habitats, as was indeed the case (Table S1). Second, the genetic diversity of wild chiltepin populations differs according to their geographical origin as shown for the 10 wild populations analyzed here [25]. Last, chiltepin populations show different levels of human management, including populations of let-standing plants, which are not sown or planted, but are tolerated or protected in anthropic habitats; and cultivated populations, in which plants are sown in home gardens or in small traditional plots.
The risk of virus disease was estimated as the prevalence of symptomatic plants. Although unapparent virus infection may affect plant fitness [40], we call here diseased plants those showing macroscopic symptoms. This is grounded in our (unpublished) observations of a fecundity reduction in symptomatic plants as compared to both infected or non-infected asymptomatic ones. However, we are aware that prevalence of symptomatic plants may underestimate disease risk, if symptom development were correlated with increased host mortality. Considering these caveats, the risk of disease was positively correlated with the level of human management of the population, being higher in cultivated than in wild populations, and intermediate in let-standing populations. Hence, results support the concept that transition of host habitat from wild ecosystems to agricultural ones results in an increase of disease risk.
A GLMM analysis of the variation of three PCs – associated with species diversity, host genetic diversity and host density – according to the level of human management, strongly suggested that the higher disease risk associated with increased human management is determined by a reduction of biodiversity, both as species diversity and host genetic diversity, and/or by an increased host density. It should be noted that habitat biodiversity and host genetic diversity, estimated as SR and He, respectively, vary along a continuum over the three levels of human management of chiltepin populations, which should avoid spurious associations. However, we cannot discard that other factors structured according to the level of human management, not specifically addressed in this work, may influence disease risk. Examples could be time of exposure to virus infection, and nutrient availability. Nevertheless, the three PCs associated to the analyzed ecological factors explained more than 95% of the variance of the analyzed variables, regardless that all populations were considered together or differentiating between wild, let-standing and cultivated populations. Therefore, although other ecological factors might have minor effects on disease risk, those here considered accounted for most of the variance within and between levels of human management, and may largely explain the emergence of viral disease associated with human management of chiltepin populations.
More specifically, both PC and model selection analyses showed that, for the All and wild data sets, species diversity of the habitat was the major predictor of disease risk (Figure 2, Tables 3 and 4). For let-standing and cultivated populations host genetic diversity was the major predictor of disease risk (Figure 2, Tables 3 and 4). The risk of infection by begomoviruses, which mostly explained symptoms, followed a largely similar pattern, except for a noticeable role of host density in determining virus prevalence in wild populations. These results support the dilution effect hypothesis for a plant-virus system. Moreover, they stress the importance of preserving biodiversity to maintain ecosystem services, a key concept in conservation biology [4]. Results also agree with most analyses of a variety of animal [5], [41] and plant systems [9]–[14], contributing to extend this hypothesis to plant virus diseases. The relationship between biodiversity and disease risk has received comparatively little attention in wild plant-infecting viruses. To our knowledge, Cereal- and Barley yellow dwarf luteoviruses (C/BYDV), which infect many species of grasses and are transmitted in a persistent manner by aphids in a highly species-specific way, is the best characterized system. In this case, most results are compatible with the amplification effect hypothesis, although the complex relationships between grass species, vector multiplication, and virus multiplication/transmission, make the effect of biodiversity on disease risk largely dependent on species composition [18]–[21], [42], [43]. Differences in life histories between luteoviruses and begomoviruses, which cause most symptoms of virus disease in chiltepin (Tables 1 and 2), could explain why effects of biodiversity vary between both systems. Begomoviruses have a narrow host range [29]–[31], and are persistently transmitted by B. tabaci, which has a wide host range [32]. Consequently, the larger the species diversity of the habitat, the larger the number of plant species in which B. tabaci can feed, and the lower fraction of meals resulting in begomovirus transmission to the focal host, resulting in host encounter reduction (sensu [5]). Interestingly, other reports of a dilution effect of biodiversity also refer to persistently transmitted viruses in which the host range of the virus is narrower than that of the vector [44], [45].
Importantly, which of the two different components of biodiversity was the primary predictor of disease and begomovirus infection risk depended on the level of human management. The reduced weight of species diversity in anthropic habitats could be explained by species diversity being largely reduced in cultivated vs. wild populations, and not varying largely among let-standing populations (Table S1). Host genetic diversity has been shown to have a negative effect on the risk of fungal diseases in crops [46]–[48]. Results from fungal pathogens were interpreted as due to differences in resistance-susceptibility among host genotypes, resulting in decreased transmission efficiency [46]–[48]. This mechanism could be also invoked to explain our results as differences in resistance to begomovirus infection have been reported among chiltepin genotypes [49]. However, genotype diversity might also reduce pathogen transmission by other mechanisms, for instance, microenvironment changes [13] or modification of the behavior of insect vectors [50].
The reduction of disease and begomovirus-infection risk with higher biodiversity was not coupled to a lower chiltepin density, as host plant density always loaded into a different PC than species or host diversity. An accepted axiom in plant pathology is that higher host density leads to higher disease risk. However, data are scarce and mostly inconclusive [6], [8], [9], and the effects of biodiversity and host density on disease risk are often difficult to differentiate [4]. The few works that attempted to differentiate these effects yielded contrasting results: density was the primary factor determining disease risk [10], [11] or there were independent and complex interactions between the effects of both factors [22]. The methodology used here avoids artificial correlations [51] and allowed disentangling the effects of these two ecological factors on disease and begomovirus infection risk.
Interestingly, infection by CMV followed a different pattern: infection risk did not depend on the level of human management, and host plant density was a relevant parameter in managed populations, but not in wild ones. The different pattern of infection risk found for begomoviruses and CMV could be due to differences in their life histories. At odds with begomoviruses, CMV is a generalist regarding the host and the vector and, perhaps more importantly, it is transmitted in a non-persistent manner. While persistent transmission is effected during feeding periods among plants that are hosts of the aphid vector, non-persistent transmission occurs during probing visits to plants that need not be hosts of the aphids, which remain viruliferous for short periods of time [52]. Thus, proximity of plants susceptible to the virus could be more important than biodiversity in determining CMV infection risk, similarly to directly transmitted fungi infecting leaves [10], [11]. Consequently, the mechanisms of transmission, in addition to the host range of the pathogen and/or its vectors, could be a primary factor in determining the relationship between biodiversity and disease risk, an unexplored issue, to our knowledge.
Finally, a larger fraction of begomovirus- or CMV-infected plants showed symptoms in managed populations than in wild ones, strongly suggesting a higher virulence of virus infection in the former, perhaps due to a higher susceptibility of plants in human-managed populations to virus infection and its effects. The relationship between host physiological condition and disease susceptibility is an underexplored subject [53]. However, we could speculate that plants of managed populations, which benefit from higher levels of water and/or nutrients than those from wild habitats, as shown by their production of about five times as much fruits (our unpublished observations), would be more competent hosts for virus vectors [20], [54]. This would encourage more frequent and longer meals, thus being under higher inoculum pressure of persistently transmitted viruses. Also, a more favorable host condition could result in higher levels of virus multiplication [21], [54], [55]. If this were the case, in addition to suffering more virulent virus infections, plants in cultivated and let-standing populations would be more competent hosts for virus vectors, virus multiplication and transmission. These factors would contribute to the higher disease risk, and thus to disease emergence, in human managed populations, regardless of the ecological factors here analyzed. However, we cannot exclude that the larger proportion of symptomatic plants in managed habitats would be the result of increased life span of infected plants due to the enhanced availability of resources in cultivated and let-standing populations, which could contribute to explain our observations.
In summary, our results show the important role of biodiversity reduction in the emergence of viral diseases associated to human management of plant populations. Our work also suggest that other ecological and genetic factors, perhaps resulting in increased virulence in anthropic habitats, need to be considered in order to fully understand the dynamics of emergence, which should be the subject of future research.
|
10.1371/journal.ppat.1000428 | The SARS-Unique Domain (SUD) of SARS Coronavirus Contains Two Macrodomains That Bind G-Quadruplexes | Since the outbreak of severe acute respiratory syndrome (SARS) in 2003, the three-dimensional structures of several of the replicase/transcriptase components of SARS coronavirus (SARS-CoV), the non-structural proteins (Nsps), have been determined. However, within the large Nsp3 (1922 amino-acid residues), the structure and function of the so-called SARS-unique domain (SUD) have remained elusive. SUD occurs only in SARS-CoV and the highly related viruses found in certain bats, but is absent from all other coronaviruses. Therefore, it has been speculated that it may be involved in the extreme pathogenicity of SARS-CoV, compared to other coronaviruses, most of which cause only mild infections in humans. In order to help elucidate the function of the SUD, we have determined crystal structures of fragment 389–652 (“SUDcore”) of Nsp3, which comprises 264 of the 338 residues of the domain. Both the monoclinic and triclinic crystal forms (2.2 and 2.8 Å resolution, respectively) revealed that SUDcore forms a homodimer. Each monomer consists of two subdomains, SUD-N and SUD-M, with a macrodomain fold similar to the SARS-CoV X-domain. However, in contrast to the latter, SUD fails to bind ADP-ribose, as determined by zone-interference gel electrophoresis. Instead, the entire SUDcore as well as its individual subdomains interact with oligonucleotides known to form G-quadruplexes. This includes oligodeoxy- as well as oligoribonucleotides. Mutations of selected lysine residues on the surface of the SUD-N subdomain lead to reduction of G-quadruplex binding, whereas mutations in the SUD-M subdomain abolish it. As there is no evidence for Nsp3 entering the nucleus of the host cell, the SARS-CoV genomic RNA or host-cell mRNA containing long G-stretches may be targets of SUD. The SARS-CoV genome is devoid of G-stretches longer than 5–6 nucleotides, but more extended G-stretches are found in the 3′-nontranslated regions of mRNAs coding for certain host-cell proteins involved in apoptosis or signal transduction, and have been shown to bind to SUD in vitro. Therefore, SUD may be involved in controlling the host cell's response to the viral infection. Possible interference with poly(ADP-ribose) polymerase-like domains is also discussed.
| The genome of the SARS coronavirus codes for 16 non-structural proteins that are involved in replicating this huge RNA (approximately 29 kilobases). The roles of many of these in replication (and/or transcription) are unknown. We attempt to derive conclusions concerning the possible functions of these proteins from their three-dimensional structures, which we determine by X-ray crystallography. Non-structural protein 3 contains at least seven different functional modules within its 1922-amino-acid polypeptide chain. One of these is the so-called SARS-unique domain, a stretch of about 338 residues that is completely absent from any other coronavirus. It may thus be responsible for the extraordinarily high pathogenicity of the SARS coronavirus, compared to other viruses of this family. We describe here the three-dimensional structure of the SARS-unique domain and show that it consists of two modules with a known fold, the so-called macrodomain. Furthermore, we demonstrate that these domains bind unusual nucleic-acid structures formed by consecutive guanosine nucleotides, where four strands of nucleic acid are forming a superhelix (so-called G-quadruplexes). SUD may be involved in binding to viral or host-cell RNA bearing this peculiar structure and thereby regulate viral replication or fight the immune response of the infected host cell.
| The SARS coronavirus (SARS-CoV) is much more pathogenic for humans than any other coronavirus. Therefore, protein domains encoded by the SARS-CoV genome that are absent in other coronaviruses are of particular interest, because they may be responsible for the extraordinary virulence. The most prominent such domain has been identified by bioinformatics as part of non-structural protein 3 (Nsp3) of the virus and appropriately named the “SARS-unique domain” (SUD) [1]. With a molecular mass of 213 kDa, Nsp3 is the largest of the non-structural proteins of SARS coronavirus (see Figure 1). Comprising 1922 amino-acid residues (polyprotein 1a/1ab residues Ala819 to Gly2740), SARS-CoV Nsp3 is larger than the entire replicase of Picornaviridae. It contains at least seven subdomains [2]: An N-terminal acidic domain (Ac, also called Nsp3a); an X-domain (also designated as ADRP, or Nsp3b); the SUD (Nsp3c); a papain-like proteinase, PL2pro (also called Nsp3d); and additional domains (Nsp3e–g) that include a transmembrane (TM) region.
At present, it is completely unclear whether and how the individual domains of Nsp3 interact with one another or with other components of the coronaviral replicase complex. Also, some of them possibly recognize proteins of the infected host cell [2]. In the absence of functional data on these domains, attempts have been made to derive their possible biological role from their three-dimensional structures (see [3] for a review). The NMR structure of an N-terminal fragment of the acidic domain (Nsp3a) has revealed a ubiquitin-like fold complemented by two additional short α-helices ([4], PDB code 2IDY). NMR chemical-shift analysis suggested that these non-canonical structural elements might bind single-stranded RNA with some specificity for AUA-containing sequences, although the KD values observed are relatively high (∼20 µM). Interestingly, a second ubiquitin-like domain occurs in Nsp3, as part of the papain-like proteinase (PL2pro, Nsp3d, [5]; PDB code 2FE8). The PL2pro cleaves the viral polyprotein after two consecutive glycine residues to release Nsp1, Nsp2, and Nsp3, respectively (The remaining cleavage reactions are performed by the coronaviral main proteinase (Mpro; [6]–[8])). In addition to its proteolytic activities on the N-terminal third of the polyproteins, the SARS-CoV PL2pro has also been shown to be a deubiquitinating enzyme [9]–[12]. Lindner et al. [13] have shown that in addition to its proteolytic and deubiquitinating activity, the SARS-CoV PL2pro acts as a de-ISGylating enzyme. Induction of ISG15 and its subsequent conjugation to proteins protects cells from the effects of viral infection [14],[15]. Since the ISG15 gene is induced by interferon as part of the antiviral response of the innate immune system, the de-ISGylation activity of Nsp3d could explain the suppression of the interferon response by the papain-like protease, in addition to a possible direct interaction between the PL2pro and IRF3 [16].
Among the subdomains of the Nsp3 multidomain protein, there is also the so-called “X- domain” (Nsp3b), which shows structural homology to macrodomains. The latter name refers to the non-histone-like domain of the histone macro2A [17]–[19]. In animal cells, such domains are occasionally physically associated with enzymes involved in ADP-ribosylation or ADP-ribose metabolism. Because of this linkage and on the basis of sequence similarity to Poa1p, a yeast protein involved in the removal of the 1″-phosphate group from ADP-ribose 1″-phosphate (a late step in tRNA splicing; [20]), it has been proposed that the coronaviral X-domains may have the function of ADP-ribose-1″-phosphatases (ADRPs; [21]). The crystal structures of X-domains of SARS-CoV [22],[23] as well as of HCoV 229E and Infectious Bronchitis Virus (IBV) [24] show that the protein has the three-layer α/β/α fold characteristic of the macrodomains.
Embedded between the X-domain (Nsp3b) and the PL2pro (Nsp3d), the SARS-unique domain (SUD; Nsp3c) fails to show sequence relationship to any other protein in the databases [1]. We have produced full-length SUD (residues 389 to 726 of Nsp3), and a more stable, shortened 264-residue version (residues 389 to 652; henceforth called SUDcore), by expression in Escherichia coli. This definition of the boundaries of the SUD is based on the structural results described here. We report crystallization of SUDcore and its X-ray structure in two crystal forms, at 2.2 and 2.8 Å resolution, respectively. The structure turns out to consist of two further copies of the macrodomain, in spite of the complete absence of sequence similarity. In addition, we demonstrate that each of the subdomains binds G-quadruplexes, both in DNA and RNA fragments, and that selected mutations of lysine residues in the first subdomain, SUD-N, lead to reduced nucleic-acid binding, whereas those in the second subdomain, SUD-M, abolish it.
Out of the many SUD constructs designed and tested by us, SUDcore (Nsp3 residues 389–652) turned out to be relatively stable and could be crystallized (Table 1). Two crystal forms were observed under identical crystallization conditions: Form-1 crystals were monoclinic (space group P21, two SUDcore molecules per asymmetric unit) and diffracted X-rays to 2.2 Å resolution; form-2 crystals were triclinic (space group P1, four SUDcore molecules per asymmetric unit) and diffracted to 2.8 Å. Both structures were determined by molecular replacement (see Materials and Methods). The r.m.s. deviations (on Cα atoms) between the models derived from the two different crystal structures are around 0.7 Å.
The models have good stereochemistry (Table 1). 94.7% of the amino-acid residues are in the favoured regions of the Ramachandran plot and 4.6% are in allowed regions. 0.6% are outliers. In all six independent copies of the SUDcore monomer, residue Val611 adopts forbidden conformational angles. This residue is located in a turn described by the polypeptide chain where it leaves the subdomain interface (see below) and reaches the surface of the molecule. The side chain makes a hydrophobic contact across the subdomain interface and is also contacting the side chain of Phe406 of a symmetry-related SUDcore dimer in the crystal lattice in the monoclinic crystal form (this also applies to two of the four monomers in the triclinic form).
SUDcore exhibits a two-domain architecture (Figure 2A). The N-terminal subdomain (SUD-N) comprises Nsp3 residues 389–517, and the C-terminal subdomain of SUDcore contains residues 525–652. We call the latter the “middle SUD subdomain”, or SUD-M, because full-length SUD has a C-terminal extension of 74 residues compared to SUDcore. The SUD-N and SUD-M subdomains have a similar fold and can be superimposed with an r.m.s.d. of 3.3–3.4 Å (based on Cα positions); they share 11% sequence identity (see Figure 2C for a structural alignment). Of the 14 amino-acid residues identical between the two subdomains, four form a conserved Leu-Glu-Glu-Ala motif at the N-terminus of helix α4. The linker between the two subdomains (residues 518–524) has no visible electron density. This is due to elevated mobility of the linker, rather than proteolytic cleavage, since we showed by SDS-PAGE of dissolved crystals that the SUDcore polypeptide (in the presence of β-mercaptoethanol) has the apparent molecular mass to be expected (∼29 kDa; not shown). In addition to the linker, SUD-N and SUD-M are connected by a disulfide bond between cysteines 492 and 623 (Figure 2B). Disulfide bonds are rare in cytosolic proteins, but in coronaviral Nsps, examples of such bonds have been reported [25],[26].
The fold of each SUD subdomain is that of a macrodomain (Figure 2A). Macrodomains consist of a largely parallel central β-sheet surrounded by 4–6 α-helices. The order of regular secondary-structure elements in SUD-N is βN1-αN1-βN2-αN2-βN3-βN4-αN3-βN5-αN4-βN6, and in SUD-M βM1-αM1-βM2-αM2-βM3-βM4-αM3-βM5-αM4-βM6-αM5. The topology of the β-strands is β1–β6–β5–β2–β4–β3, all of which are parallel except β3 (Figure 2A). Between the two subdomains, most of the secondary-structure elements are conserved with respect to their position in the three-dimensional structure, although they often differ in length. This is particularly obvious for α-helix 1, which comprises just four residues in the N-terminal subdomain but eleven in the M subdomain. Similarly, α-helix 2 has 5 vs. 10 amino-acid residues in the two subdomains. In general, the strands of the central β-sheet appear to align better between the two subdomains than do the α-helices.
Each of the SUDcore subdomains is related to the macrodomain of the histone macro2A ([18]; PDB code 1ZR3, molecule C; for SUD-N: Z-score 9.8, r.m.s.d. 2.5 Å for 112 out of 184 Cα atoms, 12% sequence identity; for SUD-M: Z-score 8.6, r.m.s.d. 2.8 Å for 115 out of 184 Cα atoms, 19% sequence identity). Called “X-domains”, single macrodomains are also found in alphaviruses, in hepatitis E virus, and in rubella virus, in addition to coronaviruses [27],[28]. The SARS-CoV X-domain (Nsp3b), the domain immediately preceding the SUD in Nsp3, shares no recognizable sequence identity with SUD-N (12%) or SUD-M (7%) (Figure 2C), but its three-dimensional structure [22],[23] (PDB code 2ACF, chain A) can be superimposed onto each of the two SUD subdomains with an r.m.s.d. (based on Cα atoms) of 2.7 Å and 2.3 Å, respectively (Figure 2D). Thus, within Nsp3, SARS-CoV has three macrodomains aligned one after the other.
In both crystal forms, SUDcore displays the same head-to-tail dimer, with the SUD-N subdomain of monomer A interacting with the SUD-M subdomain of monomer B, and vice versa. Approximately 1130 Å2 of solvent-accessible surface per monomer is buried upon dimerization (Figure 3). Due to the two-domain architecture of each monomer, the resulting four lobes give the dimer a quasi-tetrahedral shape (Figure 3A). Involving ∼10 hydrogen bonds and four well-defined salt-bridges (AspB440…ArgA554, ArgB473…GluA619, ArgB554…AspA440, and GluB619…ArgA473), interactions between the monomers are largely hydrophilic. As to be expected, the structures of the monomers are very similar to one another, with r.m.s.d. values (for Cα atoms) of 0.58 Å between monomers A and B of the monoclinic crystal form, and 0.11–0.37 Å between monomers A–D of the triclinic form. The structure of SUD-M alone is even better conserved between the individual copies of SUDcore. Also, the fold of the SUD-M subdomain is similar to the model of the SUD fragment 527–651 derived from NMR measurements, which was published very recently (r.m.s.d. ∼0.9 Å) [29].
The function of the coronaviral X-domain is still unclear; for some coronaviruses such as HCoV 229E and SARS-CoV, it has been shown to exhibit a low ADP-ribose-1″-phosphate phosphatase (Appr-1″-pase, occasionally also called “ADRP”) activity and to bind the product of the reaction, ADP-ribose [21]–[23],[30]. However, the two subdomains of SUDcore do not bind ADP-ribose, as we have demonstrated by zone-interference gel electrophoresis (Figure S1).
When we investigated possible interactions between SUD and nucleic acids by zone-interference gel electrophoresis, we found that the domain binds oligo(G) and oligo(dG) stretches with a KD of ∼1 µM, but not oligo(dA), (dC), or (dT) [31]. Single-stranded nucleotides of random sequence are only bound if they are longer than ∼15 nucleotides. Here we demonstrate that each of the two individual SUD subdomains also binds oligo(dG) (Figure 4A). With oligo(dH), where H stands for A, C, or T, but not G, only very small gel shifts, if at all, were observed. As oligo(G) stretches are known to form G-quadruplexes, i.e. four-stranded nucleic-acid structures formed by contiguous guanines [32], we also examined the binding to the oligodeoxynucleotide 5′-GGGCGCGGGAGGAATTGGGCGGG-3′, a G-rich sequence present in the bcl-2 promoter region. This oligonucleotide has been shown by NMR spectroscopy to form a G-quadruplex ([33]; PDB code 2F8U). We found that both full-length SUD and SUDcore do indeed bind this oligodeoxynucleotide and that this process is enhanced by the addition of K+ ions, which are known to stabilize G-quadruplex structures (Figure 4B). In agreement with the ability of SUD to non-specifically bind to oligonucleotides of >15 bases [31], both SUD and SUDcore were found to bind the reverse-complementary sequence, but with low affinity and, more importantly, independent of K+ ions.
As there is no evidence for SARS-CoV Nsp3 entering the nucleus and binding to DNA, we examined whether SUD would bind to an RNA known to form a quadruplex structure. Indeed, zone-interference gel shift experiments revealed major shifts for both SUD and SUDcore in the presence of the oligoribonucleotide 5′-UGGGGGGAGGGAGGGAGGGA-3′, which is a protein-binding element in the 3′-nontranslated region of chicken elastin mRNA [34] and forms G-quadruplexes [35] (Figure 4C). Furthermore, we observed a significant gel shift for SUDcore when we added the short oligonucleotide UGGGGU, which has also been shown to form a G-quadruplex ([36]; PDB code 1J8G). This shift was also enhanced by the addition of K+ (Figure 4D). Thus, SUD binds RNA (rG)-quadruplexes and DNA (dG)-quadruplexes with comparable affinity.
Inspection of the structure of the SUD dimer reveals a central narrow cleft running across the dimer surface, but distinct from the monomer-monomer interface (Figure 3C), which could be a binding site for another protein. In addition, there are several positively charged patches in the center of the dimer (Figure 3B), and on its backside (Figure 3C), which could be involved in binding to G-quadruplexes. We have prepared four sets of mutations by replacing lysine residues (and one glutamate) in these patches by alanines. The first two pairs of mutations, K505A+K506A (M1, at the end of helix αN4) and K476A+K477A (M2, in the loop between αN3 and βN5), are located on the surface of the SUD-N subdomain and lead to reduced shifts with G-quadruplexes in the zone-interference gel electrophoresis experiment, both with the G-quadruplex from the bcl-2 promoter region (Figure 5) and with (dG)10 (not shown). The second set of mutations, K563A+K565A+K568A (M3) and K565A+K568A+E571A (M4) are located in the loop connecting αM2 and βM3 of the SUD-M subdomain and abolish G-quadruplex binding altogether (Figure 5), again with both oligonucleotides.
When the SARS-unique domain was first predicted [1], the boundaries of the domain were set approximately at Nsp3 residues 352 and 726. We made major efforts to produce this protein in a stable form, but with little success. Only when we used in-vitro protein synthesis, were we able to obtain small amounts of a relatively stable preparation comprising Nsp3 residues 349–726 [31]. At the N-terminus of this construct, up to eleven residues actually correspond to the C-terminus of the preceding X-domain (Nsp3b). When we expressed a gene construct coding for SUD (349–726) in E. coli, we observed rapid proteolytic degradation of the N-terminal segment. The relatively stable intermediate obtained had its N-terminus at Nsp3 residue 389. The N-terminal segment ∼359–388 is predicted to be intrinsically unfolded by several prediction programs (not shown). Therefore, we assume segment 359–388 to be merely a linker between Nsp3b and SUD, and 389 to be the first residue of the latter. This assignment is justified by the observation that in our crystal structures reported here, the SUD-N subdomain is a complete macrodomain without any residues lacking at the N-terminus. Therefore, the protein corresponding to Nsp3 residues 389–726 is called “full-length SUD” here.
In this communication, we describe the crystal structures at 2.2 Å and 2.8 Å resolution (monoclinic and triclinic form, respectively) of the core of the SARS-unique domain (SUDcore, Nsp3 residues 389–652). SUDcore turns out to consist of two subdomains, SUD-N (Nsp3 residues 389–517) and SUD-M (525–652), each exhibiting the fold of a macrodomain. The two subdomains are connected by a flexible linker (residues 518–524) and a disulfide bond. Even though coronavirus replication occurs in the cytosol, where the environment is reductive, it is unlikely that the formation of this disulfide is an artifact owing to handling of the protein: As the linker between the SUD-N and SUD-M subdomains is very short (seven residues), and the mutual orientation of the subdomains is fixed due to the tight dimerization, cysteine residues no. 492 and 623 will be very close to one another irrespective of the exact conformation of the linker. In fact, disulfide bonds are not uncommon in coronaviral non-structural proteins (Nsps) involved in RNA replication or transcription. Among others, they have been observed in HCoV-229E Nsp9 [25] and turkey coronavirus Nsp15 [26], but in these cases, the disulfide bond connects two subunits of the homo-oligomeric proteins, whereas the occurrence in SUDcore is the first case of an intramolecular disulfide bond described in a coronavirus Nsp.
Coronavirus replication in the perinuclear region of the cell is localized to double-membrane vesicles that have been hijacked from the endoplasmic reticulum or late endosomes [37]–[40]. These vesicles are around 200–350 nm in diameter and present alone or as clusters in the cytosol [38]. The milieu inside or at the surface of these vesicles is unknown, but it is well possible that it is partially oxidative. It has also been speculated [25] that formation of disulfide bonds may be a way for the coronaviral Nsps to function in the presence of the oxidative stress that is the consequence of the viral infection [41]–[43].
Our identification of two macrodomains in SUDcore brings the number of these domains in SARS-CoV Nsp3 to three. What are the functions of these modules? The original SARS-CoV “X-domain” (Nsp3b) has been shown to have low ADP-ribose-1″-phosphate phosphatase (Appr-1″-pase or “ADRP”) activity [21]–[23]. However, this assignment is controversial. A nuclear Appr-1″-pase (Poa1p in yeast, [20]) is an enzyme of a tRNA metabolic pathway, but there is no evidence for coronavirus Nsp3 ever being translocated to the nucleus, and the other enzymes involved in this pathway are missing in coronaviruses (with the exception of the cyclic 1″,2″-phosphodiesterase (CPDase) in group 2a viruses such as Mouse Hepatitis Virus, Bovine Coronavirus, and Human Coronavirus OC43). Therefore, it has been proposed that the X-domain may be involved in binding poly(ADP-ribose), a metabolic product of NAD+ synthesized by the enzyme poly(ADP-ribose) polymerase (PARP; [23]). However, we have recently demonstrated that the X-domain of Infectious Bronchitis Virus (IBV) strain Beaudette, a group-3 coronavirus, does not have significant affinity to ADP-ribose [24]. This can be explained on the basis of crystal structures: In the X-domain (Nsp3b) of SARS-CoV [23], and in that of HCoV 229E [24], a stretch of three conserved glycine residues is involved in binding the pyrophosphate unit of ADP-ribose, whereas in the corresponding domain of IBV strain Beaudette (but not in all IBV strains, see [44]), the second glycine is replaced by serine, leading to steric interference with ADP-ribose binding [24]. In the two SUD subdomains, the triple-glycine sequence is not conserved (see Figure 2C), and hence, they do not bind ADP-ribose either.
Neuman et al. [2] reported that full-length SUD binds cobalt ions, whereas a domain called SUD-C by these authors, which is however almost identical (residues 513–651) to our SUD-M (525–652), does not. From this, they concluded that the metal-binding activity is associated with the cysteine residues in the N-terminal subdomain. We were also able to observe binding of cobalt ions to SUDcore by following the occurrence of a peak at 310 nm in the UV spectrum, which, in contrast to the data presented by Neuman et al. [2], could be reverted by addition of zinc ions. However, when we removed the N-terminal His-tag, this phenomenon could no longer be observed. Furthermore, we note that of the four cysteine residues in the SUD-N subdomain (residues 393, 456, 492, and 507), 456 and 507 are non-accessible in the interior of the subdomain, and 492 is involved in the buried disulfide bond to Cys623; therefore, Cys393 and perhaps the solvent-exposed His423 would remain the only potential ligands for cobalt ions in SUD-N. However, these residues are >12 Å apart and thus unlikely to chelate cobalt ions.
For SUD-M, a recent publication [29] reported binding to oligo(A). However, we fail to observe this (Figure 4A, lane labeled “A”). Instead, we have demonstrated that full-length SUD and SUDcore bind oligodeoxynucleotides and oligoribonucleotides that form G-quadruplexes. For full-length SUD and SUDcore, we had previously shown binding to oligo(dG) and oligo(G) stretches [31], but the demonstration here of oligo(dG) binding to the individual SUDcore subdomains, SUD-N and SUD-M, is unexpected because their overall electrostatic properties are very different from one another: SUD-N is acidic (pI = 5.3), whereas SUD-M is basic (pI = 9.0). However, even SUD-N shows surface patches with positive electrostatics that could bind nucleic acid (Figure 3B).
We have used automatic docking procedures to place the G-quadruplex found in the bcl-2 promoter region ([33]; PDB code 2F8U) into our crystal structures. One potential binding site identified is in the cleft between the SUD-M and the SUD-N subdomains within the SUDcore dimer (Figure S2A); this binding site is spatially close to the mutations M3 and M4, consistent with the observation that these mutations abolish binding completely. However, we have previously shown by Dynamic Light-Scattering that G-quadruplex binding leads to oligomerization of SUDcore [31]. Consequently, we have also constructed models based on the packing modes of SUDcore dimers observed in our crystal structures. One potential binding site for G-quadruplexes might be in a cleft between two consecutive SUDcore dimers as they occur in both the monoclinic and triclinic crystal forms (Figure S2B), but for confirmation, any of these models will have to await crystallographic determination of the complex. In summary, our mutation experiments demonstrate an involvement of several of the many lysine residues of SUD in binding G-quadruplexes, but as it is probably extended surfaces of SUDcore oligomers that participate in this process, it is not possible to pinpoint any single amino-acid residue.
The target of SUD binding could be G-quadruplexes in RNA of viral or/and cellular origin. The SARS-CoV genome contains three G6-stretches (one on the plus-strand and two on the minus-strand) and an additional two G5-sequences, which could perhaps form local G-quadruplexes. However, the G-stretch binding capabilities of SUD and SUDcore seem to have been optimized for recognition of longer G-rich sequences. By systematic variation of the length of oligo(dG), we found that SUDcore exhibits strongest affinity (KD ∼0.45 µM) for (dG)10 to (dG)14 [31]. The 3′-nontranslated regions of several host-cell mRNAs coding for proteins involved in the regulation of apoptosis and in signaling pathways contain long G-stretches and could also be targets of SUD. Examples of such mRNAs are those coding for the pro-apoptotic protein Bbc3 [45], RAB6B (a member of the Ras oncogene family, [46]), MAP kinase 1 [47], and TAB3, a component of the NF-κB signaling pathway [48]. It is conceivable that these proteins might be targets for the virus when interfering with cellular signaling. Changes in the stability and/or translation efficiency of these mRNAs due to the binding of a viral regulatory factor could result in an altered reaction of the infected cell to apoptotic signals, or it could silence the antiviral response.
The idea that coronaviral X-domains might function as modules binding poly(ADP-ribose) [23] received support from the observation that some macrodomains are connected with domains showing poly(ADP-ribose) polymerase (PARP) activity, i.e. in the so-called macroPARPs (PARP-9 and PARP-14) [49]. There are 18 human genes for members of the PARP family; the prototype enzyme, PARP-1, catalyzes the post-translational modification of many substrate proteins, including itself, in a multitude of cellular processes (DNA repair, transcriptional regulation, energy metabolism, and apoptosis) [50]–[52]. Interestingly, SUD-M and the C-terminal 74-residue subdomain (SUD-C) that is missing in our SUDcore construct together show a ∼15% sequence identity (32% similarity) to the catalytic domain of PARP-1. However, the three-dimensional structures of SUD-M (this work) and the C-terminal domain of PARP-1 [53] are different and cannot be superimposed. Another feature common between SARS-CoV SUD and PARP-1 is that the latter has recently been shown to bind to G-quadruplexes [54], although it is generally assumed that this occurs through the DNA-binding domain rather than the catalytic domain of PARP-1.
PARP-1 and most of its family members are located to the nucleus, while PARP-4 and others predominantly act in the cytoplasm [50]–[52]. PARP-4 is incorporated into vaults, RNA-containing subcellular particles in the cytoplasm [55]. Furthermore, ZAP, a human antiviral protein comprising a C-terminal PARP-like domain devoid of catalytic activity, has been shown to exhibit antiviral activity on alphaviruses [56], which contain an X-domain similar to that of coronaviruses [23],[27],[28]. In addition, ZAP contains an N-terminal zinc-finger domain, a central TiPARP (2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)-inducible PARP) domain, and a WWE domain (a protein-protein interaction module in ubiquitin and ADP-ribose conjugation proteins). In fact, ZAP appears to be part of the human innate immune system and to play a role comparable to APOBEC3G in HIV infection [57]. It is possible that this group of viruses has evolved macrodomains to counteract the antiviral activity of ZAP. Indeed, macrodomains can inhibit PARPs, as has been shown for the macrodomain of the histone mH2A1.1, which downregulates the catalytic activity of PARP-1 [58]. Having three macrodomains at its disposal, SARS-CoV may be much more efficient in knocking down the antiviral response of the host cell than other coronaviruses. Whether this involves a direct interaction between SUD and ZAP or another member of the PARP family, or competition for G-quadruplexes in viral or host-cell RNA, remains to be shown.
Full-length SUD (Nsp3 residues 389–726) and the fragment SUDcore (Nsp3 residues 389–652, previously called “SUDc5b”) of SARS-CoV strain TOR2 (acc. no. AY274119) were produced recombinantly in E. coli as described [31]. The coding regions for the SUD-N subdomain (Nsp3 residues 389–524) and the SUD-M subdomain (Nsp3 residues 525–652) were constructed by introducing an appropriate deletion into the previously described plasmid pQE30-Xa-c5b [31] using site-directed mutagenesis. Plasmids encoding SUD-N and SUD-M were prepared using primers listed in Table S1. The coding regions for four sets of mutations of SUDcore, M1 (K505A+K506A), M2 (K476A+K477A), M3 (K563A+K565A+K568A), and M4 (K565A+K568A+E571A), were constructed by introducing appropriate mutations into plasmid pQE30-Xa-c5b [31] using site-directed mutagenesis. Plasmids encoding these mutants were prepared using primers also listed in Table S1. All plasmids provided an N-terminal His-tag and a short linker sequence encoding a factor-Xa cleavage site. The coding regions of the expression plasmids were verified by DNA sequencing. E. coli M15 (pRep4) was used as expression host for these constructs. SUD-N, SUD-M, and the mutated proteins were purified according to the same protocol as for SUDcore [31].
SUDcore displayed >95% purity in SDS-PAGE, and monodispersity in Dynamic Light- Scattering. Initial crystallization screening was performed using the sitting-drop vapor-diffusion method in 96-well Intelli-Plates (Dunn Laboratories). Several commercial kits (Sigma, Jena Bioscience) were used for the screening. The protein concentration was 6 mg/ml. Using a Phoenix robotic system (Art Robbins), drops were made of 260 nl protein and 260 nl precipitant solution. The optimized crystallization condition consisted of 20% polyethylene glycol monomethyl ether 5000 and 0.2 M ammonium sulfate in 0.1 M morpholinoethane sulfonic acid (pH 6.5). Plate-like crystals grew in 3–5 days, to maximum dimensions of 0.02×0.02×0.01 mm3.
Many SUDcore crystals had to be tested for diffraction until one yielding data to 2.2 Å resolution was found. The best diffracting crystals belonged to space group P21. Under the same crystallization conditions, a second crystal form belonging to space group P1 was observed, diffracting to lower resolution of about 2.8 Å. Crystals were cryoprotected in reservoir solution that included 30% glycerol, and were harvested into a loop prior to flash-cooling in liquid nitrogen. All data were collected at 100 K from a single crystal each at beamline BL14.2, BESSY (Berlin, Germany), using an MX225 CCD detector (Rayonics), or at beamline I911-2 at MAX-lab (Lund, Sweden), using a Mar165 CCD detector (Marresearch). Data were processed with MOSFLM [59], and reduced and scaled using the SCALA [60] program from the CCP4 suite [61]. Crystals belonging to space group P21 had unit-cell parameters a = 46.36 Å, b = 68.55 Å, c = 94.21 Å, β = 99.17°, those belonging to space group P1 had unit-cell parameters a = 68.68 Å, b = 75.52 Å, c = 80.54 Å, α = 77.16°, β = 75.61°, γ = 74.48°. Data-collection statistics for both crystal forms are shown in Table 1. The asymmetric unit of the P21 form contained two SUDcore monomers, giving a Matthews coefficient [62] of 2.5 Å3 Da−1 and a solvent content of 51%, whereas the P1 crystal form had four monomers per asymmetric unit, giving corresponding parameters of 3.2 Å3 Da−1 and 63%.
We attempted to solve the structure by molecular replacement into the P21 form using the NMR coordinates of a subdomain comprising SARS-CoV Nsp3 residues 513–651; PDB code 2JWJ [29],[63]), which is almost identical to the SUD-M subdomain of SARS-CoV Nsp3. Using the program Phaser [64],[65], we could find two solutions, and the C-terminal part of SUDcore was well defined in the electron-density maps. However, for the N-terminal half, only a few segments of poly(Ala) chain could be built into the maps. This starting model was then refined in BUSTER-TNT [66] using Local Structure Similarity Restraints (LSSR) [67] as non-crystallographic symmetry (NCS) restraints to give R and Rfree values of 0.453 and 0.479, respectively. The resulting 2mFo-DFc electron density was subjected to density modification using solvent flattening, histogram matching, and 2-fold NCS-averaging using DM [68]. The averaging masks were calculated and updated using the auto-correlation procedure [69] as implemented in DM. Using the automatic building program BUCCANEER [70] together with REFMAC [71] (as implemented in the CCP4i [72] interface for CCP4) in an iterative procedure for 20 cycles resulted in a model for 501 residues in 10 chains (the longest having 208 residues), in which 448 residues were assigned both a chemical identity and a sequential residue number, while the remaining 53 residues were modeled as poly(Ala) in 8 shorter chains. The R and Rfree values resulting from REFMAC were 0.374 and 0.414, respectively. This model was refined in BUSTER-TNT, again using LSSR as NCS restraints for the common parts in the already sequenced 448 residues of the dimer, to R and Rfree values of 0.269 and 0.316. The improved electron density was again subjected to density modification using DM as detailed above, but using a lower solvent content of 35% as well as anisotropically scaled observed amplitudes as output by BUSTER-TNT. The resulting density-modified and NCS-averaged map was then used for automatic model building using the iterative BUCCANEER/REFMAC procedure described above. This produced a model with 511 residues in 5 chains with 487 residues sequenced. The R and Rfree values from REFMAC for this model were 0.289 and 0.326, respectively.
Since the refinements in BUSTER-TNT at that point showed some problematic low correlations between Fo and Fc at low resolution, the original images collected from the P21 crystal were reprocessed using XDS [73] and SCALA, applying different high-resolution cutoffs for different segments of the collected images. Details for this dataset are given in Table 1. Subsequent refinement of the P21 form with REFMAC, under application of weak NCS restraints, yielded a model with R = 0.211, Rfree = 0.264. The advanced handling of NCS restraints through LSSR in BUSTER-TNT gave a final model R = 0.211 and Rfree = 0.268. The final model in the P21 form comprises 513 residues (A389–A516; A524–A652; B393–B519; B526–B652).
Chain A of the P21 form was used for molecular replacement with the program MOLREP [74] into the P1 form. There was an unambiguous solution for four molecules in the asymmetric unit. This model was refined with BUSTER-TNT (using LSSR for NCS restraints) and rebuilt in Coot [75] to final values of R = 0.223 and Rfree = 0.240. The final model of the P1 form comprises 1014 residues.
The figures were made with PyMOL [76].
The zone-interference gel electrophoresis (ZIGE) device was adapted from Abrahams et al. [77]. ZIGE assays were performed using a horizontal 1% agarose gel system in TBE buffer (20 mM Tris, 50 mM boric acid, 0.1 mM ethylenediaminetetraacetic acid (EDTA), pH 8.3). The protein was incubated at room temperature for 30 min with different concentrations of oligodeoxynucleotides, such as (dG)10 and bcl-2 promoter region (5′-GGGCGCGGGAGGAATTGGGCGGG-3′), or oligoribonucleotides (5′-UGGGGGGAGGGAGGGAGGGA-3′ and 5′-UGGGGU-3′). The samples were mixed with dimethylsulfoxide (DMSO; final concentration 10% (v/v)) and a trace of bromophenolblue (BPB). These protein-oligonucleotide samples were applied to the small slots. Oligonucleotide with the same concentration as in the small slots was also mixed with DMSO and BPB in 1xTBE buffer and applied to the long slots of the gel (total volume 100 µl). Electrophoresis was performed at 4°C for 1 h with a constant current of 100 mA. Staining was performed as outlined in [77].
Protein Data Bank: Coordinates and structure factors have been deposited with accession code 2W2G (P21 crystal form) and 2WCT (P1 crystal form).
|
10.1371/journal.pgen.1002868 | Population Genomics of the Facultatively Mutualistic Bacteria Sinorhizobium meliloti and S. medicae | The symbiosis between rhizobial bacteria and legume plants has served as a model for investigating the genetics of nitrogen fixation and the evolution of facultative mutualism. We used deep sequence coverage (>100×) to characterize genomic diversity at the nucleotide level among 12 Sinorhizobium medicae and 32 S. meliloti strains. Although these species are closely related and share host plants, based on the ratio of shared polymorphisms to fixed differences we found that horizontal gene transfer (HGT) between these species was confined almost exclusively to plasmid genes. Three multi-genic regions that show the strongest evidence of HGT harbor genes directly involved in establishing or maintaining the mutualism with host plants. In both species, nucleotide diversity is 1.5–2.5 times greater on the plasmids than chromosomes. Interestingly, nucleotide diversity in S. meliloti but not S. medicae is highly structured along the chromosome – with mean diversity (θπ) on one half of the chromosome five times greater than mean diversity on the other half. Based on the ratio of plasmid to chromosome diversity, this appears to be due to severely reduced diversity on the chromosome half with less diversity, which is consistent with extensive hitchhiking along with a selective sweep. Frequency-spectrum based tests identified 82 genes with a signature of adaptive evolution in one species or another but none of the genes were identified in both species. Based upon available functional information, several genes identified as targets of selection are likely to alter the symbiosis with the host plant, making them attractive targets for further functional characterization.
| Facultative mutualisms are relationships between two species that can live independently, but derive benefits when living together with their mutualistic partners. The facultative mutualism between rhizobial bacteria and legume plants contributes approximately half of all biologically fixed nitrogen, an essential plant nutrient, and is an important source of nitrogen to both natural and agricultural ecosystems. We resequenced the genomes of 44 strains of two closely related species of the genus Sinorhizobium that form facultative mutualisms with the model legme Medicago truncatula. These data provide one of the most complete examinations of genomic diversity segregating within microbial species that are not causative agents of human illness. Our analyses reveal that horizontal gene transfer, a common source of new genes in microbial species, disproportionately affects genes with direct roles in the rhizobia-plant symbiosis. Analyses of nucleotide diversity segregating within each species suggests that strong selection, along with genetic hitchhiking has sharply reduced diversity along an entire chromosome half in S. meliloti. Despite the two species' ecological similarity, we did not find evidence for selection acting on the same genetic targets. In addition to providing insight into the evolutionary history of rhizobial, this study shows the feasibility and potential power of applying population genomic analyses to microbial species.
| Analyses of genome sequences can provide a nearly complete description of the nature and extent of nucleotide diversity segregating within and among species. There have been multiple investigations into genomic diversity in microbial communities using library-based and megtagenomic approaches [1] and phylogenomic studies of relatedness among microbial species [2]. By contrast, there have been few genome-wide surveys of nucleotide diversity within a prokaryotic species, and those studies have often focused on variation in genome content [3]–[5] rather than nucleotide diversity. Yet it is clear that population-genomic analyses provide an opportunity to greatly expand our understanding of the evolutionary forces shaping diversity within prokaryotic lineages [6]–[9] and identify targets of strong positive selection without bias that may be introduced when focusing on a limited number of genes or phenotypes of prior interest [10].
Prokaryotic species are often studied because they are either pathogens, of environmental or industrial importance, or because they form mutualistic associations with eukaryotes. The latter group includes members of the genera Rhizobium, Sinorhizobium (now Ensifer), Bradyrhizobium, Azorhizobium, and Mesorhizobium, collectively referred to as the rhizobia, a group of gram-negative bacteria that form symbiotic associations with legume plants. When growing in symbiosis with legumes, rhizobia convert atmospheric nitrogen (N2), which is unavailable to plants, into ammonia, which plants can use for the synthesis of amino acids. This symbiosis is estimated to contribute nearly half of all current biological nitrogen fixation [11] and is a key component of agricultural systems that are not dependent on synthetic fertilizers [12].
One of the best characterized rhizobial species is Sinorhizobium meliloti (now Ensifer meliloti). The interaction between S. meliloti and the closely related species S. medicae with the model legume M. truncatula, the genome of which was recently sequenced [13], has been the subject of extensive biochemical, molecular genetic [14]–[16], and evolutionary investigation [17]–[20]. The genomes of both S. meliloti and S. medicae consist of a single circular chromosome (∼3.65 Mb) plus two large symbiotic (sym) plasmids (∼1.3 and ∼1.6 Mb) [21], [22]. Sinorhizobium spp. also contain auxiliary plasmids, the number and identity of which varies widely among strains [23] and the functional importance of which is largely unknown. In Sinorhizobium, the genes required for forming nodules with legume hosts (including nod, exo, and nif genes) are distributed across both the chromosome and each of the two mega plasmids (hereafter referred to as plasmids) [21], [22], [24]. Bailly et al. [25] recently used low-coverage (∼0.8× average) genomic sequence data to characterize variation in gene content and nucleotide diversity on the chromosomes and two plasmids among 12 S. medicae strains. Their coverage was, however, too shallow to robustly characterize nucleotide variation along the genome or search for signatures of recent selection.
In this study we used Illumina technology to sequence the genomes of 12 S. medicae and 32 S. meliloti strains to over 100× mean depth. We aligned the Illumina data to the S. meliloti RM1021 and S. medicae WSM419 reference genomes (the chromosome and two plasmids) from each species and then used the aligned sequences to i) search for evidence of recent horizontal gene transfer between species, ii) characterize genome-wide nucleotide diversity within each species, and iii) identify genes that bear the signature of recent positive selection.
We aligned an average of ∼1,287 Mbp of sequence from each of 12 S. medicae and 32 S. melliloti strains resulting in median aligned coverage of >100 reads site−1 (Tables S1 and S2). For all six replicons (the chromosome and two plasmids of each species) the vast majority of sites were covered by either >50 or <2 reads (Figure S1). The regions with very low coverage are likely either present in the reference genome but not the resequenced strains, are <91% identical in the two strains, and thus too diverged to have aligned using our alignment parameters, or do not align to a single region in the reference genome. Because sequence reads were required to have a single alignment to the reference genome, reads that align to multiple locations were not included in final analyses. In S. medicae, an average of 95%, 79%, and 95% of the positions along the reference chromosome, pSMED02, and pSMED01 sequences, respectively, were covered by ≥10 uniquely aligned reads for each resequenced strain (Table S1). In S. meliloti, an average of 95%, 71%, and 93% of the positions along the reference chromosome, pSymA, and pSymB, respectively, were covered by ≥10 uniquely aligned reads for each resequenced strain (Table S2. Note, pSMED02 is orthologous to pSymA and pSMED01 is orthologous to pSymB). The high percentage of the reference sequence that can be aligned to the sequence data from the resequenced strains indicate that the vast majority of the sequence found in the reference genomes in each species is also found in all of our resequenced strains.
Sinorhizobium medicae and S. meliloti are closely related, have very similar host ranges, and at least partially overlapping geographic ranges [26], [27], characteristics that would provide considerable opportunity for horizontal gene transfer (HGT). Nevertheless, these are clearly distinct species; the chromosomes and plasmids from each species were reciprocally monophyletic (Figure 1) and the number of fixed differences between species greatly exceeded the number of shared polymorphisms (Table 1, Figure S3).
Although we found no evidence for interspecific transfer of whole plasmids, there are 97 genes (1 located on the chromosome, 21 on pSymB/Smed01, and 75 on pSymA/Smed02) with a ratio of shared polymorphisms to fixed differences >0.2, indicative of transferred alleles segregating within the recipient lineage (Table 1, Figure S3). Among these 97 genes (Table S3) are many with clear potential to alter the efficacy of nodulation or nitrogen metabolism including 11 fix, 13 nod, 8 nif, 2 noe, 2 nol, 5 rkp and 3 syr genes. By contrast, only 12 fix, 7 nod, 7 rkp, and no nif, noe, nol, or syr genes for which the data meet the coverage criteria had a ratio of shared polymorphisms to fixed differences <0.2.
To gain insight into the origin and fate of horizontally transferred genes we clustered the putatively transferred genes into contiguous genomic regions (horizontally transferred genes separated only by genes which did not have a putative ortholog in the reference genome of the other species or by ≤2 genes with ratios of shared polymorphism <0.2) then used neighbor joining trees to examine within and between species relationships. On pSymB/pSMED01, 20 of the 21 putatively transferred genes were found within a single 38 kb region. On pSymA/pSMED02, 6 of the putatively transferred genes are located within a 10.5 kb region of the S. medicae reference genome and 62 are located within an ∼300 kb region of the reference genomes. This 300 kb region also contains 236 genes that are present in the S. medicae genome (∼102 in S. meliloti) for which there was no identifiable ortholog in the reference genome of the other species (Table S3).
Neighbor joining trees of the large transferred regions (Figure 2), as well as other putatively transferred genes (Figure S4), suggest the history of HGT is complex. For all regions harboring genes with evidence of transfer, the majority of sequences from each species are monophyletic but the branch length separating sequences from the two species is much shorter than the length of the branch separating the two species at genes that show no signal of HGT (Figure 2). There are five regions, all of them on pSymA/pSMED02, for which the putatively transferred genes are not monophyletic (Figure S4); three for which a single S. medicae-like sequence was sampled from an S. meliloti strain, one for which a S. meliloti-like sequence was sampled from an S. medicae strain, and one for which the longest branch on the tree separates four sequences (three sampled from S. meliloti and one from S. medicae) from all other sequences. Interestingly, for this latter case, these four sequences were all sampled from Syria, suggesting geographic structuring of horizontal gene transfer and symbiotic gene alleles.
Consistent with previous multi-locus sequence [17], [20], [28] and genomic hybridization data [29] we found 1.5–2 times greater nucleotide diversity on each of the S. medicae plasmids than on the S. medicae chromosome (Table 2). For the S. medicae chromosome, Tajima's D (DT) was unimodal and centered near zero (Figure 3), a distribution consistent with a panmictic, neutrally evolving population (i.e. the standard neutral model [SNM]). Diversity within the full 32 strain sample of S. meliloti (Table 2) was two to three times greater than diversity within S. medicae but showed the same broad pattern of higher diversity on the plasmids than the chromosome (Table 2). The distribution of chromosomal DT values in the 32 strain S. meliloti sample was negatively centered (Figure 3) and the frequency spectrum of chromosomal polymorphisms revealed a mode of minor alleles found in four strains (Figure S5). This pattern is not consistent with a sample drawn from a single panmictic population and the chromosomal genealogy shows that the 32 strain sample was comprised of three distinct clades; one 24 strain clade and two 4-strain clades (Figure 1).
To remove confounding effects that population structure can have on nucleotide diversity, we recalculated diversity statistics using the sample of 24 S. meliloti strains that comprised the largest subpopulation found in our 32 strain sample (Figure 1). Unlike the 32 strain sample, for which the distribution of θW was unimodal, the distribution of θW values from the 24 strain sample was distinctly bimodal – with a considerable portion of genes having very low diversity (Figure 3). Similarly, the distribution of DT values from the 24 strain sample was multi-modal and both far more widely dispersed and more negative than the distribution in the 32 strain sample. This strongly skewed distribution appears to be largely due to genes located on the second half of the chromosome (bp 1,735,000–3,654,135); genes on this half have both low θW and low DT values (Figure 4). At both sides of this region, near the origin and terminus of replication there are sharp increases in the per-gene θW and DT values. Moreover, neither the first half of the chromosome (Figure S6) nor the plasmids show the well defined 24-strain clade seen in the whole-chromosome genealogy (Figure 1). Taken together, the lack of congruence between genealogies constructed from the two chromosome halves and plasmids as well as the sharp breaks in patterns of diversity provide evidence for transfer of large parts of the plasmids and chromosome among strains of S. meliloti.
To identify targets of recent adaptation we used the joint DTH statistic that provides a relatively powerful test of selection that is robust against demographic effects [30]. In S. medicae, 27 chromosomal, 9 pSMED01, and 4 pSMED02 genes were identified as putative targets of recent selection. Because the 32 strain sample of S. meliloti was strongly affected by population structure we searched for targets of selection in only the 24 strain sample. Moreover, because diversity in this sample was strongly structured along the length of the chromosome, we applied the DTH test to the first and second halves of the chromosome separately. These analyses identified 15 and 11 genes on the first and second halves of the chromosome, respectively, 11 pSymB, and 5 pSymA genes that harbored signatures of selection (Table S4). None of the genes identified as targets of selection were identified in both species, although fts genes, which are annotated as being involved in cell division and are down-regulated in bacteroids [31] were identified as targets of selection in both species (ftsW in S. medicae, ftsZ1 and ftsZ2 in S. meliloti). Consistent with the lack of between species overlap in the genes that harbor signatures of recent selection, between-species correlations in nucleotide diversity (θW, DT) were low for each of the three replicons (all R<0.26). Such low correlations are unexpected if selective constraints or among-gene variance in mutation rates are important determinants of nucleotide diversity and are similar in the two species.
The genes identified as putative targets of selection have a variety of annotated functions. Some of these functions are related to survival or reproduction either inside of nodules or in the soil environment, i.e. osmotic tolerance and stress (gst9, cysK2, guaB, hutH2, oxyR) and nutrient acquisition (phoU, thuR). Other putatively selected genes have functions that may be directly related to symbiosis, including hemA which is essential for symbiotic nitrogen fixation in many rhizobia, glgC and rkpJ which affect exopolysaccharide biosynthesis or export which is required that is essential for nodulation, as well as ftsW, ftsZ1, ftsZ2.
Rhizobia species are important symbionts of legume plants and this symbiosis is responsible for approximately half of all current biologically fixed nitrogen [11]. Because of this importance the biochemical and genetic basis of the symbiosis has been subject to extensive investigation [14]–[16]. To gain insight into genomic diversity segregating within rhizobial species, we sequenced to high coverage the genomes of 32 strains of S. meliloti and 12 strains of S. medicae, the two primary rhizobia symbionts of the model legume Medicago truncatula. Our analyses provide insight into the genome-scale extent of horizontal gene transfer (HGT), the structuring of nucleotide diversity within rhizobial genomes, and identify genes that have been subject to recent adaptive evolution in these species.
Previous analyses of genetic diversity within Sinorhizobium and other groups of rhizobia [20], [32]–[34] have found clear evidence that genes directly involved in nodule formation can be transferred between species; indeed, in prokaryotes genes involved in symbiosis are often found on mobile elements [35]. Consistent with these previous analyses, our whole-genome analyses revealed 97 genes, most of which are clustered on just three symplasmid regions, for which S. medicae and S. meliloti genes had a high ratio of shared to fixed polymorphisms, suggestive of recent horizontal transfer between these species. Genes with potential to alter nodulation or nitrogen fixation are overrepresented among the putatively transferred genes, suggesting that HGT may be important in the evolution of symbiosis. At the same time, the importance of HGT in shaping nucleotide diversity is largely restricted to the plasmids and appears to have very little effect on nucleotide variation in genomic regions outside of nodulation-gene islands.
Neighbor-joining trees constructed from plasmid genes show striking differences between genes that show signatures of HGT and those that do not. For genes that do not show evidence of transfer the branch separating sequences from different species is considerably longer than the branches separating sequences sampled from the same species. By contrast, genes that show evidence of transfer have comparatively short branches separating sequences sampled from the two species and relatively long branches separating sequences sampled from the same species. The short branch separating the sequences sampled from the two species suggests that transfer has occurred relatively recently, followed by the transferred sequence spreading through the recipient species. Interestingly, however, sequences from even these transferred genes are largely monophyletic. The sequence similarity of these transferred regions may facilitate ongoing transfer through homologous recombination [36] – thereby preserving these as islands of HGT. The single chromosomal gene with a high ratio of shared to fixed polymorphisms indicates that HGT of chromosomal genes is possible, even if HGT doesn't have an important effect on chromosomal sequence diversity.
The picture of diversity segregating in S. meliloti is highly dependent upon the composition of the sample. In our sample of 32 strains, the distributions of summary statistics (i.e. θW and DT) are largely consistent with a panmictic population. However, the frequency spectrum of polymorphic sites and genealogical relationships indicate that the 32 strain sample is composed of several distinct subpopulations with 24 strains forming a single well defined clade. The reasons for this substructure are not clear; the members of the 24 strain group were sampled from a wide geographic area (including France, Jordan, Syria, Tunisia), from the full spectrum of geographic locations that the 32 strains were sampled, and from multiple species of host plant (including M. truncatula, M. rigidula, and M. sativa, Table S2 and Figure S2). Regardless of the causes of the population structure, we found a close correspondence between the three major clades identified by whole-genome sequence data and relationships inferred from a 10 locus MLST characterization of population structure within S. meliloti [37], [38]. There were 17 strains in common between the two studies, 15 of these were part of the 24 strain group we identified and were monophyletic in the MLST analysis (13 were members of a single MLST group) and the two additional strains were each members of different MLST clusters. The similarity suggests that MLST data provide robust descriptions of population structure in this species.
Interestingly, patterns of diversity segregating among the 24 strain subpopulation are fundamentally different than those found in the 32 strain sample. Most strikingly, for the 24 strain sample, the two halves of the chromosome harbor distinctly different patterns of diversity, with one half of the chromosome having very low values of DT and θW relative to the other half. There are several possible causes for the two halves of the chromosome having such different patterns. One possibility is that recent HGT or balancing selection in the 24 strain sample has led to excess diversity and intermediate frequency variants on the high-diversity half of the chromosome. In fact, a neighbor-joining tree made with data from only the high-diversity half of the chromosome shows that some of the strains included in the 24 strain group cluster with strains that are not included in this group (Figure S6). The alternative possibility is that the low-diversity half of the chromosome has experienced a recent selective sweep, leading to reduced diversity and an excess of low-frequency variants. A recently introgressed region segregating at low frequency or the inclusion of a sequence from a divergent subpopulation could also explain the excess of low-frequency variants (low DT values) found on the second half of the chromosome. Either of these possibilities would cause elevated nucleotide diversity. By contrast, nucleotide diversity in this chromosomal half is reduced relative to the rest of the genome.
To determine if the data are more consistent with excess diversity on the first half of the chromosome or low diversity on the second half of the chromosome we compared nucleotide diversity on each chromosome half to diversity found segregating on the plasmids. For S. medicae the chromosome harbors 33 or 60% of the diversity segregating on pSMED02 and pSMED01, respectively. The ratio of chromosomal to plasmid diversity is similar for the S. meliloti chromosome from the 32 strain group and the first half of the chromosome from the 24 strain group, these samples harbor >30% and >50% of the diversity segregating on pSymA and pSymB, respectively. By contrast, the low-diversity, second half of the chromosome in the 24 strain group harbors 6% and 10% of the diversity segregating on pSymA and pSymB respectively. To the extent that plasmid diversity can be used as a reference, these data suggest that the distinctly different patterns of diversity found on the two chromosome halves of the 24-strain group may be due to a recent selective sweep that was strong enough to reduce diversity along the entire 1.8 Mb half of the chromosome through genetic hitchhiking [39].
If a selective sweep is responsible for the low diversity on the second half of the chromosome, the sharp borders near the origin and terminus of replication suggests that recombination at the borders is much higher than recombination along the second half of the chromosome or that selection favored the entire region, perhaps due to epistasis between genes located near the borders of the low-diversity region. If a selective sweep is the reason for the reduced diversity than it indicates that genetic hitchhiking along with selective variants could be an extremely important force shaping diversity within natural populations of prokaryotic species and may contribute to driving the divergence between prokaryotic lineages [40].
We identified 82 genes that bear a signature of recent adaptive evolution. These species have similar geographic ranges, life history, and share a common host plant, and as such they may be expected to experience similar selective forces. Nevertheless, the targets of selection in the two species show almost no overlap – no orthologous genes were identified as targets of selection in both species although fts genes, involved in cell division, are identified as targets in both species. The lack of overlap in the targets of selection suggest that these two ecologically similar species either experience very different selective forces or that selection acting similarly at the phenotypic level acts on very different genetic targets.
It is notable that no fix, nif, nod, nol, noe, or exo genes, which mutational screens identified as necessary for nodule establishment and nitrogen fixation [14]–[16], are among the genes we identified as bearing a signature of a recent selective sweep. However, nearly all of the nif, nod, and approximately one-half of the fix genes in the Sinorhizobium genome showed evidence of HGT or had no reciprocal best sequence match in the other species. Because these genes had no reciprocal best sequence matches we excluded them from our analyses of selection and therefore, their absence from the list of selected genes does not mean they have not been the subject of recent adaptation.
Population genetic analyses of nucleotide diversity segregating within Sinorhizobium medicae and S. meliloti have provided unprecedented insight into the evolutionary history of these ecologically important facultative symbionts. While previous analyses have detected evidence for horizontal gene transfer between these species, our data reveal that gene transfer is restricted almost exclusively to plasmid genes and that the plasmid regions that show evidence of transfer have less interspecific divergence than other genomic regions. Interestingly, nucleotide variation segregating within a 24-strain subpopulation of S. meliloti is highly structured along the chromosome, with one half of the chromosome harboring approximately one-fifth as much diversity as the other. The causes of the difference between the two chromosome halves may be a selective sweep coupled with extensive hitchhiking, if this is correct it would suggest that bouts of strong selection may be important in driving the divergence of bacterial species. Finally, we've identified genes that bear a signature of having evolved in response to recent positive selection. Functional characterization of these genes will provide insight into the selective forces that drive rhizobial adaptation.
We used Illumina sequencing technology to sequence the genomes of 32 strains of S. meliloti and 12 strains of S. medicae. These strains were chosen to capture diversity found within the USDA-ARS Rhizobium Germplasm Collection [38], as representative of different multi-locus genotypes [38], or because they had been recently sampled from natural populations and used in experiments to investigate interactions between Sinorhizobium and M. truncatula [41]. From each strain, DNA was extracted from culture grown cells using the Wizard Genomic DNA Purification kit (Promega Corp. Madison, WI, USA), with further purification by phenol extraction. DNA was then used to construct Illumina paired end libraries using Illumina's phusion-based library kits following the manufacturer's protocols. Insert sizes averaged 332 nt (range = 245 nt to 443 nt). Four samples were multiplexed per lane and sequenced on Illumina GAIIx machines following the manufacturer's protocols. Samples averaged just over 1 Gb of sequence (range = 724 Mb to 1584 Mb) translating into an average and minimum coverage of 174× and 108×, respectively, of the ∼6.7 Mb genome before aligning reads.
For SNP discovery, reads were aligned to the genome sequence of either S. meliloti Rm1021 [21], pSymA megaplasmid [42], pSymB megaplasmid [43] and the accessory plasmids pSmeSM11a [44], pSMeSM11b [45] and pRm1132f [46], or S. medicae WSM419 chromosome and plasmids pSMED01, pSMED02 and pSMED03 [22], using GSNAP [47] with a 91% minimum identity using the Alpheus pipeline [48]. For this work, only SNPs discovered in the alignment to the chromosome or the megaplasmids (pSymA, pSymB, pSMED01, pSMED02) were used due to poor coverage of the accessory plasmids. Nucleotide identity at a site was called only if that site was covered by ≥10 but <500 uniquely aligned reads (i.e. reads had maximum identity to only a single genomic location) and the nucleotide was supported by ≥70% of unique reads. Positions that did not meet these criteria were treated as ambiguous (N). Sequence reads are available at SRP009881, and SNP data are available for download at http://medicagohapmap.org/.
To evaluate the accuracy of base calls we PCR amplified and Sanger sequenced 25 loci from 4–6 strains (including 3 S. meliloti and 3 S. medicae) (Table S5). For the 42,953 bp of sequence for which we had both high-quality Sanger and Illumina data that met our coverage criteria there were 157 variants identified by both Sanger and Illumina, 3 variants identified in Sanger but not Illumina, and no variants identified in Illumina but not Sanger data.
Sequences of protein-coding genes were constructed using the IMG version of the WSM419 annotation downloaded on 1 December 2010 and the Rhizobase version of the Rm1021 annotation downloaded on 19 August 2010. For gene-based analyses requiring an outgroup, Rm1021 and WSM419 genes with ≥80% amino acid similarity across ≥80% of their length that were also bidirectional best hits were identified as homologous using the MaGe phyloprofile tool [49]. Gene sequences from the resequenced strains were aligned prior to analyses using the profile alignment tool in ClustalW [50].
We calculated nucleotide diversity for 10 kb non-overlapping windows and for each gene model for which we had data for >90% of the gene length for ≥80% of the strains. The number of replacement and synonymous sites for each gene within each species were estimated using the polydNdS program of the libsequence “analysis” software package [51]. Tajima's D (DT) [52], the average number of segregating sites (θW) [53] and the average pairwise number of segregating sites (θπ) [54] were all estimated using the compute program in libsequence. Fay and Wu's H (H) [55] was estimated using a custom libsequence-based program. Because DT is not defined for genes that have no polymorphism and the distribution of DT is highly skewed for genes with a single segregating site, we excluded genes with <2 segregating sites from the analysis. The number of fixed differences between S. medicae and S. meliloti were calculated on biallelic sites in alignments of orthologs using the sharedPoly program (in libsequence). Summary statistics for each of the annotated genes which met coverage criteria are in Dataset S1.
We used the joint DTH test [30] to look for genes that have experienced recent selective sweeps, considering genes in the lower 5% tail of the distribution for both DT and H as likely targets of selection. We restricted these tests of selection to genes with unambiguous nucleotide calls for >90% of the length from ≥80% of the strains and for which there was no evidence for horizontal gene transfer. For defining the 5% tails we took the ratio of genes that met the coverage and HGT requirement to the total number of genes. Genes with <2 SNPs or without a value for H were excluded.
We identified genes likely to have experienced recent horizontal gene transfer by comparing the ratio of polymorphisms that were shared between species to fixed differences between species. Based on the whole-genome distribution of this ratio (Figure S3) we identified putatively transferred genes as those with a ratio of shared polymorphisms to fixed differences >0.2.
To characterize the genealogical relationships among strains we constructed genealogies using the Neighbor joining algorithm [56] implemented in the dnadist and neighbor programs in Phylip [57] with the F84 model of DNA evolution [58]. Genealogies were constructed using concatenated gene sequences for the chromosome and each of the plasmids separately (2,741 chromosomal genes, 2,668,564 bp; 408 genes on pSymA/pSmed02, 416,009 bp and 1,049 pSymB/pSmed01 genes, 1,084,937 bp). Statistical support for clades in whole-replicon trees was evaluated using 200 bootstrap replicates. NJ trees for genes bearing a signature of horizontal gene transfer were constructed using similar methods, with statistical supported evaluated using 400 bootstrap replicates.
Several analyses were conducted separately for the first and second halves of the S. meliloti chromosome. In these cases, we used position 1,735,000 as the dividing line: this position seemed to correspond to the location of the sharp change in DT along the chromosome. This is also the location of a change in sign of the GC skew in the reference genome, indicating that the terminus of replication is located near this position [59] (Figure S7). GC skew was calculated using a custom R script [60] on all nucleotide positions in 10 kb sliding windows with a 5 kb step. The origin of replication for S. meliloti Rm1021 (the reference strain) has been experimentally determined to be near position 0 [61].
|
10.1371/journal.pntd.0006795 | Typhoid fever outbreak in the Democratic Republic of Congo: Case control and ecological study | During 2011 a large outbreak of typhoid fever affected an estimated 1430 people in Kikwit, Democratic Republic of Congo. The outbreak started in military camps in the city but then spread to the general population. This paper reports the results of an ecological analysis and a case-control study undertaken to examine water and other possible transmission pathways. Attack rates were determined for health areas and risk ratios were estimated with respect to spatial exposures. Approximately 15 months after the outbreak, demographic, environmental and exposure data were collected for 320 cases and 640 controls residing in the worst affected areas, using a structured interview questionnaire. Unadjusted and adjusted odds ratios were estimated. Complete data were available for 956 respondents. Residents of areas with water supplied via gravity on the mains network were at much greater risk of disease acquisition (risk ratio = 6.20, 95%CI 3.39–11.35) than residents of areas not supplied by this mains network. In the case control study, typhoid was found to be associated with ever using tap water from the municipal supply (OR = 4.29, 95% CI 2.20–8.38). Visible urine or faeces in the latrine was also associated with increased risk of typhoid and having chosen a water source because it is protected was negatively associated. Knowledge that washing hands can prevent typhoid fever, and stated habit of handwashing habits before cooking or after toileting was associated with increased risk of disease. However, observed associations between handwashing or plate-sharing with disease risk could very likely be due to recall bias. This outbreak of typhoid fever was strongly associated with drinking water from the municipal drinking water supply, based on the descriptive and analytic epidemiology and the finding of high levels of faecal contamination of drinking water. Future outbreaks of potentially waterborne disease need an integrated response that includes epidemiology and environmental microbiology during early stages of the outbreak.
| There was a large outbreak of typhoid fever in Kikwit, DRC, in late 2011. The outbreak started in military camps in the city but then spread to the general population. Multiple investigations were undertaken to understand how the disease spread. The worst affected areas of the city were mapped and compared to the water network. In early 2013, demographic and exposure data were collected for 320 cases and 640 controls residing in the worst affected areas, using a structured interview questionnaire to try to better understand individual risk factors. Residents of areas with water supplied via a gravity fed network were about six times more likely to have been ill with typhoid fever than residents of areas not supplied by the mains network. The most important individual risk factor was ever using tap water. Visible urine or faeces increased risk of getting typhoid but having chosen a water source because it is protected was linked to lower risk. Not handwashing and regularly sharing plates of food were also linked to less illness, but these findings may be especially subject to recall bias. The water network was also found to be heavily contaminated, including with faecal bacteria of human origin in multiple microbiological studies. Spatial, microbiological and case-control studies all implicate the water supplies in Kikwit to be unsafe and linked to spread of typhoid fever in 2011. Improvements to the mains water network in Kikwit are urgently needed to prevent future typhoid fever outbreaks.
| Typhoid fever (TF) is an infection caused by the bacterium Salmonella enterica serovar typhi (S. Typhi). The primary symptoms are fever and related malaise, but serious complications, such as intestinal haemorrhage or perforation (1–4% of all cases [1]), encephalitis, respiratory infections and metastatic abscesses can occur. In the absence of treatment, there is a case fatality rate of 10–30%, which drops to ~1% with timely treatment [2]. Data from 2010–2013 suggested that the TF disease burden in Africa was 4.3 million cases per year (95%CI 3.7–5.1 million) [3]. Data to estimate the total case fatality rate in sub-Saharan Africa are unreliable [1,3]. The mean case-fatality rate after S. Typhi caused by intestinal perforation has been reported at 19.5% (95%CI 16–22%) in African countries [4].
Typhoid is a strictly human infection and spreads from one person to another especially through faecal oral or urine oral pathways including consumption of contaminated food or water. Spread is exacerbated by poor sanitation and hygiene. Outbreaks have regularly been reported, but those occurring in low income countries are not well researched. Instead, most studies have described outbreaks in industrialized settings [2]. Consequently, opportunities to reduce transmission in these low-resource settings may not be identified or properly understood.
Repeated and sometimes severe outbreaks of typhoid have occurred in the Democratic Republic of Congo [5]. After a large outbreak of typhoid in the city of Kikwit, Bandundu Province, in the year 2006, typhoid became endemic, with low but persistent numbers of cases reported annually until another large outbreak in November 2011 to early January 2012. The 2011–2012 event resulted in 1430 identified cases. Seventy-one people developed peritonitis with perforation, and 17 people died in 2011–2012. The fatality rate was 1.5% [6]. An initial descriptive epidemiological study recognised that contaminated water supplies were likely responsible for most cases in 2011 [6], but did not elucidate on transmission pathways or other risk factors for infection in the subsequent phases of the outbreak. Patients in the 2006 and 2011 outbreaks appeared to come from the same areas of the city, which led to suspicions that an infrastructure problem or spatial feature might contribute to the risk of catching TF (S1 Protocol). It was observed that 2011 attack rates were highest in military camps within the city, especially early in the outbreak. The descriptive evidence suggested that the 2011 outbreak originated in the camps and subsequently spread to the general population. Here, we provide a quantitative investigation of exposure factors linked to infection in 2011, accompanied with recommendations both for prevention and during emergency response.
We undertook both an ecological analysis and a case-control study in order to better understand the primary transmission pathways of the epidemic and determine whether we could identify other risk factors. Based on the descriptive epidemiology and general knowledge of the epidemiology of typhoid, our primary hypotheses were 1) that the outbreak was waterborne and/or 2) that the infection spread directly from person to person.
Kikwit is the largest city in the Bandundu province of DRC. The estimated population in 2011 was 400,000 (Fig 1). Kikwit is located in the south-west of DRC and is an important commercial and administrative centre. In 2011, the city had one general referral hospital and was administratively divided into two health zones, north and south. Each health zone (“Zone de Santé”) was divided into health areas (“Aires de Santé”). There were 19 and 22 health areas respectively in north and south Kikwit at the time of the study. At the time of the outbreak, there were three military camps in the city, which accommodated about 2400 people: staff and families. Ngubu camp and Nsinga camp were in close proximity to the general population, while Ebeya camp was relatively separate from the city. The camp residents could be considered a highly mobile population and are mostly not of local origin. They had living conditions distinct from and mostly worse than the city’s settled residents. The military staff and families originate from many regions of DRC and speak many languages. Overall, living conditions in the three camps were poor and featured relatively high population density and poor hygiene and sanitation conditions. As a result, this study treated the three camps as three additional health areas which were analysed independently of the Aire de Santé where they were situated [6].
Kikwit is a hilly city with sandy soils and large erosion gullies. The city has a long wet season with an average of about 200mm of rain per month, from early September through end May. The city has centrally collected and distributed piped water, which is extracted from artesian wells, inconsistently chlorinated (S1 Report) and distributed via an aging pipe network to community tap points (standpipes, Fig 2). Water pressure is usually low throughout the network. Mains water is relatively expensive and difficult to access in parts of the city (S1 Report). Some homes have access to private wells. Surface water sources are widely available and generally free-of-charge. Surface sources are the river Kwilu, its tributaries, as well as protected and unprotected springs. No changes to the water infrastructure occurred in the time that elapsed between the 2011 outbreak and the survey dates.
Sanitation in the city is privately managed. Latrines are typically dug out by hand, and often are open air (no roof) and shared among multiple households. After a latrine fills up it is typically covered with thin soil, a mango tree is planted and a replacement latrine is dug nearby. Flooding of human waste out of active latrines is not unusual following significant rainfall events.
Study design and protocol for the spatial analysis is in S1 Protocol. During the 2011 typhoid outbreak, The Ministry of Public Health (MOPH) created a central register line list of all cases, both suspected and confirmed. The case definitions were set by the Ministry of Public Health (MOPH); “suspected case” was any person with fever ≥ 38°C for more than three days and digestive disorders, which were defined as diarrhoea, constipation or abdominal pain, as well as a negative malaria test. A “confirmed case” was a suspected case confirmed by isolation of S. Typhi from blood, bone marrow or duodenal fluid. The outbreak was confirmed by the MOPH based on the results of cultures of 50 blood and stool samples, which were tested in the University Hospital in Kinshasa. From the start of the outbreak, all TF cases had their data entered into a central electronic line list developed by the MOPH. The register included patient’s name, sex, age and address. By the end of the outbreak, the line register had 1430 cases, with illness onset dates from 19 Nov 2011 to 5 January 2012. The attack rate for each of the 41 health areas and three military camps was determined using the central register line list and estimated 2011 population data. Environmental elements were mapped [7] in early-mid 2013 for each health area in Kikwit using ArcGIS software (ESRI, California, USA). Spatial attributes were assigned to an entire health area as the predominate trait for that health area. A key environmental item of interest was the predominant water distribution network that public water taps connected to in each area. The network options were: northern pump, central pump, central gravity or off network. Other spatial attributes for each health area were population density the (average) distances from each health area to the nearest camp, market, school, springs, tarred road, water point or health care clinic. Distances to these locations were of interest because these places tend to be places where people cluster and thus might pass on infection. Data on socio-economic status, sanitation statistics or water quality (such as via the WHO Joint Monitoring Programme) were unavailable at the scale of health areas. Incident risk ratios were estimated for each health area using negative binomial regression, testing for any association between single or multiple exposures to environmental attributes in the health area and attack rate. The outcome was set to be the number of cases within each health area, the model offset was the (natural log transformation of the) total population in that same area. Distances were modelled continuously under a Poisson functional form. Categories were also available that described the water source in each health area (gravity or pumped, north or central, see data in S1 Spreadsheet).
Study design and protocol is item S2 Protocol. The case-control study focused on individual risk factors. The data were collected using structured household interviews between February and May 2013. Descriptive results of the data collected about cases during these interviews are described elsewhere [6]. Aspects of the structured survey as reported previously will be briefly recapped here.
The survey was targeted at cases and controls living in the Aires de Santé with the highest attack rates during the outbreak. The case-control study was done in only these areas primarily for logistical reasons. The attack rate for each health area was estimated using the denominator from population census data. For the case control study, we surveyed residents in the eight most affected health areas (attack rates > 0.36%) while also separating out and interviewing residents of the three military camps, because the camps all had AR > 4% (Fig 3).
To identify an odds ratio of at least 1.5 with a power of 80% for risk factors present in 30% of the control population, a sample size was set at 320 cases (25% of total), frequency matched by age and sex to 640 controls.
A structured questionnaire was developed and piloted (S1 Questionnaire). Twelve interviewers and one supervisor who spoke local languages were trained to verbally administer the questionnaire. Interviewers exercised own judgement in how to translate the questionnaire from French into other languages, when required. The interviewers were trained how to record household location and its principal water source using GPS devices (Garmin GPSMAP 76). Using the recorded addresses in the line list records, typhoid cases (from the concurrent case definition) were chosen at random, using random numbers generated in MS Excel. Cases were traced and then interviewed in the community. Serology was not used to confirm recovered case status.
Two controls of the same sex, age class of 5 years interval (0–5, 6–10, 11–15, 16–20, etc.) and health area were selected per case. Selection criterion for each control was not having been suspected of typhoid during the outbreak period. After interviewing a case, interviewers decided which end of their road to treat as the starting point, and then chose two control households by rolling a die to select the nth residences (alternating between left and right side of the street). Posited controls were asked if they had TF or were ill during the TF outbreak. Those who reported having TF or being ill were excluded as was their entire household (assumed to share same latrine and water supply). The die was rolled again if necessary to select a different household for a potential control. A maximum of one person was interviewed per household. All respondents were requested to show their household water storage containers, available soap and latrines. For children less than 13 years old, the interview was conducted with the guardian or a family member who was living in the same household and aware of the child’s condition.
During the household survey, level of awareness about the 2011–12 outbreak was very high and people claimed to be able to remember with accuracy if they had been ill during that period. Interviewers were welcome in people’s homes as representatives of MSF. No one refused to be interviewed.
All GPS readings were recorded and visualized using ArcGIS software (ESRI, California, USA). Conditional logistic regression (clogit) for risk factors were undertaken using STATA version 14.2, with data grouped by health area. Odds ratios (OR) for most risk factors were first estimated in single predictor models. Exposures and factors were excluded from univariate analysis if < 5% of responses were different from the most popular answer to a specific question. Any individual variable p-value <0.20 was carried forward into a multiple predictor case-control analysis. We endeavoured to keep all categories in the model if a variable had multiple levels. However, some variables were trialled by recoding them into binary variables where exploratory analysis found a strong association (such as having any tap water, as primary or secondary source or outside the home). Otherwise, risk factors were retained in iterative modelling as long as they had p-value < 0.05 to produce the final estimated adjusted odds ratios reported here. For purely categorical items, the reference value was set at the value with greatest frequency; for ordinal items, the reference value was set at the lowest rank answer. Using adjusted ORs and the fraction of cases receiving an exposure, the population attributable risk [8] percentage was determined for key predictors in the final model. Where few data were missing, those specific observations were excluded in the final model; where many data were missing, the variable was not used in multivariate analysis.
Study design and protocol for the water quality analysis is within S2 Protocol. Water samples of the principal source of drinking water of all interviewed cases were collected by two trained water and sanitation community workers, on 18 distinct dates from 13.3.2013 to 10.4.2013. Replicate tests were done on water samples onsite for Free Residual Chlorine (FRC) levels using the HANNA Photometer. Concentrations of ThermoTolerant Coliforms (TTC) were measured using a Delagua field kit. Tests for S. Typhi specifically were not undertaken–they seemed inappropriate so long (16 months) after the outbreak. A pathogen-specific test (such as PCR for S. Typhi) also exceeded our research budget and required equipment or laboratory facilities (for molecular biology) not available locally. In contrast, tests for TTC were useful to indicate likelihood of ongoing faecal contamination problems. Further details of the testing regime and results, including verification strategies, are in Ali et al [6]. The water quality data were used to calculate what proportion of water samples from each source could be considered high or low risk for transmission of human disease, using categories adapted from UNHCR guidelines [9].
Members of the research team changed. The procedure for selecting controls as described in the protocol was not used; a different set of procedures for selecting controls was devised, as described previously. The minimum age of independent respondents was changed from ten to 13 years old. Although specified in the protocol, the final study did not assess risk factors for disease severity (as indicated by peritonitis or intestinal perforation). No analysis of the 2006 outbreak was undertaken. Area-level education and income data were not suitable or available, so not used in the ecological analysis. Data collection dates were three months later than anticipated. The protocol also contains some factual errors because it was written prior to data collection, such as stating there were 33 Aires de Santé in Kikwit (actually there were 41, not 33), while the estimated population total was misstated to be 350,000; actual population turned out to be higher. The number of water quality tests per source was 1–4 (most often but not always 2). In the case-control study, we decided to focus on the eight most affected Aires de Santé, not the seven most affected areas as stated in the protocol. We don’t believe that any of these deviations or factual errors undermine our results or conclusions.
Ethics approval was received from the Ethical committee of the School of Public Health, University of Kinshasa (DRC) and Ministry of Higher Education, Academic and Scientific Research. Written informed consent was sought and obtained for all respondents or from their caretakers/guardians (children under 18).
None of the spatial attributes could be linked with statistical significance (p < 0.05 for risk ratio) to the 2011 TF attack rate at health area level, except for water source. Table 1 shows findings (see supporting data in S1 Spreadsheet), which reports risk ratios and attack rates for residents dependent on given water sources (p < 0.001). Residents who were dependent on the central gravity system were five times more at risk compared to residents on the northern (pumped) network (RR = 6.20 vs. 1.21), and about three times more at risk compared to those on the central pump system (RR = 6.20 vs. 2.25).
Data on occupational status of the head of household, demographic, sanitation and water quality traits identified for camp and city populations are described in greater detail in Ali et al [6]. Refer to the original survey questionnaire (S1 Questionnaire) and survey data, both raw and derived variables, (S2 Spreadsheet) for more details. Out of the 320 cases interviewed, 59 (18%) lived in the camps. Although the heads of households in the camps had more secure employment (75% in camps vs. 25% in town had work contracts), the city dwellers were more affluent, as indicated by greater access to electricity or a functioning TV. Sharing latrines with other families is normal practice in Kikwit for both military and civilian families. None of the observed latrines of the cases in camps and only 3% of cases in the general population had materials to facilitate wash hands (eg., soap and water) at a close distance (< 3 metres) from latrines. Upon request, 66% and 82% of cases in camps and general population showed the available soap in the household. This suggests that although respondents often said they washed their hands, many were in fact only rinsing their hands. Age and education profiles were similar for both military and civilian families, but households in the camps were more likely (77%) to live in a house of brick or concrete construction; most non-brick homes were made from mud. Controls were 2:1 frequency matched by age and sex to the recruited 320 cases.
The unadjusted odds ratios (OR) comparing cases and controls for individual risk factors are in Table 2. The OR in Table 2 used the factors as coded in the original survey (S1 Questionnaire and S2 Spreadsheet), although some survey elements were excluded for reasons described in the Methods, and for brevity, not all univariate results are listed in the table. Age and sex associations with case status are shown to be insignificant in Table 2, which indicates that frequency matching was successfully implemented. Seventeen possible predictive factors had odds ratios with p-values ≥ 0.20 in single variate analysis. Twenty factors were taken forward to be tried in multivariate estimations of OR (because they had p < 0.20). Some hygiene, cooking customs, and indicators of socio-economic status were among the risk factors that qualified for trial in multivariate OR estimations. At the single variate stage, intake of any tap water (OR 3.41, 95%CI 1.88–6.19), whether tap water was a primary or secondary source (OR 2.80, 95%CI 1.64–4.79), knowledge to wash hands (OR 2.36, 95%CI 1.45–3.86), assertions that they know how to avoid typhoid (OR 0.44, 95%CI 0.31–0.61), and statement of habitual washing of hands before cooking (OR 5.12, 95%CI 3.11–8.44) had the strongest association with increased disease. Those who said that they regularly shared their plates of food had reduced risk. All indicators suggestive of better handwashing behaviour (more frequent handwashing or knowledge that handwashing should reduce disease transmission), were positively associated with typhoid case status (see data in Table 2). Aspects of the home environment (topography and home construction materials) as well as habits of eating uncooked food were also significant enough to be trialled in multivariate modelling. The number of water storage containers in the household or claiming to have soap in the home did not reach the threshold to be tried in multivariate analysis.
In all cases (OR data in Table 2), those who stated that they always wash hands after defecation or before cooking and before infant care had significantly increased risk of disease. Those who stated never in response to these questions, had strongly decreased risk. Knowledge about handwashing was similarly correlated; those who said they knew they should wash hands had much increased risk. Explanations for this unexpected finding are explored in the Discussion.
To put multiple variables about washing hands behaviour or beliefs into the same model could create collinearity problems. Moreover, the information about washing hands before cooking or after childcare is incomplete because this question was only asked to female heads of household, and hence there were missing data for 253 respondents. Similarly, answers were missing for 264 respondents (56 cases and 208 controls), on whether they mentioned washing hands when asked about ways to avoid catching TF. However, there were no missing data about washing hands after defecation for any respondent. To minimise collinearity and for ease of interpretation, only the variable about handwashing habits after defecation was used to indicate handwashing knowledge or behaviour, when generating the final model.
Table 3 shows our final predictive model with all final significant predictors with adjusted odds ratios. Complete data were available for 320 cases and 636 controls. This model adjusts for age and sex for completeness, but their coefficients are not shown because their distribution was artificially imposed by the control recruitment method and therefore cannot be interpreted as risk indicators. Regularly sharing food was also linked to less illness (adj. OR 0.07, 95%CI 0.03–0.14). Contaminated mains water (adj. OR 4.25, 95%CI 2.18–8.28) was likely to be an important route for typhoid transmission in this population, either via direct ingestion or additional exposure (hand washing habits). The population attributable risk percentage (PAR%) for tap water consumption was estimated at 69.6%. Choosing a water source for perceived protected status seemed to confer reduced risk (adj. OR 0.68, 95%CI 0.48–0.95), while the indicator of visible urine/faeces in the respondent’s primary latrine area conferred increased risk of disease acquisition (adj. OR 1.43, 95%CI 1.05–1.95; PAR% = 17.3%). Other PAR values are reported in Table 3, although not for exposures that reduced risk–the PAR was not developed for that purpose.
According to the survey responses, the majority (90%) of cases in the general (not camp) population used taps at communal distribution points as their principal source of drinking water. Water sources for camp residents were more diverse. The most common sources of drinking water for cases in each camp were (Table 4): an artesian well for Ngubu camp (36%), taps at communal distribution points for Nsinga camp (34%), and an unprotected source for Ebeya camp (30%). Overall, 34% of all camp cases used communal taps as their principal source of drinking water.
Almost all the sources of tested principal drinking water were contaminated with faecal coliforms to a very high degree (see S3 Spreadsheet for original data). Free residual chlorine levels measured at the public water taps were insufficient (<0.2 mg.l-1) to zero [6]. Fig 4 indicates the main types of water source tested and the proportion of each type of each source that fell into risk categories to human health. There were 102 unique sources identified by interviewees. Protected springs were most likely to be low risk. Only one of the water taps conformed to published standards. Most respondents (892/960, 92.9%) who were asked about possible treatment methods did not report that they treated their water by chlorination, boiling or filtration (or another pathogen inactivation method); therefore, we did not include water treatment factors when estimating odds ratios and exposures.
Multiple lines of investigation tied together here establish a strong association of mains water with the spread of the outbreak from the camps to the rest of the population, conforming with the multi-level evidence recommendations for surveillance of waterborne infectious diseases made by Tillett et al. [10]. The ecological study strongly suggested that the outbreak was linked to particular parts of the city water supply and showed that attack rates for city residents were highest in the areas with a gravity-fed mains water distribution. The case-control study confirmed that using mains water was most strongly and reliably associated with the risk of disease acquisition. Descriptive epidemiological analysis found that attack rates peaked earliest and were highest overall in the military camps [6]. The outbreak appears to have started in the camps, likely due to more naïve population and poorer living conditions, where a diversity of water sources were used, as shown in this report. Microbiological analysis has repeatedly shown drinking water sources throughout Kikwit to be mostly unsafe, due to high faecal coliform counts and inadequate chlorination [6,11]. Ingress of faecal material during the outbreak period, due to low water pressure, was very plausible to expect in many parts of the network concurrently observed to be in poor repair (S1 Report). In 2011, the outdated mains water supply system in Kikwit probably played an important role in disseminating typhoid from early cases in camps to the general population.
This case-control study also adds to the descriptive epidemiology report, the information that statements by respondents about regular handwashing were linked to increased risk of disease. Handwashing with soap after toileting or prior to food preparation and after infant care, should be best practice in all settings including where endemic diseases are present, but this may not be true when the main water sources available for washing are themselves contaminated and where soap may not be available. The case-control survey did not ask questions about whether respondents used treated water for handwashing, but 91% who were asked about treatment methods, did not mention any method for treating household water. Moreover, it was observed [5] that most households lacked any handwashing soap, and only three respondents had handwashing materials close to their main latrine facility. It is very possible that many respondents said that they washed hands, when in fact they merely rinsed their hands. Self-report is prone to another bias; socially approved behaviours are usually self-reported more frequently than observed [12–14]. Bias in favour of approved behaviours could explain our apparent finding that regularly sharing food was protective. Moreover, the interview asked about current practice not those at the time of the outbreak; it is certainly plausible that people who have had typhoid would be more rigorous with handwashing practices after the outbreak, so the question could not elicit accurate information about exposure at the time of outbreak. Inefficacy in handwashing technique is an unexplored risk factor. In this resource scarce setting, it seems likely that hand-drying materials are also limited; inadequate hand drying can leave pathogens on the hands, too [14–16]. That so many respondents said they had soap but then could not display used soap when asked, supports the suggestion of bias for answers about handwashing behaviours. Consequently, given the potential biases in how the hand-washing questions were answered, we conclude that the association between reported handwashing and risk of typhoid acquisition cannot be taken as indicative of a real risk of disease in people who do (properly) wash their hands.
As for occupation of head of household, labourer with a regular work contract had the lowest quantifiable association with disease. The next lowest risk group was casual labourers, adj. OR 2.12, 95%CI 1.41–3.17. The finding may be confounded because military men fell into this occupational group with regular contracts. No further information was available about the working environment for individuals with regular contracts. Possible explanations are that this category (about one third of respondents) tended to indicate households with more financial security due to regular work, or where the head of household, because of long-term regular dirty work, had prior exposure and thus acquired immunity to multiple Salmonella species and serovars [17–20].
This study was undertaken 13–18 months after the end of the 2011 outbreak. Assuming that responses to questions asked in early 2013 can truly describe behavioural practices in late 2011-early 2012 outbreak may be suspect. The questionnaire did not ask about individual hygiene behaviour and practices during the outbreak to avoid other types of recall bias. Local staff translated the questionnaire from French to other languages as required during interviews; we did not monitor this process and it may have led to inconsistencies in how questions were asked or answered; in the Kikwit area, French and the Kituba language predominate but there are many regional languages and dialects spoken plus interviewees could have come from anywhere in the DRC, which has over 200 recognised languages. We did not use serology to confirm that controls were negative or to confirm cases. This means likely misclassification of some controls, which will have biased the odds ratios downwards in Table 3; this means our evidence for implicating water and sanitation in the spread of TF is understated. Ecological analysis was limited to only one type of geography (health areas) and only in parts of the city, and only some spatial variables (ones we could get data for). Water quality could have changed between 2011 and 2013. We measured ThermoTolerant Coliforms about 16 months after the outbreak to gauge ongoing contamination of city water supplies, rather than PCR amplification that specifically looked for S. Typhi during the actual outbreak weeks. Heavy rainfall can cause latrine overflows in Kikwit and could affect local supplies, changing preferred water sources; however, the outbreak, survey and water quality testing all took place in wet months with very similar levels of monthly rainfall (November-May period). We assumed that general state of sanitation facilities (soap, latrines) did not change since the outbreak; however, we do have considerable anecdotal information that this assumption is valid. Challenges in tracing cases were encountered due to the 13 months elapsed time since creation of the line list and survey start. Some civilian cases may have been misidentified as camp residents, due to proximity of the camps to the general population, and vice versa.
Our recommendations address both prevention and emergency responses, and also draw on observations and suggestions made by water sanitation engineers who visited Kikwit in December 2011 (S1 Report). Our key recommendation to prevent or minimise future outbreaks in Kikwit of typhoid and similar diseases, is improvement to the water network. Descriptive, spatial and case-control studies all identify the water network as instrumental in transmission of typhoid in 2011. Surveys of water supplies in Kikwit in both July and November 2015 also found widespread unacceptable faecal contamination in all drinking water sources tested; 97% of the isolated bacteria in surface waters had human origin [21]. This finding was strongly linked to outbreaks of waterborne diseases thought to affect up to 30% of the city’s population annually. A full revamp of the city’s water system would clearly be very desirable. Work is arguably most urgent in those areas fed by gravity supply, which are in the central area that also had the highest attack rates. Improvements to the mains water network could include but should not be limited to: repairs to prevent inundation (including replacing pipes and reversing soil erosion), consistent chlorination of tap water, regular monitoring of the chlorination levels, rehabilitation of unprotected springs, and closing latrines located uphill and in relative proximity to frequented water sources or water mains pipes. Hygienic harvesting of rainwater in public places could be implemented to make it easier to properly wash hands (S1 Report). It would also be desirable to improve the overall sanitation and hygiene situation in Kikwit, especially within the places that were hotspots for TF transmission in 2011 (military camps) [6]. Rehabilitating latrines, provision of ongoing resources to make handwashing safer and more effective, to enable handwashing with soap and consistent household water treatment, could be beneficial.
Recommendations as part of an immediate actionable response to an outbreak should include: creation of minimum perimeters from latrines to water sources and rapid drainage of runoff water around standpipes and hoses (S1 Report). Rapid testing of water sources and rapid ascertainment of exposure risks during an outbreak would quickly facilitate understanding how such disease was spreading. It is undesirable that the exposure data in this study were collected as late as 14 months after the outbreak. Distribution of handwashing materials with health campaigns to promote full washing, for users of all water sources, would be desirable. Emergency distribution of chlorine, either in tablets or via buckets, with usage instructions, to ensure more water treatment could be protective, although work needs to be done to make the taste of chlorinated water more acceptable to the local populace (S1 Report). Distribution of hygiene kits may well be appropriate, especially to high risk groups [16]. Vaccine-based strategies for typhoid control are recommended for school-age children in endemic countries–in this context, a targeted vaccination in the camps might be effective emergency response or short-term prevention measure [22,23].
Following high early transmission in military camps near the city of Kikwit, use of contaminated mains water was consistently and reliably, strongly associated with typhoid fever acquisition. A safer mains water network is the most valuable change that could prevent future disease. Effective measures to better protect water supplies, include but are not limited to: relocation of intake points, more consistent chlorination, preventing inundation to the distribution network, and more convenient access to treated water. Safe sources for the purposes of cooking and hand cleaning could reduce the size of TF and similar disease outbreaks in future.
|
10.1371/journal.pcbi.1005161 | Co-operation between Polymerases and Nucleotide Synthetases in the RNA World | It is believed that life passed through an RNA World stage in which replication was sustained by catalytic RNAs (ribozymes). The two most obvious types of ribozymes are a polymerase, which uses a neighbouring strand as a template to make a complementary sequence to the template, and a nucleotide synthetase, which synthesizes monomers for use by the polymerase. When a chemical source of monomers is available, the polymerase can survive on its own. When the chemical supply of monomers is too low, nucleotide production by the synthetase is essential and the two ribozymes can only survive when they are together. Here we consider a computational model to investigate conditions under which coexistence and cooperation of these two types of ribozymes is possible. The model considers six types of strands: the two functional sequences, the complementary strands to these sequences (which are required as templates), and non-functional mutants of the two sequences (which act as parasites). Strands are distributed on a two-dimensional lattice. Polymerases replicate strands on neighbouring sites and synthetases produce monomers that diffuse in the local neighbourhood. We show that coexistence of unlinked polymerases and synthetases is possible in this spatial model under conditions in which neither sequence could survive alone; hence, there is a selective force for increasing complexity. Coexistence is dependent on the relative lengths of the two functional strands, the strand diffusion rate, the monomer diffusion rate, and the rate of deleterious mutations. The sensitivity of this two-ribozyme system suggests that evolution of a system of many types of ribozymes would be difficult in a purely spatial model with unlinked genes. We therefore speculate that linkage of genes onto mini-chromosomes and encapsulation of strands in protocells would have been important fairly early in the history of life as a means of enabling more complex systems to evolve.
| Trans-acting polymerases are cooperative, because they copy neighbouring strands, and do not copy themselves directly. Inaccurate replication creates parasitic strands that act as templates but not ribozymes. It is known that in spatially distributed models with slow strand diffusion, clusters of cooperating polymerases arise that can survive in the presence of parasites provided that the error rate is less than a maximum limit (the error threshold). In the RNA World, we envisage multiple types of ribozymes working together. We would like to understand how a multi-ribozyme system could evolve from a system with a single type of polymerase ribozyme. As a first step in increasing complexity, we consider a two-ribozyme system in which there is one polymerase and one nucleotide synthetase that produces monomers for use by the polymerase. We are particularly interested to find conditions in which the chemical supply of monomers is too low for the polymerase to survive alone, but the additional monomers created by the synthetase allow the two-ribozyme system to survive where the single-ribozyme system could not. There is then a selective force for increasing the complexity of the system. Here we show that spatial clustering is sufficient to allow cooperation and survival of systems of unlinked ribozymes with different functions. Clusters form in which synthetases form fringes around the polymerases. Survival of the two-ribozyme system depends on several factors. The strand diffusion rate must be slow enough for cooperative clusters to emerge. The replication rate of the polymerase must be comparable to that of the synthetase. The diffusion rate of the monomers must be neither too slow nor too fast. The model considers the most difficult case for cooperation–unlinked genes with no compartments. The sensitivity of the two-ribozyme system that we study here suggests that evolution of a spatial system with multiple unlinked ribozymes would become increasingly more difficult as the number of components increased, and suggests that linkage and protocells would need to evolve relatively early in the history of life.
| The RNA world hypothesis proposes that in the early stages of life on Earth, RNA sequences acted both as genes and as catalysts [1–3]. The key molecule in the RNA World would be an RNA polymerase ribozyme that used another RNA strand as a template and synthesized the complementary strand to the template. A polymerase that could rapidly and accurately replicate a template of its own length would be able to sustain life in the RNA World. As each polymerase copies a neighbouring strand rather than copying itself, sustained replication of this type requires cooperation between a group of polymerases that copy one another. The behaviour of cooperative polymerases is fairly well understood from theoretical models, as we will discuss below. The aim of this paper is to understand how a replicating system based on cooperating polymerases could evolve additional functions.
Any kind of additional ribozyme with a function that supports the polymerase could potentially add to the system with a single polymerase. The most obvious second type of function to consider is a nucleotide synthetase that catalyzes synthesis of monomers used by the polymerase. The polymerase copies the synthetase sequences as well as other polymerase sequences. A polymerase can only survive alone if the monomer concentration produced by abiotic chemistry is sufficiently high. If a synthetase is also present, the monomer concentration will be increased, and the two ribozymes can potentially survive together in conditions where neither of them could survive alone. Here, we investigate under what conditions these two ribozymes can survive by mutual cooperation.
Experimental work has demonstrated that specific ribozoymes in the laboratory have many of the features that would be required to support an RNA World. Polymerase ribozymes have been gradually developed in the lab by in vitro evolution [4–9] and the maximum lengths of the templates that can be replicated are now around 200 nucleotides, which is close to the length of the catalyst. Sustained autocatalytic replication of RNAs has been demonstrated in the lab with ligases [10] and recombinases [11]. However, these systems require continued input of relatively long strands as substrates, rather than the single nucleotide substrates required by the polymerases.
In order for ribozymes to have arisen, the environment of the early Earth must have supported the synthesis of RNA strands by abiotic chemistry. Experiments that show RNA polymerization have been carried out using clay catalysts [12–14] and using wetting and drying cycles in the presence of lipid bilayers or ammonium chloride salt [15–18]. Wet-dry cycling seems to promote polymerization because polymer bond formation becomes thermodynamically favourable in the dry phase but continued polymer growth is limited by restricted diffusion. The wet phase permits repositioning of molecules and allows further polymerization to occur [19]. Partially-ordered structures of nucleotides sandwiched between lipid lamellae have been observed by X-ray scattering [20,21], which suggests that the lipid bilayers help to organize the nucleotides in a configuration that is favourable for polymerization.
Another key step necessary before the RNA World would be sequence replication via non-enzymatic template directed synthesis. This has also been observed in the lab to some extent [22–24]. Taken together, the experimental studies of ribozymes and prebiotic chemistry make it seem plausible that an RNA world could have existed on the early Earth. Here we investigate some of the problems that would be faced by RNA replicators from the standpoint of evolutionary theory.
Computational models [25–32] have been used to study the way a replicating catalytic sequence could emerge from the mixture of random sequences that would be created by prebiotic chemistry. In two-dimensional spatial models [27,28], an RNA polymerase can arise by a rare event and spread deterministically across the surface, even in the presence of other non functional RNA strands that act as parasites. When diffusion of the RNA strands is slow, clusters of polymerases form that cooperate with one another [3], whereas parasites would destroy the system if it were well mixed. The fact that spatial clustering promotes the survival of cooperating replicators is well known from a number of different types of evolutionary models [33–44].
Cooperation between ribozymes of different functions, including polymerases, nucleotide synthetases, and lipid synthetases, has been investigated in a series of papers by Ma and co-workers [45–48]. These papers use a fairly realistic but complex model with a large number of different rate parameters, whereas here we want to keep the model with as few parameters as possible, in order to facilitate a theoretical understanding of the effects of changing parameters. These papers also assume that the reactions of the RNA strands occur within protocells. However, if a replicating system can be initiated without cells, this seems simpler to begin with. It is possible to envisage a complex metabolism for RNA-based life controlled by many different types of ribozymes with different functions, all of which are copied by the same RNA polymerase. We would like to know how complex a system could become in absence of cells. It has also been suggested that ribozymes were packaged in inorganic compartments [49] prior to the origin of cells. There would have been slow diffusion and spatial clustering in such a system in the same way as for a two-dimensional surface model.
Our previous work [28] has considered the origin and spread of a polymerase in a mixture of random sequences that act as parasitic templates. In the scenario that we study here, we presume that the polymerase system already exists and we ask whether a nucleotide synthetase can add to this system. The synthetase potentially benefits the polymerase by creating additional monomers and increasing the replication rate, but it also places the burden of its own replication onto the polymerase. Thus the synthetase has the potential to be both a parasite and a cooperator. It is fairly easy to show that cooperation of these two types of ribozyme is not possible in a well-mixed model with no spatial structure. Firstly, a polymerase system alone is overrun by parasitic templates in the well-mixed case. Secondly, even if there are no parasites, then two functional ribozymes can only co-exist in the well-mixed case if they have exactly the same replication rate, which will not be true in general. In this paper, we show that in a spatial model with slow strand diffusion, stable coexistence of the two ribozymes is possible, even though they have different lengths and different replication rates. The cooperating system is stable even when inaccurate replication creates non-functional mutants of both the polymerase and the synthetase, provided the error rate is less than a finite error threshold.
The model operates on a two-dimensional square lattice. Each site may either be empty (state 0) or occupied by a single RNA strand (states 1–6), as summarized in Table 1. A polymerase (state 1) is able to use a neighbouring strand as a template to produce a strand complementary to the template. A nucleotide synthetase (state 4) is able to produce monomers that can be used by the polymerase. The complementary strands to these ribozymes (states 2 and 5) are not functional as catalysts, but they are required as templates to synthesize the ribozymes. In addition, states 3 and 6 represent mutant sequences of the polymerase and synthetase that are non-functional. The lengths of the two ribozymes are Lpol and Lsyn, and the lengths of the complementary strands and the mutants are the same as the corresponding ribozymes. All mutant strands of a given length are equivalent in this model; therefore, it is not necessary to distinguish between positive and negative strands in the mutants.
In each time step, of length δt, each lattice site is visited in a random order. If there is a strand on this site, it has an opportunity to be a template for replication. A template can only replicate if it has a neighbour that is a polymerase and another vacant neighbouring site into which the new strand is placed. The rate of replication is inversely proportional to the length of the template strand, as explained in the Methods section. Replication of the ribozymes is by alternate plus/minus strand copying—a ribozyme creates a complement and a complement creates a ribozyme. At each replication, there is a probability of a deleterious mutation that creates a non-functional mutant instead of the complementary sequence. The mutation probabilities Mpol and Msyn will in general be different for the two ribozymes because the strands can have different lengths, and the fraction of point mutations that are deleterious may depend on the structure and function of a ribozyme. Reverse mutation from a mutant sequence to a functional sequence is assumed to be negligible. Further details of the replication procedure are described in the Methods section.
Each site also has a monomer concentration. It is assumed that monomers are continually produced and destroyed by abiotic chemistry, so that an equilibrium concentration is reached. When synthetases are present, there is a local increase in the monomer concentration (see Methods). The replication rate is proportional to the monomer concentration on the template site. Monomers diffuse between lattice sites at a rate proportional to the diffusion parameter D. Strands can also diffuse by hopping into a neighbouring vacant site at rate h. Strands are treated individually and strand diffusion is a stochastic event for each strand, whereas monomer concentrations are continuous variables and monomer diffusion is treated deterministically (see Methods). Strands also break down to monomers at a rate u.
In order to understand the behavior of this model when all six types of strands are present, we will build up to this case from simpler situations. Firstly, we consider the polymerase with its complement and mutant sequences in absence of the synthetase. Secondly, we consider the two ribozymes and their complements in absence of mutation. Thirdly, we consider all six types of strand.
We first consider a situation with only polymerases and their complementary strands and mutants present (states 1–3). The only source of monomers is from abiotic chemistry, and we suppose that the monomer concentration A is fixed and constant everywhere in the lattice. Initially we set the hopping rate h to zero. This means that strands remain in the place where they are created. Clusters of strands nevertheless move across the lattice due to creation and destruction of strands. Typical equilibrium configurations are shown in Fig 1 for three different values of the mutation rate Mpol. At low mutation rates, the polymerase and its complement occupy the surface fairly uniformly. As Mpol increases, the distribution becomes increasingly more patchy. The non-functional mutants are parasites of the polymerase. Spatial clustering of the polymerases and complementary strands allows them to survive even at relatively high mutation rates, whereas in a well-mixed case, where clustering cannot occur, the system is overrun by parasites for any non-zero mutation rate.
It should be noted that the polymerization rate constant, kpol, depends on the per-nucleotide polymerization rate constant vpol, the monomer concentration A, and the length of the strand Lpol, as described in the Methods section. For the simulations in Fig 1, we fixed Lpol = 100 nucleotides, A = 500 nucleotides per site, and vpol = 3, which results in kpol = 15. However, since A is fixed and all the strands have the same length, it is only the combined value of kpol that affects the simulation, rather than the three parameters separately.
In Fig 2, time averaged strand concentrations are measured as a function of mutation rate with fixed polymerization rate. The concentrations of polymerase and complement decrease steadily with increasing Mpol, while the mutant concentration passes through a maximum. At larger Mpol all three strand concentrations go to zero at the error threshold—the maximum mutation probability that is sustainable.
Another key parameter that affects the coexistence of the polymerase with the mutant strands is the strand hopping rate h. A small value of h allows the polymerases to form clusters where they replicate neighbouring strands. However as h increases, the clusters of polymerase start to disappear. There is increased mixing of polymerase and mutants, and mutants overrun the system until they eventually kill the entire population. This is shown in Fig 3.
The results in this section are very similar to those obtained in our previous paper [28] using a slightly different model. The previous paper used local rules that we called "Two's Company, Three's a Crowd". In that case, up to three strands were allowed on one lattice site. When a polymerase was on a site with one other strand, replication was possible, which produced a third strand on the same site. When there were three strands on a site, no further replication was permitted until one of the strands diffused away. In the model of the present paper, only one strand is permitted per site, and the template strand is on the neighbouring site to the polymerase, instead of on the same site. These two models are intended to represent the same situation. The well-mixed limits of these two models are the same (see the differential equations in the Methods section and in [28]). The results of the simulations of the spatial models are qualitatively similar: there is an error threshold at a finite value of Mpol, and there is a maximum value of h above which mutants mix with polymerases and kill the system.
Having understood the case with polymerases alone, we want to build towards the case of cooperation of polymerases and synthetases. As a next step, we will consider the case with polymerases and their complements (states 1 and 2), together with synthetases and their complements (states 4 and 5), but without mutant sequences, i.e. we will set both mutation probabilities Mpol and Msyn to zero.
When synthetases are present in the model it is necessary to specify the treatment of the monomers in more detail. We suppose that monomers are created chemically from precursors at a constant rate a per site and that they break down at a constant rate proportional to the current concentration on the site. Additionally each synthetase produces monomers at a rate b. If monomer diffusion is fast, the monomer concentration will be equal on all sites. The equilibrium monomer concentration is A = a + bX4, where X4 is the fraction of lattice sites that are synthetases (see Methods). The simplest case is where b = 0, and the synthetases are just non-functional parasites. We call these parasites 'independent' to distinguish them from parasites produced by mutation of the polymerase (state 3). If it can be shown that the polymerase can coexist with synthetases when they are just parasites (b = 0), then it seems likely that coexistence will also be possible when synthetases are functional (b > 0). Therefore we start with the b = 0 case.
Fig 4 shows an example of coexistence of polymerases and independent parasites. We fix A = a = 500, vpol = 5, and Lpol = 100, so that kpol = 25, and consider varying lengths of Lsyn. The polymerization rate of a synthetase strand is ksyn = vpolA/Lsyn = kpolLpol/Lsyn. Fig 4 shows a simulation with Lsyn = 50, which means that ksyn = 2kpol. The parasites coexist with the polymerases despite the fact that they replicate twice as fast. The polymerases form clusters where strands are replicating their neighbours. The parasites can only survive on the fringes of these clusters as they need to be next to a polymerase in order to be replicated. The disadvantage to the parasites arising from spatial clustering counters the advantage they get from being shorter, faster replicators. A second possible outcome is that, if the parasite length is too long, the parasites multiply too slowly and they die out. The third possibility is where the parasites are too short. In this case, they multiply rapidly and kill the polymerases, after which they also die because they cannot replicate alone. There is thus a range of intermediate lengths of Lsyn for which coexistence of the independent parasites with the polymerases is possible.
Fig 5 shows the boundaries of this region of coexistence. The width of the coexistence region is largest when the hopping rate is h = 0. As h increases, both the upper and lower length limits for coexistence increase and move towards the length of the polymerase, Lpol = 100. The limit of large h corresponds to the well-mixed case. Coexistence in the well mixed case is only possible if Lsyn = Lpol, as can be shown from the differential equations in the Methods section. As this will not usually be the case, we expect that coexistence of two different types of strands will not be possible in a well-mixed model. In the spatial model, however, there is a wide range of coexistence, especially when the hopping rate is small.
If we now consider the case of functional synthetases (b > 0), we see that it is essentially the same as the case with b = 0. If we determine the concentration of synthetases, X4, in a simulation with b = 0 and a = a0, then we can choose any other combination of a and b such that a + bX4 = a0, and the resulting equilibrium state should be the same. In particular, if we set bX4 = a0 and a = 0, this will also be the same as the original case where a = a0 and b = 0. This is significant because it means that the combination of the two ribozymes can survive together where neither can survive on its own. When a = 0, the only monomers are coming from the synthetase, so the polymerase cannot survive alone, and the synthetase cannot survive alone because it cannot replicate.
At this point we have shown that the spatial pattern that arises in Fig 4, where the synthetases surround the fringes of the polymerase clusters, allows both sequences to survive in parameter ranges where neither could survive on its own. We might be tempted to conclude that the problem of cooperation between ribozymes of different functions is solved. This conclusion would be premature, however, because we have only considered the case with zero mutation rate. It turns out that the model considered so far with fixed monomer concentration is not stable to mutations in the synthetase. For any non-zero rate of mutation in the synthetases, the functional synthetases are gradually replaced by mutant synthetases, so there is no further production of monomers by the synthetase. What happens then depends on the value of a. If a is large enough, the polymerases can survive by themselves, and the mutant synthetases are just independent parasites, as in the b = 0 case. If a is small, then the synthetases cannot survive alone, so the whole system dies out.
In this model, mutations in the synthetase behave in a different way to mutations in the polymerase. If Msyn = 0, then the cooperating system survives with non-zero Mpol up to a finite error rate. Spatial clustering prevents the invasion of polymerase mutants (as in Figs 1 and 2), but does not prevent the invasion of synthetase mutants. The problem is that if monomer concentration is equal everywhere, the mutant synthetases multiply just as fast as the functional ones, and deleterious mutations produce more and more mutants which eventually destroy the system. In order to get a stable coexistence between the polymerases and synthetases in the presence of non-zero mutation rates in both sequences, it is necessary to include spatial variation in the monomer concentration, as we do in the following section.
We now consider the case where each synthetase produces monomers at rate b on its own site. There is a finite diffusion constant D for monomers, as described in Methods. In this case the system is stable to mutations in both polymerase and synthetase. In Fig 6 we show a case where a = 0 and monomers are only produced by the synthetase. Three different mutation rates are shown, with Mpol = Msyn in each case. Increasing mutation rate leads to increasing patchiness of the structure. Eventually an error threshold is reached. The concentrations of the strands are shown as a function of the mutation rate in Fig 7.
The monomer diffusion constant D is important in these simulations. If D is too small then the monomers accumulate on the sites of the synthetases and do not reach the sites occupied by the polymerases. This means that the replication rate of the polymerases becomes too low and the system dies out. If D is too large, monomers spread equally across the whole lattice, and this favours the multiplication of parasitic non-functional synthetases, as discussed in the previous section.
To demonstrate that it is the mutant synthetases that kill the system at high D and not the mutant polymerases, we considered simulations where mutations occurred in either the polymerases or the synthetases but not both. In Fig 8, Mpol is fixed at 0.05, and Msyn = 0. If D is small, the system dies out because monomers do not reach the polymerases. At high D the system is stable and the concentrations tend to the values they have when the monomer concentration is equal everywhere. Fig 9 differs in that Msyn = 0.05 and Mpol = 0. In this case, the system dies at low D because monomers do not reach the polymerases, as before. However, at high D the system is again unstable because mutant synthetases out-compete the functional synthetases.
The rate of replication of any template strand is proportional to the monomer concentration on the site occupied by the strand. Monomers are synthesized on the sites occupied by the synthetases and diffuse outwards. The average concentration of monomers on synthetase sites is higher than on other types of sites. This gives functional synthetases an advantage over mutant synthetases, even though deleterious mutation is driving the increase of the mutant synthetases. This explains why the model with the full treatment of monomer diffusion is stable to mutations in the synthetase as long as D is not too large.
The idea of an error threshold, i.e. a maximum rate of deleterious mutations that can be sustained by a replicating system, is important in molecular evolution. The classical treatment of the error threshold problem [50] assumes that each strand has the ability to replicate via a 'one-molecule-makes-two' process. This is used as a model of virus replication [51,52]. It is assumed that the viral RNA is being replicated by a protein polymerase that is not itself evolving. The theory considers a 'master sequence' that replicates faster than the mutants. The concentration of the master sequence decreases as a function of the mutation rate and goes to zero at the error threshold—see, for example, Fig 4 of [50], which is similar to Figs 2 and 7 in this paper. The error thresholds in this paper occur for a different reason, however. Here, we are discussing a 'two-molecules-make-three' process, where the catalytic strand is part of the evolving system. The polymerase in our models is only partly analogous to the master sequence in the standard error threshold theory. Both theories assume that the functional sequence undergoes deleterious mutations to create non-functional sequences and that back-mutations from non-functional to functional sequences are negligible with respect to deleterious mutations. The neglect of the back mutations is reasonable unless the sequences are extremely short. It is this that leads to a sharp transition at the error threshold. However, the polymerase in our model differs from the master sequence in the classical theory in that it does not have an intrinsically faster replication rate. We have considered the case where the polymerase replicates all templates at the same per-nucleotide rate. For the polymerase to survive, it has to encounter other polymerases faster than it encounters mutant sequences. This is why spatial clustering is an essential part of the models discussed here but not a part of the classical error threshold theory.
Cooperative trans-acting polymerases have been studied in several kinds of models previously [27–28,34–35,42–44]. When there is a single kind of replicating molecule, is clear that spatial clustering is effective as a means of preventing the invasion of parasites. This paper extends the question to consider cooperation between independently replicating molecules with different functions. The conclusions of this paper are positive, in the sense that we have clearly shown that spatial clustering is also sufficient to allow coexistence and cooperation between two different ribozymes with complementary functions. However, we also think that these results are significant in highlighting some of the difficulties that will occur in evolving increased complexity of replicating systems of this kind. We would like to know how to get from a single-component polymerase system to an organism with hundreds of genes with different functions. We have treated a synthetase as a single ribozyme. However, synthesis of a relatively complicated monomer like a nucleotide presumably involves many steps, and four different nucleotides are required for RNA. It is possible to imagine a system involving multiple ribozymes with different functions that contribute to RNA synthesis. Only one of these sequences needs to be a polymerase, because all the other sequences can be replicated by the same polymerase. The results of the model studied here with just two functions lead us to a better appreciation of the difficulties associated with adding multiple functions. Spatial clustering is a simple mechanism that is sufficient to achieve a certain degree of cooperation; however, we doubt that the multiple-ribozyme system envisaged here could be achievable by this means alone. We have considered the most difficult case for cooperation—independent sequences with no compartmentalization. Organisms today go beyond this by linking genes on chromosomes and keeping their molecules together inside cells. The results in this paper suggest to us that it may have been necessary to evolve chromosomes and protocells rather early.
Firstly, in absence of compartments, spatial clustering arises only if we have limited strand diffusion. We have considered a 2d model with strands bound to a surface. Other systems with restricted diffusion might also work (e.g. small pores in rocks, or spaces between stacked lipid lamellae), but unrestricted diffusion in 3d will not. Given that clustering of polymerases is required for their own survival, it is necessary for other ribozymes with secondary functions to stay closely associated with the polymerase clusters. This gives rise to the interesting patterns of co-clusters seen in Fig 6. We have not yet attempted to develop a spatial model with more than two types of functional sequence, but we suspect that it would be difficult to achieve viable co-clustering patterns if there were more than a handful of essential types of genes that all need to be next to the polymerase in order to be replicated.
We focused on the case where a = 0 in Figs 6–9 because in that case it is clear that both polymerase and synthetase are essential and there is a mutually beneficial cooperation between the two. If a is sufficiently large, then the polymerase can survive by itself. A synthestase would then be an 'optional extra'. A new kind of functional gene must arise within a system that is already stable, i.e. a new gene must always be optional at the time and place where it first appears. If there were a multi-ribozyme system in which some of the genes were optional in this way, then it would not be necessary to have every type of gene present in every small neighbourhood. This would significantly reduce the constraints that would be necessary on the co-clustering patterns, and might make a diverse system with many gene types distributed across the surface easier to evolve. However, our experience with the current two-ribozyme model is that optional genes tend to do more harm than good. For example if a is sufficient to support the polymerase, and we add a functional synthetase (with b > 0), then the concentration of the polymerase tends to go down because of the extra load imposed by the synthetase. The polymerase would be better off on its own, and the synthetase is behaving like a parasite even though it has a beneficial function. Adding optional genes that are at least partially parasitic does not seem like a good general route towards evolving complex genetic systems.
Secondly, in absence of compartments, small molecule diffusion becomes an issue. This is illustrated in the present model by considering monomer diffusion explicitly. The model shows that the two-ribozyme system only survives if D is neither too large nor too small. In a multi-ribozyme system there will be many small molecules involved in the metabolism required for nucleotide synthesis. All of these will have to have appropriate diffusion rates. Compartments, such as lipid membranes in protocells, would solve the problem of keeping useful small molecules together close to the ribozymes that synthesized them. This would reduce the problem of invasion of mutant synthetases that we discussed above. Putting different kinds of sequences together in protocells also helps to favour cooperation, as is shown by stochastic protocell models [53–55]. A recent protocell model shows that large numbers of different replicator types can be maintained in protocells by the stochastic-corrector mechanism [56]. However, including cells also introduces complications associated with cell division, segregation of the genetic material between daughter cells, and transport of molecules across membranes. The origin of cells is a key early step in the evolution of life, but it is not yet clear just how early this step is. It is quite difficult to compare models of spatial clustering and protocells because they tend to be formulated in different ways, although see [43] for a good attempt at doing this.
It is possible that the first replicating sequences were already enclosed inside protocells almost by default. The recent experiments showing that RNA polymerization is facilitated by wetting and drying cycles in the presence of lipid bilayers [15–21] suggest that the first replicating polymers may have formed in the presence of lipids. This has led to the 'coupled phases' model for the origin of life [57] in which there is an alternation of replicating polymers from a dry phase between lipid lamellae to a wet phase encapsulated in protocells. If the lipids are there in large supply from the beginning, then the first replicators do not need to synthesize them, and if the membranes are continually breaking and reforming during the wetting and drying cycles, the replicators do not need to control the division of protocells or the transport of small molecules across the membranes. It would therefore be interesting to get a better understanding of whether a physical (non-living) process of encapsulation and dispersal can also create an environment in which cooperation of functional ribozymes can evolve.
Another feature of some surface-based models is that travelling waves of replicators and parasites can arise, such as the spiral waves seen in hypercycle models [33] and the more irregular travelling wave patterns seen in the replicase-parasite (RP) system described in section 5.3.1 of reference [44]. If short, rapidly multiplying parasites are added to our model with polymerases alone, we observe irregular travelling waves similar to the RP system [44]. A reviewer has suggested that cooperation of nucleotide synthetases and polymerases would be difficult in this model in the presence of short parasites because the synthetase may not be able to join the travelling waves of polymerases. This remains to be investigated in detail, but our preliminary results suggest that there are some cases where the joint system is stable in the presence of short parasites.
The evolution of linkage between genes of different functions is also a key step that we presume evolved rather early in life. If genes are linked, then the problem of different genes replicating at different rates (illustrated in Fig 5 in this paper) is solved, because they are forced to replicate together. Also, the problem of maintaining the spatial association of the two functions is solved because they are physically linked. The downside of linkage is that a longer strand with two genes would replicate more slowly than the shorter strands with a single gene. It was shown [58] that, in a protocell model, cells containing linked genes can out-compete cells containing separate genes because their progeny have a higher chance of inheriting a full gene complement. Selection on the cell can overcome selection for fast replication of individual genes. If there were just two ribozymes linked on a single RNA strand, it is possible that they would form two structural domains that individually fold to their functional structure at the same time, but this is difficult to envisage for more than a few domains. If there were many functional sequences linked on a longer chromosome, then it seems likely that some kind of separation of function between linked sequences (chromosomes) and individual sequences (catalysts) would have to arise. Thus the evolution of linkage would also create additional problems associated with control of transcription of individual genes and distinguishing transcription from chromosome replication.
In this paper we made the assumption that a replicating RNA system required an RNA polymerase ribozyme to catalyze sequence replication; hence we assumed that the first ribozyme was a polymerase and we considered the addition of secondary ribozymes to the polymerase system. This has been the traditional view of the RNA World community, and it has motivated the search for polymerase ribozymes by in vitro evolution. However, it is worth mentioning that sequence replication could in principle occur by non-enzymatic means [22–24], and if the rate of non-enzymatic replication were faster than the hydrolysis of the templates, a polymerase catalyst would not be necessary. This would open the way to considering alternative scenarios in which some other kind of ribozyme, like a nucleotide synthetase, came first [32,45,59].
A series of papers [38–41] using the metabolic replicator model (MRM) has studied the evolution of replicating ribozyme systems on mineral surfaces, and addresses questions that are very similar to those we have considered here. The MRM model is different from ours in important respects. In the MRM, there are several ribozymes of different types that produce small molecules that contribute to the replication process. The replication rate of a strand depends on the concentration of the different metabolic ribozymes within a local region called the metabolic neighbourhood. At least one ribozyme of each type must be present in the metabolic neighbourhood in order for strand replication to occur. The metabolic ribozymes in the MRM are similar to the synthetase in our models; however, we consider diffusion of monomers explicitly, whereas the MRM simply assumes that the small molecule products of the ribozymes extend uniformly to all sites within the metabolic neighbourhood. For the majority of the results in [38–41] it is assumed that replication is non-enzymatic (possibly because it is catalyzed by the mineral surface), and that there is no specific replicase ribozyme. A few results are given in [38] and [41] for a case where a replicase ribozyme (equivalent to the polymerase in our models) is included in addition to the metabolic ribozymes. They show that the replicase can promote replication beyond what is already given by non-enzymatic means; hence the replicase can be retained as a useful addition to the system. This scenario is different from the one we have studied, in which we assume that strands cannot replicate by non-enzymatic means. The polymerase is therefore essential in our model as a first ribozyme.
Studies with the MRM have shown that fairly large sets of metabolic replicators can be supported in surface-based replication systems [40–41] in which replication is non-enzymatic. As we have argued above, we suspect that building up larger sets of ribozymes will be more difficult in our model where the replication is controlled by the polymerase. Local contacts between neighbours are more crucial in our model than the MRM for several reasons. Nearest-neighbour contact is required between the polymerase and the strand being copied in our model. Nearest neighbour contacts are not required for the metabolic ribozymes in the MRM because their effect extends over the metabolic neighbourhood. We also assumed that a new strand can only be created if there is a vacancy on a neighbouring site to the template. The MRM model allows a new strand to be placed anywhere within a replication neighbourhood that can be broader than just nearest neighbour sites. Larger replication neighbourhoods allow coexistence of larger sets of replicators [40–41]. However, a large replication neighbourhood seems rather artificial. It would seem more reasonable to create the strand on the neighbouring site and then allow diffusion of the strand along the surface. Strand diffusion has different effects in the two types of models. In the MRM, strand diffusion is beneficial because it creates a more even mixture of ribozyme types, and this increases the rate of replication. Strand diffusion is detrimental in our model because it mixes the parasites with the polymerases and leads to loss of the polymerases (as in Fig 3 above). By incorporating the diffusion of monomers explicitly in our model we uncouple the motion of the monomers from the strand diffusion.
The above discussion emphasizes that determining the rate of non-enzymatic replication is a key issue for our understanding of the RNA world. Some degree of non-enzymatic replication is required to get the replication process started, because there must be at least one plus and one minus strand for a polymerase to begin with. The non-enzymatic rate need not be sufficient to sustain continued replication if the polymerase ribozyme is the first to emerge. The emergence of a polymerase from a random mixture is considered in our previous paper [28]. Once a polymerase system is established, catalytic replication is much faster than non-enzymatic replication, and the system survives even if the non-enzymatic rate is set to zero (as it is in the current paper). In the metabolic replicator scenario, in which metabolic replicators emerge before (or instead of) a polymerase, the rate of non-enzymatic replication must be sufficient to sustain replication at all times. A key issue is whether experimental conditions exist in which non-enzymatic replication is sufficiently fast and accurate to sustain replication of strands that are long enough to be metabolic ribozymes.
It is also worth noting that, although we have presented this paper in the context of the RNA world, the models are sufficiently general to apply to replication of any kind of nucleic acid analogue polymer system in which strands act as templates for synthesis of complementary strands.
In summary, there are several different ways in which co-operation at the molecular level is essential to the way the RNA world would have functioned [3]. In this paper we have focused on the issue of cooperation between unlinked ribozymes with different functions. The mechanism of spatial clustering, which arises when strand diffusion is very slow, such as when ribozymes are bound to a surface, is known to promote cooperation when there is a single type of polymerase ribozyme. Here we have shown that the same mechanism works to promote cooperation between a polymerase and a nucleotide synthetase. The model highlights several difficulties that must be overcome. The replication of the two kinds of sequences must occur at comparable rates. Co-clusters must arise in which the two types of sequences are closely positioned spatially. Diffusion of the monomers produced by the synthetase must be sufficiently fast to benefit the polymerases and sufficiently slow to prevent invasion of parasitic sequences. All these conditions can be satisfied within the two-ribozyme system that we studied. However, these difficulties are likely to be increasingly difficult to overcome as the number of independently replicating ribozyme components increases. Therefore we conclude that additional factors such as linked genes and protocells were probably necessary relatively early in the evolution of replicating systems.
The model is simulated in discrete time steps of size δt. Strand replication, breakdown, and movement are treated as stochastic events that occur with a probability equal to the rate of that event multiplied by δt. For example, the time scale of the model is set relative to the strand breakdown rate u = 1; hence the probability of a strand breaking down to monomers is uδt per time step. A step size of δt = 0.002 was used in these simulations. In each time step, events occurred in the following order: replication and mutation; strand breakdown; monomer production and diffusion; strand hopping.
Replication and mutation—For each potential template strand, two different neighbouring sites are chosen randomly from the eight possible neighbours. Only if the first neighbour is a polymerase and the second neighbour is a vacancy, the template is replicated with a probability kδt, where k is the replication rate of the template strand, which is either kpol or ksyn (see Table 1). If replication occurs, the new strand is placed in the vacant neighbouring site. The replication rates for the two types of sequences are
kpol=vpolA/Lpol,ksyn=vpolA/Lsyn,
(1)
where A is the monomer concentration, and vpol is a constant derived below by considering a Michaelis-Menten reaction scheme. It is assumed that vpol is a property of the polymerase and is the same for all templates. Under the approximations discussed below in the section on Michaelis-Menten kinetics, the replication rates are inversely proportional to the lengths of the templates.
In some of the simulations in this paper we assume that A is a constant that is equal on all lattice sites. The mean A is controlled by a balance of monomer production and breakdown back to precursors, as well as incorporation of monomers into strands and breakdown of strands back to monomers. In the fixed-A simulations, it is supposed that all these processes are in equilibrium and that diffusion of monomers is fast enough so that the same A concentration applies across the whole lattice.
In other simulations we treat A is a variable that is different on each lattice site. In the variable-A case, the replication rates (Eq 1) of each template depend on the A of the template lattice site. When a new sequence is created, the value of A on the template site is reduced by the number of monomers in the strand being produced (either Lpol or Lsyn).
When replication occurs, there is a mutation probability M that is either Mpol or Msyn, depending on which strand is the template (Table 1). With a probability 1-M the new strand is the correct complementary sequence for the template. With a probability M, the new sequence is a mutant of the appropriate type. Reverse mutation from a mutant sequence to the corresponding functional sequence is assumed to be negligible. Secondary mutations in mutant sequences can be ignored because all mutant sequences are equivalent.
Strand breakdown—Once every strand has been given a chance to be a template, it has a chance to become degraded with probability uδt. This site then becomes a vacancy. The degradation rate u for all types of strands is assumed to be equal and is set to 1. In reality, degradation would involve multiple steps via fragments of shorter lengths. A longer strand would have more positions at which breakdown could start, but it would also take more steps to return to single monomers. The degradation rate should also be dependent on the structure of the strand. These are all complications that we have chosen to ignore in order to keep the model simple enough to be tractable. In the variable-A simulations, when a strand breaks down, a number of monomers equal to the length of the strand breaking down is added to the site previously occupied by the strand.
Monomer production, breakdown and diffusion—In the variable-A simulations, the concentration of monomers on each site is adjusted at each time step by an amount
δA=δt(a+bS−A),
(2)
where a is the rate of production of monomers from precursors by abiotic chemistry, and b is the rate of production of monomers by synthetases. The variable S is 1 if there is a synthetase on the site, and 0 otherwise. The -A term is the rate of breakdown of monomers back to precursors. In the absence of replicating strands, the mean concentration is A = a. If the concentration of synthetases is X4 and the monomers are distributed uniformly across the lattice, the mean concentration is A = a + bX4.
When monomer diffusion occurs at a finite rate, A will vary between sites. Monomer diffusion occurs deterministically. At each time step a number of monomers ADδt leaves each lattice site and is distributed equally between the eight neighbouring sites. Each site also gains monomers from each of its eight neighbours. The net change in A on one site due to diffusion is
δA=Dδt(−A+Aneighbours),
(3)
where Aneighbours is the mean concentration on the eight neighbours. In the fixed-A simulations, A is constant and the steps of monomer production, breakdown and diffusion are not necessary.
Strand diffusion—In each time step, each strand attempts to hop to one of its eight neighbouring sites with probability hδt. If the randomly chosen neighbour site is a vacancy, the strand moves to this site. It the neighbouring site is occupied, the strand stays where it is. We refer to strand diffusion as "hopping" in order to distinguish it from monomer diffusion.
Implementation—To get a visual understanding of the dynamical behaviour of the model, simulations were run using Netlogo [60]. The images shown in Figs 1, 4 and 6 use a lattice size of 150 × 150. The same model was also simulated in C in order to obtain numerical averages of quantities with a larger lattice size of 500 × 500. In both cases, periodic boundaries were used (i.e. the edges of the lattice are connected in a torus).
Well-mixed limit—If the hopping rate h is large, there is no correlation between the states of neighbouring lattice sites. This is the well-mixed limit in which we can write down deterministic differential equations for the fractions of lattice sites, Xi, in each of the states i = 0–6 (Table 1). We will give these equations in order to make it clear how the spatial lattice model is related to the well-mixed case. It should be noted, however, that co-operation of the two types of ribozyme is not possible in the well-mixed case, for several reasons that we will explain below. Here we consider the fixed-A case only so that kpol and ksyn are constants in the equations below.
It should be clear from this that the replication of type i depends on the product the template concentration Xi, the polymerase concentration X1, and the vacancy concentration X0. In the lattice model, this corresponds to the assumption that the template must have a polymerase on a neighbouring site and a vacancy on a different neighbouring site. It should also be clear that mutations occurring on replication of type 1 and 2 strands, produce type 3 strands, while mutations in types 4 and 5, produce type 6 strands.
We will briefly summarize the key properties of these well-mixed equations. Firstly, if there is no synthetase present and polymerase replication is perfectly accurate (Mpol = 0), there is a stable solution for the polymerase with positive X1 and X2, provided the polymerization rate kpol is sufficiently large. However, this state is unstable to mutation. For any non-zero value of Mpol, the polymerase is overrun by its own mutations, and the system dies out. Secondly, if both ribozymes are present and replication is perfectly accurate (Mpol = 0 and Msyn = 0), then it is only possible for the two to coexist if kpol = ksyn. As this will never be exactly true for two ribozymes with different lengths and different structures, we conclude that it is not possible for two different types of ribozymes to cooperate in a well-mixed environment. In the lattice model with small h, spatial clustering arises that allows coexistence of the two ribozymes, as we show in the Results section.
Michaelis-Menten kinetics for replication—The synthesis of a complementary strand by a polymerase involves binding of the polymerase to the template and stepwise addition of monomers. For simplicity, we treat the synthesis of a new strand as a single step. Here, we calculate the effective rate of this single step by using the Michaelis-Menten enzyme kinetics model [61]. See also reference [42] for use of this scheme with replication dynamics. The reaction scheme is
T+X1→kf←krC→kcatT+T′+X1,
(11)
where X1 is the polymerase, T is the template (any of the types of strand), T’ is the complement to the template, and C is the complex of the polymerase and template. The rates kf and kr represent the binding and dissociation of the polymerase to the template, as in the standard Michaelis-Menten scheme. The rate of the catalytic step kcat depends on the time taken for the individual monomer additions. If v1 is the rate constant for addition of one monomer, and the monomer concentration is A, then the mean time for a single addition is 1/v1A, and the mean time for synthesis of a strand of length L is L/v1A. We therefore approximate the catalytic step as a single step with rate kcat = v1A/L.
Now, following the usual Michaelis-Menten method, we assume the complex is in equilibrium with the unbound polymerase. The total polymerase concentration is X1tot=X1+C, and the complex concentration can be written as
C=kfTX1totkr+kcat+kfT.
(12)
The net rate formation of the complement T’ is
kcatC=kcatkfTX1totkr+kcat+kfT≈kcatkfkrTX1=v1kfAkrLTX1.
(13)
Above, we made the assumption that kr >> kcat + kfT, in which case X1≈X1tot. Under these assumptions, we see that the net rate of replication from a template of length L is k(L)TX1, where
k(L)=vpolAL,
(14)
and vpol = v1kf / kr. In this paper, we considered strands of two lengths, Lpol and Lsyn, which gives the two replication rate constants in Eq (1) above.
|
10.1371/journal.pntd.0006242 | Mycobacterium tuberculosis complex genotypes circulating in Nigeria based on spoligotyping obtained from Ziehl-Neelsen stained slides extracted DNA | Tuberculosis (TB) remains one of the most threatening diseases and Nigeria has one of the world’s largest burdens. This study performed high-throughput spoligotyping directly on sputum smears to describe the Mycobacterium tuberculosis strains circulating among new TB patients in Nigeria.
All State TB control programmes in Nigeria were requested to submit 25–50 smear-positive Ziehl-Neelsen (ZN) stained slides for screening during 2013–2014. DNA was extracted from 929 slides for spoligotyping and drug-resistance analysis using microbead-based flow-cytometry suspension arrays.
Spoligotyping results were obtained for 549 (59.1%) of 929 samples. Lineage 4 Cameroon sublineage (L4.6.2) represented half of the patterns, Mycobacterium africanum (L5 and L6) represented one fifth of the patterns, and all other lineages, including other L4 sublineages, represented one third of the patterns. Sublineage L4.6.2 was mostly identified in the north of the country whereas L5 was mostly observed in the south and L6 was scattered. The spatial distribution of genotypes had genetic geographic gradients. We did not obtain results enabling the detection of drug-resistance mutations.
We present the first national snapshot of the M. tuberculosis spoligotypes circulating in Nigeria based on ZN slides. Spoligotyping data can be obtained in a rapid and high-throughput manner with DNA extracted from ZN-stained slides, which may potentially improve our understanding of the genetic epidemiology of TB.
| Using a classical genotyping method designated as “spoligotyping”, which targets polymorphic repetitive DNA loci, we present a general snapshot of the genetic diversity of Mycobacterium tuberculosis complex causing tuberculosis in Nigeria. Our results were obtained on a collection of 549 DNAs, extracted from Ziehl-Neelsen stained sputum smears gathered and representative from 36 Nigerian states and during 2013–2014. We show the ubiquitous presence on the Nigerian territory of a sublineage of Lineage 4, designated as L4.6.2 or “Cameroon” Lineage, which represents almost 50% of all patterns, more prevalent where Mycobacterium africanum west African 1 (Lineage 5) is absent. We also show that Lineage 5 is geographically linked to the south-east of the country, where it is the most prevalent, and represents approximately 20% of all the samples. The last third is linked to all other L4 sublineages and to L6 (M. africanum west African 2). Our results confirm the strong phylogeographical structure of Mycobacterium tuberculosis complex in Nigeria and are suggestive of a long-term coevolution history between Homo sapiens sapiens and Mycobacterium tuberculosis complex.
| In 2015, six countries, Nigeria, India, Indonesia, China, Pakistan and South-Africa, accounted for 60% of all TB cases in the world, with Nigeria having the second largest TB burden in Africa, with an estimated 586,000 cases in 2015 [1]. Nigeria is also among the ten countries with the largest gaps between notifications of new and relapse (incident) TB cases and the best estimates of TB incidence in 2016 [2]. Despite the large number of cases, there is a paucity of information of the genetic diversity of Mycobacterium tuberculosis complex (MTC) circulating in the country and in west Africa [3–7]. In a previous study, we reported the genetic diversity of MTC in three Nigerian cities (Abuja, Ibadan and Nnewi) using a CRISPR-based fingerprinting method (spoligotyping) based on DNA extracted from clinical isolates [5, 8, 9]. Lineages 4.6.2 (Cameroon family) and Lineages 5 and 6 (Mycobacterium africanum west African 1 and 2) were the main lineages found [5]. The study however reported significant variations between the cities and it was not representative of whole Nigeria. Although it would be desirable to conduct genetic diversity studies on MTC cultures, with a wider geographical representation of patients, this would require considerable resources and would be logistically complex. Hence, in a large country like Nigeria, obtaining a representative set of cultures from all inhabited regions is difficult to achieve.
It has been reported that it is possible to conduct spoligotyping analysis directly from Ziehl-Neelsen (ZN) stained slides, scratching the material from the slides and extracting the DNA [10]. However the method has a relatively poor sensitivity and has not gained wide acceptance as a direct molecular identification technique [10].
Despite this limitation, direct spoligotyping from smears would be logistically simpler and facilitate studies with wider geographical representation, bypassing the need for culture, and allowing the direct identification of M. africanum (L5 and L6), which has specific metabolic requirements and is difficult to grow [11, 12]. In one recent study we demonstrated that the apparent disappearance of M. africanum from Burkina Faso was a sampling artefact [13, 14] and other teams in Cameroon and Benin are currently facing the dilemma on whether M. africanum is truly disappearing or not, and a study in Ghana has shown that M. africanum still represents 20% of all TB cases [15]. Indeed, an alternative explanation would be the progressive replacement of M. africanum by more modern (Lineage 4) isolates [14].
The aim of this study was to conduct a nation-wide description of the diversity of tuberculosis genotypes based on sputum smear extracts, to assess the global geographical genetic structure of MTC in Nigeria. In addition, we tested a new Nucleic Acid Amplification test (NAAT) method «TB-SPRINT», that allows simultaneous spoligotyping and prediction of drug-resistance in a sample collection based on ZN extracted DNA [16–18].
Since we wanted to provide a detailed spatial analysis of circulating MTC genotypes, we studied M. tuberculosis diversity in 36 states using ZN-stained smears representative of the country to build geographical genetic maps. A total of 929 ZN slides with grades 3+ (n = 505) and 2+ (n = 424), were collected from adults attending health facilities with cough of more than two weeks duration who had not received treatment for TB [19]. Samples were gathered as part of routine clinical work for the diagnosis of TB. Slides processed by the laboratories were then selected for the routine external quality assurance (EQA) activities conducted by the National TB and Leprosy Control Program (NTLCP) to evaluate the quality of smear microscopy in each State. These slides are routinely anonymised at the time they are submitted to the reference laboratory. Smear positive samples received by the reference laboratory for blind rechecking as part of the EQA were then selected for the study. Only date and origin of the sample were collected. Subjects did not provide informed consent as this is a routine diagnostic procedure. The Research Ethics Committee of the Liverpool School of Tropical Medicine and the Institutional Review Board of Zankli Medical Centre approved the study protocol.
ZN slides were requested via the State TB focal person of the NTLCP from all 36 States and the Federal Capital Territory of Nigeria during the years 2013 and 2014. Slides were shipped by post to Zankli Medical Centre in Abuja. ZN staining and smear grading were performed by the routine diagnostic laboratories following national guidelines [20]. DNA extraction was performed from the stained slides as follows: the mineral oil was removed with xylene, 25 μl of filtered TRIS-EDTA was added to the slides and the material was scraped off into a microcentrifuge tube. The tubes were shipped at room temperature to the Hospital Universitari Germans Trias i Pujol (Spain), where 75 μl of Chelex suspension was added. After thorough mixing, samples were incubated for 30 minutes at 95°C, sonicated for 5 minutes and centrifuged for 15 minutes at 14,000 g at 4°C. The supernatant was transferred to fresh microcentrifuge tubes and sent to the Institute for Integrative Cell Biology (Gif-sur-Yvette, France) for high-throughput spoligotyping.
High-throughput spoligotyping was performed using the microbead-based method using flow-cytometry suspension arrays (Luminex 200, Luminex Corp, Austin, TX), as previously described [21]. TB-SPOL and TB-SPRINT kits were purchased from Beamedex (Beamedex, Orsay, France; www.beamedex.com). Given the low DNA content of the samples, the PCR protocol was adjusted by increasing PCR cycles from 20 to up to 35 cycles. Each spoligotyping profile was verified at least twice and doubtful profiles were rejected. Quality control was conducted by three experts (MKG, BMM, CS) who independently assessed individual spoligotyping profiles. Full or empty squares represents the presence or absence of the 43 classical CRISPR spacers. The spoligotyping patterns were further labelled using the SITVITWEB database [22]. Lineages (L1 to L7) and sub-lineages (such as L.4.6.2, the Cameroon Lineage) were designated using the latest taxonomical standards as provided by whole-genome sequencing, using published lineage definition [23, 24]. When available, the lineage designation was complemented by the classical spoligotyping-based clade designation, e.g. Lineage 4.6.2 is also designated as the «Cameroon Clade (CAM)» [25].
An Excel result file was imported into Bionumerics version 7.5 (Biomérieux, Applied Maths, St Martens-Latem, Belgium). A minimum spanning tree (MST) was built using the Bionumerics user’s manual (Fig 1).
A Mac OSX Version of QGIS (v2.18 Las Palmas de Gran Canaria) was downloaded and installed from: http://www.kyngchaos.com/software/qgis. A licence-free Nigerian map with administrative level 1 was downloaded as a shapefile from http://maplibrary.org/library/stacks/Africa/Nigeria/index.htm; Latitude and longitude of the main Nigerian cities was downloaded from http://www.downloadexcelfiles.com/wo_en/download-list-latitudelongitude-cities-nigeria#.WUPbwYXSWDQ and http://simplemaps.com/data/world-cities.
Nigerian Population records and density was obtained by collecting data from: http://www.iplussolutions.org/isolutions-leads-consortium-streamline-patient-access-essential-treatments-nigeria-0 (assessed on November 2017, 20th) and https://www.citypopulation.de (assessed on December 2017, 8th).
Maps (Fig 2 and Fig 3) were produced using QGIS after importation of.shp and.csv files using the user’s manual (cf. S2 Table).
M. tuberculosis spoligotyping results were obtained in 549 of 929 smear samples (genotyping rate = 59.1%). Of these, 327 (64.8%) were obtained of the 505 slides with smear grades 3+ and 222 (52.4%) of the 424 slides with smear grades 2+. Spoligotyping patterns were named using Spoligotyping International Type (SIT) tags (S1 Table). An MST tree that describes the global population structure is shown in Fig 1.
One hundred and two different spoligotyping patterns were observed, among which 55 had unique patterns with one sample and 47 were clusters with two or more samples. The global distribution of patterns is dominated by the «Cameroon» family, Lineage 4.6.2, including SIT61 and derived signatures characterized by the absence of spacers 23–25 and 33–36, representing about 50% of patterns, and M. africanum, which included M. africanum west African 1 (L5; SIT431 and SIT338, signature: absence of spacer 8–12 and 37–39) and M. africanum west-African 2 (L6; SIT181, signature: absence of 8–9 and 39), which represented approximately 20% of patterns. All other lineages L1 to L4, including East African India/EAI (L1, absence of spacers 29–32 and 34), Beijing (L2, absence of spacers 1–34), Central Asia/CAS (L3, absence of spacers 4–7 and 23–34), Euro-American (L4, absence of spacer 33–36), i.e. all «T spoligotypes» and other derived sublineages (Haarlem, LAM, S, and others, absence of spacers 33–36 plus specific signatures), represented circa 30% of all patterns.
The main spoligotyping results are shown in S1 Table with Lineage assignation. Fig 1 shows the cluster analysis. Two hundred and eighty six (52.1%) of the 549 samples belonged to the L4.6.2 (Cameroon Family), of which 278 were found in 13 different clusters (SIT61, n = 232; SIT838, n = 6; SIT844, n = 3; SIT852, n = 8; SIT1204, n = 3; SIT2550, n = 3; NEW “Koro-Koro 2013”, n = 6 (described in Cameroon as CAM57); «NEW1», n = 4; «NEW12», n = 2; «NEW13*», n = 5; «NEW14*», n = 2; «NEW2», n = 2; «new-f*», n = 2), and eight patterns were orphan with a classical L4.6.2. signature. Full results are shown in S1 Table and S2 Table.
The L5 (M. africanum west African 1) genetic diversity included 87 (15.8%) isolates, of which 75 were identified in 11 clusters (SIT319, n = 6; SIT320, n = 9; SIT331, n = 23; SIT438, n = 11; SIT856, n = 7; «NEW11*», n = 5; «NEW4», n = 2; «NEW6», n = 2; «NEW7’*», n = 2; «NEW8*», n = 6; « NEW9 », n = 2), and 12 were orphan. The L6 (M. africanum west-African 2) included 15 (2.7%) samples of which 10 were found in two clusters (SIT181, n = 8; «NEW3», n = 2) and five were orphan.
The L4 (T1 subfamilies, i.e. other than L4.6.2 and L4.6) included 82 samples among which 75 were found in four clusters (SIT53, n = 68; SIT291, n = 3; SIT334, n = 2; SIT1580, n = 2) and seven were orphan. Within the T2 subfamily (Lineage 4.6), 12 samples were identified in three clusters (SIT52, n = 5; SIT742, n = 2; SIT848, n = 3) and two patterns were orphan. Lastly, a single pattern of the T3 subfamily and four additional orphan patterns were classified as belonging to the broad L4 lineage. Thirteen samples were classified as H3 (L4.1.2 or L4.5) [24]. Nine of these samples were found in three clusters (SIT49, n = 5; SIT50, n = 2; SIT307, n = 2) and four patterns were orphan. A total of seven samples were classified as belonging to Lineage 4.3 (LAM), of which five were in two LAM9 clusters (SIT42, n = 3; SIT1064, n = 2), one was orphan, and one was classified as LAM3.
Twenty samples were classified as Unknown («U»), among which 16 were found in three clusters (SIT1869, n = 2; «NEW10», n = 2; «NEW5», n = 14) and two were orphan patterns.
Finally, the genetic diversity of the remaining 19 samples was as follows: L1 (EAI5; n = 2), L1 (EAI2_Manilla; n = 1); L2 (Beijing; n = 1), L3 (CAS1_Delhi; n = 1), L4.1.1.2 (X, SIT119; n = 3), Lineage 4 (Manu_ancestor SIT523; n = 3) L6 (Mycobacterium bovis; n = 2), and five orphan patterns with no SIT or family that could be assigned by spoligotyping.
Fig 1 presents a Minimum Spanning Tree of all spoligotyping results. Relative diameters of clusters represent the relative percentage of each spoligotyping cluster. Clonal complexes, i.e. samples historically linked to a progenitor clone are easily identified and we observed that L.4.6.2 predominates. The other main clusters presented, belonging to L4 and L5 are also mentioned. The relative position of some clusters of L5 on top of Fig 1 is misleading since these clusters are not related to L4. The States information for the clustered isolates (Fig 1, right panel) shows both intra-State and inter-State clusters.
Fig 2A presents previous results in settings where investigations had been done. Fig 2B shows the list of States of Nigeria. Fig 2C and 2D described population density and absolute population data respectively (see also S2 Table). Fig 3A shows the relative percentage of L1-L6 by State. L4.6.2 was found in 36 out of 37 States (all except Anambra State). L5 was found in 28 States (23 in the south and 5 in the north of the country). L6 was found in 10 States throughout the country but underrepresented in the south, where L5 predominates. Fig 3A and 3B show that there is a predominance of L4.6.2 in the north and a predominance of L5 in the south of the country. In six states, all in the north, central, or west (Yobe, Gombe, Bauchi, Katsina, Kaduna, Niger and Sokoto) we did not detect any M. africanum. The spatial distribution of L4.6.2 in Fig 3B shows that this lineage is prevalent everywhere in the country. However, within L4, L4.6.2 was relatively less frequent compared to other L4 sublineages where L5 predominates.
L5 was present in 28 out of 37 states and was most prevalent in the south and south-east, with Enugu (75%, 15/20) Imo (66%, 4/6) Rivers (60%, 13/22), Benue (47%, 8/17), Akwa Ibom (42%, 3/7), Delta (35%, 7/20), Ebonyi (27%, 6/22) and Abia States (21%, 4/19), in decreasing order, and a marked south-east tropism (Fig 3A), with Ekiti (33%, 5/15) and Ondo (15%, 2/13) being the two most south-western states where L5 clusters were detected. Rivers was the only state with three distinct L5 clusters (SIT430, SIT331 and NEW11). Benue and Enugu States show the presence of relatively large L5 clusters (SIT856 (n = 6) and NEW5 (n = 13), respectively). L6 was found scattered from north to south and east to west in 10 states, detected at low prevalence and only in the Ebonyi state in the south-east. In three states in the north (Kebbi, Zamfara, Kano) and four in the south (Osun, Edo, Abonyi, Taraba) both L5 and L6 cases were detected.
Nigeria is one of the most ethnically diverse and demographically and economically active countries in Africa with an estimated population of 173 million in 2013 and a life expectancy of 55 years (2013) [26]. It is also a Federal country with a rich historical, cultural, ethnical, economical and anthropological diversity and many local languages. Nigerian M. tuberculosis genetic diversity had never been investigated at a national scale and recent publications attempting to describe the molecular epidemiology of TB included only a selection of States (e.g. Anambra, Cross River, Oyo, Plateau, and the FCT) [4, 5, 27–29]. Our study confirms that L4.6.2 is spread over the whole country. Further whole genome sequencing (WGS) would also allow to estimate divergence times and phylogeny within and between sublineages by estimating molecular clock rates based on single nucleotide polymorphisms (SNPs) numbers detected [30, 31].
These new results confirm our previous quantitative results obtained on MTC Lineages prevalence in Anambra, FCT and Oyo states and extends it to the whole Nigerian territory [32]. It is also evident that the L4.6.2 sublineage is not limited to Cameroon and has a much wider geographic distribution [25, 33]. Our results suggest that L6.4.2 is spreading and hence should now be the focus of WGS studies in Nigeria, whether for molecular epidemiology or to gain further knowledge on the resistome and evolutionary history of this L4 sublineage.
We found a relatively high percentage of L5-L6, especially in the south-east of the country, compared to neighbouring countries such as Cameroon and Benin, where it has been suggested that L5-L6 were disappearing [12, 34]. L5-L6 are geographically restricted and clustered in some states, especially in the south-east, where it represents an important percentage of cases. When comparing to neighbouring Benin and Cameroon on the west and east Nigerian borders respectively, the TB genetic structure is quite different. In Benin, a recent study of 100 isolates recruited in 2014, reported the following distribution: L1: 0%, L2: 8%, L3: 1%, L4: 77% (46% of L.4.6.2-CAM and 31% of others), L5: 12%, L6: 2% [34]. If we take into account the most recent results obtained, there might be an underestimation of L5 in the Benin study due to difficulty to grow M. africanum [35]. In Cameroon, the prevalence of M. africanum would have dropped in 2004–2005 to 3.3% [12]. Another more recent study performed in 2009 on 509 patients in the Adamaoua region reports 2,3% of M. africanum [36].
L5 clones (SIT438, SIT331, SIT319) are known to be more prevalent in countries around the Gulf of Guinea [6, 34, 37]. M. africanum prevalence remains stable in Ghana and it has been found significantly more common in patients of the Ewe ethnic group in this country [38]. Furthermore, there seems to be a high diversity of M. africanum in Ghana [39], which was not the result of a single outbreak, as the spoligotyping patterns from Ewe patients were quite diverse [38]. In this paper, we do not provide a formal analysis such as spatial regression proving the link of L5 to specific human subpopulations. Nevertheless, we suggest linguistic, ethnical and cultural association of some MTC genotypes (Lineage 5) with specific populations in Nigeria, that shows, as it had been shown in Ghana, a long-lasting co-evolution of M. tuberculosis and Homo sapiens sapiens [15, 40].
Recently, a study described 315 TB cases caused by M. africanum (L5-L6) in the USA during 2004–2013 [41]. Half of the L5 cases were linked to Nigerian-born patients [41]. We first tagged these spoligotypes using SITVITWEB [22]. Out of 305 patterns, 172 got a SIT label. Common clusters as well as seven new clusters were found to be shared between the USA 2004–2013 study and this present study (S3 Table). Quite often, cases depicted as being unique (orphan) in the USA matched with clustered cases found in this study (S3 Table). These results provide clues for further epidemiological and transmission studies in relation to immigration and TB history in the USA. They also demonstrate the quality of spoligotypes obtained on sputum extracts.
The limitations of this study include that culture is more sensitive than smear microscopy. However, this is not feasible presently in most of Nigeria. Transporting sputa for cultures reduces this sensitivity due to death of mycobacteria and overgrowth of contaminating flora. We also included a low number of samples collected per State (median of 25, range 13–32), which together with the suboptimal performance of spoligotyping from ZN slides and the low number of samples with results per State (median 16, range 3–23) reduced the power of the study and the possibility to conduct a more detailed geographical analysis at sub-State level. Concerning the low sensitivity of the spoligotyping on ZN extracted material, 60–65%, the classical hot ZN is said to result in lower DNA quality than the modified cold Kinyoun method. However we did not compare the quality and quantity of DNA recovered, as the former method (hot ZN) is used throughout the study. Another limitation was the lack of results on drug-resistance-linked SNPs (rpoB; katG, inhA) after PCR amplification by TB-SPRINT on the same material, a limitation that was again likely due to a suboptimal DNA extraction method. Partial results only were obtained with poor sensitivity and paucibacillary DNA-containing material, is likely to require improved DNA extraction methods such as the selective target enrichment using specific oligonucleotide coupled microspheres or other more sophisticated DNA extraction procedures [42–43].
Despite these limitations, this study generated the first nation-wide genetic diversity study of MTC in Nigeria. In three states, Oyo, FCT, Anambra this study confirms our previous quantitative results on the L4/L5-L6 relative prevalence [5]. Our results on L5-L6 prevalence are informative of the phylogeography of M. africanum in Nigeria. In addition, our high-throughput spoligotyping technique has the potential to generate more country-wide data of the genetic distribution of MTC in countries with limited resources. This approach improves the detection of M. africanum, which is often under-represented in culture-based studies [35]. With the advent of Next Generation Sequencing (NGS), more precise population-based information could easily be obtained at affordable costs, without TB culture, even in Africa. Such projects may however require samples either with a higher DNA yield or more complex DNA extraction procedures, and necessarily result in a more selected sample of patients attending secondary and tertiary hospitals. The use of slides in turn generates genetic epidemiological information at a larger scale without the need of this infrastructure. Last but not least, our results suggest that some specific TB control and/or preventive measures could be specifically taken in densely populated areas in the south-east region of the country (Fig 2C and 2D), and that assessing the drug-resistance status of the L4.6.2 lineage is an important issue for the Nigerian TB control program.
|
10.1371/journal.pntd.0005582 | “We do not bury dead livestock like human beings”: Community behaviors and risk of Rift Valley Fever virus infection in Baringo County, Kenya | Rift Valley Fever (RVF), is a viral zoonotic disease transmitted by Aedes and Culex mosquitoes. In Kenya, its occurrence is associated with increased rains. In Baringo County, RVF was first reported in 2006–2007 resulting in 85 human cases and 5 human deaths, besides livestock losses and livelihood disruptions. This study sought to investigate the county’s current RVF risk status.
A cross-sectional study on the knowledge, attitudes and practices of RVF was conducted through a mixed methods approach utilizing a questionnaire survey (n = 560) and 26 focus group discussions (n = 231). Results indicate that study participants had little knowledge of RVF causes, its signs and symptoms and transmission mechanisms to humans and livestock. However, most of them indicated that a person could be infected with zoonotic diseases through consumption of meat (79.2%) and milk (73.7%) or contact with blood (40%) from sick animals. There was a statistically significant relationship between being male and milking sick animals, consumption of milk from sick animals, consuming raw or cooked blood, slaughtering sick livestock or dead animals for consumption (all at p≤0.001), and handling sick livestock with bare hands (p = 0.025) with more men than women engaging in the risky practices. Only a few respondents relied on trained personnel or local experts to inspect meat for safety of consumption every time they slaughtered an animal at home. Sick livestock were treated using conventional and herbal medicines often without consulting veterinary officers.
Communities in Baringo County engage in behaviour that may increase their risk to RVF infections during an outbreak. The authors recommend community education to improve their response during outbreaks.
| The study focuses on the knowledge and socio-cultural practices around Rift Valley Fever (RVF) in Baringo County. It is intended to identify means through which communities in Baringo County could be exposed to RVF in the event of an outbreak. Specifically, it addresses knowledge of RVF transmission routes, practices in handling and consumption of meat, milk and blood; livestock disease management and disposal of dead animals/aborted foetuses. The study found that community members engaged in practices that would expose them to RVF in the event of an outbreak. These practices include milking and consuming milk from sick animals; consuming meat from slaughtered sick animals and those that die from disease; rarely having animals that were slaughtered at home inspected by a veterinary officer or a local animal expert before consumption; using uncertified techniques to test meat for safety of consumption; and treating sick livestock with both conventional and herbal treatments without the guidance of veterinary personnel. Further, RVF infections are likely to follow a gendered pattern based on the division of labor in livestock production. Based on their results, the study authors recommend community education to increase RVF awareness.
| Rift Valley Fever (RVF) is a zoonotic disease associated with human and livestock morbidity and mortality as well as decreased trade in livestock and derived products. It is a viral disease caused by a Phlebovirus of the Bunyaviridae family [1–3]. It is transmitted by infected aedine and culicine mosquitoes [4, 5] and through contact with infected animal tissue and secretions [1, 2]. To date, there is no evidence of human to human RVF transmission [6]. Domestic ruminants, mainly cattle, sheep and goats are susceptible to RVF [3]. Infection with the RVF virus causes distinct disease in animals and humans. In livestock, the disease is marked by mass abortions in pregnant animals and mortality in newborns [1]. In humans, RVF often manifests as a mild febrile illness that may go undetected [7]. In rare occasions, the infections develop into severe disease causing hemorrhage, encephalitis and fatalities in 1% of cases and ocular impairment in 0.5–2% [7]. Due to its public health and economic impacts, RVF is categorized as “notifiable” by the Kenya government, thereby requiring that all suspected livestock and human cases within Kenya be reported to the government, which upon confirmation must formally inform the World Organization for Animal Health (OIE) [8] and the World Health Organization (WHO)[9], respectively.
Initially, RVF outbreaks were spatially confined to Africa (including Madagascar) but in the year 2000, the disease spread to the Arabian Peninsula [10]. In East and South Africa, RVF outbreaks are associated with wet seasons with higher than normal rainfall resulting in floods [3, 11, 12] which subsequently encourage multiplication of RVF vectors [1, 5]. However, RVF can occur in the absence of rain as has been witnessed in North and West Africa where it is linked to increased mosquito populations in large rivers and dams [3]. Movement of infected vectors, persons and animals could also lead to emergence of the disease in non-endemic areas [13].
In Kenya, RVF was first characterized in 1931 [3], and has since been reported nine more times with the latest outbreak in 2006 [6]. The outbreaks have mainly occurred in northern Kenya, mainly in Ijara and Garissa areas [11,13–15]. The 2006–2007 outbreak is estimated to have cost the country US$32 million (1US$ = 65Kenya shillings) in losses of livestock, livestock productivity, trade in livestock and livestock products and allied services [14]. Of the 340 confirmed human cases of RVF in Kenya during the 2006–2007 outbreak 60% were from the northern regions including Garissa (31%), Ijara (22%) and Wajir (5%) areas [15]. A further 10% were from the coastal district of Kilifi [15]. Occurrence of RVF in Baringo County was first recorded during the 2006–2007 outbreak [15]. The outbreak occurred against the backdrop of high cattle, sheep and goat populations in the County [16] and flooding in the lowland areas around lake Baringo following the exceptionally heavy rains of 2006 [17]. The affected areas also have solanchak soils which have previously been linked to RVF in Northern Kenya [17].
The most effective method of controlling RVF in Kenya is livestock vaccination but it is done inconsistently due to irregular outbreaks [11, 18]. Further, delays in laboratory confirmation of RVF cases result in ineffective and untimely corrective interventions [17, 19]. Bans placed on livestock trading and consumption of derived products during outbreaks for disease control are not only difficult to enforce but also pose great dietary and livelihood challenges to communities [1, 20]. Most human RVF cases have been attributed more to risky handling or consumption of livestock and derived products than bites from infected vectors as exemplified in South Africa [21], Mayotte [22], Tanzania [23], West Africa [24] and Kenya [1, 15, 25].
Previous social studies on RVF in Kenya have focused mainly on Northeastern Kenya where most RVF outbreaks have occurred [14, 26–28]. In Baringo County, RVF studies have focused on RVF vectors [29, 30], and human [15] and livestock serology [17]. No studies have been done to determine how knowledge and socio-cultural practices influence community risk to RVF in the area. This paper generates additional information on the role of knowledge, attitudes and practices in the transmission dynamics of RVF in Baringo County. It also explores differences in risk to RVF infection, between men and women and among zones.
The study acquired both national and the World Health Organization (WHO) ethical clearance referenced P70/02/2013 and Protocol ID B20278 respectively. All participants were of consenting age (18 years and above) and were required to give written consent before engaging in any research activity. For illiterate participants, the researcher read out the consent form details and allowed participants to consent through provision of a thumb print instead of a signature.
The study took place in Baringo County’s Central, North and Marigat sub-counties. The research team classified the study site into four zones namely the highland, midland, lowland and riverine based on altitude. The highland has an altitude of >1500m above sea level (asl), midland >1000m-1500m asl, the lowland and riverine zones at <1000m asl. (Fig 1). The riverine zone is the area on the extreme left while the lowland zone is on the extreme right of the study site map, (Fig 1). The Tugen, a sub-tribe of the Kalenjin community, mainly inhabit the highland, midland and riverine zones while the Ilchamus who are a sub-tribe of the Maa community, are found in the lowlands. Both communities practice agriculture but the Tugen engaged in both crop and livestock farming while the Ilchamus are mainly livestock keepers. The 2006–2007 RVF outbreak occurred only in the lowland zone where the Ilchamus are found.
The study utilized a cross-sectional research design in which a Knowledge, Attitudes and Practices (KAP) survey and focus group discussions (FGDs) were conducted sequentially. Survey respondents and focus group discussants were mutually exclusive. Qualitative data on livestock production and livestock disease management practices was collected first and comprised of 26 FGDs. Due to differences in zones, sex and community distribution, iterations resulted in 26 FGDs (13 male only and 13 female only) with a total of 231 discussants. For triangulation of male and female views, four FGDs were conducted per zone, in the highland, midland and riverine areas among the Tugen. In the lowland zone, 10 focus group discussion were conducted among the Ilchamus and 4 in a rural town that had mixed communities. Purposive sampling technique was used to select FGD discussants. To qualify as a discussant, an individual had to be 18 years old and above, have lived in the area for at least one year, be a current livestock keeper or from a livestock keeping household or consumer of livestock products with previous experience in livestock keeping.
The KAP survey, whose questionnaire was informed by the FGD’s findings, targeted 560 individuals drawn from the four ecological zones. The sample size was determined through the proportion to size sampling methodology for a finite population which resulted in 383 respondents (from 20 clusters, 5 clusters per zone). A further 5% was added to cover for possible incomplete questionnaires resulting in a sample of 400 (rounded figure). Owing to a desire to increase the external validity of the survey findings, the researchers proportionately increased the sample size by 8 respondents per cluster leading to a total sample size of 560. In each zone, clusters were selected from areas with at least 30 households. Thereafter, survey respondents were identified through simple random sampling ensuring that both men and women were proportionately represented. The zoning of the study site was used to assess whether there were any differences in knowledge and practices on livestock keeping and handling of animal products.
The FGD guide was pretested in 2 separate FGDs comprising exclusively of men or women to check for its suitability. Similarly, the survey questionnaire was pretested with 40 respondents in three areas which shared similar characteristics with the sampled sites. These sites were consequently excluded from the main survey. Final adjustments were made to the FGD and survey tools prior to data collection. Only Tugen and Ilchamus speaking enumerators participated in data collection.
Survey data was analyzed in SPSS version 23 (IBM SPSS Statistics, Armonk, New York) after importation from CSPRO version 6.1 (United States Census Bureau, Washington DC) where it was entered and cleaned. Besides summary statistics, independent t-tests, one way ANOVA and Chi square tests were conducted to determine the relationship between different variables. Missing values were excluded from the analyses. The measure of statistical significance for this study was set at a p-value of 0.05. A binary logistic regression was also fitted to assess the association between respondents’ level of risk to RVF infection and demographic characteristics which comprised of zone of residence, sex, age, education level, marital status, household headship type, number of children, individual scores of knowledge of RVF transmission modes and livestock (cattle, sheep and goats) quantities measured in Tropical Livestock Units (TLUs) as guided by Cholinda and Otte [31]. The overall knowledge of possible RVF transmission routes was determined through eight questions on contact with sick animals, animal tissue, secretions and consumption of products from sick animals in the KAP survey. Each question had a Likert scale type of response where those who wholly disagreed, somewhat disagreed or neither agreed nor disagreed with a possible RVF transmission route were classified as not knowledgeable and scored a zero while those who somewhat or wholly agreed were classified as being knowledgeable and awarded a score of 1. The eight answers that a respondent gave were used to generate a cumulative score on knowledge ranging from 0–8.
The level of risk of exposure to the RVF virus was determined through 23 KAP survey questions with Likert scale responses ranging from never engaging in a given practice, or engaging very few times, sometimes, most of the times or always. The questions addressed community practices on: handling and consumption of milk, meat and blood; disposal of dead livestock; management of animals that abort; foetus disposal; and handling and treatment of sick livestock. For each question, respondents that carried out good practice were awarded a score of 1 while those that did not got 0. Individual outcomes were summed to give the total score per respondent. Respondents with scores below or equal to the mean were classified as high risk and above as low risk. The binary categorization of risk was used as the dependent variable in a binary logistic regression where the high risk category was coded as 0 and the low risk as 1. The model’s goodness of fit was tested using the Hosmer-Lemeshow test (χ2 = 0.617, df = 8, p = 1.000) and the omnibus-corpus test (χ2 = 180.799, df = 21, p<0.001) owing to the contested credibility of the Hosmer-Lemeshow test [32].
In each FGD, data was captured through note taking and audio recording. Audio files from the Tugen and Ilchamus were later transcribed directly verbatim into English by native speakers fluent in English and Swahili. Each script was verified through comparison of content with its recorded audio file and corresponding notes. Cleaned FGD data was coded into salient themes in Nvivo 10 (QSR international, Melbourne) and analyzed using the content analysis method. The emergent themes are presented together with the survey data in the results section.
The KAP survey respondents (total of N = 560), n = 266 (47.5%) were male and n = 294 (52.5%) were female. Their average age was 44 years but most (n = 147, 26.3%) were aged between 27–35 years. Slightly more than half (n = 291, 52%) had primary education and were religiously affiliated to the Christian faith (n = 554, 99%). Most respondents (n = 439, 78.4%) were in monogamous unions while the others were either in polygamous unions (n = 59, 10.5%) or single (n = 62, 11.1%). Their main income sources were crop farming (n = 266, 47.5%) and livestock farming (n = 113, 20.2%). A total of 112 men and 119 women aged between 18–84 years with an average age of 41.7 years participated in the FGDs.
In Baringo County, focus group discussants reported that livestock were considered stores of wealth; sources of food, medicine, income, manure, skins/hides, draft power, bride price, social status/prestige; and an indigenous means of predicting rainfall patterns (by “reading a goat’s stomach”) and conducting rituals. The main foods derived from livestock were meat, milk, blood, eggs and animal fat. However, meat and milk constituted a greater part of the communities’ diets compared to blood, eggs and animal fat. Among people suffering or recovering from diseases locally assumed to be severe, meat stock and milk were also used in the administration of conventional and herbal medicines. For children, medicines were ingested in or with milk whereas stock derived from boiling meat was favored for adults. Extracts from goat rumen and intestines believed to be medicinal were also used as reported by a male focus group discussants from Perkerra (lowland zone) and Borowonin (highland zone) saying, ““eyande” [a green liquid extracted from the rumen of a goat] treats chronic malaria,” and “you get uncleaned goat small intestines, cut them into pieces, mix with herbs and boil. When cooked, you take them and become well.” In the region, goat meat was most preferred, followed by beef then mutton. Discussants estimated local livestock proportions by species through a proportion piling exercise in which the moderator gave them 100 stones representing the total livestock population in their locality and asked to divide them proportionately to the livestock species they kept. The FGD participants estimated livestock populations at a median percentage of cattle 22.5% (range 13%-55%), goats 34% (range 5%-44%), sheep 20.5% (range 10%-30%), chicken 18% (range 7%-39%), donkeys 3.5% (range 0–10%), rabbits 0 (range 0–9%) and pigs 0 (0–8%).
Based on the KAP survey data, n = 481 or 86% of the respondents had heard of RVF (N = 560). Among those who had heard of RVF (n = 481), the main sources of information on RVF were radio (n = 330, 68.6%), friends and family (n = 195, 40.5%), veterinary officers (n = 166, 34.5%), community/public health officials (n = 95, 19.8%), health facilities (n = 93, 19.4%) and local animal experts (n = 86, 17.9%) as shown in Fig 2. The least utilized sources were internet (n = 2, 0.4%), text books (n = 9, 1.87%), posters/pamphlets (n = 11, 2.29%) and television (n = 13, 2.7%). According to focus group discussants from the lowland zone where the 2006–2007 outbreak occurred, there were attempts to name the disease without consensus. Proposed names were “the El -Nino livestock disease” coined from the period when the disease last occurred and “ngea na nyori” which translated to “the greenish/yellow pigment” found in cadavers. Among the Tugen, the term “kipkoloswo” which refers to yellowing characteristic of those infected with yellow fever was also used to describe RVF disease in humans.
Few KAP survey respondents were knowledgeable of RVF signs and symptoms in humans (N = 559, excluding missing values). The following signs and symptoms were identified: fever (n = 85, 17.7%), headaches (n = 78, 16.2%), jaundice (n = 69, 14.3%), vomiting (n = 66, 13.7%), diarrhea (n = 61, 12.7%), bleeding from body openings (n = 56, 11.6%), joint pains (n = 54, 11.2%) and impaired vision (n = 40, 8.3%). A male focus group discussant from Lorok in the lowland zone reported that “people infected with RVF showed some signs which resembled those of malaria; that is having very high fever, weak joints and having a headache”.
Survey respondents (N = 558) had limited knowledge of the cause of RVF. Only a third, n = 169 (30.3%) of respondents knew that mosquitoes had capacity to transmit a livestock disease to humans (Fig 3). Focus group discussants further reinforced this by implicating bad air, tsetse flies, ticks, monkeys and rains as causes of RVF. The main means through which KAP respondents believed they could be infected a with livestock zoonotic disease were through consumption of meat n = 442 (79.2%, N = 558) and milk n = 411, (73.7%, N = 558) or contact with blood n = 221, (40%, N = 553) from sick animals (Fig 3). Contact with sick animals n = 125, (22.4%, N = 559), their discharge from eyes and nose n = 166, (29.7%, N = 558), meat n = 123, (22%, N = 555), and skins/hides n = 84, (15.1% N = 555) were least associated with exposure to disease. Independents t-tests also showed that men and women had near equal mean knowledge scores of 4.54 and 4.84 respectively. There was no statistical difference between their knowledge levels t(558) = -1.383, p = 0. 167. Results from a one way ANOVA test showed that there were statistically significant differences between zones (F (3,556) = 6.571, p<0.001). Turkeys’ post hoc test further showed that there were statistically significant differences in knowledge score between the lowland and the midland (p = 0.001) and riverine zones (p = 0.003) but not the highland (p = 0.581).
The mean and median scores of the 23 questions used to determine the level or risk of exposure to RVF was 12 while the lowest score recorded was 2 and highest 20. Those that had a score of ≤12 n = 326 (58.3%) were categorized as high risk and those with 13–23 n = 233 (41.7%) as low risk (N = 559). The binary categorization of risk scores was used as the grouping variable in a binary regression model fitted to test the association between level of risk of exposure to the RVF virus and demographic characteristics. Of the variables, the highland zone, male sex and scores on knowledge of RVF transmission routes were found to be statistically significant (Table 1). People from the highland zone had 9.253 times less risk of exposure to the RVF virus compared to those from the riverine zone. The odds of men engaging in unsafe practices were 0.176 times more that of women. The odds of engaging in safe practices were higher with increased respondents’ level of knowledge of RVF transmission routes. Specifically, an increase of 1 mean score in knowledge of RVF transmission routes significantly increased the odds of engaging in safe practices by 1.144.
Comparisons of mean scores on level of engagement in RVF risk practices between men and women through an independent T-test further confirmed that women engaged less in risk practices compared to men t(555) = -8.082, p<0.001. Women had a mean score of 12.87 while men had 10.45. By zone, mean scores showed that people in the highland zone engaged in less risk practices (14.40), followed by the midland (11.20), riverine (10.79) then lowland (10.22) in increasing order. Higher scores indicated low risk of exposure to the RVF virus and vice versa. Turkey’s post hoc tests, also showed that there was a significantly statistical difference between the highland and the other zones (p<0.001).
Majority of KAP survey respondents n = 508 (91.5%), reported that they always boiled their milk before consumption (N = 555). However, n = 349 (62.4%) never consumed milk from or milked n = 326 (58.7%) sick livestock. There was a statistically significant relationship between sex and milking (χ2 = 22.146, df = 4, p<0.001) and consumption of milk from sick animals (χ2 = 53.875, df = 4, p<0.001) (Table 2). Women, whose role it was to milk, were less inclined to milk sick livestock while men showed higher tendency to consume milk from sick livestock. When animals were slaughtered at home, n = 349 (62.7%, N = 556) of respondents reported that some household members had ever consumed the raw blood while n = 424 (76.3%, N = 556) had ever consumed it cooked. The association between sex and consuming raw (χ2 = 23.970, df = 4, p<0.001) or cooked blood (χ2 = 23.556, df = 4, p<0.001) was statistically significant (Table 2). A higher proportion of men, whose role it was to slaughter livestock, consumed raw or cooked blood more often than women.
More than half n = 331 (59.7%, N = 554), of the respondents had ever eaten meat from a sick animal that had been slaughtered. A further n = 282 (50.8%, N = 555) had ever eaten meat from an animal that died from sickness. There was a statistically significant relationship between sex and slaughtering sick livestock (χ2 = 50.909, df = 4, p<0.001) or dead (χ2 = 50.358, df = 4, p<0.001) animals for consumption with higher proportions of men than women likely to engage in both practices (Table 2). Only n = 131 (23.6%, N = 556) of respondents relied on trained personnel or local experts (n = 65, 11.7%, N = 555), to check the meat for safety of consumption every time they slaughtered. Focus group discussants reported that they applied other traditional methods besides utilizing services from the experts. These included observing the health reactions of those who consumed the meat earlier and if no harm occurred they would also consume of it as exemplified in the following excerpt.
From various group discussions, different tree species used to cure meat slaughtered from sick/dead animals were identified. The tree species included “soget”/”sokonyi” (Warbugia ugandensis), “sessiat” (Acacia tortilis), “subeiwa”/“ntepes” (Acacia nubica), and “segetet” (Myrisine africana). Once meat was boiled with herbs from these tree species it was considered safe for consumption. Meat prepared in this manner was sometimes only consumed by a segment of the population. For instance, “for an animal with anthrax, men exclude women and children and they boil the meat in herbs for a long time and eat,” Female discussant, Litein-4, riverine zone.
Two ant species, “kilik” (Messor angularis) and “butbutie”, (Crematogaster sp) were used by the Tugen to test meat for safety of consumption by placing a piece of meat from the slaughtered animal near the ants’ nests then people would observe whether they (ants) would attempt to eat it or not. If the ants avoided the meat it was considered unsafe for consumption but if they did not, the meat was considered harmless hence eaten as exemplified in the following excerpts.
Another method was burying the spleen in soil and if it appeared to increase in size the meat was considered unsafe for consumption but if there was no increase the meat was considered safe for consumption as demonstrated below.
Community members used multiple methods to dispose of dead animals. Besides consumption, it was established in FGDs that animal cadavers were also buried whole, skinned then buried, skinned and given to dogs, skinned and thrown in the open or burned. Among survey respondents, only n = 153 (27.5%, N = 556) and n = 50 (9%, N = 557) reported that they always buried or burned sick animals after death, respectively. Aborted foetuses, were always buried in n = 159 (28.4%, N = 560) of cases. Up to n = 220 (40%, N = 550) of respondents reported ever leaving foetuses out in the open to rot and n = 404 (73.2%, N = 552) feeding them to dogs. There was a statistically significant association between sex and feeding aborted foetuses to dogs (χ2 = 18.114, df = 4, p = 0.001) with more men inclined to engage in the practice (Table 2). Among the Ilchamus, it was a taboo to bury dead livestock as shown in the following excerpt.
Even when they resolved to bury the dead animals, some community members would skin the animal because of the belief that “when it [a sick animal] is buried with the skin/hide on, it will cause harm to the remaining stock and they might die,” as reported by a female focus group discussant from Borowonin, the highland zone.
Management of livestock diseases was mainly left to community members and was traditionally prescribed for men. Only n = 242 (44%, N = 550) of respondents relied on a veterinary officer to treat their sick livestock most of the time whereas n = 339 (61.6%, N = 550), mostly bought veterinary medicines and treated the sick animals without the guidance of a veterinary officer. When animals aborted, n = 364 (65.7%, N = 554) of respondents often treated them with conventional veterinary medicines while n = 125 (22.6%, N = 553) used herbal treatments. Veterinary services were considered expensive most of the time by n = 343 (63.9%, N = 537) of respondents. An equal proportion of respondents n = 468 (84%, N = 557) reported often handling sick livestock and assisting deliveries with bare hands. Men were more prone to handling sick livestock with bare hands (χ2 = 11.185, df = 4, p = 0.025); treating sick livestock without consulting a veterinary officer (χ2 = 18.326, df = 4, p<0.001); consider veterinary services as expensive (χ2 = 13.210, df = 4, p = 0.010); and reporting that accessing veterinary medicines was difficult (χ2 = 32.627, df = 4, p<0.001); probably due to their experience in managing livestock diseases (Table 2). On the other hand, more women were inclined to seek a veterinary officer’s services for livestock treatment (χ2 = 17.539, df = 4, p = 0.002); think that veterinary services were easy to access (χ2 = 33.915, df = 4, p<0.001) but find it difficult to administer veterinary medicines (χ2 = 26.884, df = 4, p<0.001) probably because traditionally, the role of livestock disease management was not theirs (Table 2).
The occurrence and coverage of RVF is determined by a multiplicity of factors which include availability of susceptible hosts, competent vectors, adequate precipitation and permissive ecology besides human behavior [18, 20, 33]. The current study found that livestock farming was ranked second in importance as a livelihood activity after crop farming and farmers kept cattle, sheep and goats which are susceptible to the RVF virus.
The level of knowledge of RVF signs and symptoms in humans was low in the current study. Fever was the most known to respondents and impaired vision the least. In contrast, a study on RVF knowledge, attitudes and practices in Kilombero and Kongwa regions of Tanzania found that hemorrhage was the most known sign while joint pains/jaundice were the least known [34]. Similar to the current study, the level of knowledge of RVF in Tanzania was equally low. The disease mainly manifests as an uncomplicated febrile illness with flu-like symptoms but may involve hemorrhage, encephalitis or visual impairment [3] in <8% of cases [23]. Thus, the paucity in knowledge of RVF signs and symptoms may lead to misdiagnosis implicating other febrile ailments, such as malaria, that may be endemic in a region [34]. Inhabitants of Baringo County have been found to be knowledgeable of malaria signs the symptoms hence their ability to relate its symptomatology with that of RVF [35]. Unlike the Somalis in North Eastern Kenya who had a local name, “sandik” (bloody nose), for RVF, [26], neither the Tugen nor Ilchamus had a widely accepted term possibly because the disease was reported for the first time in 2006–2007 [17].
Good knowledge of possible RVF transmission routes was statistically associated with low risk of exposure to the RVF virus in the binary logistic regression. This finding concurs with another study on RVF knowledge, attitudes and practices in Ijara-North Eastern Kenya, where high knowledge of preventive measures was associated with high knowledge of RVF [26]. The study further showed that high knowledge of the disease was not associated with age, sex, education, marital status, household size [26]. Another study in Kilombero and Kongwa regions in Tanzania found that besides being male or coming from Kongwa, other socio-demographic characteristics had no effect on knowledge on RVF transmission, symptoms and prevention [34]. Another RVF study in Mayotte reported that having none or primary education increased the risk of RVF infections [22]. This shows that knowledge of RVF can be determined by factors which vary from place to place. In Baringo, the difference in knowledge of risk practices between the midland, lowland and riverine zone may have been as a result of differential exposure to livestock diseases and access to veterinary services.
That only few of the survey respondents knew that mosquitoes could transmit other (unspecified) diseases besides malaria was indicative of a knowledge gap. The uncertainty of the cause of RVF was further established through FGDs in which discussants implicated bad air, monkeys, tsetse flies and rain. While heavy rain is a trigger of multiplication of RVF vectors, it is not the cause. Notably, RVF outbreaks in Kenya have been associated with the El-Nino Southern Oscillation (ENSO) phenomenon which causes higher than normal rainfall, which subsequently provides ample breeding sites for RVF vectors [36].
Communities in Baringo County did engage in risky practices through handling and consumption of livestock products, management of sick animals, disposal of foetuses and dead mature stock. The proportion of people that consumed boiled milk in Baringo County was higher than reported in other studies in Africa. For example, in Sudan, a majority consumed boiled milk, while a relatively low number consumed raw milk, fermented or cooked/made into cheese in decreasing order [37]. In Ghana, most herders drank raw milk, followed by those who consumed boiled milk while those who consumed either raw or boiled milk were the least [38]. Among pastoral communities in Ijara, Kenya, only a few people consumed boiled milk [26]. The high adoption of the practice of boiling milk before consumption may have been as a result of health campaigns against Brucellosis that were reported in the region. Milking [39] and consuming raw milk from infected animals [40] have previously been identified as possible RVF virus transmission routes.
The utilization of blood as food was part of the Tugen and Ilchamus culture and was still practiced by some. The practice has been associated with pastoral communities who hold the belief that raw blood is nutritious [27]. However, contact with blood has been implicated as a risk factor of RVF virus infection during outbreaks in South Africa [21] and Kenya during the 2006–2007 [1] and 1997–1998 outbreaks [25]. Blood has been identified as part of body fluids that are highly viraemic hence very infective [39].
The use of observation, ants, herbs or the spleen to determine meat safety reflected the role indigenous knowledge played in determining community health outcomes and suggests the need for further research into the efficacy of these methods. Limited uptake of meat inspection after domestic slaughter showed that there was risk of consuming infected meat in the event of an RVF outbreak. During the 2006–2007 outbreak, the disease was transmitted to humans from infected animals through slaughtering, skinning and consumption of infected meat [17].
The use of livestock products as part of treatment courses carries potential for exposing sick people and their care givers to infection during an RVF outbreak since they will slaughter, prepare or consume animal products as exemplified in this study. Similar practices have been recorded during RVF outbreaks. For instance, fat extracted from mutton (which is derived from a highly susceptible animal to RVF), has been used in treatment of RVF symptoms like fever and hematochezia [27]. Raw blood, unpasteurized milk and fat derived from sheep have also been used in treatment of people with RVF in Ijara [28].
The processes through which foetal material are disposed can increase risk to RVF in an outbreak setting since birthing fluids contain high volumes of the infective virus [39]. Indeed, a study of RVF sero-positivity in northern Kenya conducted after the 1997–1998 outbreak established a statistical association between RVF sero-positivity and disposal of aborted foetuses [40]. In Mayotte, sero-positivity was associated with aiding livestock in delivery and contact with/disposal of aborted foetal material [22]. In this study, aborted foetuses were not always burned or buried indicating that poor disposal was a risk factor in the area.
Carcass disposal in the study site was mainly by consumption or burying. Among the Ilchamus, it was against the culture to bury a dead animal and consumption was preferred. This practice was found to be further reinforced by the belief that boiled meat was safe for consumption and carried potential for challenging the regulations provided for disposal of condemned carcasses. A similar belief was noted in Ijara, where people reported that boiled meat carried no disease [28]. In Tanzania, a study on RVF found that survey respondents skinned dead animals, buried or left it in the open [34].
Cattle, sheep and goat diseases were mainly managed by men using herbal and conventional medicines without the guidance of veterinary officers in Baringo. In addition, sick livestock were often handled with bare hands. Combined, these risky practices enhanced possibility of infection during an outbreak. Use of protective gear was also found to be a challenge in Tanzania, where only a quarter of respondents reported using them in handling dead animals [34]. The current study suggests that in the event of an outbreak, men and women would be exposed differentially, with men being at higher risk than women due to their role in treatment of sick animals and slaughtering. Similarly, Anyangu [1] and Nguku [15], reported that men were more at risk occasioned by animal related exposures through herding, slaughtering, skinning and milking of livestock during the 2006–2007 RVF outbreak in Kenya. An earlier study in Ijara district, in northern Kenya, conducted after the 1997–98 outbreak found that men had a three times more likelihood of sero-positivity compared to women due to exposure to infected vectors and animals [40]. Seufi [41], reinforced this outcome from the 2007 outbreak in Sudan that found that males aged between 15–19 years were most susceptible compared to women. Occupations such as being a farmer or housewife were also found to put individuals at risk [41]. In Mayotte, being male has been associated with RVF virus sero-positivity because men spent longer periods outdoors facilitating exposure to infected mosquitoes [22]. Thus, the role of gendered division of labour needs consideration for effective RVF risk management.
Communities in Baringo County were found to have limited knowledge on RVF causes, human signs and symptoms. Poor handling and consumption of livestock products, treatment of livestock, disposal of foetuses and carcasses were identified as possible routes of exposure to RVF virus in Baringo County. Men and women would be differentially exposed to the disease based on their gender roles in livestock farming.
The study underscores the importance of qualitative data in understanding community knowledge, attitudes and practices on diseases. In this study, it is through focus group discussions that beliefs and practices that would endanger lives such as taboos on burying dead livestock, skinning or cutting open an animal’s abdomen before disposal; traditional methods of checking for meat safety and using animal products in disease management were identified as RVF risk factors. This demonstrates that exclusive use of quantitative methods of data collection in behavioral studies can lead to loss of opportunity to gather critical insights into social problems.
The study recommends that community members should be consistently provided with information on RVF seasonality, manifestation in humans and livestock and the risk factors to strengthen their capacity in engaging in participatory disease surveillance and prevention. The health information should be tailored specifically for the local context to constructively challenge the existing myths and misconceptions. It should be relayed orally, preferably in local languages so that both the literate and illiterate community members understand. This is particularly important since loss of knowledge on RVF is possible owing to lengthy intervals between outbreaks. In addition, veterinary services within the county need to be made more accessible and affordable for effective livestock disease control.
This study was conducted nearly a decade since the first reported RVF outbreak in Baringo County. Therefore, it focused more on assessment of current knowledge and risk practices than practices conducted during the last outbreak. Adherence to good practice in livestock production and RVF control was self-reported by the study participants rather than observed by the researchers. Therefore, there is possibility that knowledgeable respondents may have stated that they engage in good practices because they know it is desirable. While qualitative data was collected through focus group discussions, the findings are very specific to the Baringo County context and cannot be generalized to other areas. These limitations notwithstanding, the study provides insights into the risk factors that would expose the community to RVF in the event of an outbreak.
|
10.1371/journal.pntd.0004786 | Evolution of the Transmission-Blocking Vaccine Candidates Pvs28 and Pvs25 in Plasmodium vivax: Geographic Differentiation and Evidence of Positive Selection | Transmission-blocking (TB) vaccines are considered an important tool for malaria control and elimination. Among all the antigens characterized as TB vaccines against Plasmodium vivax, the ookinete surface proteins Pvs28 and Pvs25 are leading candidates. These proteins likely originated by a gene duplication event that took place before the radiation of the known Plasmodium species to primates. We report an evolutionary genetic analysis of a worldwide sample of pvs28 and pvs25 alleles. Our results show that both genes display low levels of genetic polymorphism when compared to the merozoite surface antigens AMA-1 and MSP-1; however, both ookinete antigens can be as polymorphic as other merozoite antigens such as MSP-8 and MSP-10. We found that parasite populations in Asia and the Americas are geographically differentiated with comparable levels of genetic diversity and specific amino acid replacements found only in the Americas. Furthermore, the observed variation was mainly accumulated in the EGF2- and EGF3-like domains for P. vivax in both proteins. This pattern was shared by other closely related non-human primate parasites such as Plasmodium cynomolgi, suggesting that it could be functionally important. In addition, examination with a suite of evolutionary genetic analyses indicated that the observed patterns are consistent with positive natural selection acting on Pvs28 and Pvs25 polymorphisms. The geographic pattern of genetic differentiation and the evidence for positive selection strongly suggest that the functional consequences of the observed polymorphism should be evaluated during development of TBVs that include Pvs25 and Pvs28.
| Plasmodium vivax is the most prevalent human malarial parasite outside Africa. The fact that patients can relapse due to the parasite dormant liver stages, among other biologic and epidemiologic characteristics of vivax malaria, facilitates the persistence of the disease in many endemic areas. These challenges have fueled the search for new control tools, including transmission blocking (TB) vaccines targeting the parasite sexual stages. Here we study the genetic diversity of two major TB vaccine antigens, Pvs25 and Pvs28. We show that these genes are relatively conserved worldwide but still harbor diversity that is not evenly distributed across the genes. These patterns are shared by the same proteins in closely related parasite species suggesting their functional importance. We also identify strong geographic differentiation between the circulating variants found in Asia and the Americas. Finally, evolutionary genetic analyses indicate that the observed variation in both genes could be maintained by natural selection. Thus, these polymorphisms may confer an adaptive advantage to the parasite. These results indicate that the genetic variation found in these genes and their geographic distribution should be considered by vaccine developers.
| Transmission-blocking (TB) vaccines are considered an important tool for malaria control and elimination [1]. TB vaccines aim to disrupt malaria transmission by eliciting antibody mediated responses against antigens expressed during sexual or sporogonic stages of the parasite thereby inhibiting its development inside Anopheles mosquitoes. Thus far, the search of suitable targets for TB vaccines has yielded promising results. In particular, antibodies against some of the multiple parasite proteins have shown excellent TB activities [1]. Among those antigens, the ookinete surface proteins Pvs28 and Pvs25 have been considered candidates to be incorporated in TB vaccines against Plasmodium vivax. These proteins may have originated as result of a duplication event and their orthologous genes (referred to as p28 and p25) have been described in many Plasmodium species [2,3]. P28 and P25 are the two most abundant membrane proteins expressed on the zygote and ookinete surfaces; indeed, they might represent as much as 25% of the total ookinete surface proteins [4]. Their structure has been characterized as a triangular prism of EGF-like domains tethered on the cell by a glycosylphosphatidylinositol (GPI) anchor at the C-terminus [5,6]. Although their specific functions are still not clear, it is known that they are essential for the survival of ookinetes in the mosquito midgut [4]. In particular, studies in P. berghei strongly suggest that P28/P25 proteins have multiple, and partially redundant functions during ookinete and oocyst development [4].
Although they share a common origin and their functions appear to overlap, preliminary studies indicate that P25 proteins are expressed earlier than P28. Specifically, P25 is expressed prior to fertilization, achieving peak synthesis in the initial hours soon after, and then most abundantly expressed on the surface of the developing zygotes and ookinetes [7]. In contrast, P28 proteins are expressed slightly later on the ookinete surface until the young oocyst stage [7]. In the context of developing TB vaccines, antibodies against these proteins interfere both with ookinete maturation and oocyst formation [8]. In particular, mice antisera against recombinant Pvs28 and Pvs25 recognized both antigens in short term cultures of parasite sexual-stages derived from patients with P. vivax malaria, and significantly suppressed oocyst development in four Anopheles species fed with blood infected with P. vivax Salvador I strain [8]. In addition, in a preclinical trial conducted in Aotus monkeys, animal immunization with recombinant Pvs25 elicited specific antibodies able to fully block parasite infection in membrane feeding assays (MFAs) [9]. Furthermore, in a phase I clinical trial conducted with Pvs25 sera obtained from the vaccinated volunteers induced significant inhibition of P. vivax transmission in Anopheles dirus mosquitoes using an ex-vivo MFA [1,10]. Moreover, TB immunity elicited with orthologous proteins in P. falciparum and other malaria parasites has been shown as well [8,11,12]. Unlike the extensive polymorphism commonly observed in several Plasmodium blood stage surface antigens, these proteins are considered to be conserved [13–17]. Therefore, the immunogenicity, TB potential, and limited polymorphism support the use of Pvs28 and Pvs25 as suitable targets for TB vaccines.
Here, we study a worldwide sample of pvs28 and pvs25 coding alleles. We detected strong geographic differentiation between populations in Asia and the Americas with replacements at specific amino acid residues novel in the Americas. We also found that these genes can be as polymorphic as some merozoite antigens such as MSP-8 and MSP-10, with most of their variation accumulating in the EGF2 and EGF3 like domains of both proteins. Finally, our analysis indicates that positive selection may be acting on the accumulation of pvs28 and pvs25 polymorphisms.
We report pvs28 and pvs25 complete CDS sequences from geographically and temporally diverse laboratory strains provided by William Collins at the Centers for Disease Control and Prevention (CDC). We obtained pvs28 gene sequences from the following laboratory strains grouped by their geographic origin: Africa (Mauritania I), Central America (El Salvador II, Honduras III, Nicaragua, and Panama), South America (Río Meta from Colombia), Asia (Vietnam II, India VII, Thailand and Malaysia), and Oceania (Sumatra from Indonesia, Indonesia XIX, Chesson, and Harris from Papua New Guinea). We also obtained 11 pvs28 and 15 pvs25 sequences from Venezuelan archived samples [18]. In addition, we included 259 pvs28 (total of 284 in the global alignment) and 310 pvs25 (total of 325 in the final alignment) sequences available at the GenBank (release 208, June 2015). Those included data of pvs25 from Venezuelan and laboratory strains [19] and sequences from pvs28 and pvs25 from China (Yunnan Province) [13], India (Delhi, Chennai, Kamrup, Nadiad and Panna) [16], Iran [17], Korea (ROK) [15], Southern Mexico [14], Thailand (Tak Province) [12] and, Bangladesh [20].
Additionally, we report 58 sequences for p28 and p25 orthologous genes from the following species of nonhuman primate parasites (NHPPs): Plasmodium cynomolgi (p28 from: Berok, B, BX-20, Cambodian, ceylonensis, PT1, PT2, RO, Smithsonian, Gombak, and Mulligan strains; p25 from: Berok, B, BX-20, Cambodian, ceylonensis, PT1, PT2, RO, and Smithsonian strains), Plasmodium inui (p28 from: Celebes I and II, Leaf Monkey II, N34, OS, Philippine, Leucosphyrus, Perak, Taiwan I and II, and Perlis; p25 from Celebes II, Hawking, Leaf Monkey I, Leucosphyrus, Mulligan, and Perlis), Plasmodium knowlesi (p28 from: H, Hackeri, Malayan from Malaysia, Nuri from India, and the Philippine strain; p25 from: Philippine, Hackeri, and Malayan), Plasmodium coatneyi, Plasmodium fieldi (p28: Hackeri and N-3; p25: ABI and N-3 from Malaysia), Plasmodium hylobati, Plasmodium simiovale (Sri Lanka), and a parasite from African primates, Plasmodium gonderi. Information about these species and strains can be found in Coatney et al. [21].
In order to estimate the phylogenetic relationships for the genes encoding pvs28 and pvs25 and their NHPPs orthologs, we also included the published sequences in PlasmoDB, version 24 [22] and NCBI for the Asian species P. cynomolgi (PCYB_062530, PCYB_062520) and P. inui (San Antonio 1, GCA_000524495.1 and; AY639974); the Laverania group that includes P. falciparum (3D7_1030900 and 3D7_10310003), Plasmodium gaboni (Pgk strain, GCA_000576715.1), and Plasmodium reichenowi (PRCDC_1030200 and PRCDC_1030300); the human parasite Plasmodium ovale (AB051632 and AB051631); and the rodent parasites Plasmodium bergei (strain Anka, AF232051 and XM_670232), Plasmodium chabaudi (AF232048 and XM_739934), and Plasmodium yoelii (AF232055 and XM_720005). The phylogenies were rooted with the avian parasite Plasmodium gallinaceum (M96886 and J04008). In the specific case of P. cynomolgi (strain B), in addition to the p28 orthologous PCYB-062530, the two other paralogous genes were also retrieved (PCYB-062510/PCYB-007100) from PlasmoDB. To the best of our knowledge, only P. ovale [23] and P. cynomolgi [24] have one and two paralogs to p28 respectively. In the specific case of P. ovale, we only included the pos28-1 (AB051632) sequence in our phylogenetic analyses.
For all the samples processed in this study, DNA was extracted from whole blood by using QIAamp DNA Blood Mini Kit (Qiagen GmbH, Hilden, Germany). All the p28 and p25 genes reported in this study were amplified by polymerase chain reaction (PCR). PCR reactions were carried out in 50 μl volume that included 1.5 mM MgCl2, 1 X PCR buffer, 1.25 mM of each deoxynucleosidetriphosphate, 0.4 mM of each primer, and 0.03 U/μM of AmpliTaqDNA polymerase (Applied Biosystems, Roche-USA). For the pvs28 gene, we used the primers 5’-TTTGTTCATTTTTGACATACTCACTT-3’ and 5’-ATGCGCGGTGTGTTATTTGGAG-3’ with an annealing temperature (Ta) of 50°C. For P. cynomolgi, P. fieldi, P. fragile, P. inui and P. simiovale, we used the primers 5’-CCAACTGCATTATACAAAAAC-3’ and 5’-ATCTTCTTCGGCGAAAAAA-3’ (Ta: 47°C). For P. knowlesi and P. hylobati, we used 5’-TGCCACCCCTTGTTCAAAATG-3’ and 5’-GWACTGACTCTGYGADACC-3’ (Ta: 54°C). In some cases, a nested PCR was required by using the primer sets 5’-ACTTGCTCACTCGACTTAACC-3’ and 5’-CGTTTTTCTTGTCCCTTTGTCAC-3’ (Ta: 53°C) for P. vivax and 5’-ATACAAAAACGACTCCCCCTTT-3’ and 5’-CGTATGACTTGAACTGACTC-3’ (Ta: 47°C) for NHPPs.
The pvs25 gene and its orthologs were amplified with the primers 5’-CTGACTTTCGTTTCACAGCA-3’ and 5’-ACATCACAAGTCCGTAAGTT-3’ (Ta: 53°C). In the case of nested PCRs, we combined the external primers 5’-CTGACTTTCGTTTCACAGCA-3’ and 5’-CATCACAAGTCGGTAAGT-3’ (Ta: 53°C) with the internal 5’-TTCGACCGCTCAATTCGCC-3’ and 5’-CAAGTCGGTAAGTTCAGTAAAG-3’ (Ta: 55°C). The amplification conditions for both genes were as follow: 5 min at 95°C, followed by 35 cycles with 1 min of denaturation at 94°C, 1 min at the specific Ta and elongation at 72°C for 2 min. After 35 cycles, a final elongation step at 72°C for 10 min was carried out. Amplified products from the two independents PCRs were either directly purified or gel extracted and cloned in pGEM-T easy Vector Systems I following the manufactory protocol (Promega, USA). For at least two clones, both strands were sequenced using an Applied Biosystems 3730 capillary sequencer. All the sequences reported in this investigation are deposited in the GenBank under the accession numbers KU285229 to KU285332.
For both genes, p28 and p25, independent alignments of their nucleotide sequences for P. vivax and their close NHPs malaria species were performed by using the MUSCLE algorithm [25] implemented in SeaView4 [26] on translated sequences followed by visual inspection and manual editing. The protein domains (signal peptide, EGF-like domains and GPI anchor) were assigned in the alignments following the description used by Saxena et al. 2006 [5]. In the case of pvs28, the low complexity regions (LCRs) were not included in the polymorphism and phylogenetic analyses; however, those were studied separately as defined by Rich et al. 1997 [27].
We estimated the polymorphism by gene and by domain within each Plasmodium species by using the population statistics π (the average number of substitutions between any two sequences), number of segregating polymorphic sites (S), and haplotype diversity (Hd). The polymorphism was also explored by computing Tajima’s D statistic [28]. The distribution of the genetic diversity across the p28 and p25 gene-sequences was described by calculating π on a sliding-window of 50 base pairs (bp) with a step size of 10 sites. The statistic was calculated in each window, assigned to the nucleotide at the midpoint of the window and plotted against the nucleotide position. All these calculations were performed using DnaSP v5.10.01 [29].
Evidence of natural selection was explored by estimating the average number of synonymous substitution per synonymous site (dS) and non-synonymous substitutions per non-synonymous site (dN) between a pair of sequences under the Nei Gojobori method [30], with the Jukes and Cantor corrections as implemented in the MEGA6 [31]. The difference between dS and dN and its standard error was estimated by using bootstrap with 1,000 pseudo-replications, as well as a two tailed codon based Z-test on the difference between dS and dN as described in Nei and Kumar 2000 [32]. Under the neutral model, synonymous substitutions accumulate faster than non-synonymous because they do not affect the parasite fitness and/or purifying selection is expected to act against nonsynonymous substitutions (dS≥dN). Conversely, if positive selection is maintaining polymorphism, a higher incidence of nonsynonymous substitutions is expected (dS<dN). We assumed as a null hypothesis that the observed polymorphism was not under selection (dS = dN).
We also used the random effects likelihood (REL) method as implemented in HyPhy, which uses flexible, but not overly parameter-rich rate distributions [33] and allows both dS and dN to vary across sites independently. REL allows for tests of selection at a single codon site while taking into consideration rate variation across synonymous sites. It is often considered the only method for inferring selection from low divergence alignments such as pvs28 and pvs25. Evidence for natural selection was also explored in P. vivax by using the McDonald & Kreitman (MK) test which compares intra and inter-specific number of synonymous and non-synonymous changes [34]. In this analysis we compared P. vivax with their close NHPPs P. cynomolgi, P. inui and P. knowlesi for both p28 and p25 genes. Significance was assessed using a Fisher’s exact test for the 2 x 2 contingency table as implemented in the DnaSP.
In order to study the genetic relationships among worldwide haplotypes, a median joining network was estimated for a set of 284 cosmopolitan sequences of pvs28 and 325 of pvs25 genes by using Network v4.6.1.0 (Fluxus Technologies 2011). Transversions were set equal to transitions and the epsilon parameter set equal to 0 with only one round of star contraction, which collapses star-like structures in the network into single nodes. The total number of sites included in these analyses excluding gaps or missing data were 547 out of 744 for pvs28 and 558 out of 660 for the pvs25 genes. In addition, we also used DnaSP to estimate the fixation index (FST) based on haplotype-frequencies among these geographical regions.
In order to investigate whether intragenic recombination generates allelic diversity in the P. vivax ookinete genes, the genetic algorithms for recombination detection (GARD) were used to screen for the recombination breakpoints in both alignments, as implemented in Datamonkey (http://www.datamonkey.org/)[35,36]. Default parameters for the detection of recombination breakpoints and donor-recipient pairs were used with a significance cut-off of 0.05.
The evolutionary relationships among the p28 and p25 genes in Plasmodium spp. were investigated using Bayesian methods implemented in MrBayes v3.2 with the default priors [37]. A General Time-Reversible model (GTR+I+Γ) was used because it had the lowest likelihood value and possessed the fewest number of parameter that best fit the data (p28 and p25) as was estimated by MEGA6. For both phylogenies (p28 and p25), two independent chains were sampled every 200 generations in runs lasting 6 × 106 Markov Chain Monte Carlo steps, and after convergence was reached, we discarded 50% of the sample as ‘burn-in’ period. Convergence is reached when the value of the potential scale reduction factor is between 1.00 and 1.02 and the average standard deviation of the posterior probability is below 0.01 [37].
Additionally, the adaptive branch-site random effects likelihood (aBSREL) approach [38], implemented in Datamonkey, was run to detect evidence of episodic positive selection on all branches using both phylogenies (p28 and p25). It allows for different Ka/Ks ratios among sites and branches. We performed a likelihood ratio test (LRT) comparing the null model (ω = 1) against the alternative, where the branch was undergoing some form of selection (ω ≠ 1). In addition, we used BUSTED, implemented also in Datamonkey, which is an approach to identify gene-wide evidence of episodic positive selection, where the non-synonymous substitution rate is transiently greater than the synonymous rate [39]. In these analyses we selected both human malarias P. vivax and P. falciparum branches because BUSTED requires pre-specified subset of lineages.
A total of 284 and 325 sequences were studied for the pvs28 and pvs25 genes respectively. Tables 1 and 2 describe the polymorphism found in the complete gene-sequences and their subsequent domains for both genes. The overall genetic diversity, as estimated by π, revealed that both genes are relatively conserved when compared to other vaccine candidates as has been reported by previous studies analyzing a smaller sample size [3,12–17]. The pvs28 gene showed a slightly higher, but not significant polymorphism level (π = 0.0037 ± 0.0011) than pvs25 (π = 0.0023 ± 0.0010). When we compared the polymorphism observed between isolates from Asia and the Americas, we found that pvs28 samples from Asia were slightly more polymorphic (π = 0.0035 ± 0.0012) than the ones from the Americas (π = 0.0024 ± 0.0011) (Table 1). In contrast, we observed a similar genetic diversity in the pvs25 sequences from both Asia (π = 0.0017 ± 0.0008) and the Americas (π = 0.0018 ± 0.0010) (Table 2). Nevertheless, in both genes the standard errors for π overlapped when Asia and the Americas were compared.
Because of P28 and P25 EGF-like domains carry critical epitopes recognized by P. vivax TB antibodies [5,8], we also estimated π by gene-domain in addition to the geographic regions. In both genes, the EGF2 and EGF3 were the most polymorphic domains of the proteins (Tables 1 and 2). Fig 1 shows the distribution of the polymorphism across the pvs28 and pvs25 genes using a sliding window approach of the nucleotide diversity π. The polymorphism for both genes was distributed unevenly; the most conserved areas were located at the secretory signal sequence at the N-terminus, the EGF1 and EGF4 like domains, and the GPI anchor at the C-terminus. In contrast, central regions like the EGF2 and EGF3 domains accumulated higher variability.
We also studied the polymorphism found in pvs28 (Table 3) and pvs25 (Table 4) genes by country. Regardless of sampling differences, the haplotype diversity Hd was similar for both genes (pvs28-Hd = 0.765, pvs25-Hd = 0.724). Yet, the number of haplotypes found in pvs28 (H = 53) seems to be slightly higher than the ones found in the pvs25 gene (H = 35). We report estimates by country with at least 10 sequences or more in our alignment. A high haplotype diversity was observed for pvs28 in Bangladesh (Hd = 0.895), Thailand (Hd = 0.993) for Asia, and Venezuela (Hd = 0.873) from the Americas (Table 3). For pvs25, high haplotype diversity was observed in samples from China (Hd = 0.6478), Venezuela (Hd = 0.8000), and Mexico (Hd = 0.6513) (Table 4).
In order to explore how natural selection was involved in the maintenance of the observed polymorphism, we estimated the average number of synonymous (dS) and non-synonymous (dN) changes between two sequences (Tables 1–4). Overall, we found a significant excess of non-synonymous over synonymous polymorphic changes in pvs28 sequences (p = 0.0271, Table 1). Nevertheless, when we estimated the average dS and dN by region, this pattern was maintained in Asia (p = 0.0318) but not in the Americas (p = 0.4926), specifically in Thailand (p = 0.0280), India (p = 0.0090), and Bangladesh (p = 0.0238). Although there was an excess of non-synonymous (dN = 0.0027) over synonymous substitutions (dS = 0.0007) in pvs25 polymorphism (Table 2 and Table 4), the differences were not significant (p = 0.1040) (Table 2).
We examined the amino acid replacements observed in the P28 and P25 proteins by using P. vivax Salvador I strain as a reference; our observations are summarized in S1 and S2 Tables respectively. In the Pvs28 protein, we observed 44 amino acid changes, most of them in low frequency (<0.1% on 284 sequences). The EGF1 domain had the lowest number of amino acid replacements with only the replacement 52(M/L) found in 25% (S1 Table). In contrast, the EGF2 domain showed the highest number of changes (12 replacements) but most of them observed in low frequency (<1%). The most frequent replacements were at 65(T/K) found in 21.8% of the sequences, followed by 87(D/N) and 98(L/I) with 8.5% and 4.2% frequency respectively. The EGF3 domain displayed a total of 13 amino acid changes, the most common were in positions 110(N/Y) found in 5.6%, 116(L/V) in 10.9% and 140(T/S) in 16.2% of the total samples. The EGF4 domain (no including LCR of tandem repeats) and the GPI anchor region showed together only 14 changes in low frequency (<3.0%). It is important to emphasize that the polymorphisms in positions 87(D/N) and 110(N/Y) were observed only in the 40 sequences from the Americas with a frequency of 60% and 40% respectively. Here, we report these polymorphisms for the first time in Colombia, El Salvador (Sal II, [40]), Honduras, Nicaragua, Panama and Venezuela. Moreover, some of the frequent amino acid changes observed in Pvs28 (52M/L, 65T/K and 116L/V) were also present in P. cynomolgi and P. inui (S1 Table).
A similar analysis for the Pvs25 antigen (n = 325) showed a total of 34 low frequency (<1%) amino acid substitutions (S2 Table). The most frequent changes for the EGF2 were 87(Q/K) found in 12.7% and 97(E/Q) in 50.3% of the worldwide sequences. Glutamine (Q) at position 87 is one of the contacting residues involved in the binding of the transmission blocking antibody 2A8 (Fab VH domain) with the ß loop (EGF2) of the Pvs25 protein [5]. This amino acid was found to be mutated to lysine (K) in several field isolates from Iran, Mauritania, Brazil, Colombia, Mexico and Venezuela. In the EGF3 domain, changes at positions 130(I/T) in 89.1% of the sequences in addition to 131(Q/K) in 8.2% were the most common. In the case of p25 orthologous genes in close NHP malarias, different amino acid changes were also identified in all these positions (S2 Table). To show the location of the observed mutations on the Pvs28 and Pvs25 structures, the three dimensional structure for Salvador I Pvs28 was modelled by using Phyre2 [41] on the Pvs25 structure as template [5]; 65% of the Pvs28 structure was modelled with 99.4% confidence. Positions of mutations for both Pvs25 and Pvs28 were visualized using Visual Molecular Dynamics (VMD [42]). Mutations with a frequency >1% were mapped by residue location and colored according to domain (Fig 2, S1 and S2 Tables). Residues putatively under positive selection were indicated with arrows (see results from REL method explained later in the text).
The haplotype networks based on the pvs28 and pvs25 sequences are shown in Figs 3 and 4 respectively. We identified 63 distinct haplotypes among 284 pvs28 sequences from 18 regions/countries. Although the sampling effort per country did not allow us to reliably estimate and compare the haplotypes’ relative frequencies, there were some emerging patterns. In particular, we focused on the number of countries/areas where a given haplotype had been found since it is informative of its geographic range.
The pvs28 network presented two distinctive features referred here to as A and B. The feature A suggests a star-like shape consistent with an expansion of the P. vivax population for part of the network [18] while the feature B refers to the reticulated structure observed in Asia. Only 12 haplotypes (19.1%) were shared by two countries or more whereas 51 haplotypes (81.0%) were restricted to single countries. Importantly, only three haplotypes were found with a relatively broad distribution (H1, H2, H35; see Fig 3).
The haplotype denominated as H1 was the most frequent (40.1%, 114/284) and showed a worldwide distribution (Fig 3). The haplotype H35 with 59 sequences (20.8%) was the second most predominant and the most common in the Indian samples included in this study (77.3%). It was not only found in five distant Indian regions [17] but also in Bangladesh, China, Iran, and Thailand (Fig 3). The third haplotype in terms of its frequency was H2 (4.6%, 13/284) and belongs to a more divergent cluster which includes only samples from the Americas; specifically, Mexico, Colombia, El Salvador (Sal II strain), Nicaragua, Panama, and Venezuela. The network results suggest that the haplotype H2 could have originated from the most frequent H1 haplotype. Because of the study performed in Korea, which is geographically smaller, involved a large sample collected between 1996 and 2007 [15], we can speculate that haplotype H1 might be the most common in that region. Feature B of the pvs28 network (Fig 3) showed a group of haplotypes from Bangladesh, China, India, Malaysia, Thailand, and Vietnam forming reticulations. This pattern corresponds to several divergent haplotypes found in very low frequency in this set of samples.
The pvs25 haplotype network depicts 35 distinct haplotypes among 325 sequences from 15 regions/countries. We found five haplotypes in high frequency (H1, H4, H20, H23, and H25; see Fig 4). The haplotype denominated as H4 corresponded to 152 (46.8%) sequences from Bangladesh, China, India, Indonesia, Iran, Korea, and Thailand. This haplotype is related to H1, the second most predominant with 70 (21.5%) sequences distributed in China, India, Korea, Mexico, and Thailand. The other three haplotypes (H20, H23 and H25) were linked to the most frequent H1 and H4 by long branches. Finally, haplotype 25 was only found in Mexico and Venezuela.
To further determine genetic differentiation among populations, the FST fixation index was estimated. Supporting our median joining network results, FST values estimated for both genes suggest high genetic differentiation among P. vivax ookinete genes in different regions (FST > 0.15, Tables 5 and 6). Pairwise comparisons between Venezuela and Asia regions (China, Korea, India, Thailand, and Bangladesh), produced high FST values for both genes ranging from 0.426 to 0.542 in pvs28 (Table 5) and from 0.457 to 0.748 for the pvs25 gene (Table 6). As expected, a similar pattern was observed in pairwise comparisons between Mexico and Asia regions in both genes, suggesting some degree of differentiation between Asia and the Americas populations. However, when Mexico and Venezuela populations were compared, high FST values were also observed in the pvs28 (0.251) and pvs25 (0.457) coding genes. In contrast, P. vivax populations from Bangladesh compared to China and Thailand, are consistent with a minimal genetic divergence, suggesting no genetic population structure among these regions for pvs28 (FST <0.05). Tajima’s D produced consistent negative values for both pvs28 (Table 1) and pvs25 (Table 2) genes for all populations. In most cases, the results of the test were statistically significant with the exception of the American populations.
In order to investigate if recombination generated allelic diversity, the genetic algorithm recombination detection (GARD [35]) was performed. No evidence of intragenic recombination was detected in these ookinete genes.
In accordance with previous reports, the amino acid sequence alignments of the P28 and P25 proteins suggest that both are conserved among Plasmodium spp. (S1 Fig) [2,5,6,23]. All the p28 and p25 sequences included in this study have a conserved hydrophobic signal sequence at the N-terminus (residues 1–23, SignalP 4.1 Server) [43], followed by four cysteine-rich epidermal growth factor EGF-like domains and a short GPI anchor region at the C-terminus (S1 Fig). An invariable number of 20 (~9.7%) and 22 (~10%) cysteine residues were found in all NHPPs P28 and P25 proteins respectively. The EGF4-like domain in P28 proteins contains four rather than six cysteines lacking of the 5–6 disulfide bridge. P28 orthologs showed a high average content of Lys (~7.50%), Leu (~7.23%), Asn (~8.92%), Thr (~7.79%) and Val (~7.47%) (S2A Fig). Likewise, for P25 proteins we found an average content of Glu (~9.25), Gly (~6.86), Lys (~9.67), Leu (~7.54) and Val (~8.32) (S2B Fig).
We compared the polymorphism of pvs28 and pvs25 with their orthologs in P. cynomolgi, P. inui and P. knowlesi. S3 and S4 Tables show the genetic variation found in the coding sequence (CDS) and the different domains of p28 and p25 genes respectively. We found that P. cynomolgi (P28-π = 0.0340 ± 0.0049, P25-π = 0.0284 ± 0.0041) and P. inui (P28-π = 0.0400 ± 0.0045, P25-π = 0.0133 ± 0.0026) had higher genetic polymorphism than their orthologs in P. vivax. In contrast, the p28 and p25 polymorphisms observed in P. knowlesi (P28-π = 0.0023 ± 0.0011, P25-π = 0.0038 ± 0.0015) were similar to pvs28 and pvs25 (Tables 1 and 2). The P. knowlesi orthologs also have shown no polymorphism in most of the gene domains (S3 and S4 Tables). Although, the P. cynomolgi p28 paralog PCYB_007100 had similar genetic diversity (π = 0.0340 ± 0.0044) to the one considered the P. cynomolgi ortholog to pvs28, PCYB_062530 (π = 0.0340 ± 0.0049), the nucleotide diversity was different across both genes (PCYB_062530 and PCYB_007100, S3B Fig).
We estimated the average pairwise dS and dN for NHPPs orthologs to p28 and p25. In the case of p28, especially for P. cynomolgi, the diversity found in both paralogous genes was biased toward synonymous sites, a pattern consistent with purifying selection (S3 Table). This contrasts with the pattern of positive selection found in pvs28. Nevertheless, a similar pattern was found in P. inui. Although there was an excess of synonymous over non-synonymous polymorphisms in P. cynomolgi and P. knowlesi p25 CDS, the differences were not significant using the codon based Z-test (S4 Table). Interesting, the dS-dN estimated by domains suggests different patterns of selection acting in the EGF1 (negative selection) and EGF3 (positive selection) like domains in P. cynomolgi (p < 0.05, S4 Table). Again, a similar number of synonymous (dS) and non-synonymous (dN) substitutions was found for p25 in P. inui (S4 Table). The assumption of neutrality was further examined in pvs28 and pvs25 against their orthologs in P. cynomolgi and P. inui by using the MK test. This test showed an excess of nonsynonymous over synonymous polymorphisms in the pvs25 gene when divergence was compared in P. cynomolgi and P. inui (p < 0.05, S5 Table). In both cases, the neutrality indexes (NI) were bigger than 1 and the significance of the test was explained by an excess of replacement polymorphisms in pvs25. These results suggest a possible pattern of balancing selection since a preponderance of non-synonymous intra-species polymorphisms was observed. Similar trends but no significant departures from neutrality were found for pvs28 (S5 Table).
The results of the REL method are depicted in Fig 5 (see S1 Table for reference) and the estimated Bayes Factors (BF) summarize the evidence provided by the data in favor of positive selection. The method detected three codons in pvs28 where the data provided strong evidence for positive selection with BFs bigger than 50 (codon 52 with a BF of 97.69, codon 113 with a BF of 101.51, and codon 116 with a BF of 367.92; see S1 Table for reference) and five with some evidence for positive selection with BFs bigger than 10 but less than 50 (codons 53, 65, 95, 98, and 123; S1 Table). In the case of pvs25, we found only five codons (35, 97, 130, 132, and 135; see S2 Table for reference) where the data provided some evidence of those being under positive selection with BFs bigger than 10 but less than 50 (Fig 5). Residues with mutations in high frequency (>1%) were mapped on the pvs25 and pvs28 protein structures depicted in Fig 2.
The Bayesian phylogenies of p28 and p25 are depicted in Fig 6. The avian malarial parasite P. gallinaceum was included as outgroup. Overall, they are comparable with previous published phylogenies using nuclear and mitochondrial genes [19,44–47]. Briefly, in both phylogenies, three major clades were identified: 1) the Laverania subgenus, 2) the clade of rodent malarias, and 3) the P. vivax clade. Plasmodium vivax is part of a monophyletic group with closely related NHPPs found in Southeast Asia. The African primate parasite P. gonderi was consistently placed at the base of this monophyletic group in both phylogenies. The difference between p28 and p25 phylogenies was the relative position of P. ovale (Fig 6). The p28 phylogeny resembled the phylogenetic tree obtained previously based on the nuclear genes ß-tubulin, CDC-2 and the plastid gene tufA [19].
Two additional phylogenies containing all p28 and p25 strains obtained in this study for P. cynomolgi, P. inui, and P. knowlesi were estimated using P. gonderi as outgroup (S4A and S4B Fig respectively). The three p28 paralogs found in P. cynomolgi genome (B strain, PlasmoDB) were included (S4A Fig). The p28 phylogeny was slightly different to one that included all species; however, the major relationships were maintained (e. g. P. inui-P. hylobati, P. fieldi-P. simiovale and P. knowlesi-P. coatneyi). We could amplify the three different copies only for the P. cynomolgi strain Berok (S4A Fig), thus we could not confirm that all the P. cynomolgi strains have the two recent paralogs. Noteworthy, all p28 P. cynomolgi paralogs formed a monophyletic group that included the ortholog (PCYB-062530) to pvs25. This suggests duplication events in P. cynomolgi that took place after divergence from the common ancestor shared with P. vivax. To the best of our knowledge, only P. ovale and P. cynomolgi have reported paralogs to the p28 gene. However, we cannot rule out that such duplication events have occurred in others Plasmodium spp. In the case of p25 (S4B Fig), the relationship among NHP malarias from Southeast Asia were the same as those obtained in the phylogeny containing all the species (Fig 6).
Phylogenetic-based methods were used to explore the role that positive selection may have played in the divergence of these two loci. In the case of p28, no evidence of episodic diversifying selection was found in any of the 31 total branches using aBSREL (p ≤ 0.05, corrected for multiple testing). However, BUSTED revealed evidence for positive selection acting only on the P. falciparum lineage (p = 0.002, S6 Table). In contrast, no evidence of episodic diversifying selection was found in the p25 gene using aBSREL and BUSTED (S6 Table).
In order to fully characterize p28 polymorphism, we examined the LCR of short tandem repeats located at the big C-loop of the EGF4 domain [6]. The description of P28 motifs and amino acid variation is summarized in S7 and S8 Tables. This LCR is also present in all NHPPs that form part of the monophyletic group with P. vivax including P. gonderi from Africa and the Pos28-1 of P. ovale (S7 Table, S1 Fig). In contrast, such LCR is almost absent in species belonging to the Laverania clade (P. falciparum and related species) and rodent malarias with the exception of P. yoelii (S7 Table, S1 Fig). In the case of P. vivax, the consensus tandem repeat unit consists of five amino acid motif (GSGGE). The pattern from all P. vivax sequences can be summarized as [(G/E/S)S(G/R/D)GE]2–6, where the first and third positions are polymorphic (S7 and S8 Tables). Interestingly, all the amino acid changes were observed in Asia. The last tandem unit was not a repetitive motif, showing aspartic acid (D) in high frequency at the fifth position, but lower for glutamic acid (E) and glycine (G): [(G/S)S(G/D)G(D/E/G)]. This terminal unit was not included in our polymorphism estimations. Noteworthy, glycine was also the most abundant amino acid in the LCR in all NHP malarias included here (S7 Table).
Since proteins domains containing LCRs might be natural immunogenic carrying possible targets for antigenic epitopes [48], we explored the role of natural selection acting on the observed polymorphism. When the repetitive motifs were aligned among them, a significant (p < 0.05) excess of synonymous over non-synonymous substitutions was observed in P. vivax and NHP malarias (S8 Table), suggesting that the motif is conserved and its sequence might be under purifying selection.
Although pvs28 shows slightly higher polymorphism than pvs25, those differences appear not to be significant. The genetic diversity found in sequences from Asia and the Americas for both genes was similar. This pattern was observed even when there were fewer sequences from the Americas than Asia. This is consistent with studies based on mitochondrial genome sequences (potentially neutral loci) and complete genomes showing that the diversity of P. vivax population in the Americas is comparable to Asia [18,49].
Pvs28 and pvs25 showed higher genetic variability compared to other sexual stage TB antigens reported in P. vivax as pvs48/45 (π = 0,00053), the Willebrand factor A domain-related protein (WARP) (π = 0.00010) and also previous estimations of pvs25 (π = 0.00065) and pvs28 (π = 0.00000) in Korea [50]. The Korean study likely differs from ours because of its limited geographic scope. It is worth noticing that whereas the observed polymorphism is lower than those reported in many merozoite surface antigens such as AMA-1 [51], there are merozoite stage antigens such as MSP-8 and MSP-10 with similar levels of polymorphism to those reported here for the pvs28 and pvs25 genes [45].
The neighbor haplotype-network for both pvs28 and pvs25 genes formed a star-like shape consistent with the suggested underlying demographic history of a population expansion for P. vivax [18]. This could also explain the significant and negative Tajima’s D estimated for the gene (Tables 1 and 2). The low global proportion of haplotypes shared between countries for both genes suggests substantial genetic differentiation among P. vivax populations, as confirmed by high FST values (Tables 5 and 6). We also observed some degree of geographic clustering for haplotypes from the Americas; specifically, a divergent clade for the pvs28 gene characterized by the replacements located at the positions 87(D/N) and 110(N/Y) that were only found in the Americas (S1 Table). Both networks suggest that some of the haplotypes from the Americas could be derived from Asian populations [52,53]; however, the pattern is consistent with previous finding indicating that there was not a recent or single introduction of P. vivax into the continent [18].
We performed a comparative polymorphism analysis between pvs25 and pvs28 and their orthologous genes in the Asian Old World monkey parasites that are closely related: P. cynomolgi, P. inui and P. knowlesi. In contrast to the relatively low genetic diversity observed in P. vivax and P. knowlesi, the orthologs in P. cynomolgi and P. inui exhibited significantly higher variability. Similar observations have been also reported for genes expressed in asexual Plasmodium stages [19,45]. This pattern could be the result of the different demographic histories of these two parasites when compared to P. vivax and P. knowlesi. Consistently with the effect of demographic differences, the same pattern has been found in the mtDNA and other genes that are considered neutral [19, 54].
Interestingly, both Pvs28 and Pvs25 proteins showed higher variation at the EGF2 and EGF3 like domains where epitope recognition sites have been identified for blocking antibodies in Pvs25 [5], and predicted for Pvs28 [6] and Pb28 in P. berghei [55]. Noteworthy, EGF-like domains in the orthologous protein Pfs25 have shown differential immune blocking activity after being separately expressed as a yeast-secreted recombinant protein. In particular, antibodies against the EGF2 domain elicited the strongest blocking activity indicating that this domain might be a good target for TBVs [56].
The EGF-like domains in Plasmodium spp. are relatively conserved among genes and closely related species (S1 Fig). A similar pattern has been described in other EGF-like domains expressed in surface proteins from the merozoite, including MSP-4 [57], MSP-5 [58], MSP-8 and MSP-10 [45]. When we estimated the genetic diversity in P. cynomolgi, P. inui, and P. vivax, we observed regions with relatively high polymorphism in EGF2 and EGF3 in both genes (S3 Fig). How this variation affects protein folding and functionality is a matter that remains elusive. However, it has been proposed that EGFs domains can accommodate genetic changes such as gene polymorphism, mutations, insertions and deletions [55]. Consequently, structural folds in the P28 and P25 proteins may not be significantly affected by the observed amino acid changes in natural populations thereby preserving functionality.
Previous investigations suggested that the p28 and p25 coding genes were originated as result of a gene duplication event that was prior to the origin of the species included in this investigation [4]. When a duplicated gene neither adapts to a more specialized function nor is silenced by deleterious mutations and continues producing a functional protein, purifying selection could act on both paralogs keeping some level of functional redundancy [59]. Consistently, gene knockouts of either p25 (P25Sko) or p28 (P28Sko) alone in P. berghei have non-significant effects on oocyst production in infected Anopheles stephensi mosquitoes. However, concomitant disruption of both genes (Dko) strongly inhibited oocyst production up to 99% [4].
It is worth noting that duplication events have been reported for p28 in P. ovale [23] and P. cynomolgi [24] (confirmed here in the Berok strain, S4 Fig). Furthermore, in the case of P. cynomolgi, we found evidence of an excess of synonymous over nonsynonymous substitutions in the p28 paralogous gene PCYB-007100 and PCYB-062530 suggesting purifying selection (S3 Table, S4 Fig). Thus, without evidence indicating pseudogenization and patterns consistent with purifying selection, we can speculate that both p28 paralogous remain functional in P. cynomolgi.
We searched for evidence of episodic selection as a consequence of changes in ecology/vectors during the evolution of the species include in this study; however, we did not find it. Only the branch leading to P. falciparum indicated positive selection in p28, a finding that is worth exploring whenever additional Laverania species become available. We also explored the effect of selection on the pvs28 and pvs25 polymorphisms by performing the MK test and applying codon models such as REL. Their caveat is that these tests usually underperform even when adaptive evolution is present so they are regarded as conservative [60]. The MK test detected evidence for balancing selection in pvs25. A similar pattern of synonymous/non-synonymous sites within P. vivax and its divergence to P. cynomolgi and P. inui was found for pvs28 (see S5 Table), but not significant (p > 0.05). The Bayesian base method (REL), however, detected codons under selection in pvs28. In particular, the data provided very strong evidence for selection on three pvs28 codons with two of those codons, 113 and 116 (Fig 2), yielding BF factors above 100. These two codons are located in the EGF3 domain. We also found other codons in pvs28 and pvs25 where the data provided some evidence of those being under positive selection but their BF did not exceed our 50 threshold defined a priori. Those residues are indicated with yellow arrows in Fig 2.
The patterns consistent with positive selection acting on the pvs28 and pvs25 polymorphism deserve special attention. Whereas the genetic polymorphism observed on surface antigens from the asexual stage has been commonly associated to the selective pressure exerted by the vertebrate immune system [61,62], proteins expressed in the sexual stage may adapt to diverse microenvironments inside Anopheles mosquitoes where parasites have to go through in order to complete their life cycle [63]. The fact that anti-Pvs28 and anti-Pvs25 polyclonal antibodies completely block parasite transmission (Pv-Sal I) in four species of Anopheles mosquitoes [8] indicates that these proteins are essential during this phase of the parasite life-cycle. Whether pvs25 and pvs28 facilitate the Plasmodium transit through Anopheles physical barriers and by so doing, increase the parasite (and may be the vector) fitness is a matter that needs to be investigated [64,65].
The evolutionary and functional implications of LCR in P28 proteins are still elusive. In the case of asexual Plasmodium stages, they may have a role interacting with the host adaptive immune system [66,67]. However, such adaptive immune responses are absent in Anopheles vectors with the exception of some components from the vertebrate immune system contained in the blood bolus. The P28 LCR has been predicted to be part of a big C-loop, a fast evolving region forming a sheet over the ookinete surface that may affect the binding properties of the protein [6]. Furthermore, other studies have found that terminal LCR, like the one observed in P28, may confer more protein binding capacity [68]. This evidence suggests that the P28 LCR is functionally important. This possibility finds also support on the significant excess of synonymous over nonsynonymous changes on the motifs of most of the NHPPs P28 studied (p >0.05) (S7 Table), which indicate evolutionary constrains and not simply conservation from continued homogenization due to gene conversion. Nevertheless, assessing the importance of the LCR on P28 requires experimental evidence that is not currently available.
In summary, we explored the genetic polymorphism of pvs28 and pvs25, and investigated the long term evolution of the genes encoding these antigens within the genus Plasmodium. Although they were less diverse than many pre-erythrocytic and erythrocytic stage expressed antigens; their polymorphisms were comparable to others such as MSP-8 and MSP-10. We also found that these genes exhibit comparable diversities in the Americas and in Asia indicating that the use of TBVs against Pvs28 and Pvs25 will likely face similar challenges in both regions. Furthermore, we found two amino acid replacements in Pvs28 (positions 87(D/N) and 110(N/Y)) that appear to be specific for the Americas. Finally, there are polymorphisms that could be maintained by positive selection in both genes and the importance of such observation deserves to be explored. The observation that anti-Pvs28 and anti-Pvs25 polyclonal antibodies can block parasite transmission in some species of Anopheles mosquitoes [8] indicates that polymorphism in these proteins could indeed affect the parasite fitness. In particular, pvs25 and pvs28 polymorphisms could be the result of differences in vectors acting as selective pressure in some ecological contexts. Consequently, it is possible that a vaccine elicited transmission blocking immune response may not be equally effective across all vector-parasite associations in all epidemiological settings. In this context, exploring the diversity of local alleles and their interactions with specific Anopheles species could provide useful information on how to assess TBV efficacy, as well as, how to better deploy these vaccines, even partially effective ones, in the context of malaria control and elimination.
|
10.1371/journal.pntd.0003949 | Levels of 8-OxodG Predict Hepatobiliary Pathology in Opisthorchis viverrini Endemic Settings in Thailand | Opisthorchis viverrini is distinct among helminth infections as it drives a chronic inflammatory response in the intrahepatic bile duct that progresses from advanced periductal fibrosis (APF) to cholangiocarcinoma (CCA). Extensive research shows that oxidative stress (OS) plays a critical role in the transition from chronic O. viverrini infection to CCA. OS also results in the excision of a modified DNA lesion (8-oxodG) into urine, the levels of which can be detected by immunoassay. Herein, we measured concentrations of urine 8-oxodG by immunoassay from the following four groups in the Khon Kaen Cancer Cohort study: (1) O. viverrini negative individuals, (2) O. viverrini positive individuals with no APF as determined by abdominal ultrasound, (3) O. viverrini positive individuals with APF as determined by abdominal ultrasound, and (4) O. viverrini induced cases of CCA. A logistic regression model was used to evaluate the utility of creatinine-adjusted urinary 8-oxodG among these groups, along with demographic, behavioral, and immunological risk factors. Receiver operating characteristic (ROC) curve analysis was used to evaluate the predictive accuracy of urinary 8-oxodG for APF and CCA. Elevated concentrations of 8-oxodG in urine positively associated with APF and CCA in a strongly dose-dependent manner. Urinary 8-oxodG concentrations also accurately predicted whether an individual presented with APF or CCA compared to O. viverrini infected individuals without these pathologies. In conclusion, urinary 8-oxodG is a robust ‘candidate’ biomarker of the progression of APF and CCA from chronic opisthorchiasis, which is indicative of the critical role that OS plays in both of these advanced hepatobiliary pathologies. The findings also confirm our previous observations that severe liver pathology occurs early and asymptomatically in residents of O. viverrini endemic regions, where individuals are infected for years (often decades) with this food-borne pathogen. These findings also contribute to an expanding literature on 8-oxodG in an easily accessible bodily fluid (e.g., urine) as a biomarker in the multistage process of inflammation, fibrogenesis, and infection-induced cancer.
| Opisthorchis viverrini is a food-borne helminth infection that drives a strong inflammatory response in the bile duct that can result in bile duct fibrosis and bile duct cancer (intrahepatic cholangiocarcinoma). Extensive research shows that oxidative stress (OS) plays a critical role in chronic O. viverrini infection transitioning to cancer in the bile duct. OS also results in a modified DNA lesion, referred to as 8-oxodG, excreted in the urine, where it can be detected by an antibody-based test. We measured the concentrations of 8-oxodG in the urine of O. viverrini-infected individuals who had developed bile duct fibrosis or bile duct cancer and compared levels of this metabolite in urine to O. viverrini infected individuals who did not have bile duct fibrosis or cancer in Northeastern Thailand. We determined bile duct fibrosis by ultrasonography and bile duct cancer by immunohistochemistry on resected liver tissue. We then built a statistical model to quantify how well urinary 8-oxodG predicted bile duct fibrosis and bile duct cancer in O. viverrini-infected individuals. We found that individuals with elevated levels of 8-oxodG in urine had a greater probability of developing bile duct fibrosis or bile duct cancer from O. viverrini infection. This association occurred in a strongly dose-dependent manner: in other words, the O. viverrini-infected individuals who had the highest concentration of urinary 8-oxodG also had the highest risk of presenting with bile duct fibrosis or bile duct cancer. In summary, measuring levels of 8-oxodG in the urine offers a unique opportunity to develop a candidate biomarker for advanced O. viverrini induced hepatobiliary pathologies such as fibrosis and cancer. The findings also confirm our previous observations that severe liver pathology occurs early and asymptomatically in residents of O. viverrini endemic regions, where individuals are infected for years (often decades) with this food-borne neglected tropical diseases (NTD) pathogen.
| Over 750 million people (10% of the human population) are at risk of infection with food-borne trematodes, with more than 40 million people currently infected with one of three of these parasites: Clonorchis sinensis, Opisthorchis felineus, and Opisthorchis viverrini [1, 2]. O. viverrini is considered the most important of these food-borne trematodes due to its well-documented association with hepatobiliary pathologies that include advanced periductal fibrosis (APF) [3, 4] and intrahepatic cholangiocarcinoma (CCA) [5–10]. In Northeastern Thailand (Isaan), uncooked cyprinoid fish, which is the intermediate host for the parasite, are a staple of the diet, with O. viverrini infecting an estimated 10 million people in Isaan alone [8]. While infection with O. viverrini can be eliminated by chemotherapy (praziquantel), regional culinary practices result in rapid re-infection after treatment, often leading to life-long infection with the parasite [5, 7, 11] and the highest incidence of CCA in the world (85 per 100,000) [7].
In our community-based ultrasound studies in Northeastern Thailand [8, 9, 12, 13], we have identified a series of pathologic changes that occur early and asymptomatically in the bile duct in individuals resident in O. viverrini endemic areas. As individuals can be infected with O. viverrini for years (even decades), we hypothesize that a chronic cycle of tissue damage and repair ensues in the intrahepatic biliary ducts as a result of the constant immunological, mechanical and oxidative damage from the parasite, resulting in a persistent “smoldering and chronic inflammatory milieu” [14]. These processes stimulate the production of desmoplastic stroma (i.e., bile duct fibrosis), which has recently been shown to play a crucial role in promoting malignant transformation to CCA (see [15]). In both the hamster and human models of O. viverrini infection, fibrosis in the biliary epithelia routinely precedes CCA [8, 10, 16, 17].
The exact mechanism by which stromal desmoplasia (bile duct fibrosis) transforms to CCA is a topic of intense research [15, 17–19]. There is some consensus that an important component in this transformation is the genomic instability that accompanies both fibrogenesis and carcinogenesis[20, 21]. During these processes, cells are recruited to the site of damage and induce a “respiratory burst” from an increased uptake of oxygen, with the accumulation of reactive oxygen species (ROS) referred to as oxidative stress (OS) (see [22]). Both fibrotic [16] and neoplastic transformation [20, 23] have been linked to increased levels of OS by several mechanisms, including DNA damage, genomic instability, and cellular proliferation, The DNA base modifications caused by OS also result in oxidation of guanine residues to 8-oxo-7,8-dihydro-2′-deoxyguanosine (8-oxodG), which are excised into bodily fluids, such as urine, blood and saliva. Augmented 8-oxodG levels in urine have been used as a biomarker for oxidative DNA damage [20, 21, 24–28] in acute lymphoid leukemia, colorectal cancer, high grade cervical dysplasia, scleroderma fibrosis, liver fibrosis, renal cell carcinoma, lung cancers, and prostate cancer [15, 22, 29, 30].
Elevated levels of 8-oxodG have been reported in the urine of individuals chronically infected with O. viverrini and in the urine of individuals with O. viverrini-induced CCA [31–33]. However, to our knowledge, there have been no studies that measure levels of urine 8-oxodG during the interval from initial infection with O. viverrini to neoplastic transformation to CCA, a progression which occurs over years and proceeds through several well-defined hepatobiliary pathologies [12]. The most precisely detected and well documented of these hepatobiliary pathologies is APF [9]. Our community-based ultrasound studies (i.e., Khon Kaen Cancer Cohort or the KKCC) along the Chi River, Khon Kaen Province, Thailand have shown that APF is prevalent among otherwise apparently healthy residents in O. viverrini endemic areas [5, 8–10, 13]. The objective of the current manuscript was to determine if levels of 8-oxodG in urine increased when O. viverrini infection transitioned to APF [34] and if elevated levels of urine 8-oxodG associated with APF were comparable to the markedly elevated levels of urine 8-oxodG observed in individuals with O. viverrini-induced CCA [17, 28, 35]. As APF is a precursor stage to CCA, a threshold concentration of urinary 8-oxodG could then be used to risk stratify APF individuals into those more likely to develop O. viverrini-induced CCA and, as such, take advantage of recent therapeutic advances used to target bile duct fibrosis and the rich milieu it provides for neoplastic transformation to CCA [36]. More specifically, our objectives were as follows: (a) quantify the presence of urinary 8-oxodG in O. viverrini infected individuals with APF compared to O. viverrini-infected individuals without APF and O. viverrini-individuals with CCA and (b) determine the performance of urine 8-oxodG in predicting APF, including its sensitivity, specificity, and ability to predict diagnostic risk (e.g. odds ratios). These studies form the “discovery phase” for a potentially critical and easily accessible candidate biomarker that could be advanced to biomarker validation in large-scale studies in Thailand and in the neighboring countries of the Mekong River Basin, where O. viverrini-induced CCA has the highest incidence in the world [7, 12].
The participants provided written informed consent using forms approved by the Ethics Committee of Khon Kaen University School of Medicine, Khon Kaen, Thailand (reference number HE480528) and the Institutional Review Board (IRB) of the George Washington University School of Medicine, Washington, D.C (GWUMC IRB# 020864). The urine donated by participants from Group 4 was obtained from the biological specimen repository of the Liver Fluke and Cholangiocarcinoma Research Center, Khon Kaen University, Thailand using a protocol approved by the Ethical Committee on Human Research, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand (reference Nos. HE450525 and HE531061).
This study uses baseline data from a recently enrolled village in the Khon Kaen Cancer Cohort (KKCC), which samples villages along the Chi River Basin, in Khon Kaen province, Thailand. A detailed description of the KKCC and the methods used to assemble this cohort have been extensively reported [8–10, 13]. A modification to the original KKCC protocol was approved by the Division of Microbiology and Infectious Diseases (DMID) of the US National Institutes of Health, the Ethics Committee of Khon Kaen University School of Medicine, and the GWU IRB, allowing urine collection, which initiated in 2012 (Fig 1). The current sample of 221 individuals represents the most recent enrollment into the KKCC following this modified protocol (Fig 1). Participant inclusion criteria consisted of enrolling all males and females between 20 and 60 years of age (inclusive) registered with the village health outpost who were willing to participate in the study as evidenced by signing the informed consent form. Participants were excluded from the study if they attended school or worked full-time outside of the village (n = 0) or had a positive urine β-hCG pregnancy test in the case of females (n = 0). Individuals infected with O. viverrini were referred to the local public health clinic for treatment with praziquantel (PZQ) regardless of their participation in the study.
The data for the 181 KKCC participants were stratified during analyses to remove thirty two (n = 32) individuals whose feces contained eggs or larvae from other helminths endemic to the region, including hookworm (Necator americanus), Ascaris lumbricoides, or Strongyloides stercoralis (Fig 1). This reduced the study sample to 149 individuals assigned to three groups as follows: Group 1 (n = 23) individuals negative (uninfected) upon fecal examination for O. viverrini (also referred to as Endemic Controls or EN); Group 2 (n = 48) O. viverrini positive (OV+) individuals who were negative for APF as determined by US (also refereed to as Clinical Controls); and Group 3 (n = 78) O. viverrini positive individuals who were also positive for APF as determined by US (also referred to as Clinical Cases) (Fig 1). Group 4 consists of samples from 33 individuals who were not part of the KKCC, but had histologically proven O. viverrini-associated CCA, whose urine samples were stored at the biological specimen repository of the Liver Fluke and Cholangiocarcinoma Research Center, Khon Kaen University, Thailand. The samples from Group 4 were chosen from the biological specimen repository by matching them on age, sex, and nearest neighbor (i.e., residence at time of death in a village within 10 kilometers of the current study sample) to the Clinical Cases in the KKCC.
A detailed description of the ultrasonography methods used in this study can be found in the following references [8–10]. Briefly, a mobile, high-resolution ultrasound (US) machine (GE model LOGIQ Book XP) was used. Hepatobiliary abnormalities including portal vein radical echoes, echoes in liver parenchyma, indistinct gallbladder wall, gallbladder size, sludge and suspected CCA were graded and recorded. Individuals were classified as “Non-Advanced Periductal Fibrosis” (APF-) or “controls” if the US grade was 0 or 1, and “Advanced Periductal Fibrosis” (APF+) or “case” if the US grade was 2 or 3. Individuals with alcoholic liver disease, which is seen as fatty liver by US exam, were excluded from this study. Also, individuals with marked hepatic fibrosis not related to OV infection (e.g., cirrhosis from HBV or HCV) were also excluded from this study (see Fig 1).
Data analysis was performed with SAS 9.2 (SAS Institute Inc., Cary, NC, USA). The raw data used to generate this manuscript are publicly available from the Dryad Digital Repository with the accession number doi:10.5061/dryad.pd6mn.
Table 1 shows the descriptive statistics for urinary 8-oxodG levels by sex, age strata, smoking status, alcohol consumption, and the criteria used to diagnose various stages of O. viverrini-induced infection. The levels were highest among persons in the 30–39 years of age group followed by individuals in the 20–29 years of age group, which consisted of only three individuals. Smokers had higher median levels of 8-oxodG at 185.93ng/mg creatinine compared to 147.55ng/mg creatinine in non-smokers. Similarly, individuals who indicated that they consumed alcohol had higher median levels of creatinine adjusted 8-oxodG compared to those who did not, with 154.83ng/mg creatinine and 133.43ng/mg creatinine, respectively. The 8-oxodG levels increased with advancing disease status.
The Kruskal-Wallis one-way analysis of variance demonstrated that there were significant differences in the distributions of urinary 8-oxodG levels between participants separated into groups by disease progression (p < .0001). Dunn's multiple comparisons procedure, for which the family-wise error rate was set at 0.05, identified significant (p < 0.05) differences in the distributions of urinary 8-oxodG between control individuals (both EN and APF-) and O. viverrini-infected individuals (APF or CCA positive). No significant differences were observed between participants in EN and APF- groups, which, along with co-occurrence of these individuals in the same population, justified combining these participants into one group in subsequent analyses. No significant difference was observed between participants in the APF+ and CCA groups. Lack of statistically significant findings here may be explained by the presence of three individuals in the APF+ group who had extremely high levels of urinary 8-oxodG (4.66, 4.33, and 4.04 times greater than the group's median urinary 8-oxodG levels). However, the groups remained distinct in subsequent analyses due to the presence of the outlier values and also due to the specific pathological differences that exist between APF+ and CCA individuals.
Fig 2 illustrates the distribution of urinary 8-oxodG in individuals participating in this study. Panel A of Fig 2 depicts the distributions of the levels of urinary 8-oxo-dG in each of the four groups of the study participants and Panel B illustrates the distributions following the amalgamation on participants from groups 1 and 2. Superscripts above the group labels on the x-axis indicate significant findings. Following the combination of control participants into one group (HBP negative), the following results were obtained: the concentrations of 8-oxodG in both cases and controls ranged between 14.25ng/mg creatinine and 1100.97ng/mg creatinine. The range was narrower in the control group (14.25 to 476.58 ng/mg creatinine, n = 71) than among cases, especially OV+ and APF+ (28.89 to 1100.96 ng/mg creatinine, n = 78) and OV+/CCA individuals (26.74 to 933.77 ng/mg creatinine, n = 33). The median 8-oxodG levels of persons with APF or CCA were significantly higher (2.12x and 1.62x, respectively) than the median level of the control group as determined by Dunn's multiple comparisons procedure applied to the Kruskal-Wallis one-way analysis of variance, with the family-wise error rate set at 0.05. These data indicated progressively higher levels of 8-oxodG in the urine during advanced hepatobiliary disease during opisthorchiasis.
Various candidate biomarkers, including urinary 8-oxodG, serum level of IgG and IgG1 against a crude adult O. viverrini antigen extract were modeled as potential relevant predictors of progression to advanced hepatobiliary pathology during chronic O. viverrini infection. In the initial stages of model development, control individuals, APF+ individuals, and CCA individuals were included in the model to identify the potential relevant predictors of disease progression. This model identified age (p = 0.0065), serum IgG (p = 0.0002), and urinary 8-oxodG (p < 0.0001) as significant and these predictors were subsequently considered in the two models comparing control individuals to either APF+ participants or CCA participants. In this second stage of model development, urinary 8-oxodG was retained as the single significant biological predictor of an individual having ultrasound confirmed APF (p ≤ 0.0001) or histologically confirmed CCA (p ≤ 0.0014)). Age and serum IgG were no longer significant in the model of APF+ participants (p = 0.4802 and p = 0.2890, respectively). In the model of CCA individuals age remained a significant predictor (p < 0.0001), while serum IgG was eliminated as non-significant (p = 0.0587). The final parsimonious models retain the concentration of creatinine-adjusted urinary 8-oxodG as the only predictive biomarker (for diagnosis of both APF and CCA) along with age as a significant covariate in individuals diagnosed with CCA (equation 2 in S1 Text—Supplementary Equations and Definitions).
The results of a logistic regression model that included urinary 8-oxodG as a predictor were considered alongside the actual ability of the clinical (gold standard) methods to identify specific increases in the level of 8-oxodG in urine of participants in the study. Increasing odds ratios (ORs) were associated with progressively higher concentrations of 8-oxodG, indicting the increased likelihood of progressing to APF and CCA for individuals with higher measurable concentrations of creatinine-adjusted 8-oxodG in the urine. The ORs and their 95% confidence intervals (95% CIs) are summarized in Table 2.
Receiver operating characteristic (ROC) curves constructed from levels of creatinine-adjusted urinary 8-oxodG measured in study participants are presented in Figs 3 and 4. The area under the ROC curve (AUC) describes the probability of correctly identifying a positive individual as a ‘case’ and a negative individual as a ‘non-case’. An AUC of 1 would describe a diagnostic test that would correctly identify all cases and all non-cases 100% of the time. The 45-degree line in the ROC plot marks the “chance diagonal”, which corresponds to an ROC curve with an AUC of 0.50 [50]. Figs 3 and 4 are labeled with the AUC values of each ROC curve; the ROC curve generated from a diagnostic model of APF+ individuals has an AUC of 0.74 and the AUC of the ROC generated from a diagnostic model of CCA individuals is 0.88. This means that levels of urinary 8-oxodG in urine correctly identifies individuals with O. viverrini-induced APF 74% of the time and correctly identifies individuals with O. viverrini-induced CCA 88% of the time. The diagnostic positivity thresholds were established as described in the methods section and are presented in Table 3 along with the relevant diagnostic validity parameters (sensitivity and specificity), the PPV, NPV, LR+, and LR- for the urinary 8-oxodG assay.
O. viverrini-infected individuals with advanced periductal fibrosis (APF) as determined by abdominal ultrasound had markedly elevated levels of urinary 8-oxodG compared to O. viverrini-infected individuals without APF. Moreover, the concentrations of 8-oxodG in the urine of individuals with APF were comparable to the highly elevated levels of this oxidatively modified DNA lesion in individuals with O. viverrini-induced CCA [28, 31–33]. These results clearly suggest that elevated levels of this metabolite in urine are indicative of hepatobiliary fibrogenesis and tumorogenesis from chronic O. viverrini infection. Moreover, levels of 8-oxodG in the urine of O. viverrini infected individuals with APF or CCA individuals corroborated the ‘gold’ standard diagnostics used to detect both of these hepatobiliary pathologies in a dose-dependent manner: e.g., the highest 50 unit increment of 8-oxodG (200 units) indicated an increased risk of diagnosis of APF or CCA by 354% and 408%, respectively, compared to individuals with no detectable levels of urinary 8-oxodG. Furthermore, the risk models used to evaluate the utility of 8-oxodG as a biomarker were also used to identify the diagnostically relevant levels of 8-oxodG in the urine of study participants. The identification of these diagnostic threshold levels of urinary 8-oxodG, if corroborated in additional larger scale validation studies, would have important implications for urine 8-oxodG as diagnostic tool in field settings, with special significance in resource-limited settings, since they establish benchmarks that may be used to identify individuals at-risk of CCA and refer them for further testing (e.g., confirmatory abdominal ultrasound diagnosis) and preventive chemotherapy.
In order to assess the utility of 8-oxodG as a candidate biomarker for O. viverrini-induced APF and CCA, we also constructed a logistic regression risk model that initially included creatinine-adjusted urinary 8-oxodG levels along with significant demographic, behavioral, and immunological covariates associated with chronic O. viverrini infection. In this risk model, creatinine-adjusted urinary 8-oxodG emerged as the only significant predictor of APF, especially at higher concentrations of 8-oxodG. Creatinine adjusted 8-oxodG was also a significant predictor of CCA status when adjusted for age. Moreover, the risk model showed that 50 unit increases in creatinine-adjusted urinary 8-oxodG (in ng/mg) increased its ability to corroborate APF status by 137% and CCA status by 142% (after adjusting for age) and also helped to establish diagnostically relevant threshold levels of this biomarker. Moreover, elevated levels of urine 8-oxodG accurately identified progression to advanced stages of the disease (Figs 3 and 4), with high odd ratios (ORs) of APF and CCA associated with higher levels of this candidate biomarker (Table 2). As APF is precursor stage to CCA [36], a simple, non-invasive assay for 8-oxodG in urine as a biomarker for this O. viverrini-induced pathology would be of profound benefit in Southeast Asia, especially among populations residing in the resource-limited settings of the Mekong Basin region, where the incidence of intrahepatic CCA is the highest in the world [51].
The methods and standards for measuring urinary 8-oxodG have received considerable attention[24, 40], with a consensus on using urine 8-oxodG as a biomarker established by the European Standards Committee on Urinary (DNA) Lesion Analysis (ESCULA) [24]. When properly detected, measured, and analyzed, ESCULA determined urine 8-oxodG to be a robust, accurate, reproducible, and (“remarkably”) stable urine biomarker for OS [24]. The production of 8-hyrdroyguanine is almost exclusively elicited by OS, with the main attack site by oxidative radicals at the N7-C8 bond [26, 52, 53]. While there are findings that suggest that diet contributes to urinary levels of thymine glycol and 8-oxo-7,8-dihydro-guanine (8-oxoGua) [40], little or no information has been reported that this applies to urinary levels of oxidatively modified 2′-deoxyribonucleosides (8-oxodG), hence obviating diet as a potential confounder. As with other studies, the present findings showed a relatively strong association between levels of urinary creatinine and levels of urinary 8-oxodG, which (as explained by Barregard et al. [40]) are likely due to differences in body mass index (BMI), since metabolic rate is associated with lean body mass and a higher metabolic rate generates a larger amount of modified 2-deoxyribonucleoside products in urine [40]: e.g., in general, males excrete more 8-oxodG than females per kilogram body weight [40]. Hence, following the recommendations of Barregard et al. [40] for using 8-oxodG in cross-sectional studies, we normalized urinary 8-oxodG concentrations by individual creatinine concentrations as determined by the Jaffe method [38].
Oxidative DNA damage as a key event in O. viverrini-induced CCA has been studied extensively in an animal model (see [23] for review), where oxidatively damaged DNA bases formed along the inflamed intrahepatic biliary ducts, at sites adjacent to the parasite, perhaps from persistent wound repair [31, 32]. Immunohistochemical studies in a hamster model of O. viverrini-induced CCA showed that inflammatory cells surrounding the parasite in the bile duct, such as mononuclear cells and eosinophils, generate reactive oxygen species (ROS), which induce oxidative stress (OS) and increased cleavage of 8-oxodG [31, 32]. Additionally, immunohistochemical analyses on livers resected from humans with O. viverrini-induced CCA showed 8-oxodG in tumor tissue from the bile duct [20, 33]. Moreover, Thana et al. [33] observed elevated levels of 8-oxdG in the urine of individuals infected with O. viverrini compared to healthy controls as also observed in the current study (Fig 2 Panel A). However, the current study adds to the literature the observation that urine 8-oxodG can be detect in markedly higher concentration when O. viverrini infected individuals who have progressed to APF or CCA (Fig 2).
The observation that urinary 8-oxodG concentrations were similar in APF and CCA individuals support our hypothesis that common mechanisms drive bile duct fibrosis and bile duct tumorogenesis from chronic O. viverrini infection [8–10, 12, 13]. It is also in keeping with findings from other groups that chronic bile duct inflammation leads to a desmoplastic stroma (i.e., fibrosis) in the bile duct that precedes CCA (reviewed by Sirica and Gores [36]. As depicted in S1 Fig, elevated levels of 8-oxodG in the urine of APF individuals reflects ongoing tissue repair and the smoldering inflammatory milieu” [14] in the hepatobiliary epithelia from persistent injury from the parasite [15, 30, 54, 55]. This “desmoplastic reaction” to chronic O. viverrini infection provides a rich niche for cancer cells to develop and progress [1, 36, 56, 57]. Currently, therapeutic targeting to reduce desmoplastic stroma (periductal fibrosis) to prevent CCA is being investigated [36, 56, 57].
Taken together, these data show that urine 8-oxodG may be an excellent biomarker for the advanced hepatobiliary pathology that occurs prior to O. viverrini–induced CCA. In keeping with a recent position statement from the European Group on Tumor Markers [58], the current manuscript places urine 8-oxodG as a “candidate biomarker” for APF and CCA in the “discovery phase”, i.e., “when differential expression of a specific marker is shown to associate with a ‘gold’ standard clinical outcome”. The next phase of biomarker development for urine 8-oxodG is the “verification stage”, when our analyses would be extended to a much larger sample (i.e., hundreds) of individuals infected with O. viverrini and at risk of CCA. The objective of the verification stage would be to incorporate the broadest range of cases and controls in order to capture the environmental, genetic, biological, and stochastic variation in the population in the Mekong Basin Subregion. As the diagnostic sensitivity of the candidate biomarker has been established herein, the verification stage would focus on the specificity of the candidate biomarker and the utility of the diagnostic threshold levels set forth in this manuscript [58]. If validated, levels of urinary 8-oxodG may be an inexpensive way to identify at-risk individuals, who may be referred for subsequent, more demanding testing. This process would streamline diagnostics for APF and CCA and perhaps improve the utilization of the limited resources allocated to cancer screening in this region.
The findings herein confirm previous observations that severe hepatobiliary disease occurs early and asymptomatically among residents in O. viverrini endemic areas. A simple, non-invasive assay targeting 8-oxodG in urine would be of profound benefit to populations in Southeast Asia, especially in the resource-limited settings of the Mekong Basin region countries of Thailand, Laos and Cambodia, where the incidence of O. viverrini-induced CCA is the highest in the world [7]. The future plan for the candidate biomarker includes moving to a verification step to test its accuracy in a larger sample size [58].
|
10.1371/journal.pcbi.1003393 | Exploring Early Stages of the Chemical Unfolding of Proteins at the Proteome Scale | After decades of using urea as denaturant, the kinetic role of this molecule in the unfolding process is still undefined: does urea actively induce protein unfolding or passively stabilize the unfolded state? By analyzing a set of 30 proteins (representative of all native folds) through extensive molecular dynamics simulations in denaturant (using a range of force-fields), we derived robust rules for urea unfolding that are valid at the proteome level. Irrespective of the protein fold, presence or absence of disulphide bridges, and secondary structure composition, urea concentrates in the first solvation shell of quasi-native proteins, but with a density lower than that of the fully unfolded state. The presence of urea does not alter the spontaneous vibration pattern of proteins. In fact, it reduces the magnitude of such vibrations, leading to a counterintuitive slow down of the atomic-motions that opposes unfolding. Urea stickiness and slow diffusion is, however, crucial for unfolding. Long residence urea molecules placed around the hydrophobic core are crucial to stabilize partially open structures generated by thermal fluctuations. Our simulations indicate that although urea does not favor the formation of partially open microstates, it is not a mere spectator of unfolding that simply displaces to the right of the folded←→unfolded equilibrium. On the contrary, urea actively favors unfolding: it selects and stabilizes partially unfolded microstates, slowly driving the protein conformational ensemble far from the native one and also from the conformations sampled during thermal unfolding.
| The delicate equilibrium between the folded and functional structure of a protein and its unfolded state is highly dependent on environmental variables such as the solvent. For example the co-solvent urea is a well-known protein denaturant that displaces the equilibrium towards unstructured and non-functional conformations of proteins. However the molecular mechanism behind its ability remains an enigma and the interpretation of the experimental data is still ambiguous. By analyzing a set of representative proteins through extensive molecular dynamics simulations in urea, we provide a robust and consensus picture of the first stages of urea-driven protein unfolding and elucidate the role of urea in accelerating protein unfolding. Our results suggest that urea, thanks to its stickiness and slow diffusion, benefits from the intrinsic flexibility of proteins and stabilizes partially open-states, slowly driving the protein toward unfolding.
| Urea is a protein denaturant that has been used for decades in the study of protein folding/unfolding; however, after many years of research the ultimate reasons of the denaturing properties of urea remain elusive [1], [2]. The dominant paradigm for unfolding (the “direct” mechanism) claims that the denaturant properties of urea are related to its capacity to interact with exposed protein residues more strongly than water [3]–[15]. However, the nature of such a preferential interaction is not so clear. Thus, while some authors suggest that it is mostly electrostatic and related to the formation of direct hydrogen bonds [7]–[9], [16]–[17], others claim that preferential dispersion is the leading term [13]–[15]. It is also unclear whether the major destabilizing effect of urea is related to interaction with the backbone [6]–[7] or with side chains [8]–[12]. In the latter case, there is also discussion regarding the preferential side chains: polar and charged [9] or apolar [4], [10]–[12].
We recently combined multi-replica molecular dynamics (MD) simulations and direct NMR measures of ubiquitin to characterize the “urea unfolded ensemble” of this model protein [15]. Our results suggest that urea stabilizes flexible over-extended conformations of the protein, which are unlikely to be sampled in the “unfolded” state of aqueous proteins. Extended conformations of the protein with exposed hydrophobic surfaces are more urea-philic than the native globular state, due mostly to extensive London dispersion interactions (the attractive contribution in Van der Waals interactions between instantaneous dipoles) between apolar side chains and urea molecules in the first solvation shell of unfolded conformations. We believe that our results in reference 15 clarify the molecular basis of the effect of urea on the thermodynamics of the folded←→unfolded equilibrium, but unfortunately, they do not provide information on the kinetic role of urea in the unfolding process. In other words: does urea actively induce protein unfolding? Or, on the contrary, does it passively stabilize the unfolded state by selectively binding to unfolded conformations? To analyze this point, we should characterize the effect of urea in the first stages of thermochemical unfolding, when the protein structure is still close to the native conformation and internal residues are not fully exposed. Clearly, a study of this nature presents many difficulties, the most important being that the effect of urea on early stages of unfolding might be dependent on the native structure. Therefore, to obtain conclusions of general validity, all representative protein folds should be addressed. Also, results can be force-field-dependent, so if we aim to obtain robust conclusions, we should perform simulations with a variety of force-fields.
Given the typical kinetics of the folding/unfolding transitions of small globular proteins [18], microsecond (µsec) long simulations should trace the first stages of these processes. In the current work, we investigate the first stages of urea-driven protein unfolding using µsec-long atomistic simulation; to gain universality, we used 30 proteins representative of all protein folds, while to protect our conclusions from force-field-related uncertainties, we used several of the most popular force-fields. The results derived from this study provide a robust and complete picture of the role of urea in destabilizing folded states of proteins, and more importantly, on the molecular mechanisms by means of which urea contributes to accelerating protein unfolding.
We first validated our protocol using three ultra-representative proteins (in bold in Table 1), one for each of the main classes in the Structural Classification of Proteins (SCOP, [19]). We monitored the protein stability in three environments: i) in chemical unfolding conditions, in 8M urea and with a mildly high temperature (T = 368K) to speed up the observable effects; ii) in thermal unfolding condition, in water with the same high temperature; this control allowed us to distinguish the effect of urea and temperature on protein unfolding; iii) in water at room temperature as final control. Four force-fields were used (OPLSAA - ON2; CHARMM - C22; AMBER99 - P99 and P99SBILDN) for each system (see Methods for the description of the force-fields used), collecting in total 36 simulations of 1-µsec length each.
After the validation of our protocol, we extended the chemical unfolding simulations to a larger set of proteins, to avoid any bias in the conclusion due to the native structure. We performed 1 µsec of simulation in urea at high temperature (T = 398K) for 30 proteins covering all the major protein folds (Table 1 and Suppl. Dataset S1). Each system was simulated in three force-fields (C22, ON and P99), excluding P99SBILDN as reported above, and collecting in total 90 simulations. To have a more realistic picture of the native state, instead of using the crystal structure, we used as control 0.1 µsec-long simulations in water at room temperature for all the 90 systems. The analysis described here reveals some common robust trends that illustrate the effect of urea during the early stages of protein unfolding.
MD simulations with additive potentials and explicit solvent have become very popular to explore chemical unfolding of protein. There is little doubt that the use of the technique has produced sizeable advances in the field, but we cannot ignore some potential caveats in the beginning of this discussion. First, for computational reasons we (and most authors in the field) are using classical non-polarizable force-fields, which might not be accurate enough to deal with a complex process such as unfolding. Previous studies [4]–[5], [7]–[15], [25] have however demonstrated that urea/water/protein effective parameters are able to reproduce a variety of experimental observables, such as mass densities and radial distribution functions of urea/water solutions derived from neutron scattering experiments [25], the experimental water/urea transfer free energies of tripeptides [10], and the urea density around unfolded proteins found by vapor pressure osmometry measures [4], [8], [10], [13], [15], [24]. Furthermore, our recent work [15] has demonstrated that unbiased MD simulations in 8M urea reproduce very accurately the unfolded ensemble as determined from a variety of spectroscopic techniques (including SAXS and NMR) under the same conditions. Thus, despite their simplicity current force-fields reproduce reasonably well urea/water/protein mixtures. We should remember that since we are exploring a microsecond-long process, no direct experimental data is available for comparison and accordingly caution is required. This move us to use a consensus approach, running the simulations with different force-fields to extract those results that seem robust to force-field changes.
A second reason of concern is related to the stochastic nature of unfolding, where individual trajectories can show different degree of unfolding [13]. Again, by comparing different trajectories we tried to define robust findings, but we cannot ignore that the experimental result is the averaging a near-Avogadro number of trajectories. A third reason of concern, common to many experimental studies, is the generality of the results, i.e. how general are the results obtained with a few model proteins. To convince ourselves on the general validity of our results we repeated the unfolding studies for a large number of proteins representative of all prevalent folds. Despite the obvious caveat of any theoretical study, this approach provided a picture of unprecedented, to our knowledge, completeness and robustness of the early stages of urea unfolding.
Under our simulation conditions (8M urea at T = 368 K), we detected clear signals of unfolding in the microsecond range, but progress in denaturation was smaller than that reported for model proteins of reduced stability [13]. Overall, the advance in unfolding of proteins at T = 368 after 1 µsec of MD is not dependent on the fold, nor secondary structure composition, and was similar for proteins with and without disulphide bridges in the native form. No dramatic differences in the advance of unfolding were found between water and urea simulations performed at the same temperature; however, unfolding paths in the presence or absence of urea differed, since the partially unfolded structures sampled in hot water maintained the hydrophobic residues hidden in the core of the protein, while such residues were more accessible in presence of urea.
Urea solutions are more viscous than pure water, which in our simulation reduced high-frequency movements in the protein, generating an unexpected slow down of the atomic-motions. Urea residence times around protein residues were large, especially when urea molecules diffuse close to the hydrophobic core or to the interface between rigid and thermally mobile regions (hinge points). We consider that the sticky nature of urea and its preferential placement at hinge points is crucial for unfolding, since it favors the rapid trapping of residues that become exposed as a consequence of stochastic thermal motions. The stabilizing effect of urea on exposed residues slowly biases the trajectory towards the unfolded state, by decreasing the chances of microscopic refolding [12], [26]. The effect of stabilization of exposed residues is especially productive in terms of unfolding when residues are apolar, since in this case urea (but not water) traps very efficiently the residue, increasing the accessibility of other apolar residues in the vicinity. The ensuing greater recruitment of urea in the region leads to a cooperative effect resulting in the acceleration of protein unfolding. Our data shows that, similar to the unfolded state [13], [15], it is the van der Waals interactions that drive the accumulation of urea on the surface of the folded protein. However, the role of H-bonding cannot be dismissed, as these bonds are crucial for the stabilization of long-living urea interactions near hinge points, which in turn are required to bias intrinsic protein dynamics towards unfolding. Clearly, “direct” effects not only are the main factors responsible for the urea-mediated stabilization of the unfolded state [15], but are also relevant in guiding the first steps of urea unfolding.
Microscopic unfolding events are related to stochastic thermal motions, which are in principle similar to those that occur spontaneously in water at room temperature. However, urea is not a mere passive spectator that simply stabilizes the small percentage of unfolded protein coexisting within the native ensemble and leading to a displacement in the folded←→unfolded equilibrium towards the denatured state. On the contrary, urea has a dual function: i) it takes advantage of microscopic unfolding events, decreasing their chances of refolding, and favoring further unfolding [12], [26]; and ii) among these microscopic unfolding events it selects and stabilizes microstates with exposed hydrophobic regions [4] (see Suppl. Figure S7). These effects lead to a slow divergence in the temperature-unfolding pathways in water and urea, and, as shown for ubiquitin [15], to distinct unfolded states. Consequently, concepts such as folded and unfolded states or folding and unfolding pathways need to be revisited and reformulated considering the nature of the denaturant used.
As model structures for the main protein-folds we used the same structures selected in our previous work in reference 20. We first explored the early stages of urea unfolding using three ultra-representative proteins for the most populated fold in the three main classes in the SCOP database (all-α 1OPC, all-β 1CQY and α/β 1KTE; [19], [20]). Once the simulation protocols had been validated with these proteins, the study was extended to a larger set, consisting of 30 structures (110 residues on average) representative of the most populated protein folds ([20], [27] and Suppl. Dataset S1)
All starting structures were taken from the Protein Data Bank (PDB; [28]) and processed using our standard procedure implemented in the MDWeb server [29]: experimental structures were titrated to define the major ionic state at neutral pH, neutralized by ions (sodium and chloride), minimized for 1000 steps, heated up to the final temperature, and solvated using a 8M urea/water octahedron box with a spacing distance of 15 Å around the system. The box was previously equilibrated in a Monte Carlo simulation using the BOSS program [30]. The water model was taken from Jorgensen's TIP3P [31], while ion and urea force-field parameters were those considered as the default of each force-field. Urea parameters from Smith et al. [32] were used for OPLS and P99SBILDN simulations, the same charges but scaled according to the amber force-field were used in PARM 99, while Nilsson's parameters were used in the CHARMM 22 force field [33]. Systems were then pre-equilibrated for 0.5 ns with parm99-AMBER force field in keeping the backbone restrained by intra-molecular harmonic potentials and then equilibrated (0.5 ns) in each force field parameters removing backbone constraints.
For the small set of ultra-representative proteins, three sets of simulations corresponding to water at room temperature (T = 300 K), hot water (T = 368 K), and urea at high temperature (T = 368 K) were carried out. For each condition, we performed 1 µsec simulations using four force-fields: three general purpose ones (OPLSAA -ON2- [34]; CHARMM -C22- [35]; AMBER99 -P99- [36]), and a last-generation force-field able to accurately reproduce folded proteins (P99SBILDN, [37]). For the extended set of 30 proteins, control simulations in water were limited to 0.1 µsec at room temperature, while the 8M urea simulations were performed, as above, for 1 µsec at T = 368 K. Simulations for the extended set of proteins were carried out using ON2, C22 and P99. All simulations were performed using periodic boundary conditions and particle Mesh Ewald [38] corrections for the representation of long-range electrostatic effects using a 1.0 Å grid spacing and a 9 Å cutoff. All trajectories were collected with the NAMD2 [39] program. Integration of equations of motions was performed every 2 fs after removing vibrations of bonds involving hydrogen atoms using SHAKE/RATTLE algorithm [40], [41]. All simulations were carried out in the isothermal (T = 300 or 368 K)/isobaric ensemble (P = 1 atm) using the Langevin thermostat and barostats [42], [43]. The trajectories were analyzed using VMD [44] and the MdWeb server [29], as well as Flexserver which can be accessed at: http://mmb.pcb.ub.es/FlexServ/ (see also Suppl. Text S1 for a detailed explanation of the metrics).
|
10.1371/journal.pbio.2000094 | The Nuclear Receptor HIZR-1 Uses Zinc as a Ligand to Mediate Homeostasis in Response to High Zinc | Nuclear receptors were originally defined as endocrine sensors in humans, leading to the identification of the nuclear receptor superfamily. Despite intensive efforts, most nuclear receptors have no known ligand, suggesting new ligand classes remain to be discovered. Furthermore, nuclear receptors are encoded in the genomes of primitive organisms that lack endocrine signaling, suggesting the primordial function may have been environmental sensing. Here we describe a novel Caenorhabditis elegans nuclear receptor, HIZR-1, that is a high zinc sensor in an animal and the master regulator of high zinc homeostasis. The essential micronutrient zinc acts as a HIZR-1 ligand, and activated HIZR-1 increases transcription of genes that promote zinc efflux and storage. The results identify zinc as the first inorganic molecule to function as a physiological ligand for a nuclear receptor and direct environmental sensing as a novel function of nuclear receptors.
| Zinc is an essential nutrient for all life forms, and maintaining zinc homeostasis is critical for survival. However, little is known about how animals sense changes in zinc availability and make adjustments to maintain homeostasis. In particular, logic dictates there must be a mechanism for zinc sensing, but it has not been defined in animals. We discovered that the nuclear receptor transcription factor HIZR-1 is the master regulator of high zinc homeostasis in the roundworm Caenorhabditis elegans. In response to high dietary zinc, HIZR-1 activates transcription of multiple genes that encode a network of proteins that store and detoxify excess zinc. Furthermore, our results suggest HIZR-1 itself is the high zinc sensor, since it directly binds zinc ions in the ligand-binding domain that regulates transcriptional activation. These findings advance the understanding of zinc homeostasis and nuclear receptor biology. Nuclear receptors were initially characterized as receptors for hormones such as estrogen, indicating complex animals use these transcription factors to monitor their internal environment. However, nuclear receptors are present in simple organisms that lack endocrine signaling, suggesting these transcription factors might have a primordial function in sensing the external environment. Our results identify a new class of nuclear receptor ligands, the inorganic ion zinc, and a new function for nuclear receptors in directly sensing levels of a nutrient. We speculate that nutrient homeostasis mediated by direct binding of nutrients to the ligand-binding domain is a primordial function of nuclear receptors, whereas endocrine signaling in complex animals mediated by direct binding of hormones to the ligand-binding domain is a derived function of nuclear receptors that appeared later in evolution.
| Zinc is an essential nutrient for all life, including plants, animals and microbes, because zinc is involved in many different cellular events. Zn2+ binds tightly to many proteins and thereby contributes to their tertiary structure or catalytic activity [1], and Zn2+ has been proposed to function as a second messenger signaling molecule during synaptic transmission, development, and immune responses [2–4]. Zinc homeostasis is vital for human health. Inadequate dietary intake is a prevalent cause of human zinc deficiency, whereas genetic disorders that disrupt zinc uptake occur rarely. Zinc deficiency causes pathological changes in a wide range of tissues, reflecting the many uses of zinc [5–7]. Zinc excess also causes human pathology, and it may occur systemically or in specific tissues. For example, ischemic injury has been proposed to cause zinc release that mediates cell death [8,9]. Several human diseases, including diabetes, cancer, and neurodegenerative diseases, are correlated with genetic variations that affect zinc metabolism [10–12]. Elucidating zinc homeostasis is important for understanding an ancient biological process and may contribute to improving human health.
Because excess zinc is toxic, organisms require mechanisms to sense and detoxify high levels of zinc. One fundamental and evolutionarily conserved mechanism is transcriptional regulation of genes involved in zinc detoxification, such as zinc exporters and zinc sequestering proteins [13–16]. However, the regulation of high-zinc–activated transcription remains poorly understood in animals. Critical questions in this field include the following: What is the direct sensor of high zinc? And how does this sensor activate transcription of zinc homeostasis genes?
The nematode Caenorhabditis elegans is a useful model system for studies of zinc homeostasis because its simple body plan, its transparency, and the availability of powerful genetic techniques facilitate experimental analysis [13,17–21]. Studies of C. elegans are relevant to mammalian biology, since the genome encodes evolutionarily conserved zinc transporters and metallothioneins [22,23]. The transcription of these genes is regulated by dietary zinc levels in C. elegans, which is also similar to yeast and mammals [24,25]. C. elegans contains two metallothionein genes, mtl-1 and mtl-2, that are induced at the level of transcription in intestinal cells by high dietary zinc [26]. C. elegans contains 14 cation diffusion facilitator (CDF) genes that encode transporters for zinc and possibly other metal ions. The CDF genes cdf-2 and ttm-1b are transcriptionally up-regulated in intestinal cells by high dietary zinc, and these transporters promote zinc storage and excretion [14,19]. Transcriptional activation of these genes is mediated by the high zinc activation (HZA) element, a DNA enhancer [15]. However, the HZA-binding factor has not been identified.
Here we describe an unbiased forward genetic screen used to identify mediators of high-zinc–activated transcription that resulted in the discovery of the HZA-binding factor, which we named the high-zinc–activated nuclear receptor (HIZR-1). hizr-1 encodes a nuclear receptor transcription factor that has an evolutionarily conserved DNA-binding domain (DBD) and ligand-binding domain (LBD). We demonstrated that HIZR-1 is both necessary and sufficient to activate transcription of endogenous zinc-homeostasis genes in response to high dietary zinc. Thus, HIZR-1 is the master regulator of high-zinc homeostasis in C. elegans. We used genetic and biochemical approaches to analyze HIZR-1 function. The LBD directly bound zinc, which promoted nuclear accumulation and activation of the protein, indicating the LBD regulates protein activity and zinc is a physiological ligand; the DBD directly bound the HZA enhancer, which mediates transcriptional activation of multiple genes involved in zinc homeostasis [15]. These findings advance the understanding of zinc biology by identifying a sensor for high zinc in animals and elucidate homeostatic systems by defining a positive feedback loop embedded in a negative feedback circuit.
Nuclear receptors (NRs) were discovered and characterized as sensors of endocrine signals in mammals. These transcription factors contain a regulatory ligand-binding domain that interacts with hormones and a DNA-binding domain that interacts with target genes. Despite decades of effort, only about half of human NRs have identified ligands, raising the possibility that novel classes of ligands remain to be discovered [27]. NRs exist in simple organisms such as sponges that lack endocrine signaling, suggesting NRs might have primordial function in sensing external molecules [28]. However, external ligands have yet to be identified. The analysis of HIZR-1 expands the understanding of the NR superfamily by (1) identifying transition metals as a new class of physiological ligand that is distinct from previously described classes such as steroids and lipids and (2) identifying direct nutrient sensing as a new function that may represent a primordial role of NRs.
Homeostasis in response to high zinc is mediated by transcriptional activation of multiple genes in C. elegans, including the zinc transporter genes cdf-2 and ttm-1b [13,14]. To identify genes that mediate this response, we screened for mutant animals that displayed abnormal regulation of cdf-2. To visualize cdf-2 transcription, we used the method of bombardment to generate transgenic animals with an integrated, multicopy array containing a plasmid with the cdf-2 promoter fused to the coding region for green fluorescent protein (GFP) (cdf-2p::gfp) (see Materials and Methods). Transgenic cdf-2p::gfp animals displayed low-level fluorescence in standard medium and high-level fluorescence in intestinal cells in medium supplemented with zinc (Fig 1A and 1B). We identified one semidominant mutation (am285) that caused increased fluorescence in standard medium, a phenotype we named zinc-activated transcription-constitutive (Zat-c). We identified five recessive mutations (am279, am280, am286, am287, and am288) that caused reduced fluorescence in medium supplemented with zinc, a phenotype we named zinc-activated transcription-deficient (Zat-d) (Fig 1C and 1D). All six mutant strains displayed growth rates similar to wild type when cultured on standard medium. All five Zat-d mutations failed to complement one another, indicating they affect the same gene. The Zat-c and Zat-d complementation groups were positioned in the center of linkage group X (Fig 1E).
To determine how these mutations affect transcription of endogenous genes, we analyzed mRNA levels by quantitative PCR (qPCR). In wild-type animals, mRNA levels of cdf-2, ttm-1b and the metal binding metallothionein, mtl-1, are increased significantly by zinc supplementation [14,19]. By contrast, in am286 mutant animals cdf-2, ttm-1b, and mtl-1 transcript levels were significantly lower than wild type when exposed to supplemental zinc (Fig 2A–2C). Furthermore, the strain with the am285 semidominant mutation displayed significantly higher levels of cdf-2, ttm-1b, and mtl-1 transcripts compared to wild type in the absence of supplemental zinc (Fig 2D–2F). Thus, the gene affected by am286 was necessary and the gene affected by am285 was sufficient for high-zinc–activated transcription of multiple endogenous genes.
To test the hypothesis that the gene affected by am286 is functionally important for zinc homeostasis, we analyzed growth in the presence of supplemental zinc. am286 mutant animals grew similarly to wild-type animals in standard medium, demonstrating the strain is healthy in standard conditions, but displayed retarded growth compared to wild-type animals when cultured in high dietary zinc (Fig 2G). This growth defect was specific to zinc toxicity, since am286 mutant animals displayed growth similar to wild-type animals in high dietary copper (Fig 2H). Thus, the gene affected by am286 was necessary for normal growth and development in high dietary zinc.
To identify the affected gene, we performed whole genome sequencing. All six mutant strains contained mutations in the gene ZK455.6/nhr-33, which is in the mapping interval. A comparison of the predicted protein sequence to protein databases revealed homology to nuclear receptors, and we named the gene high-zinc–activated nuclear receptor (hizr-1) (S1 Fig). HIZR-1 has a conserved DBD that contains two predicted zinc-finger DNA-binding motifs, which is typical of nuclear receptors, and a conserved LBD. The recessive Zat-d alleles include three encoding substitutions of conserved residues in the DBD and two encoding truncated proteins (one nonsense and one change in a consensus splicing site). The semidominant, Zat-c allele encodes a substitution affecting the LBD (Fig 1E, S1 Table). To confirm this gene assignment, we analyzed an independently derived allele and performed rescue experiments. hizr-1(gk698405), a nonsense mutation generated by the C. elegans million mutations project [29], caused a Zat-d phenotype, as predicted (Fig 1E, S2A Fig, S1 Table). Expression of wild-type HIZR-1 protein fused to GFP rescued the am286 Zat-d mutant phenotype, as predicted (S2B and S2C Fig). The molecular and genetic analyses indicate that the five Zat-d mutations are likely strong loss-of-function or null alleles of hizr-1, whereas the Zat-c mutation is likely a gain-of-function allele of hizr-1.
We hypothesized that zinc is a ligand for HIZR-1, leading to the predictions that zinc will directly bind the LBD of purified HIZR-1 and high zinc will promote nuclear accumulation of HIZR-1 in animals. To avoid the complication of zinc binding to the DBD, which contains two predicted zinc finger motifs, we used affinity chromatography to partially purify the LBD of HIZR-1 fused to glutathione-S-transferase (GST). GST alone and GST fused to the LBD of the C. elegans DAF-12 NR, which uses dafachronic acid as a ligand [30,31], were used as specificity controls. The amino acid sequences of the DAF-12 LBD and the HIZR-1 LBD are about 15% identical, and about 47% of the residues are weakly similar. Zinc binding was analyzed using radioactive zinc-65. The LBD of HIZR-1 displayed saturable, high-affinity binding to zinc (Fig 3A). GST alone and the LBD of DAF-12 fused to GST displayed similar low-level binding, demonstrating the protein specificity of the zinc binding (Fig 3B). Nickel and manganese did not effectively compete with zinc for protein binding, demonstrating the interaction between the LBD of HIZR-1 and zinc was metal-selective (Fig 3C). However, copper did display binding to the LBD of HIZR-1. These results demonstrate a direct, high-affinity, protein-specific and metal-selective interaction between the LBD of HIZR-1 and zinc.
These data can be used to estimate the stoichiometry of binding; however, this estimation is subject to important caveats. The crystal structure of GST indicates the protein binds one molecule of zinc, suggesting the stoichiometry of binding is 1 zinc:1 GST protein molecule [32]. If we assume this stoichiometry in our binding reactions, which has not been demonstrated directly, then we estimate the zinc:HIZR-1 LBD stoichiometry is 3:1 (Fig 3A) and 4:1 (S3 Fig). The amino acids that typically coordinate zinc in proteins are histidine and cysteine, and zinc is typically coordinated by four such residues. The predicted LBD of HIZR-1 contains 15 histidine and cysteine residues, suggesting it might have the capacity to coordinate three or four zinc ions with these residues. However, the data presented here do not address the role of these residues in zinc binding.
These data can be used to estimate a dissociation constant, although this value is subject to important caveats. The calculated dissociation constant was 2.6 +/- 0.2 μM when the concentration of zinc was varied (S3 Fig) and 1.7 +/- 0.3 μM when the concentration of protein was varied (Fig 3A), which are in good agreement. However, several caveats must be considered in interpreting these calculated dissociation constant values: (1) The calculation assumes that all added zinc is available for protein binding; however, the binding reaction contained glutathionine, which is known to have zinc binding affinity and is predicted to reduce the concentration of zinc available to bind the protein. (2) The calculation assumes that every protein molecule is “active” and able to bind zinc, whereas some might have been inactive or denatured. (3) The calculation assumes that zinc and protein are bound in a 1:1 stoichiometry—as described above, HIZR-1 appears to bind more than one zinc per protein molecule. These data indicate that zinc directly binds the LBD of HIZR-1 with high affinity and metal selectivity. However, further biochemical studies are necessary to accurately define the dissociation constant, stoichiometry, and specific amino acid residues that mediate zinc binding.
To characterize regulation in animals, we analyzed the HIZR-1(WT)::GFP protein that rescues the Zat-d phenotype, indicating it is functional and expressed in a pattern similar to endogenous protein. In animals cultured with no supplemental zinc, HIZR-1(WT)::GFP displayed low-level expression in intestinal cells; it was primarily localized in the cytoplasm and occasionally in nuclei. By contrast, culture in high dietary zinc resulted in striking HIZR-1(WT)::GFP accumulation in most nuclei of alimentary tract cells, primarily in the intestine (Fig 4A,4B and 4D) [33]. High dietary copper did not significantly affect the localization, demonstrating metal specificity of this response (Fig 4D).
To analyze the am285 gain-of-function mutation, we generated animals that express mutant HIZR-1(D270N GF)::GFP protein. When cultured with no supplemental zinc, HIZR-1(D270N GF)::GFP animals displayed significantly more nuclear accumulation of GFP than HIZR-1(WT)::GFP animals (Fig 4A,4C and 4D). Thus, the D270N amino acid substitution is sufficient to promote nuclear accumulation of HIZR-1 and transcriptional activation of zinc-activated genes, highlighting the critical regulatory role of the LBD. Together, these results support the model that zinc binding to the LBD causes nuclear accumulation and transcriptional activation of HIZR-1.
We hypothesized that HIZR-1 directly binds the HZA enhancer to mediate transcriptional activation, leading to the predictions that HZA enhancer DNA will interact directly with the DBD of purified HIZR-1 and that the HZA mediates the transcriptional activation activity of HIZR-1 in animals. An electrophoretic mobility shift assay (EMSA) was conducted using partially purified, full-length HIZR-1 protein and fluorescently labeled DNA. The 35 base pair DNA sequence was derived from the cdf-2 promoter and included the 15 base pair HZA with 10 flanking base pairs on each side. HIZR-1 protein retarded the migration of the HZA DNA in the gel, indicating a direct interaction (Fig 5A and 5B). The binding was saturable, and an apparent dissociation constant of 20.4 +/- 6.8 nM was calculated (Fig 5C). Unlabeled wild-type (WT) and mutant HZA were used as specificity controls. Unlabeled WT HZA was identical in sequence to the fluorescently labeled HZA while the unlabeled mutant HZA was identical except for randomizing the order of the central 15 base pair HZA. Unlabeled mutant HZA DNA did not effectively compete for binding to HIZR-1, indicating the binding activity is sequence specific (Fig 5D). These data demonstrate a direct, high-affinity, sequence-specific interaction between HIZR-1 and the HZA enhancer.
To investigate hizr-1(lf) mutations that result in substitutions of highly conserved amino acids in the DBD, we conducted EMSA assays with mutant proteins. All three mutant proteins displayed dramatically reduced DNA binding (Fig 5A and 5B). These biochemical and genetic analyses indicate that the DBD of HIZR-1 mediates the interaction with HZA DNA and DNA binding is necessary for the transcriptional response to high dietary zinc in animals.
To investigate the role of the HZA enhancer in animals, we utilized a promoter construct that contains three copies of the HZA enhancer upstream of a basal pes-10 promoter driving expression of GFP with a nuclear localization sequence (Fig 6A) [15]. Computational analysis revealed that the basal pes-10 promoter does not contain a recognizable HZA element, and experimental analysis demonstrated that the basal pes-10 promoter is not activated by high zinc [15]. In hizr-1(+) transgenic animals, 2% displayed GFP when cultured with no supplemental zinc, whereas 92% displayed GFP in high dietary zinc, demonstrating the promoter is significantly activated by high zinc. By contrast, 0% of hizr-1(am286lf) mutant animals displayed GFP in high dietary zinc, significantly less than hizr-1(+) (p < 0.001 by Chi-squared test). Furthermore, 93% of hizr-1(am285gf) mutant animals displayed GFP induction when cultured with no supplemental zinc, significantly more than hizr-1(+) (Fig 6B–6D, S4 Fig). These data show that hizr-1 was necessary and sufficient for high-zinc–activated transcription mediated by the HZA enhancer in animals.
In C. elegans, high zinc homeostasis is mediated by a parallel negative feedback circuit; high levels of cytoplasmic zinc increase expression of CDF-2 and TTM-1B, which detoxify zinc by sequestration and excretion, respectively [13,14]. Here we show that HIZR-1 plays a pivotal role in this negative feedback circuit by sensing zinc levels, binding the HZA enhancer, and promoting transcriptional activation of these genes. We noticed that the promoter of hizr-1 contains a predicted HZA element (Fig 7A), leading to the hypothesis that HIZR-1 activates transcription of its own promoter. Consistent with this model, the level of hizr-1 mRNA was significantly increased about 4-fold by high dietary zinc in wild-type animals (Fig 7B). This regulation appears to occur at the level of transcription, since a construct containing the hizr-1 promoter driving expression of GFP (hizr-1p::gfp) was also induced by high dietary zinc (Fig 7C–7E). By contrast, hizr-1(am286lf) transgenic animals containing this construct did not display increased GFP expression in response to high dietary zinc, indicating that hizr-1 is necessary for this transcriptional activation (Fig 7D and 7E). These results identify a positive feedback circuit, since HIZR-1 protein increases levels of hizr-1 mRNA, which in turn increases levels of HIZR-1 protein. This positive feedback circuit is embedded in and promotes the parallel negative feedback circuits—increased levels of HIZR-1 protein will enhance the activation of cdf-2 and ttm-1b mRNA, promoting zinc homeostasis (Fig 8).
Sensing high and low levels of zinc is critical for homeostasis. Here we identify the nuclear receptor HIZR-1 as a high zinc sensor in an animal. The DNA-binding domain of HIZR-1 interacted directly with the HZA enhancer in purified extracts, and hizr-1 functioned through the HZA enhancer in vivo to mediate transcriptional activation in response to high zinc. Thus, HIZR-1 appears to be the HZA-binding factor that was postulated by Roh et al. (2015) when the HZA was identified as the enhancer that mediates transcriptional activation in response to high dietary zinc [15]. Nuclear receptors are typically regulated by ligand binding, which promotes nuclear accumulation and transcriptional activation [27]. We propose that a ligand for HIZR-1 is zinc, and this model is supported by two lines of evidence. First, HIZR-1 accumulated in the nucleus and activated transcription in response to high dietary zinc. Second, the ligand-binding domain of HIZR-1 interacted directly with zinc with high affinity in purified extracts. This zinc affinity of the ligand-binding domain is specific for HIZR-1, since the related ligand-binding domain of DAF-12, which uses dafachronic acid as a ligand, did not bind zinc. Furthermore, our genetic studies demonstrate that the ligand-binding domain of HIZR-1 plays a key regulatory role, since a missense mutation in the domain causes a gain-of-function phenotype characterized by constitutive nuclear accumulation and transcriptional activation. The DNA-binding domain of HIZR-1 contains two predicted zinc finger motifs that are likely to bind zinc and promote DNA binding. While the results indicate the ligand-binding domain plays a critical regulatory role, the data do not exclude the possibility that zinc interactions with the DNA-binding domain also regulate HIZR-1 activity. These results establish the function and mechanism of action of hizr-1 as the direct sensor of high zinc and the effector of high-zinc–activated transcription.
Our results raise the possibility that HIZR-1 might respond to multiple metal ions. Metal binding by the HIZR-1 LBD binding was relatively metal-specific, as nickel and manganese did not compete effectively with zinc for binding. Interestingly, copper was able to compete for binding to the HIZR-1 LBD, similar to zinc. However, multiple lines of evidence suggest that copper is not a functional ligand for HIZR-1: (i) hizr-1(lf) animals were not hypersensitive to high copper toxicity, (ii) high copper did not stimulate nuclear accumulation of HIZR-1, and (iii) high copper did not induce the transcription of cdf-2, ttm-1b, or mtl-1 [15]. Cadmium is similar to zinc, but it is an environmental pollutant rather than a physiological metal. Cadmium activates gene transcription in worms, including some genes that contain HZA elements and also respond to zinc. Further studies are necessary to determine the role of HIZR-1 in cadmium-activated transcription.
Transcription factors that play a role in zinc homeostasis have been characterized in several eukaryotic organisms. In the budding yeast Saccharomyces cerevisiae the response to low zinc is mediated by the zinc-responsive activator protein 1 (ZAP1). The ZAP1 transcription factor binds directly to a conserved DNA element in promoter regions called the zinc-responsive element (ZRE) and thereby induces target gene expression [34]. ZAP1 target genes include the zinc importers ZRT1 and ZRT2 that are induced by ZAP1 to promote zinc uptake; ZRT3 is induced to mobilize zinc stored in the vacuole [35]. ZAP1 activity is repressed by high zinc [24] and activated by low zinc; this transcription factor contains multiple zinc finger domains that may play a regulatory role. In the fission yeast Schizosaccharomyces pombe, gene repression in zinc-replete cells is mediated by the Loz1 transcription factor [36,37]. The response to high zinc has been characterized in animals based on studies of the metallothionein genes. In mammals, a zinc-finger containing transcription factor called the metal-responsive-element-binding transcription factor-1 (MTF-1) directly binds the metal response element (MRE) in the promoters of metallothionein genes [16,25,38,39]. MTF-1 is necessary for the transcriptional response to a wide range of stresses including high cadmium, high zinc, hypoxia and oxidative stress caused by reactive oxygen species. The mechanisms of MTF-1 regulation are controversial, and it is unclear whether MTF-1 senses zinc directly or is part of a system that includes another high zinc sensor [40,41]. HIZR-1 is a new type of zinc sensor, and its discovery and characterization represent important advances in understanding mechanisms of high zinc homeostasis.
In response to fluctuating zinc levels, organisms maintain zinc homeostasis by regulating the abundance and activity of metallothioneins and zinc transporters. In C. elegans, exposure to high levels of zinc causes induction of CDF-2, which sequesters zinc in lysosome-related organelles, and TTM-1B, which excretes zinc into the intestinal lumen. CDF-2 and TTM-1B function in a parallel negative feedback circuit, since single mutants display moderate or undetectable hypersensitivity to high zinc, respectively, whereas double mutants display dramatic hypersensitivity to high zinc [14]. Here we elucidate new aspects of the homeostatic system by showing that hizr-1 mediates the transcriptional response of cdf-2 and ttm-1b. Furthermore, HIZR-1 activates transcription of its own mRNA, thereby establishing a positive feedback loop: high dietary zinc increases the activity of HIZR-1 protein, which increases the levels of hizr-1 mRNA and protein, resulting in a further increase in HIZR-1 activity. Because this positive feedback loop is embedded in a negative feedback circuit, it serves to enhance the overall negative feedback circuit.
The transcription factor ZAP1 is a key part of a similar feedback circuit in response to low zinc in yeast. Zinc deficiency causes activation of ZAP1 protein, which binds the promoter and activates transcription of the Zap1 gene; this autoregulation is a positive feedback loop [42]. Activated ZAP1 protein also increases the transcription of key zinc importers such as ZRT1 and ZRT2, representing the negative feedback component of this system[43,44]. The similarities between the circuits controlled by HIZR-1 in response to high zinc in animals and ZAP1 in response to low zinc in yeast highlights the utility and importance of embedding a positive feedback loop within negative feedback circuits to maintain homeostasis.
The discovery of HIZR-1 establishes a new intersection between two important fields that were previously separate: nuclear receptors and zinc biology. Nuclear receptors were originally defined as endocrine sensors in humans, including the glucocorticoid receptor and estrogen receptor [45,46]. Genome analysis identified a nuclear receptor superfamily consisting of about 49 members in mammals. Despite decades of intensive efforts focused on ligand discovery, about half of the nuclear receptors remain “orphans” with unknown physiological ligands. This raises the possibility that novel classes of ligands remain to be discovered. Indeed, the demonstration that zinc functions as a ligand for the HIZR-1 nuclear receptor represents the first example of a new class of physiological metal ligands. All previously described physiological ligands for nuclear receptors are small hydrophobic molecules; established ligand classes include retinoids [47,48], steroids [49], sterols [30,50], fatty acid derivatives [51], and other organic molecules, such as heme [52]. Interestingly, several metals have also been reported to bind to the LBD of the estrogen receptor and appear to alter its activity; however, these metalloestrogens are typically classified as endocrine disruptors rather than physiological ligands for the estrogen receptor [53]. The demonstration that a transition metal ion is the physiological ligand establishes a new structural class of nuclear receptor ligand molecules. Furthermore, this finding raises the possibility that nuclear receptors may be sensors of high levels of other essential metal ions, such as iron, copper, and manganese.
Nuclear receptors are not present in single-celled eukaryotes, such as yeast, and appear to have evolved in primitive multicellular organisms. Ancestral nuclear receptor genes exist in sponges, animals of the phylum Porifera, which lack higher-level body organization such as tissues and organs and thus lack hormone signaling [28,54]. Therefore, nuclear receptors were proposed to have ancestral roles as environmental sensing proteins that later evolved to sense intraorganismal endocrine signals [28]. However, no such functions have been rigorously demonstrated. Our results document a nuclear receptor that responds to a nutrient, consistent with the theory that the ancestral function of nuclear receptors might have been sensing dietary and environmental stimuli, including metals such as zinc. This represents a distinct paradigm from canonical endocrinology in which a hormone like estrogen is synthesized in endocrine cells, is secreted into the bloodstream, and enters distant cells, where it binds and activates the estrogen receptor [55]. Furthermore, this is a novel demonstration that a dietary nutrient is a direct ligand for a nuclear receptor. This establishes a new paradigm for nuclear receptors as direct sensors of environmental nutrients.
C. elegans strains were cultured at 20°C on nematode growth medium (NGM) seeded with Escherichia coli OP50 unless otherwise noted [56]. The wild-type strain was Bristol N2. The Zat mutations hizr-1(am279), hizr-1(am280), hizr-1(am285), hizr-1(am286), hizr-1(am287), and hizr-1(am288) are described here. These mutations were generated by mutagenizing the high-zinc reporter strain WU1391 (cdf-2p::gfp, see Plasmid DNA construction and transgenic strain generation for details) with ethyl methanesulfonate (EMS) [56] and identified by screening for abnormal patterns of fluorescence, as described below. hizr-1(gk698405) was identified by the C. elegans million mutations project and obtained from the Caenorhabditis Genetics Center [29]. To position newly identified mutations in the C. elegans genome, we used the following mutations on linkage group X that cause visible phenotypes: unc-115(e2225), egl-15(n484) [57], and sma-5(n678). To generate the high-zinc reporter strain WU1391, we utilized the unc-119(ed3) mutation [58].
To generate the transcriptional fusion constructs for cdf-2 (cdf-2p::gfp) (pSC24) and hizr-1 (hizr-1p::gfp) (pKW11), we polymerase chain reaction (PCR)-amplified DNA fragments positioned upstream of the coding sequence using wild-type C. elegans DNA. These fragments were ligated into pBluescript SK+ (Stratagene) containing the green fluorescent (GFP) coding sequence and the unc-54 3' untranslated region (UTR). The cdf-2 promoter was amplified from the ATG start codon to 1,371 base pairs upstream of the ATG start codon. The hizr-1 promoter was amplified from the ATG start codon to 441 base pairs upstream of the ATG start codon. To analyze the HZA enhancer, we used previously described transcriptional reporter constructs containing the basal pes-10 promoter driving transcription of GFP with a nuclear localization sequence (NLS) (pes-10p::gfp-nls) (pPD107.94, a gift from A. Fire) and the basal pes-10 promoter with three copies of the HZA enhancer inserted into the promoter (3XHZApes-10p::gfp-nls) (pID24) [15].
To generate the translational fusion construct for HIZR-1, [HIZR-1(1–412 WT)::GFP] (pKW1), we PCR-amplified the hizr-1 genomic locus from the C. elegans fosmid WRM069cE11. This fragment was ligated into pBluescript SK+ (Stratagene) containing the GFP coding sequence and the unc-54 3' UTR. The hizr-1 locus was amplified from 444 base pairs upstream of the ATG start codon to the TAA stop codon. The TAA stop codon was mutated to TAT to allow translation of the C-terminal GFP. To generate the HIZR-1(1–412 D270N GF)::GFP construct (pKW8), we modified plasmid pKW1 using Agilent QuickChange II Site-Directed Mutagenesis Kit according to manufacturer’s instructions.
To integrate the cdf-2p::gfp transcriptional fusion construct into the C. elegans genome, we ligated the DNA fragment encoding the cdf-2 promoter driving expression of GFP with the unc-54 3' UTR into the plasmid pMM016 that contains the wild-type unc-119 locus (unc-119(+)) [59] (pSC24). pSC24 was bombarded into unc-119(ed3) animals [13,59], and nonUnc animals that segregated only nonUnc self progeny were selected. The cdf-2p::gfp unc-119(+) transgene is integrated on the right arm of linkage group IV and was assigned the allele name amIs10. The following transgenic strains with amIs10 were used in this study: WU1391 (cdf-2p::gfp outcrossed seven times to N2), WU1518 (cdf-2p::gfp outcrossed seven times to Hawaiian CB4856), cdf-2p::gfp; hizr-1(am279), cdf-2p::gfp; hizr-1(am280), cdf-2p::gfp; hizr-1(am285). cdf-2p::gfp;hizr-1(am286), cdf-2p::gfp; hizr-1(am287), and cdf-2p::gfp; hizr-1(am288).
Transgenic animals containing extrachromosomal arrays were generated by injecting the gonad of worms with a plasmid of interest (pKW1, pKW8, pKW11, pID24, or pPD107.94) and a co-injection marker [60]. The co-injection marker was either myo-3p::mCherry (pCJF104) [61] or the plasmid pRF4 encoding the dominant ROL-6(R71C GF) mutant protein [60]. Transgenic animals were selected by mCherry expression in body-wall muscles or by the Rol phenotype. The following transgenic strains with extrachromosomal arrays were used in this study: hizr-1(am286); HIZR-1(1–412 WT)::GFP, hizr-1(am286); HIZR-1(1–412 D270N GF)::GFP, hizr-1(+); 3XHZApes-10p::gfp-nls, hizr-1(am286); 3XHZApes-10p::gfp-nls, hizr-1(am285); 3XHZApes-10p::gfp-nls, hizr-1(am285); pes-10p::gfp-nls, hizr-1(+); hizr-1p::gfp, and hizr-1(am286); hizr-1p::gfp. All transgenic strains contained the pRF4 dominant Rol marker except for hizr-1(+); 3XHZApes-10p::gfp-nls which contained the myo-3p::mCherry marker.
To generate plasmid constructs for protein purification of full-length HIZR-1, we PCR-amplified the complete coding sequence of HIZR-1 from synthesized DNA (IDT gBlocks). This fragment was ligated into pTrcHisA (ThermoFisher). This plasmid encodes an N-terminal 6 histidine affinity purification tag (6XHis) fused to amino acids 1–412 of HIZR-1 (pKW2). The pKW2 plasmid was modified using the Agilent QuickChange II Site-Directed Mutagenesis Kit according to manufacturer’s instructions to generate the plasmids pKW4, pKW5, and pKW6 that encode full-length HIZR-1 proteins with the G23E, S30L, and R63C amino acid substitutions, respectively.
To generate plasmid constructs for protein purification of the ligand-binding domains of HIZR-1 and DAF-12, we PCR-amplified the ligand-binding domains of HIZR-1 (encoding amino acids 101–412) and DAF-12 (encoding amino acids 440–753 of the A isoform) from synthesized DNA (IDT gBlocks). These fragments were ligated into pGEX-4T-1 (GE Healthcare). These plasmids encode an N-terminal glutathione S-transferase (GST) affinity purification tag fused to either the ligand-binding domain of HIZR-1 (pKW14) or DAF-12 (pKW15).
All plasmid constructs were verified by DNA sequencing using standard methods.
To position Zat mutations on the genetic map, we utilized single nucleotide polymorphism (SNP) markers [18]. To facilitate this strategy, we introduced the integrated cdf-2p::gfp reporter from WU1391 into Hawaiian CB4856 by performing seven backcrosses to CB4856 with selection for zinc-activated GFP fluorescence. All six Zat-c and Zat-d mutations displayed tightest linkage to the SNP amP117, positioned at approximately the 9,552 kilobase pair on linkage group X (Fig 1E).
To determine how many genes were affected by the five recessive Zat-d mutations, we performed complementation experiments. All five mutations failed to complement one another for the Zat-d phenotype, indicating that all five mutations affect the same gene.
To define intervals that contain the Zat-c mutation (am285) and the Zat-d complementation group (represented by am286), we conducted three-factor mapping experiments with mutations that cause visible phenotypes [56]. We chose the X-linked genes unc-115, egl-15, and sma-5 that are located at approximately the 10,147, 11,016, and 12,005 kilobase pairs on linkage group X, respectively. The cdf-2p::gfp reporter was introduced into the double mutant mapping strains to generate cdf-2p::gfp; unc-115(e2225) egl-15(n484) or cdf-2p::gfp; egl-15(n484) sma-5(678) using standard genetic techniques. For the am285 Zat-c mutation, 0/13 Egl nonUnc and 5/5 Unc nonEgl recombinants segregated the Zat-c phenotype, indicating am285 is positioned right of egl-15. 9/20 Egl nonSma recombinants segregated the Zat-c phenotype, indicating am285 is positioned between egl-15 and sma-5. For the am286 Zat-d mutation, 0/12 Egl nonUnc and 3/3 Unc nonEgl recombinants segregated the Zat-d phenotype, indicating am286 is positioned right of egl-15. 9/30 Egl nonSma recombinants segregated the Zat-d phenotype, indicating am286 is positioned between egl-15 and sma-5. The results that 18/50 recombination events (36%) occurred between egl-15 and hizr-1 and 32/50 recombination events (64%) occurred between hizr-1 and sma-5 are consistent with the molecular distances of approximately 302 kilobase pairs between egl-15 and hizr-1 (31%) and 687 kilobase pairs between hizr-1 and sma-5 (69%).
To identify the gene affected by the newly isolated Zat mutations, we performed whole genome sequencing using DNA from the am279, am280, am285, and am286 mutant strains. Candidate mutations within the mapping interval were identified by comparing the mutant DNA sequence to wild-type DNA sequence. am279, am280, am285, and am286 all contained candidate mutations in the gene hizr-1/(ZK455.6) (Fig 1E, S1 Table). Next, the sequence of the hizr-1 genomic locus was determined by standard sequencing techniques using DNA from am287 and am288 mutant strains, which revealed mutations in the hizr-1 locus.
RNA isolation and cDNA synthesis were performed as previously described [19]. Mixed-developmental stage populations of animals were cultured on NGM dishes. Animals were washed and cultured on Noble agar minimal media (NAMM) dishes with and without supplemental zinc sulfate. NAMM dishes were seeded with concentrated OP50 E. coli. After 16–24 h of culture on NAMM dishes, animals were collected by washing for RNA extraction. RNA was extracted using TRIzol (Invitrogen), treated with DNase I, and cDNA was synthesized using the High-Capacity cDNA Reverse Transcription kit (Applied Biosystems) according to the manufacturer’s instructions. Quantitative PCR (qPCR) was performed using a 7900HT Fast Real-Time PCR system (Applied Biosystems) and the SYBR Green PCR Master Mix (Applied Biosystems) following the manufacturer’s instructions. mRNA fold change was calculated using the comparative CT method [62]. For all qPCR experiments, mRNA levels were normalized to rps-23 and error bars indicate standard deviation. Forward and reverse amplification primers were: rps-23 5'-aaggctcacattggaactcg and 5'-aggctgcttagcttcgacac; cdf-2 5'-atagcaatcggagagcaacg and 5'-tgtgacaattgcgagtgagc; ttm-1b 5'-catgggcactcacacacacac and 5'-ctcggcgacccttttgatatttc; hizr-1 5'-tcattttgcggtttcatcgtg and 5'-catcgcgtgtatctacagctac; and mtl-1 5'-ggcttgcaagtgtgactgc and 5'-cctcacagcagtacttctcac.
Synchronized embryos were placed on dishes containing NAMM that were seeded with concentrated OP50 E. coli and supplemented with zinc sulfate (ZnSO4) or copper chloride (CuCl2) and cultured for 3 d [18]. Animals were then mounted on 2% agarose pads on microscope slides and imaged with a Zeiss Axioplan 2 microscope equipped with a Zeiss AxioCam MRm digital camera. The length of individual animals (tip of head to end of tail) was measured using ImageJ software (NIH).
Live hizr-1(am286); HIZR-1(1–412 WT)::GFP or hizr-1(am286); HIZR-1(1–412 D270N GF)::GFP transgenic animals (L4 or young adult) were washed and cultured on NAMM dishes with or without zinc sulfate for 12–16 h. Transgenic animals were then immobilized and mounted onto a microscope slide with a thin pad of 2% agarose. All images were captured using a Zeiss Axioplan 2 microscope equipped with a Zeiss AxioCam MRm digital camera using identical settings and exposure times in paired experiments (Fig 4A–4C). To score alimentary nuclei per animal with detectable HIZR-1::GFP (Fig 4D), we examined live animals using an Olympus SZX12 dissecting microscope equipped with GFP fluorescence and counted GFP-positive alimentary nuclei. Animals were exposed to 200 μM zinc or 300 μM copper because they are approximately equally toxic to C. elegans [63].
Transgenic L4 or young adult animals expressing cdf-2p::gfp (Fig 1A and 1B) or hizr-1p::gfp (Fig 7C–7E) were cultured for 12–16 h on NAMM dishes with or without supplemental zinc sulfate. Transgenic animals were immobilized, mounted, and imaged as described above. GFP fluorescence intensity was quantified using ImageJ software (NIH).
To determine the percent of GFP-positive animals in a population of transgenic animals expressing P3XHZApes-10p::gfp-nls or pes-10p::gfp-nls, we examined live animals using an Olympus SZX12 dissecting microscope equipped with GFP fluorescence. Animals displaying one or more GFP-positive alimentary nuclei were classified as GFP positive (Fig 6D). Representative images were captured using a Zeiss Axioplan 2 microscope equipped with a Zeiss AxioCam MRm digital camera using identical settings and exposure times in paired experiments (Fig 6B and 6C and S4 Fig).
Plasmids encoding full-length HIZR-1(1–412) proteins with the wild-type (WT) or mutant sequence (G23E, S30L, or R63C) fused to an N-terminal 6XHis tag were transformed into BL21 E. coli cells. The empty vector pTrcHisA was transformed as a control. Cells were grown in Luria-Bertani media at 37°C, and expression was induced with 5 mM IPTG when the absorbance at 600 nm reached between 0.5–0.7. Expression was induced for 16–18 h at 16°C. Cells were then collected by centrifugation and suspended in 50 mM MOPS (pH 7.0). Cells were lysed by sonication in the presence of 0.16 mg/mL of lysozyme. Lysed cell material was pelleted by centrifugation and the supernatant was collected. HIZR-1 protein was purified from the supernatant using Clontech TALON Metal affinity resin. GST alone and the ligand-binding domains of HIZR-1 and DAF-12 fused to GST were expressed and harvested using the techniques described above; these proteins were purified using Genscript Glutathione Resin according to the manufacturer’s instructions and eluted with 10 mM L-glutathionine.
To analyze zinc binding to proteins, we used zinc-65 radionuclide (PerkinElmer, stock date 2/12/2015, specific activity 3.29 mCi/mg, concentration 5.90 mCi/mL, and radionuclidic purity 99.00%) and the purified proteins GST::HIZR-1(101–412 WT), GST::DAF-12(440–753 WT), and GST alone. Protein concentrations were quantified by Bradford assay. Equilibrium binding experiments were performed according to established guidelines [64]. Briefly, a constant amount of zinc-65 (0.01 μCi) was incubated with variable protein concentrations in 50 mM MOPS buffer with 10 mM L-glutathionine (pH 7.0) (Fig 3A and 3B). Based on the specific activity, we calculated that addition of 0.01 μCi zinc-65 results in a final concentration of zinc-65 of about 0.4 nM and a final concentration of total zinc (radioactive and nonradioactive) of about 1 μM. The half life of zinc-65 is 244 d, and it decays to copper-65. The number of decay half lives between production of the zinc-65 solution and binding experiments was less than one. Therefore, the amount of copper-65 was negligible compared to the amount of nonradioactive zinc in the source. Although purified proteins were eluted in buffer with no added zinc, we cannot exclude the possibility that the purified proteins contributed some nonradioactive zinc to the reaction mixture. Alternatively, a constant amount of protein (0.77 μM) was incubated with variable zinc concentrations (S3 Fig). Therefore, reactions consisted of (1) protein in 50 mM MOPS buffer with 10 mM L-glutathionine (pH 7.0) and (2) zinc dissolved in water. Reactions were allowed to equilibrate for 40 min, protein was vacuum blotted onto nitrocellulose membranes, membranes were briefly washed, and bound zinc-65 was quantified using a Beckman LS600 scintillation counter. At least two technical replicates were performed for each unique protein or zinc concentration. The dissociation constant was determined using GraphPad Prism software, assuming a 1:1 binding stoichiometry between zinc and the protein of interest.
To conduct the metal selectivity experiments, the binding interaction between a constant amount of GST::HIZR-1(101–412 WT) protein and zinc-65 was competed using no competitor or 500 μM of either nonradioactive zinc sulfate (ZnSO4), copper chloride (CuCl2), nickel chloride (NiCl2), or manganese chloride (MnCl2) (Fig 3C). Binding values were normalized by defining the binding interaction between the protein and zinc-65 with no nonradioactive competitor as maximal binding and setting that value equal to 1.0. Fraction maximal binding was calculated as the bound zinc-65 (CPM) for a given nonradioactive metal divided by the bound zinc-65 for no competitor (CPM).
EMSAs were conducted using the Licor Odyssey EMSA Buffer Kit according to the manufacturer’s instructions. Images were captured with a Licor Odyssey Infrared Imager, and gel bands were quantified using Licor Image Studio software. To test the effect of the Zat-d (am279, am280, and am287) mutations on DNA-binding activity, a constant amount of full-length wild-type and mutant (G23E, S30L, and R63C) protein was incubated with a constant amount of labeled HZA oligonucleotide (Fig 5A). Protein concentrations were determined by Bradford assay and confirmed to be equivalent by western blot (Fig 5B). Proteins were visualized utilizing a Pierce 6XHis eptiope tag antibody with a Licor IRDye800 goat anti-mouse antibody. Blots were imaged with a Licor Odyssey Infrared Imager.
The dissociation constant was determined by incubating a constant concentration of full-length wild-type HIZR-1(1–412 WT) protein (pKW2) with variable concentrations of labeled HZA DNA (Fig 5C). The dissociation constant was calculated using GraphPad Prism software.
To conduct sequence specificity experiments, the binding interaction between a constant amount of HIZR-1(1–412 WT) protein and labeled HZA DNA was competed using variable concentrations of either unlabeled WT or mutant HZA DNA oligonucleotides (Fig 5D). The IRDye700 labeled HZA oligonucleotide was 5'- tgtgttatcaatcataaactagaacatgtctcgag-3'. The unlabeled oligonucleotides used were wild-type HZA, 5'-tgtgttatcaatcataaactagaacatgtctcgag-3' and Mutant HZA, 5'-tgtgttatcagaacatacaacattaatgtctcgag-3'. These DNA oligonucleotides were double stranded.
All data were analyzed utilizing the two-tailed Student’s t test of samples with unequal variance except for data in Fig 6D, which was analyzed by the Chi-squared test. When displayed, error bars indicate standard deviation. p-Values less than 0.05 were considered statistically significant. All the statistical comparisons in the current study consist of comparing two values made in different experiments performed in parallel. One type of comparison is wild-type animals compared to mutant animals, a pairwise comparison of two values determined in different experiments that were performed in parallel. A second type of comparison is the same genotype of animals compared in two distinct environmental conditions.
|
10.1371/journal.pntd.0005122 | Selective Inhibitor of Nuclear Export (SINE) Compounds Alter New World Alphavirus Capsid Localization and Reduce Viral Replication in Mammalian Cells | The capsid structural protein of the New World alphavirus, Venezuelan equine encephalitis virus (VEEV), interacts with the host nuclear transport proteins importin α/β1 and CRM1. Novel selective inhibitor of nuclear export (SINE) compounds, KPT-185, KPT-335 (verdinexor), and KPT-350, target the host’s primary nuclear export protein, CRM1, in a manner similar to the archetypical inhibitor Leptomycin B. One major limitation of Leptomycin B is its irreversible binding to CRM1; which SINE compounds alleviate because they are slowly reversible. Chemically inhibiting CRM1 with these compounds enhanced capsid localization to the nucleus compared to the inactive compound KPT-301, as indicated by immunofluorescent confocal microscopy. Differences in extracellular versus intracellular viral RNA, as well as decreased capsid in cell free supernatants, indicated the inhibitors affected viral assembly, which led to a decrease in viral titers. The decrease in viral replication was confirmed using a luciferase-tagged virus and through plaque assays. SINE compounds had no effect on VEEV TC83_Cm, which encodes a mutated form of capsid that is unable to enter the nucleus. Serially passaging VEEV in the presence of KPT-185 resulted in mutations within the nuclear localization and nuclear export signals of capsid. Finally, SINE compound treatment also reduced the viral titers of the related eastern and western equine encephalitis viruses, suggesting that CRM1 maintains a common interaction with capsid proteins across the New World alphavirus genus.
| Our data demonstrate that novel selective inhibitor of nuclear export (SINE) compounds reduced viral replication of three related New World alphaviruses, VEEV, EEEV, and WEEV, indicating that CRM1 is instrumental to their life cycle. The novel CRM1 inhibitors have a large selective index and represent a potential pan-antiviral therapeutic that targets the host’s transport proteins, which are hijacked by the New World alphaviruses.
| Endemic to North, Central, and South America, the New World alphaviruses cause a febrile illness that can progress to encephalitis with accompanying high morbidity and mortality rates in humans and equines [1]. Three viruses in particular, Venezuelan, western, and eastern equine encephalitis viruses (VEEV, WEEV, and EEEV), are of concern both as naturally emerging infectious diseases and potential bioweapons [2]. There are currently no FDA-approved antivirals or vaccines for use in humans. Considerable research effort has been aimed at studying the less pathogenic Old World alphaviruses like Sindbis Virus (SINV), creating a gap in knowledge of pathogenesis and therapeutic targets of New World alphaviruses [3].
Alphaviruses belong to the single-stranded, positive sense RNA family Togaviridae. Alphaviruses are divided into New World and Old World groups based on geography, disease progression, and protein function. Severe cases of Old World alphavirus infections present with arthritic symptoms, while New World infections often become encephalitic [4]. The alphavirus genome has two distinct regions. The nonstructural region encodes a polyprotein from the genomic RNA that is cleaved into the nonstructural proteins 1–4 (nsP1-4). The structural region codes a polyprotein from the subgenomic mRNA that is cleaved into the capsid protein, envelope glycoproteins, and several other small polypeptides [5]. The functional role of the viral proteins diverges between Old and New World alphaviruses. Type I interferons (IFN), IFN-α and IFN-β, attenuate alphaviruses, and Old and New World viruses have evolved to counter the host immune response. The VEEV nonstructural proteins inhibit STAT1 activation, preventing its nuclear localization and STAT1-dependent transcription [6]. The nsP2 protein of the Old World alphavirus Chikungunya (CHIKV) inhibits STAT1 activation by blocking its phosphorylation [7]. Binding to heparin sulfate alters the cellular tropism of EEEV, allowing it to evade the host immune responses [8]. Other mechanisms involve shutting down cellular transcription globally [2]. In SINV infection, host transcription downregulation is modulated by nsP2, a critical component of the virus’ replicative enzyme complex [9]. Conversely, VEEV, as a model for the New World viruses, uses its capsid protein to halt host transcription [10]. Capsid most likely achieves this by blocking nuclear import and export [11].
The nuclear pore is a complex macromolecular gate, regulating the nuclear entry and exit of most macromolecules including proteins >40 kDa. Transit through the pore requires carrier molecules called importins and exportins. Both nuclear transport processes are highly regulated and localization within the cell is an important mechanism for regulating macromolecular function. The major human nuclear export protein is exportin 1 (XPO1, also called chromosome region maintenance 1, CRM1). CRM1 mediates the nuclear export of over 200 proteins with canonical nuclear export sequences [12], along with some proteins that use adapter molecules and a small number of RNA transcripts. CRM1, plays a role in the life cycle of several viruses, such as HIV [13], dengue virus [14], and influenza virus [15, 16], and chemically targeting it has potential as an antiviral therapeutic. By forming a tetrameric complex with the host transport proteins CRM1 and importin α/β, VEEV capsid protein disrupts functioning of the nuclear pore complex [17]. Moreover, siRNA mediated knockdown of CRM1 resulted in confinement of VEEV capsid to the nucleus, whereas loss of importin α and/or importin β resulted in VEEV capsid being primarily cytoplasmic [18]. It is assumed that the capsid of EEEV, WEEV, and other New World alphaviruses behave similarly due to the phylogenic similarity of their capsid proteins [19]. Additionally, it has been demonstrated that the capsid protein of EEEV also inhibits host transcription [20] in mammalian, but not mosquito cells [21], similar to VEEV.
Our group previously investigated disruption of the VEEV capsid/host protein complex by chemically inhibiting host nuclear import and export proteins. Targeting the nuclear pore complex with the aim of inhibiting viral replication has been proposed by others [22, 23]. Inhibitors previously demonstrated to modulate viral protein binding to importin α and β1, mifepristone [24] and ivermectin [25], and the well-documented CRM1 inhibitor, Leptomycin B [26–28], altered the cellular distribution of VEEV capsid, reduced viral replication, and increased cell survival. As expected, mifepristone and ivermectin kept capsid primarily cytoplasmic, while Leptomycin B trapped capsid in the nucleus. Altering capsid localization reduced viral titers of both the VEEV vaccine strain, VEEV-TC83, and the wild-type strain, Trinidad Donkey (VEEV-TrD). Interestingly, inhibitor treatment also increased cellular viability in infected cells compared to infected cells treated with the vehicle only [18]. However, the traditional CRM1 inhibitor, Leptomycin B, is cytotoxic, creating the need for a new generation of CRM1 inhibitors.
Karyopharm Therapeutics has discovered potent, small molecule, orally available, slowly reversible inhibitors of CRM1, termed selective inhibitor of nuclear export (SINE) compounds. The most advanced drug in this class is selinexor, which is currently under clinical investigation for the treatment of patients with advanced hematologic and solid malignancies (ClinicalTrials.gov identifier: NCT01986348). After administration to over 1400 patients as of February 2016, selinexor (KPT-330) is showing preliminary evidence of durable anti-tumor activity in multiple cancer types. SINE compounds are well tolerated in vitro and in vivo [29, 30], and some of the compounds can cross the blood-brain barrier [12]. Due to previous successes in several viral models and their tolerability in human cancer clinical trials [31], three active compounds, referred to henceforth as KPT-185, KPT-335, and KPT-350, which are analogs of selinexor, and one inactive control compound, KPT-301, were tested for antiviral activity against three New World alphaviruses.
The selective CRM1 inhibitors KPT-335, KPT-350, KPT-185, and the inactive trans-enantiomer KPT-301 were provided by Karyopharm Therapeutics (Newton, MA). KPT-185 was characterized most extensively in in vitro assays, but has poor PK properties unsuitable for use in vivo. KPT-335 (verdinexor) and -350 are less potent against CRM1 than KPT-185, but have similar specificity, good oral bioavailability and good tolerability, thereby are suitable for future in vivo studies. KPT-335 (verdinexor) has been tested in a Phase 1 healthy volunteer clinical trial and found to be safe and well-tolerated (ClinicalTrials.gov Identifier: NCT02431364). KPT-350 has a higher brain penetration ratio than KPT-335 (verdinexor).
Vero cells and mouse embryonic fibroblasts (MEFs) were maintained as described previously [18]. BHK-J cells were maintained at 37°C, 5% CO2 in Eagle’s Minimum Essential Medium (EMEM) (ATCC, Manassas, VA, 30–2003) supplemented with 7.5% fetal bovine serum (FBS) and 1% penicillin/streptomycin.
SINE compounds were dissolved in sterile DMSO, diluted in DMEM supplemented with 10% FBS, 1% penicillin/streptomycin and 1% L-glutamine, and incubated on cells for two hours prior to viral infection unless otherwise noted. Following infection, inhibitor-containing media was added back to the cells and remained for the duration of the experiment unless otherwise noted. Likewise, cells pre-treated with Leptomycin B at 45 nM prior to viral infection were also post-treated after infection, unless otherwise noted. Leptomycin B was purchased from Sigma Aldrich (L2913).
VEEV-TC83 viral stocks were produced from electroporation of in vitro transcribed viral RNA generated from the pTC83 plasmid [a kind gift from Ilya Frolov, The University of Texas Medical Branch at Galveston [32, 33]]. In brief, the viral cDNA was linearized with MluI restriction enzyme (NEB) and purified using the Minelute PCR Purification kit (Qiagen) according to manufacturer’s directions. Capped RNAs were synthesized using the Sp6 MegaScript kit (Invitrogen) with a 2:1 ratio of cap analog (m7G(5')ppp(5')G [NEB]) to GTP and treated with DNase I supplied with the kit. RNA was then isolated with the RNeasy Mini kit with a second DNAse I on-column digestion (Qiagen). RNA integrity and concentration were determined by gel electrophoresis and absorbance at 260 nm, respectively.
In vitro transcribed viral RNA was electroporated into BHK-J cells (provided by Charles M. Rice) utilizing a 2 mm gap cuvette (BTX ECM 630 Exponential Decay Wave Electroporator; Harvard Apparatus, Holliston, MA). After trypsinization, cells were washed twice and resuspended in cold Dulbecco's phosphate-buffered saline without Ca2+/Mg2+ (RNase-free) at 1.25 × 107 cells/ml. An aliquot of the cell suspension (400 μL) was mixed with 1 μg of RNA transcripts, placed into the cuvette, and pulsed once at 860 V, 25 μF capacitance, and 950 Ω resistance. Cells were then left to recover for 5 min at room temperature and resuspended in complete MEM media (Gibco-Invitrogen). Cells from three replicate electroporations were plated in three 75cm2 culture flasks for virus production. Next day [~12 hours post electroporation (hpe)], transfection media was replaced with fresh MEM media. At 18, 24 and 30 hpe, media supernatants were harvested, pooled and stored at 4° C. After the last collection, supernatants were then filtered (0.2 μM), aliquoted, and stored at -80°C. Viral titers were determined by plaque assay.
The VEEV-TC83luc virus was kindly provided by Dr. Slobodan Paessler of the University of Texas Medical Branch, Galveston [34]. VEEV-TC83 and SINV (EgAr 339), both obtained from BEI Resources, and VEEV-TC83luc, were utilized under BSL-2 conditions. VEEV-TrD, EEEV (strain GA97), WEEV (strain California 1930), and CHIKV (strain S27) were utilized under BSL-3 conditions. Working stocks were prepared by infecting Vero cells at an MOI 0.1 and collecting supernatants at 48 hours post-infection (hpi). Supernatants were clarified by centrifugation at 10,000xg for 10 minutes at 4°C to pellet cellular debris and then filtered through a 0.2 μM filter. Supernatants were aliquoted into 0.5 or 1 mL user stocks and titer determined by plaque assay. All work involving select agents is registered with the Centers for Disease Control and Prevention and conducted at George Mason University’s Biomedical Research Laboratory, which is registered in accordance with Federal select agent regulations.
When performing infections, virus was added to supplemented DMEM at a multiplicity of infection (MOI) of 1, unless otherwise stated. The viral media was added to cells and incubated at 37°C for one hour with rotation every fifteen minutes. After infection, cell cultures were washed once with sterile 1X PBS and media containing inhibitors was added back onto the cells unless noted otherwise.
Crystal violet plaque assays were used to determine viral titer as described previously [18].
To determine inhibitor toxicity, Promega’s CellTiter Luminescent Cell Viability Assay (G7571) was used to measure cellular viability 24 hours after inhibitor treatment, as described previously [18].
Vero cells were grown on cover-slips in a 6-well plate and processed for immunofluorescence analysis as described [18]. Antibodies used included anti-VEEV-capsid (BEI Resources, NR-9403) goat primary antibody (1:1000 dilution) and Alexa Fluor 568 donkey anti-goat secondary antibody (1:500 dilution). Slides were imaged using an oil-immersion 60X objective lens on a Nikon Eclipse TE 2000-U confocal microscope, with all samples subjected to four line averaging. At least three images were taken of each sample, with one representative image shown. The resulting images were processed through Nikon NIS-Elements AR Analysis 3.2 software. Digitized images were analyzed as previously described using the ImageJ version 1.47 public domain software (NIH) [18]. Briefly, Fn/c was calculated according to the formula Fn/c = (Fn-Fb)/(Fc-Fb), where Fn is the mean fluorescence determined from a typical region within the nuclei, Fc is the mean fluorescence of cytoplasm determined in the same cell from a region of cytoplasm close to the nucleus, and Fb is mean background autofluorescence determined from a mock infected sample, fixed and stained in the same fashion and imaged under the same conditions as the sample slide. See S1B Fig for an example.
Supernatants were collected to analyze extracellular viral RNA and infected cells lysed and collected in Qiagen’s RLT buffer to analyze intracellular RNA. Extracellular viral RNA was extracted using Ambion’s MagMax Viral RNA Isolation Kit (AM1836); intracellular RNA was isolated using Qiagen’s RNeasy Mini Kit (74104). Q-RT-PCR was performed as described previously [35] for viral RNA using the Applied Biosystems StepOne Plus. Primer-pairs (forward TCTGACAAGACGTTCCCAATCA, reverse GAATAACTTCCCTCCGACCACA) and TaqMan probe (5′ 6-carboxyfluorescein-TGTTGGAAGGGAAGATAAACGGCTACGC-6-carboxy-N,N,N′,N′-tetramethylrhodamine-3′) against 7931–8005 of VEEV-TC83 were utilized [36]. The absolute quantification was calculated using StepOne Software v2.3 and based on the threshold cycle relative to the standard curve. The standard curve was established using serial dilutions of VEEV-TC83 RNA at known concentrations. To assay cellular genes, mRNA was converted to cDNA using the High Capacity RNA-to-cDNA kit (Life Technologies) according to the manufacturer’s protocol. Gene expression was assayed using the following TaqMan assays: IFIT1 (Mm00515153_m1), IFIT2 (Mm00492606_m1), IFNβ1 (Mm00439552_s1), and OASL1 (Mm00455081_m1). 18S ribosomal RNA (Mm04277571_s1) was used for normalization.
Supernatants and whole cell lysates (WCL) were collected twenty-two hpi. The WCL were processed as described previously [18]. To purify viral proteins, the supernatants were subjected to sucrose density centrifugation. Briefly, supernatants were spun at 2,500 x g for 30 minutes to remove debris. 20%-60% continuous sucrose density gradients were prepared with supernatants layered on top. Samples were centrifuged at 36,000 RPM at 4°C for 4 hours. One milliliter fractions were collected and the third and fourth fractions were collected for analysis. WCL and purified supernatants in lysis buffer were separated on NuPAGE 4–12% Bis-Tris gels (Invitrogen) and transferred to PVDF as previously described [37]. The primary and secondary antibodies used were: anti-capsid (goat primary, BEI Resources, NR-9403), anti-actin (Abcam, ab49900), and donkey anti-goat (Santa Cruz Biotechnologies, sc-2020).
Cells were infected with VEEV-TC83luc at a MOI of 1 in supplemented DMEM for one hour. Luminescence was measured as an indication of viral replication using Promega’s BrightGlo Luciferase Assay System (E2610). The assay was performed according to the manufacturer’s protocol using white-walled, ninety-six well plates (Corning, 3610) seeded with 10,000 cells/well at 16 hpi, unless stated otherwise. Beckman Coulter’s DTX880 Multimode Detector measured luminescence after 100ms integration per well. CC50 and EC50 were determined by Microsoft Excel or GraphPad Prism v. 6. Nonlinear regression lines were fitted to the data then transformed using the EC50 calculator.
Vero cells were pre-treated with DMSO or KPT-185 (2.5 μM), infected with VEEV TC83 (MOI 0.1), and post-treated with compounds. Supernatants containing infectious virus were collected 1 day post infection, titered (morphology of plaques noted), and used to infect a new set of Vero cells at an MOI of 0.1. VEEV-TC83 was continuously exposed to either DMSO or KPT-185 and passaged 10 times in this manner. Viral supernatants were frozen for every passage and cells were fixed, for capsid localization microscopy analysis, at every other passage beginning with passage 2. At the end of the experiment, a fresh set of Vero cells were infected with viral supernatants (MOI 0.1) from passages 1, 6, and 10. Total RNA was harvested at 6 hpi to isolate those viral genomes that retained infectivity. RNA was isolated utilizing the RNeasy Mini kit (Qiagen) according to manufacturer’s directions. One step reverse transcription (RT)-PCR was performed with 100 ng of total RNA to generate amplicons for sequencing. The SuperScript III One-Step RT-PCR system with Platinum Taq DNA Polymerase (ThermoFisher) was utilized to amplify a 1545 bp fragment encompassing capsid from nucleotides 6970 (nsP4) to 8514 (E3). The following cycling conditions were used: cDNA synthesis, 52°C for 30 min; heat denature, 94°C for 2 min; 40 PCR cycles, 94°C for 15 sec, 55°C for 30 sec, 68°C for 2 min; final extension, 68°C for 5 min. PCR products were then agarose gel purified (MinElute Gel Purification kit, Qiagen) and sequenced.
The KPT-185 adapted mutations in capsid (T41I, K64E, and K64M) were generated in the pTC83 plasmid. These plasmids and pTC83_Cm, previously described elsewhere [38], were cloned by PCR overlap extension using standard recombinant DNA techniques. All constructs were confirmed by restriction enzyme digestion and sequencing. Plasmid and primer sequences are available upon request.
Unless noted otherwise, mean values were compared using the unpaired, two-tailed student’s t-test available on QuickCalcs, GraphPad’s free online software.
Chemically inhibiting the host’s nuclear import and export proteins alters VEEV capsid localization [18]. Using the same rationale, novel CRM1 inhibitors were tested for their ability to alter capsid localization. Four SINE compounds (S1A Fig) were first screened for toxicity in two separate cell lines, treated for 24 hours, and determined to be non-toxic at 2.5 μM (S2 Fig, [39, 40]). Next, Vero cells treated with non-toxic concentrations of active CRM1 inhibitors (KPT-185, KPT-335, KPT-350), inactive KPT-301, or DMSO were infected with VEEV-TC83, and capsid localization assessed at 16 hpi by confocal microscopy. As seen previously [17, 18], VEEV capsid localized to both the cytoplasm and nucleus in DMSO-treated cells. A similar phenotype was seen upon treatment with the inactive compound, KPT-301. The three active compounds, KPT-185, KPT-335, and KPT-350, confined capsid to the nucleus (Fig 1A and S3A Fig), yielding a localization pattern similar to that seen with treatment by the well-documented CRM1 inhibitor, Leptomycin B [18]. As a control, TC83 virus with a mutated capsid that is defective in nuclear import (TC83_Cm; [38]) was analyzed. TC83_Cm infected cells treated with DMSO, Leptomycin B, or KPT-185 all displayed capsid cytoplasmic staining (Fig 1A). To quantitate capsid localization, the nuclear to cytoplasmic fluorescence ratio (Fn/c) was measured as described previously ([18] and S1B Fig). An Fn/c value greater than one denotes predominately nuclear fluorescence, whereas a score below one indicates predominantly cytoplasmic fluorescence. Treatment with the three active compounds, KPT-185, KPT-335, and KPT-350 had statistically enhanced nuclear accumulation of capsid and Fn/c values of greater than 4 (Fig 1B). In contrast, TC83_Cm infected cells treated with DMSO, Leptomycin B, or KPT-185 had Fn/c values of less than one (Fig 1C).
To ensure the results seen with VEEV-TC83 mimicked a wild-type infection, SINE compounds were tested against the fully virulent VEEV-TrD strain. Once again, DMSO and inactive KPT-301 treatment had little effect on capsid localization. The active compounds, KPT-185, KPT-335, and KPT-350, like Leptomycin B, resulted in capsid localizing predominately to the nucleus (Fig 2A and S3B Fig) and was statistically distinct from that of DMSO treated controls (Fig 2B). These results indicate that inhibition of CRM1 with the SINE compounds confined both attenuated and wild-type VEEV capsid to the cell nucleus.
Knowing that treatment with SINE compounds resulted in increased capsid nuclear localization, it was hypothesized that this would deplete the cytoplasmic pool of capsid, thus limiting viral assembly and decreasing the amount of virus produced. To test this hypothesis, the amount of viral RNA present at 4 and 8 hpi, within cells or in the extracellular supernatants, was assessed. Viral RNA detected in the extracellular supernatants would be virion-associated, while intracellular levels would represent viral RNA to be shuttled into viral protein translation, negative-sense strand RNA replication, or encapsidation into particles. The early time points were chosen to correspond to the first round of viral replication. Extracellular viral RNA was reduced upon treatment with Leptomycin B and KPT-185 at 4 and 8 hpi relative to DMSO controls (Fig 3A), with a greater relative reduction at 8 hpi. In contrast, at 4 hpi, intracellular RNA levels were approximately equal regardless of treatment, with a decrease seen by 8 hpi in both Leptomycin B and KPT-185 treated cells relative to DMSO controls (Fig 3B). Similar results were obtained at an MOI of 10 (S4 Fig). This indicates that SINE compounds likely disrupt a late step in the virus lifecycle such as viral assembly or budding.
To further validate that SINE compounds interfere with virus assembly or budding, the amount of released capsid was compared to intracellular capsid by western blot. Vero cells were treated with SINE compounds or DMSO as described, then infected with VEEV-TC83. Media containing compounds or DMSO were replaced and extracellular supernatants and whole cell lysates were collected 22 hpi. Supernatants were purified by sucrose gradient centrifugation [38]. Capsid was present in the supernatants of DMSO and KPT-301 controls, but not in Leptomycin B or KPT-185 samples. Capsid levels were also unaffected in whole cell lysates from DMSO and KPT-301 treated cells and minimally affected in Leptomycin B or KPT-185 samples (Fig 3C). Thus SINE compounds were not inhibiting viral protein synthesis. These data combined with the altered capsid localization and the observed decrease in viral RNA in extracellular supernatants, indicate that SINE compounds can sequester capsid, removing it from the pool of available structural proteins, leading to decrease in virus assembly and/or release of mature virions.
Since CRM1 inhibitors altered capsid localization and previous work has demonstrated a concurrent reduction in viral replication [18], SINE compounds were examined for their influence on VEEV replication. VEEV-TC83luc, which contains the firefly luciferase gene cloned downstream of a duplicated subgenomic promoter [34], was used as a reporter of viral replication. The active compounds, KPT-185, KPT-335, and KPT-350, and Leptomycin B reduced luminescence at 16hpi by approximately 80%, indicating treatment significantly decreased viral replication (Fig 4A). The inactive compound KPT-301 did not have any effect on viral replication. Importantly, KPT-185 was unable to reduce luminescence when cells were infected at a higher MOI (MOI of 5 or 10) and assayed at 8 hpi (Fig 4B). These data further support the hypothesis that SINE compounds inhibit viral assembly and/or budding.
Using the luciferase assay, the EC50, the effective concentration of a compound that reduces luminescence to half that of the DMSO control, was determined. The inactive compound, KPT-301, had an EC50 value of >3 μM (Fig 4D), but the three active compounds, KPT-185 (Fig 4C), KPT-335 (Fig 4E) and KPT-350 (Fig 4F) had EC50 values that ranged from 0.62 to 0.09 μM (Table 1). The CC50 for all four compounds was greater than 10 μM and therefore 10 μM was used to calculate the selective index (SI) for each compound. All three active compounds had SI values greater than 10, indicating the compounds are good therapeutic candidates. This set of experiments demonstrated the active SINE compounds inhibited VEEV replication in a dose-dependent manner with SI values that make them good therapeutic candidates.
As the SINE compounds reduced viral replication of the reporter virus, experiments were performed to confirm the inhibition results using an unmodified virus. Supernatants from TC83 infected cells were collected after 16 hpi and viral titers calculated using plaque assays. The active SINE compounds, KPT-185, KPT-335, and KPT-350, as well as Leptomycin B, were able to significantly reduce TC83 viral titers by approximately two logs (Fig 5A). As a control, cells were alternatively infected with TC83_Cm. Minimal differences in TC83_Cm titers were observed in the presence of Leptomycin B or KPT-185 (Fig 5B). Additionally, active SINE compounds reduced VEEV-TrD viral titers by approximately two logs (Fig 5C). Conversely, treatment with inactive KPT-301 yielded viral titers similar to that observed for DMSO alone (Fig 5A and 5C). Performing a similar experiment with supernatant collected at 8, 16, and 24 hpi with TC83 yielded results analogous to the 16 hpi experiments (Fig 5D), indicating the SINE compounds have an effect early on infection as well as at later time points.
All of the previous experiments involved pre- and post-treating cells with SINE compounds; however, treatment of cells after infection is more predictive of a compound’s therapeutic potential. To this end, Vero cells were infected with VEEV-TC83 for one hour, viral inocula removed, and cells washed prior to addition of media. After four hours, active SINE compounds, the inactive compound KPT-301, Leptomycin B, or DMSO was added to the media. At 8 hpi capsid localization and VEEV titers were determined. The active SINE compounds and Leptomycin B (Fig 6A and 6B and S3C Fig) all showed capsid localization patterns similar to that seen with pre- and post-treated cells (Fig 1). Post-treatment with DMSO had no effect on capsid localization, while KPT-301 (Fig 6A and 6B) had a slight but statistically significant effect on capsid localization. However, this effect was much less dramatic as compared to the active SINE compounds. Similarly, post-treatment of Vero cells infected with VEEV-TC83 significantly reduced viral titers by approximately one log (Fig 6C). Together, these results indicate that active SINE compounds may have therapeutic, in addition to prophylactic, potential.
Type I interferon signaling is important for the innate immune response to VEEV infection. As such, SINE compounds were examined for their influence on interferon stimulated genes (ISGs) following VEEV-TC83 infection. Interferon competent MEFs were selected for this analysis. Four ISGs, IFIT1, IFIT2, IFNβ, and OASL1, were analyzed following infection in the presence of DMSO or KPT-185. These four genes were selected as they have previously been shown to be induced following VEEV infection [41, 42]. All four transcripts were strongly induced following TC83 infection (Fig 7A). Cells treated with KPT-185 also displayed increased levels of all four transcripts, with only IFNβ being significantly reduced compared to DMSO treated cells. SINE compounds were also examined for their influence on these genes in the absence of infection (Fig 7B). While there was a slight reduction of IFIT1, IFIT2, and OASL1 expression in KPT-185 treated cells as compared to DMSO after interferon addition, this was not statistical significant. These results indicate that ISGs are capable of being induced in the presence of SINE compounds.
It was hypothesized that the SINE compounds should have a similar effect on other New World alphaviruses, assuming their capsid proteins also gain access to the nucleus and interact with CRM1 to egress from the nucleus [19, 21]. Frolova’s group demonstrated that VEEV capsid contains an NLS at amino acids 64–68 and a supraNES in the region of 38–55 [17]. Using UniProt, the capsid proteins of several New World alphaviruses were aligned. The capsid proteins of VEEV, EEEV, and WEEV share critical lysine residues [17] in the positively charged NLS (Fig 8A, residues highlighted in green). A consensus NES sequence (ФxxxФxxФxФ, where Ф = L, I, F, V or M and x = any amino acid) is present in VEEV, EEEV, and WEEV capsid proteins ([17], Fig 8A). A common NES region would suggest that capsid proteins from other New World alphaviruses would interact with CRM1 in a similar manner. To test this hypothesis, SINE compounds were examined for their ability to inhibit EEEV and WEEV. A statistically significant inhibition of approximately two logs was seen with the active SINE compounds for both EEEV (Fig 8B) and WEEV (Fig 8C), while inactive KPT-301 yielded similar results to DMSO alone. This suggests that the active SINE compounds are likely acting on a common mechanism among the New World alphaviruses. Similar experiments were performed with the Old World alphaviruses, SINV and CHIKV. No inhibition was seen with SINV (Fig 8D), but there was a small and statistically significant difference seen with CHIKV (Fig 8E). It can thus be concluded that inhibition of CRM1 has a more pronounced effect on New World alphaviruses.
To determine if adaptive mutations within the VEEV capsid open reading frame (ORF) could occur in the presence of the CRM1 inhibitor, VEEV-TC83 was serially passaged ten times in the presence of DMSO or KPT-185. Viral titers and plaque morphology for three replicates at each passage are presented in S1 Table. Capsid localization was assessed every other passage by confocal microscopy (S5 and S6 Figs). Plaque morphology, viral titers, and capsid localization of DMSO treated samples stayed consistent during passages 1–9. At passage (P)10, two of the three DMSO treated replicates displayed smaller punctate plaque morphology. For KPT-185 treated samples, starting at P5 and 6 for KPT-1 and KPT-2, respectively, plaques became cloudy and reminiscent of plaques formed by TC83_Cm virus. No changes in plaque morphology were observed with KPT-3. By P10 an increase in viral titers was observed for all three KPT replicates, suggesting that KPT-185 resistant viruses had developed. With respect to capsid localization, at P2, all KPT replicates had a mixed phenotype with either predominantly nuclear or cytoplasmic staining. However, by P6, all KPT replicates displayed predominantly cytoplasmic localization that was maintained for the remainder of the passages. Sequencing of the capsid ORF from viruses obtained from passages 1, 6, and 10 was performed for all replicates to determine what adaptive mutations had occurred (Table 2). No changes were found in the capsid ORF in TC83 passaged in the presence of DMSO. In contrast, all three replicates of TC83 passaged in the presence of KPT-185 displayed substitutions within the NLS or NES motifs. By P10, KPT-1 and KPT-2 both acquired a K64 mutation located within the NLS, disrupting positive charge at this position. The K64M mutation for the KPT-1 replicate was detected at P6 when plaque morphology was cloudy and capsid localization became largely cytoplasmic. In contrast, the KPT-2 replicate acquired a three amino acid deletion (61PSA) that was not stably maintained within the population and a K64E mutation was detected instead at/by P10. Finally, for the KPT-3 replicate, which maintained normal plaque morphology, but did have altered capsid staining, a mutation within capsid’s NES motif, T41I, was detected at P10. Next, to assess whether these mutations conferred KPT-185 resistance, three capsid mutated viruses (T41I, K64E, and K64M) were constructed. Cells pre- and post-treated with KPT-185 or the vehicle were infected with either wild-type TC83 (TC83-Wt) or one of the three mutated capsid viruses (Fig 9). All three capsid mutants were significantly resistant to KPT-185 treatment as compared to TC83-Wt, indicating that these mutations were at least partially responsible for the observed resistance after KPT-185 passaging. Collectively, these results demonstrate that maintenance of VEEV in the presence of the CRM1 inhibitor, KPT-185, can result in adaptive mutations that alter capsid localization and result in drug resistance.
CRM1, which regulates nuclear export of >200 proteins, including many viral proteins, has surfaced recently as an attractive anti-viral host target [29]. CRM1 is essential for some viruses’ life cycles through regulation of viral protein nuclear export. Karyopharm Therapeutics has developed a new class of CRM1 inhibitors, SINE compounds, with drugs from this class proving to be well tolerated and active in the clinic for canine and human cancers. In addition to the potent anticancer activity of these compounds, they have been shown to have substantial anti-inflammatory and neuroprotective activity in vitro and in vivo resulting from CRM1 inhibition. CRM1 mediates nuclear export of a variety of proteins important for inflammation, including the NF-κB pathway regulatory proteins IκBα, IκBε, RelA, p100, as well as COMMD1, HSCARG, Forkhead Box, and Nrf2 transcription factors, RXRα and PPARγ nuclear receptors and the chromatin binding protein HMGB1 [43–51]. It has been demonstrated that forced nuclear retention of many of these CRM1 protein cargos with the SINE compound KPT-350 leads to anti-inflammatory and neuroprotective effects [52]. SINE compounds also decreased expression of the pro-inflammatory cytokines IFN-γ, IL-1β, IL-6 and TNF-α in influenza A H1N1-infected mouse and ferret lungs [30]. Finally, toxicological studies in rats and monkeys have been initiated in preparation for clinical development. Many alphaviruses including VEEV manipulate the same pathways, as reviewed by Steele [53] and Suhrbier [54]. We have previously shown that chemically inhibiting modulators of pro-inflammatory cytokines reduces VEEV replication cycles and has some neuroprotective outcomes [35]. Future studies will examine the anti-apoptotic, neuroprotective, and anti-inflammatory qualities of the SINE compounds in the context of an alphavirus infection.
VEEV is often used as a model organism for New World alphavirus research. We have previously demonstrated the necessity of the CRM1/VEEV capsid interaction through siRNA silencing of CRM1 [18]. Here we have demonstrated that SINE compounds alter VEEV capsid localization, disrupting viral assembly and reducing overall viral levels. Based on EEEV and WEEV sharing conserved residues within the capsid NES, we hypothesized that the mechanism of interaction between host and viral proteins is common to New World alphaviruses. Under that assumption, it was demonstrated that EEEV and WEEV could also be inhibited using SINE compounds. This suggests that the CRM1/capsid interaction is important for the replication of New World alphaviruses, and disrupting it through chemical inhibition most likely touches upon a common virus/host interaction.
In contrast, nsP2 of Old World alphaviruses, such as SINV, have been shown to localize to the nucleus and induce transcriptional shutoff as opposed to capsid [10, 55]. This is achieved in part through nsP2-mediated ubiquitination of Rpb1, a subunit of the RNA polymerase II complex, leading to its degradation in both SINV and CHIKV infections [56]. A recent study by the Herchenröder group identified an NES in capsid at amino acids 143 to 155 for CHIKV [57]. Mutating this sequence confined GFP-tagged capsid to the nucleus, as did treatment with Leptomycin B [57]. Results from this study are in tentative agreement with the results presented within, as treatment with Leptomycin B and SINE compounds significantly reduced CHIKV, but not SINV titers. However, Leptomycin B and SINE compounds inhibited New World alphaviruses (VEEV, EEEV, and WEEV) to a greater extent than CHIKV. These results suggest that Old World alphaviruses are either less or not dependent on CRM1 for viral replication. As SINE compounds will also influence host protein nuclear trafficking, the inhibition of viral replication observed could also be due to inhibition of host protein trafficking, instead of viral protein trafficking, or a combination of both. Unfortunately, the lack of commercial antibodies for CHIKV and SINV capsid and nsP2 proteins makes further analysis difficult. Tagging the proteins is beyond the scope of this paper, but invites further consideration for future studies.
Our data suggest the SINE compounds reduce viral titer by interfering with viral assembly and/or budding. The Frolov lab has demonstrated that VEEV virions can assemble directly at the plasma membrane without preassembly of nucleocapsids in the cytoplasm [58]. The amino terminal domain of capsid is dispensable for packaging but is required for RNA encapsidation [58]. Packaging signals on the encapsidated, genomic viral RNA, 4–6 stem loop structures of GGG sequences at the base, are recognized by capsid [59]. Envelope glycoproteins journey from the endoplasmic reticulum to the Golgi network before arriving at the plasma membrane. Capsid associates with the E2 glycoprotein in the plasma membrane in a process extensively studied but still not well understood [60]. From our study, Leptomycin B or KPT-185 treatment resulted in a reduction of extracellular RNA and capsid levels at early time points post infection, while intracellular levels were minimally affected at 8 hpi. This indicates viral RNA replication and translation were unaffected. These data suggest that SINE compound treatment results in accumulation of capsid in the nucleus as opposed to the cytoplasm thus delaying viral genome encapsidation, as well as interfering with capsid/glycoprotein interaction and subsequent virus particle budding.
The emergence of drug resistant mutants as well as the inherent genetic variability of viruses are significant barriers to the prevention and treatment of viral infections that rely solely on directly acting antivirals. The use of inhibitors targeting static cellular factors that are required for viral replication and may affect multiple strains/genotypes has become an attractive possibility for standalone therapies or in combination with antivirals [61–65]. However, this approach has not been without caveats, most notably, that viruses can acquire resistant mutants that circumvent their dependency on these host factors for viral replication [66–68]. How high this barrier to resistance for host-targeting antivirals is, will likely depend on the presence of naturally occurring polymorphisms [68]. From our own studies with passaging VEEV-TC83 in the presence of KPT-185, we observed the development of NES/NLS mutations (K64E, K64M, and T41I) that disrupted the trafficking of capsid to the nucleus and resulted in increased titers. For a nuclear export inhibitor to select for compensatory mutations within the capsid NLS motif as well as the NES T41I mutation to alter capsid nuclear localization suggests that there is a close genetic linkage between the capsid NES and NLS motifs. This linkage is further supported by a prior report where deletion of subdomain 2 (aa 38 to 51) of capsid (VEEV/CmΔ2) containing the NES motif, lead to a compensatory mutation within the NLS domain, K64E [58]. The K64E mutation (VEEV Δ2ad) restored cytoplasmic accumulation of nucleocapsid to levels similar to wildtype VEEV and had increased titers as compared to VEEV/CmΔ2. VEEV capsid forms a tetrameric complex with the host’s nuclear import and export proteins, importin α/β1 and CRM1, which obstructs the nuclear pore complex [17]. Given the close proximity of the NLS and NES motifs in capsid, mutation of either site may alter local secondary structure leading to disruption of this tetrameric complex overall.
Because the capsid protein shares homology [19] between the three New World alphaviruses tested, we hypothesize that SINE compounds may also disrupt viral assembly in WEEV and EEEV infections. Future studies will focus on determining if SINE treatment affects these viruses in a similar manner to VEEV, including examining capsid localization and the effect on viral assembly. It is known that the capsid proteins of VEEV [10, 69], EEEV [10, 20], and WEEV [70] contribute to the inhibition of host transcription, dampening the innate immune response. In addition, EEEV capsid is found in the nucleus at early time points after infection [71]. Therefore, we speculate that modulating nuclear trafficking may be a common mechanism of New World alphavirus capsid proteins to dampen the cellular response to infection and contribute to pathogenesis. Further, if VEEV, WEEV, and EEEV capsid proteins all target the same host proteins, i.e. nuclear import and export proteins, then a common host-targeted therapeutic should be achievable.
|
10.1371/journal.pcbi.1005880 | Sequence-dependent nucleosome sliding in rotation-coupled and uncoupled modes revealed by molecular simulations | While nucleosome positioning on eukaryotic genome play important roles for genetic regulation, molecular mechanisms of nucleosome positioning and sliding along DNA are not well understood. Here we investigated thermally-activated spontaneous nucleosome sliding mechanisms developing and applying a coarse-grained molecular simulation method that incorporates both long-range electrostatic and short-range hydrogen-bond interactions between histone octamer and DNA. The simulations revealed two distinct sliding modes depending on the nucleosomal DNA sequence. A uniform DNA sequence showed frequent sliding with one base pair step in a rotation-coupled manner, akin to screw-like motions. On the contrary, a strong positioning sequence, the so-called 601 sequence, exhibits rare, abrupt transitions of five and ten base pair steps without rotation. Moreover, we evaluated the importance of hydrogen bond interactions on the sliding mode, finding that strong and weak bonds favor respectively the rotation-coupled and -uncoupled sliding movements.
| Nucleosomes are fundamental units of chromatin folding consisting of double-stranded DNA wrapped ~1.7 times around a histone octamer. By densely populating the eukaryotic genome, nucleosomes enable efficient genome compaction inside the cellular nucleus. However, the portion of DNA occupied by a nucleosome can hardly be accessed by other DNA-binding proteins, obstructing fundamental cellular processes such as DNA replication and transcription. DNA compaction and access by other proteins can simultaneously be achieved via the dynamical repositioning of nucleosomes, which can slide along the DNA sequence. In this study, we developed and used coarse-grained molecular dynamics simulations to reveal the molecular details of nucleosome sliding. We find that the sliding mode is highly dependent on the underlying DNA sequence. Specifically, a sequence with a strong nucleosome positioning signal slides via large jumps by five and ten base pairs, preserving the optimal DNA bending profile. On the other hand, uniform sequences without the positioning signal slide via a screw-like motion of DNA, one base pair at the time. These results show that sequence has a large effect not only on the formation of nucleosomes, but also on the kinetics of repositioning.
| Nucleosomes are the fundamental structural unit of eukaryotic chromatin, composed of approximately 147 base pairs (bp) of double stranded DNA wrapped around a histone octamer [1]. Nucleosomes enable genomic DNA to be folded into chromatin and efficiently packed inside the cell nucleus [2]. At the same time, because of the tight association with histones, nucleosomal DNA cannot be usually accessed by other proteins, inhibiting transcription factor association and gene expression [3,4]. Protein binding to a DNA region originally part of a nucleosome usually requires either complete nucleosome disassembly [5] or nucleosome sliding [6] away from the target sequence. The latter mechanism does not involve the complete breakage of histone-DNA contacts.
How are nucleosome positions regulated in the cell? Nucleosome assembly is strongly dependent on the underlying DNA sequence [7], and it has been shown that sequence indeed significantly contributes to the observed pattern of nucleosome positions in vivo [8,9]. However, in the complex cell environment many other factors will determine nucleosome positions. For instance, transcription factors will compete with nucleosomes to bind their specific target sites, due to the presence of steric hindrance [4]. Furthermore, many molecular machines, called chromatin remodelers, consume ATP to actively evict nucleosomes or reposition them to new sequence locations [10]. It has also been suggested that remodelers may be particularly important to enhance structural fluctuations, enabling a rapid search of the optimal nucleosome positions [8].
Nucleosomes may also undergo spontaneous repositioning in the absence of active remodelers [11]. While the importance of this mechanism in vivo has not been carefully investigated, many in vitro studies, e.g. using 2-dimensional electrophoresis [12] and atomic force microscopy [13], confirmed the existence of spontaneous nucleosome sliding. Moreover, experimental and theoretical work suggested that repositioning may occur via a corkscrew motion of DNA [14,15]. However, different repositioning mechanisms that do not involve this kind of rotation-coupled motion of DNA, such as DNA reptation via propagation of loop defects [16], have also been proposed. DNA sequence adds complexity to this problem: Genomes are rich in both positioning and non-positioning sequence motifs that enhance or inhibit nucleosome association [17], and these motifs will likely influence the dynamics of nucleosome repositioning [18].
Here we investigated thermally-activated spontaneous nucleosome sliding dynamics by employing a molecular dynamics (MD) simulation approach. While all-atom MD simulations have been widely used to study the molecular details of nucleosome conformation and histone-DNA interactions [19–22], it would be computationally challenging to reach the time-scales relevant to observe spontaneous sliding. On the other hand, recent studies have shown the effectiveness of coarse-grained (CG) MD simulation in investigating large biomolecular complexes such as nucleosomes [23]. In particular, the combination of the AICG2+ residue-level coarse-grained model for proteins [24,25] with the three-site-per-nucleotide (3SPN) model for DNA [26] proved to be a very successful strategy with many applications to nucleosome dynamics to date [27–29]. Notably, the latest version of the 3SPN model of DNA (3SPN.2C) [30] is designed to reproduce the sequence-dependent DNA flexibility [31], making the model suitable for the study of the influence of DNA sequence on the energetics of nucleosome formation [27].
In this work, we aim to reveal the dynamics of spontaneous nucleosome repositioning using the AICG2+ and 3SPN.2C coarse-grained models, and study the influence of DNA sequence on this process. To achieve this, an appropriate representation of histone-DNA interactions is of particular importance. Firstly, we developed a novel coarse-grained representation of histone-DNA hydrogen bonds, and showed that this potential, together with excluded volume and long-range electrostatics, is necessary to generate stable nucleosomes at low ionic strength and to reproduce the experimental unwrapping behavior observed at higher salt concentrations. Then, we performed MD simulations of nucleosomes using different DNA sequences, and identified two distinct sliding modes: one, coupled to DNA rotation, for uniform and non-positioning sequences, and a second, uncoupled to rotation, for a strong nucleosome positioning sequence. Finally, using a reweighting technique, we investigated the importance of histone-DNA hydrogen bonds in controlling these two sliding modes, finding that weak bonds favor the rotation-uncoupled mode, whereas stronger bonds favor the rotation-coupled one.
We examined nucleosomes with four DNA sequences of 223 bps; 1) the so-called 601 strong positioning sequence [32] that contains 145 bps (Table 1) [33], flanked by 39-bp linker DNA of polyCG sequence, 2) the polyCG sequence, where one strand has the sequence 5'-CGCG,,,CGC-3', while the other strand has its complementary sequence, 3) the polyAA sequence made of 5'-AAAA,,, AAA-3' and its complementary sequence, and 4) a modified polyCG sequence with the addition of two TTAAA positioning motifs at the same locations as those found in the 601 sequence, which we will refer to as the polyCG-601 sequence (Table 1).
Four nucleosomal DNA sequences used in this work in the range of +- 4 super helix locations. “TA” base-step motifs appearing every ~10 bps are represented as bold.
For protein modeling, we used the crystal structure with the protein-data-bank id 1KX5 as the reference histone octamer. For the DNA, we used the package 3DNA [34] to generate the reference structures for the 3SPN.2C model to optimally model the sequence-dependent geometric features and flexibility of DNA. To model histone-DNA interactions, we used the 1KX5 and 3LZ0 crystal structures, which are respectively based on the α-satellite [35] and 601 [33] positioning sequences (see section on hydrogen bond interactions for more details).
To prepare the initial structures (Fig 1A), we used the 1KX5 and 3LZ0 crystal structures. For polyCG and polyCG-601 sequences, we used the 1KX5 nucleosome structure that contains 147-bp nucleosomal DNA and added the extra DNA linkers by aligning the last base-pair of an ideal 39-bp segment of DNA to the last base-pair at each end of the nucleosomal DNA, resulting in the addition of two 38-bp DNA linkers at each end, reaching a total of 223 bps of nucleosomal DNA. For the 601 positioning sequence, we used the 3LZ0 nucleosome structure as a template, which has 145 bps instead of 147, and added 39-bp DNA linkers, obtaining the same DNA length of 223 bps. In all the cases, these initial structures are energy minimized using the steepest-descent method before production runs.
The histone octamer is modeled according to the AICG2+ potential [24,25], where each protein residue is coarse-grained to a single bead located at the corresponding Cα atom. The portion of the histone tails not resolved in the reference 1KX5 crystal structure is modeled using a statistical potential that reproduces the residue-dependent probability distribution of angles and dihedral angles between consecutive residues as observed in a database loop crystal structures [36].
To model the double-stranded DNA we employed the sequence-dependent 3SPN.2C coarse-grained model [30]. Within this model, each nucleotide is represented by three beads corresponding to sugar, phosphate and base groups. The model has been parameterized by matching the experimental DNA melting temperature, persistence length, and average base-pair and base-step parameters for all the ten unique base-step types [30,31]. The accurate representation of sequence-dependent effects is of particular importance for our investigation, and this model has already been shown to reproduce well the experimental dependence of nucleosome formation on the underlying DNA sequence [27].
The histone octamer interacts with the DNA via excluded volume, Debye-Huckel electrostatics and a novel coarse-grained potential representing hydrogen bonds, which we develop here. The excluded volume is modeled by a r12 repulsive potential, where the particle radii are bead-type dependent and they have been estimated from the minimum distances between each pair of bead types observed in a database of protein-protein and protein-DNA complexes [37]. With respect to the original parameters in Ref. [37], the radii have been uniformly rescaled by a factor of 1.1, which prevents the histone tails from being able to insert between two DNA strands. To represent long-range electrostatic interactions between histone proteins and DNA, and within the DNA, we used the Debye-Huckel approximation with a temperature- and salt concentration-dependent dielectric constant, as described in Ref. [26]. For DNA-DNA electrostatic interactions, following the recommended settings [26,30], we set a charge of -0.6e on each phosphate bead, which takes into account the Oosawa-Manning condensation of counter ions around DNA and it results in the correct DNA persistence length. On the other hand, for protein-DNA electrostatics, we set the phosphate charges to -1e as in Refs. [37] and [27], whereas the charges on the globular part of the histone octamer have been estimated using the RESPAC method [38]. In RESPAC, the coarse-grained charges (listed in S1–S4 Tables) are optimized so that the resulting electrostatic potential provides the best approximation to the all-atom electrostatic potential of the protein in the native reference 1KX5 crystal structure. The optimization procedure has been performed at 100 mM salt concentration, but the resulting charges show very low sensitivity to this particular value, so that the same set of charges can be used to run simulations at ionic strengths used in this work. The RESPAC method is only appropriate where the protein remains close to the reference native structure during the MD simulation, therefore for the flexible histone tails (up to the first structured alpha helix in the histone) we employed the standard residue unit charges: +1e for lysine and arginine, and -1e for aspartic and glutamic acids.
Many, but not all, coarse-grained simulations of nucleosomes reported so far employed Go-like potentials for the histone-DNA interaction to ensure that the nucleosome core structure is stable and close to the observed crystal structure. While this approach is convenient for the study of many important processes such as nucleosome breathing [28] and assembly [39], it cannot be applied to the study of nucleosome repositioning, since these potentials assume specific interactions at the prefixed positioning and thus are not invariant under DNA sliding with respect to the histone core. To overcome this limitation of standard Go potentials, we introduce a sliding-invariant coarse-grained potential representing the histone-DNA hydrogen bonds that stabilize the nucleosome structure. These hydrogen bonds are formed between a set of histone residues and the DNA backbone phosphates located at the half-integer super-helical locations, where the DNA minor groove faces the octamer. We set this potential to be invariant under a rotation-coupled repositioning of the DNA, since every phosphate bound to the protein will be simply replaced by a new phosphate with the same relative orientation.
To achieve this, we could in principle create the same Go-like contact between each protein acceptor and every single phosphate in the DNA. However, the problem with this strategy is that the standard 12–10 Lennard-Jones potential normally employed is too broad, and each hydrogen bond will be counted multiple times, not only between the protein residue and the correct phosphate, but also including other neighboring phosphates.
This issue can be overcome by making the potential highly specific, using both distance- and angle-dependence to represent the formation of a hydrogen bond (HB) (Fig 1D), of which idea came from recent coarse-grained DNA modeling [26]. In our model, we define the contribution to the potential energy from these bonds to be:
Vhb=∑∑ϵf(rij−rij,0)g(θij−θij,0)g(φij−φij,0)
where the first sum runs over the list of native HBs identified in the reference nucleosome crystal structures and the second sum runs over all the DNA phosphate beads, ensuring the invariance of the potential under a rotation-coupled repositioning of DNA. ε is the energy parameter that controls the hydrogen bond strength, rij is the distance between the Cα bead of the i-th HB-forming residue and the j-th phosphate, θij is the angle between the vector connecting the i-th HB-forming residue to the phosphate and the vector connecting the two residues neighboring the bond-forming one along the polypeptide chain (see Fig 1D), φij is the angle between the HB-forming residue, the considered phosphate and the sugar bead in the same nucleotide of the phosphate. The parameters with the subscript 0 are the corresponding distance and angles of the considered HB i found in the native structure, which are used as a reference to evaluate the formation of each specific bond. The functions f and g control the distance- and angle-dependence of the potential, which take a value close to 1 where the argument is close to 0, i.e. when the distance or angle variable observed during MD is close to the reference, and decreases as the argument deviates from zero. Their precise functional forms are given by:
f(r−r0)=exp(−(r−r0)2/σ2)
g(ϕ−ϕ0)={1for|ϕ−ϕ0|⩽Δϕ1−cos2(π2Δϕ(ϕ−ϕ0))forΔϕ<|ϕ−ϕ0|⩽2Δϕ0for2Δϕ<|ϕ−ϕ0|
Where the potential widths σ and Δφ, within which a bond is considered to be formed, are respectively set to 1 Å and 10 degrees. These values have been identified by requiring the HBs in the reference 1KX5 crystal structure to be well defined to their specific phosphate groups, without including favorable energetic contributions coming from the neighboring phosphates, which would amount to a double-counting of the interactions and would occur if the widths are too large. The energy constant ε was set to 1.2 kBT, which is about the smallest value required to stabilize the nucleosome structure against large deviations from the reference crystal, while still allowing the expected nucleosome disassembly at large salt concentration (see model validation). The list of HBs and their reference distance and angle parameters have been generated from the HBs found in any of the nucleosome crystal structures with PDB id 1KX5 and 3LZ0 using the software MDAnalysis [40] with the default settings. Flexible tails are excluded from the analysis because in the potential that evaluates the formation of a hydrogen bond we are assuming a stable near-native protein conformation (the bonds formed by the tails only account to a small portion of the total histone-DNA hydrogen bonds). The reference distance and angle values of each bond are obtained from an average over the two crystal structures and the two symmetric halves of the nucleosome for each structure.
All the CG simulations were performed by the CafeMol package version 3 [41]. The simulations were conducted by Langevin dynamics with default parameters at a temperature of 300 K. To find a proper ionic strength and validate the CG model, 10 independent 108 MD-step simulations of polyCG, polyAA and 601 nucleosomes were carried out at ionic concentrations from 100 to 1000mM of mono-valent ions. For the production runs, all the system is placed into a sphere with radius 80 nm and repulsive walls. The ionic strength was set to 200 mM, and we carried 100 independent 108 MD-step simulations for the sequences polyCG, polyCG-601 and 601.
To analyze the mode of nucleosome repositioning, we considered two angular coordinates: the sliding coordinate ζ and the DNA rotation coordinate η (Fig 1B and 1C). The former is defined by the angle between the vector from the histone core center of geometry to the center of the base pair initially at the dyad and the vector corresponding to the nucleosome symmetry axis; whereas the latter, η, is defined by the angle between the vector from the DNA axis to the 1st-strand phosphate of the base pair initially at the dyad and the vector from the DNA axis at the same base pair initially at the dyad to the histone core center of geometry. To extract the position of the DNA axis at each base pair, we firstly define 10 spline lines connecting each phosphate group of residue i in the first DNA strand with the phosphates of residues i-10 and i+10 (and i-20, i+20 and so on), representing contours on the DNA tube. Then, for each 1st-strand phosphate group, we compute the closest points on each of the 10 contours amongst the set of points obtained by subdividing each spline segment into 10 equal parts. Finally, for each base pair corresponding to the phosphate group, we define the DNA axis position as the center of the circle obtained from a fit of these 10 closest points. The nucleosome symmetry axis has been obtained by fitting a straight line to the set of the centers of geometry of the symmetric pairs of Cα beads in the histones (e.g. the center of geometry of the Cα beads with residue id 51 in the 1st and 2nd H3 histones, and similarly for other symmetric residue pairs; flexible tails were not taken into account).
In order to simplify the understanding of the sliding dynamics, we convert the unit of the sliding coordinate ζ from angles to number of base pairs. To do this, we first make a table that maps base pair indexes to ζ angles as obtained from the initial nucleosome configuration. For each snapshot in the trajectories, from the table, we then find the two neighboring nucleotides closest to the obtained ζ angle. Then we calculated the number of slid base pairs corresponding to ζ via linear interpolation between the two base pair indexes.
The highly-bent nucleosomal DNA is stabilized by a strong electrostatic attraction with histones and by more local interactions primarily via a network of hydrogen bonds to the histone octamer core residues [42]. Since we cannot employ the Go-like potential for the histone-DNA interactions in this study, an alternative and accurate modeling for the histone-DNA interactions is indispensable. Here, we test our CG modeling on nucleosomes formed with the 601 positioning sequence [32]. This sequence is well-known for the presence of several TA base-step positioning motifs (see Table 1) [8] which prefer to localize at nucleosome regions where the DNA minor groove faces the histone octamer, due to the intrinsic bending of DNA [27].
We began with a simple CG model where only the electrostatic interactions are included as the attraction between histones and DNA [27]. We performed CGMD simulations using the 601 sequence flanked by 39-bp polyCG sequences in both termini at the salt concentration of monovalent ions 200mM. With this condition, the nucleosome is stable experimentally. In the initial structure, the 601 sequence is wrapped by the histone octamer while the 39-bp polyCG segments form the linker DNAs. The resulting root-mean-square-deviation (RMSD) from the reference structure 1KX5 is plotted as a function of simulation timestep in Fig 2A for three representative trajectories (blue, red, and green curves). The RMSD was calculated for the central part of DNA (the segment that is initially located between -1 and +1 super helical locations) after alignment of the globular part of the histone octamer. First, we see that the native-like state with the averaged RMSD of ~ 0.5 nm is only marginally stable (only the green trajectory stayed in this state for significant time) (see Fig 2C). Instead, all the three trajectories stayed much longer time with the averaged RMSD of ~1.5 nm. In this state, the nucleosomal DNA is still well-wrapped around the histone core, while DNA is slid by ~5 bps so that the major groove positioned around the sites where the minor groove exists in the initial structure, likely representing an artifact due to inaccurate histone-DNA interactions (see Fig 2D). On top, occasionally, nucleosomal DNA went too far away from its favorable position (Fig 2E). Thus, we conclude that the long-range electrostatic attractive interaction alone is not specific enough to stabilize the nucleosomal DNA at high precision, even though the DNA can be well-wrapped around the histone core.
Next, in order to improve the representation of the system, we added the hydrogen bond (HB) potential, as well as the electrostatic interactions, to the histone-DNA interactions. The HB potential depends not only on the distance between the amino acid and the phosphate group of the interacting pairs, but also on the two related angles (see Methods), making the potential highly specific. For the same DNA sequence and the initial structure as above, we performed CGMD simulations with the HB potential. The representative RMSD time courses depicted in Fig 2B show stable fluctuations around ~ 0.5 nm throughout the simulation time. The nucleosomal DNA resides in the positions well close to those in the crystal structures. These results suggest that combination of the HB potential and the long-range electrostatic interaction is sufficient to stabilize the nucleosomal DNA at high precision in the current CG modeling.
Notably, in the above simulations, we utilized the HB strength parameter ε = 1.2 kBT. In some preliminary tests, we found that the ε smaller than 1.0 kBT lead to transitions to ~5-bp-shifted state. Instead, ε larger or equal to 1.2 kBT resulted in small fluctuations around the crystal structure.
As the second test of the histone-DNA interactions in our CG molecular model, we examined the dependence of nucleosome unwrapping stability on the ionic strength. We performed MD simulations for the three DNA sequences of 223 bps; the 601 positioning sequence, the polyCG sequence, and the polyAA sequence. Notably, while the polyAA sequence is known to have an inhibitory effect on nucleosome positioning in vivo, its ability to be incorporated into nucleosomes is well-documented [43,44]. We prepared the initial structures where the central 147 bps are wrapped by the histone octamer. For each DNA sequence and the varying salt concentration between 100 and 1000mM, we produced 10 independent MD simulations of 108 MD steps with different stochastic forces. To examine the convergence to the equilibrium, we also performed CG simulations from fully unwrapped DNA structures with the pre-formed histone octamer: we note that, in reality, the histone octamer is known to be unstable under physiological conditions without the nucleosomal DNA and thus, the initial structure used here does not represent the realistic state. The results indicated that the structural ensemble from the fully-unwrapped and fully-wrapped structures is nearly the same after about 2x107 MD steps. Thus, in the production run of 108 MD steps, the first 2x107 MD steps were discarded.
Fig 3 plots the resulting salt-concentration dependent DNA unwrapping from the histone core. At each salt concentration the average number of DNA phosphate groups within 1.2 nm from the globular part of the histone core was numerated (we note that, in this way, some phosphate groups in the linker DNA are assigned as contacted). These titration curves show standard sigmoidal shapes, in which the critical salt concentration that causes significant DNA unwrapping from the histone core is around 500 mM. This is consistent with the range between the ionic concentrations reported by experimental studies for the 601 sequence [45,46]. We performed additional simulations using the 601 sequence with several hydrogen-bond interaction strengths at the high salt concentration of 1000 mM. The average number of histone-DNA contacts at the end of long 108 steps simulations is plotted on the inset of Fig 3. These results show that complete nucleosome disassembly occurs for a wide range of hydrogen bond strengths, and only for ε greater than 2.4 kBT the nucleosome becomes clearly overstabilized. Thus, for the nucleosome with the 601 sequence being stable near the crystal structure at 200 mM and being disassembled at 1000 mM, the acceptable range of hydrogen-bond strengths may roughly be 1.2 kBT ≤ ε ≤ 2.4 kBT.
Notably, the 601 and polyCG sequences show a similar behavior in the unwrapping simulation, allowing us to compare the spontaneous sliding of polyCG and 601 at same ionic strength. In contrast, the polyAA sequence is less stable, which is in harmony with the well-known inhibitory effect of the polyAA sequence to form nucleosomes.
In order to observe thermally-activated nucleosome sliding, we performed larger-scale MD simulations of nucleosomes with linker DNAs. To investigate the DNA sequence dependence of the sliding, we use the three distinct DNA sequences of 223 bps (Table 1); the 601 positioning sequence (the same as above), a 2-bp periodic polyCG sequence (the same as above) as a representative of uniform non-positioning sequences, and a modified polyCG sequence with the addition of two TTAAA positioning motifs at the same locations as those found in 601 (polyCG-601 sequence). In order to allow space for sliding, polyCG and polyCG-601 sequences had 38 bps of polyCG linkers flanking each side of the central 147 bps of nucleosomal DNA; similarly, the 601 sequence had 39 bps of polyCG linkers and 145 bps of nucleosomal DNA (as found in the 3LZ0 crystal structure). We expected the polyCG sequence to be one of the most mobile sequences because the nucleosome energy landscape will be invariant under a DNA shift by 2 bps, as opposed to the 601 sequence, which has several TA base-pair step positioning signals placed roughly every 10 base pairs where the minor groove sharply bends towards the histone octamer. The polyCG-601 sequence is expected to display an intermediate behavior between the two. We performed 100 independent MD simulations of 108 MD steps at the salt concentration of monovalent ion 200 mM. To quantify the sliding, we defined the sliding coordinate as the angle ζ between the nucleosome symmetry axis and the vector from the histone core center of geometry to the base pair center (see Fig 1B and 1C). For clarity, we express the sliding coordinate in base pairs, mapping the angle ζ to the number of base pairs by using the initial configuration (see Methods for details).
Fig 4A plots, for the polyCG sequence, three representative time courses of the central nucleotide position, which was initially placed at the bp index zero. We find frequent and reversible transitions, which apparently suggest one-dimensional random-walk with the step size of one bp. We also plotted the residence probability of the central nucleotide during the entire time course (Fig 4D). Since the simulation started with the bp index zero, the probability distribution has a peak at zero bps and takes a Gaussian shape.
The polyCG-601 sequence also shows frequent repositioning events, but it is significantly more stable than the polyCG sequence (Fig 4B). Notably, the bp index often comes back to the initial position, suggesting that the sliding is limited up to ±4 bps. This is due to the TTAAA motifs optimally localized at the regions of high minor groove bending. The residence probability highlights this point, exhibiting a narrower and bound distribution.
In contrast to the former two sequences, the strong positioning 601 sequence shows nearly no sliding during the entire time course for many trajectories. Only in 14 out of 100 trajectories, however, we found abrupt transitions of ~5 bps. Once the 5-bp transitions occur, we often observe a subsequent 5-bp transition either to the same or to the opposite directions (see the blue trajectory in Fig 4C). In the 100 trajectories, we obtained 7 events to reach ±10-bp states. The residence probability distribution in Fig 4F clearly shows five peaks/modes. We note that, even though 5-bp-shifted state and 10-bp-shifted state have nearly the same probabilities, a close analysis suggests differently. Since the 5-bp-shifted state is next to the zero bps initial position, this state was visited more times than the 10-bp-shifted state, while the lifetime of 5-bp-shifted state was shorter. This clearly indicates that, in terms of the free energy, the 10-bp-shifted state is more stable than the high-energy intermediate of 5-bp-shifted state.
In summary, the repositioning behavior seems to be different among the three examined sequences. While the polyCG sequence slides frequently with the step size of one bp, the 601 sequence slides much rarely with a minor step size of 5 bps and a major step size of 10 bps. To our surprise, the sliding of the polyCG-601 sequence is not “in between” the polyCG and the 601: within the same simulation time, the sliding was the most restricted for the polyCG-601 sequence. Next, we analyze the DNA sliding motions in more detail.
We analyzed the motion of DNA in detail from the configuration of the DNA base pair initially located at the dyad relative to the nucleosome core. We considered two coordinates; the sliding coordinate and the rotation coordinate. The former is defined in the previous section whereas the rotation coordinate is defined by the angle between the vector from the axis of DNA to the phosphate of the base pair initially at the dyad and the vector from the axis of DNA to the histone core centroid (Fig 1C). If the nucleosomal DNA slides around the histone core in the screw-like manner, changes in the two coordinates would be coupled, otherwise not. In the cases of polyCG and polyCG-601 intermediate sequence, the time courses of the two coordinates are fully coupled (Fig 5A and 5B). On the contrary, in the case of the sudden jumps observed with the 601 sequence, the time courses of DNA sliding and rotation did not show any coupling behavior, with the DNA orientation remaining constant over time (Fig 5C and 5F). These results show that there are two distinct sliding modes, a rotation-coupled mode for polyCG and polyCG-601, and a rotation-uncoupled mode for 601. As a supplement, we provided two movies showing the structure of polyCG and 601 nucleosomes during repositioning events, where the base pairs with the same indexes as the TA motifs in bold in Table 1 are represented with red spheres.
To show the coupling more clearly, we plotted the logarithm of the probability distribution in the two dimensions; the rotation and sliding coordinates (Fig 6). In the case of polyCG, we see a series of high probability spots corresponding to every single bp sliding (Fig 6A). Interestingly, the high probability basins lie along a diagonal line with a slope of 360 degree rotation per ~10 bps sliding, corresponding to the DNA helical pitch. This signals a perfect rotation-coupled sliding of DNA that enables repositioning without inducing any distortion to the overall nucleosome conformation (Fig 6D and 6E). The polyCG-601 sequence displays a similar probability distribution to that of polyCG, with the difference that the introduction of the TTAAA motifs breaks the rotational invariance, still allowing the rotation-coupled motion but biasing it towards the initial optimal configuration (Fig 6B, 6F, 6G and 6H).
In the case of the 601 sequence, we find two distinct motions. First, we see five isolated islands, corresponding to -10, -5, 0, +5, and +10-bp-shifted states, which all have similar rotation coordinate values. Thus, the transitions among the 5 states are not accompanied by the DNA rotation. After sliding by 10 bps the nucleosome conformation will be similar to the optimal initial state, with TA base pair steps located at the inward-bending minor groove regions. Secondly, within each island, we find the rotation-coupled sliding up to ± 2 bps, which is qualitatively similar to that observed for polyCG and polyCG-601 cases.
From the free energy profiles in Fig 6, we identify five metastable states for the 601 nucleosome, corresponding to -10, -5, 0, 5, and 10 bps repositioning from the initial configuration. For shifts by 5 bps, the configuration looks rather different from the standard nucleosome crystal structures: here the minor groove locations correspond to what was initially occupied by the major grooves. In this intermediate state, long-range electrostatic interactions are not significantly affected; however, most hydrogen bonds are broken. In order to investigate how hydrogen bond interactions influence the rotation-uncoupled repositioning mode via this intermediate state, we performed the reweighting estimate of free energies along the sliding coordinate using different hydrogen bond strengths.
It should be noted that, in this analysis, we restrict ourselves to two states only, the 0-bp and +5-bp-shifted states, because of the limited sampling. To estimate the relaxation time scale within the two states, we performed additional 50 MD simulations starting from +5-bp-shifted state, and considering the 27 trajectories that came back toward the initial configuration we estimated the time scale for the transition from the +5-bp-shifted to the 0-bp state being 1.46x107 MD steps. Since the typical time for a transition from the 0 to the +5-bp-shifted state is orders of magnitude longer, the relaxation time within the two basins is well approximated by the former and shorter time scale. Thus, in the ensemble of the 100 independent MD simulations, we discarded the first 3x107 MD steps to remove the initial-configuration bias.
The resulting free energy profiles in Fig 7A show that increasing the hydrogen bond strength drastically affects the relative populations of 0-bp and +5-bp-shifted states in the rotation-uncoupled sliding mode: with a HB strength of 1.5 kBT (only 25% higher than the setting used in our MD), the free energy difference reaches 10 kBT, strongly inhibiting the rotation-uncoupled sliding mode.
For comparison, we also estimated the free energy profile of the 601 sequence within the central island along the rotation-coupled sliding up to ±1-bp-shifted states (Fig 7B). The results show that a change in the HB strength does not significantly affect the free energy profile for the rotation-coupled repositioning mode of the 601 sequence. This is reasonable because the rotation-coupled repositioning does not break the HB except at transient states.
Put together, these results suggest that for stronger hydrogen bonds, the screw-like rotation-coupled mode may become the main mode of thermally-activated spontaneous sliding of nucleosomes, even with strong positioning sequences such as 601.
We note that the rotation-uncoupled motion of 601 nucleosomes occurs via abrupt sliding events of ~5 bp, where the DNA moves at all the contact points with the histones almost simultaneously (S1A Fig). This is in contrast to a mechanism of sliding via DNA reptation proposed in the past [16]. According to this theory, sliding is initiated by the formation of a DNA loop defect at a nucleosome end during partial unwrapping; repositioning is then completed when the loop defect diffuses to the opposite end. While our simulations show that, for the systems studied, DNA reptation does not play an important role in spontaneous repositioning, it is important to discuss the possible reasons for this observation. To this aim, we run metadynamics [47] simulations to estimate the free energy cost to create a loop defect of 10 extra base pairs accommodated at one end of 601 nucleosomes (S2 Fig). This calculations gives us the relatively high value of ~15 kBT. This energy cost is due to both breakage of histone-DNA contacts and extra DNA bending. It is also worth mentioning that for the considered time scales and physiological salt concentrations, we did not observe the large partial unwrapping of DNA necessary to form such loop defect. Furthermore, we find that for MD simulations starting with a loop defect preformed at one end, the loop is rapidly dissipated on the same end (inverse rate of 8+/-1x104 MD steps), whereas it would rarely diffuse to the opposite end (inverse rate of 7+/-1x106 MD steps, obtained via MD simulations when we prevent loop escape from the closest end, see S1B Fig).
Using a 3-state model with the starting 601 crystal configuration, an intermediate state, and the final 10-bp-shifted state, we can compare the repositioning time scales via the three possible sliding routes; the observed rotation-uncoupled mode, where the intermediate state corresponds to the 5-bp-shifted state, the unobserved DNA reptation, where the intermediate is the conformation with a loop defect at one nucleosome end, and finally the rotation-coupled mode that is only observed in polyCG, where the intermediate is a 5-bp-rotated configuration. From the transition rates between the three states (S3 Fig and S1 Text for details on these calculations), we obtain that, for 601 nucleosomes, the transition times to reposition by 10 bp via the rotation-coupled and loop-defect modes are respectively ~1.5x1011 and ~2.2x1013 MD steps, whereas via the observed rotation-uncoupled mode is ~2.4x109 MD steps. Since the DNA reptation involves the breakage of histone-DNA interactions at the intermediate states, we expect that the screw-like motion of DNA will dominate over this mechanism even for higher hydrogen bond strengths.
The sequence dependency of the dominant sliding mode is a result of a balance between DNA’s local bending energy and hydrogen bond breaking penalty. The rotation-coupled motion of DNA does not affect histone-DNA hydrogen bonds, which have to be broken only temporarily when base pairs are exchanged, but it changes the bending profile of the DNA with a periodicity of about 10 base pairs. On the other hand, rotation-uncoupled sliding does not affect DNA bending, but it proceeds via a long-lived intermediate state where most hydrogen bonds are broken.
For a uniform sequence such as polyCG, the DNA-bending energy profile is invariant under a screw-like motion of DNA. Therefore, it is natural that this type of sequences will prefer to slide via a rotation-coupled motion that enables the nucleosome to maintain histone-DNA contacts at most times. This situation changes after the introduction of nucleosome positioning elements such as the TA motifs found in the polyCG-601 and 601 sequences considered here. TA motifs are highly flexible and they prefer to localize at the histone-DNA contact points where the DNA minor groove bends inward towards the histone core [8]. Starting from the initial optimal configuration, a DNA screw-like motion by 5 base pairs would bring these positioning motifs to the locations where the minor groove bends outwards, generating an unfavorable DNA bending profile. When many of these motifs are present, as in the 601 case, the bending energy penalty due to DNA rotation may become high, and repositioning proceeds instead via the observed rotation-uncoupled route. In this way, the optimal DNA bending profile is preserved at the expenses of hydrogen bonds, which are broken to form an intermediate 5-bp-shifted state. Hydrogen bonds are then restored after a further 5-bp jump.
The dynamics of poly-CG nucleosomes observed from our MD simulations is consistent with the proposed theory of repositioning via twist defects. While the analysis the motion of DNA at different contact points does suggest the formation of these defects (S4 Fig), our data are too noisy to make conclusive statements, and we plan to perform a more accurate study on the role of these structures in the future.
It is known that some histone arginine side chains inserted into the DNA minor grooves inhibit nucleosome repositioning [48], and one may ask whether an accurate representation of these side-chains in our coarse-grained model would prevent the sliding of 601 nucleosomes via the rotation-uncoupled mode. To discuss this possibility, similarly to what was done in Ref. [49], we back-mapped the coarse-grained configurations observed during the rotation-uncoupled sliding (see S1 Text for more details). The equilibration of the all-atom structures in explicit solvent shows that the atomic clashes, which would be observed in case of a rigid movement of the DNA relative to the histone octamer, can disappear if the arginine side chains are allowed to fluctuate out of the minor groove (S5 Fig). Therefore, while we do expect these side chains to slow down repositioning, their presence should still allow sliding via the observed rotation-uncoupled mode.
In our MD simulations, to keep the model simple and to enhance the observation of repositioning events, we employed the smallest hydrogen bond strength that can stabilize the nucleosome to the observed experimental crystal structure. Using these settings, we obtained results consistent with the experimental nucleosome unwrapping profiles as a function of salt concentration. However, fine tuning of the relative strength of nucleosome interactions at this level of coarse-graining represents a very challenging problem and we should consider the possibility that stronger hydrogen bond interactions may be more appropriate to represent the behavior of the system. Therefore, we reweighted the observed repositioning free energy profiles using higher hydrogen bonds strengths, and deduced how the two repositioning modes will be affected. Hydrogen bond strength has essentially no influence on the free energy profile of the rotation-coupled sliding observed in polyCG. On the other hand, for 601 we found that increasing the hydrogen bond strength by only 25% will increase the free energy difference between the optimal nucleosome conformation and the 5-bp-shifted state up to 10 kBT, a value comparable to the DNA-bending energy penalty that 601 nucleosomes should pay to reposition via the rotation-coupled mode [27]. Therefore, screw-like sliding may be the dominant mode even in the case of strong positioning sequences, as suggested by some authors [15].
During the review process, Lequieu et al. published a closely related study [50] investigating the sequence-dependence of nucleosome sliding. Using a coarse-grained computational model similar to the one employed here, the authors found results consistent with those presented here, with strong positioning sequences such as 601 sliding via large jumps separated by high free-energy barriers and non-positioning sequences undergoing via a screw-like motion of the DNA. Notably, the free energy surfaces reported in Ref. [50] seem to display local minima corresponding to the 5-bp-shifted metastable states that we have shown to play a role in the repositioning of 601 nucleosomes. The authors also report the spontaneous formation of loop defects, which can also be observed in our MD simulations using enhanced sampling methods. However, due to their low probability, these loops should not significantly contribute to repositioning under the conditions considered here. This discrepancy is likely due to the different modeling of the histone octamer: in Ref. [50] each residue bead is centered at the side-chain center of mass and standard unit charges are employed, whereas in our case residues are centered at the Cα atom, electrostatics is defined via RESPAC charges and flexible histone tails are modeled via a statistical potential. Despite this, the overall similar behavior observed in these two independent studies suggests the robustness of our results.
While the current study successfully observed two distinct sliding modes via direct CGMD simulations, many related computational studies remain to be done for more comprehensive understanding. First, since the spontaneous sliding is inherently a rare event, one can utilize advanced sampling methods to enhance the observation of sliding events, such as transition path sampling [51] and Markov state modeling [52]. Furthermore, while the current simulations assume a stable histone octamer, the model could be improved by including histone octamer disassembly [39], which could be of particular importance to study nucleosome distortions induced by the action of active chromatin remodelers [10]. Finally, it would be interesting to apply our model to study the effect of sliding on the dynamics of large-scale chromatin fibers [53,54].
In summary, we firstly developed a novel representation of protein-DNA hydrogen bonds at the coarse-grained level and confirmed their importance for nucleosome stability. Then, we investigated the kinetics of nucleosome sliding for 3 sequences and found 2 distinct repositioning modes: rotation-coupled and rotation-uncoupled. The sequence-dependent intrinsic flexibility of DNA determines not only nucleosome position, but also the kinetics of nucleosome sliding. The underlying mechanism to switch the sliding modes is a balance between hydrogen bond interaction and DNA bending energy. For non-positioning sequences, screw-like rotation of the DNA always represents the dominant repositioning strategy. However, when DNA has a strong intrinsic bending due to the presence of positioning motifs, the bending energy penalty due to DNA rotation may become too large, and a different rotation-uncoupled mode proceeding via the breakage of all histone-DNA hydrogen bonds may dominate. For both sliding modes, due to the free energy barriers of DNA bending or hydrogen bond breakage, spontaneous sliding of positioning sequences is much slower than that of uniform sequences, suggesting that the action of active chromatin remodelers may be required only when the nucleosome sequence is rich in positioning motifs such as TA base pair steps.
|
10.1371/journal.pcbi.1004740 | Cortical Resonance Frequencies Emerge from Network Size and Connectivity | Neural oscillations occur within a wide frequency range with different brain regions exhibiting resonance-like characteristics at specific points in the spectrum. At the microscopic scale, single neurons possess intrinsic oscillatory properties, such that is not yet known whether cortical resonance is consequential to neural oscillations or an emergent property of the networks that interconnect them. Using a network model of loosely-coupled Wilson-Cowan oscillators to simulate a patch of cortical sheet, we demonstrate that the size of the activated network is inversely related to its resonance frequency. Further analysis of the parameter space indicated that the number of excitatory and inhibitory connections, as well as the average transmission delay between units, determined the resonance frequency. The model predicted that if an activated network within the visual cortex increased in size, the resonance frequency of the network would decrease. We tested this prediction experimentally using the steady-state visual evoked potential where we stimulated the visual cortex with different size stimuli at a range of driving frequencies. We demonstrate that the frequency corresponding to peak steady-state response inversely correlated with the size of the network. We conclude that although individual neurons possess resonance properties, oscillatory activity at the macroscopic level is strongly influenced by network interactions, and that the steady-state response can be used to investigate functional networks.
| When entrained using repetitive stimulation, sensory cortices appear to respond maximally, or resonate, at different driving frequencies: 10Hz in visual cortex; 20Hz and 40Hz in somatosensory and auditory cortices, respectively. The resonance frequencies are inversely correlated to the cortical volume of the respective regions, but it is unclear what drives this relationship. Here we used both computational and empirical data to demonstrate that resonance frequencies are emergent properties of the connectivity parameters of the underlying networks. The experimental paradigm stimulated large and small areas of visual cortex with different size objects made of flickering dots, and varied the driving frequency. Larger cortical areas exhibited maximum response at lower frequency than smaller areas, suggesting the inverse relationship between cortical size and resonance frequency holds, even within the same sensory modality. Computationally, we simulated cortical patches of different sizes and varied their connectivity parameters. We demonstrate that the size of the activated network is inversely related to its resonance frequency and that this change is due to the increased transmission delay and greater node degree within the larger network. The results are important for understanding the functional significance of oscillatory processes, and as a tool for probing changes in functional connectivity.
| Although oscillations are observed in the human electroencephalogram (EEG) within a wide range of frequencies (1–100 Hz), each primary sensory cortex responds maximally to a specific frequency when driven by external repetitive stimuli using the steady-state evoked potential (SSEP). The SSEP is an experimental method of ‘tagging’ cognitive processing using a repetitive stimulus where the frequency response of the tag is augmented in the EEG spectrum and varies by task condition [1,2]. The driving frequency evoking the largest steady-state response, also known as the ‘resonance frequency’ [3,4], depends on which of the sensory cortices is being stimulated. In the visual cortex, the SSEP is greatest when the driving frequency is in the range of 10-18Hz, whereas in the somatosensory and auditory cortices the maximum response is between 20-26Hz and 38-42Hz, respectively [3,5–10]. This could be a mechanism by which each of these systems is able to extract sensory information optimally from its environment [4]. There is an inverse relationship between the size of each primary sensory area and its corresponding resonance frequency range. Anatomical studies have quantified the volume of the primary visual cortex in humans (Brodmann Area 17) to be approximately 23 cm3; that of primary somatosensory cortex (Brodmann Areas 1, 3a and 3b) to approximately 13 cm3; and of the auditory cortex (Brodmann Area 41) to approximately 3 cm3 [11–13].
It is not clear whether the observed inverse relationship between the size of a sensory region and its resonance frequency is due to differing physiological characteristics of the underlying neural components within each region, or due to emergent properties of the neuronal network connecting them. In physics, Mersenne’s Law is an inverse relationship between the fundamental frequency of a vibrating string and it length [14]. It is plausible that the frequency of oscillations within a cortical network is proportional to the length of the axons that connect them. Computational results from the study of synchronisation in networks of phase oscillators support the hypothesis that topological aspects of the network affect the oscillatory behaviour of individual units. Nordenfelt et al [15] found that the length of the coupling delays and degree of connectivity were negatively correlated with the oscillatory frequency of units within the network. These results raise the possibility that a similar relationship exists between the resonance frequency of a neural network and its connectivity parameters. We explored these theoretical results in an oscillator model designed to better describe neural activity and then extended the work to include experimental validation of these findings. Comparing resonance properties of functionally different brain regions becomes problematic due to the many structural and physiological differences that exist. Therefore, in this paper we chose to investigate network resonance within the same brain area and restrict our analysis to homogenous networks.
Initially, we implemented a model of loosely-coupled Wilson-Cowan (WC) oscillators intended to simulate a small patch of human primary visual cortex [16–18] to confirm the hypothesis that network size is inversely correlated with resonance frequency. In the human brain, anatomical architecture is fixed. Functional neural networks arise transiently according to the task being performed [19]. They are themselves sub-networks within the global network of the brain or brain-region within which they reside and are constrained by the physical connections already in existence. Therefore, it is impossible to completely isolate a cortical network from all external activity. Instead, neurons fire coherently maximising the efficiency of communication between synchronous units [20]. Therefore, we activated sub-networks of increasing size using repetitive stimuli allowing us to measure resonance. The network was designed to incorporate lateral inhibition of units surrounding the activated area [21,22]. The modelling work indicated an inverse relationship between network size and resonance frequency. Since network size also affects the average length of the transmission delays and degree of connectivity in a homogenous network, we conducted further simulation studies to describe the specific effect of each of these parameters on the resonance frequency of the network. The term ‘delay’ in this manuscript is defined as the interval of time that it takes a signal to propagate along the axon from one unit to another. For the purposes of our study, we assume that the conduction velocity is fixed and therefore the delay is proportional to the distance between two connected units.
Finally, we tested the model predictions experimentally using the visual SSEP (see Fig 1 for a schematic diagram of the experimental paradigm). Stimuli of different sizes, thus activating different sized patches of the primary visual cortex, flickered at different driving frequencies. The stimuli design and their presentation ensured that there was no change in the physiological properties of the constituent neurons of each of the activated networks. This allowed us to examine intra-subject differences in network resonance thereby controlling for biophysical factors such as individual myelination. The results confirmed the prediction of the model that the larger the size of the activated visual cortical network, the lower the resonance frequency. We suggest that resonance can be utilised as a marker for change in the underlying neural networks within the brain following disease, natural development or trauma.
We studied the change in resonance frequency in networks of different sizes using a homogenous network model of loosely-coupled identical Wilson-Cowan oscillators (see Materials and Methods). We first quantified the extent to which the resonance of a sub-network within our model was affected by its size. We also considered the effects of increasing transmission delay and node degree on the resonance frequency. The model prediction was that the relationship between size and resonance frequency would be preserved in cortical networks.
Two different sized network patches were stimulated with driving frequencies between 11 and 15 Hz. In Fig 2, the mean response power for each condition over 100 trials using the bootstrap method (see Materials and Methods), normalised with respect to the total power per trial, is given in blue with the standard deviation shown as black bars. We observed a shift in the peak response from 13.5 Hz in the small network to 12.5 Hz in the larger network. A two-way repeated measures analysis of variance (ANOVA) with factors driving frequency and network size was performed on the two peak frequencies 12.5 Hz and 13.5 Hz. There was no main effect of driving frequency (F = 0.07, p = 0.83) or size (F = 70.25, p = 0.08) on response power, but a significant interaction effect was found (F = 479.03, p<0.0001). There is a trend towards a main effect of size and this is due to the greater signal-to-noise ratio achieved in the larger network.
Certain properties of our network were observed to be dependent on size; specifically the mean node degree and transmission delay (see Fig 3A). When the network is small, both of these parameters grow steeply with increasing size, while for larger networks both curves grow more slowly. This points to the presence of a microscopic regime, within which the size of the network is comparable to the width of the Gaussian associated with the extent of short-range connections, and a macroscopic regime in which the sub-network size is significantly larger than the range of local connections. Given this relationship between the size of the network and both node degree and transmission delay, we considered the effect of each of these parameters on the resonance of the network.
For simplicity of simulation we first fixed the mean node degree, along with all other parameters within the model (see Table 1), while transmission delay was varied in proportion to distance from 10 to 100 time steps. A longer delay implies a larger network in terms of distance between the units. We found that longer transmission times were associated with a lower resonance frequency (Fig 3B) (R2 = 0.62, p < 0.01).
Next, the effects of increased local node degree, the number of local connections per node or unit, were tested separately. In this simulation we assumed instantaneous transmission of information to ensure that there was no increases in mean delay as the number of connections grew. The mean node degree is defined as the average number of connections per unit. In the following simulations, we define excitatory (inhibitory) node degree to describe the average number of forward local excitatory (inhibitory) connections for each unit. We distinguished between the effects of the two types of local connections: excitatory and inhibitory by fixing one at a value of 5 and varying the other between 0 and 9. Long-range connections remained fixed. When the number of local excitatory connections per unit was fixed, the effect of increasing the inhibitory connections had a positive correlation with network resonance, Fig 3C (R2 = 0.91, p < 0.0001). Conversely, when the number of local inhibitory connections per unit was fixed the degree of excitatory connections had a strong negative relationship with frequency, Fig 3D(R2 = 0.80, p < 0.001).
We created visual stimuli of different sizes that were designed to activate specific target patches within the visual cortex in order to mirror the experimental work. This was possible due to the precise nature of retinotopic mapping of the visual field onto the cortex [23]. Four image types were presented, a Full Thick annulus (FK), a Full Thin annulus (FN) of half the area of FK, and two half annuli, Left Half (LH) and Right Half (RH) of half the area of FK (see Fig 4A). This design ensured that size was controlled for and that we could test for differences due to inter-hemispheric connectivity. Each of the stimuli flickered at ten different driving frequencies ranging from 8.2 Hz to 17.5 Hz and the response power at the driving frequency was recorded via EEG from the electrodes over the occipital cortex.
The largest response frequency for the FK stimulus was 8.8 Hz, 10.0 Hz for the FN stimulus, and 10.0 Hz for the mean response of the LH and the RH stimuli (denoted LH+RH). The small stimuli (FN, LH+RH) were all of equivalent size and it is interesting to note that they have the same mean response peak at 10.0 Hz, suggesting an undetectable effect of intra-cortical connections. These results indicate that the cortical networks within human V1 exhibit resonance-like properties which can be observed using the SSEP. Furthermore, the negative relationship between cortical network size and resonance is clearly visible. A two-way repeated measures analysis of variance (ANOVA) was performed using the two factors: driving frequency and stimulus type for individual response powers at the two peak frequencies 8.8 Hz and 10.0 Hz. There was no main effect of frequency or stimulus on response power, but a significant interaction was found for stimulus types FK as compared to FN (F = 5.5, p = 0.02) and FK compared to LH+RH (F = 3.82, p = 0.03). These results provide evidence that a larger cortical network size results in a lower resonance frequency. The grand average scalp topographies are given for the four conditions at a single stimulus frequency (9.3 Hz). The power at the stimulus frequency normalised by the total power at each electrode can be seen in Fig 4B.
The stimulus images were made up of green and red dots, one of which was slightly dominant; this formed the basis of an attentional task which ensured that the subjects stayed focussed and alert and in addition strengthened the signal [2,24,25]. Participants were asked to decide after each trial whether the object on the screen consisted of primarily green or red dots. The mean scores per stimulus type and frequency are shown in Fig 4D with the standard error given as bars. Individual scores ranged between 40% and 100% correct. A two-way repeated measures analysis of variance (ANOVA) was performed using the two factors: driving frequency and stimulus type for individual scores at all frequencies to test whether there was any significant effect of either of these on the subjects’ scores. We found a main effect of stimulus (F = 3.29, p = 0.02), no effect of stimulus frequency (F = 1.53, p = 0.13) and no interaction effect (F = 0.47, p = 0.99). The subjects scored highest with the largest stimulus (FK) which was expected. It can be seen that the highest number of correct scores for the FK condition was achieved for a stimulus frequency of 8.8 Hz, for the FN condition at 11.7 Hz and for the LH+RH condition 10.0 Hz. For the FK and LH+RH conditions, these exactly match the greatest response frequencies suggesting that there is a potential improvement in cognitive abilities for these cases.
Detailed biophysical studies have shown that even the most basic units of the neural system, neurons, possess resonance characteristics. This is due to the low-pass filtering effects of the leaky conductance channels of the cell membrane coupled with the high-pass filtering effects of the voltage-gated current systems [4]. This study was concerned with understanding whether cortical resonance is affected by global properties of the activated network. We examined the possibility that the resonance frequency of a neural network (both computational and actual) is influenced by its size. The results from our modelling work indicate that when a region of simulated cortical sheet is activated, the resonance frequency of that region is inversely proportional to its size. We found a specific relationship between resonance and the number of excitatory and inhibitory connections and the average transmission time between units within the network. The prediction of our model was confirmed experimentally in the context of the primary visual cortex using a visual steady-state evoked potential paradigm. These results suggest that resonance in the cortex is not simply the consequence of oscillatory properties of the individual neurons but that network structure plays an important role.
The modelling work indicated that the number of excitatory and inhibitory connections had the strongest effect on the resonance of the network. This is in line with recent work exploring the link between intrinsic oscillator frequency, phase-synchronisation and time-delayed coupling in a sparsely connected network of phase-oscillators [15]. Nordenfelt et al investigated the effect of increasing time-delays between units in the network and found that this had a negative effect on the mean oscillatory frequency output of the model. They also classified the average frequency of units within the system according to their node degree and found that as the number of connections increased, the average frequency of that unit decreased. These are in direct agreement with our results. While the model described in the aforementioned paper is a phase-oscillator model, the WC model that we used in our study provides a more realistic description of neural tissue dynamics.
The human visual SSEP experiment confirmed the model prediction and provided further evidence that network size is related to cortical resonance. Interestingly, there was no significant difference in maximum response frequency between the three smaller stimuli that we used; FN was designed to activate the same total area within V1 as each of LH and RH but it activated cortical sites within each hemisphere. We expected that the FN network would encompass longer average transmission delays than LH or RH and that this would be reflected in the resonance frequency of the network. It could be that this was the case but our frequency resolution was not fine enough to elucidate this difference due to technical limitations.
In our work we define local dynamics as describing the behaviour of a single macrocolumn within our model or within the brain. The changes we observed are due to the increased number of connections and transmission delays between local units as the size of the network grew. Although we use the term ‘global’ to describe these changes at the network level, we confine this definition spatially within the brain to the visual cortex. We assume that our SSEP signal is mainly generated by locally entrained oscillators within the visual cortex and this assumption is strengthened by the fact that we observed no difference in resonance frequency for the networks activated by (LH+RH) and FN. These were the same total area but the latter encompassed cross-hemispheric connections. There is evidence that SSEP brain resonances can be local and global and that they are generated by distributed sources throughout the brain [26]. For example, the visual SSEP is observed at frontal locations under specific experimental conditions [27–29]. It is likely that we are measuring a combination of these phenomena although it should be noted that sources contributing to our signal from regions external to the visual cortex due to higher-level processing are controlled for across stimulus types.
A final point to consider is that we included stimulation frequencies outside of the natural alpha range. The strong endogenous parietooccipital alpha rhythms that exist in the absence of any stimulation leave open the question of whether stimulating with frequencies within and outside of this range may have very different biophysical bases and could be a confounding variable in the data [30,31]. Global activity of this kind is thought to modulate local activity through mechanisms such as phase-coupling [32] which has implications for the observed SSEP signal. Given that we are not trying to measure or affect behavioural or perceptual changes we feel confident that our use of the SSEP as a frequency-tag is justified in this instance.
Other authors have found evidence of a link between the frequency characteristics of neural oscillations and the spatial extent of the cortical tissue over which they traverse. For example, a recent computational paper determined that the ‘spatial reach’ of low-frequency components of the local field potential (LFP) was much larger than that of the high-frequency components [33]. In vivo and in vitro experimental work in animals has shown that slower oscillations such as alpha generally engage larger networks over more distant regions whereas higher frequency oscillations, such as gamma, are confined to local networks [34,35]. Similar results have been found in human EEG studies; local visual processing was found to occur as fast gamma oscillations whereas binding activities involving disparate areas occurred in the slower frequency ranges [36]. In general, there seems to be a consensus that long-range communication is facilitated by low-frequency oscillations whereas local communication and synchrony is brought about by high frequency activity [37]. Our study is the first to demonstrate that the activation of networks of different size within the same cortical area results in a change in resonance frequency for those networks. Based on these findings, we suggest that resonance is in fact not simply a property of the neurons within a network but also an emergent feature strongly modulated by the network itself.
The observation of an inverse relationship between network size and frequency can have a diagnostic or predictive utility. The reorganisation of brain networks, either due to slow changes in the relative contribution/wiring of brain areas (plasticity) or due to fast modulation of their causal interactions (effective connectivity), underpin the early stages of a number of developmental (e.g. Dyslexia) and psychiatric (e.g. depression, psychosis), but most prominently neurodegenerative (e.g. Alzheimer’s or Parkinson’s) conditions, ahead of any observed changes in behaviour [38,39]. A shift in the peak steady-state response frequency would be a marker of change in connectivity. A shift to a higher frequency predicts a shrinking network, while a shift to a lower frequency suggests a denser network.
The results leave open the question of multi-sensory integration. The sensory cortices might be optimally tuned to collect information from their designated environment, but how and at which frequency bands would multi-sensory stimuli be integrated remains unclear. Although we have assumed a homogeneous network in this paper, for multi-sensory integration, a neural field model would be more suitable in order to account for the heterogeneity of the systems involved. It should also be pointed out that the original WC model has been shown under certain parameter values to generate negative fractional firing rates, hence can be physiologically unrealistic. Recent work has been done in this field to make this model more biologically plausible by explicitly incorporating background variables into the equations [32].
In conclusion, we have proposed a method for exploring resonance in cortical circuits using repetitive sensory stimuli. It is known that changes associated with several brain diseases, brain traumas, as well as normal development affect brain structure. We suggest that this tool could potentially be implemented to better understand these changes quantitatively. Future work in this area will incorporate stimulus systems capable of presenting stimuli at a finer temporal scale than used here in order to overcome the technical limitations mentioned above.
We used the WC model as the basic unit of our network [16,17]. The WC equations describe a simple model that essentially produces two types of output: oscillations and evoked responses. Network architecture was based on the model described in [18], modified with the inclusion of transmission delays and lateral inhibition of non-stimulated units. Structurally, the cortex consists of minicolumns that are approximately 50μm2 in size, which can be identified morphologically and physiologically [40]; in our model a single unit was intended to represent a minicolumn of the same size. Units were fixed onto a lattice-shaped two-dimensional grid with periodic boundary conditions to form a simulated patch of cortical sheet (see Fig 5 for a schematic diagram of the model). Each WC unit accounts for two coupled neural populations; an excitatory (E) and an inhibitory (I). The output of a WC unit is given as the action potential density of the excitatory population. The mathematical details of a single unit within the model are given in Eqs 1–3 and the parameter values for each simulation in Table 1. In the equations, the subindex i refers to a reference unit while the subindex j runs over the rest of the units in the network, and parameters without subindices have the same value for all units.
The strength of connections within and between the excitatory and inhibitory cell populations within a unit is described by the parameters WEE, WII, WIE, and WEI. Parameter values were fixed as WEE = 23,WII = 0,WEI = 35,WIE = −15 for all simulations as these have been shown to generate stable oscillatory behaviour when coupled with the background activity levels given below [41]. Inter-unit connectivity structure and strength are contained in the matrices CE and CI. Unless otherwise stated, the default connectivity parameters are defined as follows: when an excitatory projection exists between units i and j, CEij = 0.15 and zero otherwise; similarly, when an inhibitory projection exists between units i and j CIij = 0.10, and zero otherwise. In all cases, the matrices CE and CI are independently generated and connectivity is not necessarily reciprocal. The local excitatory and inhibitory connection strengths were set equal to those used in [18] which in turn were based on estimations from the biology in [42]. The effect of changing these variables on model output can be seen in S2 Fig.
The background activity levels are given by E0 and I0 for the excitatory and inhibitory populations respectively. These parameter values were fixed as E0 = 0.5 and I0 = −5 for all simulations.
The response function, σ, for each population is defined by a sigmoidal function, given in Eq 3, which is increasing in the interval x∈(−∞,∞); m controls the steepness of the curve and n the offset from 0. Parameter values were fixed at m = 1 and n = 4 for all simulations. These specific values were chosen to generate oscillatory output as in [41].
The WC unit has an intrinsic resonance frequency that can be controlled via the excitatory and inhibitory time constants(τE and τI) [43,44]. The spectral response of one unit as a function of the time constants is reported in the supplementary material. The time-constants for each unit were fixed for all experiments as τE = 0.014,τI = 0.013 ms.
The extrinsic noise applied to the excitatory population of each unit is denoted by ξ(t), the variance of the noise was controlled by a multiplicative factor, z, and the repetitive stimulus f(t) are all described in detail in the next sections.
Within the network there are excitatory and inhibitory local connections. The strengths of these connections are contained in the connectivity matrices CE and CI. Excitatory connections connect the excitatory population of one unit to the excitatory population of another, whereas inhibitory connections connect the excitatory population of one unit to the inhibitory population of another. The probability that a unit is forward-connected to any other was determined via a Gaussian fall-off that decreased with distance and calculated via a multivariate random number generator which selected numbers from a normally distributed set with mean μ = 0 (centred on each unit) and standard deviation ϑ = 250 μm with a maximum cut-off of 700 μm as in [18]. This was based on the fact that most local connections are found within 500 μm of the cell body (see [42] for a review of the anatomical literature). In addition, random excitatory long-range connections were generated from a uniform distribution with probability fixed such that the long range connections constituted approximately 25% of overall connections as described in [42]. The transmission delay between units was calculated as being proportional to the Euclidean distance between them on the grid.
Activity was confined to the stimulated area by simulating lateral inhibition. This was achieved by setting all excitatory connections from the activated area to the surrounding area to 0 (CEij = 0) and all inhibitory connections from the activated area to the surrounding area to 1 (CIij = 1). This was based on the experimental work described in [21,22].
We considered two model-types before embarking on this work: the neural field model and the network model. Field models treat the neural tissue as continuous in both space and time [51]. Spatial attributes such as size are intrinsic to the model and can therefore be adjusted easily allowing in-depth investigation into the effects on model behaviour. Conversely, a network model describes the neural tissue as an interconnected network of nodes. The latter method is often used to examine the specific effect of different network structures and due to its relative computational efficiency can be used to model local and global, including whole-brain, dynamics (see [52] for example). Connectivity topology within the brain consists of a mix of dense nearest-neighbour connections and sparse long-distance projections [53]. In light of this, it has been suggested that cortical models which combine these two methodologies may be the most realistic. That is, local short-range connections are thought to be extremely dense such that local properties of the brain tissue are well-described by the continuum case while long-range connections are known to be sparse and specific and therefore easier to describe using a discrete approach. There has been recent work in developing neural field models that take into account heterogeneous lateral connections [53,54]. We base our model on that proposed by [42] and implemented by [18]; a network model consisting of dense local connections which approach the continuous case with the inclusion of a percentage of sparse long-range connections between distant units. It should be noted that neural field models are a generalisation of the neural mass models which assume a continuous tissue structure rather than discretely distributed interconnected neural masses, see [55] for details. However, field models are much more computationally expensive and in the context of our study we did not feel that approach was justified.
|
10.1371/journal.ppat.1006725 | LipL21 lipoprotein binding to peptidoglycan enables Leptospira interrogans to escape NOD1 and NOD2 recognition | Leptospirosis is a widespread zoonosis, potentially severe in humans, caused by spirochetal bacteria, Leptospira interrogans (L. interrogans). Host defense mechanisms involved in leptospirosis are poorly understood. Recognition of lipopolysaccharide (LPS) and lipoproteins by Toll-Like Receptors (TLR)4 and TLR2 is crucial for clearance of leptospires in mice, yet the role of Nucleotide Oligomerization Domain (NOD)-like receptors (NOD)1 and NOD2, recognizing peptidoglycan (PG) fragments has not previously been examined. Here, we show that pathogenic leptospires escape from NOD1 and NOD2 recognition both in vitro and in vivo, in mice. We found that leptospiral PG is resistant to digestion by certain hydrolases and that a conserved outer membrane lipoprotein of unknown function, LipL21, specific for pathogenic leptospires, is tightly bound to the PG. Leptospiral PG prepared from a mutant not expressing LipL21 (lipl21-) was more readily digested than the parental or complemented strains. Muropeptides released from the PG of the lipl21- mutant, or prepared using a procedure to eliminate the LipL21 protein from the PG of the parental strain, were recognized in vitro by the human NOD1 (hNOD1) and NOD2 (hNOD2) receptors, suggesting that LipL21 protects PG from degradation into muropeptides. LipL21 expressed in E. coli also resulted in impaired PG digestion and NOD signaling. We found that murine NOD1 (mNOD1) did not recognize PG of L. interrogans. This result was confirmed by mass spectrometry showing that leptospiral PG was primarily composed of MurTriDAP, the natural agonist of hNOD1, and contained only trace amounts of the tetra muropeptide, the mNOD1 agonist. Finally, in transgenic mice expressing human NOD1 and deficient for the murine NOD1, we showed enhanced clearance of a lipl21- mutant compared to the complemented strain, or to what was observed in NOD1KO mice, suggesting that LipL21 facilitates escape from immune surveillance in humans. These novel mechanisms allowing L. interrogans to escape recognition by the NOD receptors may be important in circumventing innate host responses.
| Leptospirosis is a widespread zoonosis caused by spirochetal bacteria, Leptospira interrogans (L. interrogans). L. interrogans are primarily extracellular pathogens although some reports suggest they may replicate within macrophages. In humans, leptospirosis can cause mild or severe disease, potentially leading to death, although rats or mice, which constitute the reservoir, are asymptomatic carriers. Host defense mechanisms involved in leptospirosis remain poorly understood. Toll-Like Receptor (TLR)2 and TLR4 are crucial for the clearance of L. interrogans, but the role of the cytosolic NOD receptors in leptospirosis is unknown. Here, we report that pathogenic leptospires escape the sensing of bacterial peptidoglycan through the NOD response. We found that an outer membrane lipoprotein of L. interrogans binds to and protects the peptidoglycan from degradation into muropeptides, thereby blocking signaling through NOD proteins. Moreover, in absence of this lipoprotein, the peptidoglycan of L. interrogans is properly sensed by human NOD1 but not by murine NOD1. This is due to the near absence of muramyl tetrapeptide, the murine NOD1 agonist, in the peptidoglycan of pathogenic leptospires. These novel mechanisms of NOD avoidance may facilitate the escape of leptospires from the innate immune system of their hosts.
| Leptospira interrogans (L. interrogans) is a pathogenic spirochete responsible for leptospirosis, a globally distributed zoonosis. Infection with pathogenic leptospires leads to asymptomatic chronic carriage in kidneys of rodents, such as rats and mice, whereas in other animals, including humans, mild to severe acute disease may develop, with potentially fatal multiorgan system failure (Weil’s disease) [1]. L. interrogans colonizes a broad range of hosts, from arthropods to mammals and may have evolved strategies to escape their innate immune response.
The innate immune response relies on different humoral components, including the complement system, antimicrobial peptides and natural IgM that together participate in the elimination of pathogens. Mammalian innate immunity receptors of the Toll-like (TLR) and Nucleotide-binding Oligomerization Domain (NOD)-like (NLR) families are pattern recognition receptors (PRRs), that sense conserved microbial associated molecular patterns (MAMPs), such as bacterial lipopolysaccharide (LPS) or peptidoglycan (PG). Recognition of MAMPS by PRRs triggers a signaling cascade resulting in activation of NF-κB and other transcription factors, leading to secretion of pro-inflammatory cytokines, chemokines, antimicrobial peptides and ultimately to the attraction of phagocytes that clear pathogens [2].
Pathogenic leptospires have been shown to evade the human complement system through various mechanisms [3]. In addition, we have shown that the sensing of L. interrogans LPS by the TLR4 receptor is a crucial determinant that accounts for the difference in outcome of leptospirosis between mice and humans. Indeed, leptospiral LPS has atypical structural features [4], escapes human TLR4 recognition [5], and is unexpectedly recognized by TLR2 in human cells [6]. However, the lipid A moiety of the leptospiral LPS is sensed by murine TLR4 [5]. Consequently, mice deficient for TLR4 (TLR4KO) are sensitive to leptospirosis, while parental wild type (WT) mice are resistant to the infection [7, 8]. Interestingly, we demonstrated deleterious inflammation in infected mice deficient for both TLR4 and TLR2 (TLR2/4DKO) and MyD88KO mice, deficient for the main adaptor of TLRs. The question arose as to whether this TLR-independent inflammation could originate from leptospiral stimulation of other innate immune receptors, such as the cytosolic NLR receptors, encompassing both the inflammasome receptors (NLRP) and the NOD receptors that recognize bacterial PG. Our previous work showed that the NLRP3 inflammasome is synergistically stimulated by L. interrogans LPS and by a second cell wall component, the glycolipoprotein. This inflammasome response was dependent on both TLR2 and TLR4, and could therefore not account for the TLR-independent inflammation [7, 9, 10]. Consequently, we chose to examine whether pathogenic leptospires are recognized by the cytosolic NOD1 and NOD2 receptors.
PG is a macromolecule of the cell envelope forming a mesh-like layer that determines the shape of the bacterium and withstands the internal turgor pressure of the cell. The PG polymer is composed of antiparallel chains of two repeated acid sugars, N-acetyl glucosamine (GlcNAc) and N-acetyl muramic acid (MurNAc). A peptide of 3 to 5 amino acids is linked to each muramic acid monomer and the peptides from adjacent chains are cross-linked.
The human receptors NOD1 and NOD2 recognize muropeptides, which are fragments of PG released upon digestion by bacterial hydrolases. These enzymes participate physiologically in bacterial biosynthesis, recycling and remodeling of PG. Muropeptides can also be released by host enzymes, such as the lysozyme that hydrolyses the ß-1,4 bond between MurNAc and GlcNAc, leading to cleavage of PG and lysis of the bacteria. Lysozyme is also a very potent cationic antimicrobial peptide, which destabilizes bacterial membranes [11]. The NOD2 receptor recognizes Muramyl dipeptide (MDP), present in PG of all bacteria [12], whereas the human NOD1 receptor recognizes the Muramyl tri-peptide with a meso-diaminopimelic acid (mesoDAP) in the third position (MurTriDAP) [13, 14], usually found in PG of Gram-negative bacteria.
Leptospiral PG, which confers the helical shape of L. biflexa [15], is characterized by a mesoDAP in the third position of the peptide [16]. Other spirochetes, such as Borrelia or Treponema spp. have an ornithine residue at this position [16]. Accordingly, Borrelia spp. have been shown to be recognized by NOD2 together with TLR2, but not by NOD1 [17]. A 1996 study showed that leptospiral PG was able to stimulate human cells to secrete pro-inflammatory cytokines [18]. Together, these data suggest that the leptospiral PG could be sensed by both NOD1 and NOD2. Herein, we examined recognition of PG of pathogenic leptospires by NOD1 and NOD2 receptors. Unexpectedly, we found that an abundant lipoprotein, LipL21, of previously unknown function, is tightly bound to the PG and blocks its digestion by hydrolases in vitro, thus impairing the recognition of leptospiral muropeptides by NOD receptors. This peculiar mechanism would constitute a novel bacterial strategy to escape the NOD innate immune response.
We recently generated a bioluminescent strain of pathogenic L. interrogans serovar Manilae (MFLum1), allowing for tracking leptospires by live imaging (IVIS) [10]. In mice intraperitoneally (IP) infected with MFLum1, IVIS revealed a biphasic infection. During the first week post infection (acute phase), leptospires disseminate and replicate in blood, then disappear from the circulation and appear in urine. In the second week, leptospires progressively reappear, restricted to the kidney. One month post infection, leptospires are stably established in kidney (chronic phase), for a life-time colonization [10].
To assess the role of NOD1 and NOD2 in the clearance of pathogenic leptospires, C57BL6/J WT mice and NOD1/NOD2 double deficient mice (NOD1/2DKO) were IP infected with different doses of L. interrogans strain MFLum1 (weekly passage 14) starting from its known median lethal dose in WT mice of 5.108 to the lower dose of 4.106 bacteria per mouse (Fig 1A), using 5 times decreasing doses. Clinical observation and weight measures were performed daily. Lethality was only observed at the higher dose of 5.108 bacteria (Fig 1A) and was equivalent in WT and NOD1/2DKO mice. Indeed, 2 out of 5 WT and 3 out of 5 NOD1/2DKO mice died or were euthanized at day 2 or 3 post infection. Only mice presenting both clinical signs of acute leptospirosis (ruffled fur, prostration, hypothermia) and that lost more than 20% of their initial weight were euthanized. Considering all doses, weight losses, in accordance with clinical symptoms of leptospirosis, were proportional to the bacterial dose used for infection, and did not show any significant differences between WT and NOD1/2DKO mice (Fig 1A).
To get a more subtle insight in a potential role of NOD receptors against leptospires, bacterial loads were measured by qPCR in blood samples at 8, 24, 48 and 72 hours post infection, in urine samples at 5 and 7 days post infection, and by live imaging (IVIS) one month post infection (Fig 1B), representing acute and chronic phase of leptospirosis. Results show that bacterial loads in blood or urine of WT and NOD1/2DKO mice were similar either at the acute phase in blood or at the chronic phase in kidneys.
Together these results suggest that the NOD receptors do not play any major role in the control of leptospirosis in mice. Previous work evidenced that saprophytic Leptospira possess an atypical PG [15], a feature that could account for the absence of recognition through the NOD1/2 receptors. Therefore, we decided to examine the PG of the pathogen L. interrogans.
To study the PG of L. interrogans, we first purified the PG from the pathogenic Fiocruz L1-130 and Manilae L495 strains with a protocol routinely used in our laboratory for Gram-negative bacteria (boiling of bacteria 0.5 h in SDS, (hereafter 0.5 h)) [19]. We used mutanolysin, a hydrolase with the same specificity as lysozyme, to digest PG and obtain muropeptides that were analyzed by HPLC. The flat HPLC profiles obtained (Fig 2A) suggested that digestion of leptospiral PG by mutanolysin failed, presumably as a result of inappropriate PG purification protocol. In comparison, PG from E. coli, treated with the same enzyme, was properly digested into muropeptides that could be resolved by HPLC (Fig 2A).
Interestingly, we observed that multiple freeze/thaw cycles of the leptospiral PG increased digestion by mutanolysin (S1A Fig), suggesting that protein contaminants may have impaired the digestion. Therefore we changed the protocol and increased the boiling time of L. interrogans in SDS from 0.5 to 4 h (hereafter 4 h), as previously performed for purification of the PG of saprophytic strains of leptospires [15]. Under this condition, the PG of the pathogenic Fiocruz L1-130 and Manilae L495 strains were better digested into muropeptides (Fig 2A). Similar results were obtained with the PG of another pathogenic L. interrogans (serovar Icterohaemorraghiae strain Verdun) (S1B Fig). PG preparations from different serovars of L. interrogans presented roughly similar muropeptides profiles (Fig 2A and S1B Fig).
Next, to study the reactivity of NOD receptors towards the leptospiral muropeptides, the reporter HEK293T cells expressing the human NOD1 or NOD2 receptors were stimulated with PG of L. interrogans strains Fiocruz L1-130 and Manilae L495 prepared from 0.5 h and 4 h protocols (Fig 2B). Interestingly, the PGs 0.5 h did not stimulate the cells, consistent with the results obtained with NOD1/2DKO mice. In contrast, both PG purified from the 4 h protocol stimulated human NOD1 (hNOD1) and to a lesser extent the human NOD2 (hNOD2) (Fig 2B), showing that the 4 h protocol allowed for the release of muropeptides from PG of L. interrogans. Altogether these results suggest that the PG of L. interrogans is protected from degradation by hydrolases, which may impair recognition by NOD receptors.
Because NOD1 and NOD2 proteins were not involved in the clearance of leptospires in mice, although human NOD1 and NOD2 recognized leptospiral PG, we next compared the murine and human NOD1 responses towards the 4 h PG preparation of both pathogenic strains (Fig 2C). Interestingly, recognition by the murine NOD1 (mNOD1) receptor of both leptospiral PG was very limited compared to the hNOD1 recognition (Fig 2C). By contrast, we found that the PG from the saprophytic L. biflexa Patoc strain was recognized by the mNOD1 (Fig 2C). We did not check the murine NOD2 since we did not previously observe any species specificity of NOD2 recognition between human and mice.
To better understand the discrepancy observed between murine versus human NOD1 recognition that could be due to specific muropeptide profiles of the leptospiral PG, individual muropeptide peaks were identified by HPLC and mass spectrometry, obtained using the 4 h PG of the Manilae L495 strain, after mutanolysin digestion (Fig 2D and Table 1). An atypical profile was observed, characterized by a single dominant peak (numbered 1 on the HPLC profile Fig 2D) corresponding to muramyl tripeptide (GM3), the human NOD1 agonist. Peak 2 corresponds to deacetylated GM3. Interestingly, the peak numbered 3 corresponding to the muramyl tetra peptide (GM4) was very limited, and GM5 was not found (Fig 2D and Table 1). Since the specificity of recognition of the NOD1 receptor has been shown to differ between human and mouse, with GM4 being the agonist of mNOD1 whereas GM3 is the agonist of hNOD1 [20, 21], this result suggests that leptospiral PG possesses only a very limited amount of muropeptides able to trigger the mNOD1 recognition, although it possesses a fair amount of hNOD1 agonists. The same observations were also true for the Verdun and Fiocruz L1-130 strains (S1B and S1C Fig). The PG from the Patoc strain (S1D Fig) slightly differed from the pathogenic strain with more dimers (S1D Fig), as already shown in a previous study also indicating that GM4 represented 4% of the Patoc’s muropeptides, while the pathogenic L495 strain had only 2% of GM4 [15]. We could not confirm these proportions since our digestion only allowed visualization and mass spectrometry analyzes of major peaks (S1C Fig). Together, these results suggest that leptospires evolved strategies to escape from NOD responses, first by impairing the degradation of its PG into muropeptides, and additionally by peculiarities of the structure of their PG, for example by a reduced amount of the GM4 muropeptide, the murine NOD1 agonist.
L. biflexa are different from the pathogenic L. interrogans strains, since they have a reduced generation time (4h versus 18h), are sensitive to complement and never cause disease [1]. We previously showed using live imaging of a bioluminescent derivative of the L. biflexa Patoc strain (PFLum7) that it is mostly cleared within 24h post IP injection [10]. Since in contrast to pathogenic strains, Patoc’s PG was recognized by the murine NOD1, we tested whether NOD1/2DKO mice could be less effective to clear this strain. Therefore, we compared the clearance of 5.108 PFLum7 in WT and NOD1/2DKO (S2 Fig), but we did not evidence any significant difference in the kinetics of clearance of the bacteria, suggesting that NOD1 and NOD2 receptors do not participate in the recognition of the L. biflexa Patoc strain.
We suspected that some proteins may be linked to the PG and could have been degraded by the 4 h boiling procedure. For this reason, PG preparations from both the 0.5 h and 4 h boiling protocol were electrophoresed on a polyacrylamide gel and stained with Coomassie blue. Since PG is a macromolecule and does not enter the gel, and because muropeptides are not stained by Coomassie blue, only proteins could be detected. Indeed, one band of approximately 20 kDa was present in the Fiocruz L1-130 and Manilae L495 PGs prepared with the 4 h protocol but was missing in the PG prepared with the 0.5 h protocol (Fig 3A). This suggested that boiling L. interrogans in SDS for 0.5 h was insufficient to detach the 20 kDa protein, which was retained within the PG and did not enter the gel. The 20 kDa band analyzed after trypsin digestion was identified by mass spectroscopy as the LipL21 leptospiral lipoprotein, which was checked by immunoblotting with a polyclonal antibody directed against LipL21 (Fig 3A). LipL21 has been shown to be an abundant outer membrane surface exposed lipoprotein, of unknown function, conserved among pathogenic leptospires, and expressed during infection [22]. BLAST searches with the LipL21 protein sequence from L. interrogans serovar Manilae showed that LipL21 is restricted to Spirochetes, and present in all Leptospira species (spp.). ClustalX2 alignments of selected protein sequences show that LipL21 sequences are distributed in three statistically well-supported subgroups. Interestingly, the three sequence subgroups follow the distribution of the three pathotypes that have been described for Leptospira spp.: pathogenic, intermediate pathogenic and non-pathogenic (S3 Fig).
One mutant (M58) in the lipl21 gene, and its complemented partner (C5M58) were obtained after transposon mutagenesis performed on L. interrogans serovar Manilae strain L495, as described [23]. The expression of LipL21 was checked by immunodetection in the crude extracts. The lipl21 mutant (hereafter named lipl21-) was devoid of LipL21, and the complemented strain (called lipl21-/+), expressed slightly more LipL21 than the parental L495 strain (Fig 3B). The growth rate of lipl21- was equivalent to the parental L495 and complemented strains (Fig 3D). Interestingly, in contrast to the PG from both parental and lipl21-/+ complemented strains, the PG from the lipl21- mutant prepared with the 0.5 h protocol was effectively digested by mutanolysin and the pattern of muropeptide peaks was observed in HPLC (Fig 3C). In correlation with these results and by comparison with PG from both parental and complemented strain, PG from the lipl21- mutant also clearly signaled through hNOD1 (Fig 3E). Taken together, these results suggest that LipL21 binding to PG impairs the degradation of PG and subsequent signaling through NOD receptors.
An alternative way to assess the role of LipL21 in the protection of PG is to express the recombinant protein in a heterologous organism and study the capacity of extracted PG to be digested into muropeptides in the presence of hydrolases. Since LipL21 is an outer membrane (OM) lipoprotein of L. interrogans [22], it requires appropriate secretion across the inner membrane, to be acylated, and to reach its final OM location. To first verify that LipL21 was correctly targeted to the periplasm in E. coli, we used an alkaline phosphatase (PhoA) secretion assay. After its synthesis, the pro-PhoA is translocated across the plasma membrane towards the periplasm, with cleavage of its N-terminal sequence and maturation of disulfide bonds, essential for activity. Thus, if PhoA reaches the periplasm, it is fully active, and its activity can be monitored using a synthetic substrate (5-Bromo-4-chloro-3-indolyl phosphate, XP) that turns blue upon PhoA-dependent cleavage. We constructed a series of 4 plasmids, one carrying the full Pro-PhoA gene (positive control), one with the 5’ sequence corresponding the targeting signal removed (Δ(2–22)phoA gene, negative control, no secretion) and two plasmids carrying the full length LipL21-PhoA fusion gene, and the ΔN-LipL21-PhoA fusion gene, where the N-terminal sequence of LipL21 had been removed. E. coli carrying plasmids with these constructions were tested for PhoA activity, and results show that the LipL21-PhoA fusion is targeted to the periplasm (blue staining upon XP and IPTG addition), while the ΔN-LipL21-PhoA fusion is not (no blue staining), compared to control plasmids, indicating that LipL21 is appropriately targeted to the periplasm in the E. coli heterologous context (Fig 4A). Moreover the LipL21-PhoA fusion protein was easily detected (S4 Fig), showing that LipL21 can be correctly expressed in E. coli.
However, contrary to the PhoA fusions, different attempts to express the freestanding LipL21 in various E. coli strains were unsuccessful. Using the BL-21 Rosetta-2 E. coli strain, expressing tRNAs corresponding to rare codons, we succeeded in obtaining transformants from the pRSF-duet1. However, even without the induction of the protein expression, the generation time of the BL-21 Rosetta-2 strain harboring the LipL21 expression plasmid (pLipL21) was impaired (135 min) compared to the control strain transformed with the empty plasmid (pEmpty) (24 min), suggesting a toxic effect of the LipL21 lipoprotein expression (Fig 4B).
For improved regulation of expression, the lipl21 gene without its signal peptide was cloned in the tightly regulated pASK-IBA-6 vector, containing the signal peptide of OmpA to direct the protein to the periplasm. This allowed expression of LipL21 without deleterious effects on the bacterial growth. To limit toxicity, the protein expression was induced for only 3 hours in the BL-21 Rosetta-2 strain. After confirmation of LipL21 expression (Fig 4C), the PGs from E. coli strains, expressing or not the LipL21 protein, were prepared with both 0.5 h and 4 h protocols and digested with mutanolysin. The HPLC profiles showed that the PGs from E. coli with the empty plasmid prepared from the 0.5 h and 4 h protocols were equally well digested, whereas the PG prepared from the E. coli strain expressing LipL21 was not digested following the 0.5 h protocol, but required the 4 h boiling to be digested (Fig 4D), demonstrating that LipL21 could protect E. coli PG from digestion. These results mirror those obtained with L. interrogans PG, and demonstrate that LipL21 plays a similar function in the protection of PG in a heterologous context of E. coli.
Furthermore, the recognition by hNOD1 and hNOD2 of PG from E. coli expressing LipL21, prepared from the 0.5 h protocol, was reduced compared to the PG from the E. coli strain harboring the empty vector (Fig 4E and 4F), again mimicking the results obtained in L. interrogans. Taken together these results suggest that LipL21 binds to mesoDAP containing PG and impairs the release of muropeptides, which in turn alleviates recognition of the bacterium by the innate immune receptors NOD1 and NOD2.
We infected WT and NOD1/2DKO mice with the mutant lipl21- strain (M58) and compared the kinetics of dissemination of the bacteria in blood in the first 3 days post infection. Instead of disseminating in the blood as observed for the virulent strain MFLum1 (Fig 1A), the lipl21- mutant was cleared from blood after 72 hours post infection in the WT mice, showing that the M58 strain had lost virulence (Fig 5A). No difference in clearance was observed in NOD1/2DKO compared to WT mice, suggesting that the lack of virulence was not linked to the PG sensing through mNOD1, which was expected, but also not through NOD2 (Fig 5A). For this reason we wondered whether the lack of virulence of the lipl21- mutant could be due to a higher sensitivity to lysozyme, a potent anti-microbial peptide that can also cleave the PG. Hence, we infected with M58 parental FVB (WT) mice and LysMKO mice, deficient for the lysozyme expressed in macrophages [24]. No difference of clearance of M58 was observed between WT and LysMKO mice (Fig 5B). Moreover, an in vitro test did not show any enhanced susceptibility to lysozyme of the lipl21- mutant compared to MFLum1, expressing LipL21 (Fig 5C). Altogether these results suggest that the loss of virulence of M58 is not linked to the muropeptides sensing by the NOD receptors nor to a better access of the PG to lysozyme.
We next infected WT mice with the M58 lipl21- mutant or the complemented C5M58 lipl21-/+ strain. The complemented strain was also cleared after 3 days of infection in mice, showing that the LipL21 complementation did not restore the virulence (Fig 5D, left panel). We also did a lethal challenge in gerbils comparing the virulent MFLum1 to the lipl21- mutant and complemented strain. M58 and C5M58 strains did not kill the gerbils, in contrast to the MFLum1 (Fig 5D, right panel). A whole DNA sequencing analysis of the M58 strain showed that beside silent single nucleotide polymorphisms (SNPs), 5 non synonymous SNPs were found in M58 compared to the parental L495 strain. 3 out of these 5 SNPs were found in an unknown ORF, 1 in a transposase gene and 1 in the fliM gene, coding for the flagellar motor switch, known to be involved in virulence [25] (Table 2). Those results suggest that the M58 mutant lost its virulence independently of the lipl21 mutation, most probably in the course of the mutagenesis, and that those SNPs may be responsible for the loss of virulence.
Since we did not find an altered phenotype related to the LipL21 role in mice or gerbils, and showed that human NOD1 receptors are able to sense leptospiral PG, we next examined whether the LipL21 lipoprotein could be important in the defense against Leptospira spp. in humans. As a surrogate of the hNOD1 response, we infected mutant humanized mice expressing a transgene of the human NOD1, in a background of murine NOD1KO mice (hNOD1Tg/mNOD1KO) [21], and mNOD1KO with the M58 lipl21- mutant. Although at 8h post infection both NOD1 humanized mice and NOD1KO mice showed the same load of bacteria, the humanized mice harbored significantly fewer leptospires than mNOD1KO at 24h post infection (Fig 6A). Moreover, the infection of NOD1 humanized mice with M58 and the C5M58 complemented strain resulted in a lower bacterial load observed for M58 compared to C5M58 at 48h and 72h post infection (Fig 6B). Although these results did not reach statistical significance, this trend suggests a slight better clearance of the M58 strain compared to the C5M58 expressing LipL21, which is in contrast with the results showing no difference of clearance of M58 and C5M58 in WT mice (Fig 5A). These data suggest that the sensing of M58, devoid of LipL21, by human NOD1 is responsible for the increased clearance.
Together these results indicate that LipL21 is important for protection of PG from detection by the NOD receptors, and therefore may help Leptospira spp. to escape the immune responses of some hosts, such as humans.
In this work, we show that pathogenic leptospires avoid the innate immune response mediated by cytosolic NOD receptors through close association of the LipL21 lipoprotein with PG, thereby blocking the release of muropeptides. Moreover, a second layer of protection against murine NOD1 recognition is conferred by the peculiar composition of the leptospiral peptidoglycan, which is almost devoid of GM4, the preferential agonist of murine NOD1 [20].
Cytosolic NOD1 and NOD2 receptors recognize invasive bacteria, such as Shigella flexneri, present within the cytosol of host cells. In addition, NOD receptors are able to sense invasive intracellular bacteria residing in vacuoles, such as Salmonella enterica and also extracellular pathogens such as Helicobacter pylori that can both secrete muropeptides into the host’s cytosol using type III or type IV secretion systems, respectively [26, 27]. L. interrogans are usually thought of as extracellular bacteria and genomic studies have not identified Type III nor Type IV secretion systems [28]. However, several studies have suggested intracellular survival of pathogenic leptospires in mouse or human macrophages [29, 30], as well as in zebra fish macrophage-like cells [31], indicating that muropeptides of pathogenic leptospires could indeed be in contact with the cytosolic NOD sensors. Moreover, free muropeptides can either directly enter cells through specific transporters, such as the human intestinal peptide transporter 1, (hPEPT1) for MDP [32], or through phagocytosis of GM3 that is further processed to di- or tri-peptides and actively transported by the hPEPT2 in the cytosol [33]. Therefore, all bacteria either pathogenic or commensal can be detected by the NOD receptors.
Different types of PG modifications are known to impair NOD1 recognition, such as a change of the third amino acid in the stem peptide (mesoDAP). This is the case for Borrelia spp. that have an ornithine instead of a mesoDAP residue. The amidation of the second or third residue of the peptide stem is another mechanism that impairs the NOD1 recognition in Staphylococcus aureus and Lactobacillus plantarum. Modifications of the glycan chains are known to confer lysozyme resistance, such as O-acetylation of the MurNAc in S. aureus and in Listeria monocytogenes [34], or N-glycolylation of the MurNAc in Mycobacterium tuberculosis (reviewed in [35]). We have previously shown that N-deacetylation of the GlcNAc moiety conferred lysozyme resistance in Listeria monocytogenes [34, 36]. Therefore, to our knowledge, the 2 mechanisms found in L. interrogans of Lip21 binding on the PG and the near-absence of GM4 muropeptides have not been previously described and constitute novel bacterial strategies to escape the immune response. Moreover, for the first time we assigned a function to the leptospiral lipoprotein, LipL21. Indeed leptospires, like other spirochetes, are characterized by having many lipoproteins awaiting functional characterization [37].
The lack of difference in bacterial loads between WT and NOD1/2DKO mice upon infection with a virulent strain of L. interrogans strain L495 drove us to study the potentially atypical composition of leptospiral PG. Despite improved digestion in absence of LipL21 and the use of optimized protocols adapted to a large number of PGs from Gram-positive and Gram-negative bacteria, we never succeeded in completely digesting leptospiral PG [38]. Nevertheless, the HPLC profiles of L. interrogans muropeptides are roughly similar to those described for the PG composition of the saprophytic L. biflexa Patoc strain [15]. Indeed, HPLC digestion profiles revealed that although L. biflexa’s PG conforms to a Gram-negative chemotype, a small proportion of modified muropeptides were rarely observed in other bacteria, such as dimers lacking one of the GlcNAc groups or dimers amidated on the mesoDAP residue [15]. As in L. biflexa, we found that L. interrogans harbors modified deacetylated muropeptides, suggesting that leptospires possess peculiar enzymes such as deacetylases. Despite treatment with various enzymes, such as trypsin and chemotrypsin, to remove typical protein contaminants in preparing the PG, the amount of digested PG was always very limited compared to E. coli, and did not allow for efficient mass spectrometry analysis. This suggests that, in addition to LipL21, other mechanisms or specificities of pathogenic leptospiral PG hinder the digestion with mutanolysin, an enzyme with the same specificity as lysozyme, cleaving between the two sugars. Hence, the study of the hydrolases of Leptospira spp. could be of interest to better understand the unique structure and functions of leptospiral PG.
According to the mesoDAP composition of the leptospiral PG [16], it was expected that the leptospiral muropeptides could be recognized by hNOD1 and to a lesser extent by hNOD2 [16]. The amount of MDP present in the PG was too small to be detected by HPLC, but we found that it was released by L. interrogans PG, using HEK293T reporter cells expressing the hNOD2 receptor. The HPLC profile found for the L. interrogans PG, with only GM3, no GM5 and only minimal amounts of GM4, seems unusual since it suggests that L. interrogans rapidly processes its muropeptides to trim them in GM3. This is frequently the case for Gram-positive bacteria such as Staphylococcus carnosus [39], whereas Gram-negative bacteria usually tend to keep the muropeptides as GM4. Therefore, near-absence of GM4 is a modification of PG that to our knowledge has never been shown for other bacteria with cell walls associated with an outer membraner. However, HPLC profiles of muropeptides are available for a restricted number of bacterial spp. and this phenomenon may be more frequent than previously thought. These results suggest that the reduction of the amount of GM4 muropeptide could be an adaptation of pathogenic Leptospira spp. to avoid the murine NOD1 recognition. This would be very important because mice are, like rats, prominent reservoirs of leptospires. It would be interesting to identify the L-D peptidase responsible for the cleavage of the stem peptide. Interestingly, published data about the PG of the saprophytic strain L. biflexa strain Patoc showed a small peak of GM4 [15], suggesting that the PG of L. biflexa strain Patoc could be recognized by the murine NOD1. Indeed, using the reporter cell system, we showed that purified PG of the Patoc strain was detected in vitro by the murine NOD1. However, this strain was equally cleared by the WT and NOD1/2DKO mice, suggesting that in vivo, the NOD receptors did not recognize this saprophytic strain. Likewise for the GM2, the NOD2 agonist, the HPLC profile of muropeptides of the Patoc strain did not reveal a peak of GM4, most probably because of the poor efficiency of PG digestion. We looked for, but did not identify, proteins associated with the PG of the Patoc strain that could have impaired the in vivo recognition by mNOD1. We speculate that the distant homolog of LipL21 found in Patoc, may play the same role of PG binding, with a lower affinity for PG, explaining why we did not find it attached to the PG after SDS boiling. Further biochemical studies are required to study the potential role and affinity of the Patoc’s LipL21 to the PG. It is difficult to understand why the PG from this saprophytic strain would also avoid the NOD immune system. However, because the Patoc strain is sensitive to complement, which is a highly efficient mechanism to destroy bacteria, we cannot exclude that the NOD response may have been masked.
The species-specificity of murine NOD1 versus human NOD1 towards leptospires is reminiscent to the species specificity of TLR4, with human TLR4 unable to recognize the LPS from L. interrogans while murine TLR4 recognizes it [5]. We have previously shown that the TLR4 response is crucial to defense against L. interrogans in mice [7]. Nevertheless, L. interrogans can overcome this response and colonize the kidney [10, 40]. We have also shown that in mice NOD1 is a potent PRR in the kidney, actively participating in the control of experimental infection with uropathogenic E. coli through neutrophil recruitment [41]. We can speculate that NOD1 recognition, if active in mice, would not have allowed the kidney colonization by L. interrogans. By contrast, leptospires would not need to escape the recognition by NOD1 in humans because they avoid TLR4 recognition of their atypical LPS [5], potentially resulting in severe infection. The systematic study of species specificity of TLR and NOD receptors in different mammalian hosts towards the leptospiral MAMPs would most probably provide insights concerning the host susceptibility and characteristics of leptospirosis.
The fact that 30 min boiling in SDS was not enough to obtain a leptospiral PG sacculus prone to digestion is compelling. It suggests that the interaction between the PG and its “contaminant” is very strong, although not covalent, as it detaches after 4 hour of boiling in SDS. Moreover, the fact that repeated freeze/thaw cycles led to an increased digestion of the PG suggested that the interacting partner could be a protein, compatible with our finding of the LipL21 lipoprotein being associated with the PG. Interestingly, the only leptospiral peptides identified by mass spectroscopy from proteins released from the PG were those of LipL21. These results suggest that the leptospiral PG specifically binds to LipL21 and not to other lipoproteins colocalized with the PG layer, such as LipL32 or Loa22. The lack of co-identification of the Loa22 lipoprotein was surprising, since it has roughly the same molecular weight than LipL21, it is anchored in the outer membrane and possesses a large OmpA domain, known as a peptidoglycan-binding domain. However we did not find Loa22 associated to the peptidoglycan after SDS boiling. The finding that heterologous expression of LipL21 in E. coli reproduced the effect of LipL21 on E. coli PG indicates that the phenotype that we observed in leptospires is actually due to LipL21 rather than minor contaminants. In addition, our results show that the lipid anchor of the LipL21 lipoprotein is not required for the binding to PG, because we removed the LipL21 signal sequence including the lipobox sequence [37] to replace it by the signal peptide of OmpA, which is not a lipoprotein. Although recombinant LipL21 had previously been expressed and purified from E. coli [22], we did not obtain soluble cytosolic LipL21 despite number of trials using different expressing vectors and E. coli strains. One explanation could be linked to the presence of several cysteine residues in LipL21 that cannot form disulfide bridges in the cytoplasm. Expression of LipL21 transported in the periplasm of E. coli through its own signal peptide was not toxic when it was fused with alkaline phosphatase at the C-terminal part of the protein. This suggests that the non-anchored C-terminal part of the lipoprotein may be required for the binding to PG, or that a change of conformation because of the fusion impaired the binding. Interestingly, although the muropeptide composition is rather different between E. coli and L. interrogans, the binding of LipL21 still occurs to both types of PGs suggesting that a common feature might be key to recognition such as the mesoDAP or the sugar backbone.
Among different bacterial proteins known to bind the PG are 2 lipoproteins: the peptidoglycan associated lipoprotein (PAL), which binds non-covalently to PG, and the Braun’s lipoprotein (Lpp), a very abundant small lipoprotein of Escherichia coli, which is covalently linked to the mesoDAP residue of the PG peptide stem. Both play a role in anchoring the PG to the inner leaflet of the outer membrane (reviewed in [42]). However, we did not find any DNA or protein sequence homology between the LipL21 from L. interrogans and Lpp or PAL. Moreover, in Leptospira spp. the PG has been shown to be associated with the inner membrane [37], suggesting of a different functional role of LipL21 compared to those lipoproteins. However, LipL21 has been shown to be localized in the outer leaflet of the outer membrane [22]. It is difficult to reconcile the present results with these data, as the size of LipL21 would not allow it to be anchored in the outer leaflet of the outer membrane and at the same time linked to PG. One possibility is that LipL21 behaves as the Lpp, which has recently being shown to coexist in two forms, one residing in the periplasm linked to the PG through the C-term Lysine and another free in the outer membrane [43].
Both PG-binding proteins Lpp and PALs, as well as Loa22, are required for virulence [42] [44]. Accordingly, we showed that the lipL21- mutant M58 is avirulent in mice and gerbils. Although we showed that the lack of virulence of the M58 mutant in mice is not due to the NOD sensing, nor to an increased susceptibility to lysozyme, we did not check whether the loss of virulence could be due to an enhanced recognition by other PRRs of the peptidoglycan recognition receptors (PGRPs) family. Indeed, PGRPs are expressed in a wide range of mammalian cells, bind the PG and have an important bactericidal activity [45]. The leptospiral field remains hampered by the difficulty to achieve homologous recombination for clean on site complementation [46]. The fact that complementation, which properly restored expression of LipL21, did not restore the virulence of M58 could be due to the new mutation introduced by the random insertion of the transposon carrying the lipl21 gene. Also, we don’t know whether the lack of regulation of expression of the complemented lipl21 in trans, is important for its function. Therefore, obtaining new lipl21 mutants and complemented strains would be useful in clarifying the role of LipL21 in virulence. If not involved in virulence, LipL21 would join LipL32, the major leptospiral lipoprotein, as extremely well conserved lipoproteins in pathogenic strains that are expressed in hosts, yet whose absence strikingly does not impair virulence in mice, rats, and hamsters [47]. Given the role of LipL21 in the protection of PG and immune escape, and its leptospiral-specific distribution among all three sub-groups (pathogenic, intermediate pathogenic and non-pathogenic), it suggests a similar role of LipL21 in all Leptospira spp. It is possible that LipL21 could have different capacities to protect the PG, with a stronger association with PG in pathogenic leptospires and a weaker association in the non-pathogenic subgroup.
Importantly, we showed that LipL21 could be effective in impairing the human NOD1 recognition, using a humanized mouse model expressing the human NOD1. Our results suggest that LipL21, although not required in mice since leptospiral PG is not recognized by the murine NOD1 and very poorly by NOD2, would be important in humans to block the NOD1 sensing.
To conclude, we found no evidence of any role of NOD receptors in sensing leptospires and therefore the origin of inflammation observed in the TLR2/4DKO mice is not due to NOD activation. This total lack of NOD recognition is surprising since to our knowledge all the pathogenic bacteria yet studied, either intracellular or extracellular, are recognized by one or the other or both of the NOD receptors. The consequences of L. interrogans escape from recognition by NOD receptors will be an important subject of future studies.
L. biflexa sevorar Patoc strain Patoc (Paris) and bioluminescent derivative PFlum7, L. interrogans Icterohaemorraghiae strain Verdun, L. interrogans serovar Copenhageni strain Fiocruz L1-130, L. interrogans serovar Manilae strain L495 and bioluminescent derivative strain MFLum1 were described earlier [7, 9, 10]. Bacteria were grown in Ellinghausen-McCullough-Johnson-Harris (EMJH) medium (Bio-Rad) at 28°C without agitation.
Transposon mutagenesis was performed on L. interrogans serovar Manilae strain L495 as previously described and the location of the transposon insertion was determined by direct sequencing of genomic DNA [48]. One mutant (M58) obtained using the Himar 1 transposon with a kanamycin resistant cassette was localized in the 3’ part end of the LA_0011 (LMANv2_100108 | +2 | 972824–973384) gene, corresponding to the LipL21 lipoprotein (lipL21-). A second transposon mutagenesis using a spectinomycin resistance cassette was performed on the lipl21- mutant to reintroduce the lipl21 gene under the control of its native promoter. The complemented strain C5M58 was called lipl21-/+. The localization of the second transposon of lipl21-/+was in the hypothetical protein LA_0094.
LipL21 expression was checked by immunoblotting experiments using the polyclonal anti-rabbit serum directed against the LipL21 protein diluted 1:10000, as described [22].
Male and female C57BL/6J mice (7- to 10-week old) were used in this study and were from Janvier (Le Genest, France). FVB mice were from Charles Rivers Laboratory. LysM knock–out (KO) mice in a FVB background, mice deficient for NOD1 (mNOD1KO), transgenic mice expressing the human NOD1 and deficient for the murine NOD1, named hNOD1Tg/mNOD1KO [21], mice deficient for both NOD1 and NOD2 (NOD1/2DKO), all in a C57BL6/J background, were raised at the Institut Pasteur animal facility and were previously described [40]. Female gerbils (4-week old) were from Janvier.
Infections with L. interrogans serovar Manilae strains were conducted as described [10]. Just before infection, bacteria in late exponential phase (around 5.108 leptospires per ml) were centrifuged at room temperature for 25 min at 3250 ×g, resuspended in endotoxin-free PBS, and counted using a Petroff-Hauser chamber. Leptospires in 200 μl of PBS were injected via the intraperitoneal route (IP) into mice. Most experiments were done with a dose of 2.107 or less corresponding to a sublethal dose of the pathogenic L. interrogans Manilae. Higher doses up to 5.108 bacteria, provoking a lethal or severe infection were only used for determination of the susceptibility of NOD1/2DKO mice to leptospires. Gerbils were infected with 106 L. interrogans/ml in EMJH through the IP route. In some experiments, the virulent Manilae bioluminescent MFlum1 strain was used as control since it has been obtained by random insertion of a transposon cassette, like the lipl21- mutant [10].
Imaging was performed as described [10]. Because the dark fur of C57BL/6 mice partly blocks the emission of light, mice were shaved on the back the day before imaging. Ten minutes before imaging, 100 μl of D-luciferin potassium salt (Caliper Life Sciences, 30 mg/ml in PBS), the substrate of Firefly luciferase, was intraperitoneally injected to mice. Mice were anesthetized using a constant flow of 2.5% isoflurane mixed with oxygen and air as recommended by the manufacturer, using an XGI-8 anesthesia induction chamber (Xenogen Corp.). Mice were maintained in the anesthesia chamber for at least 5 min to allow adequate dissemination of the injected substrate. Bacterial infection images were acquired using an IVIS Spectrum system (Xenogen Corp., Alameda, CA) according to instructions from the manufacturer. Analysis and acquisition were performed using Living Image 3.1 software (Xenogen Corp.). Images were acquired using the automatic mode (acute phase) or 5 min of integration time (chronic phase and pictures) with a binning of 8 and with the emission filter in the “open” mode. All other parameters were held constant. Quantification was performed using a region of interest defined manually (kidneys) and the results were expressed as photons (P) per second (s) per cm2 per steradian (SR).
The leptospiral burden in blood and urine was determined by quantitative real-time PCR (qPCR), as described [40]. The Maxwell 16 automat was used to extract total DNA from 50 μl of blood and from a drop of urine, using the Maxwell blood DNA and cell LEV DNA purification kits (Promega), respectively. Primers and probe designed in the lpxA gene of L. interrogans strain Fiocruz L1-130 [4] were used to specifically detect pathogenic Leptospira spp. [40]. qPCR reactions were run on a Step one Plus real-time PCR apparatus using the absolute quantification program (Applied Biosystems), with the following conditions for FAM TAMRA probes: 50°C for 2 min, 95°C for 10 min, followed by 40 cycles with denaturation at 95°C for 15 s and annealing temperature 60°C for 1 min, according to the manufacturer’s instructions.
Leptospiral peptidoglycan (PG) was purified from late-exponential-phase culture as previously described [38]. Two different times, 0.5 h [38] and 4 h [15], of boiling incubation in 4% SDS were used to extract the PG. PG at a final concentration of 6 mg/mL in endotoxic free water, arising from the two different extractions were called respectively PG 0.5 h and PG 4 h. PG (250 to 500 μg) was digested with 100 U of mutanolysin from Streptomyces globulosporus (Sigma) in 12.5 mM sodium phosphate buffer (pH 5.6) for 16h at 37°C as described [38]. The reaction was stopped by boiling for 2 min. To reduce sugar moieties, sodium borohydride (10 mg/mL final) was added to the soluble muropeptide fraction in a 0.5 M sodium borate buffer (pH 9.0). After 15 min of incubation, the pH was adjusted to 2 with orthophosphoric acid to stop the reaction. Muropeptides were analyzed by high-performance liquid chromatography (HPLC) as described previously [38] with a Hypersil reversed-phase octyldecyl silane (C18) column (4.6 x 250 mm, flow-rate of 0.5 ml/min; ThermoHypersil-Keystone) at 52°C. Muropeptides are detected at a 206 nm wavelength using a Shimadzu SPD-20A-UV-Vis detector. To confirm muropeptide structures by mass spectrometry, an additional digestion of the PG prepared with the 4 h protocol was performed with chemotrypsin as described [15] to remove all protein contaminants. Individual muropeptides peaks were collected as they eluted from the HPLC column, then desalted and analyzed by LC-MS as described previously [38, 49].
Human epithelial cell line HEK293T (ATCC-CRL-3216), which does not express NOD2 and only minimal amounts of human NOD1, was transfected with a NF-κB-luciferase reporter construct together with NOD1 or NOD2 and ß-galactosidase expressing vectors as described [38]. Control muropeptides (100 nM final) from Invivogen (MurTriDAP as NOD1 agonist, MDP as NOD2 agonist, and FK156 as murine NOD1 agonist [50]) or PG, that cannot enter the cells, have been co-transfected. The activities of luciferase and ß-galactosidase were measured from cellular lysates in the presence of their substrates, using a luminometer and ELISA reader, respectively. Data are expressed as mean ± SD of triplicate values of light units, normalized to the ß-galactosidase activity.
LipL21 identity assignment on coomassie-stained 4–15% gels was confirmed by mass spectrometry (MS). MS analyzes were carried out by the Plateforme de Biophysique Moléculaire at the Institut Pasteur. Coomassie stained 21 kDa gel bands were excised using a robotic workstation ProPic Investigator. Each sample was digested by trypsin and analyzed by MALDI TOF-MS5 as previously described [51].
Cloning of full-length lipl21 in pRSF-duet and in fusion with alkaline phosphatase, as well as LipL21 devoid of its own signal sequence in pASK-IBA6, have been described in supplementary information material and methods.
All plasmids (pILL2156-FL-phoA, pILL2156-Δ(2–22)phoA, pILL2156-lipl21-phoA and pILL2156- ΔN-lipL21-phoA) were individually transformed into an MG1655ΔphoA strain (phoA748(del)::kan strain, CGSC, The Coli Genetic Stock Center, Yale). Strains were grown in the presence of chloramphenicol (25 μg/mL). Alkaline phosphatase assays were performed as described [52] either on plates or in liquid medium. Briefly, in the first case, 5 μL of an overnight preculture of each transformed strain was spread over a section of an LB plate containing IPTG and 5-Bromo-4-chloro-3-indolyl phosphate (XP), 0.1 mM and 40 μg/mL respectively. Plates were incubated overnight at 37°C and visualized. In the second case, 5 mL of LB medium was seeded with 50 μL of an overnight preculture, and then grown to an OD600 nm of 1.0 in the presence of IPTG (0.5 mM). XP was added to a final concentration of 40 μg/mL and bacteria were incubated for 30 min to 1 h at 37°C with shaking (170 rpm) until a blue coloration appeared. Strains containing pILL2156-FL-phoA were used as a positive control, while strains transformed with pILL2156-Δ(2–22)phoA were used as a negative control.
Immunodetection of LipL21 was performed according to a standard protocol using a rabbit anti-serum elicited against LipL21 [22] and a secondary antibody (anti-rabbit whole IgG coupled to horseradish peroxidase, elicited in Donkey, Amersham), both diluted 1:10000, and revealed with TMB.
Broth microdilution testing was performed using 96-well plates as previously described [53]. Lysozyme at a final concentration of 2 mg/mL, with and without EDTA (2 mM final concentration) was prepared in 100 μl per well of EMJH medium in a sterile 96-well plate. 100 μl of leptospiral suspension diluted in EMJH medium was added to each well to obtain a final concentration of 2.106 cells/mL. The plate was mixed and incubated at 28°C without agitation for 10 min to 2 hours. After PBS washes, bacteria were incubated in EMJH medium at 30°C for 5 days without agitation. Then 20 μl of 10-fold-concentrated AlamarBlue, a cell growth indicator dye, was added to the wells. AlamarBlue turns from dark blue to bright pink in response to chemical reduction of the growth medium in the presence of viable bacteria. The plate was further incubated at 30°C for 2 days in dark without agitation, and the bacterial growth was observed by the color change of the indicator.
Next-generation sequencing was performed on genomic DNA of the L495 parental and M58 strains by the Mutualized Platform for Microbiology (P2M) at Institut Pasteur, using the Nextera XT DNA Library Preparation kit (Illumina), the NextSeq 500 sequencing systems (Illumina), and the CLC Genomics Workbench 9 software (Qiagen) for analysis. Sequence reads were aligned with the sequenced and annotated L. interrogans serovar Manilae L495 genome(http://www.genoscope.cns.fr/agc/microscope/search/export.php?format=genbank&option=none&S_id=5570) by using the Burrows-Wheeler Alignment tool (BWA mem 0.7.5a) [54]. SNP calling was done with the Genome Analysis Toolkit (GATK 2.7–2) Unified Genotyper [55] by following Broad Institute best practices. To validate the call, candidate SNPs were further filtered by requiring coverage of greater than half of the genome mean coverage and 95% read agreement. All data were submitted to EMBL-EBI. A project was created in order to group M58 reads (ERR2192021) and L495 assembly (ERZ480214) associated to the genomes sequenced in this study and is available by using this accession number: PRJEB23342. (http://www.ebi.ac.uk/ena/data/view/PRJEB23342)
All protocols were reviewed by the Institut Pasteur (Paris, France), the competent authority, for compliance with the French and European regulations on Animal Welfare and with Public Health Service recommendations. This project has been reviewed and approved (# 2013–0034 and #2015–0026 to CW) and (#2016–0019 to MP) by the Institut Pasteur ethic committee for animal experimentation (Comité d’éthique d’expérimentation animale CETEA #89), agreed by the French Ministery of Agriculture.
Statistical analysis was performed using Graph Pad Prism software. The unpaired t test, (two-tailed P values) was used to compare two groups at the same time point. Values are expressed as mean ± standard error of the mean (SEM). A p value < 0.05 was considered significant. p values: *p < 0.05, **p < 0.01, ***p < 0.001.
|
10.1371/journal.pgen.0040035 | Expression Profiles Reveal Parallel Evolution of Epistatic Interactions Involving the CRP Regulon in Escherichia coli | The extent and nature of epistatic interactions between mutations are issues of fundamental importance in evolutionary biology. However, they are difficult to study and their influence on adaptation remains poorly understood. Here, we use a systems-level approach to examine epistatic interactions that arose during the evolution of Escherichia coli in a defined environment. We used expression arrays to compare the effect on global patterns of gene expression of deleting a central regulatory gene, crp. Effects were measured in two lineages that had independently evolved for 20,000 generations and in their common ancestor. We found that deleting crp had a much more dramatic effect on the expression profile of the two evolved lines than on the ancestor. Because the sequence of the crp gene was unchanged during evolution, these differences indicate epistatic interactions between crp and mutations at other loci that accumulated during evolution. Moreover, a striking degree of parallelism was observed between the two independently evolved lines; 115 genes that were not crp-dependent in the ancestor became dependent on crp in both evolved lines. An analysis of changes in crp dependence of well-characterized regulons identified a number of regulatory genes as candidates for harboring beneficial mutations that could account for these parallel expression changes. Mutations within three of these genes have previously been found and shown to contribute to fitness. Overall, these findings indicate that epistasis has been important in the adaptive evolution of these lines, and they provide new insight into the types of genetic changes through which epistasis can evolve. More generally, we demonstrate that expression profiles can be profitably used to investigate epistatic interactions.
| The effect of a genetic mutation can depend on the genotype of the organism in which it occurs. For example, a mutation that is beneficial in one genetic background might be neutral or even deleterious in another. The interactions between genes that cause this dependence—known as epistasis—play an important role in many evolutionary theories. However, they are difficult to study and remain poorly understood. We used a novel approach to examine the evolution of interactions arising between a key regulatory gene, crp, and mutations that occurred during the adaptation of a bacterium, Escherichia coli, to a laboratory environment. To do this, we measured the effect of deleting crp on the expression of all genes in the organism, providing a sensitive measure to identify new interactions involving this gene. We found that deleting crp had a dramatic and parallel effect on gene expression in two independently evolved populations, but much less effect in their ancestor. An analysis of these changes identified a number of regulatory genes as candidates for harboring beneficial mutations that could account for the parallel changes. These findings indicate that epistasis has played an important role in the evolution of these populations, and they provide insight into the types of genetic changes through which epistasis can evolve.
| Epistatic interactions are revealed when the contribution of a mutation to an organism's phenotype depends on the genetic background in which it occurs. Epistasis plays an important role in many evolutionary theories, including those seeking to explain speciation [1], the evolution of sex [2–5], and adaptation [6–10]. In practice, however, epistatic interactions are usually difficult to study and their role in the evolution of organisms therefore remains unclear.
Approaches based on quantitative-trait loci have been increasingly used to study epistasis [11–15]. Although these techniques have the advantage of being quite general, they suffer from some shortcomings including low statistical power, difficulty in detecting some types of epistatic interactions, and inapplicability to non-recombining organisms [11,16]. Recently, systems-level approaches have been developed that avoid some of these problems [17,18]. These approaches typically assess epistatic interactions by comparing the individual and pair-wise effects of large numbers of defined mutations, allowing the outline of functional biological modules and biochemical pathways to be determined [19–23]. To date, however, most systems-level studies have focused on deletion and other knockout mutations, and it is not clear whether findings of widespread epistasis are representative of mutations involved in adaptive evolution.
Bacteria and viruses are ideal organisms with which to conduct controlled evolution experiments owing to their ease of culture and short generation times, as well as the capacity to store them in a non-evolving state from which they can later be revived to allow direct comparisons between ancestral and derived states (reviewed in [24]). These experiments have allowed examination of many aspects of adaptation, including a variety of studies on the nature and extent of epistatic interactions that affect evolution [25–33]. One aspect in common to most of these studies is that they assess epistasis through the effects of mutations on fitness or some related high-level phenotype. However, at the biochemical level, it is easy to imagine that interactions might combine to create a non-linear mapping to fitness [34]. Moreover, inference of epistatic interactions from fitness alone does not usually give any insight into their underlying genetic and physiological causes.
In this study, we combine a systems-level approach with a model experimental system to examine epistatic interactions that arose during the independent adaptation of two lines of E. coli to a glucose-limited minimal medium during 20,000 generations [35,36]. Specifically, we ask whether epistatic interactions occur between a key global regulatory gene, crp, and mutations that arose during the course of the experimental evolution; and, if so, to what extent these interactions evolved in parallel in the two lines. Parallel changes in independently evolving lines are of special interest because they are often characteristic of adaptive evolution, and thus call attention to those genes and phenotypic traits that have been important targets of selection [37–43].
We chose to examine interactions at the level of transcription and involving crp for several interrelated reasons. First, CRP (cAMP receptor protein, previously known as catabolite activator protein (CAP)) is a key hub in the E. coli transcriptional network. In fact, CRP is involved in more than 200 direct regulatory interactions [44–47], which makes it a good candidate to have evolved interactions with mutations fixed during the evolution experiment. Consistent with this possibility, the evolved lines underwent substantial changes in their carbon-utilization profiles, and CRP is known to play a key role in regulating use of carbon sources in response to environmental signals including glucose concentration [48,49]. The crp gene has not itself acquired any new mutations during the evolution experiment, but new epistatic interactions may have arisen between crp and mutations that evolved in other genes [49]. Second, we have shown previously that transcription profiles changed during this evolution experiment and that at least some of these changes were caused by evolved differences in regulatory interactions [37]. Finally, mutations in regulatory genes known to depend on crp have been identified in the evolving populations [40,50].
In order to detect epistatic interactions, we compared the effect of deleting crp on the expression profiles of the ancestral and two evolved strains. Thus, we focused attention on evolutionary changes that affected the CRP regulon, even though the sequence of the crp gene itself did not change at all during the evolution experiment [37]. Expression profiles represent high-resolution phenotypes consisting of many low-level traits, thus allowing detailed comparisons of the effects of deleting crp. If the set of genes with which crp interacts remained constant during the evolution experiment, then its deletion should cause the same expression changes in the ancestral and evolved strains. However, if new epistatic interactions evolved, or if the strengths of existing interactions have changed, then we expect the crp deletion to have different effects in the evolved lines from those in the ancestor. Also, we can address the extent of parallelism in evolved changes in epistatic interactions by comparing the expression profiles of the crp mutants in the two independently evolved lines.
Deleting the crp gene caused significant reductions in the growth rate of the Ara+1 and Ara-1 evolved strains and in their ancestor (paired t-tests, all p < 0.001). The growth rate of the crp+ genotype was 1.35 (SEM ± 0.03) times higher than that of the isogenic crp− deletion mutant in the ancestral genetic background. The corresponding growth rate differentials of the crp+ and crp− genotypes were 5.92 fold (SEM ± 0.13) and 13.01 fold (SEM ± 0.57) in the evolved Ara+1 and Ara-1 strains, respectively. In both cases, these differences represent significantly more deleterious effects of the crp deletion in the evolved strains than in the ancestor (t-tests, both p < 0.0001). These results are consistent with the possibility that interactions between crp and genes elsewhere in the genome were substantially altered during the evolution experiment.
We can define the CRP regulon as the set of genes whose expression levels depend on crp. This set is defined operationally as all those genes that show significant differences in standardized expression levels depending on whether the wild-type crp+ allele or the crp– deletion allele is present. We do not attempt to distinguish whether the regulation by CRP is direct or indirect. The questions we aim to address are whether the CRP regulon changed during the evolution experiment and, if so, whether it changed in parallel ways in the two independently derived lines. To assist our interpretation of this large and complex dataset, we further define the core regulon as those genes whose expression depended on crp in all three strains, and the meta-regulon as those genes dependent on crp in any of the three strains.
The Venn diagram in Figure 1 shows all the CRP regulons inferred from analyses of gene-expression data in the three strains. The small circle containing regions D, E, F and G represents the regulon inferred from the dependence of gene expression on crp in the ancestral strain. A total of 171 genes depended on crp in this strain, which is consistent with results of a previous study using a K-12 strain of E. coli [45]. The union of all seven regions, A-G, comprises the meta-regulon inferred by combining data from the ancestral and both evolved strains. The much smaller core regulon is represented by region E only, which is the intersection of the regulons independently inferred across all three strains. Scatter plots highlighting the relationships between crp-dependent genes across the three strains are shown in Figure 2.
Several features of this analysis are striking. First, the crp meta-regulon is very large, comprising some 1,089 genes (about 25% of all genes). Second, the ancestral and core regulons are both much smaller, containing only 171 and 25 genes, respectively. Therefore, the evolved strains exhibit greatly expanded CRP regulons, compared to the ancestor. Third, and more subtly, there is much more overlap in the CRP regulons for the two evolved strains than for either evolved strain and its ancestor. The intersection for the two evolved lines is about 14% [= (B + E)/(A + B + C + D + E + F)], while the intersection for each evolved line and the ancestor is only 7% or 8% for Ara-1 and Ara+1, respectively. Of course, the Venn diagram must be interpreted cautiously because each underlying datum – a test of the difference in gene expression between crp+ and crp− genotypes – is subject to statistical uncertainty. In the analyses that follow, we demonstrate that these striking features are real, and not mere artifacts, by performing a series of more constrained and rigorous statistical tests.
Statistical analyses must, in general, consider two types of errors: false positives and false negatives. False positives arise when a test yields a nominally significant result, while in fact there is no real difference between the comparison groups. False negatives occur when there is a difference between comparison groups, but a statistical test indicates the observed difference is not significantly greater than expected by chance alone. False positives are especially important to consider when one performs numerous tests on large datasets, as in this study, where we compare expression levels between crp+ and crp− genotypes for several thousand genes in each of three genetic backgrounds. We now employ two different criteria to address whether the results summarized above and in Figure 1 could reflect false positives. Importantly, these tests do not focus on the narrow inferences of whether each particular gene is part of a CRP regulon as defined above, but instead these tests focus on the broad inference of whether the expression data, viewed as a whole, support the summary results stated above.
The first test relies on a simple comparison between the total number of genes we identified as belonging to the CRP regulon and the number of false positives that might be included in the CRP regulon given the nominal significance level. That significance level, which was 5% in our tests, describes the likelihood of a false positive arising by chance. Given that there were 4290 genes in the analysis, one could expect 4290 × (1 − (1 − 0.05)3)) ≈ 612 false positives to be distributed approximately equally among the three genetic backgrounds that comprise the entire crp meta-regulon, where the exponent reflects the fact that independent tests were performed in three strains. This number is substantially less than the 1089 genes identified as belonging to the crp meta-regulon, and the discrepancy between observed and expected proportions is highly significant based on a binomial test (p < 0.0001). Therefore, while some false positives are undoubtedly included in the meta-regulon, the expression levels for many genes thus identified do indeed depend on whether the crp+ or crp− allele is present in one or more of the three strains.
The second set of tests asks whether there is significant concordance across the strains in the directionality of differences in gene expression depending on which crp allele is present. Four such sign tests are possible; one test involves all three strains and the other three tests include the various pairs of strains (Table 1). Region E is the narrowly defined crp core regulon, which includes only the 25 genes whose expression level was significantly affected by the crp allele in all three strains. If this core regulon was a statistical anomaly, whereby measurement noise led to false positives, then we would expect half of the comparisons to show higher expression in the crp+ genotype than in the crp− genotype, and half to exhibit the opposite pattern. Given three strains, we expect only 0.53 × 2 = 25% of the genes to show the same directional effect of the crp− deletion across all three of them. In fact, 21 of the 25 genes identified as the core CRP regulon show the same directional effect of the crp deletion in all three strains, and this pattern is significantly different from the expected 25% based on a binomial test (p < 0.0001). We are confident, therefore, that the core regulon includes mostly genes that do, in fact, respond similarly across these strains to the loss of the crp-encoded function.
Region B is of particular interest because it includes those genes that were not identified as belonging to the ancestral CRP regulon, but belong to the regulon in both independently evolved lines. Of the 117 genes in this region, 115 (98%) had the same directional change in both lines with respect to the effects of the crp deletion. This parallelism is dramatically different from the 50% correspondence that one would expect if measurement noise had generated spurious associations that defined this region of overlap (p < 0.0001). Of the 115 genes that showed parallel changes in the two evolved lines, 78 (68%) had higher expression in association with the functional crp+ allele, while 37 were more highly expressed when crp was deleted. Therefore, most genes that were recruited in parallel to the CRP regulon have increased reliance on the functional crp gene for their expression (binomial test, p = 0.0002).
Using the same strategy outlined above, we can be confident that only one of the remaining two regions is biologically meaningful. Regions D and F represent the overlap between the regulons inferred for the ancestor and for the Ara-1 and Ara+1 evolved strains, respectively (Figure 1). Region D shows an excess of genes that change in the same direction; of the 30 genes in this region, 26 (86%) change in parallel (binomial test, p < 0.0001, Table 1). By contrast, the number of genes that changed in the same direction in region F was not greater than expected by chance alone (11 of 18, 61%, binomial test, p = 0.2403, Table 1). This outcome does not mean that all of the genes in region F necessarily reflect spurious associations with the CRP regulon, but the statistical tests do not support that most of the associations are real based on the criterion of concordant directionality.
It is clear that many genes were recruited to the CRP regulon in the evolved lines. But have other genes lost their dependence on crp in the evolved lines? Of the 171 genes identified as the ancestral CRP regulon (regions D, E, F, and G in Figure 1), only 25 were significantly affected by the crp deletion in both evolved lines (region E). This discrepancy may indicate that many genes became decoupled from the CRP regulon, but this inference is subject to the problem of false negatives (i.e., failing to confirm a genuine association owing to limited statistical power). We therefore reframed the question to ask whether there was a positive association between those genes that were dropped from the CRP regulon in one evolved line with those that were dropped in the other evolved line. We performed a contingency test using all the genes comprising the ancestral CRP regulon to measure this association. Indeed, there is a highly significant association between loss of a gene from the ancestral regulon in one evolved line and its loss in the other evolved line (Fisher's exact test, p < 0.0001). Of the genes in the ancestral regulon that were retained in the CRP regulon by the Ara+1 evolved line (regions E and F), only about 42% (18/43) were dropped by the Ara-1 line. But of those genes in the ancestral regulon that were dropped by Ara+1 (regions D and G), some 77% (98/128) of them were also dropped by Ara-1. We conclude that the parallel recruitment of many genes to the CRP regulon in these evolved lines was accompanied by the parallel loss of other, albeit fewer, genes from the ancestral regulon.
These analyses indicate that the CRP regulon was substantially altered in the evolved lines relative to its ancestral state, with many parallel changes in the two evolved lines. These changes could have occurred by at least three distinct mechanisms, including any combination thereof. First, both lines could have evolved changes in the pleiotropic action of CRP. These changes could occur through mutations in either crp or cyaA; the latter gene encodes adenylate cyclase, the enzyme responsible for synthesis of cAMP [49]. cAMP binds to CRP to make the active regulatory complex, cAMP-CRP. A mutation in either of these genes could change the affinity of the cAMP-CRP regulatory complex for target gene operator sequences. To test this possibility, we sequenced the upstream and coding regions of both crp and cyaA in all three strains. However, we found no mutations in either gene, indicating that the evolved changes to the CRP regulon were not caused by differences in the cAMP-CRP complex itself.
A second possibility is that mutations were substituted in the evolved lines in each gene that was recruited to the CRP regulon so as to bring that gene's expression under the control of crp. Other mutations would be required for each gene that dropped out of the ancestral regulon. This hypothesis requires hundreds of mutations in each evolved line, on the order of the number of crp-dependent genes that were either recruited to or dropped from the CRP regulon during the evolution experiment. However, both theoretical and empirical evidence indicates that the number of mutations substituted in either of these evolved lines is less than 100 [36,51], and thus much lower than the number of changes to the CRP regulon supported by our analyses (Figure 1; Table 1). Moreover, only some of the mutations that were substituted could be expected to interact with crp. Therefore, we consider this explanation to be insufficient to explain our finding of widespread evolved epistatic interactions with crp.
A third hypothesis to account for the evolved changes to the CRP regulon is that mutations were substituted in a relatively few other regulatory genes that interact with one or more genes in the existing CRP regulon. The net effect of these few substitutions would be, in effect, to rewire the interacting gene-regulatory networks. Thus, under this model, genes under the control of these intermediaries would come under the influence of crp without requiring mutations at each of the affected loci.
To explore the nature of the epistatic interactions between crp and the mutations in the evolved strains, we analyzed crp-dependent expression changes in the context of the characterized E. coli transcriptional network. The network that we used is based on a comprehensively curated database of interactions (see Materials and Methods). This network comprises 1,217 genes, including 135 regulatory genes that mediate a total of 2,333 transcriptional interactions. If the increase in the number of crp-dependent genes was mediated by changes in the interactions between crp and other regulatory genes, then we expect the evolved changes to be modular, i.e., the changes in expression should be concentrated in regulons under the control of some of those regulatory genes. To test this prediction, we quantified the expression changes within all the characterized regulons using a previously described method (see Materials and Methods section for details) [52].
As shown in Figure 3, 20 of the 135 regulons we surveyed were significantly differentially regulated by crp in at least one of the evolved lines compared to the ancestor (using a double Z-score cutoff criterion of 2.0, which corresponds to p ≈ 0.05 [52]). Of these 20 regulons, 14 changed independently in both evolved lines, which is significantly more than expected by chance (Fisher's exact test, p < 0.0001). Moreover, all 14 of these regulons were affected in the same direction in both evolved lines (binomial test, p < 0.0001). Therefore, at the regulon level, as well as at the level of individual genes, epistatic changes were largely parallel. Twelve regulons – those controlled by crp, dgsA, dhaR, flhC, flhD, fliA, galS, glnL, hdfR, malT, mlc and rbsR – evolved to become less sensitive to the crp deletion. Only two regulons, controlled by lrp and ppGpp, a small molecule whose cellular concentration is controlled by the spoT and relA genes, became more sensitive to the crp deletion in the evolved lines than in the ancestor.
Curiously, the numerical imbalance at the regulon level is opposite in direction to our earlier finding that many more genes were recruited in parallel to the CRP regulon by the evolved lines than were dropped from the ancestral regulon. In any case, this regulon-based approach confirms that not all genes were individually recruited to, or dropped from, the CRP regulon, but rather many of them came ‘bundled' in terms of their associations with other regulons, consistent with our third hypothesis. It is also interesting, and consistent with this interpretation, that three of the genes identified as candidates on the basis of this regulon approach have been shown previously to contain beneficial mutations in one (spoT [37]) or both (malT [40]; and rbsR [50]) of the evolved lines.
We deleted crp, a major regulatory gene in E. coli, and measured the effect of this deletion on the expression profile of two strains that had evolved independently for 20,000 generations and on their ancestor. The two main effects of deleting crp in the evolved and ancestral strains can be summarized as follows. (1) The CRP regulon underwent massive changes during the evolution experiment. These changes reflect a large number of genes that were recruited to the regulon, and a smaller number of genes that were lost from the regulon, during evolution. (2) A striking degree of parallelism was observed in the recruitment of genes to the CRP regulon in the two evolved strains and in the pattern of crp-dependent changes across characterized regulons. Below, we discuss the significance of these changes to the CRP regulon with respect to an assessment of epistatic interactions between crp and mutations fixed during evolution.
The large number of expression changes caused by the crp deletion makes it difficult, if not impossible, to determine the precise physiological basis for the much larger reduction in growth rate of the evolved lines with this deletion. In fact, there may well be multiple factors involved, stemming from the ‘bundling' of regulatory modules that appears to have occurred in the evolved lines as a large number of genes came under the control of crp. Here we suggest one possible mechanism, but we do not mean to exclude other possibilities.
More highly connected computational and biological networks are thought to be more robust to the disruptive effects of environmental and genetic perturbations because they buffer against those perturbations [53–56]. Elimination of a network hub may thus reduce the network's capacity to resist perturbations. A recent analysis of the transcriptional connectivity of the E. coli genome predicted that deleting crp should substantially reduce that connectivity; on average, 3.96 transcriptional connections separate two randomly chosen genes in the wild-type network, increasing to 4.48 in an otherwise identical crp− network [56]. Experiments with the ancestral strain used here showed that disruptions of hub genes tended to reduce its robustness to environmental perturbations to a greater degree than did disruptions of other genes, although this pattern was not seen for robustness to mutations [56]. Theoretical models predict that beneficial mutations will tend to have deleterious pleiotropic effects, even if they confer a net benefit [57,58]. Therefore, a plausible explanation for the increased growth-rate sensitivity of the evolved lines to the crp deletion is that, without a functional crp gene, they are less well buffered to the pleiotropic side-effects of the otherwise beneficial mutations that arose during the evolution experiment.
The individual gene and regulon based approaches we used to characterize the overall evolutionary changes to the CRP regulon gave rather different patterns. Considering the individual gene approach, genes recruited to the CRP regulon in at least one evolved line outnumbered losses in at least one line by 918 to 146 (regions A + B + C vs. D + F + G, Figure 1; Table 1). If new epistatic interactions involving regulatory genes were driving these changes, we expected that, in turn, we would see an overall tendency for regulons to show increased crp-dependence in the evolved lines. By contrast, only 7 regulons showed significantly increased crp-dependence in one or both evolved lines compared to 13 with reduced dependence (Figure 3). Regulons are simply groups of co-regulated genes, so how can it be that more genes, but fewer regulons, became more dependent on crp during evolution? We consider three explanations for this apparent discrepancy.
First, the fraction of crp-dependent genes that were recruited to the CRP regulon could differ between the subset of genes included in the regulon analysis and the complete set of genes. If so, changes in regulon-level crp-dependence would not be predictable from the overall pattern of gene recruitment and loss. To address this possibility, we repeated our earlier analysis, comparing the number of genes recruited to, and lost from, the CRP regulon during evolution, but considering only those genes that were included in the regulon analysis (∼28% of total genes). We found that the proportion of genes recruited to the CRP regulon was indeed lower in this subset (Fisher's exact test, p = 0.0009), but the number of genes recruited was still several-fold higher than the number that were lost (241 gains, 67 losses). Thus, this effect alone does not alter the expectation that an evolved increase in the number of crp-dependent genes should also increase the number of crp-dependent regulons.
Second, the regulon analysis is limited to previously characterized regulons. Several mutations substituted during the evolution of the Ara-1 and Ara+1 lines have been shown to affect the physical topology, or supercoiling, of DNA [59]. Such changes are known to influence the accessibility of regulatory protein binding sites, which can thereby substantially alter the range of genes available to be regulated [60]. Indeed, one previous study has shown that there are a number of low-affinity CRP binding sites in the E. coli chromosome that can be influenced by topological changes [61]. If evolved topological changes altered the identity of genes that comprise a known regulon, it would make it harder to identify evolved changes in that regulon, because the actual set of co-regulated genes would have diverged from the presumptive regulon that is being tested.
Third, the particular statistical method we used to assess the significance of expression change within regulons may itself be more sensitive to detecting loss rather than gain of crp-dependence in the evolved populations relative to the ancestor. The test compares the effect of deleting crp on gene expression within a regulon to a baseline covariance calculated by repeatedly sampling the same number of random genes (chosen from among the 1217 genes included in our transcription regulation network). However, deleting crp had a much greater effect on the overall expression profiles of the evolved lines than on the ancestor's profile; thus, the distribution of covariances was broader in the evolved lines, making it more difficult to detect changes in regulon crp-dependence. Even so, we could identify significant changes in three regulons with known mutations in the evolved lines, which indicates that this analysis generated a biologically meaningful signal despite this constraint on statistical resolution.
It is also difficult to know how many of the changes to the CRP regulon that occurred during the evolution of the Ara-1 and Ara+1 lines indicate new or altered molecular interactions among regulatory genes and their products, and how many reflect indirect effects of physiological changes on gene expression. In particular, it is certainly possible that some of the expression changes reflect the growth-rate differences between strains, especially given the more severe growth-rate impairment caused by the crp deletion in the evolved lines. One potential approach to test for growth-rate effects would be to compare the gene-expression profiles in these strains when they are all forced to grow at the same rate, which could be achieved by growing them in separate but identical chemostat vessels, where the dilution rate is slow enough to allow each strain to sustain itself and reach steady-state [62]. In any case, the fact that differences in growth rate may mediate some of the altered interactions demonstrated by our experiments does not alter the conclusion that these changes reflect widespread epistasis between crp and the different genetic backgrounds. In other words, it may be misguided to take a complex web of genes and their interactions, which collectively determine growth rate, and then attempt to remove the effect of growth rate in order to say that the genes do not interact.
A key advantage of the regulon approach that we used is that it can identify specific genes that may mediate the new epistatic interactions. The overlap in the identity of the regulons affected by the deletion of crp marks their regulators as candidates for having substituted mutations in the evolved lines that interact epistatically with crp. Indeed, of the 14 regulatory genes controlling the regulons identified as having changed significantly in crp-dependence in both evolved genotypes, three of them – spoT, malT and rbsR – have been shown to contain beneficial mutations [37,40,50]. Two of these genes control well-characterized regulons: rbsR regulates the expression of six genes (including itself) that control ribose catabolism, and malT encodes a positive regulator of ten genes in two operons that control maltose catabolism. A cAMP-CRP binding site is located upstream of both of these regulatory genes [63]. The mutations in rbsR involve deletions of this gene, and the majority of the rbs operon, in both evolved lines, such that the operon is no longer induced by the cAMP-CRP complex in the low-glucose environment in which the cells were grown, explaining why the regulon became less responsive to the crp deletion. The mutations in malT produce amino-acid substitutions in the coding region of this gene in both evolved lines. Although the biochemical consequences of these mutations have not yet been determined, our results are consistent with previous findings that they reduce expression of the mal operons, making them less responsive to cAMP-CRP [40].
The underlying basis of the epistasis between spoT and crp is less clear; the spoT regulon is quite large and not as well characterized as the rbsR and malT regulons. The spoT gene influences cAMP-CRP activity by changing the intra-cellular concentration of ppGpp, thus providing a mechanism by which a mutation in this gene could affect the composition of the CRP-regulon [64]. However, there is no known reciprocal link between CRP-cAMP activity and ppGpp. Therefore, it is not clear how a mutation in spoT would change the dependence of the spoT regulon on crp. In future work, we will sequence regulatory genes in the 11 other regulons that changed their crp-dependence in both evolved lines, to determine whether they also contain mutations that might explain their altered interactions with crp.
In this study, we deliberately used a mutation of large effect – a deletion of crp – to examine as fully as possible the evolved changes to the CRP regulon. However, it is also interesting to consider whether such widespread changes would result from ‘smaller' disruptions such as point mutations. Although it seems reasonable to imagine that the average effect of point mutations would be smaller, it is important to remember that the evolved epistatic interactions revealed by deleting crp were caused by mutations in other genes. We do not yet know the identity of all these mutations, but those that we have discovered include a mixture of point mutations, deletions, and insertions [37,39,40,50]. Targeted reversion of these mutations, including even ‘small' point mutations, are likely to have an effect on the extent of the evolved epistatic interactions.
Two previous studies examining epistasis in the context of this same evolution experiment compared the effects of a set of 12 transposon-insertion mutations on the fitness of the ancestor and clones isolated from an evolved line at 1,000 and 10,000 generations [28,29]. Epistatic interactions between an insertion mutation and evolved mutations elsewhere in the genome were inferred when the effect of the insertion mutation on fitness was significantly different in the ancestral and evolved backgrounds. The ability of that approach to detect epistasis is limited by the resolution of fitness measurements. Moreover, fitness integrates effects over all underlying traits, such that epistatic interactions at lower levels may be obscured. The approach used here overcomes this limitation by treating the expression level of each gene as a phenotypic trait. The large number of such traits that can be surveyed simultaneously and with considerable precision may allow the detection of more subtle differences in the effect of a mutation in different genetic backgrounds, and this approach may thus provide more sensitive detection of epistatic interactions. Furthermore, the intimate relationship between genotype and expression levels offers greater opportunity for mechanistic insights into the nature of the epistatic interactions thus discovered.
In conclusion, we have used expression arrays to analyze the extent of epistatic interactions occurring between crp and mutations that fixed during 20,000 generations of evolution in E. coli. These epistatic interactions are widespread, and many of them emerged in parallel in independently evolving lines, indicating that the derived mutations – and perhaps also the altered web of interactions – contribute to the bacteria's adaptation to the experimental environment. Indeed, some of the parallel changes in epistatic interactions involve known regulons in which key regulatory genes have been shown to harbor beneficial mutations in the evolved lines. Finally, the use of expression profiles to detect epistatic interactions has the advantage that it does not require a priori knowledge of the molecular and physiological bases of interactions that are discovered, although this approach can be integrated with such knowledge to inform our understanding of genotypic and phenotypic variation.
Twelve lines of E. coli B were founded from two ancestral types differing only by a selectively neutral marker. These lines were propagated at 37 °C in a glucose-limited minimal medium for 20,000 cell generations [35,36]. Expression profiles were obtained from crp+ clones isolated from two of these lines (designated Ara-1 and Ara+1) after 20,000 generations of evolution and from the ancestor to one of those lines. These profiles were compared to otherwise isogenic crp− derivatives of these same three clones, constructed as described below. Only one ancestral type was examined here because no difference was detected between the expression profiles of the Ara- and Ara+ ancestors in a previous study [37]. During the evolution experiment, four of the twelve lines evolved mutator phenotypes; however, the mutation rate of both lines used in this study remained at the low ancestral level [51,65].
The crp (cAMP receptor protein) gene was deleted from the ancestral and evolved strains using a suicide plasmid-mediated approach described previously [50,66]. Briefly, a 1,016-bp PCR product containing a 591-bp in-frame crp deletion allele was cloned into pDS132, a derivative of pCVD442 [50,66,67]. This plasmid was transferred by conjugation into recipient cells, in which the deletion would be introduced, and chloramphenicol-resistant cells (resulting from chromosomal integration of the non-replicating plasmid) were selected. Resistant cells were streaked onto LB + sucrose agar to select cells from which the plasmid had been lost through intra-chromosomal recombination (pDS132 encodes the sacB gene conferring susceptibility to sucrose). These cells were then screened for the presence of the crp deletion allele by PCR and hybridization experiments. Several independent crp− strains were made in each background. All replicate crp− strains exhibited the same characteristic change in colony morphology relative to their respective crp+ progenitors. The crp− strains used in this study were further verified by sequencing the crp deletion allele and surrounding regions. This sequencing confirmed the deletion allele and ruled out the possible introduction of secondary mutations in the flanking regions of the introduced allele.
Prior to growth rate assays, all strains were grown for one day under the same conditions that prevailed during the evolution experiment, in order to ensure they were similarly acclimated to the culture conditions (37 °C, minimal medium supplemented with 25 mg/L glucose). Following this pre-incubation, strains were diluted 1:100 into 300 μL of fresh media contained in a well of a standard 96-well microtiter plate. This plate was incubated with periodic shaking at 37 °C and the change in optical density at 450 nm (OD450) was monitored using a Versamax Pro spectrophotometer. The maximal growth rate of each genotype was estimated by calculating the slope of the natural log of OD450 against time during the period of most rapid growth. Reported growth rates are the average of 10 independent estimates for each strain.
Expression profiles were generated using Panorama E. coli cDNA macroarray membranes (Sigma-Genosys). Prior to RNA extraction, all strains were acclimated to the same conditions used during the evolution experiment, then diluted 1:100 into fresh medium and grown to mid-exponential phase. Cells were harvested using 0.45 μm filter units (Nalgene) and re-suspended in a 1:1 mix of buffer and RNAlater RNA stabilizing solution (Ambion). RNA was obtained using the Qiagen RNeasy system, including an additional step to remove contaminating genomic DNA using the Qiagen on-column DNAse kit. Subsequent cDNA production, labeling, and hybridization were performed according to the instructions of the macroarray membrane manufacturer. Following hybridization, membranes were washed as per manufacturer instructions and exposed to Kodak PhosphorImager screens for 24 h. Exposed screens were scanned on a STORM 840 PhosphorImager (Molecular Dynamics). Image files were analyzed using Arrayvision software (Version 6.0, Imaging Research), and the output exported to Microsoft Excel for manipulation. Three independent mRNA preparations and hybridizations were performed for each strain. All RNA isolation and hybridization steps were done in complete blocks to control for any effects of day-to-day variation on results.
The Panorama macroarray consists of 4,290 PCR products, each corresponding to an individual open reading frame from E. coli K-12, spotted in duplicate onto a nylon membrane. To measure the expression level of each gene, we subtracted the average background from the mean of the two readings to calculate its adjusted expression level. To standardize expression and thereby account for possible differences between arrays, the adjusted expression levels were normalized to the median gene expression value obtained for the array. Standardized values were log10-transformed, and paired t-tests were performed using the transformed values from replicate arrays to identify genes whose expression had changed significantly. Expression values obtained from RNA samples isolated on the same day for crp+ and crp− derivatives of each strain were paired in this analysis. These tests were performed using the paired-data analysis option of the web-based Bayesian statistical program Cyber-T (http://cybert.microarray.ics.uci.edu/) [68,69]. A sliding window of 201 genes and a confidence limit of 2 were used in this analysis. This program reduces the number of false-positive differences expected in a comparison of replicate expression profiles from two strains through the use of Bayesian estimates of variance among replicate gene measurements. The use of a formal statistical test avoids arbitrary cut-offs, but it is still expected to identify many false-positives. To account for this, additional tests were performed on subsets of the data, as described in the text, using Microsoft Excel and the SAS frequency procedure (SAS Institute V8, 2000). These tests sought to compare observed distributions of crp-dependence to those expected by chance alone. Binomial tests were used to test the directionality of crp-dependent gene expression changes, reflecting our null hypothesis that genes identified by chance as being significantly changed in two genetic backgrounds would be equally likely to change in the same or in different directions. We also performed Fisher's exact tests to compare the numbers of genes changed in each direction to expectations given observed inequalities in the number of up- and down-regulated genes in the corresponding CRP regulons. In all cases, the outcomes of these Fisher's exact tests were qualitatively consistent with the reported Binomial tests (data not shown).
A transcription interaction network for E. coli was compiled from regulatory interactions downloaded from regulonDB [63] (data were accessed and downloaded on June 22, 2006), and from relevant data presented elsewhere [70,71]. We define regulons as groups of genes directly regulated by the regulatory genes present in these datasets. Regulatory genes were defined as genes with outgoing links to at least two other genes. Because our aim was to increase the sensitivity of measuring regulatory changes caused by altered activity of any regulatory gene, those genes that were co-regulated within operons were considered as having separate links to any gene that controlled the transcription of that operon. The regulatory genes spoT and relA were considered together because they combine to control the level of (p)ppGpp, a key regulatory molecule that binds to RNA polymerase and alters its affinity for transcriptional start sites [72,73]. Although spoT and relA control the same set of genes through their action of modulating levels of (p)ppGpp, they do so in response to different environmental conditions [74]. Moreover, they are involved in distinct sets of protein-protein interactions [75,76]. Therefore, despite the apparent regulatory redundancy, we expect that a mutation altering the regulatory capacity of either gene would be discernable through changes in the expression of co-regulated genes. In all, our transcription regulation dataset comprised 2,333 interactions and included 1217 of the 4290 genes assayed on the Panorama macroarray.
To measure the extent and significance of crp-dependent expression change within the 135 regulons identified in this dataset, we used an algorithm described by Balazsi et al. [52]. This algorithm calculates a double-Z score for each regulon based on the covariance of changes in expression caused by the crp deletion among genes in the regulon relative to an average background covariance calculated over 1,000 samples of an equal number of randomly chosen genes. This score therefore provides a measure of expression change within a regulon that controls for differences between paired crp+ and crp− strains in their overall measurement variability.
We also considered more extensive interaction groups that included indirectly regulated genes (origons in the terminology of ref. 52). Results obtained using these groups were qualitatively similar to those obtained in the regulon analysis (data not shown). |
10.1371/journal.pntd.0002722 | Pharmacological Approaches That Slow Lymphatic Flow As a Snakebite First Aid | This study examines the use of topical pharmacological agents as a snakebite first aid where slowing venom reaching the circulation prevents systemic toxicity. It is based on the fact that toxin molecules in most snake venoms are large molecules and generally first enter and traverse the lymphatic system before accessing the circulation. It follows on from a previous study where it was shown that topical application of a nitric oxide donor slowed lymph flow to a similar extent in humans and rats as well as increased the time to respiratory arrest for subcutaneous injection of an elapid venom (Pseudonaja textilis, Ptx; Eastern brown snake) into the hind feet of anaesthetized rats.
The effects of topical application of the L-type Ca2+ channel antagonist nifedipine and the local anesthetic lignocaine in inhibiting lymph flow and protecting against envenomation was examined in an anaesthetized rat model. The agents significantly increased dye-measured lymph transit times by 500% and 390% compared to controls and increased the time to respiratory arrest to foot injection of a lethal dose of Ptx venom by 60% and 40% respectively. The study also examined the effect of Ptx venom dose over the lethal range of 0.4 to 1.5 mg/kg finding a negative linear relationship between increase in venom dose and time to respiratory arrest.
The findings suggest that a range of agents that inhibit lymphatic flow could potentially be used as an adjunct treatment to pressure bandaging with immobilization (PBI) in snakebite first aid. This is important given that PBI (a snakebite first aid recommended by the Australian National Health and Medical research Council) is often incorrectly applied. The use of a local anesthetic would have the added advantage of reducing pain.
| Snakebite remains a major problem worldwide causing death or serious illness in many tens of thousands of victims annually. An approach to reduce the burden of envenoming is to provide optimum first aid procedures. We have previously shown that topical application of a nitric oxide (NO) donor slowed lymph flow to similar extent in humans and rats as well as increased the time to respiratory arrest by ∼50% for subcutaneous injection of eastern brown snake venom into the hind feet of anaesthetized rats. The present study examines the use of several other topical pharmacological agents that aim to slow venom toxins reaching the circulation through the lymphatic system. The study found that the agents examined were similarly effective to that previously found for the NO donor. The fact that one of these is a commonly used topical local anesthetic may be an ideal adjunct first aid, as it provides first aid while reducing pain.
| Snake envenoming worldwide remains a major health problem with 20,000–94,000 deaths [1] and a morbidity of several 100,000 people per year [2]. Snake envenoming has been identified as a neglected tropical disease and there is a desperate need for improved treatment. First aid procedures are a major part of treating snake envenomings because envenomings nearly always occur away from hospitals leading to delays before antivenom can be administered.
In Australia, the only formally accepted snakebite first aid is pressure bandaging with immobilization (PBI) [3], [4]. Local pressure pad with compression, another mechanical method, is also seen as a potentially useful first aid [5], [6], [7]. PBI, which has long been endorsed by the National Health and Medical research Council of Australia against bites from Australian snakes, aims to limit venom entry into the circulation by preventing lymphatic transport without inhibiting arterial or venous blood flow. Both mechanical methods are based on the principle that snake toxin molecules are generally large and can't directly enter the circulation, but are readily taken up by the lymphatics [8]. To date these first aid procedures have been primarily applied for elapid snake bites of Australia and New Guinea as these venoms have limited local cytotoxicity with the main concern being death by actions once venoms enter the circulation [9].
It has been shown that PBI is often incorrectly applied particularly by untrained personnel, with reported success rates of 15% in untrained and 50% in trained personnel [10]. Thus, while PBI is highly effective at limiting toxin entry into the circulation in animal studies and mock venom studies [11], [12], its application in the clinical setting may be more tenuous. Therefore, co-application of adjunct methods that impede venom transport through the lymphatics may be beneficial.
Recently, we reported that topical application of a nitric oxide (NO) donor that is known to inhibit lymphatic pumping [13], as a potential first aid against snakebite. However, there are various other agents that are also known to inhibit spontaneous contractions of lymphatic smooth muscle and hence the intrinsic propulsion of lymph. It is therefore important to investigate whether such agents act in the same way as NO donors with the hope that they may be even more effective or provide a cheaper or more advantageous first aid. The present study compares the efficacy of the NO donor to several other topical agents that have known inhibitory actions on lymphatic function by directly or indirectly targeting the intrinsic lymphatic pump.
All experiments were approved by the University of Newcastle Animal Care and Ethics Committee for ethics approval A-2009-153 according to the Australian Code of Practice for the care and use of animals for scientific purposes as released by the National Health and Medical Research Council of Australia in 2004.
Studies were performed on urethane (1.5–1.75 g/kg i.p.) -anaesthetized male and female Wistar rats (weight range 200–550 g) at 37°C with animals euthanized without recovery at the end of the experiments. In some experiments groin lymphatic vessels were surgically exposed to facilitate measurement of foot to groin lymph transit times [13]. Envenoming was simulated by injection of snake venom (Pseudonaja textilis, Ptx; Eastern brown snake) into the feet of deeply anaesthetized rats. Our use of this venom in rats was not intended to mimic P. textilis venom action on humans, but was chosen because of its clear end point in the rat model of our studies. In the present studies we used three separate batches of freeze-dried Ptx venom (each obtained from pooled venom collected from 10–20 snakes). While two donated by the Australian Reptile Park (Somersby, NSW, Australia) showed very similar potency, the third batch from Venom Supplies (Adelaide) was less potent. Therefore the third batch was applied at proportionally higher concentrations to equate to the toxicity of the other two batches, as tested by trialing different doses until a dose was obtained that produced a similar time to respiratory arrest as for the first two batches for a dose of 1 mg/kg.
Experiments involved examining the effects of topical application of test pharmacological agents (nifedipine – Sigma N7634, lignocaine - Sigma L1026, sodium nitroprusside – Sigma S0501) dissolved in control solution (i.e. saline with 1% dimethyl sulphoxide (DMSO – Sigma 276855) added to improve skin wetting). Comparison was also made to a commercially available NO donor ointment (Care Pharmaceuticals, Sydney) containing 0.2% wt/wt of the NO donor glyceryl trinitrate (GTNO). The test agents were applied over the rat hind limb within 1 minute after dye or venom injection with agents reapplied at ∼15 min intervals. The limb was wrapped with a thin layer of tissue to ensure the leg remained exposed to the agent for all experiments and maintained at a temperature of 35±1°C. Two types of experiments were performed. The first group of experiments tested the agent's effectiveness in slowing lymph transport by measuring the arrival time of a marker dye (India ink) in surgically exposed groin lymphatics consequent to injection of 50 µl of the dye into the corresponding hind foot of anaesthetized rats. The second group of experiments tested the agent's effectiveness in increasing the time to respiratory arrest of hind foot injection of Ptx venom.
Respiration frequency was measured visually or by recording chest movement through a strain gauge connected externally to the rat chest at the level of the diaphragm. Some experiments were made while simultaneously recording blood pressure and heart rate sampling data at 1 kHz (AD Instruments; Australia). Analysis of the rate of venom-induced decline of respiration frequency was generally made from the time of venom injection. In a minority of animals there was an initial increase in respiration rate and in these cases the respiration frequency was analyzed relative to its initial plateau.
Statistical significance confirmed by one-way ANOVA followed by Holm-Sidak's multiple comparison test. Data are presented as means ± s.e.m. with n referring to the number of animals.
The effectiveness of two topical pharmacological agents, the Ca2+ channel antagonist nifedipine and the local anesthetic lignocaine at slowing the transit of lymph was tested. Nifedipine (0.1 mM) or lignocaine (10%) applied topically within 1 minute after rat hind foot dye injection, significantly (P<0.0001) slowed the transit of lymph, increasing the hind limb transit time of lymph by 500 and 390% respectively (Fig. 1).
Simulation of snakebite by injection of Ptx venom into the feet of deeply anaesthetized rats caused a linear reduction in respiration frequency. The rate of decline was reasonably constant over duration of the experiment and was dependent on venom dose with lower venom doses resulting in a slower/shallower decline in respiration frequency. Linear regression of the mean respiratory frequency plotted as a function of time for venom doses of 1.0 and 0.4 mg/kg provided respective slopes and intercepts of −0.82±0.05 & −0.26±0.05 bpm/min and 120±1 & 116±2 min (Fig. 2). Consistent with this, reducing the venom dose increased the time to respiratory arrest (Fig. 3). Both parameters showed an approximately linear dependence with venom dose over the range 1.5 to 0.4 mg/kg with respective slopes of −53±7 min/(mg/kg) and −0.6±0.1 bpm/min/(mg/kg) and intercepts of 115±7 min and −0.3±0.1 bpm/min for the time to respiratory arrest and the rate of venom-induced decline of the respiration frequency (Fig. 4). Rats survived at lower venom doses (range studied 0.1–0.2 mg/kg, n = 8) for the 2–4 h periods measured. At such doses, rats exhibited a small decrease in respiration rate which then recovered.
The effects of topical application of nifedipine, lignocaine or control solution (1% DMSO in saline) to hind limbs were examined by measuring the time to respiratory arrest and the rate of decline of the respiration frequency in anaesthetized rats consequent to foot injection of Ptx venom. Topical application of nifedipine (1 mM) and lignocaine (10%) significantly increased the time to respiratory arrest by 61% and 50% (Fig. 5). The NO donor sodium nitroprusside when applied at 10 mM or 100 mM caused a similar increase in time to respiratory arrest (75±7 min; n = 7 and 80±6 min, n = 5). In preliminary studies we observed that lower doses of nifedipine and lignocaine were similarly effective as those shown at higher doses in figure 5. Specifically, the time to respiratory arrest for nifedipine at 0.1 mM as studied on 2 animals was 87 and 95 min compared to 99±10 min (n = 8) for 1 mM nifedipine. The time to respiratory arrest for lignocaine at 5% as also studied on 2 animals was 90 and 80 min compared to 91±4 (n = 5) for 10% lignocaine. Studies comparing blood pressure or heart rate during application of Ptx venom (equivalent concentration - 1 mg/kg) under control (n = 4) and one of the experimental conditions (topical 1 mM nifedipine; n = 8) indicate there was no significant difference in heart rate and systolic or diastolic blood pressures between test and control data or during the recording period before respiratory arrest.”
The proportional effect of lymphatic inhibition by the topical agent was not dependent on the venom concentration and hence time of venom action. For example, while the time period for respiratory arrest was longer being 75±7 min (n = 6) compared to 61±4 min (n = 11) for rats injected with a lower Ptx venom concentration (near 0.65 mg/kg compared to 1 mg/kg respectively), the degree of slowing afforded by topical hind limb application of GTNO ointment occurred at a proportionally longer time of 109±9 min (n = 6) compared to 88±7 min (n = 7), with corresponding ratios (i.e. +GTNO/−GTNO) which were very similar (1.44 vs. 1.45).
The pharmacological agents tested here all inhibited the lymphatic system, as shown by a 300 to 500% increase in lymph transit times and an approximately 50% increase in the time to respiratory arrest with simulated Ptx envenomation. These data build on our previous finding that topical application of the NO donor GTNO (0.2% wt/wt glyceryl trinitrate ointment) increased the foot to groin transit time of lymph and the time to respiratory arrest consequent to simulated snakebite by similar amounts [13].
The finding that these agents had the same effect despite targeting different mechanisms provides insight into lymphatic function and the role of the lymphatics in venom absorption. The release of NO from GTNO or nitroprusside acts through the stimulation of the soluble guanylate cyclase, formation of cyclic GMP and activation of protein kinase G, the latter causing reduction in intracellular Ca2+ concentration and a decrease in the sensitivity of the contractile system to Ca2+ [14]. The net effect is both smooth muscle relaxation and an inability to pace the muscle [15], [16], which disables the lymphatic pump. Nifedipine, inhibits lymphatic pumping by blocking L-type Ca2+ channels, the voltage dependent channels that underlie generation of action potentials, entry of Ca2+ and consequent lymphatic smooth muscle contraction [17]. Therefore, nifedipine most likely produces the same outcome as the NO donors, which is to directly inhibit the lymphatic pump. In contrast, the local anesthetic lignocaine will block nerve (i.e. Na+) -mediated action potentials [18]. This suggests that transmitter release, presumably dominantly from sympathetic nerves is a key factor in maintaining spontaneous activity of the lymphatics [19]. Neurotransmitters released from sympathetic nerves include noradrenaline (i.e. norepinephrine) and the co-transmitters neuropeptide Y and ATP [20], [21]. Therefore agents that inhibit all or at least the dominant sympathetic transmitter should also be useful as a topical first aid. Our preliminary experiments measuring lymph transit times using the α-adrenoceptor blocker phentolamine support this view (data not shown).
The finding that the different lymphatic inhibitors slowed but did not block venom entry could be due to several factors. First, it is to be noted that lymph drainage occurs through both superficial lymphatics, which lie just under the skin and the deep lymphatic vessels, which drain the limb musculature and other deep tissues. Thus while venom absorption by the lymphatics will generally first occur through the superficial lymphatics following snakebite from short fanged snakes (i.e. most elapids), anastomoses also allow entry of lymph into the deep lymphatics [22]. Therefore, inhibition of lymph flow in the superficial lymphatics will not entirely block active lymph flow, which may still be mediated by the deep lymphatics, these being less influenced by the topical inhibitors. Second, there may also be passive lymph flow as will arise if there is a net positive interstitium-lymphatic pressure gradient causing lymph to flow centrally. Such flow will be aided by factors such as arterial pulsations, fluctuations of central venous pressure and skeletal muscle contractions [23], though there was no evidence of the latter in these deeply anaesthetized rats. Third, there may be low-level permeability of the vasculature to venom toxins, allowing entry of the toxins directly into the circulation.
The effects of venom dose were also examined with time to respiratory arrest found to approximately linearly decrease with increase in venom dose over the lethal dose range of 0.4 to 1.5 mg/kg studied. The fact that there was a relatively sharp cut off in the dose response such that venom doses of 0.2 mg/kg or less were not lethal whereas doses of 0.4 mg/kg or higher were always lethal highlights the importance of first aid procedures that slow and preferably limit venom entry into the circulation. It is also interesting that there wasn't a dose dependency in the effects of the pharmacological agents tested, although this was only tested in a small number of animals for lignocaine and nifedipine. This is likely to be because these doses for all three agents were already maximally inhibiting the lymphatics.
The study approach of using anaesthetized rats ensured that there was no muscle movement and hence no confounding of the data through activation of the extrinsic lymphatic pump. It also obviated ethical issues related to envenoming conscious animals. Some anesthetics have an effect on the level of sympathetic nervous activity and hence lymph flow rate. It was for this reason we chose to use urethane, as it increases sympathetic nervous activity [24], [25] and hence better represents circumstances where a snakebite victim remains still but because of fear has elevated sympathetic nerve activity. Importantly, in our previous publication on this subject, we found that GTNO ointment proportionally slowed lymphatic flow in conscious humans to the same extent as in urethane-anesthetized rats (see Fig. 1; [13]). Another observation we now make is that lymphatic inhibition had a proportional effect independent of the time course of venom action (see Results). Therefore, even if intrinsic lymphatic flow rate is modified by the anesthetic, the topical agents should still have a proportional effect and hence their effectiveness can be reasonably assessed.
The key outcome of these studies is that a range of pharmacological agents known to directly or indirectly inhibit lymphatic pumping may be of use as topical treatments in first aid for bites from snakes whose venoms are not highly cytotoxic and where death by central action is the primary concern. It is suggested that they be considered as adjunct first aid to mechanical methods such as PBI (a snakebite first aid recommended by the Australian National Health and Medical Research Council), which while highly effective for limb bites are often incorrectly applied [10]. The topical agents might also be useful for bites to the torso and hence might be used as adjuncts to the local pressure pad with compression approach, which as indicated from animal studies is effective even for such bites [7]. Our previously reported findings for NO releasing ointment [13] indicate that a reasonable approach would be to apply an ointment formulation of the inhibitor just above the bite site. PBI or a local pressure pad with compression would then be applied. Finally, while all the compounds used in this study are used on humans for other purposes, the use of a local anesthetic is probably the most compelling, as it has the added advantage of reducing pain and hence may be readily adopted. However if used then the formulation should be rapidly acting, which was not the case for the commercially available formulations we tested (unpublished), but is the case for a commercially available NO releasing ointment [13].
|
10.1371/journal.pcbi.1000250 | Identification of Mechanosensitive Genes during Embryonic Bone
Formation | Although it is known that mechanical forces are needed for normal bone
development, the current understanding of how biophysical stimuli are
interpreted by and integrated with genetic regulatory mechanisms is limited.
Mechanical forces are thought to be mediated in cells by
“mechanosensitive” genes, but it is a challenge to
demonstrate that the genetic regulation of the biological system is dependant on
particular mechanical forces in vivo. We propose a new means of selecting
candidate mechanosensitive genes by comparing in vivo gene expression patterns
with patterns of biophysical stimuli, computed using finite element analysis. In
this study, finite element analyses of the avian embryonic limb were performed
using anatomically realistic rudiment and muscle morphologies, and patterns of
biophysical stimuli were compared with the expression patterns of four candidate
mechanosensitive genes integral to bone development. The expression patterns of
two genes, Collagen X (ColX) and Indian hedgehog (Ihh), were shown to colocalise
with biophysical stimuli induced by embryonic muscle contractions, identifying
them as potentially being involved in the mechanoregulation of bone formation.
An altered mechanical environment was induced in the embryonic chick, where a
neuromuscular blocking agent was administered in ovo to modify skeletal muscle
contractions. Finite element analyses predicted dramatic changes in levels and
patterns of biophysical stimuli, and a number of immobilised specimens exhibited
differences in ColX and Ihh expression. The results obtained indicate that
computationally derived patterns of biophysical stimuli can be used to inform a
directed search for genes that may play a mechanoregulatory role in particular
in vivo events or processes. Furthermore, the experimental data demonstrate that
ColX and Ihh are involved in mechanoregulatory pathways and may be key mediators
in translating information from the mechanical environment to the molecular
regulation of bone formation in the embryo.
| While mechanical forces are known to be critical to adult bone maintenance and
repair, the importance of mechanobiology in embryonic bone formation is less
widely accepted. The influence of mechanical forces on cells is thought to be
mediated by “mechanosensitive genes,” genes which respond to
mechanical stimulation. In this research, we examined the situation in the
developing embryo. Using finite element analysis, we simulated the biophysical
stimuli in the developing bone resulting from spontaneous muscle contractions,
incorporating detailed morphology of the developing chick limb. We compared
patterns of stimuli with expression patterns of a number of genes involved in
bone formation and demonstrated a clear colocalisation in the case of two genes
(Ihh and ColX). We then altered the mechanical environment of the growing chick
embryo by blocking muscle contractions and demonstrated changes in the
magnitudes and patterns of biophysical stimuli and in the expression patterns of
both Ihh and ColX. We have demonstrated the value of combining computational
techniques with in vivo gene expression analysis to identify genes that may play
a mechanoregulatory role and have identified genes that respond to mechanical
stimulation during bone formation in vivo.
| It is widely accepted that there is a relationship between the morphology of skeletal
structures and the mechanical forces acting upon them. Such a relationship begins in
the embryo where the importance of muscle for normal bone formation has been clearly
demonstrated [1],[2]; however, it is
still not understood how biophysical stimuli are interpreted and integrated with the
genetic regulatory mechanisms guiding bone development. Presumably gene activity
within the skeletal tissues is influenced by mechanical stimulation but there is
very limited information on how this might occur in the embryo. Up- and
down-regulation of gene expression due to mechanical stimulation has been
demonstrated under certain cell culture conditions, and these genes have been called
mechanosensitive genes [3]. Most experiments revealing mechanosensitivity
have placed cells under mechanical stimulation in culture and subsequently performed
analyses to quantitatively compare the expression of many genes between stimulated
and control cells, for example using microarray analysis (e.g., [4],[5]). Using
such an approach, hundreds of potential mechanosensitive genes can be identified
simultaneously; however, these experiments do not demonstrate the mechanosensitivity
of a gene in an in vivo context. To establish that a gene plays a mechanoregulatory
role during a particular process it is necessary to examine the sensitivity of the
gene to mechanical stimulation in vivo.
It is, however, more challenging to examine candidate genes in an in vivo context. To
date, the study of Kavanagh et al. [6] is unique in
demonstrating a mechanoregulatory role for a gene during embryonic development in
vivo by altering the mechanical environment. These authors examined the expression
patterns of three signalling molecules which are implicated in regulating joint
formation; growth and differentiation factor 5 (GDF-5), fibroblast growth factor-2
(FGF-2) and FGF-4 in control and immobilised chick embryonic hindlimbs and showed
that joint line FGF-2 expression was diminished in immobilised limbs, while the
expression of the other two genes in the joint line was unaffected. They concluded
that FGF-2 has a direct mechanoregulatory role in the cavitation process. Another
approach has been to use computational modelling to identify candidate
mechanosensitive genes, where regions predicted to be under high mechanical
stimulation are correlated with the expression of certain genes. Henderson et al.
[7]
used a 2-D finite element model to predict patterns of growth-related stresses and
strains generated during the growth of a skeletal condensation for comparison with
in vivo expression patterns of “chondrogenic genes” and
“osteogenic genes”. By comparing patterns of biophysical stimuli
with gene expression data from transverse sections, they proposed that predicted
patterns of pressure correspond with expression patterns of chondrogenic genes and
that predicted patterns of strain correspond with patterns of osteogenic genes.
Their model focussed exclusively on growth related biophysical stimuli and did not,
therefore, examine the effect of embryonic muscle contractions.
Considering embryonic bone formation specifically, a number of genes involved in key
steps have been identified as mechanosensitive in in vitro cell culture assays [3],[8]. These
include genes encoding Collagen X (ColX), Fibroblast Growth Factor receptor2
(FGFr2), Indian hedgehog (Ihh) and Parathyroid hormone-related protein (PTHrP). ColX
encodes a structural protein synthesised by hypertrophic chondrocytes [9] that has
been identified as playing a role in matrix mineralization [10], and was shown to be
upregulated in in vitro cultures of bovine chondrocytes under cyclic tension and
cyclic hydrostatic pressure [11] and in ex vivo mechanical stimulation of neonatal
rabbit distal femoral condyle explants [12]. FGFr2 is a
positive regulator of chondrocyte proliferation [13], and has been
shown to be downregulated following in vitro four point bending of MC3T3-E1
preosteoblasts [5] and upregulated in in vitro mechanical stimulation
of bone marrow stromal cells [14]. Ihh is also a positive regulator of
proliferation [15], and controls the onset of chondrocyte
hypertrophy primarily via PTHrP [16]. Ihh signalling from the proliferative region is
necessary to induce the differentiation of the perichondrium into an osteogenic
tissue from which the first osteoblasts will differentiate [15]. PTHrP signalling
has been shown to negatively regulate the switch from a proliferative immature
chondrocyte to a post-proliferative mature hypertrophic chondrocyte [17]. Ihh and
PTHrP have been shown to be upregulated by mechanical stimulation; Ihh and PTHrP in
in vivo mechanical stimulation of rat mandibular condyles [18],[19], Ihh in in vitro cyclic
mechanical stimulation of embryonic chick chondrocytes [20] and PTHrP in in vitro
cyclic mechanical stimulation of rat growth plate chondrocytes [21].
In this paper, we hypothesise that mechanical forces influence embryonic bone
formation by regulating expression of mechanosensitive genes. To test this
hypothesis, the involvement of four genes in transducing mechanical information from
spontaneous muscle contractions during ossification was assessed; these are ColX,
FGFr2, Ihh and PTHrP. The genes were selected for this study based on their
importance for bone formation and evidence of their mechanosensitivity in vitro.
Using a novel approach, the potential in vivo mechanosensitivity of these genes is
initially assessed using computationally derived data on the biophysical
environment. The candidate genes were first examined by correlating their expression
patterns with patterns of biophysical stimuli across stages of development when
ossification begins. We carried out a detailed analysis of expression of the 4
candidate genes and, by using the results of finite element analyses based on 3-D
rudiment morphologies and realistic muscle loading schemes described in a previous
paper [22], we could compare the complex and time-dependant
patterns of biophysical stimuli induced by embryonic muscle contractions with gene
expression patterns at several timepoints. To corroborate the correlations found,
the direct response of both the genes and the patterns of biophysical stimuli to a
perturbation in the mechanical environment in vivo were examined. If genes whose
expression patterns could be shown to have altered expression patterns in a
perturbed mechanical environment, then this would provide strong evidence that genes
mediate a genetic regulation of the response to mechanical information during
embryonic bone formation.
Morphological and gene expression analyses were carried out on the tibiotarsal
rudiment in the hindlimb of the embryonic chick. Dissected embryos were staged
according to the Hamburger and Hamilton (HH) system [23]. Three stages
were chosen for analysis; HH30, HH32 and HH34, corresponding to roughly 6, 7 and
8 days of incubation, spanning the initiation of osteogenesis in the
tibiotarsus.
The BBSRC (Biotechnology and Biological Sciences Research Council, U.K.) ChickEST
Database (http://www.chick.manchester.ac.uk/, last accessed September
2008) and bank of Expressed Sequence Tags (ESTs) from the chick genome were used
as a source of cDNA clones from which to generate specific RNA expression probes
for the genes of interest. The database was searched for ESTs corresponding to
each gene and two ESTs were selected for each based on confirmation of perfect
alignment with the gene of interest following a Basic Local Alignment Search
Tool (BLAST [24]) analysis through the National Center for
Biotechnology Information (NCBI, http://www.ncbi.nlm.nih.gov/BLAST/, last accessed September
2008), and on the length of the EST and its position within the cDNA of the gene
of interest. ESTs of 0.5–1.0 kb were preferred. The probe generated
for ColX was produced from chEST 62e2 and aligns with nucleotides
1605–2320 on Genbank sequence ref M13496.1. The probe generated for
FGFr2 was produced from chEST 699l24 and aligns with nucleotides
1967–2716 on Genbank ref NM_205319. The probe generated for PTHrP was
produced from chEST 533c1 and aligns with nucleotides 68–734 on
Genbank ref AB175678. The Ihh cDNA clone used for probe production was a gift
from C. Tickle (Dundee) and corresponds to nucleotides 2–547 on
Genbank ref NM_204957. The probe generated for Scleraxis was produced from chEST
654f15 and aligns with nucleotides 416–1109 on Genbank sequence ref
NM_204253.1.
Each EST clone was sequenced to verify identity. Plasmid DNA carrying the EST of
interest was linearized with appropriate restriction enzymes (EcoR1 or Not1).
Antisense and sense digoxigenin-labelled RNA probes were transcribed in vitro
from 1 µg of linearized plasmid using T7 and T3 promoter sites
(according to insert orientation) in the pBluescript II KS+ vector (all
components for in vitro transcription from Roche, Germany). DNA template was
degraded by incubation of probes with RNase free DNase (Roche). The probes were
then purified on G25 columns (Amersham Biosciences, USA) according to the
manufacturer's instructions. Probe concentrations were determined by
spectophotometry and probes were stored at −20°C.
After dissection, limbs selected for in situ hybridisation were fixed in
4% paraformaldehyde (PFA) in PBS over night, and dehydrated through a
series of methanol/PBT (PBT = 0.1%
Triton X-100 in PBS; 25, 50, 75%; 1×10 minute) washes,
followed by 2×10 minutes in 100% methanol and stored at
−20°C in 30 or 50 ml tubes until needed. On the morning of
sectioning, limbs were re-hydrated through a series of methanol/PBT (75, 50,
25%; 1×10 minute) washes at 4°C. After
2×10 minutes washes in PBT, excess tissue surrounding the skeletal
rudiments was removed in order to give optimal sectioning performance. The
specimens were embedded in 4% Low Melting Agarose/PBS (Invitrogen,
UK). 80 or 100 µm sections were cut in the longitudinal direction with
a vibrating microtome (VT1000S, Leica) and stored in PBS in 12-well plates.
After 2×10 minute washes in PBT, free-floating sections were treated
with proteinase K (20 µg/ml in PBT) for 5 minutes at room temperature.
Sections were then washed twice in PBT and fixed for 20 minutes in
0.2% glutaraldehyde/4% paraformaldehyde (PFA). Fixation
was followed by washes (3×5 minutes) in PBT at room temperature, and a
further 30 minute PBT wash at 55°C. The sections were then prehybridised
at 55°C overnight in a hybridization solution containing 2%
blocking reagent (Roche), 50% formamide, 5× SSC
(Saline-sodium citrate buffer), 0.5%
3-[(3-Cholamidopropyl-[(3-Cholamidopropyl)
dimethylammonio]-1-propanesulfonate (CHAPS), 500 µg/ml
Heparin, 1 µg/ml Yeast RNA, 0.1% Tween 20 and 5 mM EDTA
(ethylenediamine tetraacetic acid) (all components from Sigma, UK, unless
otherwise stated). Antisense and sense probes were denatured at 80°C for
3 minutes and sections were then incubated at 55°C over 2–3
nights in hybridization solution containing either antisense or sense probe at
minimum concentrations of 2 ng/µl.
Post-hybridization washes were carried out at 60°C as follows:
2×10 minutes in 2× SSC; 3×20 minutes in
2× SSC/0.1% CHAPS; 3×20 minutes in 0.2×
SSC/0.1% CHAPS. The sections were then washed for 2×10
minutes in TNT (100 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% Tween 20) at
room temperature and blocked in blocking buffer (0.1 M maleic acid, 0.15 M NaCl,
3% blocking reagent (Roche)) plus 10% goat serum overnight
at 4°C. Sections were incubated overnight in fresh blocking buffer (plus
10% serum) containing a 1∶1000 dilution of anti-digoxigenin
Fab fragments conjugated with alkaline phosphatase (Roche) at 4°C, with
rocking. The sections were then washed (5×1 hour) at room temperature
in TNT and left rocking in TNT over 2 nights at 4°C. On the day the
signal was developed, sections were washed in 3 changes of NMT (100 mM Tris-HCl,
pH 9.5, 100 mM NaCl, 50 mM MgCl2) for 15 minutes each. The
chromogenic reaction was carried out in NMT containing 17.5 µg/ml
4-nitro blue tetrazolium chloride (NBT; Roche) and 6.25 µg/ml
5-bromo-4-chloro-3-indolyl-phophate (BCIP; Roche). Sections were developed in
the dark at room temperature with rocking for 6–8 hours and then fixed
in 4% PFA/PBS for 1 hour before mounting on slides with Aquapolymount
(Polysciences, Inc).
Two sets of immobilisation experiments were performed at different timepoints;
named Set A and Set B. In Set A, 120 eggs were assigned as experimental embryos,
and 80 as controls, while in Set B, 100 eggs were assigned as experimental
embryos, and 80 as controls. The eggs were incubated for 3 days, after which 4
ml of albumen was removed with a syringe so that the embryo would sink lower in
the egg and a window could be cut in the shell without rupturing the
chorioallantoic membrane. Administration of the neuromuscular blocking agent
Decamethonium Bromide (DMB) [25] began at either day 5 (Set A) or day 6 (Set
B) of incubation. Embryos assigned to the experimental group were treated daily
with 100 µl of 0.5% DMB in sterile HBSS (Hank's
Buffered Saline Solution), while control embryos were treated with 100
µl of sterile HBSS. Before administration of the drug or saline
solution, movement of the embryo was observed and recorded, and dead embryos
were discarded. After treatment, the window was sealed using wide plastic tape
and the egg returned to the incubator. The treatment was repeated daily until
the embryos were harvested at days 8, 9 and 11, corresponding to stages
HH30–32 at day 8, HH32–34 at day 9 and HH35–36 at
day 11.
All harvested embryos were stained to reveal cartilage and bone using Alcian Blue
(cartilage) and Alizarin Red (bone) using a modification of the protocol of
Hogan et al. [26], with an Alcian Blue concentration of
0.l%. After staining, the embryos were photographed, and the total
length of the tibiotarsus and the length of the bone collar were measured for
each specimen. The numbers of control and experimental specimens at days 8, 9
and 11 are detailed in Table
1. These parameters were analysed in the statistical package R (http://www.r-project.org/, last accessed September 2008), and
standard t-tests were performed in order to determine the effect of
immobilisation on the morphology of the rudiments. The right limbs of embryos
harvested at day 9 were immediately removed for preparation for sectioning and
subsequent in situ hybridisation to analyse the expression of candidate
mechanosensitive genes. Sections were compared between control and immobilised
groups to determine if the altered mechanical environment had an effect on gene
expression.
As described in detail in Nowlan et al. [22] a set of finite
element analyses of embryonic chick hindlimb skeletal rudiments were created for
stages HH30, HH32 and HH34. At HH30 and HH32, the rudiments contain cartilage
only, while the periosteal bone collar is present at the mid-diaphysis at HH34.
Anatomically accurate rudiment and muscle morphologies were obtained for each
stage using Optical Projection Tomography (OPT) [27], and two animals at
each stage were analysed to ensure results were stage dependant rather than
animal-specific. In order to characterise the biophysical environment in the
absence of skeletal muscle contractions, simulations of the immobilised state
were carried out and compared with the previously published patterns.
Immobilisation using DMB induces rigid paralysis, where muscles are in
continuous tetanus [28]. To model this situation, both ventral and
dorsal muscle forces were applied simultaneously, as opposed to the situation in
a normal embryo, where ventral muscles are active in flexion and the dorsal
muscles in extension, as shown in Figure 1. The magnitude of the force per unit area value was also
adjusted in the paralysis simulations. From the study of Reiser et al. [29],
who reported the tension development in twitch and tetanic responses in normal
and immobilised chick embryos, we deduced that the tetanic force response from
the muscles in the immobilised chicks would be 75% of the twitch
response in normal embryos. We therefore adjusted the magnitude of each of the
muscle loads to 75% of the previously applied value.
The expression patterns of ColX, FGFr2, Ihh and PTHrP illustrated in Figures 2–5 are represented schematically by stage (Figure 6) and compared with patterns of
biophysical stimuli at longitudinal sections from the finite element analyses of
normal (control) limbs as described in Nowlan et al. [22]. The predicted
fluid velocity and maximum principal strain mid-flexion, (Figure 6) underwent distinctive changes over
the three stages examined, both at the ventral and dorsal surfaces (illustrated
as solid red curves), and in a longitudinal section through the middle of the
rudiment (Figure 6,
‘Normal’ sections). At HH30, stimuli levels were
at a high level on the perichondrium at the mid-diaphysis of the rudiment. At
HH32, two concentrations of stimuli were apparent proximal and distal to the
mid-diaphysis, again on the surface of the rudiment, and by HH34, these
concentrations moved further apart along the length of the rudiment, proximal
and distal to the newly formed bone collar [22]. Two genes showed a
correlation with the patterns of biophysical stimuli; ColX and Ihh, as their
expression followed patterns of events that reflect the stimuli patterns at the
same stages. ColX was found to be expressed in the region of hypertrophic
chondrocytes and in the region of the perichondrium where bone would soon form,
spreading proximally and distally beyond the hypertrophic zone domain at the
core at HH30 and HH32 and ahead of the bone collar at HH34. Therefore its
surface expression demonstrated a correlation with the patterns of biophysical
stimuli at each of the three stages examined (Figure 6). In the earlier stages examined,
Ihh was expressed uniformly across the pre-hypertrophic zone, in one
mid-diaphyseal region at HH30, and in two bands at increasing distances proximal
and distal to the mid-diaphysis at later stages, the expression bands moving
proximally and distally in synchrony with the biophysical stimuli at the
surface. Expression of Ihh was therefore at the same longitudinal position in
the rudiment as, and adjacent to, the peak levels of biophysical stimuli (Figure 6).
In this study, we set out to test the hypothesis that mechanical forces influence
embryonic bone formation by regulating certain mechanosensitive genes. In a first
analysis, the expression patterns of four genes; ColX, FGFr2, Ihh and PTHrP, were
characterised and compared with patterns of biophysical stimuli. ColX and Ihh
expression patterns correlated with stage-matched patterns of biophysical stimuli,
whereas FGFr2 and PTHrP expression patterns did not. This identified ColX and Ihh as
potential mechanosensitive genes regulating ossification in the embryo. ColX and Ihh
expression patterns followed the same dynamic sequence of events as the patterns of
biophysical stimuli, with one peak of expression at the mid-diaphysis at the
youngest stage (HH30), and two peaks progressively more proximal and distal to the
mid-diaphysis at HH32 and HH34. The ColX expression at the surface (on the
perichondrium) correlates with the locations of peak biophysical stimuli also at the
surface, while Ihh expression in the pre-hypertrophic cartilage is at the same
longitudinal position in the rudiment as, and adjacent to, the peak levels of
biophysical stimuli. In order to corroborate the hypothesis that ColX and Ihh may
act as mechanosensitive genes for bone formation in the chick limb, an
immobilisation assay was established, where rigid paralysis was induced with the
prevention of skeletal muscle contractions. The morphological analysis of the
immobilised embryos clearly demonstrated the effect of an altered mechanical
environment on skeletal development, with immobilisation leading to shorter
tibiotarsi and decreased bone collar formation. Finite Element Analyses of skeletal
elements under rigid paralysis indicated a dramatic alteration in patterns of
biophysical stimuli both in terms of stage-dependant patterns of biophysical stimuli
and magnitudes of stimuli in comparison with the normal case. Aspects of the
expression of ColX and Ihh indeed showed altered expression patterns following
immobilisation in a proportion of specimens; (see Figures 8 and 9), corroborating their role in mechanoregulation
pathways during ossification in the chick long bone.
The identification of Ihh as mechanosensitive in vivo is of particular interest since
this gene has been shown to be a key regulator of bone formation in the mouse, and
in particular formation of the bone collar, [15]. The elevation of
expression close to the periphery of the hypertrophic zone at later stages, as
described in the Results section, was precisely
the aspect altered in a number of immobilised specimens with an earlier and more
obvious peripheral elevation when mechanical stimulation was reduced (Figure 9) – this
indicates a more complex regulation of a gene by mechanical forces than a simple up-
or down-regulation on the level of expression. Alterations in Ihh expression would
affect the switch from a proliferative to a pre-hypertrophic chondrocyte leading to
a shorter rudiment [31], and shorter rudiments were indeed found in the
treated limbs. This indicates that mechanical stimulation may play a role in
regulating the position and timing of proliferation of immature chondrocytes through
Ihh signalling. As the simulations of the immobilised embryos did not exhibit a
specific pattern in the region of the pre-hypertrophic chondrocytes that would
explain the change in the Ihh gene expression profile, these results also indicate
the involvement of one or more molecules interacting with Ihh in one or more
mechanoregulatory pathways. Alteration to the expression of ColX was observed in the
regions predicted to have highest concentrations of biophysical stimuli (Figure 8B and 8C), where the
expression in the perichondrium did not extend proximal and distal to the
hypertrophic zone. The Finite Element simulations of the immobilised limbs indicated
that peak stimuli levels at the perichondrium at all three stages were dramatically
decreased due to rigid paralysis. It is possible that ColX may promote deposition of
osteoid on the perichondrium in response to peak levels of mechanical stimulation,
which would explain, at least in part, the reduced bone formation in the altered
mechanical environment induced by immobilisation. Alternatively, it is possible that
expression in the perichondrium does not extend beyond the hypertrophic region due
to an increase in the length of the hypertrophic zone. An elevated rate of
hypertrophy would lead to a shorter rudiment, as was indeed found in the immobilised
specimens in this experiment. However, the altered expression profile of Ihh does
not suggest an increase in the number of pre-hypertrophic chondrocytes. Therefore,
it is likely that one or more other mechanoregulatory molecules are involved, and
this will be a subject for future work.
In this study, there was a certain amount of variability in the effect of the
neuromuscular blocking agent, and the change in expression patterns of candidate
mechanosensitive genes were not seen in all immobilised (drug-treated) specimens.
This variability is not unexpected since the alteration to muscle contractions is
effected by exposure to a pharmaceutical agent where the response to a set dose can
vary across individual specimens. A variable response was also evident when movement
in the experimental embryos was quantified; while movement was clearly reduced, it
was not completely removed in all specimens. However, detectable changes in gene
expression were seen for two different genes in multiple specimens, showing a
repeatable effect, and the statistically significant decrease in rudiment length and
bone formation serves as confirmation of the immobilisation treatment as a means of
altering the mechanical environment. The magnitudes of the muscle loads applied for
the embryos subjected to rigid paralysis may be an overestimation, because while we
have assumed the same volume of muscle in our simulations, it has been widely
reported that muscle mass is reduced in immobilised embryos [32]. However, as the models
are likely to overestimate the muscle forces in a completely immobilised animal,
this will only strengthen our findings of the dramatic effect on the biophysical
environment due to paralysis. Another limitation of this research is that late long
bone ossification events are significantly different in mammals and birds [33], where
the long bones of birds are formed primarily via periosteal ossification as opposed
to a combination of periosteal and endochondral ossification in mammals. However,
birds and mammals have the events preceding ossification in common, such as
hypertrophy of the chondrocytes and formation of the periosteal bone collar, and
therefore genes identified as being mechanosensitive in vivo in the chick are likely
to have a similar role in the mammal.
The study presented here has revealed the alteration of gene expression as a result
of mechanical stimulation. Even though we have identified the in vivo
mechanosensitivity of two genes in the developing limb, we do not know what
signalling cascades prompted the change in ColX and Ihh expression patterns. For
example, focussing on Ihh in particular, while it has been suggested that Ihh
regulates proliferation of chondrocytes through the activation of stretch activated
channels by mechanical stimulation [20], it remains to be discovered what transcription
factors and other intracellular molecules form the link between stretch activated
channels and upregulation of the gene. As ColX and Ihh have now been demonstrated to
be involved in mechanosensitive pathways in vivo at specific developmental
timepoints, this opens the possibility of dissecting the upstream mechanisms
involved in the response.
Many researchers have recognized the importance of the interaction between mechanical
and biological factors for bone development. A range of biophysical stimuli
parameters have been hypothesised to promote ossification, such as low levels of
hydrostatic stress and principal strain [34], local stress and
strain magnitudes [35] or low levels of octahedral shear strain and
fluid velocity [36]–[39]. The results presented
in this study suggest that biophysical stimuli promote ossification through the
action of mechanosensitive genes, but it was not possible to determine a magnitude
or level of any particular biophysical stimulus necessary for normal
mechanosensitive gene expression. Although dramatic decreases in stimuli magnitudes
were found between normal and immobilised simulations within stages, the immobilised
stimuli magnitudes at HH34 are still higher than normal values at HH30 and HH32
(Figure 6). This may suggest
that the cellular response of cells to mechanical forces in the embryo is not
constant across different stages of development. It was also not possible from this
study to conclude the precise nature of the mechanical stimulus, (such as strain or
fluid flow), causing mechanotransduction. However, with new insight into the
interactions between mechanical forces and mechanosensitive genes, computational
simulations which incorporate biological and mechanobiological influences on
ossification may now be further developed to include specific mechanosensitive
genes. Van Donkelaar and Huiskes [40] have, in fact,
already developed such a numerical model, simulating the PTHrP-Ihh control loop and
its influence on growth plate development. The results of their simulation suggest
that the mechanical stimulation of Ihh is likely to have a greater effect than
stimulation of PTHrP, a result that was also suggested in this study, by the
correlation of gene expression patterns with biophysical stimuli. Our identification
of Ihh as being mechanosensitive in vivo further corroborates the findings of van
Donkelaar and Huiskes [40], and demonstrates that, with the identification
of other mechanosensitive genes in vivo, and the subsequent development of more
complex and detailed simulations, a deeper understanding of how biophysical stimuli
are interpreted and integrated with the genetic regulatory mechanisms guiding bone
development can be gained.
The work presented here has provided a new insight into mechanoregulation of
embryonic long bone ossification. This is the first study where finite element
analyses of the embryonic limb using anatomically accurate rudiment and muscle
morphologies have enabled comparison of predicted biophysical stimuli patterns with
gene expression patterns, and the characterisation of the biophysical environment in
the growing rudiment when skeletal muscle contractions are prevented. A means of
corroborating candidate mechanosensitive genes was proposed and tested, revealing
ColX and Ihh as mechanosensitive in vivo during embryonic bone formation, and also
identifying them as potential key mediators in translating information from the
mechanical environment to the molecular regulation of bone formation in the
embryo.
|
10.1371/journal.pgen.1003916 | The Fusarium graminearum Histone H3 K27 Methyltransferase KMT6 Regulates Development and Expression of Secondary Metabolite Gene Clusters | The cereal pathogen Fusarium graminearum produces secondary metabolites toxic to humans and animals, yet coordinated transcriptional regulation of gene clusters remains largely a mystery. By chromatin immunoprecipitation and high-throughput DNA sequencing (ChIP-seq) we found that regions with secondary metabolite clusters are enriched for trimethylated histone H3 lysine 27 (H3K27me3), a histone modification associated with gene silencing. H3K27me3 was found predominantly in regions that lack synteny with other Fusarium species, generally subtelomeric regions. Di- or trimethylated H3K4 (H3K4me2/3), two modifications associated with gene activity, and H3K27me3 are predominantly found in mutually exclusive regions of the genome. To find functions for H3K27me3, we deleted the gene for the putative H3K27 methyltransferase, KMT6, a homolog of Drosophila Enhancer of zeste, E(z). The kmt6 mutant lacks H3K27me3, as shown by western blot and ChIP-seq, displays growth defects, is sterile, and constitutively expresses genes for mycotoxins, pigments and other secondary metabolites. Transcriptome analyses showed that 75% of 4,449 silent genes are enriched for H3K27me3. A subset of genes that were enriched for H3K27me3 in WT gained H3K4me2/3 in kmt6. A largely overlapping set of genes showed increased expression in kmt6. Almost 95% of the remaining 2,720 annotated silent genes showed no enrichment for either H3K27me3 or H3K4me2/3 in kmt6. In these cases mere absence of H3K27me3 was insufficient for expression, which suggests that additional changes are required to activate genes. Taken together, we show that absence of H3K27me3 allowed expression of an additional 14% of the genome, resulting in derepression of genes predominantly involved in secondary metabolite pathways and other species-specific functions, including putative secreted pathogenicity factors. Results from this study provide the framework for novel targeted strategies to control the “cryptic genome”, specifically secondary metabolite expression.
| Changes in chromatin structure are required for time- and tissue-specific gene regulation. How exactly these changes are mediated is under intense scrutiny. The interplay between activating histone modifications, e.g. H3K4me, and the silencing H3K27me3 mark has been recognized as critical to orchestrate differentiation and development in plants and animals. Here we show that filamentous fungi, exemplified by the cereal pathogen Fusarium graminearum, can use H3K27 methylation to generate silenced, facultative heterochromatin, covering more than a third of the genome, much more than the 5–8% of Neurospora or metazoan genomes. Removal of the silencing mark by mutation of the methyltransferase subunit of the PRC2 silencing complex resulted in activation of more than 1,500 genes, 14% of the genome. We show that generation of facultative heterochromatin by H3K27 methylation is an ancestral process that has been lost in certain lineages (e.g. at least some hemiascomycetes, the genus Aspergillus and some basidiomycetes). Our studies will open the door to future precise “epigenetic engineering” of gene clusters that generate bioactive compounds, e.g. putative mycotoxins, antibiotics and industrial feedstocks. Availability of tractable fungal model systems for studies of the control and function of H3K27 methylation may accelerate mechanistic research.
| Histone lysine methylation provides an epigenetic layer for transcriptional regulation, with particular methylation sites associated with active (H3K4me2/3) or repressive (H3K9me2/3 and H3K27me2/3) regions of chromatin [1]. Polycomb group (PcG) transcriptional repressors that generate and read the H3K27me3 mark were first genetically identified in Drosophila as negative regulators of Hox developmental genes [2]; they repress many additional developmental regulators by generating “facultative heterochromatin” [3],[4],[5]. Certain genes can be associated with both activating (e.g., H3K4me2/3) and silencing (e.g. H3K27me3) marks, and thus form “bivalent domains” [6] that are thought to be metastable and poised for either repression or activation during differentiation and development. The precise location of the two marks across genes is different, however, as H3K4me3 is found at the transcriptional start site (TSS) or directly downstream of it, while H3K27me3 is found both up- and downstream of the H3K4me3 peaks [7].
PcG repressive marks are opposed by activating H3K4me marks that are established by Trithorax group (TrxG) proteins in Drosophila [8]. In human [9],[10], Arabidopsis [11],[12], and yeast [13],[14], active gene promoters are associated with H3K4me3, and H3K4me2 serves as an epigenetic memory of prior transcription. H3K36me3 methyltransferases are associated with elongating RNA polymerase and generate this mark at the 3′ end of transcribed genes [15].
With this study we are beginning to uncover an important physiological role for PcG proteins in fungi. Previously, H3K27me3 had been detected in Neurospora crassa [16] but its function in this species remains unclear [17]. In our studies on centromeres of the cereal pathogen Fusarium graminearum (teleomorph: Gibberella zeae), we found that H3K27me3 was absent from pericentric regions, similar to what has been found in plants [18],[19]. By chromatin immunoprecipitation (ChIP) followed by high-throughput DNA sequencing (ChIP-seq), we found that extensive segments, covering a third of the genome, were enriched with H3K27me3 in F. graminearum. Here we show that a Fusarium homologue of the Drosophila H3K27 methyltransferase Enhancer of zeste [E(z)], an enzyme we call KMT6 in accordance with proposed nomenclature [20], generates H3K27me3 marks that cover 46% of all genes. As expected, a majority of these genes is not expressed in wildtype cells, while in the absence of KMT6 an additional 14% of all genes were induced. These were predominantly genes involved in the production or detoxification of secondary metabolites or predicted to play a role in pathogenicity. We provide a chromatin-based model for coordinated expression of many gene clusters predominantly located in extant or ancestral subtelomeric regions and containing species-specific genes of unknown function.
We searched the currently available ∼200 fungal genomes for putative homologs of PRC1 and PRC2 components (Table 1). Budding yeast, Saccharomyces cerevisiae, and fission yeast, Schizosaccharomyces pombe, have been intensely studied but no H3K27 methylation has been detected. This is consistent with the absence of genes for known PRC2 or PRC1 complex components in their genomes [21]. Several other model fungi, e.g. the human pathogen Aspergillus fumigatus, the non-pathogenic Aspergillus nidulans and the widely used plant pathogen Ustilago maydis also lack genes for known PRC components. The human pathogens Cryptococcus neoformans and Cryptococcus gattii have homologues for KMT6 and EED with other PRC components not discernable by even low-stringency BLAST searches [21]. In contrast, many filamentous ascomycetes, such as the model organism N. crassa [17] and the widely studied genus Fusarium contain a full complement of PRC2 components, with one homologue each for EZH, EED, SUZ12 and the ubiquitous Nurf-55/RbAP46/48. All fungal genomes we have investigated, however, lack Pc or other PRC1 subunits, and the components of known PRC2-targeting complexes (Table 1), suggesting that gene repression by PRC2 is mediated by a mechanism that is different from that in plants or animals [21],[22],[23].
Here we focus on F. graminearum KMT6, the homologue of Drosophila E(z) and human EZH2 [22],[23], which has been identified in N. crassa as SET-7 [17]. Both fungal KMT6-type histone methyltransferases (HMTs) are significantly longer than the metazoan proteins (Fig. S1), and outside of the recognizable pre-SET (or CXC) and SET domains essential for HMT activity [24] there is little similarity to the metazoan proteins. Esc/EED-, Su(z)12/SUZ12- and RNA-interaction motifs found in metazoan KMT6 proteins [23] are not recognizable by sequence comparisons, yet long stretches between KMT6 proteins from various fungi are conserved, suggesting the presence of fungal-specific motifs (data not shown). The KMT6 CXC domain is characterized by cysteine repeats (four Cys-X-Cys motifs interrupted by variable length spacers and single C residues), similar to the canonical CXC motif [25], which is important for substrate recognition and enzymatic activity [24]. All four proteins have similar CXC domains (Fig. S1) when compared to other HMTs [24], suggesting that their catalytic motifs are more closely related than to HMTs with different substrates, e.g. KMT1, an H3K9-specific KMT. Important conserved stretches common to all bona fide HMTs are present in both KMT6 and Neurospora SET-7 (Fig. S1), including the invariable tyrosine that is involved in catalysis [24]. The residues preceding this tyrosine (GEELFF) are more conserved within KMT6 proteins than in HMTs in general, again suggesting that KMT6 and SET-7 are most similar to E(z) and EZH2.
To understand genome organization of F. graminearum, we performed ChIP-seq with antibodies against histone modifications known to be associated with active (H3K4me2) or silent (H3K9me3 and H3K27me3) chromatin. H3K4me2 and H3K27me3 were found in large, mutually exclusive, gene-rich blocks of the genome (Fig. 1). About one third of the F. graminearum genome is associated with H3K27me3 when the fungus is grown in minimal medium with low nitrogen; global H3K27me3 enrichment is slightly reduced in high nitrogen medium. More than half (58%) of chromosome 2 is covered by this silencing mark (Fig. 1).
Comparative genome studies suggested that the four chromosomes of F. graminearum are the result of chromosome fusion events, as the most closely related Fusarium species have between 11 and 15 chromosomes each, and SNP maps between two different strains suggested recombination patterns that mark some internal regions of F. graminearum as ancestral subtelomeres [26],[27]. We thus constructed synteny maps between F. graminearum, F. oxysporum, and F. verticillioides and compared them to histone modification maps (Fig. 1). H3K4me2 (green track) is found in well-conserved regions with high synteny between the three species. In contrast, H3K27me3 (orange track) is found in non-syntenic blocks unique to F. graminearum, predominantly in subtelomeric regions. In agreement with the chromosome fusion hypothesis, we found extended internal blocks of H3K27me3 on each chromosome that may constitute ancestral subtelomeric regions. H3K27me3 colocalizes with regions of high SNP density between the reference genome strain (PH-1) and a second wild-collected strain that we re-sequenced, 00-676 [28] (Fig. 1, black histogram). In complementary experiments we observed similar patterns for the distribution of H3K27me3 in other Fusarium species, e.g. F. verticillioides, F. asiaticum and F. fujikuroi (L.R. Connolly, L. Studt, K.M. Smith, B. Tudzysnki, S.-H. Yun and M. Freitag, unpublished data). Subtelomeric regions in filamentous fungi are enriched for lineage- or species-specific genes, for example secondary metabolite gene clusters, as well as genes for secreted pathogenicity factors and detoxifying enzymes [29],[30]. Our observations led us to ask if genes enriched for H3K27me3 are repressed, and if these genes become active when H3K27me3 is removed by mutation of KMT6.
We identified the F. graminearum kmt6 gene (FGSG_15795.3) based on BLAST searches with Drosophila E(z). We used targeted gene replacement to disrupt kmt6 (Fig. S2A). PCR and Southern analyses confirmed replacement of kmt6 in colonies that were exceptionally orange in pigmentation, suffered aberrant germination patterns and stunted growth. Southern blots showed absence of the kmt6 gene and replacement with neo+ (confers G418 resistance) in the mutant transformant (Fig. S2A). A kmt6 mutant (FMF248) with perfect neo+ integration was chosen for further studies.
To test if H3K27 methylation was altered in the kmt6 mutant, we purified histones and carried out western analyses with antibodies against methylated histones. H3K27me3 was present in WT but completely absent from the kmt6 mutant (Fig. 2A). We tested several different antibodies raised against H3K27me3 peptides (Fig. 2A, Fig. S2B). All showed absence of H3K27me3 in kmt6. Levels of H3K4me2 and another activating mark, H3K36me3, are equivalent in WT and kmt6 (Fig. S2B), suggesting that lack of H3K27me3 does not result in an overall increase in H3K4 or H3K36 methylation. H3K9me3, while present, proved difficult to detect in F. graminearum by western blot (Fig. S2B), matching our expectations from ChIP-seq (data not shown). Levels of H3K27me3 were not altered in strains in which H3K9me3 was abolished by deletion of the single Fusarium Su(var3-9) homologue (kmt1) or in which Heterochromatin Protein-1 (HP1) was deleted (hpo) (Fig. 2A). We conclude that KMT6 has specificity for H3K27, and that KMT6 is the sole or predominant H3K27 methyltransferase in F. graminearum.
We repeated ChIP-seq of H3K4me2 and H3K27me3, as well as H3K4me3 and H3K36me3, in WT, kmt6 and the complemented strain under nitrogen limiting and nitrogen abundant conditions (Fig. 2B, Table S1). We used different nitrogen levels as one environmental factor that is known to affect gene regulation in many fungi. In the kmt6 mutant, H3K27me3 enrichment was completely lost, and only background genomic sequence was obtained by ChIP-seq (Fig. 2B, Fig. S3). After re-introduction of a wildtype kmt6 allele, H3K27me3 was restored to levels almost indistinguishable from WT. Our results confirm that H3K27me3 is generated by KMT6, and that restoring gene function restores H3K27me3 enrichment in all regions by de novo mechanisms that may be similar to those in animals or plants.
Certain blocks previously enriched with H3K27me3 showed acquisition of H3K4me2/3 in kmt6 mutants. No obvious differences in enrichment were observed between high and low nitrogen conditions when viewed at the whole chromosome level. Similarly, discrete H3K36me3 enrichment in genic regions was not resolvable at the whole chromosome scale. H3K4me2 and H3K4me3 were found in overlapping regions, mutually exclusive of H3K27me3.
All kmt6 mutants obtained are sterile, have morphological defects, and are altered in pigment production. We observed slower linear growth and morphological changes in kmt6 compared to WT on both minimal (MIN) and rich (YPD) media (Fig. 3A–B). Plates are covered by WT after a few days, so we used Ryan (“race”) tubes [31] to carry out long-term growth experiments. We measured linear extension of WT and kmt6 for one month (Fig. 3B). On both plates and in race tubes, linear growth on minimal medium was faster than on the richer YPD medium, but colonies on YPD grew more densely. Overall, WT growth was more than two-fold faster than kmt6 growth. We compared growth of kmt6 and WT in a wounded tomato assay [32] and found that kmt6 was unable to colonize fruit (Fig. S4), suggesting that the mutant has reduced pathogenicity.
To show that kmt6 was responsible for the defects described, we complemented a mutant strain (FMF248) with a wildtype allele of kmt6 flanked by the hph+ gene (confers hygromycin resistance; Fig. S2A). The complemented strain (FMF282) retains the neo marker at the endogenous kmt6 locus, but has an insertion of kmt6 and hph at an ectopic locus. The intensity of kmt6 probing and multiple hybridizing bands in the complemented strain suggest multiple tandem insertions of the wildtype kmt6 gene. The complemented strain showed intermediate growth rates, faster than the mutant but not fully restored to WT growth levels, and almost normal pigmentation on minimal medium (Fig. 3 A–B). On rich medium (YPD), WT grew roughly two fold faster than kmt6; on this medium the complemented strain grew as well as WT (Fig. 3B).
Fusarium graminearum is a homothallic, or self-fertile, fungus. When placed on carrot agar (CAR), WT strains undergo sexual development to generate dark pigmented fruiting bodies, “perithecia” (Fig. 3C, CAR). Depending on environmental conditions (temperature, humidity), a selfing takes 10 to 14 days. After this time, ripe ascospores are shot or ooze from perithecia in cirrhi [33],[34]. We found that kmt6 is completely infertile and does not undergo even the earliest stages of sexual development. The complemented strain initiated normal development, though production of ascospores took about twice as long as for WT strains (Fig. 3C).
We attempted to force heterokaryons between kmt6::neo+ and hph+ strains of the same lineage. When co-inoculated, heterokaryons never formed on selective medium, suggesting anastomosis defects; only the hph+ sectors were able to generate perithecia with viable ascospores. We carried out protoplast fusions to complement the kmt6 deletion by formation of [kmt6+kmt6+] heterokaryons (Fig. 3C). No sexual development was observed even after extended periods of incubation, unlike for wildtype or complemented strains, which look similar to regular selfings but take 2–3 days longer to mature. We were able to isolate double-resistant G418+ and Hyg+ colonies from both the edge and center of the [kmt6+kmt6+] colonies on carrot agar, indicating that the heterokaryon had not broken down (Fig. 3C). We conclude that the sexual differentiation defect can be complemented by transformation but surprisingly not by fusion with mycelia that should be competent for sexual development. These experiments suggest existence of dominant factors produced by kmt6 nuclei that may inhibit H3K27me3 regulation in kmt6+ nuclei or act as dominant factors inhibiting sexual development. Action of these factors cannot be easily overcome by hyphal fusion and further studies to unravel this gene regulatory developmental switch are underway.
To distinguish genes that are enriched for a particular histone modification from those with background levels of ChIP-seq reads we used EpiChIP [35], which calculates values for “normalized locus chromatin state” (NLCS) and false discovery rates for each gene. The NLCS is the area under each ChIP-seq peak, in a specified window, normalized for the length of the window and the sequencing depth. For our analysis we used genes from the current Broad Institute annotation (http://www.broadinstitute.org/annotation/genome/fusarium_group/MultiHome.html) as the window, without addition of upstream or downstream sequences. For most histone modifications we found two peaks in the NLCS distribution (Fig. 4, right panels), one for background signals (B) and one for enrichment (E). In WT, H3K27me3 enrichment extended across gene bodies but was absent near the TSS, and genes with background levels (log2∼3) are clearly distinguishable from genes with enrichment (log2∼6). In kmt6, only background signal remained for H3K27me3, resulting in a single peak in the NLCS distribution. As expected, H3K4me2 and -me3 enrichment were most pronounced near the 5′ end of genes, while H3K36me3 was found more enriched near the 3′ end of genes (Fig. 4). A single peak is observed in the distribution of H3K36me3, but unlike H3K27me3 in kmt6, the single peak in H3K36me3 NLCS distribution represents enrichment. The kmt6 mutant revealed similar patterns of enrichment for H3K4 and K36 methylation across genes when compared to WT.
We performed RNA-seq on WT and kmt6 strains in high and low nitrogen conditions to investigate whether gene expression correlates with histone modifications in the expected manner. We used Tophat to map reads obtained by RNA-seq, and cufflinks to calculate reads per kilobase of transcript per million mapped reads (RPKM), a value representative of gene expression and normalized for both transcript length and sequencing depth [36]. For each condition we plotted the RPKM of each gene from each of two biological replicates (Fig. 5A). For most genes the replicates produced similar RPKM values, and all points fall near a line with a slope of 1. Not unexpectedly, most of the variation was observed in genes with low expression. Comparing expression of genes from WT or kmt6 at low compared to high nitrogen (Fig. 5B) showed that a relatively small percentage of genes has altered expression in response to nitrogen levels. Overall we observed a trend toward decreased gene expression in low nitrogen, shown by the smooth fit regression line (Fig. S5). The kmt6 mutation caused a larger change in global gene expression than changing nitrogen availability, a well-studied environmental factor affecting expression of known metabolites [37]. The overall trend was towards increased gene expression in kmt6. The distribution of RPKM values for all genes in each condition revealed that high nitrogen caused repression of only 5–10% of all genes in both strains, while the kmt6 mutation released repression of 15–30% of all genes; many of these were repressed by high nitrogen levels (Fig. S5).
To address if histone modifications are truly predictive of gene expression in F. graminearum, we show the range of RPKM values plotted against enrichment of histone modifications, expressed as the normalized NLCS values from EpiChIP (Fig. 5C). Overall, H3K27me3 enriched genes had low RPKM values; most genes enriched for H3K27me3 (NLCS>16) had RPKM values <10, indicating very low expression. There are several genes, however, with high H3K27me3 enrichment and RPKM>100, suggesting gene expression in the presence of a usually silencing histone modification. As one would expect, genes with high values of enrichment for H3K4me2/3 also tended to have higher RPKM values, and for both H3K4me2 and H3K4me3 genes with no enrichment tended to have low expression levels, suggesting a stronger correlation between enrichment with H3K4me2/3 and expression than presence of H3K27me3 and silencing. H3K4me2 was found in far more genes than H3K4me3. H3K36me3 was found in nearly all genes, regardless of expression level.
We classified genes grown in low nitrogen as expressed or silent in both WT and kmt6 strains based on the distribution of RPKM values (Fig. 5C). When comparing WT to mutant, we found that in WT 8,855 (or 66% of all annotated 13,354) genes were expressed (Fig. 5D). Of these, 1,627 genes were not associated with any of the histone modifications investigated. More than 30% of all expressed genes (2,760 of 8,855) were significantly enriched for the silencing H3K27me3 mark, though many of these genes showed H3K27me3 enrichment just above background levels. The 4,449 silent genes (33% of 13,354 annotated genes) in WT were largely associated with H3K27me3 (76% or 3,373 of 4,499), but almost 200 silent genes also had some significant H3K4me2/3 enrichment. When H3K27me3 is lost by kmt6 mutation, the number of expressed genes jumps to 10,635 (Fig. 5D); this number does not include genes that are expressed in WT yet are more highly expressed in kmt6. Only half of these genes are newly enriched for H3K4me2/3. The other half has none of the investigated modifications. Overall we found that about 14% of the genome is derepressed by absence of H3K27me3; many additional genes are overexpressed in kmt6 compared to WT.
We immediately realized that regions of KMT6-dependent repression are home to secondary metabolite (SM) gene clusters, and thus generated heatmaps of expression changes for all primary and secondary metabolite genes to visualize the effect of high compared to low nitrogen and kmt6 mutation on expression of these genes. Growth in low and high nitrogen was compared because nitrogen is a known regulator of many SM gene clusters [37]. Primary metabolite (PM) genes were largely unaffected by either mutation of kmt6 or growth in high nitrogen (Fig. 6A). Specific sets of genes, summarized in the clustered heatmap with 8 k-means (Fig. 6A, right panel) stand out as being repressed in high nitrogen (245 genes in cluster 7), or induced by kmt6 mutation (41 genes in cluster 5 and 49 genes in cluster 2). The genes repressed in high nitrogen include six out of 17 genes in the gluconeogenesis I pathway, and several genes for carbohydrate metabolism including glycolytic enzymes (Table S2). The 90 genes induced in kmt6 are enriched for genes involved in carbohydrate binding and degradation, peptidases, and cell signaling components. Five of the 90 genes, although classified as primary metabolic genes, are part of SM clusters (carB, fus1, tri5, a kinase belonging to the zon pathway, and FGSG_10615).
In contrast to PM genes, the complete set of SM genes (Fig. 6B) was overall more derepressed in kmt6 (34% of SM genes compared to 6% of PM genes), but a smaller fraction was repressed in high nitrogen (10% of SM genes compared to 18% of PM genes). The ten genes repressed in high nitrogen (Table S2) include most of the aurofusarin cluster (aurO, aur1, aurC, aurJ, aurF, gip1 and aurS), plus an ammonium permease from the carotenoid cluster, as well as pks1 and a multidrug resistance protein, both from SM cluster FG3_38, which generates an unknown product [27]. Since our nitrogen source was ammonium nitrate, it is not surprising that an ammonium permease was downregulated. The 36 genes derepressed in kmt6 (Table S2) include carO and carX from the carotenoid cluster, nine genes from the fusarin C cluster (fus1, fus2, fus3, fus5, fus6, fus7, fus8, and two other unnamed genes), five genes from cluster FG3_20 and six genes from FG3_40, which both generate unknown products [27]. These 36 genes are the most derepressed SM genes in kmt6 (log2 fold change >4, or more than a 16-fold induction). Many other genes are derepressed to a smaller but still significant degree.
To show that SM genes are found most often in KMT6-repressed regions we mapped genome-wide changes in gene expression (log2 kmt6/WT), distribution of H3K27me3, and genes for cytochrome P450 enzymes and gene clusters containing polyketide synthases (PKS) or non-ribosomal peptide synthases (NRPS) onto F. graminearum chromosomes (Fig. 7A). We also generated heatmaps of expression data for these groups of genes (Fig. 7B). Families of cytochrome P450s and PKSs are proposed to have evolved by gene duplication and divergence [38],[39]. Most cytochrome P450, PKS, and NRPS genes were enriched for H3K27me3 (Table 2). Several, but not all, cytochrome P450s were derepressed in kmt6, notably tri4 (FGSG_03535) and tri11 (FGSG_03540) involved in deoxynivalenol (DON) synthesis, and fus8 (FGSG_07804) in the fusarin C pathway. Other cytochrome P450 genes were repressed in high nitrogen, and generally derepressed in kmt6 to a much greater degree than when comparing high to low nitrogen conditions. These include a block of contiguous genes on chromosome 1, FGSG_02111, FGSG_02113, FGSG_02114, FGSG_02117, and FGSG_02118. The products of these genes are unknown, but the neighboring gene, FGSG_02115, encodes a TRI7 (toxin biosynthesis protein) homolog and FGSG_02116 encodes an NAD-dependent epimerase or dehydratase, suggesting the existence of a novel SM cluster.
The effects of KMT6 on the expression of known SM clusters with NRPS, PKS, DTC and STC signature genes are summarized and contrasted to the effects of nitrogen (Table 2). Of 45 clusters, 35 are enriched with H3K27me3 in both low and high nitrogen conditions, compared to 21 of 45 that are repressed by high nitrogen levels. In kmt6, 32 of the 45 clusters are expressed with low (11/32), high (6/32) or either nitrogen levels (15/32). In contrast, in WT only five clusters are expressed constitutively, five are expressed in high nitrogen, 14 in low nitrogen and 21 remained silent regardless of the nitrogen level. Most of these gene clusters have unknown functions and putative compounds generated have not been defined for 29 of the 45 clusters shown. Overall, manipulation of H3K27me3 levels proved more successful for expressing these “cryptic” clusters than changes in nitrogen level.
To illustrate effects of kmt6 at the gene level, changes in histone modifications and expression are shown for two representative SM gene clusters (Fig. 8). The fusarin C (fus) mycotoxin cluster was induced in kmt6 when H3K27me3 was lost, yet H3K4me2 enrichment was barely above background levels (Fig. 8A). Nearly every gene in the fus cluster was induced more than 64-fold in both low and high nitrogen. Both fus6 (G, FGSG_07803), encoding a transporter, and fus8 (I, FGSG_07804), encoding a cytochrome P450, acquired small peaks of H3K4me2. The genes that had increased expression also lost enrichment of H3K36me3. Overexpression of the fusarin C cluster genes can cause production of various fusarins [40].
The carotenoid cluster (car) encodes the enzymes required to synthesize the pigments neurosporaxanthin and torulene [41], resulting in the orange kmt6 culture liquid and mycelium grown on plates (Fig. 8 B and C). The transcription factor gene carR (F) was induced 3-fold in low nitrogen only, but in both high and low nitrogen the biosynthetic enzymes carO (B), carB (C), carRA (D), and carX (E) were induced more than 4-fold in kmt6. The carO gene acquired some H3K4me2 in high nitrogen, but none of the other genes in the cluster were enriched for H3K4me2. The reduction in H3K36me3, seen in the other examples at genes with increased expression, was most pronounced in the car cluster at gene G (FGSG_03069, dihydrodipicolinate synthetase). WT cultures of F. graminearum produce multiple dark red pigments in nitrogen limiting conditions, but expression is repressed under high nitrogen conditions in the dark [42]. In contrast, the culture supernatant of kmt6 was reproducibly bright red in low nitrogen, and turned bright orange in high nitrogen (Fig. 8C).
The predicted secretome [43], composed of putative effector proteins required for virulence and also including plant cell wall degrading enzymes, phytotoxins and antifungals, is largely encoded in the same regions of the genome where we mapped the SM cluster genes, and secreted protein genes are overwhelmingly enriched for H3K27me3 (Table S2). In summary, the partially overlapping sets for SM gene clusters and secretome genes are localized to subtelomeric regions, enriched for H3K27me3, and induced in kmt6. Many genes with unknown function in the same regions follow these general trends, and we predict that they also function in pathogenicity or niche adaptation. The newly found ability to express many of these genes in a single mutant and in vitro represents an important step forward in the functional characterization of natural products, not just in F. graminearum but also in a wide variety of additional species.
One of our goals is to understand the genome organization of F. graminearum and the various types of chromatin associated with specific regions. To this end we carried out ChIP-seq with antibodies against di- or trimethylated H3K4 (H3K4me2/3) as proxies for nucleosomes that are associated with active chromatin segments, or H3K27me3 for facultative heterochromatin. H3K4me2/3 and H3K27me3 were found in mutually exclusive, gene-rich blocks of the genome, as reported for mammals [44]. The patterns of histone modifications we observed differ from published reports of genome-wide patterns in other fungi. As mentioned above, both budding and fission yeast lack KMT6 homologs to generate H3K27me3 [21]. The best-studied filamentous fungus, N. crassa, has H3K4me2 in nearly all gene-rich chromatin, but large, heterochromatic, gene-poor, repeat-rich blocks near telomeres and centromeres are enriched with the silencing H3K9me3 mark [16],[45],[46]. In Neurospora, H3K27me3 is found in smaller blocks that cover genes and heterochromatic repeats close to telomere ends and these are exclusive of H3K9me3 [16],[17]. Overlap with H3K4me2 distribution has not been studied in detail in Neurospora. Various species of Aspergillus seem to use H3K9me3 to silence subtelomeric gene clusters [47], although genome-wide studies have not been published. All Aspergillus species lack clear KMT6 homologues (our data and [21]). This suggests that different clades of filamentous fungi make use of different chromatin-based regulatory systems to control SM gene clusters.
We assessed distribution of histone modifications across the “average” gene. Overall, H3K27me3, H3K4me2/3 and H3K36me3 distributions across genes and proximal promoters were similar to previous results from plants, fungi and animals: H3K4me2/3 were most pronounced near the 5′ end of genes, while H3K36me3 was found more enriched near the 3′ end of genes. In N. crassa H3K4me2 is enriched uniformly throughout the gene body [48], while H3K4me3 is enriched in 5′ ends of genes [45] and H3K36me3 is enriched in 3′ ends of genes [48]. Our findings agree with published animal studies [7, with the exception of the vast extent of H3K36me3 enrichment in nearly all genes. The H3K36me3 KMT is thought to function in association with elongating RNAP and only modify actively transcribing genes [15]. However, in F. graminearum nearly all genes are significantly enriched for H3K36me3, though almost half of all genes are not expressed in WT. Preferential enrichment of activating marks in exons compared to introns observed in Caenorhabditis elegans [49] was not found in F. graminearum. H3K36me3 can regulate mismatch repair by interactions with human MutS homologues [50], suggesting additional roles for H3K36me3 beyond transcription elongation. To uncover the meaning of the strong H3K36me3 enrichment will require additional studies. Nevertheless, the overall patterns of enrichment for H3K4 and H3K36 methylation across genes were not altered in kmt6, suggesting that there is little feedback into genic distribution of these marks by H3K27me3 or other putative activities of KMT6. Enrichment of H3K4me2/3 at certain genes was altered in the absence of H3K27me3, suggesting that nucleosomes with activating marks are incorporated into chromatin in the absence of the silencing H3K27me3 modification.
While 627 genes were enriched with H3K27me3 and H3K4me2/3, a hallmark of “bivalent” regions [6], most of these genes had strong H3K4me2/3 enrichment in combination with H3K27me3 enrichment just above background levels. Examination of genes with strong enrichment for H3K4me2/3 and H3K27me3 did not reveal a functional enrichment for any particular group of genes. Thus, our data suggest that bivalent promoters or genes can occur in F. graminearum but additional work on the biological function of these regions is needed to confirm results from our genome-wide analyses. In Arabidopsis thaliana, expressed genes are associated with H3K4me3 and repressed genes are associated with H3K27me3, but 13% of genes are marked with both modifications, including genes with tissue-specific expression and for some TFs that are poised for transcription [51]. Many individual genes, however, are thought to have one or the other modification, where the observed bivalency may have been caused by fractions of mixed nuclei [51], something we cannot exclude for F. graminearum.
There were 331 genes enriched with H3K27me3 that showed at least twofold decrease in expression in kmt6. These include all four ammonium transporters, three out of six nucleoside permeases, amino acid and oligopeptide permeases, and hydrolases. As recently discussed for Drosphila [52], kmt6 and the genes for the other PRC2 components showed significant H3K27me3 enrichment while they were expressed. The observed decrease in expression of some genes upon loss of H3K27me3 may be due to indirect effects, but it remains possible that H3K27me3 is in some cases required for transcription, which will be subject to further investigation.
Lack of H3K27me3 resulted in activation of ∼14% of all predicted or known genes (1,780/13,354) that were silent in WT. It remains to be seen how many genes are activated directly (e.g. by virtue of “poised” promoters) and how many are activated indirectly (e.g. by involving additional cis- or trans-acting factors that are controlled by KMT6). Many genes (2,720/13,354 or ∼20% of the genome) remain silent even in the absence of H3K27me3, and 2,575 silent genes possessed none of the investigated modifications. This suggests that while transcription may be the default state in the absence of H3K27me3 regulation for many genes, additional activating factors may be required or some genes are subject to multiple layers of repression.
In budding yeast, H3K4me2 is found in all euchromatic genes, and H3K4me3 is found in actively transcribing or recently transcribed genes [14],[53]. While our study does not address issues of RNA stability, it is likely that many actively transcribing genes in F. graminearum lack H3K4me3 under our conditions, are associated with H3K4me2 or even with unmodified H3K4. Our results suggest that “activating” histone modifications are not absolutely required for transcription and that their deposition at transcribed regions is slow and perhaps a secondary event to transcription.
Taken together our results suggest a testable model in which absence of the silencing mark H3K27me3 removes an immediate block to transcription, allowing access to promoters by the basal transcription machinery or completion of the initiation phase of transcription by pre-assembled, or “poised”, transcription machineries on promoters. It appears that activating histone modification marks, such as H3K4me2/3 are only much later, if ever, deposited on these actively transcribing regions, as our growth experiments were carried out over several days. Curiously, the appearance of H3K4me2 appears to correlate with a reduction in H3K36me3 modification, even though all previous data supports the acquisition of both marks before or during transcription [14],[15],[53],[54].
Pigment production was very much altered in kmt6; the exact pigment profile of kmt6 grown under various conditions is the subject of an ongoing study (K.M. Smith, J. Gautschi, L.R. Connolly, M. Freitag, unpublished data). From the kmt6 expression data, some of it summarized in Table 2, it appears that repression by high nitrogen can be overridden by loss of H3K27me3. Overall, loss of H3K27me3 had more drastic effects on expression of SM gene clusters than the intensely studied regulation by nitrogen. For this study we did not measure concentrations of specific known or unknown metabolites, but previous work from several laboratories suggests that increased transcription from SM clusters by manipulation of nitrogen levels, histone H3K9 acetylation or H3K4 methylation levels results in overproduction of certain metabolites [37],[40],[55],[56].
How exactly linear growth is retarded in kmt6 may be difficult to ascertain. One possibility is that increased synthesis of pigments, other secondary metabolites and detoxifying enzymes may account for the slower growth of kmt6, either indirectly by shifting energy utilization away from primary metabolism or by direct toxic effects mediated by combinations of usually harmless metabolites. For example, the red pigment, aurofusarin, is synthesized by the aur gene cluster, which includes the polyketide synthase gene aur1/pks12 [42],[57]. A Δaur1 mutant grew faster and generated more conidia than WT on media inducing aurofusarin production [57], suggesting that costs are incurred by the production of specific metabolites. It remains to be seen how or if H3K27 methylation is altered in aur1 and similar mutants.
Attempts to activate individual silent clusters for chemical genome mining has largely focused on overexpression of cluster-specific regulators, mostly transcription factors [58],[59],[60] or heterologous expression of partial or complete clusters [61,[62],[63]. A more general approach to activate silent gene clusters involves treatment with inhibitors of DNA or histone modifying enzymes. Silent clusters were activated by histone deacetylase inhibitors or DNA methyltransferase inhibitors in Cladosporium cladosporioides [64] and A. niger [65], or by cocultivation with other organisms, e.g. bacteria or fungi, to mimic a natural environment [56],[66]. These approaches were successful in inducing a few SM clusters, but they are little different from previous attempts to find the exact culture conditions for expression of specific gene clusters. Another strategy to activate silent clusters is focused on global regulators of secondary metabolism. Mutation of selected histone-modifying enzymes predicted to be global gene regulators, e.g. the histone deacetylase HdaA [67], the CclA component of the H3K4MTase complex [55],[68] and the histone acetyltransferase EsaA [69] proved successful in affecting certain clusters in Aspergillus. Individual SM gene clusters are affected in different ways by mutating or overexpressing these enzymes and the effects are not specific to SM clusters.
The most widely studied general regulator is the “Velvet complex”, first identified in A. nidulans [70] and later also found in F. verticillioides [71], F. fujikoroi [72] and F. graminearum [73],[74]. This complex consists of the putative transcription factors VeA, VelB and a putative methyltransferase, LaeA [75] and regulates both fungal development and SM production. In A. nidulans, the complex inhibits asexual reproduction, promotes sexual development and increases SM production in dark conditions [70]. In A. fumigatus, 13 of 22 SM clusters and 20–40% of SM biosynthetic genes were expressed at lower levels in a ΔlaeA strain compared to WT [76]. The molecular mechanism of this pathway presumably involves the methyltransferase domain of LaeA [76]. LaeA controls protein levels and complex interactions between VeA and its partners [77], a function separable from its role as global regulator of SM gene clusters. The precise function of VeA, VelB, and LaeA in changing transcriptional programs has not yet been determined, but it has been suggested to involve re-programming of the constitutive heterochromatin mark, H3K9me3 [47],[78],[79].
Here we revealed a novel mechanism that links fungal development and SM expression, and that appears at least partially conserved with formation of facultative heterochromatin in plants, Drosophila and mammals by generation of blocks of H3K27me3-enriched chromatin. Essentially these blocks generate a “cryptic genome” under normal laboratory culture conditions. There is no indication that this process is dependent on members of the velvet complex. We looked for changes in expression for the “white collar” genes (i.e. the light-sensing complex that controls VeA activity), VeA, VelB, VosA and LaeA, and found no differences in expression between WT and kmt6. All of these genes were enriched for H3K4me2 and expressed in both WT and mutant. We did, however, find changes in several predicted LaeA homologs whose functions are still largely unknown. Proteins encoded by these genes were found to interact with F. graminearum VeA and named “FgVeA interacting proteins”, or VIP [73]. Conserved VIPs are FgVIP1 (FGSG_07660), FgVIP2 (FGSG_03525), FgVIP3 (FGSG_05685), FgVIP4 (FGSG_03567), FgVIP5 (FGSG_08741), and FgVIP6 (FGSG_03011). All VIP genes were enriched for H3K27me3, and loss of this modification in kmt6 caused increased transcription. Homologues of VIPs have been studied in A. nidulans, where LlmF (LaeA-like methyltransferase) interacts with velvet components and appears to shuttle the complex into the nucleus [80]. Thus it appears possible that LaeA homologs are involved in H3K27me3 regulation.
One wonders why gene family expansions and acquisition of SM clusters occurs preferentially in subtelomeric locations. Subtelomeric regions of Aspergillus species contain numerous SM clusters, and based on previous results one would expect to find large H3K9me3-enriched domains in these regions [81],[82] but this remains to be demonstrated. There is evidence, at least in Magnaporthe and Saccharomyces [83],[84],[85],[86], that subtelomeric regions are more prone to rearrangements than other regions of the genome. Published synteny maps of F. graminearum, F. verticilliodes, and F. oxysporum, as well as our preliminary results from studies with a close cousin of F. graminearum, F. asiaticum (L.R. Connolly, K.M. Smith, S.-H. Yun, M. Freitag, unpublished data), show that subtelomeric regions are hypervariable between related organisms and accumulate SNP mutations at higher rates than other regions of the genome (Fig. 1). This suggests a model in which H3K27me3 is involved in the regulation of recombination or chromosome rearrangements.
Why are SM genes in clusters? This can be explained by the “selfish cluster” hypothesis [87], at least if horizontal gene transfer is not exceedingly rare. Genes in a cluster are more likely to be transferred as a functional group if acquisition and loss of clusters is adaptive to the organism. Selective advantages to the new host organism, specifically by creation of novel clusters and maintenance of all clusters, however, remains unclear. Initially, uptake of novel DNA would not be dissimilar from invading transposable elements, which tend to be silenced by a combination of H3K9me3 and DNA methylation in N. crassa [46],[88]. Further partitioning of the genome into additional chromatin domains by making use of H3K27me3 that eventually results in coordinate regulation is a plausible hypothesis to explain the maintenance of secondary metabolite genes in clusters. One wonders if subtelomeric silencing depends on PcG proteins in other fungi. So far, we only have data for N. crassa [16],[17] and several Fusarium species, but many important animal and plant pathogens within the ascomycetes have predicted orthologues for PRC2 components.
Strains were grown in liquid YPD to collect vegetative tissue. To generate macroconidia, a small amount of frozen conidia or tissue was inoculated into 50 ml flasks containing CMC medium [89] and shaken at 150 rpm for 3–4 days at room temperature (RT, ∼22C). Conidia were collected by filtration through cheesecloth and stored at −80C in 25% glycerol. For vegetative growth assays strains were inoculated onto YPD (0.3% yeast extract, 1% bacto-peptone, 2% dextrose) or Fusarium Minimal Medium (FMM; [90]) agar plates. Crosses were performed on carrot agar at RT, taking usually ∼10 days. To assay pigment production, tissue was generated from macroconidia by shaking 100 ml cultures at 150 rpm in the dark in DVK medium (3% sucrose, 1.5% corn steep solids, 0.1% (NH4)2SO4, and 0.7% CaCO3) for three days, after which 5 ml were used to inoculate 100 ml of liquid ICI medium [91] with 6 mM or 60 mM NH4NO3 for nitrogen limiting or sufficient conditions, respectively. Cultures were grown at 25C at 150 rpm in the dark and observations made after 3 and 7 days of growth.
Replacement cassettes with the selectable hygromycin (Hyg) resistance marker (hph+), and neomycin/G418 resistance marker (neo+), encoding hygromycin and neomycin phosphotransferase, respectively, were generated by fusion PCR [92]. The 5′ and 3′ flanking regions of the kmt6 coding region were amplified from genomic DNA of PH-1 (FGSC9075, FMF1) with primers OMF1936 (5′-TCTTGGATATTGGCCAGCTC-3′) and OMF1930 (5′-GATAAGCTTGATATCGAATTCTTACTTGTGGCTfGCGGCTAATTGATGGCT-3′) or OMF1931 (5′-TGCTATACGAAGTTATGGATCCGAGCTCGTTTGGGCAGAGAAGCTTGAATA-3′) and OMF1937 (5′-GTGGAGGGAAAACTTGGTGA-3′), respectively. The loxP-neo-loxP cassette was amplified from pLC13-Tom-loxP-neo-loxP with primers OMF1148 (5′-ACAAGTAAGAATTCGATATCAAGCTTATC-3′) and OMF84 (5′-CGAGCTCGGATCCATAACTTCGTATAGCA-3′). The 5′ and 3′ kmt6 flanks were fused to the neo+ cassette by PCR with neo split marker primers OMF601 (5′-AGGCGATGCGCTGCGAATCGG-3′) and OMF1937 or OMF600 (5′-TTGAACAAGATGGATTGCACG-3′) and OMF1936. PCR-amplified fragments were gel-purified using a Qiaquick gel purification kit.
For transformations, ∼107 PH-1 conidia were inoculated into 100 ml of YPD and allowed to germinate overnight at 28C with shaking at 200 rpm. Mycelia were harvested on cheesecloth and about 1 g (wet weight) was transferred into 20 ml of 1.4 M KCl with 500 mg driselase (Sigma, D8037), 100 mg lysing enzyme (Sigma, L1412), and 1 mg chitinase (Sigma, C6137) and shaken gently at 90 rpm at 28C for 2.5 hrs to induce protoplast formation. The suspension was filtered through Nitex membrane (30 µM) and protoplasts were collected and counted. Transformants were generated by mixing ∼107 protoplasts with 1 µg of neo+ split marker fragments in 500 µl of STC and 30% PEG8000 (4∶1) and incubating at RT for 20 min. An additional 1 ml of 30% PEG was added and the mixture was incubated for another 5 min, after which 2 ml of STC were added and the mixture was combined with 87 ml of recovery medium (RM) and split between six 100 mm Petri dishes for a total 15 ml RM per dish. After 24 hrs at RT, the RM was overlayed with 15 ml RM+200 µg/ml G418. Resistant colonies were picked and purified from single conidia by generating spores in liquid CMC medium. Strains were screened for gene replacements by PCR and Southern analyses.
We generated a strain containing a wildtype kmt6 allele for complementation analyses (FMF282) by random ectopic insertion into the kmt6 deletion strain FMF248. We digested pFOLT4R4 [93] with ClaI and isolated a 4 kb fragment that contained telomere repeats. This fragment was digested with PvuII for cloning into the SmaI site of pBSII SK+ [94], generating pLC14. The SalI hph fragment of pCT74 [95] was inserted pLC14 to generate pLC15. The kmt6 gene was PCR amplified with OMF1936 and OMF1937 and inserted into pCR4-TOPO (Invitrogen) to generate pLC40. The kmt6 coding region with ∼1 kb 5′ and 3′ flanks was released from pLC40 with SpeI and inserted into the SpeI site of pLC15 to generate pLC41. This plasmid was transformed into the Δkmt6 strain (FMF248) as described above and Hyg+ transformants were screened for integration by Southern analyses.
Approximately 5×106 protoplasts of FMF 225 (heterokaryotic kmt1+/kmt1) and FMF 248 (kmt6::neo+) were mixed and plated at a density of 2.5×106 protoplasts per dish on RM with 100 µg/ml Hyg, and 100 µg/ml G418. After one week, a plug from a selected colony was transferred onto YPD agar with 200 µg/ml Hyg and 200 µg/ml G418. Plugs from this heterokaryon were transferred to carrot agar for selfings. Selfings and crosses were performed as described previously [96], with minor modifications.
To assay colonization of tomato fruits, ripe organically grown “Roma” tomatoes (Denison Farms, Corvallis, OR) were surface-sterilized by gently wiping fruit with 95% ethanol, as described previously [32]. A small region of the epidermis (∼10 mm2) was peeled back and the wound was infiltrated with 10 µl of spore suspensions containing ∼1,000 conidia (1×105 conidia/ml). Fruits were incubated at 28C above water reservoirs, increasing humidity.
Genomic DNA was isolated according to a previously published method [97], digested with HindIII, and blotted as described elsewhere [98].
Tissue for histone extractions was generated by inoculating ∼107 macroconidia into 100 ml YPD and shaking at 200 rpm at 28C for 2 days. Mycelia were harvested by filtration, frozen in liquid nitrogen, and ground to a fine powder with a mortar and pestle. Histones were acid-extracted as previously described [99]. Approximately 10 to 20 µg of total protein per lane were analyzed by SDS-PAGE. Proteins were transferred to PVDF membrane and blotted using standard procedures [100]. Primary antibodies for westerns were Millipore 07-030 for H3K4Me2, Active Motif 39159 for H3K4me3, abcam ab8898 and Active Motif 39161 for H3K9Me3, and abcam ab9050 for H3K36Me3. We used four different antibodies to detect H3K27me3, Active Motif 39535, abcam ab6002 and ab6147, and Active Motif 39155 (which resulted in high background). Secondary antibodies were HRP-conjugated goat anti-rabbit (Pierce 31460) or HRP-conjugated goat anti-mouse (Invitrogen 62-6520).
ChIP was carried out on mycelia generated by growing conidia in 100 ml DVK medium for 3 days, transferring 5 ml of the suspension to 100 ml ICI medium supplemented with 6 mM or 60 mM NH4NO3 for nitrogen limiting or sufficient conditions, respectively, and shaken at 200 rpm for 48 to 72 hrs in the dark at 28C. ChIP methods were essentially as described previously [45],[101]. Strains and antibodies used for ChIP are listed in Table S1. DNA obtained by ChIP was end-repaired and ligated to adapters as described elsewhere [102]; adapter barcodes are listed in Table S1). Fragments (300 to 500 bp long) were gel-purified and amplified by 21–24 cycles of PCR with Phusion polymerase (Finnzymes Oy, NEB) and Illumina PCR primers [102]. Libraries were sequenced on an Illumina GAII and processed with RTA1.8 and CASAVA1.7 or on a HiSeq2000 genome analyzer and processed with CASAVA1.8. Wild-collected strain 00-676 [28] was re-sequenced to identify SNPs, which were called with MAQ [103].
Total RNA was isolated from aliquots of the same tissue that was used for ChIP by a previously described method [104], and mRNA was isolated using a Poly(A)Purist MAG kit (Ambion). We removed DNA by treatment with RNase-free DNAase (Qiagen), followed by column clean-up according to manufacturer's instructions. We used Illumina TruSeq RNA Sample Preparation kits to make RNA-seq libraries; cDNA was sequenced on an Illumina HiSeq2000 genome analyzer.
ChIP-seq reads were sorted by adapter and adapter sequences were removed, then quality scores were converted to Sanger format with the MAQ sol2sanger command [103] if needed (depending on Illumina pipeline output). HTS data from ChIP- and RNA-seq were submitted to the NCBI GEO database (accession number: GSE50689). Fastq files were used as input for BWA [105] and aligned to a reformatted assembly 3 of the F. graminearum genome (http://www.broadinstitute.org/annotation/genome/fusarium_group/MultiHome.html), i.e. supercontigs were assembled into chromosomes and separated by 20 kb of Ns as placeholders for unassembled reads to match the Broad Institute v3 chromosome assembly. Sam-formatted alignment files from BWA were converted to bam format, sorted, and indexed with samtools [106] for viewing in the gbrowse2 genome browser [107]. Data are accessible at http://ascobase.cgrb.oregonstate.edu/cgi-bin/gb2/gbrowse/fgraminearum_public/. Adapter-trimmed RNA-seq reads were mapped with Tophat [108] with options -a 5 -m 1 -i 30 -I 2000 and processed in the same way as BWA output with samtools. Cufflinks was used to quantify gene expression values as reads per kilobase of exon per million reads (RPKM), and cuffdiff was used to identify differentially expressed genes between samples [36],[109]. Figures showing global RNA expression and comparison between samples were generated in R with CummeRbund [36]. Heatmaps were generated in R with pheatmap (http://cran.r-project.org/web/packages/pheatmap/pheatmap.pdf). In total, 1,628 genes in fungidb.org were found to be associated with GO term “0044238, primary metabolic process”. Of these, 1,389 had significant data from cuffdiff and were used to generate heatmaps. Secondary metabolite genes were identified from the Broad Institute database. K-means clustering with 12 centers was done in R with the default Hartigan and Wong algorithm [110] to generate “clustered” heatmaps. We found 113 cytochrome P450 genes by searching for PFAM domain “PF00067: p450” at fungidb.org, and 77 of these genes had significant values from cuffdiff and were included in the heatmap. NRPS [111] and PKS [42] genes were previously described, but we use current FGSG numbers here [112]. A near complete secretome gene set has been described [43]. The list of transcription factors from a comparative study of three Fusarium species was used [27]. Transcription factors required for perithecial development were previously identified [113].
|
10.1371/journal.pntd.0004897 | Calculation of the Average Cost per Case of Dengue Fever in Mexico Using a Micro-Costing Approach | The increasing burden of dengue fever (DF) in the Americas, and the current epidemic in previously unaffected countries, generate major costs for national healthcare systems. There is a need to quantify the average cost per DF case. In Mexico, few data are available on costs, despite DF being endemic in some areas. Extrapolations from studies in other countries may prove unreliable and are complicated by the two main Mexican healthcare systems (the Secretariat of Health [SS] and the Mexican Social Security Institute [IMSS]). The present study aimed to generate specific average DF cost-per-case data for Mexico using a micro-costing approach.
Expected medical costs associated with an ideal management protocol for DF (denoted ´ideal costs´) were compared with the medical costs of current treatment practice (denoted ´real costs´) in 2012. Real cost data were derived from chart review of DF cases and interviews with patients and key personnel from 64 selected hospitals and ambulatory care units in 16 states for IMSS and SS. In both institutions, ideal and real costs were estimated using the program, actions, activities, tasks, inputs (PAATI) approach, a micro-costing technique developed by us.
Clinical pathways were obtained for 1,168 patients following review of 1,293 charts. Ideal and real costs for SS patients were US$165.72 and US$32.60, respectively, in the outpatient setting, and US$587.77 and US$490.93, respectively, in the hospital setting. For IMSS patients, ideal and real costs were US$337.50 and US$92.03, respectively, in the outpatient setting, and US$2,042.54 and US$1,644.69 in the hospital setting.
The markedly higher ideal versus real costs may indicate deficiencies in the actual care of patients with DF. It may be necessary to derive better estimates with micro-costing techniques and compare the ideal protocol with current practice when calculating these costs, as patients do not always receive optimal care.
| Dengue fever (DF) is caused by infection with the dengue virus, which is spread by the Aedes aegypti mosquito. Although the effects of DF are usually mild, in some cases serious illness and even death may result. The average costs per case when extrapolated to society may therefore be high, particularly given the large number of people infected during an endemic year. In Mexico, relatively little is known about the average cost per case (from either the healthcare system or the patient perspective). Such information is important to guide decisions about health policy, e.g. vaccination or public education. We aimed to quantify the average cost per case of DF using a micro-costing approach, both for DF treatment according to an ideal protocol for the management of the patient (´ideal costs´) and according to current treatment practice in the health services (´real costs´). Our results were largely consistent with findings from other international studies, but showed higher ideal costs compared with real costs. We think this may point to inadequate use of laboratory tests and treatments for patients with DF in Mexico. Our cost data will be used in a subsequent publication regarding the economic impact of DF in Mexico.
| Dengue fever (DF) is caused by infection with the dengue virus, a single-stranded positive-sense RNA virus of the Flaviviridae family [1]. The virus is transmitted almost exclusively by the mosquito vector Aedes aegypti, and humans are the only known reservoir for the virus [2, 3]. Clinical presentation varies, with signs and symptoms ranging from uncomplicated fever in the case of simple DF, to bleeding and low platelet counts in the case of dengue hemorrhagic fever (DHF). According to an estimate recently published by the World Health Organization (WHO), between 50 and 100 million infections occur every year [4]. Based on a cartographic approach, Bhatt et al. [2] estimated the number of annual worldwide dengue infections to be 390 million (95% credible interval 284–528 million). The WHO estimated that 500,000 people require hospitalization each year and about 125,000 of those affected die [4].
As in most of Latin America, the disease is widespread in Mexico, although the incidence rate has varied since its reappearance in the 1970s. Peaks in the number of cases occurred in 1980, 1997, and 2009, when more than 130,000 cases were reported nationwide [5]. Between 1995 and 2011, a cumulative total of almost 600,000 cases were reported, with 11% corresponding to DHF. According to the Sub-Directorate General for Epidemiology in Mexico, 62,330, 32,021, and 26,665 cases were reported for the years 2013, 2014, and 2015, respectively [6]. Given these high incidences, the potential seriousness of infections and the considerable disease burden, it is important to have accurate estimates of the average costs associated with the disease. Such estimates will enable efficient allocation of finite healthcare resources [7], for example for vaccination [8, 9], vector control [10–12], and integrated control of dengue through vaccination and vector control [13].
When the present study was initiated, some studies of the economic consequences of dengue had been conducted in Latin American countries including Brazil [14], Colombia [15], and Panama [16]. However, substantial variations in the underlying healthcare context exist in these countries and different methodologies were used in each analysis, such that we cannot extrapolate from these data to other settings. It is thus necessary to standardize and improve the reliability of the methodology and available estimates of the costs of dengue [17, 18].
In Mexico, available data on the costs of the disease are limited. In one of the most comprehensive international studies of the cost of DF, Shepard et al. [19] used specific information from index countries (Venezuela, El Salvador, Guatemala, Panama, Brazil, and Puerto Rico) and estimated the cost of a DF case in countries of the Americas. The cost of an ambulatory case of DF in Mexico was estimated to be US$486 (of which US$264 corresponded to direct medical costs), while the estimated cost of a hospitalized case was US$1,209 (of which US$502 corresponded to direct medical costs). Clearly, the assumptions used when making such an estimate might not hold for Mexico, given the segmentation of the healthcare system and the different organization of each sector within the healthcare system. Secondly, in Mexico there are no cost centers in any of the local healthcare units or in most hospitals. It is therefore not possible to simply aggregate the costs incurred from these types of sources [20, 21]. Thirdly, the two largest public healthcare systems in Mexico are very different in terms of the package of benefits provided, organization of the medical units, as well as the level of resources available and quality of the services provided. The Mexican Social Security Institute (Instituto Mexicano del Seguro Social [IMSS]) provides coverage to all industrial workers and their families. The Secretariat of Health (Secretaría de Salud [SS]) provides coverage to uninsured individuals who do not qualify for IMSS coverage or cannot pay private insurance, and tend to be in a lower income bracket [22, 23]. The proportion of the Mexican population covered by each system is roughly equal (30.4% and 36.6%, respectively) [24].
The co-existence of two different main public healthcare delivery systems (IMSS and SS) is not the only factor that may hamper extrapolation of cost data. In Mexico, the effectiveness of the healthcare sectors has been questioned after multiple reforms have been enacted [25], with patients not always receiving optimal treatment. In particular, the SS healthcare sector, which provides coverage through ‘Popular Insurance’, has suffered from suboptimal funding and fragmented organization (in reality there are 32 different healthcare systems–one for each state in the country) [26]. A recently published study of the economic and disease burden of dengue in Mexico estimated an average cost for DF. The authors used primary data from four major hospitals in the states of Quintana Roo, Morelos, and Tabasco in Mexico [27]. Estimates of DF costs for the two healthcare systems using a larger number of medical units and data from states with endemic DF may be more accurate and will allow comparisons between the two providers.
The main objective of this study was therefore to assess the associated medical costs and cost to the individual with DF using a micro-costing approach to overcome the lack of cost center data. Chart review and interviews with patients and key personnel from 64 medical units in 16 states with endemic DF provided data on costs associated with actual treatment of the disease (‘real costs’). For further comparison we estimated the costs that would be incurred if an ideal treatment protocol were followed (‘ideal costs’).
The study was approved by the National Commission of Scientific Research for the IMSS, register number: R-2012 785–070. All participants provided written informed consent and all patient data were anonymized.
The cost per case of DF was calculated by summing the following costs: (1) direct medical costs incurred by healthcare units (including professional services, 127 medical inputs, medical drugs and related products, as well as laboratory tests); (2) costs of dengue from the patient’s perspective (direct medical costs not covered by the public healthcare services and direct nonmedical costs, eg travel expenses); and (3) indirect costs (to the patient and their family) resulting from loss of productivity and loss of earnings. The second and third components include medical and nonmedical direct costs and also indirect costs, such as loss of productivity, caregiver’s costs, etc. All costs were calculated in Mexican pesos and converted to US$ using the exchange rate on September 12, 2012 (1 US$ = 13.03 Mexican pesos) and adjusted for inflation to December 2014 when the analysis was finished.
A hierarchical micro-costing approach was used to calculate direct costs incurred by the healthcare services (known as a program, actions, activities, tasks, inputs [PAATI] analysis) [26, 27]. This type of analysis has previously been used for economic assessments of malaria [28] and hemodialysis in Mexico [26]. The method essentially involves identifying the tasks and inputs from an ideal protocol or current treatment practice for DF and then assigning a unit cost to each. The unit costs were obtained from official IMSS and SS databases. Treatment costs are different for each institution because they have different unit costs for the inputs (see Table 1; full details of the individual unit costs included in the calculation are available in the S1 Appendix).
The tasks and inputs included in the analysis were defined in two ways. In one scenario, an ‘ideal’ cost was calculated according to the tasks and inputs included in an ideal management protocol for patients with DF. Fig 1 outlines the protocol, which was validated in an expert consensus meeting held on June 5, 2012. The protocol was prepared using a systematic review of the literature, as well as guidelines available for Mexico and Latin America. These materials were also used in an expert group discussion of the issue sponsored by the Ministry of Health, in which the authors of this paper actively participated (see Betancourt-Cravioto et al. [29] for further details). Essentially, the clinical protocol is divided into three sections (diagnosis and case identification, classification and notification, and treatment), and each section is assigned a series of activities, which in turn are assigned tasks and inputs.
In the second scenario, treatment costs associated with current practice (which we denote ‘real costs’) in the healthcare services were determined by calculating costs for the tasks and inputs actually performed. These tasks and inputs were determined by chart review and patient interviews. Where data from the medical units were missing or incomplete, the information was supplemented as far as possible from interviews with treating physicians and hospital administrators. The PAATI approach is summarized in Fig 2. The cost of each activity was calculated using the average use reported per input (ie resource or cost type). We looked at each separate type of cost incurred, which allows control of the variability for each component of the healthcare process, rather than using other methodologies where the average overall cost per patient is calculated and then assigned to each activity. For example, using the patient’s clinical chart, the number of blood biometry procedures the patient had undergone was assessed and the average per patient was calculated, which in this case was 1.6 times ± 1.1 per patient. This average was then adjusted by the number of patients in each care setting (outpatient, hospitalized, or intensive care unit [ICU]) who used that resource. The resulting adjusted average resource use was finally multiplied by the unit resource cost. In this way the average cost per case was adjusted for variability by taking into account the proportion of patients who reported consumption of inputs within the sample of patients with DF, in both the patient files reviewed and the patient interviews.
Given the different structures and attributes of the two principal sectors of the healthcare system, costs were calculated for both of them. In addition, unit costs in the private sector were used to estimate the cost of the ideal protocol only, as we had no way of evaluating current treatment practice in the private sector, given the difficulties of selecting a sample from the myriad of private healthcare units in the 16 states with endemic DF. Therefore, in the present study, we only assessed operational costs or variable costs, and no estimates were made of sunk costs or other costs, such as overhead costs (eg utilities, administration, etc.).
The indirect costs from the patient´s perspective were calculated based on information derived from interviews with patients. For these data, the methodology used took into account costs reported by the patients, and these were used to calculate the average cost per case, as well as confidence intervals. A bootstrap analysis was performed to assess variability. Where possible, the patients interviewed were the same patients whose charts were reviewed. Indirect cost was defined as loss of earnings associated with the disease (patient and/or caregiver), with information taken either from direct questions or an estimate based on the number of days of work lost multiplied by the net loss of earnings per day, separately for each institution. For example, in patients who reported being employed, the average length of hospital stay in days was multiplied by the average daily salary. The same approach was used for caregivers. A cost was not assigned if the patient or caregiver did not report any income or they were students. For patients under 16 years of age, only loss of earnings for the caregiver was calculated.
We selected a non-probabilistic sample. The study sample was drawn from 32 hospitals and 32 ambulatory care units in 16 Mexican states with endemic DF. Within each state, the hospitals and ambulatory care units with the highest number of cases were selected and up to a maximum of 40 cases per hospital or unit were selected at random for chart review. To avoid selection bias reviewers were instructed to use random numbers to select records. For units with fewer than 40 cases, a census was carried out, and all cases were included in the review. Only patients with an infection occurring in 2012 were considered for chart review.
A sample size sufficient to provide reasonably robust data was calculated using the formula:
n=(Nσ2Z2)/([N−1]e2+σ2Z2)
Where n = sample size, N = total population, σ = standard deviation, e = acceptable sampling error limit (0.05), and Z = 1.96 (for a 95% confidence interval). With these parameters, a sample size of 1,440 subjects was obtained.
As discussed above, costs were calculated for both the SS and IMSS systems, and so the sample size calculation was applied to both systems separately, resulting in double the overall number of patients. This sample would be representative of overall proportions in the SS and IMSS systems at a national level, although it does not take into account state-level stratification.
For indirect costs to the patient, a bootstrap sensitivity analysis was conducted to assess the robustness of the estimation of the average input (based on the method proposed by Efron [30]). Bootstrap analyses are used to estimate the distribution, bias, or variance in a statistical sample or analysis, and to estimate confidence intervals or test hypotheses on parameters of interest when the true distribution of those parameters is unknown. For each type of cost the original sample was sampled to get 5,000 bootstrap resamples to provide estimates and 95% confidence intervals for comparison with the original estimates.
Data were collected from review of 1,293 charts (90% of the target 1,440 charts according to the sample size calculation). From these charts, it was possible to obtain clinical pathways for 1,168 (81% of the target). Some clinical pathways could not be generated because data were incomplete and could not be supplemented by interviews with treating physicians and hospital administrative personnel. In total, 1,168 patients were interviewed for assessment of out-of-pocket expenses and indirect costs. The same patients were interviewed as those with chart review in 80% of the cases. The remaining patients interviewed had no corresponding chart review.
Overall, 53% of the patients with chart review were women. The mean age of the patient population was 27 years and 34.7% were younger than 18 years old.
As shown in Table 2, the medical cost differed according to setting (based on outpatients, hospitalized patients, and patients in the ICU), regardless of whether ideal or real costs were considered or which healthcare system (SS or IMSS) was used. Of particular note were the marked differences between the ideal and real costs apparent in both systems. The main factor influencing these differences was the cost of professional services (which accounts for approximately 90% of the differences in the case of outpatients and almost 100% in the case of hospitalized patients and those in the ICU). Professional services also accounted for the largest proportion of the medical cost, while the contribution of expenditure on medicines to the overall medical cost was limited. Nevertheless, of the cost types considered, the cost of medicines showed the largest difference (in terms of percentages rather than absolute costs) between the ideal scenario and the real current treatment practice.
In both the ideal and real scenarios, the costs to the IMSS were greater (by a factor of 2–4) than the costs to the SS across all patient settings. As for differences between ideal and real costs, the main driver was the cost of professional services. Expenditure on medical consumables and drugs and related products showed very little variation between the two systems, while expenditure on laboratory tests was somewhat higher for the IMSS.
The overall costs of the ideal management of patients in the private sector was higher than the corresponding ideal costs in the IMSS for ambulatory patients (US$487.39 vs US$337.50, respectively) and hospitalized patients (US$4,077.81 vs US$2,042.54), while costs for patients in the ICU were similar (US$23,753.19 vs US$23,452.63).
Expenses from the patient’s perspective are presented in Table 3. As can be seen, here we report direct medical costs incurred by the patient as well as direct costs reported by the healthcare units. These costs increased when the patient received care in the hospital setting. The difference was not so marked between hospitalized patients and those who received care in the ICU, with the exception of patients in the SS system.
In the outpatient setting, the direct medical care costs are almost double for IMSS patients than for SS patients. It is important to remember that these costs are independent from what was spent by the healthcare unit in the treatment of the patient.
We estimated the loss of productivity based on the number of days reported in the patient interviews for general hospitalization or ICU care. The average cost for a hospitalized patient is therefore not the same as for an ICU patient because, among other factors, the average length of stay is longer for a patient who receives ICU care. Data from the national census of population and national health surveys indicate that, on average, IMSS patients are more affluent than SS patients [22, 23]. The study did not collect data on the average income of patients in the two healthcare systems, although it is important to bear in mind possible differences when interpreting the data.
The questionnaire did not address loss of income for outpatients and these data are indicated as not available in Table 3. We consider this a limitation of the study.
In general, indirect costs appeared to be higher for patients in the IMSS system than for those in the SS system, with the exception of costs for patients in the ICU, which tended to be higher for SS patients (Table 3). The indirect costs for patients with DF corresponding to loss of earnings (not related to medical costs for the patient and/or their caregivers) for patients admitted to the ICU were actually lower compared with hospitalized patients (Table 4).
Despite the fact that a non-probabilistic sampling method was employed in this analysis, the extensive chart review and direct interviewing techniques provided a robust estimation of the average cost of treatment of DF in Mexico. The study also provided information according to type of healthcare system used, thus enabling qualitative comparisons.
In this study we tried to address the limitations of other estimations of average cost per case of DF. The available cost estimates in the literature showed great variability in their methodology, using primary data, macro-costing data, patient questionnaires, administrative data, or most commonly, a combination of data sources including some primary data in a highly restricted population. In our study, the use of micro-costing provides a more detailed understanding of the direct medical costs of dengue, and that is one of its major strengths.
Although we used average costs, we did not want to make the assumption that patients from a single medical unit reflect the whole experience of a country. Therefore, in our study the average costs per case were calculated from data collected from medical audits of 64 healthcare units (1,293 chart reviews) and 1,168 patient interviews in 16 states where DF is endemic.
To compare our results with other studies, here we present the original costs and within brackets the dollars adjusted for inflation using the national consumer price index by country and the official exchange rate, annual average for The World Bank for 2014. For ambulatory cases, the direct real medical cost of 2012 US$32.60 (2014 US$35.93) in the SS was lower than the direct medical costs reported by Sheppard et al. for Brazil 2010 US$49 (2014 US$54.05) [19]), and more recently Colombia 2012 US$67 (2014 US$65.18 [15]), whereas the IMSS direct real medical cost of 2012 US$92.0 (2014 US$101.38) was lower than reported in Venezuela 2010 US$118 (2014 US$130.15) [19]) and Panama 2005 US$332 (2014 US$501.65) [16]). With a micro-costing approach, a multicenter Brazilian study reported a confidence interval of US$31 to US$89 2013, (2014 US$33.81 to 2014 US$97.06) [14], similar to the SS and IMSS ambulatory costs in our study. After we had performed our data collection, Undurraga et al. [27] published an estimation of the ambulatory costs associated with DF in Mexico, where derived costs per episode were calculated by combining patient interviews in four major hospitals in the states of Quintana Roo, Morelos, and Tabasco, macro-costing data from two major public hospitals in Tabasco, MoH health and surveillance data, WHO-CHOICE estimates for Mexico, and previous literature on dengue burden. Indirect costs were obtained based on productivity losses by age, considering both the patient and the patient’s caregivers. These authors reported a cost of 2012 US$65.53 (2014 US$72.22) per outpatient visit and the average cost for ambulatory patients was 2012 $451 (2014 US$497.05) thus lying between the SS and the IMSS costs in our study.
The direct medical costs of hospitalized patients, in relative terms, are higher in our study at 2012 US$490 (2014 US$ 539.99) and 2012 US$1,644 (2014 US$1811.71) for the SS and IMSS, respectively, compared with the estimates for Brazil 2013 US$318 (2014 US$346.80) [19]), Colombia 2012 US$330.6 (2014 US$321.63) [15]) and in the new multicenter study for Brazil 2013 US$238–479 (2014 US$259.55–522.38) [14]). Costs reported for Venezuela 2010 US$864 (2014 US$952.99) [19]) and Panama 2005 US$1065 (2014 US$1609.22) [16]) are higher than the SS costs but lower than the cost to the IMSS.
Although comparisons with other countries may be illustrative, firm conclusions cannot be drawn given the differences in economic development, population size, and healthcare systems, as well as the methodology used for the estimates.
In the case of hospitalized patients, comparisons are more difficult because no distinction is made between hospitalized and ICU settings in most Latin American studies. Interestingly, the ideal costs estimated in our study, 2012 US$587 (2014 US$646.88) for SS and US$2,042 (2014 US$2,250.31) for IMSS were closer to the extrapolated costs reported by Shepard et al. [19] for total hospitalized cost, 2010 US$1,209 (2014 US$1,333.53) in Mexico. Undurraga et al. [27] reported that the hospital cost for Mexico was 2012 US$240.04 (2014 US$261.48) per bed day and "the average cost per non-fatal dengue episode was $1,327 for hospitalized patients (2014 US$1,445.53) (direct medical: $1,010 (2014 US$1,100.22); direct non-medical: $174 (2014 US$189.54); indirect: $143 (2014 US$155.77) and $451 for ambulatory patients (2014 US$491.29) (direct medical: $253 (2014 US$275.60); direct non-medical: $92 (2014 US$100.22); indirect: $106 (2014 US$115.47) ". Thus, these data are between the SS and the IMSS costs in our study.
Finally, we are not interested in presenting the virtues or deficiencies of the Undurraga approach; we simply want to clarify the differences with our approach.
Indeed, a particularly striking feature of our results are the differences between the direct medical costs of actual treatment in clinical practice and those that would be generated if an ideal protocol, validated by experts in the field, were followed. The difference was particularly marked for outpatients. The first implication is that cost studies based on an ideal treatment protocol may not reflect clinical reality, at least in Mexico. The difference in personnel costs may reflect systematic differences in either the productivity of the personnel, their training, and treatment standards among the medical units of the institutions included in the study. The second implication is that there may be shortcomings in the Mexican healthcare systems, particularly in the outpatient setting, despite extensive reform in recent years with a view to improving the quality of care [20, 21, 31]. The main driver of the difference between real and ideal costs is the cost of professional services and the use of laboratory tests (including confirmatory tests). This may indicate that the treating physicians are not dedicating sufficient time to their patients, nor providing optimal laboratory follow-up for patients. Furthermore, we note that most of the data were collected in 2012, which was not an epidemic year. In epidemic years, it is likely that overburdening of the system would further accentuate the differences between real and ideal costs, as physicians would be forced to dedicate less time to their patients, and the demand for laboratory tests would be higher.
Another noteworthy feature of the results presented here is the higher direct medical costs incurred within the IMSS system compared with the SS system. Another Mexican study utilizing chart reviews and direct patient interviews estimated the direct medical costs in the IMSS system for patients with osteoporosis and hip fracture [24]. Compared with the SS system, they also found considerably higher cost for IMSS (US$3,891.20 vs US$1,590.70, respectively). The main driver of this difference in the present study was personnel costs, which may reflect differences in pay between the two systems. The IMSS was set up over 60 years ago and has strong unions, which may be responsible for higher staff costs. In addition, there may be a tendency to use the private sector as a guide when setting prices, particularly in the ICU setting, where the IMSS and private sector prices are very similar. As noted by Clark et al. [24] in their study of hip fracture, the SS system, which provides medical care to the lowest income groups, receives larger subsidies that are not otherwise reflected in the costs generated by a micro-costing analysis such as the present one. It is important to note that the results in our study are closer to the multicenter study in Brazil [14] that used a micro-costing approach, even though we have clear differences between our healthcare systems.
The greater availability of treatments in the IMSS system may explain the slightly higher treatment costs in the outpatient and hospitalized settings, although the more fragmented nature of the SS system may increase procurement costs. There is evidence that the heterogeneity of the state healthcare services (SS system) may lead to small variations between areas, resulting in care below that recommended in clinical guidelines. This substandard care is associated with lower costs.
Of note were the expenses reported by the patient beyond those incurred by the healthcare institutions for hospitalized patients and patients in the ICU. Expenses incurred by patients were higher within the SS system, suggesting that these patients had to supplement the care provided by the SS to a greater extent than IMSS patients. Any comparison of the costs in the IMSS and SS systems should bear in mind that the cost of each activity was calculated using the average use reported per input and the unit cost reported for every organization and state. In general, the unit costs in the IMSS system are more stable and better registered at the central level of the organization than costs within the SS system.
A limitation of our study is that loss of earnings due to days off work was only registered for patients admitted to hospital or ICU and thus was not captured for outpatients.
Indirect costs were greater for patients who attended through the IMSS, possibly reflecting their higher socioeconomic status. Interestingly, the difference was greater for hospitalized patients than those admitted to the ICU. This observation can be explained by the fact that family members may be allowed to stay with patients in a hospital ward but not with patients in an ICU setting and they will thus incur more expenses.
A second limitation was that we do not include data on management, electricity, and other utilities in the cost estimation. However, it was considered that the relevant cost is variable cost, and given the lack of data on management, electricity and other utilities, we considered that it would be misleading and add more uncertainty to the cost estimation.
As noted above, a principal weakness of the study is possible bias resulting from the non-probabilistic sampling method employed in this study. The states where DF is endemic were chosen, and, within those states, the centers with the highest number of cases were selected. Furthermore, patients requiring hospitalization and those with DHF would be more likely to attend larger reference centers, so this approach may lead to an over-representation of patients with more severe disease. A further bias may have been introduced by not ensuring that all patients interviewed corresponded to those whose charts were reviewed. However, given that 80% of the patients interviewed also had a chart review, differences would have limited impact on the final average results.
Nonetheless, by concentrating on larger centers with a larger number of cases, it was possible to collect extensive primary data that would otherwise be hard to acquire from probabilistic samples. In essence, statistical robustness was sacrificed to enable the data collected to reflect more accurately clinical practice in the sample, in contrast to many studies of costs of DF in the Latin American region. Additionally, the data available from chart review were not complete (and may not have been entirely reliable) and had to be complemented with other data sources. While this may also compromise the integrity of the data, we believe that chart review, despite its inherent limitations, ultimately provides the best approximation to what actually happens in clinical practice. It is also possible that recall bias was present in the interviews, but for most patients, DF is an event that they are unlikely to forget.
In conclusion, this study pointed to real costs associated with DF in line with those reported in other Latin American countries (with the caveat that direct comparisons should be treated with caution given the differences between countries in economic development, healthcare systems, size, and the cost methodology employed in studies). Of particular note were the large differences in the real costs derived from patient records and the ideal cost calculated from an ideal treatment protocol. This difference perhaps points to deficiencies in the care of patients with DF in Mexico. It also indicates the importance of collecting primary data when calculating DF costs to guide health policy decisions in this and other diseases that lack an appropriate estimate of its cost to the system, which poses a major hurdle for healthcare planning.
|
10.1371/journal.pntd.0006511 | Humoral and cellular immune responses to Yersinia pestis Pla antigen in humans immunized with live plague vaccine | To establish correlates of human immunity to the live plague vaccine (LPV), we analyzed parameters of cellular and antibody response to the plasminogen activator Pla of Y. pestis. This outer membrane protease is an essential virulence factor that is steadily expressed by Y. pestis.
PBMCs and sera were obtained from a cohort of naïve (n = 17) and LPV-vaccinated (n = 34) donors. Anti-Pla antibodies of different classes and IgG subclasses were determined by ELISA and immunoblotting. The analysis of antibody response was complicated with a strong reactivity of Pla with normal human sera. The linear Pla B-cell epitopes were mapped using a library of 15-mer overlapping peptides. Twelve peptides that reacted specifically with sera of vaccinated donors were found together with a major cross-reacting peptide IPNISPDSFTVAAST located at the N-terminus. PBMCs were stimulated with recombinant Pla followed by proliferative analysis and cytokine profiling. The T-cell recall response was pronounced in vaccinees less than a year post-immunization, and became Th17-polarized over time after many rounds of vaccination.
The Pla protein can serve as a biomarker of successful vaccination with LPV. The diagnostic use of Pla will require elimination of cross-reactive parts of the antigen.
| Yersinia pestis, the causative agent of plague, has been recognized as one of the most devastating pathogen experienced by mankind. It remains endemic in many parts of the world, and is considered emerging pathogen. A live attenuated Y. pestis strain EV line NIIEG has been used for decades in the former Soviet Union for human vaccination and has proven effective against all forms of plague. We began characterizing the Y. pestis-specific antibody and T cell-mediated immune responses in people immunized with live plague vaccine. The long term goal of our research is to understand the protective mechanisms underlying immunity to plague in humans and to discover novel protective antigens for their incorporation into a subunit vaccine. Here, we describe our study on immune responses in vaccinees to one of the essential virulence factors of Y. pestis, namely Pla antigen. The results of the study shed light on the development of the optimal markers to assess the correlation with vaccine-induced protection.
| Plague is known as a primary natural zoonosis but is an extremely deadly infection for humans. The disease is caused by Yersinia pestis, a gram-negative bacterium, which upon entry in the body of mammalian host is capable of establishing three major forms of plague: bubonic, septicemic, and pneumonic [1, 2]. The plasminogen activator (Pla) of Y. pestis is an outer membrane protease involved in dissemination of Y. pestis into circulation, and is one of the major virulence determinants of this pathogen [3–5]. The Pla protein is the surface-exposed trans-membrane β-barrel protease of the Omptin family with homologs found among many bacteria across family Enterobacteriacea [6]. Nevertheless, only Pla can convert plasminogen to plasmin by limited proteolysis, and this activity was likely crucial for the increased lethality of Y. pestis that developed during the course of evolution [7–9].
Detectable levels of relevant antibodies to Pla (anti-Pla Abs) have been measured in the convalescent sera of human patients who survived plague infection, as well as in mice that survived experimental plague infection [10, 11]. Moreover, anti-Pla Abs of IgG class were detected in the sera of animals and humans vaccinated with live plague vaccine (LPV) indicating immunogenicity of this outer membrane protein [12]. Immunization with purified recombinant Pla or its use in a DNA vaccine formulation provided no protection against plague in a murine model [13]. Nevertheless, partial protection was seen in mice and rats against strain of Y. pestis lacking capsular antigen F1 [14].
Besides the testing of Pla as a potential protective antigen for plague subunit vaccine formulation, there were attempts to use this outer membrane protein for immuno-diagnostic purposes. A panel of monoclonal antibodies (MAbs) to Pla was created to different epitopes that were either species-specific for Y. pestis or able to recognize other bacteria [15]. Similar studies resulted in selection of anti-Pla MAbs capable of detecting natural Y. pestis isolates, as well as modified strains of plague microbe like capsule-negative variants [16, 17].
The live plague vaccine created almost a century ago is still widely used in the former Soviet Union and China to immunize plague researchers and people at risk living in plague endemic territories [12, 18]. The advantage of the LPV over a killed plague vaccine is its ability to defend against all forms of plague, as well its ability to mimic to the plague infectious process to a certain extent, resulting in a robust protection [19]. However, this vaccine is not approved for human use in the Western countries due to the safety concerns [20]. Nevertheless, construction of rationally attenuated vaccine strains of Y. pestis has garnered attention in recent years [21], especially because the LPVs can induce both humoral and cellular immunity against plague [22–24]. Therefore, a detailed study of human immunity elicited by LPV is beneficial for both understanding the mechanism underlying the immune response to this vaccine and for future evaluation of efficacy of the next generation of plague vaccines.
In this study, we investigated antibody and cell-mediated immunity in individuals vaccinated with the live plague vaccine line EV NIIEG, which is a derivative of the well-known vaccine strain Y. pestis EV76 [12]. Here, the Pla protein was used as a model antigen, which we intended to utilize in the future as a tool for evaluation of vaccine efficacy of vaccination and as a marker of exposure to plague.
Each human volunteer provided written informed consent for blood donation. The patients in this manuscript have given written informed consent (as outlined in the PLOS consent form) to publication of their case details. This study was approved by the Human Bioethics Committee of the Saratov Scientific and Research Veterinary Institute. The Institutional Review Board (IRB) was registered with the Office for Human Research Protections (OHRP), registration number IRB00008288 (https://ohrp.cit.nih.gov/search/irbsearch.aspx?styp=bsc).
Sera from healthy 26–72 years old volunteers (n = 34, group A) of both genders who received multiple annual immunizations (2–51 injections) with the live plague vaccine line EV NIIEG (LPV), as well as from healthy individuals (n = 17, group B) who had no history of contact with either Y. pestis microbe or its antigens, were tested. We further divided group A of immunized donors into subgroups of recently vaccinated (A-RV, less than one year post-vaccination, n = 13) and early vaccinated (A-EV, more than one year post-vaccination, n = 21). The vaccination was performed by intradermal immunization (scarification), which is a standard way to immunize people with LPV in Russia [12]. This immunization was done to plague researchers in their respectful institutions, and was not performed by us. The sera were aliquoted and stored at -80°C.
Peripheral blood mononuclear cells (PBMCs) were isolated from heparinized blood by density gradient centrifugation in Histopaque (Sigma, St. Louis, MO) according to standard protocol. Cells were cultured in DMEM/F12 medium containing 10% FBS and antibiotic-antimycotic supplement for six days with or without stimulatory agent in 96 well plates (105 cells per well). The Hig-Tag-labeled Y. pestis recombinant proteins were purified as described previously for the panel of five antigens [25]. The quality of purification was evaluated with the silver stained PAGE. Soluble antigens, such as F1, were treated with AffiPrep Polymyxin resin (BioRad, Herciles, CA) to remove the traces of LPS, while partially soluble Pla was isolated in two steps. First, we isolated Pla-containing inclusion bodies, and then purified Pla using Ni2+-chromatography under denaturing conditions. The level of contaminating LPS was measured with QCL-1000 Chromogenic LAL Assay kit (Fisher Scientific). Both antigens were essentially LPS-free, as the LPS contamination was below the sensitivity level of the kit (0.1–1.0 EU/ml). Unstimulated PBMCs served as negative controls, and Concanavalin A from Canavalia ensiformis Type IV-S (ConA) (Sigma) was used as a positive control. The proliferative response was measured in quadruplicate by detection of BrdU incorporation using Cell Proliferation ELISA, BrdU chemiluminescent kit (Roche Applied Science, Indianapolis, IN) according to manufacturer’s protocol. The chemiluminescence was measured by using a BioTek Synergy HT reader (BioTek Instruments Inc., Winooski, VT). The proliferative response was expressed as a stimulation index (SI) calculated by dividing the mean relative light units per second (rlu/s) obtained for the cultured cells with a stimulant by the rlu/s of non-stimulated wells. Culture supernatants were collected on day 5 and preserved at -80°C until further use. The levels of IFN-γ, TNF-α, IL-4, IL-10, and IL-17A were measured by using commercial ELISA kits (Vector-Best, Cytokine, Russia) according to the manufacturer’s instructions. The reaction was developed using streptavidin-horseradish peroxidase with the tetra-methyl benzidine chromogen (TMB), and the optical density was measured at 450 nm.
Immulon 2 HB plates (Thermo Scientific, USA) were coated overnight at 4°C with recombinant Pla at concentration 5 μg/ml dissolved in 0.1 M carbonate buffer, pH 9.5 with 8 M urea. The remaining binding sites were blocked with 20% Newborn Calf Serum (Sigma) in Phosphate Buffered Saline (PBS). Each serum sample was two-fold serially diluted in the range of 1:50 to 1:800. Goat anti-human IgG (Fab-specific)-peroxidase (HRP) antibody (Sigma) was used as secondary antibody. The reaction was developed with the TMB substrate (Sigma). The bacterial suspension of LPV was used as a control coating antigen in ELISA. The titers were calculated as the last dilution giving values above the cut-off level that was the mean value of the blank wells (sera without antigen).
Human antibody isotyping was performed by immunoblotting technique using relevant commercial murine monoclonal subtyping IgG subclass antibodies (IgG1, IgG2, IgG3, and IgG4), as well as anti-human IgA, IgM, and IgE class specific antibodies (Rosmedbio Ltd., St.-Petersburg, Russia). The recombinant Pla antigen was separated by 12.5% SDS-PAGE, transferred to a nitrocellulose membrane, incubated with serially diluted human sera, and then probed with corresponding anti-human MAbs. Goat anti-mouse IgG (Fab-specific)-HRP Ab (Sigma) was used as secondary antibody. The substrate was TMB for the membranes (Sigma). The endpoints were determined visually with the signal considered positive when the intensity was twice over the background.
B-cell immune-reactive epitope mapping of the target antigen was performed in ELISA by using a library of 61 peptides generated from the sequence for Pla of Y. pestis CO92 (accession no. CAB53170.1) and consisting of 15-mer peptides overlapping by 10 amino acids (S1 Table). Nunc Immobilizer, Amino Modules Plates (Thermo Scientific) were coated with 20 μg of individual peptides in 0.1 M carbonate buffer, pH 9.5, overnight. ELISA was then performed as described above. The dilution of tested sera was 1:100. The interpretation of data was performed as described in a previous study with a similar design [26, 27]. Briefly, optical density (OD) values were read with a BioTek Synergy HT reader at a wavelength of 450 nm (reference wavelength, 630 nm). A signal was assigned as positive when it reached the cutoff value of twice the background OD. The background OD was the mean of the lowest 50% of all OD values obtained with that particular serum. The wells containing no peptides were used as negative controls, and recombinant Pla was used as a positive control.
GraphPad Prism 6 software was used for data handling, analysis, and graphic representation. Non-parametric tests, i.e. the Mann–Whitney test for continuous unpaired data and the Chi-square test or the Fisher’s exact test for dichotomous variables, were performed for statistical analysis. Associations were assessed using Spearman’s Rank Correlation coefficient. A P value <0.05 was considered statistically significant.
To assess the in vitro proliferative response, PBMCs isolated from study subjects were stimulated with 5 μg/ml Pla or 2 μg/ml F1 (control antigen). The SI induced by Pla was noticeably higher than that obtained in response to the control F1 antigen in both relevant vaccinated groups, such as group A-RV and group A-EV (p<0.05, p<0.0001, respectively), as well as in the group B of unvaccinated individuals (p<0.01) (Fig 1A). Although the proliferative response to Pla was pronounced, there was no significant difference between both A-RV and A-EV groups of vaccinees and control donors in the group B. Nevertheless, a moderate trend of slightly higher stimulation indexes was observed in the cohort of recently vaccinated individuals (A-RV group) compared with the donors in the A-EV group with the last vaccination occurring more than one year ago (p = 0.117).
The in vitro proliferative response to Pla was accompanied by a marked but nonspecific (p>0.05) release of a number of cytokines, such as IFN-γ, TNF-α, and IL-10 by PBMCs derived from donors of both vaccinated (A-RV and A-EV) and unvaccinated groups (Fig 1). Surprisingly, we found that production of IL-4 was significantly greater in group B than in vaccinated donors. Although there was no significant difference between stimulated and control PBMCs, the level of IL-4 was reduced by 3.4-fold in group A (ARV and A-EV) of vaccinated donors in comparison with group B of naïve donors (Fig 1B). In contrast, PBMCs obtained from donors of the A-EV group, who received multiple immunizations in the past, responded to stimulation with Pla by 14.7-fold increase (p<0.05) in making IL-17A over the naïve donors of the group B (Fig 1A, S1 Fig). This remarkable contribution of IL-17A production from the donors of the A-EV group resulted in the overall significant difference between groups A and B vaccinated and naïve donors (p = 0.043), while there was no statistical significance for the group A-RV in this category (p>0.05). The observed IL-17A release may indicate that the immune response to LPV becomes Th17-polarized over time multiple rounds of vaccination.
There was a significant modest negative correlation between number of immunizations (r = -0.475, p<0.05) and the IL-4 response in vaccinees, although corresponding correlation with post-vaccination time was negligible (r = -0.196, p>0.05) (S2 Fig and S3 Fig). Also, there was a slight positive correlation in the levels of IFN-γ (r = 0.018, p = 0.943), IL-17A (r = 0.018, p = 0.943) and TNF-α (r = 0.229, p = 0.361), and negative correlation of IL-10 (r = -0.297, p = 0.231), with the number of LPV injections. The increase in the level of IFN-γ (r = 0.079, p = 0.756), IL-17A (r = 0.147, p = 0.561), and IL-10 (r = 0.116, p = 0.646) but not TNF-α (r = -0.126, p = 0.620) may potentially correlate with the post-vaccination time (S3 Fig), although all latter cases were not statistically significant (p>0.05). Overall, we found significant association for the IL-4 cytokine, whose levels decreased in donors after an increasing number of vaccinations.
The serological immune response to Pla elicited by the LPV was investigated in the vaccinated donors of group A in comparison with the naïve donors of group B by ELISA. We detected IgG class Abs to Pla, with titers ranging from 1:50 to 1:400, in approximately half of the group A individuals. Moreover, all recently vaccinated donors in the A-RV subgroup were found to be anti-Pla positive. To our surprise, there was a significant difference in both the titers and percent of positive individuals between the subgroups A-RV and A-EV. Donors that received multiple repetitive immunizations (A-EV group) displayed a suppression of the antibody response to Pla. This observation correlates with the negative association between the level of IL-4 and number of LPV immunizations. On the other hand, 100% of the sera collected from donors in the naïve control group B reacted with the Pla antigen and exhibited titers similar to those found in the vaccinated donors of group A-RV, indicating Pla cross-reactivity (Fig 2).
We next determined the reactivity of the sera for anti-Pla antibody classes and IgG subclasses (Fig 3). Based on the ELISA results, we separated the group A donors into responders (A-Res) and non-responders (A-Non). All responders of group A and the majority of positive donors of group B demonstrated immunoreactivity with the IgG1 subclass of immunoglobulins. Only a single individual in each A and B groups showed the reaction with IgG2 subclass (Fig 3A). This one donor from the group A-Res was positive for both IgG1 and IgG2 types. Among donors of the A-Non subgroup the negative reaction was observed for anti-Pla Abs of IgG1 (p<0.05), IgG2 and IgG4 subclasses (p>0.05). Moreover, all vaccinees (group A) possessed anti-Pla Abs for IgG3 while naïve donors of the group B did not have Pla-specific Abs of this subclass (p<0.01). In contrast, anti-Pla Abs of the IgG4 subclass was found exclusively in the sera of the group B donors.
In addition to Pla-specific IgG, we also detected anti-Pla Abs of the IgA class in the sera of vaccinees but not in naïve donors. This corresponded with the increased level of IL-17A released by PBMCs from the group A donors (see Discussion). Also, we observed the presence of anti-Pla Abs of the IgM subclass in both A and B groups of donors (Fig 3B). Finally, IgE class antibodies to Pla antigen were found in sera of about one third of the vaccinated donors, both A-Res and A-Non, and only in a single unvaccinated individual. This result may be indicative of the putative allergenic potential of this antigen and the LPV vaccine in general.
A library of 61 overlapping peptides, each 15 amino acid residues in length (offset by 5 residues at a time) deduced from the entire Pla sequence was probed with 12 sera samples that exhibited the highest anti-Pla IgG titers determined by ELISA. This set included sera from eight and four donors of the vaccinated A and naïve control B groups, respectively. The results of the screening for each serum are shown in the S2 Table. We found that most of the reactive peptides interacted with Abs from both groups of donors. Nevertheless, the peptides 9, 11, 18, 30, 34, 36, 49, 52, 54, 56, 58, and 60 were specific for the donors of the A group, while peptides 19, 33, 35, 50, and 61 belonged exclusively to the group B donors. The frequency of appearance of each peptide for donor groups A and B is illustrated on Fig 4. Among the group A specific peptides, none reacted with Abs of all eight donors tested indicating the absence of the wide-range immunodominant linear epitope. Only peptide 52 reacted with 50% sera of vaccinated donors, while most of peptides reacted with one or two sera. Therefore, the formation of the antibody response to the Pla antigen with respect to the linear B epitopes might be donor-specific. The B group specific peptides appeared just once in each corresponding serum.
Two peptides, number 6 and 24, were of particular interest, because of their strong cross-reactivity with sera from both vaccinated and naïve donors. Peptide 6 showed the remarkable ability to interact with Abs from any donor of both groups, while peptide 24 reacted 100% with sera from naïve and 50% with sera from vaccinated donors (Fig 4). The existence of two broadly-reactive Pla epitopes may explain the ELISA results shown on Fig 2 in which sera of all naïve donors reacted with the entire Pla antigen immobilized in the wells of the microtiter plate.
In the current study, we investigated for the first time the T-cell recall response to the Pla antigen in human donors vaccinated with LPV. Our data indicate that the proliferative response of human PBMCs to Pla stimulus was strong in nature and even exceeded that induced by capsular F1 antigen, which is known for its pronounced immunogenic characteristics [18, 19]. However, we did not detect a statistical difference in this respect between vaccinated and naïve control donors, suggesting a nonspecific reaction likely due to the presence of cross-reacting T-cell epitope(s) within the Pla antigen. Nevertheless, there was a moderate trend (p = 0.117) in observing a slightly high stimulation index in recently vaccinated individuals (less than one year post-immunization). Therefore, we speculate that the specific T-cell response to Pla did occur in this group of vaccinees; however, it was masked by the pronounced cross-reactivity. Also, we report the presence of Pla cross-reactive linear B-cell epitopes that resulted in a strong reaction with this antigen of sera from naïve donors in ELISA. This was not totally surprising to us, since we saw an indication of this antibody cross-reactivity in our previous studies after probing Pla antigen with the panel of monoclonal antibodies [15], and also observed the Pla-reactive band on immunoblot with naïve human sera [25]. Interestingly, multiple vaccinations with LPV suppressed the antibody titers to Pla that were observed when recently (A-RV) and early (A-EV) groups of vaccinated donors were compared (Fig 2). This suppression of the antibody response to the Pla antigen is likely due to the development of a dominant immune response to other competing and more potent antigen(s) of the live Y. pestis vaccine that became enhanced overtime. If true, this may mean that multiple booster immunizations with LPV may select for the response to a few dominant antigens. These antigens may not even be protective while presenting a threat of developing an allergic reaction instead (see IgE response in Fig 3B).
The Pla protein is considered to be a good candidate for Y. pestis specific diagnostic antigen [15–17] that is expressed well at both ambient and mammalian host temperatures [28]. However, the observed Pla cross-reactivity may result in certain limitations on its use for diagnostic purposes. Therefore, we mapped the cross-reactive regions of Pla using a library of 15-mer overlapping peptides. Comparison of peptide-ELISA results with sera from eight vaccinated and four naïve donors revealed two major cross-reactive peptides, peptides 6 (IPNISPDSFTVAAST) and 24 (TDHSSHPATNVNHAN). Peptide 6 showed a particularly strong reaction for all sera tested and far exceeded the signal from any other reactive peptide in the library. There were other reacting peptides common for vaccinated and naïve donors; however, they were random and reactive with only one or two sera per group. Importantly, we found 12 peptides that specifically reacted with sera of vaccinated individuals and did not react with sera from naïve donors. Among them, only peptide 52 (TPNAKVFAEFTYSKY) reacted with 50% of sera from vaccinees suggesting that this region can potentially contain an immunodominant linear B-cell epitope recognized by the immune system of humans with different genetic backgrounds. This region may represent a good candidate to test for the purpose of creating a novel plague peptide vaccine.
We determined the distribution of Pla-reacting immunoglobulins within the IgM, IgA, and IgE classes, as well as IgG subclasses (IgG1, IgG2, IgG3, and IgG4) in the sera of vaccinated and naïve donors (Fig 3). The anti-Pla Abs of the IgM class were found in all donors tested. Generally, natural human IgM antibodies or autoantibodies play a role in maintaining the physiological homeostasis and preventing a wide range of different infections [29]. The presence of anti-Pla Abs of IgM class in naïve donors and those who received LPV immunization many years ago suggests that they derived from a constant stimulation of the immune system with cross-reacting antigens rather than from the LPV vaccination. The suspected candidates for these stimulants could be Pla-homologous proteins of the Omptin group found in many Enterobacteriaceae [6]. In contrast, vaccinated, but not naïve, donors contained anti-Pla Abs of IgA class (p<0.05) suggesting their origination from LPV immunization by dermal scarification. The existence of Pla-specific IgA correlated with our observation of marked production of IL-17A found after stimulation of PBMCs of vaccinated donors with the Pla antigen (Fig 1A), which was absent in the naïve group of donors. It was shown previously that vaccine-specific Th17 cells formed by parenteral immunization were involved in eliciting a long-term detectible level of secreted IgA [30]. Moreover, subcutaneous priming with recombinant antigen in a Th17-inducing adjuvant followed by boosting promoted high and sustained levels of IgA in the lungs. This response was proven to be associated with germinal center formation in the lung-draining lymph nodes [31]. This may comprehensively explain the high efficiency of LPV against both bubonic and pneumonic plague [12, 18, 24]. Overall, these immune response characteristics to Pla antigen suggest that Th17 polarization of the immunity to LPV can be beneficial to the host during infection [32]. The release of IL-17A in response to stimulation of PBMCs of immunized individuals could also serve as an indicative marker of successful vaccination with LPV. Nevertheless, we would like to speculate that the presence of Pla-specific antibodies of the IgE subclass in vaccinated donors only (Fig 3B) may highlight the danger of a vaccine-related trigger of an allergic response and autoimmune disease. Further studies are needed to shed light on this important issue.
It was reported previously that human immunization with killed plague vaccine induced long-lasting and mixed Th1/Th2 responses that were more polarized towards Th1 [33]. In our study, slightly elevated production of IFN-γ and diminished IL-4 in response to stimulation with Pla in the group of recently vaccinated donors (Fig 1B) also points to a Th1-biased immune response after administration of the live vaccine. This observation is supported by detection of anti-Pla Abs of IgG1 and IgG3, and absence of IgG4 subclasses in the sera of these donors [34–37].
In summary, we found that despite complications with cross-reactivity, human immunity elicited by LPV could be assessed based on analysis of the immune response to Pla antigen. Our analysis showed that LPV vaccination resulted in the response being skewed towards Th1 and Th17, while production of IL-17A by PBMCs of immunized donors in response to Pla antigen stimulation could be a good indicator of the induced immunity. Additionally, we mapped cross-reacting linear B-epitope candidates within the Pla antigen that should be helpful in developing Pla-based diagnostics for Y. pestis.
|
10.1371/journal.pcbi.1003818 | The Time Scale of Evolutionary Innovation | A fundamental question in biology is the following: what is the time scale that is needed for evolutionary innovations? There are many results that characterize single steps in terms of the fixation time of new mutants arising in populations of certain size and structure. But here we ask a different question, which is concerned with the much longer time scale of evolutionary trajectories: how long does it take for a population exploring a fitness landscape to find target sequences that encode new biological functions? Our key variable is the length, of the genetic sequence that undergoes adaptation. In computer science there is a crucial distinction between problems that require algorithms which take polynomial or exponential time. The latter are considered to be intractable. Here we develop a theoretical approach that allows us to estimate the time of evolution as function of We show that adaptation on many fitness landscapes takes time that is exponential in even if there are broad selection gradients and many targets uniformly distributed in sequence space. These negative results lead us to search for specific mechanisms that allow evolution to work on polynomial time scales. We study a regeneration process and show that it enables evolution to work in polynomial time.
| Evolutionary adaptation can be described as a biased, stochastic walk of a population of sequences in a high dimensional sequence space. The population explores a fitness landscape. The mutation-selection process biases the population towards regions of higher fitness. In this paper we estimate the time scale that is needed for evolutionary innovation. Our key parameter is the length of the genetic sequence that needs to be adapted. We show that a variety of evolutionary processes take exponential time in sequence length. We propose a specific process, which we call ‘regeneration processes’, and show that it allows evolution to work on polynomial time scales. In this view, evolution can solve a problem efficiently if it has solved a similar problem already.
| Our planet came into existence 4.6 billion years ago. There is clear chemical evidence for life on earth 3.5 billion years ago [1], [2]. The evolutionary process generated procaria, eucaria and complex multi-cellular organisms. Throughout the history of life, evolution had to discover sequences of biological polymers that perform specific, complicated functions. The average length of bacterial genes is about 1000 nucleotides, that of human genes about 3000 nucleotides. The longest known bacterial gene contains more than nucleotides, the longest human gene more than . A basic question is what is the time scale required by evolution to discover the sequences that perform desired functions. While many results exist for the fixation time of individual mutants [3]–[15], here we ask how the time scale of evolution depends on the length of the sequence that needs to be adapted. We consider the crucial distinction of polynomial versus exponential time [16]–[18]. A time scale that grows exponentially in is infeasible for long sequences.
Evolutionary dynamics operates in sequence space, which can be imagined as a discrete multi-dimensional lattice that arises when all sequences of a given length are arranged such that nearest neighbors differ by one point mutation [19]. For constant selection, each point in sequence space is associated with a non-negative fitness value (reproductive rate). The resulting fitness landscape is a high dimensional mountain range. Populations explore fitness landscapes searching for elevated regions, ridges, and peaks [20]–[27].
A question that has been extensively studied is how long does it take for existing biological functions to improve under natural selection. This problem leads to the study of adaptive walks on fitness landscapes [15], [20], [21], [28], [29]. In this paper we ask a different question: how long does it take for evolution to discover a new function? More specifically, our aim is to estimate the expected discovery time of new biological functions: how long does it take for a population of reproducing organisms to discover a biological function that is not present at the beginning of the search. We will discuss two approximations for rugged fitness landscapes. We also discuss the significance of clustered peaks.
We consider an alphabet of size four, as is the case for DNA and RNA, and a nucleotide sequence of length . We consider a population of size , which reproduces asexually. The mutation rate, , is small: individual mutations are introduced and evaluated by natural selection and random drift one at a time. The probability that the evolutionary process moves from a sequence to a sequence , which is at Hamming distance one from , is given by , where is the fixation probability of sequence in a population consisting of sequence . In the special case of a flat fitness landscape, we have , and . Thus we have an evolutionary random walk, where each step is a jump to a neighboring sequence of Hamming distance one.
Consider a high-dimensional sequence space. A particular biological function can be instantiated by some of the sequences. Each sequence has a fitness value , which measures the ability of the sequence to encode the desired function. Biological fitness landscapes are typically expected to have many peaks [29]–[31]. They can be highly rugged due to epistatic effects of mutations [32]–[34]. They can also contain large regions or networks of neutrality [20], [21]. Empirical studies of short RNA sequences have revealed that the underlying fitness landscape has low peak density [35]: around peaks in sequences.
For the purpose of estimating the expected discovery time we can approximate the fitness landscape with a binary step function over the sequence space. We discuss two different approximations (Figure 1). For the first approximation, we consider the scenario where fitness values below some threshold, , have negligible contribution; those sequences do not instantiate the desired function (either not at all or only below the minimum level that could be detected by natural selection). We approximate the rugged fitness landscape as follows: if then ; if then . The set of sequences with constitutes the target set, and the remaining fitness landscape is neutral.
The second approximation works as follows. Consider the evolutionary process exploring a rugged fitness landscape where the goal is to attain a fitness level . Local maxima below slow down the evolutionary process to attain , because the evolutionary walk might get stuck in those local maxima. In order to derive lower bounds for the expected discovery time, the rugged fitness landscape can be approximated as follows. Let be the fitness value of the highest local maximum below . Then for every sequence in a mountain range with a local maximum below we assign the fitness value . The mountain ranges with local maxima above are the target sequences. Note that the target set includes sequences that start at the upslope of mountain ranges with peaks above . Thus, again we obtain a fitness landscape with clustered targets and neutral region, where the neutral region consists of all sequences whose fitness values have been assigned to . The two approximations are illustrated in Figure 1. For the second approximation generates larger target areas than the first approximation and is therefore more lenient.
Our key results for estimating the discovery time can now be formulated for binary fitness landscapes, but they apply to any type of rugged landscape using one of the two approximations. We note that our methods can also be applied for certain non-binary fitness landscapes, and an example of a fitness landscape with a large gradient arising from multiplicative fitness effects is discussed in Sections 6 and 7 of Text S1.
We now present our main results in the following order. We first estimate the discovery time of a single search aiming to find a single broad peak. Then we study multiple simultaneous searches for a single broad peak. Finally, we consider multiple broad peaks that are uniformly randomly distributed in sequence space.
We first study a broad peak of target sequences described as follows: consider a specific sequence; any sequence within a certain Hamming distance of that sequence belongs to the target set. Specifically, we consider that the evolutionary process has succeeded, if the population discovers a sequence that differs from the specific sequence in no more than a fraction of positions. We refer to the specific sequence as the target center and as the width (or radius) of the peak. For example, if and , then the target center is surrounded by a cloud of approximately sequences. For a single broad peak with width , the target set contains at least sequences, which is an exponential function of . The fitness landscape outside the broad peak is flat. We refer this binary fitness landscape as a broad peak landscape. The population needs to discover any one of the target sequences in the broad peak, starting from some sequence that is not in the broad peak. We establish the following result.
The regeneration process formalizes the role of several existing ideas. First, it ties in with the proposal that gene duplications and genome rearrangements are major events leading to the emergence of new genes [43]. Second, evolution can be seen as a tinkerer playing around with small modifications of existing sequences rather than creating entirely new ones [44]. Third, the process is related to Gillespie's suggestion [29] that the starting sequence for an evolutionary search must have high fitness. In our theory, proximity in fitness value is replaced by proximity in sequence space. However, our results show that proximity alone is insufficient to break the exponential barrier, and only when combined with the process of regeneration it yields polynomial discovery time with high probability. Our process can also explain the emergence of orphan genes arising from non-coding regions [45]. Section 12 of the Text S1 discusses the connection of our approach to existing results.
There is one other scenario that must be mentioned. It is possible that certain biological functions are hyper-abundant in sequence space [21] and that a process generating a large number of random sequences will find the function with high probability. For example, Bartel & Szostak [46] isolated a new ribozyme from a pool of about random sequences of length . While such a process is conceivable for small effective sequence length, it cannot represent a general solution for large .
Our theory has clear empirical implications. The regeneration process can be tested in systems of in vitro evolution [47]. A starting sequence can be generated by introducing point mutations in a known protein encoding sequence of length . If these point mutations destroy the function of the protein, then the expected discovery time of any one attempt to find the original sequence should be exponential in . But only polynomially many searches in are required to find the target with high probability in polynomially many steps. The same setup can be used to explore whether the biological function can be found elsewhere in sequence space: the evolutionary trajectory beginning with the starting sequence could discover new solutions. Our theory also highlights how important it is to explore the distribution of biological functions in sequence space both for RNA [20], [21], [35], [46] and in the protein universe [48].
In summary, we have developed a theory that allows us to estimate time scales of evolutionary trajectories. We have shown that various natural processes of evolution take exponential time as function of the sequence length, . In some cases we have established strong dichotomy results for precise boundary conditions. We have proposed a mechanism that allows evolution in polynomial time scales. Some interesting directions of future work are as follows: (1) Consider various forms of rugged fitness landscapes and study more refined approximations as compared to the ones we consider; and then estimate the expected discovery time for the refined approximations. (2) While in this paper we characterize the difference between exponential and polynomial for the expected discovery time, more refined analysis (such as efficiency for polynomial time, like cubic vs quadratic time) for specific fitness landscapes using mechanisms like recombination is another interesting problem.
Our results are based on a mathematical analysis of the underlying stochastic processes. For Markov chains on the one-dimensional grid, we describe recurrence relations for the expected hitting time and present lower and upper bounds on the expected hitting time using combinatorial analysis (see Text S1 for details). We now present the basic intuitive arguments of the main results.
For a single broad peak, due to symmetry we can interpret the evolutionary random walk as a Markov chain on the one-dimensional grid. A sequence of type is steps away from the target, where is the Hamming distance between this sequence and the target. The probability that a type sequence mutates to a type sequence is given by . The stochastic process of the evolutionary random walk is a Markov chain on the one-dimensional grid .
Consider a Markov chain on the one-dimensional grid, and let denote the expected hitting time from to . The general recurrence relation for the expected hitting time is as follows:(1)for , with boundary condition . The interpretation is as follows. Given the current state , if , at least one transition will be made to a neighboring state , with probability , from which the hitting time is .
Theorem 1 is derived by obtaining precise bounds for the recurrence relation of the hitting time (Equation 1). Consider that for all (i.e., progress towards state is always possible), as otherwise is never reached from . We show (see Lemma 2 in the Text S1) that we can write as a sum, , where is the sequence defined as:(2)
The basic intuition obtained from Equation 2 is as follows: (i) If , for some constant , then the sequence grows at least as fast as a geometric series with factor . (ii) On the other hand, if and for some constant , then the sequence grows at most as fast as an arithmetic series with difference . From the above case analysis the result for Theorem 1 is obtained as follows: If , then for all , we have for some , and hence the sequence grows geometrically for a linear length in . Then, for all states (i.e., for all sequences outside of the target set). This corresponds to case 1 of Theorem 1. On the other hand, if , then it is , and case 2 of Theorem 1 is derived (for details see Corollary 2 in Text S1).
The basic intuition for the result is as follows: consider a single search for which the expected hitting time is exponential. Then for the single search the probability to succeed in polynomially many steps is negligible (as otherwise the expectation would not have been exponential). In case of independent searches, the independence ensures that the probability that all searches fail is the product of the probabilities that every single search fails. Using the above arguments we establish Theorem 2 (for details see Section 8 in Text S1).
For this result, it is first convenient to view the evolutionary walk taking place in the sequence space of all sequences of length , under no selection. Each sequence has neighbors, and considering that a point mutation happens, the transition probability to each of them is . The underlying Markov chain due to symmetry has fast mixing time, i.e., the number of steps to converge to the stationary distribution (the mixing time) is . Again by symmetry the stationary distribution is the uniform distribution. If , then from Theorem 1 we obtain that the expected time to reach a single broad peak is exponential. By union bound, if , the probability to reach any of the broad peaks within steps is negligible. Since after the first steps the Markov chain converges to the stationary distribution, then each step of the process can be interpreted as selection of sequences uniformly at random among all sequences. Using Hoeffding's inequality, we show that with high probability, in expectation such steps are required before a sequence is found that belongs to the target set. Thus we obtain the result of Theorem 3 (for details see Section 9 in Text S1).
An important aspect of our work is that we establish our results using elementary techniques for analysis of Markov chains. The use of more advanced mathematical machinery, such as martingales [49] or drift analysis [50], [51], can possibly be used to derive more refined results. While in this work our goal is to distinguish between exponential and polynomial time, whether the techniques from [49]–[51] can lead to a more refined characterization within polynomial time is an interesting direction for future work.
|
10.1371/journal.pcbi.1005815 | Automated deconvolution of structured mixtures from heterogeneous tumor genomic data | With increasing appreciation for the extent and importance of intratumor heterogeneity, much attention in cancer research has focused on profiling heterogeneity on a single patient level. Although true single-cell genomic technologies are rapidly improving, they remain too noisy and costly at present for population-level studies. Bulk sequencing remains the standard for population-scale tumor genomics, creating a need for computational tools to separate contributions of multiple tumor clones and assorted stromal and infiltrating cell populations to pooled genomic data. All such methods are limited to coarse approximations of only a few cell subpopulations, however. In prior work, we demonstrated the feasibility of improving cell type deconvolution by taking advantage of substructure in genomic mixtures via a strategy called simplicial complex unmixing. We improve on past work by introducing enhancements to automate learning of substructured genomic mixtures, with specific emphasis on genome-wide copy number variation (CNV) data, as well as the ability to process quantitative RNA expression data, and heterogeneous combinations of RNA and CNV data. We introduce methods for dimensionality estimation to better decompose mixture model substructure; fuzzy clustering to better identify substructure in sparse, noisy data; and automated model inference methods for other key model parameters. We further demonstrate their effectiveness in identifying mixture substructure in true breast cancer CNV data from the Cancer Genome Atlas (TCGA). Source code is available at https://github.com/tedroman/WSCUnmix
| One of the major challenges in making sense of cancer genomics is high heterogeneity cell-to-cell, as a tumor is typically made up of multiple cell populations with distinct genomes and gene expression patterns. The difficulty of working with such data has led to interest in computationally inferring the components of genomic mixtures. We develop a new approach to this problem designed to take better advantage of the fact that mixtures of cells across tumors or tumor regions can be expected to be highly non-uniform; samples that share greater common ancestry or progression mechanisms are likely to have more similar mixtures of cell types. We present new work on reconstructing mixtures from multiple genomic samples where the samples can be presumed to share such a pattern of similarity. Our methods automate the process of reconstructing these mixtures and the relationships between samples. We demonstrate their effectiveness on tumor genomic data in comparison to alternative methods in the literature.
| Tumor heterogeneity is now recognized as a pervasive feature of cancer biology with implications for every step of cancer development, progression, metastasis, and mortality. Most solid tumors exhibit some form of hypermutability phenotype [1], leading to extensive genomic variability as tumor cell populations expand [2]. Studies of single cells by fluorescence in situ hybridization (FISH) [3, 4] have long revealed extensive cell-to-cell variability in single tumors, an observation that has since been shown, by single-cell sequencing technologies, to occur with a far greater scale and variety of mechanisms than previously suspected (e.g., [5, 6]). Furthermore, studies of clonal populations across progression stages have revealed that it is often rare cell populations that underlie progression, rather than the dominant clones [4]. Indeed, heterogeneity itself has been shown to be predictive of progression and patient outcomes [7]. All of these observations have suggested the importance of having ways of accurately profiling tumor heterogeneity, for both basic cancer research and translational applications.
Experimental technologies for profiling tumor heterogeneity are constantly improving, but are so far impractical for systematically profiling variability genome-wide in large patient populations. FISH and related imaging technologies can profile many thousands of cells, but only at limited sets of preselected markers [4]. Single-cell sequencing can derive genome-wide profiles of hundreds to thousands of cells in single tumors [5, 8, 9], but is so far cost-prohibitive for doing so in more than very small patient populations. Furthermore, technical challenges make it difficult to develop accurate profiles of structural variations, such as copy number variations (CNVs), which are the major drivers of progression in most solid tumors [10]. Bulk regional sequencing can profile small numbers of tumor sites per patient in large patient populations [11] but provides only a coarse picture of the heterogeneity within each site. RNA sequencing (RNA-Seq) provides a measure of the quantity of RNA expression and is practical on substantially larger numbers of single-cells than DNA-Seq [9]; however, it is subject to greater noise than DNA-Seq [12] and provides a more indirect measure of clonal heterogeneity.
These technical challenges to assessing heterogeneity experimentally have led to enormous interest in computational deconvolution (also known as mixed membership modeling or unmixing) methods as a way of computationally separating cell populations from mixed samples. Originally proposed as a way of correcting for stromal contamination in genomic measurements [13], such methods were later extended to reconstructing clonal substructure [14] and subclonal evolution [15] among tumor cell populations. The past few years have seen an explosion of such methods for deconvolution of numerous forms of genomic data sources (e.g., [16–31]). All such methods, however, are limited in accuracy and capable of resolving at best a few major clonal subpopulations, a small fraction of the heterogeneity revealed by single-cell experimental studies. These limits result from an inherent difficulty of separating high-dimensional mixtures, especially from sparse, noisy data. The gap between the heterogeneity we know to be present and what we can resolve by deconvolution is enormous, suggesting a need for further methodological advances.
Genomic deconvolution is a burgeoning field in which many different approaches are now available, often differing in models, algorithms, and the kinds of data or study design for which they are well suited. Leading contemporary approaches include TITAN [19], THetA [18], THetA2 [32], PhyloWGS [33], SPRUCE [34], Canopy [35], BitPhylogeny [36], and PyClone [37], each of which we briefly discuss here. TITAN uses a graphical model to estimate subpopulations based on copy number alterations and loss of heterozygosity events for whole genome or whole exome sequencing data, assuming as input read depths and allelic ratios at single nucleotide variant (SNV) sites. THetA and its follow-up version THetA2 perform tumor composition estimation using both SNV and copy number data derived from sequence read depths. PhyloWGS uses a probabilistic model to perform deconvolution jointly with phylogeny inference specifically on low-coverage whole genome sequencing data, making use of copy number estimates and variant allele frequencies (VAFs) of simple somatic variants. SPRUCE uses SNV and CNV data similar to that of THetA/THetA2 to make inferences as to the composition of heterogeneous tumor samples, but via a combinatorial enumeration strategy to explore the space of possible phylogenies consistent with a data set. Canopy optimizes for a probabilistic model to perform joint phylogenetic inference and tumor deconvolution from a data set based on several data sources, including VAFs and allele-specific copy numbers. BitPhylogeny similarly performs joint phylogenetics and deconvolution using Markov chain Monte Carlo (MCMC) sampling, but is unusual among methods in this domain in making use of DNA methylation data. PyClone performs tumor deconvolution for multiple samples from a single patient using SNV data, CNV data, and combinations thereof as input and is designed to work specifically with targeted deep sequencing data (>1000X coverage).
In prior work, we proposed that one could better resolve genomic mixtures by taking account of extensive substructure we would expect such mixtures to exhibit [21]. That is, an individual tumor or tumor site is not likely to be a uniform mixture of all cell types observed across all tumor samples in a study. Rather, one can expect distinct samples to group into subsets that share more or fewer cells depending on how closely related they are to one another. For example, all tumor samples can be expected to share some contamination by normal cells while tumors with common subtypes can be expected to share both normal cells and cell states characteristic of those subtypes. Likewise, tumor regions might be expected to share more similarity with those nearby than those more distant in a single patient. This kind of substructure is in principle exploitable to improve our ability to reconstruct accurate mixed membership models. Specifically, by deconstructing tumor samples into subgroups with similar mixtures, one can decompose the problem of reconstructing a high-dimensional mixture into the easier problem of reconstructing several overlapping lower-dimensional mixtures.
We previously showed how to implement such an approach to substructured mixture deconvolution, adapting an earlier deconvolution strategy for uniform mixtures that was based on identifying geometric structures (simplices) of tumor point clouds in genomic space [15, 38] but subdividing these point clouds into low-dimensional subsimplices that collectively constitute a higher-level object known as a simplicial complex. This prior work used a pipeline of several sequential steps to transform a genomic point cloud into a structured mixed membership model [21]:
The resulting pipeline established a proof-of-concept for the approach, but also introduced several difficult computational challenges. For example, it required accurately pre-specifying the number of partitions and the dimensionality of each of the partitions, both difficult inference problems in themselves that require significant knowledge of the system under study.
In the present work, we improve on this proof-of-concept method by tackling several subproblems on the path to more completely automating inference of substructured genomic mixtures from populations of tumor samples. We have eliminated several nuisance parameters from the prior work, most notably by introducing methods for automated dimensionality estimation of subsimplicies and automated maximum likelihood inference of other previously user-defined parameters. We also improve upon our earlier work by proposing a model better suited to capture the uncertainty in cluster assignments through use of a fuzzy clustering representation of data points (samples) with respect to the inferred simplicial complex (and therefore the tumor phylogeny), allowing tumor samples to exhibit partial or uncertain membership in multiple phylogenetic branches. This flexibility is of particular importance when a sample is near a branch point in the simplicial structure, which corresponds biologically to a sample having a genomic profile similar to a most recent common ancestor of multiple tumor lineages. In addition, we develop a more comprehensive likelihood function, allowing us to optimize over and thus eliminate nuisance parameters from prior work.
Although the approach we introduce makes inferences as to intraturmor heterogeneity, we use information present across multiple patients (that is, intertumor heterogeneity) to make those inferences. This application assumes that commonalities in progression processes can be observed across subgroups of patients, even if the exact presentation is unique for each tumor. Because the model presumes common subgroups of tumors proceeding along similar evolutionary trajectories, an inferred mixture vertex will correspond to a coarse-grained model of a shared progression stage among a subset of tumors. That is, the vertex, would be interpreted as an approximate representation of a recurring cell type appearing in the course of progression of multiple samples. Since no two samples have exactly the same evolutionary history, however, it would be expected to reflect the common features of a cluster of similar cell types while averaging out their differences. The overall simplicial complex structure will correspond to a model of the space of evolutionary trajectories among all of these progressions stages across all observed tumor subgroups. Paths in the evolutionary tree will correspond to the recurring evolutionary pathways between the averaged progression stages represented by the vertices. Based on those reconstructions, we can then make inferences for each sample as to the relative amounts of each progression stage represented in that tumor, providing a coarse-grained inference of intratumor heterogeneity.
We validate the approach through application to breast tumor data from The Cancer Genome Atlas (TCGA) [39] and comparison with the widely-cited PyClone software [37]. We also compare with a more recent deconvolution method using DNA methylation data, providing an independent basis for comparison to the DNA copy number and RNA expression-derived deconvolution of our method [28].
In this section, we go through each step of our improved analysis pipeline, followed by a discussion of validation and application to real tumor data. We break the full inference problem into a series of sequential steps. Fig 1 provides a high-level overview of the process. The following subsections provide details on each component.
We conceptually model input data as a matrix M ∈ R s × g, where the s ∈ N rows correspond to distinct samples (which might be biopsies of tumors in a patient population, tumor sites in a single patient, or regions of a single tumor) and the g ∈ N columns correspond to probes along a genome (typically one per gene, although potentially at lower or higher resolution). Note, however, that as the underlying data types input to the method are changed, the interpretation of output is changed correspondingly. For instance, if the features used as input are not gene copy numbers, but rather SNV sites, then the components of the matrix M will be SNV VAFs for the given samples. Similarly, if samples are different regions from a single patient, the inferred phylogeny is for a single patient, rather than across a patient panel. For ease of exposition, we refer to rows as samples and columns as genes below. We use this generic matrix format because data from many sources can be preprocessed into such a matrix (e.g., array-based CNV, SNV, or expression data or whole-genome or whole-exome sequence-derived CNVs, SNVs, or expression levels). Although the basic strategy is intended to be generic with respect to platform and genomic datatype, we specifically consider here three scenarios: 1) CNV data as might be derived from array comparative genomic hybridization (aCGH) or DNA-Seq read depths, 2) RNA expression data as might be derived from expression microarrays or RNA-Seq, and 3) a heterogeneous combination of DNA CNV and RNA expression data. Our goal is to decompose the rows of M into an approximately convex combination of a smaller set of unknown mixture components (putative cell populations). More formally, we seek a decomposition
M = F V + ϵ (1)
where F ∈ R s × k are mixture proportions, V ∈ R k × g are unmixed subpopulations, k ∈ N is the number of inferred cell subpopulations, and ϵ ∈ R s × g is an error matrix. F is interpreted as the mixture fractions of the pure subpopulations, also called mixing proportions, and V as the inferred genomic profiles of the pure subpopulations, also called mixture components. This interpretation leads to natural constraints on the problem: 1) ∑i Fij = 1 for a fixed j and 2) ∀i, j: 0 ≤ Fi, j ≤ 1. Given these constraints, the formal goal of the method is to compute F and V given M, with an intermediate step of determining the mixture dimension k.
Our approach to performing this deconvolution involves constructing a more involved simplicial complex mixed membership model, which will imply F and V, through a series of discrete inference steps. While most aspects of model inference are automated, as detailed in the remainder of Materials and Methods, the following parameters and hyperparameters still require manual selection:
To begin analysis, we first pre-process M into a matrix of Z-scores:
M z = M − μ M σ M (2)
where μM is a vector of the mean copy numbers of each gene across all samples, and σM is a vector of the standard deviations of the copy numbers.
This process is altered slightly to accommodate heterogeneous DNA and RNA data that have been concatenated as features. We assume that the distributions of read counts will differ for DNA and RNA data, so instead of μ and σ for all samples column-wise, we use a μ and σ for pools of all data for each data type. That is, we evaluate the mean and standard deviation for Z-score computation for all samples and for all DNA features, and separately for all samples and all RNA features. In the RNA only case, we use the framework outlined in Eq 2. Next, to facilitate analysis of genomic point clouds, we reduce the dimension of the data using principal components analysis (PCA) [40]. While there are more sophisticated dimensionality reconstruction strategies available, we favor PCA as a simple, standard method that has relatively modest data needs. We identify a total of kupper PCs, using the Matlab pca routine in economy mode, where k u p p e r ∈ N < g is an upper bound on the number of cell subpopulations we will infer. In the present work, we use kupper = 12, intended to be approximately an upper limit on the number of distinct mixture components a method of this class might be able to infer. We denote the PCA scores, corresponding to amounts of each PC in each tumor, as S M ∈ R s × k u p p e r. Then, in order to fine-tune the automated dimensionality detection, we implement the sliver method of dimensionality estimation described in [41]. The core model proposed by that work relies on testing for the presence of “slivers”, geometric objects with poor aspect ratios, which occur when the following expression, which we call Assertion 3, is satisfied:
ν < δ j r wherer = L j j ! (3)
where
ν represents the volume of some enclosing structure,
j represents the current estimate of dimension, increasing for each time Assertion 3 is false
up until the limit of 12, and
δ represents a tolerance factor between 0 and 1.
For a quick estimate of an enclosing structure, we use the algorithm proposed in [15]. We then use the top j − 1 PCs after the algorithm terminates. To automate the selection of the δ parameter, we use all values spaced 0.05 apart between 0 and 1. The range of possible δ values is 0 to 1 for this parameter based on the approach outlined by [41]. Because some values of the parameter lead to the same estimate of the dimensionality of the dataset, we choose one representative value from each partition of the range of dimension estimate values, then choose the model that has the highest likelihood.
Lastly, we normalize the scores for each PC to a [0, 1] range, which is then assumed by the pre-clustering technique applied in the next section [42]. We compute the 0–1 normalized version of SM as
S [ 0 , 1 ] = S M − min S M max S M − min S M (4)
where the minimums and maximums are computed for each PC, taken over all samples.
We next pre-cluster data to identify initial candidate subsets of samples inferred to have drawn from the same set of mixture components. Each such subset will correspond to a distinct subsimplex of the full simplicial complex to be inferred. While this is a clustering problem, it is a non-standard one in that we seek to cluster data into distinct low-dimensional subspaces of a contiguous higher-dimensional point cloud, rather than into disjoint subclouds as is in conventional clustering. We developed a specialized clustering method for this purpose [42], based on a two-stage variant of medoidshift clustering [43]. We initially cluster in Euclidean PC space to reduce the raw data to a smaller set of representative data points. We then cluster these representatives under a negative-weight exponential kernel function using the ISOMAP distance measure [44], a form of geodesic metric measuring distance between data points through a k-nearest-neighbor graph of the input point cloud, which collectively draws on features of manifold learning and related technologies. The combination of ISOMAP distance and negative exponential kernel produces a clustering in which cluster representatives are approximately extremal points of the simplicial complex that serve to pull apart distinct subspaces of the point cloud. The initial Euclidean clustering suppresses noise, which otherwise makes the negative exponential kernel highly sensitive to outlier data. We refer the reader to [42] for full details. At the end of this process, we are left with a small set of cluster representatives M2stage, defined as the union over clusters i of a neighborhood N(xi) of points associated with each cluster representative xi:
M 2 s t a g e = ∪ i M N ( x i ) , (5)
where each representative is itself a point in S[0,1], and a corresponding clustering of all samples C = {C1, …, Cr}, S[0,1] = ⋃Ci∈CCi.
We further assess uncertainty of the cluster assignments by determining a relative statistical weight of each data point in each cluster. We use a weight function based on a folded multivariate normal distribution, where the mean of the function is a 0 vector, the covariance matrix is the identity multiplied by the distance from each cluster center to the mean of all cluster centers, and the value at which the density function is evaluated is the distance from xi to Cj in ISOMAP space. After these relative weights have been derived, we convert them to probabilities of assignment of each point to each cluster. If we denote the raw weight of the ith data point as a vector Ri, then we can define the normalized weight vector:
W i = R i − min C j ∈ C R i max C j ∈ C R i − min C j ∈ C R i (6)
In the above formula, Cj refers to an arbitrary cluster in the clustering C, over which we maximize or minimize. The clustering in principle depends on a chosen neighborhood size for the k-nearest-neighbors graph, although a scan over all possible neighborhood sizes found no sensitivity of the final model likelihood to this parameter.
We next seek to estimate the dimension of each cluster, which will correspond to the number of mixture components inferred for that cluster. The major challenge of this step is distinguishing a genuine axis of variation from random noise stemming from biological and technical limitations, particularly when working with sparse, noisy genomic measurements. Intuitively, we identify dimension by iteratively adding axes of variation via PCA until we can no longer reject the hypothesis that variance in the next dimension is distinguishable from noise.
We first build a model of expected noise per dimension by randomly sampling data points of pure Gaussian noise with mean 0 and identity covariance. We then perform PCA on this random point cloud and estimate the mean μG(i) and standard deviation σG(i) of the point cloud for each PC i ∈ 1, …, kupper. We then identify the smallest i ≤ kupper such that the standard deviation of the true data in PC i is smaller than μG(i) + κσG(i), where κ defines a significance threshold in standard deviations. In the present work, we set κ = 3 to yield effectively a significance threshold of < 0.001 for rejecting the hypothesis that the next dimension can be explained by Gaussian noise. The result of this module, then, is a vector of inferred dimensions of each of the clusters: D ∈ {1, …, kupper}r. We would expect this test to be conservative (underestimate true dimension), although less so as the size of the data set and its precision increases. We found it necessary to use a custom-made conservative dimensionality estimator, as opposed to a more standard technique (e.g., [41]), because the number of data points available in this application is much smaller than is typically assumed by methods in this problem domain. We use the approach outlined in [41] in the initial phase, as it is prior to the pre-clustering, and therefore typically has a several-fold increase in the minimum number of data points considered, bringing it better in line with the data needs of that method.
We next seek to establish an initial mixed membership model by separately unmixing each cluster, using the inferred dimension from the previous step as the number of mixture components. We establish the model by minimizing an objective function based on the noise-tolerant geometric unmixing method of [38]:
P ( θ | X ) ∝ ∏ i = 1 r ( e x p ( − ∑ j = 1 s ( | x i − F j i V j i | W j i ) ) M S T ( V j , A j ) − γ β ) (7)
Where
γ is a regularization penalty set based on an estimated signal-to-noise ratio (SNR) of the data source [21],
V are the inferred vertices,
A is the adjacency matrix,
MST is a minimum spanning tree cost,
W is the relative weight function computed above,
F are the inferred mixture components,
xi is the ith data point,
β is a BIC penalty for model complexity [45] and,
|⋅| is L1 distance.
The first term penalizes data points outside the bounding simplex via an exponentially-weighted L1 penalty. The MST term captures a form of minimum evolution model on the simplex itself intended to penalize the amount of mutation from a common source needed to explain the simplex vertices (mixture components) [21]. We optimize for the objective function via the Matlab fmincon function, fitting V and F to assign mixture components and mixture fractions to each cluster independently. In practice, we use a transformed version of the equation into negative log space, as the optimization packages are built for minimization rather than maximization, and log domain better handles underflow for small likelihoods while preserving the ordering of solutions.
We next seek to join the discrete simplices, each modeling a subset of samples as a uniform mixture, into a unified simplicial complex. We accomplish this by merging simplex vertices if we cannot reject the hypothesis that they represent distinct points in genomic space. We first establish a probability model using the k-nearest-neighbors graph on samples and vertices by modeling the set of overlapping neighbors between two vertices via a hypergeometric distribution. On the assumption two vertices draw their neighbor sets independently from the pool of all samples, the expected number of data points in common would be
| N 1 | | N 2 | N (8)
where there are N data points, |N1| nearest neighbors of the first vertex, and |N2| neighbors of the second vertex. We merge two vertices when the number of observed overlapping nearest neighbors is above expectation. We empirically determined on our synthetic data that the method is insensitive to the number of nearest neighbors for choices between 2 and N and chose k = 15 nearest neighbors arbitrarily within this range for the real data. This approach replaces computationally costly bootstrap estimates used in our prior work [21].
For those instances in which the process above does not result in a single connected simplicial complex, we add a step of post-processing to reconcile the geometric body into a single, connected simplicial complex. For those collections of bodies that do not consist of one connected component after the hypergeometric distribution correction, we iterate over all pairs of simplex vertices, merge the two vertices by creating a new vertex from the mean of the previous two vertices in all features, set the adjacency matrix to the union of the adjacency matrices of the two previous vertices, and compute the value of the objective function outlined in Cluster-wise Unmixing. We continue to merge points until there is at least one candidate consisting of a single connected component. If there are multiple such candidates, the candidate with the lowest objective function value, corresponding to the maximum of the likelihood function, is chosen. Pseudocode for this algorithm is provided in Fig 2.
To demonstrate the efficacy of the algorithm, we use breast cancer (BRCA) CNV and RNA-Seq data from The Cancer Genome Atlas (TCGA) [39]. We downloaded level 4 DNA CNV data on 2 Jun 2016 (1,080 samples) and RNA-SeqV2 data on 1 Jun 2016 (1,041 samples), of which 1,022 samples were in common, along with clinical data for this cohort. For copy number data at level 4, gene features are extracted and a list of genes is provided, in contrast to the blocking procedure required by earlier work [42]; however, the platform is flexible to represent more or less granular data.
We ran the pipeline using the following parameters: maximum number of dimensions supplied to the pre-processing sliver method: 12; number of bootstrapped replicates for pre-clustering: 1000; neighborhood size for pre-clustering: 1; number of nearest neighbors for vertex merger: 15; cutoff for dimensionality estimation: 3 standard deviations; maximum number of iterations of fmincon per simplex: 1000. The choices reflect computational resource limitations, as well as a stable number of bootstrapped replicates, and choices to ensure convergence of the methods. The neighborhood size was chosen based on assumptions implicit in our normalization technique—for full details, see [42]. The number of nearest neighbors was chosen based on the test of simulated data similar to [42] demonstrating insensitivity to this parameter up to approximately N neighbors. The 3 standard deviations chosen correspond to a p-value of approximately 0.001. The runtime of the experiments depends largely on the dimension of the maximally likely clusters (i.e., the number of subpopulations in the tumor dataset that our model chooses as most likely) and the number of iterations in the minimization phase (iterations of fmincon).
In order to assess the consistency of our method with respect to outlier data points, we conducted a sensitivity analysis using the TCGA CNV data. The sensitivity analysis was structured in an analogous fashion to 10-fold cross validation. For each of ten iterations, we excluded 10% of the data set, selected by a random uniform distribution. For the remaining data, the model was run to completion to produce a simplicial complex and assignment of mixture components and mixture fractions to the data points in that set of replicates. We then compared inferences by several measures to assess consistency across subsamples of the data.
We assessed similarity of the inferred component sets between replicates. To assess similarity of two sets of inferred vertex components A and B, we first identified for each component in A the closest matching component B, based on normalized Euclidean distance in PC space. We likewise identified for each component in B, the closest matching component in A. We assigned a score for the similarity of two vertex sets based on the mean distance between each component and its closest match relative to the mean distance between pairs of distinct components within A and within B.
To demonstrate the utility of our method, we consider three applications to data derived from the TCGA breast cancer cohort [46]. Breast tumors were chosen as an application case for two key reasons. First, there are a larger number of breast tumor samples than any other organ cancer type in TCGA, valuable for cross-cohort deconvolution. Second, breast tumors have well-documented clinical subtypes (HER2+; ER/PR+, and Triple Negative), a useful feature for validation since we would expect tumors and cell lineages within them to partition largely by subtype.
RNA-Seq data was downloaded from TCGA. The data consists of lists of gene expression in normalized counts, as well as gene name lists identifying each feature. Data from each of the samples were concatenated into a matrix of samples by genes. Using the parameters described above, the weighted unmixing procedure produces a tetrahedral simplex. Although other simplicies and simplicial complexes were considered by our algorithm, the tetrahedron was determined to be the maximum likelihood model. The results are illustrated in Fig 3, which shows the true point cloud as well as our inferred structure, where samples are colored by the clinical subtype. The DNA level 4 data consists of log2(⋅) copy number ratios, which are exponentiated and Z-scored prior to unmixing following the methods outlined above.
We also considered application to DNA CNV data from TCGA. The results are visualized in Fig 4. The decreased noise of DNA CNV technology relative to RNA-Seq technology results in a more sharply defined simplicial complex structure than was apparent with RNA-Seq data, consisting of three lines connected at a shared fulcrum. We attribute the clearer structure to the lower inherent stochasticity of DNA versus RNA data, which would be expected to better approximate the assumption that mixtures of cells will behave as linear combinations of their underlying cell types. We note that the central vertex, labeled 4, appears skewed away from the apparent junction of the three subsimplices. We attribute this skew in the position of the junction to the difficulty of accurately clustering samples near such subsimplicial boundaries, leading to imprecise positioning of the shared vertex in the distinct subsimplices that is only partly corrected when the vertices are merged.
Lastly, we considered a combination of DNA and RNA features. Because of the varying noise profiles of the data types [12], we adjusted the normalization procedure as outlined above. We have plotted the results of the unmixing below in Fig 5, using the same color code for tumor subtypes as with the RNA-only and DNA-only data. The combined data leads to a somewhat more complex structure than either individual data type alone, consisting of a tetrahedron and triangle connected at a point. The higher dimension compared to the individual data types may reflect changes in the overall noise profile or to the complementary aspects of progression that are revealed by the two data types in isolation.
We further used the TCGA CNV data to assess sensitivity of the method to subsamples of the data. We assessed reproducibility across ten replicates of 90% subsamples of the TCGA data and quantified reproducibility of inferred mixture component sets based on the ratio of Euclidean distances between best matching component pairs between replicates versus Euclidean distances within replicate sets. A score below one would then indicate general consistency between vertex sets relative to variability within each set, while a higher score would then be interpreted to mean that vertex components are highly distinct between runs relative to the variability among components within a set. Across all 45 comparisons among pairs of replicates, we found a mean distance of 0.6806 by this measure. This result suggests there is sensitivity to outliers in the simplicial inference leading to variability replicate-to-replicate, but that there is nonetheless similarity run-to-run relative to the variability in individual data sets.
To assess the functional and biological significance of the inferences made by our model in each of the three test cases, we projected the data points from PC space back into genome Z-score space. We then identified genes lists with statistically-significant increase or decrease in Z-score as assessed by Bonferroni-corrected p-values. In the RNA and combined cases, we used p = 0.01 after correction. In the DNA case, at the p = 0.01 level, DAVID [47] reported that the number of genes provided was too large to process the results. As a result, we chose a stricter threshold of p = 2.1905 × 10−11, the smallest value we could choose without producing underflows in the p-value calculation.
Those genes that were statistically significantly upregulated were then evaluated on a per-vertex basis by DAVID [47] for enrichment by functional terms corresponding to specific networks, pathways, or other functional classes. In our case, we have specific interest in enriched tissues, diseases, and disease classes, as these areas provide the ability for the database to point specifically to our dataset. As expected, the DAVID analysis revealed enrichment for several terms related to breast cancer specifically, as well as breast tissue more broadly. In Tables 1–3, we provide the most significantly enriched terms for each of RNA, DNA, and combined RNA/DNA deconvolution. Tables 1 and 2 present the ten most significantly enriched terms for RNA and DNA, respectively. Complete lists of significantly enriched genes (p ≤ 0.05) appear in Supplementary Material as S1 and S2 Tables. Only seven terms were significantly enriched for combined RNA/DNA deconvolution and therefore only those are listed in Table 3. Comparison of the method shows that DNA-only results in the largest number of distinct pathways enriched, followed by RNA-only, then combined. Combined, however, is most specifically enriched for expected term classes broadly related to cancers and breast tissue. These results may suggest that the combined data is more effective at achieving high specificity at a trade-off in sensitivity.
While there are many deconvolution tools in this domain, the variations in data assumptions of the methods make direct head-to-head comparison difficult. While TITAN [19] can make inferences from similar copy number variation data to our method, it depends on knowledge of alleleic frequency data at SNV sites unavailable to us in the present analysis. THetA [18] and THetA2 [32] perform a comparable form of inference but are tuned specifically for inference from a single tumor, making them unsuitable for comparison on a patient cohort for which our methods are designed. PhyloWGS [33] is designed for whole-genome analysis like our method, but depends on availability of variant allele fractions of novel somatic variants, a model and data type again unsuited to the kind of cross-cohort analysis performed by our method. SPRUCE [34] likewise depends on VAF data under the assumption that all samples are drawn from a single patient, making it poorly suited to the kind of data for which our method is designed. Canopy [35] likewise makes use of VAFs and allele-specific copy number data unavailable to us and poorly suited to the kind of cross-cohort analysis for which our method is designed. BitPhylogeny [36] likewise assumes a data type unavailable in our application, methylation data in that case, making direct comparison on real data infeasible.
In order to allow for some comparison to an alternative in the literature, we choose to compare to PyClone [37], as it is is a highly cited method producing similar output to our method that can in principle make inferences from a common set of data to our method. PyClone can work with copy number data and can optionally omit allele-specific frequency information (although it is designed to make use of such information if it is available). We emphasize that although PyClone can be run on a common data set to our method with some preprocessing, it is tuned for very different assumptions on those data than our method. PyClone assumes precise frequency estimates on small numbers of sites, as is appropriate to the targeted deep sequencing data for which it was designed, while our method is designed to use less precise data on large numbers of markers, as is appropriate for the whole exome or whole genome data for which it was designed. Furthermore, PyClone is also designed for multiple samples from a single tumor while ours is designed to work with cross-sectional data from distinct tumors. While we can run both methods on a common set of data, we thus cannot devise a single dataset that provides a fair test of both. Our intention in comparing the methods, then, is not to show that our method is superior to PyClone but rather that our method is filling a niche for which prior tools are not designed and to which they do not generalize well.
In order to preprocess the data in a format amenable to the PyClone system, we assume a read length of 300, and baseline copy number of 2. For this analysis, we assume a copy number of 2 for any region for which which there is no copy number alteration call in the data. We also omit analysis of sex chromosomes. We omitted allele-specific copy numbers as input to PyClone because this information is not part of the publicly-available version of TCGA data. Although both our approach and PyClone can run on SNV data, it proved computationally infeasible to include the SNVs in this dataset for the PyClone analysis, as PyClone is not designed to handle such a large marker set nor to work with markers drawn from many genetically distinct tumors.
We first attempted to run the full dataset of all level 4 gene copy number breast tumor samples from TCGA through the PyClone pipeline on a workstation equipped with an Intel i7-4770K processor at 3.5GHz per core, with 32GB of RAM. However, the approach was unable to complete in approximately 1 week of running time, which we inferred may be due to the large number of genes (> 20,000) present in the full dataset, an amount well in excess of the small targeted sequencing data assumed by PyClone, as well as by the fact the PyClone algorithm runs on a single core. We thus pruned the list of genes (features) to a subset of corresponding to known breast cancer driver genes from [48]. PyClone then successfully ran on the set of tumor data points from the TCGA breast tumor dataset. PyClone output also differs somewhat from that of our method, requiring some post-processing to facilitate comparison. PyClone outputs mean and variance scores for each sample, for each cluster of mutations, which can approximately relate to our vertices. Further, we consider a version of the means of the scores normalized to sum to one analogous to the mixture fraction scores we generate. We then test for similarities in Spearman correlation of our model’s inferred mixture fraction rank to the rank of mixture fraction provided by PyClone.
Results of comparison of our method with PyClone appear in Table 4. The PyClone comparison points (Py1 to Py4) correspond to the inferred mutational cluster prevalences. To make a fair comparison, we normalized the prevalences by their sums on a per-cluster basis to derive a fractional composition estimate based on PyClone. We then used Spearman correlation as a comparative tool to examine how similar in rank PyClone’s inferences of which clusters of genes were dysregulated are to our ranking of fractional composition with respect to inferred vertex amount. Because the vertices represent inferred pure subpopulations within tumor samples and are in PC space, the vertices are equivalent to genomic profiles of the subpopulations, where sets of genes are mutated. The correlation analysis provides a matrix of correlations where each element corresponds to the correlation between dysregulation of one Pyclone cluster and representation of one inferred subpopulation by our method. While in this case the two methods produced equal numbers of clusters, we would not expect that to be true for all data sets and the analysis does not assume the matrix dimensions are equal. There is significant (p < 0.01) positive correlation between 3 of our 4 vertices and 3 of the 4 clusters inferred by PyClone, signaling general agreement between methods in their estimate of substructure.
We further sought to validate our approach by comparing correlation of PyClone inferences to clinical labels supplied by TCGA (Table 5) with correlation of the inferences of our simplicial complex approach to the TCGA clinical labels (Table 6). Considering both positive and negative correlations, at the p≤0.0001 level, our approach has four entries significantly correlating to the clinical labels across two vertices, as compared to three entries across two clusters in the case of PyClone. Additionally, including those at a weakly significant (p ≤ 0.05) level, our approach has five entries correlating three of the four vertices’ mixture fractions to clinical subtypes, while PyClone has four entries across three of the four clusters’ mixture fractions to clinical subtypes.
To provide an additional point of comparison, we also applied our method to TCGA RNA-Seq data. PyClone is not designed to accommodate RNA-Seq data, so we provide results only for our method. Table 7 shows the results. In this test case, the simplicial complex approach retrieved significant (p ≤ 0.01) correlation at each of the inferred vertices, and for each of the subtypes, with a total of 9 significant entries.
While a perfect comparison of our method with PyClone, or any prevailing method known to us, is impossible given different data assumptions and input and output types, these comparisons provide clear evidence that our method is at least comparable in ability to identify substructure among tumor data sets when given appropriate input data to its model assumptions. On the whole, we interpret the results as being in agreement on the tested BRCA TCGA dataset, with our method providing the additional benefits of
Having the option to run on expression data, gene copy data, or heterogeneous combinations thereof,
Not requiring matched tumor-normal data or assumptions about normal samples, and
Being amenable to much larger numbers of features, such as might be derived from whole-genome data (WGS/WES), in comparison to PyClone or similar tools
In cases where the constraints of PyClone (deep targeted sequencing, matched tumor-normal samples) are well-satisfied, it may perform more accurately than the general approach we have developed, but in cases of lower read depth, datasets missing some or all normal matched samples, or with whole-genome coverage, the simplicial complex approach may be more appropriate.
We note that our method was not able to produce useful results on the trimmed list of genes we produced to yield a manageable gene set for PyClone. We speculate that the high noise in the data makes it infeasible to estimate simplicial structure from a small targeted gene set, resulting in our method fragmenting the samples into many more clusters. Our approach thus appears to be poorly suited to targeted gene sets and better to large or whole-genome data sets, in contrast to PyClone, which is well tuned for small numbers of genes but not computationally feasible for whole-genome data.
In order to further validate the approach, we examined Spearman correlation with an orthogonal data set. Onuchic et al. [28] developed a deconvolution approach based on DNA methylation data from TCGA [46]. The result of the Onuchic et al. [28] approach was a deconvolution of the data into constituent subtypes categorized into 5 cancer subgroups, a stromal group, an immune group, and a normal group. The results of correlating our results to theirs are shown in Table 8. There is significant (p < 0.01) positive correlation between what we estimate as the fulcrum of the simplicial complex—correspondening to the most recent ancestor in a phylogenetic interpretation—and the Onuchic et al. [28] estimate of stromal, immune, and normal composition. Further, our vertex 1 correlates in a statistically significant and positive way to their estimates of cancer subtype 1 and cancer subtype 5.
We have developed a novel method for taking better advantage of mixture substructure in deconvolution of mixed genomic data from heterogeneous tumor samples. This contribution is intended to advance a theoretical strategy for better resolving substructure in complex genomic mixtures, a general strategy that might be incorporated into many existing approaches for cell type deconvolution using assorted data types and inference models. The advances in the present paper bring us closer to the goal of deriving precise models of complex mixture substructure in the face of sparse, noisy genomic data without the need for extensive expert intervention. For this purpose, we have introduced new strategies for automated inference of subcluster dimensions, automated construction of a global simplicial complex structure, and better deconvolution of submixtures on small samples with uncertain subclustering. We have shown that we can automatically learn model structure from realistic sizes of data set without degrading performance of the model relative to methods requiring significantly more user intervention. We have further shown that this general approach is effective to varying degrees on CNV, RNA-Seq, and heterogeneous data sets. We have further shown that our method has comparable ability to resolve mixture structure to a leading deconvolution method, PyClone, on a common data set, while demonstrating several advantages in relaxing assumptions on data type, source, and quality.
The ultimate goal of the present work is to make sophisticated mixture deconvolution approaches more widely accessible to a non-expert community, by allowing them to be incorporated more broadly into a variety of deconvolution approaches in the literature. Much work still remains, though, both in better automating these approaches and improving inference quality. There are still several (hyper-)parameters for which the task of automated learning remains challenging. While automated dimension estimation appears valuable in improving simplicial complex models, deriving accurate estimates is a significant challenge for sparse, noisy data [49]. Integration of additional forms of genomic data into a common mixture framework is likewise a promising but challenging direction for improving inference quality. The computational framework presented here could also in principle be applied to many genomic samples from a single patient (e.g., distinct tumor regions, sites, or timepoints), although we do not explore that application here as data of this form is still scarce. The exact data needs of the method would depend on the heterogeneity across samples. We would expect this inter-sample heterogeneity to be substantially smaller for multiple samples from a single tumor than for the application to distinct tumors examined here, but nonetheless higher than is required for other tumor deconvolution methods that infer simpler underlying mixture models. Further, while we have applied this approach here to two data types and their combination, the same general strategy might be applied to many forms of genomic measurement (CNV, RNA expression, SNV, epigenetic, proteomic) and technologies for assessing them (array, sequence, or other high-throughput methods). Furthermore, as single-cell methods become more cost-effective, combinations of bulk and single-cell data may prove particularly informative. Finally, the simplicial complex models themselves require refinement to better capture the real sources of genomic mixture substructure they are meant to model, including substructure imposed by common pathways of subtype evolution, spatial constraints in the tumor microenvironment, and other sources of mixture substructure that do not conform well to our current simplicial complex model.
|
10.1371/journal.pcbi.1002502 | Virus Capsid Dissolution Studied by Microsecond Molecular Dynamics Simulations | Dissolution of many plant viruses is thought to start with swelling of the capsid caused by calcium removal following infection, but no high-resolution structures of swollen capsids exist. Here we have used microsecond all-atom molecular simulations to describe the dynamics of the capsid of satellite tobacco necrosis virus with and without the 92 structural calcium ions. The capsid expanded 2.5% upon removal of the calcium, in good agreement with experimental estimates. The water permeability of the native capsid was similar to that of a phospholipid membrane, but the permeability increased 10-fold after removing the calcium, predominantly between the 2-fold and 3-fold related subunits. The two calcium binding sites close to the icosahedral 3-fold symmetry axis were pivotal in the expansion and capsid-opening process, while the binding site on the 5-fold axis changed little structurally. These findings suggest that the dissociation of the capsid is initiated at the 3-fold axis.
| We have studied the capsid of satellite tobacco necrosis virus using large scale molecular dynamics simulations, where the atomic motions of 1,2 million particles were tracked over one microsecond. We find that the capsid swells in the simulations, and that the permeability for water increases 10-fold upon removal of the structural calcium ions. The water leaks in predominantly near the three-fold symmetry axis, suggesting that this is the spot where capsid dissociation is initiated following infection.
| Non-enveloped icosahedral viruses often contain binding sites for divalent cations, usually . The ions are typically bound between coat proteins or on the icosahedral symmetry axes. This is broadly observed in three plant virus taxa: the family Tombusviridae (and an associate satellite virus), the genus Sobemoviruses and the family Bromoviridae [1]–[4]. Binding sites for calcium ions have also been found in bacteriophages of the Leviviridae family [5], fish and insect viruses of the Nodaviridae family [6] and in the Picornaviridae family, e.g. several human rhinoviruses [7].
In many of the plant viruses it is possible to induce a conformational change in vitro by removing the ions, either by a chelating agent such as ethylenediaminetetraacetic acid (EDTA) or by exhaustive dialysis against deionized water. Ion-deprived virions reversibly expand on the order of 5–10% at neutral or slightly alkaline pH. In the swollen state internal parts of the virion as well as the RNA molecule may become susceptible to degrading enzymes [8], [9]. Chelation of the metal ions is also required for synthesis of virus proteins in cell-free translation systems [9]. Only two low-resolution crystal structures of expanded virons are available: tobacco bushy stunt virus (TBSV) at 8 Å [10] and satellite tobacco necrosis virus (STNV) at 7.5 Å [11]. The radial increases are about 11% and 4%, respectively. In addition, an expanded cowpea chlorotic mottle virus (CCMV) virion was imaged with cryo-electron microscopy at 29 Å and interpreted using rigid body fitting of the high-resolution structures of the native proteins [4]. The dynamic nature of the swelling process as well as the limited resolution of swollen virus particles structures prompted us to perform a simulation study of the capsid of STNV, with and without bound , over one microsecond. The simulations allowed us to reproduce the swelling behavior upon removal of the calcium in silico and develop an atomistic description of the process.
The T = 1 capsid of STNV consists of 60 identical coat proteins with one protein per icosahedral asymmetric unit. The coat protein is 195 amino acid residues long where residues 25–195 make up the main domain that constitutes the capsid shell. The virions readily crystallize and the major part of the coat protein has been resolved by X-ray crystallography [2], [12]–[14]. The shell domain at the C-terminus folds as a -jelly roll similar to many other single-stranded RNA plant viruses. Residues 12–24 form a helical structure that together with the helices of two neighboring subunits form a short stalk that projects inwards into the central cavity around the icosahedral 3-fold axis. The first 11 residues at the N-terminus are disordered and cannot be detected in the electron density maps – in the simulations these residues were modeled as a helix as well. This N-terminal arm and the interior surface of the capsid are lined with positively charged residues that presumably interact with the single-stranded positive-sense RNA molecule [14]. The 1239 nucleotide long genome encompasses only one open reading frame that encodes the coat protein and hence STNV is dependent on the co-infection of a helper virus (tobacco necrosis virus) for copying its RNA genome.
The capsid has three different types of binding sites (Figure 1). Type I is between two subunits close to the 3-fold symmetry axis. The protein ligands are the carboxyl groups of Asp194 and Glu25 as well as the main chain carbonyl oxygens of Ser61 and Gln64. Type II is on the 3-fold symmetry axis 8.05 nm from the center of the virion. It is coordinated by the carboxyl groups of three Asp55 residues. Type III is on the 5-fold symmetry axis 9.04 nm from the center. This is coordinated by the main chain carbonyl oxygen of five Thr138 residues. In total the capsid can accommodate 92 ions (60 at type I sites, 20 at type II sites and 12 at type III sites).
Simulations were performed of the capsid with and without at two different salt concentrations for one microsecond each (Table 1). The carboxyl groups of one of the three Asp55 residues at each of the type II calcium binding sites were protonated in the two simulations without , effectively simulating the capsid at a slightly acidic pH to mimic the conditions of the expanded capsid in the 7.5 Å crystal structure [11]. The RNA molecule was not included in our simulations since it cannot be modeled completely in the electron density maps [14], [15]. The aim of this work was to probe the dynamic behavior of a virus capsid over timescales that are more than an order of magnitude longer than what has been reported from simulations of viruses previously [16]–[18], and therewith to investigate the role of the structural calcium ions in initiating the dissolution of the satellite tobacco necrosis virus. We particularly looked into the structural features facilitating breaking up of the capsid associated with virus infection.
The capsids were stable and remained intact throughout all four trajectories. Removing the had a pronouced effect on the capsid radius. An increase in the radius of gyration () of 2.6% in and of 2.4% in was found over the course of 1 . The two trajectories with bound ( and ) showed a weak tendency to increase in size, 0.78% and 0.58% respectively (Table 1 and Figure 2).
The capsids retained their overall spherical shape with just some minor degree of elongation. Figures 3A and 3C emphasize the local anisotropy in the structural changes. The largest changes occurred in the parts of the shell close to the icosahedral 3-fold axes. This area accommodates four calcium binding sites and all of them have charged carboxyl groups as ligands. The charge repulsion induced the formation of small water-filled cavities between the proteins at this protein/protein interface (Figure 4D). An analysis where the root mean square deviation (RMSd) from the crystal structure for protein dimers, trimers and pentamers was computed is presented in Table 1. The trimers have clearly larger than average RMSd, whereas dimers and pentamers form very stable complexes. This effect is more pronounced in the simulations without .
The secondary structure elements and the overall fold of all the coat proteins were stable throughout the simulations. The N-terminal arm was the most flexible element (Figures 1, 3 and 5A). The -helix in the N-terminal arm observed in the crystal structure was stable: at the end of the simulation the number of -helical residues in the N-terminal arm was close to 11, slightly lower in the two trajectories without bound than the simulations with bound (Table 1). Residues 1–11 did not show any propensity to stay in a helical conformation. These residues were modeled as a helix in the starting structure, but they progressively lost that structure. This might be a result of the absence of the RNA molecule since the addition of molecules that mimic the phosphate backbone of RNA has been shown to promote formation of helices in the positively charged N-terminus of CCMV, that presumably plays a similar RNA-binding role [19], [20]. The number of intermolecular protein-protein hydrogen bonds was slightly lower in the STNV capsids without , in particular the number of hydrogen bonds between pairs of 2-fold and 3-fold related subunits decreased (Table 1).
The atoms in the shell domain moved on average 0.27 nm in , 0.26 nm in , 0.15 in and 0.14 in predominantly due to the overall radial expansion of the capsid (Figure 5A). The RMSd after fitting each protein to the crystal structure individually was small (0.1 nm) apart from the termini and some of the loop regions (Figure 5B). The flexibility was highest in the two termini, in the loops between secondary structure elements and in the short helix centered at residue Thr119 (Figure 5C). The – loop (using the same nomenclature as in [2]) centered at residue Thr80 and the – loop centered at residue Leu180 show high flexibility both with and without bound (Figure 5C). Both of these loops are located close to the 5-fold axis, facing the exterior of the capsid (Figure 1). The – loop at residue Asp55, the – loop at residue Thr160 and the C-terminus on the other hand show an increased flexibility mainly in the two capsids that had no (Figure 5C). These elements are all located close to the 3-fold axis and Asp55 is the ligand for the type II sites.
When the ions were removed from the capsid, the type III calcium binding sites on the icosahedral 5-fold symmetry axis were populated by either ( only) or water. had 7 out of 12 type III binding sites occupied by sodium ions throughout the simulation. The remaining type III sites as well as all the type III sites in the capsid of were occupied by one or two water molecules. The sodium ions were bound at the same position as the calcium ions – at the geometric center of the five Thr138 carbonyl groups – but half of the time, one of the five carbonyl groups pointed away from the 5-fold axis and engaged in hydrogen bonds with other water molecules. Water molecules bound to the type III sites were located on the 5-fold axis slightly exterior or interior to the binding site for cations such that they could make hydrogen bonding interactions with two or three carbonyl groups. In these cases the 5-fold symmetry of the protein subunits was broken with one (when two water molecules were bound) or two (when one water molecule was bound) carbonyl groups turned away and engaged in hydrogen bonds with water molecules from the bulk.
The two types of calcium binding sites close to the icosahedral 3-fold symmetry axis were less stable than the type III sites upon removal as reflected in the higher RMSd and RMSf of residues close to the 3-fold axis (Figure 5). In all type II sites were occupied by one or two sodium ions subsequent to the equilibration simulation (in which the protein was restrained from moving away from the crystal structure). However, at the end of the production simulation only 1 of the 20 sites was intact with all three carboxyl groups of Asp55 coordinating a sodium ion. The rest of the type II sites were broken up with the carboxyl groups facing other directions(Figure 4D). The type I site also lost its structure in the calcium-free capsids. The four residues that contributed oxygen atoms to coordinate the calcium ion had high RMSf (Figure 5C) and did not retain their relative positions. The negatively charged carboxyl side chains of all the type I sites together bound approximately 30 sodium ions in an irregular fashion.
The capsids were permeable to water in all trajectories. Assuming a homogeneous permeability across the entire surface of the capsid, an osmotic water permeability coefficient, , was calculated [21]. With bound there were about 10,000 permeation events in either direction and the capsid had an average permeability coefficient = cm/s. After removing the increased to about cm/s (Table 1). The water transport resulted in a net inflow of water in conjunction with the swelling.
In order to detect structural features associated with water permeability, the crossing events were mapped onto the virus particle. For each water molecule traversing the width of the capsid shell successfully, the closest protein residue () was determined for the water molecule half-way (in time) through, and statistics of permeation per residue were gathered. In the simulations with bound the permeation mainly occurred between the icosahedral 2-fold and 3-fold symmetry axis, at the junction between three subunits (Figures 6A–C and 4). The protein–water contacts suggest that this potential water pore is lined by five motifs: the short helix centered at residues 118–119, residues 156–158 in the – loop, the end of the – loop, the beginning of the -strand and to some degree residues 25–30 close to the flexible loop connecting the N-terminal arm to the shell domain (Figure 5D). In the simulations without bound the permeability increased at this site, as well as at the protein/protein interface at the 3-fold symmetry axis (Figures 5D, 6D–F).
If we consider the region around each icosahedral 3-fold axis to be a water pore, we can estimate the single pore permeability coefficient, , to be in the capsid with bound and without these ions (Table 1).
Many plant viruses employ the extremely low concentration of in the cytoplasm of their host cells (the homeostatic concentration of free is between 100–200 nM [22]) as a cue to initiate the replication stage of their life-cycle. The capsid will be effectively depleted of calcium as the equilibrium between occupied and unoccupied sites shifts to the latter.
Four different crystals of STNV treated with EDTA were investigated by Montelius et al., but the capsid expanded in only one of those [11] and that particular crystal diffracted to a resolution of 7.5 Å only. In our simulations we could observe a higher degree of fluctuations as well as loss of the icosahedral symmetry in the calcium-deprived capsids compared to the ones with bound calcium. Both higher flexibilty and a lower degree of symmetry would contribute to a lower crystal quality and could explain why there have not been any successful attempts to solve a high-resolution structure of an expanded capsid so far. The trajectories did not quite reach an equilibrium swollen state. The radius of the calcium-depleted capsid can be extrapolated to an “equilibrium radius” by fitting the curve (Figure 2) to = −(−). With this fit an equilibrium radius of is predicted, about a quarter of a percent higher than the value at the end of the simulation. The weak increase in the radius of the calcium-containing capsids is probably an artifact due to the lack of the RNA molecule and confirms the long-term unstable nature of the RNA-free capsid.
Swollen STNV capsids can be returned to their native radius by lowering the pH [8], suggesting that the mechanism behind the swelling is electrostatic repulsion of charged aspartate and glutamate ligands. A similar effect was deduced from electrostatics calculations of the native and expanded structures of the CCMV capsid [23]. The protonation state of the carboxyl groups is probably coupled to the magnitude of the expansion. The crystals of the expanded STNV formed at pH 6.5 and the expansion was only moderate, so the capsid may have been protonated at one of the calcium binding carboxyl groups. Since the type II site has the highest local density of negative charge, we decided to protonate one of the three carboxyl groups at each type II binding site in our simulations of the calcium-free capsid.
Analytic ultracentrifugation measurements estimate that the STNV particle can expand up to 7% when treated with EDTA [8]. The crystal structure of the expanded virion showed a radial expansion of between 0.1–0.4 nm, equivalent to 1%–5% (higher closer to the icosahedral 3-fold axis). This agrees well with what we observed in the simulations: an average increase in the of 2.5% (Table 1) and peak RMSd values of the of the shell domain above 0.35 nm at the 2-fold subunit interface, e.g. residues 96 and 122, and at the 3-fold axis, e.g. residues 25 and 195 (Figure 5A).
The osmotic water permeability of the capsid is comparable to lipid membranes, which usually have = cm/s [24]. The protein shell is thinnest at the 5-fold axis, but the calcium ion at the type III site is the one that is most difficult to chelate with EDTA [25], something which was corroborated by spectroscopic measurements showing that the type III site has a remarkably high binding affinity to the analog ( nM) [26]. In the simulations of the capsid without calcium, the coordinating cage of oxygen ligands was best preserved at the type III site. Sodium ions or water molecules replaced the divalent ion on the icosahedral 5-fold axis and obstructed the opening, resulting in minimal permeability even without . Instead, we observed extensive water permeation on the 3-fold axis and in a region close to the 3-fold axis between the 3-fold and the 2-fold axis (Figure 4). The cluster of four calcium binding sites around the 3-fold axis contains several carboxyl groups from aspartate and glutamate residues. Removing the calcium ions introduces a large amount of net negative charge that caused the subunits to move apart, creating water pockets at the 3-fold axes causing increased water permeability. If the entire region around the 3-fold axis is considered a water pore, the permeability of it is comparable to that of membrane proteins that function as water pores, e.g. mammalian aquaporins that have reported single pore permeability of [27].
The difference in size and triangulation number makes it difficult to compare the expanded structure of TBSV and STNV. The capsid of TBSV consists of 180 identical subunits in a T = 3 arrangement where each of the 60 icosahedral asymmetric units consists of three proteins in slightly different configuration. At the center of these three proteins, there is a so called quasi-3-fold axis that relates three approximately equivalent protein positions [28]. The six calcium binding sites of TBSV are located pairwise between pairs of subunits in the asymmetric unit and each site has five acidic residues from both proteins. The swollen TBSV was crystalized at pH 7.5 and the structure is about 7% larger than the native one. The most predominant structural change is that large openings appear between the quasi-3-fold and 2-fold related subunits [10]. The 3-fold axis of STNV resembles the quasi-3-fold axis of TBSV. The interfaces between the proteins around these axes contain six (TBSV) respectively four (STNV) binding sites (Figure 1), coordinated by carboxylic groups. In both capsids the largest structural differences between the expanded and native structures can be found here.
Not all viruses that have calcium binding sites in the capsid show a swelling behavior. The rhinoviruses have a calcium binding site on the 5-fold axis [7], but do not swell upon removal of these ions. Interestingly, this binding site show striking similarities to the type III calcium binding site in STNV. In both cases five backbone carbonyl oxygens point at the 5-fold axis where the ions are bound. The 5-fold axis is a region with high degree of symmetry constraints and putting an ion there solves the problem of fulfilling the symmetry at a very congested interface. We therefore propose that the carbonyl type of binding sites have a more structural role, while the carboxyl type of binding sites may play a role in the dissolution of the capsid upon infection. This would explain the low degree of structural change at the 5-five fold site in our simulation and their high propensity to bind other cations or water molecules, and this is in line with the relatively higher affinity for ions at these sites [25], [26].
There are no atomic structures of the STNV genome available, although a recently published high-resolution structure shows short fragments of RNA [14]. RNA-free virus-like particles of the STNV coat protein have not been observed so far, but the coat proteins of similar viruses readily form empty shells [29], [30]. Rather than inserting a modeled RNA structure into the capsid we decided to perform the simulations without RNA. The fact that RNA nevertheless may contribute to the stability of the capsid is evidenced from our simulations at physiological salt concentration, providing more shielding of electrostatic interactions. The radial expansion is less pronounced in this case (Figure 2) and the number of protein-protein hydrogen bonds seems to increase slightly (Table 1), a typical “salting out” effect.
Previous all-atom molecular dynamics simulations of capsids and virus-like particles focused on specific properties like the mechanical strength using force probes [17], [18] or the effect of hydration on coherent diffraction from single virus particles [31]. It has become clear though that virus simulations are sensitive to the starting conditions, like a balanced amount of water on the inside, and they require long equilibration times [18], [32]. Therefore we paid special attention to the preparation of the starting structures with an extensive equilibration protocol with part of the counter ions specifically added to the interior cavity to avoid a sudden influx of solvent ions that could disrupt the protein-protein interactions and by carefully balancing of the hydrostatic pressure inside and outside of the capsid. By doing control simulations with and be varying the ionic strength we can draw firm conclusions on the effect of removal on STNV.
The simulations presented here illustrate the mechanism by which an entire virus capsid can transform from a closed configuration into an open one with significantly increased water permeability. The magnitude of the expansion in the simulations is in good agreement with experiments. The higher flexibility and the degradation of the icosahedral symmetry in combination explain why it has been difficult to crystallize expanded capsids. Our work strongly suggests that there are two types of calcium binding sites, playing different roles in the virus lifecycle. The binding site on the 5-fold axis has a more structural role and is less involved in the capsid expansion. The binding sites between 2- and 3-fold related subunits contain many charged carboxylic side chains. Dissolution of the capsid is initiated here due to the electrostatic repulsion between these residues, if the ions are removed.
All preparations, simulations and analyses were performed with the GROMACS simulation package version 4.5.3 [33] unless otherwise stated.
The starting structures were prepared from the X-ray crystal structure of the coat protein of STNV (PDB ID: 2buk) [2]. Since residue 1–11 can not be resolved in the crystal structure, these were modeled as an -helix in a direction that did not cause steric clashes with neighboring proteins. The full capsid was generated by applying the icosahedral rotation-translation matrices in the PDB file. One ion was kept at each of the 92 calcium binding sites. The Amber99sb-ILDN forcefield [34], [35] was used for the protein combined with TIP3P water [36] in a rhombic dodecahedron simulation box with a side of about 25 nm. Water molecules on the inside of the capsid were replaced with ions to obtain a neutral system (). Additional and ions were added to obtain a system with an approximately physiological ion concentration of 150 mM (). The calcium ions were removed, a proton was added to one Asp55 side chain at each of the type II calcium bindings sites and the number of counter ions was adjusted to obtain two neutral calcium-free systems ( and respectively). Each system consisted of roughly 1.2 million atoms (Table 1).
The starting structures were energy minimized and subsequently the solvent was equilibrated for 10 ns while restraining the protein atoms and the calcium ions to the crystal coordinates with harmonic potentials with a force constant of 1000 . During the equilibration a 2 fs integration timestep was used and the neighbor lists were updated every timestep. Short range non-bonded Van der Waals (Lennard-Jones) and Coulomb interactions were calculated within a cut-off radius of 1.15 nm. The long range electrostatic interactions were calculated with the particle mesh Ewald (PME) method [37] with a grid spacing of 0.133 nm. The long range Lennard-Jones interactions were analytically corrected for in the calculation of the pressure and the energy. The pressure of the simulation box was kept at an average of 1 bar using the isotropic Berendsen barostat [38] with a time constant of 25 ps and a compressibility of . The solvent and the capsid were coupled separately to an external heat bath at 300 K with the velocity-rescaling thermostat [39] using a time constant of 0.5 ps. Water molecules were constrained using the SETTLE algorithm [40] and the covalent bonds in the proteins were constrained using the P-LINCS algorithm [33]. Boundaries were treated periodically.
After the equilibration, the position restraints were removed and an integration time step of 4 fs was used to generate 1 trajectories. In addition to constraining bond lengths, virtual hydrogen atoms were used [41] which allows slightly longer time steps. The isotropic Parrinello-Rahman barostat [42], [43] was used to keep the average pressure at 1 bar with a time constant of 1 ps. The calcium ions were tethered to the oxygen ligands using harmonic potentials with force constants of 5000 . All other simulation parameters were the same as during the equilibration. The trajectories were sampled every 50 ps for analysis. The production simulations were calculated in parallel on a Cray XE6 system over 2016 cores (612 of which for PME calculations) at a speed of 30 ns/day.
Unless otherwise stated, all trajectories were analyzed at 1 ns intervals and final values were calculated as the average over the last 100 ns. Root-mean-square displacement (RMSd) and fluctuation (RMSf) was calculated as:(1)(2)
The number of -helical residues was calculated using the g_helix program of the GROMACS package. In the water permeability analysis water molecules were counted when they had traversed the entire width of the capsid shell. Visual inspection did not imply a single-file type of permeation mechanism, which justifies that the distinction between a diffusive and osmotic permeation coefficients was not required [21]. The osmotic water permeability coefficient, , and the single pore osmotic water permeability coefficient, , was calculated as the average permeability in both directions using these formulas:(3)(4)
Where N is the number of permeation events, t is the duration, A is the area and is the concentration gradient, i.e. 55 mol/L for pure water. The capsid was approximated to have the same area as a sphere with radius 8 nm.
|
10.1371/journal.pgen.1007883 | CDI/CDS system-encoding genes of Burkholderia thailandensis are located in a mobile genetic element that defines a new class of transposon | Intercellular communication and self-recognition are critical for coordinating cooperative and competitive behaviors during sociomicrobiological community development. Contact-dependent growth inhibition (CDI) proteins are polymorphic toxin delivery systems that inhibit the growth of non-self neighboring bacteria that lack the appropriate immunity protein. In Burkholderia thailandensis, CDI system proteins (encoded by bcpAIOB genes) also induce cooperative behaviors among sibling (self) cells, a phenomenon called contact-dependent signaling (CDS). Here we describe a mobile genetic element (MGE) that carries the bcpAIOB genes in B. thailandensis E264. It is a ~210 kb composite transposon with insertion sequence (IS) elements at each end. Although the ISs are most similar to IS2 of Escherichia coli, the transposase-dependent intermediate molecule displays characteristics more similar to those of the IS26 translocatable unit (TU). A reaction requiring only the “left” IS-encoded transposase results in formation of an extrachromosomal circular dsDNA intermediate (“the megacircle”) composed of the left IS and the sequences intervening between the ISs. Insertion of the megacircle into the chromosome occurs next to a pre-existing copy of an IS2-like element, recreating a functional composite transposon. We found that BcpA activity is required for megacircle formation, and in turn, megacircle formation is required for CDS phenotypes. Our data support a model in which the bcpAIOB genes function as both helping and harming greenbeard genes, simultaneously enhancing the fitness of self bacteria that possess the same allele plus tightly linked genes that mediate cooperative behaviors, and killing non-self bacteria that do not possess the same bcpAIOB allele. Mobility of the megacircle between cells could allow bacteria invading a community to be converted to self, and would facilitate propagation of the bcpAIOB genes in the event that the invading strain is capable of overtaking the resident community.
| As social organisms, bacteria have evolved multiple ways to communicate and interact with their neighbors. Some of these interactions can be beneficial or harmful to certain members of the community, and others involve sharing of genetic material capable of transforming the recipient cell. In this study, we provide evidence for a mobile genetic element that carries the genes encoding proteins involved in bacterial killing (contact-dependent inhibition, CDI) or cooperation (contact-dependent signaling, CDS) within microbial communities. Our findings suggest the element mobilizes with a copy-out-paste-in mechanism that requires formation of a large circular DNA molecule we call “the megacircle”. We also show that production of the megacircle requires a functional CDI/CDS system and that synthesis of the megacircle is necessary for cooperation-associated phenotypes. We hypothesize that acquisition of the megacircle provides a means to transform a target cell that does not produce the same CDI/CDS system into one that is immune to inhibition via CDI, and that can participate in the cooperative behaviors of the community.
| Bacteria typically live in complex, dynamic, polymicrobial communities, and hence have evolved mechanisms to cooperate and compete with neighboring microbes to ensure efficient resource utilization and community survival [1–4]. Competitive interactions within communities are especially influential because of their contributions to evolution and genetic diversification [5]. One type of interbacterial competitive interaction is mediated by contact-dependent growth inhibition (CDI) systems [6]. CDI systems are composed of two-partner secretion (TPS) pathway proteins and are widespread among Gram-negative bacteria [6–8]. They fall into two main classes, Burkholderia-type, which are encoded by bcpAIOB genes, and Escherichia coli-type, which are encoded by cdiBAI genes [8]. The bcpB/cdiB genes encode the TpsB family outer membrane channel proteins, BcpB or CdiB, that translocate the large TpsA family exoproteins, BcpA or CdiA, to the cell surface. Delivery of the C-terminal toxin domain of BcpA or CdiA to a neighboring bacterium upon cell-cell contact results in growth inhibition or death, unless the recipient cell produces the correct BcpI or CdiI immunity protein [8–10].
A hallmark of CDI systems is their polymorphic nature. The N-terminal ~2,800 amino acids (aa) of BcpA/CdiA proteins are highly conserved, while the C-terminal ~300 aa (referred to as BcpA-CT or CdiA-CT) are variable. Distinct motifs, Nx(E/Q)LYN in BcpA and VENN in CdiA, separate the conserved and variable regions. The aa sequence of the BcpI and CdiI proteins are also polymorphic, and co-vary with BcpA-CT and CdiA-CT, respectively. Several BcpA-CT and CdiA-CT have demonstrated DNase or tRNase activity [9–12], and BcpI and CdiI proteins protect from such activity by binding to cognate (encoded by the same allele) but not non-cognate (encoded by a different allele) BcpA-CT or CdiA-CT [8–10]. Because CDI systems distinguish “self” from “non-self” neighbors based on a single allele, they have been implicated in kind selection, also known as “the greenbeard effect”. Kind selection provides a mechanism for indirect fitness in which a gene encoding a cooperative or altruistic behavior, or one that is closely linked, encodes a recognizable trait (e.g., a green beard), allowing individuals carrying the same allele to be recognized directly, irrespective of genealogy [13–16]. bcpAIOB and cdiBAI genes have been hypothesized to function as “harming greenbeard genes”; they encode proteins that cause harm to individuals that do not possess the same allele, thereby providing a fitness advantage to individuals that do possess the same allele.
Several recent studies have investigated the mechanism by which BcpA-CT and CdiA-CT are delivered to the cytoplasm of target bacteria. In E. coli, a region near the center of the CdiA protein binds to either BamA or a hetero-oligomeric complex of OmpF and OmpC, depending on the specific cdiA allele, to mediate delivery of CdiA-CT into the periplasm, and then the N-terminal half of CdiA-CT mediates translocation across the cytoplasmic membrane by interacting with a specific integral cytoplasmic membrane protein, also in an allele-specific manner [17–20]. For some CdiA proteins, another layer of specificity exists in that catalytic (toxic) activity requires an accessory protein produced by the target cell [21–23]. These data demonstrate that specificity extends beyond the interaction between BcpA-CT/CdiA-CT and BcpI/CdiI, and indicate that the only cells that are susceptible to CDI may be those that are so closely related that they also contain the same cdiBAI or bcpAIOB allele [14,24]. These observations raise the question of whether interbacterial competition is the true, or main, function of bcpAIOB- and cdiBAI-encoded proteins in nature.
We have shown that in addition to mediating competitive interactions, CDI system proteins in Burkholderia thailandensis E264 (BtE264) induce cooperative behaviors, such as biofilm formation [25]. Other phenotypes that require BcpA catalytic activity and BcpI (to, at least, prevent BcpA-CT-mediated toxicity) include production of polysaccharides that bind Congo Red (CR) dye, production of a yellow-gold color to colony biofilms that we postulate reflects production of an unidentified pigment, and aggregation of cells at the air-liquid interphase when grown in defined medium [26]. We refer to BcpA-dependent changes in gene expression resulting in biofilm formation, CR binding, pigment production, and perhaps other community behaviors as contact-dependent signaling (CDS, [26]). We hypothesize that the BcpA-CT that is delivered to a recipient cell forms a complex with BcpI, and that this complex somehow causes a change in gene expression, perhaps by binding to regulatory sites in the chromosome, catalyzing limited nicking of the chromosome, or changing the concentration of second messengers such as c-di-GMP or cAMP [14,26]. We propose that by inducing cooperative behaviors among bacteria that possess the same allele, bcpAIOB genes function as “helping greenbeard genes”.
A characteristic of greenbeard genes is linkage disequilibrium between the gene encoding the recognizable trait and the gene(s) encoding cooperative or altruistic behavior [15,16]. In bacteria, genes located on the same mobile genetic element (MGE), such as a bacteriophage, a plasmid, or a genomic island, display features of linkage disequilibrium in that they move together via transduction or transformation from one cell to another. Here, we report the serendipitous discovery of a genetic element containing the bcpAIOB genes in BtE264, and provide evidence that this element is currently mobile and defines a new class of transposon.
Using next generation sequencing (NGS) technology, we performed a re-sequencing analysis of the complete wild-type (WT) BtE264 genome. The analysis yielded a coverage graph showing how many times non-gap characters aligned to each nucleotide in the reference sequence (Fig 1A). We observed a region in chromosome I for which a high number of sequencing reads were aligned. This region has a mean coverage of 605, while the rest of chromosome I and chromosome II had mean coverages of 227.3 and 218.3, respectively. The high-coverage region spans 209,962 bps and includes 161 predicted open reading frames (S1 Table), including the bcpAIOB operon encoding the BcpAIOB CDI/CDS system. Genes annotated as insertion sequence (IS) elements are present at both ends of the region with high coverage. The “α end” (Fig 1B) contains two distinct genes annotated as ISBma1 transposable elements and are predicted to encode ISL3 family transposases. Although both genes were given the same annotation, they are dissimilar. To distinguish them, we have added a letter to the gene name and refer to them as ISBma1a (BTH_I2583) and ISBma1b (BTH_I2586). BTH_I2584 and BTH_I2585 (orfBα and orfAα) are overlapping genes predicted to encode a single transposase with similarity to the one required for transposition of IS2 in E. coli (S1 Fig, [27–32]). The “β end” of the 210 kb region also contains overlapping genes, BTH_I2744 and BTH_I2745. BTH_I2745 is identical to BTH_I2585 (orfAα), and BTH_I2744 is identical to BTH_I2584 (orfBα) except for a single silent nucleotide variation near the 3’ end. IS2 elements belong to the large IS3 family of IS elements. A common feature of this family is a programmed -1 translational frameshift that results in production of the OrfAB fusion protein, which is the functional transposase that mediates mobilization of the element [32,33]. Similar to the IS2 from E. coli, the BtE264 IS2-like transposase-encoding genes are flanked by imperfect inverted repeats, with the left (IRL) and right (IRR) repeat located 5’ to orfA and 3’ to orfB, respectively (Fig 1B and 1C; [32]). There are four additional IS2-like elements in the BtE264 genome, all of which are identical to IS2β (Fig 1A and S2 Fig). All six IS2-like elements in BtE264 are flanked by 5 bp target repeats, likely generated during integration of the element, a characteristic also observed for E. coli IS2. We did not observe increased coverage of sequences flanking any of the IS elements in our NGS analyses other than IS2α and IS2β.
Increased coverage of a contiguous region in NGS analyses could result from a duplication of the sequence compared to the reference genome. However, we have constructed several strains that contain mutations within the ~210 kb region, such as disruption of the csu Type-4 pilus-encoding operon (BTH_I2681-I2674; [26]) and deletions within the bcpAIOB locus [8]. In all cases, PCR analyses indicate that the mutations were constructed as intended, insertion borders are as expected or deleted DNA is undetected, and the mutant strains are stable and display reproducible phenotypes. These results are inconsistent with the presence of multiple, tandem copies of the ~210 kb region in the chromosome.
Alternatively, the transposition mechanism of IS2 elements could explain the presence of multiple copies of the ~210 kb region. IS2 and other IS3 family members utilize a “copy-out-paste-in” mechanism that involves formation of a circular double-stranded DNA structure [31,33,34]. The intermediate, often called a minicircle, is essentially a circularized version of a single IS2 element that is capable of inserting itself into a new location. To determine whether high coverage of the ~210 kb region resulted from the production of a double-stranded circular molecule, we performed PCR analyses using primers that anneal near the ends of the ~210 kb region (Circ1 and Circ2, Fig 2A), such that they would amplify a ~2.5 kb product spanning the junction of the circularized element. Only one PCR product, 2.2 kb in size, was generated from WT BtE264 cells (Fig 2B), and DNA sequence analysis of several independent PCR products indicated the junction contains ISBma1b and one copy of orfA and orfB (more specifically, orfBβ based on the single nucleotide variant; Fig 2C). These data are consistent with the high coverage region corresponding to a ~210 kb “megacircle” that is formed by a reaction involving the sequence between the single nucleotide variation in orfB and the IRL of IS2β, most likely within the IRL of IS2α (since transposases typically bind to, and catalyze recombination within, inverted repeats; Fig 2D; [32,33,35,36]). The junction of the megacircle is different from those of E. coli IS2 mini-circles where the inverted repeats are joined together and separated by a one- or two-base spacer [31,37], suggesting that the chemistry for the reaction involving the IS2-like elements in BtE264 is different than what has been described for E. coli IS2 elements. Multiple attempts to detect an IS2 mini-circle junction in BtE264 were unsuccessful, providing additional evidence for a distinct mechanism. Production of a circular DNA intermediate from two distant IS2 elements has not been reported. These data suggest that the BtE264 bcpAIOB-containing element may represent a previously undescribed IS2-containing composite transposon and mobile genetic element (MGE).
Many transposases, including IS3 family members, have a preference for acting on the element from which they were expressed (called cis activity), resulting from binding of the N-terminal domain of the nascent enzyme to its target sequence as it emerges from the ribosome [33,38]. The ends of the bcpAIOB-containing composite transposon are over 200 kb apart, yet they come together to form the megacircle. To determine the contribution of each IS2-like element to megacircle synthesis, we constructed mutant strains in which orfAB from IS2α or IS2β was replaced by double recombination with nptII, encoding kanamycin resistance. The inverted repeats, where transposase-mediated recombination is expected to occur, were left intact. The megacircle junction was detected by PCR in the IS2α::nptII strain, but not in the IS2β::nptII strain, indicating that the megacircle-forming reaction is a transposase-dependent event that requires the transposase encoded by IS2β, but not that encoded by IS2α (Fig 3A).
Lack of detection of a PCR product with primers Circ1 and Circ2 in the IS2β::nptII mutant provides additional evidence that the increased NGS coverage is due to the presence of extrachromosomal megacircle molecules and not tandem copies of the 210 kb region. If tandem repeats of the element were present in the genome (approximately 3 copies based on the NGS coverage), then they would each be separated by one copy of IS2β, based on our sequence analysis of the PCR product generated by Circ1 and Circ2 (Fig 3B). Mutation of the IS2β at the junction with the rest of the chromosome (the one farthest to the right in Fig 3B) would not abrogate amplification of PCR products with primers Circ1 and Circ2 as those junctions would still be present between the tandem copies (Fig 3B, bottom panel). Together, therefore, our results provide strong evidence that the bcpAIOB operon is located within a transposable element that forms an extrachromosomal circular intermediate.
Our data suggest that, despite lack of similarity between the transposases, the IS2-dependent megacircle bears mechanistic similarity with IS26-containing composite transposons. In its stationary form, a typical IS26 composite transposon is composed of two IS26 elements positioned in direct orientation and flanking passenger DNA that contains genes encoding antibiotic resistance [39–41]. IS26 does not appear to transpose as a single IS element [42], instead, it mobilizes via a circular molecule called a translocatable unit (TU). Formation of the TU is mediated by the transposase encoded by the “left” IS26 (with the IS elements oriented such that the transposases are encoded left to right), and the TU is composed of the “left” IS26 element and passenger DNA [42–44]. Thus, both the BtE264 IS2-like transposase and the IS26 transposase appear to catalyze reactions that involve distantly-located ISs to create circular intermediates that contain only the “left” IS element and the DNA intervening between the ISs (Fig 2D).
In addition to WT BtE264, our NGS studies included the ΔbcpAIOB mutant, and for this strain increased coverage in chromosome I was not observed. PCR with Circ1 and Circ2 failed to produce a product from this strain, as well as from the BcpAEKAA strain, which produces a catalytically inactive BcpA protein (Fig 4). The ΔbcpAIOB and BcpAEKAA mutants are defective for both CDI and CDS. We showed previously that a chimeric BcpA protein, in which the conserved region from the BtE264 allele is fused to the variable catalytic BcpA-CT region from an allele present in B. pseudomallei 1106a, can mediate inhibition of neighboring susceptible cells via CDI, but cannot mediate CDS [24,26]. The megacircle junction was not detected by PCR with Circ1 and Circ2 in the strain producing the chimeric BcpA protein (Bt-Bp, Fig 4). These data indicate a positive correlation between megacircle formation and CDS phenotypes.
A possible explanation for the link between the megacircle and CDS phenotypes is that circularization of the element is a result of interbacterial signaling that leads to changes in gene expression in a cell that has received a BcpA-CT from a neighboring cell, i.e., megacircle formation is a newly-identified CDS phenotype (Fig 5A, left hypothesis). Alternatively, the CDS phenotypes could be a consequence of megacircle production via an unknown mechanism (Fig 5A, right hypothesis). Characterization of the strain in which IS2β was replaced with nptII showed that these bacteria have a functional BcpAIOB system that can mediate CDI (it outcompeted the ΔbcpAIOB strain as well as wild-type bacteria, Fig 5B), but that cannot mediate CDS (it did not aggregate in minimal media, bind Congo red or produce the yellow/brown pigment, Fig 5C). Our data indicate, therefore, that BcpA activity and IS2β are both necessary for formation of the megacircle, and that the extrachromosomal megacircle molecule is somehow required for CDS phenotypes (Fig 5A, right hypothesis).
Our data indicate that the bcpAIOB-containing element forms extrachromosomal circular DNA molecules in an IS2β-dependent manner and that these are conceptually similar to the TUs formed by IS26. Our initial approach to determine if the megacircle functions as a TU was to determine if it could be transferred between cells, and we reasoned that the optimum recipient strain for testing this hypothesis would be one lacking the entire ~210 kb element. Although the element contains several genes that encode proteins that are expected to be necessary for viability (S1 Table), the essentiality of those genes has not been tested in BtE264. We therefore set out to delete the element (simultaneously testing the essentiality of genes within it), and we began by replacing the ~48 kb region encompassing genes BTH_I2587-I2630 (which we refer to as Region 1, or Reg 1), with an nptII-containing cassette, taking advantage of the fact that B. thailandensis is naturally competent and proficient at homologous recombination (Fig 6A). PCR analyses of the resulting kanamycin-resistant transformants with primers P1 and P2, which bind to BTH_I2585 (orfAα) and BTH_I2631, respectively (Fig 6A), yielded the expected ~3 kb DNA fragment (Fig 6B). However, PCR analyses to amplify genes BTH_I2604-2605, located within Reg1, yielded a DNA fragment that was identical in size to that amplified in WT Bt264 (Fig 6C), indicating the presence of those genes. The simplest explanation for this result is that the bcpAIOB-containing megacircle is, in fact, a translocatable unit that integrated into a new site in the chromosome (i.e., it had mobilized intracellularly) during the construction of the Reg1::nptII mutant because one or more genes within Reg1 are, in fact, essential.
To test the hypothesis that the megacircle had integrated at a new site, we used plasmid rescue to determine its new location by identifying the DNA sequence adjacent to BTH_I2587 (Fig 1B). Briefly, we constructed a suicide plasmid containing a 500 bp DNA fragment corresponding to the 5′ end of BTH_I2587 (a sequence unique to the mobilized IS2 element), introduced the plasmid into the Reg1::nptII mutant and selected TMP-resistant cointegrants. Genomic DNA isolated from two independent cointegrants was subjected to restriction digestion, then the gDNA fragments were ligated and transformed into E. coli, followed by selection of TMP-resistant transformants. The plasmids recovered contained a ~28 kb sequence identical to the “β end” of the bcpAIOB-containing element (Figs 1B and 6D), indicating that the megacircle had integrated in tandem to the Reg1::nptII-containing element during deletion of Reg1. This result demonstrates the ability of the megacircle to integrate into the chromosome, indicating that the megacircle is a TU and that the bcpAIOB-containing element is a newly identified composite transposon and mobile genetic element.
The Reg1::nptII merodiploid strain provided an opportunity to obtain additional evidence that the 210 kb bcpAIOB-containing region forms an extrachromosomal circular DNA molecule. We removed the nptII gene from the Reg1::nptII strain by Flp recombinase mediated recombination to create Reg1-KanS, which was then incubated with a linear DNA molecule containing the nptII gene flanked by 500 bp sequences corresponding to those 5′ and 3′ to bcpAIOB. PCR analyses indicated that one copy of bcpAIOB was replaced with nptII in the resulting KmR transformants, while one copy of bcpAIOB remained intact (S4 Fig). Only when the second copy of bcpAIOB was replaced with nptII (after removal of nptII by Flp recombinase from the site of the first replacement), were the bcpAIOB genes undetectable by PCR (S4 Fig). By contrast, when wild-type BtE264 was incubated with the same linear DNA molecule, only primers corresponding to replacement of bcpAIOB with nptII yielded a PCR product–the bcpAIOB genes were undetectable in this strain (S4 Fig). Thus, despite wild-type BtE264 containing, on average, about three copies of the 210 kb region, only one copy (the chromosomal copy) is stably maintained in the cell and susceptible to mutagenesis that is heritably maintained. These data provide further support for the 210 kb bcpAIOB-containing region existing as an extrachromosomal circular molecule. Moreover, they support the prediction, based on the absence of an identifiable ori, that the megacircle is incapable of replication, explaining why it is possible to construct strains with mutations within the element by allelic exchange, despite the presence of multiple copies of target genes within the cell at any given time; although recombination with the megacircle may occur, such recombinants cannot be selected as the megacircle is essentially a suicide plasmid.
The efficiency of obtaining kanamycin-resistant transformants upon exposure of BtE264 cells to DNA that replaces Reg1 with nptII can be used to measure megacircle-dependent mobilization of the element. In WT BtE264, transformation efficiency for the introduction of the Reg1::nptII mutation was 4.61 x10-9 (Fig 6E). By contrast, for the mutant lacking orfABβ and the strain producing the Bt-Bp BcpA chimera (which do not produce megacircles, Figs 3A and 4), the number of kanamycin-resistant colonies obtained was either zero or not significantly different from the number obtained when no DNA was included (note that the average limit of detection for this assay is 1.49 x 10−9). To rule out the possibility that the low transformation efficiency observed with the megacircle-deficient strains was caused by a defect in competence, we introduced a cassette carrying nptII by natural transformation using DNA fragments with homology corresponding to regions inside or outside the ~210 kb MGE. A similar number of transformants was obtained for these cassettes in all strains (Fig 6E), indicating that competence, or downstream processes that allow natural transformation of cells, are not affected in megacircle-deficient strains. Together, these data strongly support the conclusion that translocation of the bcpAIOB-containing MGE occurs via the megacircle.
We continued to delete the rest of the sequences within the bcpAIOB-containing MGE despite the presence of a tandem copy. The final step used regions of homology corresponding to genes outside the element, BTH_I2582 and BTH_I2746 (Fig 1B). PCR analyses of the resulting kanamycin-resistant strain (131–10) with primers P5 and P6 confirmed that nucleotides 2,945,740 to 3,156,451 had been replaced with the nptII cassette (Fig 7A and 7B). However, as we now expected, genes within the MGE were still present, as evident from PCR amplification of DNA fragments from different regions of the element (Fig 7B and S4 Fig). To determine the location of the element in this strain, we again used a plasmid rescue approach, this time with the suicide plasmid containing DNA corresponding to the 3′ end of BTH_I2743. Sequence analysis of the recovered plasmids indicated that the adjacent DNA corresponded to the 5′ end of BTH_II0368 on chromosome II, the gene adjacent to IS2β5 (BTH_II0366-0367; Fig 1A and 7C and S2 Fig). PCR analysis of strain 131–10 with primers that anneal to regions adjacent to BTH_II0365 and BTH_II0368, paired with primers Circ1 and Circ2, respectively, confirmed the MGE translocation (Fig 7C). Furthermore, the element was functional from its new location as the megacircle was detectable in strain 131–10 by PCR, and CDS phenotypes were produced (S5 Fig). These data, showing that the megacircle can integrate into a different chromosome, provide additional proof that the bcpAIOB-containing element is a composite transposon that is mobile via an extrachromosomal translocatable unit. PCR analyses of WT BtE264 with primers flanking IS2β2, IS2β3, and IS2β4 indicate that these regions are as indicated by the reference genome sequence, corroborating our findings that multiple copies of the ~210 kb region are due to extrachromosomal DNA molecules and not to an MGE integrated next to an IS2-like element (S6 Fig).
The 5-bp target duplication characteristic of mobilized IS2 elements was not observed in the plasmids recovered from strains Reg1::nptII and 131–10. Integration of the megacircle next to an existing IS2 (IS2β in strain Reg1::nptII and IS2β5 in strain 131–10) could result from homologous recombination, and we cannot rule out this possibility because introduction of the Reg1::nptII mutation, a process that requires a functional recombination machinery, has been our method of selection for bacteria with a mobilized megacircle. However, taking into account the mechanistic similarities between the IS26 TU and the megacircle, it is plausible that integration was transposase mediated. Frequency of transposase-dependent IS26 TU integration via the targeted conservative mechanism (which requires a pre-existing IS26 element at the target site and does not generate a target duplication) is much higher than random transposase-dependent integration (which results in duplication of the element and generation of the target duplication) or Rec-dependent cointegrant formation [42,44,45].
In this study, we identified a ~210 kb mobile genetic element within chromosome I of B. thailandensis E264 (BtE264) that contains the bcpAIOB genes, has IS2-like ISs at each end and defines a previously unknown class of IS2-containing composite transposon. Our data indicate that the transposon moves by a copy-out-paste-in mechanism that utilizes a double-stranded circular DNA intermediate, which we refer to as the megacircle. Only the “left” IS2-like element (IS2β) is required for megacircle formation, and only IS2β is present in the megacircle. We also showed that mobilization of the transposon to a new location within the BtE264 genome occurred next to a pre-existing IS2-like element, thus recreating the composite transposon architecture. Our data also show that megacircle formation is required for CDS phenotypes, and that in addition to IS2β, creation of the megacircle requires BcpA activity.
IS elements, the simplest transposable units in bacterial genomes, are composed of one or two transposase-encoding genes flanked by inverted repeats that serve as transposase binding sites [33]. Mobility of IS2 in E. coli occurs through a copy-out-paste-in mechanism consisting of production and integration of a double-stranded DNA (dsDNA) minicircle intermediate [31,37,46–48]. During IS2 minicircle biogenesis (the copy-out step), an active OrfAB transposase binds and cleaves the IS to generate the characteristic figure-eight structure that becomes the IS2 minicircle upon DNA replication and repair. The second step of transposition, integration of the minicircle at a target site (the paste-in step), begins with increased orfAB expression from a strong promoter formed by the abutted end repeats and spacer located in the minicircle junction. It ends with cis activity of the transposase that results in cleavage of the abutted ends to generate a reactive linear IS that can integrate into a dsDNA target site [31,49]. No definitive insertion sequence specificity has been identified for E. coli IS2, however, integration of the minicircle is not random, as it occurs in regions where the host DNA structure shifts due to abrupt changes in GC skew [50].
The predicted amino acid sequence of OrfAB encoded by the IS elements flanking the bcpAIOB-containing MGE in BtE264 is highly similar to that of IS2 OrfAB from E. coli, especially within the predicted DNA binding and catalytic domains (S1 Fig) [47]. However, we have obtained no evidence that any of the IS2-like elements in BtE264 function independently as an IS. Instead, our data indicate that IS2α and IS2β function together as a composite transposon that mobilizes via an extrachromosomal megacircle containing 158 ORFs, including the bcpAIOB genes. Although IS2 from E. coli and the IS2-like elements in BtE264 both form circular intermediates, the contents of the intermediates are different. Understanding why the transposase acts within one element in the E. coli IS and between separate elements in the BtE264 transposon awaits further investigation. However, as IS2-containing composite transposons have not been reported previously, the element in BtE264 represents the first-identified member of this class of transposon.
IS26 elements, which are members of the IS6 family, play critical roles in the dissemination of genes encoding antibiotic resistance in Gram-negative bacteria [51–54]. IS26-containing transposons have been shown to move via an excised circular element called a translocatable unit (TU, [43]). The composition of the TU is analogous to that of the BtE264 megacircle, it contains one IS plus the DNA intervening between the two IS elements in the composite transposon. Also, similar to the case for the BtE264 IS2-like element-containing transposon, only the “left” IS26 is required for TU formation and it is the “left” IS26 that is included in the TU [43]. Moreover, it also appears that IS26 elements do not transpose as single IS elements [42]. Thus, although the BtE264 IS2-like and IS26 transposases share only limited amino acid similarity (S1 Fig), they both appear to catalyze reactions that involve distantly-located sequences to create complex, extrachromosomal circular transposition intermediates.
Integration of an IS26-containing TU into a target molecule can occur randomly via an untargeted replicative mechanism involving duplication of IS26 and the target sequence, or via a targeted conservative mechanism that targets an existing IS26 element to recreate an IS26-containing composite transposon without duplication of any sequence [41,42,44,45]. The targeted conservative mechanism can occur by homologous recombination between sequences within the IS26 elements or, much more efficiently, by transposase-catalyzed recombination between sequences at the left or right ends of the IS26 elements [42]. For targeted integration involving transposase-catalyzed recombination, both IS26-encoded transposases must be active [43]. The fact that a composite transposon that is apparently identical to the one present in WT BtE264 was recreated in chromosome II in strain 131–10, or adjacent to the partially deleted MGE in the Reg1::nptII mutant, provides evidence that the BtE264 megacircle is capable of transposition via a targeted conservative mechanism, similar to the IS26-containing TU. Whether integration of the megacircle occurred via RecA-dependent or transposase-mediated recombination is currently unknown and under investigation. If it occurred via transposase-mediated recombination it would suggest that the recently discovered targeted reaction mediated by the IS26-encoded transposase is used by multiple transposases, including those with surprising little amino acid similarity.
Targeted conservative transposition of IS26-containing transposons carrying genes encoding β-lactamases facilitates amplification of the element in response to exposure of the bacterium to β-lactams [55]. Such a response to selective pressure, resulting in multiple copies of the transposon in tandem array, could explain the integration of the megacircle in strain Reg1::nptII (since Region 1 apparently contains genes essential for cell growth). In addition to tRNA-synthetases and other predicted housekeeping proteins, BcpI (the immunity protein) is essential in bacteria producing a functional BcpA protein. Interestingly, a BtE264 transposon mutant library constructed by Gallagher et al. includes mutants with transposons inserted within genes that span the bcpAIOB-containing MGE that are predicted to be essential, including bcpI [56]. Our data suggest that construction of these mutants may have been possible due to the presence and mobilization of the bcpAIOB-containing MGE.
We showed previously that a catalytically active BcpA protein is required for changes in gene expression that lead to behaviors such as biofilm formation and pigment production, a phenomenon we call CDS [26]. The mechanistic link between BcpA activity and gene expression changes, however, is unknown. We found in this study that production of the megacircle correlated directly with BcpA activity, suggesting that megacircle formation is another CDS phenotype. However, deletion of orfAB in IS2β resulted in not only lack of megacircle formation, but also lack of other CDS phenotypes, despite the BcpAIOB proteins being unaltered and functional (as evident by the fact that the ΔIS2β strain was capable of mediating CDI). These data suggest a linear relationship between BcpA activity, megacircle formation, and CDS phenotypes (i.e., active BcpA induces megacircle formation and megacircles induce CDS phenotypes). While understanding the mechanism by which megacircles induce CDS phenotypes will require further investigation, one possibility is that megacircle-dependent gene expression changes result simply from increased gene copy number. Consistent with this hypothesis, transcriptomic analyses of BtE264 cultured under CDS-inducing conditions revealed increased expression of 58 of the 161 genes within the composite transposon [26]. While some of these genes contribute directly to CDS phenotypes, such as the csu operon which is involved in biofilm formation, others, such as those predicted to encode regulatory factors, may function indirectly.
The mechanism by which BcpA induces megacircle formation is similarly unknown. Absence of megacircles in the strain producing chimeric BcpA (Bt-Bp, Fig 4) indicates that the correct catalytic activity (i.e., that of the BcpA protein encoded by the BtE264 allele) is required. One hypothesis is that the C-terminus of BcpA in BtE264, which is predicted to share structural similarity with holiday junction resolvases, is directly involved in the recombination reaction mediated by the IS2β-encoded transposase. Another possibility is that activity of the BcpA C-terminus results in a shift from production of OrfA, which inhibits activity of the IS2 element-encoded transposase in E. coli, to production of the full-length OrfAB transposase [31,32]. We are currently investigating these possibilities, and can also envisage others.
Regardless of the underlying mechanisms, megacircle formation is clearly a result of contact-dependent interactions between cells producing the same BcpAIOB proteins. We hypothesize that CDI/CDS systems function as both harming and helping greenbeards, inhibiting the growth of non-self bacteria, and inducing cooperative behaviors in self bacteria, upon cell-cell contact, with self defined by the specific bcpAIOB (or cdiBAI) allele [14]. Our current results provide evidence that the bcpAIOB genes in BtE264 are located within a mobile genetic element, which would put the bcpAIOB genes and others encoding proteins involved in cooperative behaviors in linkage disequilibrium with the rest of the chromosome, another feature of greenbeard genes. But can the bcpAIOB-containing MGE translocate from one cell to another? Our data suggest that the recipient cell would have to contain at least one IS2-like element for the megacircle to recreate the transposon. Bioinformatic analyses indicate the presence of highly conserved IS2-like elements in other Burkholderia species, including members of the Burkholderia cepacia complex (Bcc). In addition, the recipient cell would have to produce the correct outer membrane receptor and cytoplasmic membrane translocation protein for CDS to occur. Although the identities of these proteins for BcpAE264 are unknown, we showed recently that B. dolosa strain BdAU0158 can mediate CDI using BcpAIOB proteins that are nearly identical to those produced by BtE264 [57], and that BdAU0158 can induce CDS phenotypes in a BtE264 ΔbcpA mutant [26], suggesting that these strains share the receptor and translocator proteins. The recipient cell would also have to tolerate duplicate copies of the essential genes on the MGE, or have a mechanism for deleting or mutating them.
Genomic analyses have predicted that CDI/CDS system-encoding genes are located within genomic islands [58,59]. In search of evidence for transfer of bcpAIOB genes among Burkholderia species via horizontal gene transfer, we recently searched for bcpAIOB homologs and then used Mauve software to detect evidence of synteny surrounding those genes [57]. Our search identified 13 strains and six bcpAIOB alleles that differed only in the regions encoding the C-terminal ~100 aa of BcpA and the N-terminal ~150 aa of BcpI. Flanking genes were similar only in strains that contained the same allele–there was no synteny around the bcpAIOB genes among strains with slightly different bcpAIOB alleles [57]. These data suggest that if these closely-related alleles were acquired horizontally, there has been substantial evolution since that time (i.e., they do not appear to be located within the same or similar genomic islands currently). However, further comparison of the three genomes containing bcpAIOB alleles identical to that in BtE264 revealed that although the entire genomes appear to be nearly identical, strain BtE254 lacks all three IS elements (ISBma1a, IS2α, and ISBma1b) at the “α end” of the bcpAIOB-containing transposon. Moreover, these elements are flanked by direct repeats in BtE264, and there is no apparent “scar” in BtE254 (S7 Fig), suggesting that BtE264 gained these IS elements, rather than BtE254 losing them. It appears, therefore, that a relatively recent transposition event introducing ISBma1a, IS2α, and ISBma1b into chromosome I of BtE264 resulted in the formation of the bcpAIOB-containing IS2-like composite transposon, which is currently mobile, at least intracellularly.
If it occurs, interbacterial transfer of the MGE would support the selfish gene hypothesis for bcpAIOB. Our data indicate that B. thailandensis communities are composed of megacircle-producing bacteria. A non-self bacterium (one that does not contain the same bcpAIOB allele) that encounters such a community may receive a BcpA C-terminus and be killed due to the lack of the correct BcpI protein. Alternatively, the invader may receive the megacircle, and if the megacircle can insert into the chromosome, the invader can be converted into a ‘self’ cell that is not only immune to BcpA-mediated CDI, but that could produce megacircles and, consequently, proteins involved in cooperative behaviors. If the invading bacterium contains a different bcpAIOB allele, or another mechanism to kill the initial community, the invading bacterium and its descendants will eliminate the resident population and take over the niche, and the selfish bcpAIOB genes will propagate within the newly established population. Experiments to determine if the bcpAIOB-containing MGE can be transferred intercellularly are underway.
Burkholderia thailandensis E264 is an environmental isolate [60]. All plasmids and strains used in this study are listed in S2 Table in Supporting Information. Plasmids were maintained in E. coli DH5α and introduced into BtE264 through biparental matings using E. coli RHO3 as the plasmid donor [61,62]. BtE264 and E. coli strains were grown overnight with aeration at 37°C (unless indicated) in low salt Luria-Bertani (LSLB, 0.5% NaCl). Antibiotics were added to cultures at the following concentrations: 250 μg/mL (for BtE264) or 50 μg/mL (for E. coli) kanamycin (Kan), 100 μg/mL ampicillin, 200 μg/mL (for BtE264) or 50 μg/mL (for E. coli) trimethoprim (TMP), or 200 μg/mL diaminopimelic acid as appropriate. When indicated, BtE264 was cultured on M63 minimal medium (110 mM KH2PO4, 200 mM K2HPO4, 75 mM (NH4)2SO4, 16 nM FeSO4) supplemented with 1mM MgSO4 and 0.2% glucose [63].
BtE264 IS2α::nptII and IS2β::nptII were constructed by natural transformation [63]. First, a 1.4 kb DNA fragment consisting of the gene encoding kanamycin resistance, its promoter, and flanking FRT sites was amplified from pUC18miniTn7(Km) by PCR using primers containing 5’ NdeI or EcoRV restriction sites. The DNA fragment was then introduced into the blunt cloning site of pJET1.2 (Thermo Fisher), resulting in plasmids pABT62-NdeI or pABT62-EcoRV. Additionally, DNA fragments around 750 bps (for IS2β) or 1.5 kb (for IS2α) in size, and 5’ or 3’ to the IS2 elements, were amplified using BtE264 genomic DNA as template. SOEing mutagenesis was then employed to construct a single DNA product in which the 5’ and 3’ regions of homology were joined and a NdeI site was added to the middle of the PCR product. Next, the fused DNA PCR product was cloned into the blunt cloning site of pJET1.2. The resulting plasmid was confirmed by Sanger sequencing and then subjected to linearization with NdeI (NEB), so that FRT-nptII-FRT (dropped from pABT62-NdeI with the same enzyme) could be cloned into the appropriate restriction site. This gave rise to plasmids pABT78 (IS2α::nptII) and pABT66 (IS2β::nptII); which in turn were linearized with HindIII (NEB) and transformed into BtE264 WT.
Deletion of nucleotides 2,945,740 to 3,156,451 was achieved through a multi-step process involving natural transformation. First, ~750 nucleotides corresponding to the 5’ end of ISBma1b or the 3’ end of BTH_I2631 were amplified from BtE264 gDNA. Overlap PCR was performed to join the 5’ and 3’ homology sequences and form a single DNA product which included an EcoRV site in the middle. FRT-nptII-FRT (dropped from pABT62-EcoRV with the same enzyme) was then inserted into the EcoRV site of the fused sequences, generating plasmid pABT63 (Reg1::nptII). pABT63 was then introduced into WT BtE264 cells by natural transformation followed by selection with kanamycin. The kanamycin cassette was removed from BtReg1::nptII transformants by Flp-FRT recombination using pFlpTet [25,64]. The same method was used to create pABT65 (Reg3::nptII) which included the sequence 5’ to BTH_I2671 or 3’ to BTH_I2705. pABT65 was then introduced into the kanamycin sensitive BtReg1 strain; the resulting BtReg1Region3::nptII transformants was then subjected to Flp-FRT recombination to generate a kanamycin sensitive version. Next, pABT68 (Reg4::nptII, generated by joining the sequence 5’ to gene BTH_I2706 and the sequence 3’ to IS2β) was used to construct BtReg1Region3-IS2β::nptII via natural transformation using the BtReg1Region3 kanamycin sensitive strain. The kanamycin cassette was then removed, resulting in BtReg1Region3-IS2β KanS. Lastly, the sequence 5’ to IS2β and 3’ to IS2β was used to create pABT71 (Reg1-4::nptII), which was introduced into the BtReg1 Region3-ΔIS2β KanS strain to construct the final BtE264 mutant lacking the bcpAIOB-containing mobile genetic element at its native location (strain 131–10). The kanamycin sensitive BtReg1 strain was also subjected to natural transformation with linear bcpAIOB::nptII gDNA obtained from strain ΔbcpAIOB (8) resulting in strain Reg1 bcpAIOB+/–. The latter strain was then subjected to Flp-FRT recombination to generate a kanamycin sensitive version and the resulting strain was used for a second round of transformation to replace the second bcpAIOB copy with nptII (Reg1 bcpAIOB–/–).
The transformation efficiencies when a FRT-nptII-FRT cassette is introduced inside or outside of the composite transposon were determined using pABT77 and pABT79 respectively; these plasmids were constructed as follows. First, a ~1.0 kb DNA segment, with a naturally present EcoRV restriction site located ~4.5 kb upstream (outside) or ~30 kb downstream (inside) of IS2α, was amplified from BtE264 gDNA and cloned into pJET1.2. Next, the FRT-nptII-FRT cassette, dropped from pABT62-EcoRV, was inserted into the linearized pJET containing the “inside” or “outside” segment, generating plasmids pABT77 (outside of MGE::nptII) and pABT79 (inside of MGE::nptII).
Lastly, the suicide plasmids pABT73-TMP and pABT74-TMP were constructed as follows. Approximately 500 nucleotides were amplified from WT BtE264 gDNA, this sequence is identical to the region between ISBma1b and BTH_I2587 (pABT74) or the region between BTH_I2743 and IS2β (pABT73). The PCR product was then cloned into the blunt end of pJET1.2 followed by verification of the resulting plasmid. The sequence of interested was digested from the pJET1.2 backbone using BglII and then cloned into the BamHI restriction site of pEX18-TMP [65] giving rise to pABT73-TMP and pABT74-TMP. The plasmid was then moved to RHO3 cells for conjugation into BtReg1::nptII (pABT74-TMP) and strain 131–10 (pABT73-TMP), followed by selection on kanamycin- and TMP-supplemented media. At least two independent cointegrants obtained from each mating were used for plasmid rescue analyses.
Genomic DNA was isolated from WT BtE264 cells grown in liquid broth using Wizard Genomic DNA Purification Kit (Promega). Paired-end TruSeq (Illumina) gDNA libraries were generated and subjected to sequencing for 300 cycles using the Illumina MiSeq platform at the High-Throughput Sequencing Facility (HTSF) at the UNC School of Medicine. Following demultiplexing, FASTQ files were mapped to the reference genome available for BtE264 (Accession no. CP000086.1 for Chromosome I and CP000085.1 for Chromosome II) using the Geneious v. 8 standard assembler, resulting in >200x coverage. Sequencing reads can be accessed in the Sequence Read Archive (SRA); accession number PRJNA510167.
Primers Circ1 and Circ2 (S3 Table) were designed to bind at each end of, and reading in opposite direction away from, the composite transposon. Upon formation of the IS2 megacircle, Circ1 and Circ2 are in proximity and in the correct orientation to generate a product. At least 15 independent PCR products have been sequenced. To detect WT DNA, primers In1 and In2 (which bind to BTH_I2615 and BTH_I2616, respectively, to amplify a 1.0 kb fragment) were used. PCR studies were performed in 25 μL reaction mixtures with GoTaq DNA polymerase (Promega) for 25 cycles and with 3 μL of diluted over-night cultures normalized an OD600 of 1.0 as the source of template DNA. PCR conditions to detect the megacircle junction included an annealing temperature of 55°C and elongation time of 150 seconds. PCR products were analyzed on a 0.8% agarose gel containing GelRed Nucleic Acid Gel Stain (Biotium) and visualized under UV light.
Competitions between inhibitors and ΔbcpAIOB (which is susceptible to killing via CDI due to the absence of bcpI) were performed as previously described [8]. Briefly, overnight liquid cultures of inhibitors and target were diluted to OD600 of 0.2, and single inhibitors were mixed with ΔbcpAIOB at a 1:1 ratio. Next, 20 μL of the cell mixture were spotted in triplicate onto LSLB agar without antibiotic selection. Plates were incubated at room temperature for exactly 24 hours. Bacteria from the edge of the colony biofilm were harvested and suspended in PBS, then subjected to serial dilutions and plated on LSLB supplemented with appropriate antibiotics to enumerate CFU corresponding to the inhibitor and target. The competitive index (C.I.) is reported as the log of the ratio of inhibitor to target cells at 24 hours (t24) divided by the same ratio at t0. Three biological replicates were performed for each competition.
Congo red (CR) binding was determined by counting the number of CR+ and CR- colonies from strains grown on M63 minimal medium supplemented with 40 μg/mL of Congo red dye. Upon inoculation, plates were cultured at 37°C for 48 hours then incubated at room temperature for approximately three days. The ability of WT and mutant strains to aggregate at the air-liquid interphase when grown in M63 minimal medium was determined as follows. Overnight LSLB cultures were washed with PBS and diluted to an OD600 of 0.2 with M63 supplemented with 0.01% casamino acids and 0.4% glycerol in a final volume of 2 mL. Bacteria were cultured at 37°C while rotating for 24 hours, then imaged. Colony biofilm pigmentation assays were conducted as follows. Overnight LSLB cultures were washed and diluted to an OD600 of 0.2 with PBS, 20 μL of cell suspension was then spotted onto LSLB agar and air dried. Plates were incubated at room temperature for 2–3 weeks prior to imaging.
Plasmid rescue was performed using genomic DNA from two independent BtReg1::nptII::pABT74-TMP cointegrants and one 131–10::pABT73-TMP cointegrant. Genomic DNA (2 μg) was digested with 100 U of NotI (for strain BtReg1::nptII+pABT74-TMP) or SacII (for 131–10::pABT73-TMP) at 37°C for 18 hours, the reaction was then supplemented with additional 20 U of the restriction enzyme and incubated for two more hours. Next, the reaction was heat inactivated following manufactures’ recommendations. T4 ligase was added, and the reaction containing digested gDNA and ligase was incubated overnight at 16°C, then transformed into 5-alpha F’Iq High Efficiency Competent E. coli cells (NEB, C2992H), and transformants were selected on media supplemented with TMP. Lastly, “rescued” plasmids from multiple transformants were isolated and subjected to Sanger sequencing.
Natural transformation of BtE264 was used with modifications [63]. Bacteria grown overnight in LSLB were used to inoculate M63 minimal medium at a 1:20 dilution, then incubated at 37°C for 5 hours. Cultures were concentrated to an OD600 of 10 (in M63 medium) and 50 μL of cell suspension were incubated at room temperature for 30 minutes with 100 ng of linearized plasmids (pABT63, pABT77 or pABT79) or 2 μL of water. Next, 1.5 mL of fresh M63 were added to each sample and transferred to a tube to be cultured at 37°C while rotating for 20 hours. Cells were then pelleted and resuspended in 40 μL of PBS, 20 μL were plated on LSLB supplemented with kanamycin. The remaining 20 μL were subjected to serial dilutions, which were then plated on LSLB without antibiotics. Plates were incubated at 37°C for 24 hours, after which CFU were counted. Transformation efficiency was calculated by dividing the number of kanamycin resistant colonies by the number of colonies on the LSLB plates without antibiotics. Limit of detection in LSLB media supplemented with kanamycin is equal to 2.
|
10.1371/journal.pgen.1000625 | A High Incidence of Meiotic Silencing of Unsynapsed Chromatin Is Not Associated with Substantial Pachytene Loss in Heterozygous Male Mice Carrying Multiple Simple Robertsonian Translocations | Meiosis is a complex type of cell division that involves homologous chromosome pairing, synapsis, recombination, and segregation. When any of these processes is altered, cellular checkpoints arrest meiosis progression and induce cell elimination. Meiotic impairment is particularly frequent in organisms bearing chromosomal translocations. When chromosomal translocations appear in heterozygosis, the chromosomes involved may not correctly complete synapsis, recombination, and/or segregation, thus promoting the activation of checkpoints that lead to the death of the meiocytes. In mammals and other organisms, the unsynapsed chromosomal regions are subject to a process called meiotic silencing of unsynapsed chromatin (MSUC). Different degrees of asynapsis could contribute to disturb the normal loading of MSUC proteins, interfering with autosome and sex chromosome gene expression and triggering a massive pachytene cell death. We report that in mice that are heterozygous for eight multiple simple Robertsonian translocations, most pachytene spermatocytes bear trivalents with unsynapsed regions that incorporate, in a stage-dependent manner, proteins involved in MSUC (e.g., γH2AX, ATR, ubiquitinated-H2A, SUMO-1, and XMR). These spermatocytes have a correct MSUC response and are not eliminated during pachytene and most of them proceed into diplotene. However, we found a high incidence of apoptotic spermatocytes at the metaphase stage. These results suggest that in Robertsonian heterozygous mice synapsis defects on most pachytene cells do not trigger a prophase-I checkpoint. Instead, meiotic impairment seems to mainly rely on the action of a checkpoint acting at the metaphase stage. We propose that a low stringency of the pachytene checkpoint could help to increase the chances that spermatocytes with synaptic defects will complete meiotic divisions and differentiate into viable gametes. This scenario, despite a reduction of fertility, allows the spreading of Robertsonian translocations, explaining the multitude of natural Robertsonian populations described in the mouse.
| Cells have different mechanisms to assess the proper occurrence of cellular events. These mechanisms are called checkpoints and are involved in the surveillance of processes such as DNA replication and cell division. A checkpoint at the pachytene stage arrests meiosis when defects in the process of homologous chromosome synapsis and recombination are detected. In mammals, both transcriptional inactivation of chromosomal regions that are not correctly synapsed at pachytene and activation of sex chromosome genes that are normally silent during this stage could contribute to meiotic arrest. We found that when Robertsonian translocations appear in heterozygosis, many synapsis defects occur, and mechanisms that trigger transcriptional silencing of the unsynapsed chromatin are activated. However, meiotic prophase-I progression is not greatly compromised. This questions the ability of the meiotic checkpoints to halt meiosis progression when synapsis is not completed, allowing cells with synapsis defects to reach the first meiotic division. The fertility reduction of Robertsonian heterozygous mice seems to be mainly caused by errors detected by the metaphase-I checkpoint, when most of the spermatocytes die, rather than by synapsis defects. In an evolutionary context, a permissive pachytene checkpoint could contribute to increasing the chances of Robertsonian translocations to spread into natural populations.
| A series of complex processes takes place during the first meiotic division, including pairing, synapsis, recombination and segregation of homologous chromosomes. Defects in any of these processes can affect the normal progression of meiosis, causing severe fertility reduction or even sterility [1]–[3]. This is a consequence of the existence of surveillance mechanisms that monitor the accurate progression of meiotic events and promote the removal of defective cells. Two main checkpoints have been proposed to act during the first meiotic division: the pachytene checkpoint, responsible for ensuring the correct occurrence of recombination and synapsis [2],[4],[5], and the metaphase-I or spindle checkpoint, which controls the precise segregation of homologous chromosomes [6],[7].
Although the process that eliminates meiocytes in metaphase-I and II might be similar to that acting during mitosis [6],[7], a clear understanding of the mechanisms that trigger the pachytene checkpoint is still lacking. Given the interdependence between meiotic recombination and synapsis, it has been difficult to ascertain the existence of separate checkpoints for these processes in mammals. Thus, many recombination-defective mutants exhibit a delay in synapsis and/or synaptic aberrations, and meiosis is aborted during the zygotene-pachytene transition [8]–[10]. Likewise, most mutants defective for synaptonemal complex (SC) components abort meiosis at pachytene with unresolved recombination processes [11]–[15].
In addition to the accumulation of unresolved recombination intermediates, unsynapsed chromosomal regions undergo a process of transcriptional inactivation called meiotic silencing of unsynapsed chromatin (MSUC) [1], [16]–[18]. The mechanisms involved in transcriptional inactivation are particularly well characterized in mammalian male meiosis, in which sex chromosomes undergo a special case of MSUC called meiotic inactivation of sex chromosomes (MSCI) [19],[20]. This process is initiated with the accumulation of BRCA1 protein on the unsynapsed axial elements (AEs). BRCA1 is a protein involved in DNA damage repair that allows the recruitment of other factors such as ATR, promoting the phosphorylation of H2AX at serine 139 on the surrounding chromatin [21],[22]. The inactivation of sex chromosomes, which affects the unsynapsed regions of both the X and Y chromosomes, comprises an additional plethora of chromatin modifications that includes: 1) histone modification [18],[23],[24]; 2) incorporation of specific histone variants [25],[26]; 3) specific incorporation of non-histone proteins [27]–[30]; and 4) accumulation of XIST RNA [31] and other families of non-coding RNAs [32].
The initiation of MSUC seems to also operate by the action of BRCA1 and ATR [17]. Furthermore, it has been reported that many chromatin modifications detected during MSCI are also involved in the inactivation of unsynapsed autosomes. This is the case of H2AX phosphorylation [17], histone H2A ubiquitination [18], methylation of histone H3 and H4, incorporation of histone H3.3 [26] and Maelstrom protein [29]. However, the role of other chromatin modifications in MSUC remains to be demonstrated.
On these grounds, it has been proposed that MSUC may interfere with the expression of genes necessary for the completion of meiosis and this would contribute to arrest the meiotic progression of pachytene spermatocytes with synapsis defects [17]. More recently, it has been suggested that extensive asynapsis and MSUC could also interfere with MSCI [33]. Indeed, activation of some sex chromosome-linked genes that should remain inactive during meiosis has been claimed as one of the causes of meiotic failure in some mouse models [1],[17],[34],[35]. Mahadevaiah and co-workers [33] have proposed that MSCI initiation could be impeded by the sequestration of MSUC triggering proteins like BRCA1 and ATR on extensively unsynapsed autosomes, a circumstance that would preclude these proteins to relocate to the unsynapsed AEs of the sex chromosomes. MSCI abrogation has thus been proposed as the primary cause of spermatocyte death in mouse models that typically arrest meiosis at the zygotene-pachytene transition, including many recombination-defective mutants [33]. Sequestration of BRCA1 has been proposed to occur also in female meiosis [36]. However, in both cases cells seem to tolerate a certain degree of asynapsis, since both spermatocytes and oocytes with a reduced number of asynapsed chromosomes are able to progress through first meiotic prophase without interfering with of MSUC or MSCI processes [33],[36].
In the house mouse (Mus musculus domesticus), individuals that are heterozygous for Robertsonian (Rb) translocations (the fusion of two acrocentric chromosomes) show reduced fertility. This reduction is strongly correlated with impairment of spermatogenesis and loss of meiotic cells [37]–[45]. Depending on the number and complexity of Rb heterozygosity (i.e. formation of trivalents, chains or rings), meiocytes may be eliminated during prophase-I [41], [46]–[48] or during metaphase-I and II [39], [47]–[50].
The synaptic behaviour of trivalents in Rb heterozygotes has been extensively analyzed by means of electron microscopy in a wide range of mammalian species, including mouse and humans [42], [44], [45], [47], [51]–[54]. During meiosis, heterozygous mice display a high frequency of pairing abnormalities including: 1) delay in synapsis completion of trivalents; 2) existence of a variety of heterelogous synaptic situations, both within and between trivalents and between trivalents and the sex chromosomes; and 3) persistence of unsynapsed regions in the trivalents throughout pachytene. Furthermore, a reduction of the recombination frequency and a decrease of chiasma interference in these hybrids have been demonstrated [40], [55]–[58]. However, little is known about the chromatin modifications associated with these synaptic disturbances.
The aim of this study is to ascertain the extent of MSUC during meiosis in Rb heterozygous mice and to evaluate the consequences of this cellular response on the meiotic progression of spermatocytes. We used males generated by crossing individuals of a standard karyotype (2n = 40) with homozygous individuals bearing eight Rb translocations (2n = 24), collected from natural populations in Northern Italy. The resulting hybrids (2n = 32) bear eight trivalents that exhibit different degrees of asynapsis during meiosis. We have combined the analysis of synapsis and recombination progression during male meiosis with the localization of some proteins involved in MSUC, i.e., γH2AX, ATR, ubiquitinated H2A, SUMO-1 and XMR, the latter two having only been reported to act in MSCI. Our results describe the kinetics of MSUC in Rb heterozygotes and highlight the capacity of spermatocytes with synaptic defects to pass through pachytene and progress to the metaphase stage.
To characterize the progression of the first meiotic prophase, we used three main criteria: 1) the localization of SYCP3, the main component of the synaptonemal complex (SC) axial/lateral element (AE/LE), and that of RAD51, a protein related to early meiotic recombination and repair (Figure 1) that is abundantly incorporated along the chromosomes at zygotene, and then gradually disappears during pachytene and is absent at mid/late pachytene [59]; 2) the length of the pairing region between X and Y chromosomes, which extends up to 100% of the Y chromosome at early pachytene and becomes shorter as pachytene proceeds [60]; and 3) the reduction of the pairing region of sex chromosomes to the very distal end, the appearance of excrescences on the AEs of sex chromosomes, and the widening of SC attachment plates on the autosomes that identify the late pachytene stage. These criteria are comparable with those reported in recent studies carried out using RPA and MLH1 as markers of pachytene progression [61].
We found that synapsis was initiated at early zygotene in both bivalents and trivalents, but proceeded more quickly in the bivalents (Figure 1A). In fact, most trivalents were still undergoing synapsis when bivalents (b in Figure 1) were almost completely synapsed. This may be due to the fact that in trivalents, synapsis was initiated only at the distal ends of the chromosomes (Figure 1B–1B'). At this meiotic stage, the X and Y chromosomes usually lay apart from each other. At early pachytene, all bivalents and some trivalents had completed synapsis (closed configuration), although in many trivalents the chromosomal regions close to the centromeres were still unsynapsed (open configuration) (Figure 1C–1E' and Table 1).
The pattern of RAD51 localization at the early stages of prophase-I was similar to that exhibited by mice with the standard acrocentric karyotype. During zygotene, a large number of RAD51 foci appeared on both synapsed and unsynapsed AEs of bivalents and trivalents and on the X chromosome (Figure 1A–1B'). Then, during early pachytene, the number of foci started to drop, although foci remained more abundant in both trivalents and sex chromosomes than in bivalents (Figure 1C–1E'). RAD51 foci were associated with the trivalents in either the open or closed configuration and did not preferentially accumulate on the unsynapsed regions of the open trivalents (Figure 1C–1D').
At mid-pachytene, many trivalents had completed synapsis and appeared in a closed configuration, but one to four trivalents remained in an open configuration (Figure 1F–1H'). At this stage RAD51 was only present on the sex chromosomes and on some trivalents.
At late pachytene, most cells exhibited up to four trivalents with an open configuration (Figure 1I–1K'). At diplotene, when desynapsis starts and homologues initiate their separation, the proximal ends of the acrocentric chromosomes remained associated in some trivalents while they appeared clearly separated in others (Figure 1L–1M'). RAD51 was not detectable at late pachytene (Figure 1I–1K') or diplotene (Figure 1L–1M'). These results indicate that the repair of DNA might be delayed in some trivalents as it is in the sex chromosomes, but this process seems to culminate successfully in mid-late pachytene, when the signal of RAD51 disappeared, even though many unsynapsed chromosome regions are present.
Trivalents commonly engage in ectopic heterologous associations with other trivalents and/or the sex chromosomes (Figure 1C). Thus, we wondered whether these associations would involve the assembly of the SC as a tripartite structure. For this purpose we analyzed the localization of SYCP1 protein, one of the main components of the SC transverse filaments and central element (Figure 2). At zygotene, we found that trivalents could establish an end-to-end connection that did not usually involve SYCP1 (Figure 2A–2A'). However, the association of unsynapsed proximal ends of trivalents with the sex chromosomes frequently involved the formation of a short SC with either the distal region of the X chromosome, the proximal region or both (Figure 2B–2B'; see also Figure 3 and Figure S1). Furthermore, the Y chromosome was sometimes found in a self-synapsed configuration (Figure 2B–2B' and Figure S1). These situations usually occurred at early pachytene and were more rarely detectable from mid-pachytene onwards.
Heterologous synapsis was also found within each trivalent. Although it was expected that the two acrocentrics could synapse with the corresponding homologous segment of the metacentric (Figure 2C–2C'), synapsis between the heterologous proximal chromosomal regions of the acrocentrics was the most frequent configuration. Heterologous synapsis could involve either a short (Figure 2D–2D') or a long segment of both chromosomes (Figure 2E-2E') and could be maintained from pachytene until late diplotene (Figure 2G–2G'). Furthermore, we found that some unsynapsed chromosomal regions incorporated SYCP1 (Figure 2F–2F'), perhaps representing either unsynapsed regions that were about to synapse or regions of self-synapsis. Alternatively, they may reveal only a non-specific binding of SYCP1 to unsynapsed AEs, a feature that is frequently observed in the sex chromosomes [62].
To evaluate the incorporation of MSUC markers on unsynapsed Rb trivalents, we first examined the temporal localization of γH2AX (Figure 3 and Figure S1; Video S1). This protein localizes in foci at DNA double-strand breaks during DNA repair and it is also associated with the inactivation of unsynapsed chromatin in autosomes and sex chromosomes [17],[63],[64]. At leptotene, the localization of γH2AX was dispersed throughout the nucleus (Figure 3A; Video S1); then, at zygotene, γH2AX began to disappear from the synapsed chromosome regions of the bivalents and from the synapsed distal regions of some trivalents (Figure 3B). The X chromosome appears intensely labeled while the Y chromosome is usually devoid of γH2AX labeling.
When spermatocytes entered pachytene, γH2AX became restricted to the chromatin located close to the LEs of the synapsed regions of both bivalents and trivalents (see insets in Figure 3C and 3D) [65]. In the sex chromosomes, γH2AX was extended over the chromatin (Figure 3C and 3D). Interestingly, we observed that the Y chromosome is intensely labeled even when it occasionally appears self-synapsed (Figure S1). In the unsynapsed regions of the trivalents, γH2AX showed two labeling patterns: 1) occupying a wide chromatin area surrounding the unsynapsed segments, as in the sex chromosomes, and 2) occupying a more restricted chromatin area, very close to the unsynapsed AEs, as in the synapsed regions (see inset in Figure 3C). It is especially striking that in some trivalents one of the unsynapsed acrocentric chromosomes showed one of these labeling patterns while the other acrocentric showed the alternative pattern (see inset in Figure 3D).
From mid to late pachytene, γH2AX labeling appeared as a bright signal on the entire chromatin surrounding the X and Y chromosomes and the AEs of the unsynapsed regions of open trivalents (Figure 3E). It is important to stress that open trivalents showed γH2AX labeling regardless of whether they were close to the X and Y chromosomes or far from them, indicating that this labeling was not a consequence of their association with the sex chromosomes (see Video S2). At diplotene the localization of γH2AX in the sex chromosomes remained visible, and it was also detectable in the pericentromeric regions of some trivalents (Figure 3F). These regions most likely represent chromosomal segments that have remained unsynapsed during pachytene, since those that began desynapsis during diplotene, in either bivalents or trivalents, were devoid of γH2AX labeling. These results suggest that MSUC is a mechanism triggered during the early stages of prophase-I. Furthermore, it indicates that most spermatocytes carrying unsynapsed trivalents would proceed normally into diplotene.
Next, we investigated the presence of ATR in the unsynapsed regions of trivalents (Figure 4, Figure S2, and Figure S3). During zygotene, ATR labeling appeared as small foci located on the AEs/LEs in both synapsed and unsynapsed autosomes and in the X chromosome, but it was rarely observed in the Y chromosome (Figure 4A). At the zygotene/pachytene transition, ATR began to disappear from the chromosomes that had completed their synapsis (Figure 4B), although it remained as numerous and intense foci on the unsynapsed AEs. At this stage, a single ATR focus was always detected on the Y chromosome.
During early pachytene, ATR labeling appeared as a continuous line along the unsynapsed AEs of trivalents and sex chromosomes (Figure 4C). ATR localization contrasted with that of γH2AX, the latter including the whole unsynapsed chromatin (Figure 3C). This result indicates absence of colocalization of the two proteins during late zygotene and early pachytene in the unsynapsed regions (Figure S2). On the other hand, we observed some unsynapsed trivalent regions without the ATR signal (Figure 4B–4D). This observation is consistent with the absence of γH2AX in some unsynapsed trivalent regions and strongly suggests the existence of two classes of unpaired chromosome segments during early pachytene: one class that shows neither γH2AX nor ATR protein, and another class that shows the presence of both proteins.
At mid-pachytene, ATR labeling was still apparent along unsynapsed AEs and also appeared to extend to the surrounding chromatin of the unsynapsed regions of trivalents and of the sex chromosomes (Figure 4D); then, it became brighter as spermatocytes progressed to late pachytene (Figure 4E). During diplotene, ATR remained visible on the surrounding chromatin of the sex chromosomes and of the open trivalents (Figure 4F); its signal progressively faded and completely disappeared by the late diplotene stage.
In view of these results, we next analyzed the pattern of appearance and localization of three other MSUC/MSCI-related proteins: 1) monoubiquitinated H2A histone (ubiH2A) (Figure 5), known to be associated with transcriptional silencing of unsynapsed autosomes and sex chromosomes in mouse male meiosis [18]; 2) SUMO-1 (Figure 6), which is involved in SC assembly as well as in the formation of the sex body [30], [66]–[68], and 3) XMR (Figure 7), a member of the XLR gene superfamily [69], known to localize in the sex body [27].
We found that during zygotene (Figure 5A) and early pachytene (Figure 5B), ubiH2A appeared as a faint signal on the chromosome ends on both synapsed and unsynapsed LEs, as previously observed [18]. On the contrary, at mid-pachytene, an intense labeling appeared on the chromatin of unsynapsed segments of trivalents and of the sex chromosomes (Figure 5C), persisting until mid diplotene (Figure 5D), when it started to disappear.
SUMO-1 was not detected during zygotene (data not shown) and very early pachytene spermatocytes (Figure 6A). It appeared on the unsynapsed chromatin of sex chromosomes and trivalents during a temporal window between the early to mid-pachytene transition (Figure 6B–6D), indicating that its appearance was delayed compared to mice with standard karyotype [30],[70], and it remained detectable until the end of diplotene.
XMR started to accumulate on the unsynapsed chromatin of trivalents and of the sex chromosomes at early to mid-pachytene transition and it disappeared at late diplotene (Figure 7). Interestingly, the intensity of the XMR signal seemed to be lower on the unsynapsed chromatin of the trivalents that were far from the sex chromosomes compared to that on the unsynapsed chromatin of the trivalents that were close to the sex body. These results suggest that the location of XMR in the open trivalents could be influenced by their association with the sex chromosomes.
In summary, our results show that the proteins γH2AX and ATR started to appear at the beginning of prophase I (leptotene), but intense labeling of ubiH2A, SUMO-1 and XMR was detected at a later stage (early to mid-pachytene). The appearance of ubiH2A and SUMO-1, which are known to be involved in DNA repair [71],[72] slightly preceded the spread of ATR from the chromosome axes to the unpaired chromatin (see Figure S3 for a comparison between the timing of the appearance of SUMO-1 and ATR on unsynapsed chromatin). Therefore, ATR was the last protein to appear on the unsynapsed chromatin at mid-pachytene (Figure 8). All these proteins remained localized on the unsynapsed regions of trivalents and of the sex chromosomes until late diplotene, indicating an active repair of DNA on the unsynapsed chromatin of trivalents. Also, at mid-pachytene, ubiH2A and SUMO-1 might be involved in determining those chromatin modifications that would lead to the transcriptional inactivation of unsynapsed chromatin [72],[73].
The presence of trivalents with unsynapsed proximal regions throughout the first meiotic prophase raises the question of how many of these trivalents achieve a complete synapsis. To this end, we analyzed the number of completely synapsed (closed) and partially unsynapsed (open) trivalents during early, mid- and late pachytene and during early and mid-late diplotene (Table 1) in two individuals. No statistical differences were found between them. At the beginning of pachytene, only 2.22% of spermatocytes had completed synapsis in all trivalents, whereas most spermatocytes showed one to eight open trivalents, with spermatocytes having four open trivalents occurring at the highest frequency (27.41%). During prophase progression, the frequency of spermatocytes with a high number of open trivalents tended to decrease, even though, at late pachytene, the great majority (87.55%) of spermatocytes possessed open trivalents (Table 1). At diplotene, the frequency of spermatocytes with closed trivalents or with one open trivalent (recognized by the presence of γH2AX) increased slightly (18.36% and 51.20%, respectively), although not significantly when compared to that of late pachytene. On the contrary, the frequency of spermatocytes with two, three and four open trivalents slightly decreased (Table 1). These data show that most spermatocytes maintained partially unsynapsed trivalents throughout pachytene, although their number decreased towards the end of pachytene along with an increase of spermatocytes with completely synapsed trivalents.
To estimate germ cell death, we made a quantitative evaluation of the TUNEL-positive cells present in seminiferous tubule cross-sections. Confirming previous results [39],[49], TUNEL positive cells were almost exclusively present in the meiotic compartment of stage XII of the seminiferous epithelium, which contains spermatocytes at the zygotene-pachytene transition, metaphase I and II. An average of 19.44% (±4.37) of spermatocytes were TUNEL-positive, most of which were at the metaphase stage (Figure 9). When we specifically evaluated metaphase cells at stage XII, we found that 63% of them were TUNEL positive, as shown in an our previous study [39]. This suggests that 37% of metaphase cells are able to pass the spindle checkpoint and progress to further stages of differentiation. In this regard, we previously reported a mean ratio between round spermatids and pachytene spermatocytes of 1.43, corresponding to 36% of germ cell survival following meiosis in the same type of Rb heterozygous mice, although in the homozygous parentals germ cell survival is 84% and 86% for 2n = 40 and 2n = 24 karyotypes, respectively [40],[43]. Moreover, the absence of extensive cell death in other stages of the spermatogenetic cycle suggests that pachytene and diplotene spermatocytes are able to progress to meiotic divisions despite the presence of unsynapsed trivalents.
The aim of this study was to evaluate the involvement of MSUC during the meiotic progression of spermatocytes of Rb heterozygous mice. The data presented here show that the mechanisms that regulate MSUC are active during meiosis in mice heterozygous for multiple simple Rb translocations. We report that most pachytene spermatocytes bear trivalents with unsynapsed regions that incorporate, in a stage-dependent manner, proteins involved in MSUC. Our results demonstrate that although many chromosomal regions remain unsynapsed, massive cell death is not detected at pachytene. On the contrary, spermatocytes bearing unsynapsed chromosomes subject to MSUC progress into diplotene.
It has been repeatedly reported that synapsis is delayed in heterozygotes for Rb translocations [44], [45], [47], [51]–[54],[57] and other chromosomal rearrangements [46],[74],[75]. The results presented here are in agreement with these previous reports. During zygotene, while synapsis progresses rapidly in the bivalents, in the trivalents it is initiated at the distal ends and then slowly progresses to the proximal ends of the acrocentrics. Previous reports have suggested that a delay in synapsis might be influenced by architectural constraints [43],[52],[58],[76]. In fact, the centromeres and proximal telomeres of acrocentric chromosomes are located at the nuclear periphery, while centromeres of metacentric chromosomes are located more internally in the nucleus (unpublished observations). This distinct localization of centromeres is defined by the different trajectory of the metacentric chromosomes' AEs within the nuclear space compared to that of acrocentrics' AEs [76],[77]. As synapsis of trivalents progresses from their distal telomeres, metacentric centromeres tend to approach to the nuclear envelope, where acrocentric centromeres and proximal telomeres are bound. These circumstances would explain why at the beginning of pachytene, while bivalents and sex chromosomes have achieved their respective synapsis, trivalents still appear with an open configuration. The presence of many unsynapsed proximal regions in acrocentric chromosomes located at the nuclear periphery would promote their association, causing the appearance of an ectopic heterologous synapsis between them or with the sex chromosomes at early pachytene. Most of these associations tend to disappear as trivalents complete their synapsis during mid and late pachytene. In agreement with previous reports [45],[52], our results show a decrease in the number of open trivalents throughout pachytene. These results suggest that trivalents can complete synapsis during the mid and late pachytene stages, as previously reported by Moses and coworkers [52]. However, contrary to the results found in lemur Rb heterozygotes [52], in which all trivalents finally achieve complete synapsis, in mouse there is a striking persistence of trivalents in open configuration throughout pachytene and at later stages [45].
Compared to autosomal bivalents, we found that trivalents retain RAD51 until later stages; however, RAD51 foci are not specifically enriched in the unsynapsed segments of the trivalents, a finding that differs from previous studies that have reported the maintenance of this protein on asynaptic autosomal segments [78],[79]. RAD51 finally disappears from the trivalents during mid-pachytene, despite the presence of unsynapsed segments. This circumstance has two interesting implications. First, our results confirm previous data that cells can accomplish prophase-I with unsynapsed autosomes [1],[33],[36],[64],[80],[81]. Since completion of the recombination/repair process is considered necessary to bypass the pachytene checkpoint [2],[4],[5],[82] it is likely that unsynapsed segments are repaired by the end of pachytene. This behavior parallels the situation found in the sex chromosomes. Second, some trivalents probably complete synapsis after RAD51 has disappeared, indicating the existence of a mechanism that is able to complete synapsis independently of the usual recombination/repair pathway [83]. Interestingly, many of these late synapsis events culminate with the heterologous synapsis of acrocentric chromosomes within each trivalent. This process, called synaptic adjustment, has been previously reported for these and other chromosomal rearrangements [44],[45],[52],[78],[84]. An additional consequence of both the persistence of unsynapsed and the presence of non-homologous synapsed chromosome regions is the reduction of chromosome segments where reciprocal homologous recombination could take place. This could account, at least partially, for the displacement of chiasma from the centromeric regions and the overall decrease of recombination frequency observed in Rb heterozygotes [57],[85].
The results presented here show that the unsynapsed regions of trivalents incorporate many of the proteins related to MSUC, such as γH2AX, ATR and ubiH2A [17],[18] and some markers that have been previously reported only in association with MSCI, such as SUMO-1 and XMR [27],[30], supporting the idea that MSCI could be a particular case of MSUC [17],[18].
Our study on Rb heterozygotes reveals further interesting features of the MSUC process. We found that during early pachytene, some unsynapsed regions do not exhibit either γH2AX or ATR signals. This labeling is especially striking in those trivalents in which one of the open acrocentrics incorporates these markers while the other does not (see Figure 3D). This absence of either γH2AX or ATR signals might be due to a limited availability in the meiocytes of factors triggering MSUC and MSCI, like BRCA1 and ATR, as recently suggested [33],[36]. However, alternative explanations could be formulated taking into account that: 1), unlabeled unsynased chromosome segments are found in cells with either a high or a low number of open trivalents; 2), we never observed MSCI to be hampered in the sex chromosomes. Since the absence of either γH2AX or ATR labeling on some unsynapsed regions is mainly found at early pachytene, we favor the interpretation that unlabeled chromatin could represent chromosomal regions that are about to synapse and/or are asynaptic but MSUC is not initiated yet. In our model, asynapsis could not be extensive enough to exhaust MSUC/MSCI triggering factors; asynapsis in each trivalent affects just a short chromosome length, thus the total amount of unsynapsed chromatin in Rb heterozygous mice is lower than in other mouse models [33],[36]. However, given the physiological interdependence of spermatocytes in the seminiferous epithelium provided by the presence of intercellular bridges [86]–[88], it is also likely that cytoplasmic flux could compensate the mRNA/protein levels of MSUC components among different cells, buffering the effect of extensive asynapsis in some spermatocytes. These facts could determine the success of spermatocytes to have a normal MSUC/MSCI performance during the first prophase and will serve to avoid stage IV pachytene apoptosis.
Our study also adds new clues to the understanding of the sequence of initiation and spreading of chromatin modifications involved in MSUC. H2AX phosphorylation detected at late zygotene was the first modification found in unsynapsed chromatin. This was followed by the accumulation of ubiH2A, SUMO-1, XMR and finally ATR on these regions during the early-mid pachytene transition. Thus, we suggest that the modifications of the chromatin involved in MSUC occur in at least two phases (Figure 8). The first phase initiates with the phosphorylation of H2AX, resulting in chromatin silencing at leptotene/zygotene. The second phase starts at early-mid pachytene with a second round of chromatin modifications, probably driven by the persistence of ATR at unsynapsed AEs, and it involves the incorporation of ubiH2A, SUMO-1, XMR, and finally ATR into unsynapsed chromatin. Whether it also involves other histone replacements and/or modifications, such as histone H3.1 and H3.2 replacement by H3.3 and H3, and H4 methylation [26], or the incorporation of other specific proteins or RNAs, remains to be determined.
Our analysis of the temporal appearance and localization of the proteins involved in MSUC has shown that ATR starts to spread over the chromatin of unsynapsed trivalents only at mid-pachytene, after the massive accumulation of γH2AX, while ubiH2A, SUMO-1, and XMR accumulate throughout early pachytene. Previous studies have suggested that ATR is involved in phosphorylating H2AX on the surrounding chromatin at late zygotene [17],[21] and that XMR and SUMO-1 accumulate on the sex body during early pachytene [27],[30],[70]. Although the pattern of appearance of some of these proteins is not completely established and discrepancies have been reported by different authors [17],[21],[30],[70],[89],[90], the comparison of these studies with our results suggests that: 1) the incorporation of many MSUC-related factors is delayed in Rb heterozygotes compared to homozygotes; and 2) our cytological approach, and previous studies [89],[91], are not completely congruent with the role of ATR in phosphorylating H2AX at late zygotene. Since we cannot rule out that undetectable amounts of ATR are present in the unsynapsed chromatin at late zygotene, other methodological approaches would be necessary to confirm this issue.
Finally, our results indicate that in mouse MSUC is triggered during zygotene-early pachytene and that desynapsing LEs at diplotene do not incorporate MSUC markers, even if they are adjacent to regions that have remained unsynapsed during pachytene. This differs from the recently reported dynamics of sex chromosome inactivation in chicken females, in which two waves of H2AX phosphorylation, one at zygotene and other one at late pachytene, have been detected [92]. These differences in MSUC dynamics open interesting questions in an evolutionary context.
Meiotic failure has been postulated as one of the main causes of infertility in organisms bearing chromosomal rearrangements. Several models have been proposed to explain this phenomenon, including the alteration of transcriptional activity of autosomes and sex chromosomes [1], [17], [34], [93]–[95], the impairment of synapsis and recombination progression [8],[42],[45],[75],[82],[96], the alteration of nuclear architecture during prophase-I [43], and the incorrect orientation and segregation of chromosomes during meiotic divisions [39],[49],[50].
Current models postulate the existence of a pachytene checkpoint that monitors synapsis and/or recombination progression [2],[4],[5]. Pachytene arrest resulting from asynapsis has been proposed to occur as a consequence of MSUC through the inactivation of genes that are crucial to meiotic progression [17]. Additionally, it has been suggested that sequestration MSUC-related proteins like BRCA1 and ATR resulting from an excess of asynaptic chromosomes might prevent their relocation to the sex chromosomes, hampering MSCI initiation in males [1],[33] and an extensive MSUC response in females [36]. The subsequent inability to inactivate the sex chromosomes has been proposed as a primary cause of spermatocyte apoptosis in a variety of mouse models [1],[33]. The presence of many open trivalents in our model does not result in sequestration of repair factors such as ATR on unsynapsed autosomal regions, allowing the correct progression of MSCI. These results indicate that in our model asynapsis per se could not be sufficient to trigger pachytene arrest. This agrees with recent reports on human [64],[80],[81] and mouse meiosis [1],[33],[36] indicating that cells can “tolerate” a limited degree of asynapsis. Therefore, it seems likely that there is not an stringent synapsis-specific checkpoint acting during pachytene in mouse and that MSUC involvement in triggering a checkpoint during prophase-I through MSCI hampering could be limited to extreme asynaptic situations.
Nevertheless, we consider important to stress that the impairment of the meiotic progression of spermatocytes with synaptic defects could still rely on the deregulation of gene expression caused by MSUC. In this sense, MSUC effects would greatly depend on the number and/or nature of genes that are transcriptionally inactivated [17]. In Rb heterozygotes, the unsynapsed segments comprise the pericentromeric heterochromatin-rich regions and euchromatic regions meager in genes, most of which might not be critical for meiosis progression and subsequent spermiogenesis. However, while MSUC has little effect in determining pachytene arrest in this model, it is likely that the effect could be much more relevant in other models.
We found that in Rb heterozygotes meiotic failure occurs mainly during meiotic divisions, as we recorded a high proportion of apoptotic cells at stage XII of the seminiferous epithelium and very few TUNEL-positive pachytene spermatocytes. We are aware that apoptotic pachytene cells are very rapidly removed and difficult to document by TUNEL [50]. On the other hand, metaphase apoptotic cells may be difficult to eliminate from the seminiferous epithelium, causing and overestimation of cell dead at these stages [47]. However, our result are in agreement with previous reports showing that in Rb heterozygotes bearing trivalents or complex rings cell death is mainly found during meiotic divisions [47]–[49] while cell death mainly occurs during prophase-I in Rb heterozygotes bearing chromosome chains [41], [46]–[48]. Furthermore, the absence of massive cell death at the pachytene stage is also supported by our previous studies [39],[40], which showed only a slight reduction of the number of this type of spermatocytes from stage I to XI of the cycle of the seminiferous epithelium. This could account for the elimination of those spermatocytes with a high number of open trivalents, whereas those that have one to four open trivalents might be able to bypass pachytene arrest and proceed to further stages. Therefore, meiotic failure in our Rb heterozygotes seems to rely mainly on the action of checkpoints during metaphase I and II [39],[40],[45],[49]. Trivalents may have difficulties in achieving a correct orientation on the meiotic spindle, determining a delay of anaphase initiation that would lead to cell degeneration [7],[49],[97] and subsequent reduction of fertility.
Paradoxically, despite the reduced fertility of heterozygous mice, Rb translocations are very frequent in wild populations [41],[98], spread rapidly [99],[100] and represent one of the main causes of karyotype evolution in mammals [101]. We propose that the circumvention of pachytene arrest even in the presence of chromosome regions subjected to MSUC, as demonstrated in the present study, could contribute to increasing the chances of many spermatocytes to reach meiotic divisions and to differentiate into viable sperm. Although substantial cell death is produced at the metaphase stage (up to 63%), the chances of producing viable gametes are still much higher than if a more stringent pachytene checkpoint were able to eliminate up to 87% (Table 1) of pachytene spermatocytes bearing unsynapsed chromosomes.
In an evolutionary context, it must be stressed that when a chromosomal rearrangement arises in a natural population, the rearranged chromosomes must still pair, synapse, recombine and segregate from their cognate homologues. Therefore, the possibility that a chromosomal rearrangement will spread into a population would greatly depend on the meiotic defects it may cause in the heterozygotes. Thus, while Rb rearrangements may have a relatively mild effect on mouse pachytene progression, for other chromosomal rearrangements and organisms, this model cannot be applied [102].
Heterozygous Robertsonian mice (2n = 32, eight Robertsonian chromosomes in a heterozygous state) were generated by mating females of the laboratory strain CD1 (2n = 40, all acrocentric chromosomes) and males of the Milano II race (2n = 24, eight pairs of Robertsonian metacentrics in a homozygous state, Rb (2.12), Rb (3.4), Rb (5.15), Rb (6.7), Rb (8.11), Rb (9.14), Rb (10.13), Rb (16.17). Six three-month old male mice were analyzed. Mice were maintained at 22°C with a light/dark cycle of 12/12 hours and fed ad libitum. Procedures involving the use of the mice were approved by the animal ethics committees of the Faculty of Medicine, University of Chile, and the University of Pavia (Italy).
Spermatocyte spreads and squashes were obtained following the procedures described by Peters et al. [103] and Page et al. [104]. The slides were placed in PBS and incubated with the following primary antibodies: mouse anti-SYCP3 1∶100 (Abcam, Ab12452); rabbit anti-SYCP3 1∶100 (Abcam, Ab15093); rabbit anti-SYCP1 1∶100 (Abcam, Ab15087); rabbit anti RAD51 1∶50 (Calbiochem, PC130); mouse anti-phospho-histone H2AX (Ser139) 1∶1000, clone JBW301 (Upstate, 05–636); goat anti ATR 1∶80 (Santa Cruz Biotechnology, sc-1887); mouse anti ubiquityl-histone H2A 1∶15, clone E6C5 (Upstate, 05–678); mouse anti GMP-1 (SUMO-1) 1∶50 (Zymed, 33–2400); mouse RIK2D3 1∶100 that recognizes the XMR protein in the testis [27], kindly provided by Denise Escalier (Université Paris 5, Paris, France). After rinsing in PBS, the slides were incubated with appropriate secondary antibodies diluted 1∶100 in PBS: FITC-conjugated donkey anti-rabbit IgG, FITC-conjugated donkey anti-mouse IgG, TR-conjugated donkey anti-mouse IgG and FITC-conjugated donkey anti-goat IgG. Slides where then stained with 1 µg/ml DAPI. After a final rinse in PBS, the slides were mounted with Vectashield. Observations were made in a Nikon (Tokyo, Japan) Optiphot or an Olympus BX61 microscope equipped with epifluorescence optics and the images were photographed on DS camera control unit DS-L1 Nikon or captured with an Olympus DP70 digital camera. All images were processed with Adobe Photoshop CS software.
Immunolabeled spermatocytes were observed in an Olympus BX61 microscope equipped with a motorized Z-axis, epifluorescence and an Olympus DP70 digital camera. A collection of optical sections were captured using the analiSYS software (Soft Imaging System, Olympus). Images were subsequently analyzed and processed using the public domain software ImageJ (National Institutes of Health, United States; http://rsb.info.nih.gov/ij), and the output video files were edited with VirtualDub (VirtualDub, http://www.virtualdub.com).
The right testis of three mice were fixed in Bouin's fluid and embedded in paraffin wax. Five-micrometer serial transverse cross-sections were made and at least four serial sections per testis were mounted on each glass slide. One slide was stained by the periodic-acid-Schiff (PAS) reaction and counterstained with haematoxylin to identify the stages of seminiferous epithelium according to Oakberg [105]; the other slide was processed with the terminal deoxynucleotidyl transferase-mediated dUTP nick end-labelling (TUNEL) method, using an ApopTag Plus Peroxidase In Situ Apoptosis Kit (Chemicon-Millipore, Billerica, USA), according to the manufacturer's instructions. Positive and negative controls were also set up. The positive controls were established using the slides contained in the same kit and following the manufacturer's instructions. For the negative controls, sections were processed without TdT enzyme in the labelling reaction mix. The sections were counterstained with 0.5% (w/v) methyl green for 10 min at room temperature. For each animal testis, 100 cross-sectioned tubules were scored to evaluate the frequency of apoptotic tubules. A cross-section of a tubule was considered apoptotic when three or more TUNEL-positive spermatocytes were present within the seminiferous epithelium [39],[49]. The percentage of TUNEL positive cells was calculated taking into account the total number of spermatocytes per tubule section. Abercrombie's correction was applied to all cell counts [106].
We analyzed 724 and 415 spermatocytes from two three month-old heterozygous Robertsonian mice. The synapsed condition of heterologous region of Robertsonian trivalents was determined by morphological analysis identifying chromosomes with SYCP3 and the presence or absence of γH2AX positive signal in the chromatin. The data obtained from each mouse in each prophase I stage were summarized. Statistical significance between mice was assessed by the one way analysis of variance (ANOVA), followed by Tuckey post test. A Z test for two proportions was used to compare the number of spermatocytes between late pachytene, early diplotene and middle/late diplotene. In both statistical analyses a p value<0.05 was considered statistically significant with a confidence interval of 95%.
|
10.1371/journal.pgen.1005021 | HDAC4-Myogenin Axis As an Important Marker of HD-Related Skeletal Muscle Atrophy | Skeletal muscle remodelling and contractile dysfunction occur through both acute and chronic disease processes. These include the accumulation of insoluble aggregates of misfolded amyloid proteins that is a pathological feature of Huntington’s disease (HD). While HD has been described primarily as a neurological disease, HD patients’ exhibit pronounced skeletal muscle atrophy. Given that huntingtin is a ubiquitously expressed protein, skeletal muscle fibres may be at risk of a cell autonomous HD-related dysfunction. However the mechanism leading to skeletal muscle abnormalities in the clinical and pre-clinical HD settings remains unknown. To unravel this mechanism, we employed the R6/2 transgenic and HdhQ150 knock-in mouse models of HD. We found that symptomatic animals developed a progressive impairment of the contractile characteristics of the hind limb muscles tibialis anterior (TA) and extensor digitorum longus (EDL), accompanied by a significant loss of motor units in the EDL. In symptomatic animals, these pronounced functional changes were accompanied by an aberrant deregulation of contractile protein transcripts and their up-stream transcriptional regulators. In addition, HD mouse models develop a significant reduction in muscle force, possibly as a result of a deterioration in energy metabolism and decreased oxidation that is accompanied by the re-expression of the HDAC4-DACH2-myogenin axis. These results show that muscle dysfunction is a key pathological feature of HD.
| Huntington’s disease (HD) is a neurodegenerative disorder in which the mutation results in an extra-long tract of glutamines that causes the huntingtin protein to aggregate. It is characterized by neurological symptoms and brain pathology, which is associated with nuclear and cytoplasmic protein aggregates and with transcriptional deregulation. Despite the fact that HD has been recognized principally as a neurological disease, there are multiple studies indicating that peripheral pathologies including cardiac dysfunction and skeletal muscle atrophy, contribute to the overall progression of HD. To unravel the cause of the skeletal muscle dysfunction, we applied a wide range of molecular and physiological methods to the analysis of two well established genetic mouse models of this disease. We found that symptomatic animals developed muscle dysfunction characterised by a change in the contractile characteristics of fast twitch muscles and a decrease in twitch and tetanic force of hindlimb muscles. In addition, there is a significant decrease in the number of motor units innervating the EDL muscle, and this motor unit loss progresses during the course of the disease. These changes were accompanied by the re-expression of contractile transcripts and markers of muscle denervation such as the HDAC4-Dach2-myogenin axis, as well as the apparent deterioration in energy metabolism and decreased oxidation. Therefore, we conclude, that the HD-related skeletal muscle atrophy is accompanied by progressive loss of functional motor units.
| Huntington’s disease (HD) is neurodegenerative disorder in which the mutation results in the increased length of a tract of glutamines that causes the huntingtin protein (HTT) to aggregate. It is characterized by neurological symptoms and neurodegeneration that is prominent in the basal ganglia and cerebral cortex [1]. In mammals, HTT is expressed in many tissues and organs [2,3] and is involved in many critical cellular processes such as transcription, protein trafficking and vesicle transport [4]. HTT is predicted to form an elongated superhelical solenoid structure due to a large number of HEAT motifs, suggesting that it plays a scaffolding role for protein complex formation [5]. More than 200 HTT interacting partners have been identified which can be classified according to their function and include proteins that are involved in gene transcription, intracellular signalling, trafficking, endocytosis, and metabolism [6]. In mice, HTT deletion is embryonically lethal, leading to defects in all germ layers [7]. The process of mutant HTT self-aggregation is an early event in HD progression which may lead to the pathological features of HD. Insoluble polyQ aggregates are a hallmark of HD pathology and can be detected at the presymptomatic stage in HD post mortem brain [8] and can also be found in many non-central nervous system tissues in HD mouse models [9,10]. Recently, it has been shown that there are a number of factors to indicate that HD patients experience an HD-related heart pathology [11]. A recent study in the R6/2 and HdhQ150 knock-in mouse models showed that the HD-related cardiomyopathy is caused by altered central autonomic pathways and is not due to the accumulation of toxic HTT aggregates as had previously been proposed [12–14]. This was accompanied by the re-expression of foetal genes, apoptotic cardiomyocyte loss and a moderate degree of interstitial fibrosis [13]. There is also growing evidence to indicate that peripheral pathologies such as weight loss and skeletal muscle atrophy may not be a consequence of neurological dysfunction or neurodegeneration and might make a significant contribution to the disease presentation and progression [15]. Therefore it is important to identify the peripheral abnormalities that may contribute to disease progression as they may present targets for new treatment strategies.
To date, our knowledge of skeletal muscle pathology in HD is very limited as outlined in a recent review [16]. A case-study report showed that a semi-professional marathon runner (43 CAGs) developed signs of a slowly progressing myopathy with elevated creatine kinase levels many years before first signs of chorea were detected. A muscle biopsy revealed a mild myopathy with mitochondrial pathology including a complex IV deficiency [17]. Transcriptional deregulation is a typical feature of HD pathology in the brain [18] and a similar transcriptional profile in skeletal muscles (quadriceps) from R6/2 mice, HdhQ150 homozygous knock-in mice and HD patients has been identified that was consistent with a transition from fast-twitch to slow-twitch muscle fiber types [19]. Some of the molecular and physiological changes in HD muscles can even be detected in pre-symptomatic HD individuals [20–22]. At the molecular level, mitochondrial dysfunction, PPAR alpha signalling and HSF1 activation were identified as major players in skeletal muscle HD-related pathology [23,24]. Proof of concept studies have suggested that the progression of disease onset could be delayed and lifespan extended by improving muscle function in HD mouse models [25,26]. However, many aspects of HD neuromuscular transmission and muscle physiology remain unanswered and need to be studied more extensively. In this study, we have investigated the molecular and pathological features of the skeletal muscle dysfunction that develops with disease progression in mouse models of HD at the physiological level.
To test the hypothesis that mutant HTT leads to the skeletal muscle atrophy through compensatory remodelling, including the HDAC4-myogenin axis, we used two well-established HD mouse models. R6/2 mice are transgenic for a mutated N-terminal exon 1 HTT fragment [27], while the HdhQ150 mice have an expanded CAG repeat knocked-in to the mouse huntingtin gene (Htt) [28,29], which is partially mis-spliced with the result that these mice express mutant versions of both an exon 1 HTT and a full length HTT protein [30]. For R6/2 mice, we studied skeletal muscle abnormalities at symptomatic (12 weeks) and end-stage (14 weeks) disease while HdhQ150 homozygotes were compared to wild type (WT) at 22 months (end-stage disease).
We began by quantifying the change in the weight of various skeletal muscles with disease progression (Fig. 1). There was a significant decrease in the muscle mass of all muscles examined types including quadriceps, gastrocnemius/plantaris complex (G/P), tibialis anterior (TA), extensor digitorum longus (EDL) and soleus at 12 weeks of age in R6/2 mice (Fig. 1A). The HdhQ150 knock-in model showed a strikingly similar muscle mass decrease at 22 months of age (Fig. 1B). We have previously shown that HTT inclusions can be detected throughout the periphery of the R6/2 and HdhQ150 mouse models by immunohistochemistry [10,31]. More recently, we have developed the seprion-ligand ELISA, a highly quantitative method with good statistical power that can be used to measure changes in aggregate load that occur in vivo in response to pharmacological or genetic manipulations [9]. Using this assay, we were able to detect mutant HTT aggregates in the different skeletal muscle types from either R6/2 at 12 and 14 weeks of age (Fig. 1C) or HdhQ150 (Fig. 1D) mice at 22 months. Surprisingly, we observed a higher accumulation of toxic aggregates in the TA muscles in comparison to G/P and quadriceps in both HD mouse models (Fig. 1C and D). Subsequently, we used Taqman qPCR to demonstrate that the expression of the exon-1 HTT mRNA was uniform in the different types of R6/2 skleletal muscles at 12 (S1A Fig.) and 14 weeks of age (S1B Fig.) or HdhQ150 muscle at 22 months (S1C Fig.) and was not therefore the reason for the increased level of aggregates in the TA muscles in both of these models.
Next, to determine whether HD mice develop functional contractile abnormalities, we undertook isometric muscle tension experiments on TA and EDL muscles of the R6/2 mouse model and their respective WT littermates, as previously described [32], at 12 and 14 weeks of age. Both of these muscles exhibited a significant degree of atrophy as indicated above (Fig. 1) and significant alterations in their contractile function were revealed (Fig. 2A and B). The time taken for muscles to reach maximum force (time-to-peak, TTP) was significantly higher in the EDL at both time points in R6/2 mice, while TA showed a normal TTP in R6/2 mice in comparison to their WT littermates at both time points (Fig. 2A). The time taken for muscles to relax to a half of the maximum force (1/2 relaxation time) following a twitch-stimulus was significantly prolonged in both EDL and TA muscles in the R6/2 mice at 12 and 14 weeks of age in comparison to their WT littermates (Fig. 2B).
In order to corroborate the physiological findings described above, we used Taqman qPCR to quantify contractile transcipt levels that are representtive of the fast or slow type fibers. Given that global transcriptional dysregulation is a pathogenic characteristic of HD, we first performed a systematic study to identify suitable reference genes for use in the expression analysis of different skeletal muscles types from HD mouse models. We used the geNorm™ Housekeeping Gene Selection Mouse Kit and associated software to identify the three most stably expressed genes in specific muscles from R6/2 (S2 Fig.) and HdhQ150 (S3 Fig.) mice. Our relative quantification methods then used the geometric mean of these three selected reference genes for normalization, to accurately determine gene expression levels in WT, R6/2 and HdhQ150 skeletal muscle tissue. We found a significant up-regulation of slow-type contractile proteins such as Tnn1 (Troponin 1, slow) and Myh7 (myosin heavy light chain 7) in TA, EDL and G/P muscles from both HD mouse models (Fig. 2C and E). Consequently, a pronounced down-regulation of the fast-type contractile proteins like Tnn3 (Troponin3, fast) and Myh2 (myosin heavy light chain 2) was also observed in TA, EDL and G/P muscles from both HD mouse models (Fig. 2D and F). These findings indicate that there is a loss of fast-twitch muscle fibres in the EDL and TA of both models. Subsequently, we determined the expression levels of additional genes that are attributed to be altered in fast to slow twitch remodelling. TEA domain (TEAD) transcription factors and their co-activators serve important functional roles during embryonic development as well as in striated muscle gene expression and muscle regeneration [33–36]. It has been shown that striated muscle-restricted TEAD-1 expression induced a transition toward a slow muscle contractile protein phenotype, slower shortening velocity with longer contraction and relaxation times in the adult fast twitch EDL muscles [33]. We found that Tead-2 (TEA domain family member 2) (Fig. 3B) and Tead-4 (TEA domain family member 4) (Fig. 3D) were significantly up-regulated in the all diseased HD muscles in both mouse models, while Tead-1 (TEA domain family member 1) (Fig. 3A) and Tead-3 (TEA domain family member 3) (Fig. 3C) transcripts remained un-changed. The transcriptional activity of TEAD family members is highly dependent on the presence of their co-activators [37–39] and therefore, we used Taqman-qPCR to asses their transcriptional profile in the HD diseased muscles. We established that Vgll-2 (vestigial related factor 2) (Fig. 3E), Vgll-3 (vestigial related factor 3) (Fig. 3F), Vgll-4 (vestigial related factor 4) (Fig. 3G) and Yap-65 (Yes associated protein 65) (Fig. 3H) were significantly up-regulated in the TA, EDL and G/P muscles of R6/2 and HdhQ150 mice.
We also determined the maximum muscle force of TA and EDL muscles by physiological determination of single twitch (Fig. 4A) and tetanic force (Fig. 4B) in R6/2 mice. Twitch and tetanic force recordings showed that R6/2 TA muscles at 12 and 14 weeks of age were approximately 50% weaker than in their WT littermates. Moreover, physiological assessment of functional motor unit survival, revealed that there was a significant loss of motor units in R6/2 mice, which progressed from a 25% reduction at 12 weeks to over 60% loss at 14 weeks, compared to WT mice (Fig. 4C and D). These findings suggest that there is also likely to be a significant degeneration of spinal motor neurons during this period.
Physiological changes in skeletal muscle are often caused or associated with metabolic alterations. Therefore, we analysed two aspects of metabolism in the EDL and TA muscles. First we estimated the steady-state concentration of the major components of energy equilibrium that include creatine metabolites and adenine nucleotides. Analysis of ATP, phosphocreatine and related metabolites revealed a substantial depletion of the energy equilibrium in EDL and TA in both HD mouse models (Fig. 5A and B and Table 1). The phosphocreatine/creatine ratio as well as ADP and AMP levels were significantly decreased (Table 1) in both types of muscle in R6/2 and HdhQ150 mice. Besides the energy equilibrium, the total pools of the adenine nucleotides were also consistently depleted (Fig. 5 and Table 1) while changes in the redox status were less evident (Fig. 5B). A similar pattern of metabolic changes was found in the slow type soleus muscles of the HD mouse models (S1 Table). The second metabolic aspect concerned the evaluation of the substrate preference shift in these muscles. To address this, glycolysis was assessed by measuring the 13C alanine enrichment while the changes in Krebs cycle were estimated based on 13C glutamate levels after administration of 1–13C glucose. The analysis revealed that the EDL muscle showed a slower glycolytic flux from exogenous glucose as well less oxidation of glucose in both HD mouse models (Fig. 5C and D) while the TA the muscle remained unchanged (Fig. 5C and D).
To further examine the degree of skeletal muscle pathology, we determined the expression levels of additional genes that are typically altered in atrophied muscles. We found that AChR (nicotinic acetylcholine receptor) (Fig. 6A) was significantly up-regulated in all muscle types examined from mouse models. Usually, muscle atrophy is accompanied by a significant up-regulation of caspases [40]. Indeed, we found Caspase8 transcripts significantly up-regulated in the aged HdhQ150 muscles but not in those from the R6/2 mice (Fig. 6B). Similarly, we found Foxo-3 (Forkhead box O3) transcripts (Fig. 6D) to be markedly up-regulated, while Mck (muscle creatinine kinase) mRNA (Fig. 6C) was decreased in all of the muscle types examines from the R6/2 and HdhQ150 mice.
Previous studies have established HDAC4 as a critical factor that connects neural activity to the muscle remodelling program [41,42] and inactivation of HDAC4 suppressed denervation-induced muscle atrophy while increasing re-innervation [43–45]. HDAC4 up-regulation was found to be significantly greater in patients with rapidly progressive ALS (amyotrophic lateral sclerosis) and was negatively correlated with the extent of muscle re-innervation and functional outcome [46]. Similarly, an increased level of HDAC4 has been found in SMA (spinal muscular atrophy) model mice and in SMA patient muscles [47]. Consistent with this, we found that Hdac4 transcripts were significantly up-regulated in the TA, EDL an G/P muscles in the HD mouse models as compared to WT littermates (Fig. 7A). Hdac4 up-regulation was accompanied by down-regulation its direct target Dach2 (Dachshund homolog 2) (Fig. 7B) that is a negative regulator of Myogenin. Consequently, we observed a very significant up-regulation of Myogenin (Fig. 7C) and its direct target Fbxo32 (F-box only protein 32) (Fig. 7D) in HD-related muscle atrophy. Thus, one might conclude that HD-related skeletal muscle atrophy displays the typical characteristics of a denervation like muscle phenotype.
Skeletal muscle is the most abundant tissue in the mammalian body accounting for approximately 40% of body weight, and is composed of multinucleated fibers that contract to generate force and movement. In addition, skeletal muscle possesses a remarkable ability to regenerate, and can go through rapid repair following severe damage caused by exercise, toxins or diseases. The atrophy caused by degeneration of myofibers and their replacement by fibrotic tissue is the major pathological feature in many genetic muscle disorders [48,49]. Skeletal muscle atrophy in HD is a comorbidity that is observed in catabolic disease and other conditions like cancer, congestive heart failure, sepsis, denervation and disuse [16,50]. Under normal physiological conditions muscle function is orchestrated by a network of intrinsic hypertrophic and atrophic signals linked to the functional properties of the motor units that are likely to be imbalanced in HD.
In this study we aimed to provide a broad spectrum of experimental insights into skeletal muscle-associated abnormalities that develop in the R6/2 transgenic and HdhQ150 knock-in HD mouse models, in which mutant Htt is expressed under the control of the Htt promoter. We found significant alterations at the physiological level in the contractile function of the EDL and TA R6/2 muscles at 12 and 14 weeks of age. The time taken for muscles to reach maximum force (time-to-peak, TTP) and time taken for muscles to relax to half the maximum force (1/2 relaxation time) were significantly changed in R6/2 mice, indicative of a loss of fast-twitch muscle fibres in the EDL and TA muscles. In addition, transcriptional deregulation is a typical feature of HD pathology in the brain [18] and a similar transcriptional profile in the skeletal muscles (quadriceps) from R6/2 mice, HdhQ150 homozygous knock-in mice and HD patients has been identified that was consistent with a transition from fast-twitch to slow-twitch muscle fiber types [19]. Although immunohistochemistry suggested that both type I (slow) and II (fast) muscles were atrophic [31], there were more type I fibers in the R6/2 skeletal muscles. Hence, a conversion of type II to type I fibers has occurred during the process of muscle atrophy [51], most likely as a result of the loss of motor units innervating type II fibres.
Indeed, our physiological findings were also supported by the quantification of contractile transcipt levels that are representative of fast or slow type fibers [52,53]. We found a significant up-regulation of the genes encoding slow-type contractile proteins like Tnn1 and Myh7 in the TA, EDL and G/P muscles of both HD mouse models. Consequently, the transcripts for fast-type contractile proteins like Tnn3 and Myh2 (myosin heavy light chain 2) were markedly down-regulated in these muscles. This was also accompanied by the up-regulation of members of the TEAD family and their co-activators. It is well established that MCAT elements are located in the promoter-enhancer regions of cardiac, smooth, and skeletal muscle-specific genes and play a key role in the regulation of these genes during muscle development and disease [33–36].
Following a significant decrease in the muscle mass of all of the muscle types that were examined in both HD mouse models, we found that the maximum twitch and tetanic force in TA and EDL hind limb muscles of R6/2 mice were significantly reduced at the symptomatic stage, indicative of motor neuron dysfunction in these mice. Moreover, the physiological assessment of functional motor units revealed that there was a progressive loss in the number of functional fewer motor units in the EDL muscle of R6/2 mice, from ∼25% loss at 12 weeks to more than 60% loss at 14 weeks of age, as compared to their WT littermates. This finding is supported by a previous study showing that skeletal muscles of R6/2 mice developed age-dependent denervation-like abnormalities, including reduced endplate area, supersensitivity to acetylcholine, decreased sensitivity to mu-conotoxin and anode-break action potentials [51]. Moreover, the miniature endplate potential (mEPP) amplitude was notably increased while mEPP frequency was significantly reduced in the R6/2 mice [51]. In contrast, the same study showed that severely affected R6/2 mice developed a progressive increase in the number of motor endplates that fail to respond to nerve stimulation but there was no constitutive sprouting of motor neurons, even in severely atrophic muscles [51]. In fact there was no age-dependent loss of regenerative capacity of motor neurons in R6/2 mice [51]. In line with our findings, a previous study showed that the action potentials in diseased muscles were more easily triggered and prolonged than in WT littermates. Furthermore, the expression of the muscle chloride channel (ClC-1) and Kcnj2 (Kir2.1 potassium channel) transcripts were significantly reduced and defects in mRNA processing were detected [54]. These dependent denervation-like abnormalities and the highly developed muscle atrophy could be partially explained by sciatic nerve degeneration [55]. A significant decrease in the axoplasm diameter of myelinated neurons and increased number of degenerating myelinated fibers were observed; although the myelin thickness and unmyelinated fiber diameter were not affected [55]. This might be also explained by the profound localisation of mutant HTT to the neuromuscular junctions as was previously published [56]. However, it is likely that the skeletal muscle denervation-like phenotype is linked to not only to spinal motor neuron loss, but also CNS dysfunction, as previously published pathological features in relevant brain regions in both mouse models might support this hypothesis [29]. In HD patient brains, a recent meta-analysis of morphometric MRI found degenerative changes in the amygdala and insular cortex, even in the prodromal form on the disease [57].
It is well established that pronounced skeletal muscles atrophy is accompanied by altered metabolism, reviewed in [58] and our demonstration that the energy equilibrium is depleted in the skeletal muscles of HD mouse models is an important and a novel finding. The decrease in the phosphocreatine/creatine ratio and ATP/ADP ratio directly translates into lower values for phosphorylation potential and the free energy of ATP hydrolysis that might decrease the efficiency of the muscle contraction [52]. Interestingly, our study is in line with previous observations in clinical settings, as muscle ATP/phosphocreatine and inorganic phosphate levels were significantly reduced in both symptomatic and presymptomatic HD subjects [20]. In addition, HD subjects displayed a deficit in mitochondrial oxidative metabolism that might support a role for mitochondrial dysfunction as a key factor involved in the HD-related muscle pathogenesis [21]. An important aspect of this study is the identification of the mechanism underlying the decreased energy equilibrium. One possible explanation is a lack of the trophic effect of nerve stimulation [51,55] that may down-regulate the expression of energy related proteins including factors responsible for mitochondrial biogenesis [58]. Consequently, this process might lead to a decreased oxidative and substrate phosphorylation efficiency translating into a shift of energy equilibrium. Alternatively, a direct local effect of genetic alterations in the skeletal muscle that are likely to be driven by mutant HTT directly [56,59] may deregulate energy metabolism. Interestingly, a similar metabolic profile has been found in mouse embryonic stem cell (mESC) lines: Htt(−/−), extended poly-Q (Htt-Q140/7) and wild-type mESCs (Htt-Q7/7) [60]. One might conclude that the HD-related skeletal muscle atrophy is caused by loss of function in HD mouse models.
At the pathological level, the HD-related skeletal muscle atrophy was accompanied by the deregulation of AChR, Foxo-3 and Mck, typical markers of muscle atrophy and denervation in both HD mouse models [48,61]. It has been also shown that inactivation of HDAC4 suppresses denervation-like induced muscle atrophy while increasing re-innervation [41,42,45]. These findings highlight a central regulatory role of HDAC4 in activity-dependent muscle remodelling. HDAC4 up-regulation was significantly greater in patients with rapidly progressive ALS (amyotrophic lateral sclerosis) and was negatively correlated with the extent of muscle re-innervation and functional outcome [46]. An increased level of HDAC4 has been found in SMA (spinal muscular atrophy) model mice and in SMA patient muscles [47]. We found an up-regulation of the HDAC4-Dach2-myogenin axis in both HD mouse models that might be indicative of a similar activity dependent muscle remodelling in HD to that observed in ALS or SMA.
In summary, mutant HTT results in the rapid development of pathological features that would be expected to lead to a skeletal muscle contractile dysfunction e.g. leading to fast to slow fibre twitch with aberrant deregulation of contractile protein transcripts and their up-stream transcriptional regulators. In addition, HD mouse models develop a notable decrease in the twitch and tetanic force of skeletal muscles and pronounced loss of motor units, which may contribute to deterioration of energy metabolism and decreased oxidation that is accompanied by the re-expression of HDAC4-Dach2-myogenin axis (Fig. 8). Importantly, our data connects gene alterations with physiological function in HD-related skeletal muscles atrophy and might have a therapeutic potential. Recently, two key signalling pathways, i.e. those driven by insulin like growth factor (IGF) and growth differentiation factor −8 (GDF-8), have emerged to be potent regulators of skeletal muscle size. In addition, our metabolomic profile of skeletal muscles in HD mouse models might be served as a biomarker platform for prospective pre- and clinical trials.
All experimental procedures performed on mice were conducted under a project licence from the Home Office and approved by the King's College London Ethical Review Process Committee.
Hemizygous R6/2 mice were bred by backcrossing R6/2 males to (CBA x C57BL/6) F1 females (B6CBAF1/OlaHsd, Harlan Olac, Bicester, UK). HdhQ150 homozygous mice on a (CBA x C57BL/6) F1 background were obtained by intercrossing HdhQ150 heterozygous CBA/Ca and C57BL/6J congenic lines as described previously [29]. All animals had unlimited access to water and breeding chow (Special Diet Services, Witham, UK), and housing conditions and environmental enrichment were as previously described [62]. Mice were subject to a 12-h light/dark cycle. All experimental procedures were performed according to Home Office regulations.
Genomic DNA was isolated from an ear-punch. R6/2 and HdhQ150 mice were genotyped by PCR and the CAG repeat length was measured as previously described [9] and listed in S2 Table. Dissected tissues were snap frozen in liquid nitrogen and stored at −80°C until further analysis.
Total RNA from skeletal muscles was extracted with the mini-RNA kit according to the manufacturer instructions (Qiagen). The reverse transcription reaction (RT) was performed using MMLV superscript reverse transcriptase (Invitrogen) and random hexamers (Operon) as described elsewhere [63]. The final RT reaction was diluted 10-fold in nuclease free water (Sigma). All Taqman qPCR reactions were performed as described previously [64] using the Chromo4 Real-Time PCR Detector (BioRad). Stable housekeeping genes for qPCR profiling of various skeletal muscles for HD mouse models were determined using the Primer Design geNorm Housekeeping Gene Selection Mouse Kit with PerfectProbe software. Estimation of mRNA copy number was determined in triplicate for each RNA sample by comparison to the geometric mean of three endogenous housekeeping genes (Primer Design) as described [65]. Primer and probe sets for genes of interest were purchased from Primer Design or ABI.
Aggregates were captured in Seprion ligand coated plates (Microsens) and detected using the MW8 mouse monoclonal antibody (1:4000) as described [9].
Mice were injected with glucose-13C subcutaneously as a 20% solution at a dose of 3ml/kg. Two hours after glucose administration, mice were deeply anesthetized with isoflurane and EDL and TA muscles were freeze-clamped in situ with aluminum clamps pre-cooled in liquid nitrogen. Freeze dried muscles were extracted with 0.4 M perchloric acid, extracts were neutralized with 2 M KOH as described previously [66] and analysed by liquid chromatography mass spectrometry [67] using TSQ Vantage triple quadrupole mass detector linked to Surveyor chromatography system. Mass detection was carried out in fragmentation mode (Tandem MS) and 13C isotopic enrichment of fragments containing C3 of alanine or C4 of glutamate were monitored.
Mice were deeply anesthetized with isoflurane and prepared for in vivo analysis of muscle function which was performed as previously described [68]. The distal tendons of the TA and EDL muscles in both hindlimbs were dissected free and attached by silk thread to isometric force transducers (Dynamometer UFI Devices, Welwyn Garden City, UK). The sciatic nerve was exposed and sectioned proximally. The length of the muscles was adjusted for maximum twitch tension. The muscles and nerve were kept moist with saline throughout the recordings and all experiments were carried out at room temperature. Isometric contractions were elicited by stimulating the nerve to TA and EDL using square-wave pulses of 0.02 ms duration at supra-maximal intensity, via silver wire electrodes. Contractions were elicited by trains of stimuli at frequencies of 40, 80 and 100 Hz. The maximum tetanic tension was measured using a computer and appropriate software Pico Technology,Cambridgeshire, UK. The number of motor units innervating the EDL muscles was also determined as previously described [69] by stimulating the motor nerve with stimuli of increasing intensity, resulting in stepwise increments in twitch tension due to successive recruitment of motor axons with increasing stimulus thresholds. The number of stepwise increments was counted to give an estimate of the number of functional motor units (MUNE) present in each muscle. Following recording of isometric tension, the contractile characteristics of EDL and TA muscles were determined. The time to peak (TTP) was calculated by measuring the time taken (ms) for the muscle to elicit peak twitch tension and the half relaxation time (the time taken for the muscle to reach half relaxation from peak contraction) was also calculated. The tetanic contractions were recorded on a Lectromed Multitrace 2 recorder (Lectromed Ltd, UK). All parameters were measured using a computer and Picoscope v5 and v6 software (Pico Technology,Cambridgeshire, UK).
All data were analysed with Microsoft Office Excel and Student's t-test (two tailed) or ONE-WAY ANOVA with Bonferroni post-hoc test.
|
10.1371/journal.pgen.1006401 | Transcription Factors Encoded on Core and Accessory Chromosomes of Fusarium oxysporum Induce Expression of Effector Genes | Proteins secreted by pathogens during host colonization largely determine the outcome of pathogen-host interactions and are commonly called ‘effectors’. In fungal plant pathogens, coordinated transcriptional up-regulation of effector genes is a key feature of pathogenesis and effectors are often encoded in genomic regions with distinct repeat content, histone code and rate of evolution. In the tomato pathogen Fusarium oxysporum f. sp. lycopersici (Fol), effector genes reside on one of four accessory chromosomes, known as the ‘pathogenicity’ chromosome, which can be exchanged between strains through horizontal transfer. The three other accessory chromosomes in the Fol reference strain may also be important for virulence towards tomato. Expression of effector genes in Fol is highly up-regulated upon infection and requires Sge1, a transcription factor encoded on the core genome. Interestingly, the pathogenicity chromosome itself contains 13 predicted transcription factor genes and for all except one, there is a homolog on the core genome. We determined DNA binding specificity for nine transcription factors using oligonucleotide arrays. The binding sites for homologous transcription factors were highly similar, suggesting that extensive neofunctionalization of DNA binding specificity has not occurred. Several DNA binding sites are enriched on accessory chromosomes, and expression of FTF1, its core homolog FTF2 and SGE1 from a constitutive promoter can induce expression of effector genes. The DNA binding sites of only these three transcription factors are enriched among genes up-regulated during infection. We further show that Ftf1, Ftf2 and Sge1 can activate transcription from their binding sites in yeast. RNAseq analysis revealed that in strains with constitutive expression of FTF1, FTF2 or SGE1, expression of a similar set of plant-responsive genes on the pathogenicity chromosome is induced, including most effector genes. We conclude that the Fol pathogenicity chromosome may be partially transcriptionally autonomous, but there are also extensive transcriptional connections between core and accessory chromosomes.
| Eukaryotic genomes are organised. Genomic regions may differ in spatial organisation, chromatin condensation, rate of evolution, GC content, gene expression and transposon density. Many plant pathogenic fungi maintain genomic subcompartments containing specialized genes to facilitate host colonisation. These genes, for example effector genes, show concerted transcriptional up-regulation during infection. An extreme case is the tomato pathogen Fusarium oxysporum f. sp. lycopersici, which carries accessory chromosomes, of which one encodes all effector genes and can be transferred horizontally between strains. We investigated the transcriptional connections between this accessory chromosome and the core genome, particularly with respect to effector gene expression. Several of the transcription factors encoded on this accessory chromosome bind to motifs enriched on accessory chromosomes, suggesting the accessory chromosomes may be partially transcriptionally independent. Only one of these, Ftf1, can induce the expression of effector genes and binds to a motif that is enriched in their promoters. Also Sge1 –a conserved regulator of fungal lifestyle switches and required for infection–can activate the expression of effector genes. Both transcription factors induce a largely overlapping set of genes, including many of the host-induced genes on the accessory chromosome including the effector genes. This demonstrates the existence of extensive transcriptional connections between accessory and core chromosomes.
| Plant pathogenic fungi are genetically adapted to infect their host plant, but are also in a constant arms race with that host to stay virulent. For this, pathogens need to allow accelerated evolution of pathogenicity-related genes, without affecting the function of housekeeping genes. One possibility is to spatially separate these different functional groups of genes into subgenomic compartments with different rates of evolution. One of the fastest evolving determinants of pathogenicity is the effector repertoire. Definitions vary, but a typical effector is a small, secreted protein that affects the interaction between the pathogen and its host. In many plant pathogenic fungi effector genes indeed reside in specific genomic regions, generally distinguished by one or more of the following characteristics: lineage-specific or accessory, rich in transposable elements, a different GC content and/or codon bias from the rest of the genome, depleted for housekeeping genes and associated with particular chromatin modifications [1–3]. Accumulating evidence suggests that these genomic environments evolve more rapidly than the rest of the genome and facilitate adaptation [2,4,5]. In addition, in several fungi these types of regions—or parts thereof—have been shown or suggested to transfer horizontally between different strains or even species [6–11].
Another hallmark of effector genes is a plant specific expression pattern [12–14]. How coordinated expression of effector genes is regulated and how this is related to the specific genomic environment of these genes is poorly understood. For some effector genes it has been shown that their genomic environment is key for regulated expression, through histone modifications [1,15]. On the other hand, a small number of transcription factors required for effector gene expression has been identified. In Ustilago maydis the heterodimer bE/bW, the forkhead transcription factor Fox1, and the zinc finger transcription factors Rbf1 and Mzr1 are involved in transcriptional regulation of pathogenicity-related genes and/or effector genes [16–19]. In Leptosphaeria maculans and Stagnospora nodorum, homologs of StuA, a bHLH (basic helix-loop-helix) type of transcription factor, regulate expression of several effector genes [20,21]. Most is known about the role of Wor1 orthologs in effector gene expression. Wor1 is a conserved fungal transcription factor from Candida albicans, with a WOPR type of DNA binding domain [22]. In plant pathogenic fungi from the genus Fusarium (putative) effector genes and/or secondary metabolite gene clusters are regulated by an ortholog of this transcription factor: F. oxysporum f. sp. lycopersici (Sge1), F. graminearum (Fgp1), F. verticillioides (FvSge1) and F. fujikuroi (FfSge1) [23–26]. Also in the plant pathogenic fungi Botrytis cinerea (Reg1), Verticillium dahliae (VdSge1), Cladosporium fulvum (CfWor1), Zymoseptoria tritici (ZtWor1), Ustilago maydis (UmRos1) and Magnaporthe oryzae (MoGti1) deletion of the gene for this transcription factor (partially) perturbs expression of effector genes [27–32]. Mutant strains deleted for this gene are mostly non-pathogenic (except Δffsge1), although for CfWOR1 this may be a secondary effect of a developmental phenotype. In C. albicans, Wor1 was originally discovered as a ‘master regulator’ of the morphological switch from white to opaque cells. Also in Saccharomyces cerevisiae and Histoplasma capsulatum the Wor1 orthologs (Mit1 and Ryp1, respectively) regulate a morphological transition, which, both in C. albicans and H. capsulatum, is associated with differences in virulence towards humans [33]. This led to the idea that Wor1 orthologs in plant pathogenic fungi are also master regulators of a lifestyle switch, from saprotrophic to pathogenic.
In Fusarium oxysporum f. sp. lycopersici (Fol), the causal agent of Fusarium wilt in tomato, effector genes (called SIX genes for ‘Secreted In Xylem’) reside on an accessory chromosome that can be transferred horizontally between strains. Upon receipt of this accessory chromosome of Fol, a non-pathogenic strain can acquire pathogenicity towards tomato [7]. This means that effector gene expression must be ensured in different genomic environments (i.e. in the original strain and in the recipient strain). Two different but not mutually exclusive strategies to ensure effector gene expression are: i) to rely on conserved transcription factors encoded on the core genome or ii) to encode the transcription factors necessary for effector gene expression on the accessory chromosome itself. As mentioned above, Fol effector gene expression requires the presence of the core-encoded conserved transcription factor Sge1 [26]. However, also on the accessory chromosome transcription factors are encoded, 13 in total [34]. One of these transcription factor genes, FTF1, is associated with highly virulent strains of F. oxysporum f. sp. phaseoli and is up-regulated during infection [35,36]. In addition, FTF1 is present in three variants on the Fol pathogenicity chromosome and all three genes are located close to single or small groups of effector genes [34].
Although the pathogenicity chromosome of Fol is transcriptionally connected to the core genome via Sge1, the presence of numerous transcription factor genes on the chromosome itself suggests that this accessory chromosome might be transcriptionally semi-autonomous. To see whether this may be the case, we investigated the role of the transcription factors encoded on the pathogenicity chromosome of Fol in effector gene expression.
To see if the pathogenicity chromosome of Fol (chromosome 14 in reference strain Fol4287) may be transcriptionally autonomous, we inventoried the transcription factors it encodes. We found 13 predicted transcription factor genes that cluster into nine families. The transcription factor gene families were numbered TF1 to TF9 and include one homolog of EBR1 (TF8) and three homologs of FTF1 (TF1) (Fig 1, S1 Data: tab ‘TF table’) [35,37]. Most gene families encode proteins containing zinc finger DNA binding domains; four are Cys2His2 zinc finger DNA binding domains (Tf3, Tf4, Tf6 and Tf7) and two are Zn(2)Cys(6) zinc finger DNA binding domains (Tf1 and Tf8). Additionally, there are two gene families encoding transcription factors with a basic leucine zipper (bZIP) DNA binding domain (Tf5 and Tf9) and one gene family encoding forkhead transcription factors (Tf2). All transcription factor genes on the pathogenicity chromosome have a homolog on the core genome, except TF3 (Fig 1). Four of the transcription factor gene families have also expanded on other accessory chromosomes of Fol4287 (TF1, TF7, TF8 and TF9).
The F. oxysporum species complex encompasses many different formae speciales (ff.spp.), each with a specific plant host and specific accessory genomic material. All transcription factor gene families (except TF3) have one core- and one or more accessory-encoded homologs in other ff. spp. of F. oxysporum investigated. (Fig 2, S1 Fig, S1 Data:tab ‘TF table’). From here on, we refer to accessory homologs as aTF and to core homologs as cTF. The other Fusarium species analysed (F. solani, F. graminearum and F. verticillioides) each have only one homolog of most of the transcription factors; the expansion we see on the accessory regions in F. oxysporum has not occurred. Regardless of the forma specialis, core-encoded transcription factors form one clade (Fig 2, S1 Fig, indicated with a grey bar) and show little sequence divergence. In general, the F. verticillioides homologs are closest to this clade, consistent with the species phylogeny. Accessory chromosome-encoded homologs show more divergence, both within and between strains, and are also more diverse in sequence than the core encoded homologs in F. oxysporum and F. verticillioides. This suggests that either i) the expansion and divergence of the transcription factor genes on the accessory chromosomes is older than the separation between F. oxysporum and F. verticillioides, similar to what was suggested for accessory genes in general [7], or ii) the rate of evolution is accelerated in the accessory regions compared to the core genome, or iii) a combination of the two.
As transcription factor gene families have expanded and diverged on the accessory chromosomes, we wanted to investigate whether some of these genes may have neofunctionalized. To see if homologous core- and accessory-encoded transcription factors regulate different target genes, we set out to determine the DNA binding sites of each transcription factor encoded on the pathogenicity chromosome and its core-encoded homolog.
Transcription factor coding sequences were cloned from cDNA, in vitro translated as a GST-fusion and hybridised with two different oligo arrays (called HK & ME) [38]. Binding enrichments were inferred for each possible 8-mer. Of 13 transcription factors the cDNA could be amplified and cloned. For aTf1, aTf8 and aTf9, cloning of any of the homologs on the pathogenicity chromosome was unsuccessful. Possibly this was due to both low transcript levels and the presence of homologous transcripts; only hybrid PCR products were amplified. However, a partial cDNA encoding the predicted DNA binding domain from another aTf1 homolog was obtained (FOXG_17084 on accessory chromosome 6). Tf1 homologs separate in two groups based on the length of the coding sequence. The first group (~3200 bp, we refer to this as the longer coding sequence) include the core homolog, two homologs on the pathogenicity chromosome and two identical genes on other accessory chromosomes. The remaining aTF1 genes are shorter (~2800 bp), because they have a more downstream startcodon and a more upstream stopcodon (S2 Fig) [35]. All Tf1 homologs have highly similar DNA binding domains (S3 Fig). The cloned aTF1 cDNA (FOXG_17084) has a short coding sequence.
For nine of the cloned transcription factors a reliable DNA binding site could be inferred from one or both arrays (Fig 3A, S1 Data: tab ‘DNA binding assay’). In all cases both arrays yielded similar top 8-mers. For the remaining four transcription factors no significant enrichment was found. The DNA binding sites of homologous transcription factors were the same or very similar in all four families for which both homologs yielded a DNA binding site, indicating that no diversification in recognition specificity has occurred. The DNA binding site of aTf2 and its homolog cTf2 overlap with the consensus DNA binding site of forkhead transcription factors (RYMAAYA) [39]. The aTF5/cTF5 DNA binding site is almost palindromic, which is in accordance with the dimeric structure of leucine zippers. Tf1 has a Gal4-like Zn(2)Cys(6) DNA binding domain, which in Gal4 binds as a dimer to the CGGN11CCG consensus sequence [40]. The DNA binding site of aTf1 and cTf1 (TRCCG) overlaps with half of this consensus. Interestingly, the aTf1 DNA binding site overlaps with a motif found earlier to be enriched in the promoters of effector genes: aacTGCCGa [34]. Note that the top 8-mer for both aTf1 and cTf1 is perfectly contained in this motif (Fig 3A, S1 Data: tab ‘DNA binding assay’).
We attempted to determine the DNA binding site of Sge1 in a similar way, but no significant 8-mer enrichments were detected. Sge1 has an unusual fungi-specific WOPR type DNA binding domain. The DNA binding site and protein crystal structure of several homologs of Sge1 have been resolved [33,41–43]. The amino acids interacting with DNA are highly conserved among Wor1/Sge1 homologs from different fungi [23,27,29,30,41,43,44]. Consistent with this, the DNA binding sites of Sge1 orthologs in Saccharomyces cerevisiae (Mit1) and the filamentous fungus Histoplasma capsulatum (Ryp1) are the same as for Wor1 [33].
To test whether Sge1 can also bind to the Wor1-DNA binding site, we have adopted an in vivo transcriptional activation assay developed previously [42]. In short, Wor1 or Sge1 is produced constitutively in yeast together with a reporter construct consisting of the Wor1-DNA binding site and an UAS-less CYC1 promoter fused to the lacZ gene. The two SGE1 homologs in yeast (YEL007 or Mit1 and YHR177) are deleted from the yeast strain to avoid potential cross-activation. We found that Sge1 can induce the reporter gene to the same level as Wor1, and this ability is lost when the Wor1 DNA-binding site is mutated (Fig 3B). This confirms that Sge1 binds the same DNA sequence as its orthologs and is a transcriptional activator.
Determination of the DNA binding site of several transcription factors showed that DNA binding specificity has not or hardly diverged between accessory-encoded and core-encoded transcription factors. However, if accessory-encoded transcription factor homologs fulfill an important role in transcriptional regulation of the accessory chromosomes, target genes of these transcription factors could be enriched there. To test this, the number of genes with minimally one, two or three binding sites in 1 kb upstream of the annotated transcriptional start site was counted (Fusarium Comparative Sequencing Project, Broad Institute of Harvard and MIT (http://www.broadinstitute.org/), annotation 3) (S1 Data: tab ‘promoter DBS’). Because some of the sequence motifs contain relatively little information we found many occurrences throughout the genome. Nevertheless, for several DNA binding sites (of aTf1, cTf4 and aTf5) significant enrichment (hypergeometric test, P value < 0.01 after Bonferroni correction) was found on accessory chromosomes, and for the aTf2 DNA binding site enrichment was found on the core genome (Fig 3C, S1 Data: tab ‘promoter DBS’). Specific enrichment of DNA binding sites on the pathogenicity chromosome, however, was not detected. In an effort to reduce noise levels for the smaller sequence motifs we also looked at multiple occurrences, and indeed the observed enrichments were also present under these criteria. Interestingly, also the Sge1 DNA binding site is enriched on the accessory chromosomes. To test whether the observed enrichments are specific for promoter regions, we performed the same test for the 1000 bp downstream the ATG of each gene. We found no DNA binding site enrichments in the coding regions, except for the aTf1 DNA binding site, which was found enriched both in the promoter and in the coding regions of accessory genes (S4A and S4B Fig, S1 Data: tab ‘ORF DBS’).
Since the pathogenicity chromosome and the effectors encoded on it are important for the infection of tomato plants, enrichment of DNA binding sites among genes that are up-regulated during infection and those that are down-regulated was also tested [45]. Only the aTf1 and the Sge1 DNA binding sites were significantly enriched among up-regulated genes, and none of the DNA binding sites was enriched among genes down-regulated during infection (Fig 3C, S1 Data).
Taken together, this statistical analysis suggests that several transcription factor gene families on the pathogenicity chromosome may be involved in regulating targets on accessory chromosomes. In addition, aTf1 and/or cTf1 and Sge1 in particular may target genes involved in pathogenicity.
The effectors encoded on the pathogenicity chromosome are important determinants for pathogenicity and are under strict transcriptional regulation [26,46–50]. To determine whether transcription factors encoded on the pathogenicity chromosome can induce effector gene expression, strains ectopically expressing these transcription factors from a constitutive promoter—from hereon called ‘overexpressors’–were generated and tested for their ability to induce effector gene expression. For seven out of the nine transcription factor families encoded on the pathogenicity chromosome we were able to make constructs with the constitutive FEM1 promoter [51] to drive expression (cloning of aTF8 and aTF9 was unsuccessful). Three aTF1 homologs are present on the pathogenicity chromosome and one of these was selected for overexpression (FOXG_17458, long coding sequence). Also one ‘short’ aTF1 gene from another accessory region was chosen for overexpression (FOXG_17084), the same gene that was used for the DNA binding site determination.
To facilitate screening for the induction of effector gene expression, transcription factor overexpression constructs were transformed into a strain carrying a reporter gene. In this Fol strain, the reporter (GFP) replaces the ORF of the effector gene SIX1 –a representative plant-induced effector gene–so that it is expressed in locus from the SIX1 promoter [46]. For each transcription factor, 11 to 23 independent transformants were inspected for increase of GFP expression using fluorescence microscopy (S5A Fig). GFP signals varied between transformants with the same construct. The three to ten transformants that appeared to respond most strongly to each transcription factor were used to quantify GFP levels spectrophotometrically (Fig 4). Most accessory transcription factors tested did not affect SIX1 expression when expressed from the FEM1 promoter. In contrast, the long version of aTF1 very strongly induced SIX1 expression, and the shorter version conferred some induction. Also SGE1, tested in the same way, can induce SIX1 expression, albeit to a lesser extent than aTF1. To confirm overexpression of the transcription factor genes in the transformants that showed no GFP induction, transcript levels were determined with quantitative RT-PCR. In all cases an increase of transcription factor gene expression compared to the background strain was observed in at least one out of three transformants tested (S5B Fig).
Some of the strains expressing SGE1 or the long version of aTF1 from the FEM1 promoter showed slight growth retardation, and strains overexpressing aTF7 (FOXG_14275) consistently grow slower on both rich and minimal medium (S5C Fig). The same transformants were tested for their ability to infect tomato plants. Only aTF7 overexpression altered this ability; aTF7-overexpressors were less virulent, which may be explained by their retarded growth (S5D Fig).
To see whether the phenotypes caused by aTF1 or aTF7 overexpression could also be induced by overexpression of their core homologs, strains expressing cTF1 (FOXG_09390, long gene model) or cTF7 (FOXG_17774) from the FEM1 promoter were generated and tested as described above (S5 Fig, Fig 4). cTF1 can activate the SIX1 promoter almost to the levels of the long version of aTF1, and cTF7 overexpressors have a similar growth retardation and reduced virulence phenotype as those overexpressing aTF7. This suggests a similar function for core and accessory homologs at least in these two cases.
Since expression of aTF1 and cTF1 from the FEM1 promoter was shown to potently induce the SIX1 promoter, and the SIX1 promoter contains several aTf1 DNA binding sites, we proceeded to test the results of the DNA binding array in an independent system. For this we used the aTF1 homolog (FOXG_17458) used for overexpression and cTF1 in the in vivo transcriptional activation assay described above for Sge1. In this case, a short fragment was cloned from the promoter of the SIX1 effector gene. This sequence includes two Tf1-DNA binding motifs that overlap with the previously identified motif aacTGCCGa and are separated by 17 basepairs. We also constructed a version with mutated Tf1 DNA-binding sites (two basepair substitutions; Fig 5, lower panel). Both aTf1 and cTf1 are able to activate transcription from the construct with the wild type SIX1 promoter fragment, but not the mutated fragment (Fig 5). This assay was performed in the yeast strain lacking SGE1 homologs (YEL007 or Mit1 and YHR177), showing that DNA binding and transcriptional activation by aTf1 and cTf1 does not require (a homolog of) Sge1.
We found that only SGE1, aTF1 and its core homolog cTF1 bind to motifs enriched in the promoters of plant-induced genes and genes on accessory chromosomes. Also, expression of any of these three transcription factor genes from the FEM1 promoter induces effector (SIX1) gene expression. Of both SGE1 and cTF1 a role in pathogenicity has been established previously: the Fol Δsge1 mutant is no longer pathogenic and a Δctf1 (Δftf2) mutant is reduced in pathogenicity [26,45]. Expression of all three transcription factor genes is increased upon infection of tomato plants [26,35,36]. To investigate which genes are regulated by these transcription factors, we compared the transcriptomes of SGE1, aTF1 and cTF1 overexpressors with the transcriptome of the background strain. Overexpressors were preferred over gene deletion strains in this setup because effector genes are only expressed during colonization of the plant, and the plant signal that induces this up-regulation is unknown. Since the deletion mutants are not pathogenic or reduced in pathogenicity, expression cannot reliably be studied in planta with gene deletion strains either. Given that expression levels of SGE1, aTF1 and cTF1 increase during infection, we assume that strains overexpressing either of these transcription factor genes partially mimic the in planta state.
RNA was isolated from two-day old cultures growing in minimal medium. For each transcription factor two independent overexpressors were used, and for each strain three independent biological replicates were sampled. Per sample between 2.8*10^7 and 4.8*10^7 reads were paired-end sequenced (Illumina). Reads were mapped to the reference genome (Fol4287) and differentially expressed genes were called for each pairwise comparison with the background strain, using DEseq software [52]. The different pairs yielded between 396 and 691 differentially expressed genes (S6A Fig).
To compare the fold induction of expression of the overexpressed transcription factor genes to their induction during infection, we isolated RNA from Fol-infected tomato plants and from axenic cultures. Both transcriptomes of three biological replicates were sequenced and sequencing reads were mapped to the same annotation of the reference genome as above, using the same parameters. As can be seen in Fig 6, expression of each of the three transcription factor genes is increased in the respective overexpressors as well as during infection. The increase in expression is significant (p<0.05) for all three transcription factors in their respective overexpressors, but during infection only for aTF1. Q-RT-PCR confirmed a significant increase of aTF1 expression in aTF1 overexpressors and during infection, but could not confirm significant differences for cTF1 and SGE1 transcript abundance (S7 Fig).
Our inability to measure a (significant) increase of cTF1 and SGE1 transcripts above their basal expression level by Q-RT PCR was unexpected. SIX1 (GFP) induction in the independent cTF1 and SGE1 overexpressors may be caused by an only modest increase in transcript levels, as suggested by our RNAseq analysis. For the aTF1 overexpressors, expression of both SIX1 (GFP) and another effector gene (SIX3) does correlate well with aTF1 levels (S8 Fig). Also, only in transformants expressing aTF1, cTF1 or SGE1 from the FEM1 promoter did we ever observe high GFP levels in the Psix1:GFP strain (Fig 4, S5A Fig). Still, we wished to exclude the possibility that the transcriptome changes in the aTF1, cTF1 and SGE1 overexpressors could be due to spontaneous changes unrelated to the transcription factor. We therefore clustered gene expression data of significantly differentially expressed genes for all six overexpressors (S8 Fig). Both aTF1 and both cTF1 overexpressors cluster together, showing that genome wide changes in gene expression in these strains are related to the overexpressed transcription factor. The SGE1 overexpressors, however, do not form a single clade, so we could not link the transcriptional changes in these strains to SGE1 with this approach.
To further test the connection between the selected transcription factors and the changes in the transcriptome, we determined whether the DNA binding site of each transcription factor is enriched in the upstream regions of the genes that are differentially expressed in the aTF1, cTF1 or SGE1 overexpressors. For this we only considered those genes that were significantly up- or down-regulated in both overexpressors of each transcription factor to be consistent transcription factor-specific effects and this selection was used in all further analysis. Under these criteria, each transcription factor up-regulates 65 to 116 genes, and down-regulates 273 to 347 genes, when overexpressed (S6B Fig). We have analysed single occurrence of the Sge1 DNA binding site, triple occurrence of the (relatively short) Tf1 DNA binding site, and single occurrence of the longer motif found in effector gene promoters (that overlaps with the Tf1 DNA binding site) for enrichment in 1kb promoter regions or coding regions (Fig 7, S4C &S4D Fig, S1 Data: tab ‘promoter DBS’ and ‘ORF DBS’). Both the short and the long Tf1 DNA binding sites are significantly and specifically enriched in the promoters of genes up-regulated in both the aTF1 and cTF1 overexpressors, but not in the SGE1 overexpressor. The Sge1 DNA binding site, on the other hand, is specifically enriched in the promoters of SGE1- but also aTF1- and cTF1-up-regulated genes. This links the presence of DNA binding sites to changes in gene expression for all three transcription factors. For none of the down-regulated genes a significant enrichment of DNA binding sites was found, neither for down-regulated genes on the pathogenicity chromosome nor for down-regulated genes on the core chromosomes. This is consistent with the observation that all three transcription factors act as transcriptional activators.
Finally, the enrichment of the Tf1 and Sge1 DNA binding site among three other groups of genes was tested (Fig 7). The three groups are: i) genes coding for small secreted proteins; ii) genes with a (partial) miniature impala (MIMP), a non-autonomous transposable element, in the upstream region (up to 2 kb from the ATG—all Fol effector genes are associated with a MIMP, but also other, mainly accessory genes, of which around half is also up-regulated during infection (S6B Fig, [34])); iii) genes of which the protein has been detected in xylem sap of infected plants [34]. The Tf1 DNA binding site is not enriched among these categories, except the long effector motif among small secreted proteins. Remarkably, the Sge1 DNA binding site is enriched in all these categories (Fig 7, lower panel), suggesting that Sge1 may target pathogenicity-associated genes more specifically than aTf1 or cTf1.
We have shown that the genome-wide transcriptional changes in the aTF1 and cTF1 overexpressors are correlated with expression of the respective transcription factor from the FEM1 promoter, while the Sge1 binding site is enriched in upstream regions of genes upregulated in the SGE1 overexpressors and of pathogenicity-associated genes. We now compared the gene sets differentially expressed in the SGE1, aTF1 and cTF1 overexpressors and during infection (Fig 8A and S6B Fig). Strikingly, the majority of the plant-responsive genes on the pathogenicity chromosome is also up-regulated in all three overexpressors, including almost all SIX effector genes (indicated in purple text in Fig 8B). The SIX genes that are not induced are SIX13 in all overexpressors and SIX2 in the SGE1 overexpressors. On the other accessory regions (chromosome 3, 6, 15 plus small regions on chromosome 1 and 2) the overlap between plant responsive genes and Sge1-, aTf1- or cTf1-responsive genes is much smaller. Here a large group of genes is down-regulated in all transformants but not during infection. Of this group many genes are generally weakly expressed. We noticed that lower expression levels of these genes–as observed in the overexpressors–was also found in wild type strains (S9 Fig). This suggests that the apparent down-regulation of expression from this region may not be due to increased abundance of one of the three transcription factors, but rather to general variability in gene expression in this region between transformants. Another transcription factor-unrelated change in expression was observed for a part of the pathogenicity chromosome. In Fig 8B, the top rows of the enlargement of the pathogenicity chromosome show a group of genes up-regulated in one of the cTF1 overexpressors and one of the SGE1 overexpressors. Closer examination showed that these genes are part of a region on supercontig 22 that is generally higher expressed in these two transformants (S10 Fig). Given the regional nature and the lack of correlation to a particular transcription factor, we suspect these differences may be either caused by changes in chromatin state, or by ‘spontaneous’ duplication of these regions. Spontaneous duplications in accessory regions have been reported previously in F. oxysporum [53].
The majority of all plant-induced genes is located on the core chromosomes, but only a small portion of those is differentially regulated in the aTF1, cTF1 or SGE1 overexpressors. Remarkably, whereas these three transcription factors seem to target the same set of genes on the pathogenicity chromosome, on the core genome the overlap between the genes of which expression is altered by the different transcription factors is much smaller, and each mostly induces a specific set of genes, especially Sge1 (S6A Fig). Still, the genes up-regulated by the transcription factor genes on core chromosomes are enriched for plant-responsive genes (hypergeometric test, P value < 0.05 after Bonferroni correction).
To see which functional categories of genes apart from effector genes are targeted by aTf1, cTf1 and Sge1, a FunCat analysis was performed [54] (S1 Data: tab ‘FungiFun’). Of the genes up-regulated in SGE1 overexpressors, genes in the categories heme binding (including genes coding for cytochrome P450 proteins), catalase reactions and electron transport (including ATPases and oxidoreductases) are overrepresented. Among the predicted functions of the genes induced in the cTF1 overexpressors there is enrichment in secondary metabolism and C-compound and carbohydrate metabolism (including polysaccharide metabolism and protein glycosylation). The set of genes induced by aTf1 is also enriched for genes in secondary metabolism and C-compound and carbohydrate metabolism (including polysaccharide and chitin metabolism as well as pectate lyases).
In the SGE1 overexpressors a peroxisome biogenesis factor (PEX11) is up-regulated. Peroxisome function is required for pathogenicity in Fol [55]. Both aTf1 and cTf1 up-regulate expression of the shorter aTF1 gene on the pathogenicity chromosome (FOXG_16414), although not to the same level as during infection. Apart from this aTF1 homolog, expression of only one other transcription factor gene is significantly altered in any of the overexpressors: aTf1 up-regulates FOXG_04965. Strikingly, this transcription factor is required for pathogenicity [45].
The set of down-regulated genes on the pathogenicity chromosome and the core chromosomes was not significantly enriched for certain categories for aTf1 and cTf1 regulated genes. Sge1-repressed genes were enriched for genes in polysaccharide metabolism and amino saccharide metabolism. Taken together, all three transcription factors influence effector gene expression and secondary metabolism and target genes predicted to affect both the fungal and the plant cell wall (S1 Data: tab ‘FungiFun’).
Although very different transcription factors, aTf1/cTf1 and Sge1 induce expression of a large, overlapping set of genes on the pathogenicity chromosome. Together with the observation that some regions may be prone to activation that is not linked to a specific transcription factor gene, this made us wonder whether the pathogenicity chromosome is transcriptionally activated as a whole, rather than gene by gene. The accessory chromosomes are very rich in transposable elements, and we decided to investigate expression of transposable elements as a proxy for chromosome-wide transcriptional activation. Besides, we were interested to see whether general up-regulation of genes on the pathogenicity chromosome (for instance during infection) could jeopardize genome integrity by induction of transposon activity.
One problem, however, is that in our RNAseq analysis the number of transcripts derived from transposable elements was probably highly underestimated, because: i) many transposons are not annotated, ii) reads of identical transposons are distributed over all copies, obscuring activation of individual copies, and iii) reads mapping to more than ten different genomic locations were excluded. To circumvent this, a fasta file was generated where the sequence of each repetitive element, plus all previously annotated transposable elements [34] is present only once. To this file, all sequence reads from the overexpressors as well as reads from infected plant material were mapped (S11 Fig and S6D Fig). We have found no evidence for a general increase of transposable element transcription in the aTF1, cTF1 or SGE1 overexpressors or during plant infection.
We have shown that expression of Sge1, aTf1 or cTf1 from the FEM1 promoter is correlated with induced expression of a large overlapping set of genes, including effector genes. Whereas aTf1 and cTf1 are homologs and have the same DNA binding specificity, Sge1 is a very different transcription factor, with a strongly conserved, fungal specific DNA binding domain. This raises the question what causes this overlap in transcriptional activation. Expression of for example SIX1 and SIX3 is induced in aTF1, cTF1 and SGE1 overexpressors. However, whereas SIX3 has both Tf1 and Sge1 DNA binding sites present in the promoter, for SIX1 only Tf1 DNA binding sites are present. To test whether the presence of Sge1 is required for aTF1 overexpression-mediated induction of SIX1 and SIX3, we deleted SGE1 in an aTF1 overexpressing strain. Without the presence of SGE1, the expression of the effector genes SIX1 and SIX3 was reduced to low (wild type) levels, while overexpression of aTF1 itself remained unchanged (Fig 9). This shows that Sge1 is required for aTF1 overexpression-mediated activation of effector gene expression, but not necessarily via an Sge1 DNA binding site. Also, deletion of SGE1 resulted in loss of pathogenicity in both WT and in aTF1 overexpressing strains (S12 Fig).
In summary, we have demonstrated that some pathogenicity chromosome-encoded transcription factors have regulating potential on the accessory chromosomes themselves, and that aTF1 and the core-encoded cTF1 and SGE1 can induce effector gene expression and expression of other plant-responsive genes upon constitutive expression, possibly via direct binding of the transcription factors to these promoters and subsequent transcriptional activation.
This study aimed to gain insight in the transcriptional connections between core chromosomes and the accessory ‘pathogenicity’ chromosome in Fol, especially with respect to effector gene expression. We show that expression of the transcription factor genes aTF1 (FTF1, located on the pathogenicity chromosome), cTF1 (FTF2) or SGE1 (both located on the core) from the constitutive FEM1 promoter induces expression of many plant responsive genes located on the pathogenicity chromosome, particularly effector genes. We conclude that the pathogenicity chromosome is transcriptionally partially self-regulating, but not isolated from the core genome.
The large overlap in the sets of target genes of aTf1/cTf1 and Sge1 on the pathogenicity chromosome suggests these transcription factors somehow work together. This is substantiated by the observation that aTF1 overexpression-mediated induction of effector gene expression requires Sge1. Overexpression of SGE1 orthologs in F. graminearum, F. verticillioides and F. fujikuroi induces the expression of secondary metabolite genes [23–26]. Apart from SGE1, expression of secondary metabolite gene clusters depends on specialized transcription factors often physically associated with the cluster [56]. Reminiscent of this, SIX effector genes sometimes occur in mini clusters of two or three SIX genes, accompanied by a aTF1 homolog. Genes that are located between or physically close to clustered SIX genes (ORX1, SHH1) are co-induced, both during infection and in the overexpressors.
Transcriptional regulation of secondary metabolite clusters partially takes place on the chromatin level [57,58]. Some observations hint at chromatin-mediated regulation of expression of genes on the pathogenicity chromosome of Fol as well. First, there is the transcription factor unrelated up-regulation of parts of SC22 on the pathogenicity chromosome (in one cTF1 and one SGE1 overexpressor, this study). Second, the BLE resistance gene behind the constitutive GPD promoter located near the SIX1 locus is co-up-regulated in some of the overexpressors (S1 Data, tabs: ‘total mapped reads’ and ‘RPKM’. Finally, physical interactors of the Saccharomyces cerevisiae Sge1 ortholog Mit1 include histones and histone acetyl transferase complexes and are enriched for the GO terms chromosome organization, chromatin assembly or disassembly and chromatin organization (yeastgenome.org). This raises the question whether the large overlap of genes on the pathogenicity chromosome affected by expression of aTF1/ cTF1 and SGE1 from the FEM1 promoter may be caused by a shared influence on chromatin structure. It will be interesting to see if facultative heterochromatin functions as a layer of effector gene expression regulation also in F. oxysporum.
Another explanation for the overlap between the aTF1, cTF1 and SGE1 targets on the pathogenicity chromosome is that Sge1 induces effector gene expression via aTF1/cTF1 or vice versa. Despite the fact that SGE1 has potential aTf1 binding sites in its promoter and one aTF1 (FOXG_17458) has potential Sge1 DNA binding sites in its promoter, neither of the genes is constitutively up-regulated by overexpression of the other. Moreover, Sge1 is still required for effector gene expression when aTF1 is overexpressed. Also, there are clear differences between the different overexpressors, for example SIX11 and SIX2 are induced in aTF1- and cTF1-overexpressors, but not in SGE1-overexpressors. SIX11 and SIX2 indeed have no Sge1 DNA binding site in their promoter. SGE1 is, however, required for the induced expression of SIX2 upon exposure of Fol to tomato cell cultures [26].
Alternatively, Sge1 and aTf1/cTf1 may be simultaneously required at promoters for transcription activation. Sge1 and cTf1 both have basal (non-induced) expression levels, and overexpression of either might cause recruitment of the other (perhaps by direct interaction or through chromatin modification) leading to transcription activation. However, we have shown that Sge1 as well as Tf1 can bind DNA and activate transcription independent of each other. A similar situation has been described for the Sge1 homolog Ryp1 and two velvet-like transcription factors in H. capsulatum [59]. A subset of the genes induced by overexpression of Ryp1 is also induced by overexpression of either of the two velvet-like transcription factors. These shared target genes have DNA binding sites for each transcription factor in their promoter, and the three transcription factors physically interact.
To unravel the question of Sge1/Tf1 co-operation, it will be necessary to determine physical interactions, both between them and with promoters, under basal conditions and upon overexpression, in combination with analysis of chromatin structure at the same promoters.
Next to (nearly) identical DNA binding sites, several observations support an overlap in function between homologous transcription factors encoded on accessory and core genomes. aTF7- and cTF7-overexpressing strains show a similar growth retardation and reduction of virulence, and aTf1 and cTf1 both are potent inducers of effector gene expression. In addition, a deletion mutant of cTF1 is less virulent, but can still colonize the plant [45,60]. The partial loss of virulence could be a result of redundancy between cTF1 and aTF1. Deletion of either of two of the shorter aTF1 homologs (FOXG_16414 and FOXG_17123) does not affect pathogenicity [45], but recently silencing affecting both aTF1 and cTF1 was shown to reduce virulence in both f. sp. lycopersici and f. sp. phaseoli [60]. Also host-induced gene silencing of aTF1, potentially silencing the entire gene family, has been reported in F. oxysporum f. sp. cubense on banana, which led to complete absence of disease symptoms [61].
If there indeed is such a large functional overlap between core and accessory homologs, then why does Fol contain transcription factor genes on the pathogenicity chromosome at all? And why—in some cases–even multiple homologs? One possibility is that the presence of the transcription factor (and other) genes on accessory chromosomes did not result from selective pressure on a particular functional advantage but is simply provoked by certain features of accessory chromosomes, such as high transposable element density or chromatin structure. A different, but not mutually exclusive, view is a selective advantage of gene duplications because of a dosage effect. In F. oxysporum f. sp. phaseoli, a correlation between virulence and the number of aTF1 homologs has been reported [35].
It should be noted, however, that although very similar, functions of core and accessory-encoded homologous transcription factors may not be completely redundant. For example, overexpression of aTF1 (FOXG_17458) induces expression of a transcription factor required for pathogenicity, whereas overexpression of cTF1 does not. Also, the short aTF1 (FOXG_17084) homolog can bind DNA with the same specificity, but is far less efficient in inducing SIX1 expression. At present, it cannot be excluded that the short homolog might even function as a negative regulator.
It is currently unclear over how large a phylogenetic distance the pathogenicity chromosome can be transferred. Up to now, transfer as only been experimentally demonstrated within the F. oxysporum species complex [7]. In such a case the cTF1 and SGE1 homolog in the recipient strain will be nearly identical to the chromosome donor strain, only accessory-encoded aTf1 homologs on regions other than the pathogenicity chromosome will be absent or different in the recipient. Interesting in this respect is a previous observation: when the pathogenicity chromosome is transferred to the non-pathogenic strain Fo-47, this strain gains pathogenicity, but is not very aggressive. Virulence is higher in those strains that gained not only the pathogenicity chromosome, but also a second small chromosome, corresponding to the duplicated region of chromosome 3/6 in the reference strain Fol4287 [7]. This region contains few if any effector genes, but many transcription factor genes, including several homologs of aTF1, aTF4, aTF6, aTF7, aTF8 (EBR2) and aTF9.
Genome analyses also suggest at least one ancient transfer event from a Fusarium species at a phylogenetic distance from Fol somewhere between F. verticillioides and F. graminearum [7]. If such a transfer were to occur again, the cTF1 homolog in the recipient would still be very similar to the cTF1 homolog in the donor strain, at least more similar than cTF1 is to aTF1. No additional accessory aTF1 homologs would be present in the recipient. We speculate that the transcription factors on accessory chromosomes do contribute to virulence, but many of their roles could be also partially fulfilled by their core-encoded homologs within the genus Fusarium.
For SGE1 homologs, significant functional differences have been reported between closely related species, like different Fusarium spp, but also between C. albicans and S. cerevisiae [23–25,33]. As described above, the DNA binding domain at the N-terminus is very conserved, and so is–as far as it has been tested–the DNA binding site [33], whereas the C-terminal domain is very variable [24,25]. For the yeasts C. albicans and S. cerevisiae, differences in target genes are caused by differences in the presence or absence of cis-acting promoter elements [33]. For F. graminearum and Fol, however, differences in target genes are caused by differences in the C-terminal domain, and transcomplementation can partially restore pathogenicity only in some cases [24]. Also in F. fujikuroi and F. verticillioides overexpression of their respective SGE1 homologs regulates a different set of genes [23,25].
Since the differences between the Sge1 homologs between different Fusarium species are so substantial that Sge1-mediated transcription regulation is extensively rewired, this may be a limiting factor to expression of effector genes from the Fol pathogenicity chromosome in another Fusarium species. Interestingly, transcomplementation of the Δsge1 mutant of Fol with CfWOR1 from C. fulvum resulted in restoration of effector gene expression but not in restoration of pathogenicity [28].
Previously, Jonkers and co-workers have looked at differentially expressed genes in a Fol SGE1 deletion mutant compared to WT using a microarray [24]. This set (1213 genes) is very different from the differentially expressed genes in the SGE1 overexpressors that we found (168 genes); only 15 genes are present in both sets. This shows that the different approaches are complementary for identification of target genes. The genes differentially expressed between WT and deletion strain are predominantly present on the core genome. Also, the Sge1 DNA binding site is not significantly enriched among up- or down-regulated genes in the deletion mutant (104 [out of 394] down-regulated genes with a Sge1 DNA binding site and 171 [out of 820] up-regulated genes, on a total of 4870 [out of 20935] genes with a Sge1 DNA binding site), in contrast to the SGE1 overexpressor. This may be because several transcription factor genes are differentially expressed in the deletion mutant, potentially regulating secondary targets. Also, the comparison of WT and SGE1 deletion mutant was of necessity conducted under axenic growth conditions which excludes finding targets that are not or very weakly expressed under those conditions, including most genes on the accessory genome.
We have shown that aTf1 can activate effector gene expression. However, many more transcription factors are encoded on the pathogenicity chromosome. It is tempting to speculate that these transcription factors may control expression of some of the other plant-responsive genes. The core-encoded homologs of some of the other accessory transcription factors have been implicated in pathogenicity. cTF8 (EBR1) is required for full pathogenicity of Fol and F. graminearum, and orthologs of cTF4 and cTF9 are required for pathogenicity of F. graminearum (FGSG_10057, FGSG_10517 and FGSG_06651 respectively) [37,62]. Of these three transcription factors, we have only obtained a DNA binding site for cTf4, which is, like the DNA binding sites of aTf2, aTf5 and aTf7, not enriched among genes up- or down-regulated during infection. It is of course possible that there is not always a significant correlation between the presence of the DNA binding site and changes in expression during infection. This may occur when loss of pathogenicity is an indirect effect (via a second, downstream transcription factor), when only a particular combination of transcription factors regulates plant responsive genes, or when too many apparent binding sites are not functional, precluding detection of a significant association between binding site and gene regulation.
Of two transcription factors (cTf4 and aTf5), DNA binding sites are enriched on the accessory chromosomes, suggesting they may also act on genes of the accessory chromosomes. In F. graminearum, global gene regulatory network modelling revealed that species-specific genes are most often controlled by species-specific regulators, whereas genes conserved between species are controlled by conserved regulators [63]. Possibly, such a compartmentalized network structure of gene regulation may also apply to F. oxysporum.
Other putative functions of the transcription factors encoded on the pathogenicity chromosome could be downregulation of effector gene expression and returning to or maintaining a repressed, saprotrophic state. They may also be promoting horizontal transfer. It would be interesting to see if the pathogenicity chromosome can be lost, and if so, in which ways this may affect the phenotype of Fol and expression of core genes.
In order to determine the DNA binding sites of the different transcription factors, the ORFs were cloned from cDNA. For this, RNA was isolated from axenic cultures and from susceptible infected tomato plants, between 1–2 weeks after inoculation. cDNA was generated as described below. For some transcription factors alternative start codons were tried, and the longest obtained PCR product was selected for cloning. For many transcription factors, 35 PCR cycles was insufficient to amplify clonable amounts of DNA, therefore, a re-amplification was done. Primers were designed in such a way that the subsequent product could be cloned with AscI, BamHI or SbfI, in frame with a N-terminal GST-tag in an Escherichia coli T7 expression vector pTH6838. The cloned ORF was checked by sequencing.
To express the transcription factor genes in Fol, the binary vector pRW2h was modified, yielding a plasmid with a right border (facilitating Agrobacterium tumefaciens mediated transformation), the FEM1 promoter, a multiple cloning site including XbaI, AscI, StuI, SbfI, BglII and ApaI, followed by the SIX1 terminator, the HPH resistance cassette and the left border. The transcription factor ORF was then cut out of pTH6838 with the same enzymes used to clone it in, and cloned into pRW2h_Pfem_MCS_Tsix1, again with the same enzymes. For those transcription factors of which the gDNA ORF was cloned, a PCR was done on Fol007 gDNA, isolated as described in [64] using the same primers initially designed for cloning of the cDNA. The PCR product was digested and cloned directly into pRW2h_Pfem_MCS_Tsix1. The cloned transcription factor ORFs were checked by sequencing.
Fol was transformed via Agrobacterium mediated transformation, as described previously [65]. Transformants were monospored by pipetting 10–20 μl of sterile water on the emerging colony, and spreading this on a fresh PDA plate supplemented with cefotaxime and Hygromycine. After two days of growth at 25 degree Celsius, single colonies were picked and transferred to fresh plates. From these plates glycerol stocks were made and these are the transformants we worked with. Transformants were only selected on antibiotic resistance, and should contain one (or in rare cases more than one) random ectopic insertion of the T-DNA construct [26].
Cultures for RNA isolation were grown as described below, except for experiments described in S8A Fig and Fig 9. These RNA isolations were done as described under ‘Deletion of SGE1 in aTF1 overexpressor background’ later in this section. For all other cases a small cube with mycelium from a PDA plate was used to inoculate a preculture of 50 ml liquid minimal medium (1%KNO3, 3% sucrose 0.17% YNB w/o amino acids or NH4) and grown for 3 to 5 days at 25°C, shaking 150–175 rpm. From this preculture microconidia were isolated by filtering the culture over miracloth and pelleting the microconidia in the filtrate at 2000 rpm for 10–15 min.
To harvest mycelium for RNAseq analysis of transcription factor overexpressors or for quantification of transcription factor transcripts by Q-RT-PCR, microconidia were suspended in a small volume of minimal medium, counted, and used to inoculate 100 ml liquid minimal medium with 2.5 * 10^8 microconidia. This culture was grown for 2 days at 25°C, shaking 150–175 rpm before mycelium was harvested by filtering the culture over a double layer of miracloth. The mycelium in the filter was washed once with 50–100 ml sterile water, scraped from the miracloth and snap frozen in liquid nitrogen. Of each condition, three independent biological replicates were sampled.
To harvest material for RNAseq analysis of infected plants, Fol4287-infected tomato roots were harvested nine days post inoculation. Infections were performed as described below (for the bioassays). Fol4287 mycelium from axenic cultures was harvested from five day old cultures inoculated from plate in 100 ml liquid minimal medium (1%KNO3, 3% sucrose 0.17% YNB w/o amino acids or NH4), 25°C, 150–175 rpm.
RNA was isolated as described earlier, using a Trizol extraction on mycelium ground in liquid nitrogen, followed by DNase treatment and purification over RNeasy RNA purification columns, according to the instructions of the manufacturer (Qiagen).
Synthesis of cDNA was performed using 1 μg of RNA, poly dT primers, Promega RNasin (ribonuclease inhibitor) and Gibco Superscript II RNase H− Reverse transcriptase, according to instructions of Gibco.
Of 2 μg of total RNA of each biological replicate, polyadenylated RNA was amplified and ligated to adapters to make a library suitable for multiplex illumina paired-end sequencing. Each sample was barcoded and sequenced in 8 different lanes. After de-multiplexing, total reads of the different lanes were combined. This rendered one file of reads per biological replicate.
Quantitative PCR was performed with a model 7500 Real Time PCR system (Applied Biosystems) and Solis BioDyne 5x HOT FIREPol Eva Green qPCR Mix Plus (ROX). Primers used for Q-RT-PCR where designed to amplify fragments of approximately 100 bp and tested for primer efficiency and melting curve (S1 Data). 1 μL of cDNA was used per sample, two technical replicates were performed for each sample. Transcription elongation factor 1α (EF1α) gene expression was used as a reference, and RNA that was not transcribed into cDNA as a gDNA contamination control. The following formula was used to calculate the amount of DNA: [DNA] = (1/E^ Ct_sample) -(1/E^Ct_control), with E = primer efficiency, Ct_sample = Ct value of the test sample, using WT or TF overexpressor cDNA as a template and Ct_control = Ct value of no cDNA control sample (to check for gDNA contamination), with the same primer pair. The comparison with EF1α was made as follows: DNA_TF/DNA_EF1α. Standard deviations of the two technical replicates per sample were calculated with the following formula: Standard deviation = √((stdev DNA_TF /average DNA_TF)^2 + (stdev DNA_EF1α /average DNA_EF1α)^2))* (DNA_TF /DNA_EF1 α).
The Illumina reads (125 bp paired end, insert size around 200–500 bp) were mapped to the annotated genome of Fol4287 (Fusarium Comparative Sequencing Project, Broad Institute of Harvard and MIT (http://www.broadinstitute.org/), annotation 3) using CLC Genomics Workbench version 6.5.1 module (CLC bio, Aarhus, Denmark). Reads were imported as: illumina (pipeline 1.8 and later), paired-end reads, insert size 100–600 bp, remove failed reads. Imported reads were trimmed to remove any remaining adapter sequences or low quality reads. Quality scores and ambiguous nucleotides were trimmed with standard settings (limit 0.05, ambiguities 2). Adapters were trimmed by checking for the presence of the Truseq Universal adapter (minus strand) and the presence of the Truseq index adapter (plus strand) with the following parameters: mismatch = 2, gapcost = 3, cutoff ns, cutoff at end 6, action: remove adapter.
The following gene models were manually added to the annotation:
Supercontig_2.36 135864–136180 minus strand FOXG_SIX14; Supercontig_2.51 62412–62758 minus strand FOXG_SIX12; Supercontig_2.51 65216–65878 plus strand FOXG_SIX7; Supercontig_2.22 806692–807024 minus strand FOXG_SIX11; Supercontig_2.22 44647–44970 minus strand FOXG_PEG4 (Putative Effector Gene 4); Supercontig_2.36 468654–469317 minus strand FOXG_SIX8-36; Supercontig_2.51 6999–7662 minus strand FOXG_SIX8-51.
Reads were mapped to the annotated genome with parameters: Organism type = eukaryote. Exon discovery according to standard settings: relative expression level = min 0.2, min reads = 10, min length = 50 bp. Additional downstream bases = 0. Additional upstream bases = 0. Minimum length fraction = 0.9. Minimum similarity fraction = 0.95. Minimum number of reads = 10. Map only intact pairs, count paired reads as one. Unspecific match limit = 10. Expression value = Total number of mapped reads and reads per kb per million mapped reads (RPKM).
Differentially expressed genes were called for pairwise comparisons of the total number of mapped reads (three replicates), using the Bioconductor DEseq software [52]. After normalization for library size (estimateSizeFactors) and variance estimation (estimateDispersion) with parameters “method = blind, sharingMode = fit-only”, Nbinominal testing and the Benjamin-Hochberg multiple testing adjustment procedure were used [52]. For the transcription factor gene overexpressing transformants (all compared pairwise to the Fol007 Psix1GFP samples), all genes that had an adjusted p value <0.1 for both overexpressors were considered significant. For the samples from infected plants (compared to the Fol 4287 control samples) all genes with an adjusted p value < 0.05 were considered significant. The output file gives the mean normalized mapped reads per sample (mean of the three replicates), pvalue and adjusted pvalue.
For all pairwise comparisons the normalized total mapped reads were collected, and every count of zero reads was replaced by 0.1 (0.1 is roughly half of the lowest number of normalized total mapped reads in all comparisons, this allows calculation of -an approximate- fold change for each gene). Fold change and log2 fold change were calculated. To visualize expression differences in a heatmap, all genes considered differentially expressed in one of the comparisons were listed and the log2 fold change for each condition was listed in seven subsequent columns. In this list, each value not reaching the significance threshold was replaced by ‘0’ (indicating no fold change). This list was separated into three lists based on subgenome and each of these lists were clustered on gene and condition in Gene Cluster 3.0, uncentered correlation, average linkage. Results were visualized in Java Treeview.
Next to this, the differentially expressed genes were subdivided in lists of up-regulated and down-regulated genes. These lists were used to count the contributions of different subgroups to each category (for example: number of aTF1 up-regulated genes that is located on the pathogenicity chromosome). Hypergeometrical distribution tests were used to determine significant enrichments or depletions among different categories. The adjusted p value was reached by multiplying the p value with the total number of tests performed.
To determine which genes have a MIMP in their promoter, two kb upstream the ATG of each gene was searched for the presence of complete (both inverted repeats present) or partial MIMPs. Any missing SIX gene, of which a MIMP had been demonstrated in the promoter earlier [34], was manually added to the list.
The Illumina reads (125 bp paired end, insert size around 200–500 bp) were mapped to the fasta file with all repetitive/transposable elements as described above, except the following parameters: Organism type = prokaryote, unspecific match limit = 30. Reads were normalized as the number of reads per 20*10^6 uniquely mapped reads to the total genome for that particular sample.
For the more detailed analysis the same pairwise comparisons were made as described above. Every count of zero reads was replaced by 0.1, the data was log10 transformed and a T-test was performed on the log10 transformed normalized total mapped reads. All sequences that compared differently (p<0.05) were counted.
Bioassays were performed as described earlier [66]. Briefly, tomato seedlings of 10 to 11 days were trimmed at the main root and dipped in a Fol microconidia suspension of 0.5*10^7 microconidia/ml for at least 1 minute. The seedlings were potted in soil in individual pots and grown in the greenhouse at 25°C for three weeks. At the time of harvest, the above ground parts were cut off at the cotelydons and scored for fresh weight and disease index. The disease index ranges from 0 (no symptoms), 1 (thickening of hypocotyl, formation of lateral roots), 2 (one brown vessel), 3 (up to ¾ of the vessels show browning, asymmetric development) to 4 (all vessels brown, severe growth retardation, death).
Growth assays were performed by positioning a droplet of spores on the middle of a PDA or CDA plate, growing the fungus at 25°C for 5 days, and measuring the colony diameter.
GFP fluorescence was measured on a Fluostar optima platereader (BMG Labtech). For this dilutions of 10^8, 10^7, 10^6 and 10^5 microconidia per ml were made and 200 μl of each suspension was pipetted in a sterile, flat bottom, black 96 well plate (Greiner). The plate was shortly mixed and measured from the bottom, with 470–10 nm excitation and a 510–10 nm emission filter. The plates were kept o/n at 25°C, and measured again the next day, same settings. No differences were observed between the days (apart from a slight increase in fluorescence due to growth).
Details of the design and use of PBMs have been described elsewhere [38,67–69]. Here, we used two different universal PBM array designs, designated 'ME' and 'HK', after the initials of their designers, as described in [70]. Briefly, we used 150 ng of plasmid DNA in a 15 μl in vitro transcription and/or translation reaction using a PURExpress In Vitro Protein Synthesis Kit (New England BioLabs) supplemented with RNase inhibitor and 50 μM zinc acetate. After a 2-h incubation at 37°C, 15 μl of the mix was added to 155 μl of protein-binding solution for a final mix of PBS/2% skim milk/0.2 mg per ml BSA/50 μM zinc acetate/0.1% Tween-20. This mixture was added to an array previously blocked with PBS/2% skim milk and washed once with PBS/0.1% Tween-20 and once with PBS/0.01% Triton-X 100. After a 1-h incubation at room temperature, the array was washed once with PBS/0.5% Tween-20/50 mM zinc acetate and once with PBS/0.01% Triton-X 100/50 mM zinc acetate. Cy5-labeled anti-GST antibody was added, diluted in PBS/2% skim milk/50 mM zinc acetate. After a 1-h incubation at room temperature, the array was washed three times with PBS/0.05% Tween-20/50 mM zinc acetate and once with PBS/50 mM zinc acetate. The array was then imaged using an Agilent microarray scanner at 2 μm resolution. Images were scanned at two power settings: 100% photomultiplier tube (PMT) voltage (high), and 10% PMT (low). The two resulting grid images were then manually examined, and the scan with the fewest number of saturated spots was used. Image spot intensities were quantified using ImaGene software (BioDiscovery).
Calculation of spot intensities was done as described in [38]. In summary, bad spots (spots that had scratches, dust flecks or other imperfections) were flagged manually and removed from subsequent analysis. The PBM signal intensity at each spot was normalized by the corresponding amount of dsDNA. To correct for any possible nonuniformities in hybridization, these normalized PBM intensities were then adjusted according to their positions on the microarray. Each spot was considered to be at the center of a block of spots [70]. The difference between the median normalized intensity of the spots within the block and the median normalized intensity of all spots on the microarray was subtracted from the normalized intensity at that particular spot.
Calculation of 8-mer Z- and E-scores was performed as previously described [38,71]. Z-scores are derived by taking the average spot intensity for each probe containing the 8-mer, then subtracting the median value for each 8-mer, and dividing by the standard deviation, thus yielding a distribution with a median of zero and a standard deviation of one. E-scores are a modified version of the AUROC statistic, which consider the relative ranking of probes containing a given 8-mer, and range from −0.5 to +0.5, with E > 0.45 taken as highly statistically significant [67].
DNA binding sites were determined as described in [72]. The oligo-binding array returns for each transcription factor a set of eightmers and corresponding E-scores that indicate the likelihood that the protein binds this eightmer. This set of eightmers contains both the forward and reverse binding eightmers and may represent different binding motifs for a single transcription factor. Hence to infer binding motifs for a transcription factor from a set of eightmers, we need to first cluster eightmers into similar groups, where we expect at least two clusters (‘forward’ and ‘reverse’) for each transcription factor, unless the binding motif is a palindrome. First we remove unreliable eightmers (those that have a score < 0.45). We then perform pairwise Smith-Waterman alignments using the water program from the EMBOSS package (with options: -nobrief -gapopen 5.0 -gapextend 2.0) to obtain a sequence similarity measure (the Smith-Waterman score) for each eightmer-pair. We take 40.0 –the Smith-Waterman score as a pairwise distance and use hierarchical clustering as implemented in scipy (average linkage: UPGMA) to obtain a hierarchical clustering of the eightmers. We split the resulting clustering trees into two clusters and manually checked whether these correspond to ‘forward’ and ‘reverse’ strands. We find that this is the case for all transcription factors except FOXG_15625 that probably has a palindromic binding site and FOXG_04904 for which we did not find ‘reverse’ strand eightmers. In the cases where we could identify ‘forward’ and ‘reverse’ strands we added the reverse complement of ‘forward’ strand sequences to ‘reverse’ strand sequences, in other cases we simply merged both clusters as they were. We aligned the sequences and obtained a sequence logo (as shown in Fig 3A) for these alignments with WebLogo [73].
The following sequence motifs were used to search the regions 1000 bp upstream or 1000bp downstream of the annotated transcriptional start site (Fusarium Comparative Sequencing Project, Broad Institute of Harvard and MIT (http://www.broadinstitute.org/), annotation 3); aTF2: CAAAC, cTF4: AGCC[A,G,C,T]TA, aTF5: CACGT, aTF7: G[A,G,C]GGCT, aTF1:T[A,G]CCG, SGE1: TTA[A,G][A,G][G,C]TT, effector motif: AACTGCCGA. For the upstream regions we used fasta files with promoter regions that we downloaded from the Broad Institute. We used custom Python scripts to append promoter regions for SIX genes that were not part of the annotation, based on the reported locations in [34]. We used custom Python scripts to make a fasta file with sequences that correspond to the first 1000 bp downstream from the first ATG based on the transcript gtf-file downloaded from the Broad Institute or–for SIX genes that were not part of the annotation–based on locations reported in [34]. We counted the number of genes with one or more motifs in the upstream regions, forward and reverse orientation separately. We did the same for two or more motifs and three or more motifs. Significant enrichment of genes with binding sites in the upstream regions in certain subgroups (accessory genes, plant-induced genes, etc.) was tested with a Hypergeometric test, with a P value < 0.01 after Bonferroni correction.
To define TF families we used blastp to search for homologs in 12 Fusarium oxysporum species (Fusarium Comparative Sequencing Project, Broad Institute of Harvard and MIT (http://www.broadinstitute.org/)[74]. We only included hits that have an E-value < 1e-5 and for which the alignment returned by BLAST spans more than 60% of both the query and the subject sequence. We used a custom Python script to cluster all hits into families using single linkage clustering and used Clustal Omega to construct multiple sequence alignments per family [75]. We trimmed the alignments using trimAl (-strictplus) [76], inspected and manually curated them. We used PhyML (with options: -q -b -2 -v e -a e) to infer phylogenies [77]. For large families we pruned the tree such that we keep the last common ancestor of the TFs that lie on chromosome 14 and their nearest neighbour that lies on the core, as root of the pruned tree.
For each of the seven transcription factor families tested for DNA binding, we searched for the occurrence of conserved domains from the Pfam database (version 27.0) using hmmscan from the hmmer package. We manually checked for presence of residues that are conserved in the Pfam seed alignment of the DNA binding domain in the multiple sequence alignments (S3 Fig).
We compiled a list of sequences corresponding to putative transposable elements by extracting DNA sequences for elements identified in a thorough analysis of chromosome 14 [34]. We combined these elements by elements identified by running RepeatMasker (with RepBase19.11) and extracting all sequences that were not denoted as a low-complexity region or simple repeat. We filtered out multiple occurrences of identical sequences.
For the transcription activation assay in yeast, plasmids (Ptef1-expression vector [B3909] with and without WOR1, Wor1 DNA binding site–WT, mutated or empty–in front of UAS-less CYC1 promoter followed by the LacZ reporter gene [B3946]) and strains (S. cerevisiae Sigma 2000 ΔYEL007 /ΔYHR177) were a kind gift from Alexander D. Johnson and are described in [42]. To express SGE1 in yeast, the SGE1 ORF was amplified from gDNA and cloned behind the TEF1 promoter in plasmid B3909 using SpeI and XhoI restriction enzymes. To express aTF1 or cTF1 in yeast, for each gene both exons were amplified from gDNA and fused together in an overlap PCR, to create an intronless sequence. This sequence was then cloned behind the TEF1 promoter in plasmid B3909 using SpeI and SalI restriction enzymes. To clone the Tf1 DNA binding site (WT or mutated), oligo pairs were ordered corresponding to part of the SIX1 promoter (-335 to –290 relative to the ATG) containing two Tf1 motifs flanked by 5 bp on each end plus sticky ends corresponding to the XhoI restriction site. Oligos were phosphorylated using T4 PNK (Fermentas), ligated into XhoI-digested and phosphatase treated B3946 plasmid. All plasmids were sequenced to verify the insert sequence and orientation. Oligos are listed in S1 Data, tab: ‘primers’.
Plasmids containing WOR1, SGE1, aTF1 or cTF1 and LacZ reporter plasmids were co-transformed to yeast strain Sigma 2000 ΔYEL007 /ΔYHR177 according to [78]. LacZ activity was assayed as follows. Transformed yeast cells were grown in SD medium lacking Uracil and Histidine. Before cells were harvested, OD600 was measured and cells were pelleted. The cell pellet was resuspended in 150μl Z-buffer with β-mercaptoethanol (60 mM Na2HPO4, 40 mM NaH2PO4, 10 mM KCl, 1mM MgSO4, 1 mM β-mercaptoethanol, pH 7), 50 μl chloroform and 20 μl 0.1% SDS were added, tubes were vortexed for 15 seconds and 700 μl pre-warmed (30°C) Z-buffer (60 mM Na2HPO4, 40 mM NaH2PO4, 10 mM KCl, 1mM MgSO4, pH 7) with ONPG (1 mg/ml) was added at t = 0. Tubes were incubated at 30°C until the reaction started to turn yellow. The reaction was stopped with 500 μl 1M Na2CO3 and the time recorded, Tubes were centrifuged 2 min. at 14000 rpm and the OD of the supernatant was measured at 420 nm. Miller units were calculated as follows: (A420*1000)/(A600*minutes*ml culture).
To make aTF1 overexpressors in the Fol4287 wild type background, the hygromycin resistance cassette (HPH) in the plasmid described above (Pfem1aTF1-HPH cassette) was exchanged for a phleomycine resistance cassette (BLE), using XbaI plus PacI (insert) and SpeI plus PacI (vector) restriction enzymes. The resulting plasmid (Pfem1aTF1-BLE cassette) was transformed to Fol4287 and transformants were selected on zeocine. An empty plasmid (pRW1p: containing only the BLE cassette [79]) was transformed to Fol4287 in parallel, as a negative control. Ten independent zeocine resistant colonies of each transformation were monospored and checked for SIX1 and SIX3 expression. For this, liquid cultures (100 ml 1% KNO3, 3% sucrose, 0.17% YNB w/o NH4 and aa.) were inoculated from plate and mycelium was harvested after 5 days at 25°C, 150–175 rpm. RNA isolation, cDNA synthesis and Q-RT-PCR were performed as described above. SIX1 and SIX3 levels were normalized to EF1-α. Two out of ten aTF1 transformants showed induction of SIX1 and SIX3 expression. One of those was selected for deletion of SGE1.
Deletion of SGE1 in the selected Fol4287 aTF1 overexpressor was done as described in [26], using the same deletion construct and the same PCR control for deletion of SGE1. To check transformants (ectopic and in locus) for SIX1 and SIX3 expression, liquid cultures (100 ml 1% KNO3, 3% sucrose, 0.17% YNB w/o NH4 and aa.) were inoculated from plate and mycelium was harvested after 5 days at 25°C, 150–175 rpm. RNA isolation, cDNA synthesis and Q-RT-PCR were performed as described above. SIX1, SIX3, SGE1 and aTF1 levels were normalized to EF1-α. Primers used are listed in S1 Data, tab: ‘primers’.
|
10.1371/journal.ppat.1000500 | NOD2, RIP2 and IRF5 Play a Critical Role in the Type I Interferon Response to Mycobacterium tuberculosis | While the recognition of microbial infection often occurs at the cell surface via Toll-like receptors, the cytosol of the cell is also under surveillance for microbial products that breach the cell membrane. An important outcome of cytosolic recognition is the induction of IFNα and IFNβ, which are critical mediators of immunity against both bacteria and viruses. Like many intracellular pathogens, a significant fraction of the transcriptional response to Mycobacterium tuberculosis infection depends on these type I interferons, but the recognition pathways responsible remain elusive. In this work, we demonstrate that intraphagosomal M. tuberculosis stimulates the cytosolic Nod2 pathway that responds to bacterial peptidoglycan, and this event requires membrane damage that is actively inflicted by the bacterium. Unexpectedly, this recognition triggers the expression of type I interferons in a Tbk1- and Irf5-dependent manner. This response is only partially impaired by the loss of Irf3 and therefore, differs fundamentally from those stimulated by bacterial DNA, which depend entirely on this transcription factor. This difference appears to result from the unusual peptidoglycan produced by mycobacteria, which we show is a uniquely potent agonist of the Nod2/Rip2/Irf5 pathway. Thus, the Nod2 system is specialized to recognize bacteria that actively perturb host membranes and is remarkably sensitive to mycobacteria, perhaps reflecting the strong evolutionary pressure exerted by these pathogens on the mammalian immune system.
| Bacterial and viral infection stimulates production of several cytokines and chemokines that are thought to protect the host against infection. The bacterial strain known to cause tuberculosis elicits production of type I interferons, yet it was unclear how the bacteria isolated within the cell was capable of stimulating this host response. This study reveals that the bacteria use a specialized system to cause damage to these cellular compartments and release bacterial products that activate intracellular innate immune pathways. In this work, we demonstrate that Nod2, Rip2, Tbk-1, Irf3 and Irf5 proteins cooperate to produce type I interferons. Understanding how these pathways are mediated is likely to aid in the design of more effective tuberculosis vaccines.
| Mycobacterium tuberculosis (Mtb), the causative agent of human tuberculosis, is an exquisitely adapted obligate human pathogen that is thought to persist within as many as one billion individuals worldwide [1]. This bacterium's ability to survive and replicate inside a modified phagosomal compartment of host macrophages is central to the pathogenesis of this disease [2]. While residing at this site, Mtb is able to persist for decades. However, a robust cell-mediated immune response effectively inhibits bacterial replication in approximately 90% of otherwise healthy individuals, and the infection can be controlled indefinitely. Deficits in this immune response result in progressive bacterial replication, necrosis of infected lung tissue, and spread to other individuals. Thus, like many other pathogens that cause chronic infections, the long-term survival of Mtb, depends on a delicate balance between bacterial virulence and host immunity.
Immunity to tuberculosis depends on both the innate and adaptive responses of the host. Initial recognition of the bacterium is mediated by pattern recognition receptors (PRR) such as Toll-like receptors (TLRs) [3],[4] or nucleotide binding oligomerization domain (NOD)-like receptors (NLRs) [5],[6], both of which recognize conserved microbial structures known as pathogen associated molecular patterns (PAMPs). TLRs monitor the extracellular environment and endosomal compartments, and recognize a variety of microbial components including bacterial lipoprotein, peptidoglycan, CpG DNA, and double- and single-stranded RNA [4]. NLRs constitute a more diverse family of approximately 25 proteins, including the caspase-recruiting domain (CARD)-containing Nod1, Nod2 and NLRCs, the pyrin (PYR) domain-containing NLRPs and the baculovirus-inhibitor-of-apoptosis-repeats (BIRs)-containing NLRBs. Nod1 and Nod2 reside in the cytosol and recognize microbial products in this compartment [7]. While the functions of most NLR's remain undefined, the Nod1 and Nod2 proteins have been shown to respond to bacterial cell wall fragments. The Nod1 protein recognizes a fragment of peptidoglycan (PGN) containing the dipeptide γ-d-glutamyl-meso-diaminopimelic acid (iE-DAP) produced by Gram-negative and some Gram-positive bacteria. Nod2 recognizes muramyl dipeptide (MDP) present on most types of PGN [8],[9],[10],[11]. While the recognition of these common forms of peptidoglycan have been extensively studied, bacteria modify their cell walls in a myriad of ways and the effects of these modifications on Nod1/2 recognition are only beginning to be appreciated (reviewed in [12],[13],[14]). For example, Listeria monocytogenes removes a common N-acetyl moiety from the glucosamine of its peptidoglycan, which renders the cell wall resistant to host lysozyme and thereby inhibits bacterial recognition by Nod1 [15]. In contrast, mycobacteria, replace the N-acetyl group of the muramic acid of MDP with a N-glycolyl moiety[16],[17], and this modification significantly increases the potency of this compound as a Nod2 agonist (Coulombe, F. and Behr, M.A. unpublished data).
Nod1 and Nod2 functions depend on a downstream signaling component, Rip2, which belongs to a protein family currently consisting of 7 members [18]. Like the prototype Rip1, Rip2 contains an N-terminal serine threonine kinase domain followed by an intermediate region and a C-terminal caspase recruitment domain (CARD). Rip2 has been shown to be essential for cytosolic Nod1/2 signaling, and its overexpression stimulates NF-κB activity and induces apoptosis [19],[20]. We have shown that Rip2 is stably modified with ubiquitin in cells treated with the Nod2 agonist MDP [21]. This modification is required for Nod1-mediated NF-κB activation [22], indicating that stable polyubiquitination is a critical component of this signaling cascade.
Intact Mtb bacilli are recognized by both TLRs and NLRs, which cooperatively respond to infection and synergistically induce NF-κB activation [23]. However, a large fraction of the transcriptional response to Mtb, including many immunologically important proteins, such as the chemokines RANTES and IP-10, and the inducible nitric oxide synthase enzyme NOS2 that is critical for mycobacterial immunity, are induced independently of TLR2/4 and the adapter proteins MyD88, MAL and TRIF. Instead these responses rely on autocrine or paracrine signaling via type I interferons (IFNα/β), which are induced through largely undefined pathways [24].
Despite the ability of cell surface localized TLR4 to trigger IFNα and IFNβ transcription, existing evidence indicates that during genuine bacterial infections, this response instead requires the recognition of bacterial products in the cytosol. This has been most clearly demonstrated for pathogens that replicate in the host cell cytosol, such as Listeria monocytogenes and Francisella tularensis. In both cases, the bacterium must disrupt the phagosomal membrane and escape into the cytosol in order to trigger the type I IFN response in resting macrophages [25],[26],[27]. Despite its residence in the phagosome, Mtb still induces rapid and robust IFNα/β transcription, and this response depends on a specialized secretion system of the bacterium, ESX1 [28]. This system has been suggested to contribute to the perturbation of the phagosomal membrane [29],[30],[31], indicating that cytosolic recognition might be critical for IFNα/β responses to diverse bacterial pathogens including Mtb.
The primary pathways leading to IFNα/β induction upon bacterial infection remain obscure. Since transfection of DNA into the cytosol of macrophages can induce a Tbk-1 and Irf3-dependent IFNα/β response similar to that seen upon L. monocytogenes infection, bacterial DNA has been implicated as the eliciting stimulus [32]. Two different cytosolic DNA sensors have been identified, DAI [33] and AIM2 [34], but their importance during bacterial infections remains to be demonstrated. While Nod2 recognition of MDP is not absolutely required for IFNα/β production [35], it has been shown to synergize with the cytosolic DNA response and enhance IFN production during both L. monocytogenes and Mtb infection [36]. However, Nod2 stimulation alone is thought to be insufficient to induce type I IFN production [36].
In sum, while a large fraction of the macrophage response to Mtb infection depends on type I IFN [24] and therefore is likely to rely on a cytosolic signaling pathway, the bacterial products recognized and the pathways involved remain unknown. We previously found that Mtb infection of macrophages triggers Rip2 polyubiquitination in a TLR and MyD88 independent manner [21]. We now show that this stimulation is due to the ESX1-dependent entry of bacterial products into the cytosol where they are recognized by Nod2, implicating MDP as the relevant PAMP. Unexpectedly, this results in IFNα/β production that is dependent on a novel pathway consisting of Nod2, Rip2, Tbk1, and Irf5. This work is the first to implicate NLRs in IRF activation and to suggest a role for Irf5 in anti-bacterial innate immune responses. Furthermore, we found that the unusual N-glycolyl MDP produced by Mtb was 10–100 fold more potent than the commonly studied N-acetylated MDP produced by most bacteria, and that only N-glycolyl MDP could stimulate Rip2-dependent IFNα/β transcription in the absence of other stimulants. Thus, the mammalian Nod2 pathway appears to be remarkably sensitive to mycobacterial MDP and responds to infection by triggering the production of type I interferon, which is responsible for a significant component of the transcriptional response to Mtb infection.
The ability of Mtb to rapidly modify macrophage signaling and vesicular sorting pathways [2] suggests that bacterial products gain access to the cytosol soon after phagocytosis. These products are, in turn, likely to be sensed by the host and trigger the innate immune response. Previously, we demonstrated that Mtb rapidly induces the TLR2/4 independent polyubiquitination of the Rip2 protein [21], an event that could represent the initiation of cytosolic recognition. To characterize these events in more detail, we infected the mouse macrophage cell line RAW 264.7 or primary bone marrow derived macrophages (BMDM) with live or heat killed Mtb. In both cell types, we observed that infection with live, but not heat-killed, Mtb stimulated the rapid polyubiquitination of Rip2. The Mtb-induced ubiquitin modification reached maximal levels within 1 hour post-infection and declined by 4 hours (Figure 1A). Furthermore, pretreatment of cells with cytochalasin D to inhibit phagocytosis reduced Rip2 polyubiquitination in a dose-dependent manner (Figure 1B), indicating that the bacteria must be both live and intracellular to initiate this response.
Since Nod1 and Nod2 have been implicated in the cytosolic recognition of mycobacterial components [23], we sought to determine if Rip2 polyubiquitination depended on these proteins. In contrast to cells from wild type mice, inducible Rip2 polyubiquitination was not observed in macrophages derived from mice lacking Nod1 and Nod2, and was greatly reduced in cells lacking only Nod2 (Figure 1C). These data confirmed that intracellular Mtb is recognized by a Nod2-dependent pathway and that this protein is required for the stable ubiquitination of Rip2.
Live intracellular Mycobacteria were required to stimulate the Nod-Rip2 pathway, indicating that the bacterium actively participated in this process, likely via the translocation of bacterial products into the cytosol. A specialized protein secretion system, encoded by the ESX1 locus, has been implicated in the perturbation of the host membranes [29],[30],[37] and for stimulation of the type I IFN response [28] and inflammasome activation [38], suggesting that this system might contribute to cytosolic recognition via Nod proteins. In order to test this hypothesis, we infected the mouse macrophage cell line RAW 264.7 with wild type Mtb or mutants lacking ESX1 function. No induction in Rip2 polyubiquitination was observed upon infection with a strain of Mtb harboring the “RD1” mutation, which deletes a portion of the ESX1 locus [39]. Similarly, a mutant lacking espA, a distally-encoded gene that is required for ESX1-mediated secretion [40], also failed to elicit this response (Figure 2). The phenotype of the latter mutant could be complemented by the expression of espA from a plasmid vector, demonstrating that the inability to stimulate Rip2 polyubiquitination was linked to the espA mutation. Furthermore, M. bovis BCG, an attenuated vaccine strain carrying the RD1 deletion and therefore lacking ESX1 function [39], was unable to stimulate Rip2 polyubiquitination. While all of these ESX1 mutants are less virulent than wild type bacteria, the lack of Nod2-Rip2 stimulation did not appear to be a nonspecific effect of attenuation. Two unrelated bacterial mutants that are unable to grow intracellularly, a biotin auxotroph (ΔbioF [41]) and a small molecule efflux mutant (TN::rv1410c [42]), robustly stimulated this response (Figure 2). Taken together, these observations indicate that a functional ESX1 secretion system is specifically required for Nod2 stimulation.
Since the Mtb-induced Rip2 polyubiquitination required ESX1, we hypothesized that this system might be responsible for the release of Nod2 ligands into the cytosol, perhaps via the disruption of vacuolar membrane integrity. However, it also remained possible that ESX1-deficient strains simply lacked a critical PAMP or other Nod2 stimulating activity. To distinguish between these possibilities, we investigated whether ESX1 function could be complemented by two exogenous membrane-disruptive activities. Streptolysin O (SLO) is a cholesterol-dependent toxin that introduces pores directly into mammalian membranes. Pores can also be introduced by adding ATP to macrophages, resulting in stimulation of the P2X7 receptor and the subsequent opening of the hemichannel, pannexin-1 (PANX1) [43]. We observed that membrane perturbation by either of these two methods resulted in robust Rip2 polyubiquitination upon infection with espA-deficient bacteria, which were otherwise unable to induce this response (Figure 3). The involvement of PANX1 in the ATP-facilitated Rip2 ubiquitination was verified by the addition of a competitive inhibitory peptide of the PANX1 pore. This peptide, but not a scrambled control peptide, inhibited Rip2 polyubiquitination to levels observed in cells infected with the ΔespA mutant (Figure 3). While the K+ flux subsequent to membrane damage has been found to stimulate NLRs in some circumstances [5], we found that the addition of ATP or SLO alone resulted in a minimal response. These data indicate that SLO, PANX1 and ESX1 are all likely to promote Nod2 pathway activation via a similar mechanism, by facilitating the release of a stimulatory mycobacterial component into the cytosol. Since this pathway appears to be specific for peptidoglycan fragments, mycobacterial MDP-containing fragments were the most likely candidates.
The inability of ESX1 mutants to stimulate either the Nod2-Rip2 pathway or the type I IFN response [28] led us to hypothesize that the Nod2 pathway may mediate type I IFN expression in this system. To investigate a potential link between Nod2 and IFNα/β, we infected Nod2- or Rip2-deficient macrophages with Mtb, and measured the induction of IFNα and IFNβ mRNAs using real time PCR (qRT-PCR). In the absence of Rip2, IFNβ induction was reproducibly reduced approximately 3-fold, whereas IFNα induction was almost completely abrogated (Figure 4A and B). Nod2 deficiency had a similar effect on both IFNα and IFNβ transcription, consistent with its requirement for Rip2 polyubiquitination. Nod1 appears to play no role in this pathway, as nod1−/− macrophages produced wild type levels of IFNβ (Figure S1). The decreases in mRNA abundance observed in rip2−/− and nod2−/− cells were reflected in a similar decrease in protein production, as measured by ELISA (Figure 4C and D).
In order to assess the importance of Nod2 and Rip2 to the downstream IFNα/β-dependent macrophage response, we quantified the induction of RANTES mRNA, which depends on type I IFN secretion and signaling via the IFNαβ receptor (IFNAR1) in this infection model [24]. We found that in the absence of Rip2 or Nod2, Mtb infection failed to induce RANTES expression (Figure 4E). These data suggest that the effect of a Rip2 deficiency on downstream type I IFN responses may be even more pronounced than the IFNβ mRNA levels indicate. In contrast, TNFα mRNA levels were unaffected by Nod2- or Rip2-deficiency (Figure 4F) indicating that other pattern recognition pathways remained responsive to Mtb in these cells.
Consistent with previous work [28], we found that infection with ESX1 mutant bacteria induced significantly less IFNβ and RANTES expression than wild type bacteria (Figure 5). To test whether ESX1-mediated type I IFN expression was mediated solely via Rip2, we infected Rip2-deficient macrophages with ESX1 mutant bacteria and quantified IFNβ and RANTES mRNA levels. We found that in the absence of Rip2, the loss of ESX1 function resulted in a further decrease in IFNβ mRNA levels (Figure 5A), suggesting the presence of an additional host pathway(s) that contribute to IFNβ induction. However, Rip2 deletion had no significant effect in the absence of ESX1 (Figure 5B), supporting our biochemical evidence that NOD2 stimulation depends entirely the ESX1-dependent delivery of stimulants into the cytosol.
While our data indicated that a significant fraction of the IFNα/β response could be attributed to the Nod2-Rip2 pathway, it has been suggested that MDP stimulation alone is unable to induce type I IFNs and can only augment responses triggered by other pathways [36]. Indeed, we also found that the N-acetylated MDP that is commonly used to stimulate Nod2 was a very poor inducer of IFNβ and RANTES expression (Figure 6A and B). However, our preliminary studies investigating Rip2 polyubiquitination indicated that Mtb was a particularly potent stimulator of this response [21], and therefore we reasoned that this could be due to the N-glycolylated form of MDP produced by Mtb. To determine if this form of MDP was sufficient to induce type I IFN responses, we compared the ability of N-acetyl- and N-glycolyl-MDP to stimulate IFNβ expression. In contrast to N-acetyl MDP, treatment with the N-glycolylated form stimulated a robust IFNβ response, which was entirely dependent on Rip2 and Nod2 (Figure 6). In addition, at least 30-fold less N-glycolyl-MDP was necessary to stimulate the IFNβ transcription. Thus, the Nod2/Rip2 pathway alone is sufficient to induce the production of the IFN response when stimulated with this potent form of MDP.
Listeria monocytogenes infection induces a potent host type I IFN response mediated by the Tbk1 kinase and Irf3 [27],[32],[35],[44]. To test whether Mtb infection triggered similar pathways, we infected Irf3-deficient and Tbk1/Tnfr1-deficient macrophages with Mtb and measured IFN induction. The Tnfr1 deficiency was necessary to suppress the embryonic lethality of Tbk1 deletion [45]. Similar to the L. monocytogenes model, we found that IFNβ induction by Mtb infection was completely dependent upon Tbk1, and the loss of Tnfr1 had little effect (Figure 7A). However, in contrast to the complete dependence on Irf3 observed for L. monocytogenes [27],[32],[35], we found IFNβ expression was reduced, but not ablated when Irf3-deficient macrophages were infected with M. tuberculosis (Figure 7A). This partial dependence on Irf3 was not changed by varying the multiplicity of infection (Figure S2). These data prompted us to test whether other IRFs mediate Nod2-dependent type I IFN responses.
Induction of IFNβ expression is dependent on the formation of the enhancesome which includes the NF-κB, ATF-2, c-jun, Irf3 and Irf7 transcription factors [46]. Irf5 is a related family member that has also been shown to contribute to induction of type I IFN responses triggered by TLRs, and overexpression of MyD88 has been shown to synergize with Irf5 to induce IFNβ expression [47]. Based on these studies, we tested whether RIP2 collaborates with IRF5 or IRF3 to stimulate IFNβ luciferase reporter activity. HEK293 cells were transfected with an IFNβ-luciferase reporter construct, along with increasing amounts of expression plasmids encoding RIP2, MyD88, IRF3 or IRF5. RIP2 and IRF5 coexpression stimulated IFNβ promoter activity in a dose dependent manner and to a similar extent as MyD88 and IRF5 (Figure 7B). In contrast, RIP2 and IRF3 expression failed to induce this robust response (Figure 7C). RIP2 and IRF5 expression also stimulated IFNα4 promoter activity as well as a reporter construct containing multimerized ISRE elements (data not shown).
To further investigate the contribution of Irf5 to the anti-bacterial type I IFN response, we infected macrophages from Irf5-deficient mice and control littermates with either Mtb or L. monocytogenes, and measured IFNβ expression. Consistent with the luciferase reporter studies, we found that Mtb-induced IFNβ (Figure 7D) and IFNα (Figure S3) expression was impaired in the absence of Irf5. In contrast, the response to Listeria was unaffected by the loss of Irf5 (Figure 7D). While the related Rip1 adaptor protein regulates Irf7 activity in innate anti-viral signaling [48], we found that IFNβ induction after Mtb infection was unaffected by Irf7 deficiency (data not shown). To rule out the possibility that Irf3 expression levels may also be affected in irf5−/− macrophages, we verified that the Irf3 protein level was unchanged in Irf5-deficient cells (Figure S4). These results indicated that Mtb infection stimulates type I IFN expression via a pathway that depends on Nod2, Rip2, Tbk1, and Irf5. This contrasts with the pathway triggered by L. monocytogenes, which depends entirely on Irf3 and not Irf5. We reasoned that this dependence on different Irf proteins might be explained by the preferential stimulation of a Nod2-Rip2-Irf5 pathway by mycobacterial peptidoglycan. Consistent with this model, we found that the IFNβ induction triggered by N-glycolyl MDP was entirely dependent on Irf5 and independent of Irf3 (Figure 8), functionally linking Irf5 with the Nod2 pathway.
Mammals first detect microbial infections via an array of PRRs that include both cell surface TLRs and cytosolic NLRs. However, not all microbial interactions represent a pathological state, and the immune system must be able to discriminate to some degree between colonization by commensal organisms and dangerous infection. One level of discrimination is provided by the desensitization or anatomical sequestration of TLRs at sites of chronic stimulation, such as the gut, which presumably allows for tolerance to normal flora [49],[50]. Bacterial pathogens can still be recognized at these sites via NLRs, since these systems rely on the specific ability of pathogens to translocate PAMPs into the host cytosol.
The concept that NLRs are specific for pathogenic organisms that disrupt host membranes is supported in a number of bacterial systems in which the loss of specific virulence functions abrogates NLR signaling. For example, in resting macrophages, cytosolic recognition of L. monocytogenes requires the pore-forming toxin, listeriolysin O [26],[27]. Similarly, Helicobacter pylori [51] and Legionella pneumophila [32] mutants lacking a functional type IV secretion system (T4SS), and Shigella flexneri [52] or Salmonella enterica serovar typhimurium [52] mutants lacking a functional type III secretion system (T3SS) fail to stimulate NLR pathways. In each case, the virulence system in question is responsible for host membrane damage and the likely translocation of bacterial products into the cytosol where they can be recognized by NLRs and/or other cytosolic surveillance systems.
Similarly, we found that the ESX1 specialized protein secretion system of Mtb is required for Nod2 recognition. While it has been suggested that type I IFN induction via ESX1 might represent a specific immunomodulatory virulence strategy [28], analogies to these other pathogens suggests that perhaps NLR recognition is simply a byproduct of a membrane damaging function that allows bacterial products to enter the cytosol. This model is supported by our observations that other membrane perturbing agents, such as SLO and PANX1 can substitute for ESX1 function and allow cytosolic recognition. Thus, in a number of cases it appears that NLRs can be considered as sentinels for pathogens that rely on membrane damage as a pathogenic strategy.
Based on their common role in protein secretion and in facilitating cytosolic recognition, it is tempting to speculate that ESX1 and Gram-negative T3SS and T4SS function analogously to deliver effector proteins into the host cytosol. Despite these similarities, the role played by ESX1 during infection remains unclear, since no translocated effectors have been identified to date. In both Mtb and M. marinum, a related pathogen of ectotherms, ESX1 has been implicated in host membrane disruption and one of the major substrates of this system, EsxA, has been proposed to possess a membrane-lytic activity [30],[37]. This single activity could be sufficient to account for the delivery of MDP and other PAMPs to the cytosol. It remains to be determined whether perturbing host membranes is the only role played by ESX1 during infection, or if this system also serves additional functions analogous to the specialized secretion systems of other pathogens.
A major consequence of the cytosolic recognition of Mtb is the induction of type I IFN. While the importance of this response in viral defense is clear and virtually universal, its role in antibacterial immunity appears to vary. Mice deficient in the type I IFN receptor, Ifnar1, are significantly more susceptible to several Gram-positive and -negative bacterial infections [53],[54],[55],[56], indicating that IFNα/β are important for immunity to many bacteria. However, Ifnar1 mutation has the opposite effect on the outcome of L. monocytogenes infection [57], suggesting that IFNα/β can also exacerbate disease. The role played by IFNα/β in Mtb infection remains somewhat uncertain. The induction of several immunologically important genes, including NOS2, depend on IFNα/β, suggesting a protective role. Initial studies of mouse and human infections appeared to support this view [58],[59]. However, like the L. monocytogenes system, mutation of the IFNα/β receptor has in most cases been associated with decreased bacterial burden in mouse models of tuberculosis [28],[59],[60],[61],[62]. IFNα/β may fail to protect against disease because Mtb inhibits the response to these cytokines in infected macrophages [63]. The ultimate influence of IFNα/β on Mtb infection appears to depend on a number of experimental factors, which might include host species, bacterial strain, route of infection and dose. Despite these differences however, some important themes emerge from these studies. Most importantly, the effect of IFNα/β is most apparent after the onset of adaptive immunity and not before, suggesting that the major role-played by type I IFNs during tuberculosis may be to instruct the priming or maintenance of the adaptive immune response and perhaps to control the differentiation of regulatory T cells [59].
A variety of bacterial pathogens trigger the type I IFN response, and a paradigm has begun to emerge regarding the induction of this response by bacteria. One current model suggests that bacterial DNA translocated into the host cytosol is the major eliciting agent. This model is based largely on the observations that infection with L. monocytogenes or L. pneumophilla, or transfection of DNA into the cytosol induces a similar IFNβ response that is Rip2 independent, and Tbk1- and Irf3-dependent [32]. Other PAMPs, such as MDP, can provide a synergistic IFN-inducing stimulus, but have not appeared to be sufficient for induction of IFNβ in the absence of other triggers [36].
In contrast, our data support a model whereby Nod2 stimulation by Mtb infection induces the polyubiquitination of Rip2, which acts via the Tbk1 kinase to stimulate the activity of Irf5 and induce transcription of IFNα/β. This differs from the pathway triggered by other bacteria such as L. monocytogenes, which depends entirely on Irf3 in resting macrophages [32] and does not involve Irf5 (Figure 7). Although Irf5 has previously been shown to be activated by the MyD88-dependent TLR7 and TLR9 pathways, this work reveals a novel role for this protein in Nod2 signaling, and a new link between Nod proteins and the type I IFN response. Furthermore, we found that unlike the N-acetylated MDP found in many bacteria, stimulation with the N-glycolylated MDP derivative found in mycobacteria was sufficient to stimulate the IFN response in the absence of other stimuli.
A significant component of IFNβ induction remains intact upon Mtb infection of Rip2-deficient macrophages (Figures 4 and 5), indicating that additional pathways are also involved. Since virtually all IFNβ expression is ESX1-dependent, it appears that the residual induction observed in rip2−/− macrophages also depends on cytosolic recognition pathways. These pathways could certainly include a DNA sensor that acts via Irf3, as proposed for other infections, since Irf3 deficiency had a moderate effect on IFNβ expression in our experiments (Figure 7A and S2). Thus, our data do not imply that Mtb is stimulating IFNα/β in a fundamentally different manner from other bacteria. Instead, it is likely that bacterial pathogens stimulate the IFN response via multiple, partially redundant pathways, and that the relative importance of each is determined by the unique biology of the infection. In the case of Mtb, we speculate that the N-glycolylation of its peptidoglycan, and perhaps a paucity of other stimulants such as DNA, favor recognition via Nod2. It is also possible that the balance of these pathways might be affected by the activation state of the macrophage. When resting macrophages are infected with L. monocytogenes, the IFN response requires LLO and is completely Irf3 dependent. In contrast, IFNγ-stimulated cells are able to deliver this bacterium to the lysosome, where the cell wall is degraded to produce abundant peptidoglycan fragments. In this situation, a significant component of the IFNβ induction depends on Nod2 and not Irf3 [64]. While Irf5 was not investigated in this study, it is possible that this represents another situation in which robust Nod2 signaling promotes a Nod2- and Irf5- dependent type I IFN response.
While we found that loss of Nod2-Rip2 signaling only partially reduces the induction of IFNβ, Rip2 deletion completely abrogated IFNα and RANTES expression. These results can be explained by the structure of the IFN regulatory circuit. Initially, only IFNβ is expressed, and subsequently IFNα and other interferon regulated genes (IRGs), such as RANTES, are induced via an Ifnar1 and Irf7-dependent autocrine/paracrine signaling pathway [65]. Thus, it appears that the decrease in IFNβ expression that we observe is sufficient to severely impair downstream IRG induction, at least in this cell culture model.
Multiple steps of this pathway are likely to depend on stable ubiquitin modifications. Not only did we observe that Rip2 is polyubiquitinated upon infection, but we also found that a Rip2 point mutant that cannot be stably ubiquitin modified is unable to mediate IFNα/β induction in response to Mtb infection (Figure S5). Collectively, these data suggest that polyubiquitinated Rip2 is required for Mtb-induced type I IFN expression via Irf5. Interestingly, MyD88-dependent activation of Irf5 involves formation of a tertiary complex that includes the E3 ubiquitin ligase, Traf6 [66],[67]. This E3 ubiquitin ligase associates with Rip2 upon MDP stimulation, raising the possibility that a Rip2-Traf6-Irf5 complex might exist and that the activity of Irf5 might also be regulated by ubiquitin.
The specificity of the innate immune system has been shaped by the very powerful natural selection imposed by microbial pathogens. Our work suggests that upon infection with Mtb, a particularly potent form of MDP is translocated into the host cell cytosol where it triggers a novel signaling pathway leading to the robust induction of the type I IFN response. It is unlikely to be coincidental that the active component of our most potent adjuvant, complete Freund's adjuvant (CFA), consists of mycobacterial cell fragments. The specific pathway described in this work might play a major role in this adjuvant's effectiveness, since IFNα/β production is required for CFA to promote antigen-specific immune responses (55). Thus, while PAMPs are often regarded as invariant microbial components, it is clear that functionally important pathogen-specific differences exist in the composition of these molecules, and that the immune system can differentiate these subtly distinct structures.
Given the potent adjuvant activity of mycobacterial components, it is somewhat surprising that the attenuated vaccine strain M. bovis BCG, which produces the same PAMPs present in CFA, provides poor protection against pulmonary TB in adults [68],[69]. The lack of ESX1 function in this strain appears to be at least partially responsible, since the reconstitution of ESX1 improves the efficacy of this vaccine [70],[71]. While this effect has previously been attributed to either the secretion of additional antigens or altered antigen presentation, it is also possible that ESX1 activity improves immunity by delivering crucial PAMPs into the cytosol where they are optimally recognized. Understanding both the details of PAMP trafficking, as well as the precise specificity of PAMP recognition, promises to aid in both the design of improved adjuvants and more effective tuberculosis vaccines.
C57BL/6 mice ages 8–12 weeks were obtained from the Jackson Laboratory. rip2−/− mice were a kind gift from Dr. Vishva M. Dixit (Genentech, Inc. South San Francisco, CA). nod2−/− mice were provided by Dr. Peter J. Murray (Department of Infectious Diseases, St. Jude Children's Research Hospital, Memphis, TN). nod1−/− and nod1−/−nod2−/− mice were provided by Dr. Gabriel Nunez (University of Michigan Medical School, Ann Arbor, MI). irf3−/−, irf5−/−, tbk1+/+tnfr1−/− and tbk1−/−tnfr1−/− mice and their littermate controls were provided by Dr. Kate A. Fitzgerald (University of Massachusetts Medical School, Worcester, MA). Mice were housed under specific pathogen-free conditions, and in accordance with the University of Massachusetts Medical School, IACUC guidelines.
The WT strain of M. tuberculosis used in these studies was the H37Rv strain. All the mutants were derived from the wild type strain. ΔESX-1 was obtained from D. Sherman (SBRI, Seattle.WA) [39]. ΔBioF, ΔRv3616 and ΔRv3616-complemented strains have been described previously [40],[41]. TN::Rv1410 contains a himar-1 transposon inserted at nucleotide #688 of the 1557 bp predicted open reading frame [72]. All strains were cultured in 7H9 medium containing 0.05% Tween 80 and OADC enrichment (Becton Dickinson). Pre-titered stocks of Listeria monocytogenes strain 10403 stored at −80°C (kindly provided by Victor Boyartchuk) were recovered for 1 hr at 37°C in 9 ml of Tryptic Soy Broth (BD Biosciences). Bacteria were then washed and resuspended in PBS prior to infection.
Anti-Rip2 (Rabbit) and anti-ubiquitin (Mouse) antibodies were obtained from Santa Cruz Biotechnology. Anti-Irf3 antibody was obtained from Zymed. Anti-Irf5 antibody was obtained from Abcam. Anti-β-actin antibody was obtained from Sigma. MDP was obtained from InvivoGen. Mouse TNF-α was obtained from Sigma. LPS derived from Escherichia coli strain 0111.B4 was purchased from Sigma, dissolved, treated with deoxycholate, and re-extracted with phenol/chloroform as described in [73]. The pannexin-1 mimetic blocking peptides panx1 (WRQAAFVDSY) and the scrambled peptide control were synthesized by GeneScript Corporation (Piscataway, NJ) and have been described previously [74]. Streptolysin O (SLO) a pore forming protein derived from Streptococcus and Adenosine 5′- triphosphate (ATP) were purchased from Sigma. N-glycolyl muramyl dipeptide (N-glycolyl MDP) was custom synthesized (Carbohydrate Synthesis, Oxford, UK) and shown to be more than 95% pure by NMR spectrometry. This preparation was found to be free of endotoxin contamination using the Limulus amebocyte lysate assay (Pyrotell, Cape Cod Inc., MA).
Bone marrow from 8- to 10-week-old mice was harvested from femurs and differentiated into macrophages for 7 days in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% L929-cell conditioned medium, 10% fetal bovine serum, 2 mM L-glutamine and 1 mM sodium pyruvate. After 7 days in culture, bone marrow derived macrophages (BMDMs) were washed with phosphate-buffered saline (PBS) and seeded into tissue culture plates for infection. RAW 264.7 macrophage cell line was cultured in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum. All Mtb strains were cultivated in 7H9 broth, grown to exponential phase and washed thoroughly in DMEM media prior to infection. Bacterial clumps were removed by passing the washed suspension through a 5 µm syringe filter. For the peptide blocking studies, the cells were pre incubated with the desired peptides for 30 minutes followed by ATP or SLO for additional 15 minutes. Macrophages were infected at an MOI of 10 for 1 or 2 hours after which filtered cell lysates were immunoprecipitated with anti-Rip2 antibody (Santa Cruz). Heat inactivation was achieved by incubating the bacteria at 80°C for 30 minutes. Immortalized macrophage cell lines from wild type, rip2−/−, nod2−/− and nod1−/−nod2−/− mice were established by infecting bone marrow cells with a v-raf/mil and v-myc retrovirus in the presence of GM-CSF and polybrene [75],[76]. These rip2−/−, nod2−/− and nod1−/−nod2−/− macrophage cell lines express CD11b and Gr-1 and are capable of phagocytosing antibody coated beads. To determine the effect of cytochalasin D on the phagocytic function of the macrophages, we used the Vybrant phagocytosis assay kit to quantify the uptake of fluorescent E. coli. This assay was performed according to the protocol provided by the manufacturer.
For the immunoprecipitation and ubiquitination assays, cell lysates were prepared in radioimmune precipitation assay (RIPA) buffer (150 mM NaCl, 50 mM Tris-HCl (pH 7.5), 1% NP40, 0.25% deoxycholate, 0.1% SDS, 1 mM EDTA), supplemented with protease inhibitors (Roche Applied Science) and 5 mM N-Ethylmaleimide (Sigma), immunoprecipitated with anti-Rip2 antibody (Santa Cruz). Polyubiquitinated Rip2 proteins were detected by immunoblotting with an anti-ubiquitin antibody (Santa Cruz). Total immunoprecipitated Rip2 protein was measured by immunoblotting with anti-Rip2 antibodies (Santa Cruz).
HEK293 cells (2×104) seeded in 96 well plates were transfected with 40 ng of the IFNβ luciferase reporter plasmid together with a total of 100 ng of various expression plasmids using GeneJuice (Novagen). The total amounts of transfected DNA were kept constant in all experiments by adjustment with empty vector. Luciferase activity was measured 24 h later using Dual Luciferase reporter assay system (Promega). The Renilla luciferase gene (40 ng) was co-transfected and used as an internal control plasmid. IFNβ luciferase reporter activity was normalized to Renilla luciferase reporter activity. Each experiment was repeated three times. Data are expressed as mean±s.d. of three replicates.
To measure IFNα/β mRNA levels upon MDP treatment or Mtb infection, total RNA was extracted from the macrophage cultures using Trizol reagent (Invitrogen) according to the manufacturer's directions. cDNA was prepared from 2 µg of total RNA and quantitative real-time PCR performed using SYBR green as a label with the following primers: mIFNα-F, 5′-AAGATGCCCTGCTGGCTG; mIFNα-R, 5′-TTCTGCTCTGACCACCTCCC; mIFNβ-F, 5′-CGTCTCCTGGATGAACTCCAC; mIFNβ-R, TGAGGACATCTCCCACGTCA; β-actin-F, 5′-CGAGGCCCAGAGCAAGAGAG; β-actin-R, 5′-CGGTTGGCCTTAGGGTTCAG; mTNFα-F, CAGTTCTATGGCCCAGACCCT; mTNFα-R, CGGACTCCGCAAAGTCTAAG; mRANTES-F, GCCCACGTCAAGGAGTATTTCTA; mRANTES-R, ACACACTTGGCGGTTCCTTC. Results shown are representative of more than three separate infection experiments, with each PCR performed in triplicate. All values reported were in the linear range of the experiment and were normalized to β-actin values. Standard curves were generated by linear dilution of a cDNA sample generated from poly I∶C-stimulated macrophages.
IFNα protein in cell culture supernatants was performed using a custom ELISA as described previously [77]. IFNα concentrations were calculated using a recombinant IFNα (HyCult, Biotechnology, Uden, Netherlands) standard curve performed in quadruplicate using linear regression, and expressed in units per ml. IFNβ protein in cell culture supernatants was measured similarly using a custom ELISA as described in [78].
|
10.1371/journal.pgen.1003479 | A Compendium of Nucleosome and Transcript Profiles Reveals Determinants of Chromatin Architecture and Transcription | Nucleosomes in all eukaryotes examined to date adopt a characteristic architecture within genes and play fundamental roles in regulating transcription, yet the identity and precise roles of many of the trans-acting factors responsible for the establishment and maintenance of this organization remain to be identified. We profiled a compendium of 50 yeast strains carrying conditional alleles or complete deletions of genes involved in transcriptional regulation, histone biology, and chromatin remodeling, as well as compounds that target transcription and histone deacetylases, to assess their respective roles in nucleosome positioning and transcription. We find that nucleosome patterning in genes is affected by many factors, including the CAF-1 complex, Spt10, and Spt21, in addition to previously reported remodeler ATPases and histone chaperones. Disruption of these factors or reductions in histone levels led genic nucleosomes to assume positions more consistent with their intrinsic sequence preferences, with pronounced and specific shifts of the +1 nucleosome relative to the transcription start site. These shifts of +1 nucleosomes appear to have functional consequences, as several affected genes in Ino80 mutants exhibited altered expression responses. Our parallel expression profiling compendium revealed extensive transcription changes in intergenic and antisense regions, most of which occur in regions with altered nucleosome occupancy and positioning. We show that the nucleosome-excluding transcription factors Reb1, Abf1, Tbf1, and Rsc3 suppress cryptic transcripts at their target promoters, while a combined analysis of nucleosome and expression profiles identified 36 novel transcripts that are normally repressed by Tup1/Cyc8. Our data confirm and extend the roles of chromatin remodelers and chaperones as major determinants of genic nucleosome positioning, and these data provide a valuable resource for future studies.
| The genome in eukaryotic cells is packaged into nucleosomes, which play critical roles in regulating where and when different genes are expressed. For example, nucleosomes can physically block access of transcription factor to sites on DNA or direct regulatory proteins to DNA. Consistent with these roles, nucleosomes assume a stereotypical pattern around genes: they are depleted at the promoter region that marks the start of genes and assume a regularly spaced array within genes. To identify factors involved in this organization, we generated high-resolution nucleosome and transcriptome maps for 50 loss-of-function mutants with known or suspected roles in nucleosome biology in budding yeast. We show that nucleosome organization is determined by the combined effects of many factors that often exert opposing forces on nucleosomes. We further demonstrate that specific nucleosomes can be positioned independently within genes and that repositioning of nucleosomes at the start of genes may affect expression of these genes in response to environmental stimuli. Data mining of this extensive resource allowed us to show that general transcription factors act as insulators at diverging promoters to prevent the formation of cryptic transcripts, and also revealed 36 novel transcripts regulated by the Tup1/Cyc8 complex.
| Chromatin is comprised of repeating units of nucleosome particles [1], [2] consisting of approximately 147 base pairs (bp) of DNA wrapped around a core histone octamer [3], [4]. The presence of nucleosomes and their relative occupancy on DNA can influence the access of proteins to DNA and they therefore play key roles in regulating DNA transactions such as replication and transcription [5]. Indeed, many of the effects exerted by transcriptional regulators in yeast and metazoans are mediated through interactions with regulators and coregulators that change the chromatin state to a more open (in case of activation) or closed conformation (in case of repression) [6]. Correspondingly, disruption of nucleosomes has a range of cellular consequences, including cryptic transcription [7], [8], which can result from the unmasking of sequences resembling promoters, or the alteration of histone marks that affect the function of accessory factors responsible for degradation of these transcripts [9]. Identifying both the cis and trans acting determinants of nucleosome occupancy and positioning is therefore key to more fully understand transcriptional regulation.
Genome-wide nucleosome profiling studies have revealed a high degree of organization around genes [10]–[13], with several conserved features: a nucleosome depleted region (NDR) immediately upstream of the transcription start site (TSS), followed by a regularly spaced array of nucleosomes across the gene body which then gradually dissipates towards the end of the gene [14] and, for many genes, ends with an NDR in the 3′ untranslated region. The average spacing of nucleosomes in S. cerevisiae is 165 bp, with a linker region of ∼18 bp separating adjacent nucleosomes [11], [14]. The location of the 5′ NDR coincides with the promoter region and is enriched for TF binding sites [11]. Its formation in yeast appears to be driven mainly by poly(dA:dT) tracts that are structurally rigid and refractory to nucleosome assembly [15]–[17], as well as by a small set of nucleosome-excluding transcription factors (TFs) such as Rsc3, Rap1, Abf1, and Reb1, which direct NDR formation at hundreds of promoters containing their binding motifs [18], [19]. Other factors such as the Tup1 and Cyc8 co-repressors have the opposite effect and induce the formation of closed chromatin at the promoters of the genes they repress [20]–[23]. Particular attention has been focused on the attributes of the +1 nucleosome which lies immediately downstream the 5′ NDR and which is thought to have a regulatory function by controlling transcription initiation [24], [25]. The +1 nucleosome has a well-defined position relative to the TSS [12], [14] and even small lateral movements of as few as 2–3 bp can (un)mask regulatory sites near the +1 nucleosome boundary [24].
In contrast to promoter NDRs, the determinants of genic nucleosomal organization are less well understood. In vitro nucleosome reconstitution studies with purified histones and DNA indicate that as much as half of all nucleosome positions may be determined by intrinsic nucleosome-DNA sequence preferences [26], [27], however, these experiments do not reproduce the typical nucleosome periodicity relative to the TSS observed in vivo [28]–[30]. One biophysical model to explain nucleosome positioning that does not rely on underlying sequence features is the statistical positioning model [31], which predicts that nucleosomes will form regularly spaced arrays relative to a genomic barrier (such as the NDR) due solely to steric hindrance between neighboring nucleosomes. Nucleosome organization in vivo shows several features consistent with statistical positioning [10], [14], [32], but in vitro profiles obtained using varying histone to DNA ratios do not [28], [33], indicating that other factors are required to explain genic nucleosome architecture in vivo. Indeed, a recent in vitro reconstitution study demonstrated that an ATP-dependent mechanism, presumably via the action of chromatin remodeling enzymes, is required to produce periodic nucleosome patterns. Based on these observations, an alternative model was proposed in which nucleosomes are actively stacked against the NDR barrier at the 5′ ends of genes [33]. These data strongly implicate protein factors in the organization of genic nucleosomes.
Several studies have highlighted the importance of remodeler ATPases in nucleosome organization around genes [13], [33]–[36]. Combined disruption of the remodeler ATPases Chd1 and Isw1 results in a total loss of patterning [37], suggesting that their coordinated action is critical for establishing in vivo chromatin architecture. Additional factors contribute to the regulation of genic nucleosomes, e.g. histone chaperones such as Spt16 and Spt6 are thought to affect genic nucleosome distribution by virtue of their role in histone turnover and nucleosome reassembly during transcription [7]. Loss of Spt6 results in decreased levels of genic nucleosomes and a loss of genic nucleosome organization [7], [38]. The FACT component Spt16 shares many of the phenotypic effects of Spt6 [7], [39], including widespread antisense transcription defects in genes [7], [8]. Changes in nucleosome spacing have also been observed in RNA polymerase II mutants [40], suggesting that transcription also promotes nucleosome organization [28], [40].
Here, we examined a compendium of 55 mutations and conditions in S. cerevisiae for their effects on nucleosome occupancy, positioning and transcription. Loss of the remodeler ATPases Chd1, Ino80 and Isw1, the CAF-1 complex (Rlf2, Cac2, Msi1) or Spt16 leads to significant displacement of 5,886 nucleosomes on 3,616 genes such that they assume positions that are more consistent with their intrinsic DNA-binding preferences. Most rearrangements of individual genic nucleosomes are within the linker regions, with little effect on neighboring nucleosomes, indicating that there is considerable positional flexibility within genic nucleosome arrays. Changes were most apparent at distal genic nucleosomes relative to the TSS; however, we also frequently observed repositioning of the +1 nucleosome, for example, upon histone depletion and in strains lacking Ino80 and Isw1. In the case of Ino80, selected genes with +1 nucleosome shifts involved in iron and glucose homeostasis showed altered gene expression responses in response to environmental stimuli. Changes in nucleosome occupancy and positioning were coupled to genome-wide transcription changes, giving rise to antisense transcripts in CAF-1 complex mutants, while loss of Reb1, Abf1, Tbf1, and Rsc3 function resulted in cryptic transcripts in the promoter regions of their target genes. Our findings demonstrate the utility of our compendium as a valuable resource for future studies.
We examined 50 single-gene loss-of-function strains, comprised of gene deletions and temperature-sensitive (ts) or tetracycline promoter-shutoff (tet) alleles (Table S1). These genes were selected based on their known or potential role in nucleosome biology and included remodeler ATPases and chaperones, histones and histone modifiers, transcription and elongation factors, and components of RNA polymerase I and II (Figure 1A). The compendium also included 4 compounds targeting transcription and histone deacetylases, as well as a histone depletion time course performed with a strain in which H4 gene expression is exclusively under the control of a GAL1 promoter [41]–[44]. Glucose-induced repression leads to rapid nucleosome depletion in this strain. Genome-wide nucleosome occupancy profiles were generated using Affymetrix Tiling arrays with probes spaced every 4 bp [45], or next-generation sequencing. Identically prepared total RNA samples for each strain and treatment were analyzed on the same platform for strand-specific expression differences. Each compendium condition was compared to a matched wild-type (WT) reference grown in parallel (31 WT samples in total). To facilitate downstream analyses, we also prepared a manually curated set of transcript starts and ends for 5,043 yeast genes by using publicly available sequencing and tiling array data, as well as our compendium data. The nucleosome and expression profiles and gene annotations are available as a resource through the Nucleosome Compendium Browser at http://nbrowse.ccbr.utoronto.ca/mgb2/gbrowse/nucleosome/ and have been deposited in GEO.
Overall we find that most compendium mutants/conditions maintain a canonical nucleosome occupancy pattern with a depleted region (NDR) directly upstream the TSS and a regularly spaced nucleosome array across the gene when considered in aggregate (Figure 1B, Figure S1). The lack of dramatic response in most mutants is not necessarily surprising, based on a recent study which demonstrated that many chromatin mutants manifest their effects under conditions of stress [46], and may be explained in part due to inherent redundancy of factors involved in chromatin homeostasis. Each profile did, however, contain informative deviations from the WT reference. The greatest changes in nucleosome occupancy are seen for genes in the TF and nucleosome remodeler/chaperone categories. In the former category, the changes are predominantly localized to NDRs (Figure S2), which we previously showed is linked to the presence/absence of TF binding sites [18]. Changes in NDR occupancy in TF mutants are also generally correlated with expression changes of the genes with which these NDRs are associated (Figure S2). Nucleosome occupancy changes seen in mutants of nucleosome remodelers and histone chaperones occur more broadly throughout the genome, including within gene bodies and NDRs. Interestingly, the loss of histone modifiers resulted in only modest changes in nucleosome occupancy. Given that the compendium encompasses a broad range of modifiers, including those involved in histone (de)acetylation, methylation, phosphorylation, proline isomeration and ubiquitination, this observation suggests that any single histone mark plays a relatively minor role in regulating nucleosome occupancy, and by extension, genic nucleosome organization. We found 3 compendium conditions with severely disrupted nucleosome organization around genes (Figure 1B). Loss of the elongation factors Spt6 and Spt16 resulted in an almost complete loss of genic patterning, consistent with previous studies [7], [38]. We also found a progressive reduction of nucleosome patterning across the gene body within 3 to 6 hours after the shut-off of histone H4 transcription. The apparent increase in NDR occupancy after 6 hours is likely due to a normalization effect that makes NDR regions appear less pronounced as the result of a large global decrease in histone levels (see below). The fact that the global nucleosome organization around genes is maintained across most compendium conditions is consistent with cells employing redundant mechanisms that are robust against disruptions of individual chromatin modifiers and further underscores that maintenance of nucleosome occupancy is critical for cell fitness.
In addition to the impact of gain or loss of nucleosomes, DNA-associated processes are also affected by repositioning of assembled nucleosomes, for example through modulation of regulatory site accessibility [47]. We therefore expanded our analysis to assess nucleosome shifts – defined here as a change in their average center position – relative to the TSS. To this end, we applied a modified Gaussian filter to our microarray and sequencing data to determine individual nucleosome positions [48], and calculated the degree of shift for each nucleosome in each compendium condition by subtracting its estimated center position from that of the nearest nucleosome in the corresponding WT reference, located within a 100 bp window (see Materials and Methods).
Our nucleosome position analyses revealed striking patterns of genic nucleosome shifts relative to the TSS in 20 tested conditions, with median deviations ≥4 bp for at least one genic nucleosome position that were not seen in the WT reference profiles and other compendium conditions (see Figure S3 for comparison). We grouped these conditions into 7 categories based on shared biological function (remodeler ATPases, regulators of histone gene expression, the CAF-1 complex, elongation/chaperone) or experimental condition (histone depletion and transcription disruption) (Figure 2A–2F). Conditions that fell outside these broad classes were collapsed into a single category (Figure 2G) and included conditional loss-of-function mutants of the E3 ubiquitin ligase Bre1, as well as the essential transcription factors Spt15 (TBP), Abf1, Mcm1, and Gcr1. For the subset of essential genes, the observed effects may represent the combined effects of gene inactivation on chromatin and additional indirect effects. For conditions with published nucleosome occupancy maps, our nucleosome shift data is consistent with reported changes in genic nucleosome profiles. For example, a global reduction in transcription, either through loss of the RNAPII subunits Rpb2 or Rpo21, or by treatment with 6-Azauracil, which limits transcription elongation rates by reducing intracellular GTP levels [49], resulted in genic nucleosome movements away from the TSS (Figure 2F), consistent with the increased nucleosome spacing reported in Pol II mutants [40]. Likewise, we confirm genic nucleosome shifts in strains deleted for the remodeler ATPases Ino80, Isw1 and Chd1 (Figure 2A) [13], [34], [36].
The patterns of nucleosome shifts differed markedly between conditions, with the predominant effects being on either proximal or distal nucleosomes relative to the TSS, as well as a median shift towards or away from the TSS (e.g. compare Figure 2A and 2B). The progressive increase in the magnitude of shifts for nucleosomes further away from the TSS observed in many conditions is consistent with the additive effects of changes at more proximal positions. Most shifts ranged from 1 to 20 bp, with median deviations at each nucleosome position between 1 and 12 bp, placing them within the confines of inter-nucleosome linker regions, which average ∼18 bp in S. cerevisiae [11], [14, this study]. The range of lateral mobility suggests that although in vivo movements of genic nucleosomes in yeast are sterically limited by their neighbors, the linker regions do allow for positional flexibility. Notably, several conditions also exhibited hundreds of nucleosome shifts that exceeded the average size of the linker region, discussed below. We did not observe nucleosome shifts in a strain deleted for the non-essential the linker histone Hho1 (data not shown), indicating that this histone does not play a major role in setting inter-nucleosomal linker distance. This is consistent with the fact that Hho1 is present at much lower levels compared to core histones [50].
Taken together, our data suggest that remodeler ATPases, cellular histone levels, histone chaperones and transcription all exert distinct and sometimes opposite net effects on nucleosome positioning and inter-nucleosome spacing, and reveal considerable positional flexibility of genic nucleosomes. We next examined individual classes of rearrangements in more detail.
Deletion of the remodeler ATPases Chd1, Isw1 and Ino80 generally moved genic nucleosomes closer to the TSS (Figure 2A), consistent with previous studies of these remodelers [34], [36], [37]. To better characterize the changes, we identified nucleosomes with a significant (T-test p<0.05) shift of at least 10 bp in two biological replicates compared to their position in the 31 wild-type strains that were independently grown and analyzed during the course of the study. In light of the emphasis in recent studies on the range of factors that contribute to the positioning of the +1 nucleosome, and the relative dearth of study on the +2, +3, and +4 nucleosomes [33], [36], we specifically examined changes at these positions. In each condition we identified hundreds of individual nucleosome rearrangements (Figure 3A), including many that exceeded the average inter-nucleosomal distance. At every position we find both positive and negative shifts relative to the TSS. Overall, we identified 3,147 genic nucleosomes with shifts between the +1 and +4 position in the three remodeler ATPases, affecting 2,379 genes. Strikingly, most of these changes are remodeler-specific, with only 2% of the rearranged nucleosomes found in more than one condition.
One of the most prominent observations from our analysis of individual nucleosome changes is the distinctive behavior of the +1 nucleosome in the ino80-Δ and isw1-Δ strains. While distal nucleosomes move predominantly towards the TSS, most +1 nucleosomes show a strong movement away from the TSS (Figure 2, Figure 3A), confirming other recent observations of these remodelers [36]. This displacement is particularly interesting as the +1 nucleosome has been considered the most well-positioned genic nucleosome, and whose position is primarily determined by the presence of fixed barrier elements such as poly(dA:dT) tracts [10]. We also observed +1 nucleosome shifts in several other conditions (Figure 2), however, the opposing effects on the proximal vs. distal nucleosomes were unique to ino80-Δ and isw1-Δ and prompted us to examine these mutants in greater detail. A hierarchical clustering of the degree of +1 nucleosome shifts (Figure 3B) shows that loss of Ino80 and Isw1 affects distinct sets of +1 nucleosomes and can sometimes result in opposite effects on the same nucleosome. Thus, there are key differences in both the magnitude and direction of +1 nucleosome shifts in Ino80 and Isw1 mutants; this is despite the observation that these factors have a high degree of co-occupancy at the 5′ ends of genes [36]. In contrast to Ino80 and Isw1, the Chd1 deletion mutant is characterized by an invariant +1 nucleosome position, with effects on distal nucleosomes limited almost exclusively to movements towards the TSS.
Given the proximity of the +1 nucleosome to the transcription start site and its potential role in regulating transcription, we assessed the effects of +1 nucleosome shifts in the Ino80 or Isw1 deletion mutants on gene expression. In standard growth conditions, genes with significant changes in +1 nucleosome position, either positive or negative, showed only minor changes in gene expression compared to genes with stable +1 nucleosomes (Figure 3C). We also did not find any correlation between the degree or direction of nucleosome repositioning, and transcription changes (data not shown). We attribute this to the fact that the effects of nucleosome repositioning may be activating or inhibitory, depending on local sequence context, e.g. on whether binding motifs of repressors or activators are (un)masked. To provide a more detailed assessment of the effects of +1 nucleosome position changes on individual genes, we asked if any of the affected genes had an altered environmental expression response. Among the 418 genes with shifted +1 nucleosomes in ino80-Δ strains, we identified 2 genes, FRE1 (−25±15.0 bp shift) and FRE7 (−15±4.2 bp shift), that are known to be induced in iron-limiting conditions [51], [52], as well as 2 glucose-responsive transcription factors, MIG1 (+26±8.5 bp shift) and RGT1 (+19±2.8 bp shift) [53], [54]. None of these genes showed significant shifts (>10 bp) at more distal genic nucleosome positions. Consistent with our hypothesis of a regulatory function for the +1 nucleosome, we find a complete loss of induction of FRE1 and FRE7 in an ino80-Δ background compared to WT strains, following treatment with the iron chelator and transcription inhibitor 1,10-Phenantroline (Figure 3D).Similarly, the response of MIG1 and RGT1 to changes in glucose levels is altered, with a marked loss of RGT1 repression (vs. WT) upon a shift from YPD to YPG media (Figure 3E). These observations suggest that regulation of the position of the +1 nucleosome by Ino80 can affect gene expression at these loci and that these effects may have biological consequences, though this hypothesis remains to be further tested to rule out potential indirect effects of INO80 deletion.
Several scenarios could account for the nucleosome rearrangements seen upon loss of remodeler ATPases. First, there could be specific associations between remodelers and individual genes or nucleosomes. To test this, we analyzed publicly available data [36], which indicated that the affected nucleosome positions are indeed bound by remodelers, but did not reveal any difference in Ino80 or Isw1 levels at positions with shifted nucleosomes compared to other nucleosome positions (Figure S4), suggesting that they are not preferentially targeted. Secondly, we considered that the rearrangements result from changes in the number of nucleosomes following the deletion or depletion of remodeling ATPases, by assessing changes in the levels of histone H3 as a proxy for changes in global nucleosome levels. Histone H3 levels were unchanged in chd1-Δ, but decreased by 12% and 17% in ino80-Δ and isw1-Δ strains, respectively (Figure S5). This suggests that the shifts of +1 nucleosomes away from the TSS in ino80-Δ and isw1-Δ strains, but not chd1-Δ strains, may be linked to changes in global nucleosome levels, though these changes cannot fully explain the shifts at more distal nucleosome positions we observed in all three strains.
Finally, we considered the relationship between the nucleosome locations (relative to the TSS) affected by remodelers, and those locations on the DNA that are intrinsically preferred by nucleosomes, by comparing the in vivo profiles to in vitro nucleosome reconstitution data from Kaplan et al. [27]. Strikingly, upon loss of remodeler ATPase activity, there is widespread repositioning of nucleosomes towards more preferred DNA sequences (Figure 4A), which suggests that Chd1, Ino80 and Isw1 may act to disrupt these interactions to favor their organization into genic arrays. Given the potential for steric effects on neighboring nucleosomes we also examined how nucleosome shifts at each of the +1–4 positions affected other nucleosomes in the same genic array. Interestingly, while significant shifts at each genic nucleosome position were coupled to shifts of directly neighboring nucleosomes, these changes did not propagate to more distal positions in the same nucleosome array (Figure 4B). Indeed, most genes with a significant positive shift at proximal positions still show a negative shift at more distal genic nucleosomes. This further demonstrates the positional flexibility of genic nucleosomes and suggests that remodeler ATPases can exert opposite directional forces on proximal and distal genic nucleosomes.
Using a promoter shutoff strategy [41]–[44], we found that histone H4 depletion (3–6 hours) led to a progressive shift of the +1, +2, +3 and +4 nucleosomes away from the TSS, with the most pronounced effect at the +1 position (Figure 2B). More distally, there was no net change, however, the greatly increased positional variance, (Figure 2B), suggested large shifts of individual nucleosome in both directions. Indeed, when using the criteria described above, we find equal numbers of significant nucleosome shifts ≥10 bp in both directions at the +3 and +4 positions (Figure S6). Within 3 to 6 hours after shutoff of H4 transcription, there is a 15% and 76% reduction in global histone levels (Figure S5), respectively, and a decrease of 18% and 27% in assigned nucleosome positions in our tiling array data at the Gaussian score thresholds used. As we observed for the remodeler ATPases, nucleosome loss led to rearrangement of nucleosomes to positions that are more consistent with their intrinsic DNA-sequence preferences (Figure S6), suggesting that steric hindrance by neighboring nucleosomes counteracts nucleosome sequence preferences, which is consistent with findings based on histone H3 depletion [55].
We also see a reduction in global histone levels upon deletion of the histone gene regulators Spt10 (19%) and Spt21 (19%) (Figure S5), in line with the reduction of histone gene expression reported in these conditions [56]. The spt10-Δ strain in our compendium showed reduced (>3.5-fold) expression at the HTA2-HTB2 locus, encoding histones H2A/H2B, as well for as the redundant HHF1-HHT1 and HHF2-HHT2 loci (>2-fold) encoding histones H3/H4. In the spt21-Δ strain, the HTA2-HTB2 (>2-fold) and HHF2-HHT2 loci (>1.4-fold) were affected. Surprisingly, loss of the histone gene regulators Spt10 and Spt21 led to some of largest rearrangements of genic nucleosomes in the compendium (Figure 2E) despite only a modest loss of nucleosomes (Figure S5). In the case of spt10-Δ, these changes are consistent with previous reports of global disruption of chromatin structure in this mutant [57]. There are other marked differences in the nucleosome shift profiles: while prolonged depletion of histone H4 alone results in +1 nucleosome shifts, loss of Spt10 and Spt21 predominantly affects distal nucleosomes. Given that Spt10 and Spt21 regulate the levels of histones H2A, H2B and H3, in addition to histone H4, this may indicate that cells respond differently to the concerted depletion of all four histone types. Alternatively, both Spt10 and Spt21 have recently been described to affect silencing at telomeres in a manner independent of changes in histone levels [58]; our findings are therefore also compatible with potential roles for Spt10 and Spt21 in regulating global chromatin structure that go beyond the regulation of histone gene expression.
Our observations of nucleosome shifts upon histone depletion differ from those of a previous study that reported no redistribution of nucleosomes along DNA in yeast nhp6 mutants bearing deletions of NHP6A and NHP6B, despite a 20–30% reduction in histone levels [59]. Furthermore, a recent in vitro nucleosome reconstitution study in which whole cell extracts with additional ATP were used to reproduce in vivo patterning also did not find global changes in the spacing of +1–4 nucleosomes upon reduction of the histone∶DNA ratio by 50% [33]. The discrepancies between our findings and these studies may partially be accounted for by differences in the degree of nucleosome depletion between studies and in experimental setup (i.e. in vitro reconstitution vs. in vivo depletion). In addition, these studies focused primarily on large-scale rearrangements and may have missed smaller scale effects. Indeed, our reanalysis of the data for both studies shows similar shifts of the +1–4 nucleosomes upon reduction of histone levels (Figure S7), in particular for the reconstitution study [33], suggesting that with increased resolution the apparent discrepancy disappears, consistent with a recent histone H3 depletion study [55]. Taken together, these data show that there is greater positional flexibility of genic nucleosomes in response to changes in histone levels than previously assumed, with potential consequences for gene regulation.
There is an increasing shift of distal genic nucleosomes away from the TSS upon loss of the CAF-1 subunits Msi1, Rlf2 and Cac2 (Figure 2C). The CAF-1 complex is involved in nucleosome assembly [60], [61] and has also been suggested to play a role in transcription [62], [63]. The nucleosome shifts in the CAF-1 mutants may be partially due to a loss of genic nucleosomes, as there is a decrease in global histone levels in the msi1-Δ (9%) and cac2-Δ (19%) strains (Figure S5). Nevertheless, the CAF-1 profiles are distinct from those obtained after prolonged H4 depletion, with an increased shift of distal nucleosomes in the former, compared to proximal nucleosomes in the latter. In addition, the differences in the CAF-1 profiles compared to Spt6 and Spt16 (Figure 2D), suggest that the CAF-1 complex plays a distinct role in organizing genic nucleosomes. As in other compendium conditions, nucleosome position changes in CAF-1 complex mutants appear to be partially driven by their intrinsic DNA-binding specificity (Figure S8), thus the CAF-1 complex may be added to the roster of factors that oppose intrinsic nucleosome positioning in vivo.
Previous studies of Spt6 and Spt16 mutants revealed that disruption of genic nucleosome arrays results in the expression of cryptic antisense transcripts (i.e. transcripts complementary to mRNAs) [7], [8]. We therefore sought to determine whether the loss of CAF-1 complex components results in similar transcription defects by examining all compendium conditions for a significant change in expression levels (≥2-fold; p≤0.05) for RNAs of at least 80 nt in length (Table S2) on the antisense strand of annotated genes. Hierarchical clustering of expression changes in all regions associated with antisense transcripts show, as expected, a large increase in antisense transcripts in the Spt6 and Spt16 mutants and upon histone depletion (Figure 5). There is also an increase in antisense transcripts in the spt10-Δ and spt21-Δ mutants, consistent with the down-regulation of histone gene expression and changes in nucleosome spacing we observe under these conditions. Loss of the CAF-1 complex components Rlf2, Cac2 or Msi1 leads to widespread cryptic antisense transcripts, albeit to a lesser extent. The individual deletions of these components each resulted in similar antisense transcription profiles that cluster with the spt10-Δ and spt21-Δ strains. These data confirm that the CAF-1 complex plays a role in genic nucleosome positioning and functions to prevent cryptic antisense transcription, in a manner distinct from Spt6 and Spt16.
We identified 9,471 distinct regions with significant transcription changes in at least one compendium condition that did not overlap annotated strands of known features, totaling 7.0 Mb (29%) of the two strands of the genome sequence (24 Mb). The bulk of these regions were antisense to known genes (Figure 5, Table S2); however, we also observed many additional transcription changes in intergenic regions (Table S2). The breadth of these changes greatly expands on previous studies and provides a rich repository for subsequent studies. Here, we focused on transcripts originating in genomic regions that manifested changes in nucleosome positions and/or levels and found a strong link between changes in NDR nucleosome occupancy and the appearance of transcripts at the 5′ ends of genes in Tbf1, Abf1, Rsc3 and Rap1 mutants (Figure 6A), consistent with their roles in nucleosome exclusion at promoters [18], [64], [65] (Figure 6B). Other conditions in the compendium that showed strong changes in either NDR nucleosome occupancy and/or gene expression, e.g. histone depletion (Figure S9), did not show this effect, indicating that the appearance of these transcripts is a direct effect of the loss of the TFs. Given their location, we designated these transcripts as promoter associated transcripts (PATs). A complete list of all PATs and the genes they are associated with is provided in Table S3.
PATs share characteristics with cryptic unstable transcripts (CUTs) detected after disruption of the exosome complex [66], [67], such as their occurrence at gene termini. Moreover, many CUTs result from bidirectional transcription [66], [67] and similarly, we see that PATs are enriched at divergently transcribed genes (Figure 6B, Figure 6C), and originate from the NDRs of the upstream genes (Figure 6B, 6D). This suggests that most PATs in the TF mutants either result from an increase in divergent transcription from upstream promoters, or a failure to degrade these transcripts. Taken together, we conclude that the proper architecture of nucleosomes on the affected regions, established by the actions of Tbf1, Abf1, Rsc3 or Rap1, is important for preventing expression of cryptic promoter associated transcripts.
We identified the Tup1 and Cyc8 single deletion mutants as having a particularly strong relationship between changes in 5′ NDR occupancy and the expression of their target genes; almost all genes that showed an increased NDR depletion had increased expression, and vice versa (Figure S1). This prompted us to scan the genome for new transcripts with the same characteristics, to identify novel Tup1/Cyc8 regulated genes. Requiring a significant change in expression (≥2 fold; p≤0.05) across at least 80 nt, coupled with a 1.5-fold decrease in nucleosome occupancy 200 bp upstream, we identified 36 regions with significant changes in expression levels and the formation of a 5′ NDR when Tup1 and/or Cyc8 were deleted (Figure 7A, Table S4). We have designated these Tup1/Cyc8 repressed (TCR) transcripts. An example of two divergent loci identified in the subtelomeric region of chromosome 1 is shown in Figure 7B. Further examination of the 36 TCR loci in all compendium conditions confirmed that these transcripts are specifically regulated by Tup1 and Cyc8 (Figure 7C).
Among the 36 new transcripts we identified, 14 originate in subtelomeric regions, 13 are antisense to protein-coding genes and the remaining 9 are intergenic (Table S4). Seven of these transcripts overlap previously identified disabled ORFs (TCR1, TCR21, TCR5, TCR4, TCR20, TCR15 and TCR29) [68]. The subtelomeric TCR1, TCR4-6 and TCR20-21 bear striking similarity to flocculation genes, however, the presence of multiple stop codons and indels suggests that they no longer encode functional proteins. Translated blast analysis (blastx) identified TCR27 as potentially encoding a salt tolerance protein (COS3 hit, E-value 3e-30) and distant hits for TCR7, TCR8, TCR22, and TCR30, indicating that these five TCRs may be protein-coding (pseudo)genes. Finally, several of the intergenic transcripts we identified are divergently oriented relative to promoters of neighboring protein coding genes targeted by Tup1/Cyc8 (Table S4) and could therefore be the result of bidirectional transcription.
To determine if the novel Tup1/Cyc8 regulated transcripts are functional, we deleted 10 subtelomeric TCR genes and four others found in intergenic regions in a ura8-Δ (the “WT” reference) or tup1-Δ background, subjected them to a panel of 14 different stress conditions, and assessed changes in growth by serial dilution assays. The results of these assays are summarized in Figure 7D. As expected from their repression in WT conditions, none of the deletions in the ura8-Δ background showed a growth phenotype in any of the stress conditions. In contrast, several deletions exhibited growth defects in the tup1-Δ background (tup1-Δ/tcr1-Δ was 2.6, 2.8 and 6.1 times more sensitive in YPD, 0.1 mM H2O2 and 3% YPEtOH, respectively) or increased resistance (tup1-Δ/tcr11-Δ, tup1-Δ/tcr31-Δ and tup1-Δ/tcr30-Δ were 3.2, 3.3 and 2 times more resistant in 5 mg/ml 6-Azauracil, respectively) (Figure 7D), indicating that several of the loci we identified are functional.
In this study we find, by virtue of mutational inactivation and biochemical perturbation, multiple factors that act in concert to disrupt intrinsic nucleosome binding preferences and maintain in vivo positioning, and often exert opposing net forces on positioning. We report new roles in genic nucleosome positioning for Spt10, Spt21 and the CAF-1 complex, increasing the number of chromatin modifiers involved in nucleosome organization around genes. While the factors examined in this study predominantly affect different subsets of nucleosomes and genes, their loss generally leads to displacement of nucleosomes towards more intrinsically preferred sequences. Together these observations underscore the multiple ways in which cells actively counteract intrinsic nucleosome binding preferences in genic regions to achieve chromatin “homeostasis”. This holds true even for the +1 nucleosome, whose placement at the 5′ end of genes is considered extremely stable. Many of these perturbations of nucleosomes led to widespread transcription defects, which we captured by performing parallel genome-wide transcript analysis.
Of all the models that aim to capture the principles of nucleosome organization, the barrier-packing model [33] is the most comprehensive in that it accounts for most of the data published thus far. This model posits that nucleosomes are actively stacked against the TSS and was motivated by the observation that in vitro reconstitution of chromatin with reduced histone levels mainly affect the position of distal genic nucleosomes, with little effect on spacing and positioning of proximal nucleosomes [33], [59]. Our data, in contrast, reveal significant reorganization of proximal nucleosomes after depletion of histone H4 in vivo, including a pronounced shift of +1 nucleosome away from the TSS. Although this observation is not necessarily incompatible with a packing model (e.g. it could reflect a reduction in packing efficiency due to nucleosome loss), it does indicate that there are additional forces acting on proximal nucleosomes which oppose TSS stacking. Such opposing forces are also apparent in the variability in nucleosome shifts among many of the compendium mutants, with some increasing and others decreasing packing against the TSS. The overall movement of distal nucleosomes in Chd1, Ino80 or Isw1 loss-of-function mutants makes it unlikely that these remodelers are responsible for packing distal nucleosomes against the TSS, although roles for other remodeler ATPases such as Isw2 [36] are not excluded. We do find an expansion of nucleosome arrays upon loss of the CAF-1 complex, as well as Spt10 and Spt21, indicating that chaperones may contribute to packing against the TSS. Accordingly, TSS packing of nucleosomes appears to be the net outcome of multiple opposing forces, rather than the actions of any single class of factors.
The TSS packing model further predicts that shifts of proximal nucleosome should propagate throughout the array of genic nucleosomes. Although we find that local shifts can affect the positions of adjacent neighboring nucleosomes, these effects rarely spread to more distal nucleosomes. Indeed, many of the genes with a strong displacement of the +1 nucleosome away from the TSS in Ino80 and Isw1 still show an overall movement towards the TSS for more distal nucleosomes. A likely explanation for this effect is that nucleosome shifts can be buffered by changes in linker region length, which allows for a degree of local flexibility in genic arrays and decoupling of proximal and distal genic nucleosome positioning.
Within arrays, individual nucleosome shifts appear to be mainly driven by underlying sequence, with nucleosomes moving to positions that are more consistent with their intrinsic DNA-sequence preferences upon loss of Ino80, Isw1, Chd1, the CAF-1 complex, histone H4, Spt6 and Spt16 (data not shown). This extends observations previously made for Isw2 and reductions in histone H3 levels [55], [69], and suggests that there are many factors involved in disrupting intrinsic DNA sequence preferences. The movements of +1 nucleosomes away from the TSS we observed in many conditions are likely driven by the strong nucleosome-excluding properties of the NDR and confirm that many of these nucleosomes are not in their optimal intrinsic positions [28], [33], [69]. Taken together, our observations indicate that nucleosome packing is not unidirectional and strongly suggests that genic nucleosomes can and do position themselves independently.
In the perturbations interrogated here, the loss and/or repositioning of nucleosomes appears to drive the appearance of the cryptic transcripts in the parallel transcriptome maps. For example, the increased genic nucleosome spacing in strains bearing deletions of Spt10, Spt21 and components of the CAF-1 complex is accompanied by an increase in antisense transcripts. The patterns of antisense transcripts are similar to those observed in spt6 and spt16 mutants and upon histone H4 depletion, albeit to a lesser degree. The CAF-1 effects on genes are likely direct – the Msi1, Rlf2 and Cac2 subunits have been shown to be recruited to PMA1 in a transcription-dependent manner, with patterns of association with the gene body resembling those of Spt6 and Spt16 [62]. Loss of the nucleosome-excluding transcription factors Tbf1, Rap1, Abf1 and Rsc3 results in cryptic transcripts at their target promoters, most of which appear to be the result of diverging transcription from upstream promoters. These findings are consistent with the transcriptional interference previously reported upon loss of an Abf1 binding site at the ARO4/HIS7 locus [70]. Interestingly, all four of these general regulatory/transcription factors have been shown to act as strong insulators that enable neighboring chromatin domains to be regulated independently [65], [71], and further, promoters containing combinations of these binding sites are proposed to act as insulators [65]. Our results suggest that the actions of nucleosome-excluding TFs can indeed establish a boundary between adjacent promoters that prevents diverging transcription.
The discovery of novel transcripts regulated by Tup1/Cyc8 shows that new genes and/or pseudogenes can still be found, even in the well-studied budding yeast genome and transcriptome. Several of these transcripts in subtelomeric regions bear strong similarity to flocculation genes, but the presence of multiple stop codons suggests that they no longer encode for functional proteins. A blast analysis of related yeast strains shows that, in contrast to the S. cerevisiae S288c reference strain (an acknowledged evolutionary outlier [72]), many S. paradoxus and S. cerevisiae isolates bear functional copies of these genes (data not shown). Combined signatures of chromatin modifications and changes in transcript levels have been successful in identifying new functional RNAs and our results indicate that the same principles can be used to identify new functional transcripts in yeast.
Our study has focused on a core set of chromatin modifiers, transcription factors and histones, yet even within this small set we identified several new factors that impact genic nucleosome architecture and transcription. By extrapolation, it is therefore likely that many other factors that play a role in nucleosome positioning remain to be identified. Crucially, there is still a lack of understanding of the interplay between nucleosome modifiers and chaperones and how they coordinately regulate and control nucleosome architecture at genes. For example, reported differences in nucleosome positioning and spacing between species, with their attendant effects on evolution [73], may well be a reflection of differences in the balance between these factors. The data provided here should prove very useful to begin classifying the contributions of these various factors.
Microarrays were designed in collaboration with Lars Steinmetz and Ron Davis at the Stanford Genome Technology Center [45] and Affymetrix (PN 520055) and contain one set of 25-mer probes spaced every 8 bp covering the Watson strand and a second set of probes, offset 4 bp from the first set, covering the Crick strand of the Saccharomyces cerevisiae S288c genome. The design allows for 8- or 4-bp resolution hybridizations of single- or double-stranded samples, respectively.
The strains used in this study are listed in Table S1. For deletion strains, 5 mL cultures were grown overnight at 30°C in YPD. Cells were diluted to OD600 0.2/mL in 400 mL of YPD media the next morning, and grown to mid-log phase (OD600 0.8–1.0/ml) in 1 L flasks at 30°C while shaking, at which point RNA and nucleosomal DNA were isolated. All OD measurements were made using an Eppendorf BioPhotometer (SN#6131). Temperature-sensitive strains were grown at 22°C (permissive temperature) until mid-log phase. An equal volume of hot medium was then added to rapidly equilibrate the culture to 37°C (restrictive temperature), followed by a further 3 to 7 hours incubation at this temperature until a difference in OD between the mutant strain and its corresponding wild-type control became apparent. For Tet promoter-shutoff strains [74], doxycyline was added at 10 µg/ml for 24 h in order to obtain down-regulation for the gene of interest. Cells were then diluted to an OD600 of 0.2 and grown in the presence of doxycycline (10 µg/ml) until mid-log phase. These conditions were previously shown to lead to effective ablation of proteins encoded by essential genes [74].
For the histone depletion time course, the UKY403 strain harboring a deletion of both Histone H4 genes (HHF1, HHF2), a plasmid with the GAL1 promoter driving expression of Histone H4 (HHF2) and a plasmid with the GAL1 promoter driving expression of Histone H4 (HHF2), was grown in YP with 2% galactose until mid log phase [41]–[44]. Cells were then collected by centrifugation and transferred to YPD for 0, ¼,1, 3, 5, and 6 hrs. After each time point, samples were taken for nucleosome and expression analysis.
The haploid yeast strain BY4741 (parental strain of the haploid yeast deletion collection) was used for drug treatments. Cells were grown overnight and then diluted to an OD600 of 0.4. The drugs Nicotinamide (82 mM, IC50), Sodium Butyrate (20 mM, IC50), 1,10-Phenanthroline (100 µM, IC90) or 6-Azauracil (6 mM, IC90), all obtained from Sigma-Aldrich, were added to the media for 2 hours. Drug concentrations were predetermined to inhibit growth by 50% and/or 90%.
Nucleosomal DNA was prepared according to Lee et al. [11] with a modified fragment size selection step. Briefly, cells were crosslinked by direct addition of methanol-free formaldehyde (Polysciences) to a final concentration of 2% for 30 min while shaking at 30°C. The reaction was quenched by adding glycine to a final concentration of 125 mM for 5 min. Cells were pelleted, washed with 20 mL phosphate buffered saline solution once, and resuspended in 6 mL of [1 M sorbitol, 50 mM Tris 7.4] with freshly added 10 mM β-mercaptoethanol in a 15 mL conical tube. Zymolyase (20T, TakaRa Biotechnology Co., Ltd, Japan) was added to a final concentration of 0.25 mg/mL and cells were spheroplasted at 30°C while gently rolling for 30 min. After zymolyase treatment, cells were pelleted and resuspended in 4 mL of [1 M sorbitol, 50 mM NaCl, 10 mM Tris 7.4, 5 mM MgCl2, 1 mM CaCl2, 0.075% NP-40] with freshly added 1 mM β-mercaptoethanol and 500 µM spermidine. Spheroplasts were divided into 6 aliquots of 300 µL each and transferred into 1.5 mL Eppendorf tubes. Micrococcal nuclease (MNase; Worthington) dissolved in water at 0.1 U/µL stock was added to the tubes at concentrations of 0, 25, 50, 75, 100, and 150 U per sample. The digestion reactions were incubated at 37°C for 45 min and reactions were stopped by adding 75 µL of [5% SDS, 50 mM EDTA]. Remaining proteins were digested by adding 3 uL of 20 mg/mL proteinase K solution (Qiagen) to each tube for an overnight incubation at 65°C.
DNA from each MNase aliquot was isolated by phenol/chloroform extraction, concentrated via ethanol precipitation and treated with RNaseA (Fermentas) at a final concentration of 1 mg/ml for 1 hour. Samples were then separated on a 2% agarose gel and bands corresponding to ∼147 bp (mono-nucleosomal fragments) were gel-extracted using the QIAquick kit (Qiagen) for each of the digestions. Samples were then analyzed on the 2100 Agilent Bioanalyzer to determine the precise fragment size and quantity, and the samples with the greatest proportion of ∼147 bp fragments were used for hybridization.
Isolation of total RNA and hybridization onto the tiling arrays followed (Juneau et al., 2007), except that Actinomycin D was added during cDNA synthesis to prevent antisense artifacts [75]. Briefly, RNA was isolated using the acid phenol method and treated with 1 U of DNase I mix (Invitrogen, Amp Grade) for 2 hours at 37°C. Cleanup was done using the RNeasy MinElute Cleanup Kit (Qiagen Cat. No. 74204). Single strand cDNA synthesis was performed using random primers in the presence of Actinomycin D in a final concentration of 6 µg/ml.
Nucleosomal DNA and cDNA samples were prepared in batches, where treatment conditions were paired with a wild-type sample that was harvested and prepared in parallel. Samples were fragmented by nuclease digestion in a solution containing 1× One-Phor-All buffer (GE Healthcare) and 1 µL of 1∶16 DNase I mix (Invitrogen, Amp Grade) at 37°C for 2 minutes, separated on a 2% agarose gel, and stained with SYBR green (Molecular Probes) to confirm that digestion produced a distribution of fragments less than 100 bp in size and a mean of 50 bp. These fragments were then labeled with terminal deoxynucleotidyl transferase (Amersham/GE Healthcare) and biotinylated ddATP (Perkin Elmer, NEL508) at 37°C for 2 hours, hybridized on Affymetrix arrays at 45°C for 24 hours, washed and stained according to protocol EukGE-WS2v4_450 in an Affymetrix Fluidics Station 450 and scanned in an Affymetrix 7G scanner (http://www.affymetrix.com/support/technical/fluidics_scripts.affx).
Preparation of Illumina libraries of mononucleosome fragments from the rpb2-tet, rlf2-Δ, tup1-Δ, cyc8-Δ, spt16-ts, chd1-Δ, snf2-Δ, set2-Δ, swr1-Δ, msi1-Δ and tbf1-ts strains was done according to [76]. Briefly, mononucleosome fragments were end-repaired, followed by amplification-free adaptor ligation and size selection on a 2% agarose gel. Clusters were generated on a single-read flowcell using Illumina's cBot, and sequenced as 2×36 nt paired-end reads using an Illumina Genome Analyzer IIx instrument. Read counts for each sample are provided in Table S5.
Mono-nucleosomal DNA tiling arrays from all mutant and wild-type control strains were quantile-normalized together with four independent monococcal nuclease-treated genomic DNA samples, using the AffyTiling package in Bioconductor [77]. A median smoothing filter was then applied to neighboring probes in the genome with a bandwidth of 30 nt, after which each array was rescaled to the same data range. Global variations in the baseline of the normalized signals were removed using waveNorm [78] with a window size of 5 kb. Genome-wide changes in nucleosome occupancy were assessed by comparing mutant strains to wild-type controls that were cultured and processed in parallel.
cDNA tiling arrays were quantile-normalized with Affymetrix Tiling Analysis Software (TAS) v1.1 using perfect-match probes only and a bandwidth of 30 nt. Raw data from cDNA hybridizations from mutant strains were normalized against cDNA from a wild-type control sample grown in parallel (mutant/WT). Data was visualized using the Integrated Genome Browser (IGB) [79].
For the nucleosome-sequencing experiments, paired-end sequencing reads were aligned to the S. cerevisiae genome (SGD, May 2008) using Bowtie v0.12.7 [80]. The midpoint of each mapped read pair was used as an estimate of a nucleosome center position.
A Gaussian filter (Matlab Statistics Toolbox) with a bandwidth of 50 nt, a mean of zero and a standard deviation of 25 was applied to the genomic-DNA normalized microarray intensity data, and nucleosome center positions were sequentially assigned to highest local maxima in the Gaussian score until no more peaks could be identified >100 nt away from already called positions [48]. The 10% nucleosome positions with the lowest Gaussian score were excluded from further analysis. For the nucleosome-sequencing experiments we determined nucleosome positions in the same way, except that the Gaussian filter was applied to counts of the number of mapped midpoints of paired-end nucleosomal reads at each position in the genome.
Nucleosome position shifts were assessed by comparing the wild-type midpoint position to that of the closest identified nucleosome position in the treatment condition within a 100 bp window around the WT condition. This window was conservatively defined to accommodate cases where an equivalent nucleosome could not be identified, e.g. because it was lost, which could otherwise lead to erroneous calculation of shifts relative to the positions of flanking nucleosomes. Our classification of nucleosome shifts is distinct from the concept of ‘positioning’ as defined by [81], which is a measure for how well-defined the position of a given nucleosome is in a population of cells. We identified a subset of significantly shifted nucleosomes that moved by at least 10 bp in the same direction in a replicate experiment, and had a P-value<0.05 in a T-test comparing the mean nucleosome position in each treatment to the mean across 31 wild-type conditions. genic nucleosomes were numbered relative to the curated transcription start site (TSS) of S. cerevisiae genes such that the +1 nucleosome is the first nucleosome with a center position downstream of the TSS. Positive and negative numbers correspond to shifts away or towards the TSS, respectively.
A manually curated set of transcript starts and ends (Table S6) was prepared based on the May 2008 assembly from SGD, using a custom R script that allows for sequential assessment of S. cerevisiae transcripts in the context of whole-genome transcription and nucleosome occupancy data. Transcript boundaries were determined based on whole-genome tiling array hybridizations of Poly(A) and total RNA from wild-type strains in this study, which were quantile normalized using Affymetrix Tiling Array Software (v1.1), together with genomic DNA hybridizations as a control group to correct for cross-hybridization and probe GC content. Additional data sets included Poly(A) RNA-Seq data [82], [83], whole-genome Poly(A) tiling arrays [75] and whole-genome nucleosome occupancy data [11]. The nucleosome occupancy data was used to confirm the positions of the 5′ end of transcripts by assessing the presence of nucleosome-depleted regions. The R script and the data files used to determine transcript boundaries are provided as Dataset S1.
Two-sided p-values were determined for each perfect-match probe using Affymetrix Tiling Analysis Software (TAS) v1.1 and a bandwidth of 30 bp, setting the mutant strains as the “treatment” and the WT strain as the “control” channel. Differentially transcribed regions were detected by setting combined thresholds for a p-value≤0.05 and a fold-change ≥2 in mutant compared to wild-type strains. These thresholds had to be met for a minimum of 80 nt (10 consecutive probes), with a maximum gap of 48 nt between probes exceeding this threshold. At these stringent settings some larger transcribed regions were detected as fragments and an additional step was therefore included to merge neighboring transcribed fragments (transfrags) if they met the following criteria: i) maximum gap of 250 bp; ii) mapped to the same strand; iii) same direction of change (both up or both down); iv) all intervening probes showed a consistent expression change in the same direction. To exclude known transcripts, transfrags that overlapped annotated coding and non-coding genes on the same strand for more than 20% of their length were removed from subsequent analyses. Remaining transfrags were classified as “tandem-sense”, “tandem-antisense”, “diverging”, “converging” or “intragenic antisense”, relative to overlapping and neighboring genes in a 3 kb region flanking the transfrag start.
Each strain was grown in the exact same condition as described for the genome-wide assays of nucleosome occupancy and transcriptome profiling. 2.0 OD600 units of cells were pelleted, the supernatant removed and the cell pellets frozen. SDS-PAGE and immunoblotting was performed as described [84]. Extracts for immunoblotting were prepared from pelleted cells that were incubated in 0.5 mL of breaking solution (0.2 M NaOH, 0.2% β-mercaptoethanol) for 10 min on ice. Proteins were precipitated with 5% trichloracetic acid (final concentration), incubated for 10 min on ice, centrifuged in a microfuge at maximum speed for 5 min. Pellets were resuspended in 1× SDS loading buffer, incubated for 5 min at 95°C, and proteins separated on 10–20% SDS–polyacryamide gels. After electrotransfer, 0.22 micron nitrocellulose membranes (Invitrogen) were blocked for 1 h in TBST (10 mM Tris-HCl, 150 mM NaCl, 0.05% Tween 20 at pH 8.0) containing 5% dry milk powder. Antibodies used for immunoblotting were anti-histone H3 (Cell Signaling Technologies), anti-yeast PGK1 (Invitrogen), peroxidase anti-peroxidase (Sigma), developed with the Pierce Pico ECL substrate (Thermo). Blot bands were quantified using Image J [85] with the relative amount of histone protein (H3) compared to the appropriate wild-type or time 0 reference.
YJM789 wild-type cells and ino80-Δ mutant cells derived from this strain were cultured at 30°C in Yeast-Peptone-Dextrose (YPD) medium, YPGalactose medium or the presence of 1,10-phenanthroline. Cells were grown to mid-log phase (OD600∼0.8), collected by centrifuging at 3,000 rpm for 5 minutes and further grown in medium lacking the presence of glucose. Cells were collected by centrifuging at 3,000 rpm for 5 min and immediately frozen in liquid nitrogen, and stored at −80°C.
Total RNA was extracted from the cells. Reverse transcription was performed with the SuperScript™ II Reverse Transcriptase from Invitrogen. In brief, 25 ng of total RNA was used as the starting material for single strand synthesis in the presence of Actinomycin D. Quantitative PCR was performed with the SYBR Green PCR Master Mix (Applied Biosystems) in an Applied Biosystems 7900HT Fast Real-Time PCR System using Sequence Detection System software version 2.3. The following temperature profile was applied: 50°C for 2 min; 95°C for 15 min; 40 cycles of the sequence 95°C for 15 s; 58°C for 1 min; 95°C for 15 s; 60°C for 15 s; 95°C for 15 s. Primer pairs used were 5′-CCTGAACAGAAACAACTACAA-3′ and 5′-CCATTGTTGTTTGGATTTGCATTG-3′ for MIG1, 5′ GCCAGGTGGCAAGTTTTTCG-3′, 5′-CTGTACTTCATTATGTGAATCACG-3′ for RGT1, 5′ GTA TAG TGG CAA CGA TAT TAA TGT C 3′ and 5′ GTA CTA CCA TTG TCA CAC CC 3′ for FRE1, and 5′-CCTGGTAGAAGTTTCATGGC-3′ and 5′-GGTCAACACATATTCGTGAGAAC-3′ for FRE7. Fold change in the MIG1, RGT1, FRE1 and FRE7 transcript level normalized to ACT1 (5′-TGTGATGTCGATGTCCGTAAG-3′ and 5′-CGGTGATTTCCTTTTGCATT-3′) was calculated using the 2−ΔΔCt method. At least three independent replicates of each reaction were performed. Student's t-test was applied for statistical analysis (paired for control vs. treatments, and unpaired for mutants vs. wild type).
Deletion of a selection of TCR genes, TCR01, TCR03, TCR05, TCR06, TCR08, TCR11, TCR15, TCR20, TCR21, TCR22, TCR27, TCR29, TCR30 and TCR31 in the strains ura8-Δ and tup1-Δ were created using a PCR-based gene deletion strategy to generate a start- to stop- codon deletion of each of the genes in the yeast genome (Table S7). As part of the deletion process, each gene disruption was replaced with a NATR module. Deletions were verified using confirmation primers (Table S7). TCR deletions in the ura8-Δ and tup1-Δ background were then grown in a wide variety of stress conditions and a spot assay was done in the presence of 5 mM EDTA and NaCl to avoid the tup1 flocculation phenotype. Spot assays were performed on 30°C YPD, 37°C YPD, 0.1 mM H2O2, 1 M NaCl, 5% ethanol, 3% ethanol, 1% ethanol, YPGlyercol, YPGalactose, 5 mM DTT, 15 mM DTT, 0.7 mM paraquat, 1 mg/ml 6-azauracil and 5 mg/ml 6-azauracil. To determine the growth defect for the spot dilution assays on various stress conditions, high-resolution images were taken of each strain and by using ImageJ (v1.45) to determine the intensity profile of the pixels that present the yeast colonies.
Affymetrix tiling array data and next-generation sequencing data are available at GEO (record GSE44879).
|
10.1371/journal.ppat.1002826 | Identification of Site-Specific Adaptations Conferring Increased Neural Cell Tropism during Human Enterovirus 71 Infection | Enterovirus 71 (EV71) is one of the most virulent enteroviruses, but the specific molecular features that enhance its ability to disseminate in humans remain unknown. We analyzed the genomic features of EV71 in an immunocompromised host with disseminated disease according to the different sites of infection. Comparison of five full-length genomes sequenced directly from respiratory, gastrointestinal, nervous system, and blood specimens revealed three nucleotide changes that occurred within a five-day period: a non-conservative amino acid change in VP1 located within the BC loop (L97R), a region considered as an immunogenic site and possibly important in poliovirus host adaptation; a conservative amino acid substitution in protein 2B (A38V); and a silent mutation in protein 3D (L175). Infectious clones were constructed using both BrCr (lineage A) and the clinical strain (lineage C) backgrounds containing either one or both non-synonymous mutations. In vitro cell tropism and competition assays revealed that the VP197 Leu to Arg substitution within the BC loop conferred a replicative advantage in SH-SY5Y cells of neuroblastoma origin. Interestingly, this mutation was frequently associated in vitro with a second non-conservative mutation (E167G or E167A) in the VP1 EF loop in neuroblastoma cells. Comparative models of these EV71 VP1 variants were built to determine how the substitutions might affect VP1 structure and/or interactions with host cells and suggest that, while no significant structural changes were observed, the substitutions may alter interactions with host cell receptors. Taken together, our results show that the VP1 BC loop region of EV71 plays a critical role in cell tropism independent of EV71 lineage and, thus, may have contributed to dissemination and neurotropism in the immunocompromised patient.
| Human enterovirus-71 (EV71) has been the cause of major hand-foot-and-mouth disease outbreaks, particularly in the Asia-Pacific region. EV71 infection can also disseminate to the central nervous system and result in meningo-encephalitis. Despite intensive epidemiological screening, as well as experimentation in animal models, viral factors contributing to neurotropism remain ill-defined. We describe here the analysis of the full-length genomes of EV71 from different infection sites in an immunocompromised host with disseminated disease. Our data highlight a critical amino acid change within the EV71 VP1 protein that could potentially lead to dissemination and neurotropism during natural infections. This hypothesis was confirmed in vitro through reverse genetic experiments in different EV71 lineages and by in silico modelling. To our knowledge, this study provides the first genome-wide analysis of EV71 evolution and dissemination within a single human host over the course of an infection, and highlights how the emergence of mutations at critical regions of the viral genome can potentially lead to new phenotypes and neurovirulence.
| In humans, enteroviruses target a variety of different organs causing gastrointestinal, respiratory, myocardial, and central nervous system (CNS) diseases [1], [2]. The ability of enteroviruses other than poliovirus to cause neurological complications is restricted to a limited number of serotypes that include enterovirus 71 (EV71) [3], [4]. EV71 is of particular interest since it can cause major hand-foot-and-mouth disease outbreaks, such as those recently reported across the Asia-Pacific countries [5]–[8]. Nevertheless, EV71 dissemination to the CNS remains a rare event, as demonstrated by the relatively small proportion of meningo-encephalitis among millions of hand-foot-and-mouth disease cases [9]–[12].
For poliovirus, CNS invasion is thought to occur either through disruption of the blood-brain barrier or via retrograde axonal transport [8]. For EV71, experimental studies in mouse models using adapted strains suggest that the virus has the propensity to invade the CNS through retrograde axonal transport and that hematogenous transport might represent only a minor route of transmission [13]–[15]. However, the observations in mouse models do not necessarily reflect how CNS invasion occurs during human infections.
Neutrotropic enteroviruses need to escape the host defences to reach the CNS. The absence of pre-existing protective immunity, together with a relatively deficient innate immunity, is considered as the first step toward high blood viremia that will then lead to a secondary invasion of the CNS [16]. This explains why young children present more severe diseases. An inefficient immune response could also be the result of a high inoculum size, leading to an overwhelming replication and viremia. However, neurotropism is a multistep event that requires the virus not only to sustain high replication levels, but also to locate a permissive cell type within the CNS. Viral factors contributing to neurotropism have been intensively studied in vitro and in animal models in vivo using poliovirus or non-polio EVs [15]–[23], but still remain ill-defined. Until now, to the best of our knowledge, EV71 virulence factors and adaptation have not been studied directly from clinical samples during natural human infections and it remains unknown whether secondary seeding from the primary site is only a fortuitous event or if it is associated with specific viral genomic adaptation within the human host.
In this study, we analyzed the genomes of EV71 from different sites of infection in an immunocompromised host with disseminated disease. This provided a unique opportunity to investigate any potential intra-host adaptation following natural human infection and to assess whether enterovirus needs to harbor specific genomic features in order to sustain dissemination. After sequence analysis of the collected specimens, amino acid changes observed in the viral proteins VP1 and 2B and possibly associated with neurotropism were further studied both in vitro using a series of different constructs and in silico using comparative models of EV71 VP1.
A 38-year-old man with chronic lymphocytic leukemia and recently treated with four courses of chemotherapy, including rituximab, was hospitalized with fever and respiratory symptoms. Five days before admission, he developed fever (39°C), odynophagia, chills, dyspnoea with wheezing, cough and sputum. The total immoglobulin G level in blood was low at 2.74 g/L (normal range, 6.06–13.18 g/L), as were the IgM (0.1 g/L; normal range, 0.29–3.25) and IgA (0.17 g/L; normal range, 0.66–3.99) levels. Despite intravenous wide-spectrum antibiotic and antifungal treatment, fever persisted together with diarrhea. The appearance of meningeal signs prompted a lumbar puncture that revealed a slight inflammation with six white blood cells/mm3, but normal protein and glucose levels. Microbiological investigations revealed a positive enterovirus RT-PCR signal in the lower respiratory specimens (BAL), plasma, cerebrospinal fluid (CSF), and stools. Viral culture was positive for enterovirus in the respiratory tract and stools. Additional extensive microbiological investigations were all negative for any other bacterial, fungal, or viral infections. Disseminated enteroviral disease was diagnosed and the clinical condition improved rapidly after immunoglobulins were infused. This infusion was followed by a clearance of the infection in blood as shown by a negative RT-PCR assay at day 7 after infusion without relapse or evidence of persisting enteroviral infection.
The full-length enterovirus genomes were sequenced directly from BAL, stool, plasma (at days 0 and 4) and CSF specimens (Genbank accession numbers: EU414331 to EU414335). A whole genome BLAST search and a phylogenetic tree with available EV full-length genomes (http://www.picornaviridae.com/enterovirus/hev-a/hev-a_seqs.htm) revealed that this strain clusters with other EV71 serotypes within the human EV-A species. This serotyping was confirmed by immunofluorescence with an anti-EV71 monoclononal antibody applied on the BAL and the stool isolates grown in Vero cells.
This clinical enterovirus strain is related to the genogroup C1. Its full-length polyprotein sequence was then compared to publicly available full-length EV71 sequences and linked with the identified associated clinical conditions. This large-scale inter-host analysis did not identify any genomic features that could be related to specific clinical features or to disease severity. This finding indirectly supported the completion of an intra-host full-length genome analysis to find critical residues that could promote virus dissemination and invasion of the CNS.
Genomic DNA sequences and polyprotein comparisons of the five different specimens revealed two non-synonymous substitutions at positions 662 and 1050, and one synonymous substitution at position 1906 of the EV71 polyprotein (GenBank accession number: AAB39968.1). These positions correspond to amino acid 97 of VP1, 38 of 2B, and 175 of 3D (Table 1). None of these three mutations had any effect on the RNA secondary structure in the specific regions (data not shown). No other mutations were observed.
To investigate whether these two changes could play a role in immune escape, we established quantified viral stocks in Vero cells with the BAL (VP197L-2B38V) and the stool (VP197R-2B38A) isolates, respectively. Conservation of these two substitutions after cell passage was confirmed by re-sequencing. The two isolates were tested for seroneutralization in the presence of the patient's serum (sampled at day 4) at a time when the VP197R substitution was already present in plasma. Neither the BAL isolate nor the stool isolate were neutralized by the patient's serum (Table 2). A negative complement fixation assay confirmed a poor antibody response against enterovirus (data not shown). Of note, in the presence of the anti-EV71 monoclonal antibody, the growth of the VP197R-2B38A stool isolate was inhibited at a dilution <1∶30, whereas the BAL isolate remained insensitive, thus arguing against an immune escape advantage resulting from the VP197R substitution.
To investigate the potential implication of the mutations on tissue tropism, we then inoculated the VP197L-2B38V (BAL) and VP197R-2B38A (stool) isolates on three cell lines (astrocytoma [U-87 MG], neuroepithelioma [SK-N-MC], and neuroblastoma cell lines [SH-SY5Y]) previously used as references to confirm the ability of poliovirus [34], [35] or EV71 [23], [36] to infect cells of neural origin. Figure 1 shows that the stool isolate presents a strong replication advantage over the respiratory tract specimen in cells of neuroblastoma origin, whereas the two isolates replicate in similar fashion in the astrocytoma cell line (data not shown). Of note, neither of the two isolates was able to grow in neuroepithelioma cell lines, although a wild type poliovirus used as control grew easily under the same conditions (data not shown).
To assess the implication of each of the two non-synonymous substitutions governing the replicative advantage of the stool isolate in cells of neuroblastoma origin, we designed four infectious clones strictly similar to the full-length sequences of the stool or the BAL specimens, as well as two that harbor either the VP197R or the 2B38A substitutions alone (Figure 2A). We analyzed the replication efficiency of these four constructs in neuroblastoma cells (Figure 2B–2C) and other cell types (Table 3). As expected, and in contrast to the VP197L2B38V and VP197L2B38A clones that presented a delayed growth phenotype in SH-SY5Y, the virus with the stool isolate sequence and the clone with the VP197R substitution only grew very efficiently in Vero and SH-SY5Y cells, suggesting that the VP1 L97R substitution confers a replicative advantage in neuroblastoma cell line. The immunofluorescence results were corroborated by single-step replication analysis of the four derivatives in Vero and SH-SY5Y cell lines. Although the four derivatives demonstrated similar growth kinetics in Vero cells, the derivatives with the VP1L97R substitution (pClVP197R2B38A and pClVP197R2B38V) presented a strong replicative advantage in SH-SY5Y cells (Figure 2C). A significant difference linked to the VP1 sequence was also observed in cells of colorectal adenocarcinoma origin (Caco-2) with the opposite phenotype in this cell line since the growth advantage was conferred by the VP197L sequence (Table 3).
To further confirm the contribution of the VP1 L97R substitution towards the replicative advantage in cells of neuroblastoma origin, these changes were introduced individually or together in the EV71BrCr (GenBank accession number: AB204852.1) infectious clone background (kindly provided by Prof. M Arita, National Institute of Infectious Diseases, Tokyo, Japan) that presents 81% nt and 96% aa identity with the sequence of the clinical isolates. Of note, a Lys at position 98 that reduced the replication of pBrCr in Vero cells due to the introduction of a positively charged aa was substituted with a Glu to restore normal replication. Although the replication of BrCr derivatives is very low in neuroblastoma cells, the trend was similar to that observed with the four infectious clones (data not shown).
Competition experiments were performed in Vero, SH-SY5Y, and Caco-2 cells to confirm these observations. For this purpose, equimolar amounts of RNA from the infectious clones derived from the stool (pClVP197R2B38A) and BAL (pClVP197L2B38V) isolates were co-transfected in these three cell lines. The supernatant was collected and viral sequences analysed at different time points post-transfection. As early as 24 h post-transfection, the observed dominant species in Vero and Caco-2 cells was that with VP197L. Regarding SH-SY5Y, at the beginning of the competition (24 h post-transfection) the population with VP197L appeared to slightly dominate over the VP197R population. However, the situation reversed after 4 days and the VP197R population became the dominant species (between 24 and 48 h after the first passage) (Figure 3B, left panel). The fact that the VP197L sequence dominates shortly after transfection suggests that once inside the cell, this position confers a replicative advantage over the VP197R sequence. Therefore, the VP197R sequence likely presents an advantage at the cell entry stage of the viral growth cycle. Interestingly, in two of four competition experiments, the VP197R substitution was rapidly associated with a second substitution (glutamate to glycine) present at position 167 (E167G) of VP1 (VP1167G) (Figure 3B, right panel). By retrospective sequence analysis of SH-SY5Y cells infected with the stool isolate or transfected with the stool or pClVP197R2B38A derivatives, position 167 was almost always mutated into a glycine or an alanine. Alignment of the 952 VP1 sequences currently available in Genbank shows that only one other EV71 sequence (GenBank accession number: AAF13503.1) contains an alanine at position 167. Interestingly this strain also has an arginine at position 97 of VP1 (VP197R).
Finally, to investigate any potential implication of the L97R substitution regarding sensitivity to interferon beta, we co-transfected pClVP197R2B38A and pClVP197L2B38V in Vero cells (that do not produce, but are sensitive to interferon [37], [38]) pre-treated with interferon beta. The viral replication was strongly reduced by the presence of interferon beta and the VP197R substitution did not provide any advantage to the virus since the pCl VP197L2B38V construct was dominant after 24 h in Vero cells in the presence or absence of interferon (data not shown).
To determine if the substitution at VP1 residue 97 could have a structural impact and/or influence how the viral capsid interacts with cellular receptors or co-receptors, we generated and validated comparative models of EV71 VP197L and VP197R based on the known VP1 structures of 10 other closely related picornaviruses. Comparison of the energy signatures and structures of the models revealed that VP197R has no significant energetic or backbone conformational differences relative to VP197L (data not shown), suggesting that this substitution functions by influencing interactions at the capsid-host cell interface. To further assess this possibility, we aligned all known picornavirus VP1 structures that are in complex with their corresponding cellular receptors. In seven of the eight VP1-receptor structures (PDB accession codes listed in Figure 4A), the receptors bind in a canyon that contains the base of the BC loop, albeit in different orientations. One of the eight structures (PDB 3dpr: a human rhinovirus 2 [HRV2] bound to its receptor) revealed that the receptor interacts with VP1 not in the canyon, but directly above the BC loop at the five-fold axis of symmetry of VP1 (Figures 4A and C). We then aligned our EV71 VP1 models to the VP1 molecules in these structures and observed that EV71 VP197 is within 10–12 Å of the receptor surfaces. Given that amino acid sequences within the BC loop contribute to receptor selectivity among picornaviruses [32], [33] (Figure 4D), and that different strains interact with their receptor in different orientations and regions, we speculated that the positive charge introduced by the VP197R substitution could be located at the interface of human EV71 receptors and facilitate interactions with host cell receptors. Indeed, after aligning our EV71 VP1 models to the poliovirus VP1 monomers (PDB 3epf) of a complete viral capsid assembly, the arrangement of the VP1 5-mer revealed that residue 97 was close to the five-fold axis of symmetry (Figures 4B and C) in the region known to interact with host cells [39]. This model is further supported by a virus binding assay performed in Vero and SH-SY5Y cells (Figure 5). A difference in binding competence is observed in favour of pCIVP197R2B38A compared to pCIVP197L2B38V, which supports the importance of the VP197R substitution in the receptor–binding process. Of note the VP1167 position, where the compensatory E167G mutation occurred in vitro in neuroblastoma cells, lies near the interface of VP1 monomers in the capsid assembly (Figures 4B and C). Residue 167 is positioned against another negatively charged residue on the adjacent VP1 monomer and replacement of this glutamate 167 with a glycine might serve to stabilize the capsid assembly by alleviating steric and/or electrostatic interference between VP1 monomers or receptors when position 97 is mutated to an arginine.
Many investigations have focused on the molecular epidemiology of EV71 [19], [20], [31], [40]–[56], but few have attempted to identify in-host adaptation and the potential viral determinant of neurotropism or neurovirulence. Mutations in EV71 5′ UTR [23], [53], [57], VP1 gene (including the BC loop) [31], [53], and 3D polymerase [20], [22] have been shown to result in attenuation in cynomologus monkeys and in mice, but they do not change the tissue specificity in the CNS of these experimental animal models [20]–[22]. These models have many intrinsic limitations, namely, the use of adapted EV71 mouse strains and/or direct intracranial, intramuscular, or intraperitoneal inoculation. These experimental models are thus unable to mimic the natural route of infection in humans.
In this study, we analyzed the genomic differences in the EV71 genogroup C1 virus during a disseminated human disease that included meningitis. EV71 serotypes are divided into three major genetic lineages; lineage A whose prototype is the BrCr strain, and lineages B and C [31] that are further subdivided into subgenogroups B1 to B5 and C1 to C5 [8], [31], [40], [41], [43]. Studies have suggested that the C1 genogroup is rarely a cause of CNS infection [11], [41]. Our goal was to identify viral signatures that could account for dissemination or site-specific adaptation. The comparison of five full-length genomes, sequenced directly from respiratory, gastrointestinal, CSF and blood specimens, revealed a drastic non-synonymous L97R substitution in the BC loop of the VP1 capsid protein that significantly modified the resulting viral phenotype. This mutation was specifically present in the blood and CSF, but not in the respiratory tract, and was present as a mixed population in the gastrointestinal tract. In addition to the VP197R substitution, a conservative amino acid substitution at position 38 of protein 2B (V38A) was also observed in the blood, CSF, and gastrointestinal tract. Finally, a mixture of two nucleotides, both translated into a leucine in the 3D gene (3D175), was also observed in the sequence of the stool specimen.
The BC loop region of VP1 is a known dominant immunogenic site as evidenced from experimental models using laboratory-adapted poliovirus or coxsackie virus strains [24]–[29]. Seroneutralization experiments with the patient's serum failed to highlight the presence of an antibody-mediated, selective immune pressure promoting the VP1 L97R substitution. Thus, it is unlikely that the VP197R sequence present in the immunocompromised patient's blood and CSF played a critical role in viral dissemination to the nervous system via immune escape. This has to be related to the immunosuppressed condition of the patient who was previously treated with anti-CD20 antibodies, although we cannot rule out the presence of low level antibodies or cellular immunity directed against the VP197L strain that have gone undetected.
Apart from its immunogenic role, the BC loop region of VP1 was also identified as a determinant of poliovirus host adaptation [32], [33], [58]. In vitro cell tropism assays revealed that the VP197R, conferred a significant advantage to the ability of EV71 to grow in neural-derived cells, independent of the virus lineage. The results of competition experiments suggest that the advantage is probably at the cell entry step. Indeed, introduction of a positively charged amino acid in the BC loop may have a substantial impact on the interaction of host cell surface receptors with this epitope. The EV71 VP1 structure models and the virus binding assay further support this hypothesis by revealing that residue 97 is close (∼12 Å) to the interface of other known picornavirus VP1 receptors (Figure 4A), and by illustrating how the positively charged arginine side chain of VP197R on the viral capsid surface may be more accessible to certain host cell receptors than the smaller side-chain of leucine of VP197L (Figure 4C). Of note, backbone carbon atom alignment of our models to the EV71 VP1 structures [59], [60] that were released while this manuscript was under review shows less than 1.0 angstrom RMSD between models and structures (data not shown), thus further validating our modelling approach.
Interestingly, although the improved receptor-binding capacity of VP197R might be sufficient to confer a growth advantage in neuroblastoma cells, it cannot be ruled out that this substitution also confers potential advantages at various other stages (e.g., during the virion assembly process). This binding advantage was also observed to a lesser extent in Vero cells. Therefore, compensatory events must occur at one or multiple steps during virus amplification (viral genome replication, assembly, recruitment of cellular factors, others) to explain the VP197L advantage observed in the competition assay in this cell line. Of note, inoculation of the patient's stool specimen (that presented both VP197R and VP197L populations) in Vero cells resulted in virus isolate containing only the VP197R sequence. While this may seem surprising in light of the results from the competition experiment, one must note that Vero cells were transfected with RNA transcripts from pCI derivatives, thus bypassing the viral entry step. According to the binding assay, VP197R also provides a binding advantage in Vero cells that may partially explain these contradictory observations. Another possible explanation is that while the presence of a mixed population in stools was shown at the RNA level, there is no indication about the viability of the corresponding viral species in the sample. Thus, the presence of a leftover, potentially defective, viral VP197L genome that is unable to grow in culture cannot be ruled out.
Notably, in neuroblastoma cells, VP197R was frequently associated with a second mutation located in the EF loop at position 167 (E167G) of VP1 (VP1167G). In our structure models, position 167 is situated in the receptor-binding canyon near the base of the BC loop and may serve to limit the conformational flexibility of the BC loop (Figures 4A–C). Substitution of a negatively charged glutamate by the smaller neutral glycine may alleviate steric and/or electrostatic interference created by the VP197R at the VP1-receptor interface, thus serving to stabilize the VP1 interaction with the host cell receptor. This E167G substitution was absent in all of the patient's specimens analyzed, suggesting that this position is either not clinically relevant and only reflects cell-type adaptation under our experimental conditions, or that there was not enough time for it to appear during the course of infection. If more time had elapsed before treatment was administered, it is possible that the VP1167G substitution would have appeared and, in turn, exacerbated the patient's symptoms. This second hypothesis is favored by the finding of a publicly available sequence that contains both VP197R and VP1167G residues (GenBank accession number: AAF13503.1).
In vivo studies in mouse models [61]–[63] and a comparative analysis of all EV71 complete genome sequences with identified clinical backgrounds available in the Genbank database [64] both identified amino acid positions in VP1 associated with EV71 virulence, such as VP1145 in the DE loop situated on the rim of the surface canyon, or VP1164 in the EF loop situated on the slope of the canyon (Figure 4). Furthermore, after amplification of infectious clones harboring EV71 subgenogroup B3 in SH-SY5Y and RD cell lines, VP194 was recently identified and is postulated to be important for cell-type adaptation [36]. Taken together, the region surrounding the VP1 L97R mutation identified in this study likely plays an important role in cell-type adaptation and potentially neurotropism, independent of the EV71 genogroup.
In addition to the VP197R substitution, a conservative amino acid substitution at position 38 of protein 2B was observed, 2B38A. This substitution is uncommon and described in only one case among all available GenBank sequences. Whether this conservative V38A change in protein 2B may also confer new viral tropism was not substantiated in our experiments and remains an unsupported hypothesis.
Taken together, the sequence of clinical events, the genome characterization, our in vitro experiments, and our comparative VP1 structure models support the following scenario: the virus could have initially infected the respiratory tract, leading to a first viremic phase followed by invasion of the gastrointestinal tract. Alternatively, the virus may have entered simultaneously by oro-fecal and respiratory routes. High replication in the gastrointestinal tract may then have given rise to the appearance of a mixed viral population. The reduced immune response of the host then allowed a prolonged viremia, originating from a subspecies generated during the replication within the gastrointestinal tract that conferred a selective advantage for certain cell types, including neural cells. This resulted finally in neuro-invasion. In conclusion, this study provides the first genome-wide analysis of EV71 evolution and dissemination within a single human host over the course of an infection, and highlights how emergence of mutations at critical regions of the viral genome can lead to new phenotypes and neurovirulence. Further studies are underway to better define the target of the VP197R substitution and to investigate any potential effects of the associated mutation, VP1167G.
The study was approved by the institutional ethics committee of the University Hospitals of Geneva, Switzerland. Given the nature of the investigation, that none of the sampling was done for the purpose of this investigation, and that enterovirus genotyping is part of our routine surveillance activity, the requirement of written consent was waived by the ethical review board. Oral informed consent was obtained from the patient concerning the fact that the infective virus would be characterized.
The following specimens were collected for diagnostic purposes at different times in a patient hospitalized with a disseminated EV71 infection: a bronchoalveolar lavage fluid (BAL) was collected upon patient admission; a blood and stool specimen were collected 17 h and 19 h later, respectively, as well as a second blood sample after 4 days; and a cerebrospinal fluid (CSF) sample after 5 days.
RNA extraction in blood was performed with the NucliSens miniMAG method, according to the manufacturer's instructions (bioMérieux, Geneva, Switzerland). RNA extraction in BAL, CSF, and feces was performed with TRIzol (Invitrogen, Carlsbad, CA, USA). RNA extraction from infected cell and infected cell supernatant was performed with easyMAG (bioMérieux). Reverse transcription was carried out with the Superscript II RNase H− enzyme (Invitrogen) with both random hexamers and oligodT (for the most 3′ part of the genome). Real-time RT-PCR enteroviral screening was then performed with Taqman Universal Mastermix (Applied Biosystems, Rotkreuz, Switzerland) with primers and probe sequences described previously [65] and Entero/Ge/08 [66].
Amplification and detection were achieved with ABI Prism 7900 and 7000 sequence detection system (Applied Biosystems) according to methods previously described [67].
For single-step replication quantification, the Entero/Ge/08 assay was used in a one-step format using the QuantiTect Probe RT-PCR Kit (Qiagen, Hombrechtikon, Switzerland) according to the manufacturer's instructions in a 7000 Applied Biosystems thermocycler. For each derivative, viral amplicon Ct values were normalized according to the input RNA amount present in the inoculum and then to those of the endogenous RNase P gene (TaqMan RNase P Control Reagents, Applied Biosystems). Relative quantification was calculated using the 2−ΔΔCt method [68]. The quantitative Entero/Ge/08 assay was run using a 10-fold dilution series (from 5*107 to 5*104 copies/ml) of the in vitro transcribed full-length pClVP197R2B38V derivative, which was used as a quantitative reference curve for each run.
Overlapping fragments representing the complete viral genome were amplified by PCR using the AmpliTaq polymerase (Applied Biosystems) and primers designed on the basis of previously published EV71 strains. Specific primers were then designed to fill the gaps. All primers used are listed in Table S1. PCR products were purified and sequenced as previously described [67]. Each product was sequenced at least twice and analyzed by vector NTI Advance 10 software (Invitrogen). Ambiguous nucleotides were resolved by re-sequencing. To avoid the introduction of mutations by cell culture adaptation, full-length sequences were obtained directly from the clinical samples. The three nucleotide differences observed between the samples were confirmed by a new cycle of PCR and sequencing.
Diluted stool and BAL specimens were used to inoculate Vero-76 cells in 1.6 ml infection medium (Dulbecco plus 2.5% fetal calf serum, 0.2% sodium bicarbonate, penicillin, streptomycin, fungisone, gentamicin, and Hepes). Viral stocks were prepared after three cell passages. To confirm the conservation of the substitutions, partial VP1 and 2B amplifications and sequencing of the stocks were carried out with primers AN89 and AN88 [69] and primers 30 and 15, respectively, and sequenced with the PCR primers (Table S1). The stocks were quantified at day 7 post infection according to the Reed and Muench method [70] and presented a 105.7 and 105.62 TCID50/ml for the stool and lower respiratory samples, respectively.
For seroneutralization assays, 100 uL of virus diluted stocks (dilution factor 104) were incubated with 100 uL of diluted patient serum sampled at day 4 or with 100 uL of mouse anti-EV71 monoclonal antibody (MAB979; Chemicon, Temecula, CA, USA); antibody dilutions were 1/5, 1/10, 1/15, 1/20 and 1/30. The mix was incubated for 1 h at 37°C before inoculation on Vero cells and then incubated at 37°C for one week. Each dilution was performed in duplicate and the experiment was repeated twice. Control dilutions of the virus in the absence of antibody were performed in the same experiment. The cytopathic effect was read daily. The inhibition titre was defined at day 8 post inoculation. The BC loop of the 2 viral isolates was sequenced after neutralization with the patient serum to exclude reversion mutation.
Four infectious clones containing the stool sequence (pClVP197R2B38A), the lower respiratory tract sequence (pClVP197L2B38V), and two hybrid sequences (pClVP197R2B38V and pClVP197L2B38A) cloned with MluI/BamH1 in a modified pcDNA3.1 vector were ordered at Biomatik (Ontario, Canada).
In vitro transcription of pCl plasmids linearized with BamH1 were performed as previously described [71]. Vero cells were seeded at 6×105 cells in 35 mm wells of a 6-well plate. The following day, cells were transfected with 2 ug of RNA transcripts containing the different pCl derivatives using the TransMessenger Transfection Reagent kit (Qiagen). After 3 h at 37°C, the infection medium (see below) was used to replace the transfection mix. Cells were then incubated at 37°C. For pCl derivatives, cell supernatants were collected when the cytopathic effect was noticeable and passaged once in 24 cm2 cell culture flasks. Cells were incubated at 37°C until appearance of cytopathic effects (3–4 days post-transfection). Supernatants were then immediately clarified 5 min at 1000 rpm in a Multifuge 4 KR, aliquoted and stored at −70°C. The stocks were quantified according to the Reed and Muench method [70].
For competition experiments, equimolar amounts of RNA from pClVP197L2B38V and pClVP197R2B38A were co-transfected in Vero cells pretreated with 8 U/ml of Interferon beta (Rebif; Merck Serono, Geneva, Switzerland) for 12 h before transfection, SH-SY5Y cells, and Caco-2 cells. Cell supernatants were then collected at different time points and repassaged. The input RNA and the RNA extracted from cell supernatants collected at different time points were reverse transcribed and analyzed by PCR and sequencing.
Vero-76 (ATCC # CRL-1587), H292 (ATCC # CRL1848), Caco-2 cells (ATCC # HTB 37), SH-SY5Y (ATCC #CRL-2266), SK-N-MC (ATCC #HTB10), and U-87 MG (ATCC #HTB14) cell lines were used for the cell tropism assay. The infection medium for the HEp-2 cells is the same as the infection medium for Vero cells (see above). For SK-N-MC and U-87 MG cells, Dulbecco is replaced with McCoy medium for SH-SY5Y with RPMI 1640, with M100 supplemented with 10% fetal calf serum (FCS) for Caco-2 cells, and Eagle's minimum essential medium (EMEM) supplemented with 10% FCS for H292.
For pCl derivatives, replication was assessed by immunofluorescence 36 h post-infection for SH-SY5Y and after 48 h for Vero, H292, Caco-2, SK-N-MC, and U-87 MG cells. Virus isolated from SH-SY5Y and Vero cell supernatants were analysed by complete genome sequencing and quantified by the Reed and Muench method [70]. Replication in Vero and SH-SY5Y cells was further analyzed by real-time RT-PCR on RNA extracted from total cell lysates 4 h, 8 h, 12 h, 24 h and 48 h post-infection. Infections were performed in duplicate for each time point.
EV71-infected cells were labelled as follows: cells were washed twice with phosphate buffered saline (PBS) lacking Ca2+ and Mg2+ (PBS−) and fixed 1.5 h in methanol-acetone (50∶50) at −20°C. Cells were air-dried for a few minutes at room temperature before incubation with the mouse anti-EV71 (MAB979; Chemicon) primary antibody diluted 1/40 in PBS−-1% bovine serum albumin (BSA), for 45 min at 37°C in a humidity chamber. After intensive washing with PBS−, the anti-mouse IgG AB/FITC containing 0.02% Evans Blue counterstain (Millipore-Light Diagnostics, Zug, Switzerland) was added and the cells were incubated for 45 min at 37°C in the dark. After final rinsing with PBS−, coverslips were mounted in fluorotec embedding medium (BioScience AG, Rüschlikon/Zurich, Switzerland). Quantification of virus growth was calculated either manually as the percentage of positive cells or by metamorph analysis.
Images were acquired on a Zeiss AxioCam microscope with 20× and 10× objectives, leading to the calibration of 0.33 µm/pixel and 0.67 µm/pixel. Images were acquired in 2400×2500 spatial resolution and 24 bit color depth (8 bit/channel). To measure the positive markers, the following image analysis was performed with Metamorph/MetaXpress software (Molecular Devices, Sunnyvale, CA). The blue channel of the images contained DAPI-stained nuclei and the positive cells were marked with antibody-GFP. The first step of processing involved separating the two channels (blue and green); respective channels were converted from 8 to 16 bit by multiplication and the “CellScoring” tool of Metamorph software was applied to 16 bit versions of blue and green channels. Parameters used for images at 10× magnification were as follows: cell minimum width, 7; cell maximum width, 20; intensity above local threshold, 20. For images taken with the 20× objective, the respective parameters were: cell minimum width, 13; cell maximum width, 40; intensity above local threshold, 20. For both series, the positive marking was sought in the cytoplasm (parameter “Stained area”). Reported parameters include: total cell number; positive cell number; and their relative percentage.
SH-SY5Y and Vero cells were seeded at 2×104 and 4×104 cells/well, respectively, in 96-well plates. The following day, culture medium is removed and cells are washed once with cold Hanks' Balanced Salt Solution (HBSS) with CaCl2 and MgCl2 (Invitrogen). 200 µL of binding buffer (HBSS containing 1% BSA and 0.1% sodium azide) are then added and cells chilled on ice for 10 min. Supernatant is removed from cells. 100 µL of pCIVP197R2B38A and pCIVP197L2B38V stocks (amplified in Vero cells) standardized by the Entero/Ge/08 real-time RT-PCR (further confirmed by sequencing-chromatogram ratios analysis of standardized pCIVP197R2B38Aand pCIVP197L2B38V pooled stocks) are added. After 1 h of incubation on ice, unbound virus is removed by three wash steps with 200 µL/well of cold binding buffer and then cells are lysed in the wells with 200 uL of easyMAG lysis buffer. Viral RNA is extracted with the NucliSens miniMAG method and detected by real-time RT-PCR using the Entero/Ge/08 assay. RNase P quantification by Taqman assay (Applied Biosystems) was used for normalization. The virus binding assays were performed systematically in duplicate in two individual experiments for each condition.
The EV71 VP1 models were generated and evaluated using the molecular modelling suite MODELLER v9.9 [72]. As input for the modelling algorithm, our clinically isolated EV71 VP1 amino acid sequences were used to find homologous viral VP1 structures in the Protein Data Bank [PDB]. Five structures with the highest percentage identity to the EV71 VP1 amino acid sequence were selected to serve as initial model templates; bovine enterovirus VG-5-27 (PDB accession code: 1bev, 46% identity to EV71 VP1), coxsackievirus B3 coat protein (1cov, 42%), swine vesicular disease virus (1fpn, 42%), human rhinovirus serotype 2 (1r1a, 43%), and serotype 1A (1oop, 42%). The EV71 VP1 models were evaluated using the GA341 potential and the discrete optimized protein energy (DOPE) algorithms of MODELLER v9.9 [73]. While a high degree of structural variability was found in the N- and C- termini of the EV71 VP1 models, the ‘core’ VP1 sequence that is exposed and interacts with the host environment is highly similar in all template structures (core EV71 VP1 models have an overall root mean square deviation (RMSD) of ∼0.25 Å for backbone carbon atoms). In a second round of modelling, the core EV71 VP1 sequence (residues 77 to 287) was used to select 10 homologous template structures from the PDB that yielded a set containing the original five structures with human rhinovirus 16 (1aym, 42% identical to EV71 VP1 core residues), poliovirus type 2 (1eah, 42%), Sabin strain poliovirus PV3 (1pvc, 41%), swine vesicular disease virus (1mqt, 40%), and human coxsackievirus (1z7s, 40%). EV71 VP1 models were generated using these 10 template structures and assessed with GA341 and DOPE algorithms.
To investigate the VP1 models in the context of the five-fold axis of symmetry found in viral capsid assemblies, the EV71 VP1 models were structurally aligned to each VP1 monomer of the poliovirus VP1 (PDB accession code: 3epf) as arranged in the half-capsid structural coordinates found in the Virus Particle Explorer database VIPERdb2 [74] using PyMOL version 1.4.1. (The PyMOL Molecular Graphics System, Version 1.3, Schrödinger, LLC, New York, NY).
|
10.1371/journal.ppat.1001052 | Damaged Intestinal Epithelial Integrity Linked to Microbial Translocation in Pathogenic Simian Immunodeficiency Virus Infections | The chronic phase of HIV infection is marked by pathological activation of the immune system, the extent of which better predicts disease progression than either plasma viral load or CD4+ T cell count. Recently, translocation of microbial products from the gastrointestinal tract has been proposed as an underlying cause of this immune activation, based on indirect evidence including the detection of microbial products and specific immune responses in the plasma of chronically HIV-infected humans or SIV-infected Asian macaques. We analyzed tissues from SIV-infected rhesus macaques (RMs) to provide direct in situ evidence for translocation of microbial constituents from the lumen of the intestine into the lamina propria and to draining and peripheral lymph nodes and liver, accompanied by local immune responses in affected tissues. In chronically SIV-infected RMs this translocation is associated with breakdown of the integrity of the epithelial barrier of the gastrointestinal (GI) tract and apparent inability of lamina propria macrophages to effectively phagocytose translocated microbial constituents. By contrast, in the chronic phase of SIV infection in sooty mangabeys, we found no evidence of epithelial barrier breakdown, no increased microbial translocation and no pathological immune activation. Because immune activation is characteristic of the chronic phase of progressive HIV/SIV infections, these findings suggest that increased microbial translocation from the GI tract, in excess of capacity to clear the translocated microbial constituents, helps drive pathological immune activation. Novel therapeutic approaches to inhibit microbial translocation and/or attenuate chronic immune activation in HIV-infected individuals may complement treatments aimed at direct suppression of viral replication.
| Persistent activation of the immune system is a hallmark of chronic HIV/SIV infections and predicts disease progression better than either plasma viral load or CD4+ T cell count. While the causes of immune activation during chronic infection are likely multifactorial, recent work has shown that microbial translocation is associated with immune activation. However, direct, tissue level in vivo evidence of translocation and the underlying mechanisms remain unclear. Here, we sought direct in vivo evidence of translocation, and an understanding of the timing and the underlying mechanisms. We found that in RMs, microbial translocation begins during the late acute phase of SIV infection and increases progressively during chronic infection and is associated with structural damage of the GI tract. We further discovered that immune activation is temporally and causally related to microbial translocation and by the relative inability of intestinal macrophages to bind/phagocytose translocated microbial products. In SIV-infected sooty mangabeys, however, no evidence of epithelial barrier breakdown, nor increased microbial translocation or chronic immune activation were observed. Our results provide direct evidence for microbial translocation in vivo, coupled with early and progressive intestinal epithelial damage, and eventual impairment of macrophage clearance associated with dissemination of microbial products and systemic immune activation.
| Persistently elevated immune activation characterized by polyclonal B cell activation [1], increased T-cell turnover [2], increased frequencies of T cells with an activated phenotype [3], and increased levels of pro-inflammatory molecules [4] is a hallmark of disease progression in pathogenic HIV/SIV primate lentiviral infections and is a stronger predictor of disease progression than either CD4+ T-cell count or plasma viral load [5]. The importance of immune activation to disease progression in HIV/SIV infections is highlighted by the low levels of immune activation measured during the chronic phase of infection in natural hosts of SIV such as African green monkeys (AGMs) and Sooty mangabeys (SMs), which do not progress to AIDS [6].
While the consequences of immune activation in HIV/SIV infection are numerous and include increased numbers of activated CD4+ T-cell targets for the virus, attrition of the memory CD4+ T-cell pool and accumulation of high frequencies of terminally differentiated and exhausted memory T and B cells, the underlying mechanisms and sources of immune activation during infection are not well understood [7], [8]. Given accumulating evidence that persistent immune activation is at the heart of disease progression, understanding the mechanisms driving immune activation in chronic HIV disease will be important for the development of new adjunctive treatment strategies targeting this process. Although many factors may contribute to immune activation during chronic HIV/SIV infection, recent evidence has indicated that translocation of microbial products from the lumen of the intestine into the periphery may contribute importantly to this process [9], [10], [11], [12], [13], [14]. These microbial products can stimulate immune cells directly via pattern recognition receptors such as toll-like receptors. Indeed, immune activation related to microbial translocation occurs in other settings and has been implicated in many other pathological conditions. For example, the preconditioning chemotherapy and radiation prior to progenitor stem cell transplantation in individuals with hematological malignancies leads to damage of the tight epithelial barrier of the gastrointestinal (GI) tract resulting in microbial translocation [15]. These translocated microbial products can then stimulate the immune system, exacerbating graft versus host disease [16], [17], [18]. Microbial translocation leading to immune activation also occurs in inflammatory bowl disease [19], after invasive surgery [20], [21], and in pancreatitis [22].
While microbial translocation has been indirectly implicated in driving immune activation in chronically HIV-infected humans and SIV-infected rhesus macaques (RMs), the mechanisms underlying this phenomenon remain unclear, with enterocyte apoptosis [23], massive loss of GI tract CD4+ T cells [24] and/or preferential loss of GI tract Th17 cells [25], [26] all proposed as important contributing factors. Moreover, the timing of the onset of microbial translocation relative to infection has remained obscure, and direct evidence of translocation at the tissue level has been lacking. Here, using a quantitative image analysis approach to study large segments of tissue, we provide direct immunohistochemical evidence of translocation, define the timing of microbial translocation in pathogenic SIV infection of RMs and identify loss of the integrity of the intestinal epithelial barrier as a plausible mechanistic correlate of microbial translocation. The absence of translocation or associated immune activation in chronic SIV infection of SMs, which does not result in progressive disease, underscores the critical role this process plays in the pathogenesis of primate lentiviral infections and the potential value of limiting it as an approach to adjunctive therapy.
Initially, we sought evidence of microbial translocation by staining with a monoclonal antibody against LPS-core antigen in paraffin-embedded colon tissue sections obtained at necropsy from SIV-uninfected RMs (n = 6), RMs euthanized during early acute (n = 10) or late acute (n = 3) SIV infection, and chronically SIV-infected RMs euthanized at protocol specified endpoints (“Non-AIDS”; n = 8) or clinical endpoints (AIDS defining conditions, “AIDS”; n = 5, Table 1). Prior to and during early acute SIV infection, the LPS-core antigen-specific mAb stained only rare cells within the lamina propria (LP), but in dramatic contrast, in chronically infected animals both numerous LPS+ cells and abundant extracellular core LPS antigen were observed (Figures 1 and S1). Figures show a broad spectrum of microbial tanslocation and discontinuities to the structural barrier of the GI tract (discussed in detail below) that was seen in our animal cohort, from negligible (SIV-) to most severe (AIDS). Although the amount of apparently extracellular bacterial constituents within the LP varied among our chronically SIV-infected RMs; all chronically SIV-infected RMs showed readily demonstrable evidence of microbial translocation in multifocial lesions along the GI tract, findings that were absent in the SIV-uninfected animals and in early acute SIV infection (i.e. 1–10 dpi). Importantly, microbial translocation was evident in the large bowel of RMs chronically-infected with different pathogenic strains of virus (i.e. SIVmac239, SIVmac251 and SIVsmE660; Table 1), suggesting that intestinal damage leading to microbial translocation is a common feature of pathogenic SIV infections.
The specificity of our immunohistochemical staining directly documenting microbial translocation in gut tissue sections is supported by: (i) the remarkably low frequency of LPS+ cells within the LP in SIV-uninfected or RMs with early acute infection; (ii) the absence of staining with isotype-matched control antibodies in SIV-uninfected, acutely and chronically SIV-infected animals (data not shown); (iii) the lack of any evidence of non-cell associated LPS in the LP of our SIV-uninfected and early acute SIV-infected RMs, despite the abundant LPS staining in the residual luminal content of these same samples; (iv) the detection of microbial products within the LP of the colon of chronically SIV-infected RMs that were rarely seen in SIV-uninfected and early acute infected RMs, using a rabbit polyclonal antibody against Escherichia coli that recognizes many E. coli proteins (Figures 2 and S2), or using peptide nucleic acid fluorescent in situ hybridization to detect bacterial 16S RNA (data not shown).
To relate microbial translocation in the large bowel of chronically SIV-infected RMs to systemic microbial translocation, we stained sections from lymph nodes, identifying large numbers of LPS+ cells within both local draining lymph nodes (mesenteric; MesLN) (Figures 3 and S3), and remote peripheral lymph nodes (axillary; AxLN) (Figures 4 and S4) of chronically SIV-infected RMs, suggesting systemic dissemination of translocated microbial constituents originating from the GI tract. Consistent with what might be expected for the anatomic site of inductive T-cell immune responses, we found immunoreactive LPS within the medullary cords and paracortex of draining MesLN as well as peripheral AxLN during the chronic stage of infection. In contrast, in SIV-uninfected or early/acute SIV-infected RMs we found only low levels of LPS+ cells in the gut-draining MesLN and virtually no LPS+ cells in the peripheral AxLN of (Figures 3,4 S3 and S4). The primary localization of LPS within the medullary cords and sinuses, and to a lesser extent the paracortex and germinal centers, in SIV+ Non-AIDS RMs is consistent with microbial products traversing from the damaged intestine via the lymphatics. The presence of LPS within the paracortex and germinal centers suggests a) antigen presenting cells which have bound microbial antigens migrate into the T cell inductive site of the LN and b) possibly microbial product-immune complex deposition on follicular dendritic cells may be occurring. However, the biological relevance of finding microbial constituents in these anatomical sites, at this point, remains unclear. Moreover, consistent with the liver's important function as a “gatekeeper” between the intestine and peripheral circulation, we also found multifocal evidence of microbial products within the liver, in regions surrounding the hepatic portal veins and tracts in chronically SIV-infected Non-AIDS RMs with more extensive staining into the lobules of the liver in chronically SIV-infected AIDS RMs, consistent with increased intestinal damage correlating with more microbial dissemination (Figure 5 and Figure S5).
Using these approaches, we identified unequivocal direct evidence of microbial constituents within the LP of the large bowel, and in the liver and lymph nodes of all chronically SIV-infected RMs studied, but not uninfected RMs. This finding was consistent across all 32 RMs studied, including SIV-uninfected RMs (n = 6); early acute SIV-infected RMs (1–10 dpi, n = 10); late acute RMs (14–28 dpi infection, n = 3); and chronically SIV-infected RMs (56–397 dpi, n = 13 (Non-AIDS and AIDS); Table 1) and indicate that microbial translocation involves infiltration of microbial products into the LP of the GI tract during the chronic phase of SIV infection of RMs.
After demonstrating the qualitative presence of increased microbial products in the LP of large bowel, liver and lymph nodes of SIV infected RMs, we used quantitative image analysis techniques [27] to quantify the extent of microbial translocation demonstrable in high power (400×) digital scans, of tissue section whole mounts (ScanScope CS System, Aperio). Sections of colonic mucosa analyzed for each animal represented, on average, a total of 350 distinct 400× image fields per scanned tissue, providing an in-depth, systematic, assessment of segments of the GI tract that ranged from 43 to 101 linear mm of colonic epithelial lining and 16 to 39 mm2 of intestinal mucosal area. This comprehensive analysis approach, applied to randomly selected tissue sections, provided us with an unbiased and detailed evaluation of microbial translocation in representative sections of the GI tract that otherwise may not have been possible using conventional established tissue analysis methods. Using this approach, we found that the percent area of LP of the colonic mucosa containing LPS was significantly higher in chronically-infected RMs compared to uninfected and acutely (early) SIV-infected RMs (Figure 6A).
We used the same quantitative image analysis approach to evaluate the relationship between the microbial product burden within the LP of the colon and in draining (MesLN) and distant LN (AxLN) from the same animals, calculating the percent area of each tissue that contained LPS. We found a significant positive correlation between the amount of LPS within the LP of the colon and the amount of LPS within the corresponding draining MesLN (r = 0.69, P = 0.0065, Figure 6B). Moreover, we also found a significant positive correlation between LPS staining within the MesLN and within the matched AxLN (r = 0.59, P = 0.027, Figure 6C). Taken together, these data indicate that the presence of microbial products in the peripheral lymphatic tissues is intimately linked to microbial translocation from the gut.
Microbial products can directly stimulate the innate immune system via interactions with Toll-like receptors (TLR) that lead to an inflammatory cascade. To evaluate the possible relationship between translocation of microbial products and inflammation we performed double-label immunohistochemical staining for microbial products and the innate proinflammatory cytokine IFNα. Reflecting the immune activation of chronic SIV infection in RMs, IFNα expression was widespread and we consistently demonstrated co-localization of IFNα and microbial products within the LP of the colon (Figure S6), and in AxLN, and MesLN (data not shown). The majority of such co-localization occurred in the absence of detectable local viral replication, as during the chronic phase of infection productively infected cells within the LP are only rarely demonstrable by in situ hybridization in most SIV-infected RMs that have not progressed to AIDS (data not shown). Indeed, in extensive double label studies using immunohistochemical staining for IFNα and in situ hybridization for SIV RNA in tissues from chronically SIV-infected Non-AIDS RMs, there was only very limited co-localization between IFNα expression and in situ hybridization for SIV RNA in the colon or MesLN (data not shown). The overwhelming majority of the abundant IFNα immunostaining was found relatively distant from local viral replication, but in close proximity to microbial products (data not shown). Moreover, the levels of IFNα immunostaining were significantly greater than the tissue level of SIV RNA.
To evaluate further the role of translocated microbial products in driving immune activation in chronic SIV infection, we assessed the distribution and co-localization of interleukin-18 (IL-18) and LPS in the gut-draining MesLN in SIV-uninfected and chronically SIV-infected Non-AIDS RMs (Figure S7). IL-18 is produced by activated macrophages and dendritic cells in response to microbial product stimulation [28]. We found that before SIV infection there was a basal, low-level, expression of IL-18 in all structural compartments of MesLN (i.e. B cell follicles, T-cell zone and medullary cords) and that this expression was dramatically increased in SIV infection (data not shown). Although IL-18 was up-regulated in some regions where there were no visualized microbial products, high levels of IL-18+ cells were always found in close proximity to LPS in MesLN (Figure S7). Consistent with a role for translocated microbial products inducing immune activation and IL-18 expression, Ahmad and colleagues recently described significantly elevated levels of IL-18 in the serum of HIV-infected/AIDS patients compared to those of HIV-seronegative healthy individuals [29]. Collectively, these data strongly suggest that microbial products, which infiltrate the LP of the GI tract, and then spread systemically, can directly stimulate the immune system and contribute to chronic immune activation.
To determine how early during infection microbial translocation occurs, we compared microbial translocation into the LP of the colon of RMs throughout the acute stage of SIV infection (1 to 28 dpi) in vaginally-challenged animals. With the exception of one animal at 8 dpi, we found only very low levels of LPS+ cells within the LP of RMs between 1–10 dpi, observations that were indistinguishable from SIV-uninfected animals (Figures 6 and data not shown). There was a statistically significant increase in LPS seen within the LP of animals infected for between 14 and 28 days compared to uninfected or early/acute animals and evidence of microbial translocation was detected at small foci associated with breaks in the epithelial lining (Figures 6 and data not shown). Importantly, during the acute phase of infection, areas of discontinuity in the epithelial barrier were relatively infrequent, while LPS staining in the LP appeared to increase into the late acute stage of SIV infection (14–28 dpi). Interestingly the extent of lesions and discontinuities were lower than might have been expected relative to the massive enterocyte apoptosis previously described as peaking at 14 dpi in these same animals [23]. We comment below on the possible mechanisms responsible for this dissociation.
Although enterocyte apoptosis subsequently decreased, even as microbial translocation increased in the late acute stage of infection, the abundant evidence of LPS+ within the lamina propria at 28 dpi suggested that early damage to the integrity of the epithelial lining could facilitate translocation of microbial products (data not shown). Thus, we sought to determine directly if the compromise of the integrity of the epithelial barrier is a distinguishing feature of microbial translocation during chronic SIV infection, evaluating the integrity of the epithelial barrier by staining for the tight junction protein claudin-3. A similar technique has been used in studying human samples to assess the integrity of the epithelial barrier in diseases associated with discontinuities in the GI tract [30]. Examination of the integrity of the epithelial barrier by staining for claudin-3, revealed multifocal disruptions and epithelial loss of the normally continuous, epithelial barrier in tissues from the chronic stages of infection, but not in tissues from uninfected or early acute animals (Figure 7A and Figure S8). We recognize the possible contribution that undetected opportunistic enteric pathogens may play in the continuum of epithelial damage seen in our chronically SIV-infected (AIDS) RMs, and thus show in Figures 7 and S8 two examples of end-stage RMs demonstrating this dynamic spectrum of epithelial damage from multifocal epithelial loss to severe epithelial damage and ulceration. Importantly, confocal analysis showed that disruptions in the integrity of the epithelial barrier were directly associated with translocated microbial products (Figure 7B). Furthermore, quantitative image analysis from high power entire colonic tissue section scans, confirmed that chronically-infected animals had significantly more damage to the integrity of the epithelial barrier compared to uninfected and acutely-infected RMs (Figure 8A). Moreover, the degree to which the integrity of the epithelial barrier was compromised was significantly correlated with the amount of LPS within the LP (Figure 8B; r = 0.57, P = 0.032).
Loss of integrity of the epithelial barrier of the GI tract would be expected to result in several host responses including polymorphonuclear neutrophil (PMN) infiltration and increased local proliferation of enterocytes within colonic crypts in an attempt to restore the integrity of the GI tract. Importantly, we observed increased levels of myeloperoxidase+ PMNs within the lamina propria of chronically SIV-infected RMs associated with damage to the epithelial barrier (Figure S9), providing strong evidence for a tissue specific response to GI epithelial damage. To assess proliferation of GI tract enterocytes we immunohistochemically stained colon tissues with a monoclonal antibody against Ki67, a cellular marker for proliferation, and performed quantitative image analysis measuring the fraction of enterocytes along the colonic crypt that were proliferating (Ki67+). We found increased levels of proliferating enterocytes (Ki67+) in chronically SIV-infected animals compared to SIV uninfected and acutely infected RMs (Figure 9A–B and Figure S10, P = 0.016). Moreover, there was a trend towards an increase of Ki67+ enterocytes in animals between 14 and 28 days post SIV infection compared to animals infected for ∼1 week (Figure 9B, P = 0.057). These data are consistent with early (14–28 dpi) and progressive damage to the integrity of the epithelial barrier, and indicate that one mechanism underlying microbial translocation likely involves breakdown of the structural barrier of the GI tract at a rate that exceeds enterocyte proliferation and other repair mechanisms, consistent with previous reports of abnormalities within the GI tracts of chronically HIV or SIV-infected individuals [31], [32], [33], [34], [35], [36], [37], [38]. Our findings of host responses to microbial-product infiltration into the lamina propria are consistent with previous findings of fibrosis within the lamina propria of the GI tract [39]. While our data and those of previous studies are consistent with increased discontinuity of the structural barrier of the GI tract during chronic SIV infection of RMs, we cannot exclude that the structural damage we observed by immunohistochemical analysis may be attributed to increased enterocyte turnover overall, leading to an apparent increase in structural damage to the GI tract. However, the increased levels of myeloperoxidase+ PMNs within the lamina propria of chronically SIV-infected RMs associated with observed damage to the epithelial barrier (Figure S9), strongly support our conclusion of GI epithelial damage. Regardless, these data suggest that the integrity of the structural barrier of the GI tract is significantly weakened in SIV-infected individuals leading to microbial translocation.
When microbial products cross the epithelial barrier under normal, physiological conditions, they are generally phagocytosed by specialized intestinal macrophages [40]. The relatively abundant and apparently non-cell associated microbial constituents we saw in the LP of our chronically SIV-infected RMs, particularly in animals with AIDS, suggested that this might result from microbial translocation in excess of the phagocytic capacity of these macrophages. This could reflect a saturation of the maximum capacity of these macrophages to phagocytose translocated microbial constituents or alternatively, might reflect a compromise of macrophage phagocytic function resulting in a net relative defect in microbial clearance and the extracellular accumulation of microbial products. To assess this, we performed confocal microscopy of GI tract tissue from SIV-uninfected and acute and chronically SIV-infected RMs to assess the localization of microbial constituents relative to GI tract macrophages (intracellular vs. extracellular). In the rare instances where microbial products were detected in the LP of SIV-uninfected RMs, they were virtually always within HAM56+ macrophages (Figure 10 and Figure S11). In addition, during the acute phase of infection (until 28 dpi), most microbial products were mostly found within HAM56+ macrophages (Figure S11), perhaps helping to explain why LPS levels in plasma are not elevated during acute infection while sCD14 levels were moderately raised [9]. Moreover, microbial products crossing the epithelial barrier at 14 dpi were virtually always found within macrophages, whereas abundant macrophages were juxtaposed to damaged regions of the colon in late acute (28 dpi) (Figure S11). Furthermore, as infection progressed into the chronic stage of disease, the frequency of macrophages negative for microbial products increased, even though abundant numbers of macrophages were present and adjacent to microbial products (Figure 10 and Figure S11). There was no apparent change in the overall frequency of HAM56+ macrophages at the tips of the colonic crypts between SIV-uninfected, and acutely or chronically SIV-infected RMs (data not shown). The presence of high frequencies of macrophages without internalized microbial constituents, along with abundant extracellular microbial products suggested that GI tract macrophages in the later phase of acute infection and chronic stages of disease may become increasingly incapable of phagocytosing microbial products that translocate into the LP.
A distinguishing feature of non-progressive infection in natural hosts of SIV is the absence of immune activation during the chronic phase of infection [6], [41], [42], [43], [44], [45]. Because elevated plasma LPS levels are absent in both chronically SIVsmm-infected SMs and SIVagm-infected AGMs [9], [45], [46], we evaluated the structural integrity of the intestinal epithelial barrier in chronically SIVsmm-infected SMs. In contrast to findings in chronically SIV-infected RMs, and consistent with the lack of LPS within the circulation of these SIV natural host animals [9], we found no evidence of damage to the integrity of the epithelial barrier (Figure 11A) and no infiltration of microbial products into the LP of large bowel (Figure 11B) or peripheral lymph nodes (Figure 11C). These data were consistent among 7 animals studied (n = 2, SIVsmm-uninfected SMs; n = 5, chronically SIVsmm-infected SMs; Table 1). Hence preservation of the tight epithelial barrier is associated with lack of microbial translocation and immune activation in non-progressive, natural SIV infection.
Indirect evidence has implicated microbial translocation from the gut as a factor contributing to pathological immune activation in chronic HIV/SIV infection, but direct evidence of translocation and demonstration of a plausible underlying mechanism have been lacking. Using an unbiased, comprehensive approach for quantitative and qualitative immunohistologic analysis of randomly selected tissue specimens obtained from non-human primates at various times relative to SIV infection, we have shown that: 1) microbial products can be found in the LP of the large bowel, in draining and distant lymph nodes, and in the liver of chronically SIV-infected RMs; 2) microbial translocation is associated with breakdown of the integrity of the epithelial barrier of SIV-infected RMs; 3) the extent of epithelial breakdown correlates with the extent of microbial translocation; 4) epithelial barrier breakdown and microbial translocation begin to be apparent during the late acute phase of infection (14 dpi); 5) the presence of microbial products in multiple anatomical sites is associated with expression of IFN-α and IL-18 in the absence of detectable local viral replication in the LP, consistent with direct induction of immune activation; 6) macrophages in chronically SIV-infected RMs appear dysfunctional with respect to their ability to phagocytose translocated microbial products; and 7) neither epithelial barrier breakdown nor infiltration of microbial products into the LP occur during the chronic phase of SIV infection of SMs.
We provide two lines of evidence linking microbial translocation to immune activation. First, we show that in pathogenic SIV infection of RMs, damage to the integrity of the epithelial barrier of the GI tract is associated with microbial translocation, and that microbial translocation is linked to local immune activation, based on co-localization of microbial products and production of the immunoinflammatory cytokines IFNα and IL-18, including in lymph nodes anatomically distant from the GI tract. Second, in marked contrast, in SIV-infected SMs where immune activation is quickly resolved in the acute stage of infection [47], and chronic infection is not accompanied by persistent immune activation, we found neither damage to the intestinal barrier nor microbial translocation.
Recent in vitro studies with peripheral blood lymphocytes from SMs have been interpreted to suggest that a lack of type I IFN cytokine response to SIV RNA accounts for the typically nonprogressive nature of SIV infection in SMs [48]. Furthermore, these data have been used to suggest that the raised plasma LPS levels observed in chronically SIV-infected RMs and HIV-infected humans, are simply markers of damage to the GI tract and that the microbial translocation that is reflected in elevated plasma LPS levels does not contribute significantly to causing systemic immune activation [48]. However not only were LPS levels increased in chronically-infected individuals, but sCD14 and LPS binding protein levels were also increased [9], [11], [13]. These data strongly suggested that LPS was directly stimulating the immune system in vivo. In the present study, we show directly that damage to the integrity of the epithelial barrier of the GI tract allows microbial products to infiltrate into the LP and this infiltration is associated with local immune activation demonstrated by co-localization of microbial products within the LP with IFNα and IL-18 in MesLN. As only limited viral replication is demonstrable in the LP of the GI tract of chronically SIV-infected RMs, the damage to the GI epithelial lining, microbial translocation and local immune activation are unlikely to be caused by the direct effects of local viral replication. Rather, where rare infected cells are seen in the LP underlying damaged mucosa, it is more likely that the chronic immune activation, due to translocated microbial products, has provided activated CD4+ T cell targets for the virus.
Taken together, these data suggest that microbial translocation, resulting from damage to the GI tract epithelial barrier and impaired macrophage-mediated phagocytosis, results in immune activation during the chronic phase of HIV/SIV infection of humans and RMs. Importantly, we found a significant correlation between the extent of damage to the epithelial barrier of the colon and the amount of LPS within the underlying mucosa and the extent of translocated microbial products in draining and remote lymph node tissues. The extent of microbial constituents present in lymph node tissues correlated with the extent of local evidence of immune activation, further substantiating the link. Moreover, while we find that microbial translocation begins during the acute phase of infection, our previous work had indicated that elevated levels of microbial products were not seen in plasma until the chronic phase [9]. Our data suggest that microbial products are localized within tissue macrophages during the acute phase thus limiting their circulation.
The causes of damage to the integrity of the epithelial barrier of the GI tract are likely to be multifaceted, but in the chronic stages of SIV infection seem unlikely to be due to direct virotoxic effects, given the lack of association with very low levels of demonstrable local viral replication in the LP relative to the extensive epithelial damage. One possible mechanism may be related to the preferential loss of Th17 cells in the GI tract in progressive immunodeficiency lentiviral infections [25], [26], because Th17 cells produce cytokines important for enterocyte proliferation and antibacterial defensins [49], [50], [51] and IL-17 has recently been shown to suppress Th1-mediated damage to gut epithelium. Importantly, preservation of this T cell subset in the gut of chronically SIVsmm-infected SMs and SIVagm-infected AGMs [26], [52] and the sparing of the epithelial barrier of SIVsmm infected SMs we show here supports this mechanism.
We speculate that the association we found between immune activation, microbial translocation and chronic stages of SIV infection, and similarly, later stages of HIV-1 infection, reflects damage to the structural integrity of the GI tract and a potential “deficiency” of the GI tract macrophage-phagocytic system. Our observation that intestinal macrophages from SIV-infected RMs, which are generally not proinflammatory [40], are unable to clear translocated microbial products, within the LP, and could lead to increased proinflammatory responses locally are supported by several groups findings that showed: i) impaired monocyte phagocytosis in HIV-infected individuals [53]; ii) reduced LPS-mediated enhancement of phagocytosis in monocytes HIV-infected individuals compared to healthy donors [54]; and iii) significantly higher colonic mucosa proinflammatory mRNA expression levels (e.g. TNF-α, IFN-γ, and IL-6) in HIV-infected patients than in control patients [55]. These data certainly warrant further investigation into the functional properties of tissue macrophages from HIV/SIV-infected individuals and the mechanisms underlying their apparent dysfunction. While increased microbial translocation begins in the late acute stage of SIV infection, it was not until the later stages of infection that the capacity of macrophages for clearance was apparently affected, suggesting that microbial translocation has an increasing major contribution to immune activation as the host progresses towards disease. The evidence for this model comes from images of MesLN and AxLN stained for bacterial products in the chronic stage that showed dramatically increased extracellular bacterial constituents in the late AIDS stage of SIV infection, versus the mainly cellular staining of LPS at earlier stages. Taken together, these data strongly suggest that in SIV infection of RMs, and by extension, HIV infection of humans, damage to the epithelial barrier of the GI tract leads to levels of microbial translocation that exceed the capacity of host defense mechanisms to sequester away microbial constituents from secondary lymphatic tissues, resulting in persistent immune activation that contributes importantly to pathogenesis during the chronic phase of infection. Understanding the factors underlying damage to the integrity of the epithelial barrier and macrophage deficiencies that we report may lead to novel therapeutic interventions that aim to reduce microbial translocation and the deleterious effects of the consequent immune activation.
To characterize the extent of microbial translocation in the gastrointestinal tract in SIV infection, we studied tissues from an assembled cohort of SIV-uninfected and infected RMs and SIVsmm-infected and uninfected SMs originally involved in separate studies (summarized in Table 1). Tissues (colon and LNs) were obtained at necropsy from 13 rhesus macaques (Macaca mulatta) of Indian origin euthanized 1–28 days after atraumatic intravaginal infection with SIVmac251 or SIVmac239 as described elsewhere [56]. Six additional SIV-negative RMs were used as controls. In a separate study, tissues were obtained at necropsy from 13 adult RMs chronically infected with SIVmac239 or SIVsmE660 that were sacrificed either because of end-stage disease (AIDS; defined by opportunistic infections, lymphomas, or a diagnosis of wasting, based on greater than 15% body mass weight loss, n = 5) or protocol specified experimental end point (Non-AIDS; n = 8). For immunohistochemistry studies, samples were very quickly processed into fixative to avoid potential artifacts associated with post-mortem tissue changes. The post mortem interval, the time from euthanasia until collection of lymph nodes and GI tract segments were placed into fixative, ranged from ∼10–30 minutes and was consistent over four independent primate facilities contributing tissues to the present study. The GI tract segments sampled at necropsy were representative and were not selected with regard to any visually apparent lesions or other pathology. In a third study, LNs and rectal biopsies were obtained from 2 SIVsmm-negative SMs (Cercocebus atys) and 5 SMs that were naturally infected with SIVsmm as previously described [47]. Animals were housed and cared in accordance with American Association for Accreditation of Laboratory Animal Care standards in AAALAC accredited facilties, and all animal procedures were performed according to protocols approved by the Institutional Animal Care and Use Committees of the National Cancer Institute, California National Primate Research Center or Yerkes National Primate Research Center (Table 1). Unfortunately, paraffin-embedded tissue from colon, mesLN, and axLN samples were not available from all animals.
Plasma samples were analyzed for SIV vRNA by using a quantitative branched DNA (bDNA) assay [53] or using a fluorescent resonance energy transfer probe-based real-time RT-PCR (TaqMan) assay that provides a threshold sensitivity of 125 copy Eq/ml, as previously described [57]. All PCR reactions were run on ABI Prism 7700 Sequence Detection System and the fluorescent signal-based quantitation of viral RNA copy numbers in test samples was determined by ABI sequence detection software (Applied Biosystems, Foster City, CA).
Immunohistochemical staining and SIV in situ hybridization were performed as previously described [58]. In brief, unselected specimens of tissues of interest were obtained at necropsy, fixed, and paraffin embedded. Immunohistochemistry was performed using a biotin-free polymer approach (MACH-3; Biocare Medical) on 5-µm tissue sections mounted on glass slides, which were dewaxed and rehydrated with double-distilled water. Antigen retrieval was performed by heating sections in 1× DIVA Decloaker reagent (Biocare Medical) in a pressure cooker (Biocare Medical) followed by cooling to room temperature. All slides were stained using the intelliPATH FLX autostaining system (Biocare Medical) according to experimentally determined optimal conditions. This included blocking tissues with Blocking Reagent (Biocare Medical) for 10 min followed by an additional blocking step with TNB (0.1 M Tris-HCL (pH 7.5), 0.15 M NaCl, and 0.5% Blocking Reagent (NEN)) containing 2% Blocking Reagent and 100 µg/mL goat ChromePure IgG (Jackson Immunoresearch) for 10 minutes at room temperature. Endogenous peroxidase was blocked with 1.5% (v/v) H2O2 in TBS (pH 7.4). Primary antibodies were diluted in TNB containing 2% Blocking Reagent and 100 µg/mL goat ChromePure IgG for 1 h at room temperature. Mouse or rabbit MACH-3 secondary polymer systems (Biocare Medical) were applied for 20 minutes each. Double immunohistochemical staining was performed on colon and lymph node sections with either mouse monoclonal anti-LPS-core and rabbit polyclonal anti-IL18 antibodies or mouse monoclonal anti-IFNα and rabbit polyclonal anti-E.coli using the MACH-2 multiplex staining system (Biocare Medical) according to manufacturer's instructions. Sections were developed with ImmPACT DAB (Vector Laboratories) and/or Vulcan Fast Red chromogen (Biocare Medical), counterstained with hematoxylin, and mounted in Permount (Fisher Scientific).
All stained slides were scanned at high magnification (400×) using the ScanScope CS System (Aperio Technologies, Inc.) yielding high-resolution data from the entire tissue section. Bacterial PNA FISH was performed using the universal bacterial 16 s ribosomal RNA specific (UniBac) FITC conjugated PNA probe (AdvanDx, Inc.) according to manufacturer's instructions, with the exception that a heat induced epitope retrieval pretreatment step was performed in 1× Diva retrieval buffer (Biocare Medical) for 20 minutes in a 95°C water bath prior to hybridization. FISH samples were examined and imaged using a Nikon 80i upright fluorescent microscope (Nikon Instruments, Inc) equipped with a BrightLine multiband bandpass FITC/Texas Red filter (Semrock; data not shown). Primary antibodies used were: mouse anti-human Interferon-α (clone MMHA-2; PBL InterferonSource), mouse anti-LPS core (clone WN1 222-5; Hycult or provided by Dr. Robin Barclay), mouse anti-macrophage (clone HAM56; Dako), mouse anti-cytokeratin (clone MNF116; Dako), polyclonal rabbit anti-E. coli (Dako), polyclonal rabbit anti-IL-18 (Sigma Prestige Antibodies by Atlas Antibodies) and polyclonal rabbit anti-Claudin-3 (Labvision).
Immunofluorescent confocal microscopy was performed on treated slides as above, but stained overnight with primary antibodies at 4°C, washed, stained with fluorescently conjugated secondary antibodies for 1 h in the dark, counterstained with DAPI (Molecular Probes), mounted in AquaPoly mount (Polysciences, Inc.) and imaged using a Olympus FluoView FV1000. Z-stack images were taken for each high power field that spanned the entire 5 µm tissue section and representative 3-D projections from z-stack images were generated using Imaris 7.0.0 software (Bitplane Inc.). Secondary antibodies used were donkey anti-mouse IgG Alexa Fluor 488 and donkey anti-rabbit IgG Alexa Fluor 555 (all from Molecular Probes).
To quantify microbial translocation (LPS) into the LP and the extent of epithelial barrier damage (claudin-3), 5-µm thick sections were cut from paraffin blocks of unselected tissue sections obtained at necropsy and stained with either monoclonal antibody against LPS-core or polyclonal antibody for claudin-3 and counter stained with heamotoxalin. High power (400×) whole tissue scans were obtained using an Aperio ScanScope as described above and imported into Photoshop CS3 (Adobe Systems Inc., Mountain View, California, USA). Images were manually trimmed to remove the submucosae, muscularis and residual luminal content, leaving only the LP mucosae to analyze. The percent area of the LP staining for LPS was determined essentially as previously described using Photoshop CS3 tools with plug-ins from Reindeer Graphics [39], [47]. The percent area of LN staining for LPS was determined from whole LN scans as above but without the need to trim the image. The proportion of the epithelial barrier that was damaged during SIV infection was first determined by manually tracing (in red) the area of the lumen/GI epithelial tract interface that had no claudin-3 staining epithelial cells, using the brush tool in Photoshop CS3. The remaining claudin-3 staining intact epithelial cell regions were then manually traced (in black). The percent damage was calculated by determining the proportion of the image that was red (lack of claudin-3 stain) compared to the total epithelial surface area (red+black) using plug-in tools from Reindeer Graphics.
Spearman's rank correlation and Mann-Whitney tests were performed using Prism 4.0 software (Prism, San Diego, CA).
|
10.1371/journal.pgen.1003535 | Juvenile Hormone and Insulin Regulate Trehalose Homeostasis in the Red Flour Beetle, Tribolium castaneum | Insulin/IGF-1 signaling (IIS) has been well studied for its role in the control of life span extension and resistance to a variety of stresses. The Drosophila melanogaster insulin-like receptor (InR) mutant showed extended life span due to reduced juvenile hormone (JH) levels. However, little is known about the mechanism of cross talk between IIS and JH in regulation of life span extension and resistance to starvation. In the current study, we investigated the role of IIS and JH signaling in regulation of resistance to starvation. Reduction in JH biosynthesis, JH action, or insulin-like peptide 2 (ILP2) syntheses by RNA interference (RNAi)-aided knockdown in the expression of genes coding for juvenile hormone acid methyltransferase (JHAMT), methoprene-tolerant (Met), or ILP2 respectively decreased lipid and carbohydrate metabolism and extended the survival of starved beetles. Interestingly, the extension of life span could be restored by injection of bovine insulin into JHAMT RNAi beetles but not by application of JH III to ILP2 RNAi beetles. These data suggest that JH controls starvation resistance by regulating synthesis of ILP2. More importantly, JH regulates trehalose homeostasis, including trehalose transport and metabolism, and controls utilization of stored nutrients in starved adults.
| Both juvenile hormone (JH) and Insulin/IGF-1 signaling (IIS) regulate life span and starvation resistance in insects. Regulation of longevity and starvation resistance by IIS has been well studied, yet little is known about the underlying mechanisms and cross talk between these two hormones. The red flour beetle, Tribolium castaneum, is a good model to study cross talk between JH and IIS because both of these pathways are important in regulation of life span and starvation resistance. The starved male beetles with either reduced JH or ILP2 levels live longer due to a lower rate in lipid and carbohydrate metabolism when compared with the control beetles. Juvenile hormone regulates starvation survival through regulation of synthesis of ILP2, trehalose transporter (TRET), and trehalase. Reduction in JH levels or its action or ILP2 expression decreased trehalase levels in the fat body, resulting in a slower rate of conversion of trehalose to glucose. Reduction in JH levels or its action also caused a decrease in TRET levels in the alimentary canal leading to a lower rate of uptake of trehalose into this tissue resulting in more trehalose available in the hemolymph. Trehalose likely regulates various processes to protect beetles from stress.
| Many biological functions of juvenile hormone (JH) in regulation of almost every aspect of an insect's life have been reported since its discovery in 1965 [1], [2]. To maintain the larval state, JH induces the expression of the genes coding for transcription factors such as Kr-h1 to prevent metamorphosis; knockdown in the expression of the gene coding for Kr-h1 by RNAi in larvae leads to precocious metamorphosis that cannot be rescued by exogenous JH application [3]. During the larval stage, JH suppresses imaginal disc growth promoted by nutrition [4], and the nutritional signals mediated by insulin/IGF signaling (IIS) can override JH suppression [5]; but, in the absence of JH, the wing disc grows despite severe starvation. Interestingly, the wing disc growth is well correlated with trehalose levels during the larval stage until the critical weight is reached; starvation causes a decline in hemolymph glucose and trehalose and cessation of wing imaginal disk growth, which can be rescued by injection of trehalose. After reaching the critical weight, the trehalose response to starvation disappears and the action of insulin becomes decoupled from nutrition. The wing disks also lose their sensitivity to repression by JH [6].
To direct reproductive maturation in Drosophila melanogaster and Tribolium castaneum, JH regulates the production of male accessory gland proteins in the male and vitellogenin (Vg) in the female [7]–[9]. A basic helix-loop-helix (bHLH) per-Arnt-Sim (PAS) family transcription factor, methoprene-tolerant (Met) interacts with other members of this family including steroid receptor co-activator (SRC) and Cycle; binds to both JH and JH response elements (JHRE) present in the promoters of JH-response genes [10]–[15].
The function of JH has been well studied in the regulation of molting, metamorphosis, and reproduction. However, mechanisms of JH action in regulation of life span and starvation resistance are still unclear. In the monarch butterfly, migrant adults live longer than summer adults when both are maintained under standard laboratory conditions. Interestingly, the longevity of migrant adults is restored to that of summer adults by treatment with JH I, and the life span of summer adults is increased by 100% when the corpora allata are surgically removed [16]. Similarly, in the InR mutant of D. melanogaster, life span extension is due to reduced JH levels [17]. These studies showed that lower levels of JH could extend the life span under certain conditions. However, the underlying mechanisms of JH action on the longevity and the cross talk between JH and IIS pathway are still not well understood.
The IIS function in life span, longevity, and stress resistance has been thoroughly investigated because of evolutionarily conserved function from yeast to mammals [18]. These functions include regulation of cellular adaptation to stress stimuli, such as nutrient-poor conditions [19] and oxidative stress [20], [21], promoting autophagy [22] and regulation of metabolism [23].
T. castaneum is a good model for these studies because of efficient functioning of RNAi and rapid JH response. In the previous studies, we showed that JH regulates ILP2 and ILP3 synthesis; ILP2 and ILP3 in turn regulate Vg synthesis [9]. These studies provided a good model to explore the interplay between JH and IIS signaling pathways. Here, we investigated the effects of JH and IIS on survival and carbohydrate metabolism in adults under starvation. RNAi, topical application of JH III, and injection of bovine insulin were used to modify JH and/or insulin levels in the adults of T. castaneum to study the cross-talk between JH and IIS signaling in regulation of resistance to starvation.
To determine the role of JH in the survival of the starved T. castaneum, the newly emerged male adults were injected with malE (dsRNA prepared using a bacterial gene malE as a control), JHAMT (a key enzyme in JH synthesis), and Met (JH receptor). The control-starved beetles began to die on seventh day post-adult emergence (PAE), and all beetles died by the fourteenth day PAE. However, a block in JH synthesis or its action by knockdown in the expression of genes coding for JHAMT (mean survival 12.8 days) or Met (mean survival 12.7 days) extended survival of the starved beetles by one day (control mean survival 11.7 days, P = 0.00002 in log rank test, Fig. 1A).
To determine whether or not ILPs are involved in regulation of starvation survival, ILP1, ILP2, ILP3, and ILP4 dsRNA were injected into newly emerged adults, and the survival of RNAi beetles was monitored under starvation conditions. All four dsRNAs caused more than 80% reduction in their target mRNA levels (Fig. S1). As shown in Figure 1B, only ILP2 knockdown extended life span (mean survival 12.9 days) similar to that in JHAMT or Met RNAi beetles when compared with the control (12.1 days mean survival, P = 3.39E-06 in log rank test). While ILP1 knockdown shortened the survival, ILP3 and ILP4 knockdown did not show any significant effect. In addition, injection of bovine insulin decreased survival of JHAMT (12.9 days mean survival) and ILP2 RNAi beetles (12.6 days mean survival) to that in control insects (12.1 days mean survival, P = 0.005, Fig. 1C). The application of JH III decreased the survival of JHAMT RNAi beetles (from 11.6 to 11.0 days mean survival, P = 0.038), but not ILP2 RNAi beetles (11.5 days mean survival for both, P = 0.467). (Fig. 1D). These data suggest that both IIS and JH may work through similar or overlapping mechanisms to regulate survival of starved adults and that JH may work upstream to the IIS pathway.
To determine the major energy source for starved beetles, the total lipid, carbohydrate, and protein levels were measured in fed and starved beetles. In the fed beetles, the levels of all three nutrients did not change significantly during days 3–8 PAE (Fig. 2A). In contrast, in the starved animals, the levels of all three nutrients gradually decreased from day 3 to day 8 PAE (Fig. 2A). These data suggest that the beetles use all three sources of nutrients during starvation. To determine whether JH or IIS regulate metabolism of these macromolecules, the levels of these macromolecules were determined in JHAMT or ILP2 RNAi beetles. Strikingly, higher protein, carbohydrate, and lipid levels were detected in JHAMT and ILP2 RNAi beetles when compared to the levels in the control beetles injected with malE dsRNA (Fig. 2B). These data suggest that the life span extension during starvation in either JH or IIS deficient animals could be due to the reduced metabolism.
To determine whether the trehalose, a major sugar in most insects, or the glucose, a major sugar in most animals, is utilized during starvation, trehalose or glucose were fed to the starved beetles. When beetles were fed on non-nutritional cellulose diet or cellulose diet supplemented with 10% trehalose or 10% glucose, the beetles fed on a trehalose-supplemented diet lived significantly longer when compared to the other two groups (P = 0.001, Fig. 3A). There was no significant difference in the survival of cellulose-fed or cellulose+10% glucose-fed beetles (Fig. 3A). These data suggest that major insect sugar trehalose is important for survival of starved beetles. Moreover, the ratio of glucose and trehalose in the hemolymph increased in the control beetles upon starvation from day 4 to day 6, suggesting more glucose is needed during starvation for the energy supply. However, this ratio decreased by 77–81%, 37–93%, and 70–89% in starved ILP2, JHAMT, or Met RNAi beetles respectively, when compared to the levels in control beetles (Fig. 3B). These data suggest that JH and ILP2 regulate trehalose levels in starved beetles.
Trehalose homeostasis is controlled by trehalose-6-phosphate synthase (TPS), the main enzyme involved in the synthesis of trehalose in the fat body [24]; Trehalose transporter (TRET), the direction of transport depends on the concentration gradient of trehalose [25]; and the trahalase, the major enzyme involved in conversion of trehalose to glucose in various insect tissues [26]. To determine the relative contribution of TRET, TPS, and trehalase in extending life span in starved beetles, we identified genes coding for trehalase (G04791), TRET (G13653), and TPS (G07883) based on sequence similarity with their homologs in other insects. These genes are highly conserved among insects (Fig. S2). We injected trehalase, TRET or TPS dsRNA into newly emerged beetles. The dsRNA injected beetles were starved for 13 days and life span changes were monitored. The TRET RNAi beetles showed a slight but significant increase by 0.26 day of mean survival in life span when compared to the control beetles (P = 0.042, Fig. 3C). Trehalase RNAi beetles showed no differences from the control, and the TPS RNAi beetles showed a decrease by 0.21 day of mean survival in life span when compared to the control beetles (P = 0.05, Fig. 3C). Similarly, the ratio between glucose and trehalose decreased by 44–66% in TRET RNAi beetles and increased by 1.2–3.9-fold in TPS RNAi beetles during starvation (Fig. 3B). RNAi studies showed that the mRNA levels of TRET and trehalase, but not TPS, decreased in beetles injected with JHAMT dsRNA, suggesting that JH is required for expression of genes coding for TRET and trehalase during starvation (Fig. 3D).
Studies on expression of genes coding for TPS, TRET, and trehalase in male adults showed that gene coding for TPS is predominantly expressed in the testis, gene coding for TRET is predominantly expressed in the alimentary canal, and gene coding for trehalase is expressed in the fat body, head, and alimentary canal (Fig. 4A). Comparison of TPS, TRET, and trehalase mRNA levels in starved and fed adults showed that TRET mRNA levels are higher in the starved beetles when compared to their levels in fed beetles. In contrast, the TPS mRNA levels are higher in the fed beetles than in the starved beetles. However, trehalase mRNA levels did not vary between starved and fed beetles (Fig. 4B).
TRET mRNA levels decreased in JHAMT and Met RNAi insects when compared to their levels in control insects in the alimentary canal but not in the fat body or head of starved insects, suggesting that JH regulates the expression of this gene in the alimentary canal (Fig. 5A). Moreover, topical application of JH III induced the expression of the gene coding for TRET in the alimentary canal but not in the fat body or head (Fig. 5B). Similarly, the mRNA levels of trehalase in the fat body decreased in JHAMT, Met, and ILP2 RNAi insects (Fig. 5C). A decrease in mRNA levels of trehalase was observed in the alimentary canal isolated from JHAMT and Met RNAi beetles (Fig. 5C). Similarly, head tissue dissected from JHAMT, ILP2, and Met RNAi insects showed a decrease in trehalase mRNA levels (Fig. 5C). Topical application of JH III induced the gene coding for trehalase in the fat body but not in the alimentary canal or head (Fig. 5D). Injection of insulin into starved males on day 5 induced trehalase gene expression by 2.2 and 1.9-fold in the fat body and head respectively when compared to the levels in the same tissues dissected from control beetles (Fig. 5D). These data suggest that both JH and insulin regulate expression of the gene coding for trehalase and JH but not insulin regulates expression of the gene coding for TRET.
The first major contribution of the current study is the discovery that JH and ILP2 regulate trehalose homeostasis in starved beetles. RNAi-aided knockdown in the expression of genes coding for JHAMT (a key enzyme in JH synthesis) and Met (JH receptor) or ILP2 extended the survival of starved beetles (Fig. 1&S4). Injection of bovine insulin rescued the effects of both JHAMT and ILP2 RNAi on starvation survival. In contrast, topical application of JH III restored starvation resistance to the control level in starved JHAMT RNAi adults, but not in the ILP2 RNAi adults (Fig. 1C & D). RNAi-aided knockdown in the expression of genes coding for JHAMT or Met in male adult beetles caused a decrease in expression of ILP2 suggesting that both JH and its receptor are required for expression of this gene (Fig. S3). In addition, topical application of JH III induces expression of ILP2 in male beetles (Fig. S3). Moreover, JH titer could have been higher in starved beetles than the titers in the fed beetles as suggested by both JHAMT and Kr-h1 mRNA levels (Fig. S5). Taken together, these data suggest that JH regulates starvation resistance at least partially working through ILP2. Similar results on the role of JH in starvation resistance and extending life span have been reported in the burying beetles including Nicrophorus orbicollis, N. tomentosus, and Ptomascopus morio [27]; in D. melanogaster [17], [28]; and in the monarch butterfly [16]. In both T. castaneum [9] and Apis mellifera [29], JH induces expression of ILPs. In T. castaneum JH induces expression of ILP2 and ILP3 in females and regulates expression of Vg genes through insulin pathway. In A. mellifera, JH works through ILP1 and regulates carbohydrate metabolism when worker bees shift from nursing to foraging. The conserved roles of IIS pathway have been well studied in regulation of life span and reproduction from yeast to mammals [18], [29], [30]. In D. melanogaster, partial ablation of the insulin producing cells, the median neurosecretory cells in the brain, has extended life span, reduced fecundity, altered lipid and carbohydrate metabolism and increased oxidative stress resistance [31], [32]. These previous studies and our data reported in this paper suggest that JH regulates metabolic and reproductive processes at least partially through IIS signaling pathway.
How does JH regulate carbohydrate metabolism? In T. castaneum males, JH regulates expression of genes coding for trehalase and TRET, the two proteins critical for trehalose metabolism and transport (Fig. 5). Knockdown in expression of genes coding for JHAMT, Met, or ILP2 in the starved T. castaneum caused a decrease in trehalase mRNA levels in the fat body, which is the major tissue for storage of nutrients (Figs. 3 & 5). This would have caused a decrease in metabolism of trehalose to glucose, resulting in an increase in trehalose and decrease in glucose in the hemolymph. In T. castaneum, JH regulates expression of the gene coding for trehalase through the ILP2 and IIS pathway.
Interestingly, JH but not ILP2 regulates expression of the gene coding for TRET in T. castaneum, suggesting that JH may recruit Met to bind the promoter region of the gene coding for TRET, which is a different mechanism from that described for trehalase regulation (Fig. 5). Studies are in progress to test this hypothesis. Our data suggest that IIS is not involved in transcriptional regulation of the gene coding for TRET. However, it is possible that IIS may regulate translocation of TRET protein similar to insulin regulation of glucose transporter 4 (GLUT4) in humans by stimulating translocation of GLUT4 to the plasma membrane [33]–[35]. In type II diabetes patients, expression levels of the gene coding for GLUT4 and its translocation influence glucose transport [36], [37].
Insulin regulation of trehalose levels has been reported in Caenorhabditis elegans, D. melanogaster, and Bombyx mori [31], [38]–[41]. Insulin signaling regulates trehalose homeostasis by controlling expression of the gene coding for trehalase in the silkworm B. mori [42], [43] by a direct molecular interaction with trehalase in Tenebrio molitor [26] and by regulating the trehalose synthesis in C. elegans [39]. However, in the starved male T. castaneum, knockdown in the expression of the gene coding for either ILP2 or JHAMT did not affect TPS mRNA levels, suggesting that TPS is not under the control of JH or IIS in starved male beetles. It is possible that trehalose synthesis, an energy consuming process, is not active during starvation.
The second major contribution of the current studies is the discovery that trehalose plays an important role in starvation resistance in T. castaneum. Feeding trehalose but not glucose extended the starvation survival, suggesting that trehalose plays an important role in starvation resistance in addition to being an energy source. Trehalose alters the life span as shown in both IIS-reduced C. elegans and JH-deficient fruit fly [38], [39]. In addition to the main function as an energy source [24], [44], trehalose could be acting as a chemical chaperone or as a metabolism modifier in protection of beetles from death.
Insect hemolymph as a “sink” or “reserve” carries a variety of metabolites [45]–[47]. The hemolymph composition of metabolites reflects nutrient intake and serves as a feedback signal to regulate food intake [48] and the rate of energy expenditure [49]. It is also possible that the trehalose distributed to the tissues and organs could help protect cells against heat, cold, desiccation, anoxia, and oxidation and retard age-associated decline in survivorship and extend life span [50]. In addition, trehalose induces autophagy, independent of TOR, clears the aggregate-prone proteins associated with Parkinson's [51] and Huntington's [52] diseases.
The trehalose homeostasis including trehalose levels and trehalose distribution play important roles in starvation resistance. Here, we found that both IIS and JH signaling pathways are involved in controlling starvation resistance via regulating trehalose homeostasis. The detailed molecular mechanisms that govern the cross-talk between these two major signaling pathways in regulation of trehalose homeostasis are the focus of intense research in several laboratories around the world.
Strain GA-1 of T. castaneum was reared on organic wheat flour containing 10% yeast at 30±1°C under standard conditions. New adults were separated within 6 hours post-adult eclosion (PAE) and staged from then onward.
Total RNA was isolated using the TRI reagent (Molecular Research Center Inc., Cincinnati, Ohio). The DNA was eliminated from the total RNA using DNase I (Ambion Inc., Austin, Texas) and 2 µg of total RNA for each sample was used for cDNA synthesis. Primers used in quantitative reverse transcriptase PCR (qRT-PCR) are listed in Table 1 or previously published [9], [12]. QRT-PCR reactions were performed using a common program as follows: initial incubation of 95°C for 3 min was followed by 40 cycles of 95°C for 10 s, 55°C for 1 min, Relative levels of mRNAs were quantified in triplicates and normalized using an internal control (ribosomal protein 49, RP49 mRNA).
For dsRNA synthesis, genomic DNA was used as a template to amplify fragments of genes in Table 1, and the PCR products and the MEGA script RNAi Kit (Ambion Inc., Austin, Texas) were employed for dsRNA synthesis. Genomic DNA was extracted from T. castaneum adults and purified using the DNeasy Tissue Kit (QIAGEN, Valencia, CA). All the primers used for dsRNA synthesis and real time PCR are shown in Table 1. For annealing dsRNA, the reaction mixture was incubated at 75°C for 5 minutes and cooled to room temperature over a period of 60 minutes. After treatment with DNase, dsRNA was purified by phenol/chloroform extraction followed by ethanol precipitation. The dsRNA concentration was determined using a Nano Drop 2000 (Thermo Scientific, Pittsburgh, PA). The dsRNA was prepared using 808 bp PCR fragment of E. coli malE gene amplified from 28iMal vector (New England Biolabs, Ipswich, MA) was used as a control.
Newly hatched male adults (within 6 hours after emergence) were anesthetized with ether vapor for 4–5 minutes and lined on a glass slide covered with two-sided tape. The dsRNA was injected into the dorsal side of the first or second abdominal segment using an injection needle pulled from a glass capillary tube using a needle puller (Idaho Technology, Salt Lake City, UT). About 0.8–1 µg (0.1 µl) dsRNA was injected into each new male adult. The malE dsRNA was used as a control. The injected beetles were removed from the slide and reared in whole wheat flour at 30±1°C.
To restore the starvation survival by topical application of JH III or injection of bovine insulin, 0.5 µl of 10 mM JH III in acetone or acetone alone was topically applied to the males injected with malE, ILP2, or JHAMT dsRNA on day 3, day 5, and day 7. 0.2 µl 25 mM HEPES, pH 8.2, or 10 mg/ml bovine insulin solution in 25 mM HEPES (Sigma Aldrich, St. Louis, MO), was injected into malE, ILP2, or JHAMT RNAi males on day 5 PAE.
Total amount of carbohydrates was determined using an Anthrone-based method [53]. A 1 µg/µl solution of glycogen was used as the standard, from which 0–200 µg calibration series were prepared. Three adults were placed in each tube and crushed with a homogenizer in 1 ml of Anthrone reagent. Standards and samples were heated at 92°C for 17 minutes. The samples were allowed to cool to room temperature and optical density (OD) was measured at 625 nm. The amount of total lipids was estimated using the vanillin reagent method [54]. A 1 µg/µl solution of commercial vegetable oil was used as a standard by preparing 0–400 µg calibration series. Three male adults were placed in each tube and crushed with a homogenizer in 500 µl mixture of chloroform–methanol. Samples were kept in a heating block to evaporate the chloroform–methanol. After evaporating the solvent, 200 µl of sulfuric acid was added, and samples were heated in a heating block at 99°C for 10 minutes. The samples were cooled to room temperature, and 800 µl of vanillin reagent was added to each tube and mixed well. Standards and samples were incubated for 30 minutes, and ODs of samples were read at 490 nm. Total protein levels were determined using the Bradford reagent (Sigma Aldrich, St. Louis, MO), and a series of dilutions of bovine serum albumin were used to prepare the standard curve.
To extract hemolymph from the beetles on days 4, 5, and 6 after injection of malE, ILP2, JHAMT, Met, TRET, or TPS dsRNA on day 0 newly emerged adults, the wings were removed and a few holes were poked into the body with forceps. The wings were placed in a microfuge tube containing 250 µl 0.25 M Na2CO3 buffer. The supernatant was collected after centrifugation at a full speed for 10 minutes. Trehalose is a non-reducing sugar resistant to 100°C. The hemolymph in the Na2CO3 buffer was incubated in a 95°C water bath for 2 hours to inactive all enzymes. 150 µl 1 M acetic acid and 600 µl 0.25 M Na-acetate (pH 5.2) were added, and the solution was centrifuged (10 minutes, 12,000 rpm, 24°C). One hundred microliters of supernatant were incubated overnight at 37°C with 1 µl porcine kidney trehalase (Sigma Aldrich, St. Louis, MO) to convert trehalose into glucose. Thirty microliters of this solution were added to 100 microliters of a glucose reagent solution (Sigma Aldrich, St. Louis, MO) and incubated 20 minutes at 37°C. Glucose concentration was quantified at 340 nm with a spectrophotometer. The trehalose dihydrate (Sigma Aldrich, St. Louis, MO) was used as a control and also used to prepare reference curves.
Newly emerged beetles were injected with dsRNA and reared without diet in the 96-well plate individually at 30°C incubator and checked at 5:00 pm every day. Male adults were used in all the experiments.
All the data were analyzed using the SPSS 13.0. The Kaplan-Meier program was used to analyze the survival time and the Log rank analysis was performed to compare the effect of JH III and insulin treatment. To compare nutrient levels, mRNA levels or the ratio between glucose and trehalose, the one-way ANOVA was used, for all the data analysis, the P-value for statistical significance is defined as P<0.05.
|
10.1371/journal.pgen.1007048 | Role of Neuropilin-1/Semaphorin-3A signaling in the functional and morphological integrity of the cochlea | Neuropilin-1 (Nrp1) encodes the transmembrane cellular receptor neuropilin-1, which is associated with cardiovascular and neuronal development and was within the peak SNP interval on chromosome 8 in our prior GWAS study on age-related hearing loss (ARHL) in mice. In this study, we generated and characterized an inner ear-specific Nrp1 conditional knockout (CKO) mouse line because Nrp1 constitutive knockouts are embryonic lethal. In situ hybridization demonstrated weak Nrp1 mRNA expression late in embryonic cochlear development, but increased expression in early postnatal stages when cochlear hair cell innervation patterns have been shown to mature. At postnatal day 5, Nrp1 CKO mice showed disorganized outer spiral bundles and enlarged microvessels of the stria vascularis (SV) but normal spiral ganglion cell (SGN) density and presynaptic ribbon body counts; however, we observed enlarged SV microvessels, reduced SGN density, and a reduction of presynaptic ribbons in the outer hair cell region of 4-month-old Nrp1 CKO mice. In addition, we demonstrated elevated hearing thresholds of the 2-month-old and 4-month-old Nrp1 CKO mice at frequencies ranging from 4 to 32kHz when compared to 2-month-old mice. These data suggest that conditional loss of Nrp1 in the inner ear leads to progressive hearing loss in mice. We also demonstrated that mice with a truncated variant of Nrp1 show cochlear axon guidance defects and that exogenous semaphorin-3A, a known neuropilin-1 receptor agonist, repels SGN axons in vitro. These data suggest that Neuropilin-1/Semaphorin-3A signaling may also serve a role in neuronal pathfinding in the developing cochlea. In summary, our results here support a model whereby Neuropilin-1/Semaphorin-3A signaling is critical for the functional and morphological integrity of the cochlea and that Nrp1 may play a role in ARHL.
| Neuropilin-1 is a member of the neuropilin family acting as an essential cell surface receptor involved in semaphorin-dependent axon guidance and VEGF-dependent angiogenesis and lies within our previously identified ARHL GWAS interval. In this study, we investigated the role of Neuropilin-1/Semaphorin-3A signaling in the functional and morphological integrity of the cochlea, specifically the innervation and vascularization patterns. Detailed analyses of the cochleae of 4-month-old Nrp1 CKO mice showed abnormalities in ribbon synapses, innervation of the hair cells, and microvessels of the stria vascularis. We show also that Neuropilin-1/Semaphorin-3A signaling plays an important role in cochlear innervation.
| Age-related hearing loss (ARHL), or presbycusis, is a progressive bilateral symmetrical sensorineural hearing loss [1] characterized by four types of pathology: (1) sensory deficits resulting from loss of outer hair cells as seen in loss of high frequency auditory brainstem response, (2) neural deficits from auditory nerve degeneration resulting in poor speech recognition, (3) degeneration of the stria vascularis leading to flat audiometric losses across frequencies; and (4) cochlear conductive deficits associated with increased stiffness of the basilar membrane resulting in evenly sloping audiometric losses [2]. Familial studies of presbycusis have attributed approximately half of audiometric variances to hereditary factors; however, the highly variable age of onset, disease progression, and severity of ARHL demonstrate the current uncertain contribution of individual genetic factors to cochlear integrity [3]. Our group has recently demonstrated that ARHL in humans is a polygenic trait [4]. Human genetic studies suggest associations between ARHL and several genes including GRHL2, ITGA8, IQGAP2, GRM7, PCDH15, PCDH20, APOE, EDN1, ESRRG [2]. Although very little is known about ARHL in humans, numerous studies have been published on ARHL in mice. A genetic component to ARHL in inbred mice has been described with approximately 18 Mendelian loci reported to date [5–8]. It has been our overriding hypothesis that true ARHL in mice, as in humans, is a polygenic trait with the composite phenotype resulting from genomic variation at multiple loci likely different from the Mendelian loci described thus far.
To define the genetic architecture of ARHL in mice, we undertook a genome-wide association study (GWAS) using a meta-analysis strategy by combining data sets from five cohorts containing 937 samples in total [9]. The results of the meta-analysis led us to an approximately 2 Mb interval containing Nrp1, a gene that is involved in cardiovascular and neuronal development and is closely related to Neuropilin-2 (Nrp2), a gene involved in cochlear epithelial innervation [10]. Neuropilin-1 is a transmembrane receptor type I protein that is known to bind both vascular endothelial growth factor beta (VEGFb) and semaphorin classes including subtypes 3A, 3B, 3C, and 3D. Semaphorin-3A is involved in axonal guidance via chemorepulsion [11]. Semaphorin-3A -induced neuronal growth cone collapse has been shown to require neuropilin-1 in conjunction with Plexin-A co-receptors [12]. Previous cardiovascular studies have shown that altered endothelial cell migration, abnormal blood flow, and enlarged vessels are major defects caused by targeted inactivation of the Nrp1 gene in mice. Additionally, homozygous Nrp1 mutant mice are known to have a perinatal lethal phenotype due to impaired heart development [13].
In this study, using an inner ear-specific knock out, we investigated the role of Nrp1 in the functional and morphological integrity of the cochlea in mice. The results of this study suggest Nrp1 may be involved in ARHL.
We first used in situ hybridization to characterize the expression of Nrp1 and its ligand Sema3a at different stages of cochlear development. Between E13.5 and P1, the SGNs migrate along the extending cochlear duct then extend peripheral axons toward the cochlear epithelium and central axons toward the brainstem [14]. In situ hybridization of WT cochleae at E13.5, E15.5, E18.5, and P1 showed weak expression of Nrp1 at E16.5 and E18.5 with more robust expression starting at P1 (Fig 1A and 1C). Our in situ hybridization data also showed that Sema3a expression started around E13.5 and continued at E16.5 and E18.5 on the abneural side of the cochlear epithelium and SGNs (Fig 1D–1F). These data suggest that both Nrp1 and Sema3a are expressed in the cochlea during time points when SGNs begin to innervate the organ of Corti.
We next performed immunostaining using P5 cochleae to determine the precise distribution of neuropilin-1 and semaphorin-3A in the postnatal cochlea. As shown in Fig 2, neuropilin-1 is visible in SGNs and the SV (Fig 2A, 2B, 2D and 2E), but not expressed after conditional deletion of Nrp1 (Fig 2C and 2F). Semaphorin-3A protein was visible within the organ of Corti (Fig 2G and 2H) and SGNs (Fig 2J and 2K), but not after the semaphorin-3A antibody was pre-adsorbed by the blocking peptide (Fig 2I and 2L). Overall, these data suggest that Nrp1 is expressed at minimal levels by SGNs and cells of the stria vascularis before birth while Nrp1 levels become elevated in these locations after birth. Sema3a is expressed by cells within the cochlear epithelium and SGNs, which is complementary to the expression patterns of Nrp1. The simultaneous expression of both neuropilin-1 and semaphorin-3A shortly after birth suggested these factors may be involved in the process of SGN pruning and refinement, which occurs during this time in cochlear development. These findings prompted further investigation of Neuropilin-1/Sema-3A signaling in cochlear innervation.
Much of the genetic data from our meta-analysis GWAS came from the original backcrossing data (C57BL/6J x DBA/2J) during the mapping of Ahl8 [6]. In their mapping study of ahl8, a locus on chromosome 8 was also identified. This led us to determine the possibility of Nrp1 expression variation in the cochlear tissue of C57BL/6J and DBA/2J mice. Real-time PCR showed 1.78-fold higher Nrp1 expression for DBA/2J mice (1.96) when compared to C57BL/6J (1.09) (S1 Fig) supporting our gene selection (p<0.01).
To investigate the function of neuropilin-1 in the cochlea, we generated an inner-ear specific conditional knockout mouse using Pax2Cre and loxp-driven Nrp1 removal (see Methods for details). Using this line, we first wanted to determine whether Nrp1 is required for the formation or maintenance of ribbon bodies, which represent glutamatergic synapses connecting hair cells and SGNs. To visualize and quantify ribbon bodies, cochlear whole mount preparations from WT, Nrp1fl/+;Pax2Cre and Nrp1fl/fl;Pax2Cre mice at P5 (n = 4 per group) and 4 months (n = 5 per group) were immunostained with antibodies that bind to ribeye, a splice variant of CtBP2. The tissue samples were stained with Hoechst33342 to confirm whether the ribbon bodies were located on either OHCs or IHCs. The time points described above were chosen so that we could track any possible changes in synaptic connectivity from just after birth to full maturity within the cochlea. CtBP2 counts of both IHCs and OHCs at P5 showed no differences between WT and Nrp1 CKO samples (Fig 3). No significant changes in immunostaining of IHCs was observed between Nrp1fl/fl;Pax2Cre and WT at 4 months; however, at this time point, the synaptic ribbon density (CtBP2 counts) in the OHC region in 4-month-old mice was significantly reduced (p<0.01) for Nrp1fl/fl;Pax2Cre mutants (1.9 puncta/cell) compared to the controls (1.4 puncta/cell). Given this reduction in ribbon synapses, we next wanted to determine whether Nrp1 loss also conferred a loss of SGNs (possibly through apoptosis). Thus, we quantified SGN density at the apical, middle, and basal turns of the cochlea at P5 (n = 4 per group) and at 4-months (n = 5 per group) for WT, Nrp1fl/+;Pax2Cre and Nrp1fl/fl;Pax2Cre mice (Fig 4). As expected from our ribbon synapse counts, at P5 no significant changes in SGN density were observed among the different genotypes or regions of the cochlea (Fig 4E). However, the SGN counts in 4-month-old mice (Fig 4F) showed decreased density of the neuronal cell bodies in Nrp1fl/fl;Pax2Cre mice compared to WT controls (apical turn p = 0.03, middle turn p = 0.03, and basal turn p = 0.002). Interestingly, the number of SGNs lost in the absence of Nrp1 is unexpectedly high compared to the number of IHC ribbon synapses lost (see Fig 3 and Discussion). Nevertheless, the loss of OHC synaptic ribbons and diminished SGN density in 4-month-old Nrp1fl/fl;Pax2Cre mice, suggests that Nrp1 may play a role in maintaining SGN integrity during postnatal stages.
To examine the mechanism of SGN cell loss, we performed caspase-3 immunostaining. Caspase-3, a molecule necessary for the cellular apoptotic cascade, was identified by immunostaining in WT and Nrp1fl/fl;Pax2Cre cochleae to ascertain the fate of the SGNs once mice reached 4 months of age. Caspase-3 positive neurons were found in Nrp1fl/fl;Pax2Cre mutants but not in WT mice, suggesting that the loss of OHC ribbon synapses resulted from pruning or the apoptosis of mature neurons (Fig 4C and 4D). Taken together, these data suggest a gradual loss of contacts between OHCs and SGNs in the absence of Nrp1, which correlates with the death of SGNs around 4 months of age.
We next wanted to see if Nrp1fl/fl;Pax2Cre mutants also showed defects in cochlear innervation patterns to determine the extent to which Nrp1 may function in axon guidance in the cochlea. Cochleae from WT and Nrp1fl/fl;Pax2Cre mutants at P5 and 4 months were immunostained with TUJ1 antibodies and assessed as whole-mount preparations. At P5, disorganized outer spiral bundles (type II fibers) were evident in cochleae of the Nrp1fl/fl;Pax2Cre mice at basal, mid, and apical turns (n = 3 per group), but the radial fibers (mostly type I SGNs) appeared normal (Fig 5A and 5B). Compared to controls, we also observed significant disruptions to the normal patterns of innervation in cochleae from 4-month-old Nrp1fl/fl;Pax2Cre mice (Fig 5C and 5D). In a normal cochleae, 90–95% of the SGNs innervate IHCs; the remaining 5–10% of neurons travel beyond IHCs to innervate OHCs in an en passant fashion [15]. TUJ1 immunostaining of cochlear nerve fibers extending into the hair cell region in 4-month-old Nrp1fl/fl;Pax2Cre mutants (basal turn) revealed aberrant axons with abnormal innervation of OHCs. Mid-modiolar cross-sections of the cochlea of the Nrp1fl/fl;Pax2Cre mice (4-month-old) also showed disorganized innervation of the outer hair cells (Fig 6).
Neuropilin-1 can be activated by both secreted semaphorins and Vascular Endothelial Growth Factors (VEGFs) [16]. Given this, we wanted to ask next whether mice with a variant of Nrp1 that fails to bind secreted semaphorins (Nrp1sema-) showed cochlear innervation defects similar to the Nrp1 CKO line [17]. We first used anti-TUJ1 antibodies to examine the overall distribution of nerve fibers in mutant and WT cochleae at E16.5 (Fig 7A–7F). For each sample, we also performed anti-Myo6 and anti-Sox2 immunostaining to identify the hair cells and supporting cells, respectively. Compared to cochleae from WT littermates (Fig 7A–7C), cochleae from Nrp1sema-/sema- mice showed nerve fibers in great excess with many that extended past the OHC region and even sometimes past the Deiters’ cell region (Fig 7D–7F). Cochleae from Nrp1sema-/sema- mice showed a normal distribution of hair cells and supporting cells (Fig 7B, 7C, 7E and 7F) indicating the innervation defects here were not due to changes in organ of Corti formation. Using E18.5 samples from the Nrp1sema-/sema- mice, we next delineated the distribution of SGN afferents using Syt1 antibodies and cochlear efferents using Gap43 antibodies [10]. Compared to the apex and middle regions of control cochleae, Nrp1sema-/sema- cochleae showed a significant increase in Syt+ fibers in the OHC region (Fig 7K and 7O), but no changes in the distribution of Gap43+ efferent fibers. At the base, we found no significant increases in Syt+ fibers in the OHC region of Nrp1sema-/sema- cochleae overall (Fig 7) but did often see unusual nerve bundles that were both Syt1- and Gap43-positive (see arrowheads in 7M and N). These unusual bundles often took torturous paths toward the organ of Corti and terminated in the hair cell region or just beyond. In addition, the anatomical origins of these neurons were not clear in that the processes appeared to come from outside of the cochlea and not Rosenthal’s canal where the SGN cell bodies are located. Nevertheless, cochleae from Nrp1sema-/sema- mice showed innervation defects that, to a large extent, mirrored the phenotypic defects in the Nrp1 CKO mice. This indicates that neuropilin-1 receptor activation by secreted semaphorins is necessary for normal cochlear innervation.
To further investigate the role of Neuropilin-1/Sema-3A in mediating SGN migration and refinement, we used small interfering RNAs (siRNAs) targeting Nrp1 mRNA to reduce neuropilin-1 protein levels in SGNs in cell culture. A transient transfection with predesigned siRNA oligonucleotides decreased neuropilin-1 protein expression as measured by Western immunoblotting (Fig 8). The SGN explant culture showed that semaphorin-3A, at a concentration of 250 ng/mL, mediated axonal repulsion (Fig 8B). The neurite outgrowth experiment was continued using the concentration of Nrp1 siRNA oligonucleotide (50nM) that produced the greatest knockdown (approximately 60% decrease from the control). Nrp1 knockdown caused by siRNA transfection decreased neurite outgrowth and abolished the ability of Sema3a to decrease neurite outgrowth, suggesting that Sema3a repulses SGNs in an Nrp1-depended manner (Fig 8I). In contrast, the negative control scrambled siRNA neither decreased Nrp1 protein nor abolished Sema3a activity. These experiments further support a key role for Nrp1/Sema3a signaling in cochlear innervation.
To investigate the interaction between semaphorin-3A and SGNs, we established whole cochlear cultures from E17.5 mice and placed them in media containing either control IgG-Fc or semaphorin-3A-Fc (20nM). To determine whether semaphorin-3A altered hair cell innervation, the tissue samples were labeled with TUJ1 and Myo6 antibodies and imaged by confocal microscopy. Compared to control samples that showed robust hair cell innervation (Fig 8G), samples treated with semaphorin-3A-Fc showed significantly reduced innervation of the sensory epithelium (Fig 8H). To quantify this change in innervation, high-resolution confocal z-stacks were taken from the volume of tissue occupied by the HCs. Compared to controls, semaphorin-3A decreased innervation around the sensory epithelium by nearly 60% (Fig 8J). These data suggest a possible role for semaphorin-3A in inhibiting SGN outgrowth.
For a detailed analysis of the entire auditory pathway in Nrp1fl/fl;Pax2Cre mutants, we next evaluated OHC activity using DPOAE and neuronal responses by ABR wave I peak-to-peak amplitudes. DPOAEs, cochlear responses generated after two simultaneous pure tone frequencies, are objective indicators of OHC functional status [18]. OHC function was determined to be normal in Nrp1 mutants as DPOAE levels for Nrp1fl/fl;Pax2Cre, Nrp1fl/+;Pax2Cre, and WT groups did not differ significantly at 2 and 4 months of age.
ABR test results of the 2-month-old mice showed that the hearing thresholds of the Nrp1fl/fl;Pax2Cre group were significantly higher than WT controls at 4kHz, 8kHz, 16kHz, 24kHz, and 32kHz. At 4 months of age, Nrp1fl/fl;Pax2Cre mice developed elevated hearing thresholds at all tested frequencies except 12 kHz when compared to WT controls (Fig 9).
Peak-to-peak analysis of wave I was calculated from the ABR data described above. Wave I is thought to indicate the summed activity of SGN contact with hair cells, so a normal DPOAE with a diminished wave I peak would suggest dysfunction of the SGNs, IHCs, or the synapses between them [19]. At 2 months of age, no significant changes in wave I amplitude were found among the three groups of mice; however, the Nrp1fl/fl;Pax2Cre mutants at 4 months of age recorded significantly lower wave I amplitudes than WT mice at 8 kHz, 12kHz, and 32 kHz. Paired with our immunostaining results of IHC defects in 4-month-old Nrp1fl/fl;Pax2Cre mice, the reduced wave I amplitude suggests the contribution of cochlear neural damage in the hearing loss seen in 4-month-old Nrp1 mutants (Fig 10).
The composition of endolymph and the maintenance of the endocochlear potential are determined by ion balance regulated by the stria vascularis (SV) [20]. We hypothesized that abnormal vascularization of the SV could lead to electrolyte imbalance, resulting in abnormal hearing thresholds. To test this hypothesis, we investigated the morphology of the micro-vessels of the SV at the basal turn of the cochlea (n = 3 per group). The lectin immunostaining of the Nrp1fl/fl;Pax2Cre cochleae at P5 and 4-months-old demonstrated grossly enlarged SV microvessels (Fig 11). The minimum and maximum microvessel diameter of the Nrp1fl/fl;Pax2Cre mice were 4.17μm and 43.67μm at P5, and 3.22μm and 95μm in 4-month-old mice, respectively. The minimum and maximum microvessel diameters of the WT mice were 5.32μm and 23.90μm at P5, and 5.2μm and 26.15μm in 4-month-old mice, respectively. Overall, the maximum microvessel diameter in 4-month-old Nrp1fl/fl;Pax2Cre mice was 3.6 fold higher than WT controls. Thus, future endocochlear potential studies are needed to pinpoint the effect of Nrp1 knockout on normal functioning of auditory hair cells.
There exists a growing literature supporting the notion that ARHL may be associated with degenerative changes in the cochlear nerve and its synapses [21, 22]. Although this phenomenon and that of the classically defined neural presbycusis are now well studied histologically, little is known about the molecular basis for this pathology. Using a meta-analysis GWAS approach we have defined several candidate regions for ARHL in mice, one of which included Nrp1 [9]. Nrp1 is a well-known factor in neuronal and cardiovascular development [13, 17]. Its homolog, Nrp2 has been shown to be involved in inhibiting type I SGNs from the OHC region in the developing cochlea [10]. While substantial data exists for the role of Nrp1 in tumorigenesis and embryonic development, to date, the role of Nrp1 in the postnatal development of the cochlear apparatus remains unclear. According to previously published cochlear nerve microarray data, Nrp1 expression in the spiral ganglion shows a peak at E16.5 followed by a dip between E16 and P0, and a general trend of increased expression up to two weeks into postnatal development [23]. These findings are consistent with our data as we found upregulated Nrp1 postnatal expression in the organ of Corti and in the SGNs in the first postnatal week, a critical period for maturation of hair cell innervation which suggests a role for Nrp1 in this process.
In the peripheral auditory system, the type I SGN afferent fibers undergo significant reorganization during embryonic development in mice [10]. In this study, we have identified Nrp1 to be a critical component of this reorganization process. Previously, mice lacking normal Nrp1 function (Nrp1Sema- mutants) showed pathfinding defects in vestibular ganglion neurons [17], defasciculation of the intercostal nerves, and crossing bundles to neighboring nerves of the Nrp1Sema- mutants [24]. Many previous studies have also demonstrated a role for semaphorin-3A in axonal chemorepulsion, including repulsion of sensory and cortical axons [11, 25]. Our SGN explant and semi-intact cochlear cultures also demonstrated semaphorin-3A repels SGNs, which suggests semaphorin-3A can normally inhibit SGN outgrowth. Complementary to these findings, Nrp1sema-/sema- cochleae showed dramatically enhanced innervation by Syt-positive fibers during developmental stages, suggesting excessive innervation by SGNs possibly due to the absence of a repulsive signal. We do not yet know whether this was due to increased numbers of SGNs, increased complexity of individual SGNs, or ectopic innervation of the cochlea by a different population of neurons (e.g. vestibular neurons). During embryonic development, Type II SGNs pass by the IHCs and reach the OHC area then extend toward the base of the cochlea forming en passant contacts with 3 to 10 OHCs within the same row. These projections gather beneath the rows of OHCs to form 3 outer spiral bundles [15]. Here, we demonstrate that Nrp1 conditional mutants show disorganized outer spiral bundles at early neonatal stages (P5) and that these pathfinding defects are apparent in the older (4-month-old) mice. In addition, Nrp1sema-/sema- cochleae showed excessive numbers of SGNs present in the OHC region in late embryonic stages.
There were several pieces of evidence shown here that implicate Nrp1 in age-related hearing loss. First, we found that SGN density was lost in the Nrp1fl/fl;Pax2Cre cochleae over time (Fig 4). Since most of these neurons are likely type I SGNs, this potentially explains why the Nrp1fl/fl;Pax2Cre mice showed elevated ABR thresholds and wave 1 amplitude reductions. Larger ABR wave I amplitude shifts at equal sound pressure levels are associated with greater auditory nerve threshold elevation [26]. Second, Nrp1fl/fl;Pax2Cre cochleae showed conspicuous defects in the stria vascularis, which normally maintains the ionic composition of the endolymph and promotes normal auditory transmission. Although we did not detect any significant changes to otoacoustic emissions in the Nrp1fl/fl;Pax2Cre mice (suggesting normal OHC function), it is possible that these mice have defects in IHC function that also, like the loss of SGNs, contributes to their altered hearing thresholds. The Nrp1fl/fl;Pax2Cre mice did show reduced numbers of OHC ribbon synapses, but type II SGNs do not contribute to the canonical auditory pathway [27], so it is unlikely that this phenotypic defect contributes to the changes in hearing thresholds.
One curious finding here was that the Nrp1fl/fl;Pax2Cre mice showed a profound loss of SGNs and only a mild loss of IHC ribbon synapses. Although overall the differences between controls were not statistically significant (Fig 3C), we did observe an almost 50% decrease in IHC ribbon synapses in 3 out of 5 of the 4-month-old Nrp1 mutants examined. Since the majority of SGNs are type I and terminate on IHCs, it would be expected that there would be a commensurate reduction in IHC ribbon bodies. One obvious cause for this discrepancy is that many of the CtBP2-positive bodies in IHCs from the Nrp1fl/fl;Pax2Cre cochleae may represent orphaned synapses that lack a postsynaptic terminal. A second less likely possibility is that some remaining type I SGNs extend collateral processes that form synapses with the IHCs.
Overall, these data suggest that reduced SGN density in 4-month-old Nrp1 mutants, in addition to abnormal axonal pathfinding, lead to impairment of the OHCs synaptic integrity. Consistent with the abnormal neuronal phenotype in 4-month-old Nrp1 deficient mice, we also observed a decline in ABR wave I amplitudes as they matured.
Nrp2, the other member of the neuropilin family, is responsible for encoding a transmembrane receptor protein with sequence homology to Nrp1 but with different ligand binding affinities. Neuropilin-2 receptors bind Sema-3 subtypes 3C and 3F, VEGF-A and VEGF-C isoforms [28]. While Nrp2 plays a role in neuronal pathfinding, it has not been linked to neuron survival [10]. We have shown, however, that Nrp1 likely plays a role in long-term neuronal survival and maintenance throughout life as SGN cell counts diminish with age in Nrp1 CKO mice. These results were consistent with previously published data showing that Nrp1 is an essential factor in survival of the GnRH and trigeminal neurons by interacting with vascular endothelial growth factor (VEGF) ligands [29, 30].
While our results support a model in which Nrp1 signaling is necessary for the establishment of SGN projections in the postnatal period, Nrp1 also appeared to play an essential role in normal vascular development of the cochlea. Our results show that Nrp1 deletion leads to enlarged vessels in the stria vascularis of early postnatal and adult mice, which may have an impact on the maintenance of the endocochlear potential [31].
Animal procedures were performed at the Zilkha Neurogenetic Institute in accordance with the guidelines of the Institutional Care and Use Committee (IACUC) of the University of Southern California. Nrp1fl/fl mice of mixed backgrounds (CBA/CaJ x C57BL/6J) strains were kindly provided by Dr. Henry Sucov. Pax2Cre mice of mixed backgrounds strains (CBA/CaJ x C57BL/6J) were kindly provided by Dr. Takahiro Ohyama. In Pax2Cre mice, Cre mRNA is detectable in the otic placode starting at the late presomite stage [32]. Nrp1 CKO mice of either sex were obtained by crossing Nrp1fl/fl mice to Nrp1fl/+;Pax2Cre mice. For postnatal collections, P0 was defined as the day of birth. For genotyping of Nrp1 knockout mice, polymerase chain reaction (PCR) was performed using the following primers: Nrp1 forward 5’- AGGTTAGGCTTCAGGCCAAT-3’, Nrp1 Reverse 5’ GGTACCCTGGGTTTTCGATT-3’; Pax2Cre Forward 5’-GCCTGCATTACCGGTCGATGCAACGA-3’, Pax2Cre Reverse 5’-GTGGCAGATGGCGCGGCAACACCATT-3’. The Nrp1sema- line [17] was kindly provided by Dr. Alex Kolodkin of Johns Hopkins University. Nrp1sema- mice were bred and maintained at either the Porter Neuroscience Research Facility (Bethesda, MD) under the guidelines of the NIDCD IACUC or at the Division of Comparative Medicine (Washington, DC) under the guidelines of the Georgetown University IACUC. Nrp1sema- mice were maintained on a C57BL/6J background. Male and female heterozygous mice were bred to generate homozygous mutants and littermate controls. Genotyping was performed using the following WT and Nrp1sema- specific primers: AGGCCAATCAAAGTCCTGAAA ACAGTCCC and AAACCCCCTCAATTGATGTTAACACAGCCC.
Six-week-old C57BL/6 mice (n = 8) and DBA/2J mice (n = 7) were euthanized, and bilateral inner ears were harvested. Cochlear tissues were collected, and left and right ear samples were combined and immediately processed with RNAqueous Total RNA Isolation Kit (Life Technologies) according to manufacturer’s instructions. Total RNA was then converted to cDNA using the SuperScript III First-Strand Synthesis SuperMix (Life Technologies). PCR was performed using the primer pairs acquired from applied biosystems: assay ID: Mm00435379_m1. Each sample was run in triplicate along with the housekeeping gene, GAPDH. Relative quantities of the transcripts were determined using the 2−ΔΔCt method using GAPDH as a reference.
Cochlear whole mount sample preparation: Mouse cochleae were dissected after the second hearing measurement and were fixed with 4% PFA overnight. Fixed samples were decalcified using 10% EDTA, and dissected using the mouse cochlear dissection method from Eaton Peabody Laboratories at the Massachusetts Eye and Ear Institute website (http://www.masseyeandear.org/research/otolaryngology/investigators/laboratories/eaton-peabody-laboratories). For the Nrp1sema- mice samples, embryonic cochleae were fixed for 30 minutes in 4% PFA and then rinsed extensively in 1X PBS before dissection and immunostaining.
Cochlear frozen section sample preparation: Fixed heads were sequentially dehydrated in 15% and 30% sucrose, embedded in Tissue-Tek O.C.T. compound (Sakura Finetek) and snap frozen on dry ice. Blocks were sectioned (12 μm thickness) on a Leica 3050 S cryostat in a cranial-to-caudal coronal direction.
SGN explant three-dimensional culture: Sensory epithelia of cochleae with attached SGNs were dissected from the E16.5 embryos and placed in Leibovitz’s L-15 medium (Invitrogen). The isolated SGNs were cut into four equal pieces starting from one turn away from the apex of cochlea. Each SGN explant was transferred in a drop of phenol red-free Matrigel (Corning) and placed on poly-D-lysine (50 μg/ml)-coated glass coverslips in a 24-well plate. After complete solidification of the Matrigel, DMEM/F12 medium supplemented with 10% fetal bovine serum, 1% N2 supplement, and 0.3 mg/ml ampicillin were added and maintained in the culture for 3 days. After fixation, dissected cochleae or tissue sections were permeabilized with 0.2% TrionX-100 followed by incubation in 10% blocking serum for 2 hours at room temperature. The samples were incubated with the primary antibody at 4°C for 24 to 48 hours and exposed to secondary antibodies for 2 hours at room temperature. Using a Carl Zeiss LSM 780 laser scanning microscope (AxioObserver.Z1), 3 representative images were taken for each slide, and the total, average, and maximal neurite lengths per SGN explant were measured using Metamorph software (Molecular Devices) [33, 34]. Antibodies used in this study were as follows: Alexa 488-conjugated mouse anti-neuron specific class III beta tubulin (anti-TUJ1) (1:300; Covance), rabbit anti-Neuropilin-1 (1:50;Abcam), rabbit anti-Semaphorin-3A (1:100; Abcam), mouse anti-CtBP2 (1:200; BD Biosciences), mouse-anti-TUJ1 (1:1,000, Covance), rabbit-anti-myosin VI (1:1,000, Proteus Biosciences), goat-anti-Sox2 (1:300, Santa Cruz Biotechnology), chicken-anti-synaptotagmin-1 (1:1,000, Aves Labs), mouse-anti-GAP43 (1:2,000, Chemicon), Alexa Fluor-488 anti-mouse (1:500; Life technologies), Alexa Fluor 594 anti-goat (1:500; ThermoFisher). Fluorescent dye Hoechst 33342 (0.1 μg/mL; Southern Biotech) was used for DNA labeling. Blocking peptide for Anti-Semaphorin-3A antibody (Abcam) was used for Sema-3A immunostaining (negative controls). For synaptic ribbon-to-hair cell ratios, tissue sections at the basal turn of the cochlea were selected, and the number of synaptic ribbons was compared separately to the number of inner hair cells and to the number of outer hair cells per section. The synaptic ribbon-to-hair cell ratios for WT and Nrp1fl/fl;Pax2Cre 4-month-old mouse cochleae (n = 5 per group) and P5 mouse cochleae (n = 4 per group) were assessed. Spiral ganglion cells of the P5 and 4-month-old mice were counted at apical, mid, and basal turns of the cochlea (n = 5 per group at each turn). To quantify numbers of Syt1+ fibers in the OHC region of Nrp1sema-/sema- cochleae and their littermate controls, we determined the number of fiber tracks extending into the OHC region and normalized that value to the longitudinal distance of the cochlea within that region. Sample sizes: 9 WT cochleae and 13 Nrp1sema-/sema- cochleae.
In situ hybridization was performed as previously described [35]. Briefly, embryonic day E15.5 heads were fixed in 4% paraformaldehyde in PBS overnight at 4°C, sunk in 30% sucrose in PBS at 4°C, incubated in Tissue-Tek O.C.T. compound (Sakura Finetek) at room temperature for 10 min and frozen on dry ice. Sections, 14μm thick, were cut using a Leica 3050 S cryostat. RNA probes for mouse Nrp1 (GE Dharmacon, Clone ID 6409596) and mouse Sema3a (GE Dharmacon, Clone ID 30532393) were synthesized, labeled with digoxigenin, and hydrolyzed by standard procedures. In situ hybridization images were obtained under bright-field microscopy (BZ9000; Keyence, Osaka, Japan).
SGN explants in antibiotic-free medium were transfected with 50 nM predesigned Nrp1 siRNA oligonucleotides (Santa Cruz Biotechnology) using Lipofectamine 3000 (Invitrogen) as per manufacturer’s instructions. Some explants were treated with 250 ng/ml Sema-3A-Fc (R&D Systems). We used a scrambled siRNA oligonucleotide that did not exhibit homology to any known encoding region as a negative control. The siRNA-mediated knockdown efficiency was determined by Western immunoblotting. After Nrp1 siRNA transfection and semaphorin-3A treatments, the Matrigels covering SGN explants were removed with cell recovery solution (Corning) and the explants were homogenized in RIPA lysis buffer supplemented with a cocktail of protease inhibitors (Santa Cruz Biotechnology). Explant extracts were electrophoresed on 7.5% SDS-polyacrylamide gels and the separated proteins were transferred to a polyvinylidene fluoride membrane. The membranes were blocked with 5% dry milk and then incubated with rabbit-anti-Neuropilin-1 (1: 1000 dilution; Abcam) overnight, followed by incubation with a goat anti-rabbit secondary antibody conjugated with horseradish peroxidase. The immunoreactive proteins were visualized by the Supersignal West Femto Maximum Sensitivity Substrate kit (Thermo Scientific). The bands were quantitated by densitometric scanning using the Omega 12 IC Molecular Imaging System and the Ultra Quant software. β-Actin loading control was carried out.
E17.5 cochleae were dissected in chilled HBBS/HEPES, and the cochlear capsule and stria vascularis were removed. The tissue samples were placed on polycarbonate filters and incubated at 37°C. The culture medium: DMEM supplemented with 10% fetal bovine serum, 0.2% N2 (Thermo-Fisher Scientific), and 0.0001 ciprofloxacin (Sigma). Either purified Sema-3A-Fc (R&D Systems) or human IgG-Fc (Jackson Immunoresearch) was added to the medium for a final concentration of 20nM. After 20hr of incubation, the cochleae were removed from the filters, fixed for 15 min in cold 4% PFA, then processed for immunostaining (anti-TUJ1 and anti-Myo6) and confocal imaging. To image the tissue samples, confocal z-stacks were acquired starting at the “top” of the hair cells (lumen of the cochlear duct) through the “bottom” of the supporting cells (past the basement membrane). To quantify how Sema-3A-Fc altered cochlear innervation, maximum-intensity z-projects were generated only from the z-planes that included the hair cells. Using ImageJ, a region of interest (ROI) around the inner and outer hair cells was selected from each sample and the area occupied by TUJ1-positive neurons was normalized to the area of the ROI. For each sample, equivalent thresholding was used.
ABR and DPOAE testing was performed for Nrp1fl/fl;Pax2Cre (n = 8), Nrp1fl/+;Pax2Cre (n = 8), and WT mice (n = 11) at 2 and 4 months of age. DPOAE testing was performed for Nrp1fl/fl;Pax2Cre and WT mice at 4 months of age. Stainless-steel electrodes were placed subcutaneously at the vertex of the head and the right mastoid with a ground electrode at the base of the tail. Auditory signals were presented as tone pips with a rise and fall time of 0.5 msec and a total duration of 5 msec at the frequencies 4, 8, 12, 16, 24, and 32 kHz. Tone pips were delivered below threshold and then increased in 5 dB increments until the goal of 100 dB was reached. Signals were presented at a rate of 30/second. Responses were filtered with a 0.3 to 3 kHz pass-band (x10000 times). For each stimulus intensity, 512 waveforms were averaged. Hearing threshold was determined by inspection of ABR waveforms and was defined as the greatest intensity at which no reliable ABR waveforms were generated. Data were stored for offline analysis of peak-to-peak (P1-N1) values for wave 1 amplitudes. ABR thresholds were determined by an independent observer who was blinded to the genotypes of the mice. Peak-to-peak (P1-N1) values for wave 1 amplitudes were calculated using the ABR Peak Analysis software (Bradley Buran, Eaton-Peabody Laboratory). Distortion product otoacoustic emissions (DPOAEs) were recorded with a custom-designed transducer and data-acquisition system based on a National Instruments PXI hardware configuration developed at the Eaton-Peabody Laboratories, Massachusetts Eye and Ear Infirmary. The 2f1-f2 DPOAE was measured with an f2/f1 ratio of 1.22 for f2 = 8, 11, 16, 22 and 32 kHz at primary tone levels (L2) presented from 20 to 70 dB SPL in 10 dB steps to form an input/output function. DPOAE threshold was determined from the input/output function as the extrapolated primary tone level (L2) required to produce a DPOAE of 0 dB SPL. DPOAEs with at least 3 dB signal-to-noise (SNR) ratio were included in all analyses; however, when the SNR was < 3 dB, the DPOAE was assigned the level of the noise floor and included. This was done to avoid excluding the very lowest level DPOAEs, which carry information by virtue of their reduced amplitude. DP-grams (DPOAE amplitude across frequency, 8–32 kHz) were also constructed at L2 = 70 dB SPL for all mice; only points > 3 dB SNR were included in this analysis.
Statistical analysis for ABR, and wave I amplitude were performed using analysis of variance (one-way ANOVA) and post hoc comparisons with Fisher’s LSD test. For neurite outgrowth analysis, Kruskal-Wallis Test, and Dunn's Test were used. Synaptic ribbon count, and spiral ganglion neuron counts were performed using the student’s t-test. GraphPad Prism 7 software was used to perform the tests. Continuous variables with normal distribution were expressed as mean +/- standard deviation (SD). A 2-tailed p value less than 0.05 indicated statistically significant differences.
The results of this study support a potential role for Nrp1 expression variation to be associated with ARHL in mice. Furthermore, our findings suggest Nrp1 plays a role in the postnatal development and maintenance of the cochlear neuronal network. Our findings also support the role of Nrp1 in normal development of the stria vascularis microvasculature. Taken together, these data suggest that both neuronal and vascular abnormalities of Nrp1 deficient mice contribute to abnormal hearing in 4-month-old mice. To further understand the impact of Neuropilin-1/Sema-3A on ARHL, future studies will likely include investigation of tissue-specific knock outs of Sema3a and VEGF-B in the inner ear, neuronal cell type-specific immunostaining of these selected animal models, and assessment of cochlear hemodynamics and electrophysiology of the Nrp1 deficient mice.
|
10.1371/journal.pgen.1002878 | The Drosophila Mi-2 Chromatin-Remodeling Factor Regulates Higher-Order Chromatin Structure and Cohesin Dynamics In Vivo | dMi-2 is a highly conserved ATP-dependent chromatin-remodeling factor that regulates transcription and cell fates by altering the structure or positioning of nucleosomes. Here we report an unanticipated role for dMi-2 in the regulation of higher-order chromatin structure in Drosophila. Loss of dMi-2 function causes salivary gland polytene chromosomes to lose their characteristic banding pattern and appear more condensed than normal. Conversely, increased expression of dMi-2 triggers decondensation of polytene chromosomes accompanied by a significant increase in nuclear volume; this effect is relatively rapid and is dependent on the ATPase activity of dMi-2. Live analysis revealed that dMi-2 disrupts interactions between the aligned chromatids of salivary gland polytene chromosomes. dMi-2 and the cohesin complex are enriched at sites of active transcription; fluorescence-recovery after photobleaching (FRAP) assays showed that dMi-2 decreases stable association of cohesin with polytene chromosomes. These findings demonstrate that dMi-2 is an important regulator of both chromosome condensation and cohesin binding in interphase cells.
| The packaging of DNA into chromatin is critical for the organization and expression of eukaryotic genes. Nucleosomes repress transcription by blocking the access of transcription factors and other regulatory proteins to DNA. Levels of chromatin organization above the level of the nucleosome—including chromosome folding, pairing, and looping—can also have profound effects on gene expression. However, the mechanisms by which higher-order chromatin structure is regulated and used to control gene expression remain poorly understood. Using Drosophila as a model organism, we have discovered an unanticipated role for dMi-2, a well-characterized ATP-dependent chromatin- remodeling factor, in the regulation of higher-order chromatin structure and cohesin dynamics in vivo. The inhibition of dMi-2 function causes chromosomes to condense, while elevated expression of dMi-2 triggers the decondensation of polytene and mitotic chromosomes and also destabilizes cohesin binding. Our findings suggest that dMi-2 may regulate transcription and cellular differentiation in other organisms, including humans, by altering higher-order chromatin structure or cohesin dynamics.
| The packaging of DNA into chromatin is critical for the organization and expression of eukaryotic genes [1], [2], [3]. The basic unit of chromatin structure, the nucleosome, contains the core histones H2A, H2B, H3 and H4. The association of nucleosomes with histone H1 and other linker histones facilitates their packaging into 30 nm fibers, which in turn are packaged into increasingly compact higher-order structures. Nucleosomes and other components of chromatin can repress transcription by blocking the access of regulatory proteins and the basal transcriptional machinery to DNA. There is growing evidence that levels of chromosome organization above the level of the nucleosome – including chromosome folding, pairing and looping – also play important roles in the regulation of gene expression. For example, condensin and cohesin, which were initially identified by their roles in mitosis and meiosis, modulate transcription by promoting long-range chromosomal interactions and DNA looping in interphase cells [4].
The repressive effects of nucleosomes on transcription are modulated by two general mechanisms: the covalent modification of nucleosomal histones and ATP-dependent chromatin remodeling [1], [5]. By altering the structure or positioning of nucleosomes, ATP-dependent chromatin-remodeling factors play critical roles in transcription and other nuclear processes. Dozens of chromatin-remodeling factors, including members of the SWI/SNF, ISWI, CHD and INO80 families, have been identified in organisms ranging from yeast to humans. By contrast, relatively little is known about how higher-order chromatin structure is regulated and exploited to control gene expression and other nuclear processes.
A major barrier to the identification of factors that regulate higher-order chromatin structure is the difficulty of visualizing the decondensed interphase chromosomes of diploid cells. This barrier can be overcome through the use of Drosophila melanogaster as a model organism. During Drosophila development, many tissues undergo multiple rounds of DNA replication in the absence of cytokinesis, leading to the formation of huge polytene chromosomes containing hundreds of aligned sister chromatids. These transcriptionally active chromosomes are indistinguishable from the interphase chromosomes of diploid cells in most respects. Genetic studies in Drosophila have identified numerous factors that regulate polytene chromosome structure, including ISWI, an ATP-dependent chromatin-remodeling factor. The loss of ISWI function leads to the decondensation of salivary gland polytene chromosomes, possibly due to failure to assemble chromatin containing the linker histone H1 [6], [7], [8]. This striking phenotype led us to investigate the potential involvement of another ATP-dependent chromatin-remodeling factor, Drosophila Mi-2 (dMi-2), in the regulation of higher-order chromatin structure.
dMi-2 functions as the ATPase subunit of multiple chromatin-remodeling complexes, including the NuRD (Nucleosome Remodeling and Deacetylase) complex and dMec (Drosophila MEP-1 containing complex) [9]. NuRD is highly conserved in metazoans and is thought to repress transcription via its chromatin-remodeling and histone deacetylase activities [10], [11], [12]. dMec is the most abundant dMi-2 complex in Drosophila and has been implicated in SUMO-dependent transcriptional repression [13], [14]. Mi-2 plays an important role in cell fate specification in organisms ranging from nematodes to vertebrates. For example, Mi-2 helps maintain the distinction between the germline and soma during C. elegans embryogenesis [15]; regulates the terminal differentiation of B lymphocytes into plasma cells in mammals [16]; and participates in the transcriptional repression of HOX genes by Hunchback and Polycomb in Drosophila [17]. dMi-2 is also required for the efficient expression of heat-shock genes in Drosophila, indicating that its function is not limited to transcriptional repression [18].
Here we report an unanticipated role for Mi-2 in the regulation of higher-order chromatin structure in Drosophila. The loss of dMi-2 function causes salivary gland polytene chromosomes to lose their characteristic banding pattern and appear more condensed than normal. Conversely, the increased expression of dMi-2 in the salivary gland disrupts interactions between sister chromatids and triggers the decondensation of polytene chromosomes. Consistent with these findings, dMi-2 disrupts the association of cohesin with polytene chromosomes. Our studies reveal that dMi-2 is an important regulator of both chromosome condensation and cohesin binding in interphase cells.
As a first step toward investigating the role of dMi-2 in the regulation of higher-order chromatin structure, we investigated phenotypes resulting from either the loss or gain of dMi-2 function. We could not examine the polytene chromosomes of individuals homozygous for loss-of-function dMi-2 alleles since they die relatively early in larval development [17], [19]. We therefore used the GAL4 system [20] to examine the consequences of expressing a dominant-negative form of the dMi-2 protein in the salivary gland. For this study, we used a mutation that deletes amino acids 932-1158 of the dMi-2 protein, including its bipartite ATP-binding site. This deletion eliminates the ATPase activity of dMi-2 in vitro [21] without disrupting its interaction with other proteins (Figure S1). We therefore reasoned that the expression of dMi-2Δ932-1158 should have a strong, dominant-negative effect on dMi-2 function in vivo, as has been observed for comparable mutations in other chromatin-remodeling factors [7], [22], [23].
Transgenic fly strains bearing GAL4-inducible transgenes encoding wild-type (UAS-dMi-2+) or dominant-negative (UAS-dMi-2Δ932-1158) dMi-2 proteins were generated by P element transformation. Transformants were crossed to GAL4 driver lines to generate progeny that express the dMi-2 transgenes in stage or tissue-specific patterns. The GAL4 system is inherently temperature-sensitive; higher levels of transgene expression are observed at higher temperatures [24]. This allowed us to modulate the expression of dMi-2 transgenes by varying the temperature at which progeny were reared.
The expression of dMi-2Δ932-1158 transgenes under the control of a ubiquitously expressed GAL4 driver (da-GAL4) had a dramatic effect on viability. At 25°C, the majority of UAS-dMi-2Δ932-1158 6-5/+; da-GAL4/+ individuals completed larval development, but only 5% survived to adulthood (Table 1). The lethality resulting from the expression of dMi-2Δ932-1158 was moderately enhanced by a hypomorphic allele of dMi-2 (dMi-2f08103) and strongly enhanced by the null allele dMi-24 (Table 1). At 29°C, no individuals expressing dMi-2Δ932-1158 under the control of the da-GAL4 driver survived beyond the first or second larval instar (data not shown). This lethal phase is identical to that resulting from the complete loss of zygotic dMi-2 function [17], [19]. These findings confirmed that the expression of the catalytically inactive dMi-2Δ932-1158 protein has strong, dominant-negative effects on dMi-2 function in vivo.
We also investigated the effect of expressing wild-type dMi-2 protein on Drosophila development using the da-GAL4 driver. As expected, the UAS-dMi-2+ transgene rescued the recessive lethality of dMi-24; 74% of UAS-dMi-2+ 3-3/+; dMi-24 da-GAL4/dMi-24 survived to adulthood (Table 1). Surprisingly, the over-expression of dMi-2+ in a wild-type dMi-2 background resulted in larval lethality; only 12% of UAS-dMi-2+ 3-3/+; da-GAL4/+ individuals survived beyond the third-larval instar (Table 1). The lethality resulting from the expression of the UAS-dMi-2+ transgene was suppressed by a twofold reduction in the level of endogenous dMi-2 function; over 80% of UAS-dMi-2+ 3-3/+; dMi-24/da-GAL4 individuals survived to adulthood (Table 1). Thus, even a modest increase in the level of dMi-2 can be lethal.
The above findings demonstrated that the GAL4 system can be used to analyze phenotypes resulting from either the gain or loss of dMi-2 function. We therefore proceeded to characterize chromosome defects resulting from the expression of wild-type or dominant-negative dMi-2 proteins. We initially used the da-GAL4 driver to induce the expression of UAS-dMi-2Δ932-1158 and UAS-dMi-2+ transgenes in salivary gland nuclei. The increased expression of wild-type dMi-2 protein dramatically altered the structure of salivary gland polytene chromosomes; fixed polytene chromosomes squashes appeared much larger than normal; their banding pattern was severely disrupted; and it was often difficult to distinguish individual chromosome arms (compare Figure 1A and 1B). This phenotype was never observed in wild-type larvae (data not shown) or UAS-LacZ/+; da-GAL4/+ control larvae (Figure 1A). Chromosome defects were even more pronounced in larvae bearing two copies of the UAS-dMi-2+ transgene in addition to the da-GAL4 driver (Figure 1C). Similar results were obtained with a Sgs3-GAL4 driver which is expressed only in the salivary gland of third-instar larvae (Figure S2).
To gain a more accurate impression of the effect of increased dMi-2 expression on chromosome structure, we used live analysis to visualize the chromosomes of larvae expressing a fluorescently tagged core histone (His2Av-GFP). The salivary gland chromosomes of third-instar larvae expressing one or two UAS-dMi-2+ transgenes under the control of the da-GAL4 driver displayed a clear banding pattern but were much larger than normal (compare Figure 1D to 1E and 1F). The increased expression of dMi-2 led to a two to fourfold increase in chromosome volume (Figure 1G). The severe disruption of the chromosome banding pattern evident in fixed polytene chromosome squashes was not observed by live analysis (compare Figures 1B and 1C to 1E and 1F). This difference suggests the over-expression of dMi-2 makes chromosomes unusually sensitive to mechanical stress, leading to their disruption during the squashing procedure. Taken together, these observations suggest that the elevated expression of dMi-2 increases the size of polytene chromosomes while compromising their structural integrity.
We next examined the consequences of expressing the dominant-negative dMi-2Δ932-1158 protein in salivary gland nuclei. A significant fraction of UAS-dMi-2Δ932-1158 6-5/+; da-GAL4/+ individuals survive to adulthood at 25°C; at 29°C, they do not survive beyond the first or second larval instar (Table 1 and data not shown). To reduce dMi-2 function without blocking larval development, individuals of this genotype were shifted from 25°C to 29°C before the end of the first larval instar (0–48 hours after egg laying). Under these conditions, the loss of dMi-2 function disrupted the banding pattern of salivary gland chromosomes and caused them to appear much thinner than normal (compare Figure 2B and 2D to 2A and 2C). Similar results were obtained using a GAL4-regulated transgene encoding a catalytically inactive form of the dMi-2 protein (dMi-2K761R) in which a conserved lysine in the ATP-binding site is replaced with an arginine (MM and AB, unpublished data). Live analysis revealed chromosome defects similar to those observed in fixed chromosome squashes (compare Figure 2F and 2H to 2E and 2G). The ability of dMi-2 to increase the size of salivary gland polytene chromosomes is therefore dependent on its ATPase activity.
Drosophila kismet (KIS) was identified in genetic screens for Polycomb antagonists and subsequently shown to encode a member of the CHD subfamily of chromatin-remodeling factors [25], [26], [27]. The chromosomal distributions of KIS-L and dMi-2 are highly overlapping; both proteins are enriched at less condensed, transcriptionally active regions of chromatin (Figure S5A) [28]. These similarities prompted us to investigate whether the over-expression of KIS-L also alters the structure of salivary gland polytene chromosomes. Transgenic fly strains bearing GAL4-inducible transgenes encoding wild-type KIS-L protein, UAS-KIS-L+, were generated by P element transformation; the da-GAL4 driver was used to drive the ubiquitous expression of these transgenes. As observed for dMi-2, the over-expression of the wild-type KIS-L protein is lethal; UAS-KIS-L+ 20-7/+; da-GAL4/+ individuals failed to survive beyond late larval or early pupal stages of development (data not shown). Unlike dMi-2, however, the increased expression of KIS-L did not cause obvious changes in the structure of salivary gland polytene chromosomes (Figure 1H and 1I). The ability of dMi-2 to alter chromosome structure therefore does not appear to be a property common to all CHD proteins.
The chromosome defects described above resulted from the continuous expression of wild-type or dominant-negative dMi-2 proteins over a five to seven day period. To help distinguish between primary and secondary consequences of dMi-2 over-expression, we used a modification of the GAL4 system that permits the precise temporal regulation of GAL4-responsive transgenes through the use of a temperature-sensitive GAL80 repressor (GAL80ts) [29]. At the permissive temperature (18°C), GAL80ts binds to and inhibits the GAL4 activator. At the restrictive temperature (29°C), GAL4 function is restored, leading to the rapid activation of the GAL4-responsive target gene. The use of this system allowed us to monitor the time course of changes in chromosome structure resulting from increased dMi-2 expression.
UAS-dMi-2+ 3-3/+; UAS-dMi-2+ 15-1/da-GAL4 tubP-Gal80ts individuals were shifted from 18°C to 29°C at the middle of the third larval instar to activate UAS-dMi-2+ expression. RT-PCR was used to monitor the expression of the endogenous dMi-2+ gene, the UAS-dMi-2+ transgenes and the overall level of dMi-2 RNA during the subsequent 24 hours. The expression of the UAS-dMi-2+ transgenes, but not the endogenous dMi-2 gene, was rapidly activated after the shift to the restrictive temperature, leading to roughly a twofold increase in the total level of dMi-2 RNA within seven hours (Figure 3A). The level of dMi-2 RNA remained constant for the remainder of the experiment. Changes in chromosome structure were evident in squashed polytene chromosomes within ten to twelve hours after the shift to 29°C and became increasingly pronounced over the following twelve hours (Figure S3). The relatively short lag time between the increase in dMi-2 expression and alterations in chromosome morphology suggests that dMi-2 may directly regulate higher-order chromatin structure.
Salivary gland polytene chromosomes consist of more than a thousand closely aligned sister chromatids formed by repeated rounds of DNA replication in the absence of cytokinesis. Additional rounds of DNA replication could therefore account for the huge size of salivary gland polytene chromosomes in larvae that express increased levels of dMi-2. To address this possibility, the above experiments were repeated using larvae expressing a core histone tagged with RFP (His2Av-mRFP1); this allowed us to use live analysis to directly compare changes in chromosome volume, DNA content (as estimated by the level of His2Av-mRFP1 fluorescence) and chromosome compaction (the ratio of chromosome volume to DNA content) resulting from increased dMi-2 expression. UAS-dMi-2+ 3-3/His2Av-mRFP1; UAS-dMi-2+ 15-1/da-GAL4 tubP-Gal80ts individuals were shifted from 18°C to 29°C to activate UAS-dMi-2+ expression late in the third larval instar. 27 hours after the shift to 29°C, live analysis revealed a significant increase in polytene chromosome volume, but not His2Av-mRFP1 fluorescence, compared to larvae that had been maintained at 18°C, resulting in almost a twofold decrease in chromosome compaction (Figure 3B–G). Similar results were obtained when we compared the area of salivary gland polytene chromosome squashes to DNA content as estimated by DAPI fluorescence (data not shown). Increased dMi-2 expression thus triggers the decondensation of polytene chromosomes as opposed to additional rounds of replication.
We next examined the effect of the dominant-negative dMi-2Δ932-1158 protein on chromosome compaction using the GAL4-GAL80ts system. UAS-dMi-2Δ932-1158 6-5/His2Av-mRFP1; +/da-GAL4 tubP-Gal80ts individuals were shifted from 18°C to 29°C to activate dMi-2Δ932-1158 expression late in the third larval instar. 48 hours after the shift to 29°C, live analysis revealed significant decreases in both polytene chromosome volume and His2Av-mRFP1 fluorescence relative to larvae maintained at 18°C (Figure 4A–D). A modest increase in chromosome compaction was also observed (Figure 4E); this effect was relatively striking when nuclei with similar DNA content were compared (Figure 4F). These findings provide additional evidence that dMi-2 is an important regulator of chromosome compaction in vivo.
Activation of the hsp70 heat-shock (HS) genes is accompanied by the formation of transcriptionally active HS puffs at loci 87A and 87C, a process that is generally attributed to chromosome decondensation. dMi-2 is recruited to HS genes following their activation [18]. We asked if loss of dMi-2 function would interfere with HS-induced chromosome puffing. We subjected third-instar larvae to a 20 minute HS at 37°C and visualized puff formation at 87A and 87C by indirect immunofluorescence using a RNA Pol II antibody. Under these conditions, all the polytene chromosomes of control larvae exhibited strong (67%) to moderate (33%) decondensation at 87A and 87C as judged by a loss of DAPI signal and an accumulation of RNA polymerase II (Figure 5, upper panels). By contrast, the extent of puff formation and the amount of polymerase II signal was significantly reduced in transgenic larvae expressing catalytically inactive dMi-2 (dMi-2K761R, raised at 25°C prior to heat-shock; Figure 5, lower panels); no puffs or only moderate puffing was observed in 14 and 53% of the chromosomes of these larvae, respectively. This result suggests that dMi-2 function is important for a local and rapid chromosome decondensation event.
The reduced expression of histone H1, a linker histone that promotes the formation of 30 nm fibers, causes polytene chromosome defects that are remarkably similar to those resulting from the increased expression of dMi-2 [8], [30]. This similarity suggested that dMi-2 might regulate chromosome structure by antagonizing the assembly of chromatin containing histone H1. To investigate this possibility, we examined whether the over-expression of dMi-2 decreases the association of histone H1 with chromatin. The live analysis of larvae expressing GFP-tagged histone H1 revealed that the elevated expression of dMi-2 increased the size of polytene chromosomes relative to control larvae without reducing the levels of bound histone H1 (Figure 6A–C). The over-expression of dMi-2 also did not reduce ISWI expression or the level of histone H1 in chromatin extracted from salivary glands as assayed by Western blotting (Figure 6D; Figure S4). The expression of dominant-negative dMi-2 in salivary gland nuclei caused a slight decrease in histone H1 levels (Figure S4). Thus, dMi-2 does not promote chromosome decondensation by repressing histone H1 expression or assembly.
The disruption of interactions between aligned chromatids could also account for the changes in polytene chromosome structure caused by elevated dMi-2 expression. To test this hypothesis, we used the LacO/LacI-GFP system [31] to monitor the effect of dMi-2 over-expression on the structure of a specific locus by live analysis. In this system, a GFP-tagged Lac repressor (LacI-GFP) binds to tandem copies of Lac operator (LacO) sequences inserted at a single chromosomal location (Figure 7A). This assay has been successfully used to detect disruption of the normal parallel alignment of the chromatids in polytene chromosomes [32]. For this study, we used an insertion of a LacO transgene in an interband on the second chromosome and a transgene encoding LacI-GFP repressor under the control of the Hsp83 promoter. Control UAS-dMi-2+ 3-3/LacO LacI-GFP; UAS-dMi-2+ 15-1/da-GAL4 tubP-Gal80ts individuals were reared at 18°C to the end of the third larval instar (240 hours) and heat-shocked for 60 minutes at 37°C to activate LacI-GFP expression. To over-express dMi-2, individuals of the same genotype were reared at 18°C for (9 days); shifted to 29°C for 27 hours to activate the UAS-dMi-2+ transgenes; and heat-shocked for 60 minutes at 37°C to activate LacI-GFP expression. Polytene chromosomes of control larvae showed a compact band of GFP fluorescence signal perpendicular to the chromosome axis (Figure 7B and D). In larvae expressing elevated levels of dMi-2, however, the GFP signal was dispersed, with dots of LacI-GFP fluorescence scattered over a fivefold larger area (Figure 7C and E). This striking phenotype suggests that dMi-2 may play an unanticipated role in chromosome cohesion in vivo.
Cohesin is essential for sister chromatid cohesion in eukaryotic cells, and also plays important roles in gene expression, DNA repair, genome stability and chromatin organization [33], [34], [35], [36]. Four subunits of cohesin–Smc1, Smc3, Rad21 (Scc1/Mcd1) and stromalin (SA/Scc3/Stag2)–form a ring-like structure that encircles chromosomes from telophase until mitosis [33], [37]. Cohesin is loaded onto chromosomes by the kollerin complex containing the Nipped-B adherin protein, and is removed by the releasin complex and the separase protease to allow chromosome separation during anaphase [38].
As a first step toward investigating potential interactions between dMi-2 and cohesin, we compared the distributions of dMi-2; the SA cohesin subunit, and the Nipped-B adherin on salivary gland polytene chromosomes of wild-type larvae. Interestingly, the distributions of both SA and Nipped-B were nearly identical to that of dMi-2 (Figure 8B and C). All three proteins were preferentially associated with less condensed, transcriptionally active regions of polytene chromosomes, as evidenced by the co-localization of dMi-2 and elongating RNA polymerase II (Pol II Ser2) (Figure 8A and Figure S5 A–F) [28].
We used genetic experiments to test the possibility that Nipped-B and dMi-2 functionally interact, as suggested by their co-localization. Heterozygous null Nipped-B407 mutations reduce adult wing blade size relative to wild-type alleles, including the wild-type allele on the parental chromosome (P57B) on which Nipped-B407 was induced (Figure 8D) [39]. The reduced wing size mirrors overall reduced body size, similar to that seen in mouse and human Nipped-B (Nipbl, NIPBL) heterozygotes (MG, DD unpublished) [40], [41], [42]. The heterozygous dMi-24 mutation suppresses the reduced wing size caused by heterozygous Nipped-B407. This genetic interaction suggests that dMi-2 antagonizes the function of the Nipped-B cohesin loading factor.
We examined whether the over-expression of dMi-2 alters polytene chromosome structure by altering the expression of cohesin subunits. The over-expression of dMi-2 for twenty-four hours using the GAL4-GAL80ts system caused the dramatic decondensation of polytene chromosomes, but did not reduce the level of the Smc1 protein as assayed by western blotting (Figure 9A). The over-expression of dMi-2 slightly increased Smc1, SA, and Rad21 RNA in the salivary gland, as assayed by RT-PCR (Figure 9B–D). Although the cause of this increase is not known, it is clear that reduced cohesin expression is not responsible for dMi-2-dependent chromosome decondensation.
We next investigated whether dMi-2 antagonizes binding of cohesin to polytene chromosomes. Fluorescence recovery after photobleaching (FRAP) assays were used to analyze interactions between Smc1 and polytene chromosomes in third-instar larvae that express modest levels of EGFP-tagged Smc1 in addition to normal levels of endogenous Smc1 [43]. A previous study defined the existence of three dynamic forms of the cohesin complex in salivary glands: unbound, weakly bound and stably bound to chromosomes, with short and long equilibration half-lives for the weak and stable forms, respectively [43]. Moderate over-expression of dMi-2 with two transgenes at 25°C had a dramatic impact on the association of cohesin with polytene chromosomes. In UAS-dMi-2+ 3-3/EGFP-Smc1; UAS-dMi-2+ 15-1/da-GAL4 larvae, there was a much faster equilibration between the bleached and the unbleached halves of the nucleus (Figure 10A), and the chromosomal half-life of the stable binding form decreased from 425 to 173 sec (Figure 10B). In addition, the fraction of stable EGFP-Smc1 binding decreased from 13% of the total to 7%, with a corresponding increase in the unbound fraction (Figure 10C). As expected, the nuclei with moderate dMi-2 overexpression were 1.6-fold larger in volume (9,784±627 µm3) on average than the control nuclei (5,986±337 µm3). Given the unusually long chromosomal residence time of stable cohesin, nuclear volume does not significantly affect the measured residence time through recapture of released cohesin [43]. A very large volume increase would be expected to increase recapture and cause an apparent increase in the measured chromosomal half-life instead of the observed decrease. These findings support the idea that dMi-2 promotes the dissociation of cohesin from polytene chromosomes. Reducing Nipped-B gene dosage similarly decreases the amount of stable cohesin, but in contrast to dMi-2 overexpression, does not alter its chromosomal residence time [43]. Thus it is unlikely that dMi-2 promotes cohesin dissociation by reducing Nipped-B activity.
To investigate whether dMi-2 affects an aspect of chromatin organization unique to polytene chromosomes, we examined whether its increased expression alters the structure of mitotic chromosomes in wing imaginal discs, a diploid tissue that is particularly well suited for cytological studies. The elevated expression of dMi-2 in UAS-dMi-2+ 3-3/+; UAS-dMi-2+ 15-1/da-GAL4 larvae caused mitotic chromosomes to appear much less compact than normal (data not shown) without altering the mitotic index (1.4 vs. 1.3 in control larvae). To characterize this phenotype in more detail, we examined the metaphase chromosomes of imaginal discs treated with colchicine to induce mitotic arrest prior to fixation. In colchicine-treated wing discs of larvae over-expressing dMi-2, 20% of the metaphase chromosomes appeared disorganized and less compact than normal (compare figure 11B and C to A). Similar defects were not observed in colchicine-treated imaginal discs of control larvae. Roughly one-third of the abnormal metaphase chromosomes were also significantly longer than normal (third chromosome length = 3.3±0.4 µm (n = 43) vs. 5.4±1.0 µm (n = 63); compare figure 11C to A). These striking defects indicate that dMi-2 plays a general role in the regulation of higher-order chromatin structure in both diploid and polytene cells.
Several lines of evidence suggest that the chromosome decondensation resulting from the increased expression of the dMi-2 protein reflects its normal function. The inducible transgene used in our studies rescued the recessive lethality of a dMi-2 null allele indicating that it encodes a fully functional dMi-2 protein. Even a modest (two to threefold) increase in dMi-2 levels triggered chromosome decondensation; this effect was relatively rapid and was dependent on dMi-2 ATPase activity. The expression of a dominant-negative dMi-2 protein had the opposite effect on chromosome structure; the loss of dMi-2 function made polytene chromosomes appear more condensed than normal and blurred the distinction between bands and interbands. Taken together, these observations provide strong evidence that dMi-2 promotes chromosome decondensation in vivo.
HS puff formation exemplifies an unusually rapid and localized chromatin decondensation event. The precise mechanisms driving puff formation are not clear. HS gene activation is accompanied by nucleosome depletion, activation of PARP, extensive PARylation and the recruitment of topoisomerase, HSF, RNA polymerase II and several elongation factors as well as dMi-2 [18], [44], [45], [46]. Expression of dominant-negative dMi-2 protein reduces the transcriptional activation of HS genes but the underlying mechanisms have not been resolved [18]. Here, we have found that expression of a dominant negative dMi-2 protein leads to the formation of HS puffs with reduced size. This suggests that dMi-2 effects on chromosome structure are not always global. Instead, our results indicate that dMi-2 can be recruited to specific chromatin regions by environmental stimuli to contribute to the rapid local decondensation of chromatin. It is conceivable that the compromised decondensation of HS loci in dMi-2 mutants is one of the reasons for the observed loss in transcriptional output [18].
To date, only one other Drosophila chromatin-remodeling factor, ISWI, has been implicated in the regulation of higher-order chromosome structure [7], [47], [48]. ISWI promotes the association of the H1 linker histone with interphase chromosomes [6], [8]. The loss of ISWI function leads to chromosome decondensation accompanied by the loss of histone H1, and the elimination of histone H1 causes chromosome defects that are remarkably similar to those resulting from dMi-2 over-expression [6], [8]. Although this similarity suggested that dMi-2 and ISWI have antagonistic effects on higher-order chromatin structure and histone H1 assembly, we failed to detect decreases in ISWI and histone H1 expression or histone H1 assembly following dMi-2 over-expression. dMi-2 and ISWI therefore appear to regulate distinct aspects of higher-order chromatin structure.
Cohesin has been the topic of intensive study due to its critical role in sister chromatid cohesion during mitosis, and its roles in gene regulation and DNA repair. The complex forms a ring-like structure that encircles chromosomes beginning in telophase, and mediates sister chromatid cohesion upon DNA replication [33]. Cohesin binding is dynamic, but unusually stable compared to most DNA-binding proteins. Interphase cohesin is continuously loaded by the kollerin complex containing Nipped-B and released from chromosomes by the releasin complex containing Pds5 and Wapl [38], [43]. Our studies revealed an intriguing connection between dMi-2 and the cohesin complex, and argue that dMi-2 facilitates removal of cohesin from chromosomes during interphase. This activity is not restricted to situations in which dMi-2 is expressed at unusually high levels, since a twofold reduction in dMi-2 dosage counteracts the developmental consequences of reduced dosage of Nipped-B. Our findings add dMi-2 to the list of factors that regulate cohesin binding.
Cohesin regulates transcription by multiple mechanisms, including long-range interactions between insulators, enhancers and promoters via the formation of DNA loops, repression in collaboration with Polycomb proteins, and controlling transition of paused polymerase to elongation [34], [36], [49]. The observed suppression of a dominant Nipped-B mutant phenotype by reduced dMi-2 gene dosage suggests that regulation of cohesin chromosome binding may be one mechanism by which dMi-2 controls gene expression.
The live analysis of a LacO array tagged with GFP in living cells is consistent with a potential role for dMi-2 in chromosome cohesion. The array is organized in a compact disc due to cohesion between precisely aligned chromatids. The over-expression of dMi-2 caused the LacO array to disperse into hundreds of discrete foci, presumably due to the disruption of interactions between sister chromatids. The over-expression of dMi-2 also disrupted the organization of mitotic chromosomes along their longitudinal axes, possibly by interfering with chromosomal interactions in cis that contribute to the organization of chromosome shape [50], [51].
Our findings show that dMi-2 plays unanticipated roles in both the regulation of higher-order chromosome structure and cohesin dynamics. Is there a causal relationship between the two activities? The sudden removal of cohesin in late larval development by targeted proteolysis does not dramatically alter polytene structure [52] and thus cohesin may not be critical for maintenance of polytene structure once fully established. However, genetic studies of pds5 have revealed a role for both cohesin binding and sister chromatid cohesion in forming the normal structure of polytene chromosomes [53]. A pds5 null allele and an allele encoding an N-terminally truncated protein alter polytene chromosome structure in distinctive ways, but in both cases the size and normal banding pattern are disrupted [53]. Taken together, the above considerations prevent us from concluding that dMi-2 promotes chromosome decondensation by destabilizing cohesin binding. However, because dMi-2 over-expression causes a large reduction in both the amount of stable cohesin and its chromosomal residence time, we can conclude that cohesin binding has been reduced, and that it is also likely that cohesion is affected.
The internal diameter of cohesin is ∼35 by 50 nm; it can therefore encircle only one 30 nm or two 10 nm chromatin fibers [54]. Interactions between cohesin complexes are thought to contribute to chromatid cohesion and presumably anchor chromatin loops to form “hubs” of high transcriptional activity [55], [56], [57]. The destabilization of cohesin binding therefore may be a secondary consequence of changes in chromatin structure catalyzed by dMi-2. Further work will be necessary to test this possibility and clarify the causal relationship, if any, between changes in chromosome structure and cohesin binding catalyzed by dMi-2.
It is intriguing that dMi-2, an antagonist of cohesin binding and well-characterized transcriptional repressor, co-localizes with cohesin at sites of active transcription. Although cohesin subunits and Nipped-B were not identified as stable subunits of dMi-2 containing complexes in cultured cells [13], [58], the extensive overlap between their chromosomal distributions suggests that chromatin structure and gene activity may be dependent on a fine balance of opposing dMi-2 and cohesin activities. Cohesin selectively binds and regulates active genes that have paused RNA polymerase, and can both positively and negatively regulate these genes by multiple mechanisms, including controlling the transition of paused polymerase to elongation [49]. It is possible that dMi-2 may also influence this transition by regulating cohesin binding and the chromatin structure at the pause sites. Intriguingly, mouse Mi-2ß and the NuRD complex bind active and poised gene promoters in thymocytes, and have both negative and positive effects on expression of these genes [59]. The Mi-2/NuRD complex regulates the expression of genes involved in lymphocyte differentiation [16], [60] and is also involved in stem cell renewal and determination [61], [62]. As in Drosophila, mammalian cohesin also regulates many genes critical for growth and development [34]. Our findings raise the interesting possibility that Mi-2 may regulate cellular differentiation in vertebrates by modulating chromosome condensation and cohesin activity.
Flies were raised on cornmeal, agar, yeast and molasses medium, supplemented with methyl paraben and propionic acid. Mutations, chromosome aberrations, strains and abbreviations used in this study are described in Table S1 or Flybase (www.flybase.org) unless otherwise indicated. The amorphic dMi-24 allele is a frameshift mutation that blocks the production of functional dMi-2 protein [17]. The hypomorphic dMi-2f08104 allele contains a PiggyBac transposon insertion in the first intron of the dMi-2 gene. The GAL4 and GAL4/GAL80ts system and sgs-GAL4, ey-GAL4 and da-GAL4 drivers were used to activate the expression of UAS transgenes (Table S1) [20], [29], [63].
DNA fragments encoding C-terminal Flag-tagged dMi-2+ and dMi-2Δ932-1158 proteins were amplified from dMi-2 cDNA clones [21] by PCR using the primers described in Table S2, subcloned into the NotI and XbaI sites of pVL1392, and transferred into the P-element transformation vector pUAST. To generate a GAL4-regulated transgene encoding N-terminal Flag-tagged KIS-L, we first subcloned the DNA fragment generated by the hybridization of two oligonucleotides shown in Table S2 into the EcoR1 and Not1 sites of pUAST to generate the pUAST-NFLAG vector. A 2.3 kb DNA fragment containing a Not1 site immediately upstream of the KIS-L start codon was amplified from the kis2 cDNA [27] by PCR using the primers shown in Table S2. The resulting 2.3 kb Not1-EcoR1 fragment, together with the partially overlapping kis2, kis30 and kis40a cDNA clones [27], was used to generate an approximately 17 kb Not1-Kpn1 fragment containing the entire KIS-L coding region and subcloned into the Not1 and Kpn1 sites of pUAST-NFLAG. To generate a GAL4-regulated transgene encoding C-terminal GFP-tagged histone H1, a fragment was amplified from Drosophila genomic DNA using the primers shown in Table S2, subcloned into the EcoR1 and Xho1 sites of pENTR1A and transferred into the P-element transformation vector pTWG [20], [64], [65], [66]. Transformants were generated using by P-element-mediated transformation using the Df(1)w67c2 strain [67]. Homozygous viable transformants used in this study include UAS-dMi-2+ 3-3, UAS-dMi-2Δ932-1158 6-5, UAS-KIS-L+ 20-7 and UAS-H1-GFP 2-1 on the second chromosome and UAS-dMi-2+ 15-1 on the third chromosome (Table S1).
To determine the viability of individuals bearing dMi-2 mutations and transgenes, progeny of the crosses shown in Table 1 were scored. For each cross, ∼400 embryos were collected on grape juice/agar plates, transferred to vials and cultured at 25°C. Percent survival to each stage was calculated as the ratio between the number of individuals counted and the number of individuals expected for each genotype. The data for each of the crosses shown in Table 1 represent the average of at least two experiments.
To analyze polytene chromosomes, salivary glands of late third-instar larvae were dissected in 0.7% NaCl and fixed in 1.85% formaldehyde/45% acetic acid as previously described [6]. Mitotic chromosomes of wing imaginal discs were examined as described previously for Drosophila neuroblasts [68].
To analyze the effect of dMi-2Δ932-1158 expression on chromosome structure, the progeny of da-GAL4 females crossed to UAS-dMi-2Δ932-1158 6-5 males were shifted from 25°C to 29°C 48 hours after egg deposition. To analyze the effect of dMi-2 over-expression on chromosome structure, da-GAL4 females, Sgs3-GAL4 females or ey-GAL4 females were crossed to UAS-dMi-2+ 3-3 males, UAS-dMi-2+ 15-1 males, or UAS-dMi-2+ 3-3 UAS-dMi-2+ 15-1 males; the resulting progeny were raised at 29°C. To analyze the effect of Kismet over-expression on chromosome structure, the progeny of ey-GAL4 females crossed to UAS-KIS-L+ 20-7 males were raised at 25°C. To analyze the effect of dMi-2K761R expression on puff formation, third-instar larvae from da-GAL4 females crossed to UAS-dMi-2K761R males were shifted from 25°C to 37°C to induce the heat shock. Heat shock puff formation was induced by incubating two to four third-instar larvae in a 0.2 ml PCR tube for 20 minutes at 37°C; salivary glands were immediately dissected from the larvae in 0.7% NaCl pre-warmed to 37°C. Polytene chromosomes were fixed, squashed and stained with DAPI or antibodies [6].
Fixed preparations of polytene and mitotic chromosomes were analyzed with a Zeiss Axioskop 2 plus fluorescent microscope equipped with an AxioCam HRm CCD camera and images were acquired with a HCX PL APO CS 63.0×1.40 OIL objective and Axiovision 4.6.3 software [Zeiss]. For DNA quantification, images of DAPI stained chromosomes were captured using identical exposure times. Volocity 5.4 software (Perkin Elmer) was used to calculate the sum of pixel intensities within chromosomes boundaries and to calculate the Pearson's correlation coefficient for the co-localization of dMi-2 with Stromalin and Nipped-B.
Live analysis of polytene chromosomes was performed as previously described [8] in larvae bearing transgenes encoding GFP or mRFP1-tagged histone H2Av [69]. The level of histone H1 associated with chromosomes was analyzed using a transgene encoding GFP-tagged histone H1 (UAS-H1-GFP; Table S1) under the control of the ey-GAL4 driver. Volocity software (release 5.4, Perkin Elmer) was used for three-dimensional reconstructions. For volume calculation we imaged sections of polytene nuclei every 0.5 µm. The change in chromatin compaction was established by calculating the ratio of nuclear volume to DNA content. The ratio for control samples was normalized to one.
Antibodies used in this study include rabbit polyclonal antibodies against dMi-2 (dMi-2C and dMi-2N) [70], Drosophila histone H1 (dH1 S72) [71], rabbit polyclonal Drosophila ISWI 196 [72], rabbit polyclonal Drosophila Rpd3 [70], Drosophila Smc1 [53], and histone H3 (Abcam, ab1791); a mouse monoclonal antibody against RNA PolIIoser2 (H5, Covance); and guinea pig antibodies against Stromalin and Nipped B [53], [73]. The preparation of salivary gland chromatin extracts, SDS-PAGE and protein blotting were performed as previously described [6]. Baculoviral expression of dMi-2, dMi-2Δ932-1158, CAF1p55, dp66 and dMBD2/3 in Sf9 cells, extract preparation and anti-FLAG co-immunoprecipitation was carried out as described [13].
UAS-dMi-2+ 3-3/+; UAS-dMi-2+ 15-1/da-GAL4 Gal80ts and UAS-dMi-2Δ932-1158 6-5/+; +/da-GAL4 Gal80ts individuals were shifted from 18°C to 29°C at the middle of the third larval instar to activate UAS-dMi-2+ and UAS-dMi-2Δ932-1158 expression, respectively. The time course of changes in chromosome structure following the shift to 29°C was monitored via both live analysis and the analysis of polytene chromosome squashes as described above. RNA levels in the salivary gland of UAS-dMi-2+ 3-3/+; UAS-dMi-2+ 15-1/da-GAL4 Gal80ts individuals were monitored by RT-PCR using primers specific to the RNAs encoded by Smc1, SA, Rad21, the endogenous dMi-2+ gene or the UAS-dMi-2+ transgene. Primers from the dMi-2 coding region were used to measure the overall increase in total dMi-2 RNA levels (Table S2). RNA was isolated from salivary glands of ten UAS-dMi-2+ 3-3/+; UAS-dMi-2+ 15-1/da-GAL4 Gal80ts female larvae at specific times after the shift to 29°C and from ten control larvae maintained at 18°C using NucloSpin RNA XS (Clontech). cDNA was synthesized using the Superscript III First-Strand Synthesis System for RT-PCR kit (Invitrogen) and quantified with a nano-spectrophotometer using the NANODROP software. Identical amounts of cDNA were used for the PCR amplification of each sample.
The progeny of LacI-GFP LacO; da-GAL4 GAL80ts females crossed to UAS-dMi-2+ 3-3; UAS-dMi-2+ 15-1 males were allowed to develop to the third larval instar at 18°C (eight days). The expression of dMi-2 transgenes was activated by raising the temperature to 29°C for 27 hours, after which larvae were heat-shocked for 1–2 hours at 37°C to express LacI-GFP. After one hour of recovery at 29°C, salivary glands were dissected and polytene chromosomes were subjected to live analysis as described above. 0.2 µm sections of nuclei were imaged by confocal microscopy.
Wings of adult progeny of crosses reared under uncrowded conditions at 25°C were mounted on microscope slides in Permount (Fisher Scientific) under coverslips. Digital images of the mounted wings were obtained using a 4X objective on a Nikon Microphot microscope (calibrated using a slide micrometer) and the blade areas measured using NIH ImageJ software (rsbweb.nih.gov/ij/). Statistical calculations and boxplots were generated using SSP software (economics files.pomona.edu/StatSite/ssp.html).
The EGFP-Smc1 fusion protein and its functional characteristics were described previously [43]. Fluorescence recovery after photobleaching (FRAP) and data analysis was performed as previously described using salivary glands from late third-instar UAS-dMi-2+ 3-3/Chip-EGFP-Smc1; UAS-dMi-2+15-1/da-GAL4 larvae raised at 25°C [43].
|
10.1371/journal.pcbi.1006340 | An enormous potential for niche construction through bacterial cross-feeding in a homogeneous environment | Microorganisms modify their environment by excreting by-products of metabolism, which can create new ecological niches that can help microbial populations diversify. A striking example comes from experimental evolution of genetically identical Escherichia coli populations that are grown in a homogeneous environment with the single carbon source glucose. In such experiments, stable communities of genetically diverse cross-feeding E. coli cells readily emerge. Some cells that consume the primary carbon source glucose excrete a secondary carbon source, such as acetate, that sustains other community members. Few such cross-feeding polymorphisms are known experimentally, because they are difficult to screen for. We studied the potential of bacterial metabolism to create new ecological niches based on cross-feeding. To do so, we used genome scale models of the metabolism of E. coli and metabolisms of similar complexity, to identify unique pairs of primary and secondary carbon sources in these metabolisms. We then combined dynamic flux balance analysis with analytical calculations to identify which pair of carbon sources can sustain a polymorphic cross-feeding community. We identified almost 10,000 such pairs of carbon sources, each of them corresponding to a unique ecological niche. Bacterial metabolism shows an immense potential for the construction of new ecological niches through cross feeding.
| Biodiversity can emerge in a completely homogeneous environment from populations with initially genetically identical individuals. This striking observation comes from experimental evolution of bacteria, which create new ecological niches when they excrete nutrient-rich waste products that can sustain the life of other bacteria. It is difficult to estimate the potential of any one organism for such metabolic niche construction experimentally, because it is challenging to screen for novel metabolic abilities on a large scale. We therefore used experimentally validated models of bacterial metabolism to predict how many novel niches organisms like Escherichia coli can construct, if a novel niche must be able to sustain a stable community of microbes that differ in the nutrients they consume. We identify thousands of such niches. They differ in their primary carbon source and a secondary carbon source that is excreted by some microbes and used by others. Because we restricted ourselves to chemically simple environments, we may even have underestimated the enormous potential of microbes for niche construction.
| With as many as one trillion predicted species, microbial diversity on our planet is enormous [1]. To understand the origins of biological diversity in general and microbial diversity in particular is a central goal of ecology and evolutionary biology. For many decades, most biological diversity was thought to arise in allopatry, that is, when populations become physically subdivided [2]. More recently, biologists have increasingly accepted that populations can also diversify in sympatry, that is, without any physical barriers [3–9]. Examples of sympatric diversification include insect populations that adapt evolutionarily to different plant hosts [9], stickleback populations that evolve reproductive isolation at least partly in sympatry [6], Midas cichlid populations that originated in a small volcanic crater lake in Nicaragua [7], and bacteriophage lambda that specializes on different bacterial hosts [8]. In bacteria, sympatric divergence has been observed both in nature [10,11] and during experimental evolution [12–15].
Sympatric diversification is easiest in heterogeneous environments [16,17]. Because such environments provide multiple ecological niches, organisms can easily diversify when they specialize and adapt to these niches. Such diversity can then be maintained according to the niche exclusion principle–the principle states that different organisms cannot occupy the same niche [18]. Examples include the spatial structure of an unshaken growth medium, which facilitates morphological diversification in experimental evolution of Pseudomonas fluorescens [12]; spatial (free-living or particle-associated) and temporal (spring and fall) resource partitioning, which triggers sympatric speciation in bacterioplankton [19]; the divergence that occurs as a result of host shifts from hawthorn to domestic apples in apple maggot flies [9]; as well as the specialization of bacteriophages to Escherichia coli expressing different membrane proteins [8,9].
In apparent contradiction to the niche exclusion principle, sympatric diversification can also occur in homogeneous environments [6,7,10,11,13,14,20]. Perhaps the most striking example involves stable genetic polymorphisms that can originate in E. coli populations cultured in the homogeneous and well-mixed environment of a batch culture or a chemostat, a device in which a cell culture is kept in a constant nutrient environment by continually supplying it with nutrient medium [13,14,20–23]. For example, over a mere 800 generations of laboratory evolution in a glucose-limited chemostat, initially isogenic populations of E. coli can diversify into multiple genetically different strains [13,21–23]. These strains stably coexist in the chemostat as a result of cross-feeding [22].That is, one strain consumes the primary carbon source glucose and excretes a secondary carbon source (acetate or glycerol), whereas the other strain feeds on the secondary carbon source. These phenotypic differences result from regulatory DNA mutations in transcription factors and cis-regulatory regions. They include a cis-regulatory mutation affecting the expression of acetyl CoA synthetase, an enzyme that catalyzes the transformation of acetate to acetyl CoA, which enters the tricarboxylic acid cycle to produce energy. They also include a structural mutation in the glycerol-3-phosphate repressor, which can result in constitutive expression of glycerol utilization genes [21]. Experiments like this suggest that E. coli may readily diversify genetically and metabolically in a completely homogeneous environment.
The emergence of cross-feeding is an example of niche construction, a process where organisms change their environment in ways that can affect the evolutionary dynamics of themselves and of other organisms [24–29]. Prominent examples of niche construction include animals that construct artifacts such as webs, nests and burrows [24]; earthworms and plants that alter the fertility, humidity and chemical composition of soil [28,30,31]; and bacteria that construct biofilms and excrete antibiotics as well as metabolic by-products [32]. Constructed niches can affect evolution even on the short time scales of experimental evolution, where populations of Pseudomonas fluorescens become dependent on their own modifications of their chemical environment [33,34].
The origin of new niches associated with bacterial cross-feeding is not easy to detect experimentally: Except for differences in colony morphology, cross-feeding polymorphisms generally lack phenotypes that are both macroscopically visible and highly specific. However, computational analysis can help predict the conditions under which cross-feeding polymorphisms can originate and persist. Some authors use small biochemical networks to search for the conditions that promote genetic diversification through cross-feeding interactions [35–40]. Others use digital organisms with evolvable genomes and metabolic networks [41]. Yet others simulate individuals in an evolving population where random mutations can change nutrient consumption rates in a model of E. coli central carbon metabolism, and show that glucose-acetate cross-feeding can originate in such a population. Most recently, a genome scale metabolic network of E. coli was used to study cross-feeding and other metabolic dependencies that emerge as a result of evolution under gene loss [42] or amino-acid leakage [43].
Here we go beyond this work and evaluate the general potential for the construction of new niches associated with cross-feeding that is inherent to the metabolism of E. coli and to complex metabolic systems in general. That is, we ask how many different kinds of ecologically stable cross-feeding interactions can emerge in an initially homogeneous population, where one bacterial strain feeds on a primary carbon source and produces a secondary carbon source that sustains the other strain. To answer this question, we take advantage of a well-studied and experimentally validated [44] genome-scale model of E. coli metabolism. We use Flux Balance Analysis (FBA), an experimentally validated computational technique [45], to characterize the production of secondary carbon sources that can help cross-feeding polymorphisms emerge. We then use dynamic flux balance analysis (dFBA) [46,47], a variant of FBA that uses genome-scale metabolic information to predict the ecological dynamics of microbial communities and how they change their chemical environment over time. We use dFBA to study the conditions under which two cross-feeding strains can establish a stable community in a chemostat. After having reproduced the experimentally observed glucose-acetate cross-feeding polymorphism [13], we then identify additional pairs of primary and secondary carbon sources that can lead to the establishment of stable cross-feeding communities. We find thousands of such pairs, both in E. coli and other metabolic reaction networks of similar complexity. Our work demonstrates the great potential of metabolic systems to construct new ecological niches.
Our first analysis prepares the ground by examining the conditions under which a glucose-acetate cross-feeding polymorphism can be stably maintained by two E. coli strains. We studied this specific polymorphism, because it is experimentally well documented [13,21–23], and aimed to reproduce it in silico. Specifically, we simulated the dynamic of a community composed of two cross-feeding E. coli strains (or ecotypes [48]), a producer strain P that produces a secondary carbon source as a by-product of feeding on some primary carbon source, and a consumer strain C that consumes this secondary carbon source. We use the same genome scale metabolic network of E. coli iJO1366 [44] to model both strains. This means that the metabolic networks of both strains comprise exactly the same reactions and metabolites. This modeling decision reflects the observation that cross-feeding strains can emerge from a single E. coli ancestor in little evolutionary time [22]. The metabolic differences between cross-feeding strains do not result from differences in their complement of enzyme-coding genes, but from regulatory mutations that affect how much of a specific carbon source each strain can consume or produce [49].
We model these differences phenomenologically, through differences in the flux through two specific reactions in strains P and C. Specifically, we model the secondary carbon source production of strain P by imposing a non-zero production flux pscs,P for this carbon source via the exchange reaction that transports the secondary carbon source out of the cell. And we model the secondary carbon source consumption of strain C by limiting the strain’s primary carbon source consumption. This modeling decision is motivated by the experimental observation that when cross-feeding emerges in E. coli [22], the consumer strain’s ability to consume its primary carbon source becomes impaired. One might argue that increasing the consumption of the secondary carbon source might be biologically more sensible. However, the two approaches are equivalent. Here is why. Since we simulate a chemostat culture, once steady state is reached, the strains in the chemostat grow at a constant rate. The consumer strain C achieves this growth rate by consuming both primary and secondary carbon sources. If consumption of the primary carbon source increases, consumption of the secondary carbon source becomes reduced by an equivalent amount (Eq 3, S4 Text), such that the steady state is unaffected.
For the well-studied cross-feeding interaction of acetate producer and consumer strains, regulatory mutations in specific genes are known to bring forth the metabolic behavior of producer and consumer strains [21,23,50]. Since the objective of our work was to study not only glucose-acetate cross-feeding but multiple other cross-feeding interactions we decided not to incorporate assumptions about specific mutations in specific genes into our model. By imposing general constraints on the production and consumption of specific carbon sources, we allowed for the possibility that our modeled strains could achieve these constraints in different ways, depending on the carbon source considered. The specific mutations that may underlie our strains’ metabolic behavior will be the subject of future work.
In the first part of our analysis, we focus on glucose as a primary carbon source, and on acetate as a secondary carbon source. The secondary carbon source is excreted by the producer strain P at a rate pac,P, and consumed by the consumer strain C (Fig 1A). We initially assume that the consumer strain C cannot consume glucose (cglc,C = 0), an assumption that we relax below (S3 Text). To find out whether both strains can coexist in a stable chemostat community, we first use Flux Balance Analysis [45](FBA, Methods) in the form of dynamic FBA [46,47] (Methods).
To mimic typical experimental conditions, we performed all simulations with a dilution rate D, the rate at which culture is replaced with fresh medium, of D = 0.2 h-1[13]. At this dilution rate, the maximum rate at which E. coli cells can produce acetate (pacmax) without being eventually flushed out from the chemostat is 50.3 mmol gDW-1 h-1 (S2 Text). (Here and below, all units of metabolic flux are given in mmol gDW-1 h-1). To ensure survival of the producer strain P, we simulated chemostat dynamics at an acetate production rate pac,P that is equal to 5% of this maximum (2.6 mmol gDW-1 h-1). We initialized the chemostat in the presence of only the acetate producing strain P, and once this strain had reached steady-state, which occurred after no more than 50 hours, we introduced the acetate consuming strain C. We then monitored the joint dynamics of both strains until they had reached steady-state or until one strain had gone extinct.
Fig 1B shows the change in biomass of P and C over time. Only three carbon-containing metabolites–glucose, acetate, and carbon dioxide–change their concentration (Fig 1C).The concentration of glucose (Fig 1C, grey) decreases as the acetate producer P consumes glucose and this decrease is concurrent with an increase in P's biomass (Fig 1B, blue). Strain P metabolizes glucose partially to carbon dioxide (Fig 1C, green) and partially to acetate (Fig 1C, black), which is why the concentration of both metabolic by-products increases. Once the acetate consumer strain C is introduced into the chemostat at 50 hours (Fig 1B, red), the chemical environment contains a substantial amount of acetate, which strain C metabolizes to carbon dioxide to synthesize biomass. By 100 hours, the system has reached a new steady state, in which a stable polymorphism of the acetate producer (P) and the acetate consumer (C) strain is maintained as a result of their cross-feeding interaction.
We next wanted to find out how the population’s behavior changes if the amount of acetate excreted by the producer strain P varies. We thus varied the acetate production rate pac,P up to the maximum beyond which the producer goes extinct. Not surprisingly, the steady-state biomass of strain P is reduced as its acetate production increases (Fig 1D, blue), because of the metabolic cost incurred by acetate production. In contrast, the steady-state biomass of the consumer strain C has a unimodal distribution, with a maximum biomass reached at approximately 80% of the maximal acetate production rate. The reason is that C's biomass reflects the acetate concentration in the chemostat, and this concentration depends not only on the amount of acetate produced per unit of producer strain (pac,P), but also on the amount of producer biomass. As acetate production pac,P increases, the amount of acetate produced per unit biomass increases but the amount of producer biomass decreases. The joint effect of these opposing patterns is a unimodal distribution of consumer biomass.
The total (community's) biomass (Fig 1D, black) decreases with increasing acetate production and has its maximum (0.78 gDW/l) in the absence of acetate production. The reason is that part of the acetate excreted into the chemostat environment is removed through the dilution flux D and not available for usage. In addition, even if all produced acetate were available, its production and later consumption are associated with losses in terms of energy and carbon atoms.
Regardless of the acetate production of the producer strain P, the producer and consumer strains stably coexist, as has also been found experimentally [22]. Additional simulations show that the eventual steady-state composition of the chemostat does not depend on the initial biomass of either strain or the time at which C is introduced (S1 Fig). In contrast, a higher dilution flux D will result in higher biomass for the producer P, but a lower biomass for the consumer C. This can be intuitively understood if we consider the concentration of the nutrients that support growth of each strain: The higher the dilution rate is, the more similar the composition of the chemostat is to that of the fresh medium, which contains high amounts of glucose but no acetate.
When the consumer strain C can also metabolize the primary carbon source (cglc,C > 0), the two strains compete for this carbon source, and coexistence is no longer guaranteed. However, analytical calculations supplemented by simulations show that the two strains can stably coexist under a broad range of glucose consumption and acetate production rates (S3 and S4 Texts, S2 and S3 Figs). When they do, they occupy distinct ecological niches in their nutrient environment [51,52], as shown in Fig 1E. One can visualize the ecological niche space in our chemostat environment as a multidimensional space, where each axis of the space corresponds to the availability or consumption of a nutrient available in the environment. Because in our analysis only two carbon sources are present, the producer strain P and the consumer strain C can compete only for these two carbon sources, which renders our niche space two-dimensional (Fig 1E). Its axes correspond to glucose and acetate consumption rates in metabolic steady state. The ecological niches for our two strains can overlap at the level of glucose consumption (Fig 1E and S3 Fig), but the strains cannot consume identical amounts of glucose without losing their metabolic differences. Thus, their ecological niches cannot overlap completely, consistent with the competitive exclusion principle from ecological theory [18]. We note that this conception of a niche is consistent with the geometric framework of nutritional niche representations [51,52], where niches correspond to the “blend and ratio of nutrients that maximize fitness”.
Our analysis so far reproduced the experimentally observed construction of the glucose-acetate niche [14,22], and identified the conditions under which two E. coli strains can coexist in this niche (Fig 1D, S3 and S4 Texts, S2 and S3 Figs). We next turn to secondary carbon sources other than acetate. Although only glycerol has been experimentally identified as an additional secondary carbon source in cross-feeding experiments [14,22], E. coli cells can produce many other metabolites when growing on glucose [53]. These metabolites, as well as possibly additional, still unknown metabolites, might serve as secondary carbon sources. To identify all possible secondary carbon sources, we first identified all carbon containing metabolites in the iJO1366 metabolic network that can be transported across the cell wall (i.e., metabolites containing an associated exchange reaction). We used FBA to identify which of these molecules can sustain E. coli growth when present as the sole carbon source, which is a prerequisite for a molecule’s usefulness in the cross-feeding interactions we study. FBA predicts 180 metabolites (whose acronyms are given in the circle of Fig 2A) that can sustain growth of E. coli when used as sole carbon sources (Methods). (See Supplementary Table 3 in [44] for a list of standard metabolite acronyms used in Fig 2A).
For each of these 180 metabolites, we determined whether E. coli can produce the metabolite when it is provided with glucose as the sole carbon source (See methods). Fig 2A shows a graphical representation of the answer, where an arc connects glucose (grey arrow) to another carbon source if that carbon source can be produced when glucose is the sole primary carbon source. (This means that the enzymes needed to transform glucose into the carbon source are present in E. coli). There are 58 such secondary carbon sources, acetate (black arrow) being one of them. In other words, E. coli cells growing on glucose can modify the environment by producing 58 alternative nutrients, each of which can sustain the life of other E. coli individuals.
The secondary carbon sources differ greatly in the maximum rate (pmmax) at which they can be produced (Fig 2B), which ranges from 4.75 mmol gDW-1 h-1 for N-Acetyl-D-glucosamine(anhydrous)N-Acetylmuramic acid to 178 mmol gDW-1 h-1 for formate. Acetate’s maximal synthesis rate is 50.3 mmol gDW-1 h-1, about twice the mean production rate of 24.7 mmol gDW-1 h-1, and second highest among all secondary carbon sources (together with glycolate). This maximum production rate reflects the cost of producing a carbon source: The costlier the production of a secondary carbon source is, the smaller is its maximum production rate (S4 Fig). In addition, the secondary carbon sources differ in their specific biomass yield α, which is the growth rate that can be achieved per unit of carbon source consumed (see Fig 2B and S2 Text and S7 Fig green dots). This yield varies from 0.0014 to 0.30 (in gDWmmol-1of carbon source). Acetate’s biomass yield equals 0.025 and is thus low, less than half of the mean value of 0.068. Thus, while acetate is not costly to produce, its low biomass yield also does not allow for much biomass production in strains that consume it. (See S1 File for biomass yields, maximum production rates, and production costs for all secondary carbon sources).
Following our analysis of the glucose-acetate niche (Fig 1), we then asked whether glucose, in combination with each of these individual metabolites, could lead to a stable cross feeding polymorphism. In other words, are there values of the production rate of metabolite x (px,P) and the consumption rate of glucose (cglc,C) that lead to a stable cross-feeding polymorphism? We found that all secondary carbon sources other than formate can sustain a stable community of two strains, even though these communities vary greatly in the amount of total biomass that they contain (Fig 2C).
If strain P completely respires glucose to carbon dioxide and thus does not excrete any secondary carbon source, the total community biomass is equal to the biomass of P, and reaches a maximum value of 0.78 gDW/l, which is indicated by a dashed-dotted line in Fig 2C. For any one secondary carbon source excreted by strain P, the steady-state biomass of the community will change with the carbon source’s excretion rate, up to a sustainable maximum (pmmax) beyond which the producer grows so slowly that it will eventually be flushed out from the chemostat. The figure indicates this change by one grey line (and superimposed colored circles) for each of the 54 secondary carbon sources, along with percentages that indicate the percentage of the maximum rate pmmax at which strain P produces the secondary carbon source. (Displaying secondary carbon source excretion as a percentage of this allowable maximum has the advantage of displaying the same cost for all producers, regardless of which secondary carbon source they produce, such that at any given percentage of this maximum, producers of all secondary carbon sources reach the same steady-state biomass.) As the amount of a secondary carbon source produced by P increases, the total community biomass decreases, i.e., the circles in Fig 2C become further removed from the dashed-dotted line. We already observed this behavior for acetate (Fig 1C, black line in Fig 2C), but Fig 2C illustrates that it holds for all secondary carbon sources.
The absence of a stable community in the glucose-formate niche space is a consequence of formate’s low biomass yield, which requires high formate consumption (112 mmol gDW-1 h-1) to support a growth rate greater than the dilution rate D = 0.2 h-1. This level of consumption is impossible under our assumed transport limit (Vmax = 20 mmol gDW-1 h-1). Higher transport limits or lower dilution rates would, however, permit the existence of a stable community.
The steady-state biomass changes of both strains P and C with increased production of a secondary carbon source (Fig 2C) are analogous to what we observed in Fig 1D, and they exist for the same reason. To support higher production fluxes of any secondary carbon source, P needs to consume more glucose and therefore reaches lower steady-state biomass. To understand the change in steady-state biomass of strain C with an increasing production rate of the secondary carbon source, one has to take into account two factors. The first is the maximal production rate of the secondary carbon source (pmax), which affects the carbon source’s availability for C’s consumption. The second is the biomass yield α of the secondary carbon source, which affects the growth rate achieved per unit flux of consumed carbon source. The product of production and yield (αmpmmax) determines the steady-state biomass of C. If a metabolite is costly (with low maximal production) one can expect low excretion, but a high biomass yield of the same metabolite may compensate for its low production and permit a higher steady-state biomass of C. The colors in Fig 2 indicate the magnitude of αmpmmax for all 58 secondary carbon sources. The figure illustrates that secondary carbon sources whose product of production and yield (αmpmmax) is low will lead to communities with lower total biomass.
The product of production and yield αacpacmax for acetate does not have an unusually large value (equals 1.26 h-1). About half of the secondary carbon sources (29 out of the 58) have a higher maximal growth rate αmpmmax than acetate (αmpmmax > 1.26, blue dots in Fig 2D), and therefore support higher community biomass.
The observations in Fig 2D are based on the assumption that strain C consumes only the secondary carbon source, but not the primary carbon source glucose (cglc,C = 0). However, relaxing this assumption to cglc,C > 0 also allows for stable coexistence of P and C (S6 Fig). The key difference is that stable coexistence then becomes possible for each of the 58 secondary carbon sources, including formate. If C’s glucose consumption is so high that it covers at least 90% of the energy and carbon required for persistence in the chemostat, formate can supply the remaining energy and carbon needed. In this case, coexistence of a formate producer strain and a glucose-formate consumer strain would be possible.
Because glucose is not the only primary carbon source that can sustain E. coli, we extended the previous analysis of searching for secondary carbon sources from glucose to all 180 primary carbon sources. We began by identifying the number of potential secondary carbon sources that can be produced when E. coli grows on each primary carbon source. This number ranges from 54 to 62, depending on the primary carbon source (blue circles in S7 Fig and S8A Fig).
Most secondary carbon sources can be produced from all primary carbon sources, as is the case for acetate, but some secondary sources can be produced from just a few primary carbon sources (red circles in S7 Fig and S8B Fig).
Our analytical results (S4 Text) reveal that coexistence is possible for each pair of primary and secondary carbon sources, as long as two conditions are met. The producer strain must produce the secondary carbon source, and the consumer strain C must be able to persist in the chemostat by consuming both the primary and the secondary carbon source (not just the primary carbon source alone). If this were not the case, that is, if the consumer strain was able to persist in the chemostat by consuming only the primary carbon source, then it would have an advantage over the producer strain, which uses part of the consumed primary carbon source to produce the secondary carbon source. In this case, the producer strain would go extinct.
In total, our analysis finds 83 different secondary carbon sources and 9913 unique pairs of primary and secondary carbon sources that allow stable coexistence of a producer strain P and a consumer strain C (Table 1). Taken together, these observations imply that the synthesis of by-product metabolites by E. coli can create an enormous number of new ecological niches whose identity depends on the primary carbon source available in the environment.
The metabolism of any one organism is the product of a long evolutionary history. E. coli’s enormous potential for the creation of novel metabolic niches could be an accident of this evolutionary history, or it could be a more general property of the chemical reaction networks that constitute a metabolism. To find out, we performed several additional analyses. First, we analyzed the niche construction potential of two microbes different from and not closely related to E. coli, i.e., the soil bacterium Bacillus subtilis (model iYO844 [54]) and the yeast Saccharomyces cerevisiae (model iMM904, [55]). (See methods for a detailed description of the procedure.) The analysis revealed (Table 1) that these organisms also have a large niche construction potential. They can form stable cross-feeding communities with more than 1000 pairs of primary and secondary carbon sources (Table 1).
Each of these three organisms has its own evolutionary history which molded its metabolic network. The observation that they all share a large potential to construct new metabolic niches hints that this potential is a general property of metabolic systems, and not just a peculiarity of the organisms studied and their evolutionary history. To exclude the influence of this history more rigorously, we repeated our analysis with metabolic networks that are not the product of evolution, but that we created in silico with an algorithm that produces random viable networks. These are biochemical reaction networks that produce all essential biomass molecules in a given chemical environment, but contain an otherwise random complement of biochemical reactions drawn from the known “universe” of such reactions. We obtained these networks through a previously published [56,57] Markov Chain Monte Carlo (MCMC) procedure that samples such networks from a vast space of metabolic networks (See Methods).
We note that simpler sampling methods, such as “brute force” uniform sampling of a given number of reactions from a reaction universe is very unlikely to yield viable networks [56]. In contrast, MCMC sampling can yield not only viable networks but viable networks whose reaction complement is effectively random beyond the requirements imposed by viability, as shown by previous work [56,58].
We used this method to create samples of 500 random viable networks viable on glucose as a sole carbon source and that have the same number of reactions (2251) as the E. coli network. We made these networks permeable to all 330 metabolites to which E. coli is permeable. In other words, these random viable metabolisms have the potential to consume and produce the same metabolites as E. coli.
In our sampling procedure, we only required these networks to be viable on glucose, but as a result of complex correlations between metabolic phenotypes [59,60], they are usually also viable on additional primary carbon sources (Fig 3A). Specifically, the number of primary carbon sources on which each sampled metabolic network is viable ranges from 1 to 52 (mean 32±10) (Fig 3A). We observe 218 primary carbon sources on which at least one of these random networks is viable.
When exposed to glucose as the primary carbon source, these networks produce between 0 and 22 secondary carbon sources (mean 11±5) that can sustain a two-strain community (Fig 3B). Taking all sampled metabolic networks we find 84 secondary carbon sources that are produced by at least one random viable network (26 more than produced by E. coli). Most secondary carbon sources are produced by more than one random viable network. Fig 3C shows all secondary carbon sources that are produced by any random network, ranked by the fraction of the 500 random viable networks that produce them. Acetate and glycerol, the secondary carbon sources found experimentally when growing E. coli on glucose are among the top-ranked carbon sources, with respective ranks of 19 and 13.
When exposed to not just glucose but to each of its primary carbon sources in turn, a random viable metabolic network can produce on average 14±6 secondary carbon sources (ranging from 0 to 26). The number of primary-secondary carbon source pairs varied greatly between networks, ranging from 0 to 1065 (mean 404±236). In total we observe 15685 different primary-secondary carbon source pairs that could serve as the foundation of a stable community in at least one random viable network.
In sum, because even random viable metabolisms show high niche construction and cross-feeding potential, this potential is likely an intrinsic property of metabolic systems.
Our last analysis complements the previous analysis by constructing a pan-metabolic network that contains all metabolic reactions from a known and curated “universe” of metabolic reactions (Methods). For various reasons, such a network could never be realized in any one organism, but it provides another way to inform us which kind of cross feeding interactions are metabolically possible. (S10 Fig illustrates how this pan-metabolism analysis relates to our previous analysis of random viable networks.) The pan-metabolic network we analyzed comprises 7222 reactions and 5625 metabolites. As in our analysis of random viable metabolic networks, we only allowed those 330 metabolites to enter and leave a cell that can also enter or leave E. coli. This focuses our analysis on novel biosynthetic abilities rather than on novel transport as a reason for the production or consumption of novel carbon sources. It also implies that we may underestimate the numbers of primary and secondary carbon sources, and perhaps dramatically so.
The pan-metabolic network harbors 221 metabolites that can be used as primary carbon sources, 41 more than E. coli. The minimum number of secondary carbon sources produced per primary carbon source is 85 (S8C Fig), compared to the 54 in E. coli (S8A Fig). In total, the pan-metabolic model can produce 109 secondary carbon sources. The total number of primary-secondary carbon source pairs almost doubles relative to E. coli (18959 and 9913 pairs for the pan-metabolic and the E. coli network, respectively). See Table 1.
In sum, the numbers of primary-secondary carbon source pairs that can sustain stable communities is greatest in the pan-metabolic network. In different organisms, different subsets of such pairs may be suitable for cross-feeding induced niche construction.
When an isogenic bacterial population grows in a homogeneous environment with a single nutrient or carbon source, the organisms in the population initially behave similarly and consume this nutrient. They may also excrete by-product metabolites that accumulate in the environment. If they can express the necessary enzymes, they may switch to consume the by-products once most of the initial nutrient is consumed. In such a population, DNA mutations may arise that alter metabolic properties like enzyme activities or expression permanently. As a result, ancestor and mutant strains may compete for the original nutrient, and one of them may eventually be excluded from the population. Alternatively, a mutant may specialize in the consumption of the ancestor’s by-products. Our work focuses on this commensal or mutualistic scenario, which can help ancestor and mutant to coexist stably, and thus permanently increase genetic and metabolic diversity.
We searched exhaustively for by-product or secondary carbon sources that can be excreted when a microbial strain grows on some primary carbon source, and that can themselves sustain microbial life. We performed this search with the metabolism of three non closely related organisms: E. coli, B. subtilis and S. cerevisiae; with 500 randomly sampled metabolic networks that we required to be viable on at least glucose and that contained the same number of biochemical reactions as E. coli; and with a pan-metabolic network containing 7222 biochemical reactions known to occur in extant organisms. For each of these metabolisms we identified thousands of possible cross-feeding interactions where one strain produces a carbon source that can sustain the other strain. Through a combination of analytical calculations and simulations of the ecological dynamics of two-strain chemostat communities, we demonstrated the existence of 9919 unique cross-feeding niches in E. coli alone that can sustain a stable two-strain community. Each niche corresponds to a unique pair of primary and secondary carbon sources. Our observations suggest an enormous potential for population diversification through niche construction and cross feeding.
Although it may seem puzzling that an organism would dispose of metabolites that could advance its own growth, it is not an unusual phenomenon. The causes are multiple, and include membrane leakage, overflow metabolism, genetic mutations, and cells fermenting carbon sources even in the presence of oxygen [36,61,62]. In addition to acetate, for example, E. coli frequently releases formate, lactate, succinate and ethanol into the environment as a result of fermentation or membrane leakage [61]. Various microorganisms, including Escherichia coli, Corynebacterium glutamicum, Bacillus licheniformis and Saccharomyces cerevisiae, excrete a broad diversity of more than 30 metabolic intermediates and amino acids [53]. Detecting such secondary carbon sources may promote the experimental discovery of new cross-feeding interactions.
Our work differs in various ways from previous studies on microbial metabolic interactions that include competition, commensalism and mutualism [47,63–71] in general, and cross-feeding in particular [35–42,72–74]. The most closely related studies [42,47,63] use a metabolic model of E. coli to study the various cross feeding interactions that can emerge in co-culture after single gene knock-out [63] or extensive gene loss [42]. Our work, in contrast, shows that even without such alterations to its reaction complement E. coli can create many niches. In addition, we also analyzed other organisms, as well as random viable metabolisms to demonstrate that this niche construction potential is not just a property of E. coli or closely related organisms, but a generic property of complex metabolic systems. Other authors have demonstrated that microbes from different species that are cultured together can show new biosynthetic abilities [47]. In contrast, our work shows that new niches and stable communities can emerge from within a population of initially identical individuals. And perhaps most importantly, we have not merely reproduced a single experimentally demonstrated niche construction process, but found that metabolic systems can give rise to myriad new niches through cross-feeding.
Our analysis has several limitations. First, we rely on current knowledge about the metabolism of E. coli, B. subtilis, S. cerevisiae and on reactions in the pan-metabolic network. Future research is likely to discover additional reactions in these networks. They may allow the consumption of additional primary carbon sources, or the synthesis of additional secondary carbon sources. In either case, such additional reactions can only increase, not decrease, the niche construction potential of metabolism.
Second, whereas different organisms can import or excrete a different spectrum of molecules, our analysis of random viable networks and the pan-metabolic network allowed only those metabolites to enter or leave a cell that can also enter or leave E. coli. Even so, we found thousands of potential niches. Had we opened cellular transport to further molecules, the number of niches would have increased as well and perhaps dramatically so.
Third, we varied only carbon sources. Similar analyses could be conducted for sources of other chemical elements, such as nitrogen or sulfur. Again, the potential for niche construction could only increase in this case, because different sources of a chemical element can facilitate to the production of novel secondary metabolites. In sum, these limitations, when overcome, would strengthen our conclusion.
Fourth, random viable metabolic networks and the pan-metabolic network may contain thermodynamically infeasible ATP producing cycles [75–77] that can alter biomass growth. For this reason, it would not have been sensible to simulate cross-feeding dynamics for these metabolic networks. However, our analytical calculations show that the conditions for coexistence hold generally and independently of any one metabolism.
Our observations raise the question why the only known cross-feeding polymorphisms that have been detected in E. coli chemostats involve acetate and glycerol as secondary carbon sources. One candidate reason is that many other such polymorphisms exist but have not been detected, because currently no systematic screen for cross-feeding interactions exists. Cross-feeding polymorphisms are usually manifest in different colony morphologies on agar plates, and substantial biochemical and genetic work is needed to prove that such polymorphisms result from cross feeding [13,14,22]. A second candidate reason is that in many such polymorphisms, one of the strains may constitute a small fraction of community biomass, which would make its detection even harder. For instance, we showed (Fig 2D) that half of the secondary carbon sources that E. coli can produce in a glucose environment cause a high metabolic cost to the producer strain or little biomass gain to the consumer strains, which leads to an even lower biomass of the consumer strain than for glucose-acetate cross-feeding. Third, perhaps not all cross-feeding polymorphisms we predict can be biologically realized. For example, on some primary carbon sources multiple regulatory mutations may be needed before a strain produces or consumes some of the secondary carbon sources we predict. Even though such combinations of mutations may arise in large populations of bacteria, the respective secondary carbon sources will be less frequently produced than carbon sources for which single mutations suffice. Characterizing the regulatory mutations needed to bring forth specific secondary carbon sources is a complex undertaking that we will focus on in future work.
Our work focuses on bacterial populations, but similar phenomena may occur elsewhere. For example, they may help explain a hallmark of cancer, the metabolic heterogeneity within tumors [78]. Many tumors occupy low oxygen-environments, because they grow faster than blood vessels can form. As a result, they synthesize fermentation products like fumarate or succinate [79]. In addition, even when oxygen is available, tumor cells exhibit the Warburg effect [80], the fermentation of glucose to lactate. It is possible that these phenomena may help create new nutritional niches that may be colonized by tumor cells.
Like most biodiversity, bacterial diversity may have arisen through repeated adaptive radiations, in which a single lineage rapidly diversifies to occupy multiple ecological niches [12,81–83]. Usually, species created during adaptive radiations are thought to occupy pre-existing niches, but the rapid emergence of extensive cross-feeding in homogeneous environments [13,14,20] raises the possibility that many niches are constructed during a radiation. That is, when a bacterial population excretes one or more energy-rich by-product metabolites, it creates niches that can be occupied by mutant strains that are well-adapted to these niches. By excreting their own specific metabolites, these strains can then become stepping stones towards further diversification. In this process, the new metabolic niches into which a population radiates are constructed by the population itself. Because any one bacterial strain can excrete a broad spectrum of metabolites, and because our work identified thousands of niches that could sustain stable communities, the potential for such diversification should not be underestimated. We hope that our observations will motivate experimental work that identifies the extent to which this potential is realized.
Flux balance analysis (FBA) is a computational method to predict metabolic fluxes–the rate at which chemical reactions convert substrates into products–of all reactions in a genome-scale metabolic network [45]. FBA requires information about the stoichiometry of chemical reactions in a metabolic network. It makes two central assumptions. The first is that cells are in a metabolic steady-state. The second is that cells effectively optimize some metabolic property such as biomass production (growth). Additional constraints can be incorporated into the optimization problem that FBA solves, in order to account for the thermodynamic and enzymatic properties of a network’s biochemical reaction. The optimization problem that FBA solves can be formalized as a linear programming problem [45,46] in the following way:
Maximizevgrowth
s.t.Sv=0
li≤vi≤ui
Here, S is the stoichiometric matrix, a matrix of size m × r that mathematically describes the stoichiometry of the network’s metabolic reactions. The integer m denotes the number of metabolites and r denotes the number of biochemical reactions in the network. These reactions include all known metabolic reactions that take place in an organism, which are called internal reactions. They also include reactions that represent the exchange (import or export) of metabolites with the external environment. Furthermore, they include a biomass growth reaction, which is a “virtual” reaction that reflects in which proportion biomass precursors are incorporated into the biomass of the modeled organism [44–46]. Each entry Sij of the stoichiometric matrix contains the stoichiometric coefficient with which metabolite i participates in reaction j. The vector v is a vector (of size r) that harbors the metabolic flux through each reaction in the network. vgrowth specifies the flux through the biomass growth reaction. Fluxes through biochemical reactions are restricted by lower and upper bounds that constrain the flux through each reaction in the network. These bounds are given by the variables l and u, respectively, which are vectors of size r. We performed FBA optimization with the GNU Linear Programming Kit (GLPK; http://www.gnu.org/software/glpk).
Strains P and C will grow at rates μP and μC, a process that will increase their respective biomasses XP and XC. In a chemostat, fresh medium is continuously added and culture is continuously removed at a dilution rate D. Such dilution leads to a decrease in biomass inside the chemostat. Overall, the change in biomass for P and C can be expressed by the following system of ordinary differential equations:
dXPdt=(μP−D)XP
(1)
dXCdt=(μC−D)XC
(2)
The concentration of any one metabolite (M) in a chemostat also varies. If a metabolite is present in the fresh medium at concentration M0, the metabolite's concentration will increase at a rate DM0 as a result of fresh medium continually being added to the chemostat. In addition, the metabolite’s concentration will also increase if it is produced by strain P (with flux JM,Pout) or C (with flux JM,Cout). The total rate at which M is produced will then equal JM,PoutXPΔt+JM,CoutXCΔt. Conversely, the concentration of M will decrease due to removal of old medium at a rate DM, and possibly also due to consumption by P (with flux JM,Pin) and C (with flux JM,Cin), at a total rate JM,PinXPΔt+JM,CinXCΔt. We denote the net flux of metabolite M as JM=JMout−JMin, i.e., which results in a positive net flux JM if the metabolite is produced and negative otherwise. Overall, the change in concentration for each metabolite present in the chemostat is then described by the differential equation
dMdt=D(M0−M)+JM,PXP+JM,CXC
(3)
We performed FBA to compute instantaneous growth rates (μP and μC) in h-1), as well as consumption and excretion fluxes of each metabolite by strains P and C (JM,P and JM,C, in mmol gDW-1 h-1). We used the values thus computed in dynamic FBA [46] to determine the changing amounts of biomass (expressed in gDW/l) of our microbial strains, as well as the abundance of all metabolites (in mM), nutrients, and waste products in our simulated chemostat.
Dynamic FBA (dFBA) [46] is an FBA-based method to describe the temporal growth dynamics of microbes and how this dynamics affects the microbes’ chemical environment. It has been used, for example, to describe chemical growth and by-product secretion of E. coli in batch and fed-batch cultures [46], to study the dynamics of a two-species microbial ecosystem in batch culture [47] and to simulate the growth and metabolic dynamics of microbes in time and space [84].
Briefly, dFBA starts from some initial time point and performs Flux Balance Analysis (FBA) iteratively at each time point during a given time interval to compute the maximally possible growth rate for each strain in the chemostat environment. As microbial strains grow, they consume nutrients and excrete waste products (including, possibly, secondary carbon sources) and thus change their growth environments. Dynamic FBA takes these changes into account by computing the chemical composition of the environment at each time point. In doing so, dynamic FBA predicts how the biomass of bacterial strains and the chemical composition of the environment can change over time.
We used dFBA to predict the temporal behavior of a microbial population composed of acetate producer strain P and consumer strains C in a chemostat. We next describe in detail how we used dFBA to simulate population growth in a glucose-limited minimal medium. We note that our procedure can be applied to any other carbon source by substituting glucose with the desired carbon source.
Our simulations used the following parameters and initial conditions. We chose a dilution rate of D = 0.2 h-1 to mirror conditions from previous experiments that had identified cross-feeding interactions [13]. We set the glucose concentration in the fresh medium to 1 mM, close to the 0.7 mM used in [13]. We assumed that ammonium, calcium, chloride, cobalt, copper, iron, magnesium, manganese, molybdate, nickel, oxygen, phosphate, potassium, protons, sodium, sulphate and zinc are present in non-limiting amounts. The initial concentrations of all metabolites in the chemostat are identical to those of the fresh medium. Unless otherwise stated, we initialized the chemostat with 0.01 gDW/l of strain P and 0.001 gDW/l of strain C which corresponds to an overall cell density of approximately 107 cells/ml [85–87]. We chose these initial biomass values arbitrarily, except that their unequal values are well-suited to ask whether the consumer strain C, when introduced in small amounts into a culture of strain P, can invade the culture. However, we also show that changing initial biomass values, the ratio of the biomass values, and the time of introduction of C into the chemostat have no effect on the biomass of P and C once steady-state is reached (S1 Fig).
After initializing our simulations, using these parameters, we discretized time into short intervals of 0.1 h and performed dynamic FBA [46] by iterating the following three steps (described in more detail below): calculation of maximum nutrient uptake rates, FBA, and calculation of environmental composition.
1. Calculation of maximum uptake rates. The uptake of a nutrient by an organism is limited by two factors: the capacity to transport the nutrient across the cell wall (transport limitation) and the availability of the nutrient in the environment (nutrient availability limitation). To determine the nutrient transport limit (for a nutrient at concentration M), we assumed Michaelis-Menten kinetics (VmaxM/(kM + M)) with parameter values set to Vmax = 20 mmol gDW-1 h-1 and kM = 0.05 mM. These parameters are based on data in the Brenda enzyme database [88,89] and have been used in related analyses [47].
At the beginning of a simulation, the nutrient concentration is high and the biomasses of P and C are low. Therefore, nutrient consumption is initially limited by transport. As biomass grows the nutrient begins to be scarce and nutrient availability rather than transport become limiting for nutrient consumption. In other words, the transport limit shapes the transient biomass dynamics but the availability limit determines the steady-state biomass. This also means that Vmax and kM can vary over a wide range without affecting the steady state. In S1 Fig we demonstrate the chemostat dynamics for various values of Vmax and kM to exemplify how these parameters modify the transient biomass dynamics, but not the steady state.
To determine the nutrient availability limit, we divided the nutrient concentration M by the nutrient consuming biomass. This biomass depends on how much nutrients the strains consumed in the immediately past (according to Eqs (1) and (2) in S1 Text). We describe and justify our procedure to calculate the nutrient availability limit, which differs from that of some other authors, in depth in the supplementary material (S1 Text).
Once we had determined a strain’s transport limit for a nutrient and the nutrient’s availability limit, we set the uptake rate of the nutrient to the minimum of both, which ensures that organisms do not consume more of a nutrient than is physiologically feasible and available to them.
2. FBA. Once we had calculated maximum uptake rates of nutrients as just described, we performed FBA for each strain independently. The calculation yielded growth rate values (μP and μC) for both strains, as well as consumption or excretion rates of each metabolite M for both strains (JM,P and JM,C).
3. Calculation of environmental composition. With the results of FBA in hand, we used Euler’s method [90] to determine the environmental change caused by nutrient consumption, waste production, and biomass growth. We did so in accordance to Eqs (1), (2) and (3), using the conditions from the beginning of this section and a time increment of 0.1 h.
We repeated these three steps until at most 1000 h (104 time steps) had elapsed or until the chemostat had reached steady state. We assumed that steady state had been reached if the standard deviation of growth rates determined over 50 consecutive time steps was smaller than 10−5 for both strains. We carried out these simulations using MATLAB (Mathworks Inc.).
We searched for all metabolites that could serve as carbon sources in the following way. To identify primary carbon sources, we first considered all metabolites in the E. coli model iJO1366 [45] a candidate primary carbon source, if it contained at least one carbon atom and if E. coli had an exchange reaction for this carbon source. Second, we used FBA to determine E. coli’s maximal biomass production when each of these primary carbon sources was available as the sole carbon source. (We assumed that ammonium, calcium, chloride, cobalt, copper, iron, magnesium, manganese, molybdate, nickel, oxygen, phosphate, potassium, protons, sodium, sulphate and zinc can be consumed without constraints). Third, if any one carbon source was able to sustain non-zero biomass production, we considered it an actual primary carbon source. Here and below, we viewed only biomass production fluxes above 10−5 mmol gDW-1 h-1 as being different from zero.
Our approach identified 180 primary carbon sources (Fig 2A). On about half of these carbon sources, growth of E. coli has been demonstrated experimentally [91,92]. No experimental data is available for multiple other carbon sources. Metabolic reconstruction errors may account for the discrepancies between computational predictions and experimental observations for some other carbon sources, but at least for the well-studied E. coli, they may be a minor cause compared to regulatory constraints that are not incorporated by most genome-scale models analyzed with FBA [92]. Such regulatory constraints, where enzymes are encoded by a genome but are not expressed when needed, can be easily broken. That is, even on the short time scales of laboratory evolution, microbial populations can adapt to grow on a novel carbon source in accordance with FBA predictions [93]. Because regulatory evolution can occur during the long-term cultivation of E. coli that we model, we assume that regulatory constraints can be by-passed, and thus use all 180 primary carbon sources on which FBA predicts growth in our analyses.
We considered a metabolite a secondary carbon source if (i) it can serve as a primary carbon source and (ii) if it can be produced as a metabolic by-product when another metabolite serves as a primary carbon source. The first condition ensures that the metabolite can sustain growth of a strain consuming it, and the second condition ensures that the metabolite can be produced. Note that all primary carbon sources are potential secondary carbon sources, but only some of them may be produced as metabolic by-products in a given environment. Most importantly, whether a carbon source is produced depends on the available primary carbon source. To identify actual secondary carbon sources and distinguish them from potential ones, we iterated through all pairs of primary carbon sources and potential secondary carbon sources, and performed FBA. More specifically, we used the primary carbon source as the sole carbon source (uptake rate: 10 mmol gDW-1h-1), maximized the production of the potential secondary carbon source, and constrained biomass production in FBA to be greater than zero. If the potential secondary carbon source could be produced at a rate greater than zero under this constraint, we considered the carbon source an actual secondary carbon source.
We used the same method described in the previous paragraphs to search for primary and secondary carbon sources in the genome scale metabolic networks of B. subtilis (model iYO844 [54]) and S. cerevisiae (model iMM904, [55]), modifying only the chemical environment. Specifically, for B. subtilis we used an environment composed of ammonium, calcium, carbon dioxide, iron, magnesium, oxygen, phosphate, potassium, protons, sodium and sulphate. We constrained the ammonium, phosphate and sulphate uptake rates of B. subtilis to a maximum of 5 mmol gDW-1h-1. For S. cerevisiae, we used a medium consisting of ammonium, iron, oxygen, phosphate, potassium, protons, sodium and sulphate, and constrained the oxygen uptake rate to a maximum of 2 mmol gDW-1h-1. We obtained all three metabolic models (iJO1366, iYO844 and iMM904) from the BiGG Database[94].
The pan-metabolic network is a network containing all metabolic reactions with well-defined stoichiometry that are known to take place in some organism. For our analysis, we extended a previously used pan-metabolic network comprising 5484 metabolites and 6892 reactions [59] by adding the 141 metabolites and 330 reactions from E. coli iJO1366 that were not already present in this network. This amended pan-metabolic network includes 5625 metabolites and 7222 reactions.
We found that 3070 reactions (43%) in the pan-metabolic network are unconditionally blocked [95]. That is, they cannot carry non-zero flux without violating FBA’s steady state assumption when all metabolites to which E. coli is permeable can freely enter and leave the cell. We note that if more metabolites where allowed to enter and leave the pan-metabolic network the number of blocked reactions would decrease. For reference, in the well-curated iJO1366 model 227 reactions (9%) are unconditionally blocked. In terms of absolute numbers, the pan-metabolic network contains 1569 more reactions that can carry nonzero flux than the iJO1366 model.
Genome scale metabolic network reconstructions often contain spurious energy producing cycles that violate the second low of thermodynamic [75–77]. Unless removed from the network, these cycles can spuriously increase biomass production. (For example, removal of these cycles causes an approximately 25% reduction of biomass production in 92% of the networks analyzed in [77]). The pan-metabolic network that we use contains reactions that create spurious ATP producing cycles, allowing ATP to be produced even in the absence of nutrients. However, because 67 metabolites in addition to ATP must be produced for biomass growth, biomass cannot be produced in the absence of nutrients. We emphasize that we evaluated the pan-metabolic network’s viability on specific carbon sources and its ability to produce secondary carbon sources without any quantitative evaluation of fluxes, for which spurious cycles might be a problem.
We wanted to study the niche construction capacity not only of E. coli, but of multiple networks of similar complexity that do not share E. coli’s or any other organism’s evolutionary history. Any one metabolic network can be thought of as a subset of reactions drawn from the set of all metabolic reactions feasible in a living organism, i.e., the pan-metabolic network. In the enormous space of all possible metabolic networks only a tiny fraction is viable on any one carbon source, i.e., they can produce biomass when this carbon source is the sole carbon source. We focused on such viable networks, and to sample them from the space of such networks, we used a technique based on Markov Chain Monte Carlo (MCMC) sampling [56,57], which samples networks during long random walks in the space of all metabolic networks of a given size. The statistical theory behind MCMC sampling [96] shows that its random walks are ergodic, i.e., roughly speaking, they are equally likely to visit all metabolic networks in a connected region of the space of such networks. In previous work, we have shown that in the space of all possible networks, networks viable on a specific carbon source form indeed a subset connected by single reaction changes [97]. One requirement of the method is that random walks have a sufficiently long burn-in period to ensure that any “memory” of the starting network of such a random walk has decayed. In previous work [56], we determined that a burn-in period of 5000 reaction changes is sufficient for this purpose. When this requirement is met, the method essentially ensures that the sampled networks contain a random complement of reactions, with no similarity to the starting network in excess of that required for viability on a specific carbon source.
The method starts from an initial network, which we chose as a network that is viable on glucose as a sole source of carbon and that has the same number of reactions (2583) as E. coli.(The computational cost of MCMC sampling prevented us from exploring other primary carbon sources.) To create this initial network, we first performed Flux Balance Analysis on the pan-metabolic network, with glucose as the only source of carbon. Of all reactions in the pan-metabolic network, 1263 reactions showed non-zero flux and were included in the initial network, which ensured viability on glucose. We chose the remaining (1320) reactions needed to arrive at an equal number of reactions as E. coli at random from the pan-metabolic network. From this initial network the MCMC method creates a long sequence of modified networks. In our implementation of the method, a new network is created by a reaction swap, in which a reaction from the existing network is randomly chosen for deletion, while a randomly chosen reaction (that is not yet present in the network) from the pan-metabolic network is added to the network. If the network remains viable after this reaction swap, the swap is accepted, and the network is modified with a further reaction swap. In contrast, if the reaction swap disrupts viability on glucose, the swap is rejected and a new swap is tried. Modifying metabolic networks through reaction swaps ensures that the number of reactions in the network remains constant (and equal to the number of reactions in the initial network). As the number of reaction swaps increases, the number of reactions that the altered networks share with the initial network becomes smaller and smaller, until the complement of reactions has become effectively randomized after 5000 successful swaps [95]. We stored such a randomized network (which is still viable on glucose) for further analysis after 5000 successful swaps.
We performed 500 independent such random walks, thus creating 500 metabolic networks all viable on glucose and containing as many reactions as the network of E. coli does. Each of them is the end point of a sequence of 5000 successful reaction swaps. This procedure is very time consuming, and to accelerate it, we first determined the reactions that are essential for growth on glucose in the pan-metabolic network, and did not subject these (169) reactions to deletions.
We note that we did not alter the exchange reactions of the starting network, which ensures that in the randomized networks the same metabolites can be exchanged with the environment as in E. coli. We also note that random networks contain a large number of reactions (1197±36) that cannot carry non-zero flux in any of the environments we consider, i.e., they are unconditionally blocked. Even though all networks analyzed in this work contain unconditionally blocked reactions, the numbers observed for the random viable networks are especially large.
Because we created random networks by sampling reactions from the pan-metabolic network, these networks may also contain spurious cycles [75–77]. For this reason we analyzed them similarly to the pan-metabolic network, evaluating only viability on specific carbon sources and the ability to produce secondary carbon sources, without any quantitative evaluation of fluxes which are most affected by spurious cycles.
We implemented our sampling procedure in python, using the cobrapy package [98] to perform flux balance analysis to check for viability on glucose of the networks created after each reaction swap.
|
10.1371/journal.ppat.1003803 | The TFPI-2 Derived Peptide EDC34 Improves Outcome of Gram-Negative Sepsis | Sepsis is characterized by a dysregulated host-pathogen response, leading to high cytokine levels, excessive coagulation and failure to eradicate invasive bacteria. Novel therapeutic strategies that address crucial pathogenetic steps during infection are urgently needed. Here, we describe novel bioactive roles and therapeutic anti-infective potential of the peptide EDC34, derived from the C-terminus of tissue factor pathway inhibitor-2 (TFPI-2). This peptide exerted direct bactericidal effects and boosted activation of the classical complement pathway including formation of antimicrobial C3a, but inhibited bacteria-induced activation of the contact system. Correspondingly, in mouse models of severe Escherichia coli and Pseudomonas aeruginosa infection, treatment with EDC34 reduced bacterial levels and lung damage. In combination with the antibiotic ceftazidime, the peptide significantly prolonged survival and reduced mortality in mice. The peptide's boosting effect on bacterial clearance paired with its inhibiting effect on excessive coagulation makes it a promising therapeutic candidate for invasive Gram-negative infections.
| Bacterial infections, especially sepsis, are worldwide a major cause of morbidity and mortality. Sepsis is characterized by an excessive and uncontrolled immune and coagulation response caused by bacteria and bacterial products, which eventually leads to multiple organ failure. Despite supportive treatments and administration of antibiotics, the incidence of sepsis is rising. Development of antibiotic resistance among bacteria, and the inability of antibiotics to target dysregulated host responses during severe infections and sepsis, motivates the search for novel anti-infective treatment modalities. Here, we describe a therapeutic potential of the peptide EDC34, derived from the C-terminus of tissue factor pathway inhibitor-2 (TFPI-2). The peptide's boosting effect on bacterial clearance paired with its inhibiting effect on excessive coagulation makes it a promising therapeutic candidate for invasive Gram-negative infections.
| Sepsis is a major cause of death in the western world, with mortality ranging between 30 and 70% [1]. The disease is characterized by an excessive and dysregulated immune and coagulation response to microbial infections, leading to capillary leakage, lung damage, and finally multiple organ failure [2], [3], [4]. Several clinical trials targeting coagulation as well as pro-inflammatory responses have been conducted, including administration of activated protein C [5], [6], antibodies against TNF-α [7], [8], [9], interleukin-1 receptor antagonists [10], [11], interleukin-6 antagonists [12], anti-endotoxin antibodies [12], PAF receptor antagonists [13], antithrombin III [14], [15], [16] and other agents [17], [18], [19], [20]. Even though animal experiments showed promising results, all drug candidates tested so far have failed to show clinical efficiency. Consequently, today's treatment for sepsis is largely based on antibiotics in combination with supportive measures, illustrating the need for new therapeutic approaches.
Antimicrobial peptides constitute one group of compounds, which have recently attracted attention as new anti-infectives. Due to their preferential interactions with prokaryotic and fungal membranes, these peptides provide a rapid and broad-spectrum response towards both Gram-negative and Gram-positive bacteria, as well as fungi [21], [22], [23], [24]. Antimicrobial peptides also mediate diverse immunomodulatory roles, including anti-inflammatory effects as well as stimulation of chemotaxis, chemokine production, wound healing and angiogenesis [25], [26], [27], motivating the term host defense peptides (HDP). The molecular diversity of HDPs has provided a wide range of structures of potential interest for the anti-infective field. For example, immunomodulatory peptides such as IDRs (Innate Defense Regulator) selectively protect against bacterial infection by chemokine induction and neutrophil recruitment, while reducing pro-inflammatory cytokine responses [26], [27]. Further, the lactoferrin-derived peptide hLF1-11 differentiates monocytes, which enhances clearance of pathogens, a feature currently utilized in the development of therapies for infections in immunosuppressed patients [28]. Additionally, C-terminal peptides of human thrombin, exerting anti-inflammatory, anti-coagulative and antimicrobial effects, are effective in ameliorating LPS-induced shock and Pseudomonas sepsis in experimental settings in animals [29], [30], [31], further exemplifying that endogenous HDPs may have therapeutic potential.
The present work is based on the finding that the highly positively charged C-terminus of tissue factor pathway inhibitor-2 (TFPI-2) encodes for antimicrobial activity [32]. Previous data demonstrated that C-terminal TFPI-2 fragments are released in human wounds, and can be generated by neutrophil elastase in vitro. A direct antimicrobial effect of a prototypic 34 amino acids long C-terminal TFPI-2 peptide, EDC34 was furthermore demonstrated in vitro [32]. In this work, utilizing various in vitro and in vivo models aimed at characterizing effects on complement, coagulation, and bacterial clearance, we demonstrate a therapeutic potential of the peptide for the treatment of Gram-negative infections.
Initial experiments utilizing E. coli and P. aeruginosa isolates demonstrated that EDC34 displayed significant antibacterial activity in physiological buffer, and enhanced activity in the presence of human plasma (Figure 1A and Figure S1A and B). Heat inactivation of human plasma abolished this potentiating effect. A control peptide (DAA14) derived from the N-terminal part of TFPI-2 did not show any antimicrobial effects in buffer or plasma (data not shown). In contrast, the cathelicidin LL-37 was partially inhibited by the addition of native as well as heat-inactivated plasma (Figure S1A and B), which was compatible with previous observations showing that the peptide's activity is compromised in presence of plasma due to protein binding [33]. Thus, these data suggested that the antibacterial effect of EDC34 is dependent on additional bactericidal systems in normal plasma, such as complement. As described for LL-37, it was observed that EDC34 showed reduced antibacterial activity in heat-inactivated plasma, particularly against the E. coli strains 25922, 49.1 and 47.1 when compared to buffer (Figure 1A and Figure S1A), possibly related to scavenging effects of plasma proteins. In human whole blood, EDC34 demonstrated potent antimicrobial effects against various E. coli and P. aeruginosa isolates at doses as low as 0.3 µM, except for P. aeruginosa PA01, which was killed at 3 µM (Figure S1C and D). Notably, 10–100 times higher doses were required for LL-37 mediated killing (Figure S1C and D). Kinetic studies in the presence of plasma demonstrated that the bacterial killing mediated via EDC34 (3 µM) occurred within 10–20 min for E. coli, and 40–60 min for P. aeruginosa (Figure S2A and B). Contrary to results for a complement-susceptible E. coli strain (Figure 1A, left panel), no EDC34-mediated enhancement of bacterial killing was observed for the E. coli O18:K1 strain, previously shown to be resistant to complement-mediated lysis [34] (Figure 1A, right panel). Next, antibacterial and hemolytic effects of EDC34 were simultaneously analyzed in human blood. Consistent with the complement-dependent action of EDC34, this peptide was active against Gram-negative E. coli and P. aeruginosa in whole blood, while exhibiting little or no effects against Gram-positive S. aureus and S. pyogenes AP1 (Figure 1B, left panel). No significant increase in hemolysis was observed at peptide doses up to 120 µM (Figure 1B, right panel). Next, the fact that the human EDC34 sequence differs from the related murine sequence prompted us to compare the effects of human EDC34 with those of the related peptide of murine origin (DAC31), derived from a corresponding C-terminal region of murine TFPI-2 (Table S1). Both peptides exerted similar antimicrobial effects in buffer as well as in human and mouse plasma (Figure 1C). Taken together, the data imply that bacterial killing by EDC34 in plasma and blood is enhanced by the presence of an intact complement system.
In order to study the complement boosting effect of EDC34 further, western blot and FACS analyses were used. EDC34 enhanced binding of C1q to E. coli (Figure 2A), and increased the formation of the membrane attack complex (MAC), as evidenced by increased binding of C9 and related high molecular weight compounds (Figure 2A). A significant generation of C3a was also observed after addition of EDC34 (Figure 2B–D and Figure S3). After subjecting P. aeruginosa to plasma, in contrast to the results with E. coli above, an activation of complement by the bacteria per se was observed (Figure 2A–D and Figure S3). Nevertheless, quantification of C1q after addition of EDC34 detected an increase of this molecule in association with P. aeruginosa (Figure 2C and Figure S2). Additionally, fragments corresponding to bacterial-bound C3a, as well as peptides of higher molecular weight containing the C-terminal epitope of C3a, were detected, particularly after addition of EDC34 (Figure 2B–D). Electron microscopy studies on fibrin slough from a patient with a chronic wound infected by P. aeruginosa were furthermore performed to explore a possible co-localization between the C-terminal TFPI-2 region and C3a in vivo. Using immunogold-labeled antibodies against the C-terminal part of TFPI-2 and against C3a, the evaluation of 50 bacterial profiles indeed showed that ∼70% of C-terminal TFPI-2 peptide epitopes were closely associated with C3a (Figure 2D). Taken together, these results indicate that EDC34 promotes complement activation in response to E. coli and P. aeruginosa.
TFPI-2 inhibits TF-VII-mediated coagulation and affects factor XIa and plasma kallikrein [35], [36], effects which are thought to be mediated by the Kunitz-type domains of TFPI-2 [37]. However, no evidence has so far been presented that the C-terminal region of TFPI-2 may influence coagulation. Clotting assays using human citrate plasma revealed that EDC34 interfered with the intrinsic pathway of coagulation, as illustrated by a dose-dependent prolongation of the activated partial thromboplastin time (aPTT) (Figure 3A). At 50 µM, the increase in aPTT was about 4-fold in human plasma (Figure 3A) and 20-fold in mouse plasma (Figure 3C). In contrast, the other parts of the coagulation system, such as the extrinsic pathway of coagulation, monitored by measuring the prothrombin time (PT), and thrombin-induced fibrin-network formation (thrombin clotting time; TCT), were not affected or to a lower extent at doses of 20–50 µM (Figure 3B and C, and data not shown). The activation of the intrinsic system takes place at negatively charged surfaces, such as kaolin or bacteria, and involves activation of FXII, which then leads to the activation of plasma kallikrein (PK) and FXI [38]. Analysis of PK activity at the surface of kaolin showed that EDC34, but not the control peptide from the N-terminal region of TFPI-2 (DAA14), was able to block the PK activity (Figure 3D). In order to determine whether EDC34 inhibits PK activity also on bacterial surfaces, P. aeruginosa and E. coli bacteria, sensitive to the peptide in plasma and blood, were chosen for further studies. Bacteria were pre-incubated with EDC34, followed by incubation with plasma, and the effect of EDC34 and the control peptide recorded by measuring the PK activity. The results showed that only EDC34 was able to inhibit PK activation (Figure 3E). Since high molecular weight kininogen (HMWK) is a substrate for PK [38], western blots were utilized to assess possible degradation of HMWK in plasma. Compatible with the results of the PK assay, EDC34 blocked the degradation of HMWK (Figure 3F). PK-cleaved HMWK releases bradykinin (BK), a potent pro-inflammatory mediator [38], and corresponding to the above, EDC34 also significantly inhibited the generation of bradykinin (Figure 3G). The N-terminal TFPI-2 control peptide DAA14 neither inhibited PK activation nor bradykinin generation (Figure 3D–G). These results show that EDC34 mainly inhibits activation of the intrinsic pathway of coagulation, leading to reduced HMWK degradation and bradykinin release.
To assess possible anti-coagulative effects in vivo, 12 mg/kg of LPS was injected into the mice followed by intraperitoneal administration (i. p.) of 0.5 mg of EDC34 after 30 min. Measurements of whole blood clotting, aPTT, PT, and TCT at 4 h and 8 h after LPS challenge clearly demonstrated an anti-coagulant effect of EDC34 in this model (Figure 4A). This effect was confirmed by determining thrombin-antithrombin complexes (TAT) in mouse plasma 8 h after LPS injection. TAT complexes were reduced in EDC34-treated animals (Figure 4B). Previous studies have demonstrated that EDC34 binds LPS similarly to LL-37, compatible with its direct antibacterial effects on bacteria [32]. To test whether LPS-binding confers any anti-inflammatory effects to EDC34, mouse macrophages were stimulated with LPS in presence or absence of EDC34. GKY25, a thrombin-derived LPS-binding and anti-inflammatory peptide, was used as positive control [29], [31]. In contrast to GKY25, which blocked the LPS response at 1–5 µM, EDC34 did not exert any endotoxin-blocking effects, even at significantly higher concentrations (Figure S4). Previously published work showed that GKY25 (at 0.5 mg) significantly improved survival as well as inhibited cytokine responses in a mouse model of endotoxin shock [29], [31]. In contrast, and corresponding to the absence of anti-inflammatory effects in vitro, the same dose (0.5 mg) of EDC34, when administrated i. p. 30 minutes after endotoxin exposure, was not able to block IL-6, MCP-1, and TNF-α responses, although it affected the production of the anti-inflammatory IL-10 at an early time point (Figure 4C), and reduced IFN-γ after 8 h. Thus, these results show that EDC34 largely modulates coagulation during LPS-induced shock in vivo.
Taken together, the above data provided a rationale for using the human peptide EDC34 in murine Gram-negative infection models. Hence, mice were infected i. p. with E. coli, followed by i) immediate i. p. treatment with EDC34, ii) delayed i. p. treatment with the peptide given after 1 h, or iii) s. c. treatment after 1 h, in order to separate peptide and bacteria, and minimize direct peptide effects during i. p. administration. Notably, in these models, immediate as well as delayed EDC34 treatment administered i. p. or s. c. yielded significant improvements in survival when compared to controls (Figure 5A). Although all treated mice displayed a significant weight loss, it was observed that the i. p. treated animals recovered faster (Figure 5B). The i. p. model using immediate treatment was selected in order to evaluate peptide effects in greater detail. Mice were infected i. p. with E. coli, followed by treatment with EDC34. Two hours post-infection, a significant reduction of bacterial levels in the peritoneal cavity was detected only in peptide-treated animals (Figure 5C). To assess the importance of complement activation for the observed antibacterial effects of EDC34 in vivo, C3a levels were measured in peritoneal fluid and plasma (Figure S5A and B). In peritoneal fluid, treatment with EDC34 indeed yielded increased C3a levels, in particular after 2 hours post-infection (Figure S5A). These results were compatible with the observed reduction of bacterial counts in the peritoneum of these EDC34-treated animals (Figure 5C). Concomitantly, C3a levels in plasma of peptide-treated animals were reduced, reflecting the local boosting effect of EDC34 at the site of infection leading to overall reduced bacterial levels and hence, relatively less complement activation systemically (Figure S5B). The importance of an intact complement system was further demonstrated in experiments employing pretreatment with cobra venom factor (CVF), used in order to deplete animals of complement factors [39] before infection and peptide treatment. In this infection model, the antibacterial effect of EDC34 was significantly compromised in CVF-treated mice when compared to control mice (Figure S5C). Taken together, these experiments are compatible with the previous in vitro experiments presented in Figure 2, and firmly demonstrate that EDC34 mediates its antibacterial effect in vivo by boosting of complement activation. Having shown this, a second set of in vivo experiments with E. coli, aimed at studying peptide effects on bacterial dissemination, cytokines, and coagulation parameters after 2, 4, and 8 h post-infection were performed. The results showed that treatment with EDC34 yielded significantly lower bacterial levels in the spleen, kidney, liver and blood when compared to the controls (Figure 5D). Notably, infected mice showed a reduced clotting capacity and exhibited prolonged aPTT and PT times, along with an increase in TAT complexes in their plasma. Treatment with EDC34 however, resulted in an improved clotting function, as evidenced by reduced coagulation times (Figure 5E) and TAT complex formation (Figure 5F). These data implied that the peptide, by enhancing bacterial clearance and blocking bacteria-induced coagulation, reduced the excessive consumption of coagulation factors in this animal model, thus improving hemostasis function. It is of note that time-dependent cytokine responses were detected also in EDC34 treated animals, although at lower levels when compared with non-treated infected animals. (Figure 5G). However, this reduction of cytokine responses was not unexpected, since the bacterial levels were reduced 10–100 times by EDC34 (Figure 5D). Hence, the lower, but retained cytokine response was compatible with the noted absence of significant anti-inflammatory effects of EDC34 in the LPS-shock model above (Figure 4C).
To further investigate the functions of EDC34 in a mouse model of Pseudomonas-induced sepsis, two strains of P. aeruginosa, PA01 and 15159 (the latter a clinical isolate), were used. The bacteria were injected i. p., and the peptide was immediately administered either by i. p. injection (1×1) or s. c., either 1 h (1×1) or 1 and 7 h (1×2) after bacterial injection. EDC34 yielded significant improvements in survival after immediate treatment (Figure 6A), but failed to improve survival after delayed treatment (not shown). Corresponding to the survival results, a reduced number of bacteria was observed in spleen, kidney, and liver compared to control mice, for both strains, after immediate treatment with EDC34 (Figure 6B). Delayed treatment s. c. yielded no significant reduction of bacterial levels. However, and compatible with the peptide's anti-coagulative actions in vitro and in vivo, scanning electron microscopy (SEM) analyses of the lungs from mice infected with P. aeruginosa PAO1 showed that EDC34, irrespective of administration route, reduced fibrin deposition and pulmonary leakage of proteins and red blood cells (Figure 6C).
Although EDC34 lowered initial bacterial levels after i. p. administration, the peptide did not completely eradicate bacteria in the above infection models, particularly noted for the clinical isolate. The activation of the intrinsic coagulation system by bacteria underlies the excessive coagulation and bradykinin-induced vascular leakage, and is therefore of interest to target during P. aeruginosa infection. Since antibiotics do not inhibit these bacterial effects on coagulation, we decided to explore the multiple effects of EDC34 in combination with ceftazidime, an antibiotic often used in Gram-negative infections. In an infection model mimicking a potential clinical situation, bacteria were injected i. p., followed by treatment 90 min and 4.5 h after bacterial challenge with either ceftazidime (300 mg/kg) or EDC34 (0.5 mg s. c.) alone, or the antibiotic in combination with EDC34 (doses as above). In contrast to animals treated with ceftazidime alone, the combination treatment significantly improved survival of the animals (Figure 6D). In contrast, the peptide alone did not increase survival in this model (not shown). Further, bacterial levels were monitored in the spleen, kidney, and liver. Ceftazidime-treated animals had 100-fold less bacterial levels, which were further slightly reduced after addition of EDC34 (Figure 6E), while EDC34 alone did not significantly reduce bacteria. Nevertheless, in spite of similar bacterial levels (Figure 6E), the generation of TAT complexes was reduced in EDC34-treated mice compared to controls (Figure 6F), and thrombocyte levels were increased after peptide treatment. This increase in thrombocytes was also noted in combination with the antibiotic, indicating a reduced activation of the coagulation system due to peptide treatment (Figure 6G). This observation was supported by reduced fibrin deposition and pulmonary leakage of protein and red blood cells in the lungs of animals treated with EDC34 alone, or in combination with ceftazidime as judged by SEM (Figure 6H) and confirmed by histochemistry and scoring of lung damage (Figure 6I and Figure S6). It was noted that animals treated with ceftazidime alone showed lung damage changes, involving reduction of alveolar space, increased cell infiltration and wall thickness, and formation of thrombi, similar to those observed in infected untreated animals (Figure 6I and Figure S6). EDC34 did not significantly reduce cytokine levels when given alone. Nevertheless, cytokine levels were reduced after treatment with the antibiotic as well as with the combination with EDC34 (Figure 6J). EDC34 at a total dose of 2 mg did not significantly affect coagulation (aPTT, PT, TCT), thrombocytes, and cytokines after 12 h when administrated into healthy mice (Figure S7).
The development of novel anti-infective treatments has been largely addressing the bacteria itself, or the subsequent dysregulated host response, while efforts to control the dysregulated host response have so far failed. Therefore, there is a need for new therapies that address new targets in the complex interplay between microbes and the host. HDPs with multiple roles, such as those here defined for the TFPI-2-derived EDC34, targeting both bacteria and coagulative mechanisms, should therefore be of therapeutic interest. The main objective in this work was to clarify bioactive effects of EDC34 in vitro, and to what extent these can be utilized in anti-infective therapy in vivo. Challenging in this perspective was the fact that the in vitro effects, as observed in isolated experimental systems, were not independent of each other in vivo. For example, reduced bacterial levels could also be linked to concomitant reductions in pro-inflammatory cytokines in vivo. In such cases, an attempt was made to reduce complexity, and to study isolated factors or events, such as the peptide's anti-inflammatory effects in vitro in relation to LPS-stimulation of macrophages, or in vivo, in relation to endotoxin shock. In other cases, we highlight unique and peptide-dependent effects, such as abolished fibrin deposition irrespective of bacterial load and antibiotic usage in the animal models, clearly linking the anti-coagulative effects in vitro including blocking of bacteria-induced kallikrein activation and bradykinin release, to those observed in vivo. Furthermore, interference by EDC34 with the coagulation system in vivo was demonstrated not only by reduced fibrin deposition but also evidenced by reduced TAT levels, along with increased thrombocyte levels. This is of importance, since sepsis- and coagulation-related acute lung injury is considered to be a critical feature compromising the clinical outcome during sepsis [40], [41], [42]. In response to LPS challenge, the coagulation cascade is activated, leading to an excessive activation of the coagulation system, followed by consumption of coagulation factors in the blood resulting in prolonged clotting times [43]. In line with this, LPS-injected mice showed a reduced clotting capacity and exhibited prolonged aPTT and PT times in their plasma (Figure 4). Treatment with EDC34 however, resulted in a partially normalized clotting function, as evidenced by reduced coagulation times (Figure 4). Hence, these data showed that the peptide, by blocking coagulation as shown in Figure 3, reduced the excessive consumption of coagulation factors in this animal model. Also of importance, was the observation that no signs of bleeding or prolongation of coagulation was noted in animals treated with EDC34 only (Figure S7A and B). This should be of value, since other agents tested in the clinic, such as activated protein C, mainly affect the extrinsic or primary pathway of coagulation, and thus, may increase the risk of bleeding complications [5], [6]. As mentioned above, EDC34 did not present any significant anti-endotoxin effects in vitro or in vivo, in spite of its binding to LPS [32]. Such absence of direct anti-inflammatory effects may be advantageous, particularly in situations where an anti-coagulative action is of importance, while maintaining a normal LPS immune response.
The complement system is crucial for bacterial clearance, indicating that strategies based on manipulating complement activation and thus bacterial clearance, may be of therapeutic interest. It is therefore notable that EDC34 boosts complement activation rapidly ex vivo in relation to the Gram-negative E. coli and P. aeruginosa. Thus, E. coli did not significantly induce activation of the classical complement pathway in absence of EDC34 in plasma in vitro, and the boosting effects were dependent on EDC34-binding to the bacteria. Although EDC34 was active against Gram-positive S. aureus in low salt buffer conditions [32], the results in human blood, where EDC34 was particularly active against E. coli and P. aeruginosa, further underscores the importance of the peptide's dependence of an active complement system for its actions in the in vivo models. Notably, EDC34-mediated formation of C3a, an anaphylatoxin exerting antimicrobial effects [44], was increased not only in vitro, but also locally at the site of infection in the animal E. coli infection model. In this context, it was interesting to note that although C3a was increased intraperitoneally, the overall levels of C3a in blood of the animals were reduced after peptide treatment. At a first sight, this may appear paradoxical, however, it is not unexpected when considering the significant reduction of bacteria systemically in peptide-treated animals. Also relevant in this context is that analyses of levels of C3 in E. coli infected animals showed that the total levels of this complement factor were unaffected after treatment i. p. with EDC34 (not shown), indicating that the modulation takes place locally, and that the total levels of C3 may buffer a potential local, initial consumption of the protein in this particular animal model. Furthermore, animals with a compromised complement function due to CVF treatment [39] did not exhibit a boosting of E. coli clearance by EDC34 in vivo. These results were compatible with the in vitro studies using heat-inactivated (hence complement-inactivated) plasma (Figure 2), elegantly demonstrating the importance of an intact complement system for the action of EDC34 in vitro and in vivo.
From a biological perspective, it remains to be investigated whether TFPI-2 adds to host defense in vivo. Although these studies are currently initiated and TFPI-2−/− mice have been generated and are currently validated and characterized, this work is clearly beyond the scope of this present work aimed at utilizing the TFPI-2 peptide as a therapeutic molecule. Nevertheless, it is notable that emerging evidence suggests an involvement of TFPI-2 in host defense. Thus, in vitro, stimulation of human endothelial cells with inflammatory mediators such as LPS, and TNF-α significantly increases TFPI-2 expression [45]. Analogously, in vivo in a murine model, TFPI-2 expression is dramatically upregulated in the liver and in the lungs during LPS stimulation [46], [47]. These data, together with previous findings on release of C-terminal fragments by neutrophil elastase and their presence in human wounds [32], and as shown herein, particularly in association with bacteria (Figure 2D), suggest that TFPI-2 fragments may exert physiological roles during infection.
From a clinical perspective infections with Gram-negative bacteria such as E. coli and P. aeruginosa contribute to morbidity and mortality during abdominal surgery, peritoneal dialysis, burn wounds or in immunocompromized patients [48], [49], [50], [51]. An initial phase of effective bacterial clearance is crucial for patient outcome, and therefore, approaches based on substitution with EDC34 in combination with antibiotics may be of therapeutic interest. In order to verify such possibilities, experimental models are crucial, and it is notable that EDC34 exerted similar antibacterial effects in mouse and in human plasma, which should facilitate further development of experimental models incorporating readouts involving lung and organ damage.
From a toxicological perspective, many HDPs cause side effects involving damage to eukaryotic cell membranes. It is therefore notably that EDC34 did not show toxicity at therapeutic doses in vitro [32], that the peptides complement boosting activity was only recorded in presence of, and in relation to, bacteria, and that EDC34 was well tolerated in non-infected animals, neither significantly affecting coagulation parameters nor thrombocyte and cytokine levels (Figure S7).
In summary, the effects of EDC34, involving blocking of contact activation, and bacterial killing directly or by boosting of complement activation, while maintaining a functional cytokine response (Figure S8), represents a potential new approach for boosting bacterial clearance while inhibiting the deleterious effects of excessive pro-coagulative responses during infection with Gram-negative bacteria.
The use of human blood was approved by the Ethics Committee at Lund University, Lund, Sweden (Permit Number: 657-2008). Written informed consent was obtained from the donors. The animal experiments were conducted according to national guidelines (Swedish Animal Welfare Act SFS 1988:534) and were approved by the Laboratory Animal Ethics Committee of Malmö/Lund, Sweden (Permit Numbers: M228-10, M252-11)
The peptides EDC34 (EDCKRACAKALKKKKKMPKLRFASRIRKIRKKQF), DAC31 (DACHRACVKGWKKPKRWKIGDFLPRFWKHLS) and DAA14 (DAAQEPTGNNAEIC) were synthesized by Biopeptide Co., San Diego, CA, whereas LL-37 (LLGDFFRKSKEKIGKEFKRIVQRIKDFLRNLVPRTES) was obtained from Innovagen AB, Lund, Sweden. The purity (>95%) of these peptides was confirmed by mass spectral analysis (MALDI-ToF Voyager).
The bacterial isolates E. coli DH5α, E. coli ATCC 25922, P. aeruginosa ATCC 27853, Staphylococcus aureus ATCC 29213 and S. pyogenes AP1 were obtained from the American Type Culture Collection. The P. aeruginosa strain PA01 was a generous gift from Dr. B. Iglewski (University of Rochester). E. coli O18:K1 was a kind gift from Dr. C. van 't Veer (University of Amsterdam). The clinical isolates E. coli 49.1, E. coli 47.1 and P. aeruginosa 15159 were obtained from the Department of Bacteriology, Lund University Hospital, Sweden.
Animals were housed under standard conditions of light and temperature and had free access to standard laboratory chow and water. The animals were purchased from Charles River or the animal facility at Lund University.
E. coli, S. aureus and S. pyogenes AP1 strains were grown to mid-exponential phase in Todd-Hewitt (TH) broth. P. aeruginosa strains were grown in TH broth overnight. Bacteria were washed and diluted in 10 mM Tris, pH 7.4, containing 0.15 M NaCl, either alone or with 20% normal or heat-inactivated citrate-plasma or 50% citrate blood. Fifty µl bacteria (2×106 cfu/ml) were incubated at 37°C for 2 h with the C-terminal TFPI-2 derived peptide EDC34, LL-37 or DAC31 at the indicated concentrations. Serial dilutions of the incubation mixture were plated on TH agar, followed by incubation at 37°C overnight and cfu determination.
Human citrate blood was diluted (1∶1) with PBS prior to addition of 2×108 cfu/ml bacteria. The mixture was incubated with end-over-end rotation for 1 h at 37°C in the presence of peptides (60 and 120 µM). Two percent Triton X-100 (Sigma-Aldrich) served as positive control. The samples were then centrifuged at 800×g for 10 min and the supernatant was transferred to a 96-well microtiter plate. Hemoglobin release was determined by measuring the absorbance at 540 nm and is expressed as % of Triton X-100 induced hemolysis.
Bacteria (1–2×109 cfu) were incubated for 30 min or 1 h at 37°C with human plasma alone or supplemented with TAMRA-labeled EDC34 (at 3 µM). Samples were then prepared for FACS analysis as previously described [52]. For visualization of the complement proteins, rabbit polyclonal antibodies against either LGE27, a C-terminal epitope of human C3a, or rabbit polyclonal antibodies against C1q (both at 1∶100) in combination with a secondary goat anti rabbit IgG FITC-labeled antibody (1∶500, Sigma) were used. Flow cytometry analysis (Becton-Dickinson, Franklin Lakes, NJ) was performed using a FACS Calibur flow cytometry system. The bacterial population was selected by gating with appropriate settings of forward scatter (FSC) and sideward scatter (SSC). Controls without primary antibodies were included. Total positive cells present and fluorescence index (FI) (positive cells present multiplied with mean) are presented in the figures.
Human citrate plasma was supplemented with bacteria (1–2×109 cfu) and incubated alone or with EDC34 at 3 µM for 30 min or 1 h at 37°C, centrifuged and supernatants and the bacterial cells were collected. The pull down assay, to extract bound proteins from the bacteria was performed as described previously [52], except two additional wash steps with acetone prior to sample separation by SDS page using 16.5% Tris-tricine gels (C.B.S Scientific) under reducing conditions. Proteins and peptides were transferred to nitrocellulose membranes (Hybond-C), blocked by 3% (w/v) skimmed milk, washed, and incubated with rabbit polyclonal antibodies against the C-terminal part of C3a (LGE27 antibodies) (1∶1000), rabbit polyclonal antibodies to C1q (1∶1000) (Dako) or rabbit polyclonal antibodies to C5b-9 (1∶1000) (Abcam, England). The proteins were detected by using HRP-conjugated secondary antibodies (1∶2000) (Dako) and an enhanced chemiluminesent substrate (LumiGLO) developing system (Upstate cell signaling solutions).
All clotting times were analyzed using a coagulometer (Amelung, Lemgo, Germany). For determination of prothrombin time (PT, thromboplastin reagent (Trinity Biotech)) and thrombin clotting time (TCT, Thrombin reagent (Technoclone)), 50 µl of fresh citrate plasma, together with indicated concentrations of EDC34 or DAA14 were pre-warmed for 60 sec at 37°C before clot formation was initiated by adding 50 µl clotting reagent. To record the activated partial thromboplastin time (aPTT), 50 µl of a kaolin-containing solution (Dapttin, Technoclone) was added to the plasma-peptide mix and incubated for 200 sec before clot formation was initiated by adding 50 µl of 30 mM fresh CaCl2 solution. To determine the blood clotting time, 50 µl of citrated blood were pre-warmed to 37°C for 60 sec, before 50 µl of 30 mM fresh CaCl2 were added to initiate coagulation. Thrombin/antithrombin complexes (TATc) were determined in mouse citrate-plasma by ELISA (USCN Life Sciences Inc.).
Bacteria were grown in TH broth to mid-exponential phase (OD620∼0.5), washed twice in 50 mM Tris/HCl, pH 7.4 and resuspended in 50 mM Tris/HCl, pH 7.4+50 µM ZnCl2 to a final concentration of 2×109 cfu/ml. Hundred microliter of bacteria were incubated with 10 µl of EDC34/DAA14 or buffer for 60 sec before the addition of 100 µl human citrate plasma. Samples were incubated for 35 min at 37°C on rotation followed by centrifugation. The bacterial pellets were washed once in 50 mM Tris/HCl, pH 7.4 and resuspended in 100 µl 50 mM Tris/HCl, pH 7.4+50 µM ZnCl2 buffer containing 2 mM of the chromogenic substrate S-2302 (Chromogenix). Samples were incubated for 30–60 min at 37°C, centrifuged and the absorbance was measured at 405 nm in the bacterial supernatants. In other experiments 100 µl of Dapttin (Technoclone) were incubated with 10 µl of EDC34/DAA14 for 60 sec prior to the addition of human citrate plasma. Samples were incubated for 3 min at RT, centrifuged and the pellet was washed twice in 50 mM Tris/HCl, pH 7.4 before suspension in 100 µl 50 mM Tris/HCl, pH 7.4+50 µM ZnCl2 buffer containing 2 mM of the chromogenic substrate S-2302. After 30 min incubation at RT samples were centrifuged and the absorbance of the supernatant was determined (A405 nm). Samples containing 50 mM Tris/HCl, pH 7.4 instead of citrate plasma served as negative controls.
Bacteria and Dapttin samples were prepared as described in the chromogenic substrate assay. Samples were incubated for 15 min at RT, with shaking. Dapttin samples were centrifuged at 10.000 rpm for 2 min and supernatants were stored at −20°C. Bacterial samples were washed twice in 50 mM Tris/HCl, pH 7.4 and resuspended in 55 µl 50 mM Tris/HCl, pH 7.4+50 µM ZnCl2 buffer followed by 15 min incubation at RT before centrifugation at 10.000 rpm for 2 min and storage of supernatants at −20°C. Samples were separated by SDS page and analyzed by western blot using antibodies against HK and its degradation products as previously described [53].
Human citrate plasma was incubated with 50 µM of EDC34 or DAA14 for 2 min at RT before the addition of Dapttin and incubation for 2 min at 37°C. Samples were kept on ice, diluted with deproteinising buffer (1∶5), centrifuged (at 4°C) and the pellet was mixed with equal amounts of assay buffer. Samples with H2O instead of peptide served as positive controls. All buffers were provided together with the ELISA kit used to quantify the released bradykinin according to manufactures instructions (Markit-M-Bradykinin Kit; DS Pharma Biomedical co.Ltd).
Male C57BL/6 mice (8 weeks) were i. p. injected with 5 mg/kg of E. coli O111:B4 LPS (Sigma). Thirty minutes after LPS challenge mice were treated with 0.5 mg EDC34 (i. p.). For analysis of cytokines and coagulation parameters, mice were sacrificed 4 and 8 h post-LPS injection and the blood was collected by cardiac puncture.
E. coli DH5-α bacteria were grown to mid-exponential phase (OD620∼0.5), harvested, washed in PBS and diluted in the same buffer to 2×109 cfu/ml. Hundred fifty microliter of the bacterial suspension was injected intraperitoneally (i. p.) into male BALB/c mice immediately followed by 0.5 mg EDC34 or PBS. Mice were sacrificed 0.5, 2, 4 and 8 h post-infection to evaluate cfu, cytokines, coagulation parameters, blood counts and histology of the lungs. In another set of experiments mice were treated with 0.5 mg EDC34 injected i. p., immediately, or after 1 h, or subcutaneously 1 h post-infection. The animal status and weight was followed daily for up to 7 days. Mice sacrificed before day 7, according to predefined endpoint criteria, were counted as non-survivors. In another experiment, male BALB/c mice were depleted of complement factors by i. p. injection of 4.8 U of cobra venom factor (CVF) (Quidel) [39]. After 16 h mice were infected with E. coli DH5-α (i. p.) and immediately treated with 0.5 mg EDC34 or PBS. Mice were sacrificed 6 h post-infection and cfu were evaluated in spleen and liver, respectively.
P. aeruginosa 15159 or P. aeruginosa PA01 bacteria were grown to mid-exponential phase (OD620∼0.5), harvested, washed in PBS, diluted in the same buffer to 2×109 cfu/ml, and kept on ice until injection. Hundred microliter of the bacterial suspension was injected (i. p.) into male C57BL/6 mice. EDC34 (0.5 mg) or buffer alone was administered i. p. immediately after bacterial injection, or s. c. either as one dose after 1 h, or two doses at 1 h and 7 h. In another experiment, mice were treated twice s. c with a combination of 300 mg/kg ceftazidime and 0.5 mg of EDC34, 1.5 h and 4.5 h after bacterial infection. Data from three independent experiments were pooled. The survival data were obtained by following the animals daily up to 7 days monitoring status and weight. Mice reaching the predefined endpoint-criteria were sacrificed and counted as non-survivors.
For histological evaluation of lungs derived from the in vivo P. aeruginosa infection model, tissues were collected at indicated time points, fixed in 4% formaldehyde for 24 h, embedded in paraffin, sectioned and stained with hematoxylin and eosin. Assessment of differences in alveolar space, cell infiltration, thickness of alveolar septa (cell wall thickness) and thrombi (Figure S6) was performed by scoring of at least five view fields per section by three blinded independent observers (score 1–4; where 1 indicates no change, 2 minor, 3 medium, and 4 significant change).
For transmission electron microscopy analysis of the presence of TFPI-2 fragments in vivo, fibrin slough from a patient with a chronic venous ulcer (CWS) was fixed and processed as previously described [52]. For immunostaining [52], rabbit polyclonal antibodies against the C-terminal of TFPI-2 (CAKALKKKKKMPKLRFASRIRKIRKKQF) alone, or in combination with rabbit polyclonal antibodies against the C-terminal part of C3a (LGE27 antibodies) (1 µg/ml) (Innovagen AB) were utilized. Controls without primary antibodies were also included. For simultaneous detection of TFPI-2 and C3a, 1 µg/ml EM rabbit anti-goat IgG 20 nm Au (BBI) and 1 µg/ml EM goat anti-rabbit IgG 10 nm Au (BBI) were used. All samples were examined with a Jeol JEM 1230 electron microscope operated at 80 kV accelerating voltage. Images were recorded with a Gatan Multiscan 791 charge-coupled device camera. For scanning electron microscopy, lungs were collected at 12 h after injection of bacteria and fixed in 2.5% (v/v) glutaraldehyde in 0.15 M sodium cacodylate buffer, pH 7.4, overnight at room temperature and further treated as described previously [31]. Specimens were examined in a JEOL JSM-350 scanning electron microscope. To quantify pulmonary lesions, lung samples from 30 different fields covering an entire lung section were made, and the percentage of fibrin deposits and fields exhibiting hemorrhage were determined.
The cytokines IL-6, IL-10, MCP-1, IFN-γ, and TNF-α were measured in mouse plasma using the Cytometric bead array; Mouse Inflammation Kit (Becton Dickinson AB) according to the manufacturer's instructions.
The number of platelets in mouse blood anti-coagulated with EDTA was determined using the VetScan HM5 (TrioLab).
Values are shown as mean with SEM. For statistical evaluation of two experimental groups the Mann-Whitney U-test was used and for comparison of survival curves the log-rank test. To compare more than two groups One-Way or Two-Way ANOVA with Bonferoni post-test were used. Viable count data are presented as mean with SD. All statistical evaluations were performed using the GraphPad Prism software 5.0. with *p-<0.05, **<0.01 and ***p<0.001 and ns = not significant.
|
10.1371/journal.pntd.0005918 | Effectiveness and economic assessment of routine larviciding for prevention of chikungunya and dengue in temperate urban settings in Europe | In the last decades, several European countries where arboviral infections are not endemic have faced outbreaks of diseases such as chikungunya and dengue, initially introduced by infectious travellers from tropical endemic areas and then spread locally via mosquito bites. To keep in check the epidemiological risk, interventions targeted to control vector abundance can be implemented by local authorities. We assessed the epidemiological effectiveness and economic costs and benefits of routine larviciding in European towns with temperate climate, using a mathematical model of Aedes albopictus populations and viral transmission, calibrated on entomological surveillance data collected from ten municipalities in Northern Italy during 2014 and 2015.We found that routine larviciding of public catch basins can limit both the risk of autochthonous transmission and the size of potential epidemics. Ideal larvicide interventions should be timed in such a way to cover the month of July. Optimally timed larviciding can reduce locally transmitted cases of chikungunya by 20% - 33% for a single application (dengue: 18–22%) and up to 43% - 65% if treatment is repeated four times throughout the season (dengue: 31–51%). In larger municipalities (>35,000 inhabitants), the cost of comprehensive larviciding over the whole urban area overcomes potential health benefits related to preventing cases of disease, suggesting the adoption of more localized interventions. Small/medium sized towns with high mosquito abundance will likely have a positive cost-benefit balance. Involvement of private citizens in routine larviciding activities further reduces transmission risks but with disproportionate costs of intervention. International travels and the incidence of mosquito-borne diseases are increasing worldwide, exposing a growing number of European citizens to higher risks of potential outbreaks. Results from this study may support the planning and timing of interventions aimed to reduce the probability of autochthonous transmission as well as the nuisance for local populations living in temperate areas of Europe.
| Larvicides are a key tool to prevent the growth of mosquito populations and decrease both the risks of outbreaks of mosquito-borne diseases and the nuisance deriving from bites. In order to assist municipalities from temperate areas in Europe in effectively planning vector control programs, we modelled the effect of larviciding in public areas on populations of Aedes albopictus using mosquito collection data from 10 municipalities in Northern Italy, over two years with very different temperature conditions. We then evaluated the resulting probabilities of potential outbreaks of chikungunya and dengue and their expected final sizes, and we compared the intervention costs to the economic and health benefits due to the avoidance of clinical cases. By assessing several different intervention strategies, we found that the optimal timing should be centred on the month of July, corresponding to the period of maximal growth of the mosquito population. Municipality-wide interventions are likely to be cost-effective in small-to-medium towns (below 35,000 inhabitants) even where mosquito infestation is moderate, whereas for larger cities a neighbourhood-based intervention should be considered. The involvement of citizens to apply larvicides within private premises resulted effective but generally too costly.
| During the last decade, Europe has faced outbreaks of mosquito-borne diseases (MBD) such as dengue and chikungunya, following the continuous importation of human cases in areas with established competent vectors such as the invasive mosquito Aedes (Stegomyia) albopictus (Skuse) [1]. Vector control interventions can be implemented by local authorities to keep in check mosquito abundance and consequently reduce the epidemiological risk. Adulticide spraying rapidly reduces the number of mosquitoes, but its effect is short-lived [2]. For this reason, it is particularly indicated in situations where the transmission risk needs to be reduced drastically and quickly, such as when an individual is diagnosed with an MBD, to prevent or curtail an outbreak [3]. Since the effectiveness of reactive measures decreases with the delay between outbreak initiation and implementation of control [4], a better approach may consist in preventive interventions. Treatment of potential breeding sites with larvicide products has a delayed impact in reducing adult populations [3], but experimental studies show that their effect lasts for several weeks [5], making them better suited for preventive routine control. The main limit to larviciding as a control option is the proportion of breeding sites that are actually accessible to interventions by public health authorities. To overcome this limit, education campaigns may be carried forward to encourage citizens to remove and treat potential breeding sites from their private premises during the mosquito season [6, 7]. Mathematical modelling of MBD associated with cost-effectiveness analyses can help optimizing routine vector control interventions [8] with respect to constraints in human and financial resources [9].
With the aim of assisting European municipalities in planning and timing preventive vector control, we assessed the potential epidemiological impact on chikungunya and dengue, and the ensuing economic benefits for the health system, produced by routine larviciding against Ae. albopictus within urban sites in temperate climates.
Mosquito monitoring via adult trapping was carried out in ten municipalities from the Northern Italian provinces of Belluno and Trento, characterized by a temperate climate [10]. Mosquitoes were collected using Biogents (BG) Sentinel traps (Biogents AG, Regensburg, Germany) baited with lures and CO2 from dry ice. After each trapping session, mosquitoes were killed by freezing at -20°C, identified using taxonomic keys [11, 12] and confirmed by PCR if found in a location for the first time [12].
We simulated the transmission dynamics associated with chikungunya and dengue using a standard SEI-SEIR approach [13] in which mosquitoes develop lifelong infection after an (extrinsic) incubation period since the bite to an infectious human (SEI sub-model), whereas humans develop temporary infection, followed by the development of immunity, after an (intrinsic) incubation period since the bite from an infectious mosquito (SEIR sub-model). We considered temperature-dependent extrinsic incubation periods and per-bite transmission probabilities for dengue [14], whereas only temperature-independent estimates were available for Chikungunya [15, 16]. The transmission model was initialized with a single infectious human, representing an imported case at a date sampled uniformly between January 1st and December 31st. The population size of female Ae. albopictus mosquitoes over time in the transmission model was estimated by fitting a population model to capture data collected in the absence of larvicidal treatments, following the same approach already adopted in [13, 17]. The model considers four developmental stages of mosquitoes (eggs, larvae, pupae and adults) and reproduces their life cycle by means of temperature-dependent parameters regulating the stage-specific rates of mortality and development. Free model parameters (i.e. the site- and year- specific habitat suitability and the capture rate of BG traps) were estimated via a Monte Carlo Markov Chain approach based on a Poisson likelihood [13, 17].
We then included the effect of routine larviciding in the population model. Experimental studies of several available commercial larvicide products show that 99% of existing larvae and hatching eggs are killed within a given breeding site, with constant efficacy of about 30 days, independently of the specific product used [5, 18, 19]. We considered a standard approach targeting breeding sites in publicly accessible spaces (e.g., catch basins placed in public parks and along the road system), and an additional strategy where public interventions were integrated by the involvement of citizens in treating and removing breeding sites within private premises. The latter was parametrized on results from a pilot project conducted in two municipalities within the same area of this study [7], in which larvicide products were delivered door-to-door and free of charge to house dwellers, who were sensitized and educated to mosquito control interventions. A key determinant of effectiveness is the fraction of existing breeding sites in a given area that are actually treated (coverage). We adopted a coverage range between 30% and 50% for larviciding of public catch basins only, and between 60% and 75% for interventions that additionally involve citizens. These ranges were computed from available data on the density and proportion of reachable breeding sites in public and private premises [7]. Other strategies aimed at extending the coverage (e.g. removal of other breeding sites such as water buckets, plant saucers, tarpaulins, etc.) were not considered.
Since the effect of larvicides is transitory, treatment of catch basins may be repeated multiple times within a given season. We considered several different starting dates and from 1 to 4 applications of larvicide treatments within a given mosquito season (hereafter referred to as “effort level”), implemented with monthly frequency.
To evaluate the economic acceptability of the two considered strategies, a cost-utility analysis for the prevention of dengue and chikungunya was conducted, taking the number of infections as input from the transmission model. Disability Adjusted Life Years (DALYs) averted and net costs were derived comparing an intervention scenario to the case in which no control programs were put in place (baseline). The baseline was set to reflect a municipality where only the monitoring of mosquito presence via ovitraps was performed [7].
The analysis was conducted from a public healthcare system perspective through the maximization of the net health benefit (NHB) [20]. This measure is defined as the difference between the DALY averted and the incremental cost due to the intervention, the latter divided by the willingness to pay (WTP) by public authorities for each DALY averted. Following WHO recommendations [21], we assumed such value approximately equal to the Gross Domestic Product (GDP), which is about 35,000 euro per capita in our study area [22].
Probabilities of each infected case of being symptomatic, notified, severe, hospitalized and of dying, and the length of stay in hospital, were derived from published studies [23, 24] and from analyzing data from the Italian Hospital Discharge System (Schede di Dimissioni Ospedaliere), accounting for all hospital admissions for chikungunya and dengue recorded in Italy. The cost of illness was estimated according to expert opinion. The costs of intervention were estimated from actual costs during control activities against Ae. albopictus recently performed in two municipalities from the study area [7].
For all the considered scenarios, the NHB was computed on a set of 100,000 stochastic realizations accounting for the uncertainty in both the transmission and the economic model’s parameters. Full details on this analysis are provided in S1 Text.
To assess the feasibility and sustainability of public interventions, we used responses from a questionnaire administered in 2013 to municipalities of the province of Trento, aimed at collecting information on the actual public expenditure on vector control activities.
The estimated density of adult female mosquitoes (averaged between April 10th and September 30th) was between 4 and 88 per hectare in 2014 and between 9 and 198 in 2015, depending on the municipality (see Table 1). The higher abundance in 2015 is mostly due to the much higher temperatures recorded during summer. The initial reproduction numbers and the threshold for autochthonous transmission of chikungunya and dengue over time were estimated in a previous study [17]. Here, for each site and year, we computed the probability of autochthonous transmission of chikungunya and dengue originated by an imported infection in the absence of interventions. Higher vector densities during 2015 resulted in an increased risk of local transmission for both infections, compared to the previous year. The probability of observing at least one secondary case was estimated to be up to 30% for chikungunya and 15% for dengue in highly infested towns in 2015. Corresponding maximum probabilities in 2014 were around 20% for chikungunya and 5% for dengue. This means that 7 importations of chikungunya and 15 importations of dengue in towns most at risk would have a >90% probability of causing at least one secondary case in 2015. Sporadic transmission (less than 10 secondary cases) is by far the most likely scenario, especially for dengue (Fig 1). However, we found a low, but non-negligible, probability (up to 2.7%) that an uncontrolled chikungunya outbreak would produce more than 50 cases in several sites during 2015.
Routine preventive larvicide treatments can reduce significantly mosquito populations and consequently the probability and size of outbreaks triggered by sporadic importation of infected cases. To evaluate the overall effectiveness, we considered the expected number of total secondary infections per imported case. Under the baseline scenario of no control interventions, this index ranged from 0.1 to 5.2, depending on the site and year; corresponding numbers for dengue were everywhere below 0.5. Because of the smaller epidemiological risk of dengue, we discuss only the cost-effectiveness analysis on chikungunya, leaving corresponding results for dengue to the S1 Text.
For each site and year, and for each timing, effort level and assumed coverage, we evaluated the relative reduction in the expected number of secondary infections per imported case as a measure of effectiveness. Fig 2 and Table 2 show that all interventions with optimal effectiveness covered the month of July, which corresponds to the estimated period of steepest growth of the adult Ae. albopictus population in both years. We selected for further analyses only interventions with optimal timing for each effort level (Table 2; the reduction in mosquito abundance corresponding to the optimally timed interventions is reported in the S1 Text). We found that an increase in the effort level does not proportionally reduce the expected number of cases (Fig 3). In particular, an expansion in the coverage of breeding sites from 30% to 50% would be more effective than doubling the effort level while keeping the coverage at 30%. In general, interventions are most beneficial when the baseline risk is highest.
Towards an optimal allocation of resources, the benefits of reducing the potential number of transmitted cases needs to be compared with the intervention costs. Taking into account all possible clinical outcomes, including the probability of severe illness and of hospitalization, the estimated average cost per infection is 424.9 euros (95% CI 342–533) for chikungunya and 275.88 euros (95% CI 151–422) for each dengue infection. The corresponding average DALY loss per case is higher for chikungunya (0.45, 95% CI 0.10–1.12) than for dengue (0.29, 95% CI 0.15–0.44). In Fig 4, we show the relative probability that each effort level (including the no-intervention scenario) will maximize the NHB for each site, year, and coverage. Three main outcomes can be identified. The first is represented by larger cities (Trento, Belluno and Rovereto, all above 35,000 inhabitants) where non-intervention has the highest likelihood of being optimal. In these sites, the poor economic effectiveness of larviciding depends on the relatively low number of expected secondary cases even in the absence of treatment (Fig 3), combined with the high intervention costs due to the extent of the area to be covered. The second group consists of smaller towns where intervention is always beneficial (Povo, Santa Giustina, Tenno and Tezze, all below 10,000 inhabitants) and where higher effort levels have the highest probabilities of being optimal. Strigno (about 3,400 inhabitants) represents an exception to this rule, where the low intervention costs are counterbalanced by a very small transmission risk in the absence of interventions. Nonetheless, even in Strigno a low-effort intervention (single treatment) might be beneficial because of its low cost. The third situation occurs in towns of intermediate size (Feltre and Riva del Garda, between 20,000 and 35,000 inhabitants) where both the intervention costs and the transmission risks are high. In these cases, depending on the larviciding coverage, absence of intervention might be the optimal strategy in seasons of lower mosquito abundance (2014) while a low-to-moderate effort (1 to 3 treatments) might be the best choice in years of high infestation (2015). Overall, the probability that a more intensive intervention will be optimal increases with the coverage and with higher transmission risk (2015 vs. 2014). We also tested the cost-effectiveness of expanding the coverage by involving private citizens [7]. We found that this type of intervention might achieve significant additional reductions in the expected number of secondary cases and probability of local transmission (reported in the S1 Text). However, they are rarely optimal from the economic perspective because they require labour-intensive activities. Fig 5 reports results of the NHB analysis for a single larvicide treatment, but qualitative inferences are similar for more intensive efforts (see S1 Text). The only two instances where involvement of citizens was found to be economically beneficial were Povo and Tezze and only during the 2015 mosquito season, i.e. only where the urban size is small enough to keep intervention costs low and where the transmission risk at baseline is sufficiently high.
Two municipalities under study, Trento and Riva del Garda, had responded to a previously administered questionnaire on public expenditure on vector control, declaring an overall budget of 0.254 euro and 0.532 euros per inhabitant, respectively. In Trento, the most cost-effective activity predicted by our model was monitoring by ovitraps (Fig 4), which has an estimated average cost of 0.016 euro per inhabitant; in Riva del Garda, one or two larvicide applications per year would be likely optimal and would cost between 0.256 and 0.512 euros per inhabitant. Therefore, the most cost-effective strategies are sustainable with respect to the current allocated budget. We provide full details on questionnaires, municipality-specific answers and intervention costs in the S1 Text.
In this work, we evaluated the effect of routine larviciding against dengue and chikungunya, two viruses transmitted by bites of Ae. albopictus mosquitoes. We used data from two seasons of entomological surveillance in multiple sites from northern Italy to parametrize a mathematical model of mosquito population dynamics and control. The population model was coupled with a transmission dynamics model and a cost-effectiveness analysis to identify suitable routine vector control strategies for temperate climate municipalities in Europe. We found that, in the absence of interventions, the risk of autochthonous dengue transmission was low and limited to sporadic transmission in both years, because of the relatively low competence of European strains of Ae. albopictus. On the other hand, the risk of a chikungunya outbreak was estimated to be up to 30% in 2015, with a non-negligible probability of observing outbreaks larger than 50 cases in most sites.
We found that the most effective interventions in reducing the amount of expected locally transmitted cases were those for which the window of larvicide efficacy covered at least the month of July (Fig 2, Table 2). Larviciding reduced the probability of secondary cases only moderately, but it had an important impact in avoiding larger outbreaks. Our analysis included two seasons that were representative of a broad range of mosquito abundances, due to the remarkable temperature differences. The cost-effectiveness of larviciding depends on the actual mosquito abundance in a given year; however, general rules could be identified independently of the considered year: small villages (<10,000 inhabitants) with moderate-to-high mosquito abundances will maximally benefit of intense larviciding efforts made of season-round monthly treatment of public catch basins. For medium-sized towns (20–35,000 inhabitants) with high infestation rate, the benefits are partially offset by the higher cost of intervention; in these cases, a moderate larviciding effort (1 to 3 treatments within the season) is recommended. Larger cities in our study (>35,000 inhabitants) were characterized by a low or intermediate transmission risk, and the high costs of an intervention covering the entire urban area made it economically disadvantageous. In these situations, treating specific neighbourhoods with highest mosquito abundance (called ‘hot spot' approach [25]) may be cost-effective. In order to evaluate such a scenario, however, it would be necessary to model the complex effect of the urban layout on the spatial distribution of breeding sites and on the dynamics of mosquito populations [7], which is out of the scope of our study. Treatment of private breeding sites via the direct involvement of citizens by door-to-door visits was recommended only in small towns with high mosquito infestation. A survey on the allocated budget for mosquito control programs across different municipalities showed that expenses required for the most cost-effective interventions are sustainable for the considered area.
These results need to be contextualized with respect to our simplifying assumptions. First, all results are given conditionally on a uniform probability of importation of an infectious individual within a given epidemiological year. For comparison, in the considered provinces of Trento and Belluno, three imported cases of dengue and one imported case of chikungunya were recorded in 2014 (C. Rizzo, personal communication); however, the actual importation rate may vary significantly by year and time of the year, depending on spatio-temporal patterns of global epidemics and international travel. We did not consider reactive interventions that are implemented when a case of chikungunya or dengue infection is detected or after an outbreak has started (e.g., insecticide air spraying in the neighbourhood of the index case [26]). In addition, our results are relative to the prevention of arboviral transmission; however, there may be other purposes in vector control activities, such as the reduction of nuisance for citizens, which were not included in our analysis. For what concerns the economic assessment, we did not consider the impact of local transmission detection on the blood supply chain. Upon clinical confirmation of a locally transmitted arboviral infection, restrictions on the usage of blood bags collected in the region are applied to prevent transmission via transfusions, and screening tests on available blood supplies are implemented [26]. These additional interventions are quite expensive, and savings associated to the reduction of transmission risk granted by larvicides may dramatically offset the cost-benefit balance in favour of the intervention. However, these costs are difficult to estimate because of the lack of sufficient data.
We did not include other arboviroses transmitted by Ae. albopictus because of their lower epidemiological relevance to the considered area. For example, the risk of Zika virus transmission was found to be close to zero in the study region, even under conservative scenarios [17]. Nonetheless, we note that larvicides produce simultaneous benefits in preventing multiple diseases transmitted not only by Ae. albopictus but also by other affected mosquito species (e.g. West Nile virus associated to Culex pipiens L.). Furthermore, larviciding may assist in limiting the spread of other invasive mosquito species such as Aedes (Hulecoeteomyia) japonicus (Theobald) and Aedes (Hulecoeteomyia) koreicus (Edwards) [1, 27]. An interesting research question is how the balance of ecological interactions between mosquito species [28] may be offset by such interventions.
Other studies [2, 6, 7] have investigated the effectiveness of vector control in Europe using different approaches. The cost-effectiveness of larvicidal treatment against Ae. albopictus in temperate climates has been evaluated only in combination with other interventions during an ongoing outbreak [29, 30]; other studies were based on endemic (extra-European) settings where transmission is mainly mediated by Aedes (Stegomyia) aegypti (Linneus) [31, 32]. Overall, results from different studies and approaches, including our own, are consistent in highlighting the potential of larviciding towards reducing mosquito populations; however, this reduction will not result in a complete elimination of the risk of local chikungunya or dengue transmission. Additional strategies may integrate the control of risks from mosquito-borne diseases, including source reduction methods (e.g. identification and removal of breeding sites), mass trapping (e.g. via lethal ovitraps) and approaches leveraging ecological interactions (such as the use of Wolbachia bacteria or the release of genetically sterilized male mosquitoes). A comprehensive review of the potential for these strategies can be found in [9], but specific cost-effectiveness studies are needed to identify optimal strategies for vector control. European municipalities with temperate climate where Ae. albopictus is established may take advantage of results from this study when planning and timing routine larviciding interventions aimed to prevent or reduce epidemiological risks. Temperate European areas share with our study collection area similar temperature suitability for the transmission of arboviruses [33] and similar abundances of Ae. albopictus [34], so that results on the epidemiological effectiveness of larviciding should not differ significantly. More caution should be paid when extrapolating cost-effectiveness conclusions to different countries, given potential differences in health and intervention costs and in the choice of the WTP. Finally, we suggest that the proposed methodological approach may also be extended to European areas with different climates, conditional on the availability of local data on mosquito abundances estimated via entomological surveillance activities.
|
10.1371/journal.pgen.1005454 | In Vivo Evidence for Lysosome Depletion and Impaired Autophagic Clearance in Hereditary Spastic Paraplegia Type SPG11 | Hereditary spastic paraplegia (HSP) is characterized by a dying back degeneration of corticospinal axons which leads to progressive weakness and spasticity of the legs. SPG11 is the most common autosomal-recessive form of HSPs and is caused by mutations in SPG11. A recent in vitro study suggested that Spatacsin, the respective gene product, is needed for the recycling of lysosomes from autolysosomes, a process known as autophagic lysosome reformation. The relevance of this observation for hereditary spastic paraplegia, however, has remained unclear. Here, we report that disruption of Spatacsin in mice indeed causes hereditary spastic paraplegia-like phenotypes with loss of cortical neurons and Purkinje cells. Degenerating neurons accumulate autofluorescent material, which stains for the lysosomal protein Lamp1 and for p62, a marker of substrate destined to be degraded by autophagy, and hence appears to be related to autolysosomes. Supporting a more generalized defect of autophagy, levels of lipidated LC3 are increased in Spatacsin knockout mouse embryonic fibrobasts (MEFs). Though distinct parameters of lysosomal function like processing of cathepsin D and lysosomal pH are preserved, lysosome numbers are reduced in knockout MEFs and the recovery of lysosomes during sustained starvation impaired consistent with a defect of autophagic lysosome reformation. Because lysosomes are reduced in cortical neurons and Purkinje cells in vivo, we propose that the decreased number of lysosomes available for fusion with autophagosomes impairs autolysosomal clearance, results in the accumulation of undegraded material and finally causes death of particularly sensitive neurons like cortical motoneurons and Purkinje cells in knockout mice.
| Autophagy is a degradative pathway for the removal and subsequent recycling of dysfunctional intracellular components. The material destined for degradation is initially enclosed by a double membrane, the autophagosome. In autolysosomes, which result from fusion of autophagosomes with lysosomes, the material is finally broken down. Recent in vitro data suggested that the protein Spatacsin plays a pivotal role in the regeneration of lysosomes from autolysosomes. Spatacsin is encoded by SPG11, the most common gene mutated in autosomal recessive hereditary spastic paraplegia. Here we show that mice devoid of Spatacsin develop symptoms consistent with spastic paraplegia and progressively loose cortical motoneurons and Purkinje cells. In these mice degenerating neurons have a reduced number of lysosomes available for fusion with autophagosomes and consequently accumulate autolysosome-derived material over time. In the long term this causes death of particularly sensitive neurons like cortical motoneurons and Purkinje cells.
| Hereditary spastic paraplegias (HSPs) are a group of movement disorders characterized by length-dependent degeneration of upper motoneuron axons resulting in leg weakness and spasticity [1]. More than 70 genetically distinct forms (SPG1-SPG72) are currently recognized [2]. SPG11 represents a complicated form of HSP with cognitive decline, thinning of the corpus callosum, white matter lesions and cerebellar signs among other symptoms very similar to SPG15 [3]. While SPG11 is caused by SPG11 mutations [4], mutations in SPG15/ZFYVE26 underlie SPG15 [5]. Suggesting that SPG11 and SPG15 are pathophysiologically linked, the protein products of both SPG11 and SPG15, Spatacsin and Spastizin respectively, associate with the adaptor protein complex 5 (AP-5), which belongs to a group of tetrameric protein complexes involved in vesicular transport [6–8]. Interestingly, mutations in AP5Z1 encoding the ζ-subunit of AP-5 underlie SPG48 [6], which shares several clinical features with SPG11 and SPG15 [3].
The subcellular localization of the proteins and their suggested respective functions are quite controversial. DNA repair [6], cell division [9], autophagy [10], axon outgrowth [11], and endolysosomal trafficking have been proposed [12,13]. The latter was suggested because knockdown of individual AP-5 subunits in HeLa cells caused the cation-independent mannose 6-phosphate receptor to become trapped in clusters of early endosomes [12]. Also pointing to this direction degenerating neurons in a recent Spastizin knockout mouse model accumulated autofluorescent material in Lamp1-positive vesicular structures [13] and fibroblasts from both SPG11 and SPG15 patients displayed an enlarged Lamp1-positive compartment [14]. Because autophagosome numbers were increased in fibroblasts of SPG15 patients and in knockdown studies with primary mouse neurons, it was further proposed that the fusion of autophagosomes with lysosomes is impaired [10]. This concept has recently been challenged by in vitro studies on HeLa cells showing that Spatacsin and Spastizin are essential for the regeneration of lysosomes from autolysosomes, a process known as autophagic lysosome reformation (ALR) [15], which has so far only been observed in vitro [16]. According to this model impaired ALR is expected to lead to exhaustion of lysosomes available for fusion of autophagosomes and accumulation of autolysosomes.
We here provide data that neurons in Spatacsin knockout mice accumulate abnormal autolysosomes and autolysosome-related autofluorescent material. Autolysosomes also accumulate in knockout mouse embryonic fibroblasts (MEFs), while their lysosome numbers are decreased. Upon starvation lysosomes are depleted in MEFs of both genotypes, but only recover in wild-type during prolonged starvation in accordance with a defect of the regeneration of lysosomes from autolysosomes. Consistently, lysosomes are reduced in knockout Purkinje cells and cortical motoneurons, even before accumulation of autofluorescent material and overt neurodegeneration. The loss of particularly susceptible neurons like cortical motoneurons and Purkinje cells finally causes the complex neurological phenotype of Spatacsin knockout mice.
To model SPG11 in vivo, we injected cells of the ES cell clone EUCE0085_F05 from the European conditional mouse mutagenesis program (EUCOMM) into donor blastocysts. The resulting chimeric mice were mated with C57Bl6 wild-type (WT) mice to obtain mice with a heterozygous trapped locus. Because the gene-trap cassette is integrated into intron 1 of Spg11 (Fig 1A), the targeted locus is predicted to encode a cytoplasmic fusion protein of the 82 N-terminal amino acids of Spatacsin with βgeo under the control of the endogenous Spg11 promoter, while the following part of Spatacsin is lost.
The β-galactosidase activity of βgeo allowed us to assess the expression of Spg11 by LacZ staining of tissue sections of heterozygous trapped mice, which supports a broad expression pattern including cortex and hippocampus (Fig 1B–1E), cerebellum (Fig 1F), neurons of the brain stem (Fig 1G), and the spinal cord (Fig 1H). To get more information on the expression of Spg11 in different types of neurons we co-stained tissue sections for β-galactosidase and various marker proteins including NeuN, a broad neuronal marker, Ctip2, which preferentially labels layer V neurons [17], and parvalbumin, which is expressed in a subset of interneurons. From these co-stainings it appears that Spg11 is broadly expressed in different types of neurons including principal cells and inhibitory neurons (S1 Fig).
From matings of heterozygous trapped mice we obtained homozygous targeted offspring in the expected Mendelian ratio. While Northern analysis with a probe corresponding to part of exon 30 of Spg11 detected transcripts of the expected size in RNA isolated from WT brains, the transcript was absent in RNA isolated from homozygous trapped mice (Fig 1I). To detect the Spatacsin protein we generated monoclonal antibodies directed against an epitope within the α-solenoid domain of Spatacsin (Fig 1A) and affinity-purified the resulting antiserum. Confirming its specificity, the antibody detected a polypeptide of the predicted size of Spatacsin of 273 kDa in brain lysates from WT mice, which was absent from protein lysates isolated from brains of homozygous trapped mice (Fig 1J). Though the antibody was suited for Western blot analysis, it did not detect endogenous Spatacsin in immunostainings.
Because it was shown that Spatacsin co-precipitates with Zfyve26 and subunits of the AP-5 complex [6], we assessed whether Zfyve26 levels are changed in brain lysates of homozygous trapped mice. Indeed, Zfyve26 levels were reduced. In contrast levels of the β-subunit of the AP-5 complex (Ap5b1) were not changed (Fig 1K, though it has been reported that siRNA mediated knockdown of Spatacsin in HeLa cells caused a decrease of levels for the μ5-subunit of the AP-5 complex [12]. Decreased levels AP5B1 were also reported for fibroblasts isolated from SPG11 patients [15].
Our results are consistent with the assumption that the trapped Spg11 locus corresponds to a null allele and that mice homozygous for the trapped allele represent Spatacsin knockout (KO) mice. Further on, homozygous trapped mice are therefore also referred to as Spatacsin KO mice.
Spatacsin KO mice younger than 12 months of age did not show any obvious motor phenotype compared to WT littermates. Subsequently, KO mice developed a progressive gait disorder. For quantification of the motor phenotype we measured the foot-base-angle (FBA) at toe-off positions of the hind-paws, which decreased with age in KO mice (Fig 2A–2C). Moreover, the latency of KO mice to fall off an accelerating rotating rod decreased in aged mice (Fig 2D). Further suggesting a motor coordination defect the number of falls in the beam walking test was significantly increased at around 13 months of age (Fig 2E). Around this age the body weight of KO mice decreased, consistent with a deterioration of the overall health status of KO mice (Fig 2F). Thus, Spatacsin KO mice show a progressive worsening of motor performance compatible with complex HSP.
Suggesting a systemic neurodegenerative disorder the brain size (Fig 3A and 3B) and weight (Fig 3C) did not differ around 2 months of age but was reduced in 16-month-old KO mice. Quantification of NeuN-positive neurons of the motor cortex revealed a loss of large projection neurons in cortical layers V to VI at 16 but not at 8 months of age (Fig 3D–3F). This was further confirmed by staining with Ctip2 (Fig 3D and 3E). Neuron numbers in layers I-III, where most of the commissural neurons reside [18], were unchanged, which is consistent with the intact corpus callosum in aged Spatacsin KO mice (Fig 3G and 3H).
Large diameter axon fibers were reduced by roughly 50% at 8 and by roughly 75% at 16 months of age in the lumbar corticospinal tract (L4, Fig 3I–3K). In the cervical corticospinal tract it was reduced by 56% at 16 months of age (n = 3; Student’s t-test: p<0.001). In contrast to previous results from zebra fish [19] the overall structure of motor-endplates was not altered in KO mice (S2 Fig).
Purkinje cells were drastically reduced (Fig 3L–3N), while numbers of pyramidal cells in the hippocampus and spinal cord motoneurons were not changed in 16-month-old KO mice (S3 Fig).
Neuron loss in the cortex (Fig 4A and 4B) and the cerebellum (Fig 4E and 4F) was preceded by intraneuronal accumulation of autofluorescent material (emission wavelength 460–630 nm) and paralleled by the activation of astrocytes as evident from GFAP stainings (Fig 4C, 4D, 4G and 4H). Cells accumulating autofluorescence co-labeled with β-galactosidase, Ctip2, SatB2, parvalbumin or calbindin suggesting that principal cells as well as inhibitory interneurons are affected (S4 Fig). Though autofluorescent material also accumulated in other regions of the central nervous system like hippocampus (Fig 4I and 4J), different nuclei in the brain stem including the vestibular nuclei and the inferior olivary nucleus (Fig 4M–4P), and spinal cord neurons (Fig 4U and 4V), there was no evidence for astrocyte activation in these regions (Fig 4K, 4L, 4Q–4T, 4W and 4X).
To characterize the intraneuronal autofluorescent material in more detail, we stained brain sections for different subcellular marker proteins. Different from WT (Fig 5A–5A”), autofluorescent spots in Purkinje cells of KO mice were large, often clustered and were surrounded by membranes positive for the lysosomal marker protein Lamp1 (Fig 5B–5B”). As the contents of these vesicular structures stained for p62 (Fig 5D–5D”), a receptor for cargo destined to be degraded by autophagy, the autofluorescent deposits likely represent undegraded autolysosomal material.
Consistent with a defect of autolysosomal clearance, the number of autolysosomes, defined as vesicles positive for both Lamp1 and p62 (Fig 6A, 6G and 6K), and LC3-II levels (Fig 6B and 6B’) were increased in KO MEFs compared to WT. LC3-II levels further increased upon treatment with bafilomycin A1, which inhibits autolysosome acidification and hence autolysosomal degradation (Fig 6B and 6B’). Western analysis of Lamp1 levels did not reveal an alteration of overall Lamp1 levels in KO MEFs (Fig 6C and 6C’). Lysosomal pH as an important determinant for the activity of lysosomal proteases also did not differ between genotypes (Fig 6D) and the ratio between the mature and the precursor forms of the lysosomal protease cathepsin D was unchanged (Fig 6E and 6E’). The number of lysosomes, however, as defined as Lamp1-positive vesicular structures that did not co-stain for p62 was reduced in KO MEFs (Fig 6F). Though lysosomes were depleted upon induction of autophagy by starvation for 6 h in both WT and KO MEFs, only in WT lysosome numbers recovered to baseline after 14 h of ongoing starvation while they remained diminished in KO lysosomes (Fig 6G–6N and S5 Fig).
We asked whether the results obtained in MEFs also apply to neurons. As in cultured MEFs, overall Lamp1 levels were unchanged (Fig 7A and 7A’) and the processing of cathepsin D was intact in KO samples as judged from Western blot analysis of brain lysates (Fig 7B and 7B’). Similar results were obtained for region specific lysates of cortex, hippocampus, cerebellum, and spinal cord (S6 Fig). As a correlate of impaired autolysosomal degradation the ultrastructural analysis of Purkinje cells revealed membrane-bound vesicles filled with heterogeneous material including organelle-like structures in Spatacsin KO but not in WT mice (Fig 7C and 7D). In KO samples we further observed an accumulation of electron-dense deposits of irregular shape reminiscent of lipofuscin interspersed between abnormal autolysosomes, while only some typical lipofuscin particles were found in controls of the same age (Fig 7C and 7D). Similar membranous bodies and lipofuscin-like material were also found in cortical, hippocampal, and spinal cord neurons of aged KO mice (S7 Fig). Levels of p62 were strongly elevated in the Triton X-100 insoluble fractions of KO whole brain lysates as well as in lysates of selected brain regions (S6 Fig), whereas levels of Beclin-1, one of the key proteins for the initial steps of autophagosome formation [20], were not changed (Fig 7E and 7E’). We next quantified Lamp1-positive and p62-negative vesicles in Purkinje cell somata of brain sections from 2-month-old and 11-month-old mice, respectively (Fig 7F–7H). Strikingly, the number of lysosomes was already decreased in Purkinje cells of Spatacsin KO mice at 2 months of age before we observed autofluorescent deposits or any signs of Purkinje cell degeneration. This alteration was preserved at 11 months of age. Similar results were also obtained for cortical motoneurons (S8 Fig).
To study the pathophysiology of SPG11 we generated a Spatacsin KO mouse model. Consistent with HSP and proving the assumption that SPG11 is caused by Spatacsin loss-of-function, axons of cortical motoneurons degenerate in Spatacsin KO mice. Similar to human patients affected by SPG11, KO mice developed a progressive spastic gait disorder with cerebellar ataxia during the course of the disease [21], while we did not observe a thinning of the corpus callosum, which is one of the main features of SPG11 patients [21]. This discrepancy may be explained by the fact that corpus callosum phenotypes strongly depend on the respective mouse strain [22,23].
Spatacsin KO mice progressively loose large diameter axons of the corticospinal tract similar to other mouse models for HSP. In contrast to mouse models for pure HSP [24–26] but similar to our findings for Zfyve26 [13] cortical motoneurons and Purkinje cells finally die. This neuron loss is paralleled by activation of astroglia as reported for other mouse models with neurodegeneration [27], while this is not the case in regions without overt neuron loss like hippocampus or spinal cord.
Since we did not observe structural changes or motor defects in KO mice younger than 12 months of age, disruption of Spatacsin does not entail obvious neurodevelopmental defects, as might have been expected from compromised axon outgrowth in neurons derived from induced pluripotent stem cells from SPG11 patients and upon siRNA-mediated Spatacsin knockdown in mouse cortical neurons [11].
Neuron loss in Spatacsin knockout mice was preceded by accumulation of intracellular autofluorescent material, which was associated with Lamp1-positive membranes. This is reminiscent of neuronal ceroid lipofuscinosis (NCL), lysosomal storage disorders characterized by lysosomal accumulation of autofluorescent ceroid lipopigments and neurodegeneration [28]. Because the autofluorescent deposits in Spatacsin KO mice also stained for p62, a receptor for material delivered into autophagosomes, these structures rather represent abnormal autolysosomes instead of lysosomes. Moreover, the ultrastructural analysis revealed membrane-bound vesicles filled with heterogeneous material including organelle-like structures at different stages of degradation consistent with autolysosomes in neurons from different regions. Along this line, membranous structures reported in sural nerve biopsies [29] and iPSC-derived neurons from SPG11 patients [11] may represent abnormal autolysosomes. Their accumulation indicates that autophagic clearance is impaired in Spatacsin KO, while the fusion of autophagosomes with lysosomes still occurs.
In agreement with our findings in Purkinje cells, the number of autolysosomes, characterized as vesicles labeled for both Lamp1 and p62, was increased in KO MEFs, which fits with previous results from siRNA-mediated knockdown of Spatacsin in HeLa cells [15]. Moreover, levels of LC3-II, the lipidated form of LC3 recruited to autophagosomal membranes, were increased as well. Since LC3-II levels in MEFs further increased upon treatment with bafilomycin A1, which inhibits lysosomal acidification and hence autolysosome clearance [30], autophagy does not appear to be completely blocked in Spatacsin deficient cells. Because fibroblast proliferation was unchanged in SPG11 patients [14], compromised autophagy upon disruption of Spatacsin may be less critical for fibroblasts than for postmitotic cells like neurons. Notably, it was reported that disruption of either Atg5 or Atg7 in neurons, which nearly abolishes autophagy completely, caused a loss of cortical neurons and Purkinje cells within the first 6 postnatal weeks while other types of neurons were less sensitive [31,32]. Thus the milder phenotype in Spatacsin KO mice, in which motoneurons and Purkinje cells are preserved at 2 months of age, is compatible with a partial impairment of autophagy. It appears that cortical motoneurons and Purkinje cells are particularly sensitive to autophagy defects. The long axonal projections of cortical motoneurons and the complex dendritic arbors of Purkinje cells may render these cells particularly sensitive for secondary transport defects because of accumulation of autophagy substrates. Indeed, axonal transport was compromised in neurons derived from induced pluripotent stem cells obtained from SPG11 patients and upon siRNA mediated Spatacsin knockdown in cortical mouse neurons [11].
The defect of autophagic clearance observed upon disruption of Spatacsin could arise from a primary lysosomal defect, as Spatacsin has been shown to interact with the adaptor protein complex AP-5, which was suggested to play a role for endosomal sorting [7,12]. Accordingly mistargeting of proteins normally destined for lysosomes or missorted cargo proteins may accumulate within lysosomes. Both situations may result in lysosomal dysfunction and hence a diminished turnover of autolysosomes as suggested for different lysosomal disorders [33–35]. Consistent with data obtained upon knockdown of Spatacsin in HeLa cells [15], a major lysosomal defect in Spatacsin KO cells is rather unlikely, because the processing of the lysosomal protease cathepsin D and the lysosomal pH were unchanged. Instead of a lysosomal defect we observed a depletion of lysosomes in both MEFs and Purkinje cells of KO mice. Lysosomes can either be generated through the endosomal pathway via the trans-Golgi network [36] or can be regenerated from autolysosomes via a process called autophagic lysosome reformation (ALR). The latter process is characterized by budding of “protolysosomal tubules” from autolysosomes, which finally separate and mature into functional lysosomes [16,37]. In HeLa cells depleted for either Spatacsin or Spastizin these tubules did not evolve upon serum starvation and hence it was proposed that impaired ALR may underlie SPG11 and SPG15 [15]. Our finding that in Spatacsin KO MEFs lysosomes are not recovered after prolonged starvation together with diminished lysosome numbers in cortical motoneurons and Purkinje cells support this conclusion for SPG11.
Taken together, our observations provide first evidence that ALR, which has so far only been observed in vitro [16,37], is also relevant in vivo. Along this line we propose that a reduction of lysosomes available for fusion with autophagosomes upon disruption of Spatacsin causes a defect in autolysosomal degradation, a consecutive accumulation of undegraded material and finally neuronal death.
To disrupt Spatacsin in mice, we used the EUCE0085_F05 embryonic stem cell clone E14 (EUCOMM) harbouring of a genetrap cassette in the first intron of the Spg11 gene. This clone was injected into C57BL/6 donor blastocysts and transferred into foster mice. The resulting chimeric mice were mated with C57BL/6 mice to obtain heterozygous gene-trapped mice, which were subsequently mated to obtain homozygous trapped mice. For genotyping genomic DNA was isolated from tail biopsies. The primers “for” (cggctgcgggcagtctccaagtgc), “rev” (gggatgggaaaggttccgagaggc), and “cas_rev” (cgactcagtcaatcggaggactgg) were used in a single PCR reaction. The primer pair for/rev amplified a 256 bp fragment for the wild-type allele and the primer pair for/cas_rev a 167 bp fragment for the trapped allele. Experiments were performed on a mixed 129SvJ/C57BL/6 background in the 4th to 6th generation. Mice were housed in a 12 h light/dark cycle and fed on a regular diet ad libitum.
All animal experiments were approved by the Thüringer Landesamt für Lebensmittelsicherheit und Verbraucherschutz (TLLV) (application number 02-016/13) and were conducted under strict accordance with the ARRIVE guidelines.
Beam-walking and coordination test were performed on a horizontal plastic beam (1,000 mm long, 40 mm broad, 20 cm elevated from the ground) leading to the home cage as previously described [13]. For Rotarod analysis mice were placed on the rotating rod of the apparatus (Ugo basile). After constant speed (4 rpm) for a maximum of 2 min the speed was continuously accelerated (4–40 rpm in 5 min), and the latency until mice fell off the beam was recorded. The mean from two independent trials per day was used for statistical analysis.
For the probe we amplified a 602 bp cDNA fragment (part of exon 30 of Spg11) from mouse brain cDNA with the forward primer 5’-gcaaacactaacacacactccgcagtgg-3’ and the reverse primer 5’-gcaacaccagcactagatcctggc-3’. Northern blot analysis was performed as described previously [38].
Monoclonal antibodies were raised against the epitope EKLSSGSISRDD (amino acids 1400–1411) of the Spatacsin protein in BALB/C mice (c346, Abmart). The affinity-purified antibody was used in a dilution of 1:50. Our polyclonal rabbit anti-Zfyve26 antibody described previously [13] was used at a dilution of 1:50.
The following commercially available antibodies were used: mouse anti-Calnexin (1:1,000, BD Biosciences); goat anti-Ap5b1 (1:500, Santa Cruz); rabbit anti-β-Galactosidase (1:250, Chemicon); rabbit anti-Calbindin D-28K (1:1,000, Millipore); mouse anti-parvalbumin (1:5,000, Swant); rat anti-Ctip2 (1:200, Abcam), mouse anti-SatB2 (1:100, Santa Cruz); α-bungarotoxin conjugated with Alexa Fluor555 (1:500, Life Technologies); mouse anti-NeuN (1:1,000, Millipore); mouse anti-GFAP (1:1,000, Millipore); rat anti-Lamp1 (1:500 for immunofluorescence studies; 1:1,000 for immunoblots, BD Pharmigen); rabbit anti-LC3 (1:500 for immunoblots, Novus Biologicals); mouse anti-p62 (1:250 used for immunofluorescence studies; 1:1,000 used for immunoblots, Abcam); rabbit anti-Beclin-1 (1:500, Santa Cruz); goat anti-CtsD (1:500, Santa Cruz); rabbit anti-β-actin (1:2000, Abcam); goat anti-Gapdh (1:500, Santa Cruz). Horseradish peroxidase—labelled secondary antibodies for Western blotting: goat anti-rabbit and goat anti-mouse (both 1:4,000, Amersham Bioscience); goat anti-rat (1:2,000, Santa Cruz); rabbit anti-goat (1:1,000, Sigma-Aldrich). Fluorescently labelled secondary antibodies: goat anti-rabbit, goat anti-mouse, or goat anti-rat coupled with Alexa 488 and Alexa 546, respectively (1:1,000, Life Technologies); goat anti-mouse, goat anti-rabbit or goat anti-rat coupled with Cy5 (1:1,000, Jackson ImmunoResearch Laboratories). Nuclei were counterstained with Hoechst-33258 (1:10,000; Molecular Probes).
For immunoblotting brain tissue lysates were prepared as described [13]. Triton-X 100 insoluble fractions from total brain and spinal cord as well as from brain specific regions like cortex, hippocampus, and cerebellum were prepared from three mice per genotype as described previously [39]. For the hippocampus 6 hippocampi per genotype were pooled to prepare protein lysates. All samples were denatured for 5 minutes at 95°C in Laemmli buffer and separated by SDS PAGE and blotted onto PVDF membranes (Roche), which were blocked with 2.5% (w/v) milk powder and 2.5% (w/v) BSA in TBS-T (137 mM NaCl, 2.7 mM KCl, 19 mM Tris base, 1% (w/v) Tween). Proteins were either detected with the ECL Plus Western Blotting Detection System (GE Healthcare) on a LAS 4000 system (GE Healthcare) or based on fluorescence using a LI-COR Odyssey detection system.
Animals were anaesthetized with isoflurane (Actavis) and perfused transcardially with 4% PFA in 1xPBS. Brains were removed and post-fixed in 4% PFA overnight at 4°C.
LacZ stainings of tissue sections were performed as described [40].
For histological analysis tissues were either embedded in paraffin or in Tissue-Tek (Sakura). Sections from paraffin embedded tissues were 8μm and cryosections 20μm thick. For histological analysis sections were stained either with hematoxylin/eosin or cresyl violet acetate (Nissl) according to the manufacturers’ protocols (Sigma-Aldrich). Images were captured with an Olympus DP70 microscope and further analysed by ImageJ. Pyramidal neurons in the Stratum pyramidale of the hippocampus were counted for corresponding regions and normalized to the respective area. The quantification of alpha-motoneurons and large diameter corticospinal axons was performed on semi-thin cervical and lumbar sections stained with Richard’s Blue [41]. Large diameter axons defined by a diameter > 4μm were counted. Alpha-motoneurons were identified because of the location in the ventral horn and their characteristic morphological appearance. Neuromuscular junctions were stained with α-bungarotoxin according to the manufacturer’s protocol (Life Technologies) in 20μm cryosections of either the gastrocnemius muscle from the hindlimb or the triceps brachii muscle from the forelimb, respectively. For quantification of cortical neurons 40μm free floating sagittal brain cryosections were stained for NeuN and mounted. Images of the motor cortex were taken with a Leica TCS SP5 confocal scanning fluorescence microscope. Neurons were quantified with the cell counter plug in and the area measurement tool of ImageJ.
Free floating 20μm sections of the brain or primary cells were rinsed three times with 1xPBS, then fixed for 15 minutes in 4% PFA in 1xPBS at room temperature and washed three times for 10 min in 1xPBS. 0.25% Triton-X in 1xPBS was used to permeabilize the cells. After rinsing the cells once with 1xPBS, blocking solution (5% goat serum in 1xPBS) was added. Primary and secondary antibodies were applied in blocking solution. Images were taken with a Leica TCS SP5 confocal scanning fluorescence microscope with the Z-stack module.
To analyze whether the autofluorescent deposits co-localize with subcellular markers in brain tissue sections the fluorescent signal of deposits and the Cy5 secondary antibodies were recorded and further analyzed by linear unmixing as described previously [13].
As lysosomes we defined Lamp1-positive but p62-negative vesicular structures, while autolysosomes are characterized by the presence of Lamp1 and p62. In order to analyze the number of lysosomes 40 μm thick sagittal brain sections were co-stained for Lamp1 and p62. Only sections of somata of Purkinje cells not extending beyond the image boundary and hit vertically in respect to the nucleus were selected. The images were recorded with the Leica TCS SP5 confocal scanning fluorescence microscope. The number of free lysosomes were counted and normalized to the area of the cell soma with ImageJ. Co-localization between Lamp1 and p62 as well as between fusion proteins and subcellular markers were performed in BioImageXD as described [42].
For semi- and ultrathin sections, 2 animals per genotype were perfused with 50 ml fixative (4% paraformaldehyde, 1% glutaraldehyde). Brain and spinal cord were removed and post-fixed overnight at 4°C. 150μm sagittal and coronal sections of brain and spinal cord were cut with a vibratome (Leica Microsystems) and processed as described [41]. Semithin sections were stained with Richard’s blue. Ultrathin 80 nm sections (Ultratome III, LKB Instruments) were mounted on filmed copper grids (100 mesh), post-stained with lead citrate, and studied in a transmission electron microscope (EM 900, Zeiss) at 80 kV.
Mouse embryonic fibroblasts (MEFs) were prepared from E13.5 mouse embryos as described [13]. In order to assess the number of free lysosomes MEFs were cultured on 13 mm diameter coverslips (Marienfeld) in 24-well plate (Greiner) and maintained in DMEM medium (Life Technologies) with or without (starvation condition) 10% FBS and 2mM L-glutamine as described [37]. Cells at baseline conditions and cells starved for 6 and 14 h were fixed with 4% PFA in 1xPBS, rinsed with 1xPBS, and co-stained with anti-rat-Lamp1 and anti-mouse-p62 as described above. Images were digitally acquired by a Leica TCS SP5 confocal scanning fluorescence microscope and the number of Lamp1-positive and p62-negative lysosomes quantified with ImageJ. To inhibit autophagy cells were incubated with medium containing 100 nM Bafilomycin A1 (Santa Cruz) for 16h.
Lysosomal pH measurements were carried out as described previously in [43]. More than 1,000 lysosomes from 3 independent experiments were analysed per genotype.
For repeated experiments two-way ANOVA followed by Bonferroni post-hoc tests were used to compare between genotypes. For morphological and quantitative western blot analysis Student’s two-tailed t-test was used. Data are shown as mean±SEM if not indicated otherwise.
|
10.1371/journal.pgen.1004545 | Phosphorylation of a Central Clock Transcription Factor Is Required for Thermal but Not Photic Entrainment | Transcriptional/translational feedback loops drive daily cycles of expression in clock genes and clock-controlled genes, which ultimately underlie many of the overt circadian rhythms manifested by organisms. Moreover, phosphorylation of clock proteins plays crucial roles in the temporal regulation of clock protein activity, stability and subcellular localization. dCLOCK (dCLK), the master transcription factor driving cyclical gene expression and the rate-limiting component in the Drosophila circadian clock, undergoes daily changes in phosphorylation. However, the physiological role of dCLK phosphorylation is not clear. Using a Drosophila tissue culture system, we identified multiple phosphorylation sites on dCLK. Expression of a mutated version of dCLK where all the mapped phospho-sites were switched to alanine (dCLK-15A) rescues the arrythmicity of Clkout flies, yet with an approximately 1.5 hr shorter period. The dCLK-15A protein attains substantially higher levels in flies compared to the control situation, and also appears to have enhanced transcriptional activity, consistent with the observed higher peak values and amplitudes in the mRNA rhythms of several core clock genes. Surprisingly, the clock-controlled daily activity rhythm in dCLK-15A expressing flies does not synchronize properly to daily temperature cycles, although there is no defect in aligning to light/dark cycles. Our findings suggest a novel role for clock protein phosphorylation in governing the relative strengths of entraining modalities by adjusting the dynamics of circadian gene expression.
| Circadian clocks are synchronized to local time by daily cycles in light-dark and temperature. Although light is generally thought to be the most dominant entraining cue in nature, daily cycles in temperature are sufficient to synchronize clocks in a large range of organisms. In Drosophila, dCLOCK is a master circadian transcription factor that drives cyclical gene expression and is likely the rate-limiting component in the transcriptional/translational feedback loops that underlie the timekeeping mechanism. dCLOCK undergoes temporal changes in phosphorylation throughout a day, which is also observed for mammalian CLOCK. However, the role of CLOCK phosphorylation at the organismal level is still unclear. Using mass-spectrometry, we identified more than a dozen phosphorylation sites on dCLOCK. Blocking global phosphorylation of dCLOCK by mutating phospho-acceptor sites to alanine increases its abundance and transcriptional activity, leading to higher peak values and amplitudes in the mRNA rhythms of core clock genes, which likely explains the accelerated clock speed. Surprisingly, the clock-controlled daily activity rhythm fails to maintain synchrony with daily temperature cycles, although there is no observable defect in aligning to light/dark cycles. Our findings suggest a novel role for clock protein phosphorylation in governing the effective strengths of entraining modalities by adjusting clock amplitude.
| A large variety of life forms manifest circadian (≅24 hr) rhythms in behavior and physiology that are driven by endogenous cellular clocks or pacemakers [1], [2]. Perhaps the most biologically relevant property of circadian clocks is that they can be synchronized (entrained) to local time by external time cues, a feature that endows organisms with the ability to anticipate environmental changes and hence perform activities at optimal times during the day. The main environmental synchronizing agents of circadian clocks in nature are the daily cycles in light/dark and ambient temperature. In general, photic cues are the most potent synchronizing agent for organisms, whereas thermal entrainment is less powerful [3], [4]. Work in the last 20 years using a variety of model organisms has revealed that the molecular logic underlying circadian clock mechanisms is highly conserved [2]. Circadian clocks are based on intracellular mechanisms that involve a core transcriptional translational feedback loop (TTFL) composed of central clock proteins that drive daily oscillations in their own gene expression as well as downstream clock-controlled genes (ccgs). Daily oscillations in the transcript levels of ccgs ultimately drive many of the rhythmic behaviors and physiologies manifested by organisms.
The rate-limiting component of the main TTFL in Drosophila is the basic-helix-loop-helix (bHLH) PAS domain containing transcription factor termed dCLOCK (Drosophila CLOCK; dCLK) [5], which forms a heterodimer with CYCLE (CYC), another bHLH-PAS containing clock transcription factor [6]. The dCLK-CYC heterodimer binds to E box DNA elements inducing the expression of the clock genes period (per) and timeless (tim), in addition to other clock genes and ccgs. Subsequently, the PER and TIM proteins interact in the cytoplasm and after a time-delay translocate to the nucleus where they function with other factors to inhibit the transcriptional activity of dCLK-CYC. Eventually, the levels of PER and TIM decline in the nucleus, facilitating another round of dCLK-CYC-mediated transcription. In a “secondary” stabilizing TTFL, the dCLK-CYC heterodimer induces the expression of PAR domain protein 1ε (pdp1ε) and vrille (vri), whose protein products (i.e., PDP1ε and VRI) in turn activate and repress the expression of dClk, respectively, leading to daily cycles in dClk mRNA levels [7], [8]. Mammalian clocks also use a CLOCK-based transcription factor in their main TTFL, which involves a heterodimer comprised of mCLOCK (mammalian CLOCK; mCLK) and BMAL1 (homolog of CYC) that governs rhythmic expression of the negative regulators Per1-3, in addition to other clock genes and ccgs [9].
Although TTFLs constitute a major molecular framework for the oscillatory behavior of cellular clocks, posttranslational modifications of clock proteins are central to maintain proper timekeeping functions by regulating clock protein stability, sub-cellular localization and activity [10]–[14]. A well-studied example of clock protein phosphorylation is the progressive phosphorylation of PER, which has a critical role in setting the pace of the clock and controlling temporal changes in dCLK-CYC-mediated transcription by regulating PER stability, timing of nuclear entry and how long it persists in the nucleus [15]–[22]. Newly synthesized PER is present as non-to-hypo-phosphorylated isoforms in the late day/early night and undergoes progressive increases in the extent of phosphorylation, culminating in the appearance of mostly or exclusively hyper-phosphorylated isoforms in the late night/early day that are recognized for rapid degradation by the 26S proteasome [10]. Numerous PER-relevant kinases have been identified, with DOUBLETIME [DBT; homolog of vertebrate casein kinase Iδ/ε (CKIδ/ε)] [21], [23] operating as the major kinase regulating temporal changes in the stability of PER. Other kinases include SHAGGY [SGG; homolog of vertebrate glycogen synthase kinase 3β (GSK3β)] [24], casein kinase 2 (CK2) [25], [26] and NEMO [15], [27].
dCLK also undergoes circadian changes in phosphorylation state, but in a manner different from that of PER [28], [29]. dCLK is present in a mostly intermediate phosphorylated state throughout the day, converting to largely hyper-phosphorylated isoforms in the late night/early day. DBT stably interacts with PER throughout most of its daily life-cycle and this association likely facilitates the ability of DBT to regulate dCLK [28]–[31]. Although the role(s) of dCLK phosphorylation is not clear it appears that hyper-phosphorylated isoforms have decreased stability and possibly reduced transcriptional activity [28]–[30]. In addition to DBT, several kinases such as protein kinase A (PKA), CaMKII, MAPK, and NMO have been implicated in regulating the activity and/or levels of dCLK [27], [32]. More recently CK2 was reported to act directly on dCLK, stabilizing it while reducing its activity [33]. The mammalian CLOCK protein also manifests circadian oscillations in phosphorylation in vivo [34], [35], which is triggered by hetero-dimeric complex formation with BMAL1 [34], [35]. Mass spectrometric analysis of purified mCLK from the mouse liver identified Ser38, Ser42, and Ser427 as sites phosphorylated in vivo [36]. Ser38 and Ser42 are located in the bHLH region and phosphorylation of those residues down-regulates transcriptional activity of mCLK via decreasing binding activity to E box element [36]. Phosphorylation of Ser427 is reported as being dependent on GSK-3β activity and relevant for degradation of mCLK [37]. PKG and PKC have been implicated as mCLK kinases regulating phase resetting [38], [39]. Despite these advances using several animal model systems, it is still unclear how CLOCK phosphorylation impacts the function of circadian timing systems at the organismal level.
In this study, we used a simplified Drosophila S2 cell culture system in combination with mass spectrometry to map phosphorylation sites on dCLK. Our results indicate that dCLK is highly phosphorylated (at least 14 phospho-sites). In S2 cells, mutated versions of dCLK where all the mapped Ser/Thr sites were switched to Ala (herein referred to as dCLK-15A) manifested increased E box dependent transcriptional activity without affecting interactions with other core clock partners such as CYC and PER. In flies, dCLK-15A protein is exclusively hypo-phosphorylated suggesting that we identified, at the very least, a major portion of the total phosphorylation sites found on dCLK in flies. Expression of dCLK-15A rescues the arrhythmicity of Clkout flies yet with an approximately 1.5 hr shorter period. Consistent with a role in regulating protein stability, the levels of dCLK15A are substantially higher compared to the control situation, which along with increases in transcriptional activity likely explains the faster pace of the clock. The daily peak levels in per/tim mRNA and protein reached higher values in dCLK-15A expressing flies, further supporting the notion that dCLK levels are normally rate-limiting in the clock mechanism. Surprisingly, the clock-controlled daily activity rhythm in dCLK-15A mutant flies fail to maintain synchrony with daily temperature cycles, although there is no defect in aligning to light/dark cycles. Together, our findings indicate that in animal systems, the post-translational modification of a master circadian transcription factor plays a critical role in setting the pace of the clock and regulating circadian entrainment.
As an initial attempt to better understand the role(s) of dCLK phosphorylation we sought to map phosphorylation sites using recombinant protein production in cultured Drosophila S2 cells. This simplified cell culture system was successfully used to identify physiologically relevant phosphorylation sites on Drosophila PER [15], [16], [18], [40]. Prior work showed that production of recombinant dCLK in S2 cells leads to significant shifts in electrophoretic mobility that are due to phosphorylation [29]. Thus, we established S2 cell lines stably expressing HA-dClk-V5 under the inducible metallothionein promoter (pMT). Total cell lysates were subjected to immunoprecipitation with anti-V5 antibody, followed by multi-protease digestion, titansphere nanocolumn phosphopeptide enrichment, and tandem mass spectrometry, as previously described [16], [41]. We identified 14 phosphorylation sites on dCLK, all of which are at Ser residues, with the possible exception of Tyr607 (Table 1). Many of the identified phosphorylation sites are in the C-terminal half of dCLK, which contains several Q-rich regions that might function in transcriptional activation (Figure 1A). Phosphorylation was also detected at sites close to the N- and C-terminus of the dCLK protein. Interestingly, no phosphorylation sites were found in any of the known functional domains of dCLK; e.g. bHLH, PAS domains and Q-rich regions (Figure 1A and Table 1).
In preliminary studies we individually mutated each of the identified phosphorylated Ser residues to Ala residues but did not see major effects on dCLK electrophoretic mobility, except for the S859A mutant version of dCLK, which manifested slightly faster electrophoretic mobility. (Figure S1A and B). The transcriptional activities of most single site mutants were somewhat increased (≤2 fold), except for the S924A mutant version of dCLK, which manifested a slight but reproducible decrease (Figure S1C). Overall, our initial studies in S2 cells were not able to identify whether certain individual phospho-sites are particularly significant in regulating dCLK metabolism or activity. While ongoing work is aimed at better understanding the roles of individual phospho-sites, in this study we focused on more global aspects of dCLK phosphorylation by generating a mutant version wherein all the Ser phospho-acceptor sites identified by mass spectrometry were switched to Ala. Since the mass spectrometry data did not unambiguously identify which Ser among amino acids 209–211 is phosphorylated, we switched all 3 Ser to Ala. In addition, although Tyr 607 or Ser 611 is phosphorylated, to focus on Ser phosphorylation, we only mutated Ser 611 to an Ala. By using site-directed mutagenesis, we serially mutated the aforementioned 16 Ser to Ala (dCLK-16A). The electrophoretic mobility of dCLK-16A is indistinguishable from that of λ-phosphatase treated wild-type dCLK and was not altered by λ-phosphatase treatment (Figure 1B), indicating that we mapped most, if not all, the sites on dCLK phosphorylated by endogenous kinases in S2 cells.
dCLK-16A interacts with either CYC or PER proteins to a similar extent as that observed for the wild-type version (dCLK-WT), demonstrating that dCLK-16A is not grossly misfolded (Figure 1C). In addition, our findings suggest that the phosphorylated state of dCLK is not a major signal regulating interactions with core clock partners. Consistent with prior work, we observed increases in non/hypo-phosphorylated isoforms of dCLK when dPER is co-expressed (Figure 1C, compare lane 1 and 6) [31]. With regards to transcriptional activity, dCLK-16A is more potent compared to dCLK-WT in stimulating E-box dependent transcription (Figure 1D), while still maintaining its sensitivity to inhibition by dPER (Figure 1E). Earlier findings showed that hyper-phosphorylated dCLK is less stable and that DBT might contribute to this instability, although the exact role of DBT is not clear [28]–[30]. We compared the stabilities of dCLK-16A and dCLK-WT under a variety of conditions, including overexpressing DBT, but did not detect a significant difference (Figure S2A and B), suggesting we did not map one or more phosphorylation sites critical for regulating dCLK stability and/or the pathway for dCLK degradation in S2 cells is not identical to that in flies (see below). Taken together, the results obtained using well-established S2 cell based assays indicate that dCLK-16A retains key clock-relevant biochemical functions and suggest that global phosphorylation of dCLK reduces its transactivation potential.
To investigate whether the dCLK phosphorylation sites we identified play a physiological role in the Drosophila circadian timing system, we first evaluated the ability of a novel wildtype dClk transgene to rescue behavioral rhythms in the arrhythmic Clkout genetic background (herein, termed as p{dClk-WT};Clkout). Clkout is a newly described arrhythmic null mutant that does not produce dCLK protein (Mahesh et al., submitted). The transgene was constructed with a 13.9 kb genomic fragment that contains the dClk gene, which we modified by introducing a V5 epitope tag at the C-terminus of the dClk open reading frame for enhanced protein surveillance. Flies were exposed to standard entraining conditions of 12 hr light∶12 hr dark cycles [LD; where zeitgeber time 0 (ZT0) is defined as lights-on] at 25°C, followed by several days in constant dark conditions (DD) to measure free-running behavioral periods. In the behavioral analysis, p{dClk-WT};Clkout flies manifested robust locomotor activity rhythms with normal ∼24 hr periods (Table 2, Mahesh et al., submitted), indicating that the circadian clock system functions properly in these flies.
Next, we sought to generate transgenic flies harboring a dCLK-16A version of the dClk rescue transgene. However, because of technical difficulties in generating a version that also included replacing Ser5 with Ala, we made a dClk version wherein the other 15 Ser residues were switched to Ala, termed dClk-15A. In S2 cells, dCLK-15A behaves similar to dCLK-16A, including no observable effect of phosphatase treatment on electrophoretic mobility and enhanced E-box dependent transcriptional activity (Figure S3). Although phosphorylation of Ser5 might affect dCLK function in a manner that is not revealed in the S2 cell based assays we used, the CLK-15A protein contains the majority of phosphorylation sites and should address if global phosphorylation of dCLK plays an important role in the circadian timing system. Two independent lines of transgenic flies harboring the dClk-15A transgene were obtained and circadian behavior was monitored in the Clkout genetic background (referred to as p{dClk-15A}, 2M; Clkout and p{dClk-15A}, 6M; Clkout). In sharp contrast to flies harboring the control version of dClk, the two independent lines of p{dClk-15A};Clkout flies manifested generally weaker behavioral rhythms that are approximately 1.5 hr shorter compared to their wild-type counterparts (Table 2).
Under standard conditions of LD at 25°C, D. melanogaster exhibits a bimodal distribution of activity with a “morning” and “evening” bout of activity centered around ZT0 and ZT12, respectively. Although p{dClk-15A};Clkout flies manifest the typical bimodal distribution of locomotor activity, the onset of both the morning and evening bouts of activity were earlier (Figure 2, compare panels B and C to A), consistent with the shorter free-running periods. The Clkout flies only showed a “startle” response to the lights-on transition but no rhythmic behavior (Figure 2E). In constant dark conditions, the downswing in evening activity is clearly earlier in p{dClk-15A};Clkout flies, in agreement with the shorter free-running period (Table 2, Figure 2 and S4). We also examined the locomotor behavior of flies harboring the dClk-WT transgene in a wild-type genetic background, resulting in flies with four copies of the dClk gene (herein referred to as p{dClk-WT};+/+). The circadian period was shortened to approximately 23 hr (Table 2 and Figure 2D), which is well correlated with previous reports demonstrating that increasing the copy number of dClk shortens the circadian period of behavioral rhythms [42], [43].
A hallmark property of circadian rhythms is that the period length is very constant over a wide range of physiologically relevant temperatures, termed temperature compensation [44]. To investigate whether phosphorylation of dCLK might have a role in temperature compensation, we analyzed behavioral rhythms at three standard temperatures (i.e., 18°, 25° and 29°C). Although we noted a decrease in rhythmicity for dClk-15A;Clkout flies at 29°C, the periods were quite similar over the temperature range tested (Table 2), suggesting that global phosphorylation of dCLK does not play a major role in temperature compensation.
We examined the temporal profiles of dCLK protein by analyzing head extracts prepared from p{dClk-WT};Clkout and p{dClk-15A};Clkout flies in LD conditions (Figure 3A and C). dCLK-WT protein undergoes daily changes in phosphorylation that are consistent with earlier results probing endogenously produced dCLK; namely, hypo- to medium- phosphorylated isoforms present during the mid-day/early night (e.g., ZT 8 to ZT 16) and mostly hyper-phosphorylated isoforms present in the late night/early day (e.g., ZT20 to ZT4) (Figure 3A) [28], [29]. However, the mobility of dCLK-15A was similar throughout a daily cycle (Figure 3A), and co-migrated with λ phosphatase treated dCLK-WT (Figure 3B). Thus, similar to results in S2 cells, dCLK-15A exhibits little to no phosphorylation in vivo, suggesting that the phospho-sites we identified by mass spectrometry comprise, at the very least, a major portion of the total phosphorylation sites found on dCLK in flies (it is also possible that one or more of the phospho-sites we mutated are required for phosphorylation at other sites, but this would still result in a mainly hypo-phosphorylated dCLK protein). Intriguingly, the levels of dCLK-15A were substantially higher compared to dCLK-WT throughout a daily cycle. Quantification of immunoblots indicated that the average daily levels of dCLK-15A are about 2.5 times more than those of dCLK-WT (Figure 3C).
To examine whether the high levels of dCLK-15A proteins results from elevated mRNA abundance, we measured dClk transcript levels. As reported previously, although the overall daily abundance of dCLK-WT protein is essentially constant throughout a daily cycle, dClk-WT mRNA levels oscillate with peak amounts attained during the late night-to-early day and reaching trough values around ZT12 [45], [46]. The daily oscillation in dClk-15A mRNA abundance is similar to the wild-type situation and even seemed to have lower peak levels (Figure 3D). These results indicate that in general global phosphorylation of dCLK decreases its stability in vivo, consistent with prior findings using S2 cells [28], [29].
To further examine the status of the clockworks, we measured the daily profiles in per and tim transcripts and protein levels. Both per and tim mRNA levels in p{dClk-15A};Clkout flies were reproducibly higher, especially during the daily upswing that occurs between ZT4 – 12 (Figure 4A and B). These result further support the notion that dCLK levels are normally rate-limiting for circadian transcription and suggest that despite the increased abundance of dCLK-15A there is sufficient PER to engage in normal repression of dCLK-15A/CYC activity. Indeed, PER protein levels were reproducibly higher in p{dClk-15A};Clkout flies (Figure 4C and D), consistent with the increased transcript levels. In p{dClk-15A};Clkout flies, TIM protein levels were slightly but reproducibly increased (Figure 4E and F). The increased daily upswing in tim mRNA levels in p{dClk-15A};Clkout flies might have a smaller effect on overall TIM protein levels because light induces the rapid degradation of TIM [47], possibly limiting the ability of TIM to accumulate during the early night prior to the start of transcriptional feedback repression. Taken together, we show that the stability of dCLK in flies is strongly increased by blocking phosphorylation at one or more sites. Moreover, augmenting the total abundance of dCLK accelerates the daily accumulation of per/tim transcripts and increases their peak levels, indicative of higher overall dCLK-CYC-mediated transcription. In addition, increased in vivo transcriptional activity of dCLK-15A may also contribute to higher dCLK-CYC-mediated transcription, as is the case in S2 cells (Figure 1D). These results demonstrate that dCLK phosphorylation plays a key role in setting the amplitudes of the per mRNA and protein rhythms, molecular oscillations that are central to the primary TTFL and circadian speed control in Drosophila [16], [48].
Besides light-dark cycles, daily changes in temperature can also synchronize (entrain) circadian rhythms in a wide variety or organisms [49]. Prior work showed that D. melanogaster can entrain to daily cycles of alternating 12 hr ‘warm’/12 hr ‘cold’ cycles that differ by as little as 2–3°C [50]–[52]. To determine if flies expressing dCLK-15A have a defect in entraining to temperature cycle, flies were kept in constant darkness, exposed to 12 hr∶12 hr temperature cycles of 24°C∶29°C (TC) and locomotor activity rhythms analyzed (Table 3 and Figure 5). The daily distribution of activity in p{dClk-15A};Clkout flies is strikingly different compared to the wild-type control. As previously observed for wildtype strains of D. melanogaster entrained to daily temperature cycles [50]–[52], p{dClk-WT};Clkout flies exhibit the classic “anticipatory” rise in activity just prior to the low-to-high and high-to-low temperature transitions, similar to what is observed in light-dark cycles around ZT0 and ZT12 (Figure 5 and S5; there is a “startle” response at the transition from low-to-high temperature that is also observed in Clkout flies, analogous to the transient burst in activity at lights-on in a LD cycle). In sharp contrast, during the beginning of the temperature entrainment regime although p{dClk-15A};Clkout flies also manifest two activity peaks, they occur much earlier at around the middle of the warm- and cryo-phases (Figure 5B).
Interestingly, while the timing of the “startle” response at the transition from low-to-high temperature remained constant in p{dClk-15A};Clkout flies, the timing of the major activity peak occurring during the mid-warm phase appeared to progressively advance on subsequent days in TC (Figure 5B). Analysis of individual activity records also confirmed this trend (Figure 5E). The abnormal behavioral pattern under temperature cycles for p{dClk-15A};Clkout flies was also observed when flies were exposed to a temperature cycle after first treating them with constant light for 6 days to abolish the circadian timing system (Figure S6). Thus, the defective entrainment of p{dClk-15A};Clkout flies to TC is not dependent on the status of the clock at the time that the temperature entrainment was initiated. Furthermore, although the main activity bout in p{dClk-15A};Clkout flies advanced on subsequent days during TC, the rate of advancement was clearly greater during free-running conditions following TC (Figure S5), suggesting partial entrainment during TC. Following entrainment to TC, the free-running period of dCLK-15A producing flies is about 1.5 hr shorter compared to the wild type dCLK-WT control (Table 3 and Figure S5). The faster running clock in p{dClk-15A};Clkout flies during free-running conditions after exposure to TC is consistent with results obtained following entrainment to LD (Table 2). Thus, it appears that when exposed to daily temperature cycles p{dClk-15A};Clkout flies can adopt some alignment with the entraining conditions, albeit without a normal phase relationship, but that this entrainment is weak and the flies partially free-run at their shorter endogenous periods, leading to progressive advances in their behavioral rhythm relative to the 24 hr entraining regime. Although not as dramatic, the timing of the warm-phase activity bout in p{dClk-WT};+/+ flies also advanced during TC (Figure 5F), whereas this was not the case for p{dClk-WT};Clkout flies (Figure 5D). In addition, the clock in p{dClk-WT};+/+ runs about 1 hr faster than the control situation, strongly suggesting that augmenting dCLK levels (Figure S7) impairs the ability of the circadian timing system to entrain to daily temperature cycles.
Temperature cycles can entrain behavioral rhythms in Drosophila exposed to constant light (LL) despite the fact that LL normally abolishes circadian rhythms [51], [53], [54]. Intriguingly, constant light exposure rescues the ability of the p{dClk-15A};Clkout flies to maintain a more stable 24-hr phase relationship with the temperature cycle (Figure 5, panels G–J), further supporting the notion that entrainment to temperature but not light is specifically impaired in these flies. Taken together, these data suggest that in the absence of light, the dCLK phosphorylation program is required for the proper entrainment of behavioral rhythms to daily temperature cycles and reveal an unanticipated role for a central clock transcription factor in modality-specific entrainment.
In p{dClk-WT};Clkout flies, hypo/intermediate-phosphorylated dCLK isoforms are present throughout the thermo phase in TC (Figure 6A, lane 2 and 3), while hyper-phosphorylated dCLK isoforms are only observed during the latter half of the cryo phase (Figure 6A, lane 5 and 6). This temporal pattern in dCLK phosphorylation is similar to that observed in LD cycles and is consistent with prior work showing that the circadian clock mechanism in Drosophila can be synchronized by daily temperature cycles [50], [51]. As expected and similar to results using LD cycles, dCLK-15A attains higher overall daily levels and does not exhibit significant phosphorylation in p{dClk-15A};Clkout flies exposed to temperature cycles (Figure 6A).
Since p{dClk-15A};Clkout flies display altered entrainment to TC cycles that becomes progressively more abnormal with prolonged duration, we tested whether the molecular clock might also exhibit a more defective status with increasing time by measuring the levels of the tim mRNA at both early (e.g., day 3) and later (e.g., day 6) days of exposure to TC. We chose tim mRNA levels as a surrogate marker for clock dynamics because it normally has a robust high amplitude rhythm (Figure 4A, B; [55]), facilitating measuring changes in molecular oscillations over the course of several days. Although the daily average levels in tim mRNA were higher in p{dClk-15A};Clkout flies on day three of TC compared to the wild-type situation (Figure 6B), consistent with findings in LD (Figure 4B), both genotypes showed similar and robust cycling patterns. However, by day six of TC, the tim mRNA oscillation pattern in p{dClk-15A};Clkout flies became significantly different from that observed for p{dClk-WT};Clkout flies (Figure 6C). Most notably, while tim mRNA cycling still manifested high-amplitude cycling in p{dClk-WT};Clkout flies on day six of TC, tim mRNA levels during the upswing phase were significantly higher in p{dClk-15A};Clkout flies, resulting in an abnormal cycling pattern. Although we did not establish a causal relationship between the observed loss in normal tim mRNA cycling and the defective behavioral entrainment in p{dClk-15A};Clkout flies during TC, the results clearly show that prolonged exposure to TC is not only associated with increasingly altered phasing of rhythms at the behavioral level (Figure 5) but also at the molecular level.
As with behavioral rhythms prior work showed that circadian molecular cycles can be synchronized to TC in the presence of constant light [51], [53]. In agreement with the observation that constant light exposure enabled p{dClk-15A};Clkout flies to more robustly synchronize to temperature cycles (Figure 5), daily rhythms in the levels of tim mRNA for both genotypes were quite similar even after six days in constant light during TC (Figure 6D and E), indicating the clock in p{dClk-15A};Clkout flies is functioning in a more wild-type manner under these conditions. Taken together, while this molecular analysis is of limited scope, it suggests that constant light exposure facilitates the ability of p{dClk-15A};Clkout flies to entrain to TC by enhancing normal clock function.
Phosphorylation of clock proteins plays diverse roles in circadian oscillatory mechanisms by regulating numerous aspects of clock protein metabolism/activity, including time-of-day dependent changes in stability, transcriptional activity and subcellular localization [10]–[12]. Although dCLK, the master transcription factor in the Drosophila circadian clock [43], [56], undergoes daily changes in phosphorylation, the physiological role of dCLK phosphorylation was not clear. As a means to address this issue, we first identified phosphorylation sites on dCLK purified from cultured Drosophila S2 cells (Table 1 and Figure 1A). To examine the in vivo significance of dCLK phosphorylation, we generated transgenic flies expressing dCLK-15A wherein 15 serine residues that were identified as sites (or possible sites) of phosphorylation were switched to alanine, and examined circadian behavior in a Clkout genetic background. Our results indicate that global phosphorylation of dCLK is an important aspect of setting clock speed by regulating the daily levels and/or activity of dCLK. This is consistent with earlier work suggesting dCLK is the rate-limiting component in the central transcriptional/translational feedback loop (TTFL) in the Drosophila clock mechanism, and that increasing the levels of dCLK lead to shorter behavioral periods [5], [42], [43]. A surprising finding is that entrainment to daily temperature cycles but not light-dark cycles are highly dependent on dCLK phosphorylation. These results suggest a novel role for phosphorylation in circadian timing systems; namely, the effective strength of an entraining cue can be modulated by adjusting the dynamics of the TTFL via controlling the levels/activity of a master circadian transcription factor (see below).
In this study, we show that dCLK undergoes multi-site phosphorylation. Among the phospho-sites identified, seven serine residues are situated immediately N-terminal to a proline, indicating a major role for the CMGC group of kinases. Indeed, studies using cultured S2 cells suggested that dCLK is a potential target of several distinct CMGC kinases [32]. More recent work also identified the pro-directed kinase NEMO as a dCLK-relevant kinase [27]. Ongoing work is aimed at identifying the kinases responsible for targeting the different phospho-sites on dCLK. It should be noted that in this study we mapped phosphorylation sites on dCLK expressed in S2 cells, which when resolved by SDS-polyacrylamide gel electrophoresis is mainly observed as two major electrophoretic mobility bands corresponding to non/hypo-phosphorylated isoforms and an ‘intermediate’ more highly phosphorylated slower migrating species [29]. Although DBT is endogenously expressed in S2 cells, the addition of exogenous DBT and/or the inhibition of protein phosphatases leads to the detection of hyper-phosphorylated isoforms of dCLK in S2 cells [29]. Thus, it is likely that we did not identify all the phospho-sites on dCLK. However, we cannot rule out the possibility that there were minor levels of hyper-phosphorylated dCLK in our preparations that were above the detection limit for phospho-site mapping by mass spectrometry. Irrespective, the phospho-sites that we identified in S2 cultured cells make a clear contribution to the daily dCLK phosphorylation program in flies and contribute to the circadian timing system.
Elimination of phosphorylation sites from dCLK (dCLK-15A) leads to significant increases in the overall daily levels of dCLK in flies, which is well correlated with previous reports in S2 cells showing that hyper-phosphorylated dCLK is sensitive to degradation [28], [29]. In general, global phosphorylation appears to reduce the stabilities of clock proteins by generating one or more phospho-degrons that are recognized by E3 ubiquitin ligases, which ultimately leads to the accelerated degradation of the phosphorylated isoforms via the proteasome pathway [13]. The E3 ligase termed CTRIP appears to directly regulate the levels of dCLK (and possibly PER), although the role of dCLK phosphorylation in this mechanism, if any, is not clear [57]. When assayed in S2 cells the stability of dCLK-16A was similar to that of dCLK-WT (e.g., Figure 1 and Figure S2). Because differences in transcript levels cannot explain the significantly higher levels of dCLK-15A in flies compared to dCLK-WT (Figure 3), it is almost certain that dCLK-15A is a more stable protein in clock cells. Thus, it appears that S2 cells do not fully recapitulate the in vivo role of phosphorylation on dCLK degradation. If we did miss mapping some sites on hyper-phosphorylated dCLK that are critical for regulating stability it is possible that these sites can still be phosphorylated on dCLK-15A expressed in S2 cells but not in flies. For example, hyper-phosphorylation of dCLK might depend on prior phosphorylation at one or more of the 15 phospho-sites we identified, and this dependency might be more strict in flies compared to the S2 cell over-expression system. Hierarchical phosphorylation has been demonstrated for other clock proteins, such as Drosophila PER and mammalian CLK [15], [37], [40]. Future work will be required to determine if there are other phospho-sites besides those we identified that regulate dCLK stability in flies.
Besides regulating the stability of core clock transcription factors, phosphorylation modulates trans-activation potential [36], [37], [58]–[61]. dCLK-15A expressed in S2 cells exhibited normal binding to CYC (and PER) but exhibits more potent transcriptional activity, at least in the context of a simple E-box driven expression (Figure 1D). Consistent with this, the levels of dper and tim mRNAs in p{dClk-15A};Clkout flies are higher compared to the control situation (Figure 4A, B). Of course, phosphorylation also affects the levels of dCLK-15A in flies, so at this stage it is not possible to determine how much the increased per/tim transcript levels are due to changes in the levels or activity of dCLK-15A. Nonetheless, our results strongly suggest that in wild-type flies the levels and/or activity of dCLK act in a rate-limiting fashion during the daily accumulation phase of per/tim transcripts and possibly other targets. In addition, the phospho-sites that we identified do not seem to be play a major determinant in feedback repression by PER and associated factors. Strong repression was observed in S2 cells for the dCLK-15A version (Figure 1E) and the normal daily downswing in per/tim levels occurred in p{dClk-15A};Clkout flies (Figure 4A and B). However, it is possible that we missed some phospho-sites that more specifically regulate the transcriptional activity of dCLK.
At the behavioral level, p{dClk-15A};Clkout flies exhibit short period rhythms, consistent with prior work showing that increasing the dosage of dClk quickens the pace of the clock [42], [43]. In light-dark cycles, p{dClk-15A};Clkout flies maintain a stable phase relationship with the entraining environment, displaying the typical anticipatory bimodal activity pattern (Figure 2). Moreover, in a daily light-dark cycle the timing of the morning and especially evening peak of activity is shifted in flies with different endogenous periods, appearing earlier in fast clocks and later in slow clocks [52]. Indeed, the p{dClk-15A};Clkout flies follows this trend as the evening (and morning) bout of activity in LD is earlier compared to control flies (Figure 2). Together, these results indicate that although global phosphorylation of dCLK is an important determinant in setting clock speed, it plays little to no role in photic entrainment.
Surprisingly, the elimination of phosphorylation sites on dCLK strongly influences circadian behavior in daily temperature cycles (Figure 5). Temperature cycles with amplitudes of only 2° to 3°C robustly synchronizes circadian rhythms in Drosophila and other organisms [51], [52], [62]–[66]. When exposed to temperature cycles of 24°C/29°C, control p{dClk-WT};Clkout flies manifested the typical bimodal activity pattern with bouts of activity anticipating the two temperature transition points, similar to that occurring during entrainment to LD cycles (Figure 5A and S5A). However, even during the first days in TC, p{dClk-15A};Clkout flies already exhibit a very abnormal phase alignment with ‘morning’ and ‘evening’ bouts of activity that occur much earlier, around the middle of the cryo- and thermal-phases, respectively (Figure 5B and S5B, C). The advanced timing of the morning and evening bouts of activity is much earlier than would be expected based solely on the 1.5 hr shorter circadian period in p{dClk-15A};Clkout flies (Table 3). That entrainment to TC is highly defective in p{dClk-15A};Clkout flies is even more dramatically underscored by the progressive advances in the evening component of activity on subsequent days (Figure 5). Although not as apparent, flies with increased dosage of dClk (p{dClk-WT};+/+ flies) also showed progressively earlier evening activity bouts in thermal cycles (Figure 5C and F) but not LD cycles, further suggesting that increased levels/activity of dCLK are causally linked to the inability of maintaining a stable phase relationship with TC. Because the timing of the evening activity in both p{dClk-15A};Clkout and p{dClk-WT};+/+ flies occurs progressively earlier during TC, our results strongly suggest that these flies are only weakly synchronized to TC and are partially free-running at their faster endogenous periods.
In trying to determine why p{dClk-15A};Clkout flies might exhibit a defect in temperature entrainment but not photic entrainment, it is important to note that several lines of evidence support the notion that light is a more potent synchronizer of the clock in D. melanogaster compared to temperature entrainment, including the use of out-of-phase light/dark and temperature cycles [4]. In addition, lowering the levels/function of the key photic entrainment photoreceptor CRYPTOCHROME (CRY) increases the ability to synchronize to TC [67], suggesting the dominance of light input under normal conditions. Also, it takes many more days to shift the phase of the clock via TC compared to LD cycles [50]. The overall strength of light in D. melanogaster entrainment is not surprising given the ability of light pulses to evoke the rapid degradation of TIM and the great sensitivity of Drosophila CRY/TIM to light [68].
Indeed, constant light rescues the ability of TC to stably entrain behavioral rhythms in p{dClk-15A};Clkout (Figure 5, G–J), presumably by maintaining the clock in a more normal state (Figure 6). Intriguingly, prior work showed a similar pattern for the classic perS and perL mutants that display short (19 hr) and long (29 hr) endogenous rhythms, respectively [66]. That is, while wild-type flies entrain to TC in DD or LL, but perS and perL flies only entrain to TC in LL [66]. This suggests that alterations in the PER protein rhythm might preferentially disrupt thermal entrainment. In the case of p{dClk-15A};Clkout flies the amplitude of the PER abundance cycle is increased reaching higher peak values (Figure 4). Clocks with higher amplitudes are more resistant to entrainment by weak zeitgebers [69]–[71]. Relevant to this discussion, reducing CLOCK activity in mice decreased the amplitude of the circadian pacemaker and per gene expression, enhancing the ability to evoke phase shifts in behavioral rhythms [72], [73]. Thus, a simple model for our results is that the increased per mRNA and protein rhythms in p{dClk-15A};Clkout flies leads to an increase in pacemaker amplitude minimizing their ability to synchronize to weaker entraining signals such as TC. However, it should be noted that higher amplitude rhythms of cycling mRNAs are highly suggestive but not definitive proof of an increase in oscillator. A standard approach to infer the relative amplitude of a clock is to increase the strength of the entraining signal, which should enhance its entrainment potential [69], [70], [74].
Although a change in the amplitude of the clock in p{dClk-15A};Clkout flies offers a plausible explanation for the preferential defect in temperature entrainment, there are other possibilities. For example, CRY-positive clock cells are more important for entraining to LD cycles, whereas CRY-negative clock cells are more important for TC entrainment [4]. Thus, dCLK-15A could have preferential effects in CRY-negative cells to lessen their contribution, impairing TC entrainment. Another more speculative idea is that the phosphorylation of dCLK can act as a thermal sensor, although this would be specific to temperature entrainment as temperature compensation appears normal in the p{dClk-15A};Clkout flies (Table 2). Clearly, future studies will be required to better address the mechanism underlying the impaired synchronization of p{dClk-15A};Clkout flies to temperature cycles. However, our findings reveal that phosphorylation of a key rate-limiting circadian transcription factor is critical for entrainment to daily temperature cycles. Indeed, the CLOCK protein in zebrafish [65] was shown to be regulated by temperature, suggesting a universal role for CLOCK in the adaptation of animal circadian clocks to thermal cues.
The pMT-dClk-V5, pMT-HA-dClk-V5, pMT-HA-dClk, pAct-per, pAct-per-V5 and pMT-dbt-V5 plasmids were described previously [20], [29], [31]. pMT-dClk15A-V5 and pMT-dClk16A-V5 were generated by serially changing codons for Ser to those of Ala by using a Quick Change site-directed mutagenesis kit (Stratagene). All final constructs were verified by DNA sequencing.
Hygromycin-resistant stable Schneider 2 (S2) cell lines expressing pMT-HA-dClk-V5 were established for dCLK purification. dClk expression was induced by adding 500 µM CuSO4 to the medium and cells were harvested 24 hr post-induction. 200 ml of culture (3×106 cells/ml) was used and harvested cells were lysed using modified-RIPA buffer (50 mM Tris-HCl [pH 7.5], 150 mM NaCl, 1% NP-40, 0.25% Sodium deoxycholate) with the addition of a protease inhibitor cocktail (Roche) containing 1 mM EDTA, 25 mM NaF, and 1 mM Na2VO3. To extracts, anti-V5 antibody (Invitrogen) was added and incubated overnight with gentle rotation at 4°C followed by the addition of Dynabeads Protein A (Invitrogen) with a further overnight incubation. Beads were collected using DynalMPC. dCLK was eluted with modified Laemmli buffer (150 mM Tris-HCl [pH 6.8], 6 mM EDTA, 3% SDS, 30% Glycerol) supplemented with 50 mM reducing agent TCEP (Calbiochem) at 65°C for 20 min. Alkylation was performed by adding 0.5M IAA (iodoacetamide) for 20 min at room temperature in the dark. The eluate was resolved using 8% SDS-PAGE, and all the detectable dCLK bands of differing electrophoretic mobility excised (which under the conditions used was mainly the ‘intermediate’ phosphorylated band), subjected to protease digestion and analyzed by mass spectrometry. Mass spectrometry was performed as described in Schlosser et al. 2005. Data analysis was performed as described previously [41].
S2 cells were obtained from Invitrogen and transfected using effectene following the manufacturer's protocol (Qiagen). Luciferase (luc) reporter assay was performed as described previously [29], [75]. Briefly, S2 cells were placed in 24-well plates and co-transfected with 0–100 ng pMT-dClk-V5 and pMT-dClk-16A-V5 along with 10 ng of perEluc, 30 ng of pAct-β-gal-V5/His as indicated. dPER mediated repression of dCLK dependent transactivation was measured by transfecting 0–20 ng of pAct-dper together with 2 ng of pMT-dClk-V5 or pMT-dClk-16A-V5. One day after transfection, dClk expression was induced with 500 µM CuSO4 (final in the media), and after another day cells were washed in phosphate buffered saline (PBS), followed by lysis in 300 µl of Reporter Lysis Buffer (Promega). Aliquots of cell extracts were assayed for β-galactosidase and luciferase activities using the Luciferase Assay System and protocols supplied by the manufacturer (Promega).
Clkout flies were generated in one of our laboratories (P.E.H.) as follows: 5.2 kb deletion of dClk exon 1 and upstream sequences was generated by FLP-mediated recombination between FRT sites in the pBac Clk[f06808] and pBac Clk [f03095] [76], [77]. Flippase (FLP)-induced recombination was induced by a daily 1 h heat-shock at 37°C given to hsFLP;;f06808/f03095 larvae and pupae. Three recombinants were recovered, and each produced a deletion rather than a duplication of intervening dClk sequences. The remaining pBac insert was excised via pBac transposase induced transposition resulting in white-eyed flies harboring the deletion [78]. A DNA fragment containing the deleted sequences was amplified using primers situated upstream of the f03095 insertion site (5′ CGGAATATTGGACAACAAACAG 3′) and downstream of the f06808 insertion site (5′CAGCAGTGGAATCTTAATACAG 3′), and sequenced to confirm the endpoints of the deletion. This new dClk deletion allele was named Clkout.
To generate transgenic flies that produce wild-type dCLK tagged with V5 at the C-terminus, dClk-containing P[acman] transgene was generated using recombineering-mediated gap repair [79]. To prepare the P[acman] vector, homology arms were amplified from genomic DNA with primers clkLA-f (ATGTGGCGCGCCGCCCCAAAAATCCATAAATGCT) and clkLA-r (GTGTTGGATCCAGGGGTGTTATAGAGAGGGACA) for the left arm and clkRA-f (GTGTGGATCCGCAGAGTGAAACCTGTGCAA) and clkRA-r (ATATATGTGCGGCCGCTCCCGGTTATGAGTTTTTCG) for the right arm via PCR, and cloned as AscI-BamHI and BamHI-NotI fragments into AscI and NotI digested attB-P[acman]-ApR vector (modified to remove the SphI site) to form attB-P[acman]ClkLARA. Recombination-competent SW102 cells harboring BAC clone RP98 5K6 (BACPAC Resource Center, Oakland, Ca, USA), which contains the dClk genomic region, were transformed with the attB-P[acman]ClkLARA vector (linearized with BamHI). Recombinants containing 15.5 kb of genomic sequence beginning ∼8 kb upstream of the dClk translation start and ending ∼2.5 kb downstream of the dClk stop codon were verified by PCR and sequencing and termed attB-P[acman]-Clk. To introduce a V5 epitope tag at the C-terminus of the dClk open reading frame (ORF), a 3′ genomic fragment of dClk (from 351 bp upstream to 1580 bp downstream of the translation stop) was cloned into pGEM-T vector (Promega, Madison, WI). Sequences encoding V5 were introduced in-frame immediately upstream of the dClk stop codon using the Quickchange site directed mutagenesis kit (Stratagene, La Jolla, CA) to create pGEM-T-dClk3′V5. The 3′ dClk genomic fragment in attB-P[acman]- dClk was swapped with the 3′ fragment in pGEM-T-dClk3′V5 using SphI and NotI to form attB-P[acman]- dClkV5. This transgene was inserted into the VK00018 attP site on chromosome 2 via PhiC31-mediated transgenesis [79], [80].
Transformation vector containing a genomic dClk wherein the codons for the 15 identified phospho-serine were switched to those for alanine was generated in multiple stages as follows: A genomic dClk sub-fragment from NheI to SphI site was isolated from P[acman]-dClk-V5 and subcloned into pSP72 vector where the multi-cloning sites were mutagenized to introduce a NheI site, and named this plasmid as pSP72-dClk(NheI/SphI). Next, we obtained a dClk sub-fragment spanning from the NcoI to SphI sites by restriction digestion of pSP72-dClk(NheI/SphI), subcloned the released fragment into pSP72 where the multi-cloning sites were mutagenized to introduce a NcoI site, and named this plasmid as pSP72-dClk(NcoI/SphI). We performed serial site directed mutagenesis with pSP72-dClk(NcoI/SphI) and finally made pSP72-dClk(NCoI/SphI)-S11A wherein codons for the serine residues at amino acids 209, 210, 211, 444, 450, 487, 504, 611, 645, 859, 902 on dCLK were all switched to those for alanine residues [(GenBank accession number NP_001014576)]. We purified the dClk(NCoI/SphI)-S11A insert by restriction enzyme digestion of pSP72-dClk(NCoI/SphI)-S11A and replaced the wild-type dClk(NcoI/SphI) insert, generating pSP72-dClk(NheI/SphI)-S11A. Next, a more 3′ genomic dClk sub-fragment from the SphI to NotI sites was subcloned into pSP72 vector where the multi-cloning sites were mutagenized to include NotI and NheI sites, and named this plasmid as pSP72-dClk(SphI/NotI). We performed serial site directed mutagenesis with pSP72-dClk(SphI/NotI) and made pSP72-dClk(SphI/NotI)-S4A wherein codons for the serine residues at amino acids 924, 934, 938, 1018 were switched to those for alanine. Finally, the genomic dClk(SphI/NotI)-S4A fragment was ligated with pSP72-dClk(NheI/SphI)-S11A generating pSP72-dClk(NheI/NotI)-S15A, and then dClk(NheI/NotI)-S15A fragment was switched with wild-type dClk(NheI/NotI) fragment in pacman-dClk-V5 plasmid yielding P[acman]-dClk-15A-V5. Transgenic flies were generated by BestGene Inc. (CA, USA). P[acman]-dClk-15A-V5 transformation vector was injected into flies carrying the VK00018 attP docking site on the second chromosome for site-specific integration [79]. Two independent germ-line transformants bearing the dClk-15A-V5 transgene in a wild-type background were obtained and then crossed into a Clkout genetic background to yield dClk-15A-V5;Clkout.
The locomotor activities of individual flies were measured as previously described using the Drosophila Activity Monitoring system from Trikinetics (Waltham, MA). Young adult flies were used for the analysis and exposed to 4 days of 12 h light followed by 12 h dark [where zeitgeber time 0 (ZT0) is defined as the time when the light phase begins] at 25°C and subsequently kept in constant dark conditions (DD) for 7 days. Temperature entrainment (temperature cycle, TC) was performed in constant dark condition and in some cases, in the presence of constant light (>2000lux). Temperature cycles were 12 h of 24°C (cryo phase) followed by 12 h of 29°C (thermal phase) (where ZT0 is defined as the time when the cryo phase begins) for 4 days and subsequently kept at 24°C for 7 days. The locomotor activity data for each individual fly was analyzed using the FaasX software (Fly Activity Analysis Suite for MacOSX), which was generously provided by F. Rouyer (CNRS, France). Periods were calculated for each individual fly using chi-square periodogram analysis and pooled to obtain a group average for each independent transgenic line or genotype. Power is a quantification of the relative strength of the rhythm during DD. Individual flies with a power ≥10 and a ‘width’ value of 2 or more (denotes number of peaks in 30-min increments above the periodogram 95% confidence line) were considered rhythmic. Actogram represents the locomotor activity data throughout the experimental period. Vertical bars in the actogram represent absolute activity levels for each 30 min intervals averaged for each given genotypes of flies. The strength of this measurement can be manipulated by using the function called hash density, which represent the number of times fly need to make beam crossing to be registered as one vertical bar. The hash density of the actogram was varied for better comparison depending on the activity levels of given genotypes of flies.
Protein extracts from S2 cells were prepared as previously described [31]. Briefly, the cells were lysed using modified-RIPA buffer (50 mM Tris-HCl [pH 7.5], 150 mM NaCl, 1% NP-40, 0.25% Sodium deoxycholate) with the addition of protease inhibitor cocktail (GeneDEPOT) and phosphatase inhibitor cocktail (GeneDEPOT). For detection of dCLK recombinant protein, extracts were obtained using RIPA buffer 25 mM Tris-HCl [pH 7.5], 50 mM NaCl, 0.5% Sodium deoxycholate, 0.5% NP40, 0.1% SDS) and were sonicated briefly as previously described [29]. Flies were collected by freezing at the indicated times in light-dark (LD) or temperature cycles (TC) and total fly head extracts prepared using modified-RIPA buffer or RIPA buffer with sonication (for dCLK). Extracts were resolved by 5% polyacrylamide gels or by 3–8% Tris-acetate Criterion gel (Bio-Rad) in some case for dCLK, transferred to PVDF membrane (Immobilon-P, Millipore), and immunoblots were treated with chemiluminescence (ECL, Thermo). Primary antibodies were used at the following dilutions; anti-V5 (Invitrogen), 1∶5000; anti-HA (12CA5, Roche), 1∶2000; anti-OGT (H-300, Santa Cruz), 1∶3000; anti-PER, (Rb1) 1∶3000; anti-TIM (TR3), 1∶3000; anti-dCLK (GP208) 1∶3000. Quantification of band intensity was performed using image J software.
For immunoprecipitation, cell extracts from S2 cells were prepared and 3 µl of anti-HA (12CA5) or anti-V5 antibody was added depending on the target protein sought, and incubated for overnight at 4°C with gentle rotation. The next day, 20 µl of Gammabind-sephase bead (GE healthcare) was added with a further incubation of 3 hr at 4°C. The immune complexes were eluted with 1X SDS-PAGE sample buffer. For λ- phosphatase treatment, the purified immune complexes were resuspended in λ protein phosphatase buffer (50 mM Tris-HCl [pH 7.5], 0.1 mM EDTA, 5 mM DTT, 0.01% Triton X-100, 2 mM MnCl2, and 0.1 mg/ml bovine serum albumin), divided into two equal aliquots. One aliquot of bead was treated with 200 units of λ protein phosphatase (NEB) and no addition was made to the other aliquot. Both aliquots were incubated for 30 min at 30°C with occasional shaking, and immune complexes analyzed by immunoblotting.
Total RNA was isolated from frozen heads using QIAzol lysis reagent (QIAGEN). 1 µg of total RNA was reverse transcribed with oligo-dT primer using Prime Script reverse transcriptase (TAKARA) and real-time PCR was performed in Corbett Rotor Gene 6000 (Corbett Life Science) using Quantitect SYBR Green PCR kit (Qiagen). Primer sequences used here are as follows; dper forward: 5′-GACCGAATCCCTGCTCAATA-3′; dper reverse: 5′-GTGTCATTGGCGGACTTCTT-3′; tim forward: 5′-CCCTTATACCCGAGGTGGAT-3′; tim reverse: 5′-TGATCGAGTTGCAGTGCTTC-3′; dClk forward: 5′-CAGCCGCAATTCAATCAGTA-3′; dClk reverse: 5′-GCAACTGTGAGTGGCTCTGA-3′. We also included primers for the noncycling mRNA coding for CBP20 as previously described, and sequences are as follows; cbp20 forward: 5′-GTCTGATTCGTGTGGACTGG-3′; cbp20 reverse: 5′-CAACAGTTTGCCATAACCCC-3′. Results were analyzed with software associated with Rotor Gene 6000, and relative mRNA levels were quantitated using the 2−ΔΔCt method.
|
10.1371/journal.pntd.0004253 | Mycobacterium ulcerans in the Elderly: More Severe Disease and Suboptimal Outcomes | The clinical presentation of M. ulcerans disease and the safety and effectiveness of treatment may differ in elderly compared with younger populations related to relative immune defficiencies, co-morbidities and drug interactions. However, elderly populations with M. ulcerans disease have not been comprehensively studied.
A retrospective analysis was performed on an observational cohort of all confirmed M. ulcerans cases managed at Barwon Health from 1/1/1998-31/12/2014. The cohort included 327 patients; 131(40.0%) ≥65 years and 196(60.0%) <65 years of age. Patients ≥65 years had a shorter median duration of symptoms prior to diagnosis (p<0.01), a higher proportion with diabetes (p<0.001) and immune suppression (p<0.001), and were more likely to have lesions that were multiple (OR 4.67, 95% CI 1.78–12.31, p<0.001) and WHO category 3 (OR 4.59, 95% CI 1.98–10.59, p<0.001). Antibiotic complications occurred in 69(24.3%) treatment episodes at an increased incidence in those aged ≥65 years (OR 5.29, 95% CI 2.81–9.98, p<0.001). There were 4(1.2%) deaths, with significantly more in the age-group ≥65 years (4 compared with 0 deaths, p = 0.01). The overall treatment success rate was 92.2%. For the age-group ≥65 years there was a reduced rate of treatment success overall (OR 0.34, 95% CI 0.14–0.80, p = <0.01) and when surgery was used alone (OR 0.21, 95% CI 0.06–0.76, p<0.01). Patients ≥65 years were more likely to have a paradoxical reaction (OR 2.06, 95% CI 1.17–3.62, p = 0.01).
Elderly patients comprise a significant proportion of M. ulcerans disease patients in Australian populations and present with more severe and advanced disease forms. Currently recommended treatments are associated with increased toxicity and reduced effectiveness in elderly populations. Increased efforts are required to diagnose M. ulcerans earlier in elderly populations, and research is urgently required to develop more effective and less toxic treatments for this age-group.
| Mycobacterium ulcerans is an infection that can affect all age-groups. It causes necrosis of skin and soft-tissue often resulting in severe outcomes and long-term disability. However, due to the majority of infections worldwide occurring in children and young adults, there is a paucity of information available in elderly patients. It is important that elderly patients are not neglected as the clinical presentation and treatment outcomes may differ significantly from younger patients related to relative immune defficiencies, co-morbidities and increased potential for drug interactions. We specifically examined patients with M. ulcerans disease aged ≥ 65 years and showed that they comprise a significant proportion of patients affected in Australian populations. They present with more severe and advanced disease forms, and suffer from increased toxicity and reduced effectiveness of the currently recommended treatments. Therefore, our study demonstrates that increased efforts are required to diagnose M. ulcerans disease earlier in elderly populations, and that research is urgently required to develop more effective and less toxic treatments for this age-group.
| Mycobacterium ulcerans (M. ulcerans) is an infection that causes necrotizing lesions of skin and subcutaneous tissue. The majority of cases are reported from west and central Africa, but unlike Africa where the disease occurs mainly in children[1,2], in south-eastern Victoria, Australia it occurs mainly in adults with a large proportion aged > 50 years.[3] Reported rates of disease in Australian populations are up to 7 times higher in those ≥55 years of age[4]. Current M. ulcerans treatment guidelines recommend combined antibiotics for 8 weeks with surgery as an adjunctive treatment[5,6].
The clinical presentation of M. ulcerans disease (Buruli ulcer), as well as the safety and effectiveness of treatment, may differ in elderly compared with younger populations. It is known that immune function reduces with senescence, and as the immune system plays a vital role in the control of M. ulcerans[7,8], this may lead to an increase in the incidence and severity of disease as well as reduced effectiveness of treatment. There may also be altered health-seeking behaviours in older people who may find accessing healthcare more difficult or neglect skin lesions, or for whom there is a potentially increased prevalence of alternative causes of ulceration (eg venous disease) resulting in misdiagnosis. These aforementioned issues could lead to delays in diagnosis with increased disease severity. Furthermore, increased rates of co-morbidities in elderly patients may adversely affect immune function, but may also lead to increased drug interactions and the potential for increased toxicity associated with antibiotic treatment[9].
In our practice we have observed significant numbers of elderly patients developing M. ulcerans disease. Our earlier published experience has suggested that populations older than 60 years of age may have had increased prevalence of multiple M. ulcerans lesions at presentation[3], reduced rates of treatment success with surgical treatment[10], and increased rates of antibiotic related paradoxical reactions[11]. However, populations aged ≥65 years with M. ulcerans have not been comprehensively studied. We therefore undertook to describe in an Australian cohort the proportion of patients aged ≥65 years affected by M. ulcerans and compare them with younger patients with respect to their clinical presentation, and the safety and effectiveness of treatment.
A retrospective analysis was performed on data from a prospectively collected cohort of all confirmed M. ulcerans cases managed at Barwon Health from 1/1/1998-31/12/2014. A M. ulcerans case was defined as the presence of a lesion clinically suggestive of M. ulcerans plus any of (1) a culture of M. ulcerans from the lesion, (2) a positive PCR from a swab or biopsy of the lesion, or (3) histopathology of an excised lesion showing a necrotic granulomatous ulcer with the presence of acid-fast bacilli (AFB) consistent with acute M. ulcerans infection. Lesion size was determined by measuring the extent of lesion induration with a ruler and a WHO category was assigned according to published definitions.[5] Elderly age was defined as ≥65 years in line with the accepted definition for most developed countries[12].
Drug dosages for adults included rifampicin 10 mg/kg/day (up to a maximum of 600 mg daily), ciprofloxacin 500 mg twice daily, moxifloxacin 400 mg once daily, clarithromycin 7.5 mg/kg twice daily (up to 500 mg twice daily) and ethambutol 15 mg/kg/day. A complication of medical therapy was defined as an adverse event attributed to an antibiotic that required its cessation. In cases where it was not possible to determine which antibiotic of a combination was responsible for the complication, both antibiotics were attributed with a complication. Immune suppression was defined as current treatment with immunosuppressive medication (eg. prednisolone) or active malignancy.
Treatment failure was defined as patients developing disease recurrence within 12 months of initiating treatment. Recurrence was defined as a new lesion appearing in the wound, locally, or another part of the body that met the case definition for M. ulcerans disease within 12 months of initiating treatment. Paradoxical reactions were defined by the presence of one or both of the following features: a) clinical: an initial improvement on antibiotic treatment in the clinical appearance of a M. ulcerans lesion followed by deterioration of the lesion or its surrounding tissues, or the appearance of a new lesion(s), and b) histopathology examination of excised tissue from the clinical lesion showing evidence of an intense inflammatory reaction consistent with a paradoxical reaction[11].
Data was collected prospectively using Epi-info 6 (CDC, Atlanta) and analysed retrospectively using STATA 12 (StataCorp, Texas, USA). Outcome data were censored at the time of death, disease recurrence or after 12 months of follow-up from initiation of antibiotics. Categorical variables were compared using 2x2 tables and the Chi-squared test. Medians of non-parametric variables were compared using the Wilcoxon rank sum test. Odds ratios were calculated using the Mantel-Haenszel test.
This study was approved by the Barwon Health Human Research and Ethics Committee. All previously gathered human medical data were analysed in a de-identified fashion.
There were 327 patients treated for M. ulcerans at Barwon Health between 1/1/1998-31/12/2014 and all were included in the study. The median patient age was 58 years (IQR 35–74 years); 131 (40.0%) were ≥65 years and 196 (60.0%) were <65 years. 165 (50.5%) were male and 162 (49.5%) female.
Three hundred and eight (94.2%) patients had 1 M. ulcerans lesion, 10 (3.1%) had 2 lesions, 6 (1.8%) had 3 lesions, 1(0.3%) had 10 lesions and 2 (0.6%) had 13 lesions. 84.9% of lesions were ulcerative and 79.9% were classified as WHO category 1. The median duration of symptoms prior to diagnosis was 42 days (IQR 28–70 days). (Table 1)
There were some significant differences in baseline characteristics between the age-groups. (Table 1) Patients in the age-group ≥65 years were less likely to be male (OR 0.60, 95% CI 0.38–0.94, p = 0.02), the median duration of symptoms prior to diagnosis was significantly shorter (35 compared to 42 days, p<0.01), and there was a higher proportion of patients with diabetes (p<0.001) and immune suppression (p<0.001).
Patients in the age-group ≥65 years were more likely to have lesions that were multiple (OR 3.49, 95% CI 1.26–9.54, p<0.01) and classified as WHO category 3 compared with category 1 and 2 combined (OR 4.89, 95% CI 1.95–12.25, p<0.001). They also had a higher proportion of oedematous compared to non-oedematous lesions (11.5% compared with 6.2%, p = 0.09). (Table 1)
Two hundred and eighty (85.6%) patients received antibiotic treatment for a median of 56 days (IQR 49–83 days). 115 (87.8%) of those ≥ 65 years received antibiotics and 165 (84.2%) of those < 65 years received antibiotics (p = 0.97).
There were 284 antibiotic treatment episodes in 280 patients (4 patients received a second antibiotic course—three due to disease recurrence, and 1 for a late paradoxical reaction). Initial antibiotic combinations used were rifampicin/ciprofloxacin in 162 (57.0%), rifampicin/clarithromycin in 95 (33.4%), rifampicin/moxifloxacin in 8 (2.8%), rifampicin/clarithromycin/ethambutol in 6 (2.1%), clarithromycin/ciprofloxacin in 3 (1.1%) and other varied combinations in 10 (3.5%) treatment episodes.
Overall 69 (24.3%) antibiotic treatment episodes were associated with a complication severe enough to require cessation of at least one antibiotic. There was an increased incidence of antibiotic complications in those aged ≥ 65 years compared with those aged <65 years (OR 5.29, 95% CI 2.81–9.98, p<0.001).
Including antibiotics commenced as second-line treatment following cessation of one or more of the initial antibiotics due to complications, 276 (97.5%) treatment episodes included rifampicin, 174 (61.5%) ciprofloxacin, 127 (44.9%) clarithromycin, 13 (4.6%) ethambutol, 10 (3.5%) moxifloxacin and 9 (3.2%) amikacin. Rifampicin was associated with complications in 47 (17.0%) treatment episodes in which it was used, and this was more common in those aged ≥ 65 years compared to < 65 years (OR 4.87, 95% CI 2.36–10.07, p<0.001). (Table 2, Fig 1) Ciprofloxacin was associated with complications in 32 (18.4%) treatment episodes in which it was used, and this was more common in populations ≥ 65 years (OR 2.92, 95% CI 1.26–6.75, p<0.01). Clarithromycin was associated with complications in 24 (18.9%) treatment episodes in which it was used, and this was increased in populations ≥ 65 years (OR 3.38, 95% CI 1.30–8.78, p<0.01). (Table 2, Fig 1)
The specific complications associated with each antibiotic are listed in Table 3. In 11 patients hospitalization was required to manage the antibiotic complication; 9/49 (18%) in those ≥65 and 2/20 (10%) in those <65 years (OR 2.03, 95% CI 0.39–10.55, p = 0.39).
210 (64.2%) patients had surgery; More patients in the ≥ 65 years age-group had surgery compared to the < 65 years age-group [92 (70.2%) compared to 118 (60.2%), (p = 0.06)]. 62 (29.5%) had surgery alone and 148 (70.5%) had surgery plus antibiotics. There was no difference in the proportions who had surgery alone between the age-groups (p = 0.96).
Treatment outcomes for first M. ulcerans lesions could be determined for 323 (98.8%) patients; 2 were lost to follow-up, 1 was transferred out, and 1 had an unclear outcome. At the time of submission, 300 (92.9%) had their outcomes determined after 12 months of follow-up and 23 (7.1%) patients after 9 months of follow-up.
There were 4 (1.2%) deaths (Table 4). The median age of those who died was 91.5 years (IQR 71.5–94.5 years), with significantly more deaths in the age-group ≥ 65 years compared to the age-group < 65 years [4 (3.1%) compared to 0 (0.0%), p = 0.01]. Only one of the deaths (#2) was felt to be directly attributable to M. ulcerans infection as a result of skin sepsis and secondary decompensated cardiac failure.
For the remaining 319 patients, the overall treatment success rate was 92.2% (Table 5); there was a reduced rate of treatment success in the age-group ≥ 65 years compared to the age-group <65 years (OR 0.34, 95% CI 0.14–0.80, p<0.01). There was a significantly reduced treatment success rate for the age-group ≥ 65 years when surgery was used alone (OR 0.21, 95% CI 0.06–0.76, p<0.01). Rates of treatment success for surgery plus antibiotics or antibiotics alone were similar between the age-groups. (Table 5)
67/275 (24.4%) patients experienced antibiotic-associated paradoxical reactions; 36 (32.4%) patients ≥ 65 years and 31 (18.9%) patients < 65 years. Patients ≥ 65 years were significantly more likely to have a paradoxical reaction compared with those < 65 years (OR 2.06, 95% CI 1.17–3.62, p = 0.01). This was independent of the WHO category of the lesion (42% v 22% for category 1, 56% v 44% for category 2 and 60% v 50% for category 3 when comparing age ≥65 to <65 years).
In describing a large cohort of patients aged ≥65 years, we have studied for the first time this unique population with M. ulcerans disease. Cohorts described from Africa, where most M. ulcerans cases are reported, mainly involve children with few numbers of patients aged ≥65 years[1,2]. Additionally, studies reported from Australia have focused on cohorts across all age-groups[3,13,14]. This study therefore provides important new information pertaining to the epidemiology, clinical characteristics, treatment and outcomes in elderly populations.
Patients aged ≥65 years represent an important subgroup in our cohort with M. ulcerans disease comprising two out of every 5 patients. Previous reports suggest that they may have an increased incidence of disease[4]. Additionally, our study suggests that they have more advanced and severe disease at presentation with an increased rate of multiple, large and oedematous lesions. Early non-ulcerative lesions (plaques or nodules) were infrequently reported. This is not due to late presentation as in our study the time from reported symptom onset to presentation for care was reduced in elderly patients. Instead this may be related to reduced immunity in older populations that inhibits the control of M. ulcerans leading to larger and oedematous forms of disease and the dissemination of lesions to other sites. This would be similar to the effect of HIV induced immune suppression which is associated with more severe M. ulcerans disease with an increase in the size, number and proportion of advanced lesions[15]. The reduced immunity in elderly populations may relate to the increasing immune suppression associated with senescence, and the increased presence of immunosuppressive conditions such as diabetes and malignancy or the increased likelihood of receiving immunosuppressive medication.
Our study demonstrates that treatment of M. ulcerans disease in elderly populations is associated with increased toxicity and reduced effectiveness. Nearly one-half (42%) of patients aged ≥ 65 years had to cease an antibiotic due to complications at a rate 5 times higher than younger populations, and complications were more severe with nearly one-fifth (18%) requiring hospitalization. However this cannot be avoided by treating without antibiotics as treatment with surgery alone resulted in a 79% increased failure rate in this age-group. Furthermore, there is a two-fold increase in antibiotic-associated paradoxical reactions which can cause significant morbidity and complicate treatment[11,16]. Therefore there is an urgent need to develop less toxic and more effective treatments for elderly populations.
All of the most commonly used oral antibiotics active against M. ulcerans have significant drug interactions. Rifampicin induces, and ciprofloxacin and clarithromycin inhibit, the cytochrome P450 enzyme system[17,18] leading to interactions with many commonly used medications. Clarithromycin and fluoroquinolones can prolong the QT interval creating a potential for serious arrhythmias if combined with other medical conditions or medication who do the same. This makes treatment more difficult and increases the risk of toxicity in elderly populations who are frequently prescribed multiple other medications. In addition, the pharmacokinetics of the antibiotics may differ with increasing age potentially resulting in toxic levels with currently recommended doses. For example it has been shown that elderly patients have higher serum concentrations and a longer half-life for antibiotics due to either increased bioavailability (ciprofloxacin) and reduced renal function with age (ciprofloxacin and clarithromycin) [19,20].
We advocate that pharmacokinetic and pharmacodynamic studies of frequently used antibiotics be performed in elderly patients to explore the safety and effectiveness of current and lower doses of antibiotics, including intermittent dosing regimens (e.g. thrice weekly). Further research should also be performed on the safety and effectiveness of shorter duration antibiotic regimens[21]. It would be worthwhile exploring the use of alternative antibiotics such as the new anti-tuberculous agent bedaqueline, which shows strong bactericidal activity against M. ulcerans in mouse models[22], and avermectins which have shown promising in vivo activity against M. ulcerans[23], as these may be equally effective but potentially less toxic in elderly populations.
Elderly patients were also found to have increased rates of antibiotic-associated paradoxical reactions, independent of lesion size. These likely occur due to the reversal of mycolactone induced immune suppression and the increased antigenic stimulus provided by dying mycobacteria when antibiotics are administered[24,25]. The increased rate in elderly patients in theory could relate to an increased organism load secondary to their relatively weakened immune systems which provides a greater antigenic stimulus combined with a greater potential for rapid immune function improvements when the inhibitory effects of mycolactone toxin are removed with antibiotics[11]. Increased paradoxical reactions contribute to the increased toxicity associated with antibiotics in elderly patients, and research is required to try and understand the reasons for their increased incidence in this age-group and to try and minimise their impact. Our early experience is that pre-emptive corticosteroids commenced at the initiation of antibiotics may prevent paradoxical reactions in elderly patients with oedematous lesions[26] and this should be further studied.
Finally it should be noted that M. ulcerans is not without mortality in elderly patients where sepsis secondary to skin ulceration, or complications of treatment, can contribute to death in patients with significant co-morbidities or frailty due to age.
We acknowledge the limitation that this is an observational study and as treatments were not randomized between groups there may be unmeasured confounders that may have influenced the results. However the cohort is large, data is collected prospectively and rates of follow-up are very high supporting the validity of our findings.
In conclusion, elderly patients comprise a significant proportion of M. ulcerans disease patients in Australian populations and present with more severe and advanced forms of disease. Currently recommended M. ulcerans treatments are associated with increased toxicity and reduced effectiveness in elderly populations. Increased efforts are required to diagnose M. ulcerans earlier in elderly populations, and research is urgently required to develop more effective and less toxic treatments for this age-group.
|
10.1371/journal.pgen.1000277 | A PI3-Kinase–Mediated Negative Feedback Regulates Neuronal Excitability | Use-dependent downregulation of neuronal activity (negative feedback) can act as a homeostatic mechanism to maintain neuronal activity at a particular specified value. Disruption of this negative feedback might lead to neurological pathologies, such as epilepsy, but the precise mechanisms by which this feedback can occur remain incompletely understood. At one glutamatergic synapse, the Drosophila neuromuscular junction, a mutation in the group II metabotropic glutamate receptor gene (DmGluRA) increased motor neuron excitability by disrupting an autocrine, glutamate-mediated negative feedback. We show that DmGluRA mutations increase neuronal excitability by preventing PI3 kinase (PI3K) activation and consequently hyperactivating the transcription factor Foxo. Furthermore, glutamate application increases levels of phospho-Akt, a product of PI3K signaling, within motor nerve terminals in a DmGluRA-dependent manner. Finally, we show that PI3K increases both axon diameter and synapse number via the Tor/S6 kinase pathway, but not Foxo. In humans, PI3K and group II mGluRs are implicated in epilepsy, neurofibromatosis, autism, schizophrenia, and other neurological disorders; however, neither the link between group II mGluRs and PI3K, nor the role of PI3K-dependent regulation of Foxo in the control of neuronal excitability, had been previously reported. Our work suggests that some of the deficits in these neurological disorders might result from disruption of glutamate-mediated homeostasis of neuronal excitability.
| Use-dependent downregulation of neuronal excitability (negative feedback) can act to maintain neuronal activity within specified levels. Disruption of this homeostasis can lead to neurological disorders, such as epilepsy. Here, we report a novel mechanism for negative feedback control of excitability in the Drosophila larval motor neuron. In this mechanism, activation by the excitatory neurotransmitter glutamate of metabotropic glutamate receptors (mGluRs) located at motor nerve terminals decreases excitability by activating PI3 kinase (PI3K), consequently causing the phosphorylation and inhibition of the transcription factor Foxo. Foxo inhibition, in turn, decreases neuronal excitability. These observations are of interest for two reasons. First, our observation that PI3K activity regulates neuronal excitability is of interest because altered PI3K activity is implicated in a number of neurological disorders, such as autism and neurofibromatosis. Our results raise the possibility that altered excitability might contribute to the deficits in these disorders. Second, our observation that group II metabotropic glutamate receptors (mGluRs) activate PI3K is of interest because group II mGluRs are implicated in epilepsy, anxiety disorders, and schizophrenia. Yet the downstream signaling pathways affected by these treatments are incompletely understood. Our results raise the possibility that the PI3K pathway might be an essential mediator of signalling by these mGluRs.
| Negative feedback processes, which can enable maintenance of neuronal homeostasis, are widely observed in neuronal systems [1]–[3]. For example, neuronal silencing via tetrodotoxin application both in vivo and in vitro increases excitability [4]–[6]. This effect occurs in vitro via both increased sodium currents and decreased potassium currents. However, the signaling pathways responsible for these excitability changes remain unclear.
The mammalian group II metabotropic glutamate receptors, which are G-protein coupled receptors activated by glutamate, are well positioned to mediate negative feedback. When localized presynaptically, these receptors can act as autoinhibitors of glutamate release [7]–[10]. Because these receptors are located outside of the active zone [11], activation is thought to occur only during conditions of elevated glutamate release and might serve to prevent glutamate-mediated neurotoxicity. Agonists for these receptors are proposed for treatment of schizophrenia, anxiety and epilepsy, among others [12],[13], but the mGluR-dependent signaling pathways that underlie these disorders remain unidentified. Furthermore, although many of the acute effects of group II mGluR activation on neuronal physiology have been elucidated [14],[15], possible long term effects on neuronal function, such as through changes in ion channel gene expression, remain essentially unexplored.
In Drosophila, the single DmGluRA gene encodes a protein most similar to the mammalian group II mGluR [16]. DmGluRA is located presynaptically at the neuromuscular junction (nmj), which suggests that DmGluRA might regulate transmitter release from motor neurons. Elimination of DmGluRA by the null mutation DmGluRA112b, or by RNAi-mediated DmGluRA knockdown specifically in motor neurons, increases neuronal excitability [16]. Given that glutamate is the excitatory neurotransmitter from Drosophila motor neurons, the increased excitability of DmGluRA mutants raised the possibility that DmGluRA decreases motor neuron excitability upon activation by glutamate released from motor nerve terminals. In this view, DmGluRA would mediate an activity-dependent negative feedback on excitability. However, the mechanism by which this negative feedback is accomplished was not elucidated.
Here we show that glutamate-mediated activation of DmGluRA decreases neuronal excitability by activating the lipid kinase PI3 kinase (PI3K), which promotes growth and inhibits apoptosis in various cell types. In particular, we report that transgene-induced inhibition of PI3K in motor neurons confers neuronal excitability phenotypes similar to DmGluRA112b, whereas transgene-induced activation of PI3K confers the opposite excitability phenotypes. We also show that PI3K activation in motor neurons suppresses the increased excitability of DmGluRA112b, and glutamate application to motor nerve terminals activates PI3K in a DmGluRA-dependent manner. Finally, we show that altered PI3K activity regulates both axon diameter and synapse number, and that these effects on neuronal growth are mediated by the Tor/S6 kinase pathway, whereas the effects of PI3K on neuronal excitability are mediated by the transcription factor Foxo. We conclude that negative feedback of Drosophila motor neuron excitability occurs via the glutamate-induced activation of DmGluRA autoreceptors, causing the PI3K-dependent inhibition of Foxo and a consequent decrease in neuronal excitability. A similar negative feedback operating in the mammalian CNS might underlie neuronal disorders involving the group II mGluRs or PI3K.
The increase in neuronal excitability conferred by the DmGluRA112b null mutation is manifested by an increased rate of onset of a form of synaptic plasticity termed long-term facilitation (LTF) [16],[17], which is induced when a motor neuron is subjected to repetitive nerve stimulation at low bath [Ca2+]. At a certain point in the stimulus train, an abrupt increase in transmitter release and hence muscle depolarization (termed excitatory junctional potential, or ejp) is observed (Figure 1A). LTF not only increases ejp amplitude, but also ejp duration, indicative of prolonged and asynchronous transmitter release (Figure 1A). This abrupt increase in the amount and duration of transmitter release is caused by an abrupt increase in the duration of nerve terminal depolarization and hence Ca2+ influx, and reflects a progressive increase in motor neuron excitability induced by the repetitive nerve stimulation: when an excitability threshold is reached, LTF occurs [17]–[19].
In Drosophila, many genotypes that increase motor neuron excitability by decreasing K+ currents or increasing Na+ currents increase the rate of onset of LTF. For example, altered activities of frequenin and Hyperkinetic, which act via K+ channels, or paralytic and pumilio, which act via Na+ channels, each increase the rate of onset of LTF [18]–[24]. By increasing motor neuron excitability, the genotypes described above apparently bring excitability closer to the threshold required to evoke LTF and consequently decrease the number of prior nerve stimulations required to reach this threshold. In these genotypes, the prolonged nerve terminal depolarizations that triggered LTF were revealed by recording ejps and simultaneously recording extracellularly electrical activity within the peripheral nerves during LTF onset. It was found that LTF onset was accompanied by the appearance within peripheral nerves of supernumerary action potentials occurring at about 10 msec intervals following the initial induced action potential [18]–[25]. Several lines of evidence suggested that these supernumerary action potentials arose in motor axons and were responsible for the increased transmitter release underlying LTF. First, the number of these supernumerary action potentials correlated with ejp duration, and second, these supernumerary action potentials often preceded depolarizing steps in the asynchronous, multi-step ejps that occurred after LTF onset. Similar supernumerary action potentials were observed following nerve stimulation in the eag Sh double mutant, in which two distinct K channel α subunits are simultaneously eliminated, and which consequently exhibits extreme neuronal hyperexcitability. In the eag Sh double mutant, these supernumerary action potentials arise in the motor nerve terminals and exhibit retrograde propagation [25]. It was suggested that the supernumerary action potentials were caused by, and also prolonged, motor nerve terminal depolarization, and thus participated in a positive feedback loop prolonging depolarization [25]. This positive feedback loop presumably underlies the abrupt, threshold-like onset of LTF.
The observation that mGluR112b increases the rate of onset of LTF suggested that DmGluRA112b increases motor neuron excitability as well [16]. To confirm this suggestion, we simultaneously recorded peripheral nerve electrical activity and ejps during LTF induced by 10 Hz stimulus trains. As previously observed in the hyperexcitable genotypes described above [18]–[25], we found that the abrupt onset of LTF in mGluRA112b was accompanied in the nerve by the appearance of supernumerary action potentials (Figure S1). This observation confirmed that LTF onset in mGluRA112b was caused by prolonged motor nerve terminal depolarization, and hence that mGluRA112b increases neuronal excitability. Thus, as suggested previously [16], it appears that DmGluRA mediates an activity-dependent inhibition of neuronal excitability. In this view, glutamate release from motor nerve terminals downregulates subsequent neuronal activity by activating presynaptic DmGluRA autoreceptors, which then decrease excitability. Elimination of DmGluRA disrupts this negative feedback and prevents the decrease in excitability from occurring.
In addition to increasing neuronal excitability, DmGluRA112b also decreases arborization and synapse number at the larval neuromuscular junction [16]. This phenotype is also observed in larval motor neurons with decreased activity of PI3K [26]. This observation raised the possibility that DmGluRA might exert its effects on neuronal excitability as well as synapse formation via PI3K activity. To test the possibility that PI3K mediates the effects of DmGluRA on neuronal excitability, we used the D42 Gal4 driver [27],[28] to overexpress transgenes expected to alter activity of the motor neuron PI3K pathway. We found that inhibiting the PI3K pathway by motor neuron-specific overexpression of either the phosphatase PTEN, which opposes the effect of PI3K, or the dominant-negative PI3KDN [29], each significantly increased the rate of onset of LTF, similarly to that of DmGluRA112b (Figure 1A and 1B). In contrast, we found that activating the PI3K pathway by expression of the constitutively active PI3K-CAAX [29], or via RNAi-mediated inhibition of PTEN, decreased rate of onset of LTF (Figure 1A and 1B). As was described above for mGluRA112b, LTF onset was accompanied by the appearance of supernumerary action potentials in the nerve (Figure S1) demonstrating that altered excitability is responsible for the altered rate of onset of LTF in these genotypes.
The rate of LTF onset described above was measured in the presence of the potassium channel blocking drug quinidine, which moderately increases neuronal excitability and hence rate of onset of LTF in the larval motor neuron. Quinidine application sensitizes the motor neuron to the effects of the nerve stimulation and enables LTF to occur reliably in genotypes with low excitability, even at lower stimulus frequencies. To demonstrate that altered PI3K activity does not alter rate of onset of LTF by altering sensitivity to quinidine, we compared the timing of LTF onset in the absence of quinidine in wildtype larvae and in larvae with inhibited PI3K. We found that inhibiting PI3K activity in motor neurons significantly accelerated LTF onset even in the absence of quinidine (Figure S2) demonstrating that altered sensitivity of motor neurons to quinidine does not underlie the altered onset rate of LTF that we observe.
In addition to effects on LTF, mutations that alter motor neuron excitability can alter basal transmitter release and hence ejp amplitude at low bath Ca2+ concentrations, at which Ca2+ influx would be limiting for vesicle fusion to occur. For example, mutations in ether-a go-go (eag), which encodes a potassium channel α subunit, increase transmitter release about two-fold [25], whereas a mutation in the sodium channel gene paralytic decreases transmitter release by increasing the frequency at which nerve stimulation failed to evoke any vesicle fusion, termed “failure” of vesicle release [30]. Presumably altered excitability affects the amplitude or duration of the action potential and consequently the amount of Ca2+ influx through voltage-gated channels. We found that DmGluRA112b also increased ejp amplitude and hence basal transmitter release at three low bath Ca2+ concentrations tested (Figure 1C), which is consistent with increased motor neuron excitability in this genotype. We found that decreasing PI3K pathway activity via motor neuron overexpression of PI3KDN or PTEN also increased transmitter release to levels similar to DmGluRA112b, whereas increasing PI3K pathway activity via overexpression of PI3K-CAAX decreased basal transmitter release (Figure 1C).
The DmGluRA112b mutation also decreased the frequency at which failures of vesicle release occur, particularly at the lower Ca2+ concentrations tested (Figure 1D). This observation confirms that the effect of DmGluRA112b on ejp amplitude is presynaptic. We also observed a decreased frequency of failures when the PI3K pathway was inhibited by motor neuron expression of PI3KDN or PTEN (Figure 1D). In contrast, motor neuron overexpression of PI3K-CAAX increased the frequency of failures (Figure 1D). Therefore, with three electrophysiological readouts, the DmGluRA112b mutant phenotype was mimicked by decreased activity of the PI3K pathway, whereas increasing PI3K pathway activity conferred opposite effects.
These observations support the notion that loss of DmGluRA increases motor neuron excitability by preventing the activation of PI3K. If so, then motor neuron expression of PI3K-CAAX, which will be active independently of DmGluRA, is predicted to suppress the DmGluRA112b hyperexcitability. To test this possibility, we drove motor-neuron expression of PI3K-CAAX in a DmGluRA112b background and found a rate of onset of LTF and ejp amplitude that was very similar to what was observed when PI3K-CAAX was expressed in a wildtype background, but significantly different from DmGluRA112b (Figure 1B, Figure 1C). In addition, motor neuron-specific expression of PI3K-CAAX increased failure frequency at the two lower [Ca2+] tested to the same level in DmGluRA112b larvae as in wildtype (Figure 1D). We conclude that hyperexcitability of the DmGluRA112b mutant results from inability to activate PI3K.
The results described above suggest that glutamate release from motor nerve terminals as a consequence of motor neuron activity activates PI3K within motor nerve terminals via DmGluRA autoreceptors. To test this possibility directly, we measured the ability of glutamate applied to the neuromuscular junction to activate PI3K within motor nerve terminals. To assay for PI3K activity we applied an antibody specific for the phosphorylated form of the kinase Akt (p-Akt), which is increased by elevated PI3K pathway activity. The usefulness of this antibody for specific detection of Drosophila p-Akt has been previously demonstrated [31]–[33]. The ability to detect p-Akt in larval motor nerve terminals overexpressing PI3K-CAAX, but not in wildtype (Figure 2), further validates this antibody as a PI3K reporter.
We compared p-Akt levels in wildtype versus DmGluRA112b motor nerve terminals immediately prior to or following a 1 minute application of 100 µM glutamate. We found that glutamate application strongly increased p-Akt levels in wildtype larvae, but not in the DmGluRA112b larvae (Figure 2), demonstrating that glutamate application increases nerve terminal p-Akt levels, and that DmGluRA activity is required for this increase.
We found that DmGluRA activity was required presynaptically for this p-Akt increase: motor neuron-specific expression of a DmGluRA RNAi construct [16], blocked the ability of glutamate to increase p-Akt levels (Figure 2). In [16] it was reported that expression of DmGluRA RNAi decreased, but did not eliminate, mGluRA immunoreactivity, suggesting that this transgene decreases, but does not eliminate, glutamate-mediated signalling via mGluRA. The ability of this transgene to block glutamate-mediated induction of p-Akt suggests that activation of PI3K by glutamate is sensitive to mGluRA levels and requires a minimum level of mGluRA expression. In contrast, expression of the DmGluRA RNAi construct in the muscle, with use of the 24B Gal4 driver, did not inhibit p-Akt levels: p-Akt intensity following 1 minute of glutamate application was not significantly different from wildtype (17.6+/−2.9, p = 0.59).
To determine if PI3K activity was required presynaptically for this glutamate-induced p-Akt increase, we inhibited PI3K activity by motor neuron-expression of PI3KDN, and found that this transgene significantly inhibited the ability of glutamate to activate p-Akt (Figure 2). Thus, presynaptic DmGluRA and PI3K activity are both necessary for glutamate to increase p-Akt.
Many effects of the PI3K pathway are mediated by the downstream kinase Akt. Activated Akt phosphorylates targets such as Tsc1/Tsc2, which regulates cell growth via the Tor/S6 Kinase (S6K) pathway [34], Foxo, which regulates apoptosis [35], and GSK3 [36], which mediates at least in part the effects of altered PI3K pathway activity on arborization and synapse number [26]. All of these Akt-mediated phosphorylation events inhibit activity of the target protein.
If PI3K pathway activity decreases neuronal excitability by inhibiting Foxo, then Foxo overexpression is predicted to mimic the hyperexcitability observed when PI3K pathway activity is blocked in motor neurons, whereas loss of Foxo is predicted to mimic the hypoexcitability observed when PI3K-CAAX is expressed in motor neurons. To test these predictions, we measured neuronal excitability in larvae carrying the heteroallelic Foxo21/Foxo25 null mutant combination [37] and in larvae overexpressing Foxo+ [38] in motor neurons. We found that overexpression of Foxo+ increased the rate of onset of LTF, basal transmitter release and frequency of successful ejps to a level very similar to that observed when PI3K pathway activity was decreased (Figure 3) whereas in Foxo21/Foxo25 larvae, the rate of onset of LTF, basal transmitter release and frequency of successful ejps were decreased to levels very similar to those observed when PI3K-CAAX was expressed in motor neurons (Figure 3). These observations support the notion that PI3K activity decreases excitability by downregulating Foxo activity.
If the hyperexcitability conferred by motor neuron expression of PI3KDN results from Foxo hyperactivity, then the Foxo21/Foxo25 null combination will suppress this hyperexcitability and confer motor neuron hypoexcitability similar to what is observed in Foxo21/Foxo25 larvae in an otherwise wildtype background. We confirmed this prediction: larvae carrying the Foxo21/Foxo25 null combination and expressing PI3KDN in motor neurons exhibited a rate of onset of LTF, basal transmitter release, and failure frequency very similar to what was observed in the Foxo21/Foxo25 null mutant alone (Figure 3), or in larvae expressing PI3K-CAAX in motor neurons. We used the OK6 motor neuron Gal4 driver rather than D42 for ease of stock construction in experiments involving Foxo21/Foxo25. OK6 confers motor neuron phenotypes indistinguishable from D42 in our assays (Figure 3B and not shown).
In addition, if the hypoexcitability conferred by motor neuron expression of PI3K-CAAX results from decreased Foxo activity, then co-overexpression of Foxo+ will suppress this hypoexcitability and confer hyperexcitability similar to what is observed when PI3KDN, PTEN or Foxo+ alone are expressed in motor neurons. We confirmed this prediction: larvae co-expressing Foxo+ and PI3K-CAAX in motor neurons exhibited rate of onset of LTF, basal transmitter release and failure frequency very similar to what was observed when PI3KDN, PTEN, or Foxo+ alone were expressed in motor neurons (Figure 3). Thus, eliminating Foxo reverses the hyperexcitability conferred by blocking PI3K pathway in motor neurons, whereas overexpressing Foxo+ reverses the hypoexcitability confered by activating PI3K in motor neurons. These epistasis tests support the notion that PI3K activity decreases motor neuron excitability by inhibiting Foxo.
In contrast, we found that altering the Tor/S6K pathway had little effect on motor neuron excitability. In particular, motor neuron expression of neither the dominant-negative S6KDN nor the constitutively active S6KAct transgene [39] had any effect on the rate of onset of LTF (Figure 4). In addition, except for the appearance of some enhancement at the lowest stimulus frequency applied, expression of S6KDN had no effect on the ability of PI3K-CAAX to decrease the rate of onset of LTF (Figure 4). Furthermore, expression of S6KDN had no effect on basal transmitter release, and did not affect the ability of PI3K-CAAX to depress basal transmitter release (data not shown). Therefore we conclude that the Tor/S6K pathway does not mediate the effects of PI3K on neuronal excitability.
Because altered PI3K pathway activity alters motor neuron arborization and synapse number [26], it seemed possible that a causal relationship existed between the PI3K-mediated excitability and neuroanatomy defects. To test this possibility, we evaluated the roles of the Tor/S6K and Foxo pathways in mediating the effects of altered PI3K activity on synapse number. We found that motor neuron-specific expression of S6KAct increased synapse number to an extent similar to PI3K-CAAX, and motor neuron expression of S6KDN partially suppressed the increase in synapse number conferred by PI3K-CAAX (Figure 5A and 5B). These observations suggest that S6K mediates in part the effects of PI3K on arborization and synapse number. However, the ability of S6KDN to suppress only partially the effects of PI3K-CAAX overgrowth suggests that both Tor/S6K and a second, PI3K-mediated, pathway (presumably involving GSK3) regulate synapse formation. A role for the Tor/S6K in the control of synapse number was previously reported by Knox et al. (2007). In this report, null mutations in S6K decreased synapse number as well as muscle size at the larval nmj. However, it was further reported that activation of the PI3K effector Rheb, which activates Tor/S6K, increased synapse number at the larval nmj even when Tor activity was inhibited by rapamycin [40], raising the possibility that Rheb activates synapse formation via multiple redundant pathways, including Tor/S6K.
In contrast to the effects of altered S6K on synapse formation, we found that Foxo+ overexpression had no effect on synapse number (data not shown) and failed to suppress the growth-promoting effects of PI3K-CAAX (Figure 5A and 5B).
We found that the PI3K pathway also affects axon diameter. In Drosophila peripheral nerves, about 80 axons, including about 35 motor axons, are wrapped by about three layers of glia, as shown in the transmission electron micrograph from cross sections of peripheral nerves in Figure 5C. We found that motor neuron specific expression of PTEN decreased motor axon diameter, whereas motor-neuron specific expression of PI3K-CAAX increased motor axon diameter. Tor/S6K, but not Foxo, mediates this growth effect. In particular, motor neuron-specific expression of S6KAct increased axon diameter to an extent similar to PI3K-CAAX, and motor-neuron-specific expression of S6KDN decreased motor axon diameter to an extent similar to PTEN and also partially suppressed the growth-promoting effects conferred by PI3K-CAAX. In contrast, Foxo+ overexpression did not have a significant effect on the ability of PI3K-CAAX to increase axon diameter (Figure 5D). Therefore, Foxo mediates the excitability effects, but not the growth-promoting effects, of altered PI3K pathway activity, whereas the Tor/S6K pathway mediates in part the growth promoting effects but not the excitability effects of altered PI3K pathway. We conclude that the excitability and growth effects are completely separable genetically and thus have no causal relationship.
Depending on the system, neuronal activity can either restrict or promote synapse formation [41]. The Drosophila eag Sh double mutant, in which two distinct potassium channel subunits are simultaneously disrupted, displays extreme neuronal hyperexcitability [25], and a consequent increase in synapse number [42],[43]. This activity-dependent increase in synapse number does not require DmGluRA activity [16], suggesting that excessive glutamate release is not necessary for this excessive growth to occur. To determine if PI3K activity is required for this overgrowth, we compared synapse number in wildtype larvae, in larvae expressing dominant-negative transgenes for both eag (eagDN) and Sh (ShDN) [44],[45] in motor neurons, and in larvae co-expressing eagDN, ShDN and PI3KDN. We found that co-expression of eagDN and ShDN in motor neurons increased synapse number similarly to what was observed previously [42], and that this increase was completely blocked by simultaneous expression of PI3KDN but not by lacZ (Figure 6). Thus, the activity-dependent increase in synapse formation requires PI3K activity. The observation that glutamate activation of DmGluRA is not necessary for this increase raises the possibility that another PI3K activator contributes to synapse formation at the larval nmj. Insulin is a plausible candidate for such an activator because both insulin and insulin receptor immunoreactivity are present at the nmj [46].
The effects on neuronal excitability of altered DmGluRA, PI3K, and Foxo activities are consistent with a model in which glutamate released from motor nerve terminals as a consequence of motor neuron activity activates motor neuron PI3K via DmGluRA autoreceptors, which then downregulate neuronal excitability via inhibition of Foxo (Figure 7). Foxo, in turn, might regulate excitability via transcription of ion channel subunits or regulators. Although such putative Foxo targets have not been identified, one potential target might be the translational repressor encoded by pumilio (pum): pum expression is downregulated by neuronal activity, Pum decreases transcript levels of the sodium channel encoded by para, and both para overexpression and pum mutations increase rate of onset of LTF in a manner similar to that described here [19],[21],[23],[24].
The DmGluRA-dependent negative feedback reported here co-exists with several other negative feedback systems operating at the Drosophila nmj. In addition to altered excitability, these systems include alterations in the vesicle release properties of the motor nerve terminal and density of the muscle glutamate receptors [1]. Presumably, these diverse feedback systems, acting in parallel, regulate specific aspects of neuronal function. The DmGluRA-dependent feedback system reported here differs in several respects from some of the other feedback systems reported. For example, this DmGluRA-dependent feedback apparently involves transcriptional changes, suggesting that this system operates on a long time scale and thus will be responsive to chronic, rather than acute, changes in neuronal activity. In addition, this system is likely to be motor neuron-cell autonomous, and will not involve participation of additional cells, such as target muscles or adjacent glia. Furthermore, this system is predicted to link mechanistically several PI3K-dependent processes, including activity-dependent downregulation of neuronal excitability and upregulation of neuronal growth. In this regard, the PI3K-dependent inhibition of Foxo might protect neurons from excitotoxic effects of prolonged stimulation; such protection would not be accomplished by the other feedback systems operating.
Both mGluRs and PI3K have been previously implicated in regulation of ion channel activity. For example, ligand activation of group I mGluRs trigger Gαq-mediated release of Ca2+ from intracellular stores, and consequently activate Ca2+-dependent K+ channels and nonselective cation channels [47], whereas activation of group II mGluRs inhibit transmitter release via inhibition of P/Q Ca2+ channels [48],[49]. PI3K activation can promote ion channel insertion into cell membranes [50]–[52] and can mediate the decrease in excitability conferred by application of leptin, the product of the obese gene, by activating Ca2+-dependent K+ channels [53]. However, to our knowledge, effects of mGluR or PI3K activation on ion channel transcription in the nervous system have not been reported.
PI3K also regulates ion channel activity in non-excitable cells. PI3K mediates the ability of insulin growth factor to activate the Eag channel, and the ability of serum to activate the intermediate-conductance Ca2+-activated K+ channel in breast carcinoma and lymphoma cells, respectively [54],[55]. Interestingly, in these non-excitable cells, activation is accomplished by both acute effects on channel activity as well as long term effects as a consequence of increased channel transcription. Therefore, PI3K can regulate channel activity over different time courses, and via distinct mechanisms, presumably via distinct effector pathways.
Human orthologues of group II mGluRs and PI3K are implicated in several neurological disorders. For example, group II mGluRs are potential drug targets for schizophrenia, epilepsy, and anxiety disorders [12]–[14], raising the possibility that altered excitability of glutamatergic neurons might play a role in these disorders. In addition, levels of phospho-Foxo, a product of PI3K/Akt activity, are increased following induction of seizures in rats, and in the hippocampi of epileptic patients [56]. This activity-induced increase in phospho-Foxo was interpreted as a mechanism to protect neurons from the excitotoxic effects of excessive glutamate release because Foxo is more likely than phospho-Foxo to promote apoptosis. Our results raise the possibility that this increase in phospho-Foxo levels occurs via glutamate-induced PI3K activation mediated by group II mGluRs, and interpret this increase as a negative feedback on excitability. A role for PI3K activity in inhibiting epileptic seizures is further supported by the recent observation that application of leptin, a known PI3K activator, inhibits seizures in a PI3K-dependent manner [57]. Increased insulin/IGF levels and increased PI3K activity are also implicated in autism spectrum disorders [58],[59]. These increases are generally hypothesized to affect neuronal function by increasing arborization and synapse formation, but our results raise the possibility that altered neuronal excitability might also contribute. Thus, the results reported here might have significance for several human neurological disorders.
The mechanism by which glutamate-activated DmGluRA activates PI3K remains unknown. Although mammalian group I mGluRs activate PI3K via the Homer scaffolding protein and the PI3K enhancer PIKE [60], Drosophila mGluRA, similar to mammalian group II mGluRs, lack Homer binding motifs [61] and thus would not be predicted to activate PI3K by this mechanism. Alternatively, although the inhibition of glutamate-induced p-Akt activation by PI3KDN expression demonstrates that PI3K activity is required for this activation, it remains possible that glutamate increases p-Akt levels by activating an enzyme in addition to PI3K. For example, Akt is reported to be phosphorylated and activated by Calmodulin-dependent kinase kinase [62]. Additionally, glutamate-activated DmGluRA might activate PI3K in motor nerve terminals indirectly by triggering Ca2+ release from stores, leading to release of insulin and hence activation of PI3K by well-established mechanisms. Further experiments will be required to address these issues.
Fly stocks were maintained on standard cornmeal/ agar Drosophila media at room temperature. D42 and OK6 express Gal4 in motor neurons and were provided by Tom Schwarz, Boston, Massachusetts, and Hermann Aberle, Tubingen, Germany respectively. The UAS-PI3KDN (D954A) and UAS-PI3K-CAAX transgenes were provided by Sally Leevers, London, UK, the UAS-Foxo+ transgene was provided by Marc Tatar, Providence, RI, the Foxo21 and Foxo25 lines were provided by Heinrich Jasper, Rochester, NY, the UAS-S6KDN and UAS-S6Kact transgenes were provided by Ping Shen, Athens, GA, and the UAS-DmGluRA-RNAi and the DmGluRA112b lines were provided by Marie-Laure Parmentier, Montpelier, France. All other fly stocks were provided by the Drosophila stock center, Bloomington, IN.
FITC conjugated antibodies against horseradish peroxidase (HRP) were raised in goat (Jackson ImmunoResearch) and were used at 1∶400 dilution. Antibodies against Drosophila p-Akt (Ser505) were raised in rabbit (Cell Signaling Technologies) and were used at 1∶500 dilution. Rhodamine Red conjugated goat anti-rabbit (Jackson ImmunoResearch) was used at a dilution of 1∶1000. For arborization measurements, larvae were dissected in PBS-T and fixed in 4% paraformaldehyde. Images were taken on a Zeiss 410 laser scanning confocal microscope (LSM) with a 20× objective. ImageJ was used to obtain surface area measurements of muscle 6 from abdominal segment A3, and the number of boutons was counted manually. For p-Akt measurements, larvae were dissected in Grace's insect cell culture media (Gibco). When glutamate was applied,100 µM glutamic acid monosodium salt monohydrate (Acros Organics) dissolved in Grace's insect cell culture media was added to the well of the dissection plate. 1 minute after glutamate addition, larvae were rapidly washed in standard saline (0.128 M NaCl, 2.0 mM KCl, 4.0 mM MgCl2, 0.34 M sucrose, 5.0 mM HEPES, pH 7.1, and 0.15 mM CaCl2), and then immediately fixed in 4% paraformaldehyde. For the 10 min wash, the larvae were washed in Grace's insect cell culture media and placed on shaker for 10 minutes before fixing. Care was taken to treat all samples identically during this procedure. Images were taken on a Zeiss 510 LSM with a 20× objective. Z-stacks were compiled from 2 µm serial sections to a depth adequate to encompass the entire bouton thickness for each sample (from 8–20 µm). Muscles 7 and 6 from either abdominal segments A3 or A4 were used for measurements. ImageJ software was used to analyze p-Akt intensities. In particular, 2D projections were created using the median pixel intensity from each stack at each coordinate point. Neuronal structures, marked by anti-HRP, were traced using the freehand selection tool and the selection was transferred to the anti-p-Akt image where the mean pixel intensity value was measured. Background was obtained with a selection box encompassing the non-neuronal area of muscles 6 and 7 in the particular abdominal segment, the mean pixel intensity was measured and subtracted from the mean p-Akt pixel intensity.
The D42 motor neuron driver was used to express transgenes for all experiments, except that the OK6 driver was used for experiments the Foxo21/Foxo25 genotype was included. OK6 is located on a different chromosome from Foxo, which simplifies stock construction. Larvae were grown to the wandering third-instar stage in uncrowded bottles at room temperature and dissected as described (17, 18) in standard saline solution (128 mM NaCl, 2.0 mM KCl, 4.0 mM MgCl2, 34 mM sucrose, 5.0 mM HEPES, pH 7.1, and CaCl2 as specified in the text). Peripheral nerves were cut posterior to the ventral ganglion and were stimulated using a suction electrode. Muscle recordings were taken from muscle 6 in abdominal sections 3–5. Stimulation intensity (5 V for approximately 0.05 msec) was adjusted to 1.5 times threshold, which reproducibly stimulates both axons innervating muscle cell 6. Recording electrodes were pulled using a Flaming/Brown micropipette puller to a tip resistance of 10–40 MΩ and filled with 3M KCl. LTF and ejp amplitude data are reported as geometric, rather than arithmetic means, because the data show a positive skew. For extracellular recordings of neuronal action potentials, a loop of nerve near the nerve terminal was introduced into a suction electrode and nerve activity recorded with a DAM-80 differential amplifier.
Larvae were grown to the wandering third-instar stage in uncrowded bottles at room temperature. Dissections and preparation for microscopy were performed as previously described [63]. Nerve cross sections close to (within about 10 µm from) the ventral ganglion were obtained and analyzed. Axon diameter measurements were taken from the five largest axons from five different nerves from at least two different larvae.
|
10.1371/journal.pntd.0005666 | Visceral leishmaniasis in an environmentally protected area in southeastern Brazil: Epidemiological and laboratory cross-sectional investigation of phlebotomine fauna, wild hosts and canine cases | Leishmaniasis is a rapidly expanding zoonosis that shows increasing urbanization. Concern exists regarding the role of wildlife in visceral leishmaniasis (VL) transmission, due to frequent natural or anthropogenic environmental changes that facilitate contact between wildlife, humans and their pets. The municipality of Campinas, in southeastern Brazil, initially recorded VL in 2009, when the first autochthonous case was confirmed in a dog living in an upscale residential condominium, located inside an environmentally protected area (EPA). Since then, disease transmission remains restricted to dogs inhabiting two geographically contiguous condominiums within the EPA.
We conducted a cross-sectional study of the VL focus to investigate Leishmania spp. infection in domestic dogs, wild mammals and sand flies using molecular tools and recommended serological techniques. Canine seroprevalences of 1.5% and 1.2% were observed in 2013 and 2015, respectively. Six insect species, confirmed or suspected vectors or potential transmitters of Leishmania, were identified. Two specimens of the main L. (L.) infantum vector in Brazil, Lutzomyia longipalpis, were captured in the EPA. Natural infection by L. (L.) infantum was recorded in one Expapillata firmatoi specimen and two Pintomyia monticola. Natural infection by L. (L.) infantum and Leishmania subgenus Viannia was also detected in two white-eared opossums (Didelphis albiventris), a known reservoir of VL. Geographical coordinates of each sampling of infected animals were plotted on a map of the EPA, demonstrating proximity between these animals, human residences, including the dogs positive for VL, and forest areas.
The EPA, which is inhabited by humans, has an active VL focus. The risk of establishing and maintaining disease transmission foci in similar scenarios, i.e. wild areas that undergo environmental modifications, is evident. Moreover, different epidemiological profiles of VL must be included to elaborate prevention and control measures that consider the particularities of each transmission area.
| Leishmaniasis are neglected tropical diseases that represent a major public health problem in the world. New outbreaks have been recorded in the outskirts of some cities and phlebotomine vectors show geographic expansion, potentially associated with environmental and climate changes. We investigated a recent focus of visceral leishmaniasis (VL) in dogs from two residential condominiums located in an environmentally protected area in the municipality of Campinas, São Paulo, southeastern Brazil. The purpose of the study was to investigate Leishmania (L.) infantum infection, the etiological agent of the disease, in vector insects (phlebotominae), domestic (dog) and wild hosts. Antibody prevalence was determined in 1.5% and 1.2% of dogs examined in 2013 and 2015, respectively. In addition, natural infection by the VL agent was confirmed in two species of sand flies and two white-eared opossums. All positive animals were captured near human residences and in forest areas. Scenarios like these, with changes in ecosystems and the proximity of wild and domestic animals and humans, can promote the occurrence of zoonoses. The participation of wildlife in the transmission cycle of zoonoses has assumed greater importance in recent years as a result of the consolidation of the One Health concept.
| In the last 35 years, visceral leishmaniasis (VL) has reemerged as an important public health problem in different parts of the world, with the expansion of transmission and the occurrence of new cases [1].
In the Americas, the disease occurs within a complex network of relationships involving the etiological agent, Leishmania (Leishmania) infantum (Synonym: L. chagasi) and its different populations: sand flies, mainly of the species Lutzomyia longipalpis [2]; domestic or wild vertebrates, sources of vector infection; and susceptible hosts. All of these can be present in a certain time and space, interacting with the particularities of the environment.
Although the dog is considered the main reservoir in the urban environment, it is necessary to consider the possibility of other non-human hosts participating in the maintenance of L. (L.) infantum in the endemic environment. These include several species belonging to different orders of wild and/or synanthropic animals, which are naturally infected in several regions and ecotypes [3].
In Brazil, until the mid-1980s VL was considered a disease of primarily wild and rural environments. From the late 1980s onward, epidemics have occurred in the urban environment, together with rapid expansion of transmission foci in hundreds of municipalities, including populous urban centers and several capitals [4–6]. The rural epidemiological pattern has been modified by increasing urbanization, while the geographic expansion of the disease has been observed in previously safe municipalities [7–10]. Expansion of the spaces occupied by Lu. longipalpis has also been verified in endemic urban environments in several regions of the country [11,12].
In the state of São Paulo, southeastern Brazil, VL transmission is also in full expansion. Over a period of 17 years, from 1997 to 2014, the presence of the vector Lu. longipalpis has been confirmed in 177 municipalities in the state, with 2,467 autochthonous human cases, 214 deaths and a mortality rate of 8.7% [13,14].
It is currently possible to recognize at least three different transmission patterns and epidemiological profiles of the disease in the state of São Paulo: the first occurs in 76 municipalities and involves both human and canine VL cases, besides the presence of the vector Lu. Longipalpis. The second occurs in 47 municipalities, seven of which without records of Lu. longipalpis, and involving only canine cases, with no record of human autochthony. Finally, a third pattern involves only human cases, which occurs in nine municipalities, three of which without reports of the vector [13].
Little is known of the dynamics of the expansion of new outbreaks, the environmental and epidemiological determinants of the disease, or the dispersion and behavior of vectors and hosts potentially involved in the different VL transmission cycles. What is known, however, is that increasing environmental degradation of natural and/or anthropogenic origin is a concern in several parts of the world.
The destruction or replacement of the original vegetal cover of a region can alter the composition and interaction of its fauna [15], which can lead to modifications and adaptations in the biology of vectors and hosts of diseases like leishmaniasis.
The initial VL focus in Campinas was identified in 2009, based on an autochthonous canine case reported in an environmentally protected area (EPA) within the municipality [16,17]. This area has undergone intense environmental changes since the 1970s, with deforestation for road and residential condominiums construction.
The epidemiological investigation conducted by the Municipal and State Health Department for the first case of canine VL (CVL) in 2009 determined an anti-Leishmania seroprevalence of 2.0% (4/198) in dogs and none of the 40 wild mammals captured in this occasion was positive for L. (L.) infantum by polymerase chain reaction (PCR). Sand flies were present in 16/85 (18.8%) residences and Lu. longipalpis was verified in three of the 16 (18.8%) houses with the presence of sand flies, being 22.2% of the females positive for L. (L.) infantum by PCR [17].
Since it was the first suspected case of autochthony for VL and in order to confirm the occurrence of L. (L.) infantum for the first time in this geographic area, at that occasion biological samples collected from dogs diagnosed positive in serological tests were sent to a reference laboratory, the Adolfo Lutz Institute (IAL), which confirms the occurrence of the parasite through isolation in culture medium and molecular tools.
From 2009 to 2016, no human cases have yet been diagnosed in Campinas and notification of canine cases remains geographically restricted to the two contiguous residential condominiums located inside the EPA.
In order to identify the different components involved in the transmission cycle of this CVL focus in Campinas, we conducted a broad epidemiological investigation involving the capture of sand flies and free-living wild animals in other areas of the EPA and integrated this research with the results of canine serological surveys.
This research was approved by the Ethics Committee on Animal Use (ECAU) of Campinas State University (protocol no. 3296–1), the ECAU of the Adolfo Lutz Institute and the Pasteur Institute (protocol no. 01/2013) and by the Brazilian Institute of Environment and Renewable Natural Resources, through the Biodiversity Authorization and Information System (IBAMA, SISBIO, no. 42926-1/2).
The municipality of Campinas (22°53’20” S, 47°04’40” W) has approximately one million inhabitants and is located in the southeast region of São Paulo, 100 km from the state capital. In 1993, the Mayor’s Office of the municipality of Campinas delimited an area of 223 Km2 in the east of the city as an EPA, located between latitudes 22°45’00” and 22°56’00” S and longitudes 46°52’30” and 47°00’00” W [15,18]. This area contains the largest number of forest fragments in the municipality, including several Atlantic Forest remnants, as well as fauna diversity of 68 species of wild mammals, including several endangered species [15,19].
Despite the cultural and biological richness, over the last few decades, the EPA has undergone significant environmental changes, with the implantation of upscale horizontal condominiums, which opened up new roads, established new inhabitants and altered the delimitation of the urban perimeters [15,17,20].
In 2009, the first autochthonous case of CVL in the municipality of Campinas was reported in a dog that lived in one of these residential condominiums inside the EPA, close to residual forest fragments, where contact between the wild fauna, humans and their pets can occur [16,17]. Since then, new canine cases have been reported in this same condominium and in another that is geographically contiguous and shares the same environmental characteristics. The proximity of human residences to these green areas is shown in Fig 1, together with the area of CVL focus in the EPA.
We set traps at 18 different points throughout the EPA to capture vectors and wild animals. These capture points were defined based on criteria that included easy of access by roads, proximity to water courses, size of the forest area, forest connections with urban perimeters and proximity to the CVL focus. They were defined with the help of Google Earth software (Google, version 7.1.2.2041, compiled in 2013).
A serological survey was conducted among domestic dogs in the condominium residences where the CVL focus was located. The sampling points, obtained by GPS (Global Position System), are shown in Fig 2.
The forest fragments and water resources of the EPA were mapped by the Municipal Health Department of Campinas using a geographic information system (GIS) with the MapInfo software, using a matrix layer of an aerial photograph of the region, obtained in 2007. A vector layer was superimposed, which showed the remnants of Atlantic Forest in the state of São Paulo, produced by the project “SOS Mata Atlântica”, obtained from the National Institute of Space Research website (INPE, http://www.inpe.br). ArcGIS software, v. 10.1 (Environmental System Research Institute, CA), was used to plot the geographic coordinates and draw the maps with all geographic objects and the cartographic base of the municipality available from the Brazilian Institute of Geography and Statistics (IBGE).
The survey area of canine cases in the EPA was defined based on the VL Surveillance Program of the state of São Paulo, which establishes a radius of 200 m from the confirmed canine case for conducting surveillance and control actions, and is expanded, whenever necessary, until samples from at least 100 dogs are obtained [21]. In this study, we included blood serum samples collected from all dogs resident in the two condominiums where the CVL focus was located, in addition to areas adjacent to the same (Fig 2).
The serological survey was conducted as part of the planned monitoring and control actions of the Municipal Health Department of Campinas and the IAL, performed in the dog population residing inside the EPA.
After authorization from the owners, data pertaining to the dogs were recorded in a database and serum samples were collected. The serological tests used were those recommended by the National Program for Surveillance and Control of Leishmaniasis: the Dual Path Platform rapid immunochromatographic test (TR DPP, Bio-Manguinhos, Brazil) for screening and the enzyme-linked immunosorbent assay (ELISA, Bio-Manguinhos, Brazil) to confirm the diagnosis, following the manufacturers’ protocols.
In this study, we included the surveys conducted in 2013 and 2015, considering that in 2014, there was no supply of kit for canine serological diagnosis from Brazilian Ministry of Health to the Municipal Health Department of Campinas, making it impossible to conduct the serological survey in this year, since this is the official technique for dog survey in Brazil.
Sand flies and mammal captures were made monthly from April 2014 to March 2015, for three consecutive nights per month, simultaneously in three different points selected in EPA: one inside the condominiums where there is the CVL focus, another in the forest fragments with no contact with urban areas and a third randomly selected point.
For the collection of sand flies we used two or three modified CDC light traps (Horst, Brazil) per point of collection for two or three consecutive nights. The light traps were set up about one meter from the forest ground [22]. Species were classified as proposed by Galati [22] and the abbreviation scheme of genus names was that described by Marcondes [23].
Due to their high morphological resemblance, taxonomic identification of Brumptomyia females ended at the genus level. Evandromyia sallesi and Ev. cortelezzii were considered as Ev. cortelezzii-sallesi, based on the intermediate morphological forms observed and the morphological similarity of their females.
Wild mammals were captured for three consecutive nights per month, also from April 2014 to March 2015, using sixty tomahawk traps (Gabrisa, Brazil) with dimensions of 30 x 21 x 20 cm and 55 x 20 x 20 cm, baited with banana and chicken simultaneously at three different points in the EPA. A total of 20 traps per point (totally 60 traps) were distributed along the soil of forest region in linear transects with 10 to 20 meters between the traps [24].
The captured mammals were identified with microchip implants (Microchips Brasil, Brazil), and blood samples were collected by venipuncture during manual containment. Samples were stored at -20°C until DNA extraction.
The capture, containment, handling and sampling of the captured mammals were all carried out according to guidelines for the use of wild mammals in research, as recommended by the American Society of Mammologists [25].
DNA extraction of total blood samples from wild mammals was performed within 48 h of the conclusion of each capture, using a QIAamp DNA mini kit in a QIAcube DNA extractor (Qiagen, Netherlands).
To measure the quantity and purity of DNA from the samples, we used the NanoDrop ND-1000 (Thermo Fisher Scientific, EUA) equipment. To evaluate the endogenous quality, DNA integrity and the presence of inhibitors in blood samples, we performed PCR reactions using primers for the conserved gene of the interphotoreceptor retinoid-binding protein (IRBP), IRBP-CF-FWD (5'-TCCAACACCACCACTGAGATCTGGAC-3') and IRBP-CF-REV (5'-GTGAGGAAGAAATCGGACTGGCC-3') [26], or for the β1 (5’-ACCACCAACTTCATCCACGTTCACC-3’) and β2 genes (5’-CTTCTGACACAACTGTGTTCACTAGC-3’) [27].
The female sand flies were grouped in pools (1–6 specimens) according to species and location. DNA was extracted using a DNeasy Blood and Tissue kit (Qiagen, Netherlands). The quantity and purity of the DNA samples was measured using an Epoch spectrophotometer (BioTek, Winooski, Vermont).
PCR was performed using the primers LITSR (5’ CTGGATCATTTTCCGATG 3’) and L5.8S (5’ TGATACCACTTATCGCACTT 3’) to amplify a 300–350 base pair (bp) fragment from the intergenic region of the Leishmania DNA, internal transcribed spacer-1 (ITS-1) [28] and primers Lch14 and Lch15, which amplify a 167 bp of the kinetoplast minicircle DNA of L. (L.) infantum [29].
As reaction controls, we used ultrapure water and extracted DNA from in vitro cultures of standard strains of L. (L.) infantum (MHOM/BR/2002/LPC-RPV), L. (V.) braziliensis (MHOM/BR/1975/M2903) and L. (L.) major (MHOM/IL/1980/FRIEDLIN).
The reaction mixture contained 1.3 μL of buffer (50mM KCl, 20mM Tris-HCl pH 8.4), 0.4 μL MgCl2 (1.6 μM), 0.25 μL of each oligonucleotide (0.2 μM), 0.25 μL dNTP (0.2 mM), 0.25 μL of Platinum Taq DNA polymerase (Invitrogen, Brazil) and 8.3 μL of ultrapure water, with 1 μL of extracted DNA with a minimal concentration of 10 ng/μL. Thermal cycling conditions followed those of El Tai et al. (2000) [28] and Silva et al. [29]. The amplified products were identified by 1.5% agarose gel electrophoresis containing 1.0 μL/10 mL of SYBR safe DNA gel stain (Invitrogen, Life Technologies, USA).
Amplified products were purified with the Illustra GFX PCR DNA and Gel Band Purification kit (GE Healthcare, UK). Genetic sequencing by the Sanger method was performed in a Genetic Analyzer 3500 automated sequencer using a BigDye Terminator v 3.1 Cycle Sequencing kit (Applied Biosystems, Life Technologies, USA).
The sense and antisense sequences were visualized using Chromas v 2.1.1 software (Technelysium Pty Ltd, Australia), and were submitted to global alignment using MEGA5 software [30] and compared with sequences deposited in the GenBank, using the nucleotide basic local alignment search tool (BLASTn, http://www.ncbi.nlm.nih.gov/BLAST).
The annual anti-Leishmania canine seroprevalences in the study area and the respective 95% confidence intervals (95%CI) were calculated. The frequency and percentages of sand flies and wild mammals infected, with their respective 95%CI, were also described. Statistical analysis was performed in Stata software, v. 11.0 (StataCorp LP, USA).
Human residences were observed in close proximity to the forest fragments, where the CVL focus is located inside the Campinas EPA (Fig 1). The distribution of trap sites for capturing sand flies and wild animals is shown in Fig 2, besides domestic dogs sampled.
Fig 3 shows the spatial distribution of seropositive dogs, sand flies and wild mammals naturally infected by species of Leishmania, confirmed by molecular techniques.
In the canine surveys of 2013 and 2015 were evaluated, respectively, 590 and 571 blood samples from domestic dogs, performing 1,161 examinations (S1 Table). The seroprevalence determined and the respective confidence interval are shown in Table 1.
Four hundred and seventy-seven specimens of sand flies were collected, 291 of which were male and 186 female (S2 Table). Six species known or suspected of being vectors were identified as follows: Nyssomyia whitmani (107 males, 47 females), Migonemyia migonei (81 males, 34 females), Pintomia fischeri (21 males, 12 females), Nyssomyia neivai (12 males, 13 females), Pintomyia pessoai (3 males, 1 female) and Lu. longipalpis (2 males) (Table 2).
The ITS-1 PCR detected a 300–350 bp fragment in three samples (1.6%; 95%CI 0.3–4.6%) belonging to Ex. firmatoi (1/10; 10.0%; 95%CI 0.2–44.5%) and Pi. monticola (2/25; 8.0%; 95%CI 1.0–26.0%) that characterized the samples as positive for Leishmania genus.
These three samples, one of Ex. firmatoi and two of Pi. monticola, were also amplified with Lch14/15 primers, producing the expected 167 bp products, later confirmed as L. chagasi (synonymous L. (L.) infantum) by genetic sequencing that showed 96 to 100% similarity with kinetoplast minicircle sequences deposited in the GenBank (Accession numbers AF308682.1, E values = 2e-37; 2e-50 and 2e-49).
Eighty-two wild mammals from six species were sampled: 1 Mazama gouazoubira (brocket deer; 1.2%), 1 Sciurus (Guerlinguetus) aestuans (squirrel; 1.2%), 8 Callithrix jacchus (white-tufted-ear marmoset; 9.8%), 11 Didelphis aurita (black-eared opossum; 13.4%), 18 Callithrix penicillata (black-tufted-ear marmoset; 22.0%) and 43 Didelphis albiventris (white-eared opossum; 52.4%). Six (6/82; 7.3%) D. albiventris were identified by a microchip reading in a second capture and sampled again (S3 Table).
Amplification of an expected 300–350 bp product was observed in 2/88 (2.3%; 95%CI 0.3–8.0%) samples from D. albiventris (2/43; 4.7%; 95%CI 0.6–15.8%). The first amplicon was confirmed as L. (L.) infantum and the second as Leishmania subgenus Viannia, making both the first records of infection by Leishmania species in wild mammals in this region [31].
In a third D. albiventris sample, a 500 bp product was confirmed as Trypanosoma rangeli by genetic sequencing (98% similarity; E value = 0.0, GenBank accession number AY230237.1 and others). No sample was positive when using the primers Lch14/15.
The presence of natural infection by L. (L.) infantum in female sand flies and natural infection by two Leishmania species in a mammal species considered to be a potential reservoir (D. albiventris) provides strong evidence that sylvatic Leishmania transmission cycle is occurring in the EPA. However, it was not possible to confirm whether VL in Campinas occurred due to a wild enzootic cycle of transmission. Since dogs cohabit with infected wild animals, the hypothesis of the infection of wild mammals by infected dogs cannot be ruled out.
In the investigation conducted into the first case of CVL in the EPA in 2009, 40 wild mammals belonging to the species Nectomys squamipes, D. albiventris, C. penicillata and Gracilianus agilis were captured in wooded areas of the EPA and samples were examined by PCR, but none presented positive for VL [17]. In our investigation, infection among wild fauna by different species of Leishmania [31] was confirmed for the first time in this region, together with another species of trypanosomatid, T. rangeli. In addition, at the time of the first investigation, proximity of 100 m between human residences and forest areas where sand flies and wild mammals were captured was demonstrated [17].
In our study, although the capture of wild animals and phlebotominae was conducted in forest fragments in several locations within the EPA, the occurrence of L. (L.) infantum in these animals was observed only in locations close to seropositive domestic dogs. This spatial distribution suggests the involvement of these animals in the CVL focus, even though low Lu. longipalpis density and a small number of infected wild animals was observed.
Some authors suggest that the public health impact of VL infected dogs in urban areas is greater than in rural and wild regions, and that attention should be given to individuals living or frequenting these locations. Restricted contact between wildlife and domestic dogs has also been proposed to reduce the probability of VL transmission to humans and wild animals from dogs [32,33].
Discussion concerning the participation of wild species in the transmission of zoonotic parasites has become particularly important in recent years with the consolidation of the One Health concept [34,35]. Anthropogenic changes in ecosystems are particularly important in this context, resulting in greater proximity between wild and domestic animals and humans [36].
Proximity to forest areas and wildlife with human residences is a worrying scenario. In the state of Rio de Janeiro, dogs living within 100 m of forests presented a 3.5-fold higher risk of acquiring VL. The presence of opossums in the peridomicile increased the chances of infection in dogs 2.6-fold, with a 30.0% prevalence in these wild animals [37].
Another study detected a 20% prevalence of VL in dogs from rural areas in the state of Minas Gerais, located up to 2 Km from five EPAs that contained fragments of Atlantic Forest. The risk factors for these dogs were different from those living in urban areas [33].
In the Chaco region of Argentina, Barroso et al. [38] surveyed 77 dogs living in a forest area where two cases of human VL had occurred and determined a 13% prevalence. The authors suggested that the emergence of cases in dogs and humans in a wild region that was endemic for tegumentary leishmaniasis, with a relatively low canine prevalence, is clearly compatible with the involvement of wild mammals as reservoirs and that parasite transmission occurs from these animals. This scenario is very similar to the CVL focus located in the Campinas EPA.
In the EPA, the presence of 15 phlebotomine sand flies species, even at low frequencies, shows a diversity of fauna, as previously reported in forest environments [39,40].
One interesting factor is the low density of Lu. longipalpis in the study area. Based on analysis of sexual pheromones secreted by males, it is accepted that Lu. longipalpis is a species complex [41]. The chemotype population of Lu. longipalpis found in Campinas, cembrene-1, is different to the chemotype population found in the western region of the state of São Paulo, (S)-9-methylgermacrene-B [42]. The remarkable differences between the epidemiological situations, population size and sibling complex of Lu. longipalpis in Campinas, corroborates the findings of Casanova et al. [43], who suggested there are different vectorial capacities and competence between siblings.
In this study, two species of phlebotomine sand flies, Pi. monticola and Ex. firmatoi, were found naturally infected with L. (L.) infantum. These species are essentially sylvatic and highly anthropophilic [44,45]. The participation of these species in the transmission of L. (V.) braziliensis and L. (L.) infantum has been suggested in the municipalities of Divinópolis, Minas Gerais, and in Rio de Janeiro, respectively, with Pi. monticola and Ex. firmatoi [46,47].
Females of different species were also found naturally infected by L. (L.) infantum in other endemic areas, such as specimens belonging to the cortelezzii complex in the Brazilian States of Minas Gerais and Mato Grosso do Sul—which are not amongst the incriminated leishmaniasis vector—and may be involved in a wild or rural cycle of L. (L.) infantum transmission in these areas [48,49]. Mg. migonei–a known vector of cutaneous leishmaniasis—was also found naturally infected in the Brazilian States of Pernambuco and Ceará. Some authors suggested that this species could act as a potential vector in VL transmission, particularly in areas where Lu. longipalpis is absent [50,51]. It is important to note that recent studies have demonstrated the high susceptibility of Mg. migonei to infection with L. (L.) infantum, reinforcing this hypothesis [52,53].
The detection of Leishmania DNA in a sand fly species does not prove vector competence [54]. We cannot exclude the persistence of DNA without any infective role, particularly because there are infected dogs and wild animals in the vicinity of these infected sand flies [55]. However, these results reinforce the need for further studies to investigate the vectorial capacity of these species, especially experimental studies.
Although canine VL cases usually precede the occurrence of the disease in humans, no cases of human VL have been recorded in Campinas city to date. This fact could be associated with the local transmission characteristics related to the diversity and competence of the vectors, the presence of wild reservoirs and hosts, the characteristics of the parasite (not investigated), as well as the pattern of human contact, with transmission in sparsely populated area and in luxury condominiums.
The hypothesis that in some new VL foci in the state of São Paulo, subpopulations of Lu. longipalpis could be present and account for the existence and perpetuation of L. (L.) infantum enzootic wild cycles should be considered, particularly in areas around preserved forest environments or in recent residential projects.
On the other hand, Motoie et al. [56] identified the presence of genetically distinct populations of L. (L.) infantum in São Paulo, indicating that the inherent characteristics of the parasite itself could also be responsible for different epidemiological patterns of VL observed in new foci in the state.
In addition to factors directly associated with the transmission cycle, the occurrence of CVL cases in the Campinas EPA seems to be related to the occupation process, in which anthropic activities changed the natural landscapes, which has forest fragments containing wild fauna. In these regions, the exposure of humans and their pets to parasites and vectors of diseases to which they have not been previously exposed can result in disease outbreaks from natural enzootic foci.
The process of urban expansion within the EPA in the 1970s and 1980s was associated with an outbreak of cutaneous leishmaniasis in human inhabitants of the EPA in 1993 and 1994 [20]. In our study, an opossum was infected with a species of the subgenus Viannia, known to be responsible for tegumentary leishmaniasis in Brazil, which confirms the current risk of infection in this area and, again, the possible involvement of the wild fauna in the maintenance of the transmission cycle.
In addition, environmental modifications seem to be related to the adaptation of phlebotomine vectors to urban environments due to a decrease in the availability of wild animals as a food source, making dogs and humans more accessible alternatives to the vector [21,57].
The VL Surveillance and Control Program in Brazil supervises the main actions to reduce morbidity and lethality, aimed at early diagnosis and treatment of human cases, vector control and identification and elimination of seropositive domestic dogs [57]. Although there is good theoretical support for these measures, there is no evidence of their effectiveness in reducing the prevalence in endemic regions nor in the expansion of new outbreaks [58–60].
In Europe, although the domestic dog is considered the main reservoir in endemic areas, wild reservoirs have been proposed as potentially responsible for the lack of success in VL control [61]. In Brazil, the zoonotic transmission cycle is concentrated in areas where human habitation is located close to the wild cycle of the disease [2].
Although serological tests may be subject to false positive and/or negative results, due to its sensitivity and specificity, their use is valued for epidemiological surveillance purposes [22, 53].
Despite the serological tests used in this study, Grimaldi Jr. et al. [62]reported that DPP displayed high specificity (96%) but low sensitivity (47%) in identify asymptomatic dogs, but the sensitivity was significantly higher (98%) in diseased cases, indicating that this convenient test may be useful to identify the most infectious dogs.
In another study, Laurenti et al. [63] report that DPP detected both asymptomatic and symptomatic dogs in equal proportions. Using this test, 42/47 (89.4%) symptomatic dogs were detected positive, besides 35/38 (92.1%) asymptomatic dogs, with good accuracy of 92.7%. The ELISA BioManguinhos was positive in 43/47 (91.5%) symptomatic dogs and 34/38 (89.5%) asymptomatic dogs, with accuracy of 84.3%. Although, the combination of the two tests, as used in our study and in Brazilian surveillance activities, results in 99.1% sensitivity and 73.9% specificity, showing that it is a useful serology protocol for surveys.
In our study, the proximity of seropositive dogs housing sites with capture sites of wild mammals and sandflies infected with Leishmania spp. reinforces the assumption that they may be related. However, as shown in Fig 1, the human residences were constructed in the middle of the vegetation, being difficult to delimit the occurrence of independent transmission cycles in peridomiciliar or wild environment.
The need to elaborate distinct control and prophylaxis strategies, according to the characteristics of each transmission area, including the rural and wild areas in the control programs is now becoming evident. Studies that clarify the patterns of infection in different locations are useful for the success of such actions. Thus, investigating a new VL focus in all its distinct aspects contributes to our understanding of the key elements of the transmission dynamics and disease control. Actions that do not consider these particularities are less likely to succeed.
|
10.1371/journal.pgen.1007681 | Coronary artery disease genes SMAD3 and TCF21 promote opposing interactive genetic programs that regulate smooth muscle cell differentiation and disease risk | Although numerous genetic loci have been associated with coronary artery disease (CAD) with genome wide association studies, efforts are needed to identify the causal genes in these loci and link them into fundamental signaling pathways. Recent studies have investigated the disease mechanism of CAD associated gene SMAD3, a central transcription factor (TF) in the TGFβ pathway, investigating its role in smooth muscle biology. In vitro studies in human coronary artery smooth muscle cells (HCASMC) revealed that SMAD3 modulates cellular phenotype, promoting expression of differentiation marker genes while inhibiting proliferation. RNA sequencing and chromatin immunoprecipitation sequencing studies in HCASMC identified downstream genes that reside in pathways which mediate vascular development and atherosclerosis processes in this cell type. HCASMC phenotype, and gene expression patterns promoted by SMAD3 were noted to have opposing direction of effect compared to another CAD associated TF, TCF21. At sites of SMAD3 and TCF21 colocalization on DNA, SMAD3 binding was inversely correlated with TCF21 binding, due in part to TCF21 locally blocking chromatin accessibility at the SMAD3 binding site. Further, TCF21 was able to directly inhibit SMAD3 activation of gene expression in transfection reporter gene studies. In contrast to TCF21 which is protective toward CAD, SMAD3 expression in HCASMC was shown to be directly correlated with disease risk. We propose that the pro-differentiation action of SMAD3 inhibits dedifferentiation that is required for HCASMC to expand and stabilize disease plaque as they respond to vascular stresses, counteracting the protective dedifferentiating activity of TCF21 and promoting disease risk.
| Coronary artery disease (CAD) is the worldwide leading cause of death. The majority of risk for CAD is genetic in nature, i.e., a feature of the genetic information that is transmitted to each individual from both parents, and primarily affects the disease processes in the blood vessel wall that regulate the disease molecular pathways. Modern genetic approaches have allowed mapping of the regions of the human genome that encode information that mediates this risk. The SMAD3 gene has been identified through these studies, a known master regulatory of other genes and molecular pathways, and we have investigated the functions of this gene that are important for disease risk. SMAD3 affects basic functions of a cellular component of the vessel wall, the smooth muscle cell (SMC), that is responsible for responding to vascular stresses to heal the lesions that are produced in conjunction with elevated lipids and other classic risk factors. Studies reported here show that SMAD3 actually inhibits the cellular processes that allow SMC to repair the vascular lesions, and its expression is promoted by the disease related variable sequences in the disease associated regions of the genome. SMAD3 is opposed by another CAD gene, TCF21, that functions to block the effects of SMAD3 expression, and these studies identify genetic mechanisms by which this is done. Thus, these studies identify an interactive pathway that directly contributes to disease risk, and the ability to block SMAD3 or promote TCF21 function could be exploited to inhibit vascular events such as myocardial infarction.
| Coronary artery disease (CAD) is the worldwide leading cause of death [1]. Numerous genetic loci have been associated with CAD with genome-wide association studies (GWAS) [2–8] and point to common inherited variation that mediates the genetic risk for this disease. Unfortunately, a majority of the identified causal variation resides outside of protein coding exons, in regulatory regions of the genome that are poorly understood [9], and further efforts are required to understand the mechanisms of disease association. Thus, ongoing efforts are required to identify causal genes in these loci and link them to fundamental signaling pathways that may be targeted for therapeutic benefit.
One molecular pathway that appears to be highly represented among the causal genes identified in CAD loci is constituted with members of the TGFβ superfamily [10]. Also, our unbiased genome-wide studies of chromosomal accessibility and epigenome mapping in human coronary artery smooth muscle cells (HCASMC) have identified a significant enrichment for CAD loci in chromosomal regions affected by TGFβ signaling in this cell type [11]. TGFβ signaling controls a diverse set of cellular processes, including proliferation, cell-cell recognition, differentiation, and specification of developmental fate, during embryogenesis as well as in mature tissues [12–18]. One TGFβ family member, SMAD3, has been experimentally linked to atherosclerosis [19, 20], recently associated with CAD, and implicated with genomic and functional studies as the causal gene at 15q22.33 by this and another laboratory [11, 21]. These studies implicate this critical component of the canonical TGFβ pathway in CAD, although the disease relevant cell type and mechanism of effect remain unclear.
There are considerable data that link SMAD3 to developmental and disease processes in the smooth muscle cell (SMC) component of the coronary circulation. In embryonic development, TGFβ signaling plays a key role in SMC differentiation from the mesoderm and the intimately related process of vascular wall development [22–25]. Importantly, TGFβ mediates epithelial mesenchymal transition (EMT) in development of the coronary circulation, promoting epicardial cell migration into the myocardium and formation of coronary artery smooth muscle cells [26–28]. Data also suggests that SMAD3 regulates fundamental SMC processes that are relevant to vascular disease risk. SMAD3 mutations have been linked to the syndromic disease Aneurysms Osteoarthritis Syndrome that is characterized by large vessel aneurysms that primarily result from loss of SMC and related structural matrix components [29]. Relevant to atherosclerosis and CAD, SMAD3 has been shown to directly bind the SMC lineage determining transcription factor myocardin (MYOCD) to regulate transcription of differentiation factors [30] and linked to differentiative and anti-proliferative effects in a number of smooth muscle cell models [23–25, 31]. Interestingly, these actions appear to directly oppose those of another CAD associated gene TCF21. TCF21 promotes dedifferentiation, proliferation and migration of HCASMC, broadly promoting phenotypic switching, an epigenetic process that is hypothesized to reduce disease risk [32–34]. Taken together, these findings suggest that cell state changes in HCASMC are a critical aspect of CAD pathophysiology, that dedifferentiation is a critical protective function against vascular destabilization, and predict that SMAD3 and TCF21 have opposing functional roles in regulating the phenotype of HCASMC.
In studies reported here, we have focused on SMAD3 as a key factor in the pathophysiology of CAD, because of its localization in a CAD associated locus [4, 35], its central role as an effector in the TGFβ pathway [36–40], and its function as a transcriptional regulator [30, 39, 41, 42]. In vitro functional studies along with RNAseq and ChIPseq analyses establish that SMAD3 elicits a pro-differentiation phenotype in HCASMC, opposing the functions of the CAD associated factor TCF21 to promote CAD risk [32–34].
Given the role of TGFβ in differentiation of epicardial precursor cells to the coronary smooth muscle cell lineage during embryogenesis [26–28], we specifically investigated the expression of HCASMC marker genes in SMAD3 siRNA knockdown experiments. Comparison of HCASMC transfected with a specific SMAD3 siRNA or an RNA with scrambled sequence showed that a significant decrease in SMAD3 mRNA levels (1.0 vs. 0.30, p<0.001) and a 65% decrease in protein expression by quantitative western analysis (Fig 1A). This was associated with decreased mRNA levels for SMC markers ACTA2 (1.0 vs. 0.4, p<0.001,) TAGLN (1.0 vs. 0.43, p<0.01), and CNN1 (1.0 vs 0.27, p<0.001) as well as decreased protein levels for ACTA2 and TAGLN (Fig 1C, 1D and 1E, S1A Fig). Similar experiments were performed with over-expression of SMAD3 in HCASMC, achieved by transfection of these cells with a SMAD3 expression construct, as assessed by mRNA levels (1.0 vs. 223.3, p<0.0001) with an average 11-fold increase in protein expression as determine by quantitative western analysis (Fig 1B). Confirming an opposite effect on ACTA2, TAGLN, and CNN1 expression, increases were observed in mRNA levels for these two genes (1.0 vs. 2.53, p<0.01, 1.0 vs. 1.50, p<0.01, and 1.0 vs. 2.24, p<0.05 respectively), as well as increased protein levels (Fig 1C, 1D and 1E, S1A Fig). In a separate type of assay, fluorescence was quantified for HCASMC expressing either increased or decreased SMAD3. These studies also indicated that decreased levels of SMAD3 were associated with decreased expression of ACTA2 (17.7, vs. 4.3, p,0.001), TAGLN (28 vs. 7, p<0.001), and CNN1 (14 vs. 7.3, p<0.05) while increased SMAD3 promoted increased expression of ACTA2 (15.0 vs 23.7, p<0.01), TAGLN (12.7 vs 29.7, p<0.01) and CNN1 (6.67 vs, 16.7, p<0.01) marker genes (Fig 1F, 1G and 1H, S1B Fig and S1C Fig).
The effect of SMAD3 on cellular migration and proliferation of HCASMC was also investigated. Employing a wound closure assay of migration, SMAD3 siRNA knockdown significantly decreased the surface area covered by the cells after 24 hours of incubation (60 vs. 36.67 for control cells, p<0.01) (Fig 1I, S1D Fig). Over-expression of SMAD3 produced a significant increase in migratory activity (49 vs. 95.8 control cells, p<0.01). Although seemingly inconsistent with the differentiation effect in HCASMC, SMAD3 has been previously shown to promote migration in culture model systems [43]. The ability of SMAD3 to regulate cell cycle in the cultured HCASMC was evaluated with a high sensitivity EdU assay. Knockdown of SMAD3 did not show an effect while over-expression produced a significant decrease in cellular proliferation (15.67 vs. 4.67 control cells, p<0.001) (Fig 1J, S1E Fig).
To gain insights into the role of SMAD3 expression in HCASMC, and support the in vitro functional assays, we performed genome-wide transcriptomic studies, as recently described [44]. We employed RNA sequencing (RNAseq) on cells transfected with either non-silencing scrambled control (SCR) or small interfering SMAD3 (siSMAD3) RNAs to characterize genes and pathways that are regulated by this transcription factor in HCASMC. We identified 493 differentially expressed (DE) genes (FDR ≤ 0.05), as assessed with the DESeq algorithm [45]. Investigation of these DE genes with the Ingenuity Pathway Analysis (IPA) software (Qiagen) identified several overlapping canonical pathways, including “regulation of epithelial mesenchymal transition” (p = 1.15e-05), “axonal guidance signaling” (p = 1.17e-05), and “semaphorin signaling” (p = 1.36e-05) (Table 1). The highest degree of association with causal networks was identified for those regulated by the CAD associated CXCR4 cytokine (p = 3.32e-15) [6, 7], the chromatin regulating KAT2B (p300) lysine acetyltransferase (p = 3.55e-15), and the CAD associated soluble guanylate cyclase signaling molecule GUCY (p = 1.24e-14) [7]. IPA identified a high degree of enrichment for SMAD3 knockdown DE genes among those associated with “cardiovascular disease” (p = 1.62e-4–8.37e-9) (Table 1, S1 Table), including those associated with vascular developmental syndromes such as “abnormal morphology of vasculature,” “abnormal morphology of blood vessels”, “abnormal morphology of artery,” as well as those associated with atherosclerotic vascular disease, “occlusion of blood vessel”, and “atherosclerosis”. These functional categories were enriched for genes in the endothelin signaling pathway, including CAD associated gene EDNRA, as well as EDN1, EDNRB, and ECE, smooth muscle cell differentiation factor MYOCD and vascular development factor ANGPT1 (Fig 2).
In the highly relevant Physiological System Development and Function analysis in IPA, the top three terms identified for the SMAD3 knockdown DE genes were identical to those identified in the similar analysis conducted with the CAD GWAS associated genes identified in a recent meta-analysis [7], “cardiovascular system development and function, “organismal development” and “organismal survival.” (S2 Table). While these category terms are quite broad, they suggest significant overlap between SMAD3 regulated genes and those identified in CAD associated loci. To rigorously test the significance of the overlap between the two gene lists, we employed the full GWAS catalog gene list as background and obtained a p-value of 0.000106 when employing the Fisher exact test. Cardiovascular system category terms (p = 1.12e-18) included “development of vasculature,” “angiogenesis,” and “vasculogenesis” (Table 1, S3 Table). Prominent genes in these categories included SPRY4, FGF1, and HGF, as well as various semaphorin and ephrin factors known to have roles in neuronal and vascular development [46–50]. Also, of interest were terms related to smooth muscle functions including “mean arterial pressure,” “proliferation of vascular smooth muscle cells”, and “contraction of blood vessel.” Top molecular and cellular functions included “cell movement” (p = 1.67e-04–8.21e-17), “gene expression” (p = 5.87e-07–2.82e-14), and “cell death and survival” (1.19e-04–5.56e-12) (Table 1, S4 Table).
Taking the 89 genes in the functional subcategory “vasculature development” in the Cardiovascular Development and Function category, we used well-curated molecular interactions in the IPA Knowledge Base to build a gene network (Fig 2, S5 Table). This subcategory was felt to be particularly pertinent to the molecular and cellular basis of CAD, based on the highly significant enrichment for such terms in the IPA analysis of the CAD GWAS meta-analysis genes [7], and the close match to terms in this category found here with SMAD3 knockdown transcriptomic analysis. The interactions of the network DE genes were visualized with Cytoscape to highlight the mechanism of interaction between nodes and the various features of the relationship of each gene to the network as a whole (Fig 2). This network highlights the interaction of SMAD3 with highly connected node genes that reflect integration of SMAD3 with important vascular functions, including the endothelin pathway (EDN1, EDNRA, EDNRB, ECE), the HGF-MET signaling axis, CXCL8 inflammatory pathway, and RBPJ component of Notch signaling.
Interestingly, many of the key vascular developmental genes were noted to also be regulated by TCF21 [34, 51], but in the opposite direction, including EDNRA, EMP2, FGF7, ID1, ID3, IL24, ITGB8, SEMA3A, SEMA5A, etc. Also, a number of matrix genes were differentially regulated by SMAD3 and TCF21 in opposite directions, including COL1A1, COL1A2, COL3A1, COL5A2, THBS1, and ITGAV among others. Some of these differences were visualized by mapping the SMAD3 differential gene expression values onto the TCF21 “cardiovascular disease” transcriptional network (S2 Fig) [34]. This comparison also identified some genes that are regulated in the same direction, including some matrix genes (MMP2, MMP3, FBN1) and vascular development genes (SEMA3D, NRP1, ANGPT1). A number of the TIMP and MMP genes that were differentially regulated by SMAD3 and TCF21, and of interest in vascular disease processes, were further investigated by qRT-PCR. These studies documented differential regulation of TIMP1, TIMP3 and MMP10 by SMAD3 expression (S3 Fig).
To identify genes that are directly regulated by SMAD3, and to link this CAD gene with other genes that are associated with CAD, we performed chromatin immunoprecipitation sequencing (ChIPseq) with cultured HCASMC. SMAD3 ChIPseq identified 30,292 total binding sites in the HCASMC genome. The ChIPseq findings in HCASMC were validated by ChIP-PCR of representative well documented SMAD3 target loci, TAGLN, CNN1, COL1A1, and SERPINE1, verifying SMAD3 binding that was increased with TGFβ1 stimulation (S4A Fig). For comparison to the HCASMC data, we used the same analysis pipeline and analyzed the published ChIPseq data obtained for the A549 lung cancer cell line [52]. Interestingly, intersection of the 24,587 A549 and HCASMC peaks, requiring overlap of at least one basepair identified sharing of only 1076 of the HCASMC peaks and 1393 of the A549 peaks. Increasing the binding site by 1 kb on either side of the peak only increased the number of overlapping HCASMC SMAD3 and A549 SMAD3 peaks to 1826 and 2303 respectively. Further, use of the HOMER “de novo” algorithm identified several related SMAD3 motifs in the HCASMC data which were specific variants of the three well characterized binding sequences (Fig 3A) [52]. Taken together these analyses suggest that SMAD3 binds a unique repertoire of regions in the HCASMC genome, consistent with the unique integral role of TGFβ in the transcriptional regulation of the phenotype of this cellular lineage [30, 53, 54].
SMAD3 peaks were assigned to genes with the Genomic Regions Enrichment of Annotations Tool (GREAT) [55], and this collection of target genes were employed in gene ontology analysis using DAVID (Fig 3B, S4B Fig). Terms identified by Biological Process analysis included highly relevant significant terms including “vasculogenesis”, “transcription from RNA pol II promoter”, and “regulation of cell differentiation.” Significant KEGG pathways included “Wnt signaling,” “vascular smooth muscle contraction,” and “TGFβ signaling pathway” terms. Disease enrichment terms included a number of significantly relevant terms highlighting target gene association with CAD, including “myocardial infarction,” coronary disease,” and “coronary artery disease.” Molecular function pathways were largely related to transcriptional regulation and SMAD-DNA binding (S4B Fig).
Given the marked functional dichotomy between SMAD3 and TCF21, which promotes a de-differentiation program in HCASMC [34], we were interested to investigate possible interactions that might reflect coordinated regulation of HCASMC phenotype, and thus disease risk. First, we intersected the SMAD3 ChIPseq and our standard TCF21 HCASMC ChIPseq datasets (see Methods) [56] to identify regions of the genome that mediate binding of both transcription factors. Scanning the overlapped SMAD3 peaks for known binding motifs with the HOMER “known” algorithm, we found the CAGCTG TCF21 binding sequence to be the most highly ranked (p = 1e-12), suggesting that TCF21 binds in proximity to SMAD3 in a significant number of colocalizing loci (Fig 3C). Also, the binding sequence for TCF12, a TCF21 obligate heterodimer was also enriched in these SMAD3 peaks (p = 1e-9). Further, we investigated characteristics of the genomic overlap patterns of the two ChIPseq datasets to better understand how they might be clustered on a whole genome level. First, we investigated the possible overlap of binding at a gene level. For this analysis, we intersected a restricted set of data for both SMAD3 (fold change>5, -logQ>10) and TCF21 (fold change>15, -logQ>200), assigning genes to peaks with GREAT. With this approach, we found 4,647 genes were assigned to the SMAD3 peaks, with 3,143 or 68% of these genes also identified among the 8632 total genes assigned to TCF21 peaks (Fig 3D).
Further, by intersecting the full SMAD3 dataset and the standard TCF21 dataset, 979 of 30,392 SMAD3 peaks were found to overlap 935 of 29,533 TCF21 ChIPseq peaks (Fig 3E). The apparent disparity between the results from these two analyses (Fig 3D and 3E) was due to the fact that although the peaks reside in the same loci, and therefore assigned to the same genes, they are distant enough to not share basepairs except for 979 SMAD3 peaks and 935 TCF21 peaks. We further investigated the colocalization of SMAD3 and TCF21 binding in these ~1000 loci with density plots for SMAD3, TCF21, and the negative control TF HNF1A, along with HCASMC ATACseq regions of open chromatin in HCASMC (Fig 3F). This analysis revealed an overlap of binding sites for SMAD3 and TCF21 in ATACseq open chromatin regions, compared to HNF1A. Further, we investigated the relationship between SMAD3 and TCF21 binding and epigenetic features in restricted areas of colocalization for loci encoding genes of interest in disease pathophysiology (Fig 3G and 3H). In loci encoding the SERPINE1 and COL1A1 genes, we identified overlap of SMAD3 and TCF21 peaks, in regions of HCASMC open chromatin as identified with ATACseq. Further, there was colocalization with the H3K27Ac histone mark as identified in HCASMC and in ENCODE samples. For these two genes, there was overlap of HCASMC SMAD3 peaks with some of those identified by ChIPseq in A549 epidermal lung cancer cells, although there were also significant differences in the binding patterns.
Finally, to investigate the biological pathways represented by the genes in those loci where SMAD3 and TCF21 co-localize, we assigned genes to this set of peaks with GREAT and performed GO analysis. This analysis identified terms quite similar to those found for the SMAD3 ChIPseq dataset alone (Fig 3B, S4C Fig). However, the overall number of genes identified for each term was smaller and the p-values were in general smaller than the results with the full SMAD3 targeted gene list. For example, in the KEGG pathways, there was a significantly smaller p-value for “pathways in cancer” and “Hippo signaling,” and in the Biological Processes category for “cell migration” and “transcription from RNA pol II promoter.” Surprisingly, in the Disease Enrichment category CAD related terms “myocardial infarction” p-value were less significant while “coronary artery disease” and other related term p-values were unchanged.
Given the opposite effects of SMAD3 and the CAD associated transcription factor TCF21 on the differentiation state of HCASMC [34], and evidence that these two transcription factors colocalize in a number of shared binding regions of the genome where they may regulate the same genes, we were interested to determine if there is a functional relationship in shared loci. To investigate binding patterns suggestive of interactions between SMAD3 and TCF21 at the level of protein binding to DNA, we analyzed the relative binding of each factor by normalizing the number of reads in the peaks to background counts in the region (see Methods). We focused on those binding sites which showed a greater than two-fold difference in normalized read counts, as a measure of relative binding (Fig 4A and 4B). For the directly overlapping ChIPseq binding sites for SMAD3 and TCF21, more than half showed a 2-fold discrepancy in binding. Out of 583 biased binding sites, 358 (358/583, 61%) showed higher relative binding of TCF21 and 225 (225/583, 39%) showed higher relative binding for SMAD3 (Fig 4A). This pattern of SMAD3 compared to TCF21 binding at shared loci is quite distinct from the expected pattern for factors that bind with equal affinity, as shown for JUN and JUND binding, where only a small percentage of the binding was at sites with a greater than 2-fold difference (Fig 4B). These findings suggested that one of these TFs might inhibit binding of the other at these shared loci, a mutual inhibitory interaction, or opposing responses to external signaling pathways.
To investigate the possibility that TCF21 inhibits SMAD3 binding at these loci, we performed ChIP-PCR at two representative loci, encoding the SERPINE1 and COL1A1 genes. PCR primers were designed to amplify regions where both SMAD3 and TCF21 bind in areas of open chromatin and active histone mark configuration (Fig 3G and 3H). As a control, ChIP-PCR was also conducted for the ACTB gene at a SMAD3 binding site where there is no evidence of TCF21 binding. HCASMC treated with scrambled, control, siRNA showed significant enrichment for SMAD3 binding as expected at loci identified by ChIPseq studies. SMAD3 binding increased significantly at the SERPINE1 (0.018 vs 0.084, p<0.01) and Col1A1 loci (0.014 vs 0.114, p<0.05) loci with effective knockdown of TCF21 expression (Fig 4C). Employing the assay for transposase-accessible chromatin coupled with quantitative PCR (ATAC-PCR), we further investigated whether TCF21 modulation of SMAD3 binding might be associated with changes in local chromatin accessibility. The ATAC-PCR studies showed that knockdown of TCF21 produced significantly increased accessibility at the SMAD3 binding site in both SERPINE1 (6.67 vs. 8.02, p<0.05) and COL1A1 (8.40 vs 12.41, p<0.01) (Fig 4D). These data are consistent with TCF21 indirectly inhibiting SMAD3 binding in regions of colocalization through an epigenetic mechanism affecting chromatin accessibility.
Finally, to investigate possible direct interactions between SMAD3 and TCF21 regulation of gene expression, we performed reporter gene transfection studies with an enhancer region in the SERPINE1 gene where putative binding sites for these TFs are separated by 78 basepairs (Fig 3G). This region ~9kb upstream of the transcription start site of the human SERPINE1 gene was cloned into a reporter plasmid, and co-transfected into a human umbilical vein smooth muscle cell line (HUVSMC) and primary cultured HCASMC, with expression plasmids for SMAD3 and/or TCF21 (Fig 4E and 4F). The enhancer alone increased expression of the basal reporter, and transcription was further increased with SMAD3 over-expression. TCF21 transfection significantly inhibited expression of the basal enhancer, and mitigated the increase that was seen with SMAD3 over-expression in both HUVSMC (3.27 vs 1.30, p<0.01) and HCASMC (2.00 vs 1.46, p<0.01). Thus, at this enhancer where SMAD3 and TCF21 binding colocalize, TCF21 can significantly inhibit the transcriptional effect of SMAD3. Since the reporter constructs were not integrated into the chromatin where epigenetic modification by TCF21 could affect SMAD3 function, these data suggest a separate mechanism, with these TFs producing independent opposing effects on the basal transcription apparatus, or possibly TCF21 blocking SMAD3 binding through direct protein-protein interaction.
While there has been much interest and study of the role of the TGFβ pathway in vascular disease pathophysiology, there remains much debate regarding the direction of effect for the pathway and specifically SMAD3 [19, 20, 57, 58]. Limited public expression quantitative trait loci (eQTL) datasets have suggested that SMAD3 expression is associated with disease risk [11, 21]. We have extended these studies to investigate the causality and directionality of the SMAD3 gene in CAD with three additional strategies. First, we assessed gene expression at the putative causal variant rs17293632 with disease relevant eQTL data from human coronary artery smooth muscle cells [59], and found that the risk C allele increases expression of SMAD3 (p<0.05), indicating SMAD3 is a CAD promoting transcription factor and suggesting that this gene is active in this cell type in vivo (Fig 5A).
In a second approach employing the HCASMC eQTL data, we assumed an additive model of CAD risk for alleles located in the SMAD3 locus. For this analysis, we selected CAD associated SNPs from the latest CARDIOGRAM+C4D meta-analysis that were associated with CAD at a p-value cutoff of 1.0e-6 and that were located within 100kb away from the start and end of the composite SMAD3 gene [7]. To improve the correlation in the local regions of strong linkage disequilibrium we used an algorithm (see Methods) to resolve the local haplotype structure in each HCASMC sample and average expression on those samples that possess identical haplotype profiles in the tested region. This reduced the variability by eliminating the component of the variance that arises from inter-individual variability and technical issues. By employing this approach, the regression coefficient was 0.45, with a nominally significant p-value of 0.08, suggesting that SMAD3 expression increases with a greater number of cis-acting risk alleles as per an additive model (Fig 5B).
Finally, we have investigated the correlation between allelic risk and gene expression in mammary artery (MAM) tissue samples from individuals with coronary artery disease in the Stockholm-Tartu Atherosclerosis Reverse Networks Engineering Task study (STARNET) [60] (Fig 5C). We observed a strong positive correlation (r = 0.894; p = 3.76e-09) for the effect sizes (log OR) of CAD associated variants (p<1e-04) with those also associated with SMAD3 gene expression (cis-eQTLs) in STARNET. Notably we did not observe a similar trend in atherosclerotic aortic tissues, in which TCF21 was identified as a strong cis-eQTL gene. These findings leveraging natural genetic variation from GWAS and SMC-enriched artery tissues from CAD individuals further implicate SMAD3 as a pro-atherosclerotic gene, consistent with findings in HCASMC.
Having recently established the likely causality of SMAD3 for CAD [11], the overall goal of the work reported here was to determine the mechanism and direction of effect for this gene, and thus integrate it into a causal framework that regulates disease pathophysiology. We have demonstrated with various genetic and genomic approaches that a significant portion of the genetic risk for CAD resides in the smooth muscle cell lineage and that the TGFβ pathway in particular regulates genomic features in disease loci, and have thus focused here on HCASMC [11, 59]. These studies confirm previous work in other SMC models that SMAD3 promotes a differentiation program in these cells, as evidenced by upregulation of lineage markers [22, 61, 62]. Further, we have shown that SMAD3 binds in regions of the genome that also harbor TCF21 binding sites, and provide several lines of evidence that point to antagonistic effects between DNA binding and transcriptional action of these two TFs, primarily for TCF21 inhibition of SMAD3. These data suggest that TCF21 may mediate this effect through modifying the epigenome and also possibly directly inhibiting SMAD3 binding to DNA. Finally, we have provided the most compelling evidence to date, employing eQTL data from HCASMC and vascular tissues, that expression of SMAD3 is directly causal for CAD. We propose that the pro-differentiation function of SMAD3 inhibits HCASMC dedifferentiation, phenotypic modulation, of these cells as they respond to vascular stress and expand to stabilize the plaque, and that SMAD3 function is directly opposed at the transcriptional level by the disease protective expression of TCF21, which promotes dedifferentiation and phenotypic modulation.
Our results showing SMAD3 promotion of HCASMC marker expression and proliferation are consistent with previous studies investigating TGFβ signaling in other types of SMCs [24, 61], and are consistent with a pro-differentiation paradigm. Although previous in vitro studies have shown a migratory effect of TGFβ signaling [43], our data showing that SMAD3 can promote HCASMC migration would seem at odds with the marker gene and proliferation effects, and raise the question of whether SMAD3 is truly promoting differentiation of HCASMC or merely acting as a transcriptional regulator in these experiments. At a molecular level, it is known that SMAD3 binds the SMC transcription factor myocardin to promote this transcriptional program [30]. Myocardin is widely regarded as a lineage determining factor, fundamentally specifying the SMC fate, and thus serving as more than a transcription factor that activates contractile marker expression [63]. Also, it is well known that TGFβ signaling has a role in embryonic vascular development, specifically promoting induction of SMC markers in mesodermal cells that become phenotypic SMC, and also promoting their migration to the forming vascular structure [22]. Thus, there is precedent for SMAD3 signaling jointly promoting SMC differentiation and migration. Further, for these studies, HCASMC were evaluated in media containing a number of growth factors and cytokines found in serum and growth additives. Whichever factors promote migration in this setting may not be functional in vivo in the disease setting. Migratory activity would seem to be under the control of a number of factors with competing programs. In the context of SMAD3 effects on disease risk, pro-differentiation and inhibited cell division are at odds with the protective effects of TCF21 in the vessel wall and thus likely disease promoting, while increased migration is also promoted by TCF21 and likely protective. Fortunately, this conundrum is resolved by human genetic data that clearly indicate that expression of SMAD3 increases risk for human CAD, possibly by providing a stimulus for SMC to remain differentiated and oppose phenotypic modulation that appears to be protective.
The TGFβ pathway has been linked to a number of vascular diseases. These include syndromic diseases associated with aortic aneurysms, including Marfan’s and Loeys Dietz Syndromes (LDS), due to mutations in TGFβ signaling genes FBN1, TGFBR1, TGFBR2, TGFB2, and TGFB3 [64, 65]. Also, SMAD3 mutations have been linked to the syndromic disease Aneurysms Osteoarthritis Syndrome [29]. In general, these aortopathies are characterized by large vessel aneurysms that primarily result from loss of SMC and related structural matrix components [29]. A paradox in the field is that disease mutations that should be amorphic, including SMAD3 mutations associated with aortopathy, actually produce increased TGFβ signaling as the mechanism of effect [65]. The relationship between the TGFβ pathway and common complex vascular diseases such as CAD has been more difficult to establish. The literature is replete with reports of in vitro and in vivo model systems studies of this pathway in vascular disease, but there have been much debate regarding the directionality of effect for TGFβ signaling, and SMAD3 function in particular, on disease initiation and progression [19, 20, 57, 58]. Interest in the role of TGFβ signaling in CAD has been renewed because of the CAD GWAS association of numerous loci that harbor TGFβ signaling molecules [10, 11].
In addition to SMAD3, a striking recent finding in GWAS meta-analyses has been the identification of the TGFB1 locus as a CAD associated region of the genome [6–8]. Although TGFB1 has not been identified as the causal gene in this locus through mechanistic studies, and there are not highly informative eQTLs in vascular cells or tissues, algorithms that integrate available gene expression and CAD association data, Summary data-based Mendelian Randomization [66] and MetaXcan [67], provide compelling support for the causality of TGFB1. These data suggest that TGFB1 expression promotes CAD risk, which is consistent with the fact that canonical TGFβ1 signaling occurs primarily through the risk-promoting SMAD3 pathway. While much remains to be learned regarding the role of TGFβ family members in vascular disease, mutations associated with Mendelian aortopathies as well as common variation associated with CAD, appear to result from increased TGFβ signaling.
A number of previous studies have provided evidence that TGFβ signaling promotes the differentiated phenotype, and that SMAD3 is involved at a molecular level in this process through direct interaction with the lineage determining factor myocardin (MYOCD) [30]. However, given the unique embryonic origin, the singular phenotypic characteristics of coronary artery smooth muscle cells, and the importance of this question in the context of recent GWAS studies linking SMAD3 to CAD risk, we sought to investigate in detail how SMAD3 affects the cell state of HCASMC, and how it might interact at a molecular level with other factors that are involved in the relationship between SMC phenotype and CAD. In vitro studies provided clear evidence that SMAD3 expression promotes expression of differentiation markers, and inhibits proliferation of HCASMC, and this phenotype contrasts to that promoted by CAD associated factor TCF21 which inhibits HCASMC differentiation [34]. ChIPseq data indicated that the majority of SMAD3 binding loci also bind TCF21 (Fig 3D), and that TCF21 binds adjacent to SMAD3 in ~3% of loci (Fig 3E). Detailed analysis of the normalized read depth at binding sites in these loci revealed an inverse correlation of SMAD3 and TCF21 binding, suggesting that these TFs might inhibit the binding of each other in these regions of the genome. This hypothesis was supported by ChIP-PCR studies in TCF21 depleted HCASMC which showed greater SMAD3 binding in two loci that harbor adjacent colocalization of these two factors. The observed interaction was shown to be due in part to local regulation of chromatin accessibility by TCF21. It is known that TCF21 recruits HDACs, and that SMAD3 recruits coactivators such as histone acetyltransferases (HATs) p300 and CREB binding protein (CBP), and may also recruit primarily repressive histone deacetylases (HDACs) [68, 69]. Further, reporter gene transfection studies with a SERPINE1 enhancer region that binds both SMAD3 and TCF21 revealed that TCF21 expression inhibited the SMAD3 positive effect on transcription, an effect that could represent independent effects mediated through individual TF binding sites, or could be due to direct interaction of these TFs affecting the binding of one another. Taken together, these data suggest two possible mechanisms by which TCF21 affects the transcriptional function of SMAD3, through modification of the epigenome in regions where they colocalize, and also that TCF21 can directly affect/inhibit SMAD3 binding and transcriptional promotion of the SMC differentiation program, across small genomic distances as has been described for other TF interactions [70].
Given the work reported here for the SMAD3 transcription factor, and growing information for GWAS loci genes that are expressed and functional in HCASMC [33, 71–73]it is possible to begin to establish a disease related transcriptional network for CAD in this cell type (Fig 6). Findings from these RNAseq and ChIPseq studies show that a number of SMAD3 downstream targets are validated or leading candidate causal genes in CAD associated loci, including CDKN2B, LMOD1, EDNRA, and SEMA5A. SMAD3 binds the loci for two of its upstream receptors, TGFβR1 and TGFβR2, but RNAseq data did not show their differential regulation in cultured HCASMC. Interestingly, these genes are also targets of TCF21, with the direction of effect primarily being the opposite of SMAD3, primarily inhibitory. In addition to SMAD3, TCF21 binds and regulates numerous TGFβ pathway factor genes, including TGFB1, TGFBR1 and TGFBR2. Although not directly related to TGFβ signaling, TCF21 inhibits expression of the CAD PDGFD gene, as well as PDGFB and PDGFRB receptor genes. The putative SMAD3-TCF21 interactions are consistent with data from various types of experiments presented here which indicate that TCF21 has an inhibitory role toward SMAD3 regulation of gene expression. While directionality of disease effect is in general difficult to establish, compelling data from HCASMC and vascular eQTL data presented here suggest that expression of SMAD3 is directly correlated to disease risk, and importantly these findings are opposite to those indicating that expression of TCF21 has a protective role [33, 59]. Further, given that expression of SMAD3 appears to promote disease risk along with HCASMC differentiation, and TCF21 promotes de-differentiation of this cell type and inhibits disease risk, these findings together argue that the process of phenotypic modulation is protective toward disease risk [33, 34, 74].
A limitation of the work reported here is the absence of correlation with appropriate in vivo experiments in appropriate animal models, such as genetic mouse models of atherosclerosis. For instance, combination of conditional deletion in the SMC lineage with concomitant lineage tracing of the targeted cells would be predicted to show that loss of SMAD3 expression would lead to increased exit of SMC from the media. Also, a concomitant increase in the number of SMC-derived cells in the plaque, and possibly at the fibrous cap, would be anticipated. Correlation of histological findings in diseased and normal human tissues samples from individuals of known genotype could provide compelling corroborative data regarding the molecular mechanism and direction of effect. These and other mechanistic studies are expected to support the human genomic and genetic data derived from experiments reported here.
Primary human coronary artery smooth muscle cells (HCASMCs) were purchased from three different manufacturers, PromoCell, Lonza and Cell Applications at passage 2 and were cultured in smooth muscle cell basal media along with hEGF, insulin, hFGF-B and fetal bovine serum (FBS) (Lonza # CC-3182) according to the manufacturer’s instructions. HCASMCs between passages 5–8 were used for all the experiments. For the ChIPseq and ChIP-PCR studies, cells were serum starved for 48 hrs and then treated with 5 ng/ml human recombinant TGFβ1 (R&D Systems) for 6 hrs before crosslinking.
SMAD3 (s8401 and s8402) silencer select siRNAs were purchased from Life Technologies. siRNA transfection was performed using Lipofectamine RNAiMAX (Life Technologies). For each well treated with the SMAD3 siRNA or scrambled control (Life technologies, #4390843), the final concentration was 20 nM. HCASMCs were seeded in 6 well plates and grown to 75% confluence before siRNA transfection. HCASMCs were transfected with the SMAD3 siRNA or scrambled control for 12 hours and subsequently collected and processed for RNA isolation after 48 hrs of transfection using the RNeasy kit (Qiagen). For the SMAD3 overexpression study, HCASMCs were transduced with 5ug of pRK5F-SMAD3 cDNA (Addgene plasmid# 12625) or control pCDNA3.1 DNA (ThermoFisher Scientific, plasmid# V79020) using the Amaxa Basic Nucleofector kit for primary mammalian smooth muscle cells (Lonza #VPI-1004) at a density of 1x106 cells per 100 μL sample using Nucleofector Program U-025. Cells were changed to medium with supplements 24hrs after transfection and cultured for an additional 48 hrs. The transduction efficiency was assessed by transducing HCASMCs with 2 μg of pmax-GFP cDNA and quantifying the percentage of GFP positive cells by quantitative fluorescence microscopy.
For evaluation of the effect of SMAD3 on HCASMC lineage marker expression, cells were subjected to knockdown or overexpression of SMAD3 as outlined above, and gene expression quantitated as below by qRT-PCR and western blotting. For quantitative immunofluorescence assay of marker expression, adherent HCASMCs were fixed with 4% paraformaldehyde (PFA) for 20 minutes, permeabilized with 0.5% Triton-X 100/PBS for 10 minutes and blocked with 3% BSA/PBS for 60 minutes at room temperature. Primary antibody incubations were performed at 4°C overnight followed by incubation with an Alexa Fluor-tagged secondary antibody for 60 minutes at room temperature. Nuclei were counterstained with Hoechst (1 μg/ml, Life Technologies). Images were acquired with a Zeiss Axioplan 2 microscope using the LASX software.
Migratory effects of SMAD3 expression were evaluated with a gap closure assay. The gap closure assay (Cell Biolabs #CBA-126) was conducted according to the manufacturer’s protocol. Briefly, 10,000 lentivirus transduced cells were seeded per well and let attach overnight. After gel removal 3 wells per condition were directly stained with crystal violet and imaged while the remaining 9 wells per condition were incubated for 12h before crystal violet staining. The covered area per well was quantified using ImageJ v1.47.
Proliferation of HCASMC with SMAD3 knockdown and over-expression was evaluated with an in vitro EdU assay (Thermo-Fisher). HCASMC were serum starved for 24 hours. Following starvation, the cells were exposed to serum for 24 hrs, with treatment during the last 3 hrs of this period with EdU from the Click-iT EdU Alexa Fluor 488 Imaging Kit (Life technologies, Carlsbad, CA; Cat# C10377) at a concentration of 20uM. The cells were incubated with EdU for 3 hours, then fixed and permeabilized using 4% PFA and 0.5% Triton-X in PBS, respectively. This was followed by incubation with Click-iT reaction cocktail, including CuSO4 and Alexa Fluor Azide, and then with nuclear staining with DAPI solution. Using a Leica inverted microscope, the number of total nuclei, and the number of co-stained nuclei were counted using the ImageJ (NIH) software on ten consecutive 10x fields for each condition.
RNA for all samples was extracted using the RNAeasy mini kit (Qiagen). HCASMC RNA (500 ng) were reverse transcribed using the High capacity RNA-to-cDNA Synthesis kit (Applied Biosystems). Quantitative PCR of the cDNA samples was performed on a ViiA7 Real-Time PCR system (Applied Biosystems) and gene expression levels were measured using SYBR green assays using custom designed primers and normalized to PBGD and GAPDH levels (Table 2).
Protein samples were harvested at 4°C using 1X RIPA buffer containing fresh protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific). Protein concentrations were determined using the Pierce BCA Protein Assay Kit. 50μg of each denatured HCASMC sample was loaded onto a 4–15% gradient SDS-PAGE gel (Bio-Rad). Samples were transferred to polyvinylidene difluoride membrane (Life Technologies) for 2h at 100V at 4°C and blocked with 5% milk in Tris-buffered saline and 0.05% TWEEN 20 (TBS-T, Sigma) for 1h at room temperature. Membranes were incubated with the following primary antibodies, mouse anti-GAPDH antibody (ab8245) was used as the loading control in all experiments. Anti-rabbit HRP (Sigma) or anti-mouse HRP (Sigma) secondary antibodies were used at a concentration of 1:10000 and diluted in 5% milk containing 0.05% Tween 20. Bands were detected using ECL western Blotting detection reagents (Pierce) per manufacturer’s instructions on the LI-COR Odyssey imaging system.
HCASMC (line 1508) were grown and SMAD3 expression knocked down as above. Three experimental and three control samples were generated and sequenced on a HiSeq 4000 machine, 125 bp paired end reads. Reads were processed using rnaSeqFPro, a workflow for full processing of RNASeq data starting from fastq files. In brief, the quality control was performed using FastQC, mapping to the human genome hg19 was performed using STAR second pass mapping to increase the percentage of mapped reads, and counting was done with featureCounts using GENCODE gtf annotation. Next, rnaSeqFPro performed differential analysis using DESeq2, conducted principal component analysis and hierarchical clustering using standard R functions, plotPCA and heatmap.2 and generated graphs using gglot2. DESeq2 gave 493 differentially expressed (DE) genes (FDR ≤ 0.05).
The differentially expressed genelist was used to interrogate the Ingenuity Knowledge Base, identifying Canonical Pathways and Cardiovascular Disease, and Molecular and Cellular Functions category enrichments. There was also significant enrichment for Cardiovascular System Development and Function terms, and the genes attributed to the top subcategory “development of vasculature” were employed to build a SMAD3 HCASMC network. Using only genes in this list, a network was created with connectivity supplied by the curated molecular interaction database of IPA. Visualization of this network was performed using Cytoscape open source software. Node color was mapped to log2 fold change with red representing genes that are downregulated along with SMAD3 and green representing genes that are upregulated, and node size mapped to the number of interactions with other genes. Edges were colored to distinguish types forms of functional interactions.
Briefly, approximately 2e6 HCASMC cells were fixed with 1% formaldehyde and quenched by glycine. The cells were washed three times with PBS and then harvested in ChIP lysis buffer (50 mM Tris-HCl, pH 8, 5 mM EDTA, 0.5% SDS). Crosslinked chromatin was sheared for 3x1 min by sonication (Branson SFX250 Sonifier) before extensive centrifugation. Four volumes of ChIP dilution buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100) was added to the supernatant. The resulted lysate was then incubated with Dynabeads Protein G (Thermo Scientific, 10009D) and antibodies at 4°C over-night. Beads were washed once with buffer 1 (20 mM Tris pH 8, 2 mM EDTA, 150 mM NaCl, 1% Triton X100, 0.1% SDS), once with buffer 2 (10 mM Tris pH 8, 1 mM EDTA, 500 mM NaCl, 1% Triton X100, 0.1% SDS), once with buffer 3 (10 mM Tris pH 8, 1 mM EDTA, 250 mM LiCl, 1% NP40, 1% sodium deoxycholate monohydrate) and twice with TE buffer. DNA was eluted by Chip elution buffer (0.1 M NaHCO3, 1% SDS, 20 μg/ml proteinase K). The elution was incubated at 65°C over-night and DNA was extracted with DNA purification kit (Zymo D4013). The purified DNA was assayed by quantitative PCR with ABI ViiA 7 and Power SYBR Green Master Mix (ABI 4368706) (Table 3).
ChIP was performed using the SMAD3 Abcam antibody ab28379 and HCASMC line 1508, and library was prepared using standard procedures. Briefly, DNA was prepared for end repair (Lucigen Endi-it, ER0720) and “A” tailing (NEB Klenow, M0212S), adaptor ligation (Promega, M180A), and library amplification (NEB, M0531S). ChIP-seq libraries were sequenced on HiSeq X10 for 150-bp paired-end sequencing. Quality control of ChIP-seq data was performed using Fastqc, and then low-quality bases and adaptor contamination were trimmed by cutadapt. After quality control and data filtering, data was mapped to hg19 using BWA mem algorithm. Duplicate reads were marked by Picard Markduplicate module and removed with unmapped reads by samtools view -f 2 -F 1804. MACS2.1.1 was used for peaks calling with default parameters and input as control. We utilized the Genomic Regions Enrichment of Annotations Tool (GREAT 3.0) to analyze the detected peaks, with the parameter “Basal plus extension”, which is proximal: 5 kb upstream, 1 kb downstream, plus Distal: up to 1000 kb. Gene ontology from GREAT output were analyzed by DAVID. KEGG pathways, biological processes, molecular functions, and GAD disease enrichment analysis was carried out using default settings. The HOMER findMotifsGenome.pl script was employed to search for known TRANSFAC motifs and to generate de novo motifs. The intersecBed was used to find overlapped peaks between SMAD3 and TCF21. The filter used to cut off SMAD3 peaks is: fold change> 5 and -logq_value>10. Pooled TCF21 peaks were cut off with three different thresholds, liberal: fold change>5, -logq_value>25; standard: fold change>10, -logq_value>60; stringent: fold change>15, -logq_value>200. The fastq files of SMAD3 ChIPseq in A549 cell lines was extracted from SRR1014002 by fastq-dump. Similar methods were used in quality control, alignment, peak calling and intersection with SMAD3 peaks in HCASMC.
To test the relative binding of SMAD3 and TCF21 transcription factors in regions of the genome where their binding was colocalized by ChIPseq studies, we compared the two ChIPseq datasets by calculating overlapping regions using bedtools and creating distributions of normalized fold changes, i.e., relative read counts in peaks compared to background, with a relative scale 0–100 for overlapping or adjacent binding sites. The results were presented as a normalized fold change correlation plot indicating relative binding of the two transcription factors. This process was automated in an algorithm ChIPSeqCompare (https://github.com/milospjanic/ChIPSeqCompare), which investigates differential binding by two transcription factors that are hypothesized to interact via epigenetic modification or protein-protein-DNA interactions.
HCASMCs (passages 5–6) were cultured in normal media. Approximately 50,000 fresh cells were collected by centrifugation at 500g for 5min and washed with cold PBS. Nuclei-enriched fractions were extracted with cold lysis buffer (10 mM Tris–HCl, pH7.4, 10 mM NaCl, 3 mM MgCl2 and 0.1% IGEPAL) and the pellets were resuspended in transposition reaction buffer (20 mM Tris-Cl pH7.5, 10 mM MgCl2, 20% Dimethylformamide) and Tn5 transposase (Illumina Nextera). Transposition reactions were incubated at 37°C for 30 min, followed by DNA purification using the DNA Clean-up and Concentration kit (Zymo D4013). The genomic DNA was extracted using Quick-DNA Microprep Kit (Zymo D3020). The purified DNA was quantified by qPCR with ABI ViiA 7 and Power SYBR Green Master Mix (ABI 4368706) and normalized by genomic DNA.
To test directionality of effect of SMAD3 cis-risk alleles in HCASMC on SMAD3 gene expression, we investigated SMAD3 eQTL data emanating from a genome-wide association of gene expression with imputed common variation identified in 52 HCASMC studied with whole genome RNA sequencing and 30x whole genome sequencing [59]. Significance of association of variance in SMAD3 expression with genotype was evaluated by regression analysis, the results were visualized with HCASMCeQTLviewer (https://github.com/milospjanic/HCASMCeQTLviewer), a combined bash/R/awk script that outputs the directionality of effect for any SNP-gene association in HCASMC.
To test causality and directionality of the SMAD3 gene in CAD in the context of multiple cis-acting eQTLs, we investigated the directionality of change of expression level with a number of risk GWAS SNPs. This analysis employed CAD GWAS data from a recent meta-analysis [7] to select risk SNPs and define risk alleles and HCASMC eQTL data to perform regression analysis. To facilitate this analysis, we developed a custom algorithm to analyze whole genome sequencing and expression data in HCASMC. We developed GeneCausalityTest for coronary artery disease (https://github.com/milospjanic/GeneCausalityTestCAD), a combined bash/awk/R script for defining causality of a gene for a given trait, in this case CAD, given the directionality of change of expression level with the increasing number of risk GWAS SNPs. This correlation was improved by resolving local haplotype structure, creating a sample correlation matrix and averaging on samples with unique local haplotype profile. This approach improved correlations made by GeneCausalityTest for defining causality of a gene for CAD, especially in the local regions of strong linkage disequilibrium. This analysis provided output for the directionality of expression change with the increasing number of risk SNPs and used CAD GWAS data (Nelson et al.) and HCASMC eQTL data for regression analysis. For correlating the effect sizes of SMAD3 variants and SMAD3 gene expression, the CAD GWAS summary data from Nelson et al. and cis-eQTL summary data from STARNET were obtained from www.cardiogramplusc4d.org and dbGaP (phs001203.v1.p1), respectively [7, 60]. The files were loaded and processed in R using subset and merge functions to obtain an overlapping variant list at GWAS P<1E-04 and nominal cis-eQTL P<0.05. The beta coefficients (negative log odds ratio (OR)) of these variants were plotted using ggplot and a linear mixed model was used to compute a smooth local regression. Pearson correlation coefficient r and p-value of significance were calculated using cor.test in R. This method has been standardized in an algorithm named UniqueHaplotypeTestCAD (https://github.com/milospjanic/UniqueHaplotypeTestCAD).
A dual Luciferase assay (Promega, #E1910) was used to measure the luciferase and renilla activity in the transfected cells. Five hours after transfection, growth media was removed, and cells were washed with 1X PBS. For cell lysis, cells were incubated with 100 μl of 1X Passive cell lysis buffer for 20 minutes at room temperature. 10 μl of the cell lysate was transferred to 96 well white flat bottom assay plates (Costar, #3912). Luciferase assays were performed with SpectramaxL Microplate luminometer with SoftMax Pro software. First, 100 μl of Luciferase assay substrate was injected into the lysate and luciferase activity was measured for 10 sec. Subsequently, 100 μl of Stop and Glo reagent was injected to the lysate to stop the luciferase activity and catalyze the renilla reaction. Renilla activity was measured for 5 seconds following 1.6 second incubation. HCASMCs and A7R5s were transfected with either SMAD binding element-luciferase reporter plasmid (pSBE4 luc), pGL3-hSM22-325-luc plasmid, pGL3-luc or the pBV-luc constructs and stimulated with SMAD3 (pRK5F SMAD3) and TCF21 (pCMV6 XL4) cDNAs. The Renilla luciferase reporter plasmid was used as the internal control of transfection efficiency.
Experiments were performed by the investigators blinded to the treatments/conditions during the data collection and analysis, using at least two independent biological replicates and treatments/conditions in technical triplicate. For the statistical analyses not discussed above, methods were as follows. R or GraphPad Prism 7.0 was used for statistical analysis. For motif and gene enrichment analyses, we used the cumulative binomial distribution test. For overlapping of genomic regions or gene sets, we used Fisher’s exact test and/or the hypergeometric test, as indicated. For comparisons between two groups of equal variance, an unpaired two-tailed Student’s t-test was performed or in cases of unequal variance a Welch’s unequal variances t-test was performed, as indicated. P values <0.05 were considered statistically significant. For multiple comparison testing, one-way analysis of variance (ANOVA) accompanied by Tukey’s post hoc test were used as appropriate. All error bars represent standard error of the mean (SE).
|
10.1371/journal.pntd.0003102 | Mycetoma: Experience of 482 Cases in a Single Center in Mexico | Mycetoma is a chronic granulomatous disease. It is classified into eumycetoma caused by fungi and actinomycetoma due to filamentous actinomycetes. Mycetoma can be found in geographic areas in close proximity to the Tropic of Cancer. Mexico is one of the countries in which this disease is highly endemic. In this retrospective study we report epidemiologic, clinical and microbiologic data of mycetoma observed in the General Hospital of Mexico in a 33 year-period (1980 to 2013). A total of 482 cases were included which were clinical and microbiology confirmed. Four hundred and forty four cases (92.11%) were actinomycetomas and 38 cases (7.88%) were eumycetomas. Most patients were agricultural workers; there was a male predominance with a sex ratio of 3∶1. The mean age was 34.5 years old (most ranged from 21 to 40 years). The main affected localization was lower and upper limbs (70.74% and 14.52% respectively). Most of the patients came from humid tropical areas (Morelos, Guerrero and Hidalgo were the regions commonly reported). The main clinical presentation was as tumor-like soft tissue swelling with draining sinuses (97.1%). Grains were observed in all the cases. The principal causative agents for actinomycetoma were: Nocardia brasiliensis (78.21%) and Actinomadura madurae (8.7%); meanwhile, for eumycetomas: Madurella mycetomatis and Scedosporium boydii (synonym: Pseudallescheria boydii) were identified. This is a single-center, with long-follow up, cross-sectional study that allows determining the prevalence and characteristics of mycetoma in different regions of Mexico.
| Mycetoma is a chronic, subcutaneous granulomatous disease that usually begins after traumatic inoculation with causative microorganisms. Based on its etiology, mycetoma is referred to eumycetoma when the infection is caused by filamentous fungi, and actinomycetoma when the infection is due to aerobic actinomycetes (in Mexico predominantly Nocardia brasiliensis). Establishing the etiology is extremely important since it impacts treatment regimens. Mycetoma typically presents around the Tropic of Cancer between latitude 15° South and 30° North (also known as “mycetoma belt”) affecting poor populations in Africa, Asia, and Latin America, including Mexico, which represents a highly endemic area with higher frequencies of actinomycetomas. Mycetoma usually affects males (male∶female ratio of 3∶1), agricultural or rural workers (age range 20–40 years) that typically do not have access to protective equipment. The main clinical presentation is as soft tissue swelling with sinus tract formation draining grains, which leads to diagnosis. The foot is the most commonly affected localization; however, when disease presents in high risk areas, such as the trunk, it can disseminate to the lungs and spinal cord. This report represents a single center study which provides epidemiologic, clinical, and microbiological data of mycetoma cases in different regions of Mexico.
| Mycetoma is a chronic granulomatous disease, associated with a progressive, inflammatory reaction that clinically presents as tumor-like soft tissue swelling with sinus tract formation that drains purulent material containing grains. Mycetoma usually results of traumatic implantation of soil organisms on subcutaneous tissue; can be classified as eumycetoma or actinomycetoma depending on whether the infection is caused by filamentous fungi or aerobic filamentous actinomycetes, respectively [1], [2], [3].
Mycetoma represents a classical neglected disease that primarily affects the poorer populations and rural regions of Africa, Latin America, and Asia at latitudes defined as the “mycetoma belt” where higher mycetoma frequencies are observed. This region is located around the Tropic of Cancer, between latitudes 15° South and 30° North, encompassing the countries with the highest rates of infection including Sudan, Somalia, Senegal, India, Yemen, Mexico, and Venezuela [1], [4], [5]. The predominant climate of the “mycetoma belt” is subtropical and dry tropical with an annual average rainfall of about 500–1000 mm and temperatures ranged from 10–20°C to 20–40°C, respectively. This region is characterized by low humidity and low annual rainfall with well-defined alternating rainy and dry seasons. Actinomycetomas caused by Nocardia spp. occur mostly in regions with higher humidity, while actinomycetoma caused by Actinomadura spp. and Streptomyces spp. or eumycetoma occur in drier areas with low relative humidity [1], [3], [4], [5].
Most causative agents of mycetoma, including fungi and actinomycetes, have been isolated from soil, decaying organic matter, plants and thorns; and, the disease is usually associated with traumatic injury followed by inoculation of the microorganism propagule. There are three main factors associated with the establishment of disease: inoculum size, immune status of the host, and hormonal adaptation (based on the observation that men typically develop the disease) [1], [3], [6], [7], [8], [9].
Epidemiological data from different areas demonstrate that males are more affected (sex ratio 3–4∶1), ranging in age between the third and fourth decades of life (20–40 years). Some studies have reported that 3–5% of cases affect children committed to field-work [6], [10], [11]. Mycetoma is common in persons that work in rudimentary conditions without protective garments or shoes leading to the presentation of the illness primarily in poor rural workers or homemakers that participate in outdoor activities. Nearly all cases affect the lower limbs (75%), especially the foot and lower limbs. The nature of the patient's occupation also influences disease presentation, for example lumberjacks and sugarcane carriers generally present with mycetoma on the back [6], [10], [11].
The incubation period is unknown, disease symptoms present months to years after traumatic inoculation, depending on the inoculum size, strain virulence, and the host's immune response [7]. Because reporting mycetoma cases is not mandatory, the worldwide incidence is unknown; however, a recent published meta-analysis by van de Sande [11] reported incidence rates of 3.49 and 1.81 per 100,000 habitants in Mauritania and Sudan, respectively; while Mexico showed the highest rate in Latin American. Although a separate study reported that the number of cases per year in Sudan, Mauritania, and Mexico were 106, 80.7, and 73, respectively [6], several cases are probably not reported.
The objective in this study was to provide epidemiological, clinical and microbiological data of mycetoma in different regions on Mexico, presenting at a public single center, the General Hospital of Mexico (specialty hospital).
The Institutional Review Board approved the retrospective (cross-sectional) analysis of the database and clinical records of the Mycology Department of the Dermatology Service at the General Hospital of Mexico, patients were enrolled between January 1980 and December 2013 (34 years). We included all cases of mycetoma confirmed by microscopic observation of grains by direct examination with 10% potassium hydroxide (KOH), saline solution, and lugol solution. The culture media used were Sabouraud dextrose agar and Yeast extract agar, however, when infection by Actinomadura madurae was suspected, Lowenstein-Jensen agar, and BHI agar (Brain Heart Infusion) were used. Histological examination was performed in some cases using hematoxylin and eosin (H&E), Grocott's methenamine silver (GMS), and Periodic acid–Schiff (PAS)
Actinomycetes identification was carried out using micro morphological criteria (Gram and Kinyoun stains) as well as biochemical and major phenotypic tests such as urease production, hydrolysis of casein, gelatin, tyrosine, xanthine, hypoxanthine substrates and, growth at 45°C [12], [13]. Fungal agent identification was based on morphological and reproductive form criteria and on biochemical tests. Some strains were identified using molecular techniques (by amplification and sequence analysis of ribosomal DNA. the internal transcribed spacer region (ITS), the translation elongation factor 1 alpha (EF1-α), the partial beta tubulin gene (TUB), and the small subunit of the nuclear ribosomal RNA gene [nucSSU]). General epidemiologic and clinical data were extracted from clinical records. All patients remained anonymous and descriptive statistics were used to analyze the data.
A total of 482 mycetoma cases were included in the present study, one patient presented simultaneously two different mycetomas. Demographic data are described in Table 1.
Figure 1 illustrates the age distribution of mycetoma, and Figure 2 the geographic mycetoma distribution in Mexico. The Pacific Ocean zone including Morelos, Guerrero, and Oaxaca accounted for 284 cases (58.92%) and the Gulf of Mexico zone (including Hidalgo and Veracruz) accounted for 162 cases (33.60%). The remaining 36 cases presented in other states including Chiapas, Tabasco, Puebla, Michoacan, Colima, Jalisco, San Luis Potosi, Durango, Chihuahua, and Baja California.
Most cases were actinomycetomas (92.11%) with a male to female ratio of 2.8∶1. However, this ratio changes in cases due to Actinomadura madurae, of the 42 cases 13 (30.06%) were males and 29 (69.04%) were females with a male∶female ratio of 1∶2.2. Age ranges were classified in decades, the mean age was 34.5 years old (range 7–92 years). Pediatric cases (<18 years old) were 20/482 (4.14%) and 5/482 cases (1.083%) were younger than 15 years old.
Lower limbs were affected in 341 cases (70.74%). Three hundred one cases (62.44%) occurred in the foot, 70 cases (14.52%) affected the upper limbs (36 cases on the hands [7.46%] and 34 on the arms [7.05%]). The trunk was involved in 49 cases (10.16%), 38 (7.88%) of those included back and shoulders. All cases involving multiple sites were associated with multiple traumatic inoculations. In regards to clinical presentation, mycetomas were classified as follows: 468 cases as tumor-like with draining sinuses (97.1%); eight cases as tumor–like without sinuses (1.65%); four cases as verrucous plaque (0.82%) and two as cystic form (0.41%). Eight patients (1.6%) presented lymphatic spread from the original mycetoma lesion: six from the foot to the inguinal area and two from the back to the axillary region. (Figure 3) One patient presented with two mycetomas, each with a distinct causative agent: the one affecting the right foot was caused by Madurella mycetomatis, and the second one affecting the left foot was caused by Fusarium solani complex.
Etiological agents were identified in 472/482 cases (98.34%). The agent was found in all of the actinomycetoma cases (n = 444 cases). In 430 cases (89.2%) the microorganisms were isolated and identified, and the remaining 14 cases were classified according to the grains observed during direct examination and/or histopathologic analysis. Two cases had double concurrently causative agents: N. asteroides s. l.+N. brasiliensis [14] and N. brasiliensis+A. madurae. Thirty-eight cases were eumycetomas and in 30/38 cases the etiological agents were isolated. Of the remaining seven cases (four hyaline and three melanized-type) only grain observation at direct microscopy was detected without identifying the causative fungi (Table 2) [15], [16].
Mycetoma is a chronic granulomatous disease generally affecting low-income people including agricultural workers, peasants, or rural workers laboring with limited or no protective garments and soiled tools. The majority of cases (62%) described in this report affected the foot, supporting previous reports [3], [5], [6], [10] and one meta-analysis [11] that described foot as the most common site of infection (68.7% of cases). Since mycetoma presented most often on the feet of individuals living in the Indian endemic region, it explains why initial reports mentioned it as “Madura foot” [11], [17], [18], [19]. Due to the predilection for feet, mycetoma control could be achieved by using appropriate footwear and clothing that protects the limbs. However, it should be emphasized that people living in endemic regions sometimes wear open-toed shoes, mainly due to the warm climate and therefore are less protected against potential trauma [1], [5], [20], [21].
Mycetoma is associated with high morbidity and low mortality; however, the socioeconomic impact is significant; therefore patients are unable to work, resulting in decreased family income. In addition, treatments are expensive and difficult to maintain due to prolonged course of the disease. Almost all countries located in the “mycetoma belt” do not provide free quality health services or medical insurance [5], [20], [21].
The classical clinical presentation of mycetoma should lead to simple diagnosis based on the identification of a swelling zone with multiple sinus tracts; however, there is a significant lack of information between patients and clinicians, leading to delayed diagnosis and late referral to hospital, and consequently inadequate therapeutic response [5], [7], [8].
This study examined nearly 500 mycetoma cases from a single public hospital, helping to control for variables and facilitated the isolation and identification of the causative agents (the causative agent was classified in most of the cases). Ninety two percent were actinomycetoma and 8% were eumycetoma, in accordance with previous studies [10]. This result differed from one report [6] that identified 3.5% eumycetoma cases this is probably due to the difficult diagnosing these cases until they are finally referred to specialty hospitals. Reports from Latin America [22], [23], [24], [25] and, particularly, studies conducted in Mexico show a predominance of actinomycetoma, in contrast to those made in Africa, India, and Asia where cases of eumycetoma predominate [17], [20], [26], [27], [28], [29]. This epidemiological difference can be explained by differences in climate and other environmental factors. The effect of climate is observed in mycetoma cases reported in India [17], [18], [19] where the majority of actinomycetoma occur in the northern region, where the climate is subtropical and has a higher annual rainfall; while, eumycetomas occur more often in the southern where the climate is dry tropical, has a low relative humidity, and more constant temperatures. In Mexico, eumycetomas occur in drier areas. This study provides a more accurate number of cases of mycetoma in Mexico (73 new cases per-year) [6], we believe that mycetoma remains difficult for clinicians to diagnose, for that several cases may be under diagnosed and therefore underreported [1], [3], [11].
In our study, we observed that mycetoma primarily affects men, with a male∶female ratio of almost 3∶1 in our study, in concordance with a previous report about mycetoma incidence in Mexico [6]. This male predominance of micetoma can be attributed to occupational and hormonal aspect [8], [9]. The role of hormones in disease susceptibility may be explained by the few cases of mycetoma in children and the rapid growth of the lesions and increased in severity during pregnancy. Interestingly, the observed change of male∶female ratio in cases of mycetoma caused by A. madurae (male∶female ratio of 1∶2.2) is partly due to this microorganism is not affected by progesterone and testosterone as with Nocardia brasiliensis [8], [9].
Figure 1 shows mycetoma is most prevalent in the third decade of life (63.26%), which represents the most productive ages.
The mean age was 34.5 years, similar to observations were made by van de Sande [7]. Some cases were reported in elderly, however we must consider that the infection may have started many years ago, suggesting that these individuals may have acquired the disease in youth. Moreover, only 4% of the cases were reported in patients <18 years old similar to previously reported studies [6], [30]. The percentage of children infected in our study (and other reports [6], [30]) differed from a report by Fahal et al. [31] that described a 15% infection rate (n = 722) in children in Sudan, this was probably due to their outdoor work activities. However, the same study reported trauma in only 22.5% patients suggesting that different mechanisms of infection that deserve to be clarified [1], [31].
The main clinical presentation was tumor-like with draining sinuses; cases presented as tumor-like without sinuses and cystic form were all eumycetomas; verrucous-plaque presentation was rare, the last one is very important since its differential diagnosis include verrucous-tuberculosis, chromoblastomycosis and nontuberculous mycobacterial diseases. Although lower limbs (predominantly feet) were most commonly affected by mycetoma in our population (similar to the majority of previous reports [6], [10], [25]) it is interesting that the trunk was affected in about 10% of cases (predominantly back and shoulders). A previous Mexican report [6] described and incidence rate of the trunk in 19% of the cases, which was significantly different from the 1.4% rate described for these cases in Sudan [11]. These differences in anatomic regions affected may be explained due to occupation differences; patients in Mexico usually carry wood, sugarcane, or diverse materials on their backs. Mycetoma affecting the trunk should be considered of poor prognosis because of the proximity of lungs, spinal cord, and viscera [32]. It is also important to emphasize that cases presenting with multiple infections were described in immunocompetent patients that suffered multiple traumatic inoculations. Cases associated with lymphatic spread (1.65%) are typically seen in immunocompromised patients (malnutrition, immunosuppressants, malignant tumors, and chronic alcoholism).
Nocardia spp. was the main etiological agent (82.32% of cases), being N. brasiliensis the predominant species (78.21%). The strains were identified using phenotypic tests that lead to the identification of two more species: N. asteroides complex and N. otitidiscaviarum (formerly N. caviae) [14], [33]. Molecular biology techniques such as PCR and sequencing of 16S rRNA and the hsp65 gene allow the correct identification and classification of Nocardia species [13]. For example, N. mexicana, N. harenae, and N. takedensis were isolated from Mexican patients [34], [35], [36]. The second most common causative agent was A. madurae; distributed worldwide [11], which is easily identified due to the larger (1–3 mm) white-yellowish, soft, wide-fringed border grains. Regardless, A. madurae is more difficult to isolate than Nocardia species, which regularly grow in rich culture media such as Lowenstein-Jensen, BHI-agar. Morphologic analyses leads to the identification of the causative agent, however, confirmation is carried out using phenotypic and temperature tests [12]. Other etiologic agents include A. pelletieri and S. somaliensis commonly found in Africa and Asia. Remarkably, the two cases presented with mixed infections following traumatic inoculation were due to N. brasiliensis+N. asteroides s.l. and N. brasiliensis+A. madurae, respectively, both presented in immunocompetent patients [14].
The causal agents of eumycetoma represented only 8% of our series, melanized fungi was the most commonly observed (26 cases, 5.39%). Of these, Madurella mycetomatis was the foremost isolated fungi, found in 15 cases, and also considered responsible for 25% of cases worldwide, mainly in Africa and Asia [11]. Some cases have a well-defined history of trauma (e.g., thorn pricks) prior to mycetoma development; however, in some cases a well-defined traumatic event was not identified, de Hoog [37] recently noted that M. mycetomatis is a close relative to dung-inhabiting fungi and suggested that the natural habitat of this fungus could therefore also be dung; trauma or repeated contact with cattle dung could act as an adjuvant for inoculation of causative agents of mycetoma. Identification of Madurella species has been hampered by the absence of sporulation leading to confusion during the identification process. The use of molecular techniques for the identification of specific regions and genes (e.g., rRNA, ITS, parcial β-tubulin gene, RNA polymerase II subunit 2 gene) has defined Madurella species as a cryptic complex belonging to the order Sordariales, consists primarily of M. mycetomatis, M. fahalii, M. tropicana, and M. pseudomycetomatis [38]. Identification of the infecting agent is critical since differences in thermal adaptation and susceptibility to antifungal agents exist between strains. M. grisea has been reclassified and now belongs to the order Pleosporales and named Trematospheria grisea. It should be noted that the latter was more frequently reported as a cause of mycetoma than M. mycetomatis in a recent report from Mexico [6]; however, difficulties in morphological and phenotypic identification could have led to confusion. Other melanized fungi further characterized by molecular biology techniques was Exophiala jeanselmei and Cladophialophora bantiana, which are common agents of phaeohyphomycosis [15]; and Cladophialophora mycetomatis, considered new specie [16]. Regarding hyaline fungi, Scedosporium boydii (Syn: Scedosporium apiospermum, Pseudallescheria boydii), was the foremost isolated strain (similar to other studies) [6], [11], [24], [26]. Mycetomas due to Fusarium have been previously described [39], [40], we found F. solani in two cases; interestingly, one of them was microscopically classified as F. chlamydosporum, but was reclassified using molecular biology as part of the Fusarium solani complex (CBS 135554). One case resulted from infection with Aspergillus nidulans, an agent rarely reported [41] and identified morphologically for the presence of Hülle cells. A case resulting from Microsporum canis infection was classified as pseudomycetoma [42] because of the rarity of the agent as a mycetoma pathogen and usually development as consequence of a chronic tinea capitis, typically seen in immunocompromised patients in the absence of traumatic inoculation [42], [43].
The study has limitations inherent to its design, however, provides important information about the status of mycetoma in Mexico. The study results can be generalized only to our population (Mexico); although the geographical areas studied has similarities with other world regions in terms of climate, distribution of etiologic agents and sociocultural conditions.
Mycetoma fulfills all the criteria of a neglected tropical disease [20], [21], [44]. It is extremely important to monitor cases and their causative agents, as a mean to understand the epidemiology of the disease, and to establish interventions for prevention, treatment and rehabilitation.
|
10.1371/journal.ppat.1003196 | Modeling Host Genetic Regulation of Influenza Pathogenesis in the Collaborative Cross | Genetic variation contributes to host responses and outcomes following infection by influenza A virus or other viral infections. Yet narrow windows of disease symptoms and confounding environmental factors have made it difficult to identify polymorphic genes that contribute to differential disease outcomes in human populations. Therefore, to control for these confounding environmental variables in a system that models the levels of genetic diversity found in outbred populations such as humans, we used incipient lines of the highly genetically diverse Collaborative Cross (CC) recombinant inbred (RI) panel (the pre-CC population) to study how genetic variation impacts influenza associated disease across a genetically diverse population. A wide range of variation in influenza disease related phenotypes including virus replication, virus-induced inflammation, and weight loss was observed. Many of the disease associated phenotypes were correlated, with viral replication and virus-induced inflammation being predictors of virus-induced weight loss. Despite these correlations, pre-CC mice with unique and novel disease phenotype combinations were observed. We also identified sets of transcripts (modules) that were correlated with aspects of disease. In order to identify how host genetic polymorphisms contribute to the observed variation in disease, we conducted quantitative trait loci (QTL) mapping. We identified several QTL contributing to specific aspects of the host response including virus-induced weight loss, titer, pulmonary edema, neutrophil recruitment to the airways, and transcriptional expression. Existing whole-genome sequence data was applied to identify high priority candidate genes within QTL regions. A key host response QTL was located at the site of the known anti-influenza Mx1 gene. We sequenced the coding regions of Mx1 in the eight CC founder strains, and identified a novel Mx1 allele that showed reduced ability to inhibit viral replication, while maintaining protection from weight loss.
| Host responses to an infectious agent are highly variable across the human population, however, it is not entirely clear how various factors such as pathogen dose, demography, environment and host genetic polymorphisms contribute to variable host responses and infectious outcomes. In this study, a new in vivo experimental model was used that recapitulates many of the genetic characteristics of an outbred population, such as humans. By controlling viral dose, environment and demographic variables, we were able to focus on the role that host genetic variation plays in influenza virus infection. Both the range of disease phenotypes and the combinations of sets of disease phenotypes at 4 days post infection across this population exhibited a large amount of diversity, reminiscent of the variation seen across the human population. Multiple host genome regions were identified that contributed to different aspects of the host response to influenza infection. Taken together, these results emphasize the critical role of host genetics in the response to infectious diseases. Given the breadth of host responses seen within this population, several new models for unique host responses to infection were identified.
| Influenza A virus (IAV) (orthomyxoviridae) is a negative sense RNA virus which causes severe, acute respiratory disease. Worldwide influenza infections cause several million cases annually, with severe pandemics (such as the 1918 pandemic) causing high levels of morbidity and mortality [1]. Among infected individuals there is significant variation in the clinical disease caused by IAV ranging from an asymptomatic infection to severe and acute respiratory distress syndrome [2]–[8]. Population-wide disease variation applies not only to clinical disease, but also to individual immune responses mounted in response to IAV infection [9], [10], as well as long-term complicating pathologies and co-infections [2], [11]–[13]. Despite the importance of understanding the underlying mechanisms of IAV-associated disease, the sources of the observed disease variation are unclear.
Like many viruses, IAV engages in a large number of complex interactions with various host proteins [14], [15]. It is less clear how polymorphisms in these and other host genes/proteins cause variation in the disease process following infection with IAV. A study of survival data from the 1918 IAV pandemic showed that host genetic variation [16] contributes to IAV disease variation. However, in contrast with other pathogens [17]–[21], human polymorphisms have not yet been identified that contribute to variable responses to IAV infection, although there have been some suggestions of polymorphisms in HLA contributing to IAV recovery [22], [23]. As IAV disease severity is likely due to a combination of viral, host, demographic and environmental factors [7], [13], [24]–[26], this complexity has interfered with reductionist approaches to evaluating the role that host genetic variation plays in regulating different IAV-associated disease outcomes across the population.
Mouse models of IAV infection have provided novel insights into the role of host genetics on IAV disease outcomes. This approach led to the discovery of the naturally polymorphic, interferon inducible Mx1 gene, which inhibits IAV replication and limits disease [27]. Subsequently, most studies of host genetic contributions have used naturally defective Mx1 mouse strains, such as C57BL/6J to study the effect of gene knock-outs on the host response to influenza. These studies have shown that many genes contribute to the host response [28]–[35] (reviewed in [36], [37]), and knock-outs often affect clinical disease primarily by altering the host inflammatory response [28], [29], [32], [35], [38]–[40]. Comparisons between inbred mouse strains [41]–[44] have confirmed that natural variation contributes to differential host responses. Given that most polymorphisms within the human population will be those that alter expression and/or function, rather than whole gene knock-outs, studies comparing naturally occurring polymorphisms are more relevant to human disease. Several recent studies [42], [45], [46] using classical recombinant inbred (RI) panels have identified a number of quantitative trait loci (QTL) contributing to host responses following IAV infection. However, traditional mouse genetics systems have limitations on their ability to accurately model the genetic structure and diversity of outbred populations, like humans [47], [48].
We developed a new model that captures host responses to IAV infection across a genetically diverse host population by using incipient lines from the Collaborative Cross (CC) octo-parental RI panel, known as the pre-CC population [49]–[51]. This population is highly genetically diverse (∼40 million single nucleotide polymorphisms (SNPs) evenly distributed across the genome), with up to eight functionally variant alleles at any given locus [52]. The pre-CC population exhibited a broad range of phenotypic outcomes, including unique combinations of disease phenotypes following infection, and we identified three novel QTL associated with multiple aspects of influenza induced disease. Furthermore, we identified a novel Mx1 allele in the CAST/EiJ mouse strain and sequenced the associated haplotype. By integrating QTL mapping with whole genome sequence information, we significantly reduced the number of candidate genes within each QTL. Our findings provide a clarification of the importance of genetic variation in the host's response to IAV infection, and a foundation of support for the hypothesis that genetically complex mouse models such as the CC will provide a robust platform for studying the role of host genetic variation in regulating the host response to infection.
We used 155 pre-CC mice, each from an independent, incipient CC line, as well as sets of mice (n = 5–11) from each of the eight CC founder strains. Mice were infected with a dose of the mouse adapted A/PR/8/34 (PR8) IAV that was known to cause severe disease in several of the CC founder strains (C57BL/6J, 129s1/SvImJ, A/J), and we assessed IAV-induced weight loss (measured as a percentage of starting weight) and clinical disease daily through four days post infection (D4), at which point the mice were euthanized and lung tissue assessed for viral replication, virus-induced inflammation and pathology, and (pre-CC mice only) transcriptional profiles by microarray analysis within the lungs (Table 1, Table S1, Dataset S1). This D4 timepoint was chosen to allow severe pathology to develop in susceptible lines, while minimizing the animals lost in this study due to humane euthanasia conditions. We examined the weight changes and clinical scores animals experienced through the course of this experiment, and found that weight loss and clinical scores of animals were highest at D4. We therefore limited our analysis of weight and clinical scores to this timepoint. Importantly, in analyzing the phenotypes of the pre-CC mice, we found no evidence for effects of age, generation of inbreeding, block effects or starting weights on gathered phenotypes, and therefore did not include these variables in our analysis.
The infected founder strains varied significantly for all measured phenotypes, including D4 weight, log titer, virus induced inflammation and pathology, except for variation in alveolar debris (p-values ranging from 0.15 to 1.37×10−9, Figure 1, Table S1). Founder strains could be grouped into susceptible (high viral titer, inflammation and weight loss) or resistant (low viral titer, little inflammation and weight loss) groups (Figure 1, Figure S1). As with the founders, many aspects of IAV associated disease were correlated with each other across the pre-CC population (correlation coefficients ranging from −0.78 to 0.78, Figure 1, Table S2), with the exception of alveolar immune cell infiltration as well as gross edema and hemorrhage at time of harvest, which were not strongly correlated with the rest of the host response to infection. Pre-CC mice often showed unique combinations of disease-associated phenotypes (e.g. high levels of viral replication but low inflammation and weight loss, no replication but significant weight loss, Figures 1 and 2). Therefore, though the pre-CC population recapitulated the range of variation within any given phenotype (Table S1), we observed new phenotypic combinations not seen in the parental lines.
The unique combinations of disease-associated phenotypes across the pre-CC population led us to investigate the relationships between viral replication and immune cell infiltration on weight loss, a long standing question within the IAV field. Since the large number of pre-CC mice we had in this study lacked the genetic structure of the founder strains, this population was uniquely positioned to evaluate the relationships between these disease parameters. Both log titer and airway inflammation (the cellular infiltrate most clearly related to infection status) were significant predictors (p-values<2.2×10−16 and 1.56×10−9, respectively) of D4 weight (Figure 1, Table S3). However, log titer and airway inflammation together were significantly better predictors of D4 weight than either variable alone (based on both partial F tests and Akaike Information Criterion (AIC), Table S3).
In addition to measuring disease associated phenotypes, we also assessed host transcript levels within the lungs at four days post infection. Of the 155 pre-CC mice used in this study, 99 had RNA of sufficient quality to use for RNA microarray analysis (see GEO, accession GSE30506 for full microarray dataset). A total of 11,700 genes passed quality control processing, and did not have a SNP across the eight founder lines which could impact their intensity on the array. Out of these 11,700, we identified the 6000 most variable and interconnected genes across this population and used weighted gene co-expression network analysis (WCGNA) to cluster these transcripts into twelve modules, labeled A–L (Figure 3, Table S4). Seven modules (B, D, F–I, K) were enriched for specific gene ontology (GO) terms (Table S5), including cellular signaling (module G), cell growth and biosynthesis (module D) and immune responses (module K). There was little to no overlap between the enriched categories across modules. We used the eigengene, an idealized representation of module transcription levels for each individual mouse, to correlate module expression levels with disease phenotypes as eigengene expression has been used previously to simply describe the sets of transcripts within a module [53]. We found that eigengene values for eight of the twelve modules (modules A–C, F, H, and J–L) were correlated with multiple disease-related phenotypes. Modules E and G correlated with aspects of Virus-induced inflammation and module D correlated with D4 weight (Figure 3). These results suggest that in this genetically diverse population severity of influenza infection is associated with wide-scale variation in a large number of biological processes within the lung.
Given the variation in disease, virologic, inflammatory, pathologic, and transcriptional phenotypes observed in the pre-CC animals, we conducted QTL mapping (as in [49]) to identify host genome regions contributing to variation in IAV-induced phenotypes (Figures S2, S3, S4, Table 2). Previous studies [27] have identified a large effect IAV resistance gene, Mx1, on chromosome 16, and our a priori expectation (see below) was that we would identify a QTL over Mx1, validating our mapping approaches within the CC. Consistent with this, we identified a highly significant QTL on chromosome 16, HrI1 (Host response to Influenza) that contributed to a number of these disease-associated phenotypes (D4 weight, D4 clinical, log titer, IHC score, airway inflammation and airway damage). HrI1 explained 41.67% of the variation in weight loss (i.e. the adjusted R-squared for a model with all 8 strain effects), and similar amounts of variation in other phenotypes, and was located in a 0.71 Mb region (1.5 LOD interval: 97500418-98213493) annotated as containing 10 genes and one non-coding RNA, including the known anti-influenza gene Mx1. We also conducted QTL mapping on the eigengenes of the expression modules to identify mQTL (module QTL). Three modules (B, C, and K) had QTL that overlapped with HrI1 (Table 2). These modules included ones enriched for a number of cell-adhesion and morphogenesis/development transcripts (module B) and immune system response phenotypes (module K), while module C was not enriched for any specific functional categories.
We grouped the eight founder alleles at HrI1 by their estimated effects on each phenotype [49], [54], [55]. Alleles from five strains (A/J, C57BL/6J, 129S1/SvImJ, NOD/ShiLtJ and WSB/EiJ) affected the host response similarly and were associated with decreased influenza resistance (i.e. higher titers, higher weight loss, more pathology), increased module K and decreased expression of modules B and C. A/J, C57BL/6J and WSB/EiJ had previously been identified [27], [56] as having nonfunctional Mx1 alleles. In contrast, the NZO/HILtJ and PWK/PhJ alleles within the pre-CC population shared similar effects and increased influenza resistance. Previously, CAST/EiJ was characterized as having a full length Mx1 allele, based on analysis of portions of the Mx1 locus [56]. Despite the presence of a full length transcript, the effect of the CAST/EiJ allele across the pre-CC population was intermediate in conferring resistance in our QTL models. Pre-CC mice with the CAST/EiJ allele showed low-to-intermediate weight loss. In contrast, these animals had viral titers that were intermediate between animals with nonfunctional Mx1 alleles and those with a NZO/HILtJ or PWK/PhJ allele (Figure 4). These three functional groups held true when considering animals from the eight founder strains. Importantly, animals from the CAST/EiJ strain showed intermediate weight loss, but had viral titers no different than founder strains with nonfunctional Mx1 alleles (Figure 1). We conclude that three Mx1 alleles segregate in the pre-CC population, with the CAST/EiJ allele being functionally distinct from the classical protective Mx1 allele, where this allele confers limited protection from viral replication, but does protect from virus-induced weight loss. We found no significant differences in Mx1 mRNA gene expression in the lung at two days post-infection using one strain from each allele group (C57BL/6J, CAST/EiJ and PWK/PhJ, Figure 4). C57BL/6J had the highest mean level of up-regulation, with CAST/EiJ intermediate and PWK/PhJ having the lowest level of expression. This suggests that the differences between the CAST/EiJ and PWK/PhJ alleles are due to coding changes within the gene and not variation in gene expression.
The sequence variation at the Mx1 locus in mouse is poorly understood despite its well-known role in influenza susceptibility. This is due in part to the presence of a deletion in the C57BL/6J strain (the mouse reference genome, see Materials and Methods) and the subsequent effect on the annotation of several of Mx1 exons in the mouse assembly (mm9) and in the Sanger Institute's Mouse Genomes sequencing project [57]. Therefore, we identified the genetic variants in all Mx1 exons in each of the eight founder strains by sequencing the Mx1 exons from each strain (see Methods). We found five distinct haplotypes (Figure 4, Table S6). The most common haplotype in the CC contains a large (>2 kb) deletion that spans three coding exons (9, 10, and 11) that are highly conserved among placental mammals. As previously described [27], this deletion leads to a frame shift and early stop codon in exon 12. This haplotype results in the presence of the same non-functional Mx1 gene in A/J, C57BL/6J, 129S1/SvImJ and NOD/ShiLtJ. We confirmed that WSB/EiJ also has a non-functional allele due to a nonsense mutation in exon 10 [56]. The other three strains have full length ORFs and each has a distinct protein sequence due to the different combinations of alleles at two non-synonymous SNPs. However, the genetic variants identified in our analysis and the regional assignment of sub-specific origin [47], [58] demonstrate that the CAST/EiJ strain has a divergent haplotype of Mus musculus castaneus origin while PWK/PhJ and NZO/HILtJ both have haplotypes of M. m. musculus origin. The three functional haplotypes are characterized largely by synonymous variation and variation in untranslated regions of the gene. There was a single amino acid substitution identified in NZO/HILtJ relative to PWK/PhJ and CAST/EiJ (Gly616Arg), and a single amino acid substitution identified in CAST/EiJ relative to NZO/HILtJ and PWK/PhJ (Gly83Arg). Although we cannot preclude transcriptional differences at different time points during infection from having a role in the functional differences between the three Mx1 alleles, the nonsynonymous substitution that is unique to the CAST/EiJ haplotypes is a strong candidate to explain the intermediate phenotype of the CAST/EiJ Mx1 allele. These results demonstrate our ability to identify: 1) a known IAV resistance locus with only one mouse per line, 2) the multiple phenotypes regulated by this locus, and 3) previously unidentified allelic variants due to the multiple alleles segregating within the pre-CC population.
We returned to our transcriptional data in an attempt to better understand how the CAST/EiJ Mx1 allele could contribute to protection from weight loss while showing high titers and severe inflammatory responses. Of the 11,700 transcripts that passed QA/QC and were not SNP impacted, we determined that 2156 transcripts (18.4%) had their expression levels significantly impacted by genotype at the most significant Mx1 marker (i.e. these transcripts had an expression, or eQTL at Mx1, Table S7), confirming the large role that Mx1 has on regulating the response to IAV infection. We grouped these transcripts based on their allele effects, specifically looking for those transcripts where (a) CAST/EiJ Mx1 alleles grouped with the resistant PWK/PhJ or NZO/HILtJ Mx1 alleles, or (b) where CAST/EiJ allele grouped with susceptible A/J, C57BL/6J, 129S1/SvImJ, NOD/ShiLtJ and WSB/EiJ Mx1 alleles. A total of 307 transcripts with an eQTL at Mx1 (14.2%) had allele effects consistent with CAST/EiJ grouping with the resistant PWK/PhJ and NZO/HILtJ alleles, while 1207 transcripts with an eQTL at Mx1 (55.9%) had allele effects consistent with CAST/EiJ grouping with the susceptible A/J, C57BL/6J, 129S1/SvImJ, NOD/ShiLtJ and WSB/EiJ alleles.
Those transcripts where CAST/EiJ grouped with the susceptible alleles showed significant enrichments for a large number of GO terms (Table S8), including biosynthesis and biogenesis processes (upregulated in the lines with a PWK/PhJ or NZO/HILtJ allele) and a highly diverse array of inflammatory, apoptotic, chemotactic, cell growth and hematologic-based terms (upregulated in the lines with A/J, C57BL/6J, 129S1/SvImJ, NOD/ShiLtJ, CAST/EiJ and WSB/EiJ alleles). In contrast, those transcripts where CAST/EiJ was grouped with the resistant PWK/PhJ and NZO/HILtJ alleles showed a much more limited enrichment, with mainly cytokine and T-cell processes (downregulated in lines with CAST/EiJ, PWK/PhJ and NZO/HILtJ alleles) being enriched.
Despite the large effect of Mx1 on influenza response, there was large phenotypic variation within both the functional and non-functional Mx1 allele classes. This suggested the presence of modifier alleles segregating in the pre-CC population. To find these modifiers, we conducted additional genome scans after accounting for genotype at the most significant HrI1 marker (see methods), thereby controlling for Mx1 allele. This model accounted for the large effect of HrI1 and resulted in a single significant QTL, HrI2, on chromosome 7 (1.5 LOD interval: 89130587-96764352), that explained 9.7% of the variation in D4 weight (Figure S2). This region is annotated as containing 69 genes and 10 non-coding RNAs. Analysis of the allelic effects at HrI2 suggests that animals with an A/J allele showed less weight loss than other animals, and animals with a 129S1/SvImJ allele showed more weight loss than other animals.
To eliminate epistatic effects of the protective Mx1 genotype, we analyzed those Mx1-/- individuals in our pre-CC population, where this group consisted of 99 mice defined as having two Mx1 alleles coming from any of the A/J, C57BL6/J, 129S1/SvImJ, NOD/ShiLtJ or WSB/EiJ strains. Although this susceptible subpopulation still showed a wide range of phenotypes (Table S9), it was skewed towards increased disease-associated phenotypes. The correlations between weight loss, viral replication, pathology and aspects of the immune cell infiltrate were weaker than those seen across the whole population (Table S10). Specifically, while aspects of pathology and immune cell infiltrate remained correlated with each other, we observed reduced correlations between titer and pathology, titer and inflammation, and clinical disease and pathology. We also reexamined the relationship between titer, airway inflammation and weight loss, to determine if our earlier observation, which linked both titer and airway inflammation as significant predictors of weight loss was independent of Mx1 status. Despite the reduced strength of relationships across the population, both titer and airway inflammation were still significant predictors of weight loss. Again, knowledge of both titer and airway inflammation was a better predictor of weight loss than either variable alone (based on both partial F tests and AIC, Table S3).
RNA of high quality was recovered from 60 mice within this Mx1-/- population, and we used WCGNA to cluster the 6,000 most variable transcripts across this population (4,933 of these 6,000 transcripts were also identified in the whole population analysis). Even in the absence of a large effect resistance gene, Mx1, we were able to group transcripts into functionally relevant co-expression modules. In total, this analysis identified eleven modules labeled M-W (Table S11). Again, modules were enriched for a wide range of functional terms, and showed little overlap between categories (Table S12). Eight modules (M-O, Q, T-W) were significantly correlated with clinical disease and/or viral replication, being enriched for T-cell processes (module M), inflammatory responses (module N), and signaling processes (module O). Module Q (enriched for cell cycle processes) was exclusively associated with some virus-induced inflammation, and two modules had no clear relationships with any phenotypes (enriched for sensory and neurological processes (module U) as well as metabolic and biosynthesis (module V), Table S13). Absence of an mQTL overlapping the Mx1 region (see below) indicates that, as expected, the effect of the Mx1 locus on the coexpression network has been ameliorated in this population. This indicates, along with the continued, albeit, weaker associations between the modules and phenotypes that performing a WGCNA analysis conditioning on the Mx1 allele group provides a way to highlight additional diseases-associated genetic regulation of transcript expression.
When we conducted QTL mapping in the Mx1-/- subpopulation, we identified a significant QTL, HrI3, which explained 29.73% of the variation in Pulmonary Edema on chromosome 1 (7.31 Mb, 21767867-29085401) annotated as containing 24 genes and 11 non-coding RNA (Table 2, Figure S3). Additionally, we identified a suggestive QTL, HrI4, which explained 22.77% of the variation in airway neutrophils on chromosome 15 (77427235-86625488), a 9.19 Mb region annotated as containing 206 genes and 35 non-coding RNAs (Table 2, Figure S4). In contrast to our results with the whole pre-CC population, we were not able to identify any mQTL contributing to variation in module expression within this Mx1-/- subpopulation.
In order to confirm the role of HrI3 in contributing to control of pulmonary edema, we challenged a new set of female animals from a small set of completely inbred CC lines with IAV. These animals were homozygous for various founder alleles across the entire candidate region for HrI3 and founder strain alleles were each represented by two CC lines (e.g. two lines that vary across the rest of their genome both share the WSB/EiJ allele at HrI3). We examined the severity of pulmonary edema in these animals at four days post infection. Founder strain alleles at HrI3 significantly affected pulmonary edema (F3,16 = 8.48, p = 0.0013, Figure 5), validating the role of this genome region in the host response to IAV.
Having identified three novel QTL our next objective was to narrow down QTL regions to specific candidate genes/features. We used the estimated allele effects along with the whole genome sequences of the founder strains [57] to narrow the list of candidate genes within each interval (see methods).
When a particular allele underlies a QTL, it is due to a causal genome feature (e.g. SNP, insertion/deletion) corresponding to that allele, contrasted with the other alleles in the cross. In the simplest case of one allele contrasted with the other seven this means that a private genome feature in the single strain is causative for the QTL. Under more complex scenarios (e.g. two strains contrasted with six), because CC mice share common ancestry due to their natural history and the unique history of laboratory mice [47], [58], we assume that causal alleles are often shared across mouse strains. That is, if two strains are segregating from the other six, it is likely due to a common feature these two strains privately share. We used the allele effects plots (Figures S2, S3, S4) to group founder strains underneath QTL peaks into two groups based on the largest difference between groups (Note that for completely inbred lines, phenotype-by-genotype plots would provide similar information. For the incompletely inbred pre-CC mice, with up to 36 allele combinations at each locus, PxG plots are difficult to interpret). For each group, we identified the regions in which all strains were identical or nearly identical (≥98%). Then we excluded regions that were not unique to the allele group (e.g. where two causative alleles had different SNP patterns). Using this approach, we narrowed the candidate regions for HrI3 (Pulmonary Edema: NZO/HILtJ and WSB/EiJ alleles reducing edema), from 7.31 Mb to 1.01 Mb, containing 10/24 genes and 1/10 annotated non-coding RNAs (Table 3). HrI4 (Airways Neutrophils: C57BL6/J, NZO/HILtJ and PWK/PhJ increasing infiltration) was similarly reduced from 9.19 Mb to 91 kb, including 12/206 genes and 2/35 non-coding RNAs (Table 3). HrI2 represented a case where a single founder allele associated with either increased resistance (A/J) or susceptibility (129S1/SvImJ) contrasted with the other six strains showing an intermediate phenotype. We were therefore looking for individual SNPs (and not regions of difference) that differentiated A/J or 129S1/SvImJ from the other strains. We identified 144 private A/J SNPs or small in/dels, and 611 private 129S1/SvImJ SNPs or small in/dels (out of a total of 106,684 SNPs or small in/dels in the region). These SNPs occur in or near 28 genes (7 genes unique to A/J, 13 unique to 129S1/SvImJ, and 8 overlapping between the two, Table 3).
In all three of these cases, the high priority candidate genes we identified covered a range of biological functions, including a large number with no annotated functions. While no obvious candidates jump out with HrI3, HrI4 includes Grap2, involved in leucocyte specific signaling [59]. Similarly, HrI2 includes the chemoattractant/T-cell modulator Il16 [60] as well as Nox4, which is potentially involved in production of reactive oxygen species and interacts with the TLR4 pathway [61].
The host response to infection represents a complex set of interacting phenotypes, where variation in these phenotypes is likely influenced by interactions between multiple polymorphic genes as well as other factors (specific virus-host interactions, environment, exposure, age). While reverse genetics approaches have afforded insight into the role of viral genes in infection [62]–[64], well defined models do not exist for understanding how polymorphic host genes interact to regulate host response to infection. A number of mouse models including gene specific knockouts and transgenic lines [28]–[30], [32], [34], [35], [39], [65], panels of genetically distinct mouse lines [44], [66], and classical RI panels [42], [45], [46], have been used to provide key insights into the role of specific genes in pathogenesis, however, these systems do not accurately reflect the situation in outbred populations. While these systems either interrogate the role of specific genes in the context of a single genetic background (e.g. knockouts) or analyze the impact of two variant alleles (e.g. classic RI panels) on disease pathogenesis, in genetically complex populations, such as humans, disease outcomes are likely determined by interactions between multiple polymorphic genes, with multiple polymorphic alleles at these loci. Therefore, we chose to use the pre-CC population to assess how genetic polymorphisms impact the host response to influenza infection in a population of animals that more closely represents the genetic diversity found in outbred populations. Our results show that even within the constraints of this pre-CC study (i.e. one animal/incipient line, single time point), we were able to uncover underlying relationships between host responses to infection, identify new disease phenotype combinations not present within the founder strains, and identify novel QTL impacting aspects of the host response to infection, suggesting that the CC panel represents a powerful system for studying pathogen interactions within genetically complex populations.
Host genetic polymorphisms have been shown to contribute to differential disease outcomes, and evidence exists for influenza [16], [22], [23], [67] that suggests host genetic variants are important regulators of influenza pathogenesis. The identification of a QTL of major effect sitting over the anti-influenza gene Mx1 was not surprising. Mx1 is known to strongly inhibit influenza virus replication, limiting the resultant IAV-induced disease symptoms in mice [68]. As was to be expected, within the pre-CC population, functional Mx1 alleles reduced viral titers, weight loss and clinical disease, inflammation and pathology. Mx1 also acted to influence the expression levels of a large number of transcripts. While it is unlikely a direct transcriptional regulator, Mx1's potent ability to inhibit IAV replication likely alters the signaling environments and host response pathways triggered in response to infection.
Within human populations, it is possible for multiple alleles to exist at any given locus. Similarly, at any locus within the pre-CC population, up to eight distinct alleles exist. In addition to increasing the probability of having functionally variant alleles segregating within the population, multiple alleles at a locus can give rise to distinct phenotypic outcomes across the population. The effects of this allelic variation can clearly be seen when considering Mx1. A total of 5 distinct Mx1 haplotypes exist in the pre-CC population, and they can be grouped into three functionally distinct alleles based on their effects during influenza infection. Of particular interest is the CAST/EiJ allele, which disassociates the effects of Mx1 on control of viral replication from its' ability to protect from a clinical disease aspects. We utilized the large number of transcripts that had an eQTL at Mx1 to better understand potential ways in which the CAST/EiJ allele might provide clinical protection while being unable to control IAV replication. We identified a set of transcripts that were significantly upregulated in those individuals with defective Mx1s, but were downregulated in individuals with CAST/EiJ, PWK/PhJ and NZO/HILtJ Mx1 allele). These transcripts included sets of inflammatory and immune related transcripts, such as SOCS3, Irf1, and Interferon-gamma. GTPases, such as Mx1, are important in a number of signaling and protein production activities [69], [70], and it is possible that the signaling activities of Mx1 are responsible for regulating specific transcripts, such as those mentioned above, in limiting clinical disease independent of Mx1's previously described anti-IAV activities. However, additional studies are needed to better define whether the differential effect of the CAST/EiJ Mx1 allele are due to Mx1-associated signaling or more subtle effects on viral replication which subsequently affect inflammation and disease.
In a more general sense, these results illustrate the advantages provided by using a system such as the CC compared to using classical inbred strains such as the founder strains. Because of the inherent genome structure of the founder strains [47], it would be difficult to differentiate between the disassociated effects of the CAST/EiJ Mx1 allele we found in the pre-CC population and the alternate hypothesis that CAST/EiJ's Mx1 was completely non-functional, and that there was another polymorphic gene within CAST/EiJ that provided some protection from clinical disease. It is only through the recombination present within the CC, and the use of large populations of unique lines that can be evaluated within the CC that such hypothesis can be differentiated. This lack of structure across the genomes also allowed us to gain new insight into the relationships between coexpressed transcripts and disease outcomes, as well as the relationships between specific disease processes (see population-wide pattern section below).
Though the identification of Mx1 served to validate our mapping study, the presence of a large effect allele within genetically variable populations can mask those with smaller effect sizes. Indeed, this genetic architecture is common to pathogen resistance, as several other large effect genes have been identified for viral (e.g. flaviviruses [71], mouse CMV [72], norovirus [20], HIV [18]), bacterial (NRAMP [73]) and parasitic (malaria [74]) diseases. Due to these genes of large effect, many studies of these pathogens have been conducted within susceptible models (e.g. mouse models of influenza infection are almost universally Mx1-/- models). It is likely that there are specific alleles influencing disease processes that only act in the context of the presence or absence of major resistance alleles (e.g. an allele that affected the degree of tissue repair would only act when there had been significant tissue damage, which a functional Mx1 allele would prevent). To address this issue, we conducted further genome mapping while accounting for Mx1 status using two complementary approaches: asking whether there are loci that act in addition to Mx1 in regulating disease-associated phenotypes (HrI2), and also asking if there are loci that act only in the highly susceptible Mx1 negative population (HrI3, HrI4). By doing so, we were able to identify three additional loci influencing weight loss, pulmonary edema and neutrophil infiltration into the airways, further validating the role of genetic variation at HrI3 in contributing to pulmonary edema differences in a separate set of fully inbred CC animals. The development of pulmonary edema [6], [75], as well as neutrophilic infiltration [76] have been shown to contribute to disease severity and long-term lung disease in the human population. Our results show that host genetic variation not only contributes to direct responses to infection, but also to other aspects of the host response that can lead to long-term complications following infection.
In addition to allowing us to identify novel QTL that impacted the host response to IAV, our mapping using only Mx1-/- animals also allowed us to compare our study to other studies using QTL mapping within Mx1-/- panels [42], [45], [46]. While these studies all identified QTL contributing to IAV responses, there was no overlap between these QTL and the ones we discovered. This result is unsurprising given the differences in virus strains, phenotypes measured, and polymorphisms within the two panels. Nevertheless, the aggregate of these studies further strengthens the idea that virus-host interactions are highly complex, and that polymorphic host genes are critical for numerous responses. Our results further emphasize the unique genetic interactions that occur in specific sub-populations of infected individuals that regulate disease processes. While we limited our analysis to the Mx1-/- animals in the pre-CC population, due to sample size, it is likely that similar QTL can be identified in future work using animals with functional Mx1.
Ultimately, the goal of QTL mapping studies is to identify the causal polymorphic genes or genome features responsible for variation in disease processes. A variety of studies have used transcriptional data [42], [49] to narrow QTL regions into candidate genes. However, transcriptional analysis can be confounded by the dynamic nature of transcriptional responses, and can also be confounded by SNPs residing underneath the expression probes that can impact binding [77], [78]. An alternate approach, taking advantage of the allelic complexity of the CC is the use of the sequence of the founder strains [57] to interpret the mapping results and to prioritize candidate genes within QTL. This approach is quite powerful; as causative polymorphisms must be contained within QTL regions, and has been used effectively in other pre-CC studies [49], [55]. However, Mx1 provides a cautionary note for regions of the genome in that there is a large in/del differentiating the eight CC founders. In fact, C57BL/6J, and thus the assembly, has the deletion. In such regions, functional genomic features may be misannotated and more importantly the genetic variants present in founders without the deletion is currently not annotated. This lack of annotation makes it difficult to conclusively dissect the polymorphisms within these regions that might cause phenotypic variation across the population, and currently requires more intensive sequencing efforts on a case-by-case basis. The Mx1 result illustrates the importance of improving the annotation of genetic variants in the mouse genome. It also suggests that in addition of the processed lists of SNPs and in/dels available at the Mouse Genome Projects from the Sanger Institute, the analysis of the allele effect in QTL intervals should be analyzed by searching for signatures of structural variation that might be present in the raw reads. As further QTL analyses are undertaken within the CC system, approaches to narrow down onto candidate genes and polymorphisms will need to be further developed, likely integrating both transcriptional and refined analysis of sequence data to account for other potential causative genome features. These approaches will be facilitated by the interrogation of transcriptional activity at multiple time points in completely inbred CC lines.
The pre-CC experiment also allowed us to identify specific CC lines that might be useful models for specific disease phenotypes (e.g. animals with high titers but little/no clinical disease as super-spreaders), as well as other interesting relationships between disease components across the pre-CC population. For example, while lung hemorrhage is a clinically important influenza-associated phenotype [79], we found that lung hemorrhage was only correlated with alveolar inflammation and not with other metrics of viral spread or disease severity. This result bears further study, but it suggests that while hemorrhage indicates a severe response to influenza infection, hemorrhage is governed by processes that are largely disassociated from those processes contributing to overall severity of infection at least through day 4. Results such as this further highlight the complexity of the host response to infection, and the need to consider genetically diverse populations when attempting to understand disease processes. While the individual pre-CC mice used within this study were insufficient to develop these new models of disease processes, as CC lines become increasingly available [52], utilization of specific CC lines with unique responses to infectious diseases might well become a critical resource for uncovering avenues of viral pathogenesis in specific subpopulations.
In addition, we identified transcriptional modules that correlated with overall disease severity, or with specific aspects of the host response to infection (e.g. inflammatory components, clinical disease). Recent efforts from a number of groups [10], [80] have focused on identifying markers associated with different disease states (e.g. protective vaccine responses, asymptomatic individuals) across human cohorts. While the nature of the pre-CC study (restricted to a single time-point) makes it difficult to draw broad conclusions about these results, it does suggest that there are unique transcriptional signatures relating to different aspects of the host response to infection. Future studies leveraging the full power of the CC (identical animals at different time points, compared to baseline transcriptional levels) will provide the opportunity for identification of molecular signatures of different disease-associated phenotypes, informing us both of the mechanisms through which these processes are occurring, as well as providing non-invasive diagnostic markers of various disease-related phenotypes.
The findings of this study also provide new insights into the relative contribution of viral replication versus virus-induced inflammation in the pathogenesis of influenza infection. There is conflicting evidence from a variety of in vivo studies as to the importance of virus-induced inflammation [28], [39], [64], [81], [82] and control of viral load [30], [39], [66] on disease severity. However, these studies have all used different mouse strains, influenza strains and experimental conditions, making direct comparisons difficult. The novel allele combinations in the pre-CC population allowed new insight by dissociating phenotypes that were correlated in the founder strains. This allowed us to assess the relative contribution of inflammation and viral replication on disease outcome. Consistent with the complex nature of virus-induced disease, we found that both viral replication and levels of inflammation were predictive of disease outcome independently of one another. Although the overall correlations between viral titer, weight loss and inflammation were reduced within the Mx1-/- subpopulation of pre-CC animals, we again found that both viral replication and levels of inflammation together were better predictors of disease outcome than either was alone. Consistent with this analysis, we identified CC lines that showed high levels of replication and weight loss, but little inflammation as well as lines with excessive inflammation and weight loss, but low viral titers suggesting that multiple pathways can lead to similar clinical disease outcomes across a genetically diverse population. Future studies utilizing this model with fully inbred CC lines, will allow us to more fully evaluate how the kinetics, magnitude, and duration of viral replication and inflammation contribute to disease outcome over the temporal course of the infection. These results illustrate the potential and the power of using genetically diverse mice to study the relative contribution of specific aspects of the pathogen and host response which together drive disease outcome.
There is an increased appreciation for the role that host polymorphisms play in the host response to infectious diseases [83]. However, for a number of reasons, Genome Wide Association Studies (GWAS) of responses to acute infectious diseases within the human population have been difficult to conduct [84]. While future studies using the CC panel will allow for the evaluation of multiple animals/line and allow for integration of information across multiple timepoints, for this study we had access to only a single animal/line within the pre-CC population making it similar in design to GWAS and raising concerns about our ability to identify host response QTL within this study. However, the identification of several disease associated QTL, including a QTL containing the known IAV associated resistance gene Mx1, even when using a single mouse/time point, suggests that the CC lines represent a robust system for identifying polymorphic genes that regulate host responses to infectious diseases. As the CC can be recreated and manipulated, it will increasingly become a useful tool to a) identify candidate genes and pathways for more targeted association studies within human populations, and b) allow us to increase our understanding of how critical demographic and environmental factors, as well as specific genetic subpopulations impact some of the variability in GWAS studies of acute infectious diseases in humans.
Host genetic variation clearly plays an important role in regulating differential response phenotypes to infectious disease progression. Herein, we provide proof of concept and a framework for identifying the role of polymorphic genes on microbial pathogenesis using a genetically diverse population: underlying relationships between different disease phenotypes, genetic control of phenotypes following infection (both those of large effect, as well as those that modulate the host response), and transcriptional profiles that related to specific disease-associated phenotypes. In summary, this study shows that a genetically complex in vivo model represents a useful system for modeling pathogen interactions within genetically diverse populations and identifying novel genetic loci controlling multiple aspects of disease pathogenesis. Though this study had clear limitations, the pre-CC population provided the appropriate framework to develop the methodological approaches that resulted in the identification and prioritization of genes within novel disease loci. These results strongly support the hypothesis that studies using the fully inbred CC lines, with the use of replicate animals and evaluation of phenotypic variation during influenza infection over time, will be even more successful in identifying polymorphic genes that regulate multiple disease associated phenotypes including those phenotypes associated with adaptive immune responses and disease recovery. Furthermore, through careful selection of CC lines, studies can be designed to specifically investigate how interactions between allelic variants in two or more genes interact to influence complex phenotypic outcomes during infection.
Mouse studies were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All mouse studies were performed at the University of North Carolina (Animal Welfare Assurance # A3410-01) using protocols approved by the UNC Institutional Animal Care and Use Committee (IACUC). All studies were performed in a manner designed to minimize pain and suffering in infected animals, and any animals that exhibited severe disease signs was euthanized immediately in accordance with IACUC approved endpoints.
8–16 week old female animals from the 8 founder strains (A/J, C57BL/6J, 129S1/SvImJ, NOD/ShiLtJ, NZO/HILtJ, CAST/EiJ, PWK/PhJ, and WSB/EiJ) were derived from the Jackson labs (jax.org), and bred at UNC Chapel Hill under specific pathogen free conditions. 8–12 week old female Pre-CC mice were bred at Oak Ridge National Laboratories under specific pathogen free conditions, and transferred directly into a BSL-3 containment laboratory at UNC Chapel Hill. Inbred CC mice were bred at UNC Chapel Hill under specific pathogen free conditions. All experiments were approved by the UNC Chapel Hill Institutional Animal Care and Use Committee.
The mouse adapted influenza A strain A/PR/8/34 (H1N1) was used for all infection studies. A/PR/8/34 stocks were made by infection of 10-day old embryonated chicken eggs. MDCK cells grown in high glucose Dulbecco's modified Eagle's medium (10% FBS, 1% Penicillin-Streptomycin) were used for titering virus.
Animals were lightly anesthetized via inhalation with Isoflurane (Piramal, Bethlehem, Pa). Following anesthesia, animals were infected intranasally with 5×10∧2 pfu of PR8 in 50 µL of phosphate buffered saline (PBS), while mock infected animals received only 50 µL of PBS. Animals were assayed daily for morbidity (determined as % weight loss), mortality and clinical disease scores. At 4 days post infection, animals were euthanized via Isoflurane overdose and cardiac puncture, animals were assessed for gross pathology (lung hemorrhage and edema) and tissues were taken for various assays.
MDCK cells were seeded into 96 well plates at a density of 1.5×10∧5 cells/well in DMEM (10% FBS, 1% Pen-strep) and incubated at 37 degrees overnight. Cells were washed 2 times with PBS, before addition of 100 µL of DMEM to each well. Media was removed from all wells in the 1st column of the plate, and 146 uL of lung homogenate in DMEM was added to these wells (each biological sample was added to 4 wells). Serial dilutions of 46 µL (0.5 log dilutions) were carried out across the plate. Plates were incubated at 37 degrees C for 1 hour, inoculum was removed and 150 µL of serum free DMEM with 1 µg/mL of trypsin was added to each well. Plates were then incubated at 37oC for 3 days. Media was then removed, and wells were stained with a 1% Crystal Violet solution. The stain was washed off with water. Titer is determined as follows:(1)Where Xp is the last dilution where all of the replicates of a given sample are positive, D is the serial dilution log and Sp is the sum of the proportion of replicates at all dilutions where positives are seen (starting with the Xp dilution).
The right lung was removed and submerged in 10% buffered formalin (Fischer) without inflation for 1 week before being submitted to the UNC Linberger Comprehensive Cancer Center histopathology core for processing. Two 5 micron thick Hematoxylin and Eosin stained lung sections (step-separated by 100 microns) were blind-scored by microscopic evaluation performed by two independent scorers for a variety of metrics relating to the extent and severity of immune cell infiltration and pathological damage on a 0–3 (none, mild, moderate, severe) scale.
For detection of influenza virus antigen, we used serial sections from formalin-fixed, paraffin-embedded lung samples. After deparaffinization and rehydration, antigen retrieval was performed using 0.1% protease (10 min at 37°C). Endogenous peroxidase was blocked with 3% hydrogen peroxide and slides were briefly washed with phosphate-buffered saline (PBS)/0.05% Tween 20. Mouse anti- influenza virus nucleoprotein (clone Hb65, ATCC) and horseradish peroxidase labeled goat anti-mouse IgG2a were used for 1 h at room temperature. Peroxidase activity was revealed by incubating slides in 3-amino-9-ethylcarbazole (AEC, Sigma) for 10 minutes, resulting in a bright red precipitate, followed by counterstaining with hematoxylin. Tissue sections from non-infected BALB/c mice and mouse IgG2a isotype antibody (R&D) were used as negative controls. The extent of influenza viral antigen spread across these slides was then scored in a blinded fashion on a 0–3 scale.
At 4 days after infection, mice were killed and lung tissue harvested and placed in RNAlater (Applied Biosystems/Ambion, Austin, TX) and stored at −80°. The tissues were subsequently homogenized, and RNA extracted as previously described [85]. RNA samples were spectroscopically verified for purity, and the quality of the intact RNA was assessed using an Agilent 2100 Bioanalyzer. cRNA probes were generated from each sample by the use of an Agilent one-color Low Input Quick Amp Labeling Kit (Agilent Technologies, Santa Clara, CA). Individual cRNA samples were hybridized to Agilent mouse whole-genome oligonucleotide 4×44 microarrays according to manufacturer instructions. Samples from individual mice were evaluated to enable examination of animal-to-animal variation as part of the data analysis. Slides were scanned with an Agilent DNA microarray scanner, and the resulting images were analyzed using Agilent Feature Extractor version 8.1.1.1. The Agilent Feature Extractor software was used to perform image analysis, including significance of signal and spatial detrending and to apply a universal error model. For these hybridizations, the most conservative error model was applied. Raw data were then loaded into a custom-designed laboratory information management system (LIMS). Data were warehoused in a Labkey system (Labkey, Inc., Seattle, WA). Raw array data are available from GEO with accession GSE30506.
The Agilent arrays were background corrected by applying the Normal-Exponential convolution model [86] and normalized using quantile normalization [87] with the Agi4×44PreProcess Bioconductor package (www.bioconductor.org). The probes were filtered requiring that all probes meet specific QC requirements (probe intensity had to be found, well above background, not saturated, and not be nonuniformity or population outliers as defined by the standard parameters in Agi4×44PreProcess package) for all samples. Differential expression analysis was performed using the LIMMA Bioconductor package [88], and the false discovery rate was calculated using the qvalue Bioconductor package [89]. Probes were mapped to the mm9 genome using BLAT [90] requiring at least 98% identity. Probes that did not map, mapped to multiple locations equally well, or contained a high confidence single nucleotide polymorphism (SNP) from one of the eight progenitor strains from the Sanger Institute/Wellcome Trust mouse sequencing project [57] in the probe sequence were excluded from analysis. There were 11,700 probes passing QC and not potentially impacted by a SNP. The Gene Ontology (GO) analysis was performed using the standard hypergeometric test from the Gostats Bioconductor package [91] with a universe consisting of the unique genes from the probes entered into the DE analysis. Only the Biological Process subset of the Gene Ontology was used for testing. The Benjamini and Yekutieli false discovery rate (FDR) [92] was computed for the P-value distribution for this analysis to address dependencies inherent from the hierarchical/nested structure of the GO categories.
For both the full analysis and the Mx1-/- analysis, six thousand probes were chosen to be entered into the analysis based on both high variability across samples as well as a measure of how connected they were [93]. Arrays were preprocessed separately for both analyses. These probes were used for the formation of coexpression modules through the weighted gene coexpression network analysis (WGCNA) procedure [93], [94]. Module formation was signed [95] and was carried out using the dynamicTreeCut R package [96] with pruning carried out based only on the dendrogram. All modules were checked for statistical significance through a permutation procedure whereby the mean topological overlap of those probes within a module was compared to the mean topological overlap of 10,000 random modules of the same size chosen from the initial set of 6,000 probes. Using the WGCNA package the module eigengene (first principle component of the expression matrix) for each module was computed [97]. The module eigengene can be viewed as the representative profile that summarizes the module expression profile. The module eigengene was first correlated with the clinical traits using Pearson's correlation with P-values provided as Student's asymptotic P-value. The module QTL scan was carried out similar to below but using the eigengene for each module as a phenotype. Specifically, each eigengene was regressed on the expected haplotype contribution from each of the eight founding inbred strains. Significance was assessed using the –log10 P-values (using a Bonferroni type correction (α = 0.05)) from the model and support intervals were computed using the 1.5 LOD drop method [54]. This method of defining an mQTL is essentially the same as a previous study using F2 intercrosses [53]. A related approach looking at overrepresentation of eQTLs in a module [98] could potentially be sensitive to significance cutoffs and module size and necessitates a full eQTL scan.
Genotyping and haplotype reconstruction were done as described in [49]. Briefly, each pre-CC animal was genotyped using Mouse Diversity [47] test A-array at 181,752 well performing SNPs which were polymorphic across the founder strains. Once genotypes were determined (Dataset S2), founder strain haplotype probabilities were computed for all genotyped loci using the HAPPY algorithm [99]. Genetic map positions were based on the integrated mouse genetic map using mouse genome build 37 [100].
Genome scans were run as described in [49]. Briefly, QTL mapping was conducted using the BAGPIPE package [101] to regress each phenotype on the computed haplotypes in the interval between adjacent genotype markers, producing a LOD score in each interval to evaluate significance. Genome-wide significance was determined by permutation test, with 250 permutations conducted per scan.
A more complex model was also used to control for Mx1 status, whereby the null model included the haplotype information from the most significant marker at the Mx1 locus (JAX00072951). LOD scores are then computed for each haplotype interval based on the increase in fit of genotype to phenotype when Mx1 haplotype is already taken into account.
For the likely regions of identified QTL peaks, SNP data for the eight founder strains from the Sanger mouse genomes project was downloaded, and filtered to include only homozygous calls. In the case where a single founder strain allele drives a QTL peak, all private SNPs for that strain are candidates for the observed phenotype. In the case where multiple founder strains drive a QTL peak, the most likely hypothesis is that the causative polymorphism exists in a region of shared ancestry between these founder strains. SNPs were categorized into 3 classes: Consistent with a shared ancestry (SNPs where the driver strains share a private SNP), Inconsistent with a shared ancestry (SNPs where the driver strains share different alleles with other strains), and Uninformative with regards to ancestry (SNPs private to a single strain and SNPs shared by driver strains as well as others). Candidate regions were defined as regions containing at least one consistent SNP, and were bounded by the 1st nucleotide after the last inconsistent SNP until the last nucleotide before the next inconsistent SNP, and also had to exceed 100 base pairs in length. We identified all annotated genes and non-coding RNAs that were within 500 bases of, or in consistent regions and classify these as our likely candidates.
To characterize the genetic variation at the Mx1 locus we combined data from the mouse genome assembly, the Sanger Institute's Mouse Genomes sequencing project and a mouse full-length Mx1 cDNA clone (CT010406). By aligning the cDNA to the Mx1 genomic locus we identified three missing exons (exons 9, 10 and 11). We used this information to design primers to amplify and sequence every exon and the span the deletion boundaries in each strain (Table S14). The deletion occurs between positions 97674078 and 97674079 of the reference on chromosome 10. The deletion is 921 bp upstream from exon 12 and 302 bp downstream of exon 8. Chromosome walking was used to partially sequence the introns missing in the assembly. All sequence variants have been submitted to NCBI (GenBank Accession numbers: JQ860141-JQ860220).
Whole lung RNA from 8 week old female C57BL/6J, CAST/EiJ, and PWK/PhJ that had been either mock or flu infected were isolated using Trizol (Invitrogen, Carlsbad CA), and following their protocol. One microgram of total isolated RNA from each sample was reverse transcribed using MMLV-RT (Promega, Madison WI), and following their protocol. We ran TaqMan real time PCR with two primer-probe pairs (Applied Biosystems Foster City, CA): Mm01217999_m1 to amplify the 5′ gene region of Mx1 transcripts, and Hs03928985_g1 to amplify 18s mRNA. Fold-induction was calculated as the difference in expression levels for infected animals as compared to their strain-matched mock animals.
|
10.1371/journal.ppat.1007338 | Klebsiella pneumoniae infection of murine neutrophils impairs their efferocytic clearance by modulating cell death machinery | Neutrophils are the first infiltrating cell type essential for combating pneumoseptic infections by bacterial pathogens including Klebsiella pneumoniae (KPn). Following an infection or injury, removal of apoptotic infiltrates via a highly regulated process called efferocytosis is required for restoration of homeostasis, but little is known regarding the effect of bacterial infection on this process. Here we demonstrate that KPn infection impedes the efferocytic uptake of neutrophils in-vitro and in-vivo in lungs by macrophages. This impaired efferocytosis of infected neutrophils coincides with drastic reduction in the neutrophil surface exposure of apoptosis signature phospholipid phosphatidyserine (PS); and increased activity of phospholipid transporter flippases, which maintain PS in the inner leaflet of plasma membrane. Concomitantly, pharmacological inhibition of flippase activity enhanced PS externalization and restored the efferocytosis of KPn infected neutrophils. We further show that KPn infection interferes with apoptosis activation and instead activates non-apoptotic programmed cell death via activation of necroptosis machinery in neutrophils. Accordingly, pharmacological inhibition of necroptosis by RIPK1 and RIPK3 inhibitors restored the efferocytic uptake of KPn infected neutrophils in-vitro. Importantly, treatment of KPn infected mice with necroptosis inhibitor improved the disease outcome in-vivo in preclinical mouse model of KPn pneumonia. To our knowledge, this is the first report of neutrophil efferocytosis impairment by KPn via modulation of cell death pathway, which may provide novel targets for therapeutic intervention of this infection.
| Inflammatory diseases caused by infectious or sterile injuries are often characterized by pathological accumulation of dead or dying infiltrating cells. Pneumonic sepsis caused by Klebsiella pneumoniae (KPn), an opportunistic pathogen, has similar etiology, however, the underlying mechanism remains unknown. Here we report that KPn infection subverts a protective host process termed efferocytosis, by which the phagocytic cells engulf and clear dead/dying cells thereby resolving inflammation and infection. Our results show that KPn infected neutrophils are cleared less efficiently via efferocytosis than the uninfected neutrophils. Mechanistic studies implicated a reduced exposure of “eat me” signal phosphatidyleserine (PS) via increased flippase activity and skewing of cell death pathway toward necroptosis in impaired efferocytosis of infected neutrophils. Accordingly, pharmacological reversal of PS exposure by flippase inhibition, treatment with necroptosis inhibitors restored the efferocytic clearance of KPn infected neutrophils, and improved the disease outcome in a preclinical model of pneumonic sepsis. To our knowledge this is the first report of KPn subversion of efferocytic clearance of neutrophils by impairing pro-efferocytic apoptotic signatures and activation of necroptosis machinery. This could lead to novel therapeutic targets against KPn infection and associated inflammation in pneumonic sepsis.
| Pneumonia is the most frequent cause of sepsis [1–3], which is one of the oldest and most elusive syndromes and a major challenge in medicine [4]. With no effective therapies there are over 750,000 cases of sepsis each year in the United States alone, which accounts for 10% of all ICU patients, leading to a mortality rate between 20–50% depending on certain risk factors [5, 6]. In particular Klebsiella pneumoniae (KPn), an opportunistic pathogen, accounts for 5–20% of all Gram-negative sepsis cases [1, 3]. A notable emergence of antibiotic resistant strains of KPn in clinical settings has caused concerns over an already dwindling armamentarium of antibiotics. Thus, an understanding of host immune responses and pathogen-mediated manipulation thereof will likely provide novel therapeutic targets. In this regard, neutrophils are the first cell types to infiltrate the site of infection and contribute to the initial protective response. Indeed, in murine models of KPn infection, neutrophil-mediated responses are shown to be essential for initial control of the infection [7, 8]. We and others have shown that persistent accumulation of neutrophils and their over activation causes perpetuation of inflammation in pneumoseptic KPn infection [9, 10] [11–15]. Moreover, neutrophils have been reported to constitute a reservoir for this pathogen and aide in systemic dissemination of this infection [16]. This underscores the importance of neutrophil turnover in KPn pneumonia and sepsis.
Clearance of neutrophils by phagocytic cells, mainly macrophages, occurs via efferocytosis, which is a highly regulated receptor-dependent process [17, 18]. Aided by the action of phospholipid translocases such as flippases, apoptotic cells increase exoplasmic exposure of phosphatidylserine (PS), which is recognized as “eat-me” signal by macrophage cell surface receptors initiating their engulfment. The swift efferocytic clearance of infected and uninfected apoptotic cells prevents the release of pro-inflammatory mediators from dead cells as well as controls pathogens not destroyed through phagocytosis [19–21] [9, 22]. Apoptosis is thus considered an “immunologically silent” cell death mechanism [23]. Alternative cell death modalities, on the other hand, are typically characterized by rupture of outer membrane and release of intracellular contents recognized by the immune cells as danger signals eventuating in inflammatory responses [24]. Hence the immunological consequences of different cell death modalities differ, not only by virtue of their morphological features but also, by dictating the engagement of efferocytic cells. In that regard, modulating cell death modality may be a virulence mechanism of pathogens by which pathogens can subvert resolution of inflammatory response in their favor. Necroptosis is a programmed cell death which is mediated by receptor interacting protein kinase-3 (RIPK3) and its substrate mixed lineage kinase like (MLKL) [25]. Several studies have shown necroptosis in macrophages and epithelial cells to underlie inflammation and lung disease [26–32], however, activation of necroptosis in neutrophils and its implication in neutrophil turnover is not well understood and remains under active investigation. Moreover, while there is considerable research done on the efferocytosis pathways and its effect on sterile inflammatory disease development, studies on KPn subversion of this important protective host response are extremely limited. Lastly, whether KPn infection skews cell death to necroptosis in neutrophils and how it affects KPn pathogenesis and disease outcome is currently unknown.
In this study, we present key evidence to support several novel findings, which show that KPn infection impairs efferocytic uptake of neutrophils by inhibiting surface exposure of apoptosis signature PS exposure via modulation of flippase activity and activation of necroptosis pathway in neutrophils. With the use of pharmacological enhancers and blockers in-vitro and in-vivo in a preclinical mouse model of KPn pneumonia, our studies present evidence for a novel virulence mechanism of KPn which, by modulating programmed cell death, causes neutrophil turnover deficit. This defect can be targeted for therapies to treat KPn infection in the face of constant emergence of antibiotic resistant strains of this bacterium.
In order to determine the effect of KPn infection on efferocytosis of neutrophils, we compared the uptake of Carboxyfluorescein succinimidyl ester (CFSE) labelled uninfected or KPn infected neutrophils by macrophages using flow cytometry. Fig 1A (lower panel) shows the gating scheme using F4/80 and Ly6G antibodies to enumerate Ly6G-F4/80+CFSE+ macrophages that have internalized the CFSE-labelled neutrophils. Gating on singlets and Ly6G-negative cells allowed exclusion of adhered neutrophils since engulfed neutrophils will not be accessible to Ly6G antibody. As shown in Fig 1B, after 2 hrs of incubation with CFSE labelled neutrophils, 24.6% ± 4.1% macrophages stained positive for uptake of uninfected neutrophils. On the other hand, only 10.5% ± 1.8% macrophages had efferocytosed KPn infected neutrophils showing a 2.3 fold reduction of efferocytic uptake upon infection. This showed that infection of neutrophils with KPn caused a significant reduction in their efferocytic uptake by macrophages. Importantly, this infection-induced impairment was largely dependent on live bacteria and/or active bacterial replication as the neutrophils exposed to heat-killed KPn were efferocytosed more efficiently (efferocytic index of 70.9 ± 2.3, normalized to percentage uptake of uninfected neutrophils) as compared to those exposed to the live bacteria (efferocytic index 42.8 ± 1.4) (Fig 1C). Since bacterial factors likely involved in efferocytic impairment are not the focus of current study, we restricted further comparison between uninfected neutrophils and those infected with live bacteria only.
To investigate the physiological relevance of these findings we corroborated these data in a controlled in-vivo setting by intranasal administration of CFSE-labelled uninfected or KPn infected neutrophils in mouse lungs followed by flow cytometry analysis of alveolar macrophages (schematic shown in Fig 1D). Consistent with our ex-vivo results, significantly reduced numbers of Ly6G-F4/80+CFSE+ alveolar macrophages were recovered from the lungs of mice instilled with KPn infected neutrophils as compared to those that received uninfected neutrophils (Fig 1E). Together, these data strongly suggested that KPn infection significantly reduced the efferocytic clearance of neutrophils by macrophages in-vitro as well as in-vivo.
Since PS externalized to the plasma membrane serves as an “eat me” signal for efferocytosis of apoptotic cells by phagocytes [33], we examined if KPn infection modulates PS exposure on infected neutrophils. For this we compared the kinetics of PS externalization in uninfected and KPn infected neutrophils at various times by flow cytometry using Annexin V which specifically binds to PS. Propidium iodide (PI) was used to stain necrotic cells. As shown in Fig 2A and 2A’, approximately 20% of the uninfected neutrophils exhibited surface exposure of PS (Annexin V+PI- cells) after 30 min. of incubation, which increased over a period of 4 hrs, albeit the change over time was not statistically significant, which is typical of spontaneous apoptosis measured during first 4–6 hrs [34]. The KPn infected neutrophils, on the other hand, exhibited similar levels of PS exposure (Annexin V+PI- cells) as their uninfected counterparts up to 2hrs post-infection. However, by 3 hrs post-infection, a drastic reduction in PS exposure was observed in KPn infected neutrophils, as shown by significantly reduced numbers of Annexin V+PI- cells, which remained significantly lower than those in the uninfected cells at 4 hrs (Fig 2A-red bars and 2A’). On the other hand, the numbers of KPn infected late apoptotic (Annexin V+PI+ double positive cells) were significantly reduced at 4hrs, but not 3hrs p.i. (Fig 2A- green bars and 2A’). Together, these data suggested that uninfected neutrophils undergo spontaneous apoptosis indicated by kinetic increase in the numbers of early and late apoptotic cells expressing exofacial PS. KPn infection, on the other hand, arrests this process. Kinetic measurement of efferocytosis of uninfected and infected neutrophils revealed a direct correlation between the reduced PS exposure at 3hrs p.i. with a reduction in their efferocytic uptake, where the uptake of infected neutrophils was reduced by 2–3 fold at 3hrs and 4hrs p.i. compared to the uninfected cells (Fig 2B and 2B’). This suggested that reduced surface exposure of PS on KPn infected neutrophils correlates with a corresponding decrease in their efferocytic uptake by macrophages.
Flippases translocate PS from outer to inner plasma membrane leaflet in live cells and apoptotic PS exposure is accompanied by loss of flippase activity [35]. In order to understand the mechanism underlying the reduced exofacial PS exposure in KPn infected neutrophils, we examined flippase activity in these cells by measuring the internalization 1-oleoyl-2-{6-[(7-nitro-2-1,3-benzoxadiazol-4-yl)amino]hexanoyl}-sn-glycero-3-PS (NBD-PS). As shown in Fig 3A, the phospholipid flippase activity was significantly higher in KPn infected neutrophils compared to their uninfected counterparts at 3hrs and 4hrs p.i. This coincided with reduced PS exposure observed at these time points shown in Fig 2A. To confirm that increased flippase activity and concomitant reduced PS exposure is responsible for retarded efferocytosis of KPn infected neutrophils, we treated the cells with flippase inhibitor N-ethylmaleimide (NEM), which triggers flipping of PS to outer plasma membrane of cells [36–38]. Indeed, treatment with NEM dramatically increased the PS exposure on these KPn infected neutrophils to similar levels as observed in uninfected neutrophils (Fig 3B). As compared to 9.9% of the untreated KPn infected neutrophils, 52.1% ± 7.7% of KPn infected cells bound Annexin V after NEM treatment (Fig 3B). Having established that NEM treatment increased the surface exposed PS in KPn infected neutrophils, we next tested if efferocytic uptake of KPn infected neutrophils was improved following NEM treatment. As shown in Fig 3C, reversal of PS externalization upon NEM treatment restored the efferocytic uptake of KPn infected neutrophils (26.5 ± 1.7%) to the levels observed with uninfected neutrophils (24.7 ± 1.0%). Together these data showed that KPn infection likely causes increased flipping of PS resulting in its diminished externalization and thus an impairment of PS-dependent recognition and efferocytic uptake of infected neutrophils by macrophages.
Exofacial PS is a characteristic feature of apoptosis [35]. Our observation of reduced PS exposure and increased flippase activity in KPn infected neutrophils prompted us to examine apoptosis activation in these cells. Since executioner caspases -3 and -7 are universally activated during apoptosis and have also been shown to inactivate flippases to facilitate PS externalization [39, 40], we performed live cell imaging to compare caspase3/7 activation in uninfected and KPn infected neutrophils at various times post-infection. Neutrophils treated with apoptosis inducer staurosporine served as experimental positive control which showed increased activation of Caspase-3 and -7 over time as determined by fluorogenic cleavage of CellEvent Caspase-3/7 Green Detection Reagent (DEVD peptide) by activated caspases (Fig 4A, middle panel). Uninfected neutrophils also exhibited an increase in caspase activation albeit to a lesser extent than the staurosporine treatment, indicating spontaneous apoptosis, characteristic of primary neutrophils (Fig 4A upper panel). In contrast, significantly fewer neutrophils exhibited Caspase 3/7 activation upon KPn infection, as compared to the uninfected neutrophils or the positive control (Fig 4A, lower panel). Quantitation of the mean fluorescence intensity by flow cytometry analysis confirmed these observations and showed a significantly lower fluorescence detection at 3hrs and 4hrs post-infection in KPn infected neutrophils, compared to the uninfected or staurosporine treated neutrophils (Fig 4B). Taken together, these data suggested that KPn infection reduces activation of Caspases 3/7 which likely spares flippase activity to allow increased retention of PS to the inner leaflet of plasma membrane eventuating in impaired recognition and efferocytosis of these cells by macrophages.
Despite the impediment of apoptosis as evident by an impaired caspase 3/7 activation and reduced PS exposure, we found a higher number of KPn infected neutrophils stained positive with an amine-reactive dye, at 3hrs and 4hrs p.i. as compared to Annexin V+PI- cells with PS exposure at that time (Fig 5). This staining method captures global cell death without discriminating between various cell death types. This data suggested that instead of inhibiting cell death, KPn infection likely reprograms the cell death modality. To test this hypothesis, we analyzed necroptosis activation in uninfected and KPn infected neutrophils. For this, we first examined caspase 8 which is centrally positioned to control the extrinsic pathway of apoptosis by activating caspase 3 and 7 and by degradation of RIPK1 [41, 42]. Western blot analysis of Caspase 8 in uninfected and KPn infected neutrophils showed a significantly reduced processing of procaspase 8 into active Caspase 8 in the infected neutrophils (Fig 6A and 6A’). This coincided with the reduced degradation of RIPK1 in KPn infected neutrophils as compared to their uninfected counterparts, indicating inhibition of apoptosis and activation of necroptosis (Fig 6B and 6B’). Indeed, phosphorylation of RIPK3 as well as MLKL (Fig 6C and 6C’) was significantly higher in KPn infected neutrophils as compared to the uninfected cells strongly suggesting activation of necroptosis in these cells upon infection. As a positive control, neutrophils treated with TNF-alpha in combination with SMAC mimetic and caspase inhibitor Q-VD-OPh were analyzed for necroptosis activation in parallel with all samples (Fig 6 +ve control). A hallmark of necroptosis activation is the translocation of MLKL oligomers to the plasma membrane, following RIPK3-mediated phosphorylation of MLKL [43]. Confocal imaging of uninfected and KPn infected neutrophils at 3hrs post-infection showed a strong phospho-MLKL signal on the plasma membrane of infected neutrophils which was similar to that observed with the positive control cells (Fig 6D). Together with the western blot analysis, these data strongly suggested that KPn infection skews the cell death in neutrophils toward necroptosis and away from apoptosis, a cell death mode favorable for efferocytic clearance.
To establish the physiological relevance of modulation of cell death to necroptosis by KPn, we examined the effect pharmacological inhibition of necroptosis on efferocytosis of neutrophils. For this we treated uninfected and KPn infected neutrophils with Necrostatin 1s (Nec1s), a specific RIPK1 inhibitor shown to suppress necroptosis [30]. As shown in Fig 7, a significantly lower percentage of KPn infected neutrophils were efferocytosed as compared to the uninfected cells. Pretreatment with Nec1s reversed this effect of KPn infection and restored the efferocytic uptake of infected neutrophils to the levels similar to uninfected cells (Fig 7A and 7A’). Of note, Nec1s treatment did not significantly alter the efferocytosis of uninfected neutrophils undergoing spontaneous apoptosis. To further confirm the role of necroptosis in KPn infection-driven efferocytosis inhibition, we also tested the effect of RIPK3 inhibitor GSK’872. Indeed, GSK’872 treatment of KPn infected neutrophils significantly increased their efferocytic uptake in comparison with the cells treated with the vehicle alone (Fig 7B and 7B’). The treatment of infected neutrophils with Nec1s or GSK’872 did not increase PS exposure in these cells (Fig 7C). This indicated that activation of necroptosis and modulation of PS exposure are likely two independent mechanisms underlying the impaired effrocytosis of KPn infected neutrophils.
We found that efferocytosis of KPn infected neutrophils is impaired in the lungs of mice (Fig 1D). Having established necroptosis as one of the underlying mechanisms, we sought to examine the effect of necroptosis inhibition by Nec1s treatment on the disease outcome in mice undergoing KPn pneumonia. For this, we treated mice 2hrs before infection and every 4hrs up to 12hrs post-infection with Nec1s as described [44]. Flow cytometry analysis of lungs at 3dp.i., the septic phase of infection exhibiting peak neutrophil accumulation [11, 14, 45], showed that the number of CD11b+Ly6G+ neutrophils was significantly reduced in KPn infected mice upon Nec1s treatment as compared to untreated or vehicle treated KPn infected mice (Fig 8A and 8A’). This indicated that neutrophilia was attenuated by necroptosis inhibition. The bacterial titers in the lungs of Nec1s treated mice was also significantly reduced as compared to the untreated or vehicle treated animals (Fig 8B), further supporting the protective effect of necroptosis inhibition. Overall, these data showed that necroptosis has pathological consequences in promoting pneumonia during KPn infection and inhibition of this pathway improves the disease pathology.
During an acute injury, neutrophils are promptly recruited to the site which, upon the resolution of insult, undergo apoptosis and are cleared by efferocytosis without inducing overt inflammation [9]. Apoptotic cells are swiftly removed by efferocytic cells by recognition of surface “eat me” signal phosphatidylserine (PS), exposed as a result of caspase-mediated inhibition of flippase activity on apoptotic cell membrane [46]. Here we report that Klebsiella pneumoniae, an opportunistic pathogen, prevents efferocytic clearance of neutrophils by reducing surface exposure of PS via modulation of flippase activity. We also show that KPn infection skews the programmed cells death away from “efferocytosis favorable” apoptosis toward necroptosis, which is likely the associated with increased neutrophilia and poor disease outcome. Concomitantly, reversal of PS exposure and blockage of necroptosis improves efferocytic clearance of neutrophils as well as disease outcome in murine inhalation model of KPn pneumonia. Our study provides important insights into pathogenic mechanisms that can be targeted for future antimicrobial therapies for this infection.
Neutrophils are short-lived cells that undergo constitutive or spontaneous apoptosis following which these cells are cleared by phagocytes via efferocytosis, to prevent secondary necrosis and release of their noxious cargo that may cause bystander tissue damage. Induction of apoptosis, typically characterized by activation of executioner caspases and externalization of phosphatidylserine (PS), is thus considered the first step of initiation of efferocytosis process. Efferocytic phagocytes recognize specific “eat-me” signals on apoptotic cells in a receptor dependent manner [33]. Exofacial PS is considered the most well-characterized “eat-me” signal recognized by the phagocytes for removal of apoptotic cells [47]. Owing to an efficient and immunologically silent nature of this event, many viruses and parasites have been reported to use PS mimicry by concentrating PS on their surface and hijack efferocytic machinery of host cells to promote their internalization and cell-to-cell spread [48–50]. In contrast to this, pathogen -mediated skewing of efferocytic clearance of immune cells via PS recognition is much less studied. Staphylococcus aureus is shown to inhibit efferocytic clearance of neutrophils via upregulation of CD47, a “don’t eat me” signal and an alternative cell death pathway of necroptosis [51]. In our study we instead found that KPn infection results in a drastic down modulation of exofacial PS, reduced apoptotic caspases and increased activation of necroptosis machinery. The relevance of these events was confirmed by reversal of PS exposure and inhibition of necroptosis on KPn infected neutrophils, which rescued their efferocytic clearance in-vitro and reduced the bacterial burden as well as neutrophilia in-vivo in murine pneumoseptic KPn infection. Although delayed apoptosis in KPn infected neutrophils [52, 53]; and indirect evidence of necroptosis activation in KPn infected macrophages [32] was recently reported, we present the first evidence of the mechanism underlying reduced PS exposure; and physiological relevance of KPn mediated skewing of cell death modality in neutrophils to avoid their efferocytic clearance. PS-dependent clearance of immune cells results in the release of anti-inflammatory and pro-resolving factors which downregulate the inflammation [54]. In addition, efferocytic uptake of apoptotic neutrophils regulates granulopoiesis and peripheral neutrophilia via an IL-23/IL-17/G-CSF axis [55]. On the other hand, neutrophils undergoing necroptosis are less amenable to efferocytic clearance and amplify inflammation [56, 57]. It is tempting to speculate that reduced activation of apoptosis and PS exposure in KPn infected neutrophils, in combination with activation of necroptosis machinery, likely result in the loss of the negative regulatory axis thus contributing to neutrophilia and inflammation characteristic of this infection. Studies to examine this line of inquiry are currently underway in our laboratory.
PS distribution on the lipid bilayer in biological membranes is controlled by the activity of ATP-dependent flippases and scramblases [58]. Flippases are responsible for removing PS from the external leaflet by ATP-dependent active transport; the relevant proteins are members of the type IV subfamily of P-type ATPases. Constant surveillance of the external leaflet by these enzymes establishes a normal plasma membrane distribution in which virtually all of the PS is in the inner leaflet. The second enzymatic activity termed scramblases, regulating the distribution of PS between leaflets catalyzes rapid and nonspecific exchange of phospholipids between the two sides of the bilayer. In the apoptotic caspase cascade, caspases 3 and 7 cleave the flippases to inactivate them leading to irreversible PS exposure [35]. We found a significantly reduced caspase 3 and 7 activation in neutrophils upon KPn infection, suggesting that KPn mediated impairment of PS exposure involves modulation of flippase activity. Indeed, KPn infected neutrophils exhibited enhanced flippase activity, which likely contributes to restriction of PS to inner leaflet resulting in its reduced exofacial exposure. While the involvement of scramblases and the identity of specific flippase is an ongoing area of investigation in our laboratory, our results show, for the first time, that KPn downregulates PS exposure on neutrophils by modulating phospholipid translocase activity.
Necroptosis is defined as a caspase-independent non-apoptotic form of regulated cell death which requires RIPK1 and RIPK3-mediated activation of pseudokinase, mixed lineage kinase-like protein (MLKL) [59]. Extrinsic pathway of apoptosis activated by death ligands such as FasL or TLR ligands promotes processing of caspase-8 which degrades RIPK1 and drives apoptotic cell death [60]. Conversely, inhibition of caspase 8 spares RIPK1 to promote RIPK3 autophosphorylation and recruitment of MLKL to form a necroptosome complex [61]. Phosphorylation of the C-terminal pseudokinase domain of MLKL by RIPK3 promotes MLKL translocation to plasma membrane, which eventually disrupts the membrane integrity [62]. Our results showed reduced processing of caspase 8 in KPn infected neutrophils indicating the apoptosis arrest in these cells. This coincided with increased RIPK1 levels, RIPK3 phosphorylation and MLKL phosphorylation as well as translocation of phospho-MLKL to plasma membrane, strongly suggesting the activation of necroptosis pathway in KPN infected neutrophils. This supports the notion that KPn drives cell death away from immunologically silent apoptosis which favors efferocytosis, toward inflammatory necroptosis [57]. We show the physiological consequence of this phenomenon by increased efferocytic uptake of infected neutrophils upon treatment with NEM to increase PS exposure as well as with inhibitors of necroptosis. The observation of increased global cell death at 3hrs in KPn infected neutrophils (Fig 5) as compared to much lower early and late apoptotic/necrotic cells at that time point (Fig 2) is intriguing. This may be reflective of higher sensitivity of amine-binding dyes compared to propidium iodide, or that there may be a certain interval/time lag between the MLKL localization and lysis of cells. As such, the precise events following integration of MLKL into plasma membrane leading up to plasma membrane rupture are still poorly understood. Whether the increased RIPK1, RIPK3 and MLKL activation marks a typical necroptotic, lytic event during infection or if KPn utilizes the necroptosis machinery to modulate cellular functions in its favor, or a combination of both, requires further investigation. Nevertheless, our report presents convincing evidence that necroptosis machinery is activated upon KPn infection which, along with reduced PS exposure via increased flippase activity, contributes to reduced efferocytic clearance of neutrophils.
Necroptotic fibroblasts and monocytic cell lines were recently reported to expose PS on their surface in pMLKL-dependent fashion [63, 64]. This elegant study implicated plasma-membrane associated active MLKL in rapid PS exposure that occurs within 5 minutes of RIPK3 and MLKL activation. In our study we found a reduced PS exposure, but increased pMLKL in primary neutrophils after 3 hrs of infection with bacterial pathogen KPn. PS exposure was increased in KPn infected neutrophils early during infection (30 min). However, we did not find activated RIPK3 or p-MLKL at that time in uninfected or infected neutrophils. This suggests that different pathways of PS exposure and necroptosis are activated in primary neutrophils in response to bacterial infection versus the fibroblasts and monocytic cell lines activated by different stimuli used in the reported study. This is further supported by our observation that inhibition of necroptosis does not restore PS exposure in infected neutrophils. In light of our results showing the reversal of efferocytosis inhibition upon increased PS exposure as well as upon inhibition of necroptosis, it is highly likely that KPn utilizes two independent strategies to inhibit the clearance of neutrophils and to drive the neutrophilia and inflammation characteristic of pneumonic infection. Improved disease outcome in terms of reduced bacterial burden and decreased neutrophil accumulation in lungs of pneumonic mice, upon blockage of necroptosis supports a detrimental effect of cell death modulation by this pathogen. Our results are in line with elegant studies showing the pathological effect of necroptosis in development of bacterial pneumonia and possibility of using necroptosis inhibitors as an adjunct therapy for bacterial infections [30, 32, 44, 65].
KPn has recently gained attention as a “successful” pathogen owing to an emergence of hypervirulent strains as well as antibiotic resistance [66]. The wide range of infections caused by this pathogen in immunocompromised and immune-competent individuals have become increasingly difficult to treat owing partly to the arsenal of virulence factors exhibited by this pathogen that it utilizes to protect itself from host immune response [67]. Based on these virulence factors KPn has been categorized as an “evader” rather than an “offender” pathogen. Our results presented here provide new insights into the strategies employed by this pathogen to actively suppress an important host defense i.e. efferocytic clearance of neutrophils by modulation of cell death pathway and PS externalization. Although the identity of putative virulence factor/s involved in the process are currently being worked out in our lab, our data showing downregulation of PS exposure by increased flippase activity, reduced activation of apoptosis executioner caspases and reprogramming of cell death toward inflammatory necroptosis pathway in neutrophils infected with KPn, highlight virulence strategies employed by this pathogen to impair efferocytic clearance of these cells by macrophages. Given its host protective consequences, subversion of efferocytosis may be advantageous for this pathogen to establish infection. Also of importance in this regard, a recent report showed neutrophils as vehicles for KPn in its dissemination to establish liver abscess, a severe clinical complication of KPn infection [16]. Efferocytic clearance of KPn-infected neutrophils thus may aid in circumventing these distant metastatic complications, as has been shown in case of mycobacterial infection [22]. Notwithstanding the identity of virulence factors involved, our findings open new avenues to treat and prevent the systemic spread of this infection. As antibiotic resistance is a serious problem associated with Klebsiella infections, elucidation of mechanism by which this pathogen manipulates efferocytosis, as reported here, might provide novel therapeutic targets.
The KPn (ATCC strain 43816) were grown to log phase in LB medium at 37°C. For isolation of cells and in-vivo experiments 6–8 weeks old wild-type C57BL/6 bred in the animal facility of the University of North Dakota were used. The animals were handled according to the institutional and federal guidelines.
Peritoneal neutrophils and macrophages were isolated 12-16hrs and 5 days respectively after intraperitoneal injection of sterile 4% thioglycollate (BD Biosciences, San Jose, CA) (10, 12). Purity of the cells was ascertained by flow cytometry analysis (Ly6G+ neutrophils 80–85%; F4/80+ macrophages 85–90%). Isolated neutrophils were left uninfected or infected with 10 MOI of KPn for 3 hrs followed by labelling with Carboxyfluorescein succinimidyl ester (CFSE; CellTrace CFSE Cell Proliferation Kit from Invitrogen). Macrophages seeded on 6 well plates (0.5x106 cell/ml) were incubated with CFSE-labelled uninfected or KPn infected neutrophils at a ratio of 5:1. After 2hrs of efferocytosis, non-internalized neutrophils were removed by washing thoroughly. Macrophages were scraped and stained with F4/80 and Ly6G antibodies for flow cytometry. Gating scheme to quantitate Ly6G-F4/80+CFSE+ efferocytic macrophages that had internalized labelled neutrophils is shown in Fig 1A. For some experiments, uninfected and infected neutrophils were pretreated with 5 mM of N-ethylmaleimide (NEM) or with necroptosis inhibitors necrostatin-1s (100 μM) (BioVision, California) or GSK’872 (3 μM) (Millipore Sigma, St. Louis, MO) for 30 min before infection. For calculating efferocytic index, the percentage of Ly6G-F4/80+CFSE+ macrophages engulfing KPn infected neutrophils was normalized to those that engulfed uninfected neutrophils (taken as 100%) [68, 69].
Uninfected or KPn infected (MOI 10) peritoneal neutrophils labelled with CFSE (repeated) were administered intranasally (35μL/3.5x106 cells) into mice anaesthetized using a mixture of 30mg/ml ketamine and 4 mg/ml xylazine in PBS. Lungs were lavaged 2 hrs after the instillation as described by us [11, 13]. Efferocytic uptake by macrophages was quantitated by flow cytometry using PE-Cy7 conjugated anti-F4/80 and APC conjugated anti-Ly6G antibodies (BioLegend, San Diego, CA) to enumerate Ly6G- F4/80+ CFSE+ alveolar macrophages. Efferocytic index was calculated as described above.
Peritoneal neutrophils were left uninfected or infected with 10 MOI of KPn for 3 hrs. A FITC Annexin V Apoptosis Detection Kit (BD Biosciences, San Jose, CA) was used according to manufacturer’s instructions followed by flow cytometry analysis to enumerate percentage of Ly6G+ PI- Annexin V+ cells with surface exposed PS using BD LSR II flow cytometer (Becton Dickinson, San Jose, CA). To enumerate global cell death, neutrophils uninfected or infected with K. pneumoniae for various times as described above were stained using LIVE/DEAD Fixable Near-IR Dead Cell Stain Kit (Invitrogen) in serum free PBS for 30 minutes on ice per manufacturer’s instructions. Cells were then washed with PBS containing 1% FBS and analyzed by flow cytometry. For caspase 3/7 activation staining, CellEvent Caspase-3/7 Green Detection Reagent (Invitrogen) kinetic assay was used according to the manufacturer’s instructions. Briefly, peritoneal neutrophils were loaded with optimized concentration of the detection reagent for 30 minutes at 37°C. Cells were then either left uninfected or infected with K. pneumoniae (at 10 MOI) in complete RPMI medium. At indicated times, uninfected and infected cells were analyzed for Caspase3/7 activation by flow cytometry to quantify Mean Fluorescence intensity (MFI) or were imaged using Olympus upright phase-contrast fluorescent microscope. Staurosporine (5 μM, Sigma-Aldrich), an apoptosis inducer was used a positive control for each time point.
Flippase activity in uninfected and KPn infected neutrophils was determined as described previously [70, 71]. Briefly, peritoneal neutrophils were incubated with 1 μl of 1 mg/ml of 1,2-dioleoyl-sn-glycero-3-phospho-L-serine-N-(7-nitro-2-1,3-benzoxadiazol-4-yl) (NBD-PS) in 1 ml HBSS (Hank’s balanced Salt Solution) containing 1 g/liter glucose for 20 min at 37°C. Cells were then washed, pelleted and suspended in the RPMI-1640 without or with K. pneumoniae (10 MOI) for various time points. After each incubation period, neutrophils were washed with 1 ml PBS containing 5% fatty acid free BSA. To measure the NBD-PS in the inner leaflet of the membrane, 40 μl of freshly prepared 1 M sodium dithionite prepared in 0.5 M Tris was added to sample, in order to bleach the NBD-PS on the outer leaflet. Mean fluorescence intensity (MFI) of KPn infected and uninfected neutrophils was then measured by flow cytometry. Flippase activity of each sample was quantified by comparing NBD-PS fluorescence before F(o) and after F(x) bleaching according to the following Equation.
The percent Flippase activity was calculated using the following equation where Fflip (100) is the NBD-PS stained cells.
For detection of necroptosis markers, peritoneal neutrophils uninfected or infected with 10 MOI KPn for 3 hours were lysed with RIPA buffer containing protease and phosphatase inhibitors (Sigma) as described previously by us and others [44, 72, 73] and probed with anti-mouse Caspase 8, RIPK1 antibodies (Cell Signaling Technology, Danvers, MA) and phosphor-RIP3 S232 & T231, phosphor-MLKL S345 antibodies (Abcam, Cambridge, MA) by immunoblot analysis. Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as loading control. Peritoneal neutrophils treated with 1 μM SMAC mimetic AT‐406 and 20 μM Q‐VD-OPh for 30 min prior to stimulating with 100 ng/mL mouse TNF-α were used as positive control for necroptosis induction [74]. Densitometry analysis on blots was performed using Image J software and band intensities were represented as the ratio of the protein of interest and the internal control levels. For phosphorylated proteins, the band intensities were represented as the ratio of phosphorylated and total protein.
For plasma membrane localization of phospho-MLKL, uninfected, infected or positive control neutrophils were washed with PBS and were fixed on Poly-L-lysine coated glass slides with 4% paraformaldehyde. Cells were then processed for immunostaining as described by us [11, 14, 15, 75] using mouse monoclonal anti-phospho-MLKL (0.1μg/ml) (Millipore Sigma) conjugated to Alexa Fluor 488 (Molecular Probes, Eugene, OR) and rat monoclonal anti-Gr1 (eBioscience) followed by goat anti-rabbit Alexa Fluor 546 secondary antibody. Plasma membrane localization of proteins was analyzed using Zeiss LSM 510 meta confocal microscope and images were processed using ImageJ software.
Mice were anaesthetized with a mixture of 30mg/ml ketamine and 4 mg/ml xylazine in PBS and were infected intranasally with 3.0 x 104 CFUs in 20μl of saline, of KPn or with 20 μl of saline alone. In some instances, mice received intraperitoneal injections (100μl) of necrostatin-1s (100μM) or vehicle alone (DMSO) 2hrs before infection and every 4hrs up to 12hrs post-infection as described [44]. No toxicity or overt pathology was observed at this dose. The mice were euthanized at 3dp.i. and lungs were aseptically homogenized in cold PBS with Complete protease inhibitor cocktail (Roche Diagnostics, Germany). For the bacterial burden analyses, serially diluted homogenates were plated on LB agar and incubated at 37°C overnight.
For enumeration of neutrophils in mouse lungs, lungs cells were harvested from mice at 3 days p.i. and processed as previously described by us [11, 14, 45]. Quantitation of neutrophils by flow cytometry (using a BD LSR II, Becton Dickinson, San Jose, CA) was done by quantitating Ly6G+CD11b+ cells stained with Pacific Blue anti-mouse CD11b (Clone M1/70) and APC anti-mouse Ly6G (Clone 1A8) antibodies (Biolegend, San Diego, CA). FlowJo (Tree Star) software was used to analyze the data.
Statistical analyses were performed using the Student t test (SIGMA PLOT 8.0, Systat Software, San Jose, CA) for two-group comparisons. A p value of ≤ 0.05 was considered statistically significant. Nonparametric ANOVA and Dunn's post hoc analyses were used for multiple-group comparisons.
Mice were cared for according to the recommendations of the NIH, published in the Guide for the Care and Use of Laboratory Animals. All techniques used were reviewed and approved by the University of North Dakota Institutional Animal Care and Use Committee (IACUC) under the protocol 1503–3.
|
10.1371/journal.pntd.0003038 | Shifting Patterns of Aedes aegypti Fine Scale Spatial Clustering in Iquitos, Peru | Empiric evidence shows that Aedes aegypti abundance is spatially heterogeneous and that some areas and larval habitats produce more mosquitoes than others. There is a knowledge gap, however, with regards to the temporal persistence of such Ae. aegypti abundance hotspots. In this study, we used a longitudinal entomologic dataset from the city of Iquitos, Peru, to (1) quantify the spatial clustering patterns of adult Ae. aegypti and pupae counts per house, (2) determine overlap between clusters, (3) quantify the temporal stability of clusters over nine entomologic surveys spaced four months apart, and (4) quantify the extent of clustering at the household and neighborhood levels.
Data from 13,662 household entomological visits performed in two Iquitos neighborhoods differing in Ae. aegypti abundance and dengue virus transmission was analyzed using global and local spatial statistics. The location and extent of Ae. aegypti pupae and adult hotspots (i.e., small groups of houses with significantly [p<0.05] high mosquito abundance) were calculated for each of the 9 entomologic surveys. The extent of clustering was used to quantify the probability of finding spatially correlated populations. Our analyses indicate that Ae. aegypti distribution was highly focal (most clusters do not extend beyond 30 meters) and that hotspots of high vector abundance were common on every survey date, but they were temporally unstable over the period of study.
Our findings have implications for understanding Ae. aegypti distribution and for the design of surveillance and control activities relying on household-level data. In settings like Iquitos, where there is a relatively low percentage of Ae. aegypti in permanent water-holding containers, identifying and targeting key premises will be significantly challenged by shifting hotspots of Ae. aegypti infestation. Focusing efforts in large geographic areas with historically high levels of transmission may be more effective than targeting Ae. aegypti hotspots.
| We carried out a comprehensive study of the long-term trends in household-level Aedes aegypti spatial distribution within a well-defined urban area endemic for dengue virus. By using a dataset consisting of 13,662 household entomological visits performed in two neighborhoods in Iquitos, Peru, we quantified the ∼3 year spatial clustering patterns of Ae. aegypti among houses and the temporal persistence of vector abundance hotspots. Our results provide strong support for the conclusion that Ae. aegypti distribution is highly focal and that hotspots of high vector abundance at the level of small groups of houses are common, but temporally unstable. Results from our study have implications for understanding the spatio-temporal patterns of Ae. aegypti abundance and for the design of surveillance and control activities that are based on household-level entomological data.
| Despite decades of vector control efforts, dengue has become the most important mosquito-borne viral disease of humans. Estimates indicate that ∼390 million dengue virus (DENV) infections occur annually throughout the tropical and subtropical world [1], [2]. In the last twenty years, dengue epidemics have increased in number, magnitude and severity, due in part to range expansion of the mosquito vector Aedes aegypti, geographic spread and evolution of DENV, and increased urbanization and international travel [3], [4]. The emergence of DENV as a public health problem has been influenced by the interplay of multiple factors, including the abundance, dispersal and blood feeding patterns of female Ae. aegypti; complex interactions among multiple virus serotypes and genotypes; environmental factors (i.e., temperature, humidity and rainfall); herd immunity in human populations; and human density, age structure and movement [4]–[8].
Aedes aegypti's ecology and behavior contribute to its efficient transmission of DENV and spatio-temporal patterns of human DENV infections. They bite during the daytime when human hosts are active, are highly anthropophilic, are well adapted to human habitations, and tend to be relatively sedentary with limited dispersal tendencies; they seldom disperse beyond 100 m [7], [9]–[12]. Results from mathematical and simulation models indicate that such traits can have strong effects on DENV transmission dynamics, due to their influence on contact between humans and mosquito vectors [13], [14]. Empiric evidence from entomologic field surveys and population genetics studies support the notion that Ae. aegypti abundance is spatially heterogeneous and that some areas and larval habitats are likely to produce more adult mosquitoes than others [15]–[21]. Given the current emphasis on spatially-based interventions (where reactive control is performed based on the proximity to residences of dengue cases, [22]), identifying and predicting the occurrence of vector hotspots (small groups of houses with disproportionately high productivity, vector abundance and potential for DENV transmission) is a logical next step for assessing current control recommendations and devising innovative concepts for Ae. aegypti control and dengue management.
Understanding patterns of Ae. aegypti distribution at fine spatial (e.g., at the household level and within a neighborhood) and temporal scales (e.g., within and across consecutive seasons and years), is curtailed by the difficulty of collecting adequate information at those levels of resolution. Most studies describing within-city patterns of Ae. aegypti distribution are performed at aggregated spatial scales (neighborhoods or census districts, e.g. [23], [24]) or by analyzing information from a network of traps spaced over several hundred meters (e.g. [19], [25]). In one detailed analysis, Getis et al. [15] reported that, in Iquitos, Peru adult Ae. aegypti were aggregated up to 30 meters, but pupae did not cluster beyond the household. Their findings are in agreement with the focal nature of Ae. aegypti dispersal [9] and have been validated in rural Thai villages [20], [26], [27], in a northern Argentina community neighborhood [28], in coastal Ecuador [29], and through complex simulation models [30], [31]. Implicit in the finding of spatially clustered populations is the notion that hotspots of high vector abundance could be the focus of targeted vector control interventions. Theoretical models support the idea that control interventions targeting hotspots can disproportionately reduce pathogen transmission in comparison to blanket or random interventions [32]–[38]. Furthermore, targeting vector control interventions with greater or equal efficacy to blanket interventions could also result in reduced pesticide usage and operational costs (e.g., [38], [39]).
Most of the published research on the spatial pattern of household-level Ae. aegypti distribution covered short temporal scales (either cross-sectional or a single season), analyzed data on vector presence but not abundance, lacked measures of the variability of clustering estimates, and did not consider how persistent (or predictable) spatial clusters were. Before considering whether Ae. aegypti hotspots could be considered as rational targets for vector control, information on their temporal variability and persistence is required to better inform where and when interventions should occur. To fill this knowledge gap, we used a detailed longitudinal entomologic dataset to quantify long-term patterns of Ae. aegypti spatial distribution at the household level. Specifically, our objectives were to (1) quantify the spatial clustering patterns of adult female Ae. aegypti and pupae counts per house over nine sampling surveys separated by approximately four months, spanning a 3-year period (2009–2011) in two Iquitos neighborhoods differing in Ae. aegypti infestation levels; (2) determine overlap between clusters of Ae. aegypti females and pupae; (3) quantify the spatial and temporal stability of clusters over the nine entomologic surveys.
Strict protocols for household enrolment study were followed, including contacting homeowners and asking for their permission to have their house and patio being inspected for pupae and adult mosquito presence and abundance. The procedures for enrollment of households in the entomologic and demographic surveys were approved by University of California, Davis (2007.15244); NAMRU-6 (NMRCD 2007.0007), which included Peruvian representation; and Emory University (IRB9162) Institutional Review Boards.
Our study was performed in the Maynas and Tupac Amaru neighborhoods of the city of Iquitos, the largest urban center (population ∼370,000) in the Peruvian Amazon. The two neighborhoods were described previously [24], [40], [41] and were chosen for comparison because they differed epidemiologically and entomologically. Maynas has higher DENV prevalence rates and Ae. aegypti infestation levels than Tupac Amaru [15]. Additionally, Maynas is older, more centrally located within Iquitos, more urbanized and wealthier than Tupac Amaru [24], [40], [41].
Data were collected using standardized household entomological surveys performed approximately every 4 months from 2009 to 2012 (9 consecutive surveys). Methods for surveys and mosquito collection are described in detail elsewhere [15], [24]. Briefly, Ae. aegypti productivity was assessed by pupal surveys performed for all containers at each surveyed house [24]. Indoor and outdoor adult mosquito abundances were measured by using Prokopack mosquito aspirators [42]. Two-person survey teams were rotated over time to limit temporal and collector bias [24]. Aspiration collections were conducted in each room of the house as well as in the patio. All containers found to be holding water were measured, classified, and scored for sunlight exposure, fill method (actively via faucet or passively by rain), and presence of a cover. Collected adults and pupae were taken to the field laboratory for species identification. Pupae were counted and placed in plastic vials labeled with a unique house number, container code, and date. Each subsequent day, adults that emerged were collected and placed in a −20°C freezer. After 30 minutes to 1 hour, they were identified to species, counted by sex, and data were recorded on the entomology collection sheet.
All collected data were linked to a household level Geographic Information System (GIS) for the city of Iquitos (described in [15], [24], [43]). Survey data was imported into ArcGIS 10.1 (ESRI, Redlands, CA) and linked to the Iquitos GIS by the house code (a unique alpha-numeric code painted on every house's door and used throughout the Iquitos field studies). Data were then projected in Universal Transverse Mercator and WGS −84 DATUM and used to map the raw field data as well as results of spatial statistics tests.
Spatial analyses were performed on the number of water-holding containers per house (a measure of habitat availability), the number of Ae. aegypti pupae per house (a measure of productivity) and number of Ae. aegypti adult males and females per house (a measure of DENV entomologic risk). Details of each test formula, expected values, and calculations were thoroughly described by Getis et al. [15]. Below, we provide a brief description of each statistical test and its implementation within the context of this study.
Global spatial statistics were implemented to detect the presence of spatial clustering of Ae. aegypti infestation anywhere within each study neighborhood [15]. To account for bias introduced by the clustered pattern of households within a block we compared the increments in the observed clustering of houses (k-function) with the pattern and of Ae. aegypti presence (k-function) and abundance (weighted k-function) as described by Getis et al. [15]. A bivariate k-function test [44] was implemented to detect the spatial scale up to which pupae and adult presence were related to each other. The function was an extension of the k-function for univariate data and compared the scales up to which infestation in one event (e.g., pupae) were more clustered than the distribution of the two events combined (infestation of pupae and adults). Given that points located on the edges were more likely to cluster because they had fewer neighbors than central points, an edge effect correction was included in the formulae of all k-functions [15].
Local Getis hotspot analysis (Gi*) was applied to map the occurrence of clusters of high Ae. aegypti abundance and water-holding container numbers [15]. Houses that were members of clusters were identified using a z-score of ±3.706 as a cutoff for cluster membership (Bonferroni-corrected z-value). To account for overdispersion in the data (which can dramatically affect Getis Gi* test), analyses were performed on the log-transformed mosquito abundance (Log[number of pupae/adults+1]) and the log-transformed number of water-holding containers (Log[number of containers+1]). Once members of significant clusters were identified, the distance up to which clustering occurred around each house was identified as by Getis et al 2003 [15]. Local analyses were performed separately for each neighborhood, entomologic survey and Ae. aegypti infestation measure. For all spatial analysis tests, clustering distances between 1–5 meters were considered to occur within the household (the average width of a house lot in Iquitos is 4.6 m) whereas clustering distances beyond 5 m were considered to be between households. Weighted K-function analysis was performed on the aggregated number of entomologic surveys that a house was member of a high Ae. aegypti abundance cluster (range of values, 1–9 surveys) to determine whether some houses or areas within each neighborhood were consistent hotspots of vector abundance. In the context of our study, we define an Ae. aegypti hotspot as a distinct house or group of houses with significantly higher mosquito densities than surrounding houses [33], [36].
Clustering distances of adult Ae aegypti at the household level (from the Getis Gi* test) aggregated across all entomologic surveys were used to calculate the cumulative probability distribution of clustering; i.e., the probability of finding clusters with an extent equal to or less than d meters. Maximum Likelihood techniques were applied to fit various statistical distributions (e.g., exponential, power law) to the cumulative probability distribution of the distance of local clustering. This kind of functional relationship described the probability of finding spatially correlated populations at increasing distances from a household. Curve fits were performed independently for each neighborhood and for both neighborhoods combined.
Analyses were performed using the Point Pattern Analysis (PPA) program developed by Arthur Getis with assistance from Laura Hungerford, Dong-Mei Chen, and Jared Aldstadt (available online at http://www.nku.edu/~longa/cgi-bin/cgi-tcl-examples/generic/ppa/ppa.cgi) and the packages splancs ([45]) and fitdistrplus ([46]) of the R statistical software (ver 2.15 [47]). Curve fitting procedures were performed using Matlab (Mathworks, Natick, MA) curve fitting function.
From March 2009 to October 2011, 13,662 household entomological inspections were performed (7,156 in Maynas and 6,506 in Tupac Amaru; Table 1). A total of 1,226 and 1,068 unique houses were inspected in Maynas and Tupac Amaru, respectively (Table 1). On average (SD), 884 (59) and 832 (73) households were visited on each entomologic survey in Maynas and Tupac Amaru, respectively. Seventy-seven percent (SD = 7.5%) of buildings surveyed in both neighborhoods were residential, followed by stores (mainly houses used as neighborhood stores) (mean = 17.5%; SD = 1.8%) (Figure S1). Surveys lasted on average (SD) 22 (7) days in Maynas and 18 (5) days in Tupac Amaru (Table 1). On average (SD), each house was surveyed 5.8 (2.4) times throughout the study period. The average (SD) number of residents per house in both neighborhoods was 6.0 (3.1). Percentage of households using two of the most important water sources (piped and rain water) ranged from 91% to 95% for piped and 2% to 7% for rain water.
Maynas households had a significantly higher average number of water-holding containers than Tupac Amaru (Two-sample Wilcoxon test, W = 224207, P<0.001). Across both neighborhoods, most (98.9%) houses had at least one water holding container throughout the study period. The proportion of houses with positive containers ranged between 0.04–0.12 in Maynas and 0.03–0.10 in Tupac Amaru (Figure S2). A total of 5,833 Ae. aegypti pupae (3,192 in Maynas and 2,641 in Tupac Amaru) and 8,709 adult males and females (5,671 in Maynas and 3,038 in Tupac Amaru) were collected over the nine surveys. Forty-nine percent of all Ae. aegypti adults collected were females. Adult abundance was highly overdispersed, 91% of all Maynas houses and 94.9% of all Tupac Amaru houses were infested with 5 or less adult Ae. aegypti mosquitoes. The median number of adult Ae. aegypti per house was significantly higher in Maynas than in Tupac Amaru (Figure S3, W = 1544194, P<0.001). There was, however, considerable variation embedded in these estimates (Figure 1 and Figure 2). The number of adults and pupae collected per house ranged from 0 to 163 and from 0 to 681, respectively. Infested houses were found throughout the study neighborhoods (Figure 1 and Figure 2).
Table 2 summarizes the results of the k-functions applied to pupae and adult presence and abundance in each neighborhood. Pupal collections showed a low degree of spatial clustering in both neighborhoods (Table 2). Clusters of pupae presence and abundance were observed in 44% and 11% of surveys, respectively, for Maynas and in 55% and 33% of surveys in Tupac Amaru (Table 2). The estimated overall mean ± SD clustering distance was 16.6±5.0 m for pupae presence and 10.3±7.8 m for pupae abundance. Average clustering distances did not differ between neighborhoods (17.5±5.5 m in Maynas vs 16.0±5.0 m in Tupac Amaru). The spatial distribution of adult Ae. aegypti showed a stronger pattern, with clusters found on every survey. Clustering distances in both neighborhoods ranged from 1 m (within the household) to 40 m, with mean ± SD clustering values across both neighborhoods for adult presence and abundance of 16.3±10.7 m and 17.4±12.9 m, respectively (Table 2).
Figure 3 shows the results of the bivariate k-functions applied to test the scale up to which pupae and adult infestation were associated during nine entomologic surveys between 2009 and 2011. Given the lack of difference in global clustering between neighborhoods, results were pooled to show the overall scale up to which pupae and adults are associated. When the observed value (solid line) is higher than the random expectation (dashed line) spatial association between variables occurs at such a distance (Figure 3). For 4 out of 9 surveys (44%), pupae and adults were clustered within the household (at a distance of 5 m or less) (Figure 3). A very focal level of association between pupae and adults was found when clustering occurred beyond the household (Figure 3); the average ±SD clustering distance was 11.4±5.4 m. Collections performed in December-January (surveys 3, 6 and 9) had higher extent of association between pupae and adults (15–17 meters) compared to the remaining surveys (up to 5 m), indicating that during those months either the extent of populations is larger or the abundance of Ae. aegypti is more patchily distributed.
The number of water-holding containers (log+1)-transformed did not show any strong spatial pattern. Out of an average of 722 houses per survey in Tupac Amaru, only 5 unique households were members of significant clusters (GI*(d)>3.71; P<0.05) of high container numbers (one in survey 2, two in survey 4, one in survey 5 and one in survey 6). In Maynas, only three houses were members of clusters (all in survey 6). Clustering distances in all cases did not exceed the household (<5 m). This indicates that, whereas water-holding containers are very common, they do not show any spatial structure within both neighborhoods.
Given the low probability of finding pupae clusters, local spatial analyses were performed on Ae. aegypti adult abundance data only. Hotspot analysis maps are presented in Figure S3 and summaries of clustering measures in Table 3. On average, 3.1% of Maynas and 1% of Tupac Amaru households were members of a cluster of high adult abundance (Table 3 and Figure S3). An average of 51.8% (range = 32–68%) of all adults collected in Maynas and 28.7% (10–54%) of all adults collected in Tupac Amaru were found within the identified spatial clusters. In Maynas, an average of 30.4% (range = 0–53%) of clusters of adult abundance occurred beyond the household, whereas in Tupac Amaru the proportion of clusters occurring beyond the household increased to 50.7% (range = 14–86%) (Table 3).
There was no obvious consistent temporal pattern of adult clusters in both neighborhoods; i.e., the location of clusters in one survey differed from the location of clusters in future or prior surveys. The temporal instability in Ae. aegypti hotspots is shown in Figure 4. Most houses in Maynas and Tupac Amaru (80.9% and 87.9%, respectively) were identified as hotspots only once in the 9 survey periods. The maximum number of survey dates when a house was identified as a hotspot was 3 (out of 9 surveys) in Maynas and 2 (out of 9 surveys) in Tupac Amaru (Figure 4). The spatial location of hotspots did not follow any apparent spatial pattern; the distribution of hotspots within both neighborhoods did not differ from a random distribution (Figure 4).
We used the maximum distance of adult Ae aegypti local clustering in each neighborhood to estimate the cumulative probability distribution for finding spatially correlated populations at increasing distances from a household (Figure 5). A value of 0–5 meters in the X-axis of Figure 5 indicates that clustering did not exceed the household whereas values higher than 5 meters indicate that Ae. aegypti abundance was spatially correlated beyond the household. The probability of finding spatially correlated adult populations decreased significantly with increasing distances from the house, with patterns for both neighborhoods better explained by a negative exponential model of the form (Table 4). Model fit was very high (R2Maynas = 0.91; R2Tupac Amaru = 0.86; R2Both = 0.89). When data from both neighborhoods was combined, the probability of finding adults clustering beyond the household (>5 m) was 42% (95% CI,57.8–25.8%) and the finding of clusters of high adult abundance with an extent of 100 m was rare (6.1%, 95% CI, 0.0–21.0%) (Figure 5 and Table 4). Predicted values were very similar between neighborhoods (Table 4 and Figure S5).
Trends over a 3-year study period in household-level spatial distribution within a well-defined urban area provide strong evidence for highly focal distribution of Ae. aegypti. Hotspots of high mosquito abundance in small groups of houses were common, but temporally unstable.
Theory predicts that interventions targeting super-spreaders can disproportionately impact pathogen transmission in comparison to blanket or non-targeted interventions [32], [34], [35]. For certain vector-borne diseases, locations are more important than individual persons with regard to their contribution to transmission; disease ‘hotspots’ or ‘key-locations’ dominate the spatial dynamics of various vector-borne diseases [33], [36], [38], [48]. For dengue, the concept of key locations has been studied in terms of productivity of larval habitats, leading to the identification of key-premises [49] and, via spatial analyses, identifying potential vector or virus hotspots. Some researchers have concluded that targeting vector control at hotspots of high Ae. aegypti productivity will be a more effective and efficient use of available resources than traditional, more evenly applied interventions [20], [28], [29], [50].
The occurrence of shifting hotspots of Ae. aegypti abundance imposes a significant challenge to intervention strategies targeting vector control on households. Because adult mosquito hotspots observed during one of our surveys did not predict hotspots at the same location during prior or subsequent surveys, we do not expect identifying and targeting key-premises [49] to be operationally practical in all DENV endemic settings. In most DENV endemic areas, the availability and type of containers that can produce adult Ae. aegypti are affected by the reliability of piped water services, a factor that tends to be highly variable in space and time [51]. Container management practices by the occupants of the property, coupled with the range of Ae. aegypti flight dispersal and Ae. aegypti egg-laying behavior further contribute to the spatially and temporally heterogeneous pattern of vector larval productivity and adult distribution [9], [18], [20], [50], [52]. Thus, in a city like Iquitos, where there is a relatively low percentage of Ae. aegypti in permanent water holding containers [52], and where container availability is high across households, a strategy of identifying and targeting key premises will be significantly challenged by shifting hotspots of Ae. aegypti infestation.
Using a grid of 19 BG-Sentinel traps uniformly distributed at ∼130 m intervals and surveyed every 3 weeks, Barrera [19] described the distribution of adult Ae. aegypti abundance as temporally stable, with some traps consistently being members of clusters of high mosquito abundance. Similarly, analysis of weekly sentinel ovitrap data aggregated at the block or neighborhood levels indicated high levels of persistence in Ae. aegypti infestation patterns [23], [25]. Such patterns differ dramatically from the observed lack of persistence in infestation clustering reported when analyzing household-level Ae aegypti abundance data; i.e., this study and Garelli et al. [28]. Both studies indicate that, although hotspots of Ae. aegypti abundance are common, their specific location within a study neighborhood is different in every entomologic survey. Thus, whereas Ae. aegypti abundance appears to be spatially autocorrelated within weeks and at aggregated geographic units, over longer time scales (months) and at fine spatial resolution (household) the occurrence of shifting rather than temporally stable hotspots appears to be a common feature of vector distribution. By integrating our results with the ones found at aggregated spatial units (neighborhoods or census districts) we postulate that focusing efforts in large geographic areas with historically high levels of transmission within a city may be more effective than targeting households statistically identified as Ae. aegypti hotspots.
By following the same sampling and statistical methodologies, and by using information from roughly the same households as the ones studied by Getis et al. [15], we are able to confirm that Ae. aegypti adult distribution is highly focal, with average clustering not exceeding the household and its immediate neighbors. Also similar to Getis et al. [15], our study shows that clusters of high pupal abundance were rare and, when present, they rarely exceeded beyond a single household. These findings are in agreement with reports from Thailand indicating average local clustering values of 15 m [26] and from Ecuador with clustering values for pupae and adults of up to 20 m and 10 m, respectively [29], but differ from a recent report from Argentina reporting clusters of pupal abundance extending up to 400 m [28]. One of the main factors explaining the difference between studies relates to the methodology used to assess clustering. For example, Garelli et al. [28] analyzed data using a test that does not account for the inherent clustered pattern of houses within blocks. In our study, like Getis et al. [15], we accounted for such bias by comparing the distribution pattern of mosquitoes to the background distribution of households. The focal nature of Ae. aegypti distribution imposes important challenges to the integration of household-level information into predictive models of city-wide dynamics of vector distribution. An unresolved issue concerns tradeoffs in the cost and predictability of different strategies for assessing and responding to city-wide entomologic risk for DENV infection. For example, would it be more appropriate to implement (both in isolation and combined) quick, imperfect and spatially widespread entomologic indexes such as ovitraps or would it be better to use time consuming, more precise and spatially constrained indices, such as detailed adult/pupal indices?
Counting the absolute number of pupae in each larval development site has been recommended as a method for prioritizing containers requiring treatment in targeted larval development-site reduction strategies [52], [53]. Pupal counts are also considered a representative approximation of local adult mosquito populations [24], [53], and the pupae per person index is a frequently cited indicator for calculating a minimum threshold of pupal infestation for DENV transmission risk. Our study extends previous assessments of the association between pupae and adult abundance by showing that both indices rarely correlate with each other at spatial scales beyond the household and, when they do, they do so within 15 m of a house. Overall, the lack of proper consideration of spatial and temporal scales at which entomological measures are valid, as well as the limited inclusion of environmental, biological and human behavioral drivers of human-mosquito contacts, are important knowledge gaps in our ability to derive the maximum benefit out of entomological measures for surveillance and control programs [22].
The spatial pattern of Ae. aegypti distribution we detected was consistent across two neighborhoods that differed in mosquito infestation levels and DENV transmission. Maynas had high Ae. aegypti abundance and DENV transmission levels. Tupac Amaru had lower vector abundance and one of the lowest sero-incidence levels in the city of Iquitos [40], [41]. In both neighborhoods, however, Ae. aegypti populations were spatially clustered, clustering occurred at similar distances, and hotspots had a weak temporal persistence. Because most spatial analysis tests focus on relative rather than absolute patterns (i.e., compare observed values at location i to the overall mean), the finding of similar patterns in both neighborhoods may point to similar mechanisms driving Ae. aegypti population dynamics in them.
Ae. aegypti control is generally reactive (applied after the detection of local human DENV infections) and tends to rely on a geographic-based design in which interventions are applied at a given distance from a dengue case's residence [22]. Most programs use 100 m [54] as operational thresholds to deliver insecticides or other interventions, based on the premise that this distance represents the upper limit for Ae. aegypti dispersal. What this distance threshold does not take into account is that infected people can quickly move the virus well beyond 100 m of their home [6], [55], [56]. The lack of an empirically derived dispersal kernel (the probability of a given mosquito dispersing d meters away) for Ae. aegypti has further encouraged adoption of 100 m as the threshold for control measures. The focal pattern of Ae. aegypti adult distribution at the household level derived from our study suggests that adult flight beyond 30 m would be a rare event, provided food and habitat are available within such a radius. By integrating information on the extent of clustering of adult Ae. aegypti in two neighborhoods and over 9 entomologic surveys, we estimated the probability of finding spatially correlated populations, which could emerge due to dispersal and mixing of adult populations located in neighboring premises. We consider such estimates as a proxy of a dispersal kernel for adult Ae. aegypti. Our analysis indicates that, regardless of the background infestation levels in a neighborhood, the probability of finding Ae. aegypti adults dispersing beyond the house decreases exponentially with distance, being very low (∼6%) at 100 m. Our observations are in agreement with mark-release-recapture data suggesting that most individual adult Ae. aegypti do not fly far from the household where they developed as larvae (or were released as adults) [9], [57]–[59]. In addition to not accounting for longer range movements by virus infected humans (6), our results indicate that vector control activities applied at 100 m from a case's house will be a highly inefficient use of resources because it dramatically overestimates the actual extent of entomological risk associated with a potential transmission hotspot.
|
10.1371/journal.pcbi.1003637 | Essential Plasticity and Redundancy of Metabolism Unveiled by Synthetic Lethality Analysis | We unravel how functional plasticity and redundancy are essential mechanisms underlying the ability to survive of metabolic networks. We perform an exhaustive computational screening of synthetic lethal reaction pairs in Escherichia coli in a minimal medium and we find that synthetic lethal pairs divide in two different groups depending on whether the synthetic lethal interaction works as a backup or as a parallel use mechanism, the first corresponding to essential plasticity and the second to essential redundancy. In E. coli, the analysis of pathways entanglement through essential redundancy supports the view that synthetic lethality affects preferentially a single function or pathway. In contrast, essential plasticity, the dominant class, tends to be inter-pathway but strongly localized and unveils Cell Envelope Biosynthesis as an essential backup for Membrane Lipid Metabolism. When comparing E. coli and Mycoplasma pneumoniae, we find that the metabolic networks of the two organisms exhibit a large difference in the relative importance of plasticity and redundancy which is consistent with the conjecture that plasticity is a sophisticated mechanism that requires a complex organization. Finally, coessential reaction pairs are explored in different environmental conditions to uncover the interplay between the two mechanisms. We find that synthetic lethal interactions and their classification in plasticity and redundancy are basically insensitive to medium composition, and are highly conserved even when the environment is enriched with nonessential compounds or overconstrained to decrease maximum biomass formation.
| Plasticity and redundancy are essential features of biological systems, from brain to genome, that underlie the ability of organisms to survive. In metabolic networks, these essential mechanisms are unveiled by the analysis and categorization of synthetic lethal pairs of reactions. We find that plasticity acts as a backup mechanism that reorganizes metabolic fluxes, while redundancy corresponds to a simultaneous use of different flux channels that increases fitness. Both capacities ensure viability and are highly insensitive to environmental conditions, but plasticity seems a more sophisticated mechanism requiring a more complex functional organization. Our results have clear implications for biotechnology and biomedicine, since targeting a plasticity or a redundancy synthetic lethal pair will certainly have different requirements and peculiar specific effects.
| Homeostasis in living systems balances internal states over a possibly wide range of internal or external variations, and organisms unable to maintain stability in front of internal disruptions or changing environments can experience dysfunction and even collapse. At the level of cellular metabolism, multiple regulatory mechanisms control homeostasis, including allosteric or posttranslational activation of enzymes and modulation of enzymatic activity by transcriptional regulation [1]. Research efforts have typically focused on elucidating the molecular basis of such controls, but we are still far from understanding the complex functional strategies that explain homeostasis at a systems level, even when this concept is loosened to that of maintaining viability in front of perturbations.
Beyond the molecular level, plasticity and redundancy are large-scale strategies that offer the organism the ability to exhibit no or only mild phenotypic variation in front of environmental changes or upon malfunction of some of its parts. In particular, these mechanisms protect metabolism against the effects of single enzyme-coding gene mutations or reaction failures such that most metabolic genes are not essential for cell viability. However, some mutants fail when an additional gene is knocked out, so that specific pair combinations of non-essential metabolic genes or reactions become essential for biomass formation. As an example, double mutants defective in the two different phosphoribosylglycinamide transformylases present in Escherichia coli –with catalytic action in purine biosynthesis and thus important as crucial components of DNA, RNA or ATP– require exogenously added purine for growth, while single knockout mutants do not result in purine auxotrophy [2].
These synthetic lethal (SL) combinations [3]–[6] have recently attracted attention because of their utility for identifying functional associations between gene functions and, in the context of human genome, for the prospects of new targets in drug development. However, inviable synthetic lethal mutants are difficult to characterize experimentally despite the high-throughput techniques developed recently [7]. We are still far from a comprehensive empirical identification of all SL metabolic gene or reaction pairs in a particular organism [8], even more when considering different growth conditions. Metabolic screening based on computational methods, such as Flux Balance Analysis (FBA) simulating medium dependent optimal growth phenotypes, becomes then a powerful complementary technique particularly suited for an exhaustive in silico prediction of SL pairs [8], [9] in high-quality genome-scale metabolic reconstructions [10].
With this perspective, we unveil how functional plasticity and redundancy are essential systems-level mechanisms underlying the viability of metabolic networks. In previous works on cellular metabolism [11], [12], plasticity was some times associated to changes in the fluxes of reactions when an organism is shifted from one growth condition to another. Instead, we discuss here functional plasticity as the ability of reorganizing metabolic fluxes to maintain viability in response to reaction failures when the growth condition remains guaranteed. On the other hand, functional redundancy applies to the simultaneous use of alternative fluxes in a given medium, even if some can completely or partially compensate for the other [13]. We perform an exhaustive computational screening of synthetic lethal reaction pairs in E. coli in a glucose minimal medium and we find that SL reaction pairs divide in two different groups depending on whether the SL interaction works as a backup or as a parallel use mechanism, the first corresponding to essential plasticity and the second to essential redundancy. When comparing the metabolisms of E. coli and Mycoplasma pneumoniae, we find that the two organisms exhibit a large difference in the relative importance of plasticity and redundancy. In E. coli, the analysis of how pathways are entangled through SL pairs supports the view that redundancy SL pairs preferentially affect a single function or pathway [3]. In contrast –and in agreement with reported SL genetic interactions in yeast [14]– essential plasticity, which is the dominant class in E. coli, tends to be inter-pathway but concentrated and unveils Cell Envelope Biosynthesis as an essential backup for Membrane Lipid Metabolism. Finally, different environmental conditions are tested to explore the interplay between these two mechanisms in coessential reaction pairs.
We use FBA [15] (see Materials and Methods) under glucose minimal medium to compute all SL reaction pairs in the iJO1366 E. coli metabolic network reconstruction [16], and also in the iJW145 M. pneumoniae model [17], [18] for a comparative analysis. A reaction pair deletion is annotated as inviable, and so as a synthetic lethal, if the double mutant shows a no-growth phenotype. We preliminarily reduce the space of reactions to be considered in forming potential SL pairs to the set of reactions that can be active but not essential in glucose minimal medium, irrespective of the level of attainable growth (see Materials and Methods). This ensemble, formed of reactions in E. coli (see Supplementary Data Table S1) and in M. pneumoniae (see Supplementary Data Table S9), is a subset of the original reconstruction that includes but that is not limited to the set of FBA active reactions under maximum growth constraint [19], [20]. Next, we analyze in detail the classification of identified SL reaction pairs into plasticity and redundancy subtypes (Fig. 1).
We found that of all reaction pair deletions in E. coli are in silico synthetic lethals and can be separated in two different subtypes. In the biggest group, having a relative size of , one of the paired reactions is active in the medium under evaluation while the second reaction has no associated flux. The rest of SL reaction pairs are formed by two active reactions.
As discussed previously in [8], some FBA computationally predicted SL pairs can be inconsistent since they contain at least one gene reported as essential in vivo. In our screening, and in accordance with results in [8], this situation corresponds to of all identified SL pairs (see Materials and Methods). Apart from these inconsistencies, active-inactive coessential reaction pairs are referred to as plasticity synthetic lethality (PSL) pairs (Fig. 1a). We found PSL reaction pairs, of all diagnosed SL pairs in iJO1366 (Fig. 2) (see Supplementary Data Table S2). Coessential inactive and active reactions in these pairs have zero and non-zero FBA flux respectively. When the active reaction is removed from the metabolic network, fluxes reorganize such that the zero-flux reaction in the pair turns on as a backup of the removed reaction to ensure viability of the organism, even though the growth is generally lowered. In contrast, the level of growth is unperturbed when the inactive reaction is removed. As an example, the SL pair valine-pyruvate aminotransferase and valine transaminase form a PSL pair, the second reaction being the backup of the first (their simultaneous knockout produces auxotrophic mutants requiring isoleucine to grow [21]).
While the single activation of one of the reactions in a PSL pair is enough to ensure viability in front of single reaction disruptions, the parallel use of both coessential reactions may happen in other cases. We name redundancy synthetic lethality (RSL) reaction pairs those in which both reactions are active and used in parallel (Fig. 1b). Of all SL reaction pairs in iJO1366, we found that () are RSL (Fig. 2) (see Supplementary Data Table S2). We checked indeed that for of the RSL pairs the simultaneous use of both reactions increases fitness as compared to the situation when only one of the reactions is active (fitness is here understood as the maximal FBA biomass production rate for the organism). For the remaining two pairs growth remains unchanged. As an illustrative example of parallel use, oxygen transport combines with reactions in the ATP forming phase of Glycolysis to form RSL reaction pairs. If Oxydative Phosphorylation is blocked by the absence of oxygen and no alternative anaerobic process like Glycolysis is used, the energy metabolism of E. coli collapses and so the whole organism.
Network distance (see Materials and Methods) between reaction counterparts is slightly shorter in RSL pairs than in PSL pairs. Indeed, not all reactions in RSL or PSL pairs are directly connected through common metabolites. Direct connections happen for and of pairs respectively, while the rest can be separated by up to four other intermediate reactions so that the average shortest paths are and , respectively (the average shortest path of the whole metabolic network is ). Both essential plasticity and redundancy display overlap in reactions and associated genes. In the RSL pairs, we identified different reactions controlled by genes or gene complexes. The PSL pairs involve different reactions controlled by genes or gene complexes.
Although our analysis refers to reactions, specific signatures of enzyme activity may be worth stressing in connection with our analysis of coessential reaction pairs. For some of the identified SL pairs, we found direct experimental evidence reported in the literature [2], [21] and other results that support the buffering activity of reactions in some SL pairs, like in the aerobic/anaerobic synthesis of Heme [22], [23] and in the oxidative/non-oxidative working phases of the Pentose Phosphate Pathway [24]. In other cases, we found that enzymatic degeneracy can be responsible for explaining two of the in silico detected RSL reaction pairs in E. coli. One RSL reaction pair –that produces isopentenyl diphosphate and its isomer dimethylallyl diphosphate, biosynthetic precursors of terpenes in E. coli that have the potential to serve as a basis for advanced biofuels [25]– is catalyzed by a single enzyme encoded by an essential gene (one-to-many enzyme multifunctionality (Fig. 1f)). Conversely, isoenzymes are encoded by different genes but can catalyze the same biochemical reactions. This many-to-one relationship ensures that single deletion mutants lacking any of the genes encoding one of the isoenzymes can still be viable (Fig. 1f). We found that this case happens in RSL reaction pair catalyzed by isoenzymes encoded by nonessential genes associated to transketolase activity in the Pentose Phosphate Pathway [26].
Finally, a comparative study shows that coessential reaction pairs are times more abundant in a much simpler genome-reduced organisms of increased linearity and reduced complexity such as M. pneumoniae [27], [28]. We found that of all potential candidate reaction pairs in M. pneumoniae are synthetic lethals (see Supplementary Data Table S10), vs solely the in E. coli. Inconsistencies are also much more abundant relatively to E. coli and the balance of RSL vs PSL reaction pairs is also different (Fig. 2). Parallel use happens as frequently as the backup mechanism in coessential reactions, with of all synthetic lethals being RSL pairs and being PSL pairs. As compared to results reported in [17] for the synthetic lethality of genes, our methodology detects the same SL gene pairs and new SL gene pairs. Since the different genes in these pairs form two different complexes of and genes and one gene remains isolated, the SL gene pairs reduce to just SL reaction pairs (in the RSL and RSL I categories) sharing one of the reactions. The reactions involved in the pairs are uptake of G3P, G3P oxidation to dihydroxyacetone phosphate, and uptake of orthophosphate. As reported in Ref. [17], two independent routes through third-party pathways connect glycolysis to lipid biosynthesis. The first two reactions above, R1 and R2, are involved in one of the routes, while the last reaction R3 influences the flux through the other route. When R1 and R3 or R2 and R3 are removed from iJW145 model, the organism collapses due to the simultaneous failure of both routes.
To investigate further the role of essential plasticity and redundancy in the global organization of metabolic networks, we studied the entanglement of biochemical pathways through synthetic lethality. We annotated all reactions in synthetic lethal pairs in terms of the standard metabolic pathway classification and counted the frequencies of dual pathways combinations both for plasticity and redundancy subtypes. In Fig. 3, we give a visual summary of pathways entanglement through essential plasticity and redundancy using a graph representation where pathways are linked whenever they participate together in a SL interaction (discontinuous lines represent redundancy SL interactions (Fig. 3c) and continuous arrows stand for plasticity SL interactions (Fig. 3d)). The frequency of a given pathway combination in RSL or PSL pairs defines the weight of the corresponding link.
In E. coli (Fig. 3a), we observe that the synthetic lethality entanglement of pathways is in general very low, with the exception of the entanglement between Cell Envelope Biosysthesis and Membrane Lipid Metabolism. Redundancy SL pairs are basically intra-pathway, with only 3 of 15 being inter-pathway. Of all intra-pathway RSL pairs, concentrate in the Pentose Phosphate pathway. Interestingly, the distribution of PSL reaction pairs avoids that of RSL pairs and, in contrast, tends to be inter-pathway. Of all PSL pairs, include zero-flux reactions in Cell Envelope Biosysthesis and active reactions in the Membrane Lipid Metabolism, which unveils Cell Envelope Biosysthesis as an essential backup for Membrane Lipid Metabolism. Intra-pathway plasticity coessentiality amounts to of PSL pairs and is concentrated in Cofactor and Prosthetic Group and Cell Envelope Biosynthesis.
In M. pneumoniae (Fig. 3 b) pathways entanglement through coessentiality of reactions is very low as in E. coli. Redundancy SL pairs can be intra-pathway ( of ) or inter-pathway ( of ) and PSL pairs are basically intra-pathway ( of ). Redundancy SL pairs denote the parallel use of reactions in Folate Metabolism and reactions in Nucleotide and Cofactor Metabolism. These two pathways, Folate and Nucleotide Metabolism, are also linked by PSL pairs with non-essential reactions in Folate Metabolism and essential reaction backups in Nucleotide Metabolism. Nucleotide Metabolism is also the pathway that concentrates most PSL pairs. Both RSL and PSL reaction pairs unveil Nucleotide and Folate Metabolism as the most entangled pathways.
We finally investigate whether the subtype of a coessential pair switches between plasticity and redundancy depending on the growth condition under evaluation. Environmental specificity of genes and reactions has been explored experimentally [16], [29], [30] and in silico [31] for different organisms and for random viable metabolic network samples, and it has also been extended to multiple knockouts in yeast [9], [12] and E. coli [32].
We investigate the sensitivity of SL reaction pairs in E. coli to changes in minimal medium composition. We focus on the SL pairs detected in glucose minimal medium, and we check their classification over minimal media (formed of mineral salts and one variable source of carbon, nitrogen, sulfur and phosphorus [16], see Materials and Methods). In Fig. 4a, we display the SL reaction pairs ranked by the fraction of media in which the pairs are synthetically lethal. We find that for most pairs, coessentiality is not specific of an environment and only a minimal number of pairs shows environmental specificity. In particular, coessential pairs are lethal in all media and are lethal in more than of environments. For each SL pair, we count the number of media in which the SL pair is classified in the plasticity subtype as compared to the total number of media in which the pair is predicted to be coessential. Results are shown in Fig. 4b. Nearly all SL pairs, , are in the plasticity subclass for more than of the media, while pairs display a switching behavior between plasticity and redundancy. Noticeably, these pairs are intra-pathway and share common metabolites. Of them, pairs contribute to biosynthesis of amino acids (Valine, Leucine, and Isoleucine Metabolism and Glycine and Serine Metabolism) and pairs belong to the Pentose Phosphate Pathway and are related to the production of carbon backbones used in the synthesis of aromatic amino acids. Finally, reaction pairs maintain the redundancy subclass across all conditions in which are coessential.
We also explored the behavior of E. coli in amino acid-enriched medium (see Materials and Methods). Comparing with glucose minimal medium, our first observation is that 223 of the 234 SL pairs detected in glucose minimal medium are also found to be lethal in amino acid-enriched medium (see Supplementary Data Table S3), which means that 11 pairs are rescued. Of the 11 RSL pairs in amino acid-enriched medium, 8 are conserved and 3 switch from plasticity in the minimal to redundancy in the amino acid-enriched medium. On the other hand, of the PSL pairs are conserved and 4 change from redundancy in the minimal to plasticity in the amino acid-enriched medium. Noticeably, only in of the conserved PSL pairs the pattern of activity changes from the reductase reaction producing dimethylallyl diphosphate to the isomerization of the less reactive isopentenyl pyrophosphate. In addition, a set of lethal reaction pairs () occur, all of them involving however one essential reaction in glucose minimal medium that in amino acid-enriched medium becomes nonessential and instead takes part in a SL pair. Apart from those, no other new SL pairs are found. In addition, we redid our simulations taking into consideration a LB based rich medium [33], [34] (see Supplementary Data Table S4). In this rich medium, we only found new rescues when compared to the minimal medium (two new rescues as compared to the amino acid-enriched medium) and only SL pairs change their plasticity/redundancy category (see Supplementary Data Table S5).
We found that plasticity and redundancy are still conserved when the growth maximization requirement is loosen (see Materials and Methods). If growth is relaxed in E. coli to of its maximum value in glucose minimal medium, in silico essentiality of individual reactions does not change but activation of reactions increases. We find however that the effect of this reorganization is indeed mild for plasticity and redundancy. When biomass production is relaxed by overconstraining the upper bounds of the uptake rates of mineral salts, all SL pairs are conserved and of them maintain their PSL or RSL classification. The absolute number of RSL pairs increases from to since of RSL pairs in the reference conditions given by glucose minimal medium change to plasticity in the overconstrained medium, and at the same time PSL pairs change to RSL. On the other hand, SL pairs of in the reference medium remain as PSL pairs in the overconstrained condition. However, the pattern of activity in the pair has switched in of the PSL pairs in this case, which indicates that the specific selection of the active reaction in a PSL pair can have an impact in the level of attainable growth (see Supplementary Data Table S6). If instead of limiting the uptake of mineral salts we overconstrain the uptake rates of basic nutrients providing sources of carbon, nitrogen, phosphorus and sulfur, the effect is even softer and indeed negligible as compared to the reference medium. The number of active reactions only increases in , the essentiality of individual reactions and SL pairs is conserved, and of them maintain their PSL or RSL classification with only SL pairs that switch class and only PSL pair that changes the active reaction (see Supplementary Data Table S7).
Synthetic lethals are complex functional combinations of genes or reactions that denote at the same time both vulnerability in front of double deletions and robustness in front of the failures of any of the two counterparts. Although synthetic lethal genes could be associated to the plasticity and redundancy categories, approaching directly pairs of reactions without the (sometimes multifunctional) scaffold of enzymes and genes allows us to determine in a clean and systematic way the minimal combinations of reactions that turn out to be essential for an organism. Working at the level of reactions, we showed that this synthetic lethality is meditated by two different mechanisms, essential plasticity and essential redundancy, depending on whether one reaction is active for maximum growth in the medium under consideration and the second inactive, or in contrast both reactions have non-zero flux. Plasticity sets up as a sophisticated backup mechanism (mainly inter-pathway in E. coli) that is able to reorganize metabolic fluxes turning on inactive reactions when coessential counterparts are removed in order to maintain viability in a specific medium, while redundancy corresponds to a simultaneous use of different flux channels (mainly intra-pathway in E. coli) that ensures viability and besides increases fitness. Apparently, it could seem extremely improbable that the removal of an inactive reaction together with a non-essential active one, like in PSL pairs, could have any lethal effect on an organism. However, we found that this situation is indeed overwhelmingly dominant in E. coli as compared to redundancy synthetic lethality, and it is still relatively frequent even in a less complex organism like M. pneumoniae.
Synthetic lethal mutations have been assumed to affect a single function or pathway [3], which reinforces the idea that pathways act as autonomous self-contained functional subsystems. In contrast, other investigations in yeast [14] report that synthetic-lethal genetic interactions are approximately three and a half times as likely to span pairs of pathways than to occur within pathways. In our work, we found that RSL pairs in E. coli are predominantly intra-pathway while PSL pairs, more abundant, tend to be inter-pathway although concentrated in the entanglement of just two pathways, Cell Envelope Biosynthesis and Membrane Lipid Metabolism. The comparative study here shows that although pathways entanglement through coessentiality of reactions is low in both organisms, RSL pairs in M. pneumoniae can be intra-pathway or inter-pathway, linking Folate Metabolism and Nucleotide and Cofactor Metabolism, and PSL pairs are basically intra-pathway and located in Nucleotide Metabolism. Taken together, these results indicate that Folate and Nucleotide Metabolic pathways preserve most rescue routes for reaction deletion events, in accordance with results in [17]. The fact that the proportion of plasticity SL pairs is considerably decreased in M. pneumoniae as compared to E. coli could be indicative that, even if both plasticity and redundancy serve an important function in achieving viability, essential plasticity is a more sophisticated mechanism that requires a higher degree of functional organization, using at the same time less resources for maximum growth. At the same time, this can also be explained by the relative unchanging environmental conditions of M. pneumoniae in the lung, that could have induced the elimination of pathways not required in that medium [17]. This suggests that the adaptability of M. pneumoniae is very much reduced and its SL behavior could not be resilient to environmental changes.
We also found that SL reaction pairs and their subdivision in plasticity and redundancy are highly conserved independently of the composition of the minimal medium that acts as environmental condition for growth, and even when this environment is enriched with nonessential compounds or overconstrained to decrease the maximum biomass production. These environment unespecific SL pairs can thus be selected as potential drug targets operative regardless of the chemical environment of the cell. We admit that large-scale computational screenings might be biased both by model details and by the quality of the experimental data used to feed the metabolic reconstruction, which could affect the identification of SL pairs and their categorization in plasticity and redundancy. However, we believe that FBA and the genome-scale metabolic reconstructions that we use in this work have proven to be reliable and highly congruent with empirical observations and that, with few exceptions, our results will pass the test of new modeling approaches that may become the reference in the future.
Our work was intended as an advancement in setting the basis for an analysis of essential plasticity and essential redundancy in metabolic networks. The elucidation of these capacities through synthetic lethal interactions has clear implications for biotechnology and biomedicine, since targeting a PSL or a RSL pair will certainly have different requirements and implications. In human cells, the analysis of PSL and RSL pairs can help the design of new drug targets, once we solve the enormous tasks challenging the accurate understanding of cell metabolism in humans, e.g., the lack of association between the majority of genes and a recognized function and the deficiency in reconstruction models for each specific human cell type. Beyond metabolic networks, plasticity and redundancy are also very important concepts for other biological complex systems, like the brain. Evolution has been postulated as the main explanation for the plasticity and redundancy abilities observed in the brain, with the hypothesis that when new parts of the brain evolved and took over the old brain functions, the old areas that mediated those functions still retained some control [35]. Whether essential plasticity and essential redundancy are adaptive in cell metabolism or, as it has been argued for metabolism in changing environments [12], [36], they are rather a byproduct of the evolution of biological networks toward survival, these regulatory mechanisms are key to understand how complex biological systems protect themselves against malfunction.
Preliminary to the identification of SL pairs using FBA, we filter the set of reactions in the metabolic reconstruction of the organisms to keep only those potentially relevant. As a first step, we only take into account reaction pairs with associated genetic information. Hence, spontaneous reactions or reactions with unknown associated genes are discarded, leading to the elimination of 117 reactions for E. coli and 48 reactions for M. pneumoniae.
To identify potential synthetic lethal reaction pairs, first we compute which reactions can have a non-zero flux in a particular medium composition (glucose minimal medium/amino acid-enriched medium/rich medium). To do this, we follow an approach similar to Flux Variability Analysis (FVA) [20]. FVA consists on computing the minimum and maximum values of the fluxes that each reaction can have, taking into account that the growth rate must be fixed to a value that ensures viability. However, being conservative, we are interested in capturing all the possible scenarios independently of the value of the flux of the biomass reaction, since in this way we can take also into account non-optimal/low-growth scenarios. Therefore, we modify FVA to compute the minimum and maximum possible values of the flux of each reaction regardless of the value of the biomass formation rate. To this end, we do not constrain the value of the flux of the biomass reaction and we just allow any positive value, . Under this condition, we determine which reactions have maximum or minimum values of the fluxes different from . By doing this we obtain the maximal set of reactions which can be active in the particular medium under consideration independently of the rate of biomass formation of the organism.
Flux Balance Analysis is a method to compute the fluxes of reactions in metabolic networks at steady state. In its standard version, FBA optimizes biomass production without using kinetic parameters [15]. This method is able to predict the growth rate and the fluxes of a metabolic network [37], [38] with high accuracy. In particular, FBA was shown to predict gene essentiality with an accuracy of 90% [36], [39]. It was also used for in silico prediction of SL pairs [8], [9].
We implement FBA using GNU Linear Programming Kit (GLPK) to compute synthetic lethality of reaction pairs. Once we have determined the space of reactions to be considered in potential SL pairs using FVA (see Materials and Methods subsection above “Identification of the space of potential reactions in SL pairs: biomass unconstrained Flux Variability Analysis”), we apply an exhaustive search checking all potential pairs. For each possible combination, we compute FBA on the mutant obtained by removing the corresponding pair of reactions from the metabolic network model. We annotate the double deletion as inviable, and so as a synthetic lethality pair, if FBA shows a no-growth phenotype. Notice that reversible reactions are treated as two coupled reactions to account for the forward and reverse fluxes.
In order to validate our methodology (including the FVA reaction set selection), we computed in silico single essential reactions and SL reaction pairs for the iAF1260 model of E. coli in glucose minimal medium. Our results are in perfect agreement with those reported in [8].
Pairs, formed by reactions, both having an associated gene or genetic entity, are checked according to their experimental essentiality. For E. coli, we use the information given in Ref. [16]. For M. pneumoniae, we use results from a genome-wide transposon study in M. genitalium given in Ref. [40]. A functional ortholog in M. genitalium can be assigned to 128 metabolic genes in iJW145, and we associate the essentiality of that ortholog to the corresponding gene in M. pneunomiae. The other genes are assumed (similarly to Ref. [17]) to be not essential for growth due to their absence in M. genitalium and the high similarity of the metabolic networks of both organisms (Ref. [18]). Three cases may occur:
Besides, detected SL pairs with an active and an inactive reaction (PSL pairs) associated to isoenzymes (2 occurrences) and multifunctional enzymes (1 occurrence) are classified as inconsistencies.
We model metabolism as a bipartite directed network [28], where directed links connect metabolites with reactions in which they participate as reactants or products. The shortest path length measures the topological distance between reactions in this network representation. It is calculated as the minimal number of different intermediate nodes (metabolites and reactions) visited when going from one reaction to another following the directed links. In practice, we use Dijkstra's Algorithm [41]. to compute these network-based distances.
|
10.1371/journal.ppat.1000001 | A Mycobacterial Enzyme Essential for Cell Division Synergizes with Resuscitation-Promoting Factor | The final stage of bacterial cell division requires the activity of one or more enzymes capable of degrading the layers of peptidoglycan connecting two recently developed daughter cells. Although this is a key step in cell division and is required by all peptidoglycan-containing bacteria, little is known about how these potentially lethal enzymes are regulated. It is likely that regulation is mediated, at least partly, through protein–protein interactions. Two lytic transglycosylases of mycobacteria, known as resuscitation-promoting factor B and E (RpfB and RpfE), have previously been shown to interact with the peptidoglycan-hydrolyzing endopeptidase, Rpf-interacting protein A (RipA). These proteins may form a complex at the septum of dividing bacteria. To investigate the function of this potential complex, we generated depletion strains in M. smegmatis. Here we show that, while depletion of rpfB has no effect on viability or morphology, ripA depletion results in a marked decrease in growth and formation of long, branched chains. These growth and morphological defects could be functionally complemented by the M. tuberculosis ripA orthologue (rv1477), but not by another ripA-like orthologue (rv1478). Depletion of ripA also resulted in increased susceptibility to the cell wall–targeting β-lactams. Furthermore, we demonstrate that RipA has hydrolytic activity towards several cell wall substrates and synergizes with RpfB. These data reveal the unusual essentiality of a peptidoglycan hydrolase and suggest a novel protein–protein interaction as one way of regulating its activity.
| Mycobacteria, like all peptidoglycan-containing bacteria, must extend and cleave the surrounding structurally rigid layer of peptidoglycan to grow and divide. The peptidoglycan hydrolases responsible for this cleavage often have redundant functions, both revealing their importance and making them difficult to study. Furthermore, such hydrolases must be tightly regulated, due to their potentially lytic nature. We recently demonstrated the interaction between a lytic transglycosylase (Rpf) and an endopeptidase (RipA) at the septum of dividing bacteria. To investigate the role of these two hydrolases, we generated conditional mutants of each and were surprised to find that depletion of ripA resulted in long chains of cells. This phenotype was reversed upon induction of ripA, indicating that cell wall expansion and septum formation can be decoupled from the process of septum resolution. In addition, we present data showing that the combination of Rpf and RipA results in enhanced hydrolysis of peptidoglycan in an in vitro assay, suggesting protein–protein interactions as one potential mechanism of regulation.
| Though not formally considered virulence factors, genes required for bacterial cell division clearly are necessary for the growth, and thus, pathogenesis, of bacteria. The distinction between homeostatic and virulence genes is blurred when nonessential genes involved in vegetative cell division become essential under specific stressful conditions encountered inside a host. Such an example is seen with the resuscitation-promoting factors (Rpf) encoded by many different bacteria, including mycobacteria. These proteins are named for their ability to resuscitate nonreplicating dormant bacteria. The single Rpf-encoding gene in Micrococcus luteus [1] is essential, but as many as three of the five genes encoding RpfA-E can be deleted in Mycobacterium tuberculosis without markedly affecting in vitro growth. However, one single deletion (rpfB) and several of the triple combinations yielded strains unable to grow or divide in stressful conditions in vitro and in vivo [2],[3]. This suggests that certain potential cell division proteins that appear to play nonessential roles in homeostatic processes can become vital in conditions of stress.
The vital processes of cell growth and division involve the temporal and spatial coordination of events such as peptidoglycan and cell wall extension, DNA replication, chromosomal partitioning, Z-ring assembly, septum formation, and cytokinesis. Much of the mechanism behind this coordination involves inhibiting and stabilizing proteins that regulate the eventual assembly of a Z-ring at the midcell of bacteria [4],[5]. This Z-ring consists primarily of a polymerized ring of tubulin-like FtsZ on the cytoplasmic side of the plasma membrane, stabilized by membrane-associated and integral membrane proteins. Assembly occurs in an ordered fashion that is not entirely linear, with some components assembling before joining the Z-ring [6],[7],[8]. Some of the last proteins to be recruited to the Z-ring are thought to be the peptidoglycan hydrolyzing enzymes, such as AmiC [9] and EnvC [10] in E. coli. These enzymes digest the peptidoglycan layers connecting two recently developed daughter cells in the final stage of cell division [11]. While crucial to cell division, the regulation of these potentially lethal enzymes is poorly understood. It is thought that protein-protein interactions play a role in regulating activity and localization [12]. Studying cell wall hydrolases has proven difficult due to the large number encoded in most bacterial genomes and the high degree of functional redundancy. For example, seven hydrolases can be deleted from a strain of E. coli without loss of viability [13].
Little is known about the hydrolases involved in mycobacterial cell division. CwlM and Rv2719c were both shown to be mycobacterial cell wall hydrolases [14],[15]. Two recently identified hydrolases, RpfB and RpfE, were shown to interact with rpf-interacting protein (RipA), a peptidoglycan endopeptidase [16]. Rpf proteins constitute a family of lytic transglycosylase enzymes capable of hydrolyzing the glycosidic bonds in the essential stress-bearing, shape-maintaining peptidoglycan layer [17]. RpfB has a structure similar to the E. coli soluble lytic transglycosylase 70 (Slt70) [18] and is known to hydrolyze the beta-1,4-glycosidic bond between N-acetyl muramic acid and N- acetyl glucosamine [19]. RipA has been shown to be a peptidoglycan hydrolase [16]. It is predicted to function as an L,D-endopeptidase, capable of hydrolyzing D-glutamyl-meso-diaminopimelic acid [20], two amino acids that are part of the crosslinking peptides vital for keeping peptidoglycan rigid and stable [21]. Both RpfB and RipA localize to the septa of dividing bacteria [16] and thus may play a role in the late stages of mycobacterial cell division, possibly during regrowth from a stressed state.
Here we show that depletion of ripA in a strain of M. smegmatis results in a significant decrease in growth, formation of long, branched chains, and increased sensitivity to a cell wall–targeting antibiotic. These defects can be functionally complemented with the M. tuberculosis allele of ripA. We demonstrate that the peptidoglycan hydrolytic activity of RipA synergizes with RpfB. Thus, this protein is an unusual example of an essential peptidoglycan hydrolase whose activity may be partially regulated through protein-protein interactions.
RipA interacts with RpfB, a lytic transglycosylase, and colocalizes at the septum [16]. We hypothesized that the RipA-RpfB complex may be involved in degrading peptidoglycan at the septum during cell division. To further investigate the function the individual components of this complex, we attempted to make deletion strains of ripA and rpfB in M. smegmatis. Though disruption of rpfB was successful, we were unable to disrupt the ripA gene in M. smegmatis. We have previously reported that the ripA gene in M. tuberculosis appears to be essential for in vitro growth [22], suggesting that it might also be essential in M. smegmatis.
To test this possibility we constructed depletion strains of M. smegmatis in which ripA (MSMEG3153) or rpfB (MSMEG5439) are transcribed from an inducible tetracycline promoter (Ptet, Figure 1A). We found that the ripA depletion strain had dramatically reduced growth in media lacking inducer (Figure 1B), while the rpfB depletion strain had normal growth in the absence of inducer. Both ripA and rpfB depletion strains grew normally in the presence of inducer. The growth phenotype seen with ripA depletion, as measured by optical density, was dose-dependent. While the optical density of ripA depleted cultures failed to increase in the absence of inducer, we did observe that bacteria formed visible clumps that increased in size during incubation. These clumps (due to filamentation) failed to suspend well and, therefore, were poorly measured using spectrophotometry. Remarkably, this phenotype is reversible, as addition of inducer to growth-arrested, ripA depleted cells resulted in the resumption of normal growth (Figure 1C). This indicates that the frequency of septum resolution can be uncoupled from septum formation and cell elongation. High levels of inducer did not result in gross morphological changes or lysis.
To further confirm the requirement of ripA for growth, we plated ripA-depletion or rpfB-depletion strains of M. smegmatis on plates containing a gradient of inducer. The ripA-depletion strain grew in a Tet-dependent manner, with the highest growth in the center corresponding to the highest concentration of inducer and no growth near the edges of the plate where inducer was the lowest, confirming the requirement of ripA for growth. Conversely, the rpfB-depletion strain grew similarly throughout the plate in a Tet-independent manner, indicating that rpfB is not required for growth under these conditions.
While rpfB depleted cells had normal morphology, ripA depleted cells had markedly abnormal shape (Figures 1C and 2A). They grew in long, branched chains that account for the clumps seen grossly. Staining with the fluorescent membrane dye, TMA-DPH, revealed periodic septa along the chains of bacteria. DNA staining with SYTO 9 revealed nucleoids along the length of the chained bacteria, separated by septa, indicating that DNA segregation and septum formation processes are intact (Figure 2B). Occasional patches where the cell wall appeared pinched or partially degraded were also observed. It is not clear if these represent defective division sites or locations where bacteria had begun to lyse.
Branches, not observed in wild-type cells, were seen in almost all ripA-depleted bacteria visualized (>95%). Interestingly, 91% (203/223) of the branches visualized originated directly adjacent to septa.
The M. tuberculosis ripA gene encodes a 472 amino acid protein that has been shown to degrade peptidoglycan [23]. Its C-terminal 105 amino acids contain a putative endopeptidase domain, which has 40% identity with the Listeria monocytogenes p60 protein (Figure 3A). p60 has been shown to be a cell wall endopeptidase by its ability to degrade cell wall [24],[25]. ripA is the first gene in a bicistronic operon. rv1478, the downstream gene, encodes a 241-amino acid protein, consisting of a signal sequence followed by a sequence homologous to the C-terminal half of RipA (70% putative hydrolase domain identity and 27% overall amino acid identity). Both genes in the apparent ripA operon encode predicted endopeptidase domains similar to a known p60 hydrolase [24] (Figure 3A). Because the inserted tetracycline-inducible promoter lies upstream of the operon, presumably transcription of both is dependent on the presence of inducer. Thus, either gene could be responsible for the observed phenotype. There are at least five p60-like genes in many of the mycobacterial species, including M. tuberculosis, M. smegmatis, and M. bovis BCG. These include rv0024, rv1477 (ripA), rv1478, rv1566c, and rv2190c. These genes lie in various genomic contexts and there is no function discernable from genomic synteny (Figure 3B).
To identify the responsible gene we tested if the M. tuberculosis allele of ripA (ripA-mtb) was able to complement the M. smegmatis with diminished native ripA (ripA-smeg) production. We expressed ripA-mtb from a zeocin-marked episomal plasmid. A ripA-smeg depletion strain containing the ripA-mtb construct grew similarly to wildtype in the absence of inducer, while a strain carrying an empty plasmid formed chains when ripA-smeg was depleted (Figure 3C). In contrast, the M. tuberculosis allele of the ripA paralogue, rv1478, was not able to complement an M. smegmatis ripA depletion. These results confirm that ripA is sufficient for complementing the strain depleted of the ripA (ripA-MSMEG3154) operon and show that ripA-mtb is functionally similar to ripA-smeg.
Because depletion of ripA may have a marked effect on cell wall structure, we reasoned that strains with diminished expression of ripA might have altered susceptibility to antibiotics that target the cell wall. To test this, we grew the M. Smegmatis regulated ripA strain in the presence of inducer, then washed and spread on plates containing various amounts of inducer (0, 10, 100 ng/ml Tet). A sterilized disc of Whatman paper was placed in the middle of the plate and 10 ul of a single antibiotic was added to the disc. After incubating for 3–4 days, the size of the zone of inhibition was measured (Figure 4). Complete depletion of ripA resulted in a remarkably high level of susceptibility to β-lactams (carbenicillin). It is unclear why depletion of ripA results in such an increase in susceptibility to β-lactams, which target the transpeptidase reaction required for cross-linking peptidoglycan during cell elongation and division. Susceptibility to cycloserine, an analog of D-alanine that inhibits the formation of the cytoplamsic pentapeptide that is eventually transported across the cell membrane and used to cross-link PG strands, was independent of induction of ripA. While increased permeability is often attributed to an observed increase in susceptibility to an antibiotic, depletion of ripA did not affect susceptibility to cycloserine, suggesting a more specific defect.
RipA has been predicted and shown to degrade M. luteus cell wall material [25],[26]. To test if RipA hydrolyzes peptidoglycan and cell wall material from other species of bacteria, we determined the enzymatic activity of RipA using a variety of FITC-labeled, cell wall–derived substrates. We expressed RipA as a fusion protein with GST in E. coli and purified the fusion protein using affinity chromatography. We found that GST-RipA, but not GST alone, was able to hydrolyze cell wall derived from M. smegmatis as well as peptidoglycan purified from Streptomyces, and had minimal activity against M. luteus cell wall (Figure 5B–D). Therefore, RipA is capable of hydrolyzing cell wall material from several bacterial species.
The predicted activity of RipA is to cleave the peptide cross-linkages in peptidoglycan and is distinct from Rpf, which is predicted to cleave glycosidic bonds in peptidoglycan (Figure 5A). Given the close proximity of these predicted cleavage sites, we hypothesized that the interaction of RpfB with RipA may result in enhanced hydrolytic activity. To test this possibility we expressed a portion of RpfB as a fusion protein with GST in E. coli and purified it using affinity chromatography.
Using the same assays described above, we found that GST-RpfB alone had minimal ability to degrade cell wall extracts or purified peptidoglycan. However, when GST-RpfB was combined with GST-RipA, activity was more than the sum of individual enzyme activities. The same result was found with all substrates tested (Figure 5B–D). No increase was detected when GST was combined with GST-RpfB or GST-RipA (data not shown). Addition of twice as much Rpf yielded no increase in hydrolysis, while twice as much RipA yielded twice the hydrolysis (data not shown) indicating the assay is in the linear range.
In this work, we demonstrate that RipA is essential for normal cell division in M. smegmatis, with its depletion resulting in long, branched filaments and increased susceptibility to a specific cell wall targeting antibiotic. Furthermore, RipA cleaves peptidoglycan and synergizes with RpfB. Taken together, these data support a model where RipA is 1) required for the final stage of cell division, where daughter cells are separated and 2) has peptidoglycan hydrolytic activity that may be modulated by RpfB under certain conditions.
It is unusual that RipA is essential for normal cell division in M. smegmatis and, apparently, M. tuberculosis [22]. Because bacteria encode a number of hydrolytic enzymes that are, at least in part, functionally redundant, strains carrying deletions of single hydrolase genes are generally viable, though combinations of mutations can result in lack of viability [13]. In M. smegmatis and M. tuberculosis, ripA does not appear to be redundant. Conversely, while M. marinum strains carrying mutations in the homologous gene, iipA, do have abnormal morphology, they are still able to divide. Mutations in the hydrolytic domain of IipA abolished complementation of the defect, confirming the importance of the hydrolytic activity of IipA [27]. In M. marinum, different rip paralogues might be able to complement for loss of iipA.
None of the rpf genes appears to be essential in M. tuberculosis and combinations of at least three rpf genes can be deleted in M. tuberculosis strains while still maintaining normal in vitro vegetative growth [2]. We demonstrate that RpfB is also not essential in M. smegmatis. It logically follows that the interaction between RpfB and RipA must not be essential for RipA function during vegetative growth. Of course, it is possible that another Rpf protein is able to compensate for the absence of RpfB, resulting in increased RipA-dependent activity. For example, RpfE is able to interact with RipA [16]. It is also possible that the RipA-RpfB interaction, and subsequent enhanced hydrolytic activity, is required only under special circumstances, such as growth under specific conditions of stress. As noted, RpfB is required for resuscitation of M. tuberculosis in a reactivation mouse model [3]. Likewise, deletion of several combinations of three rpf genes results in viable bacteria that are unable to resuscitate from in vitro and in vivo resuscitation assays [2]. Thus, the RipA-RpfB interaction may be necessary under certain conditions.
There are several models that might explain the cooperativity seen between RipA and RpfB. One protein might allosterically activate the other, resulting in increased peptidoglycan degradation. Alternatively, both proteins might be fully active, but their association might bring their active sites in close proximity, thus producing cleavage of bonds located near to one another in the peptidoglycan. Since peptidoglycan is a highly cross-linked polymer, nearby cleavages are more likely to effectively degrade peptidoglycan and release fragments.
Several of the most effective antibiotics, including many important antimycobacterial agents, target cell wall synthesis. RipA appears to represent a particular vulnerability for M. tuberculosis. In addition to its possible role in reactivation through interaction with Rpf, RipA is essential for normal cell division and is accessible to drugs, given its external localization. Inhibiting the enzymatic activity should block the ability of daughter cells to separate from one another, while blocking protein-protein interactions could result in dysregulation of activity. Thus, RipA is an attractive target for antimycobacterial drug development.
E. coli XL-1 (Stratagene) strains were used for cloning and E. coli BL21 (DE3) (Stratagene) was used for expression of recombinant proteins from the pET41a (Novagen) or pMal (NEB). Mycobacterium smegmatis (mc2155) and Mycobacterium tuberculosis (H37Rv) strains were grown at 37°C in Middlebrook 7H9 broth supplemented with ADC and Tween80 and antibiotic when appropriate.
The E. coli expression strain, BL21(DE3) was used to synthesize each protein following the Novagen manual protocol. Protein concentrations were measured using the Bradford assay, normalized, and confirmed by coomassie-stained polyacrylamide gels.
Protein samples were combined with 4× Laemmli's SDS PAGE buffer and boiled at 100°C for 10 minutes. Proteins were separated on 10% Tris-tricine polyacrylamide gels by electrophoresis, transferred to nitrocellulose, and probed with specific antibodies using standard techniques.
M. smegmatis cell wall was prepared as previously described [14]. Streptomyces peptidoglycan and lyophilized M. luteus cell wall were both obtained from Sigma. The fluorescein isothiocyanate (FITC)-labeled bacterial cell wall was prepared by covalently linking FITC to amine groups in the cell wall. 10 mg FITC (Molecular Probes) was used to label 10 mg of insoluble peptidoglycan or cell wall material following the protocol from Molecular Probes.
Recombinant M. tuberculosis proteins were incubated with several FITC-labeled cell wall substrates and assayed for activity by measuring FITC release. 25 µg of Rpf or RipA alone or 25 µg of Rpf and 25 µg RipA combined, were added to 25 µl of 2 mg/ml substrate and 25 µl 4× reaction buffer (50 mM Tris, 10 mM MgCl, 50 mM KCl, 2 mM MnCl, 0.01% Chaps, 100 mM KH2PO4, pH 5.75). The final volume was brought to 100 µl with H2O. As a control, 50 µg of lysozyme was added to M. smegmatis cell wall. Similar combinations with GST were also tested. GST alone, as well as buffer alone, were used to determine background release of FITC. After incubating at 30°C with enzyme and buffer for 3–5 days, the insoluble substrate was centrifuged (18,000×g) and soluble FITC was measured with filters for excitation 485 nm and emission 538 nm.
Depletion strains were generated as previously described [28]. Briefly, M. smegmatis, with the tetracycline repressor gene integrated into the attB site, was transformed with a suicide vector containing the first 600 nucleotides of M. smegmatis ripA gene under control of the tetracycline operator/promoter system (Ptet). Transformants were selected for hygromycin resistance. Appropriate recombination was confirmed using forward primers to Ptet and Prip (native ripA promoter) paired with a reverse primer to the 3′ end of ripA. The presence of a product of appropriate size for the former and lacking in the latter, confirmed the desired strain. Attempts to disrupt the ripA gene in M. smegmatis using a nonreplicating suicide vector designed to recombine into the middle of the gene were unsuccessful (though control knockouts, such as rpfB, were successful).
The ripA and rpfB depletion strains were initially grown in 7H9 media containing kanamycin (selecting for TetR) and hygromycin (selecting for inserted pTet) as well as anhydrotetracycline (Tet). Once cultures reached late log-phase or stationary phase, they were centrifuged (2500×g for 5 minutes), washed once with PBS, and resuspended in media with varying amounts of Tet. To test recovery of ripA depleted cells, Tet was either added directly to cultures grown without Tet or to fresh media inoculated with cells depleted of ripA. To test complementation, the ripA gene and its native promoter from M. tuberculosis was amplified and cloned into an episomal plasmid containing the zeocin gene as a marker. This construct, or the isogenic empty vector, was transformed into the ripA depletion strain of M. smegmatis. Strains were grown in the presence of tetracycline inducer, washed and inoculated into media lacking inducer. Cultures were monitored by OD600 and microscopy. To confirm the essentiality of ripA, depletion strains of M. smegmatis were grown on a plate with a gradient of inducer generated by placing 10 µl of 10 ng/ml Tet on a paper disc in the center of the plate, resulting in a concentration of inducer highest at the middle of the plate and lowest at the edges.
The ripA depletion strain of M. smegmatis was spread on LB agar plates containing different amounts of anhydrotetracycline inducer (ng/ml concentrations) to regulate the amount of ripA expressed. A filter disc with 10 µl of carbenicillin (100 mg/ml), isoniazid (10 mg/ml), or cycloserine (100 mg/ml) was placed in the center of plate and the diameter of inhibition of growth was measured after 4 days of growth.
M. smegmatis strains were centrifuged at 2500×g for 2 minutes, washed with 1ml PBS, and resuspended in 20 µl of PBS containing 50 nM TMA-DPH or 5 µM SYTO 9 for staining membranes or DNA, respectively. Samples were imaged with a Nikon TE-200 100× (NA 1.4) objective and captured with an Orca-II ER cooled CCD camera (Hamamatsu). Final images were prepared using Adobe Photoshop 7.0. |
10.1371/journal.pcbi.1004708 | A Sensory-Driven Trade-Off between Coordinated Motion in Social Prey and a Predator’s Visual Confusion | Social animals are capable of enhancing their awareness by paying attention to their neighbors, and prey found in groups can also confuse their predators. Both sides of these sensory benefits have long been appreciated, yet less is known of how the perception of events from the perspectives of both prey and predator can interact to influence their encounters. Here we examined how a visual sensory mechanism impacts the collective motion of prey and, subsequently, how their resulting movements influenced predator confusion and capture ability. We presented virtual prey to human players in a targeting game and measured the speed and accuracy with which participants caught designated prey. As prey paid more attention to neighbor movements their collective coordination increased, yet increases in prey coordination were positively associated with increases in the speed and accuracy of attacks. However, while attack speed was unaffected by the initial state of the prey, accuracy dropped significantly if the prey were already organized at the start of the attack, rather than in the process of self-organizing. By repeating attack scenarios and masking the targeted prey’s neighbors we were able to visually isolate them and conclusively demonstrate how visual confusion impacted capture ability. Delays in capture caused by decreased coordination amongst the prey depended upon the collection motion of neighboring prey, while it was primarily the motion of the targets themselves that determined capture accuracy. Interestingly, while a complete loss of coordination in the prey (e.g., a flash expansion) caused the greatest delay in capture, such behavior had little effect on capture accuracy. Lastly, while increases in collective coordination in prey enhanced personal risk, traveling in coordinated groups was still better than appearing alone. These findings demonstrate a trade-off between the sensory mechanisms that can enhance the collective properties that emerge in social animals and the individual group member’s predation risk during an attack.
| Many social species coordinate their movements as they travel together, which is generally considered to be an adaptive behavior predominantly mediated by vision. Collective coordination lowers the individual’s chances of becoming separated from the group and can reduce the time spent in risky areas. More broadly, information is believed to spread more rapidly among individuals as group level order increases. Yet, little attention has been payed to how collective order impacts a predator’s perception of their prey during an attack. Motion detection improves with the coherency of visual stimuli, which suggests that the sensory mechanisms that promote coordinated motion in prey may also improve predator perception and, consequently, attack success. Here we explore this conundrum using an interactive game with humans acting as surrogate predators attacking collections of virtual prey. By isolating visually mediated behaviors in both prey and predator we demonstrate the potential for a sensory-driven trade-off to arise between a group’s collective order and the individual’s risk of being captured.
| Organisms are constantly challenged by the need to effectively extract and process pertinent information from uncertain and often dangerous environments. Many species manage to reduce their uncertainty through social information. When animals gather together and coordinate their activities the members can enhance their own perception through collective vigilance, which increases the speed and accuracy of individual decisions and reduces predation risk [1–3]. Predators can also be cognitively challenged when presented with multiple targets (the confusion effect), thereby lowering their ability to process information and, consequently, further reducing a prey’s risk of being killed. It is clear that social interactions can influence perception in both social prey and their predators, yet we still know little of how these opposing processes can interact to influence predator-prey encounters.
Collective vigilance is an important anti-predatory benefit of social life that depends upon how group members pay attention to one another. Pulliam’s original mathematical argument (1973) assumed that once one member of the group is aware of a threat that the information is immediately public. In reality information needs to propagate across members and transmission rate is limited by individual perception, predominantly through auditory and/or visual monitoring [4–6]. For animal groups on the move, the quality of their social communications is generally measured by how well they can coordinate their actions. The costs or benefits of such collective coordination are likely to be particularly acute during transition phases, as when groups initially coalesce and respond to a perturbation. Unfortunately, the mechanisms for how animals integrate social cues and coordinate their activities remain elusive, but data indicate that individuals often rely on vision for rapid responses to changes in their neighbors’ movements (bees, [7]; fish, [8, 9]; humans, [10]). Recent theory suggests that selective attention to motion cues can expedite the sensory integration process in animal groups, substantially enhancing the speed and accuracy with which social animals coordinate their actions [11].
Motion is a critical factor in predation, although a prey’s speed and turning behavior can have varying effects on a predator’s capture ability. An animal’s speed can initially attract a predator’s attention, thereby increasing the risk of detection and attack [12, 13], yet once driven to action animals will display impressive escape speeds to avoid capture [14, 15]. Increased turning rates are believed to reduce predation risk by disrupting a predator’s ability to visually track their quarry and predict an intercept course [16], although evidence for this is mixed [17–19]. Prey turning behavior also appears to be less influential in predator attacks than prey speed and the effect of either of these velocity components can be reversed when a predator must consider multiple targets at once [20, 21].
Prey found in groups can further reduce predator capture success through the confusion effect, although when and how group-level movements contribute to this process remains unclear. Multiple competing stimuli divide a predator’s attention, overwhelming its ability to select any one prey, thereby reducing its capture ability [2, 4, 22]. The confusion effect is generally associated with changes in either the number or density of prey [23–26], yet confusion can begin with a pair of prey [20] and density effects are context dependent [27–29]. Jones and her colleagues [17] demonstrated that randomly moving particles were harder for human subjects to capture when these asocial particles moved more erratically and increased in number. What remains to be tested is how even simple social feedbacks will impact a prey’s risk during an attack and, more importantly, whether any predator confusion derives from the targeted individual’s socially driven movements, visual distractions caused by neighbors, or both.
How coordinated social movements influence predation risk remain largely unexplored. Social coordination statistically reduces movement variability, which should be detrimental for moving prey since predators can be very efficient at visually tracking targets [18, 19]. Additionally, random-dot assays demonstrate that visual perception of motion improves with the coherency, or coordination, of collective stimuli (fish, [30]; birds, [31]; primates, [32]; humans, [31]). If coordinated motion reduces movement variability and improves capture ability in visual species, then we should expect a trade-off between enhancing information transmission among group members and reducing an individual’s risk of being captured during an attack.
In this study we looked for evidence of a trade-off between social coordination and predation risk by testing the hypothesis that increases in prey social coordination can reduce visual confusion in a predator. Specifically, we explored how socially influenced directional feedbacks in social prey impact predator capture ability. To test our hypothesis we used a visual-based social model to generate virtual prey in a targeting game in which human participants acted as surrogate predators. The prey model incorporates motion-guided attention into a self-organizing process, whereby individuals react to the movements of their neighbors based on those visual cues that exceed a sensory threshold [11]. We then projected our virtual prey onto a computer screen and participants were tasked with capturing a targeted prey within a group using a mouse. Our virtual prey could only perceive one another and could not respond directly to the attack, which avoids confounding responses based on personal vs. social information in the prey and reflects scenarios in which most of the group depends on social information when responding to disturbances [33, 34]. Capture ability was measured by recording player capture latency (time until clicking on a prey item) and accuracy (distance from the prey item when clicked).
We first determined how the prey’s visual sensory thresholds, their speed, and initial state influenced capture ability. Tuning the preys’ sensory thresholds allows us to explore links between prey visual perception and its impacts on a predator’s visual performance. To address the potential for any ‘behavioral oddity’ that may impact risk profiles we then independently varied the target’s sensory thresholds from those of the remaining group members. In nature risk profiles may also vary based on the initial state of the prey at the start of an attack. To address this we controlled the initial speed of the targeted prey and the overall state of the group (disordered, ordered), which enabled us to explore very different ecological conditions. Lastly, we used a novel means of testing for visual confusion by repeating the above scenarios, but veiling the target’s neighbors. By adopting this approach we controlled for the presence of visual distractions created by neighboring prey, while retaining their collective influence on the target’s movements.
Pilot trials were conducted at various locations in the United States, including government buildings and scientific society meetings, while the experimental data presented here were collected from 30 consenting adult subjects from l’Université de Lausanne, Switzerland. U.S. trials were approved by Portland State’s Institutional Review Board (#122144) and ethical approval at l’Université de Lausanne was provided by the Secrétariat de la Commission cantonale (VD) d’éthique de la recherche sur l’être humain. Experimental protocols were identical across locations.
We simulated groups of prey by modifying a particle model in which social coordination stems from how individuals visually perceive and react to the movement decisions of their neighbors [11]. The model’s basic structure builds upon earlier works to study how basic cognitive mechanisms can influence animal movement mechanics in dynamic environments [35]. Here, the value of each neighbor’s movements is weighted according to an observer’s visual perspective:
v j , i = ω j , i · v ^ j , i , (1)
where v ^ j , i is neighbor j’s current direction relative to subject i and ωj,i scales the influence of that vector based on the observer’s perception (see section S1.1 in the S1 Text for further details). Based on what individuals can see we then assume that all of this information is filtered, so that each group member selectively responds to a set of Ni neighbors whose motion cues are strong enough to stand out in their field of view:
j ∈ N i if ω j , i ω ¯ ≥ m (2)
Here ω ¯ is the mean speed of the individual’s optic flow and parameter m is a motion threshold that tunes how sensitive individuals are to the changes in the relative speeds of their neighbors. Eq 2 represents a generalization of the signal-to-noise property of Weber’s Law, in which the perception of a stimulus is some constant proportion of the background. Individuals then attempt to adjust their velocity according to this social information, v i s, which is represented by the average degree of relative motion that has grabbed their attention:
v i s = v i + ⟨ ω j , i · v ^ j , i ⟩ N i (3)
The individual’s motion threshold therefore acts as a sensory mechanism that dictates its level of social interaction, and therefore, the overall degree of coordination that emerges. For brevity we will simply refer to m as the prey’s social threshold from here on. Groups composed of individuals with low social thresholds display highly coordinated motion (Fig 1a), which decays rapidly as individual thresholds increase because members pay less attention to one another’s actions (Fig 1a inset). Note that when individual thresholds are essentially unconstrained (m = 0) the resulting degree of social coordination initially improves before decaying as m continues to increase. This pattern aligns with the hypothesis that an unconstrained sensory range within a group can actually reduce directional coordination, since distant neighbors are less likely to behave in a similar fashion [2].
Each individual’s future step is then an attempt to adapt to its current social settings, whereby individuals effectively scale their reaction to any social directional cues according to their own internal state:
v i ( t + Δ t ) = v i s1 - γ ( v i s )v ^ i s( t ) (4)
Here v i s and v ^ i s represent the speed and directional components of an individual’s socially adjusted velocity, v i s. Function γ() is a biological extension of the dampening term used in the application of Langevin dynamics in social force models [36]. For our purposes we’ve generalized this physical term to a cost function to reflect the tendency for animals to modify their speeds in response to both energetic and ecological stressors (see S1 Fig and section S1.2 in the S1 Text). Individuals track their current speed relative to an expected optimum (v*), which reflects a population-level behavior (average travel speed) that stems from individual-level capabilities. When traveling as a group such speed control not only ensures that all members travel at the same speed, but effectively causes group members to accelerate when they fall behind and decelerate when they pull away from their neighbors (vi < v* vs. vi > v*, respectively). Recent empirical work has demonstrated that such speed dynamics are an important component in collective motion and information transfer [37–39].
Individuals then update their positions discretely based on their desired velocity:
x i ( t + Δ t ) = x i ( t ) + v i ( t + Δ t ) + U v i , η Δ t (5)
Function U adds stochasticity to the process by independently adjusting speed and direction by a magnitude of η, where η ∈ [0, 1]. Speed varies by ± η ⋅ v(t), while the heading v ^ ( t + Δ t ) is rotated by an angle between [−ηπ/2, ηπ/2]. Prey turning rate was limited to (π/2)/Δt. Time is given by each simulation step t and distance units are generic, being normalized to particle diameter (D = 2r). See S1 Table for additional details).
Directional preferences in the model are predominantly driven by the degree to which individuals respond to changes in the relative velocities of their neighbors (Eq 3). By including an ecological feedback to control individual velocity (Eq 4) we can generate new dynamics to explore an important consequence of social behavior, namely, the interaction between the individual’s social threshold (here, sensitivity to neighbor motion), their resulting velocity, and, finally, the emergent degree of social coordination. Consider a group of foragers departing an area by moving off at random. With little social feedbacks in their directional choices, each individual may show only minor deviations in direction (given by U in Eq 5) as they accelerate to their expected travel speed (Fig 1b, open circles). Such asocial behavior can also occur during predation events and is typified by the so-called flash expansion effect observed in fish schools under attack [14]. However, when individuals mimic the decisions of their neighbors this behavior causes a directional feedback that propagates across the group and delays the average individual’s departure speed (Fig 1b). Here the extent of this departure lag is driven by each individual’s social threshold, m in Eq 2. As m drops, individuals take longer to achieve a directional consensus as they pool information across an increasing number of neighbors and forward momentum is delayed. While the final speed achieved is unaffected by the underlying social interactions, the initial lag in departure speed during self-organization may increase a targeted individual’s risk of being captured by a predator. Given the feedback between social coordination and group speed, the initial state of the group should also affect predation risk. If prey are already aligned at the start of a predator’s attack, such as a group of fish aligned in a current, then any socially driven effects on the average individual’s speed due to turning behaviors should be dampened because directional differences are either ignored or marginalized (Fig 1c; Eq 5).
The use of interactive games that rely on virtual prey and or surrogate predators has steadily grown as a practical means of minimizing unnecessary harm, or controlling for confounding factors like the state of the individuals (fish, [21]; birds, [40]; humans, [17, 25, 28, 41]). Here we designed an interactive game to test how a targeted prey’s initial escape speed, social threshold, and the initial organization of its group members could affect a predator’s capture ability. The game began by displaying instructions outlining the goal and the rules. Participants were asked to sit behind a portable screen and wear earplugs to reduce both visual and auditory distractions. Prey appeared in the center of the screen either alone or in a group of 25 and a player’s goal was to click on the designated target before it escaped by moving off-screen. The target was highlighted in red for 250 ms before turning the same color as the remaining prey (for an example see S1 Video). Players had to pass a practice round to become familiar with the mouse and screen settings, and could not proceed onto the data collection phase unless they captured 4/5 practice targets. If a target moved off-screen before a player could capture it the session was counted as a miss and terminated. We recorded the movements of the mouse and all virtual prey every 20 ms, and these data were used to calculate player capture latency and accuracy. Failure to record a mouse click during a trial was flagged and the activity of the virtual prey and player’s mouse movements were reviewed. There were 13 such trials, which together constituted less than 1% of the data collected and we corrected for any spurious effects prior to our analyses (See section S2.1 in the S1 Text for details).
Each player completed 144 trials of different parameter combinations and on average the game took 10 minutes to complete (including training), with each trial lasting between 240–5,767 ms (median time = 1,141 ms, or 1.14 s). Games were run on a 15.4” MacBook Pro under 1440 x 900 resolution (110 pixels per inch). At the start of each trial the average prey member accelerated up to a mean speed of 0.400 D/t, where D was equivalent to 1 pixel and t is a simulation step. Prey speed was limited to a minimum of 0.1 D/t because pilot trials indicated that slower speeds were too easy for participants. Animations were rendered at 50 frames per second, yielding particle speeds ranging from 0.12 (min.), 0.46 (mean) and 1.15 (max.) cm/s. All prey particles were rendered to appear approximately 0.95 cm long.
We organized our procedures into three distinct sessions (I–III). The conditions and parameter settings for each of these sessions were determined a priori based on earlier pilot trials conducted during the summer of 2013. As such, all parameter combinations were presented to players in a randomized, full factorial design rather than incrementally in distinct experiments (Table 1). The graphical interface of the game was coded in the Java-based Processing platform (version 1.5.1) and all data processing and statistical analyses were done in R (version 3.0.2) [42].
Players were presented with two replicates of each parameter combination, and we used linear mixed effects models (LMM) to determine if our primary factors (ve, m{T,G}, ρ0, and veiled) significantly affected the latency (PL) and accuracy (PA) with which players captured their targets. Latency represents the time (ms) taken in a trial to click on the target. Accuracy represents the normalized distance between the mouse cursor and the target’s center of mass when the player attempted to capture the target by clicking on or near it. For analyses, latency was log transformed, and accuracy was transformed so that PA = 1 − (log(dT + 1)/max(log(dT + 1))), where dT was the distance from the mouse to the target in pixels [29]. All dT values were reviewed for edge effects and adjusted if necessary to avoid spurious PL or PA values (See S2 Fig and section S2.1 in the S1 Text). PA values range from 0–1 (furthest to closest recorded values). The social threshold values, m, were also transformed for the LMM analyses as L ( m ) = log ( m + 1 ). Variance within player was included as a random effect, and all initial models included the predictor variables relevant to each of the above experimental conditions and all possible interactions. We generated final models by removing insignificant (P > 0.05) terms, beginning with highest-order interactions and working to main effects, using analysis of deviance tests on nested models when needed [43].
To better understand why our factors influenced player capture ability we conducted a secondary set of analyses to explore how the underlying kinetics of prey motion affected predator latency and accuracy. Kinetically derived properties (i.e., those derived from the prey’s movements) were partitioned into local and global functional groups to explore the potential for differences between these scales. Local metrics included spacing around the target, along with its speed and turning behavior. Correspondingly, global metrics included average nearest neighbor distance, group speed and collective order (Table 1). Local spacing around each target was measured using its Voronoi polygon area, vpa, which was calculated using the sp and deldir libraries in R [44, 45]. A target’s vpa represents its personal space, whereby every point within the polygon is closer to the target than to any of its nearby neighbors. This metric effectively measures a prey’s risk of being captured by a predator [46]. Global spacing was measured using the average nearest neighbor distance across all particles [29] and group speed was simply the mean speed of all prey. A target’s turning behavior was measured by the tortuosity of its movements, tor. Tortuosity is a dimensionless metric that is inversely related to how much an object’s movements deviate from a straight line [47]:
t o r = 1 - | x final - x 0 | ∑ l t , (6)
where the numerator is the net displacement between positions from start to finish and the denominator is the length of the path traveled in that time period. The more often the target turns, the closer tor approaches 1. Group speed was calculated as the mean speed across all group members. Collective order, ρ, measures directional organization as the degree of alignment across all headings within the group:
ρ ( t ) = 1 25 ∑ i v ^ i ( t ) , (7)
where ρ at time t is given by the average magnitude of the orientation vectors for each prey in the group, v ^ i. In the game the initial degree of directional organization is used as a factor, denoted by ρ0, whereas organization itself is denoted merely as ρ.
We again used linear mixed effects models in which variance within player was modeled as a random effect to account for our repeated measures design. When analyzing the relationship between our response variables (PL, PA) and the preys’ kinetic properties we were more interested in the relative contribution of these metrics to explaining any observed patterns than in their predictive abilities. As such, we standardized all kinetic metrics as z - ( z ¯ / z) and scaled them to unit variance to compare their relative effects on player capture ability [48]. We checked for collinearity issues using both pairwise correlations and variable inflation factors (VIF)[49] and the only notable concern was a high correlation (Spearman r = 0.673) between target speed and group speed (See S3 Fig). To correct this we compared the fit of each of these metrics to PL and PA for each experimental session. We then used sequential regression to replace the values of the worse fitting predictor with the residuals obtained by regressing it onto the better predictor (determined using AIC)[50]. VIF values were then used to confirm the absence of any remaining collinearity in each session prior to model fitting.
Players were slower to click on those targets adopting the faster of the two escape speeds (P = 0.003; Fig 2a; S2 Table). Player latency, PL, also increased with increases in the prey’s social threshold (P < 0.001; Fig 2b), yet the initial degree of organization in the prey had no significant effect on latency. The kinetic mechanisms driving the observed patterns in latency were primarily the overall speed of all prey (global effect), along with the spacing around the target and its turning behavior (local effect; Fig 2c). Taken together, these three kinetic properties accounted for over 73% of the overall effect strength in the LMM. Interestingly, while players took longer to capture targets in faster moving groups (vG, P < 0.001), the speed of the target itself had no effect. Players also took longer to click on targets that either turned more frequently (tor, P < 0.001) or were more spatially isolated (vpa, P < 0.001; d ¯ 1, P < 0.001). Although local spacing significantly interacted with the targets’ speed and turning behavior, these interactions themselves had only a marginal impact on player capture latency (Fig 2c). Additionally, while increases in both local and global spacing had similar effects on player latency, changes in local spacing had a much stronger effect than did changes in global spacing (23% vs. 9% for vpa and d ¯ 1, respectively; see S2 Table).
Player accuracy, PA, improved when targets adopted slower escape speeds (ve, P < 0.001; Fig 3a, 3b; S3 Table), regardless of prey social threshold or the initial organizational state of the group. Accuracy decreased significantly as the prey’s social threshold increased and prey movements became less coordinated (P < 0.001). There was also a significant interaction between prey social thresholds and their initial overall degree of organization (m x ρ0, P = 0.002). Those virtual prey that were already organized (i.e., aligned) at the start of the attack were more difficult to catch than those that were disorganized at the start of the attack (ρ0, P < 0.01), and any socially derived effects on PA dampened out relative to those observed when groups were initially disorganized (e.g., random initial orientations). In other words, while prey with higher social thresholds (less attention to group members) were always harder to capture than those with lower social thresholds (more attention to group members), the benefits of reduced social interactions were largely context dependent and were significantly reduced when groups were already organized (S4a Fig). In terms of the kinetic mechanisms driving these effects, target movements primarily drove player capture accuracy; the speed and turning behavior of the target accounted for 65% of the overall effect strength in the final LMM (Fig 3c; S3 Table).
Here we explored what happened when the target’s social threshold differed from that of the remaining group members, mT ≠ mG, and we found that such sensory heterogeneity among the prey had no significant effect on player latency or accuracy. As in the trials of session I, players took longer to click on those targets that had higher social thresholds (P < 0.001, S4 Table). These prey paid less attention to their fellow group members, which resulted in less social coordination. Also, player accuracy was again affected by the interaction between the prey’s social threshold and the initial global organization of the group. While accuracy was lower for prey within groups that were initially organized (P < 0.001), and declined as the targets paid less attention to their group members (P < 0.001), these effects were much stronger for targets found in groups that were initially disorganized (P = 0.029; See S4b Fig and S5 Table). We found no substantive differences in the underlying kinetic metrics for either capture latency or accuracy, suggesting that the same movement behaviors driving the response variables in session I were doing the same here (S4 and S5 Tables).
Whether the targets appeared to be moving alone or within a group did not, by itself, have an overall effect on player latency. There was however a significant interaction between the targets’ social threshold and the veiled condition on player latency (P = 0.002). Although players were still slower to click on their targets as the prey’s social threshold increased (P < 0.001), this effect relied upon the visual presence of their neighbors and disappeared when the prey appeared to be traveling alone (Fig 4a; S6 Table). The behavioral kinetics driving player latency under these conditions were again mean group speed and target turning behavior (together making up 51% of the effect strength in the final LMM), and to a lesser degree, local spacing and individual speed (comprising another 32% of the effect strength).
In contrast to player latency, accuracy significantly increased simply because targets appeared to be moving alone (P < 0.001). While player accuracy retained its negative relationship with the prey’s social threshold, decreasing as m increased and prey showed less collective coordination, this effect was stronger for targets that appeared to be alone compared to those appearing in groups (P < 0.001; Fig 4b, S7 Table in the S1 Text). Player accuracy was again context dependent, improving when targets were found in groups that were initially disorganized rather than organized at the start of the attack (P < 0.001). The two most influential behavioral kinetics were target velocity (P < 0.001) and tortuosity (P < 0.001), which together accounted for 73% of the effect strength in the final LMM. The remaining influential kinetic factors included local spacing around the target (vpa, P < 0.001) and mean group velocity (v G ¯, P < 0.001).
In general, the results of this last experiment indicate that an attacker’s visual confusion manifests at different stages of the attack sequence based on the degree of social coordination displayed by the prey—a finding that parallels previous scale-dependent effects of density on predator attacks [29]. Asocial motion, reminiscent of a flash expansion in fish, causes a visually driven delay in capture, but has little effect on the accuracy of the attack (asocial condition, m = 10, in Fig 4a vs. Fig 4b). In contrast, while any visible, coordinated motion around a targeted prey does little to delay an attack, it invariably leads to a visually driven reduction in capture accuracy, relative to when the targets were visually isolated (Fig 4, social conditions, m ≠ 10).
The results from this experiment also highlighted the visual component beneath a weak, but persistent, association between player latency and accuracy. In general, player accuracy got worse with time, as it was negatively correlated with capture latency (sessions I-III; Spearman r = −0.23, P < 0.001). The strength of this association varied across experimental sessions, but the trend remained consistent and weak. Here we found that any association between latency and accuracy disappeared when players were only able to see the targeted prey and not their surround neighbors (See S5 Fig).
The coordinated movements of social animals have long been seen as providing several adaptive advantages to group members, including enhanced prey awareness and increased predator response times [2–4]. Our results demonstrate for the first time that a visual sensory mechanism likely to enhance reflexive coordination in social animals can actually increase individual predation risk during a targeted attack. We also showed that changes in prey social coordination can visually confuse an attacker without requiring explicit changes in group size or density.
We controlled the social interactions in groups of virtual prey by tuning their visual sensory thresholds, which modulated their attention to neighbor activity and thereby influenced the degree of coordinated motion that emerged (Fig 1). We subsequently found a positive relationship between increases in coordinated, or polarized, motion and the individual’s risk of being captured (Figs 2b and 3a). A reduction in capture ability when grouped prey suddenly move with little to no regard to their neighbor’s decisions aligns well with expectations, given that such ‘flash expansions’ in fish schools are often associated with aiding in the confusion of a predator [14]. Yet the positive relationship between coordinated motion and an attacker’s capture ability deviates from general expectations. Coordinated collective motion is thought to benefit group members by enhancing the speed at which passive information is transmitted across individuals and has therefore been presumed to reduce predation risk [14, 51, 52]. Recent empirical evidence supports this to some degree. Ioannou and his colleagues showed that coordinated motion in virtual prey, meant to mimic water fleas, Daphnia spp., reduced the likelihood of being attacked by Bluegill sunfish, Lepomis macrochirus[21]. However, this effect was primarily attributed to reducing the virtual preys’ exposure time to a hover-predator–a finding that is likely paralleled in our study by the interaction between capture ability and the initial degree of organization in the prey at the start of the attack (discussed further below). The findings of [21] also highlight a conflict across scales of organization, as those virtual prey in their study that showed more erratic turning behavior within a group were less prone to attack, yet at the group level the prevalence of such behaviors resulted in less mobile swarms that were preferentially attacked. In their model, social interaction influenced the virtual prey’s turning behavior, while speed was held constant.
In our study reductions in player capture ability were generally attributed to increases in prey speed, while changes in prey turning behavior did not show a consistent relationship with capture ability. Less social prey initially moved more quickly than their more social counterparts, yet these same phenotypes also displayed very different turning behaviors from one another (see Fig 5a and 5b). Those prey with a modicum of social interaction displayed the most tortuous movements, while the asocial prey displayed the least (m = 2.5 vs. 10, respectively). In contrast, both of these sensory phenotypes resulted in consistently faster departure speeds in the window leading up to capture events, with asocial members moving the fastest. Speed also impacted player capture ability differently at different levels of organization. Capture latency was driven by the average speed of all group members, while accuracy was primarily affected by the target’s speed.
Players displayed behaviors that are typically consistent with confusion. They would occasionally track the wrong prey, or showed signs equivalent to the pass-along effect [5] in which the targeted prey’s movements lead an attacker to miss and accidentally hit one of the target’s neighbors (See S2d Fig). When our virtual prey appeared to be alone, socially mediated effects on capture latency were eliminated, despite the fact that the targets displayed identical movement patterns in the corresponding trials when their group members were visible to the player (Fig 4a). These results provide clear evidence that the loss of coordinated motion within a group of fixed size creates a visual confusion effect that impacts capture latency. More importantly, the data demonstrate that the degree of erratic motion in a targeted prey’s trajectory alone had no effect on how long it took players to capture their target. In contrast, changes in social movement patterns were sufficient to impact capture accuracy, although this effect was dampened by the presence of visual distractions (Fig 4b). So, as discussed briefly in session III of the results, we see that even though increases in social coordination enhance individual risk within groups, such behavior is none-the-less better than traveling alone. Additionally, in a drawn out engagement with a predator, maintaining some degree of social coordination may outweigh the benefits of the asocial condition because individuals would be less likely to become isolated from the group and preferentially targeted by either another predator, or a second attack.
In effect there are likely to be two opposing factors at play in predator confusion once a target has been selected from the group. The first is the loss of the target through misidentification (e.g., the pass along effect, S2d Fig). For example, [20] presented evidence that three-spined sticklebacks, Gasterosteus aculeatus L., were more prone to attack pairs of water fleas, Daphnia magna, when the prey moved in parallel compared to when their paths crossed. Another potential factor is that coordinated motion in a group of objects can enhance visual tracking. Consider that the simplest mechanism for visual tracking is to keep the target’s image steady on the retina [53], and this process is either reinforced by, or relies upon, visual saccades [54]. Visual saccades are rapid eye or head movements used to correct gaze errors, maintain fixation, or jump to predicted positions (insects, [18, 54]; crustaceans, [55]; fish, [56]; humans, [57]). Since target acquisition is temporarily lost during saccades, it is likely that target trajectories are more easily reacquired when any background stimuli move in a similar pattern [58]. Given our findings on what factors could lead to visual confusion in a predator, it would be informative for future efforts to adapt the current framework and explore how predators can compensate for these effects by switching targets and, if so, how this impacts prey behaviors or phenotypes.
As mentioned earlier, coordinated motion in our prey could reduce an attacker’s success since we found context dependent differences in the speed and accuracy of player capture ability. While targeting latency was robust to large changes in the initial state of the group (ρ0), capture accuracy was not and targets displaying escape speeds in polarized groups reduced player accuracy and likely drove the overall significant reduction in performance (Fig 3, S3 Table). Differences in targeting latency and accuracy patterns may be linked to the timing of events and the added importance that local spacing played in capture latency. Prey in our simulations self-organized quickly from an initially disorganized state (≈1 s; S6 Fig), and so these kinetic patterns occurred well within the time frame required by humans to both visually respond to stimuli (100–150 ms) and physically react through hand movements (≈300 ms; [59, 60]). Along with prey speed and tortuosity, local spacing was also an important factor driving capture latencies, and each of these three kinetic properties evolved differently over time (Fig 5). Differences in the speed and turning behavior of the prey emerged early during the attack sequence and diverged along the preys’ social thresholds, while the resulting impacts of these movements on local spacing (vpa) generally did not become apparent until after a time by which nearly half of the attacks were already over. Patterns in local spacing leading up to the capture window were also robust to differences in the initial conditions, while differences in speed and turning behavior were not (Fig 5, S6 Fig). In contrast, the resulting pattern in capture accuracy aligns well with our expectations, given the importance of collective speed and the relative reduction in capture accuracy across prey social thresholds (Figs 1b, 1c; 3a and 3b). When prey were already aligned at the start of an attack any socially mediated differences in target speeds were less apparent and differences in their respective tortuosities took longer to emerge (S6 Fig). Local spacing, it seems, plays a much more important role in the timing of an attack than in capture or handling.
Spacing has long been thought to play an important part in predation risk within social groups, particularly with regards to being more exposed than one’s neighbors and being singled out for attack [46]. The greatest potential risk in adopting a different movement strategy during an attack is becoming separated from the group, although prey on the outskirts can still benefit from the behavior of the group [41]. However, our results support the findings of [28], where increases in local prey density around a target served to increase the speed at which it was captured, rather than diminish it. In addition we found no evidence of a behavioral oddity effect, since there was no interaction between prey social thresholds within the groups (mT x mG) and predator capture ability. Jones et al. [17] came to a similar conclusion in a study using human players to explore the effect of movement heterogeneity in asocial prey that were clumped together.
The approach presented here provides a means to control and test how sensory processes in both prey and predator can interact to influence their encounters. We have demonstrated a positive relationship between the degree of coordinated motion in social prey and the individual’s predation risk during a targeted attack. It is the degree of coordinated motion in a group of social prey that drives any visual delays in capture, while the accuracy of an attack relies primarily on the speed and turning actions of the targets themselves. Coordinated motion in animal groups can potentially reduce the time spent in dangerous areas and help individuals to avoid becoming isolated, yet such movement patterns can also alleviate predator confusion during a directed attack. The benefits of coordinated motion are therefore context dependent, which would help explain why social animals that move collectively display such a rich array of emergent behaviors during the course of an attack.
|
10.1371/journal.pcbi.1004478 | Evolution of Intra-specific Regulatory Networks in a Multipartite Bacterial Genome | Reconstruction of the regulatory network is an important step in understanding how organisms control the expression of gene products and therefore phenotypes. Recent studies have pointed out the importance of regulatory network plasticity in bacterial adaptation and evolution. The evolution of such networks within and outside the species boundary is however still obscure. Sinorhizobium meliloti is an ideal species for such study, having three large replicons, many genomes available and a significant knowledge of its transcription factors (TF). Each replicon has a specific functional and evolutionary mark; which might also emerge from the analysis of their regulatory signatures. Here we have studied the plasticity of the regulatory network within and outside the S. meliloti species, looking for the presence of 41 TFs binding motifs in 51 strains and 5 related rhizobial species. We have detected a preference of several TFs for one of the three replicons, and the function of regulated genes was found to be in accordance with the overall replicon functional signature: house-keeping functions for the chromosome, metabolism for the chromid, symbiosis for the megaplasmid. This therefore suggests a replicon-specific wiring of the regulatory network in the S. meliloti species. At the same time a significant part of the predicted regulatory network is shared between the chromosome and the chromid, thus adding an additional layer by which the chromid integrates itself in the core genome. Furthermore, the regulatory network distance was found to be correlated with both promoter regions and accessory genome evolution inside the species, indicating that both pangenome compartments are involved in the regulatory network evolution. We also observed that genes which are not included in the species regulatory network are more likely to belong to the accessory genome, indicating that regulatory interactions should also be considered to predict gene conservation in bacterial pangenomes.
| The influence of transcriptional regulatory networks on the evolution of bacterial pangenomes has not yet been elucidated, even though the role of transcriptional regulation is widely recognized. Using the model symbiont Sinorhizobium meliloti we have predicted the regulatory targets of 41 transcription factors in 51 strains and 5 other rhizobial species, showing a correlation between regulon diversity and pangenome evolution, through upstream sequence diversity and accessory genome composition. We have also shown that genes not wired to the regulatory network are more likely to belong to the accessory genome, thus suggesting that inclusion in the regulatory circuits may be an indicator of gene conservation. We have also highlighted a series of transcription factors that preferentially regulate genes belonging to one of the three replicons of this species, indicating the presence of replicon-specific regulatory modules, with peculiar functional signatures. At the same time the chromid shares a significant part of the regulatory network with the chromosome, indicating an additional way by which this replicon integrates itself in the pangenome.
| Regulation of gene expression is recognized as a key component in the cellular response to the environment. This is especially true in the microbial world, for two reasons: bacterial cells are often under severe energy constraints, the most important being protein translation [1] and they usually face a vast range of environmental and physiological conditions; being able to efficiently and readily react to ever changing conditions can most certainly give a selective advantage over competitors and give rise to specific regulatory networks.
Transcription is mainly regulated by proteins, called transcription factors (TF), which usually contain a protein domain capable of binding to specific DNA sequences, called TF binding sites (TFBS). Depending on the position of the TFBS with respect to the transcriptional start site of the regulated gene, the TF can act either as a transcriptional activator or a repressor, mostly because of its interaction with the RNA polymerase and sigma factors [2, 3]. The binding of the TF to its cognate TFBS is based on non-covalent interactions whose strength is indicated by the so-called affinity constant. Since TFBS can have variations around a preferred sequence, the affinity of a TF for its TFBSs covers a continuous range of values; however, since the TF binding strength appears to follow a sigmoid behaviour, it is possible to distinguish between ‘weak’ and ‘strong’ TFBSs [4].
As opposed to eukaryotic species, prokaryotic TFBSs are usually distinguishable from the ‘background DNA’, and they tend to have a simpler structure and a close proximity to the transcription start site [5]. The application of information theory concepts to TFBS identification and analysis, revealed that specificity of the TF for a certain TFBS depends on the length, variability and composition of the TFBS itself with respect to the overall genomic background (i.e. the sequence composition). Intuitively, the minimum information content able to provide specific recognition of the TFBS by the TF mostly depends on the genome size and its composition; increasing the size of the genome clearly increases the number of putatively non-functional TFBSs, and when the TFBS bases composition is close to the background DNA composition it may be impossible to discern a true functional TFBS from the surrounding DNA. Transcription factors recognizing TFBS characterized by low information content usually control the transcription of many genes across the genome; alternative sigma factors usually belong to this class, and their TFBSs also show larger variability between species [5]. Gene targets of these TFs are harder to reliably predict, for the presence of many non-functional sites along the genome. The high gene density of bacterial genomes and its organization in operons results in specific expression or repression of whole functional pathways in response to stimuli. Furthermore, the presence of several TFBSs in the upstream region of a gene can result in a complex transcriptional response that recall the behaviour of logic gates [6].
Prediction of TFBSs in a genome usually relies on the availability of a position specific scoring matrix (PSSM) storing the frequency of each nucleotide at each position of a TFBS. PSSM modelling the variability of a TFBS can be built by identifying enriched DNA patterns in promoter regions of genes that are known to be under the control of the TF under analysis, better if guided by other assays, like the binding of the TF to synthetic nucleotides. Several algorithms have been developed to use such PSSM to search for TFBSs in nucleotide sequences, such as the MEME suite [7], RSAT [8–10] and the Bio.motif package [11]. A recent alternative method relies on the construction of a hidden markov model (HMM) from an alignment of nucleotide sequences, which can then be used to scan a query nucleotide sequence [12–14]. Since all these methods and their implementations have different weaknesses, it has been advised to use their combination to run predictions [15].
Regulatory networks evolve rapidly, making the comparisons between distant organisms difficult [16–19]. At broad phylogenetic distances, it has been shown that the conservation of a TF is lower than its targets [16]. Additionally, species with similar lifestyles tend to show conservation of regulatory network motifs, despite significant variability in the gene composition of the network, suggesting an evolutionary pressure towards the emergence of certain regulatory logics [16].
The fluidity of most transcriptional regulatory connections is well known and documented, not only at large phylogenetic distances, but also at the level of intra-species comparisons too [20–23]. Experiments have shown that Bacteria have high tolerance towards changes in the regulatory circuitry, making them potentially able to exploit even radical changes to the regulatory network, without extensive changes in phenotypes [24]. However, this is strongly dependent on which regulatory interaction undergoes changes, since there are also examples where a single change determines an observable difference in phenotype [25, 26]. Bacteria have therefore a mixture of robust and fragile edges in their regulatory networks and evolution can play with them at different extent to explore: i) the function of new genes, by integrating them in the old gene regulatory network, and ii) if genes that are part of the gene regulatory network can be removed without harm to the physiology of the cell. The extent of variability and evolution of the regulatory network inside a species is, however, still poorly understood.
The aim of this study is a comparative genomics analysis of regulatory networks, to understand the impact of regulatory network variability on pangenome evolution. We decided to use the Sinorhizobium meliloti species, the nitrogen-fixing symbiont of plants from the genus Medicago. S. meliloti has been deeply investigated as a model for symbiotic interaction and an extensive knowledge on its TFs is present in the literature [27, 28]. This species presents a marked genomic difference with respect to other well-know bacterial model species, such as Escherichia coli, since S. meliloti genome comprises three replicons of comparable size: a chromosome, a chromid [29] and a megaplasmid, characterized by functionally and evolutionary distinct signatures [30, 31]. This arrangement raises the question of how TF targets are distributed over the replicons. Recent reports have shown that there are only two genes essential for growth in minimal media and soil encoded in the S. meliloti chromid [32], even though the chromid harbours many genes shared by all sequenced strains of S. meliloti species. Moreover, S. meliloti has several genomes sequenced to date [23, 30, 33–39] and the potential for biotechnological and agricultural applications, which could benefit from this analysis. At the comparative genomics level, different strains show quite a high level of variation. Indeed, the pangenome (the collection of all genes from different strains [40]) of this species has an abundant fraction of genes common to all members of the species (termed core genome, as opposed to the strain-exclusive and/or partially shared fraction, called accessory genome) of around 5000 gene families; approximately 40% of the genome belongs to the accessory fraction [31, 35]. A preliminary analysis revealed that some of the TFs of the core genome also control genes of the accessory genome [23]. This allowed to propose that, when comparing the same regulon in different strains, we can define a panregulon, including a set of core (shared) target genes and an accessory (variable) regulon fraction [23]. It should be noticed that while the core regulon is necessarily formed by genes belonging to the core genome, the opposite can also be true (i.e. that a gene belonging to the core genome belongs to the accessory regulon). However, the dynamics of the panregulon in relation to the evolutionary rules controlling the variability of the accessory regulon fraction are still not understood.
We have therefore constructed the regulatory network of the S. meliloti species, using the PSSMs of 41 TFs collected from the literature and public databases. We have applied a combination of TFBS prediction methods, combining their output with information about the core and accessory gene families. We have also predicted the presence of the same TFBSs in five other closely related rhizobial species (termed ‘outgroups’: Rhizobium leguminosarum bv. viciae, Rhizobium etli, Mesorhizobium loti, Sinorhizobium fredii and Sinorhizobium medicae). This regulatory network has been used to highlight the different behaviours that are present within and between species. Our predictions and other comparative genomics observations are publicly available (https://github.com/combogenomics/rhizoreg/).
Based on COG annotations, all the 51 S. meliloti strains analysed in this study, have been found to encode a similar number of predicted TFs (an average of 522); a similar number has been also found in the five outgroups (an average of 533). This is in accordance with previous reports correlating genome size with the number of TFs [41]. Rhizobia belonging to the Alphaproteobacteria class (alpha-rhizobia), which are known to have larger genomes compared to other bacteria from the same class [42], have then one of the largest collection of TFs in the known bacterial kingdom. As the accessory genome accounts for about 40% of the proteome size [31, 35], it is reasonable to expect that a similar proportion of TFs will belong to the accessory genome. Indeed, about 70% of the TFs encoded in the S. meliloti pangenome belong to the core genome, while the remaining TFs are present in 1–3 genomes only; this orthologous genes distribution is similar to the one observed for the whole pangenome [43] (S1 Fig). However, most of the 41 TFs analyzed in this study were found to belong to the core genome (37), with the only notable exception represented by RhrA, the activator of the rhizobactin regulon, which is absent in 35% of the strains under study, confirming previous analysis [23, 44, 45]. More interestingly, recent reports have demonstrated how the presence of the rhizobactin operon confers competitive advantage over other S. meliloti strains in iron limited environments [32]; we could therefore speculate that a significant fraction of the S. meliloti strains have a competitive disadvantage in environments with limitation in iron bioavailability. Surprisingly, an ortholog of FixJ (the component of the global two-component system FixJL, which turns on nitrogen-fixation genes in microaerobiosis during symbiosis) was not predicted in two S. meliloti strains (A0643DD and C0438LL); the absence of the gene was further confirmed by PCR. Even though such an important regulator has been found to be absent in these two strains, another gene with similar domains (orthologous group SinMel7252, containing gene SMa1686 from the reference strain Rm1021) was found to belong to the core genome. SMa1686 was shown to be regulated by RirA [46], but to the best of our knowledge no indications of its relationships with microaerophilic growth conditions and symbiosis are present. Consequently, we cannot a priori exclude that the regulatory functions of FixJ may be carried on by homologs (as for instance orthologs of SMa1686) in strains A0643DD and C0438LL. Indeed, previous works have indicated that several target genes of FixJ lack a direct symbiotic function, suggesting the presence of functional redundancy in the genome [47].
Sixteen TFs were absent in at least one of the outgroups. Of these, 6 are encoded by pSymA, the symbiotic megaplasmid, including two copies of NodD, FixJ, RctR, SyrM and RhrA (S1 Fig). Such difference between intraspecific and interspecific TF gene content may anticipate a similar difference at the downstream regulatory network, for the absence of cross-regulatory links.
To minimize the number of false positives in our predictions, we selected PSSMs with relatively high information content (over the reference strain minimum information content, see Materials and Methods) A wide range of information gain for PSSMs was observed; of the starting 83 TFBSs retrieved from literature and databases, 41 have been found to have enough information content to reliably predict their TFBSs (Fig 1a, S1 Table). For FixJ, two separate motifs acting together have been described [48], one above and one slightly below the threshold: both motifs have been used.
We have applied a novel TFBS prediction approach to overcome common problems associated with the prediction algorithms and to maximize accuracy and sensitivity [3], including operon predictions to recover most of the downstream regulated genes (see Materials and methods). The predictions accuracy was determined with a comparison with the downstream regulons reported in the literature, when available (Fig 1b and 1c); the average accuracy of the predictions was found to be around 55%, with a tendency to positively correlate with the motif information gain (S2 Fig). This behaviour may be explained by the fact that most regulons have been defined on the basis of gene expression data and therefore contain both direct and indirect targets of the TF; our strategy is then not able to recover the indirect targets which might explain the relatively low accuracy. An example of a known regulatory interaction predicted by our approach is rem (SMc03046), a putative transcriptional regulator involved in the control of motility in S. meliloti Rm1021 [49], which was predicted to be under the control of MucR in our analysis (S1 Material).
To provide additional validation to our predictions, we used a compendium of S. meliloti gene expression data from the Colombos database [50] (see Materials and Methods). The full compendium contained 424 conditions and was used to calculate average correlation coefficients among the genes of i) the same predicted regulons, ii) the regulons reported in the literature and iii) random groups of genes sampled from the genome (Fig 1d and S2 Material). We have selected the conditions maximising the average correlation for a group of genes using a genetic algorithm (see Materials and Methods). Correlations for our predictions were not significantly different from the experimentally defined regulons; genes belonging to predicted regulons had a slight tendency to be higher than the random regulons, but if this difference was not significant (p = 0.09). We further experimentally confirmed some of the predictions on a subset of predicted promoters of the NodD regulon (S2 Table).
Predicted TFBSs in upstream regions against TFBSs predicted in coding regions were considered as signal to noise ratio (upstream hits on total hits) to measure the predictions quality (Fig 1e); for more than 70% of the analysed TF the observed ratio was above 50%, with a very poor correlation with the motif information content.
Taken together these results show that our predictions are of fairly good quality.
Little variability in the number of genes under the control of each TF was observed among different strains (Fig 2 and Table 1). Each TF was predicted to control the transcription of 12 genes on average, with RirA showing the largest regulon (with an average of 71.6 genes) and SyrM the smallest one (with an average of 1.1 genes). TFs with lower information content TFBSs showed a tendency to control a larger number of genes (S2 Fig), which confirms the influence of the information content on motif recognition. The predicted regulons were found to have comparable sizes in the outgroups; therefore the regulon is conserved in size between different species; this might be the result of the conservation across the species of the TFBS or of more general energy constraints on transcription/translation.
Besides similar regulon sizes, we found that an average 40% of genes belonging to a regulon belong to the accessory genome (Table 2); this implies that although variable, each TF recruits a similar number of genes under its control, at least in the species analysed here. Obviously, the variability of the regulons is related with both the variability in upstream regions of core genes and the presence of genes from the accessory genome (whose presence varies across and between the species) in the regulons.
Predictions for TFs with low information content TFBSs showed a very poor accuracy and precision when compared to experimental data found in the literature; an efficient search strategy for such TFBSs using PSSM has still to be developed. However, from an evolutionary point of view, since those TFs are predicted to bind rather aspecifically to many sites along the genome, this would result in even a larger divergence of regulons between strains, as recently reported in comparison among species [51].
To clarify if the patterns of variability of the regulatory network are related to the phylogenetic distance among strains a comparison between divergence of panregulons and divergence of pangenomes was performed.
Following the pangenome analysis, we calculate three sets of distance matrices among the genomes under analysis (see Materials and Methods): the first was obtained from the alignment of core genes (hereinafter the core distance), the second from alignments of the upstream regions of the core genes (the upstream distance), and the third is instead based on the presence/absence profiles of accessory genes (gene content distance). The three distances were then compared with the regulatory network distance of the corresponding strains/species, which was calculated with the same metric defined by Babu and collaborators [16]. Intuitively, the divergence in upstream regions should be paralleled by divergence in the regulatory network, since the former will at some point determine a loss/gain of TFBSs affecting the structure of the regulatory network. Similarly, a larger difference in gene content should also be mirrored by a higher variability in the regulatory network, since new genes may be recruited in the regulatory network and/or TFs may be lost/gained. On the other hand, we don’t expect to observe a strong correlation between core and regulatory network distances; this is also due to the lower divergence at the coding level between strains, implying that regulon diversity inside a species could be driven by gene content variability and upstream sequences variability.
These hypotheses on patterns of correlations between pangenome differences and regulatory divergence were confirmed at the species level (Fig 3a and 3b). The comparison between S. meliloti strains showed that the regulatory network distance is correlated with both the upstream distance and with gene content distance. The core distance showed no significant correlation with the regulatory network distance (Fig 3b). When considering the outgroup species, all three distances were found to be similarly correlated with the regulatory network distance (Fig 3c). Since the divergence in coding sequences cannot directly influence transcriptional regulation (with the exception of non-synonymous mutations in the DNA binding domain of a TF), we propose that the most likely explanation of the observed correlations is the overall genome divergence between species, which is ultimately reflected by a higher divergence at the regulatory network level. This is also confirmed by the high correlation coefficients among the three distances. We then concluded that the patterns of regulatory network variation are paralleled, at the species level, by changes in promoter sequences and by the variation in the accessory genome composition, at least in S. meliloti. These two fractions of the pangenome could then be used as bona fide predictors of the extent of rewiring in regulatory networks. However, from these data we cannot confirm a direct causative explanation for the observed regulatory network variation, as this analysis has been focused on the whole pangenome. The striking difference between the slow rate of coding sequence evolution versus the much larger difference in the regulatory networks is however worth noting.
Regulatory network evolutionary dynamics showed interesting differences within and between species. Each observed regulatory interaction in the two datasets (S. meliloti and the outgroups) and its state across all strains was used to build a hidden markov model to infer the preferred state transitions in our predictions (see Materials and methods), that corresponds to the ways the gene regulatory network can grow and shrink. The possible states of a target gene depend on the presence of the TF, the target gene itself and the upstream TFBS. Therefore, each target gene can be found in one of six different states (Fig 4a). The “plugged” state being the only functional one, which corresponds to a target gene with a TFBS in its promoter region when the TF is present in the genome. The other five are non-functional states but may represent transitory states during the evolution of gene regulatory networks. Each of these states lack: i) the TFBS (“unplugged”), ii) the TF (“ready”), iii) both the TF and the TFBS (“not ready”), iv) the regulated gene (“absent”) or v) both the TF and the gene itself (“missing”). This HMM can be used to estimate the probability for state transitions, that is the probability of observing a change from one state to another between two strains. This results in a model that is able to provide a general description of the evolution of regulatory networks within and between bacterial species. Since the models is based on observed states in the available strains, we consider it as a “snapshot” of the regulatory network evolution, and not an equilibrium model.
According to the model, the most represented state in the S. meliloti regulatory network is the “plugged” one, indicating conservation of regulatory interactions at the species level (Fig 4b and S3 Table). More interestingly, the model predicts that the “unplugged” genes are mostly seen recruited by the regulatory network and that the regulatory link is then maintained with high probability. Very little probability was given to the “plugged” to “missing” and “plugged” to “absent” transitions, indicating that genes belonging to the gene regulatory network are rarely removed from the genome. On the other hand, genes with no TFBS and its cognate TF are more frequently found to undergo loss (“not ready” to “missing”), suggesting that regulatory interactions are important for gene conservation at the species level. When considering a wider phylogenetic level (the outgroups), the broader variability in TF gene targets resulted in the “plugged” and “missing” state as equally probable, indicating that regulons might evolve by adding and removing new elements to a conserved kernel of gene targets (Fig 4c and S3 Table). This is also reflected in a smaller probability that a target gene i) remains in the “plugged” state when compared to the S. meliloti species level, and ii) that it acquires a TFBS. On the other hand, the same probability as within the S. meliloti species was observed for the transition “not ready” to “missing”, which seems to confirm the importance of regulatory features in explaining the accessory genome fraction evolution. Consequently, a different evolutionary dynamics of regulatory circuitry changes seems to be present in relation to the taxonomic ranks; at the species level, robust networks are formed and they tend to include new genes from the species pangenome, which then may be conserved. On the contrary, when comparing wider taxonomic ranges, regulatory networks are less conserved and genes are apparently included in each species’ genome directly with their regulatory features (in a sort of plug-and-play model).
Transcription factors with replicon preference were found to have functional signatures in accordance with the functions encoded in the three main replicons of S. meliloti. This aspect has been evaluated by mapping each draft genome on the S. meliloti replicons (see Materials and methods) and considering the presence of each gene in the replicons for each of the 51 strains analysed here. Using a clustering approach on normalized gene hits on each replicon we have found that 19 TFs preferentially regulate genes belonging to one of the three replicons: five to the chromosome (NtrR, OxyR, NesR, ChvI and SMc03165), six to the pSymB chromid (SM-b21706, SM-b20667, ChpR, RbtR, SM-b21598 and SM-b21372) and eight to the symbiotic megaplasmid pSymA (SyrM, NodD3, RhrA, NodD1, NodD2, FixJ, FixK1 and NifA) (Fig 5a); these TFs are also encoded by the same replicon.
The six TFs encoded by the pSymB chromid (whose regulon is also preferentially located on pSymB) appear to mostly regulate the transport and metabolism of various carbon and nitrogen sources, including ribitol (RbtR), tagatose, sorbitol and mannitol (SM-b21372), ribose (SM-b21598), lactose (SM-b21706) and tartrate, succinate, butyrate and pyruvate (SM-b20667). The eight TFs present in the symbiotic megaplasmid pSymA (with regulons preferentially located on pSymA) were found to be involved in the regulation of key symbiotic processes, including nitrogenase synthesis and functioning through micro-aerophilia (FixJ, FixK1 and NifA), nod-factors biosynthesis (SyrM, NodD1, NodD2 and NodD3), and iron scavenging (RhrA).
A functional enrichment analysis using COG annotations (S3 Fig) on genes belonging to the regulons of the replicon-biased TFs confirmed this general observation: no functional category was enriched in the chromosome. The G category (carbohydrate metabolism and transport) was enriched in genes regulated by pSymB encoded TFs, in agreement with the role of chromid pSymB in providing metabolic versatility to S. meliloti. The C (energy production and conversion), U (intracellular trafficing and secretion) and T (Signal Transduction) categories were enriched in genes under the control of pSymA-harboured TFs, which show some relationship with the establishment on the plant symbiosis. This analysis allowed us to depict a scenario where a significant part of the regulatory network is replicon-specific, with a tendency to maintain the functional signature of the host replicon, thus confirming earlier reports on the evolutionary independence of chromids and megaplasmids in S. meliloti [29, 31, 32].
Interestingly, a fraction of TFs have target genes which span over different replicons, and show a preference for cross regulation between the chromosome and the chromid (Fig 5b). The presence of cross-replicon regulons, may indeed allow a stabilization of genomic structure, genetically and metabolically connecting chromosome encoded functions with those present in the other two S. meliloti replicons. In the evolutionary model of the chromid [29, 31, 32], its stabilization within the host genome is related to the acquisition of essential (core) genes in a previously introgressed megaplasmid which gained niche-specific genes. Here, we found that for TFs encoded on the chromosome (as AglR, GlnBK, IolR, BetI, LsrAB, MucR, PckR, RirA, NesR) a variable number of target genes are present on pSymB (S1 Material). The preference for cross-regulation between the chromosome and the chromid, as opposed to the megaplasmid uncovers an additional mechanism by which a chromid integrates itself in bacterial pangenomes.
Regulatory networks are key components of cell’s response to environmental and physiological changes. In the past years, several works have highlighted a high transcriptomic variability in strains or individuals from the same species [52, 53], in addition to genomic variation. Consequently, regulatory network variation might have profound impact on local adaptation and fitness of organisms. Recent studies have confirmed that bacterial regulatory networks are able to tolerate the addition of new genes [24], which in turn can serve as raw material for selection to operate. Using our original combined search strategy, we indeed found variability in regulon composition within the S. meliloti species, which in fact accounted on average on 40% of the regulon of each strain. On the other hand the regulon size was found to be conserved even outside the species boundary. This could suggest that even though the genes under the control of a TF vary between strains, there is a general constraint on the size of the transcriptional response. Whether this is due to energy constraints or being simply an effect due to the genome base composition is yet to be clarified.
We found that the regulatory network distance (as defined in [16]) correlates with the upstream distance and also with the gene content distance. This correlations may suggest that regulatory network composition is influenced by both promoter variability and accessory genome variability. Indeed, we may speculate that the sequence divergence in upstream regions can result in the appearance or disappearance of TFBSs, thus changing the regulatory network content. Moreover, gene content dynamics may also have a strong impact on the regulatory network, with the introduction of new gene cassettes containing TFBS recognized by resident TFs. We can consequently hypothesize that the evolution of bacterial regulatory networks, as that of the pangenome, may be influenced by mechanisms of gene acquisitions, such as lateral gene transfer, and it’s not only linked to mutations in upstream regions.
The observed changes in the regulatory network also show interesting features with respect to pangenome composition. Indeed, even if a significant difference in the state transitions of regulatory links inside and outside the species boundary has been shown, for genes that lack both a TFBS and their cognate TF, we have observed a similar tendency to disappear from the pangenome. This observation may suggest that the dynamics governing pangenome evolution within a species could depend in part on a ‘gene fitness’ related to being wired into the regulatory network. We can then propose that regulatory networks have an important role in shaping the bacterial gene content and can contribute to gene fitness, which in turn may be linked to environmental adaptation.
Moreover, the preference of nineteen TFs for target genes on one of the three replicons of S. meliloti indicates that in multipartite bacterial genomes, similarly to replicon-dependent patterns of evolution in gene and functions content [31], a replicon-specific transcriptional regulation is to be expected. At the same time, a significant number of cross-links between the chromosome and the chromid suggest for the first time an additional mechanism by which new replicons can be integrated into a bacterial pangenome.
The 51 genomic sequences belonging to Sinorhizobium meliloti and the five genomic sequences from closely related symbiotic species are listed in S4 Table.
The orthology relationships inside the 51 S. meliloti strains has been computed using the Blast-BBH algorithm implemented in the DuctApe suite (version 0.13.0) [54], using default parameters. The same analysis has been conducted on the five closely related species with the addition of the Rm1021 reference strain, using the BLOSUM62 scoring matrix to account for their greater sequence diversity.
The number of regulators present in each genome has been estimated using COG annotations. The similarity of each protein against the COG database has been measured with a rpsblast scan [55], using an E-value threshold of 1e-10. Each protein mapped to the COG category K (Transcription) has been considered as a putative regulator.
To confirm the absence of the fixJ gene in strains A0643DD and C0438LL, PCR primers amplifying a large portion (from nucleotide position 32 nt to 595 out of 615 nt total) of the coding sequence of fixJ gene have been designed on the basis based on the ortholog sequence in strain BL225C (SinmeB_6173) with Primer3Plus (fw: 5′-ACGAAGAGCCGGTCAGGAAGTCGCTGGCATTCATGCTG-3′; rv 5-CGGCGAGAGCCATGCGAACGAGATGGGGGAGGCTC-3) [56]. PCR has been performed with the Maxima Hot Start Green Master Mix (Thermo Fisher) in 20 microL total volume by using 10 ng of DNA, purified from liquid culture with FAST DNA Kit (QBiogene) and 10 pmols of each primer. Cycling conditions were as follows: 5′ 94°C, followed by 30” 94°C, 30” 55°C, 1′ 72°C repeated for 35 cycles. PCR products were resolved after agarose gel electrophoresis (1.5 w/v) in TAE buffer with ethidium bromide (10 microg/ml) as staining agent.
The 83 regulators whose PSSM has been extracted from the various sources are listed in S1 Table. For those PSSMs retrieved from the literature, we collected the upstream regions of the regulated genes and (when available), the consensus binding sites from bibliographical records; the upstream regions have then been analysed with the meme program [7](version 4.9.0), using the model that retrieved the PSMM with higher similarity to literature. Twenty-two motif files have been generated using the information retrieved from the RhizoRegNet database [27]. Fifteen motif files have been generated using the information retrieved from the RegTransBase database [57]. For the 5 regulators having more than one predicted motif, for instance those having a variable length (FixJ, RpoD, RpoE2, RpoH1 and RpoH2), one motif file for each motif length has been generated. All the retrieved PSSMs have been converted to HMM models using the hmmbuild program from the HMMer suite [12–14](version 3.1b1), using the alignments present in the MEME motif file. It has been previously shown that in bacterial genomes TFBS can be reliably distinguished from background DNA only if their information content is higher than the minimum information content for the target genome, which depends on the genome size and composition [5] (this simplification of course ignores other factors such accessibility or proximity of the RNA polymerase). The information gain of the TFBS with respect to the genome is calculated using the Kullback-Leibler divergence between the corresponding nucleotide frequencies [58], and it has been shown to correlate with the motif length and base composition of the motif with respect to the surrounding genome sequence. TF motifs with sufficient information content also tend to show less variability in their regulon composition between species [51]; by focusing our analysis on such TFs we ensured a more precise analysis. The information content of each motif has been calculated as suggested by Wunderlich et al [5], using the Rm1021 reference genome for the calculation of the minimum information content; given the dependence of this variable on genome size and the fact that all the S. meliloti strains have similar genome size, there has been no need to calculate a strain specific threshold. PSSMs whose information content was found to be lower the minimum information content have been discarded with exception of FixJ, which has two distinct PSSM, one of which is above the threshold. In the presence of more than one source for a regulator (literature, RhizoRegNet or RegTransBase), the PSSM having the highest information content has been considered in the final analysis.
For each genome, background k-mers frequencies have been calculated using the fasta-get-markov program from the MEME suite (version 4.9.0) [7], using 3 as the maximum value for k. Each regulatory motif has been searched inside each genomic sequence using four scanning algorithms. The mast program from the MEME suite (version 4.9.0) [7] has been used with an E-value threshold of 100 and the use of a genome-specific background file. The matrix-scan program from the RSAT suite [8–10] has been used with a P-value threshold of 0.001, the background file and a pseudocount of 0.01, as suggested by Nishida et al. [59]. The Bio.motifs package from the Biopython library (version 1.62b) [11] has been used with a false negative rate threshold of 0.05 and a pseudocount of 0.01, as suggested by Nishida et al. [59]. The nHMMer program from the HMMer suite (version 3.1b1) [12–14] has been used with an E-value threshold of 100 and with all the heuristic filters turned off. Each regulatory motif hit has been parsed, separating the hits being present in the upstream region of a gene from the others. The upstream region has been defined as the intergenic region (not overlapping any coding sequence) in front of the first codon with a maximum size of 600 bp. In the case of a palindrome motif, the motif orientation has been ignored.
The distributions of the raw scores has been tested using a normality test, as implemented in the SciPy library (version 0.13.3) [60][61]. The score threshold has been determined through the calculation of the raw scores quartiles (Q1 and Q3) and defining the score threshold (τS in Eq 1) in order to consider only the upper outliers [62].
τ S = Q 3 + ( 1 . 5 ( Q 3 - Q 1 ) ) . (1)
For the Biopython method the bit score has been used, while for the RSAT, HMMer and MEME methods the negative base 10 logarithm of the E-value has been considered. The regulatory motifs predicted by at least three methods have been considered for further analysis.
The compendium of gene expression data for S. meliloti str. Rm1021 from the Colombos database [50] was used to calculate correlation coefficients among genes in the regulons reported in the literature, our predictions and random sets of genes. Random regulons were produced by random sampling groups of genes of size 5, 10 and 15, for which 500 sets were produced. Correlation was quantified by the squared uncentered correlation coefficient, which was calculated using Matlab, as the square of 1 − cos distance. Values plotted in Fig 1d are averages over the entire set of genes under analysis. We have implemented a strategy allowing to select the conditions maximizing the average squared correlation within a group of genes, since many of the conditions of the compendium are likely not related to our predictions. Selection of the conditions was performed using the genetic algorithm implemented in the GA Matlab function, with default tolerances (TolCon = 10−6, TolFun = 10−6). We let the algorithm select the conditions minimizing 1 R 2 where R is the uncentered correlation averaged over all pairwise comparisons made within the group of genes under analysis. Since we noticed that correlations are strongly and inversely correlated with the number N of included conditions, especially when N ≤ 20, we discarded all cases where the number of conditions was less than 20 (final N = 950). All conditions containing missing data in at least one of the genes under analysis were discarded before starting the procedure. For some of the known and predicted regulons, correlations were not calculated as the available number of conditions after removing missing data was less than 30 before the optimization.
Upstream sequences from selected putative target genes of NodD regulon were analysed (see S2 Table). Sequences (approximately 400 nt upstream the translation start site of the gene) were amplified from crude lisates of S. meliloti strains with AccuPrime Pfx DNA Polymerase (Thermo Fisher) and cloned into pTO2 vector (which carries GFPuv as reporter gene [63]) by using SalI and KnpI restriction sites. Recombinant clones of E. coli S17-1 strain were selected by gentamycin resistance and verified by sequencing of inserted fragments. Positive clones were used for transferring recombinant pOT2 vectors to S. meliloti Rm1021 by bi-parental conjugation by using previously described protocols [64][65]. S. meliloti Rm1021 recombinant strains were then tested for GFP fluorescence after incubation of a 5 ml culture grown at the mid-exponential phase with 1 microM luteolin (Sigma-Aldrich) in liquid TY medium at 30°C for 3h. GFP fluorecence was measured on a Infine200 Pro plate reader (Tecan). Measures were taken in triplicate and normalized to cell growth estimates as absorbance to 600nm.
The operons belonging to the 56 genomes of this study have been predicted using the Operon Prediction Software (OFS, version 1.2) [66], using a beta threshold of 0.7 and a probability threshold of 0.5. The number and length of the predicted operons in each strain are listed in S5 Table.
Each contig of the 44 S. meliloti draft genomes has been mapped to the seven complete genomes using CONTIGuator (version 2.7.3) [67], using a 15% coverage threshold and considering blast hits over 1000 bp in length. A contig has been considered mapped to a replicon when it has been found mapped to the replicon in at least five complete genomes, or when it has been mapped to the replicon in at least one complete genome and to no replicon in the others. Knowing that very few portions of the S. meliloti genome are shuffled between replicons [31], we assessed the quality of this mapping procedure by checking whether the S. meliloti orthologs were found to be mapped to more than one replicon; for each orthologous group the genes not mapped to any replicon have been removed, and the relative abundance of the most representative mapped replicon has been computed. A relative abundance of 1 means that the orthologs have all been mapped to the same replicon in all the strains. The vast majority of the orthologous groups was found to map to a single replicon (S4 Fig).
The number of average gene hits has been divided for each replicon (either from a complete genome or a draft genome) and normalized by the number of genes belonging to each replicon in the Rm1021 reference strain. Regulators with preferential regulatory hits in a specific replicon have been highlighted performing a k-means clustering (k = 5, selected using an elbow test [68]) and plotted using the two principal components of the proportion of hits in each replicon, using the scikits-learn package (version 0.14.1) [69]. Only the three main replicons (chromosome, pSymB and pSymA) have been considered. COG categories enrichments have been tested using a Fisher’s exact test, as implemented in the DendroPy package [70].
Phylogenetic distance inside the S. meliloti pangenome and the pangenome of the five related species has been computed as described in a previous work [31]. The pangenome has been divided in three fractions, allowing the use of three distinct phylogenetic distances. The “core” distance has been calculated through the alignment of all the nucleotide sequences of each core gene, discarding those genes where at least one sequence was 60bp shorter or longer with respect to the other sequences. The “upstream” distance has been calculated through the alignment of the core genes upstream regions, discarding sequences below 5bp in length. The alignments have been calculated using MUSCLE (version 3.8.31) [71] and the bayesian tree has been inferred using MrBayes (version 3.2.0) [72]. The distance matrix for both distance categories has been computed from the phylogenetic tree using the textitBio.Phylo package inside the Biopython library (version 1.62b) [73]. The “accessory” distance has been calculated through the construction of a presence/absence binary matrix for all the accessory genome OGs; the distance between each strain has been then calculated using the Jaccard distance measure, as implemented in the SciPy library (version 0.13.3) [61].
The distance between each strain inside the S. meliloti and the other five related species regulatory network has been computed using the distance in the presence/absence of regulatory interactions as suggested in the work of Babu and collaborators [16]. The distance between strain A and B is computed using Eq 2.
D A B = 1 - c o r e A B t o t a l A B , (2)
where coreAB and totalAB represent the number of conserved and total regulatory interactions, respectively.
Pearson and Spearman correlation coefficients between the pangenome and the regulatory network distance have been calculated using the implementations of the SciPy library (version 0.13.3) [61], removing the outliers using a Z-score threshold of 3.5 on the mean absolute deviation of the distances.
The state transitions of the regulatory network has been inferred by encoding them in a hidden markov model. Each one of the regulatory links observed in at least one strain has been tested for their state in each organism, following the labelling of Fig 4a. Specifically, each regulatory link in the network of each organism could belong to one of the following categories:
Plugged: regulator, gene and TFBS present
Unplugged: regulator and gene present, TFBS absent
Ready: gene and TFBS present, regulator absent
Not ready: gene present, regulator and TFBS absent
Absent: regulator present, gene and TFBS absent
Missing: regulator, gene and TFBS absent
The hidden markov model has been constructed using the Baum-Welch algorithm [74], as implemented in the GHMM python library. For each observed regulatory link in the regulatory network, the observed transition between each permutation of pairs of strains has been used to train the HMM and then compute the states and transitions probabilities. The transition probability has been defined for each state as the probability of observing the transition between two strains. Since each state has different transition probabilities and their sum is one for each state, we do not observe symmetrical probabilities.
Regulatory motifs data has been analysed and visualized using the NumPy [75] and matplotlib [76] libraries inside the iPython environment [77]. Regulatory networks have been built using the networkx library [78] and visualized using Gephi [79].
Genomic sequences, regulatory motif files and search and analysis scripts are available as separate git repositories. The rhizoreg repository (https://github.com/combogenomics/rhizoreg/), contains the input data; the regtools repository (https://github.com/combogenomics/regtools/) contains the main scripts used to conduct the analysis.
|
10.1371/journal.ppat.1002657 | Entry of Human Papillomavirus Type 16 by Actin-Dependent, Clathrin- and Lipid Raft-Independent Endocytosis | Infectious endocytosis of incoming human papillomavirus type 16 (HPV-16), the main etiological agent of cervical cancer, is poorly characterized in terms of cellular requirements and pathways. Conflicting reports attribute HPV-16 entry to clathrin-dependent and -independent mechanisms. To comprehensively describe the cell biological features of HPV-16 entry into human epithelial cells, we compared HPV-16 pseudovirion (PsV) infection in the context of cell perturbations (drug inhibition, siRNA silencing, overexpression of dominant mutants) to five other viruses (influenza A virus, Semliki Forest virus, simian virus 40, vesicular stomatitis virus, and vaccinia virus) with defined endocytic requirements. Our analysis included infection data, i.e. GFP expression after plasmid delivery by HPV-16 PsV, and endocytosis assays in combination with electron, immunofluorescence, and video microscopy. The results indicated that HPV-16 entry into HeLa and HaCaT cells was clathrin-, caveolin-, cholesterol- and dynamin-independent. The virus made use of a potentially novel ligand-induced endocytic pathway related to macropinocytosis. This pathway was distinct from classical macropinocytosis in regards to vesicle size, cholesterol-sensitivity, and GTPase requirements, but similar in respect to the need for tyrosine kinase signaling, actin dynamics, Na+/H+ exchangers, PAK-1 and PKC. After internalization the virus was transported to late endosomes and/or endolysosomes, and activated through exposure to low pH.
| Human papillomavirus type 16 is the main etiological agent of cervical cancer. Despite advances in our understanding of transformation and cancer progression, as well as preventative vaccination strategies, the early events in papillomavirus infections are incompletely understood. Here, we investigated which strategies and cellular mechanisms the virus uses to enter epithelial cells. Entry was slow and asynchronous likely due to several structural alterations, which needed to occur on the cell exterior. Interestingly, the virus hijacked a potentially novel pathway of endocytosis for entry, which was distinct from classical macropinocytosis in regards to vesicle size, cholesterol-sensitivity, and GTPase requirements, but similar in respect to tyrosine kinase signaling, actin dynamics, Na+/H+ exchangers, PAK-1 and PKC requirements. This cellular mechanism may also be used by other viruses such as influenza A virus, echo virus 1, and choriomeningitis virus.
| Human Papillomaviruses (HPVs) are a family of small non-enveloped viruses that induce mostly benign papillomas. However, several high risk HPV types, most prominently HPV-16, cause cervical cancer and other epithelial tumors. While the molecular biology of transformation has been studied in some detail [1], the cell biology of HPV-16 entry is still a subject of scientific debate [2]. Transmission is conferred by virions that contain two structural proteins, L1 and L2. These provide an icosahedral (T = 7) capsid of 52–55 nm that protects the closed, circular, double-stranded DNA genome. HPVs initially enter basal cells of stratified epithelia [3]. The productive life cycle of HPVs requires differentiating human epidermal tissue allowing only limited infectious virus production in vitro. As a surrogate model, entry into a variety of cell lines has been studied using HPV virus-like particles (VLP) or pseudovirions (PsV) [4]–[9], i.e. viral capsids harbouring a reporter plasmid [10].
HPV-16 can bind to cell surface receptors and to the extracellular matrix (ECM). Binding to cells occurs through interaction with glycosaminoglycans (GAGs), mostly heparan sulfate proteoglycans (HSPGs, [11]–[14]). Binding to the ECM may additionally involve laminin-332 [15]. Binding results in conformational changes in the capsid that allow partial externalization of the inner virion protein L2 by cyclophilin B. This, in turn, leads to cleavage of the N-terminus proximal sequence of L2 by furin [16]–[20]. Transfer from HSPGs to a putative secondary receptor has been proposed to precede infectious internalization [21].
The literature on papillomavirus entry does not allow a generalized view regarding the mode of endocytosis. Several studies with HPV-16, HPV-31, and BPV-1 have relied on the small compound inhibitor chlorpromazine to suggest entry by clathrin-mediated endocytosis (CME) [5], [6], [22]. In addition, dynamin-2 has been suggested as a mediator in primary endocytic vesicle (PEV) formation for HPV-16 [4]. However, recent studies challenge an unambiguous role of CME in entry of papillomaviruses: HPV-31 entry into keratinocytes has been attributed to caveolar/lipid-raft mediated endocytosis [5], [8], and Spoden and colleagues [9] suggested that HPV-16 entry into 293TT and HeLa cells occurs through tetraspanin-enriched microdomains, and is clathrin-, lipid raft- and dynamin-independent. As a minimal consensus, all studies agree that HPV-16 entry is insensitive to cholesterol depletion, but sensitive to lysosomotropic agents. Some of the confusion may be explained by the use different VLP/PsV preparations and cell lines.
In this study, we addressed HPV-16 entry by systematically perturbing the function of numerous cellular key factors implicated in the various endocytic mechanisms known to date using chemical inhibitors, siRNA silencing, and overexpression of dominant negative (DN) proteins. Immunofluorescence analysis, live cell imaging, and thin section electron microscopy were used to analyze viral colocalization or cotrafficking with cellular factors, and to visualize the morphology of virus-containing vesicles. To circumvent some limitations that cellular endocytic ligands may pose for a comparative analysis such as monovalency, lesser or higher sensitivity to pertubations, different time courses, etc., we used viruses with known endocytic requirements as controls, i.e. simian virus 40 (SV40), Semliki Forest virus (SFV), influenza A virus (IAV), vesicular stomatitis virus (VSV), and vaccinia virus (VV). The results indicated that HPV-16 entry was clathrin-, caveolin-, flotillin-, cholesterol-, and dynamin-independent and did not involve the glycosphingolipid enriched endocytic carrier (GEEC) pathway, the Arf6 pathway, and the IL-2 pathway. In contrast, HPV-16 made use of a potentially novel endocytic pathway possibly related to macropinocytosis, which was dependent on actin dynamics and tyrosine kinase signaling. The viruses were carried to late endosomes/lysosomes and underwent slow acid-dependent penetration.
First, we analyzed how rapidly infectious cell-bound HPV-16 particles were internalized by endocytosis in HeLa and HaCaT cells. To distinguish between external and internalized viruses, we used a high pH wash to inactivate the PsV located on the cell surface (see Figure S1 A, B, and material and methods for details). HPV-16 PsV were bound to cells in the cold, unbound virus was washed away, and the cells were warmed to 37°C. At different times post warming, cells were subjected briefly to a high pH wash and infection was scored after 48 h by the detection of GFP expression from the PsV reporter plasmid using flow cytometry.
This way the internalization exclusively of infectious particles was probed.
When using a low multiplicity of infection (MOI, 0.1 tdu/cell), so that a single infectious particle per cell caused infection, we found that about 2 h at 37°C were needed for the first viruses to be internalized. The half-time for the entire population was 11 h (Figure 1A, black). When a higher MOI (10 tdu/cell) was used, infectious internalization by the first (fastest) of several infectious particles caused infection providing information regarding the fastest possible internalization kinetics. The fastest particles provided a half time of internalization of 4 h post warming with no apparent lag time (Figure 1A, white). Similar internalization kinetics were observed in HaCaT cells (Figure 1B).
When internalization of fluorescently labeled HPV-16 particles was analyzed by scoring cell-associated fluorescence after removing cell surface-bound viruses by protease digestion at different times post warming, similar kinetics as compared to the high MOI infectious internalization were observed (data not shown). This suggested that all particles, infectious and non-infectious, followed a similar itinerary.
With half times of internalization ranging from 4–12 h p.i. for the fastest and average particles, our data was consistent with previous observations, where bulk internalization of papillomaviruses and PsV was shown to be extremely slow [6], [23], [24]. To our knowledge, this describes the slowest internalization kinetics observed for a virus.
Endocytic internalization of individual viruses in live cells can be visualized using total internal reflection microscopy (TIRFM) by the loss of fluorescence as vesicles leave the evanescent field [25]. We used fluorophore labeled PsV that can be easily followed by microscopy and retain their infectivity after labelling [7]. As papillomavirus entry has been mostly attributed to CME, we analyzed internalization of AF488-labeled virions in cells expressing clathrin light chain tagged with mRFP (CLC-RFP) to follow any potential cointernalization.
After binding to cells, HPV-16 particles exhibit periods of lateral diffusion and directed transport, after which movement stops [7]. Such confinement is often associated with the interaction of plasma membrane receptors with the actin cytoskeleton or endocytic coat elements [26]. We analyzed how quickly the fluorescence of HPV-16 PsV particles was lost after confinement, an event consistent with internalization. The majority of particles stayed confined and were not internalized during data acquisition possibly due to HPV-16 anchoring to ECM factors and plasma membrane receptors [15], [27]. However, internalization of several particles (30 min p.i., n = 17; 2 h p.i., n = 31) could be observed, and when that happened it occurred on average within 120 s post confinement (Figure 1C, Video S1). The fastest event occurred 93 s after confinement. The formation and endocytosis of clathrin coated pits was readily observed by TIRFM in HeLa cells. Consistent with previous reports, the full cycle lasted about 60 s (Figure 1C) [28], [29]. Cointernalization of HPV-16 and clathrin was not observed. Thus we found that single virus internalization events were fast, but occurred only sporadically over a period of many hours. The absence of any clathrin signal during internalization suggested a clathrin-independent endocytosis event.
This prompted us to reinvestigate the endocytic requirements of HPV-16 entry. We used small compound inhibitors, siRNA-mediated depletion, and overexpression of DN proteins to identify the cellular factors required or dispensable for infectious endocytosis.
To reduce potential cytotoxic effects of cellular inhibitors, the inhibitors were replaced after the entry phase was over by a less toxic inhibitor such as NH4Cl that blocked acid-activation of viruses in endosomes (Figure S2A). Swapping for a second inhibitor allowed us to assess the effect of an inhibitor of interest on pre-acidification in the entry process such as binding, endocytosis and perhaps intracellular routing (compare Figure S2A, Table S1, and material and methods). The infection data was normalized to cells in which only the secondary inhibitor was used. In addition, we excluded any data that exhibited significant cytotoxic effects, i.e. a reduction of cell number by more than 50% for a given pertubation. Infection and cytotoxicity was analyzed by automated microscopy and computational analysis of cell number and infection (see Figure S2B and material and methods for details).
To minimize erroneous interpretation of data, and to exclude pleiotrophic effects beyond cell toxicity, we controlled for the specificity and efficiency of inhibition using prototype virus infection experiments using the same inhibitor. In addition, cell biological controls such as transferrin uptake for CME were more sensitive to pertubation than virus infection (data not shown). Here, we used SFV and VSV as controls entering by CME, and SV40 using caveolar/lipid-raft dependent endocytosis. Further controls included infection by VV for macropinocytosis, and IAV for penetration of late endosomal compartments. To stay within one experimental system for comparison of the different viruses, we used HeLa cells but verified key findings for HPV-16 in HaCaT cells. Since the dominant effect of mutant proteins often depends on the level of overexpression, we analyzed infection of transfected cells by flow cytometry and grouped the data into low, medium, and high mutant protein expressing cells (Figure S2C).
First, we analyzed whether infection of HPV-16 in HeLa cells was clathrin-dependent. To inhibit CME, we used siRNA to deplete the clathrin heavy chain or the μ subunit of the clathrin adaptor AP2. A depletion to 20% or less of the protein levels in control cells resulted in efficient decrease in SFV infection (Figure 2A, Figure S3A), a virus that enters exclusively by CME [30]–[32]. HPV-16 infection was not significantly affected (Figure 2A), indicating that HPV-16 entry was clathrin- and AP2-independent. This was consistent with the TIRFM results described above.
Chlorpromazine, generally considered an inhibitor of CME, abolished HPV-16 infection at 10 µg/ml in line with earlier studies [5], [8], [22]. At this concentration, chlorpromazine also partially affected infection of SV40, a virus that makes use of CME-independent endocytosis (Figure 2B). Since efficient clathrin and AP2 depletion did not affect HPV-16 infection, we suspected chlorpromazine to cause CME-independent inhibition of entry.
To test whether HPV-16 was using a caveolin/lipid raft-mediated pathway, we depleted cells of cholesterol with nystatin/progesterone, a combination of drugs often used to inhibit fomation of lipid rafts in the plasma membrane [33]. SV40 infection, which uses a combination of a caveolar and a caveolin-independent lipid raft-mediated mechanism [34]–[36], was efficiently inhibited (Figure 2C). SFV infection was also affected most likely due to its cholesterol-dependent membrane fusion activity [37]. However, HPV-16 entry was, if anything, increased (Figure 2D). Similarly, cholesterol extraction by methyl-ß-cyclodextrin had only little effect (Table S1). We found, moreover, that HPV-16 infected HeLa cells depleted of caveolin-1 and caveolin-1 knockout cells with the same efficiency as wild-type cells (Figure 2D, E, Figure S3B). Since SV40 can make use of a caveolin-independent lipid-raft dependent pathway [34], and since cholera toxin B, that can enter cells by caveolar endocytosis, can make use of multiple pathways [38], meaningful positive controls could unfortunately not be employed.
To test whether endocytosis of HPV-16 would involve flotillins, proposed to define a distinct endocytic mechanisms [39], we depleted cells of flotillin-1 or -2, or both by siRNA mediated knockdown. Since no viral cargo exists for flotillin endocytosis, and since cholera toxin B can use multiple pathways, again, no direct positive controls could be used. Independent or combined depletion of flotillin-1 and -2 to 20% protein levels or less of control cells had no effect on HPV-16 infection, whereas SV40 infection was enhanced after flotillin-2 but not -1 knockdown (Figure 2F, Figure S3C).
We observed no significant colocalization of HPV-16 with caveolin-1 or flotillin-1 at 2 h post warming by confocal microscopy (Figure 2G–J). At 8 h, the degree of colocalization between HPV-16 and caveolin or flotillin increased slightly (Figure 2H–J). The colocalizing objects were mostly located intracellularly and perinuclearly, which suggested that they belonged to a population of endosomal vacuoles (Fig. 2G, H; [35], [40]).
In summary, infectious HPV-16 entry was independent of clathrin, AP2, caveolin, lipid rafts and flotillin, and HPV-16 did not associate with caveolin or flotillin at the cell surface. This indicated that HPV-16 entry occurred by a distinct endocytic mechanism.
Dynamin-2 facilitates membrane fission to generate endocytic vesicles in several endocytic pathways [41], [42]. To test its involvement in HPV-16 entry, we used overexpression of the DN K44A mutant, the inhibitor dynasore, and siRNA depletion of dynamin-2 as methods of pertubation. Overexpression of K44A dynamin-2 deficient in GTP hydrolysis reduced infection of SFV or SV40 (Figure 3B, C) in a dose dependent manner to 40% or 50% of control, respectively,whereas HPV-16 entry was only slightly perturbed (Figure 3A). Dynamin-2 depletion had no significant effect on HPV-16 infection (Table S1). Dynasore, a noncompetitive inhibitor of dynamin-2, reduced infection of SFV and SV40 to 30% and 40%, respectively (Figure 3D). In contrast, HPV-16 entry was enhanced by up to 50% relative to control (Figure 3D). Our data suggested that dynamin-2 was not required for HPV-16 endocytosis.
Actin polymerization plays a role in several endocytic pathways. During phagocytosis and macropinocytosis it causes outward protrusion of the plasma membrane. In addition, actin polymerization but not outward protrusion of the plasma membrane is required in caveolar/lipid raft endocytosis, the IL-2, the glycosphingolypid enriched endocytic compartment (GEEC), and the Arf6 pathways [41], [42].
To test the involvement of actin, we used the actin depolymerizing agent cytochalasin D and the stabilizing agent jasplakinolide. As expected, the drugs strongly inhibited SV40 infection, whereas SFV infection was unperturbed by jasplakinolide but strongly enhanced by cytochalasin D (Figure 4A, B). Since HPV-16 infection was efficiently inhibited by both drugs (Figure 4A, B, Table S1), we concluded that actin dynamics facilitated infection, in line with previous studies on HPV-16 uptake in BHPE cells or HPV-33 entry into COS-7 cells [23], [43].
Since actin polymerization is most critical during macropinocytic/phagocytic protrusion of the plasma membrane, we checked if HPV-16 infection would be affected by inhibition of the Na+/H+ exchanger, considered to be a hallmark of macropinocytosis [44]. SFV and SV40 infection were only mildly affected upon EIPA treatment, whereas HPV-16 infection was effectively reduced (Figure 4G), so that we concluded that the Na+/H+ exchanger is involved.
During macropinocytosis, actin polymerization events are regulated by the Rho-like GTPases Cdc42 and Rac1, whereas phagocytosis and caveolar/lipid-raft endocytosis depend on RhoA [42], [44], [45]. We first inhibited RhoA by C3 toxin, and all Rho-like GTPases by toxin B. C3 toxin, as expected, partially inhibited SV40 infection, but did not affect HPV-16 entry (Figure 4C). Toxin B inhibited SFV infection and SV40 infection, but HPV-16 infection was unperturbed (Figure 4C).
Since Rho-like GTPases are key regulators of actin polymerization, we wanted to verify these results. We overexpressed Cdc42, Rac1, and RhoA, and their DN (T17/19N) or constitutively active (CA, Q61L, G14V) mutants. None of the mutants affected HPV-16 entry (Figure 4D). However, when we infected Cdc42 expressing cells with vaccinia virus (IHD-J/GFP), which enters cells by macropinocytosis [46], [47], the DN mutant blocked infection by about 50% (Figure 4E). Hence, we concluded that HPV-16 infection occurred by an endocytic mechanism dependent on actin polymerization but independent of signaling via the classical Rho-like GTPases.
Arf6, another small GTPase, has been implicated in macropinocytosis and in an elusive endocytic pathway that occurs in HeLa cells [48], [49]. Arf6 also has the potential to stimulate actin polymerization [50]. Hence, we tested whether Arf6 might be regulating actin dynamics during HPV-16 endocytosis. HPV-16 entry was unaffected by overexpression of the CA (Q67L) or DN (T27N) mutant (Figure 4F). In addition, siRNA mediated silencing of Arf6 did not affect HPV-16 entry (Table S1). We concluded that HPV-16 endocytosis did not involve an Arf6 mediated pathway, and that actin dynamics during HPV-16 entry were not regulated by Arf6.
Also, actomyosin contractility was not required for entry, as inhibitors of myosin II and myosin light chain kinase only mildly reduced infection consistent with their role in transport of HPV-16 along filopodia by actin retrograde flow (Table S1, [7]).
Specific cellular signals are required for activating different endocytic pathways [41], [42], [51]. Hence, we analyzed the role of major kinases and phosphatases involved in endocytic uptake and their effect on HPV-16 entry. Genistein-mediated inhibition of tyrosine kinases blocked HPV-16 entry (Figure 5A). As expected, SV40 infection was likewise affected (Figure 5A, [34], [36]), while SFV infection was unaffected (Fig. 5A). Similarly, inhibition of serine/threonine kinases with H-7 blocked HPV-16 and SV40 infection but not SFV infection (Figure 5B). When we inhibited phosphatases by the broad inhibitor orthovanadate, entry of HPV-16, SV40 and SFV were inhibited (Figure 5C). However, when the phosphatase classes PP1, PP2A, and PP2B were inhibited by okadaic acid, or PP1 by tautomycetin, SFV infection was largely unaffected whereas HPV-16 and SV40 infection was blocked (Figure 5C, Table S1).
To further analyze signaling in the context of HPV-16 endocytosis, we inhibited major kinases involved in macropinocytosis such as EGF receptor tyrosine kinase (EGFR), protein kinase C (PKC), PI3 kinase (PI3K), and p21-activated kinase (PAK-1) [44]. Inhibition of EGFR signaling by Iressa blocked HPV-16 and VV but not SFV entry (Figure 5D). Also, PKC signaling was required for HPV-16 infection as indicated by the inhibitory effects of calphostin C and rottlerin (Figure 5E, Table S1). Inhibition of PI3K by wortmannin or PI-103 reduced HPV-16 infection by 50–60%, which indicated some involvement (Figure 5F, G). Inhibition of PAK-1 by IPA-3 blocked HPV-16 and VV but not SV40 infection (Figure 5H). Also, HPV-16 and VV but not VSV entry were reduced in PAK-1 depleted cells (Figure 5I).
In summary, HPV-16 infection required activation of cellular signaling. The signaling factors included EGFR, phosphatases, PKC and PAK-1, possibly additional tyrosine and serine/threonine kinases, and potentially PI3K. This list of signaling components is similar to the signaling activating macropinocytosis.
To test whether the required cell factors were also critical in more physiologically relevant cells, we infected pharmacologically perturbed HaCaT cells, i.e. spontaneously immortalized keratinocytes. There was virtually no difference between HeLa and HaCaT cells suggesting that HPV-16 entry occurred by a similar endocytic mechanism (Figure S4).
To test whether any of the inhibitors affected HPV-16 binding to cells we measured the association of fluorescent viruses with perturbed HeLa cells. As a positive control, we blocked cell binding by preincubation of PsV with soluble heparin, as previously described [11], [13], [52]. Alternatively, chlorate treatment of cells was used, which results in undersulfation of GAG chains that the virus uses as attachment factors. In both cases, a dramatic reduction in binding was observed (Figure 6A). When we used a variety of inhibitors that did or did not perturb HPV-16 entry, no significant reduction in binding was observed (Figure 6B). Only cholesterol depletion of the plasma membrane by methyl-β-cyclodextrin showed a minor reduction in binding that correlated with a small reduction in infection, which suggested a general membrane-receptor perturbing effect of the drug as nystatin/progesterone had no effect.
To analyze which factors contributed to endocytic internalization as opposed to intracellular events prior to acid-activation we employed the internalization assay for infective particles and compared it to the infection assay with inhibitor swap (compare Figure 1A, Figure S1, S2, material and methods). The relative amount of internalized infectious particles was analyzed at 12 h, when the drug was washed out, and a high pH wash inactivated extracellular virus, so that infection could only be caused by already internalized virus. Infection was continued the absence of an inhibitor for further 36 h. To control for the general reversibility of drug treatment and for cytotoxicity, we included samples where we did not subject cells to a high pH wash. Any samples where reversibility of the inhibitors of interest was significantly affected were excluded. As a positive control, we used chlorate treatment.
Infectious internalization and infection were found to be similarly reduced by cytochalasin D (actin polymerization), jasplakinolide (actin depolymerization), EIPA (Na+/H+ exchanger), genistein (tyrosine kinases), PI-103 (PI3K), IPA-3 (PAK-1), and H-7 (serine/threonine kinases) (Figure 6C). Although infection was reduced, internalization was only slightly perturbed by treatment of cells with orthovanadate (phosphatases), or unaffected by bafilomycin (endosomal pH), and nocodazole (microtubules) suggesting a role in intracellular routing of the virus. Nystatin/progesteron (cholesterol), blebbistatin (myosin II), and ML-7 (myosin light chain kinase) did not affect HPV-16 internalization or infection, as expected. We concluded that endocytic internalization of HPV-16 required actin polymerization dynamics, the function of the Na+/H+ exchanger, tyrosine and serine/threonine kinases, PAK-1, and PI3K signaling. Intracellular steps required phosphatases, microtubules, and acid-activation.
Then, we assessed internalization of HPV-16 particles (as opposed to infectious units) using a trypan blue based assay [35]. Addition of the membrane impermeable trypan blue to cells quenched the fluorescence of AF594 labeled particles bound to cells, whereas the fluorescence of internalized particles was unaffected (Figure S5). Microscopic analysis showed that e.g. cytochalasin D rendered almost all HPV-16 particles susceptible to trypan blue consistent with a block in internalization, whereas e.g. NH4Cl and nocodazole did not (Figure 6D). Since this data correlated with our infectious internalization analysis (Figure 6C, D; data not shown), we concluded that HPV-16 particles were not internalized and not channeled into a non-infectious uptake pathway but stalled at the cell surface.
Using thin section electron microscopy, we analyzed the morphology of HPV-16 internalization in HeLa and HaCaT cells. Bound viruses were readily detectable at the plasma membrane (Figure 7A). Occasionally, viruses appeared to cause a slight indentation in the plasma membrane (Figure 7B). From 1–24 h post warming, viruses were mainly observed in indentations 65–120 nm in diameter without a visible coat (Figure 7D, H, I) but not in clathrin coated pits (Figure 7F, data not shown). Occasionally, these indentations were wider or tubular and held several particles (Figure 7E, not shown). The particles had a distance of about 5–10 nm from the membrane. About 10–20% were associated with membrane protrusions resembling filopodia, which we have previously shown to be capable of binding and transporting HPV-16 (Figure 7H, [7]). Virus containing indentations were occasionally located at the base of such protrusions (Figure 7H). In close proximity to the plasma membrane, viruses were observed in small vesicles (70–140 nm in diameter) that held one or sometimes two particles (Figure 7C, D, J). Our data suggested that viruses were internalized via such small uncoated pits and vesicles that could accommodate one or possibly several particles.
When virus internalization was blocked using cytochalasin D to inhibit actin polymerization, virus particles were found in deep, narrow, almost tubular invaginations of the plasma membrane that varied in length from a few 100 nm to microns where viruses occurred in tight rows (Figure 7G, K). Thus, the endocytic pits containing viruses seemed to grow into long tubules unable to separate from the plasma membrane by scission. In addition to implicating actin polymerization in scission, this suggested that viruses were internalized after sinking into plasma membrane rather than by protrusive engulfment, which would often be seen during macropinocytosis.
After endocytic internalization, most incoming animal viruses are delivered to endosomes and their penetration into the cytosol occurs either in early or late endosomes (EE, LE). If the viruses are endocytosed by macropinocytosis or related mechanisms, the agenda is less well characterized. Macropinosomes can recycle to the plasma membrane or fuse with endosomes or lysosomes at different levels of the pathway [44]. They acidify and it is possible that they undergo a maturation of their own before fusing with lysosomes.
Previous studies have shown that entry of a variety of papillomaviruses is blocked by agents that raise the pH in endosomes and lysosomes [5], [6], [8], [22], [53]. We confirmed that HPV-16 infection, both in HeLa and HaCaT cells, is reduced by the weak base NH4Cl, which neutralizes endosomal pH, and by other inhibitors of endosomal acidification such as bafilomycin A1 and monensin (Table S1, not shown). Hence, endocytosis carries viruses to acidic compartments, where low pH-based activation for penetration or uncoating likely occurs.
To determine when the virus passed the acid-activated step during entry, we added NH4Cl and bafilomycin A1 at various times post warming and scored for infection. If infection would no longer be perturbed at a certain time, the viruses must have passed the acid-activation step. Our results indicated that about half of the virus population had passed the low pH dependent step by 13–16 h in HeLa and HaCaT cells (Figure 8A). This means that the critical acid exposure occurred 2–3 h after endocytosis (compare Figure 1A, B). We confirmed previous findings that infection can be blocked by nocodazole-induced depolymerization of microtubules suggesting a need for intracellular transport of endocytic vacuoles or the virus itself (Figure 9A, [6], [23]). Since nocodazole add-on experiments followed a similar time course (not shown), we assume that microtubule-based intracellular sorting of endosomes is required for successful HPV-16 infection.
Progression in the endocytic pathway involves the risk that viral particles are degraded. We infected cells at a low MOI in the presence of NH4Cl and subsequently washed out NH4Cl at various times. The virus remained infectious for prolonged times exhibiting 80% infectivity even after a washout at 24 h p.i. (Figure 8B). Thus, the viruses were not rapidly routed to compartments, where they were inactivated after acidification.
To analyze the average time that viruses require for acid-activation, we infected cells in the presence of NH4Cl. After 12 h, we washed out NH4Cl which results in a fast re-acidification of endosomes and lysosomes [54]. At various times after washout, we re-added NH4Cl, and thus created a time window of virus exposure to low pH. When infectivity was quantified, we found that the halftime of activation was 6 h (Figure 8C), a surprisingly long time period suggesting that to be infectious the virus had to be exposed to low pH, or an enzyme requiring low pH for several hours. That the long time period was specific to HPV-16, is shown by the fact that the low pH based activation of VSV occurred almost completely within a time window of 15 min. (Figure S6).
Since most transport steps in the endocytic pathway are regulated by Rab GTPases, we analyzed which of several RabGTPases were required for HPV-16 infection. We overexpressed GFP- or RFP-tagged Rab GTPases either as wildtype, or in the form of DN or CA mutants.
The DN and CA mutants of Rab5 reduced HPV-16 infection when highly overexpressed (Figure 9B). To control for the efficiency of the dominant effect, we used SFV and VSV as controls, which both require Rab5-mediated transport to EEs for infection. Since the infectivity of both was similarly reduced (Figure 9B), we concluded that HPV-16 required Rab5. No requirement for Rab7 was detected (Figure 9C, left panel). As a positive control, we infected cells with IAV that requires Rab7 for transport to the LE compartment [55]. Since infection of IAV was reduced by the DN mutant, we concluded that HPV-16 was Rab7-independent (Figure 9C, right panel). Overexpression of Rab1, Rab4, Rab6, Rab11 and their mutants did not affect infection (Figure S7).
When the localization of HPV-16 with EE markers was analyzed, we found no significant colocalization of virus with the EEA1 from 30 min -2 h or at 8 h (Figure 9D, E, data not shown). However, we readily found AF488 labeled particles that co-migrated with Rab5 positive structures in live cells (Video S2). The time period spent in the Rab5-positive compartments may be fairly limited as exemplified in Video S2 by a virus entering a Rab5-positive, virus containing vacuole that loses the Rab5 signal about 90 s after this event.
Colocalization of HPV-16 with the LE and lysosome marker LAMP-1 was detectable at 8 h p.i. (Figure 9D, F). Since the localization of LAMP-1 in the crowded perinuclear space may introduce artifactual values for colocalization, we analyzed colocalization with giantin, a Golgi marker. No significant colocalization was observed (Figure 9D, G). We concluded that PsV were trafficked to perinuclear, LAMP-1 containing, acidic compartments. Virus was transported to these vacuoles in the presence of NH4Cl (Figure 9D, G). Combined with the fact that after NH4Cl washout the virus remained infectious (compare Figure 8B), this result suggested that the transport occurred to LEs or endolysosomes, and that this represented the infectious route.
EM showed the virus frequently in multivesicular bodies (Figure 10B, C) and lysosomes (Figure 10D), and occasionally in large endosomal vacuoles that lacked internal vesicles (Figure 10A). They were not observed in the Golgi complex, the endoplasmic reticulum (ER), or the cytosol. Interestingly, virus particles were detectable even late during infection in lysosomal compartments, which suggested that despite the fact that BrdU labeled DNA of PsV become accessible to antibody staining [56], they were not degraded and remained as recognizable particles.
In summary, our data were consistent with intracellular transport of HPV-16 to the LE/lysosomes. During transport the virus-containing vacuoles underwent a transient Rab5 association.
The cell biological experiments presented here indicated that entry of HPV-16 into tissue culture cells involves strategies different from those described for most other viruses. Instead it involves signaling-dependent formation of small invaginations and vesicles at the plasma membrane devoid of detectable coats. The cellular factors needed to generate the PEVs are similar but not identical to those involved in macropinocytosis. After endocytosis, the viruses are deposited into acidic organelles with similarities to classical endosomal compartments and macropinosomes. Clearly, the uptake process in HeLa and HaCaT cells does not depend on CME as previously proposed [5], [6], [8], [22].
Internalization and intracellular trafficking of HPV-16 occurs exceptionally slowly and asynchronously. In agreement with previous reports [6], [23], [24], infectious internalization of a pre-bound cohort of viruses lasted over a period of no fewer than 20 h with a t1/2 of 11 h. Judging by the timing of the critical acid-exposure step, a microtubule-dependent step, and the arrival in LAMP-1-containing vacuoles, the average time that viruses spent in endocytic organelles before penetration into the cytosol was 2-3 h. For efficient infection, the viruses had to be exposed to acidic pH in the endosomal pathway for several hours. In comparison, viruses such as VSV and SFV are internalized within minutes and acid-activated penetration takes place a few minutes thereafter (compare Figure S6) [37], [57]. Failing to penetrate, they follow the classical endosome pathway and begin to be degraded in lysosomes within 30 min. The only other viruses with comparatively slow entry kinetics are the polyomaviruses. For example SV40 takes 6–12 h to reach the ER where penetration occurs [58], [59].
Initial binding of HPV-16 particles to the cell surface occurs rapidly and with high affinity [60]. In a mouse vaginal model, binding occurs to the basement membrane and depends on HSPGs [14], [20], whereas in human tissue culture cells ECM binding can in addition involve laminin-332 (previously named laminin-5, [15]). Proteoglycan-bound viruses undergo a structural change that subsequently allows further changes conferred by cyclophilins and furin [16], [17]. A conserved epitope in the minor capsid protein L2 is exposed [18]. This is crucial for infection, and probably followed by transfer of the virus to an unidentified secondary cell surface receptor [18], [21].
Live cell microscopy showed that the internalization event as such is rapid. It occurs on average within 2 minutes once lateral movement of the virus has stopped. It is possible that viruses can only trigger infectious internalization once they are fully activated and have been transferred to the new receptor. This means that the changes induced by cyclophilin and furin, or the transfer to the secondary receptor may be rate-limiting for internalization. This step-wise process would thus occur asynchronously followed by a fast endocytic internalization. Viruses would then be endocytosed one at a time causing a protracted internalization time course seen in the virus population. Since the structural changes occur when the viruses are bound to the ECM as shown in mouse vaginal tissue [20], transfer from the ECM to the cell body may also take a long time. The transfer from ECM to cells is probably promoted by actin- and myosin II-mediated transport of viruses along filopodial cell surface protrusions as observed in tissue culture cells [7].
Since internalization was strongly dependent on cell signaling, it was most likely a virus-induced process. Inhibitor analysis showed that the signaling cascade included the EGFR as well as other tyrosine and serine/threonine kinases such as PAK-1 and PKC. In addition, PI3K signaling was required, which was previously identified to be required for entry of HPV 6b, 18, 31, and 35 or for uptake of HPV-16 into langerhans cells [61], [62]. The involvement of the EGFR and the other downstream signaling kinases was reminiscent of macropinocytosis, which is often activated by receptor tyrosine kinases and requires a series of further kinases. Another kinase that may additionally be involved in HPV-16 endocytosis is the FAK [63]. That the signaling cascade was modulated, was suggested by the requirement for cellular phosphatases.
Endocytosis and infection were clearly independent of CME. Clathrin, AP-2, and dynamin were not needed, and HPV-16 particles did not co-internalize with clathrin-light chain-RFP, or co-localize with endogenous clathrin heavy chain (not shown). Internalization took place in smooth, uncoated pits. That chlorpromazine blocked infection and internalization of HPV-16 particles may be explained by pleiotrophic effects of the drug. Namely, chlorpromazine can interfere with the formation of phagosomes or macropinosomes [64], [65], cause inhibition of PLC-regulated actin rearrangements [66], [67], and change plasma membrane fluidity [68]. Also, HPV-16 entry did not occur by a dynamin-, caveolin-, flotillin-, lipid raft-, or Arf6-mediated mechanism, thus excluding a role for other often reported micropinocytic pathways (compare [41], [51]). Two previous studies implicated caveolin-1 and dynamin-2 in HPV-16 infection in HaCaT cells [4], [69]. However, the observed effects of dynamin pertubation were marginal, cell toxicity due to caveolin-1 depletion high, and no positive and negative controls were employed [4], [69].
Actin polymerization and depolymerization was critical for HPV-16 endocytosis. That long, virus-filled, tubular invaginations were formed in the plasma membrane of cytochalasin D-treated cells suggested that actin was required for scission of endocytic vesicles and not for the generation of membrane protrusions as observed for macropinocytosis. The lack of protrusions was supported by electron microscopy data, and the lack of ruffling or bleb formation by light microscopy ([7], not shown). Moreover, the dynamics of actin were not regulated by the classical Rho GTPases, Cdc42, Rac1, and RhoA. In contrast to macropinocytosis which occurs via large, irregularly shaped, fluid-filled vacuoles, the HPV-16 was observed in small invaginations in the plasma membrane (65–120 nm in diameter). These lacked any visible coat structure. Scission resulted in the formation of small endocytic vacuoles that held one or occasionally two virus particles.
Obviously, many details in the process of vesicle formation remain unclear. We do not know how membrane curvature beneath the virus is induced, how growing pits are stabilized, how and when constriction and membrane scission occur, and how these events are coordinated in time and space. Numerous cellular proteins are likely to be involved in this process. Tetraspanin-enriched microdomains have been suggested as entry platforms for HPV-16 endocytosis [9], and may serve as scaffolds for the recruitment of cytosolic proteins and as a signaling platform.
After endocytosis, the viruses were briefly associated with Rab5-containing vacuoles, and Rab5 was also required for infection. Whether the Rab5-containing structures were EEs is not clear, because no significant colocalization with another marker of EEs, EEA1, could be observed. The viruses were further transported to LE and lysosomes positive for Rab7 and LAMP1. It is possible that the virus bypassed EEs, and like macropinosomes and phagosomes entered the endocytic pathway at the level of LEs or endolysosomes [70].
Two to three hours after internalization the virus passes through a low pH-dependent step critical for infection. It is unclear at this point, whether it involves structural changes triggered in the virus triggered directly by low pH, a block in endosome trafficking and fusion [71], or the impaired function of endosomal proteases [72].
The combination of low pH-dependent activation and action of endosomal proteases after proper trafficking may result in a major uncoating step. There is evidence the viral genome and the minor capsid protein L2 are exposed in a process most likely triggered by the shedding of at least a part of L1 [56]. Remarkably, even after 18 h within LE/lysosomal compartments virus particles appeared to be intact by EM. Potentially, the virus could fully uncoat only after or during translocation into the cytosol. Penetration through the vacuolar membrane has been suggested to involve exposure of a membrane destabilizing peptide in L2 [73]. Before they can enter the nucleus, the viruses use microtubule-dependent transport mediated by an interaction between L2 and dynein [74].
Taken together, the cell biological features of HPV-16 endocytosis and intracellular trafficking showed several similarities with macropinocytosis. In Table 1, we have listed some of the known properties of macropinocytosis as well as their manifestation during vaccinia virus entry. When compared with the results on HPV-16 endocytosis, the strong dependence of HPV-16 on actin dynamics, receptor tyrosine kinases, PKC, PAK-1, and the Na+/H+ exchanger are consistent with a macropinocytic mechanism [44]. In particular, the requirement for receptor tyrosine kinases, PAK-1 and the Na+/H+ exchanger suggest that the signaling pathway is shared [44]. However, HPV-16 endocytosis did not involve formation of ruffles or other types of plasma membrane protrusions. The role of actin was rather connected to the pinching off of endocytic vesicles. Cholesterol-rich domains, myosin II and Rho GTPases were not involved, and the HPV-16 containing endocytic pits and vesicles were small and relatively homogeneous in shape and diameter. Another difference was that endocytosis of HPV-16 occurred over prolonged periods of time and seemed to activate the process locally, whereas e.g. a single vaccinia virus has been shown to activate macropinocytosis transiently for about 30 min and on large areas of the plasma membrane. We have to conclude that HPV-16 endocytosis did not occur by normal macropinocytosis. The data also did not conform with what is known about phagocytosis (Table 1).
HPV-16 is most likely not the only virus using this particular endocytic mechanism for entry. Although IAV can use CME for uptake, electron and light microscopy have indicated the presence of a parallel, macropinocytosis-like alternate pathway involving small pits and vesicles devoid of a clathrin coat [75], [76]. Recent analysis of this pathway revealed dependence on a set of cellular factors similar to those used by HPV-16 (Table 1, [77]). Note, that like HPV-16 this pathway does not depend on Rho GTPases. Other viruses that may follow similar pathways include echo virus 1, a picorna virus, and lymphocytic choriomeningitis virus, an arena virus [78], [79]. A more detailed cell biological analysis of these viruses will be needed to define whether the mechanims are the same, whether they too represent variants of macropinocytosis different enough to be considered a distinct endocytosis pathway altogether.
In respect to papilloma viruses, it will be important to determine if the pathway described here applies to HPV-16 infection in epithelial tissue cells in situ. The common findings on the role of actin [23], [43] and on the requirement of PI3K [61], [62] in different cell types and with HPV subtypes, together with divergent results on the use of lipid rafts and caveolin for e.g. HPV-31, warrant a comparative study of different HPV subtypes and papillomaviruses from warts and raft cultures to adress whether the majority of HPV subtypes makes use of the same pathway in a variety of cell types.
HeLa and HaCaT cells, and mouse embryonic fibroblasts (caveolin-1 wildtype and knockout, [35]) were cultured in DMEM (Invitrogen) containing 10% fetal calf serum. For a full list of antibodies, inhibitors, plasmids, siRNAs, or reagents used, please refer to Table S2.
HPV-16 PsV containing the pRwB or pCIneo-EGFP plasmid were produced with the p16sheLL plasmid by the propagation method described by Buck and colleagues [80]. The PsV were matured for 24 h in the presence of RNAse to maximize the purification of PsV containing the reporter plasmid, resulting in an improved particle to infectivity ratio [80]. All plasmids and production methods are fully described on the Schiller laboratory's website (http://ccr.cancer.gov/staff/staff.asp?profileid=5637). IAV (H1N1, A/PR 34/8), SFV, SV40, recombinant VSV (Indiana) expressing GFP (VSV-GFP), recombinant VV (IHD-J) expressing GFP were produced as described [32], [47], [57], [59], [81].
HPV-16 PsV were covalently labeled with fluorophores as described [7]. Briefly, purified HPV-16 PsV were incubated for 1 h at room temperature in PBS with a ten-fold molar excess of Alexa Fluor (AF) succinimidylesters (Molecular Probes) over the major capsid protein L1. PsV were separated from the labeling reagent by size exclusion chromatography using NAP5 columns (GE Healthcare) and stored at 4°C.
For transfection, cells were trypsinized, pelleted, washed with PBS and transfected with expression plasmids in Nucleofector solution R (Amaxa) utilizing program I13 of the Amaxa Nucleofector according to the manufacturer's instructions. Cells were seeded in 12-well plates or on 18 mm coverslips and used for infection or live cell imaging experiments, respectively, at 6–14 h post transfection.
Microscopy was performed on a custom modified Olympus IX71 inverted microscope. Modifications included a heated incubation chamber that surrounded the microscope stage set to 37°C, an objective-type TIRFM setup from TILL Photonics (Grafeling, Germany), and a monochromator for epifluorescence excitation with a controller allowing hardware-controlled fast switching between total internal reflection fluorescence and epi-fluorescence excitation and acquisition (TILL Photonics). Images were acquired using a TILL Image QE chargecoupled device camera and TILLVISION software (both from TILL Photonics). The total internal reflection angle was manually adjusted for every experiment. Live HeLa cells on 18-mm coverslips were mounted in custom-made chambers. 5,000 HPV-16 PsV/cell labeled with AF dyes were added into the 0.5 ml of medium on the stage. Movies were recorded at a rate of 2–5 frames per s.
Live cell confocal microscopy was performed on a Zeiss 510 microscope with a confocal laser scanning setup. The confocal microscope was equipped with a heated incubation chamber set at 37°C. Cells were mounted on 18-mm coverslips in custom-made chambers, and cells were incubated with normal growth medium. HPV-16 PsV (500–1,000 particles/cell) were added into the 0.5 ml of medium on the stage. After 30 min – 4 h post addition of virus particles to cells image acquisition was performed. Movies were recorded at a rate of 2 frames per s.
For siRNA mediated knockdown experiments, 3,000 cells were reverse transfected in optical bottom 96-well plates (Nunc) with the indicated amount of siRNA oligos using Lipofectamine RNAimax (Invitrogen), and infected 48 h post transfection of siRNAs. In case of oligos directed against clathrin heavy chain and the AP2 μ subunit, cells were transfected for a second time 24 h post initial transfection. For small compound inhibition studies, 5,000 cells were seeded 24 h prior to experimentation in 96-well optical bottom well plates. Inhibitors were added at indicated concentrations 30 min or as indicated 12–16 h prior to infection (Figure S2A). Cells were infected with HPV-16, SV40, SFV, or VV to result in about 20% infection. For inhibitor studies, the primary inhibitors were exchanged at 12 or 10 h p.i. against 10 mM NH4Cl or 5 mM DTT in case of HPV-16 or SV40 infection, respectively, to reduce cytotoxicity (Figure S2A). After 4% paraformaldehyde fixation, cells were stained with Hoechst 33258 for cell nuclei and in case of SFV, SV40, and IAV with antibodies directed against SFV glycoproteins, T-antigen, and nucleoprotein, respectively, and secondary antibodies conjugated to Alexa Fluor 488 (Molecular Probes) directed against the primary antibodies. Nine images per well were acquired for each, nuclear stain (Hoechst 33258) and infection stain (GFP, or IF staining for SV40 T-antigen, SFV glycoproteins), on a Pathway 855 automated microscopy station (Becton Dickinson) using a 10× objective (Olympus) employing a laser-based autofocus every image. Cell numbers and raw infection indices for each well were determined using a MATLAB-based infection scoring procedure (The Mathworks) described previously ([35], Figure S2C). The raw infection indices were normalized to solvent treated control cells or AllStar negative siRNA transfected control cells to result in relative infection percentages.
5×104 cells seeded in 12-well plates were either transfected as described above or treated with small compound inhibitors 30 min prior to infection. Cells were infected with HPV-16 dsRed or GFP, VSV-GFP, VV-GFP, or IAV to result in 20% infection. Cells were trypsinized, fixed in 4% paraformaldehyde, and in case of IAV, immunofluorescence staining for the nucleoprotein was performed. Transfected cells were analyzed by flow cytometry for the level of transgene expression, grouped into low-, medium-, and high-expressing cells (Figure S2C), and the subpopulation was subsequently analyzed for the number of infected cells. The relative number of infected cells was normalized to cells expressing only the XFP-tag according to expression levels. Inhibitor treated cells were normalized to solvent treated control cells.
5×104 cells were seeded on 18 mm coverslips and infected 16–24 h post seeding with 500 AF488-labeled HPV-16 particles/cell for 2 or 8 h. Cells were fixed with 4% paraformaldehyde or in ice-cold methanol, stained against cellular structures with primary antibodies (Table S2) and AF594 labeled secondary antibodies (Molecular Probes). Cells were mounted in Citifluor AF1, and analyzed on a Zeiss 510 microscope with a confocal laser scanning setup. Per experiment, at least 10 fields of view were imaged in 5 confocal slices with 3–8 cells/field of view. The degree of colocalization between virus and cellular structures was assessed using Bioimage XD (www.bioimagexd.net). The thresholded pixel-per-pixel colocalization was analyzed in an automated fashion where significance of colocalization was attributed to a Costes P-value≥0.95 [82].
30,000 Hela cells were seeded into optical bottom 96-well plates 24 h prior to experimentation to reach confluency. Cells were pretreated with inhibitors as in the infection studies. AF488 labeled HPV-16 (5,000 particles/cell) was added to each well, and incubated for 1 h at 37°C, when cells were fixed in paraformaldehyde and stained with Hoechst 33258 for cell nuclei. Cellular and cell associated fluorescence was analyzed using a Safire2 plate reader (Tecan), normalized to solvent controls, and depicted as relative cell associated fluorescence (relative binding).
5×104 HeLa cells were seeded on glass coverslips 24 h prior to experimentation. Cells were infected with HPV-16 AF594 (500–1000 particles/cell) in the presence or absence of inhibitors. 6 h post addition, coverslips were mounted in custom made chambers, and the amount of fluorescent viruses/cell was analyzed in live cells on a Zeiss Observer Z1 microscope equipped with a Yokogawa spinning disc confocal unit and a heating chamber surrounding the microscope by acquiring Z stacks. After the initial acquisition, trypan blue (0.4% (w/v), Invitrogen) was added to live cells in a 1∶50 dilution. This immediately shifted the emission spectrum of viruses on the cell-surface and led to a loss of detectable fluorescence, whereas the fluorescence of intracellular viruses remained unaltered. Images were visually analyzed and representative examples were depicted as maximum intensity projections.
5×104 HeLa or HaCaT cells were seeded 24 h prior to experimentation in 12 well plates. HPV-16 PsV were added to cells in the cold. After 4 h the plates were shifted to 37°C. At different time points post warming, cells were incubated with high pH buffer (PBS, pH 10.5) for 1 min, which was replaced by normal growth medium. The treatment did not cause any cell toxicity (not shown) but resulted in disassembly of surface bound virions rendering them non-infectious (Figure S1). Alternatively, HeLa cells were infected in the presence or absence of inhibitors with HPV-16 to result in about 20% infection of the unperturbed control (10–30 particles/cell) At 12 h p.i., cells were incubated for 2 min with the high pH buffer, which was replaced with normal growth medium without inhbitor causing infection by already internalized virus. 48 h p.i., cells were trypsinized, fixed in 4% paraformaldehyde, and analyzed for infection (GFP expression) by flow cytometry.
For thin-section EM, HeLa or HaCaT cells plated onto 12-mm coverslips were incubated with HPV-16 PsV (10–20,000 particles/cell) 37°C for 30 min, 2, 6, 12, and 24 h before fixation with 2.5% glutaraldehyde (in 0.05 M sodium cacodylate pH 7.2, 50 mM KCl, 1.25 mM MgCl2, and 1.25 mM CaCl2, 30 min at RT), followed by 1.5 h of incubation in 2% OsO4 on ice. Dehydration, embedding, and thin sectioning were performed as previously described [58]. After transmission electron microscopy (Zeiss EM 91 microscope), images were exported as 8-bit TIFF files.
If not otherwise mentioned, all data is depicted ± standard deviation from at least three independent experiments.
|
10.1371/journal.pbio.1000116 | Nudel and FAK as Antagonizing Strength Modulators of Nascent Adhesions through Paxillin | Adhesion and detachment are coordinated critical steps during cell migration. Conceptually, efficient migration requires both effective stabilization of membrane protrusions at the leading edge via nascent adhesions and their successful persistence during retraction of the trailing side via disruption of focal adhesions. As nascent adhesions are much smaller in size than focal adhesions, they are expected to exhibit a stronger adhesivity in order to achieve the coordination between cell front and back. Here, we show that Nudel knockdown by interference RNA (RNAi) resulted in cell edge shrinkage due to poor adhesions of membrane protrusions. Nudel bound to paxillin, a scaffold protein of focal contacts, and colocalized with it in areas of active membrane protrusions, presumably at nascent adhesions. The Nudel-paxillin interaction was disrupted by focal adhesion kinase (FAK) in a paxillin-binding–dependent manner. Forced localization of Nudel in all focal contacts by fusing it to paxillin markedly strengthened their adhesivity, whereas overexpression of structurally activated FAK or any paxillin-binding FAK mutant lacking the N-terminal autoinhibitory domain caused cell edge shrinkage. These results suggest a novel mechanism for selective reinforcement of nascent adhesions via interplays of Nudel and FAK with paxillin to facilitate cell migration.
| Cell migration is an essential process in both single-cell and multicellular organisms. In higher animals, cell migration is important for many biological processes, including embryonic development, the immune response, and wound healing. Cancer cell invasion into healthy tissues occurs as a result of inappropriate cell migration. As can be easily visualized when cultured in the lab, mammalian cells attach to surfaces through focal adhesions, cellular structures characterized by complexes of the transmembrane protein integrin and intracellular proteins including paxillin and focal adhesion kinase (FAK). In order for cells to move, they must coordinate two processes: extension of the front edge of the cell and retraction of the back edge. To accomplish this, a cell first protrudes membranous structures from the front edge and then establishes adhesion structures known as nascent adhesions to hold the extensions in place. At the same time, the focal adhesions that hold a cell in place must be disrupted in order for the back edge of the cell to retract. Here, we show that a protein called Nudel is enriched at the front edge of moving cells, where it interacts with paxillin but is not detected in focal adhesions. We further show that the focal adhesion protein FAK is able to abolish the Nudel-paxillin interaction, leading to repression of the formation of nascent adhesions and to the loss of cell extensions. We therefore propose a model in which modulation of paxillin interactions in nascent adhesions and in focal adhesions is critical for coordinated cell movement: the Nudel-paxillin interaction enhances the strength of nascent adhesions to promote the attachment of membrane protrusions at the front edge of the cell, whereas FAK prevents the Nudel-paxillin interaction in focal adhesions in order to facilitate retraction of the back edge of the cell.
| In order to achieve efficient migration, cell adhesion and detachment must be properly coordinated. Cells attach to the substratum via punctate focal contacts (FCs). FCs contain integrin family members of transmembrane receptors and a variety of intracellular “adhesion” proteins and function to connect the extracellular matrix (ECM) to the actin cytoskeleton [1],[2]. During cell migration, membrane protrusions at the leading edge are triggered after activation of the Rho-family small GTPases Cdc42 and Rac1 [3]. Activated integrin dimers situated at the tip of protrusions then search for and bind to their ligands in the ECM to form nascent adhesions [4]. Nascent adhesions can mature into focal complexes (FXs), which are located mainly at the base of lamellipodium [5],[6]. FXs associate with branched F-actin and are thought to facilitate the propulsive effect of the lamellipodium. Some FXs then further evolve into the largest form of FC, namely focal adhesion (FA). FAs associate with the termini of F-actin bundles, or stress fibers, which provide cells with contractile forces [1],[6]–[8].
It is known that only moderate concentrations of the ECM are associated with maximal cell motility [9],[10]. Conceptually, fast migration would require efficient adhesion of leading-edge protrusions and rapid retraction of the trailing side [10],[11]. These two requirements could be satisfied if nascent adhesion sites exhibit stronger ECM-binding strengths than adhesion sites in FXs as well as FAs. Stronger adhesions at nascent sites would both promote the stabilization of membrane protrusions and facilitate persistency of the leading edge during cell retraction to allow efficient forward movement. In fact, tension on each contact site within FAs, which are caps of stress fibers [1],[7], is relatively constant in a cell [12]. Therefore, smaller FAs can only transmit weaker traction forces [12]. In contrast, compared to FAs, nascent adhesions, despite their submicroscopic sizes, have been shown to transmit stronger traction forces [13]. This would physically require a stronger integrin-ECM interaction at nascent adhesion sites than at adhesion sites of FXs and FAs. Whether mechanical strengths of different adhesion sites are indeed modulated and the underlying molecular mechanism(s), however, remain unclear.
FCs are dynamic structures. They are assembled through a hierarchical process. Paxillin and talin are believed to bind directly to integrin at adhesion sites [2]. Other proteins such as vinculin and focal adhesion kinase (FAK) are then recruited to form dot-like FXs, whereas FA formation is accompanied by the appearance of zyxin [6],[14]. FAK can be activated by multiple signaling pathways and is crucial for FC dynamics and membrane protrusion [2]. Its FA-targeting (FAT) domain, located at the C-terminus, interacts with talin and the LD domains of paxillin [2],[15]. In addition to assembly, FCs are subjected to dynamic disassembly as well [7]. Both nascent adhesion sites and FXs can be rapidly disassembled if they failed to evolve [6],[14]. FAs are relatively long-lived. Their disassembly often occurs at the trailing side of migrating cells. Moreover, FA formation can be promoted by internal and external tensions [12],[16]–[18]. Tensions on stress fibers can also lead to a net disassembly of distal adhesion sites and assembly of proximal sites, resulting in centripetal movement of FAs [19].
Mammalian Nudel (also named Ndel1) and Lis1 are essential for cell viability [20],[21] and for functions of the microtubule (MT)-based, minus end–directed motor cytoplasmic dynein in diverse processes including mitosis, neuronal migration, and intracellular transport [20],[22]–[27]. In addition, Nudel can also stabilize active Cdc42 by sequestering a negative regulator, Cdc42GAP, at the leading edge during migration of NIH3T3 cells [28]. Nudel confers homodimerization and Lis1 binding through its N-terminal coiled-coil region, whereas its C-terminus is able to interact with dynein heavy chain, Cdc42GAP, and other proteins [23],[26],[28]–[30].
In this report, we describe a novel mechanism we identified that regulates adhesivity of integrin-mediated adhesions. Our results indicate that Nudel selectively strengthens FC sites in nascent adhesions through a direct interaction with paxillin to facilitate stabilization of membrane protrusions at the leading edge, whereas structurally activated FAK can displace Nudel from paxillin in a kinase-independent manner, thus reducing the strength of FC sites in FXs and FAs to promote retraction of the trailing side.
We have previously shown that Nudel knockdown markedly inhibited pseudopodial formation in mouse fibroblast NIH3T3 cells [28]. To clarify whether this is solely related to defects in membrane protrusion, human epithelial ECV304 cells were chosen for analysis because they migrated with typical fan-shaped lamellipodia (Figure 1A; Videos S1 and S2). For convenient identification of live transfectants, the interference RNA (RNAi) constructs, pTER-Nudi for Nudel and pTER-Luci as a control [31], were modified to coexpress green fluorescent protein (GFP) or red fluorescent protein (RFP). As in NIH3T3 cells [28], Nudel RNAi in sparse ECV304 cells significantly repressed membrane protrusions and thus migration (Figures 1A, S1A, and S1B; Videos S1 and S2). Overexpression of Nudel with an RNAi-resistant construct (Nudel-R) rescued both lamellipodial formation and cell motilities (Figure S1C–S1E), thereby excluding a possible off-target effect of the RNAi construct.
Nudel RNAi has been shown to cause inactivation of Cdc42 [28], which could in turn repress Rac1 activity [32],[33]. If the lack of lamellipodia in Nudel-depleted cells (Figure 1A) was simply due to inhibition of Rac1, introduction of a constitutive active form of Rac1 (Rac1CA) should be able to fully restore lamellipodium formation [3],[34]. Consistent with a previous report [34], 76% of GFP-Rac1CA–positive cells cotransfected with pTER-Luci-RFP (n = 233) became flat and circular in shape, due to extensive formation and spreading of lamellipodia (Figure 1B, panels 1 and 2). In contrast, although 67% of pTER-Nudi-RFP transfectants overexpressing GFP-Rac1CA (n = 316) formed lamellipodia, as judged by the existence of F-actin–rich membrane ruffles, they failed to spread extensively (Figure 1B, panels 3 and 4, arrows). Quantitation also indicated that they generally exhibited obviously reduced circularity and area as compared to control cells (Luci+Rac1CA) (Figure 1C). To corroborate these results, we applied a dominant-negative Cdc42 (Cdc42DN) to repress Cdc42 activity (unpublished data) [35] and found that as expected, its overexpression failed to repress cell spreading stimulated by Rac1CA (Figure 1B and 1C). Therefore, the spreading defect associated with Nudel depletion is not solely due to inhibition of Cdc42 and Rac1.
We then performed time-lapse microscopy to examine why Nudel-depleted cells failed to fully spread even in the presence of Rac1CA. The control transfectants, which were much larger in size than surrounding untransfected cells, showed vigorous membrane ruffling at cell edges (Figure 1D; Video S3) [34]. In contrast, although GFP-Rac1CA induced active membrane protrusions in Nudel-depleted cells (Figure 1E vs. 1A), the protrusions were not persistent and usually retracted back within a few minutes (Figure 1E; Video S4), indicating lack of stable attachment to the substratum. As a result, the cells failed to spread even when monitored for more than 500 min (Figure 1E; Video S4).
We further excluded the possibility that Nudel RNAi repressed lamellipodial formation through inhibition of dynein because NudelC36, a deletion mutant whose overexpression inhibits dynein [22],[23], failed to affect ECV304 cell migration (Figure S2A and S2B). Normal lamellipodial formation was seen as well in cells overexpressing either GFP-tagged NudelC36 or another dynein inhibitor, p50dynamitin (Figure S2C) [36],[37].
Taken together, these results strongly suggest a critical role of Nudel in stable attachment of nascent membrane protrusions to the substratum. Importantly, such a role is distinct from the previous ones in regulation of Cdc42 and dynein [28], therefore defining a novel function of Nudel in cell migration.
We then examined detailed distributions of FCs and F-actin in ECV304 cells with Nudel knockdown. Indeed, compared to the typical arc-like lamellipodial formation in most sparse transfectants of pTER-Luci-GFP (71.0%; n = 356) (Figure 2A, panels 1 and 2), transfection with pTER-Nudi-GFP resulted in severe cell edge shrinkage in both subconfluent cells (63.1%; n = 388) and confluent cells scratched to induce migration [38],[39] (Figures 2A, panels 3 and 4, and S3A). Moreover, robust FAs and stress fibers at the cell periphery were seen (Figures 2A, panels 3 and 4, and S3A). Similar phenotypes were also observed in HeLa cells, independent of cell densities (Figure S3B).
The FAs/stress fibers can develop in response to forces provided either intrinsically through contraction of myosin on stress fibers or externally by mechanical strains [12],[17],[18],[40]. To better understand the phenotypes of Nudel RNAi, we disrupted the intrinsic contractile forces using blebbistatin, a small-molecule inhibitor of myosin II ATPase activity [41]. After blebbistatin treatment for 45 min, FAs and stress fibers were mostly disassembled in control cells, as expected (Figure 2B) [5],[41]. Nevertheless, they were still largely preserved in Nudel RNAi cells (Figure 2B), suggesting that the robust FAs/stress fibers in Nudel-depleted cells (Figures 2A and S3) were formed in response to tensions from the collapsing cell edges in order to resist further shrinkage, instead of from the contractile forces of myosin II.
To understand why cell edges tended to shrink upon Nudel RNAi, we examined FCs in Nudel-depleted cells overexpressing Rac1CA. In control cells, Rac1CA induced typical FX around the entire cell periphery (Figure 2C and 2D) [34]. In contrast, although FXs were readily observed in pTER-Nudi-RFP transfectants overexpressing Rac1CA, they only appeared in less than half of the cell periphery in approximately 82% of cells (Figure 2C and 2D), indicating a markedly reduced efficiency of FX formation. We then treated such cells with blebbistatin for 25 min to block maturation of their nascent adhesions into FXs [5]. In contrast to the appearance of a rim of tiny, dense nascent adhesions within the lamellipodium in control cells (Figure 2E) [5], Nudel RNAi cells overexpressing Rac1CA showed little accumulation of nascent adhesions around the cell periphery (Figure 2E), though vigorous membrane protrusions still occurred (Figure 2E) as in untreated cells (Figure 1B, panels 3 and 4, and 1E). Therefore, the negative effect of Nudel RNAi on stabilization of membrane protrusions (Figures 1 and 2A) is attributed to poor formation of nascent adhesions.
To understand how Nudel could affect nascent adhesions, we performed a screen for its partner(s) in FCs. FLAG-Nudel coexpressed with a GFP-tagged FC protein such as vinculin, paxillin, or FAK was subjected to coimmunoprecipitation (co-IP). Ponceau S staining revealed GFP-paxillin as the major protein associated with FLAG-Nudel (Figure 3A and 3B, lane 6), strongly suggesting a direct interaction. GFP-paxillin was also associated with FLAG-NudelN20, a mutant lacking Lis1-binding activity [22], but not with FLAG-NudelC36 (Figure 3A and 3B, lanes 7 and 8). The failure of NudelC36 to interact with paxillin was also consistent with the results that, unlike the wild-type Nudel (Figure S1D and S1E), NudelC36 overexpressed from an RNAi-resistant construct failed to restore the motility of Nudel-depleted cells (Figure S2A and S2B). FLAG-Nudel was able to associate with Tyr/Ser/Thr-phosphorylated isoforms important for physiological functions of paxillin (Figure 3C) [42]–[44], further suggesting a functional interplay between the two proteins.
To confirm their direct interaction, GST-paxillin and FLAG-Nudel were expressed in Escherichia coli. Glutathione S-transferase (GST)-pulldown assays indeed indicated their interaction (Figure 3D, lane 6). Moreover, when paxillin mutants containing either the LD domains or the LIM domains (Figure S4A) [15] were assayed, only PaxLIM interacted with Nudel (Figure 3D, lanes 8). Reciprocal experiments also support a direct Nudel-paxillin interaction (Figure S4B). In contrast, vinculin, a paxillin-associated FC protein [15], failed to bind directly to Nudel (Figure S4B). As paxillin exists in all types of FCs and is a scaffold/adaptor protein critical for cell migration [6],[15], it may serve as the target of Nudel in cell adhesion. Consistently, FLAG-Nudel formed a complex with endogenous paxillin and vinculin (see below).
We then examined localization of Nudel and paxillin in ECV304 cells migrating into an artificial “wound” [38],[39]. As in NIH3T3 cells [28], Nudel was enriched at the leading edge, and colocalized with paxillin there (Figure 3E, arrowheads). Moreover, both proteins were enriched in areas of cell protrusions, indicated by the presence of active actin polymerization (Figure 3E) [4],[45]. In contrast, Nudel did not show colocalization with the paxillin puncta, which typically represent FXs and FAs (Figure 3E) [46]. Quantitation analyses also indicated a significant correlation between Nudel and paxillin at the leading edge (Figure 3F–3H). These results imply interaction of both proteins in early stages of FC formation and are consistent with the role of Nudel in nascent membrane adhesion (Figures 1 and 2).
We then tried to assess whether the Nudel-paxillin interaction indeed contributed positively to nascent cell adhesion. As integrin-mediated nascent adhesion sites are submicroscopic structures and only represented a portion of total adhesion sites (Figures 2 and 3) [7], direct assays on them would not be feasible. We thus reasoned that a fusion protein, paxillin-GFP-Nudel (PGN), would make all adhesion sites Nudel-containing, thus allowing convenient examination of Nudel's effect on adhesion. Although such a construct is somewhat artificial, a similar strategy has been successfully used in other studies [47]. Similar to Pax-GFP (Figure 4A and 4B) [48], PGN was also located in FCs (Figures 4A, 4B, S5A, and S5B). Moreover, PGN still bound to Lis1 (Figure S5C), a protein associated with the N-terminal portion of Nudel [30]. Therefore, both paxillin and Nudel in the fusion protein are still functional.
To verify whether PGN stabilized the cell–substratum adhesion, we first examined FA motilities [19], which may reflect the stability of individual adhesion sites of FAs against tension. For easy comparison, image sequences at 0, 10, and 20 min were pseudocolored red, green, and blue, respectively, and merged. Motile FAs would thus display rainbow colors, whereas nonmotile ones would be white [19]. Upon overexpression of Pax-GFP, FAs in both nonmotile (Figure 4A) and motile cells (Figure 4B) exhibited similar active centripetal movement, as judged by the appearance and orientation of rainbow colors. The average velocity was 0.0434 µm/min (Figure 4C), about 3-fold lower than that of 3T3 fibroblasts [19]. It should be noted that Smilenov and colleagues [19] considered cells just after division as “migrating” cells and defined the remaining population as “stationary” cells. Therefore, the population analyzed herein is equivalent to the “stationary” population in the previous study [19]. Just as in ECV304 cells (Figure 1), this population of fibroblasts is in fact not truly stationary [28].
In cells overexpressing PGN, FA motilities were largely reduced, as judged by the obvious appearance of white color (Figure 4A and 4B). The average velocity of FAs was reduced by approximately 3-fold (0.0146 µm/min) as compared to that in cells overexpressing Pax-GFP (Figure 4C). Moreover, as FCs close to the cell edges where membranes are dynamic showed obvious turnover in PGN-positive cells as well (Figure 4B), the reduced FA motility is unlikely due to defects in FA disassembly. Rather, it suggests an increased strength of the FC sites.
To further corroborate the above results, we investigated whether PGN-positive cells exhibited enhanced adhesion on a laminin- and fibronectin-coated surface against different shear forces [49]. HEK 293T cells were used instead of ECV304 because the latter cells tended to aggregate, thus precluding sorting with fluorescence-activated cell sorter (FACS) to eliminate untransfected cells. At the wall shear stress of 1 and 2 dyne/cm2, PGN-expressing cells accumulated more rapidly than Pax-GFP–positive cells (Figure 4D). They maintained a higher resistance to the increasing shear stress from 2 to 16 dyne/cm2 (Figure 4D).
Taken together, these results strongly suggest that the Nudel-paxillin association can enhance adhesion strength of FC sites. This further explains why Nudel is critical for stabilization of nascent membrane protrusions (Figures 1 and 2).
As Nudel was not seen in either FXs or FAs (Figure 3E), we speculated that it might be displaced by certain paxillin-binding protein(s) that are recruited during the maturation of nascent adhesion sites [6]. Indeed, co-IP results indicated that overexpression of GFP-tagged FAK, but not vinculin, both of which are paxillin-binding FC proteins [15], abolished the interaction between GFP-paxillin and FLAG-Nudel (Figure 5A and 5B, lanes 1 and 2 vs. 9 and 10). FAK mutants lacking the paxillin-binding FAT domain [2], e.g., FAKΔFAT and FAKKin, were completely ineffective (Figure 5A and 5B, lane 7 vs. 15; Figure S6A and S6B, lane 1 vs. 4). In contrast, mutants containing FAT or even FAT alone, e.g., FAKΔFERM, FAKFAT, and FRNK, a naturally occurring FAK isoform [50], were all potent in disrupting the Nudel-paxillin interaction (Figure 5A and 5B, lanes 4 and 8 vs. 12 and 16; Figure S6A and S6B, lanes 2 vs. 5). When the paxillin-interacting ability of FAKΔFERM was abolished by point mutations V954A/L961A [51] or I936E/I998E [52], the resultant mutants, FAKΔFMPX1 and FAKΔFMPX2, failed to disrupt the Nudel-paxillin interaction (Figure 5A and 5B, lanes 5 and 6 vs. 13 and 14). In contrast, the kinase-dead mutant FAKKD carrying a K454R mutation [53] still showed strong activity in competing for paxillin binding (Figure 5A and 5B, lane 3 vs. 11). Therefore, FAK can, in a kinase-independent manner, disrupt the Nudel-paxillin interaction through its physical association with paxillin.
To understand how FAK abolishes the Nudel-paxillin interaction, GFP-tagged PaxLD and PaxLIM (Figure S4A) [15] were tested for their ability to bind Nudel in the presence of GFP-FAKΔFAT or FAKΔFERM (Figure 5C). As expected (Figure 5A and 5B), the Nudel-paxillin interaction was not affected by FAKΔFAT but was disrupted by FAKΔFERM (Figure 5C, lanes 3 and 4 vs. 9 and 10). PaxLIM associated with Nudel in both cases (Figure 5C, lanes 1 and 2 vs. 7 and 8), whereas PaxLD failed to do so in either case (Figure 5C, lanes 5 and 6 vs. 11 and 12). These data further confirm that Nudel interacts with paxillin via the LIM domains (Figure 3D). Moreover, competition by FAK is mediated through its direct interaction with the LD domains of paxillin [2],[15].
To further investigate whether FAK indeed regulates the Nudel-paxillin interaction at physiological conditions, we performed co-IP experiments to check whether a decrease in endogenous FAK levels could affect the Nudel-paxillin interaction. As we were not able to detect endogenous paxillin in co-IP experiments using anti-Nudel IgY, possibly due to a steric effect of the antibody, we overexpressed in HEK293T cells low levels of FLAG-GFP-Nudel (at 3–6-fold of endogenous Nudel level) through the internal ribosome entry site (IRES) and performed co-IP assays with anti-FLAG resin (Figure 5D). The levels of endogenous FAK were reduced sequentially through transfection of increasing amounts of the RNAi plasmids (Figure 5D, lanes 1–6). Indeed, association of endogenous paxillin with FLAG-GFP-Nudel was markedly enhanced following the reduction of endogenous FAK levels (Figure 5D, lanes 7–12). Association of vinculin was also detected, with its levels paralleling those of paxillin (Figure 5D, lanes 7–12). Therefore, endogenous FAK can regulate the Nudel-paxillin interaction as well.
If the Nudel-paxillin interaction was indeed critical for nascent membrane adhesion, according to the above results (Figure 5A and 5B), overexpressing FAK or any FAK mutant that binds paxillin should displace Nudel prematurely from nascent adhesion sites and consequently impair cell spreading. Indeed, overexpression of any FAT-containing deletion mutant, i.e., FAKΔFERM, FRNK, or even FAKFAT, resulted in high incidences (≥72%) of cell shrinkage: affected cells usually lacked FXs and were typically polygonal in shape, with cell edges supported by F-actin bundles and FAs (Figures 6A, 6B, and S6C). In contrast, mutants lacking the FAT domain, e.g., FAKΔFAT and FAKKin, or containing FAT but lacking paxillin-binding activity, e.g., FAKΔFMPX1 and FAKΔFMPX2, only generated background levels of shrunken cells (Figures 6A, 6B, and S6C). Moreover, although FAKΔFAT and FAKKin failed to show FA localization, FAKΔFMPX1 was still efficiently targeted to FAs (Figures 6A and S6C) [51]. FAKΔFMPX2 exhibited weak, but clear, FA localization as well (Figure S6C). Because FAKΔFMPX1 and FAKΔFMPX2 do not bind to paxillin, their localization to FA is probably mediated by talin [2],[51],[52],[54]. Thus, the paxillin-binding activity of FAK is essential for the induction of cell edge shrinkage, whereas its kinase domain is dispensable.
In contrast to the intact FAT-containing mutants, full-length FAK only showed a mild effect. Although GFP-FAK–positive cells with the shrinkage phenotypes were approximately 2-fold higher in percentages than surrounding untransfected cells, the majority of cells overexpressing GFP-FAK (66.4% in average) showed normal lamellipodia (Figure 6A and 6B). The kinase-dead mutant GFP-FAKKD had a similar effect (Figure 6A and 6B), whereas FAKOpn, a mutant containing two point mutations (Y180A/M183A) that abrogate the autoinhibitory effect of the FERM domain [55] potently induced cell edge shrinkage upon its overexpression (Figure 6A and 6B). Similar phenotypes were seen in ECV304 cells grown on fibronectin- and/or laminin-coated substratum as well as in CV1 and NIH3T3 cells (Figure S7; unpublished data). Therefore, the “open” structure of full-length FAK is important for both full activation of FAK [55] and induction of cell edge shrinkage.
FAK is believed to promote cell migration through its kinase activity [2]. To understand why both FAKOpn, which exhibits robust kinase activity in cells (unpublished data) [55], and FRNK, which is a dominant-negative mutant on kinase activity of endogenous FAK [50], caused similar cell shrinkage phenotypes (Figure 6A and 6B), we monitored behaviors of live cells. Consistent with results in fixed cells (Figure 6A and 6B), ECV304 cells overexpressing GFP-FAKOpn were narrow or polygonal in shape (Figure 6C; Video S5). Whereas surrounding untransfected cells migrated through typical arc-like lamellipodia, these transfectants extended long processes rich in transient filopodium-like projections (n = 19/20) and migrated like fibroblasts (Figure 6C and 6D; Video S5) [28]. In contrast, cells overexpressing GFP-FRNK showed markedly reduced motilities (Figures 6C and S6D; Video S6). Such cells also failed to show active membrane protrusions (n = 21/21) (Figure 6D), consistent with the lack of FAK kinase activity. Therefore, although cells overexpressing FAKOpn or FRNK showed different motilities, they share similar shrinkage phenotypes.
These results identify a novel kinase-independent role of FAK in cell spreading. As this role of FAK depends on its interaction with paxillin, the “shrunken” phenotype caused by FAK overexpression is attributed to poor adhesions of nascent membrane protrusions due to premature disruption of the Nudel-paxillin interaction at the leading edge.
We have previously shown that Nudel is required for membrane protrusions in NIH3T3 cells [28]. Here, we further showed that Nudel depletion markedly repressed lamellipodial formation in ECV304 cells (Figure 1A; Video S2), indicating a general requirement of Nudel in membrane protrusion. Lamellipodial formation requires Rac activity, whereas Cdc42 can activate Rac [3],[32],[34]. Therefore, the protrusion defect upon Nudel depletion is consistent with inhibition of Cdc42 activity [28]. By contrast, although Nudel is also essential for dynein functions [22],[23], dynein activity is not important for lamellipodial formation and free migration of ECV304 cells (Figure S2).
Nudel is also critical for stabilization of membrane protrusions by facilitating nascent adhesion formation. First, Nudel depletion by RNAi primarily resulted in cell edge collapse (Figures 2A and S3). As the robust stress fibers in Nudel RNAi cells were not sensitive to blebbistatin treatment (Figure 2B), their formation is unlikely due to increased contractile forces on stress fibers, e.g., through activation of Rho GTPase [34],[40]. Rather, it is attributed to mechanical strains caused by cell edge shrinkage because stress fibers induced by mechanical forces do not depend on myosin II activity [18],[40],
Second, although overexpression of Rac1CA rescued the membrane protrusion defect of Nudel depletion, cells still failed to fully spread due to poor adhesions of their protruded membranes (Figure 1B and 1E; Video S4). Importantly, this phenotype is not caused by the inactivation of Cdc42 per se, as coexpression of a dominant-negative form of Cdc42 with Rac1CA failed to repress cell spreading (Figure 1B and 1C).
Third, the markedly reduced formation of FXs as well as nascent adhesions upon Nudel depletion even in the presence of Rac1CA (Figure 2C and 2E) further indicates a positive role of Nudel in nascent adhesions.
The Nudel-paxillin interaction further substantiated the role of Nudel in nascent adhesion because paxillin can bind directly to integrin and is thus one of the earliest intracellular proteins at nascent adhesions [2],[5],[6]. As the interaction between exogenous Nudel and paxillin was readily detected even by Ponceau S staining after co-IP (Figure 3A), these two proteins appear to interact with high affinity. Moreover, they interacted directly through their C-terminal domains (Figures 3 and S4). The interaction between endogenous paxillin and FLAG-Nudel expressed at a relative low level can also be detected in vivo, especially upon knockdown of FAK expression (Figure 5D).
Our results suggest that Nudel interacts with paxillin in nascent adhesions. Complex formation of Nudel with both paxillin and vinculin (Figures 5B, lane 9, and 5D) suggests its localization in certain FCs. Nevertheless, it was not detected in FXs or FAs, but enriched and colocalized with paxillin at the leading edge in areas of active membrane protrusions (Figure 3E–3H), where nascent adhesions occur [4],[5]. Moreover, a localization of Nudel in nascent adhesions is also consistent with its functions there (Figures 1 and 2).
We provided evidence showing that the presence of Nudel in FCs can indeed stabilize integrin-ECM interactions using PGN (Pax-GFP-Nudel) (Figures 4 and S5). PGN was specifically located in FCs (Figure S5B) and reduced FA motilities by approximately 3-fold, compared to Pax-GFP (Figure 4A–4C), suggesting elevated stability, or strength, of individual adhesion site. Furthermore, the elevated adhesiveness of PGN-positive cells over Pax-GFP-positive ones, measured through their resistance to shear forces (Figure 4D) [49], further supports the increase in adhesion strengths. Although PGN is an artificial protein and may not precisely reflect situations in vivo, its effects on FA motility and cell adhesion are well in agreement with other results that suggest a role of Nudel in stabilization of nascent adhesions (Figures 1–3). Consistently, paxillin-deficient cells exhibit a delayed rate of spreading [56]. Nevertheless, it is currently not known whether the Nudel-paxillin interaction stabilizes integrin-ECM ligations by modulating the integrin conformation or by regulating other intracellular adhesion molecules.
We demonstrated that FAK is a key regulator of the Nudel-paxillin interaction. FAK was able to disrupt the interaction via direct binding to paxillin (Figure 5). Such a competition effect may be mediated through steric hindrance. Alternatively, given that the FAT domain alone, which covers only one-eighth of FAK, was already sufficient to disrupt the Nudel-paxillin interaction (Figure 5A and 5B), FAK binding may induce in paxillin a conformational change that abrogates Nudel binding. That FAK and Nudel bound to distinct regions of paxillin (Figures 3D and 5C) [15] also supports the latter speculation.
In addition to its known kinase-dependent functions in cell migration [2], we found that FAK can negatively regulate nascent adhesions. Overexpression of FAK resulted in an approximately 2-fold increase in incidence of cells with shrunken edges comparing to surrounding untransfected populations (Figure 6A and 6B). Deleting the FERM domain (e.g., FAKΔFERM) or abolishing its autoinhibitory role through point mutations (e.g., FAKOpn) [55] considerably augmented incidences of the shrunken phenotype (Figure 6A and 6B). Such a phenotype, however, is not correlated with the kinase activity of FAK because it is similar in cells overexpressing either the hyperactive (e.g., FAKOpn and FAKΔFERM) [55] or the dominant-negative (e.g., FRNK and possibly FAKFAT) [50] mutants (Figures 6A, 6B, and S6C). In addition, in the absence of the FERM domain, the potency of FAK to induce cell edge collapse is only correlated with its interaction with paxillin (Figures 6A, 6B, and S6C). Localization of FAK in FCs, however, is not sufficient: the point mutants FAKΔFMPX1 and FAKΔFMPX2 localized to FAs but failed to cause the shrunken phenotype (Figures 6A and S6C).
We therefore propose a model to explain how paxillin, Nudel, and FAK cooperate to modulate integrin-mediated adhesivity in cell migration (Figure 7): during membrane protrusion, activated integrin molecules located on polymerizing F-actin [4] bind to ECM to form nascent adhesion sites containing paxillin [6]; association of Nudel with paxillin strengthens such sites; upon formation of the open conformation in response to external signals, possibly through interaction of integrin and/or growth factor receptors [55], FAK displaces Nudel from paxillin; adhesion sites now exhibit a lower strength than those containing Nudel.
The antagonizing roles of Nudel and FAK in adhesivity provide a mechanism for cells to properly coordinate adhesion and migration. The positive effect of Nudel on adhesion strength can stabilize nascent adhesion sites and thus facilitate stabilization of membrane protrusions at the leading edge. Stronger adhesiveness would also allow nascent sites to transmit stronger traction forces [13] and to resist retraction. On the other hand, because FAs are large in size, a decreased strength of their individual FC sites would facilitate FA movement (Figure 4) [19] and retraction of the trailing side.
Our findings also help to understand how cells orchestrate different events in migration. As formation of the open structure of FAK depends on upstream signals and serves as a prerequisite for activation of the kinase [55], disruption of the Nudel-paxillin interaction, thus down-regulation of adhesivity at nascent adhesions, is likely to precede other events associated with the kinase activity of FAK [2]. Such an ordered sequence of action appears important for cell migration because premature disruption of the Nudel-paxillin interaction and/or interference with the kinase activity of FAK affects cell motility. For instances, although excess wild-type FAK failed to interfere with lamellipodial formation in the majority of cells (Figure 6A and 6B), overexpression of FAKOpn to prematurely disrupt the Nudel-paxillin interaction (Figures 5, 6, and S6) while provoking a hyperactive kinase activity [55], impaired the arc-like lamellipodium formation in ECV304 cells and resulted in cell migration through transient filopodium-like membrane projections (Figure 6; Video S5). In contrast, overexpression of FRNK to similarly abrogate the Nudel-paxillin interaction (Figure S6) while also inhibiting endogenous FAK activity [50],[57] caused cell shrinkage but poor migration (Figures 6 and S6; Video S6) [57]. Furthermore, FAK-null cells have been shown to exhibit robust FC formation at the cell periphery [58],[59], reminiscent of enhanced cell edge adhesions to the substratum. These cells also show poor migration [58],[59].
We have previously shown that Nudel can stabilize active Cdc42 at the leading edge by sequestering Cdc42GAP in NIH3T3 cells [28]. Nudel also contributes to dynein functions at the leading edge [28],[60]. Moreover, similar to paxillin (Figure 3), both Cdc42GAP and dynein heavy chain bind to the C-terminus of Nudel [22],[28]. How these functions of Nudel are coordinated is not yet clear. One possibility is that Nudel interacts with different partners for different functions or in different cell types. Another possibility is that these partners use Nudel as a common platform to achieve orchestration of different functions. Interestingly, in co-IP experiments, we found that associations of Cdc42GAP, paxillin, and dynein with Nudel were significantly enhanced upon overexpression of both paxillin and Cdc42GAP (Figure S8). If such a synergetic effect on Nudel binding occurred at the leading edge due to enrichment of these proteins there (Figure 3) [28],[60], the Nudel-paxillin interaction and the regional activation of Cdc42 and dynein would become spatiotemporally coupled events to eventually facilitate establishment of a polarized lamellipodium. These issues will be worthy of future investigations.
Expression plasmids for human Nudel, its mutants, and p50dynamitin were described previously [22],[28],[61]. pLV-IRES-FLAG-GFP and pLV-IRES-FLAG-GFP-Nudel were constructed from a lentiviral vector (a gift from Qiwei Zhai, Institute of Nutritional Science, Shanghai Institutes for Biological Sciences [SIBS]) for low-level expression of FLAG fusion proteins via the internal ribosome entry site (IRES). pTER-Nudi, a Nudel RNAi construct, and a control construct pTER-Luci [31] were further modified to coexpress GFP or RFP. The RNAi-resistant Nudel constructs contained three silent mutations in the short hairpin RNA (shRNA)-target region. The expressed proteins, despite unchanged amino acid sequences, were named Nudel-R and NudelC36-R sheer for presentation purposes. To silence FAK expression, pTER-FAKi1 and pTER-FAKi2 were constructed and cotransfected at a 1∶1 ratio. Their targeting sequences are 5′GGTACTGGTATGGAACGTTCT3′ and 5′GCCTTAACAATGCGTCAGT3′, respectively. Expression plasmid for GFP-vinculin was kindly provided by Benjamin Geiger (Weizmann Institute, Israel). pGFP-hPaxillin and pVSV-mFAK/FRNK were gifts from Kenneth M.Yamada (National Institute of Dental and Craniofacial Research [NIDCR], National Institutes of Health [NIH]). To express fusion proteins paxillin-GFP or Paxillin-GFP-Nudel, the coding sequence of paxillin was amplified by PCR and inserted in-frame between the NheI and AgeI sites of pEGFP-C1 or pEGFP-C1-Nudel. FAK and paxillin mutants were created by PCR as well. Plasmids for expression of GFP-tagged Rac1, FLAG-tagged Cdc42, and mutants were from Xiaobing Yuan (Institute of Neuroscience, SIBS) and Michiyuki Matsuda (Osaka University, Japan). Plasmids containing PCR fragments were subjected to sequencing confirmation.
Mouse monoclonal antibodies (mAbs) to α-tubulin, vinculin, FLAG, and phospho-Tyr, and rabbit antibodies to FAK and FLAG were purchased from Sigma-Aldrich. mAbs to paxillin and phospho-Ser/Thr were from BD Biosciences Transduction Laboratories. Rabbit antibody to GFP was from Santa Cruz Biotechnology. Anti-GST mAb was from Wolwo Biotech. Anti-Nudel IgY was generated from chicken and affinity-purified [23]. Secondary antibodies conjugated with peroxidase or Alexa Fluor-405, -488, -546, or -647 were purchased from Invitrogen. Phalloidin-Alexa-647 was from Invitrogen. Phalloidin-TRITC and blebbistatin were from Sigma-Aldrich.
All cells were cultured at 37°C and 5% CO2 in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% (v/v) bovine serum (Sijiqing). Human embryonic kidney (HEK) 293T cells were transfected by using the conventional calcium phosphate method. This cell line was used for assays involving immunoprecipitation due to its high transfection efficiency. Human bladder epithelial ECV304 [62] and cervical carcinoma HeLa cells were transfected with Lipofectamine2000 (Invitrogen). In overexpression experiments, cells were harvested approximately 48 h posttransfection for biochemical assays or fixed approximately 20 h posttransfection for microscopy. In RNAi experiments, they were used 48–72 h posttransfection. To determine RNAi efficiency in ECV304, GFP-positive transfectants were enriched to approximately 90% by using a BD FACSAria cell sorter 48 h after transfection. To prevent cell aggregation, Latrunculin A (0.1 µg/ml; Invitrogen) was added prior to sorting. Transfectants were cultured for an additional 24 h and then collected for immunoblotting (IB).
Approximately 1×107 HEK293T cells were lysed in co-IP buffer (20 mM Tris-HCl [pH 7.5], 100 mM KCl, 0.1% NP-40, 1 mM EDTA, 10% glycerol, 1 mM DTT, 50 mM NaF, 10 mM Na-pyrophosphate, 1 mM Na-Vanadate, and protease inhibitors cocktail [Calbiochem]) by repetitive pipetting through a 1-ml tip. After centrifugation at 10,000 g for 10 min to remove debris, lysates were incubated with anti-FLAG M2 agarose beads (Sigma) for 2 h on a rotator at 4 °C. The beads were then washed with the buffer for three times, followed by elution with synthetic FLAG peptide [63]. For pull-down assays, bacterial lysates containing GST fusion proteins or FLAG-Nudel were premixed for 1 h and then incubated with glutathione or anti-FLAG agarose beads (Sigma-Aldrich) for another 1 h at 4°C with agitation. Proteins binding to the beads were then boiled in SDS-sample buffer and subjected to IB. When necessary, membranes were stripped and blotted with different antibodies. Experiments were repeated at least three times.
Unless indicated, cells were grown sparsely on sterile glass coverslips without pre-coating of ECM. They were fixed with 4% paraformaldehyde (Sigma-Aldrich) for 15 min, followed by permeabilization with 0.5% Triton X-100 (v/v) for 10 min. For scratch wound assays, confluent cell monolayers cultured in serum-free medium for 12 h were scratched with yellow tips [28] and then cultured in serum-containing medium for an additional 3 h prior to fixation.
Immunofluorescence staining was performed with appropriate combinations of antibodies. F-actin was decorated with fluorochrome-labeled phalloidin. Images were captured with a Leica TCS SP2 laser-scanning confocal microscope. Grayscale images were converted to pseudocolor using Adobe Photoshop. Statistical data were presented as mean±standard deviation (SD) from at least three experiments. Cell area and circularity (4π×area/perimeter2) were measured using ImageJ (NIH). To quantify fluorescent colocalizations along the leading edge, intensity profiles were obtained using ImageJ. Cross-correlations and Pearson correlation coefficients of the intensity profiles were calculated with Matlab (MathWorks) [4].
ECV304 cells were cultured in L-15 medium (Invitrogen) supplemented with 10% (v/v) bovine serum. Image sequences for cell migration were collected by using an Olympus IX81 microscope with 37 °C-incubation chamber, motorized stage, and Evolution QEi CCD camera (Media Cybernetics), or a Leica AS MDW workstation with a heating hood and a CoolSNAP HQ CCD camera (Roper Scientific) [28],[64]. For FA motility assays, cells were imaged by using an Olympus FluoView 1000 inverted confocal microscope with a heating stage at 5-min intervals. ImageJ (NIH) was used for measurement. Migration tracks were determined as tracks of nuclei [28]. Average velocity of a sparse cell was calculated using its track length of free migration.
Flow chamber assays were performed basically as described [49]. A polystyrene Petri dish coated with purified human laminin and fibronectin (12.5 µg/ml each; Sigma-Aldrich) was used as the lower wall of the chamber. HEK293T transfectants were trypsinized and sorted. GFP-positive cells were diluted to 1×106/ml in complete culture medium and infused into the flow chamber immediately. Cells were allowed to accumulate for 30 s at 0.3 dyne/cm2 and for 10 s at 0.4 dyne/cm2. Shear stress was then increased every 10 s from 1 dyne/cm2 up to 32 dyne/cm2 in 2-fold increments. The number of cells remaining bound at the end of each 10-s interval was counted.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.