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.0002683 | Evaluation in Mice of a Conjugate Vaccine for Cholera Made from Vibrio cholerae O1 (Ogawa) O-Specific Polysaccharide | Protective immunity against cholera is serogroup specific. Serogroup specificity in Vibrio cholerae is determined by the O-specific polysaccharide (OSP) of lipopolysaccharide (LPS). Generally, polysaccharides are poorly immunogenic, especially in young children.
Here we report the evaluation in mice of a conjugate vaccine for cholera (OSP:TThc) made from V. cholerae O1 Ogawa O-Specific Polysaccharide–core (OSP) and recombinant tetanus toxoid heavy chain fragment (TThc). We immunized mice intramuscularly on days 0, 21, and 42 with OSP:TThc or OSP only, with or without dmLT, a non-toxigenic immunoadjuvant derived from heat labile toxin of Escherichia coli.
We detected significant serum IgG antibody responses targeting OSP following a single immunization in mice receiving OSP:TThc with or without adjuvant. Anti-LPS IgG responses were detected following a second immunization in these cohorts. No anti-OSP or anti-LPS IgG responses were detected at any time in animals receiving un-conjugated OSP with or without immunoadjuvant, and in animals receiving immunoadjuvant alone. Responses were highest following immunization with adjuvant. Serum anti-OSP IgM responses were detected in mice receiving OSP:TThc with or without immunoadjuvant, and in mice receiving unconjugated OSP. Serum anti-LPS IgM and vibriocidal responses were detected in all vaccine cohorts except in mice receiving immunoadjuvant alone. No significant IgA anti-OSP or anti-LPS responses developed in any group. Administration of OSP:TThc and adjuvant also induced memory B cell responses targeting OSP and resulted in 95% protective efficacy in a mouse lethality cholera challenge model.
We describe a protectively immunogenic cholera conjugate in mice. Development of a cholera conjugate vaccine could assist in inducing long-term protective immunity, especially in young children who respond poorly to polysaccharide antigens.
| Cholera is a severe dehydrating diarrheal illness of humans caused by organisms Vibrio cholerae serogroups O1 or O139 serogroup organisms. Protective immunity against cholera is serogroup specific. Serogroup specificity in V. cholerae is determined by the O-specific polysaccharide (OSP) of lipopolysaccharide (LPS). Generally, polysaccharides are poorly immunogenic, especially in young children. Unfortunately, children bear a large burden of cholera globally. Here we describe a novel cholera conjugate vaccine and show that it induces immune responses in mice, including memory responses, to OSP, the T cell-independent antigen that probably is the target of protective immunity to cholera. These responses were highest following immunization of the vaccine with a novel immunoadjuvant, dmLT. We also show that immunization of mice with this conjugate vaccine protects against challenge with wild-type V. cholerae. A protectively immunogenic cholera conjugate vaccine that induces long-term memory responses could have particular utility in young children who are most at risk of cholera.
| Cholera is a severe dehydrating diarrheal illness of humans caused by organisms Vibrio cholerae O1 or O139 serogroup organisms. V. cholerae O139 has largely disappeared and is reported from just a few Asian countries [1]. Cholera affects 3–5 million people each year, killing ∼100,000 annually, and cholera is endemic in over 50 countries [2]. V. cholerae O1 can be distinguished genotypically and phenotypically into classical and El Tor biotypes [2] and Ogawa and Inaba serotypes. Ogawa differs from Inaba only by the presence of a 2-O-methyl group in the non-reducing terminal sugar of O-specific polysaccharide (OSP) [3]–[5]. Currently, the global cholera pandemic is caused by organisms V. cholerae O1, El Tor, organisms, with the prevalent serotype fluctuating during cholera outbreaks, switching between Ogawa and Inaba [1].
Protection against cholera is serogroup specific. Previous infection with V. cholerae O139 provides no cross-protection from cholera caused by V. cholerae O1, and vice versa [6]–[8]. Serogroup specificity is largely determined by the O-specific polysaccharide (OSP) of lipopolysacharide (LPS). OSP is attached to lipid A that is part of the outer membrane of V. cholerae [9]. We have previously shown that a synthetic neoglyconjugate cholera vaccine containing a hexasaccharide of V. cholerae O1 Ogawa is protectively immunogenic in mice [10]–[12]. We were therefore interested in evaluating whether a cholera conjugate vaccine containing native OSP recovered from V. cholerae O1 would also be immunogenic.
The use of animals complied fully with relevant governmental and institutional requirements, guidelines, and policies. This work was approved by the Massachusetts General Hospital Subcommittee on Research Animal Care (SRAC) – OLAW Assurance # A3596-01; Protocol #2004N000192. The work adheres to the USDA Animal Welfare Act, PHS Policy on Humane Care and Use of Laboratory Animals, and the “ILAR Guide for the Care and Use of Laboratory Animals”.
V. cholerae O1 El Tor Ogawa strain X25049 [13] was used to prepare LPS for use in vaccine preparation and immunological assays, in addition to vibriocidal assays, and wild-type classical V. cholerae O1 classical Ogawa strain O395 [10] was used in vibriocidal assays and the neonatal challenge. Strains were grown in Luria-Bertani broth.
LPS was recovered from X25049, and OSP-core (OSPc) was derived from LPS as previously described [9], [14]. As a carrier protein, recombinant tetanus toxoid heavy chain fragment (TThc) was used [15], [16]. TThc was prepared as a 52,108 Da recombinant protein in E. coli BL21 (DE3) Star with a self-cleaving intein tag using affinity and size exclusion chromatography, as previously described [17].
Conjugation was carried out as previously described [14]. Briefly, 3,4-dimethoxy-3-cyclobutene-1, 2-dione (4.0 mg) was added to a solution of Ogawa O-SP–core antigen (8.0 mg) in pH 7 phosphate buffer (0.05 M, 400 µL) contained in a 2 mL V-shaped reaction vessel, and the mixture was gently stirred at room temperature for 48 h. The solution was transferred into an Amicon Ultra (4 mL, 3K cutoff) centrifuge tube and dialyzed against pure water (centrifugation at 4°C, 7,500× g, 8 times, 35 min each time). The retentate was lyophilized to afford the O-SP–core squarate monomethyl ester as white solid (7.4 mg, 91%).
TThc (3.2 mg) and the methyl squarate derivative of the Ogawa O-SP–core antigen described above (7.4 mg) were weighed into a 1 mL V-shaped reaction vessel and 240 µL of 0.5 M pH 9 borate buffer was added (to form ∼5 mM solution with respect to the antigen; antigen/carrier = 20∶1). A clear solution was formed. The mixture was stirred at room temperature and the progress of the reaction was monitored by SELDI-TOF MS at 24, 48, 72, 96, and 168 h, when no more increase of antigen/carrier ratio could be observed. The mixture was transferred into an Amicon Ultra (4 mL, 30 K cutoff) centrifuge tube and dialyzed (centrifugation at 4°C, 7,500× g, 8 times, 8 min each time) against 10 mM aqueous ammonium carbonate. After lyophilization, 4.6 mg (83%, based on TThc) of conjugate was obtained as a white solid. On the basis of the molecular mass of the carrier (52,108 Da), conjugate (90,000 Da, determined by SELDI TOF MS) and average MW of the OSP antigen of 5,900 Da [14], the antigen/TThc ratio was 6.4∶1 (conjugation efficiency, 32%) (figure 1). A corresponding conjugate was made of OSP: bovine serum albumin (BSA; Sigma #A-4503) using the same approach as described above for use in immunologic assays. The OSP:BSA product contained 4.8 moles OSP per BSA.
For these experiments, we used dmLT, a double mutant derivative of Escherichia coli heat labile toxin (LT), as an immunoadjuvant. dmLT (R192G/L211A) retains immunoadjuvanticity with markedly reduced enterotoxicity [18]. dmLT was prepared as previously described [18], [19].
We immunized cohorts of 10–15, three to five week old Swiss Webster female mice intramuscularly with OSP:TThc or OSP (10 µg sugar per mouse; total 3 doses) with or without dmLT (5 µg). Mice were immunized on days 0, 21, and 42. We collected blood samples via tail bleeds on days 0, 21, 28, 42, 49 and 56. Samples were collected, processed, aliquoted, and stored as previously described [10], [11]. For the memory B cell assay, splenocytes were isolated after day 56 and processed for ELISPOT as previously described [20].
We quantified anti-LPS and OSP IgG, IgM and IgA responses in serum using standard enzyme-linked immunosorbent assay (ELISA) protocols [10], [11]. To assess anti-LPS antibody responses, we coated ELISA plates with V. cholerae O1 Ogawa LPS (2.5 µg/mL) in PBS [10], [11]. To assess anti-OSP antibody responses, we coated ELISA plates with OSP:BSA (1 µg/mL) in PBS. To each well, we added 100 µL of serum (diluted 1∶25 in 0.1% BSA in phosphate buffered saline-Tween) and detected the presence of antigen-specific antibodies using horseradish peroxidase-conjugated anti-mouse IgG, IgM or IgA antibody (diluted 1∶1000 in 0.1% BSA in phosphate buffered saline-Tween) (Southern Biotech, Birmingham, AL). After 1.5 h incubation at 37°C, we developed the plates with a 0.55 mg/mL solution of 2,2′ 0-azinobis (3-ethylbenzothiazoline-6-sulfonic acid) (ABTS; Sigma) with 0.03% H2O2 (Sigma), and determined the optical density at 405 nm with a Vmax microplate kinetic reader (Molecular Devices Corp. Sunnyvale, CA). Plates were read for 5 min at 30 s intervals, and the maximum slope for an optical density change of 0.2 U was reported as millioptical density units per minute (mOD/min). We normalized ELISA units (EU) by calculating the ratio of the optical density of the test sample to that of a standard of pooled sera from mice vaccinated with cholera vaccine from a previous study run on the same plate. We characterized a responder as a ≥2-fold increase in anti-LPS and OSP EU kinetic responses.
We assessed serum vibriocidal antibody titers against V. cholerae X25049 in a micro-assay as previously described [21], [22]. We inactivated endogenous complement activity of mouse serum by heating it for 1 hr at 56°C. We then added 50 µl aliquots of two-fold serial dilutions of heat-inactivated sera in 0.15M saline (1∶25 to 1∶25,600) to wells of sterile 96-well tissue culture plates containing 50 µl/well of V. cholerae X25049 (OD 0.1) in 0.15M saline and 22% guinea pig complement (EMD Biosciences, San Diego, CA). The plates were then incubated for 1 hr at 37°C. 150 µl of brain heart infusion broth (Becton Dickinson, Sparks, MD) was added to each well, and plates were incubated for an additional 1.5 h at 37°C, when optical density at 600 nm was assessed. We calculated the vibriocidal titer as the dilution of serum causing 50% reduction in optical density compared with that of wells containing no serum [23], [24]. We characterized a responder as a ≥4-fold increase in vibriocidal titer.
We assessed memory B-cell assays after the third round of immunization based on previously described methods [20]. Briefly, we treated splenocytes from mice with 1 ml erythrocyte lysis buffer (Sigma) and resuspended them in RPMI supplemented with 10% fetal bovine serum (FBS) (Hyclone, Logan, UT), beta-mercaptoethanol (Sigma, St. Louis, MO), R595 LPS (Alexis), ConA stimulated supernatant and antibiotics (penicillin, streptomycin). The ConA stimulated supernatant was made from naïve mice splenocytes cultured with 2.5 ug/ml ConA and 20 ng/ml PMA for 48 hours at 37°C in a humid atmosphere with 5% CO2. We then cultured spleen cells in 96 well round-bottom plates containing 1×107 cells/mL irradiated syngeneic spleen cell feeders (1200 rad) from naïve mice, and 1×105 cells/well from immunized mice in a total volume of 200 µl. Plates were then incubated at 37°C in a humid atmosphere with 5% CO2. After 6 days in culture, cells were harvested and antigen-specific memory B cell responses were measured by conventional ELISPOT method. We assessed antigen-specific OSP and total IgG ELISPOT assays on these cultured cells. Specifically, nitrocellulose bottom plates (MAHAS4510, Millipore, Bedford, MA) were coated with OSP:BSA (100 ng/well) or with goat anti-mouse IgG (Southern Biotech, Birmingham, AL) or with keyhole limpet hemocyanin (KLH; Pierce Biotechnology, Rockford, IL) (2.5 µg/mL, negative control). After we blocked the plates with RPMI supplemented with 10% FBS, we added the cultured cells to the wells and incubated the plates for 5 h at 37°C in a humid atmosphere with 5% CO2. We then added biotinylated anti mouse IgGγ (Southern Biotech, Birmingham, AL) antibody at 1∶1000, detected IgG antibody expressing cells using horseradish peroxidase-conjugated avidin-D (5 mg/ml, Vector Labs), and developed plates with AEC (3 amino-9-ethyl-carbozole; Sigma). We used unstimulated samples as negative controls and assessed responses to KLH. We characterized a responder as having >2 times total IgG cells with stimulation versus no stimulation and >3 anti-OSP spots.
To assess protection afforded by immunization, we used a cholera neonatal mouse challenge assay, as previously described [10], [11], using wild-type O1 Ogawa V. cholerae O395. In brief, we removed three to five days old un-immunized CD-1 suckling mice (n = 20 mice/cohort) from dams two hours prior to inoculation. We then administered to pups a 50 µl inoculum comprised of 2.3×109 CFU of V. cholerae O395 mixed with a 1∶250 dilution of pooled day 56 serum from mice intramuscularly immunized with the conjugate vaccine OSP:TThc with dmLT, or immunized with dmLT alone. Following oral challenge, we kept neonates separate from dams at 30°C and monitored animals every 3 hr for 36 hr, after which surviving animals were euthanized.
We compared data from different groups using Mann-Whitney U tests. Within each group, comparisons of data from different time points to baseline data (day 0) were carried out using Wilcoxon Signed-Rank tests. Kaplan-Meier and log rank analysis were carried out to compare survival curves in the neonatal challenge study. All reported P values were two-tailed, with a cutoff of P<0.05 considered a threshold for statistical significance. We performed statistical analyses using GraphPad Prism 4 (GraphPad Software, Inc., La Jolla, CA).
We determined progress of conjugation and average carbohydrate content/carbohydrate–protein ratio of OSP:TThc by Surface-Enhanced Laser Desorption–Ionization Mass Spectrometry (SELDI) [25]. Similar to the matrix assisted variant (MALDI) [26], this technique determines average degree of incorporation of carbohydrate onto protein, as well as molecular weight distribution in glycoconjugates. The SELDI analysis showed that the average molecular mass of the conjugate was 90,150 Da. Subtracting from that value the molecular mass of the recombinant protein TThc carrier, 52,108 Da [17][27] the conjugate product molecular mass increased by 38,042 Da. Based on the difference between m/z values of subpeaks within the SELDI peak [14] ; also [28] the molecular mass of the polymolecular OSP–core was determined to average ∼5,900 Da, representingattachment of various lengths of OSP to core. The molecular mass of the conjugate determined by SELDI, 91,150 Da, then indicated the molar ratio of OSP–core:TThc to be ∼6.4∶1.
Following the first injection, we detected significant anti-OSP serum IgG antibody responses in mice receiving OSP:TThc with or without adjuvant (figure 2). Higher magnitude and response rates (P<0.01) were observed in the cohort of animals receiving conjugate vaccine with dmLT (response rate after two doses: 100%). No anti-OSP IgG responses were detected at any time in animals receiving un-conjugated OSP only, with or without immunoadjuvant, or in animals receiving immunoadjuvant alone. Mice receiving OSP:TThc with or without immunoadjuvant and mice receiving OSP alone developed anti-OSP IgM responses (figure 3). IgM responses were only detected following a minimum of two immunizations, and response frequency and magnitude were highest in animals receiving OSP:TThc with adjuvant. No significant IgA anti-OSP antibody was detected in any group (not shown).
Significant serum anti-LPS IgG responses developed following a second immunization in mice receiving conjugate with or without adjuvant (figure 2). Anti-LPS IgM responses were detected in all vaccine cohorts except in mice receiving immunoadjuvant alone (figure 3). No significant anti-LPS IgA responses developed in any group (not shown).
Low-level vibriocidal responses (magnitude and response frequency) were detected in animals receiving unconjugated OSP with or without adjuvant (figure 4). Administration of the immunoadjuvant alone did not elicit any vibriocidal response in animals.
Antigen-specific IgG memory B-cell responses are shown in table 1. OSP IgG specific memory B cell responses were detected in 65% of mice immunized with conjugate vaccine and adjuvant. 18% and 22% of mice immunized with OSP in the presence of dmLT or OSP:TThc alone developed detectable OSP specific memory B cell responses, respectively. No OSP memory response was detected in mice receiving dmLT alone.
We found a significant difference in survival between mice challenged with wild-type V. cholerae O1 Ogawa O395 mixed with sera collected from mice immunized with conjugate and adjuvant (95% survival at 36 hours), compared to mice challenged using sera from mice immunized with adjuvant alone (0% survival at 30 hours; 95% protection; P<0.0001) (figure 5).
In this study, we demonstrate that a cholera conjugate vaccine containing OSP recovered from V. cholerae is protectively immunogenic and induces anti-OSP memory B cell responses in mice. There is a growing body of evidence that anti-OSP responses may be a prime mediator of protective immunity against cholera. Protective immunity to cholera is serogroup specific. Previous infection with V. cholerae O1 provides no protection against O139 and vice versa. This is despite the fact that O1 and O139 express essentially identical cholera toxins (CT) and that O139 is thought to be a derivative of an O1 El Tor strain with high-level homology of most genes in O1 El Tor and O139 [29], [30]. O139 differs from O1 in its genes encoding OSP and in the presence of capsule. The capsule of O139 is comprised of a polysaccharide whose repeating unit is identical to the O139 OSP [15]. The core moieties of O139 and O1 are identical [31]. These data suggest that protection from cholera may be mediated by the serogroup OSP of LPS.
Analysis of anti-OSP responses in cholera patients and their potential role in protection has only recently been initiated [9], [32]. There is however significant evidence that anti-LPS responses correlate with protection from cholera [33], [34]. The vibriocidal response correlates with protection [35] and is largely comprised of anti-LPS IgM responses [36]. We have recently shown that the vibriocidal response can be largely adsorbed away by OSP [9]. Anti-LPS IgA responses in serum and stool have also been associated with protection against cholera among household contacts of cholera patients in Bangladesh [34]. Anti-LPS memory B cell responses similarly correlate with protection against cholera [33].
Currently, two oral killed cholera vaccines are WHO-prequalified and commercially available [37]. One contains approximately 1011 killed V. cholerae O1 classical and El Tor strain organisms (Ogawa and Inaba) and is supplemented with 1 mg of recombinant non toxic B subunit of cholera toxin (WC:rBS; Dukoral, Crucell, Sweden). The other is bivalent, containing killed classical and El Tor O1 organisms as well as an O139 strain, and it does not contain supplemental cholera toxin subunit (Shanchol, Shantabiotechnic-Sanofi, India). Following two doses, these vaccines are 40–85% effective for 6–60 months [13], [37]–[39]. The level of response and duration of protection is particularly decreased in children younger than 5 years of age, compared to older children and adults [38], [40], [41], with booster doses of Dukoral being recommended every 6 months for children under 5 years of age [42]. In comparison, wild-type cholera is associated with high level (90–100%) of protective immunity for at least 3 years in volunteer challenge studies [43] and 3–10 year protection in population-based studies [44]. The level and duration of protection afforded by previous wild-type cholera appears to be the same in young children and in older individuals [45], [46].
We have previously shown that wild-type cholera is associated with a pro-inflammatory response even in young children in Bangladesh, but that vaccination of Bangladeshi children with WC-rBS induces a T-regulatory response [47]. We have also shown that wild-type cholera induces anti-LPS memory B cell responses, even in young children [45], but that children and adult recipients of WC-rBS do not develop such responses [13], [41]. In addition, induction of memory B cell responses correlates with the magnitude of early T cell responses in older recipients of WC-rBS [47], but younger child recipients do not develop T cell responses [47]. These observations may in part explain the lower level and shorter duration of protection afforded by WC-rBS in young children compared to that induced by wild-type disease. Unfortunately, children bear a very large burden of cholera, especially in endemic areas [48], [49]. For instance, 40–80% of children in Bangladesh develop serologic evidence of previous exposure to V. cholerae by the age of 15 years [35], [50], and in areas of India, there is an estimated cholera incidence of 7 per 1000 for children less than 5 years of age, compared to 2.19 in older children and 0.93 in adults (>14 years age) [51]. There is thus a need for improved cholera vaccines or immunization strategies capable of inducing high-level and long-term immunity, especially in young children.
Immune responses targeting OSP may be critical in determining protective immunity from cholera. Since OSP is a T cell-independent antigen, and because young children do not develop prominent responses to polysaccharide antigens administered alone, we are particularly interested in developing a cholera conjugate vaccine. Here we show that a cholera conjugate vaccine is protectively immunogenic in mice and induces memory B cell responses against OSP. Previous prototype cholera conjugates have been developed [52][53][54] Our work contains a number of innovative features. The conjugation process is carried out using squaric acid chemistry, linking the glucosamine present in core oligosaccharide to carrier protein via single point attachment [14]. This takes advantage of the core oligosaccharide, effectively using it as a linker and resulting in a sun-burst display of OSP in a manner that may mimic that present on the surface of V. cholerae. Recent data suggest that the way LPS antigen is presented can impact subsequent immune responses [55]. The fact that the resulting conjugate in our analysis is not cross-linked and, therefore, easier to characterize, together with conjugation methodology that produces conjugates in a predictable manner [56], maximizes the likelihood that vaccine generated in this way and its immunological properties would be reproducible, which is not the case with a number of conjugate vaccines for cholera reported to date. Of note, we do not think that core oligosaccharide contributes significantly to the protective immunity that we observed since previous infection of humans with V. cholerae O1 does not provide protection from V. cholerae O139 and vice versa, despite the presence of identical core oligosaccharides. We also employed as carrier a recombinant immunogenic fragment of tetanus toxoid that could be used as carrier in other vaccines as well. Individuals at risk of cholera are often the most globally disenfranchised and impoverished and may not have received all recommended immunizations, including tetanus vaccine. In addition, we used a novel immunoadjuvant, dmLT [18]. A number of derivatives of the ADP-ribosylating LT molecule of E. coli have been developed and evaluated in humans [18], [57], [58]. These molecules have in common their retained immunoadjuvanticity but markedly diminished enterotoxicity [18]. We have previously shown that transcutaneously applied CT or LT can act as an immunoadjuvant [10], [11]; here we show that low-dose dmLT can also be safely administered parenterally in mice.
Our study is encouraging, but many questions remain. Would an Inaba-based vaccine result in comparable results? Would a response targeting Ogawa OSP cross protect against Inaba? Previous human suggests it may not [59] Would an Inaba-based vaccine protect against Ogawa-associated disease? Could a bi/multi-valent conjugate vaccine be developed? Our vaccine induced vibriocidal responses. Is this a reflection of the fact that a significant component of the vibriocidal response can be adsorbed with OSP [9] or are additional purification steps required? How do conjugates using purified OSP compare to glycoconjugate vaccines prepared from synthetic carbohydrates, which are also under development [10], [11], [60], [61]? Could other immunoadjuvants be used? Is it possible to induce mucosal responses, or would a parenteral cholera vaccine be sufficient when most humans at risk of cholera are also at high risk of tropical or environmental enteropathy with attendant leaking of serum antibodies into the intestinal lumen?
Despite these questions, it is notable that previously produced killed LPS-based whole cell parenteral cholera vaccines were associated with up to 80% protection against disease in humans [62]. Our data suggest that an improved parenteral cholera conjugate vaccine can be developed, one that induces immune responses, including memory B cell responses, to a normally T cell independent antigen (OPS) that is the major target of protective immunity to cholera. Furthermore, this conjugate vaccine can protect against wild-type challenge in animals. Such a conjugate vaccine could have particular utility in young children who are most at risk of cholera.
|
10.1371/journal.ppat.1003619 | Real-Time Whole-Body Visualization of Chikungunya Virus Infection and Host Interferon Response in Zebrafish | Chikungunya Virus (CHIKV), a re-emerging arbovirus that may cause severe disease, constitutes an important public health problem. Herein we describe a novel CHIKV infection model in zebrafish, where viral spread was live-imaged in the whole body up to cellular resolution. Infected cells emerged in various organs in one principal wave with a median appearance time of ∼14 hours post infection. Timing of infected cell death was organ dependent, leading to a shift of CHIKV localization towards the brain. As in mammals, CHIKV infection triggered a strong type-I interferon (IFN) response, critical for survival. IFN was mainly expressed by neutrophils and hepatocytes. Cell type specific ablation experiments further demonstrated that neutrophils play a crucial, unexpected role in CHIKV containment. Altogether, our results show that the zebrafish represents a novel valuable model to dynamically visualize replication, pathogenesis and host responses to a human virus.
| Chikungunya, a re-emerging disease caused by a mosquito-transmitted virus, is an important public health problem. We developed a zebrafish model for chikungunya virus infection. For the first time, rise and death of virus-infected cells could be live imaged in the entire body of a vertebrate. We observed a widespread wave of apparition of newly infected cells during the first day after inoculation of the virus. We then found that infected cells died at a strongly organ-dependent rate, accounting for the progressive shift of virus localization. Notably, the virus persisted in the brain despite apparent recovery of infected zebrafish. We found this recovery to be critically dependent on the host type I interferon response. Surprisingly, we identified neutrophils as a major cell population expressing interferon and controlling chikungunya virus.
| Chikungunya virus (CHIKV) is a mosquito-transmitted virus that causes serious illness and has reemerged in Africa and Asia since 2000, causing outbreaks with millions of cases after decades of near-absence [1]. The epidemic spread to previously CHIKV-free areas, such as La Reunion Island in the Indian Ocean, probably as a consequence of the adaptive mutation of the virus to a new vector species, Aedes albopictus, the tiger mosquito [2], [3], [4], [5]. Unlike traditional CHIKV vectors such as A. aegypti, A. albopictus can produce cold-resistant eggs and is a major invasive species of temperate countries [6], and as it also seems to better transmit the virus [7], CHIKV is now threatening to invade many new territories including the Caribbean, southeast USA and southern Europe. There is currently no commercial vaccine or efficient treatment available for this disease [1].
CHIKV infection is often debilitating and may last from weeks to months; its symptoms in humans include acute fever, rash, joint and muscle pain, chronic arthralgia and, more rarely, severe complications with a fatality rate of about 1 in 1000 [1], [8], [9], [10]. However, CHIKV infection in humans is generally self-limiting, with a short but intense viremia lasting about one week, controlled by type-I interferons (IFNs) [8]. Specific antibodies become detectable shortly after and contribute to virus clearance [11].
CHIKV tropism in vivo, and host innate immune responses are only starting to be characterized [8], [9]. In humans, the virus displays a wide cellular tropism in vitro, infecting fibroblasts, endothelial, epithelial, muscle cells, and to a lower extent, myeloid cells like macrophages [12], [13]. Severe encephalopathies have been reported in CHIKV-infected humans, mostly in infants - more than half infected newborns [14], compared with ∼0.1% in adults [15] - yet CHIKV neurotropism remains controversial [16], [17]. It is still debated whether CHIKV may persist in some cellular reservoirs after the early viremic phase and be responsible for painful relapses that may persist for months.
Murine and macaque models that recapitulate to some extent the human disease have been developed [18], [19], [20], [21]. These models have greatly improved our understanding of the disease, but they do not allow the visualization of infection dynamics and host antiviral and inflammatory responses at the whole body level.
Recently, the zebrafish Danio rerio has emerged as a new model for host-pathogen interactions, largely because their small, transparent larvae are highly suited to in vivo imaging. Zebrafish possess an innate and adaptive immune system akin to that of mammals, but its free-swimming larva relies solely on innate immunity for the first month of its life, allowing the specific dissection of innate immune responses [22]. At the larval stage, cellular immunity consists of myeloid cells only, with neutrophils and macrophages being the main effector cells [23], [24]. As in mammals, antiviral immunity is orchestrated by virus-induced IFNs, of which the zebrafish possess four (IFNφ1-4) [25], [26], structurally similar to mammalian type I IFNs [27]. Zebrafish type I IFNs have been divided into two groups: I (IFNφ1 and φ4) and II (IFNφ2 and φ3), that signal via two different heterodimeric receptors, CRFB1/CRFB5 and CRFB2/CRFB5, respectively. As IFNφ2 is expressed only in adults and IFNφ4 has little activity, the IFN response is mediated by IFNφ1 and IFNφ3 in zebrafish larvae [26], [28].
Since CHIKV infects both mammals and insects, and since other members of the alphavirus genus naturally infect salmonids [29], [30], we hypothesized that the zebrafish free-swimming larva might be sensitive to CHIKV, allowing live imaging of infected cells and dynamics of host-virus relationship in the entire animal. Here we describe a new CHIKV infection model in zebrafish larvae and analyze the dynamics of infection, cell death and host responses. Type I IFNs were critical for survival of CHIKV-infected zebrafish and we identified an unexpected role for neutrophils in both the production of type I IFNs and control of CHIKV infection.
We first asked whether zebrafish were sensitive to CHIKV infection. Larvae aged 3 days post-fertilization (dpf) were injected intravenously (Figure 1A) with ∼102 TCID50 CHIKV, using a strain from the 2005–2006 Reunion Island outbreak (CHIKV-115) [13] or a closely related strain engineered to express GFP (CHIKV-GFP) [3]. Both CHIKV-115 and CHIKV-GFP established infection and replicated in vivo, with production of infectious virions peaking at 24–48 hours post-infection (hpi) (>105 TCID50/larva; i.e., >108 TCID50/gram of tissue) (Figure 1B). Using qRT-PCR with E1-specific primers, we found similar kinetics (Figure 1C). These primers amplify both the genomic and subgenomic transcripts, hence mainly reflect the level of the latter, which is more abundant in alphavirus-infected cells [31], although the ratio of genomic to subgenomic transcripts may vary widely among alphaviruses. Predictably, similar kinetics were obtained for virus-encoded, subgenomic promoter-driven GFP transcripts (Figure 1C). Symptoms, most obvious at 3 days post-infection (dpi), were mild compared to other zebrafish viral infection models [28], [32], [33], [34], the most consistent one being opacification of the yolk (Figure S1A in Text S1). Other less frequent signs included delay in swim bladder inflation, slowing of blood flow, irregular heartbeat, edema, loss of equilibrium and sluggish response to touch (Table S1 in Text S1). These signs were generally transient and by 5 dpi, >90% of infected larvae had apparently recovered, surviving until at least 7 dpi (not shown).
We monitored organs and cells of live CHIKV-GFP infected zebrafish. GFP patterns varied through time (Figure 1D) and between individuals (Figure S1B in Text S1). GFP was detected in liver, jaw, gills, vascular endothelium, eyes, fins, blood cells, muscle fibers, brain, spinal cord, swim bladder and the yolk syncytial layer. Similar patterns were observed in CHIKV-115 infected zebrafish after fixation and immunohistochemistry (IHC) with a capsid-specific antibody (not shown). We quantified the distribution of infected cells in the entire organism over time to establish the kinetics of viral dissemination (Figure 1E). The amount of infected cells peaked by 1–2 dpi in most organs (jaw, fins, liver, vessels, musculature). This peak was followed by a sharp decrease both in the frequency of larvae showing infection in a given organ, and the number of infected cells per organ. By 4 dpi, CHIKV was cleared from most organs. In contrast, infection in the brain parenchyma became visible at 2 dpi in most animals and persisted at least until 5 dpi (Figures 1D and 1E), suggesting that the brain may represent a viral reservoir in zebrafish. At 7 dpi, the latest time point testable, infection in the brain was still strong (Figure S1C in Text S1); in addition, double staining of CHIKV-GFP infected larvae with anti-GFP and anti-capsid antibodies showed that almost all capsid-positive cells also expressed GFP, indicating that GFP expression was a reliable indicator of the infection, even into late stages.
Confocal imaging of IHC-labeled CHIKV-infected larvae showed infection in various cell types (Figure 2A), namely fibroblasts in fins (Figure 2B) and jaw (not shown), endothelial cells (Figure 2C), muscle fibers (Figure 2D) and hepatocytes (Figure 2E, and Figure S2 in Text S1). Infection also occurred occasionally in red blood cells (Figure S2 in Text S1) but not in macrophages or neutrophils (not shown). In zebrafish brain, CHIKV was detected in both neurons and glial cells (Figure 2F, and Figure S2 in Text S1).
To assess the dynamics of CHIKV infection and its cytopathic effects, we performed time-lapse imaging of CHIKV-GFP infected larvae (Figures 3A and 3B) and compiled the appearance and death of GFP+ cells (Figures 3C–E). 88% of newly infected cells appeared before 24 hpi in one major wave (Figure 3C). The median time of appearance of new GFP+ cells was 14±2 hpi with similar kinetics in all cell types (Figure 3D). Death of GFP+ infected cells presented apoptosis features such as membrane blebbing and cellular fragmentation (Figure 3B and Movies S1 and S2). It was frequent from 24 hpi onwards (Figure 3C), with an overall median death time of 67±4 hpi, but dependent on cell type (Figure 3E). For instance, liver cells were highly susceptible to CHIKV cytopathic effects, with a median occurrence of death at 41±5 hpi, implying that hepatocytes survive for ∼27 h following GFP detection, compared to a ∼53 h survival period for the general cell population. In contrast, almost all infected brain parenchyma cells survived at least until 72 hpi. These results demonstrate that the apparent shifting tropism of infection towards brain (Figure 1E) is largely due to differential cell survival.
Type I IFN signaling is critical for control of CHIKV in mammals [18], [20]. In zebrafish larvae, CHIKV triggered high mRNA levels of ifnφ1 (NM_207640, secreted isoform transcript) and ifnφ3 (NM_001111083), and of various IFN-induced genes including viperin/vig-1/rsad2 (NM_001025556) (Figures 4A–C and not shown). Ifnφ1 and viperin levels, peaking at 17–24 hpi, remained high for at least 4 days, correlating with viral burden. These levels were higher than previously observed with fish viruses in zebrafish [28], [32], [34]. Ifnφ3 induction was less prominent in breadth and duration. To assess the role of the IFN response, we knocked down receptors for all IFNφs with antisense morpholino oligonucleotides (MO) directed to the CRFB1 (NM_001079681) and CRFB2 (NM_001077626) subunits [26]. When IFN receptor expression was impaired (CRFB1+2 MO), the disease was particularly severe (Table S1 in Text S1), as measured by a disease score (defined in Table S2 in Text S1) (Figure 4D). Among CRFB1+2 morphant fish, >90% died from infection (Figure 4E), while virus burden was increased up to 100-fold when compared to infected control morphants (Figure 4F). Upstream of IFN signaling, sensing of CHIKV through the cytosolic pathway was important as knockdown of MAVS (IPS-1/CARDIF/VISA) (NM_001080584) (Figure S3 in Text S1) also led to an increase in disease severity and mortality, as well as in virus burden (Figures 4D–F), consistent with results obtained in mice [35], [36]. As expected, knockdown of CRFB1 and CFRB2 did not affect ifnφ1 production but blocked viperin expression, whereas in MAVS morphants, both ifnφ1 and viperin levels were significantly reduced (Figure S3 in Text S1). Altogether, these results show that the type I IFN pathway controls CHIKV replication and pathogenesis in zebrafish.
To identify the source of IFN, we first performed whole-mount in situ hybridization (WISH) using an antisense probe for ifnφ1 at the peak of the response. In CHIKV-infected larvae, ifnφ1 expression was detected in the liver and in scattered cells with a morphology and distribution evoking leukocytes (Figure 4G). To better visualize the spatiotemporal dynamics of IFN production, we designed a transgenic IFNφ1 reporter zebrafish, in which the ifnφ1 promoter drives expression of the mCherry red fluorescent protein. In uninfected 3–6 dpf transgenic larvae, mCherry was detected in very few (10–30) cells, all with leukocyte morphology and mostly residing in the caudal hematopoietic tissue (CHT), but upon CHIKV infection, the number of mCherry+ cells dramatically increased (Figure 5A). Starting from 2 dpi, two main populations of mCherry+ cells were detected: hepatocytes and motile leukocytes (Figure 5A and Movie S3). The mCherry+ leukocytes were dispersed throughout the body except the CNS, mostly in the anterior region and the CHT, and persisted until at least 4 dpi. This pattern of expression of the reporter transgene was similar to that of the endogenous ifnφ1 gene (Figure 4G), but appearing later, a delay apparently due to the time required for protein expression and maturation, since at 24 hpi mCherry fluorescence was still low despite mCherry mRNA expression (Figure 5B). Thus, the reporter transgene is faithful but somewhat delayed compared to endogenous ifnφ1. Notably though, viral GFP and mCherry were not detected in the same cells, suggesting that IFN release occurs mostly in uninfected or non-productively infected cells.
We further characterized IFN-producing cells. We FACS-sorted mCherry+ cells from infected ifnφ1:mCherry zebrafish at 3 dpi and analyzed their mRNA expression profile (Figure 5C). As expected, expression of ifnφ1 was highest in sorted mCherry+ cells. These cells did not notably co-express ifnφ3. Among leukocyte genes, the macrophage marker c-fms/csf1r (NM_131672) was increased in mCherry+ cells, but the strongest enrichment was for myeloperoxydase (mpx, NM_212779), a specific neutrophil marker in zebrafish [23], [24]. The hepatocyte marker fabp1a (NM_001044712) was also expressed, consistent with some hepatocytes producing IFN. Sorted mCherry− cells expressed lower but significant ifnφ1 levels – especially if compared to naïve larvae, which express it to an extremely low level -, likely due to the aforementioned delay. Both mCherry+ and mCherry− expressed the IFN-inducing transcription factors irf3 (NM_001111083) and irf7 (NM_200677), with the latter being enriched among mCherry+ cells.
To confirm the involvement of neutrophils, we crossed neutrophil reporter mpx:GFP with ifnφ1:mCherry zebrafish. In double transgenic CHIKV-infected zebrafish, either uninfected or CHIKV-infected, more than 80% of mCherry+ leukocytes expressed GFP (Figures 6A–C). Their morphology, distribution, speed, and presence of refractile moving granules, as assessed by live Nomarski microscopy, were also consistent with neutrophil identity [24] (not shown).
The number of mCherry+ neutrophils strongly increased by 48 hpi and remained high until at least 96 hpi (Figure 6D, and Figure S4A in Text S1). Other mCherry+ leukocytes (mostly mpeg1+ macrophages, not shown) were also increased at 48 hpi, but in lower numbers, and notably in the CHT where they transiently made up about half the mCherry+ population (Figure 6E, and Figures S4B and S4C in Text S1). Neutrophil numbers, quantified by Sudan Black staining, peaked at 72 hpi (2001±312 cells/larva compared to 945±234 cells/larva in uninfected controls) (Figure 6F); both mCherry+ and mCherry− neutrophils increased (Figure 6D and not shown). Nevertheless, neutrophil distribution was not obviously perturbed: they did not accumulate at infection foci and were absent from the CNS, like in uninfected fish [24]. Interestingly, knockdown of IFN receptors blocked neutrophil increase, indicating that it is dependent on the IFN response (Figure 6G).
We next addressed the role of neutrophils, macrophages and hepatocytes in the control of CHIKV infection by cell depletion strategies.
First, we blocked myelopoiesis by knocking down PU.1/spi1 (AF321099), resulting in reduced neutrophil and, even more deeply, macrophage populations [37] (Figures S5A and S5B in Text S1; note that head images reflect the impact on mature cells, while tail images include the hematopoietic region to assess depletion of precursors). PU.1 knockdown dramatically increased disease severity (disease score of 10.8±3.4 compared to 2.3±1.6 in control morphants) and mortality (Figures 7A and 7B), and correlated with an increase in viral transcripts (Figure 7C). Therefore, myeloid cells largely control CHIKV in zebrafish.
To distinguish the roles of these two leukocyte types, we first selectively depleted macrophages with a transgenic drug-inducible cell ablation system [38] (Figures S5C and S5D in Text S1). Macrophage-depleted CHIKV-infected larvae exhibited a small increase in disease severity (disease score of 4.8±2.6 compared to 2.3±1.5 in control transgenics) (Figure 7D) but almost no mortality (Figure 7E), despite modestly increased virus amounts (Figure 7F). This suggests that macrophage depletion plays a minor role in the phenotype of PU.1 morphants.
Comparable specific depletion of neutrophils was not available, however csf3r/gcsfr (NM_001113377) knockdown has been shown to affect neutrophil populations more than macrophages [39]. Indeed, at 3 dpf, our csf3r morphants displayed no significant reduction of mpeg1+ macrophage numbers, while mpx+ neutrophils were severely depleted (Figures S6A and S6B in Text S1); neutrophil depletion lasted until 6 dpf (Figure S6C in Text S1). In infected animals too, csf3r knockdown led to a stronger reduction of neutrophils than macrophages, in contrast to PU.1 (Figure S6D in Text S1). Csf3r morphants were highly susceptible to CHIKV, with a high disease score (Figure 7G), mortality starting 3 days after infection (Figure 7H), and strongly increased virus transcripts (Figure 7I).
In addition, we attempted to block the increase in neutrophil numbers by knocking down nos2a (zebrafish iNOS) (NM_001104937), a strategy recently described to block infection-induced granulopoiesis in a bacterial infection system [40]. The neutrophil population was not reduced in nos2a morphants before the infection (not shown), but its increase was effectively prevented (Figure S6E in Text S1), and this was associated with increased disease scores (Figure 7J), mortality starting at 4 dpi (Figure 7K), and an increase in viral transcripts (Figure 7L).
Altogether, these experiments provide independent and convergent evidence consistent with neutrophils being the major population controlling CHIKV, in agreement with their predominance among ifnφ1-expressing leukocytes (Figure 6C).
Finally, transient hepatocyte depletion using a Tomm22 (NM_001001724) MO [41] (Figure S7 in Text S1) also led to higher disease severity and more virus production (Figures 7M and 7O) but no increased mortality (Figure 7N), indicating that hepatocytes do not play a role as important as leukocytes in controlling CHIKV.
In this study, we establish zebrafish as a new model for the study of the pathogenesis of CHIKV. The overall course of viral spread in zebrafish larvae was close to that observed in mammals, with an early peak of viremia followed by a decline, similar targeted cell types, and a critical dependence on the host IFN response for the control of the virus. In addition, the powerful in vivo imaging techniques available in zebrafish revealed new features of the infection.
We could image the onset of infection in individual cells throughout the body. Almost all new infected cells appeared during one major wave during the first 24 hours following injection of the virus, with relatively little difference between the various targeted organs. Because we could not detect cells with strong GFP expression before the rise of this first wave of infected cells, we presume it reflects the initial set of cells infected by the inoculated virions. The significant inter-individual variation that we observe may be a consequence of a larger number of susceptible cells than of inoculated virions, resulting in a stochastic initial pattern of infection. The decline of appearance of newly infected cells shortly followed the onset of the host IFN response, suggesting that by the time the initial wave of infected cells produce new infectious virions, the host response has made most other cells refractory to the virus. We also observed and quantified infected cell death events, which typically presented apoptosis characteristics. The timing of death of CHIKV+ cells was strongly organ-dependent. The differential survival of infected cells accounted for the apparent shift of tropism towards the brain parenchyma, where infection persisted even after clearance from the rest of the body.
The longer persistence of CHIKV in brains of zebrafish suggests that neurons may constitute a previously overlooked reservoir for the virus. However, this is likely to be mostly the case in infant humans, since encephalitis is a feature of chikungunya disease in newborns rather than in adults. CHIKV potential reservoirs are a matter of conjecture because many patients display chronic arthralgia in the months following CHIKV infection despite resolution of viremia, and it is unclear whether this is due to long-lasting auto-inflammation triggered by the initial infection or to stimulation by persistent virus [8], [9], [10]. In adult macaques, CHIKV was suggested to persist in macrophages, not CNS [19]. In infected neonate mice, CHIKV was not found to persist in the brain [18], [42]. Moreover, in this model, CHIKV was found to infect leptomeningeal and choroid plexus cells, but not brain parenchyme. Yet, mouse brain parenchymal cells may be infected by CHIKV, as shown after intranasal infection [43] or on primary cell cultures [17].
The zebrafish model also allowed us to dynamically image and FACS-sort the cells that are responding to the virus by expressing the ifnφ1 gene. Based on gene expression profile, morphology, and co-expression of the mpx:GFP transgene, two main populations were shown to express the ifnφ1:mCherry transgene: neutrophils and hepatocytes. Interestingly, while both irf7 and irf3 are ISGs in fish [44], and therefore expected to be induced in all cells of infected fish, irf7 was expressed at a higher level in sorted mCherry+ than mCherry− cells. This would be consistent with constitutively higher expression of irf7 by cells specialized in ifnφ1 expression in zebrafish – mirroring key properties of plasmacytoid dendritic cells of mammals [45]. The cell types, however, were different. Although not viewed as a specialized source of IFN, hepatocytes have been found to be prominent producers in some cases, for example during Thogoto virus infection of a mouse IFNβ reporter cell line in vitro [46]. By contrast, neutrophils are so far not considered to represent an important source of IFN [47], [48]. Nevertheless, in zebrafish larvae, neutrophils were found to represent 80% of ifnφ1-expressing leukocytes. In this respect, it should be stressed that our main marker, mpx, not entirely neutrophil-specific in mammals, has been shown to be strictly neutrophil-specific in zebrafish [23], [24]. In addition, our depletion experiments were consistent with neutrophils being a key population controlling CHIKV infection in zebrafish, whereas macrophages and hepatocytes made a minor contribution to this control. Macrophage depletion having little consequences, no synergy of macrophages and neutrophils seems required to control CHIKV. However, until a truly neutrophil-specific depletion method becomes available in zebrafish, we cannot rule out the possibility of a significant additive contribution of a minor csf3r-dependent macrophage subpopulation to that of neutrophils; compensation mechanisms following depletion are also an important caveat to consider.
Besides IFN production, other mechanisms may be responsible for the observed protective role of myeloid cells, especially neutrophils, against CHIKV pathogenesis. The role of neutrophils in protecting against viral infections is not fully deciphered [49]. Neutrophil extracellular traps (NETs) were recently shown to protect host cells from myxoma virus infection in mice [50] and to capture HIV-1 and promote its elimination through the action of myeloperoxidase and α-defensin in humans [51]. Zebrafish neutrophils, which share many functional characteristics with their human counterparts, including the production of NETs [52], avidly engulf bacteria on surfaces [53] and scavenge dying infected cells in mycobacterial disease [54], but their function during viral infection was so far unknown. It will be worth further studying the role of zebrafish (and human) neutrophils in sensing of CHIKV-infected cells and the mechanisms mediating viral clearance.
Neutrophil numbers were increased with CHIKV infection, a response we found to be dependent on the IFN response. This was contrary to our expectations, as acute IFN induction by viral infection is known to cause granulocytopenia [55], and even in fish, granulocyte numbers were found to be reduced during a viral infection [56]. Interestingly, neutrophilia has been reported in CHIKV-infected humans with a high viral load [57], suggesting that CHIKV may stimulate neutrophils in an unusual manner. Remarkably, this increase was found to depend on nos2a (zebrafish iNOS), as had been observed in a Salmonella infection model in zebrafish [40]. Depending on the experiment settings, iNOS has been found to favor [58], [59] or counteract [60] neutrophil infiltration in inflamed organs in mice. It would be worth investigating the contribution of iNOS to the inflammatory response induced by CHIKV in mammals.
Comparing patterns of infection and of IFN response, it may be significant that virus persistence - dictated by survival of infected cells - was inversely correlated with local production of IFN. The organ where infected cells died fastest was the liver, which was also a major local source of IFN. Conversely, infected cells persisted much longer in the brain, an organ from which neutrophils are excluded, whereas they patrol other tissues in zebrafish [24]. Assessing the relative contribution of the cell autonomous – such as autophagy [61] – and non-cell autonomous (mostly, IFN-driven) events underlying sensitivity of the cells to the cytopathic effect of CHIKV in vivo will be one of our future goals. IFNs have been shown to induce apoptosis of virus-infected cells [62]. It is possible that infected brain neurons and glial cells persist due to the blood brain barrier (BBB) blocking IFN access to this organ. Zebrafish brain endothelial cells express BBB markers Claudin 5 and ZO-1 as early as 3 dpf and brain parenchymal vessels are impermeable to horseradish peroxidase (44 kDa) and rhodamine-dextran (10 kDa) at this stage [63]. It is therefore likely that zebrafish IFNφs (∼20 kDa) cannot reach the brain parenchyma, which would prevent brain-infected cells from undergoing apoptosis. It has also been suggested that less “renewable” tissues and cells, such as post-mitotic neurons, respond to type I IFNs differently from other cell types [62].
Imaging studies detailing the dynamics of single virus-infected cells in vivo are very recent and remain scarce [64], [65], [66]. The zebrafish model offers the unique opportunity to visualize and characterize in real time the rise and death of infected cells, throughout the body. To our knowledge, this study represents the first analysis of the fate of single virus-infected cells in a whole organism. Combined with the ability to image IFN-producing cells and to perform host gene silencing, mutagenesis or drug screening, our work establishes the zebrafish as a new valuable host for the study of human pathogenic viruses.
All animal experiments described in the present study were conducted at the Institut Pasteur according to European Union guidelines for handling of laboratory animals (http://ec.europa.eu/environment/chemicals/lab_animals/home_en.htm) and were approved by the Direction Sanitaire et Vétérinaire de Paris under permit #B-75-1061.
Zebrafish embryos were raised as previously described [67], [68]. Wild-type AB fish were initially obtained from ZIRC (Eugene, OR, USA). The following transgenic lines were used: Tg(gata1a:DsRed)sd2 [69], Tg(elavl3:EGFP)knu3 [70] referred to as HuC:GFP in the text, Tg(gfap:EGFP)mi2001 [71], Tg(fabp10:dsRed)gz4 [72], Tg(mpx:EGFP)i114 [73], Tg(mpeg1:mCherry)gl23 and Tg(mpeg1:Gal4FF)gl25 [74], and Tg(UAS-E1b:Eco.NfsB-mCherry)c26 [38] referred to as UAS:NfsB-mCherry in the text. For imaging purposes, embryos were generally raised in 0.003% 1-phenyl-2-thiourea (PTU) from 24 hpf onwards to prevent melanin pigment formation.
CHIKV was produced on BHK cells. CHIKV-115 is a clinical strain isolated in 2005 from a young adult in La Réunion [2] and its entire genome sequence is available (#AM258990). This virus has been passaged three times since cloning. CHIKV-GFP corresponds to the CHIKV-LR 5′GFP virus generated by insertion of a GFP-encoding sequence controlled by the CHIKV subgenomic promoter between the two main genes of the CHIKV genome, using the LR backbone (#EU224268) derived from the OPY1 strain, a 2006 clinical isolate from La Réunion; GFP expression has been found to be retained in >80% infected cells for up to 8 serial passages in mammalian or mosquito cells [3]. The CHIKV-GFP virus we used previously went through two to three passages.
We generated two independent lines of ifnφ1 reporter transgenics, Tg(ifnphi1:mCherry)ip1 and Tg(ifnphi1:mCherry)ip2 with indistinguishable transgene expression (not shown), and both are referred here as ifnφ1:mCherry fish. The 6.5 kb SpeI-PstI fragment from PAC clone BUSMP706A0151Q01 (IMAGENE) covering the ifnφ1 promoter was cloned ahead of the ORF for a farnesylated version of mCherry in a Tol2 derivative vector to yield vector pTol2pIFNL1mC-F. The fragment includes exon 1 including the first codons of the zebrafish ifnφ1 ORF. This construct was co-injected with tol2 mRNA into 1-cell stage eggs of AB origin [75].
Injections and handling of larvae were performed as described [68]. Briefly, zebrafish larvae aged 70–72 hpf were inoculated by microfinjection of ∼102 TCID50 CHIKV (∼1 nl of supernatant from infected BHK cells, diluted to 108 TCID50/ml) in the caudal vein or aorta. Larvae were then distributed in individual wells of 24-well culture plates, kept at 28°C and regularly inspected with a stereomicroscope. Clinical signs of infection were assessed first on aware animals, which were then anesthetized for better observation. Quantitative assessment of the clinical status was based on a precise list of criteria (see Table S2 in Text S1) assessed blindly, yielding a disease score ranging from 0 (no disease sign) to 15 (dead or terminally ill).
Infected larvae were snap-frozen and kept at −80°C before homogenization in ∼100 µl of medium; samples were then titrated as TCID50/larva on Vero cells [13].
RNA extraction, cDNA synthesis and quantitative PCR were performed as previously described [34]; externally quantified standards were included to provide absolute transcript amounts. The following pairs of primers (sense and antisense) were used: GFP: CCATCTTCTTCAAGGACGAC and CGTTGTGGCTGTTGTAGTTG; ef1α: GCTGATCGTTGGAGTCAACA and ACAGACTTGACCTCAGTGGT; ifnφ1 (secreted isoform): TGAGAACTCAAATGTGGACCT and GTCCTCCACCTTTGACTTGT; ifnφ3: GAGGATCAGGTTACTGGTGT and GTTCATGATGCATGTGCTGTA; viperin: GCTGAAAGAAGCAGGAATGG and AAACACTGGAAGACCTTCCAA; E1-CHIKV: AARTGYGCNGTNCAVTCNATG and CCNCCNGTDATYTTYTGNACCCA (these primers match positions 10921–10943 and 11167–11189, respectively, of the CHIKV genome acc#AM258990, and include degenerate bases, labeled according to the IUPAC convention, making them indifferent to silent mutations); irf3: GAGCCAAATCTGGCGACATT and GGCCTGACTCATCCATGTT; irf7: TCTGCATGCAGTTTCCCAGT and TGGTCCACTGTAGTGTGTGA; mpx: ATGGAGGGTGATCTTTGA and AAGCTATGTGGGATGTGA; mpeg1: CCCACCAAGTGAAAGAGG and GTGTTTGATTGTTTTCAATGG; fabp1a: AGACAGAGCTAAAACTGTGGT and AGCTGAGAGTGTTACTGATAG; mCherry: CCCGCCGACATCCCCGACTA and GGGTCACGGTCACCACGCC. To normalize cDNA amounts, we used the housekeeping gene ef1α transcripts, except in specified cases where results were normalized to viral burden using E1-CHIKV.
Larvae were anesthetized and laid on the bottom of an agarose-coated, sealed Petri dish, and imaged as described [34]. To assess efficiency of depletion strategies, Z-stacks with 22 µm steps of anesthetized larvae were taken with a Leica Z16 APO A macroscope and quantification performed using ImageJ software. For Figures S6A and S6B in Text S1 quantification was performed as described before [76].
For in vivo time-lapse imaging, 4–6 larvae, anaesthetized with 112 µg /ml tricaine, were laterally positioned and immobilized in ∼1% low melting point agarose in the center of a 54-mm plastic bottom Petri dish, then covered with 2 ml water containing tricaine. Multiple field transmission and fluorescence imaging was performed using a Nikon Biostation IMQ, using a 10× objective (NA 0.5) and a DSQi camera. Imaging was typically performed at 26°C and Z-stacks with 10 µm steps were acquired at least every 30 minutes. Imaging sessions typically lasted 6–24 hours; control uninfected larvae were always included. Cell emergence and death data were concatenated from multiple imaging sessions covering the 0 to 72 hpi time frame.
IHC was performed as described [77]. Primary antibodies used were: mouse mAb to alphavirus capsid (1∶200) [78], rabbit polyclonal to DsRed (1∶300, Clontech) which also labels the mCherry protein, mouse monoclonal to GFP (1∶500, Invitrogen), chicken polyclonal to GFP (1∶500, Abcam), mouse monoclonal (FIS 2F11/2) to gut secretory cell epitopes (1∶400, Abcam). Secondary antibodies used were: Cy3-labeled goat anti-rabbit or anti-mouse IgG (1∶300, Jackson Immunoresearch), Alexa 488-labeled goat anti-mouse or anti-chicken (1∶500, Invitrogen). Nuclei were stained for 30 min at room temperature with Hoechst 33342 at 2 µg/ml (Invitrogen).
Fixed embryos were progressively transferred into 80% glycerol before imaging. Confocal images of IHC-processed fish were taken with a Leica SPE inverted confocal microscope equipped with a 16× (NA 0.5), 63× (NA 1.30) oil immersion objectives and a 10× (NA 0.30) dry objective. Images of larvae stained by WISH or Sudan Black B were taken with a Leica MZ16 stereomicroscope using illumination from above. Whole-body images of IHC-treated larvae were taken with a Leica Z16 APO A macroscope. Images were processed with the LAS-AF (Leica), ImageJ and Adobe Photoshop softwares. Cells with amoeboid morphology were scored as “leukocytes”.
Morpholino antisense oligonucleotides (Gene Tools) were injected into 1–4-cell stage embryos as previously described [68]. crfb1 splice morpholino (CGCCAAGATCATACCTGTAAAGTAA) (2 ng) was injected together with crfb2 splice morpholino (CTATGAATCCTCACCTAGGGTAAAC) (2 ng), knocking down all type I IFN receptors [26]. Other morpholinos: mavs splice morpholino (ATTTGAATCCACTTACCCGATCAGA) (4 ng); tomm22 translation morpholino [41] (GAGAAAGCTCCTGGATCGTAGCCAT) (2 ng); pu.1 translation morpholino (GATATACTGATACTCCATTGGTGGT) [37] (20 ng in 2 nl); csf3r translation morpholino (GAAGCACAAGCGAGACGGATGCCAT) [74] (4 ng); nos2a splice morpholino (ACAGTTTAAAAGTACCTTAGCCGCT) [40] (6 ng). Control morpholinos with no target: #1 (GAAAGCATGGCATCTGGATCATCGA) (2–6 ng); #2 (TACCAAAAGCTCTCTTATCGAGGGA) (20 ng); #3 (CCTCTTACCTCAGTTACAATTTATA) (4 ng).
Embryo dissociation was performed as described elsewhere [79]. Sorted cells were collected in lysis buffer and RNA was extracted using a RNAqueous Micro kit (Ambion). Cell preparations were performed in a BL3 facility; the cell sorter, located under a plastic tent within a BL2 facility, was flushed for several hours with diluted bleach following the sorting.
WISH was performed as described before [80], with a hybridization temperature of 55°C. To generate the ifnφ1 antisense probe, we RT-PCR amplified a 503 bp fragment of zebrafish ifnφ1 cDNA from CHIKV-infected larvae using a T3-modified antisense primer (GAATTCATTAACCCTCACTAAAGGGAGATTGACCCTTGCGTTGCTT) and a normal sense (TCTGCAGAGTCAAAGCTCTG). PCR products were purified with QIAquick PCR purification kit (Qiagen) and the probe was transcribed in vitro with T3 polymerase (Promega). Unincorporated nucleotides were removed by purification on NucAway spin columns (Ambion).
Neutrophil granules were stained as in [24], allowing neutrophils to be counted easily with a dissecting scope.
Metronidazole-mediated depletion was performed as described in [38]. Briefly Tg(mpeg1:Gal4FF)gl25/− [74] were crossed to Tg(UAS-E1b:NfsB-mCherry)c264/c264 [38] to generate double-positive transgenics and single-positive sibling controls. Embryos were placed, from 48 hpf to 70 hpf, in a 10 mM Metronidazole, 0,1% DMSO solution to induce specific depletion of NfsB-mCherry-expressing macrophages. Embryos were then rinsed 3× with embryo water.
To evaluate difference between means, a two-tailed unpaired t-test or an analysis of variance (ANOVA) followed by Bonferroni's multiple comparison test was used, when appropriate. Normal distributions were analyzed with the Kolmogorov-Smirnov test. Non-Gaussian data were analyzed with a Kruskal-Wallis test followed by Dunn's multiple comparison test. P<0.05 was considered statistically significant (symbols: ***P<0.001; **P<0.01; *P<0.05). Survival data were plotted using the Kaplan-Meier estimator and log-rank tests were performed to assess differences between groups. Statistical analyses were performed using Prism software.
|
10.1371/journal.pbio.2006347 | Transcriptional outcomes and kinetic patterning of gene expression in response to NF-κB activation | Transcription factor nuclear factor kappa B (NF-κB) regulates cellular responses to environmental cues. Many stimuli induce NF-κB transiently, making time-dependent transcriptional outputs a fundamental feature of NF-κB activation. Here we show that NF-κB target genes have distinct kinetic patterns in activated B lymphoma cells. By combining RELA binding, RNA polymerase II (Pol II) recruitment, and perturbation of NF-κB activation, we demonstrate that kinetic differences amongst early- and late-activated RELA target genes can be understood based on chromatin configuration prior to cell activation and RELA-dependent priming, respectively. We also identified genes that were repressed by RELA activation and others that responded to RELA-activated transcription factors. Cumulatively, our studies define an NF-κB-responsive inducible gene cascade in activated B cells.
| The nuclear factor kappa B (NF-κB) family of transcription factors regulates cellular responses to a wide variety of environmental cues. These could be extracellular stimuli that activate cell surface receptors, such as pathogens, or intracellular stress signals such as DNA damage or oxidative stress. In response to these triggers, NF-κB proteins accumulate in the cell nucleus, bind to specific DNA sequences in the genome, and thereby modulate gene transcription. Because of the diversity of signals that activate NF-κB and the ubiquity of this pathway in most cell types, cellular outcomes via NF-κB activation must be finely tuned to respond to the initiating stimulus. One mechanism by which NF-κB-dependent gene expression is regulated is by varying the duration of nuclear NF-κB; some signals lead to persistent nuclear NF-κB, while others lead to transient nuclear NF-κB. Consequently, time dependency of transcriptional responses is a unique signature of the initiating stimulus. Here we probed mechanisms that generate kinetic patterns of NF-κB-dependent gene expression in B lymphoma cells responding to a transient NF-κB-activating stimulus. By genetically manipulating NF-κB induction, we identified direct targets of RELA, a member of the NF-κB family, and provide evidence that kinetic patterns are established by a combination of factors that include the chromatin state of genes prior to cell activation and cofactors that work with RELA.
| The family of nuclear factor kappa B (NF-κB) transcription factors regulates inducible gene transcription in response to diverse stimuli. Signals from receptors ultimately activate inhibitor of NF-κB kinases (IKKs) that phosphorylate a variety NFKB inhibitors (IκBs), targeting them for degradation and leading to accumulation of NF-κB family members in the nucleus. The “classical” pathway, via IKK2 activation, results in nuclear translocation of RelA- or Rel-containing NF-κB proteins, whereas the nonclassical pathway, via IKK1 activation, results in nuclear accumulation of RelB-containing NF-κB proteins [1,2]. Some signals only activate IKK1 (such as B cell activating factor [BAFF]/BAFF receptor [BAFF-R]), others only activate IKK2 (such as tumor necrosis factor alpha [TNFα] and interleukin 1 beta [IL-1β]), and yet others activate both IKK1 and IKK2 (such as CD40/CD40L interaction). The cellular response to NF-κB activation therefore depends on the nature of the stimulus and the associated pattern of NF-κB proteins that are driven to the nucleus. Despite identification of a handful of well-accepted NF-κB target genes (such as NFKBIA, TNFAIP3, and MYC), genome-wide transcriptional responses mediated by NF-κB remain poorly defined.
There are several reasons for this. First, the time course of NF-κB induction varies greatly depending on stimulus. For example, classical NF-κB activation by TNFα or IL-1β is rapid and transient, whereas activation via Toll-like receptor 4 (TLR4) is slower and more sustained [1,3]. Second, NF-κB proteins consist of several family members. Nuclear factor kappa B subunit 1 (NFKB1), RELA, and REL proteins respond primarily by the classical pathway, whereas nuclear factor kappa B subunit 2 (NFKB2) and RELB respond to nonclassical activation. Thus, a comprehensive analysis must include gene targets of each family member. This inherent complexity is compounded by observations that some genes that encode Rel family members, such as NFKB1, NFKB2, and RELB, are themselves targets of NF-κB. Third, NF-κB responses vary depending on the cell type as well as the initiating stimulus. Cell type specificity of NF-κB targets in monocytes/macrophages has been proposed to be conferred by regulated access of induced NF-κB to a subset of genomic sites by tissue-specific transcription factors [4–6]. Stimulus specificity has been explored largely in the context of TLR signaling and attributed to differences in dynamic patterns of NF-κB induction [7–9]. Yet, connections between NF-κB dynamics and transcriptional output are not well understood.
For genes to be classified as NF-κB targets, they must change in expression as a consequence of NF-κB binding in cells that have received 1 or more NF-κB-inducing stimuli. This requires integrating at least 3 variables: transcriptional outcomes, NF-κB binding, and the consequences of abolishing NF-κB binding. The first 2 criteria have been explored by microarray or RNA sequencing (RNA-Seq) and by chromatin immunoprecipitation and sequencing (ChIP-Seq) to assay transcription factor binding genome-wide. For NF-κB, the majority of ChIP-Seq studies used TNFα as the NF-κB-inducing stimulus in HeLa cells or endothelial cells, with stimulus times ranging from 1 to 6 h. In response to TNFα, RELA bound to approximately 1,200–12,500 sites genome-wide in different studies, with the majority of binding occurring at sites other than gene promoters [10–19]. One of the earliest RELA ChIP-Seq studies also noted that the factor bound at many genes whose expression was unaffected by TNFα treatment [20]. Sites of RELA binding were enriched not only for the κB motif (GGGRNYYYCC) but also for recognition sites of other transcription factors, especially activator protein 1 (AP1). The latter observation corroborated previously reported interactions between NF-κB and AP1 [21,22]. Additionally, E2F and forkhead box M1 (FoxM1) transcription factor motifs have been identified within RELA binding regions [20,23]. The RelA response has also been characterized in murine macrophages treated with lipopolysaccharide (LPS) for 1–3 h [4,5,24,25]. A substantial proportion of sites to which RelA was recruited in these cells were found to have prebound PU.1, a macrophage-enriched transcription factor. In addition, recognition motifs of interferon regulatory factors (IRFs) and AP1 were enriched at RelA-bound sites in macrophages. Such differences have been proposed to underlie selectivity of NF-κB responses.
The third criterion, that of establishing that transcriptional outcome is a consequence of NF-κB binding, remains largely underexplored. The most precise way to causally connect binding events to gene expression requires mutating these sites in genomic DNA followed by transcriptional analyses. This is virtually impossible to do on a genome-wide scale. A more feasible, yet meaningful, alternative is to monitor transcriptional consequences of depleting the transcription factor, such as the use of RelA-deficient macrophages to validate the role of RelA in the LPS response [24]. Additionally, most studies do not account for the impact of dynamic patterns of RELA induction on inducible gene transcription. For this, both NF-κB binding and RNA levels must be interrogated at multiple time points in response to a stimulus. This is especially true for stimuli that activate NF-κB transiently.
Here we probed transcriptional responses to a transient NF-κB-inducing stimulus by combining data from kinetic analyses of RELA binding, RNA polymerase recruitment, transcriptional output, and perturbation of classical NF-κB activation. Using BJAB B lymphoma cells, we demonstrate that NF-κB/RELA is transiently recruited to nearly 3,000 sites genome-wide in response to pharmacological mimics of B cell antigen receptor activation. From these sites, we identified several hundred genes that were direct transcriptional targets of NF-κB. Most of these genes were not found in databases of putative NF-κB target genes. Though the majority of functional NF-κB target genes were up-regulated by RELA, we also identified genes whose expression was suppressed by RELA binding. RELA target genes displayed different transcriptional kinetics, and most recruited RNA polymerase II (Pol II) in response to cell activation. In querying the basis for kinetic differences, we found that late-activated NF-κB target genes required extracellular signal–regulated kinase (ERK) activity, whereas rapidly induced NF-κB target genes were marked by Pol II–containing loops in unactivated cells. Despite relatively short activation times used in our experiments, we also identified many “indirect” targets of NF-κB. These genes appeared to be regulated by NF-κB-induced transcription factors and thereby represented downstream effects of NF-κB activation. Consequently, transcriptional responses of such genes were delayed compared to direct NF-κB target genes. Taken together, our studies define the first steps of an NF-κB-responsive inducible gene cascade in activated B cells and highlight mechanisms by which kinetic patterns of NF-κB-dependent gene induction are established.
The availability of well-curated lists of NF-κB responsive genes and their time-dependent expression in response to activating stimuli is an essential prerequisite to elucidate transcriptional consequences of NF-κB activation. Because functional characterization of NF-κB responses in lymphocytes is especially scarce, we initiated studies to understand the kinetics of NF-κB-dependent inducible gene transcription in B lymphoid cells activated via the pharmacological equivalent of B cell antigen receptor signaling.
NF-κB inducibility in BJAB human B lymphoma cells closely paralleled that seen in primary B cells activated via the B cell receptor (BCR) (S1A Fig) [26]. Hallmarks of this response were rapid nuclear translocation of RELA followed by exit from the nucleus within 4–6 h after stimulation. Over a 4 h time course of phorbol 12-myristate 13-acetate (PMA) and ionomycin (P+I) treatment, approximately 1,000 genes were up-regulated and 1,000 genes were down-regulated more than 2-fold in BJAB cells (Fig 1A). We used k-means clustering based on correlation as the distance metric to partition up- and down-regulated genes into 6 distinguishable patterns (Fig 1B, S1B Fig). Hierarchical clustering of RNA expression profiles also revealed categories similar to those determined by k-means analysis (S1C and S1D Fig). Additionally, Gene Ontology (GO) analyses revealed distinct biological functions of genes in each category of up-regulated genes (S1E Fig). Expression of the largest subset of genes increased over the 4 h time course (Fig 1B, patterns 1A, 3A, and 4A). Smaller subsets of genes were rapidly up-regulated at 1 h and then either leveled off at 4 h or were subsequently down-regulated at 4 h (Fig 1B, patterns 2A, 5A, and 6A). Analogously, most genes that were down-regulated by P+I decreased continuously from 0 to 4 h. Examples from these categories are shown in Fig 1C. Amongst these diverse patterns of altered expression, our goal was to identify genes that were responding directly to NF-κB activation.
To identify genes that bound inducible NF-κB, we carried out ChIP-Seq using anti-RelA antibodies with unactivated cells or cells activated with P+I for 1 or 4 h. We focused only on those ChIP-Seq peaks that exceeded a threshold peak score of 100 (after peak annotation in HOMER) and were replicated in 2 independent ChIP-Seq experiments (S1F Fig). We reasoned that these stringent criteria would increase focus on robust RELA interactions genome-wide, despite decreasing the total numbers of peaks being studied. We identified 345 RELA binding sites prior to cell activation. The number of RELA-bound sites increased to nearly 3,000 after 1 h of activation and thereafter fell back to approximately 600 sites at 4 h (Fig 1D), demonstrating that genome-bound RELA closely paralleled total nuclear RELA levels. Sequences related to the κB motif (recognition site of NF-κB) were enriched at sites of RELA binding in all conditions (S1G Fig). Most inducible RELA binding at 1 h occurred in parts of the genome annotated as introns and intergenic regions (Fig 1D), a tendency that was noted in previous studies. We identified approximately 500 gene promoters that were newly targeted by RELA in activated cells.
As a first step towards identifying functional targets of RELA, we used HOMER to associate RELA peaks with genes. Peaks that were located outside annotated gene promoters and introns usually fell within 50 kb of the transcriptional start sites (TSSs) of assigned genes [20]. Applying these criteria to the approximately 1,000 genes whose RNA levels increased ≥2-fold with activation, we found inducible RELA bound to 354 genes; conversely, RELA was associated with 201 (out of 900) genes whose expression decreased ≥2-fold upon activation (Fig 1E). The majority of RELA-binding up-regulated genes increased in expression between 1 and 4 h of activation (Fig 1B, indicated in red). Conversely, expression of most RELA-binding down-regulated genes also decreased between 1 and 4 h, the period during which nuclear RELA levels were falling. Within the group of RELA-binding up-regulated genes were recognizable NF-κB target genes (such as NFKBIA, TNFAIP3, and RELB), as well as others that had not been previously associated with NF-κB (such as RILPL2) (Fig 1F).
The 354 up-regulated and 201 down-regulated genes (such as CYTH4, Fig 1F) with inducible RELA binding constituted a working list of putative NF-κB target genes in activated B cells. Only 106 out of 354 up-regulated genes and 36 of 201 down-regulated RELA-binding genes identified in our analysis were present in a list of 1,992 putative NF-κB-responsive genes compiled from the “NF-κB Target Genes” list maintained by Thomas Gilmore’s lab (http://www.bu.edu/nf-kb/gene-resources/target-genes/), NF-κB target gene sets (https://www.yumpu.com/en/document/view/8327926/the-nfkb-target-gene-sets-are-listed-below-broad-institute), and other recent publications [24,27]. The newly identified 248 up-regulated and 165 down-regulated putative NF-κB targets are shown in S1 Table.
To directly identify functional targets of inducible RELA (that is, genes whose transcriptional changes depended on RELA binding), we attenuated classical NF-κB activation by expressing a degradation-resistant, mutated dominant negative IκBα (dnIκBα) [28,29]. This form of IκBα is expected to quench the release of NF-κB proteins from all cytosolic IκBs via the posttranslational pathway. For this, we generated 2 clones of BJAB cells in which dnIκBα could be induced by tetracycline (Tet) treatment (Fig 2A). In these clones, nuclear RELA induction in response to P+I was similar to that of control BJAB cells in the absence of Tet but was abolished in cells that had been pretreated with Tet for 24 h (Fig 2B). To determine the effects of dnIκBα expression on inducible gene expression, each clone was either pretreated with Tet for 24 h (to induce dnIκBα) or not, followed by activation with P+I for 0, 1, or 4 h. Replicate experiments were carried out with each clone, and total RNA was prepared for RNA-Seq. Basal gene expression was not affected substantially in the presence or absence of Tet (S2A and S2B Fig). We compared inducible gene expression in each clone in the presence or absence of Tet and focused only on those inducible genes whose response to dnIκBα was replicated in both clones.
We identified 806 genes whose inducible expression was significantly reduced (false discovery rate [FDR] ≤ 0.05) by dnIκBα at either 1 or 4 h (Fig 2C, middle) in both Tet-inducible clones. Of these, 304 genes inducibly bound RELA in our ChIP-Seq experiments and were therefore considered to be direct transcriptional targets of RELA. These genes varied in their kinetic responses to cell activation (Fig 2C, right and S2C Fig) and were enriched for NF-κB binding motifs at sites of RELA binding as well as in their promoters (S2D Fig). We found that NF-κB target genes with different kinetic expression patterns enriched for different biological functions (S2F Fig). GO analyses showed that transiently activated genes (pattern 3Ad) were associated with processes such as “negative regulation of cellular processes,” “leukocyte activation,” and “inflammatory response.” By contrast, processes that scored high among genes whose expression continued to increase between 1 and 4 h activation (patterns 2Ad and 4Ad) included “ribosome biogenesis,” “immune response,” regulation of “Type 1 interferon production,” and “cellular response to cytokines.” These observations indicate that kinetic patterns were associated with distinct functional categories of NF-κB target genes. Additionally, de novo motif analysis in HOMER revealed distinct transcription factor motifs associated with RELA peaks in different kinetic patterns. RELA peaks in patterns 2Ad and 4Ad genes were enriched for the κB motif as well as the motif for transcription factor AP1, whereas the latter motif was not evident in RELA peaks of pattern 3Ad genes (S2E Fig). This list of 304 NF-κB target genes included many that had not been previously identified as being NF-κB responsive (S2 Table).
To probe the NF-κB response, we focused on 130 of the 304 direct target genes that were induced more than 2-fold in the absence of Tet (S2 Table). Out of these 130 genes, 74 have not been previously categorized as NF-κB targets. We found that virtually all transiently induced genes were amongst these 130 most robustly induced, RELA-binding, and dnIκBα-sensitive genes (Fig 2C, right, red numbers). This category included genes such as TNFAIP3 and HERPUD1 (Fig 2D, left). The prevalence of genes with this expression profile was consistent with reports showing that many NF-κB target genes have short-lived mRNAs [30]. For the majority of these genes, RELA binding occurred close to TSSs (S4D Fig). Second, most (95 out of 130) of these genes were contained in the set of 354 putative target genes identified in Fig 1. The remaining 259 (out of 354) genes that were not substantially affected by dnIκBα, despite robust inducible RELA binding and altered gene transcription, reaffirmed the idea that inducible RELA binding and inducible transcription are insufficient criteria to identify functional targets of NF-κB. Third, mRNA levels of many genes continued to increase within the time frame of our studies (Fig 2C, right, patterns 2Ad and 4Ad). These included genes such as STAT5A and NR1D1 (Fig 2D, right). We found that RELA was recruited to these genes at 1 h, but most of it was lost by 4 h. These observations indicated that continued increase in mRNA was RELA-independent, suggesting that RELA was required to initiate but not maintain transcription of these genes.
To further explore the basis for continued transcription of NF-κB target genes after chromatin-bound RELA was depleted, we drew upon the observation that the AP1 motif was enriched in RELA peaks associated with genes whose expression levels continued to increase between 1 and 4 h (patterns 2Ad and 4Ad, S2E Fig). We tested the possible involvement of this transcription factor family by pharmacologically inhibiting ERK, a mitogen-activated protein kinase (MAPK) required for activation of AP1-like factors. RNA isolated from BJAB cells treated with P+I in the presence or absence of the ERK inhibitor PD0325901 was assayed by quantitative real-time PCR (qRT-PCR) for expression of genes from patterns 2Ad and 4Ad. We found that inducible expression of 3 genes from these categories was suppressed by ERK inhibition (Fig 3A, top line). To rule out that PD0325901 affected NF-κB activation by some unanticipated pathway, we also assayed genes whose expression kinetics coincided with RELA induction by P+I (pattern 3Ad in Fig 2C). These genes were not substantially affected by PD0325901 (Fig 3A, lower line). We conclude that ERK-dependent transcription factors confer continued transcriptional activity to a subset of NF-κB target genes after induced nuclear RELA levels dissipate. The kinetically delayed response of these genes is consistent with a “priming” role for RELA, followed by an activator role for AP1-like factors. Such priming may involve recruitment or stabilization of additional transcription factors or coactivators by transiently bound RELA.
RELA did not bind the remaining 502 genes whose expression was reduced by dnIκBα in both clones (Fig 2C, middle, gray). Transcriptional responses of these genes in the presence or absence of dnIκBα also clustered into patterns similar to those seen for direct target genes, including 2 in which gene expression was inducibly down-regulated upon cell activation (Fig 2C left, S3A Fig). A pattern that was prominently missing in this gene set compared to direct RELA targets was one in which RNA levels increased transiently in response to activation. To understand the basis for these genes being affected by dnIκBα in the absence of RELA binding, we looked for shared transcription regulatory features in this set. HOMER analysis of promoter regions of these genes revealed an enrichment for the binding motif of transcription factor MYC (S3B Fig). The MYC motif was found in the promoters of 247 of these 502 genes, including BRIXI, DDX18, and AKAP1, and the majority of these genes (226 out of 247) were previously shown to bind MYC in ChIP-Seq assays [31]. Because MYC is a known target of NF-κB (S3C Fig), we hypothesized that this set of genes was induced by NF-κB-activated transcription factors. Among 304 direct RELA target genes, we found 37 that encoded DNA-binding transcriptional regulators (S3D Fig, left). These included genes for KLF10 and IRF1 that were previously linked to NF-κB. We also found many other genes encoding factors such as HES1 and ZNF267 that had not been previously associated with NF-κB (Fig 3B). We confirmed dnIκBα-sensitive RELA recruitment to promoters of these genes by chromatin immunoprecipitation (ChIP) followed by quantitative PCR (qPCR) (S3E Fig). Thus, many transcription factor genes were induced in activated cells via NF-κB. To further substantiate the hypothesis that NF-κB-induced transcription factors contributed to dnIκBα-sensitive gene expression, we evaluated whether promoters of indirect NF-κB target genes contained recognition motifs for NF-κB-regulated transcription factors.
For the promoter analysis, we focused on 78 (out of 502) genes (S3 Table) that were dnIκBα-sensitive, did not bind RELA, and were changed more than 2-fold by P+I treatment in the absence of Tet. We searched for transcription factor motifs that were present in more than 20% of these promoters (S3F Fig). This analysis revealed recognition sites for kruppel-like factor (KLF), zinc finger protein (ZNF), and ETS-domain transcription factors. The correspondence between transcription factor genes induced by NF-κB and motifs enriched in promoters of indirect NF-κB target genes supports the notion that the set of dnIκBα-sensitive genes that did not bind RELA were targets of transcription factors activated by NF-κB. However, in the present study, we did not directly evaluate binding of such NF-κB-induced transcription factors genome-wide. We will refer to such genes as indirect targets of RELA. In keeping with their proposed dependence on NF-κB-induced transcription factors, RNA levels of such indirect targets increased at later times compared to direct targets (Fig 3C).
GO analyses of the dominant patterns (2Ai and 4Ai) of indirect NF-κB target genes showed that they were enriched for genes involved in RNA processing, ribosome biogenesis, and RNA-associated metabolic processes (S3G Fig). These processes were largely distinct from those associated with direct NF-κB target genes, thereby identifying a hierarchy of biological consequences associated with NF-κB activation. Because MYC has been implicated in activating ribosomal genes [32], we surmise that many of the identified processes are the consequence of NF-κB-directed MYC expression in activated BJAB cells. Taken together, our kinetic analyses identified gene targets at which inducible RELA binding activated transcription (direct targets) and, additionally, revealed secondary transcriptional consequences of NF-κB activation in B lymphoblastoid cells via NF-κB-induced transcription factors (indirect targets). We also uncovered a mechanism by which RELA induced persistent transcriptional activity despite its transient nuclear induction.
We identified 263 gene transcripts that were up-regulated by dnIκBα expression in both Tet-inducible BJAB clones. Eighty-five of these genes bound RELA in activated cells (Fig 4A, green circle); our interpretation is that RELA binding reduced expression of this subset of genes. Such RELA-repressed genes had diverse expression profiles, including genes that were up- or down-regulated in response to activation (Fig 4A right, patterns 6Rd and 1Rd, respectively, S4A Fig left). Some examples are shown in Fig 4B (see also S4B Fig for complete RNA time courses). Of the 53 (out of 85) RELA-repressed genes whose expression changed more than 2-fold in response to activation in the absence of Tet, 36 were not found in NF-κB-related databases and thus represent novel targets of NF-κB activity (S4 Table). In contrast to RELA-activated genes that contained canonical κB motifs within RELA peaks, sequence motifs underlying RELA peaks of RELA-repressed genes were enriched for AP1 binding sites (S4C Fig). Additionally, RELA binding was scattered throughout these genes rather than being enriched in promoter regions (S4D Fig). Our interpretation is that RELA was recruited to these regions primarily by association with DNA-bound AP1 factors rather than direct DNA binding by RELA itself. Such interactions may attenuate transcriptional activation by AP1, thereby resulting in gene repression.
In total, 178 genes up-regulated by dnIκBα expression did not bind RELA (Fig 4A, left; Fig 4C). We surmised that up-regulation of these genes by dnIκBα was also a secondary consequence of NF-κB activation. That is, such genes were either activated by transcription factors that were up-regulated by dnIκBα expression or attenuated by factors that were direct or indirect RELA targets. We found examples of each category in our RNA-Seq database. Among the 85 RELA-repressed genes, we found 23 that encoded transcriptional regulators (S3D Fig) whose increased expression in the presence of dnIκBa could be responsible for activating a subset of up-regulated genes that did not bind RELA. To further identify factors that indirectly activated gene transcription by dnIκBa, we screened promoter motifs present in 57 (out of 178) genes whose expression changed more than 2-fold in the absence of Tet. Most of these overlapped with motifs present in genes that were indirectly activated by NF-κB (S5 Table, S4E and S4F Fig). However, a few motifs were selectively associated with dnIκBα-activated genes, such as those for signal transducer and activator of transcription (STAT) and SRY-box (SOX) factors, and those for nuclear hormone receptors (S4F Fig). As shown above, STAT5 is a direct target of NF-κB in these cells and may negatively regulate a subset of these indirect RELA-repressed genes. We also found SOX8 and PPARG mRNAs to be up-regulated by P+I in dnIκBα-expressing cells (S4G Fig); however, the associated mechanism(s) have not been further addressed in this study. We conclude that NF-κB activation also initiates a cascade of transcriptional down-regulation, both by directly interacting with a subset of genes and by modulating expression of other transcription factors.
GO analysis of RELA-repressed genes revealed some interesting features. As noted for RELA-activated genes, there was relatively little overlap between the 6 patterns for the top 10 biological processes (S4H Fig). Among genes that were directly repressed by RELA, we found 1 pattern (6Rd) to enrich for genes involved in transcription termination by Pol II. One interpretation is that NF-κB proteins elevate gene expression by both activating transcription initiation and inhibiting transcription termination. The latter mechanism may apply to genes for which NF-κB has been proposed to push prebound RNA polymerase from abortive initiation state to productive elongation mode. Other pathways that featured in this gene set included modulation of biosynthetic and metabolic processes and negative regulation of cellular processes. Prominent among genes that were indirectly repressed by RELA were those involved in autophagosome organization and assembly and posttranslational protein modifications (S4I Fig). By up-regulating essential autophagy genes such as ATG5 and 7 [33] and suppressing others involved in autophagosome assembly, NF-κB may fine-tune the autophagic response. The emerging patterns reveal synergistic use of RELA-dependent activation and suppression of gene expression to optimize cellular responses.
The preceding analysis started by identifying genes whose inducible expression changed significantly in the presence of dnIκBα. While focusing on functional targets of RELA, this approach did not fully utilize our time-dependent RNA analyses. In particular, we missed out on the broader genomic landscape of NF-κB recruitment in response to B cell activation, especially where RELA binding did not affect mRNA levels after dnIκBα induction. Such sites were of potential interest because they vastly outnumbered those where RELA binding had functional consequences and may therefore contribute to B cell biology in unanticipated ways. We did this in 2 steps. First, we identified genes whose expression did not change in both dnIκBα-inducible clones across all time points of activation with or without Tet. Approximately 600 out of 8,000 such genes inducibly bound RELA (S5A Fig). These binding sites were associated with canonical κB and AP1 motifs (S5B and S5C Fig). Second, we used k-means clustering with correlation parameter to visualize gene expression patterns of the remaining genes in each clone in the absence or presence of dnIκBα (S5D Fig). Patterns with similar expression characteristics in both clones were combined into 4 patterns of inducible gene expression (S5E Fig).
Patterns I and II corresponded to genes whose expression decreased or increased, respectively, in response to dnIκBα expression. Analysis of these gene sets largely recapitulated the conclusions from Figs 2–4. Patterns III and IV provided insights into patterns of inducible expression that were not identified in the preceding analysis. These groups contained genes that were either up- (Pattern III) or down-regulated (Pattern IV) with activation but whose expression was not affected by dnIκBα (S5E and S5F Fig). Numerous genes in each set bound RELA (green circles), and NF-κB and AP1 binding sites were again enriched in sequences underlying RELA peaks (middle column). These observations reinforced the idea from Figs 1 and 2 that inducible binding coupled with transcriptional changes was an insufficient criterion to identify functional targets of transcription factors. Differential regulation of these gene sets was also evident from their promoter architecture. Binding motifs for ETS-domain proteins and the transcription factor Yin Yang 1 (YY1) were enriched in gene promoters that were up-regulated with activation (S5E Fig, right column), whereas the motif for IRFs was enriched weakly amongst genes that were down-regulated with activation.
Rapid gene induction in response to cell stimulation is programmed in different ways. In the classic example of c-Fos induction, phosphorylation of a promoter-bound transcription factor in unactivated cells triggers RNA synthesis after cell activation [34,35]. In other instances, RNA polymerases bound at promoters in unactivated cells can be pushed into elongation mode by phosphorylation of their C-terminal domain in response to stimuli [36,37]. This mode of activation has been implicated at some NF-κB-dependent target genes [38–40]. To gain more insight into inducible gene expression by RELA, we carried out ChIP-Seq with antibodies directed against Pol II. We used a threshold peak score of ≥100 in HOMER and reproducibility in replicate experiments to assign Pol II occupancy with confidence (S6A Fig). Using these criteria, we found that 50 out of 130 direct NF-κB target genes contained prebound Pol II at their promoters prior to activation (S6B Fig, S6A Table). However, these genes recruited additional Pol II after cell activation, which was evident in the average profile across all direct target genes (S6C and S6D Fig). Proportionally fewer indirect target genes (13 out of 78 genes) had prebound Pol II prior to P+I treatment, and inducible Pol II recruitment was clearly evident at these promoters in response to activation (S6E Fig). We conclude that Pol II recruitment is a major mechanism of inducible gene transcription by NF-κB.
Presence or absence of Pol II at the basal state did not readily explain kinetic differences in patterns of NF-κB target gene induction. To further probe for a possible mechanism, we performed chromatin interaction analysis by paired-end tag sequencing (ChIA-PET) in cells prior to stimulation. This assay scores for interaction of Pol II–bound sequences with other parts of the genome [41,42]. Biological replicates were processed using ChIA-PET tool software [43], and the data were divided into 4 groups (Fig 5A). Over the entire dataset, genes that lacked Pol II had the lowest RNA levels at baseline, genes with Pol II–bound promoters with no loops had intermediate RNA levels, and genes whose Pol II–bound promoters engaged in looping interactions had the highest levels of RNA (S6F Fig). Similar trends were observed in previous ChIA-PET studies [42]. Additionally, we confirmed several looping interactions that had been identified in earlier studies, indicating that our assay scored for functionally relevant Pol II–associated interactions (S6G Fig).
We then examined chromatin interactions in the context of NF-κB target genes. Approximately half of the 130 robustly induced (≥2-fold) direct RELA target genes had Pol II loops in unactivated cells (Fig 5B, left; S6B Table). On average, RELA target genes with preformed loops reached close to maximum levels of inducible expression rapidly compared to those without loops (Fig 5B, middle), and most of the transiently induced target genes identified in Fig 2 (22 out of 30) fell in this category (Fig 5B, right, pattern 3Ad). Conversely, RELA target genes that did not have preformed loops were enriched for genes that were induced more slowly and whose expression increased continuously over the time course of activation (Fig 5B, right, 34 out of 54 genes in pattern 2Ad and 23 out of 43 genes in pattern 4Ad). One example of a prelooped RELA target gene is shown in Fig 5D (left).
For indirect target genes, the trends were reversed (Fig 5C). Approximately 30% of genes in this category (24 out of 78 robustly induced genes) had preformed Pol II loops (S6B Table). Looped genes in this set had higher basal RNA levels but did not show kinetic differences in RNA induction compared to nonlooped genes (Fig 5C, middle). Instead, indirect target genes that had preformed loops in unactivated cells achieved higher levels of induced RNA compared to genes with no loops. Many indirect target genes with preexisting loops tended to be induced early in the presence of dnIκBα but crashed thereafter (18 prelooped genes in patterns 3Ai, 4Ai in Fig 5C, right). One example of a prelooped indirect RELA target gene is shown in Fig 5D (right). We propose that preformed loops regulate kinetic patterns of direct RELA target genes, whereas they determine the maximal RNA output for indirect target genes.
While RELA target genes were typically involved in single promoter interactions (Fig 5A, pattern III), we also found approximately 1,000 genes that were involved in multiple promoter interactions (ChIA-PET category 4). These genes yielded GO terms such as “purine triphosphate metabolic process,” “pyrimidine nucleotide biosynthetic process,” and other comparable metabolic pathways (S6H Fig). Amongst these, we found interactions involving PPP4C, ALDOA, and histone genes (S6I Fig). It is likely that linking gene promoters via Pol II interactions provides a mechanism to coregulate genes that are involved in a common biological pathway [42]. In contrast, the need for greater flexibility in output of RELA-responsive genes depending on the stimulus and cell type may preclude their connection in a preformed network.
By combining time-dependent transcriptional responses with genome-wide recruitment of RELA and Pol II and perturbation of classical NF-κB activation, we sought to identify mechanisms by which kinetic patterns of NF-κB-dependent gene expression are established. In these studies, a pharmacologic equivalent of antigen receptor signaling was used to activate human B lymphoblastoid BJAB cells over a time course during which nuclear NF-κB was transiently induced. Three interesting features emerged from a consideration of RELA target genes. First, close to 60% of the 130 most robustly induced target genes identified here had not been previously connected with NF-κB responses. Our list also contained well-established NF-κB targets such as MYC, TNFAIP3, and NFKBIA, attesting to the validity of our analyses. The incompleteness of current lists of NF-κB target genes was further accentuated when we included the additional 174 targets identified here that were induced less robustly. Of these 174 genes, 80% were not present in NF-κB-related databases, while the remaining 20% included genes such as TP53, TNIP1, and TAP1 that were previously linked to NF-κB. We surmise that NF-κB target genes identified here that are also present in earlier lists may represent more “universal” targets that respond regardless of cell type or stimulus. In contrast, genes uniquely identified in our study may represent cell type–or stimulus-specific responses. While use of a lymphoma cell line for these studies makes it difficult to draw direct connections to transcriptional responses of primary human B cells, the hundreds of new functionally curated NF-κB target genes identified here constitute a unique pool of possible mediators of NF-κB activity in B lymphoid cells.
We also identified many genes that we refer to as indirect NF-κB targets. Such genes were sensitive to dnIκBa expression but did not bind RELA. We hypothesize that such genes were activated (or repressed) by NF-κB-induced transcriptional regulators. Even within the relatively short time course of our kinetic studies, we identified 37 genes among the 304 direct targets that encoded transcriptional regulators. In addition to previously identified targets such as MYC and IRF1, this list included many new NF-κB-regulated factors such as hes family bHLH transcription factor 1 (HES1) and ZNF267. Additional studies are needed to directly evaluate the contribution of such factors in regulating indirect target gene transcription. It is possible that a subset of genes that we classified as indirect RELA targets are controlled by RELA bound to sites that did not score in the program used to connect binding sites to genes.
From the studies presented here, we cannot specify the subunit composition of RELA-containing homo- or heterodimers that activate transcription of the identified NF-κB target genes. Because most of the RELA genome binding occurred at 1 h after cell activation, our working hypothesis is that functional NF-κB measured in these assays was generated from cytosolic pools by the classical posttranslational pathway. By electrophoretic mobility shift assays, this form consists largely of p50/RELA heterodimers; however, sequential ChIP is required to verify this model. Additionally, the contribution of REL to dnIκBa-sensitive gene transcription was not experimentally evaluated by depleting REL in BJAB cells. Studies to address this question are in progress using Rel-deficient murine B cells.
Second, 2 dominant kinetic patterns of inducible expression emerged for the 130 genes that were most strongly induced. A small number (30 out of 130) were transiently induced and included genes such as TNFAIP3 and NFKBIA. The expression pattern of such genes can be easily explained by transcriptional activation when bulk RELA is nuclear (at 1 h), followed by transcriptional inactivation when RELA moves back into the cytoplasm (at 4 h), together with rapid degradation of the encoded mRNAs. The short half-life of many of these transcripts has been previously highlighted [30]. More surprising was the observation that inducible expression of the majority of these genes (97 out of 130) continued to increase between 1 and 4 h of activation. This time period coincided with down-regulation of RELA from the nucleus, and indeed, we found that RELA was lost from gene promoters over this period. For a subset of genes that we tested, ERK activity was required for sustained RNA synthesis after loss of nuclear RELA, pointing to involvement of the AP1 family of transcription factors. Our working hypothesis is that RELA binding “primes” the promoter for subsequent binding and transcriptional activation by ERK-dependent transcription factors. However, continued RELA binding is not required for transcriptional activity, thus distinguishing this mode of gene regulation from synergistic promoter activity by cobound factors.
Our observations also provide a novel perspective on the phenomenon of assisted loading, a term used to describe cooperative recruitment of transcription factors that co-occupy gene regulatory sequences [44]. Regarding NF-κB/RelA, it has previously been shown that binding of IRF5 or STAT3 to a subset of genomic sites in activated hepatocytes or macrophages, respectively, requires simultaneous RELA activation [5,45]. In the examples presented here, we show instead that RELA does a hit-and-run on gene promoters that have a characteristic kinetic transcription profile. Although RELA is lost from these promoters, our experiments do not distinguish whether the κB site remains empty or is occupied by other factors. For example, p50 homodimers associated with IκBξ, which have been previously proposed to confer transcriptional activity [46], may substitute RELA-containing complexes at such sites. Alternatively, other proteins that recognize κB motifs may provide transcriptional activity in the absence of RELA. Nuclear factor of activated T cell (NFAT) proteins are a distinct possibility because they are induced in P+I-activated B cells [47], have been shown to bind to κB elements [48], and function in collaboration with AP1 factors.
Earlier studies have analyzed the connection between AP1/ERK and kinetics of gene induction by NF-κB. Natoli and colleagues showed that a subset of inflammatory gene promoters recruited RELA in response to LPS only after being marked by serine phosphorylated histone H3 (H3S10P) via p38 MAPK activity [49]. The 2-step process delayed RELA recruitment and transcriptional induction of these genes compared to other inflammatory genes to which RELA bound without requiring H3S10P. By contrast, RELA recruitment was not delayed even at late-induced genes in the studies described here, thereby invoking a novel mode of intersection between NF-κB and MAPK pathways. We note several differences that may underlie mechanistic variations observed for NF-κB-inducible transcription in the 2 studies. These include the different cell types used (dendritic cells versus B cells) that could differentially mark RELA recruitment sites, different initiating stimuli (LPS versus P+I) that induce distinct cytosolic signaling milieus, and the nature of NF-κB activation (sustained versus transient) that could influence gene expression outcomes. Further studies are needed to understand rules by which NF-κB tunes cellular responses to diverse stimuli.
In a more recent study, Brasier and colleagues identified AP1 and SP1 motifs in regions surrounding RELA peaks in TNFα-induced A549 (human pulmonary epithelial) cells [14]. Genes with SP1 motifs reached maximal inducible expression 30 min after activation, whereas those with AP1 motifs reached maximal levels 60 min after activation. Two interesting features emerged from a comparison of our data with those of Yang and colleagues [14]. First, de novo motif analysis did not reveal SP1 sites near RELA peaks of our most rapidly induced genes (Fig 2, pattern 3Ad). Thus, rapid gene induction by NF-κB comes in different flavors. One possibility is that SP1 and NF-κB cooperate at promoters where inducibility via NF-κB is coupled with relatively high basal-level expression via SP1. Second, in TNFα-treated A549 cells, RELA levels at a prototypical NF-κB/AP1 promoter continued to rise even when RNA levels were falling. By contrast, we found that at ERK-sensitive genes in activated BJAB cells, RELA levels fell, while RNA levels continued to rise. These distinctions yet again emphasize the variety of ways in which kinetic patterns of NF-κB-dependent transcription are achieved.
Third, half of the strongly induced direct target genes had preformed Pol II–containing loops in unactivated cells. Genes that contained such loops were induced more rapidly on average than genes without loops and reached close to maximal expression levels at 1 h post activation. In contrast, RNA levels of unlooped genes continued to increase in the interval between 1 and 4 h. We propose that kinetic patterns of NF-κB-dependent transcription are determined in part by a poised state reflected in such preformed loops. Interestingly, most of the transiently induced targets (22 out of 30) fell in the looped category, likely reflecting the need for these genes to reach maximum expression as soon as possible while RELA is still in the nucleus. These genes were also more evolutionarily conserved than prelooped genes that were induced more slowly (S6K Fig)[50]. However, several transiently induced genes (such as TNFAIP3 and NFKBIA) did not have looped configurations in unactivated cells. One possibility is that these genes have “simple” NF-κB-dependent promoters that do not require interactions with distal regulatory sequences to modulate expression levels. Hao and Baltimore recently demonstrated that genes that are rapidly transcriptionally induced undergo rapid splicing to produce cytoplasmic mRNA [30]. Such mRNAs are also relatively unstable, resulting in transient gene induction. Five out of 7 genes that were shared between our dataset of transiently induced NF-κB target genes and that of Hao and Baltimore were found to have looped configurations in unactivated BJAB cells. Thus, a prelooped configuration may also assist in increasing splicing efficiency of rapidly induced genes. We note the caveat that the transformed state of BJAB cells may affect the distribution of genes with preformed loops.
We also identified many genes whose inducible expression increased in the presence of dnIκBα. A subset of these genes bound RELA in our ChIP-Seq analyses, suggesting that RELA binding attenuated transcription of these genes. Identification of AP1 motif as the prominent sequence at sites of RELA binding leads us to hypothesize that RELA is recruited by protein–protein interactions with transcription factors bound at these sites rather than by DNA binding. In doing so, RELA may reduce transcription activation function of the DNA-bound factor. Interestingly, genes encoding several AP1 motif-binding factors, such as FOS, FOSB, MEF2B, and MEF2C, were in this set of RELA-repressed genes, possibly indicating some form of regulatory feedback. Dual-specificity phosphatase 1 (DUSP1), a regulator of ERK activity, was also in this list, again suggesting cross talk between NF-κB and AP1 signaling cascades. At other genes, DNA-bound RELA might interfere with the progression of RNA polymerases, thereby reducing transcriptional output. Though the mechanism and functional importance of RELA-dependent down-regulation of gene expression remain largely speculative at this time, our studies highlight a mode of gene regulation by this transcription factor that has been largely overlooked.
BJAB cells were cultured in RPMI 1640 medium supplemented with 10% FBS (HyClone), Penicillin-Streptomycin-Glutamine (Invitrogen), and 2-mercaptoethanol. For activation, cells were exposed to 50 ng/ml PMA and 2 μM ionomycin (Sigma-Aldrich) for 1 and 4 h. To inhibit ERK signaling, cells were pretreated with 0.33 nM PD0325901 (Selleckchem) for 1 h before P+I stimulation. To generate dnIκBα-inducible clones, BJAB cells were transfected with pcDNA6/TR (Invitrogen) to express Tet repressor (TetR), and stable clones were selected in 15 μg/ml blasticidin (Invitrogen) for 6 d. Stable single clones with the highest levels of TetR expression were subsequently transfected with full-length dnIκBα (S32A and S36A) cloned into pcDNA4/TO. Stable clones were selected in the presence of both blasticidin and 600 μg/ml zeocin (Invitrogen) for 6 d. dnIκBα expression was induced with 1 μg/ml Tet (Invitrogen) for 24 h before P+I treatment. All cell lines were maintained at 37°C with 5% CO2.
The following antibodies were used for ChIP, ChIP-Seq, or western blot: RelA (sc-372, Santa Cruz); Pol II (Covance Cat # MMS-126R); hnRNP (sc-32301, Santa Cruz); IκBα (sc-371, Santa Cruz); β-actin (sc-47778, Santa Cruz). Horseradish peroxidase–coupled goat anti-mouse IgG and goat anti-rabbit IgG (Santa Cruz) were used for immunoblotting.
Proteins were separated by electrophoresis through 10% SDS-PAGE and electrophoretically transferred to nitrocellulose membrane (Millipore). After blocking with 5% dried milk in Tris-HCl-buffered saline/0.05% Tween (TBST) for 1 h, membranes were incubated with primary antibodies overnight, washed in TBST, and incubated for 1 h with horseradish peroxidase–coupled secondary antibodies (Santa Cruz). Proteins were detected by using the enhanced chemiluminescence (ECL) systems (Pierce) and Syngene Imaging System.
ChIP experiments were performed as described by [51]. Briefly, cells were washed twice with PBS and then were fixed at room temperature with either 1% formaldehyde in PBS for 10 min (for Pol II ChIP) or 1.5 mM EGS (Pierce Cat # 21565) for 30 min followed by 1% formaldehyde (Sigma-Aldrich) at room temperature for 15 min in PBS (RELA ChIP). Reactions were quenched by adding glycine to a final concentration of 0.125 M, and cells were washed twice with cold PBS. Nuclei were isolated and lysed in buffer containing 50 mM Hepes-KOH, pH 7.5; 150 mM NaCl; 1 mM EDTA; 1% Triton X-100; 0.1% sodium deoxycholate; 0.1% SDS; and protease inhibitors. The crosslinked chromatin was subjected to fragmentation by sonication (Branson Sonicator). ChIP was performed with 2.5 μg anti-RelA antibody (Santa Cruz) and 2 μg anti-Pol II antibody (Covance) prebound to 50 μl Protein A or G Dynabeads (Invitrogen). Sonicated chromatin was added to antibody-bound beads and incubated at 4°C overnight. Beads were collected by centrifugation, washed, and incubated at 65°C for 4 h in elution buffer (50 mM Tris-HCl, pH 7.5; 10 mM EDTA; 1% SDS) to reverse cross-linking. ChIP DNA was purified by phenol-chloroform extraction followed by ethanol precipitation. For qPCR quantitation of ChIP, the signal from gene-specific amplicon was compared to an amplicon from the H19 locus according to the equation RE = 2- (CT(target gene)-CT(H19). Primer sequences are listed in Table 1.
For sequencing, adapters were ligated to the precipitated DNA fragments or the input DNA to construct a sequencing library according to the manufacturer’s protocol (Illumina, San Diego, CA, United States). Adapters with a T overhang were ligated to the DNA fragments and size selected (approximately 200–350 bases) on a 4.5% agarose gel. Eighteen cycles of PCR amplification were performed to enrich for fragments with an adaptor on both ends. These samples were bound to an Illumina single-read Flowcell, followed by cluster generation on the Illumina Cluster Station and sequencing with Illumina Genome Analyzer (GA-II). Two biological replicate ChIP-Seq experiments were carried out with each antibody. ChIP-Seq data are available on the GEO website (http://www.ncbi.nlm.nih.gov/geo/) (Accession number GSE117259).
Bowtie2 software [52] was used to map quality-filtered reads from demultiplexed FASTQ files to human genome assembly GRCh37/hg19 with the default options. RELA ChIP-Seq peaks were called using standard parameters in MACS 2.1.0 [53] with input as the control and activated samples as the treatment. Peaks were called at an FDR ≤ 0.05. Peak annotation and motif finding were carried out with HOMER (http://homer.ucsd.edu/homer/). The HOMER program annotatePeaks.pl was used to annotate peaks with default parameters (promoter regions were defined from −1 kb to +100 bp). In our analysis, most intergenic peaks were located within 50 kb of TSSs. Two biological replicate experiments were carried out, and peaks with peak score ≥ 100 that were common to both replicates were used for all further analysis. All samples were normalized to 10 million reads for visualization. The programs findMotifsGenome.pl and findMotifs.pl were used to identify transcription factor binding motifs within peaks or promoter regions (−400 to +100 bp from TSSs). The program findGO.pl was used to assess the enrichment of various categories of gene function (GO).
Total RNA was extracted (2 × 106 cells) using the Qiagen RNeasy Mini Kit (Qiagen). cDNA was synthesized with the SuperScript First Strand Synthesis System (Invitrogen Life Technologies). RT-PCR was performed in duplicates using the ABI PRISM 7000 (Applied Biosystems, Carlsbad, CA, US). Expression of NFKB1, IL4I1, MAPK6, RND1, TNFAIP3, and NFKBID was normalized to GAPDH mRNA on the same PCR plate. Relative expression (RE) of individual genes was calculated by the equation RE = 2- (CT(target gene)-CT(GAPDH)). Primer sequences are listed in Table 1.
Total RNA purified from BJAB and transfected derivatives was used for bar-coded library preparation and sequencing at the Johns Hopkins Deep Sequencing & Microarray Core.
Two independent dnIκBα-inducible single-cell clones were treated with Tet for 24 h or not, prior to activation with P+I for different times. This experiment was performed twice for each clone, and RNA was prepared for sequencing.
RSEM [54] was used to align RNA-Seq reads to the human genome and to quantify transcript abundance. EBSeq [55] was used to compare the aligned reads from multiple conditions to find differentially expressed genes using a cutoff of FDR ≤ 0.05. k-means analysis of RNA expression data was carried out in MATLAB using normalized read counts, with correlation as the distance metric, the number of times to repeat clustering set to 5, and other parameters set to default. All samples were normalized to 1 million aligned reads for visualization. Integrative analysis of gene expression in relation to ChIP-Seq data was done by ngs.plot (https://github.com/shenlab-sinai/ngsplot). RNA-Seq data are available on the GEO website (http://www.ncbi.nlm.nih.gov/geo/) (Accession number GSE117259).
Heatmaps of gene expression in S1C Fig (left), S1D Fig (left), S2C Fig (left), S3A Fig (left), and S4A Fig were generated using the package “gplots” in R program (https://CRAN.R-project.org/package=gplots) by log2-transformed normalized read counts after adding a pseudocount of 1. Colors represent standardized gene expression for which each gene is standardized across samples to have zero mean and unit standard deviation. The row color bar marks the cluster membership of each gene from the previous k-means clustering results. In addition, hierarchical clustering was applied based on the standardized gene expression. The results are shown in S1C (right) and S1D (right) Fig, S2C Fig (right), and S3A Fig (right) as heatmaps in which the rows are reordered by the hierarchical clustering results, and the row color bar represents the k-means clustering results.
The silhouette for each gene in which the expression is up-regulated ≥2 fold (1,021 genes in Fig 1B) in BJAB cells is based on correlation as the distance metric [56]. The silhouette value ranges from −1 to 1, where a larger value means that the gene is better matched to its own cluster than the neighboring clusters. As a baseline, we permuted cluster memberships and calculated the silhouette, which is shown in S1B Fig as the random group.
RNA Pol II ChIA-PET was performed as previously described [41,42]. A total of 109 BJAB cells were treated with 1.5 mM EGS (Pierce Cat # 21565) for 30 min, followed by 1% formaldehyde at room temperature for 15 min and then neutralized using 0.2 M glycine. Chromatin was sheared by sonication, and anti-Pol II monoclonal antibody 8WG16 (Covance, MMS-126R) was used to enrich Pol II–bound fragments. A portion of ChIP DNA was eluted from antibody-coated beads for quantitation using Picogreen fluorimetry and for enrichment analysis using qPCR. Two biological replicate ChIP samples were used for ChIA-PET library construction [42].
ChIA-PET data analysis was carried out as described by [43] (https://github.com/GuoliangLi-HZAU/ChIA-PET_Tool). Final peak calling was done at FDR ≤ 0.05. After filtering for paired-end tag (PET) clusters ≥ 2 counts and FDR ≤ 0.05, approximately 6,000 loops per replicate were obtained. The overlap between two replicates was 80%; data analysis was carried out from 1 replicate. ChIA-PET data are available on the GEO website (http://www.ncbi.nlm.nih.gov/geo/) (Accession number GSE117259).
All RNA-Seq, ChIP-Seq, and ChIA-PET data were visualized by preparing custom tracks for the WashU EpiGenome Browser [57,58]. All analyses were performed on Biowulf or Helix computer clusters at NIH.
Evolutionary conservation of NF-κB binding sites was calculated using a method similar to Iwanaszko and colleagues [50]. Homologous genes were obtained from NCBI HomoloGene (https://www.ncbi.nlm.nih.gov/homologene) for human, chimpanzee, rhesus, cattle, dog, mouse, and rat, followed by identification of promoter sequence (upstream 1,000 bp to TSS) of each gene for each species using R package biomaRt [59] based on the Ensembl database [60]. Promoter sequences of the other species (i.e., chimpanzee, rhesus, cattle, dog, mouse, and rat) were compared to the promoter sequence of human using R package msa [61]. We used R package TFBSTools [62] to identify conserved NF-κB binding sites between human and the other species. Four NF-κB family motifs (i.e., NFKB1: MA0105.2, NFKB2: MA0778.1, REL: MA0101.1, and RELA: MA0107.1) obtained from JASPAR [63] were used for the analysis. In accordance with Iwanaszko and colleagues [50], the threshold of the motif mapping score was set to 80%, and the conservation cutoff was set to 40%, based on a window size of 51 bp. To obtain the percentage of conserved NF-κB binding sites for each gene, the total number of NF-κB motif sites in the human genome was obtained based on the 4 NF-κB family motifs. The number of conserved NF-κB motif sites was calculated by comparing human and every other species, respectively. The percentage of conserved NF-κB binding sites is calculated as the ratio between the conserved NF-κB motif sites and the total number of NF-κB motif sites for each gene.
In Fig 5B and 5C, data are represented as the mean with the SEM. Comparisons between two groups at each time point were assessed by a 1-way ANOVA Kruskal-Wallis test. A p-value of ≤0.05 was considered statistically significant.
|
10.1371/journal.pgen.1004609 | Recovery from an Acute Infection in C. elegans Requires the GATA Transcription Factor ELT-2 | The mechanisms involved in the recognition of microbial pathogens and activation of the immune system have been extensively studied. However, the mechanisms involved in the recovery phase of an infection are incompletely characterized at both the cellular and physiological levels. Here, we establish a Caenorhabditis elegans-Salmonella enterica model of acute infection and antibiotic treatment for studying biological changes during the resolution phase of an infection. Using whole genome expression profiles of acutely infected animals, we found that genes that are markers of innate immunity are down-regulated upon recovery, while genes involved in xenobiotic detoxification, redox regulation, and cellular homeostasis are up-regulated. In silico analyses demonstrated that genes altered during recovery from infection were transcriptionally regulated by conserved transcription factors, including GATA/ELT-2, FOXO/DAF-16, and Nrf/SKN-1. Finally, we found that recovery from an acute bacterial infection is dependent on ELT-2 activity.
| Infections by bacterial pathogens often produce substantial tissue damage and alter metabolism in the host that, if left unchecked, could be detrimental to overall fitness. The cellular and systemic responses that resolve these alterations in the host are not well defined. Here, we examine transcriptional networks in an animal host that are modulated during the resolution phase of an intestinal infection treated with an antibiotic. Up-regulation of genes involved in detoxification and cellular homeostasis during the resolution phase is controlled by the conserved endodermal GATA transcription factor ELT-2. GATA transcription factors are known to be involved in the development, differentiation, and function of a diverse array of metazoan tissue types. Therefore, our results ascribe a new role to GATA transcription factors in recovery from an acute infection. Fully characterizing the host response during resolution of an infection will lead to a better understanding of human health concerns related to recurrent infections, wound healing, autoimmune diseases, and chronic inflammatory disorders.
| The course of human bacterial infections is controlled by a combination of immune responses, physiological changes, and, if necessary, antibiotic treatment. To recover from an infection and return to homeostasis, the host must activate mechanisms capable of controlling the damage caused by pathogen virulence factors, inflammation, and a potentially toxic antibiotic exposure. If these alterations in host physiology are not handled appropriately, the host risks entering a state of reduced fitness. This reduced fitness manifests in the form of recurrent infections, inappropriate wound healing, autoimmune diseases, and chronic inflammatory disorders. While the mechanisms involved in the recognition of microbial pathogens as such and the subsequent activation of the immune system have been extensively studied, the pathways involved in host recovery after an infection remain understudied.
To examine the biological changes that take place during the recovery phase of an acute bacterial infection, we decided to use the nematode Caenorhabditis elegans as a simple model host. Various human bacterial pathogens, including Pseudomonas aeruginosa, Salmonella enterica, Staphylococcus aureus, and Enterococcus faecalis, have been shown to colonize and kill C. elegans using conserved virulence mechanisms [1]–[4]. Moreover, C. elegans responds to infections using an inducible innate immune system that is controlled by several evolutionary conserved signaling cascades including the p38-MAPK (PMK-1), insulin-IGF (DAF-16), GATA (ELT-2), and TGF-B (SMA-6) pathways [5]–[8]. It is therefore likely that investigating C. elegans recovery from bacterial infection would shed light on host responses that reestablish homeostasis post-infection.
In this study, we established a C. elegans-S. enterica pathogenesis system as a model of acute infection by infecting nematodes with S. enterica and subsequently resolving the infection by treatment with the antibiotic Tetracycline. Using this acute infection model, we profiled gene expression changes in the host over the course of the infection and during the recovery phase of the infection. We found that during recovery, certain components of the host innate immune response were dampened, while mechanisms involved in xenobiotic detoxification, redox regulation, and cytoprotection were activated. A large number of the genes altered during recovery corresponded to intestinal genes regulated by ELT-2, which is a conserved GATA transcription factor that plays a key role in the control of intestinal functions in C. elegans. Further studies indicated that the recovery from acute S. enterica infection required ELT-2, indicating that ELT-2 controls not only induction of innate immune response genes but also genes that play a crucial role in the resolution of an infection.
Although host responses that limit microbial infection have been extensively studied, the mechanisms involved in the recovery phase of an infection are incompletely characterized at both the cellular and physiological level. We decided to use Caenorhabditis elegans as a simple model host for assessing biological changes during the recovery phase from an acute infection. C. elegans is propagated in the laboratory by feeding them E. coli strain OP50. E. coli is effectively disrupted by the C. elegans pharyngeal grinder and essentially no intact bacterial cells can be found in the intestinal lumen of young, immunocompetent animals. However, pathogenic bacteria such as Salmonella enterica are capable of killing C. elegans by infectious processes that correlate with the accumulation of bacteria in the intestine. As in mammalian hosts, a small inoculum of S. enterica is capable of establishing a persistent infection in C. elegans that does not require constant exposure to bacteria and cannot be prevented by transferring the infected animals to plates containing E. coli [2], [9].
To determine whether a long-lasting, chronic S. enterica infection could be easily reversed by antibiotic treatment to model a short, acute infection we used fer-1(b232ts) animals, which are fertilization defective at the restrictive temperature. This prevents losing track of the initially infected animals in the morass of progeny that would be otherwise generated following an acute infection. We first established that transferring S. enterica-infected animals to plates containing the bacteriostatic antibiotic Tetracycline and seeded with TetR E. coli was sufficient to significantly reduce bacterial burden (Figure S1). Subsequently, we decided to use 50 µg/ml Tetracycline treatment to reduce S. enterica burden to model an acute infection in C. elegans.
We monitored bacterial accumulation over the course of a 120 hour infection in synchronized larval stage 1 (L1) fer-1(b232ts) animals continuously grown on plates seeded with S. enterica-GFP or transferred to Tetracycline-containing plates seeded with TetR E. coli. Consistent with previous findings indicating that C. elegans larvae are highly resistant to pathogen-mediated killing and that death does not occur during the first several days of an S. enterica infection [9], [10], we observed that only 4.1% of the animals exposed to S. enterica-GFP starting at the L1 stage were colonized 72 hours later (Figure 1A). In contrast, at 96 and 120 hours post-exposure, 41.9% and 71.4% of the animals were colonized by S. enterica-GFP (Figure 1B–C). We found that transferring animals from S. enterica at 72 or 96 hours to Tetracycline-containing plates for 24 hours reduced bacterial burden (Figure 1B–C). Quantification of the number of live bacteria in animals that were infected with S. enterica-GFP for 72 hours and treated with Tetracycline for 24 hours showed a significant reduction of bacterial burden (Figure 1D), confirming that Tetracycline treatment can prevent S. enterica from persistently colonizing the C. elegans intestine and causing a chronic infection.
Even though our results indicate that Tetracycline can prevent S. enterica from causing a persistent colonization of the C. elegans intestine, it was unclear whether acute pathogenic challenge would damage the animal and translate into an associated reduction in survival. As shown in Figure 1E, we found that the survival of animals infected with S. enterica and then treated with Tetracycline is significantly higher than that of animals continuously infected (Figure 1E; yellow vs. red lines). Also, survival of infected and then Tetracycline-treated animals is nearly equivalent to animals that were never infected (Figure 1E, yellow vs. black lines). Treatment with Tetracycline in the presence of killed bacteria only increased C. elegans mean lifespan from 14.2 to 14.9 days (Figure S2). Taken together, these studies show that an S. enterica infection can be resolved by treating the animals with Tetracycline and indicate that this type of treatment can be used to model an acute S. enterica infection that progresses towards chronicity if the animals were to remain untreated.
To investigate cellular mechanisms potentially involved in recovery after an infection, we utilized Agilent C. elegans gene expression microarrays to identify changes in gene expression during infection and changes that take place after the infection is reversed by treatment with Tetracycline (Figure 2A, Tables S1 and S2). Initially, we focused our analysis on animals that were infected with S. enterica for 96 hours vs. animals that were infected for 72 hours and treated with Tetracycline for 24 hours to resolve the infection. At 96 hours, 99% of the animals were alive in both conditions (Figure 1E, Day 1). Overall, 243 genes, or approximately 1% of the C. elegans genome, were altered more than 2-fold (p<0.05) when comparing the 96-hour cohorts. Of these altered genes, 126 were down-regulated and 117 were up-regulated (Table S2).
To identify related gene groups that are transcriptionally controlled by pathways potentially involved in the changes that take place after infection, we performed an unbiased gene enrichment analysis using the database for annotation, visualization and integrated discovery (DAVID, http://david.abcc.ncifcrf.gov/) [11]. The 10 gene ontology (GO) clusters with the highest DAVID enrichment score are shown in Figure 2B and Table S3. For the subset of down-regulated genes that respond to the resolution of the infection, the 2 top-scoring GO clusters, c-type lectins and lysozyme groupings, have previously been described as part of an inducible C. elegans immune response to a variety of pathogens [12]–[15]. For the subset of up-regulated genes that respond to the resolution of the infection, 4 of the top 10 highest scoring ontology clusters are associated with xenobiotic detoxification, redox regulation, or cytoprotection [16], [17]. These results indicate that the activation of the innate immune system of C. elegans by S. enterica infection is attenuated once the infection is resolved and that certain cellular homeostatic pathways are activated during recovery.
Since only 42% of the animals exposed to S. enterica for 96 hours exhibited visible bacterial colonization (Figure 1B), we decided to examine gene expression profiles of animals that were infected for 120 hours, which exhibited an even higher degree of bacterial colonization (Figure 1C). A comparison of gene expression profiles from animals that were infected with S. enterica for 120 hours vs. animals that were infected for 96 hours and treated with Tetracycline for 24 hours identified 57 and 72 genes that are down- or up-regulated greater than 2-fold (p<0.05), respectively (Table S2). Analysis of GO terms in these gene sets via DAVID gives a shorter but similar list of enriched gene clusters (Figure 2B and Table S3), confirming that, as the infection resolves, marker genes of immune activation are down-regulated while genes that correspond to cellular homeostatic pathways are up-regulated. Moreover, the significant overlap between 96 and 120 hour treatment gene sets indicates that the changes that take place after an infection is resolved are reproducible and that similar transcriptional profiles are elicited at different times (Figure 2C and Table S4). The smaller number of genes down- and up-regulated by the resolution of the infection at 120 hours compared to 96 hours could be a consequence of the higher heterogeneity of S. enterica colonization in the 120-hour population (comparison of Figures 1B and 1C). It is also possible that as the infection progresses, the animals suffer irreversible damage that makes them less responsive to antibiotic treatment.
To validate the results of the microarrays, we performed quantitative real-time PCR (qRT-PCR) on a subset of the 243 genes that change upon Tetracycline treatment of infected animals. This subset includes 11 up-regulated genes and 6 down-regulated genes that were either present in a high scoring GO cluster, were highly misregulated, or both. We performed qRT-PCR on RNA harvested from C. elegans that were subjected to the same conditions as in the microarray studies. As shown in Figure 3A and 3B, the changes in gene expression as assessed by qRT-PCR were comparable to those observed by microarray profiling. Further analysis indicated that 16 of the 17 genes had statistically significant expression changes during Tetracycline-mediated recovery from S. enterica infection (Figure 3A and 3B). Thus, the microarray data accurately reflects the majority of gene expression differences between treated and non-treated animals.
Resolution of S. enterica infection by treatment with Tetracycline results in the down-regulation of genes that are markers of innate immunity and the up-regulation of genes that function in xenobiotic detoxification, redox regulation, and cytoprotection (Figure 2B). While the resolution of the infection may be responsible for altering the expression of these genes, it is also possible that Tetracycline is directly inducing these changes. To distinguish between these two possibilities, we compared changes in gene expression due to Tetracycline alone vs. changes in gene expression due to recovery from infection by treatment with Tetracycline. We found that expression of 8 out of 16 tested genes were significantly different (Figure 3C–D), highlighting the role of these 8 recovery genes in pathways that are altered during the resolution of the S. enterica infection. Considering that the genome wide microarray shows that 243 genes change their expression upon recovery at 96 hours, we estimate that approximately 122 genes are regulated by recovery from infection independently of Tetracycline while the remaining genes are regulated by the inclusion of Tetracycline alone. This suggests that Tetracycline may be directly inducing gene expression changes in the host that may help clear an infection independently of its antimicrobial activity.
To confirm the finding that a subset of genes is altered upon resolution of an S. enterica infection independent of Tetracycline, we performed equivalent experiments using the antibiotic Kanamycin. These studies indicate that Kanamycin alone did not alter the expression of 9 tested genes in uninfected animals (Figure S3A–B). Furthermore, 8 of the 9 alterations in gene expression seen in infected animals treated with Kanamycin are similar to those seen in infected animals treated with Tetracycline (Figure S3C–D).
To provide further insight into the behavior of genes altered during antibiotic-mediated recovery, we examined gene expression profiles over the course of the 96-hour infection (Table S1). We focused our analysis on qRT-PCR-confirmed genes that are known markers of immune activation and genes that correspond to cellular homeostatic pathways. The expression of innate immunity genes diminished significantly after the infection was resolved by Tetracycline treatment (Figure 3C and E). In contrast, genes involved in regulating cellular homeostasis were significantly up-regulated upon recovery from infection (Figure 3D and F). As a control, the expression of 3 select housekeeping genes remained relatively constant both during the course of the infection and during recovery (Figure 3G). Overall, these studies suggest that as the infection resolves, cellular homeostatic mechanisms are activated while elements of the immune response are attenuated.
Several of the GO clusters identified in the set of genes up-regulated during resolution of the infection correspond to genes whose products are involved in detoxification. We therefore hypothesized that the reduction of the pathogenic insult during the recovery phase of an infection may trigger processes involved in detoxification and clearance of immune effectors that, while necessary to combat pathogens, can have deleterious effects on the host. Recently, it was demonstrated that reactive oxygen species (ROS), a component of the C. elegans immune response to S. enterica and other pathogens [18]–[20], contributes to infectious pathogenicity (i.e., damage to the host). Thus, we decided to study gsto-1, which is an up-regulated gene that encodes an omega-class glutathione S-transferase that protects C. elegans from oxidative stress under non-infected conditions [21]. We found that survival of gsto-1(RNAi) animals infected with S. enterica and treated with Tetracycline was not significantly different from that of control animals (Figure S4).
The lack of a significant effect by gsto-1 RNAi could be attributed to incomplete RNAi or to functional redundancy among the multitude of detoxification genes that are up-regulated during recovery (Figure 2B and Table S3). The gsto-1 locus is transcriptionally regulated by the GATA transcription factor ELT-2 [21], leading us to consider a role for ELT-2 in controlling the expression of a set of genes required for resolution of an infection. Consequently, we applied several in silico approaches to determine whether ELT-2-regulated genes are present in the genes whose expression changes by the resolution of an infection. We compared the set of genes altered during recovery to previously identified ELT-2-regulated gene sets and to other control data sets. The ELT-2-regulated gene sets were among the 10 data sets with the strongest overlap with our recovery gene set (Figure 4A and Table S5). As ELT-2 regulates the expression of genes in the C. elegans intestine via trans-acting activity at TGATAA (extended GATA) cis-regulatory motifs [22], [23], we looked for the presence of TGATAA binding sites in the putative promoter regions of the down- and up- regulated genes. Approximately 63% of the 243 genes regulated by recovery contain at least 1 TGATAA site within the 1.5 kb sequence upstream of their transcriptional start site (Figure 4B). By comparison, only 54% of genes in 3 randomly selected gene sets (n = 243 each) have at least 1 TGATAA sequence in the equivalent 1.5 kb region (Figure 4B). Additionally, we observed that at least 1 TGATAA site is present in the putative promoter region of 7 of the 8 recovery genes verified by qRT-PCR (Table S6). While gsto-1 does not contain a TGATAA site in this 1.5 kb region, it does have a single site 3.8 kb upstream of the transcriptional start site. Moreover, it has been experimentally demonstrated that ELT-2 regulates the transcription of gsto-1 [21].
Consistent with the post-developmental role of ELT-2 in the regulation of adult intestinal functions [5], [6], [24]–[26], another in silico approach showed that 17 out of 32 (53%) recovery genes with at least 1 TGATAA site and for which the data is available are expressed in the intestine (Table S7). Only 1 out of these 32 genes is expressed in the hypodermis where other GATA transcription factors function [27]. This analysis also showed that 4 out of 8 recovery genes verified by qRT-PCR are expressed in the intestine (Table S7). Taken together, our in silico analyses leads to the hypothesis that ELT-2 controls the expression of a subset of genes during the recovery phase of an infection.
To further substantiate a role for ELT-2 in the transcriptional regulation of genes during recovery, we studied the effect of elt-2 RNAi on the expression of recovery genes. As ELT-2 is essential for C. elegans larval development [22], RNAi was performed on late larval stage 4 (L4) animals. This approach has been used successfully to inhibit elt-2 expression for at least 6 days [28], [29]. As shown in Figure 4D, RNAi of elt-2 inhibited the expression of the 5 studied genes that are up-regulated during recovery from the infection by treatment with either Tetracycline (Figure 3D) or Kanamycin (Figure S3B). Inhibition of elt-2 by RNAi also further down-regulated the expression of ilys-3 and lys-9 (Figure 4C). However, RNAi of elt-2 does not result in the unselective down-regulation of recovery genes as acdh-1 is not down-regulated (Figure 4C). In addition, certain ELT-2-controlled immunity and structural genes [12], [28] are not significantly altered during recovery from S. enterica infection (Figure S5A–B). We further confirmed by qRT-PCR that transcript levels of clec-67, which is a known marker of immunity controlled by ELT-2 [12], are not altered upon recovery (Figure S5C). We conclude that expression of a specific intestinal gene program during resolution of an infection is dependent upon the action of the GATA transcription factor ELT-2.
To test whether ELT-2 is required for recovery after infection, we studied the survival of elt-2(RNAi) animals infected with S. enterica and treated with Tetracycline. RNAi inhibition of elt-2 starting at the L4 stage did not alter the survival of animals growing on live E. coli (Figure 5A; black lines), nor did it alter survival in the presence of Tetracycline (Figure S6). This data indicates that L4 elt-2(RNAi) animals are not sick merely due to disruptions in basal immunity or intestinal function. However, RNAi of elt-2 prevented the recovery of infected animals by treatment with Tetracycline (Figure 5B; yellow lines), highlighting the role of ELT-2 during the recovery phase of the infection. In agreement with previously published reports that ELT-2 regulates innate immunity [12], [28], RNAi of elt-2 did significantly reduce survival of animals continuously infected with S. enterica (Figure 5A; red lines). To address whether genes crucial for immunity are generally required for recovery, we studied pmk-1, which encodes a p38 mitogen-activated protein kinase that is a major regulator of innate immunity in C. elegans [14], [30], [31]. Even though RNAi of pmk-1 elicited sensitivity to S. enterica-mediated killing (Figure 5C; red lines), it did not prevent the recovery of infected animals by treatment with Tetracycline (Figure 5D). Taken together, these results indicate that ELT-2 is required for both early immune responses against pathogens and responses that are activated upon recovery from an infection by S. enterica.
Using gene expression profiling, in silico analysis, and reverse genetic approaches, we have defined a novel post-developmental role for the GATA transcription factor ELT-2 during the resolution of an infection. ELT-2 was originally identified as a key regulator of C. elegans intestinal specification during development [22]. However, it is now becoming clear that ELT-2 has an extensive post-developmental role in the regulation of a plethora of adult intestinal functions. Under the control of ELT-2, the 20 cells of the adult intestine in C. elegans function in nutrient uptake, synthesis and storage of macromolecules, epithelial immunity, and host-microbial communication [5], [6], [24]–[26]. There are several additional GATA transcription factors encoded in the C. elegans genome, including ELT-4 and ELT-7, which regulate intestinal gene expression programs [32]. Further studies will be required to determine their possible contribution to the recovery process.
Owing to the multi-functional nature of the intestine and due to the fact that ELT-2 regulates nearly all intestinal genes [33], it is not surprising that half of the genes in the C. elegans genome have putative ELT-2 binding sites (Figure 4B). Thus, it is logical to conclude that the specification of different functional outputs that takes place in the C. elegans intestine during the complete course of an infection is controlled by additional co-factors that act together with ELT-2. Recent work has demonstrated that GATA transcription factors, including ELT-2, act coordinately with the insulin-IGF pathway transcriptional regulator DAF-16 in a cell-autonomous manner to regulate lifespan extension in C. elegans [34]. It is therefore plausible that DAF-16 acts with ELT-2 to co-regulate genes important for infection resolution. Indeed, we observed a significant enrichment of both ELT-2- and DAF-16-controlled targets in our set of genes altered during recovery from infection (Figure 4A and Table S5). An emerging theme is that coordinated transcriptional activity of DAF-16 and ELT-2 would be necessary for the modulation of cytoprotective pathways that, in turn, are required for a majority of cellular stress response pathways [16].
Another candidate factor that may regulate damage response genes in conjunction with ELT-2 and/or DAF-16 is the Nrf1/SKN-1 transcription factor. Previous work has demonstrated that signaling cascades downstream of reactive oxygen species (ROS) induce a cytoprotective SKN-1 pathway [35], [36]. SKN-1 might regulate cytoprotective genes downstream of or in parallel to ELT-2 and/or DAF-16 to mediate resolution of an infection. Indeed, SKN-1-positively regulated targets are significantly enriched in the set of genes that are up-regulated during infection resolution (Table S5).
Mounting evidence indicates that ELT-2 activity is modulated under a variety of environmental conditions or physiological states. ELT-2-mediated immunity to a variety of pathogens is activated by currently unknown mechanisms. Interestingly, a paper by Lee and colleagues demonstrates that the intestinal pathogen B. pseudomallei can actively target and degrade ELT-2 to prevent host immune responses [29]. We failed to observe any alterations in ELT-2 protein localization or abundance caused by S. enterica infection or recovery. These pathogens, which kill C. elegans at a distinctly different rate, must utilize different mechanisms to overcome the host immune system.
Signals during the initial decline in infection may function to reprogram the transcriptional activity of ELT-2 from an innate immune program to a cytoprotective one. These unidentified signals may be bacterial- and/or host-derived. Specific bacterial-derived signals, such as those involved in biofilm formation or quorum sensing, may be the primary trigger for the ELT-2 switch [26]. These bacterial-derived signals might act directly on ELT-2 or they may transit through host-encoded genes. Alternatively, host-encoded regulators that normally function during development, such as the END-1/END-3 specification factors, might be re-activated during the resolution of infection to direct the transcription of detoxification genes by ELT-2. Interestingly, the END-1/END-3 system lies downstream of the oxidative stress (ROS) response protein SKN-1 in the development of the C. elegans intestine [24]. Alternatively, changes in ELT-2 activity may be controlled by local chromatin remodeling in a manner similar to the regulation of DAF-16 transcriptional activity [37].
In summary, our results identified a new, key role for ELT-2 during recovery from a bacterial infection. We revealed that during recovery from an infection, genes that are markers of innate immunity are down-regulated, while the expression of genes involved in xenobiotic detoxification, redox regulation, and cytoprotection is enhanced. Interestingly, a number of genes encoding antibacterial factors (ABFs) are up-regulated during the course of the S. enterica infection (Table S1). However, the expression of abf genes is not down-regulated once the infection is resolved. This could be due to a mechanism used by C. elegans to maintain high levels of abf genes throughout reproductive adulthood. It is also possible that ABFs have a high specificity for damaging prokaryotic cells, having little or no impact on host cells. Thus, there would be no immediate need to reverse their up-regulation once the infection is resolved, unlike the case of lysozyme-encoding genes, which could potentially damage host cells. The ELT-2 interaction with the aforementioned co-factors may dictate the specificity of the expression profile during the different phases of an infection. A number of microbial killing pathways and cellular homeostatic pathways are controlled by the nervous system in infected C. elegans [38]–[46]. An important question that remains to be evaluated is whether the nervous system also plays a role in the control of the mechanisms involved in recovery after infections have been cleared.
C. elegans strain HH142 fer-1(b232ts) was provided by the Caenorhabditis Genetics Center. C. elegans were maintained at 15°C on NGM—OP50 plates without antibiotics. The following bacterial strains were used for experiments: Escherichia coli strain OP50-1 [SmR] [47], E. coli—dsRed strain OP50 [AmpR, CbR] [47], E. coli strain HT115 [TetR] [48], E. coli strain HT115 pL4440 [AmpR, TetR] [48], E. coli strain DH5α pSMC21 [KanR] [49], Salmonella enterica enterica serovar Typhimurium strain 1344 [SmR] [50], S. enterica—GFP strain SM022 [SmR, KanR] [51]. Bacteria were grown overnight for 14 hours in 3 ml LB broth at 37°C.
fer-1(b232ts) animals were synchronized by treating gravid adults with sodium hydroxide and bleach. About 2,000 synchronized L1 animals were grown on full lawn S. enterica—GFP plates at 25°C for 36, 72, 96, or 120 hours. At designated transfer time points, animals were rinsed off S. enterica—GFP plates, washed with M9 (4 changes ×15 minutes), concentrated, and then transferred to plates with or without 50 µg/ml Tetracycline that were seeded with E. coli HT115 or S. enterica-GFP. At designated visualization time points, animals were picked to an NGM—OP50 plate for 1 hour before being picked to a new NGM—OP50 plate. Animals were then visualized at 20× using a Leica MZ FLIII fluorescence stereomicroscope. In heavily colonized animals (heavy) GFP fluorescence was visible in the presence of halogen white light set at 60%, while in weakly infected animals (weak) GFP fluorescence was only visible in the absence of white light. Animals where GFP fluorescence was not detected even in the absence of white light were scored as not infected (none).
For the quantification of colony forming units (CFUs), fer-1(b232ts) animals were synchronized by treating gravid adults with sodium hydroxide and bleach. About 2,000 synchronized L1 animals were grown on full lawn S. enterica—GFP plates at 25°C for 72 hours. At the designated transfer time points, animals were rinsed off S. enterica—GFP plates, washed with M9 (4 changes ×15 minutes), concentrated, and then transferred to S. enterica-GFP or E. coli plus 50 µg/ml Tetracycline plates. At designated CFU time points, animals were picked to 3 NGM—OP50 plates (20 minutes each) before being picked to a 1.5 ml eppendorf tube with 50 µl of PBS plus 0.1% Triton-X-100. A total of 10 animals per condition were mechanically disrupted using a mini-pestle. Serial dilutions of the lysates were spread onto LB/Kanamycin (50 µg/ml) plates to select for S. enterica—GFP cells and grown for 24 hours at 37°C.
Bacteria – E. coli HT115 or S. enterica were grown overnight for 14 hours in 3 ml LB broth at 37°C. A total of 50 µl (scoring) or 500 µl (exposure) of the resulting cultures were spread onto modified (0.35% peptone) NGM plates with or without 50 µg/ml Tetracycline and allowed to grow for 1–2 days at 25°C to produce a thick lawn. fer-1(b232ts) animals were synchronized by treating gravid adults with sodium hydroxide and bleach. Synchronized L1 animals were grown on full lawn S. enterica—GFP plates at 25°C for 72 hours before being transferred to the appropriate (treatment or not) plates. The assays were performed at 25°C. Animals were scored every day and were considered dead when they failed to respond to touch. Animals were transferred to fresh plates every other day for the entire length of the experiment. Survival was plotted using Kaplan-Meier survival curves and analyzed by the logrank test using GraphPad Prism (GraphPad Software, Inc., San Diego, CA). Survival curves resulting in p values of <0.05 were considered significantly different. A total of 60 animals per condition per experiment were used.
E. coli was grown as described above. A 50- µl drop of the bacteria was plated on a 3.5-cm plate of modified NGM agar containing 40 µg/ml fluoro-deoxyuridine with or without 50 µg/ml Tetracycline. A total of 100 animals per condition per experiment were used. The assays were performed at 25°C. Survival curves were analyzed as described above.
gsto-1(RNAi). E. coli HT115(DE3) bacterial strains expressing double-stranded RNA [48] were grown for 9 hours in 5 ml LB broth containing Ampicillin (50 µg/ml) at 37°C. The resulting cultures were seeded onto NGM plates containing Carbenicillin (50 µg/ml) and isopropyl-1-thio-β-D-galactopyranoside (3 mM). dsRNA-expressing bacteria were allowed to grow for 2 days at 25°C to produce a thick lawn. fer-1(b232ts) L4 animals were placed on RNAi or vector control plates for 5 days at 15°C until F1 animals developed. fer-1(b232ts) F1 L4 animals were placed on a second RNAi or vector control plate and incubated for another 5 days at 15°C until adult F2 animals developed. Gravid F2 RNAi animals were picked to full lawn E. coli or S. enterica—GFP plates and allowed to lay eggs for 3 hours at 25°C to synchronize a third generation population. These third generation animals were kept on E. coli or S. enterica—GFP plates for 72 hours before being transferred to plates with or without 50 µg/ml Tetracycline that were seeded with E. coli or S. enterica-GFP. unc-22(RNAi) was used as positive control in all experiments to account for RNAi efficiency. The gsto-1 (mv_C29E4.7) RNAi vector was verified by DNA sequencing. A total of 60 animals per condition per experiment were scored for survival.
elt-2(RNAi) and pmk-1(RNAi). Production of RNAi plates was the same as described above. Gravid fer-1(b232ts) animals were allowed to lay eggs for 3 hours at 25°C on NGM-HT115 plates. Gravid animals were removed and the eggs/plates were incubated for 36 hours at 25°C. Synchronized L4 animals were then transferred to RNAi or vector control plates and incubated for an additional 36 hours at 25°C. Young adult RNAi or vector control animals were then transferred to and grown on full lawn E. coli or S. enterica—GFP plates for 36 hours at 25°C. Adult worms were then transferred to plates with or without 50 µg/ml Tetracycline that were seeded with E. coli or S. enterica-GFP. unc-22(RNAi) was used as positive control in all experiments to account for RNAi efficiency. The elt-2 (mv_AAC36130) and pmk-1 (sjj_B0218.3) RNAi vectors were verified by DNA sequencing. A total of 60 animals per condition per experiment were scored for survival.
fer-1(b232ts). Animals were synchronized by treating gravid adults with sodium hydroxide and bleach. Synchronized L1 animals were grown on full lawn E. coli or S. enterica—GFP plates for 72 hours. At 72 hours, animals were rinsed off E. coli or S. enterica—GFP plates, washed with M9 (4 changes ×15 minutes), concentrated, and then transferred to E. coli, E. coli plus 50 µg/ml Tetracycline, E. coli plus 50 µg/ml Kanamycin, or S. enterica—GFP plates. At 24 hours post-transfer, animals were rinsed off plates, washed with M9 (4 changes ×15 minutes), and flash-frozen in Trizol (Life Technologies, Carlsbad, CA). Total RNA was extracted using the RNeasy Plus Universal Kit (Qiagen, Netherlands).
elt-2(RNAi); fer-1(b232ts). Animals were synchronized by treating gravid adults with sodium hydroxide and bleach. Synchronized L1 animals were grown on full lawn E. coli plates for 36 hours. At 36 hours, animals were rinsed off E. coli, washed with M9 (4 changes ×15 minutes), concentrated, and then placed on RNAi or vector control plates for 36 hours. At 72 hours, animals were rinsed off these plates washed with M9 (4 changes ×15 minutes), concentrated, and then placed on S. enterica—GFP plates for 36 hours. At 108 hours, animals were rinsed off S. enterica—GFP plates, washed with M9 (4 changes ×15 minutes), concentrated, and then transferred to E. coli, E. coli plus 50 µg/ml Tetracycline, or S. enterica—GFP plates for 24 hours. Animals were rinsed off plates, washed with M9 (4 changes ×15 minutes), and flash-frozen in Trizol (Life Technologies, Carlsbad, CA). Total RNA was extracted using the RNeasy Plus Universal Kit (Qiagen, Netherlands). All studies were performed at 25°C.
Total RNA was obtained as described above. A total of 1 µg total RNA was oligo(dT) primed and reverse transcribed in a 50 µl volume using the SuperScript III First-Strand Synthesis System (Life Technologies, Carlsbad, CA). Reactions without the addition of reverse transcriptase (RT) were also performed and served as controls for contaminating genomic DNA in quantitative PCR experiments. Two µl of the resulting plus or minus RT reactions served as templates in quantitative PCR experiments using Power SYBR Green PCR Master Mix (Life Technologies, Carlsbad, CA) and the StepOnePlus Real-Time PCR System (Life Technologies, Carlsbad, CA). For each sample, 3 technical replicates were performed. Pan-actin-normalized Ct values were determined using the StepOnePlus Software (Life Technologies, Carlsbad, CA). Primer sequences are available upon request. When applicable a one or two variable t-test was performed.
fer-1(b232ts) animals were synchronized by treating gravid adults with sodium hydroxide and bleach. Synchronized L1 animals were grown on full lawn E. coli OP50 (uninfected) or full lawn S. enterica plates at 25°C for 36, 72, 96, or 120 hours. At designated transfer time points, animals were rinsed off S. enterica plates, washed with M9 (4 changes ×15 minutes), concentrated, and then transferred to S. enterica or E. coli plus 50 µg/ml Tetracycline plates. At designated harvesting time points, animals were rinsed off plates, washed with M9 (4 changes ×15 minutes), and flash-frozen in Trizol (Life Technologies, Carlsbad, CA). Total RNA was extracted using the RNeasy Plus Universal Kit (Qiagen, Netherlands).
Total RNA was assessed for quality with an Agilent 2100 Bioanalyzer G2939A (Agilent Technologies, Santa Clara, CA) and a Nanodrop 8000 spectrophotomer (Thermo, Wilmington, DE). 100 ng of total RNA was converted to 1.65 µg Cy-3-labeled, linearly amplified cRNA using the Low Input Quick Amp (LIQA) Labeling One-Color Microarray-Based Gene Expression Analysis Kit (Agilent Technologies, Santa Clara, CA). cRNA was fragmented and added to 44 K feature Agilent C. elegans Gene Expression Microarray V2 slides (Agilent Technologies, Santa Clara, CA). Hybridization was performed in the Agilent rotisserie Hybridization Oven for 17 hours. Arrays were subsequently washed and scanned with the Agilent B scanner according to standard Agilent protocols (Agilent Technologies, Santa Clara, CA). Scanned data was log2 transformed and quantile normalized using Partek Genomics Suite (St. Louis, MO). Analysis of variance (ANOVA) t tests and fold-change calculations were also performed using Partek Genomics Suite (St. Louis, MO). For each of the 5 time points, 2 biological replicates were assessed. The microarray data was deposited in the Gene Expression Omnibus database: GSE54212.
Gene lists were culled from the literature and passed through WormBase Converter [52] using the WS220 genome release as the output (references are noted in Table S5). A total of 20,834 WS220 genes are represented by 1 or more probes in the Agilent C. elegans V2 array (Agilent Technologies, Santa Clara, CA). Gene ontology analysis was performed using the DAVID Bioinformatics Database (david.abcc.ncifcrf.gov/). The most significant gene ontology term in each DAVID functional annotation cluster was set as the significance of the overall cluster. Statistical significance of the overlap between two gene sets was calculated using the following on-line program: nemates.org/MA/progs/overlap_stats.html. Representation Factor represents the number of overlapping genes divided by the expected number of overlapping genes drawn from 2 independent groups. A background gene list of 20,834 was used for the calculation. P values were calculated using the hypergeometric probability. 1.5 kb cis-regulatory sequences were identified using WormMart (wormbase.org). Expression patterns were determined using WormMine (wormbase.org). Detailed bioinformatics protocols are available upon request.
|
10.1371/journal.pcbi.1006759 | RedCom: A strategy for reduced metabolic modeling of complex microbial communities and its application for analyzing experimental datasets from anaerobic digestion | Constraint-based modeling (CBM) is increasingly used to analyze the metabolism of complex microbial communities involved in ecology, biomedicine, and various biotechnological processes. While CBM is an established framework for studying the metabolism of single species with linear stoichiometric models, CBM of communities with balanced growth is more complicated, not only due to the larger size of the multi-species metabolic network but also because of the bilinear nature of the resulting community models. Moreover, the solution space of these community models often contains biologically unrealistic solutions, which, even with model linearization and under application of certain objective functions, cannot easily be excluded. Here we present RedCom, a new approach to build reduced community models in which the metabolisms of the participating organisms are represented by net conversions computed from the respective single-species networks. By discarding (single-species) net conversions that violate a minimality criterion in the exchange fluxes, it is ensured that unrealistic solutions in the community model are excluded where a species altruistically synthesizes large amounts of byproducts (instead of biomass) to fulfill the requirements of other species. We employed the RedCom approach for modeling communities of up to nine organisms involved in typical degradation steps of anaerobic digestion in biogas plants. Compared to full (bilinear and linearized) community models, we found that the reduced community models obtained with RedCom are not only much smaller but allow, also in the largest model with nine species, extensive calculations required to fully characterize the solution space and to reveal key properties of communities with maximum methane yield and production rates. Furthermore, the predictive power of the reduced community models is significantly larger because they predict much smaller ranges of feasible community compositions and exchange fluxes still being consistent with measurements obtained from enrichment cultures. For an enrichment culture for growth on ethanol, we also used metaproteomic data to further constrain the solution space of the community models. Both model and proteomic data indicated a dominance of acetoclastic methanogens (Methanosarcinales) and Desulfovibrionales being the least abundant group in this microbial community.
| Microbial communities are involved in many fundamental processes in nature, health and biotechnology. The elucidation of interdependencies between the involved players of microbial communities and how the interactions shape the composition, behavior and characteristic features of the consortium has become an important branch of microbiology research. Many communities are based on the exchange of metabolites between the species and constraint-based metabolic modeling has become an important approach for a formal description and quantitative analysis of these metabolic dependencies. However, the complexity of the models rises quickly with a growing number of organisms and the space of predicted feasible behaviors often includes unrealistic solutions. Here we present RedCom, a new approach to build reduced stoichiometric models of balanced microbial communities based on net conversions of the single-species models. We demonstrate the applicability of our RedCom approach by modeling communities of up to nine organisms involved in degradation steps of anaerobic digestion in biogas plants. As one of the first studies in this field, we compare simulation results from the community models with experimental data of laboratory-scale biogas reactors for growth on ethanol and glucose-cellulose media. The results also demonstrate a higher predictive power of the RedCom vs. the full models.
| Microbial communities are of major importance for human health [1,2], geochemical cycles [3,4] and biotechnological processes [5–7]. Despite of their importance, most microbial communities are still poorly understood due to their complex nature. Mathematical modeling can help to uncover the interactions and dependencies of the members of these communities. Different modeling formalisms have been used to simulate microbial communities including stoichiometric models, which can be analyzed by constraint-based methods [8–18]. An increasing number of stoichiometric community models considers balanced growth as a key assumption stating that all organisms must grow with the same growth rate in a stable community [11,15,16]. One central goal of these models is the characterization and prediction of possible community compositions and the analysis of the different modes of cross-feeding between the involved organisms.
Stoichiometric models of microbial communities with balanced growth usually result in bilinear models, where, in some equations, independent variables are multiplied with each other. Thus, apart from their increased size, these models have a more complex nature than the linear metabolic models of single species. To make bilinear models amenable to established constraint-based modeling approaches, they can be linearized by fixing either the community growth rate [16] or the community composition [11,15]. In this study, we first provide a unified framework for setting-up, analyzing, and linearizing community models. Even in linearized community models, the application of certain constraint-based techniques becomes quickly infeasible with an increasing number of organisms. Furthermore, one shortcoming of existing methods for modeling of communities is that the solution space often contains unrealistic solutions (where, for example, a species behaves unrealistically altruistic to produce substrates needed by other community members). We therefore introduce a new approach, RedCom, to build reduced community models. The main principle of RedCom is similar to what has been suggested by Taffs et al. [10], namely to compute, in a first step, relevant net conversions of the single-species models which serve as reactions for the reduced model. This reduced model can then be used to identify suitable combinations of single-species net conversions to obtain community-level conversions. However, while Taffs et al. [10] used elementary modes to describe the single-species net conversions, RedCom is based on the more general concept of elementary flux vectors [19,20]. This will be required to ensure balanced growth in the community model and to appropriately account for flux bounds and other (e.g. proteome allocation) constraints. Reduced community models obtained with RedCom do not only focus on most relevant solutions but allow for a comprehensive characterization of solution spaces also for communities with more than only two or three species. In the following, we apply the proposed techniques for different community models with increasing complexity from three up to nine species. The investigated communities are capable of degrading different substrates to biogas, a renewable energy source. Community models of the biogas process give insights on interdependencies and feasible community compositions and may contribute to increase productivity and stability of this process. As one of the first studies, we also compare simulation results from the community models with experimental data of laboratory-scale biogas reactors for growth on ethanol and glucose-cellulose media.
Constraint-based (stoichiometric) modeling of metabolic networks [21] relies on the assumption of a steady-state for internal metabolite concentrations leading to the mass balance equation:
Nr=0
(1)
The structure of the network is captured by the stoichiometric matrix N storing the stoichiometric coefficients of the metabolites (rows) in the metabolic reactions (columns). As consequence of eq. (1), steady-state flux vectors r fulfill the condition that no net accumulation or depletion of internal metabolites occurs. Additionally to the steady-state condition, reversibility constraints (2), flux bounds (3) and other types of inhomogeneous linear constraints (4) can be included:
rj≥0forj∈Irrev
(2)
αj≤rj≤βj
(3)
Ar≤b.
(4)
The set Irrev contains the indices of irreversible reactions. If only the steady-state (1) and the irreversibility constraints (2) are taken into account, the solution space forms a polyhedral (flux) cone; with any constraint of type (3) or (4) its shape becomes a (flux) polyhedron.
In order to create a community model combining all (n) single-species models, herein referred to as full model, a compartmented approach is usually employed [9,11,12,15,22,23]. Each organism represents one compartment and an additional exchange compartment allows for exchange of metabolites (substrates/products) between organisms and with the medium (Fig 1). With the new exchange compartment, the former external (unbalanced) metabolites become now internal ones and must be balanced in eq. (1). Exchange metabolites used by several species are combined such that they exist only once in the community model.
As described in [15] the units of the (specific) single-species reaction rates must be adapted to refer to the total community (instead of single-species) biomass. Accordingly, the units of all reaction rates change from mmol/gDWi/h to mmol/gDWc/h. Exceptions are the n biomass synthesis (growth) reactions producing the species biomasses BMi from a (species-specific) set of precursors:
γi,1pi,1+γi,2pi,2+⋯+γi,qpi,q→1BMi[gDWi](i=1…n)
(5)
In the single-species models, the specific (growth) rates μi (i = 1…n) of these n reactions referred to unit 1/h, which is now changed to gDWi/gDWc/h. We indicate the changed units of these reaction rates in the community model by the symbol μ˜i(i=1…n).
Furthermore, n new pseudo-reactions are introduced in the community model to describe the integration of the n species biomasses into the community biomass BMc (Fig 1):
1BMi[gDWi]→1BMc[gDWc](rate:rBMi→BMc[gDWi/gDWc/h])(i=1…n)
(6)
Finally, a new community growth reaction is introduced “exporting” the synthesized community biomass to the medium (Fig 1); the rate of this reaction is the community growth rate μc [1/h]:
1BMc[gDWc]→(rate:μc[1/h])
(7)
Note that, in steady state, μ˜i=rBMi→BMc and ∑i=1nμ˜i=∑i=1nrBMi→BMc=μc. The obtained structure of the whole community network is captured in the community stoichiometric matrix Nc and the reaction rates in the community flux vector rc (with units as described above). As for the single-species models, we demand steady-state for the metabolites (including all metabolites in the exchange compartment):
Ncrc=0.
(8)
In a stable continuous culture, the growth rate of microorganisms is typically equal to the dilution rate. We assume that the same is true for a microbial community cultivated in a continuous process. In that case, the growth rates μi of all organisms (each normalized to the respective specific biomass) must be identical and equal the community growth rate μc:
μ1=μ2=⋯=μn=μc.
(9)
This concept of balanced growth of microbial communities has previously been proposed by Khandelwal et al. (2012) and is also an underlying principle of the OptDeg [15] and the SteadyCom [16] approach. It has been argued that, even if there is no steady state in a continuous cultivation, the specific growth rates of the organisms need to be the same on average because otherwise the fastest organism would outgrow the others. With constant growth rates, also the fractional biomass abundances
Fi=BMiBMc
(10)
of each species i in the community biomass BMc must be constant. The fractions Fi define the community composition F = (F1,…,Fn) and sum up to unity:
∑i=1nFi=1.
(11)
With balanced growth, the fraction Fi of species i is given by the ratio of the specific biomass production rate of species i (normalized to the community biomass) and the community growth rate:
Fi=BMiBMc=rBMi→BMcμc=μ˜iμc
(12))
Note that the fractional contributions to the synthesis of the community biomass (μ˜i=rBMi→BMc; normalized to BMc) are not identical over the species, hence, the μ˜i need not fulfill (9). However, for the specific growth rates μi (referring to BMi) it holds that μi=μ˜i/Fi=μc and thus (9) is indeed satisfied. For each species, we can rewrite (12) to the following constraint:
rBMi→BMc=Fiμc.
(13)
(Alternatively we could also use μ˜i instead of rBMi→BMc in this equation). In the optimization problems considered below, constraints of type (13) need to be included only for n−1 species, because (6), (7), and (11) already imply (13) for the n-th species: rBMn→BMc=μc−∑i=1…n−1rBMi→BMc=μc−∑i=1…n−1Fiμc=μc−(1−Fn)μc=Fnμc.
Due to the re-normalization of the reaction rates from specific to community biomass, as the last step in assembling the community model we also need to adjust the normalization of the original flux bounds (3) and other inhomogeneous conditions (4) by multiplying them with the fractional abundances:
Fiαij≤rijc≤Fiβij
(14)
where αij and βij are the lower and upper bounds for reaction j in organism i and rijc is the reaction rate of reaction j in organism i in the community model. Likewise, constraints (4) are adjusted for each organism to
Airic≤Fibi
(15)
(Ai, bi correspond to the respective variables in (4) for species i). The irreversibility constraints for the reaction rates are kept from the single-species models:
rijc≥0forj∈Irrevi.
(16)
To exclude solutions with non-zero fluxes rijc≠0 in organisms that are not present in the community (Fi = 0), we assume that every flux in species i is bounded (i.e., αij and βij are bounded). With (14), a non-zero flux rijc then implies Fi>0. In principle, with the chosen constraints, one can also consider the case where the community is not growing (μc = 0), i.e., where dependencies arise exclusively from the maintenance metabolism of the participating species. However, if the community is growing (μc>0), a non-zero flux rijc≠0 in species i implies again Fi>0 and, due to (13), then also rBMi→BMc=μ˜i>0.
In analogy to classical flux balance analysis (FBA) in single organisms, we may formulate a (linear) objective function maximizing certain (combinations of) reaction rates in the community model:
MaximizecTrcs.t.(8),(11),(13)−(16)
(17)
Due to the multiplication of (independent) variables in constraints (13), the community model and the associated optimization problem become bilinear. While non-linear solvers can be employed to solve the optimization problem (e.g., to search for maximum community growth rates or to scan feasible ranges of fluxes or community compositions; see below), a linearization can be applied to enable application of standard linear programming solvers and methods routinely used in (linear) constraint-based modeling.
Two approaches have been used to linearize bilinear community models and to simplify its analysis (Fig 2). In the first approach (utilized in SteadyCom [16]), the community growth rate μc is fixed to a constant (known) value. The constraints (13) become then linear and the optimization problem (17) thus treatable with standard linear programming (LP) solvers. Linearization by fixing the community growth rate is useful, for example, to analyze which community compositions are feasible for a given community growth rate. Repeating these analyses (in discrete steps) for the feasible range of community growth rates yields a more complete picture of the whole solution space.
An alternative linearization method was used in community FBA [11] and in the OptDeg approach [15]. Here, instead of the community growth rate, the community composition, i.e. all the fractional abundances Fi, are fixed. Eq (13) becomes then again linear allowing the utilization of LP solvers. With given fractional abundances, constraint (11) can be removed from the optimization problem (17).
This second linearization approach is useful to scan, for example, the feasible community flux space for a given community composition. However, with a growing number of organisms, this scanning becomes very expensive in terms of the number of linear programs to be solved [16]. In this study, we therefore linearize community models by fixing μc as proposed in the SteadyCom approach. We used an iterative approach to find the maximum community growth rate μc,max in these linearized models. First, we set μc to a value of 0.005 h-1. If a feasible flux distribution exists (here, any (including a zero) objective function can be used in Eq (17)), we double μc and check again for a feasible flux distribution. We repeat these steps until no feasible flux distribution is found. We then take the average of this μc and the last feasible μc (or zero if the first μc did not yield a flux distribution). These steps are repeated (check for feasibility, use average of latest feasible and infeasible μc as new constraint and check again for feasibility) until the difference between the last feasible and infeasible μc is smaller than 0.00001 h-1.
Generally, for both linearization variants, apart from the FBA-like optimization in (17), other constraint-based methods like flux variability analysis (FVA) or metabolic pathway analysis based on elementary flux modes or elementary flux vectors can be carried out (see below).
The described approaches for modeling communities under balanced growth can be used to define and analyze community solution spaces. However, these solution spaces often include unrealistic solutions on the species-level (e.g., a species synthesizes, without any benefit for its own growth, products required by another species in the community [15]). Consequently, the predicted ranges for community compositions or growth rates may become very large as they include many non-relevant phenotypes. FBA could be used to find community compositions fulfilling certain optimality criteria, but the question of suitable objective function in communities arises. In single-species models, a typical objective function is maximization of the growth rate. In community models we can also maximize the community growth rate [11]. But, again, even these optimal solutions may represent unrealistic community compositions in which some organisms waste substrate to ensure survival of the others [15]. We therefore proposed previously an optimization approach to minimize a weighted sum of substrate uptake rates to find community compositions in which all organisms grow with their maximum biomass yields [15]. This approach enabled us to narrow down the solution space to community compositions in which all organisms grow optimally with their maximum biomass yields at a given community growth rate. When introducing our model reduction approach below, we will use a similar method to exclude unrealistic community flux distributions.
Elementary flux modes (EFMs) are non-decomposable flux vectors fulfilling Eqs (1) and (2) [24]. EFMs represent balanced pathways or cycles and have become an important tool for exploring metabolic networks [20,25–28]. However, one shortcoming of EFMs is that inhomogeneous constraints (Eqs (3) and (4) in the single-species models and (14) and (15) in the community model), such as non-growth associated ATP maintenance demand and substrate-uptake limits, cannot be considered. We therefore make use of the concept of elementary flux vectors (EFVs), a generalization of EFMs which can account for inhomogeneous constraints [19,20]. From the theory of EFVs, it is known that the flux polyhedron P resulting from a set of linear constraints is generated by convex combinations of bounded EFVs pk and conic linear combinations of unbounded EFVs xi and yj:
P={r∈ℜn|r=∑k∈Kγkpk+∑i∈Iαixi+∑j∈Jβjyj,γk≥0,∑k∈Kγk=1,αi≥0}
(18)
Due to combinatorial explosion, EFVs can usually only be calculated in medium-scale metabolic networks and, thus, only in smaller community models combining the central metabolism of two or three species.
We present RedCom, a new method to generate community models of reduced size and with reduced solution spaces excluding unrealistic community behaviors. The main idea of the reduction approach, which has some similarities with but is not identical to an approach presented by Taffs et al. [10], is to describe the metabolism of each organism by certain net conversions taken from the EFVs of the single-species models (Fig 2). Since we are mainly interested in community compositions and metabolic interactions (exchange reactions) between the community members, it is often sufficient to focus only on net conversions of the respective species instead of taking its whole metabolic reaction network explicitly into account. Furthermore, from the list of all net conversions of a species we select only those that obey certain optimality criteria avoiding unrealistic phenotypes in the community model. The selected net conversions are used as reactions in the reduced community model to be built. The construction of reduced community models with the RedCom approach is described in the following, a detailed example is given in S1 Text in the Supplements.
All models presented in the Results section were implemented and analyzed with CellNetAnalyzer version 2018.1, a MATLAB package for structural and functional analysis of metabolic and signaling networks [29,30]. CPLEX was used as a solver for linear optimizations and efmtool for computation of EFVs. For solving bilinear problems, we used the fmincon solver for nonlinear optimization in MATLAB. The solver needs an initial flux distribution that we retrieved from the linearized model.
Experimental data from a laboratory-scale biogas reactor on a defined glucose-cellulose medium were published earlier [31] and used for a comparison with predictions from the nine-species biogas producing community (see Results). The data were taken from steady-state conditions [31]. We calculated the average methane and CO2 production rates over a course of 100 days. To achieve steady-state conditions, the reactors were operated under similar conditions for 190 days prior to this time period. Additionally to the data already published, we estimated biomass dry weights by measuring protein concentrations with the Lowry Assay [32] and dividing them by the factor 0.64 (assumed fraction of protein of the total biomass in the model). We used these data to calculate specific production and consumption rates for comparison with simulation results.
A detailed description of the procedures applied for inoculation, feeding, and sample analyses along with cultivation setup and parameters is given in the S6 Text. Briefly, two 1.5 L bioreactor systems were inoculated with sludge from the aforementioned enrichment and fed with the same medium containing 14.6% (v/v) ethanol as main carbon source instead of glucose and cellulose. After an adaption period, continuous cultivation mode was initiated using constant feeding rates and volume control. In the following, different dilution rates were sampled at steady-state conditions, starting from 5.3∙10−4 h-1 further increasing until the biogas production collapsed. Sampling and subsequent analyses comprised pH, biomass protein content, biogas composition and biogas volume produced. In addition, samples were analyzed for residual ethanol and accumulated organic acids. Finally, taxonomic analysis was carried out using an established MS-based metaproteomic workflow (see S8 Text).
In the following, we first describe the construction of metabolic models of nine organisms capturing major degradation steps in the biogas process. Subsequently we combine these single-species models to community models of increasing complexity containing three, six, and nine strains. For each considered community, we construct, analyze and compare three different types of models (bilinear model, linearized full model, reduced model obtained with RedCom) as described in the Methods section. For the six- and nine-species communities, we compare model predictions with experimental data.
Using KEGG [33] and MetaCyc [34] as well as various literature references we manually constructed single-species models of the central metabolism of nine different organisms all being relevant for the biogas process: four primary fermenting bacteria (Acetobacterium woodii, Escherichica coli, Clostridium acetobutylicum, Propionibacterium freudenreichii), three secondary fermenting bacteria (Syntrophomonas wolfei, Syntrophobacter fumaroxidans, Desulfovibrio vulgaris) and two methanogenic archaea (Methanospirillum hungatei and Methanosarcina barkeri). As suggested by Taffs et al. [10], we consider each of these organisms as one functional guild in the biogas process with certain metabolic properties. More specifically, under anaerobic conditions, E. coli produces ethanol as well as different organic acids like formate, lactate, acetate and succinate from glucose, glycerol and gluconate. A. woodii is an homoacetogenic organism that can either ferment sugars like glucose and fructose but also lactate, formate or hydrogen and CO2 to produce acetate via the Wood-Ljungdahl pathway [35,36]. P. freudenreichii can ferment glucose, glycerol and lactate to succinate and propionate. The organism uses the methyl-malonyl-CoA pathway to produce propionate. Some organisms using the methyl-malonyl-CoA pathway like Pelobacter propionicus are also capable of using ethanol as a substrate [37]. Since we aimed to represent the functional guild of propionate producing bacteria using the methyl-malonyl-CoA pathway, we also added ethanol oxidation to propionate to the model. C. acetobutylicum ferments glucose and glycerol to different organic acids and solvents like acetate, butyrate, ethanol, butanol and aceton. The organism is known to grow in two different phases [38]. In the first phase, the organism produces organic acids like acetate and butyrate. These pathways have high ATP yields but the acids produced lower the pH in the medium. In the second phase, acids are taken up and solvents like butanol and aceton are the main product. C. acetobutylicum represents primary fermenting bacteria in our community model and we assumed that mainly production of formate, acetate, butyrate and ethanol is relevant in anaerobic digestion. We therefore disabled production of the other solvents in the community model. D. vulgaris is a sulfate-reducing bacterium that can grow on organic substrates like pyruvate, lactate and ethanol using sulfate or thiosulfate as an electron acceptor. In the absence of electron acceptors, the organism can also grow in syntrophic associations with hydrogen utilizing organisms. The products formed by D. vulgaris are either acetate and hydrogen plus CO2 or formate (in syntrophic cultures) or acetate plus hydrogen sulfide (when sulfate is present). Additionally, the organism can utilize hydrogen with acetate as a carbon source and sulfate as an electron acceptor. S. fumaroxidans can grow on propionate in syntrophy or with sulfate as an electron acceptor [39]. In pure culture the organism can grow on fumarate, fumarate plus propionate or succinate, formate or hydrogen plus sulfate [39]. S. wolfei is a secondary fermenting bacterium that can degrade saturated fatty acids from butyrate through octanoate either to acetate and hydrogen (even number of C-atoms) or to acetate, propionate and hydrogen (odd number off C-atoms) in syntrophic cultures [40]. Growth of S. wolfei is also possible on crotonate in monoculture [41].
The methanogenic organism M. hungatei (cytochrome-free) produces methane from formate or from hydrogen plus CO2 while M. barkeri (possesses cytochromes) can use hydrogen plus CO2, acetate, methanol and methylamines for methanogenesis. In addition to different substrates utilized by the methanogens they also differ in ATP yields and substrate affinities. M. barkeri has higher ATP yields but lower substrate affinity for hydrogenotrophic methanogens. In our M. barkeri model we only implemented methanogenesis from acetate, methanol, and hydrogen with CO2.
A summary of the single-species models with model dimensions (number of metabolites and reactions) and constraints is given in Table 1. The models of D. vulgaris, M. barkeri and M. hungatei were published before [15]. We estimated flux bounds for substrate uptake and product formation from experimental data or existing models, partially also from closely related organisms (see S3 Text). Maintenance coefficients (rATPmaint) were taken from literature data but the reported values varied by more than one order of magnitude between the different species (Table 1, S3 Text). Below we will therefore carry out a sensitivity analysis to investigate the influence of the maintenance coefficients on simulation results. For model validation, we also compared model predictions with measured biomass yields reported in the literature (see S4 Text). All models are listed (and also provided in SBML format) in S1 Table in the Supplements.
For the simulations performed in this work, we focused on ethanol (three and six-species community) and glucose (nine-species community) as the only available substrates and switched the uptake of other substrates (glycerol, gluconate, methanol, fructose, sulfate) off to reflect the composition of media used in the experiments.
We investigated a three-species community model (Table 2) consisting of D. vulgaris, M. hungatei and M. barkeri. This community can convert ethanol to methane, CO2, and acetate and thus covers the last two steps of anaerobic digestion. A similar community was experimentally investigated by Tatton et al. [42] and simulated with FBA in a previous study [15]. In analogy to the study of Tatton et al. [42], the uptake of external CO2 was allowed to also include solutions in which the acetoclastic methanogen is non-essential.
We extended the three-species community model to a model with six of the nine model organisms by additionally integrating A. woodii, P. freudenreichii, and S. fumaroxidans (Tables 1 and 2). The three additional organisms were chosen according to their potential of being part in an ethanol-degrading community; they represent functional guilds that extend the capability of the three-species community investigated above by additional pathways for homoacetogenesis and propionate fermentation. Growth of the other (remaining) three organisms (CA, SW, EC; see Table 1) is not supported with ethanol as substrate and they have therefore not been included yet. Note that, at this initial point, no experimental data have been used yet to adjust the composition of the community model; this will later be done when including metaproteomic data from a concrete enrichment culture.
We finally simulated a community capable of growth on glucose. Here, all of our nine guild organisms can potentially be involved in the process and are thus part of the community model (Tables 1 and 2). In addition to the six-species community studied above, this model included E. coli, C. acetobutylicum and S. wolfei. We first simulated the community with the bilinear model to predict the maximum community growth rate as well as ranges for substrate uptake, product excretion, biogas composition and methane yield (Table 5). As already observed for the six-species model, a reliable prediction for μc,max was thus not possible with this model (with the iterative approach in the linearized models we found that μc,max = 0.23 h-1) In contrast, the predicted ranges for reaction rates and yields seem reasonable.
We then compared predictions of the linearized full model and the reduced model with experimental data (Table 5) from an enrichment culture grown on glucose-cellulose medium ([31]; see also Methods). Data were available for two duplicate experiments with identical dilution rate. Since hydrolysis of cellulose is not included in the model, we used glucose as a starting point and assumed that cellulose is converted to glucose by hydrolytic enzymes. We set the community growth rate to 0.00067 h-1, which corresponds to the dilution rate of the experiment and derived the corresponding linearized full community model and the reduced community model. EFV computation was possible with the reduced model (213689 EFVs) but not with the full model where we computed only ranges for biomass compositions, exchange rates, and methane yield via flux variability analysis (Table 5 and Fig 8).
Confirming findings from the three- and six-species models, we observed that the predicted ranges, especially of exchange rates and community compositions, are again considerably smaller in the reduced model compared to the linearized full model. In fact, the calculated ranges of exchange rates of the linearized full model are almost identical to the ones from the bilinear model, although the latter did not consider a fixed growth rate. The measured exchange rates were only slightly smaller than the minimum rates predicted by the models. The predicted ranges of the reduced model lie on the lower end of the range of the linearized full model and are thus closer to the experimental data indicating that the organisms use their substrate efficiently as assumed by our model reduction approach (Table 5 and Fig 8). The slight overestimation of the rates could again be a consequence of overestimating maintenance coefficients or an underestimation of ATP yields in the models. Furthermore, we noticed a relatively high variance of the measurements for the exchange rates which may partially explain deviations between data and model predictions. We also measured higher methane to CO2 ratios and lower methane yields than predicted by the models. Typically, we would expect a ratio of 1 methane to one CO2 for carbohydrates like glucose. However, some of the released CO2 might have been lost due to its better solubility in water (compared to methane).
Microbial communities are of major importance for health, nature, and biotechnological applications. Constraint-based stoichiometric modeling helps to obtain a better understanding of interrelationships in these communities and to make quantitative predictions. However, compared to classical constraint-based modeling of single-species metabolic networks, analysis of community models based on the favored concept of balanced growth is hampered by four major technical difficulties:
Our introduced RedCom approach, where reduced community models are constructed from net conversions of the linear single-species models, addresses three of the above four issues ((2)-(4)). Taffs et al. [10] also published an approach where EFMs (instead of EFVs) of single-species models were used as input for the community model (“nested pathway consortium analysis approach”). While the basic principle is the same, our RedCom approach uses EFVs instead of EFMs which is mandatory to guarantee balanced growth of the community and to allow the consideration of flux bounds, maintenance coefficients, and other inhomogeneous constraints. A necessary pre-processing step is the calculation of EFVs in the single-species models for the fixed community growth rate followed by the selection of relevant EFVs projected onto their exchange fluxes. Different optimization or selection criteria can be used for selecting the relevant single-species behaviors. We decided to use all EFVs representing minimal conversions of exchange metabolites, which, as one particular advantage, ensures exclusion of unrealistic (altruistic) community behaviors of the respective species (see point (3)). Dependent on the application, other criteria could be used as well. In the three-, six-, and nine-species community models considered herein, the RedCom approach led to community models with desired properties: the models (a) are much smaller than the full (linearized) models, (b) exclude many spurious solutions, and (c) are amenable for detailed EFV analysis enabling the extraction of many important features of the community while avoiding an elaborate scanning of the solution space. There are two potential disadvantages of the reduction approach. First, the reduced community model contains information on the exchange fluxes while the internal flux distributions are not visible. However, in most applications of community models, the focus is indeed on predictions on the exchange fluxes, product yields, and feasible community compositions, which can all be derived from the respective flux vector of the reduced model. Furthermore, internal flux distributions of single-species could be “unpacked” from particular community net conversions whenever needed. A second potential disadvantage concerns the calculation of EFVs from the single-species models, which is usually not feasible if the latter are at genome-scale. However, with the typical application focus on exchange fluxes, single-species metabolic network models at the level of the central metabolism seem to be sufficient in many cases. Third, since the reduced community model requires eventually only the (minimal) net conversions of the single-species models, the (direct) calculation of elementary conversions might be a feasible approach even in genome-scale models [45].
We applied our RedCom approach to build community models of up to nine species relevant for the biogas process. We used a compartmented approach where each functional guild in anaerobic digestion is represented by a core model (central metabolism) of one organism. For the respective communities, we analyzed the maximum community growth rate and the feasible ranges of exchange rates, yields and fractional abundances of the involved species—with the bilinear as well as with the linearized and the reduced community model. Results were always consistent (in bilinear models, as long as the solver could reliably compute the respective minima and maxima). However, the reduced models obtained with the RedCom approach show a significantly narrower solution space by excluding solutions from the single-species models that are physiologically very unlikely resulting in more conclusive model predictions. While bilinear community models are usually linearized to make them amenable to constraint-based analysis techniques, we found that they can, in principle, be used to roughly gauge the community’s solution space. However, in larger models, some solutions found by the solver, especially for the determined maximum community growth rate, depended on starting values used for the solver pointing to potential issues with finding the global optimum in this non-linear optimization problem. Whenever the community growth rate can be fixed (e.g., to the maximum growth rate or to the dilution rate used in an experiment), the bilinear model becomes linear making its analysis and calculations simpler. With increasing numbers of organisms the computational costs, e.g. for an FVA-based scanning of the solution space, increase drastically also in the linearized (full) community model and EFVs could only be computed for models of up to four organisms. With the reduced community models, we were able to compute and analyze EFVs also for the largest community consisting of nine organisms.
In order to compare simulation results with experimental data from biogas communities and to investigate which solutions of the solution space are the most relevant in a concrete culture, we carried out experiments with an ethanol enrichment culture for different dilution rates. First, we compared experimental data with predicted ranges for specific substrate uptake and product formation rates as well as for biogas composition and methane yields obtained from the linearized full and the reduced six-species model. The predicted ranges of the specific rates covered the measured values but were very large and thus of low predictive power, especially in the full model. Confirming earlier findings [15], the maintenance coefficient of the different species has a tremendous impact on many properties of the community, especially on the predicted rates and community composition. Therefore, the maintenance coefficient should be determined as precisely as possible to obtain valid community models. As many other microbial communities, anaerobic digestion communities usually have a low growth rate implying that a relatively large fraction of the metabolism is devoted to maintenance processes. Generally, our results for the anaerobic digestion community indicate that the best agreement of model predictions and experimental data can be achieved when the maintenance coefficients of all species are approximately set to 1 mmol/(gDW∙h). In contrast to rates and community compositions, the predicted ranges for methane yields and biogas composition were much smaller and appeared to be less sensitive to the maintenance coefficients making these model predictions generally more reliable. In fact, methane yields and biogas compositions from the experiments were close to the predicted values for both the reduced and the full model. The predicted maximum growth rate of the full and the reduced six-species community model were identical but considerably higher than the maximum dilution rates that supported a stable process with the enrichment culture. Here, maximization of the community growth rate might not be a suitable objective function for communities in a realistic continuous process. In particular, maximum substrate uptake rates used in the models are usually derived from single-species cultures under their respective optimal conditions and it is likely that process conditions do not support optimal conditions and maximum growth rates for all organisms. The slowest (essential) species will then limit the overall community growth rate.
We used metaproteomic data from enrichment cultures for growth on ethanol to find out, which taxonomies and pathways were present in these cultures and to use this information to build a more constrained community model for this culture. The most abundant taxonomic orders identified in the experiments were Methanosarcinales, Methanomicrobiales, Methanococcales, Methanobacteriales and Desulfovibrionales, which correspond to the guilds represented by M. barkeri, M. hungatei and D. vulgaris in our six-species model. Furthermore, we found enzymes for ethanol oxidatation in Desulfovibrionales, acetoclastic and hydrogenotrophic methanogenesis in the archaeal superkingdom. There was little to no evidence for syntrophic acetate oxidation, homoacetogenesis and ethanol oxidation to propionate, which agrees well with the taxonomic analysis. In a last step, we used that information to further constrain the reduced six-species model and explored options to predict community compositions from the remaining solution space. The model predicted M. barkeri to be the dominant species in the community and D. vulgaris to be the least abundant organism. In fact, Methanosarcinales was also the taxonomic order with the highest spectral count abundance in the experiments while D. vulgaris had the lowest abundance confirming the model predictions. The experimental data also indicated that mainly Methanosaeta species were involved in acetoclastic methanogenesis. These organisms grow with lower biomass yield but higher substrate affinity compared to Methanosarcina species. Therefore, Methanosaeta should be added as a separate guild to the community model for future studies. Overall, to the best of our knowledge, the presented model-driven analysis of metaproteomic data from communities involved in anaerobic digestion is the biggest of its kind reported so far and demonstrates the high potential of a computer-aided approach to investigate properties and to assess experimental data of microbial communities.
|
10.1371/journal.pgen.1006738 | DAF-18/PTEN signals through AAK-1/AMPK to inhibit MPK-1/MAPK in feedback control of germline stem cell proliferation | Under replete growth conditions, abundant nutrient uptake leads to the systemic activation of insulin/IGF-1 signalling (IIS) and the promotion of stem cell growth/proliferation. Activated IIS can stimulate the ERK/MAPK pathway, the activation of which also supports optimal stem cell proliferation in various systems. Stem cell proliferation rates can further be locally refined to meet the resident tissue’s need for differentiated progeny. We have recently shown that the accumulation of mature oocytes in the C. elegans germ line, through DAF-18/PTEN, inhibits adult germline stem cell (GSC) proliferation, despite high systemic IIS activation. We show here that this feedback occurs through a novel cryptic signalling pathway that requires PAR-4/LKB1, AAK-1/AMPK and PAR-5/14-3-3 to inhibit the activity of MPK-1/MAPK, antagonize IIS, and inhibit both GSC proliferation and the production of additional oocytes. Interestingly, our results imply that DAF-18/PTEN, through PAR-4/LKB1, can activate AAK-1/AMPK in the absence of apparent energy stress. As all components are conserved, similar signalling cascades may regulate stem cell activities in other organisms and be widely implicated in cancer.
| Adult stem cell proliferation rates respond to the needs for their differentiated progeny. For example, the lung stem cells of a smoker will divide more frequently than those of a non-smoker to constantly replenish smoke-damaged tissues. Molecularly, stem cell proliferation rates vary according to growth factor levels, some of which, such as insulin, are regulated systemically. What remains unclear is whether and how a more or less rapid turnover of differentiated cells interferes with growth factor signalling to ultimately set the stem cells’ proliferative rate. We show here that this is accomplished though a novel cryptic signalling pathway that locally antagonizes the effects of insulin through inactivating a more locally-acting growth factor pathway (ERK/MAPK signalling). Interestingly, a ubiquitous protein kinase (AMPK) that has primarily been implicated in energy stress response constitutes an integral part of this feedback mechanism. This suggests that an abundance of differentiated cells is molecularly interpreted as being similar to energy stress at the level of the stem cells, likewise leading to the inhibition of their growth and proliferation.
| Stem cells can divide symmetrically to generate more copies of themselves or asymmetrically to produce one stem cell and a daughter cell destined for differentiation. Whether a stem cell will proliferate or differentiate is governed by niche signaling, which locally prevents differentiation [1]. Stem cell proliferation rates on the other hand, are largely controlled by growth factors that act in parallel to niche signaling to stimulate stem cell activity independent of fate [2–9]. Some of these growth factors are systemically controlled, such as insulin/IGF-1 signalling (IIS), which relays nutrient uptake information to cell growth/proliferation, largely through stimulating protein synthesis by activating the mTOR complex [10–12].
In C. elegans (Ce), activated IIS stimulates germline stem cell (GSC) proliferation throughout postembryonic life [2,3,9]. Genetic inactivation of IIS during early larval development leads to entry into a developmentally-arrested stage called dauer in which GSCs are fully quiescent [2,13], while IIS inactivation at later stages of development or adulthood dramatically slows GSC proliferation rates [3,9]. Dauer entry is naturally promoted by food deprivation, which causes energy stress and leads to a down-regulation of IIS and activation the AMP-activated protein kinase AAK (Ce AMPK) [14], a master intracellular energy-stress sensor [15]. In this context, the proper establishment of GSC quiescence downstream of insulin receptor inactivation requires DAF-18 (Ce PTEN), PAR-4 (Ce LKB1; an AMPK-activating kinase [15]), as well as germline AAK activity [2,14].
In addition to being subject to systemic growth factor regulation, the proliferation rates of certain stem cell populations are also affected by the demand for their differentiated progeny. This was first demonstrated in Drosophila intestinal stem cells, the proliferation of which can be stimulated by cytokines released by damaged intestinal epithelial cells [5]. In Drosophila hematopoietic progenitor cells, proliferation was found to be inhibited by differentiated hemocytes through an adenosine deaminase growth factor A (Adgf-A) signal [16]. In mammals, hair follicle stem cell proliferation is stimulated by their transit-amplifying progeny once these begin producing Sonic hedgehog (Shh) [17]. Interactions between stem cells and their differentiated progeny is thus a conserved phenomenon that is likely to be involved in the regulation of diverse types of stem cells.
Recently, we found that the proliferation rates of the C. elegans GSCs were negatively influenced by oocyte accumulation in sperm-less or sperm-depleted hermaphrodites [9]. C. elegans hermaphrodites produce a finite number of sperm during larval development, then switch to oocyte production during adult life, which continues until sperm stores are depleted. Sperm constitutively secrete major sperm proteins (MSPs) that activate cAMP signaling in the proximal somatic gonad. cAMP leads to activation of OMA-1 and OMA-2 in the proximal-most oocyte, triggering oocyte maturation (i.e. the transition between diakinesis and metaphase of meiosis I, which is accompanied by nuclear envelope breakdown, rearrangement of the cortical cytoskeleton, and meiotic spindle assembly) and ovulation [18–21]. Thus, when sperm is absent or depleted, oocytes do not mature and begin to accumulate in the proximal gonad, eventually leading to the inhibition of further oocyte production, as well as the inhibition of GSC proliferation at the distal end [9,19,22]. Indeed, the GSCs of sperm-less or sperm-depleted hermaphrodites enter G2/M quiescence with stochastic transient bouts of proliferation, leading to an overall drastically reduced mitotic index (MI) [9,23].
Feedback inhibition of GSC proliferation by oocyte accumulation requires DAF-18 activity, which acts to locally antagonize systemic IIS information and block GSC proliferation, specifically in the gonad arm that is sperm-depleted and has accumulated oocytes. In this situation, DAF-18 does not block GSC proliferation by directly inhibiting IIS through antagonizing AGE-1 (Ce PI3K) activity because the transcription factor that AGE-1 eventually inhibits, DAF-16 (Ce FOXO), remains inactivated in sperm-less animals and is dispensable for feedback GSC regulation [9]. Thus, we reasoned that DAF-18 was inhibiting GSC proliferation in parallel to IIS via another, yet unidentified, signalling mechanism. Our work here demonstrates how DAF-18 negatively regulates MPK-1 (Ce MAPK) function through activating PAR-4 and AAK-1 (Ce AMPK α1-catalytic subunit), in the absence of any apparent stress, to locally couple GSC proliferation rates to the need for their differentiated progeny.
We first sought to identify the regulators that, together with DAF-18 and independently of DAF-16, promote GSC quiescence in sperm depleted animals. Because PAR-4 and AMPK function with DAF-18 to promote C. elegans GSC quiescence independently of DAF-16 during dauer development [2], we tested whether PAR-4 and/or AMPK were similarly required to promote GSC quiescence in sperm-depleted, adult day 4 (A4) hermaphrodites. We found that GSC quiescence was not induced in sperm-depleted A4 animals bearing either a strong par-4 loss-of-function allele [24], or null mutations in both AMPK α-catalytic subunits (aak-1 and aak-2) (Fig 1A and 1B). These mutations similarly disrupted the promotion of GSC quiescence at A1 in sperm-less mutants, such that the GSC MI of fog-1; par-4 and fog-1; aak-1; aak-2 animals was indistinguishable from that of wild-type animals or fog-1; daf-18 double mutants (Fig 1A and 1C). We further noticed that both fog-1; par-4 double mutants and fog-1; aak-1; aak-2 triple mutants did not accumulate a large number of unfertilized oocytes in their proximal gonad, a feature of fog-1 single mutants. Accordingly, the proximal-most oocyte in these animals spontaneously matured in the absence of sperm and was ovulated (Figs 1D and S1), a phenotype also observed in fog-1; daf-18 mutants [9]. These results indicate that PAR-4 and AMPK are, like DAF-18, required to promote oocyte and GSC quiescence in adult hermaphrodites following sperm depletion.
During dauer entry, both AMPK α-catalytic subunits have a similar and additive effect on the induction of GSC quiescence [2]. We wondered if they acted likewise for the establishment of oocyte and GSC quiescence following sperm depletion in adult animals. Surprisingly, we found that the removal of aak-1 alone fully recapitulated the loss of daf-18 or par-4, both in terms of quiescent oocyte accumulation and GSC MI regulation (Fig 1A, 1C and 1D). The loss of aak-2 only had a marginal, yet statistically significant, effect on oocyte accumulation (Fig 1C and 1D). Therefore, AAK-1 is required and sufficient for the induction of both oocyte and GSC quiescence in the absence of sperm. Together with the existing literature [2,9,14], these results suggest that upon sperm depletion, downstream or in parallel to DAF-18, PAR-4 phosphorylates and activates an AAK-1-containing AMPK complex that prevents oocyte maturation. This leads to oocyte accumulation in the proximal gonad arm, which subsequently promotes GSC quiescence in the distal end. The difference in aak catalytic subunit requirements between dauer (both aak-1 and aak-2) and adult feedback (aak-1 only) regulation of GSC proliferation supports the notion that the two responses are mechanistically different.
All of the mutations uncoupling GSC proliferation from sperm availability also prevent oocyte quiescence, retention, and accumulation (Figs 1C, 1D and S1; [9]). Although we had previously suspected that oocyte accumulation triggers the inhibition of GSC proliferation, no condition had allowed us to directly ascertain this. We reasoned that OMA-1/2 were likely to be required for the ectopic oocyte maturation that we had observed in sperm-less or sperm-depleted aak-1 mutants, thus potentially allowing us to uncouple oocyte accumulation and feedback signalling. In these oma-1; oma-2 double mutants, sperm is normally produced and activates MSP signaling in the proximal somatic gonad, but the oocytes fail to mature and they accumulate. As in sperm-less fog-1 or fog-2 mutants (we did not observe any phenotypic differences between fog-1 and fog-2 for any of the measured parameters [9], and thus these two backgrounds were used interchangeably), GSC proliferation is also reduced in oma-1; oma-2 A1 double mutants (Fig 2A–2E; [9]). oma-1; oma-2 double mutants thus phenotypically resemble fog mutants although they accumulate a smaller number of variably-sized oocytes that do not appear as compressed as those in the germ lines of fog mutants [9,18] (also refer to Fig 3A–3C).
We thus generated aak-1; oma-1; oma-2 triple mutants to see if this would prevent oocyte maturation and induce oocyte accumulation in animals defective for feedback signalling to the GSCs. As expected, we found that oocytes did not undergo spontaneous maturation and accumulated in aak-1; oma-1; oma-2 triple mutants. Notably, despite the presence of accumulated oocytes, the GSC MI remained elevated at A1 in these animals (Fig 2E), indicating that oocyte accumulation requires AAK-1 activity to promote GSC quiescence. Strikingly however, oocyte production did not slow down and these animals became filled with arrested, diakinesis-stage oocytes within 1–2 days (Fig 2A–2D). Diakenesis-arrested oocytes in aak-1; oma-1; oma-2 triple mutants accumulated sometimes in multiple, disorganized rows in the gonad arm, filled-up the uterus, and could eventually breach the gonad proper (Fig 2A–2C). Thus, we conclude that AAK-1 is required at three steps for feedback control of GSC proliferation: 1) to permit oocyte accumulation by preventing their spontaneous maturation and ovulation in the absence of sperm/MSP signal, 2) to slow-down oocyte production upon oocyte accumulation, and 3) to inhibit GSC proliferation.
AMPK is activated by direct binding to 5’-AMP, the concentration of which can dramatically increase during energy stress, but also under many other stressful cellular circumstances that affect ATP usage/production. Significant AMPK activation however requires phosphorylation on a conserved activating residue, which is generally performed by LKB1, a constitutively active kinase [25]. Thus, AMPK has emerged as a master metabolic stress sensor that is turned ON in response to various kinds of stresses and acts to implement a modified energy homeostasis balance [15]. Well-fed sperm-less hermaphrodites (fog mutants) have normally activated IIS [9] and a wild-type lifespan [26], suggesting that they do not experience systemic energy stress. However, specific cells could still be subjected to stress during oocyte accumulation, leading to localized, cell-autonomous AMPK activation. Notably, the accumulating unfertilized oocytes in fog mutants undergo visible compaction (Fig 3A and 3B; [19,27]) and could be experiencing mechanical stress. Consistent with this, large ribonucleoprotein (RNP) foci (also known as P-bodies or stress granules) form in the compressed and arrested oocytes of fog mutants [27]. Similar RNP foci also form in the non-arrested (non-compressed) oocytes of sperm-bearing wild-type animals during heat shock, osmotic stress, or anoxia, suggesting that formation of these RNP foci is a general response of oocytes to stress [27]. Thus, AAK-1 activity could be locally triggered in accumulated oocytes due to a localized stress response and act to block the production of further oocytes and GSC proliferation.
To test this hypothesis, we monitored the localization of MEX-3, a germline protein known to accumulate in RNP granules upon stress [27], in fog mutants that were devoid of feedback signalling. We found that RNP foci did not form in the proximal oocytes of fog-1; daf-18, fog-1; par-4 or fog-1; aak-1 mutants (S2A–S2D Fig). However we could not exclude that this was a consequence of the ectopic oocyte maturation phenotype that we observed in these mutants, which prevents oocyte accumulation (and compression). We therefore monitored MEX-3 localization in the arrested oocytes of aak-1; oma-1; oma-2 triple mutants, which accumulate but do not undergo oocyte maturation. We found that RNP foci were absent from arrested oocytes of aak-1; oma-1; oma-2 triple mutants but, to our surprise, they were also absent from the arrested oocytes of oma-1; oma-2 double mutants (Fig 3A–3D). This suggests that the arrested oocytes of oma-1; oma-2 double mutants are not experiencing any stress, unlike the seemingly compressed ones of fog mutants. Thus, the formation of RNP foci in arrested oocytes is not required for feedback regulation of GSC proliferation. These results suggest that AAK-1 is functioning in this pathway as a signalling kinase rather than as an energy or stress sensor.
In both pre-dauer and adult germ lines, DAF-18 is required to suppress GSC proliferation, but can do so independently of DAF-16, the ortholog of FOXO that functions as the main IIS effector in C. elegans [2,9,26,28]. DAF-18 was reported to impinge on the ERK/MAPK pathway independently of DAF-16 during vulva development and meiotic progression [29,30]. While the C. elegans ortholog of ERK/MAPK, MPK-1, is present throughout the germ line, its doubly-phosphorylated active form (dpMPK-1) is detectable in two distinct regions of the adult hermaphrodite germ line: late pachytene-stage germ cells (termed “zone I”) and in growing and maturing oocytes (termed “zone II”) [30–32]. Accordingly, mpk-1 is required for germ cell progression through the pachytene stage and is involved in oocyte growth and maturation [19,31]. Interestingly, it was also reported that null mpk-1 mutants, which can grow into sterile vulva-less adults, have a small germ line, which is compatible with reduced GSC proliferation [31].
We reasoned that if DAF-18 activity acts to down-regulate MPK-1 following sperm depletion, DAF-18 could potentially block GSC proliferation as well as germ cell progression through the pachytene stage. To test this hypothesis, we first verified that mpk-1 was indeed required for optimal GSC proliferation. As expected, we found that the GSC MI of A1 null mpk-1 mutant hermaphrodites was significantly lower than that of wild-type animals, and similar to that of A1 fog-1 or fog-2 mutants (Fig 4A). Thus, mpk-1 activity is required to maintain a high level of adult GSC proliferation. It was previously reported that dpMPK-1 levels decrease in the germ line of sperm-depleted animals, first in the proximal oocytes, and subsequently, but only after oocytes have accumulated, in pachytene-stage germ cells [19,30,31]. To test if the inactivation of MPK-1 is required for sperm depletion to promote GSC quiescence, we up-regulated MPK-1 signalling in fog-1 mutants using a conditional gain-of-function (gf) allele in the gene encoding the C. elegans Ras ortholog, let-60 [33,34]. Under restrictive conditions, A1 fog-1; let-60(gf) double mutants had a GSC MI that was intermediate to that of A1 fog-1 mutants and A1 wild-type animals (Fig 4B). Consistent with a previously-reported phenotype of let-60(gf) mutants [31], fog-1; let-60(gf) double mutants accumulated a greater number of smaller oocytes (Fig 4C). These results indicate that hyperactivation of Ras can alleviate GSC quiescence upon oocyte accumulation and suggest that inactivation of MPK-1 is required for oocyte accumulation to block GSC proliferation.
If daf-18 and mpk-1 were acting as part of the same signalling pathway during feedback control of GSC proliferation, we would expect the effect of the two mutations not to add up, but that one (acting downstream) would cancel the effect of the other. We thus asked if a null mutation in daf-18 modified the low GSC MI phenotype of mpk-1 null mutants. We found that A1 mpk-1; daf-18 double null mutants had a low GSC MI that was indistinguishable from that of mpk-1 single mutants (Fig 4A), indicating that MPK-1 acts downstream of (or in parallel to) DAF-18 to maintain elevated GSC proliferative activity. To confirm that this DAF-18-to-MPK-1 axis of signalling that controls GSC proliferation is acting in parallel to IIS, we tested whether IIS inhibition would be additive to the loss of MPK-1 activity. Strikingly, we found that the GSCs of A1 daf-2 mpk-1 double mutants do not divide and have a null MI (Fig 4A), indicating that adult GSC proliferation absolutely requires at least one of these two growth factors and that they largely act in parallel.
To determine whether the stimulatory effect that the DAF-18-to-MPK-1 axis of signalling has on GSCs depends on IIS, we measured the GSC MI at A1 in both daf-2; par-4 and daf-2; let-60(gf) double mutants. We found that these two mutations, unlike a mutation in daf-18 [9], did not significantly restore the GSC MI in daf-2 mutants (S3 Fig), indicating that the DAF-18-to-MPK-1 axis of signalling does not directly inhibit IIS in wild-type adults, but specifically antagonizes activated IIS in sperm-less hermaphrodites. While loss of either daf-18 or par-4 impairs the establishment of GSC quiescence during dauer development [2], we found that the ectopic activation of MPK-1 signalling in let-60(gf) dauer animals had no significant effect on GSCs (S3 Fig). These results suggest that control of GSC proliferation during dauer formation does not require MPK-1 inhibition, and that MPK-1 inhibition is thus specifically required for feedback control in adults. Altogether, these results suggest that upon sperm depletion, DAF-18 inhibits MPK-1 activity through PAR-4 and AAK-1 to antagonize IIS independently of DAF-16, and block GSC proliferation in response to oocyte accumulation.
To further test this hypothesis, we evaluated the levels MPK-1 phosphorylation on activating residues when oocytes accumulate in the absence of sperm. In wild-type animals, dpMPK-1 levels peak in the -1 oocyte, and the high level of proximal staining is dependent on MSP signalling [18,19]. In young adult oma-1; oma-2 double mutants producing their first few oocytes, dpMPK-1 is transiently detectable in the first oocytes, indicating that they do receive the sperm MSP signal, however it is not maintained thereafter [18]. Thus, MPK-1 activation in zone II requires MSP signalling, but maintenance of MPK-1 activity is further dependent on oocyte maturation and/or the presence of OMA proteins. As previously reported, we found that dpMPK-1 levels in zones I-II decreased to background upon oocyte accumulation in A1 fog-1 and fog-2 mutant hermaphrodites, in oma-1; oma-2 double mutants, as well as in unmated, sperm-depleted A4 wild-type hermaphrodites (Figs 4D–4F, 4M, S4 and S5; [18,19,31]). In contrast, dpMPK-1 levels remained elevated in zones I-II in continually mated A4 wild-type hermaphrodites (Figs 4G, S4 and S5). We found that the decrease of dpMPK-1 levels in zones I-II of Fog and Oma animals requires DAF-18, PAR-4 and AAK-1 activities (Figs 4H–4N, S4 and S5). Of note, loss of DAF-18 led to MPK-1 upregulation ectopically throughout the gonad in A4 animals (Figs 4I, S4 and S5; [30]). These results, in combination with the existing literature, suggest that when sperm is depleted, MPK-1 is first inactivated in zone II of the germ line [19,20,31], and that oocyte accumulation then triggers a feedback signal that further inhibits MPK-1 more distally in zone I. Oocyte accumulation-mediated MPK-1 inactivation, first in zone II and then in zone I, would then be required to block adult GSC proliferation, either directly or indirectly.
To gain insight into where MPK-1 inactivation may be required to inhibit GSC proliferation, we removed mpk-1 activity from gld-3 nos-3 double mutant animals, in which GSC differentiation is completely blocked, preventing the formation of zones I-II altogether, and leading to the formation of germline tumours [35,36]. We found that the lack of mpk-1 activity suppressed the growth of germline tumours in gld-3 nos-3 animals (S5 Fig), indicating that during development, MPK-1 functions either cell autonomously in the GSCs and/or non-cell autonomously in the soma to promote GSC proliferation. Interestingly, we did not measure a significant difference in the GSC MI between gld-3 nos-3 and gld-3 nos-3; mpk-1 A1 animals (S6 Fig). This suggests that during adult life, MPK-1 activation in differentiated germ cells is required to promote GSC proliferation in mpk-1(+) animals. We conclude that upon sperm depletion, DAF-18, PAR-4 and AAK-1 are required to prevent OMA-1 and OMA-2 activation in the absence of MSP signalling and thus permit the accumulation of quiescent oocytes in the proximal gonad. Upon oocyte accumulation, the activation of DAF-18, PAR-4 and AAK-1 induces the inhibition of MPK-1 activity in zones I-II of the germline, which antagonizes IIS signals and induces GSC quiescence while also blocking meiotic progression and the production of further oocytes.
The consensus recognition motif for AMPK is well defined and conserved [15,25], and there is an overlap between a subset of potential AMPK consensus phosphorylation sites and that of the signalling adaptor protein 14-3-3 [25,37,38]. The functional relevance of this motif overlap has been previously demonstrated. For example, phosphorylation of Raptor by AMPK in cultured mammalian cells has been demonstrated to induce 14-3-3 binding, which in turns leads to Raptor inactivation [37]. Interestingly, PAR-5 (the C. elegans germline-expressed ortholog of 14-3-3 [39]) was found to be required for the maintenance of oocyte quiescence in an RNAi screen designed to isolate regulators of oocyte maturation [40]. This raised the possibility that following sperm depletion, binding between PAR-5 and AAK-1-phosphorylated MPK-1 regulators could lead to the inactivation of this pathway to couple both oocyte and GSC quiescence to sperm availability. To test this hypothesis, we first asked if par-5 was required to couple GSC proliferation with oocyte needs in C. elegans adults. We found that RNAi inactivation of par-5 in either fog-1 or fog-2 mutants prevented feedback inhibition of GSC proliferation (Fig 5A). Similarly, GSC proliferation was not significantly inhibited in A1 fog-1; par-5 double mutants (Fig 5B). Thus, sperm depletion requires PAR-5 to inhibit GSC proliferation. As was the case for daf-18, par-4 and aak-1, inactivation of par-5 caused spontaneous oocyte maturation and ovulation in fog-1 mutants, preventing their accumulation (Fig 5C) and stress granule formation (S2 Fig). This suggests that all four genes act in the same signalling pathway to retroactively couple GSC proliferation to oocyte needs.
To determine if PAR-5 is required to inhibit MPK-1 activity in zone I following sperm depletion, we measured the levels of dpMPK-1 in A1 fog-1; par-5 loss-of-function mutants. As for fog-1; daf-18, fog-1; par-4 and fog-1; aak-1 mutants, we found that the germ line of A1 fog-1; par-5 double mutants had increased levels of dpMPK-1 in zones I-II relative to the background, albeit levels in zone II were variable between gonads (Figs 4O, S4 and S5). Thus, PAR-5 is required to inactivate MPK-1 in late-pachytene stage germ cells following sperm depletion. Given these results and the functional overlap between the AMPK consensus site and that of 14-3-3, these results suggest that PAR-5 may act to mediate the effects of AMPK phosphorylation to ultimately inhibit MPK-1 activity in the pachytene area of the germ line to promote GSC quiescence.
Our results uncover a novel signalling pathway that links, in two sequential steps, oocyte needs to the regulation of GSC proliferation in C. elegans adult hermaphrodites (Fig 6). We previously demonstrated that DAF-18 was required to inhibit GSC proliferation downstream of sperm depletion, and that in the absence of sperm in these animals, unfertilized oocytes spontaneously activated and were ovulated [9]. Here, we demonstrate that the lack of PAR-4, AAK-1 or PAR-5 causes a similar phenotype, disrupting oocyte quiescence and feedback regulation of GSC proliferation. Preventing oocyte activation in a feedback-defective mutant also revealed an additional coupling step, this time between oocyte maturation and oocyte production. In aak-1; oma-1; oma-2 triple mutants, loss of OMA-1 and OMA-2 prevents oocyte maturation, triggering their accumulation, while the loss of AAK-1 blocks feedback signalling to the rest of the germ line. In this case, neither GSC proliferation nor oocyte maturation is blocked and we observe hyper-accumulation of oocytes. Thus, in feedback-defective animals, the accumulation of diakinesis-arrested oocytes cannot block GSC proliferation or oocyte production. This rules out the possibility that oocyte accumulation is physically sufficient to promote GSC quiescence, in which case the accumulated oocytes in aak-1; oma-1; oma-2 triple mutants would have been expected to promote GSC quiescence.
We favor the possibility that oocyte accumulation indirectly promotes GSC quiescence through first causing an arrest in oocyte production, by blocking pachytene progression upstream in the germ line. Stalling of germ cell progression through meiosis would concomitantly inhibit GSC proliferation further upstream, in a chain of interdependent retrograde events, as we previously proposed [9]. In this scenario, AAK-1 would act as the gear that couples oocyte accumulation to the arrest of oocyte production (and other feedback steps). The newly demonstrated involvement of mpk-1 in feedback GSC regulation fits well such a mode of operation. Indeed, our results (combined with those of others [9,19,30,40]) are consistent with a model in which sperm depletion blocks oocyte maturation and ovulation, causing their accumulation. After oocytes have accumulated up to a certain point, levels of dpMPK-1 in the pachytene zone of the germ line drop, blocking meiotic progression and the production of additional oocytes. Our results indicate that this inactivation of MPK-1 in pachytene-stage germ cells following oocyte accumulation requires DAF-18, PAR-4, AAK-1 and PAR-5 activities, and that ectopic activation of mpk-1 signalling prevents oocyte accumulation from blocking GSC proliferation. Lastly, we found that inactivation of mpk-1 suppressed the effects of a daf-18 null allele on GSC proliferation, indicating that MPK-1 functions downstream of, or in parallel to, DAF-18. Thus, we propose that following oocyte accumulation, DAF-18 activity leads to MPK-1 downregulation in pachytene-stage germ cells through activation of PAR-4 and AAK-1 (PAR-4 can directly phosphorylate and activate AAK-1/AMPK [14,15]), followed by AAK-1 phosphorylation of yet unidentified targets and their recognition by PAR-5, to inhibit MPK-1 signalling (Fig 6). How this inactivation of MPK-1 exactly inhibits GSC proliferation remains to be elucidated and is a focus for future investigation. However, our analysis of GSC proliferation in tumorous animals that lack differentiated germ cells imply that mpk-1 activity may promote GSC proliferation partly in the soma and/or GSCs, as well as in differentiated germ cells in the adult, possibly in zone I. Despite their anticipated germline requirement, the precise site(s) of PAR-4/LKB1 and AAK-1/AMPK action in feedback control of GSC proliferation remains to be determined.
Our results indicate that the regulation of larval and adult GSC proliferation is mechanistically distinct. Feedback regulation of GSC proliferation by oocyte accumulation requires PAR-4/LKB1 and AAK-1/AMPK. These genes are also required for inhibition of GSC proliferation during dauer development but in this context they function downstream of insulin and/or TGF-β signals [2]. Our work reveals that PAR-4 and AAK-1 control GSC proliferation upon oocyte accumulation in the absence of changes in IIS during adulthood. The molecular context of feedback regulation of GSC proliferation by oocyte accumulation is thus mechanistically distinct from inhibition of GSC proliferation during dauer development, despite some overlap in gene requirements. Our results suggest that the inhibition of MPK-1 signalling in differentiated germ cells by AAK-1 may be an adult-specific mechanism through which GSC proliferation is controlled. While this mechanism must permit localized regulation of GSC proliferation, it appears to function only in animals that have activated IIS, perhaps because the inhibition of IIS also inactivates MPK-1 in germline zone I [30].
The involvement of AAK-1/AMPK in this feedback signal is particularly interesting because AMPK has been historically predominantly implicated in cellular stress response, typically inhibiting cell growth and proliferation in response to nutrient stress [2,15,25]. Hormones that increase intracellular Ca2+ can however activate AMPK in the absence of stress via phosphorylation by the calmodulin-dependent protein kinase CaMKKβ, but this is independent of PAR-4/LKB1 [25,41–43]. To our knowledge, the feedback mechanism that we herein report is thus the first documented occurrence of AMPK activation in a PAR-4/LKB1-dependent context in the absence of any apparent stress, but in response to a specific signal. Although we did not provide molecular evidence of AAK-1/AMPK activation, we assume it is the case due to the concomitant requirement for the upstream AMPK activating kinase PAR-4/LKB1. As for the regulation of lipid reserves in dauer larvae [44], only one of the two catalytic subunit isoforms is sufficient to fulfill this function. Thus, an emerging role of AAK-1/AMPK is its function as a signalling molecule downstream of DAF-18/PTEN and PAR-4/LKB1 to inhibit stem cell proliferation through PAR-5/14-3-3-dependent MPK-1/MAPK inhibition.
Cancer is emerging as being a disease of the stem cells. For instance, cancer incidence was proposed to be related to the frequency of stem cell divisions [45], suggesting it is the stem cells themselves that can turn ill after going through a number of divisions and become tumorigenic. Understanding how stem cell division rates are regulated in vivo is thus immediately relevant to cancer biology. In this regard, our results reveal that some tumour suppressor genes, such as PTEN and LKB1, may prevent cancer by limiting stem cell divisions when these divisions are not needed. Indeed, defective feedback inhibition of intestinal stem cells may well underlie the growth of the related benign hamartomas in Cowden’s and Peutz-Jegher’s syndromes, which are caused in humans by germline mutations in PTEN and LKB1, respectively [46–48]. As such, we predict that the mechanisms we have identified in feedback GSC regulation in C. elegans may underlie cancer development in humans.
Animals were maintained at 15°C on standard NGM plates and fed E. coli bacteria (OP50) unless otherwise indicated [49]. The Bristol isolate (N2) was used as wild-type throughout. The following alleles, deficiencies and transgenes were used. LGI: fog-1(q785). LGII: gld-3(q730), nos-3(q650), cpIs42[mex-5p::mNeonGreen::PLCδ-PH::tbb-2 3’UTR + unc-119(+)] (generously provided by Drs Dan Dickinson and Bob Goldstein [50]). LGIII: daf-2(e1370), aak-1(tm1944), mpk-1(ga117). LGIV: daf-18(nr2037, ok480), oma-1(zu405te33), par-5(it55), unc-22(e66), let-60(ga89)gf. LGV: fog-2(oz40), par-4(it57), oma-2(te51), qIs56[lag-2p::GFP; unc-119(+)]. LGX: aak-2(ok524). The following rearrangements were used. qC1[dpy-19(e1259) glp-1(q339) qIs26[rol-6(su1006)gf; lag-2p::GFP]]III, hT2[bli-4(e937) let-?(q782) qIs48](I;III), mIn1 [mIs14 dpy-10(e128)] II, nT1[unc-?(n754) let-?] (IV;V). RNAi was induced by feeding from the L1 stage as previously described [51].
GSC mitotic indexes were evaluated as previously described [9], by transferring late-L4 stage animals from 15°C to a new plate at 25°C and allowing them to grow for an additional 24 hours. A1 animals were then harvested and their gonads were dissected and stained as described below. 4'6-diamidino-2-phenylindole (DAPI) was used to highlight germ nuclei. Anti-HIM-3 antibodies were used to mark (and exclude) differentiated GSC progeny (HIM-3 positive), and anti-phospho[ser10]-histone H3, to mark M-phase nuclei. Undifferentiated germ nuclei counting in 3 dimensions was partially automated as described [11], with an ImageJ plugin developed by Dr Jane Hubbard’s laboratory. For every genotype/condition for which the MI was previously published (unmated A1: N2, fog-1, fog-1; daf-18, oma-1; oma-2, daf-2; [9]), we did not observe a significant difference between the older and newer MI datasets (P>0.05; two-tailed t-test).
A1 animals were generated as above, fixed in Carnoy’s solution and stained with DAPI as previously described [2]. The number of diakinesis-stage oocytes (having a least one condensed bivalent chromosome [52]) per gonad arm was then determined. In some backgrounds, the chromosomes in one to a few of the proximal-most oocytes sometimes appeared all condensed together and these were included in the analysis. Endomitotic oocytes were however excluded.
Whole animals were fixed in Carnoy’s solution and stained with DAPI [2]. For germline immunofluorescence, gonads were dissected out of the animal in a drop of PBS on a cover slip, which was then placed against a poly-L-lysine coated slide, submitted to a freeze-crack and stained as previously described [9,53]. Primary mouse monoclonal anti-phospho[ser10]-histone H3 (1:200, Cell Signaling #9706), mouse monoclonal anti-double-phospho[Thr202/Tyr204]MAPK (anti-MAPKYT) (1:100, Cell Signaling #9106), rabbit polyclonal anti-HIM-3 (1:500, a generous gift from Dr Monique Zetka) [53], mouse monoclonal anti-MEX-3 (1:100, a generous gift from Dr Jim Priess) [27], and secondary A488-conjugated goat anti-mouse, or A546-conjugated goat anti-rabbit antibodies (both at 1:500, Invitrogen) were used. DAPI was used as a counterstain.
Images were acquired at a 0.75 μm (Figs 1 and 2B) or 1 μm (Figs 2C, 3, 4, S1, S2, S5 and S6) intervals using a 20x (whole-worm) or a 60x (germ line) objective on either of two DeltaVision microscopes, deconvolved (whole-worm and distal germ lines), maximally-projected, stitched (whole-worm and whole germ lines), and thresholded using ImageJ. All images are maximal projections, except DIC images, which show a single focal plane. For display and ease of comparison purposes, whole-worms and germ lines were computationally straightened using ImageJ, except for S5 Fig. Fluorescence intensity profiles were obtained and processed essentially as previously described [30] using ImageJ, except that they were obtained from stitched maximal projections instead of single focal planes, and that the background from each stitched image was individually subtracted. Images were reconstructed by stitching multiple, overlapping acquisitions of the same sample using the “Grid/collection stitching with sequential images” plugin with default parameters in ImageJ [54]. Samples in Figs 4 and S4D–S4I, S4K and in S4J, S4L–S4O were acquired on a different microscope/setting that gave rise to a lower signal-to-noise ratio. To partly (and conservatively) circumvent this issue, we defined the background germline signal as the highest signal intensity observed with the better signal-to-noise microscope/setting in animals without dpMPK-1 signal in zones I-II (e.g. fog-2, A1).
|
10.1371/journal.pgen.1000601 | The Population and Evolutionary Dynamics of Homologous Gene Recombination in Bacteria | In bacteria, recombination is a rare event, not a part of the reproductive process. Nevertheless, recombination—broadly defined to include the acquisition of genes from external sources, i.e., horizontal gene transfer (HGT)—plays a central role as a source of variation for adaptive evolution in many species of bacteria. Much of niche expansion, resistance to antibiotics and other environmental stresses, virulence, and other characteristics that make bacteria interesting and problematic, is achieved through the expression of genes and genetic elements obtained from other populations of bacteria of the same and different species, as well as from eukaryotes and archaea. While recombination of homologous genes among members of the same species has played a central role in the development of the genetics and molecular biology of bacteria, the contribution of homologous gene recombination (HGR) to bacterial evolution is not at all clear. Also, not so clear are the selective pressures responsible for the evolution and maintenance of transformation, the only bacteria-encoded form of HGR. Using a semi-stochastic simulation of mutation, recombination, and selection within bacterial populations and competition between populations, we explore (1) the contribution of HGR to the rate of adaptive evolution in these populations and (2) the conditions under which HGR will provide a bacterial population a selective advantage over non-recombining or more slowly recombining populations. The results of our simulation indicate that, under broad conditions: (1) HGR occurring at rates in the range anticipated for bacteria like Streptococcus pneumoniae, Escherichia coli, Haemophilus influenzae, and Bacillus subtilis will accelerate the rate at which a population adapts to environmental conditions; (2) once established in a population, selection for this capacity to increase rates of adaptive evolution can maintain bacteria-encoded mechanisms of recombination and prevent invasion of non-recombining populations, even when recombination engenders a modest fitness cost; and (3) because of the density- and frequency-dependent nature of HGR in bacteria, this capacity to increase rates of adaptive evolution is not sufficient as a selective force to provide a recombining population a selective advantage when it is rare. Under realistic conditions, homologous gene recombination will increase the rate of adaptive evolution in bacterial populations and, once established, selection for higher rates of evolution will promote the maintenance of bacteria-encoded mechanisms for HGR. On the other hand, increasing rates of adaptive evolution by HGR is unlikely to be the sole or even a dominant selective pressure responsible for the original evolution of transformation.
| For many species of bacteria, recombination in the form of the acquisition and expression of genes and genetic elements acquired from other bacteria, eukaryotes, and archaea, HGT is an important source of variation for adaptive evolution. Not so clear is the contribution of recombination of homologous genes to adaptive evolution and as a selective pressure for the evolution and maintenance of HGT. Using computer simulations, we explore the role of HGR to adaptive evolution and selection for the evolution and maintenance of HGT. We demonstrate that under realistic conditions by shuffling genes within a bacterial population, HGR will increase its rate of adaptive evolution. Once established, this capacity to increase the rate of adaptive evolution can serve as a selective force for the maintenance of HGT. On the other hand, HGR cannot provide an advantage to a population when its density is low or when the recombining population is rare relative to non-recombining competitors. Thus, we postulate that it is unlikely that the only bacteria—rather than plasmid (or phage)—determined mechanism of HGR, transformation, evolved in response to selection for higher rates of evolution by gene shuffling.
| Recombination in the form of the receipt and incorporation of genes and genetic elements from other strains and species of bacteria [1] as well as archaea and eukaryotes [2],[3],[4],[5],[6],[7] plays a prominent role as a source of variation for the adaptive evolution of many species of bacteria [8],[9],[10],[11],[12],[13],[14],[15],[16],[17],[18]. Because of this ability to acquire genes and genetic elements from other organisms, horizontal gene transfer (HGT), the pace of adaptive evolution in bacteria need not be limited by the standing genetic variation within a population or the slow rate by which adaptive genes are generated by recurrent mutation. Through single HGT events bacteria can obtain chromosomal genes and gene clusters (islands) as well as plasmids, transposons and prophage bearing genes that have successfully traversed the gauntlet of natural selection in some other population of their own or other species. In this way, bacteria can expand their ecological niches; colonize new habitats and hosts, metabolize new energy sources, synthesize essential nutrients, survive toxic agents like antibiotics, and alas, increase their virulence to human and other hosts.
Less clear are the ecological and evolutionary consequences of more mundane HGT events, such as homologous gene recombination (HGR) among members of the same population. In accord with classical population genetic theory, meiotic recombination of can increase the rate at which populations adapt to new environments by assembling in single organisms combinations of adaptive mutations occurring in different members of their population and by reducing the rate at which populations accumulate deleterious mutations, (“Muller's Ratchet”). For superb reviews of this classical theory and some of its more recent extensions see [19],[20],[21].
There are good theoretical reasons to anticipate that recombination among chromosomal genes already present and those generated by mutation within a bacterial population can augment its rate of evolution in a variety of ecological situations [22]. There is also experimental evidence in support of this prediction. Transformation occurs at measurable rates in B. subtilis maintained in a more or less natural setting and the resulting HGR appears to promote adaptive evolution in these populations [23]. Two recent experimental studies comparing rates of evolution among recombining and non-recombining populations provide direct evidence that capacity for F-plasmid mediated recombination in E. coli [24], and transformation-mediated recombination in Helicobacter pylori [25], can increase the rate at which these bacteria adapt to culture conditions. However, there is also evidence from studies with experimental populations of E. coli [26] and Acinetobacter baylyi [27] indicating that there are conditions where there are no differences in the rates at which recombining and non-recombining populations adapt to environmental conditions.
On first consideration, it would seem that if recombination increases the rate at which populations adapt to their environment, the capacity for shuffling homologous genes within a population would provide an advantage to the recombining strain when competing with populations without this capacity. Not so clear are the conditions under which selection will favor recombining populations in this way. When will a recombining population prevail over non- or more slowly- recombining populations, and do so in the face of fitness costs associated with the capacity for recombination?
Here, we present the results of a study using computer simulations of mutation, recombination, selection and inter-population competition to explore the conditions under which: i) recombination augments rates of evolution in bacterial populations and, ii) when the capacity for HGR will be favored in competition with non-recombining populations. We demonstrate that under broad conditions, HGR occurring at rates in a range estimated for E. coli, H. influenza, S. pneumoniae,, and B. subtilis can increase the rate of adaptive evolution in bacterial populations. We show that this capacity for increasing rates of evolution by shuffling chromosomal genes can provide a recombining population a selective advantage in competition with populations without this capacity even when the recombining population has a lower intrinsic fitness. On the other hand, we also demonstrate that because the rate of recombination in bacteria depends on the density of the recombining population, the conditions under which recombination can provide a population a selective advantage in competition with non-recombining populations are restricted to when the recombining population is relatively common and the total population density is high. Even in the absence of a fitness cost, when the recombining population is rare, it will not be favored despite its ability to acquire genes from the dominant non-recombining population. We discuss the implications of these simulation results to the role of recombination in the adaptive evolution of bacteria and the evolution and maintenance of different mechanisms for homologous gene recombination in bacteria.
We open our analysis with a consideration of the contribution of recombination to the rate of increase in the average fitness of single populations of bacteria. In these simulations, the five loci contribute equally to fitness and the three alleles at each locus, 1, 2 and 3 contribute additively. Mutation and recombination are random processes with all five loci equally likely to change in any given mutation or recombination event. In the case of mutation, the change in allelic state is independent of the genetic structure of the population. For recombination, the likelihood of a particular change in the allelic state of any of the five genes in recipient is proportional to the frequencies of those alleles in the population at large. Selection is a deterministic process with fitness being proportional to the frequency of high index alleles (see Figure 2). (For more details we encourage the reader to at least peruse the METHODS section, which we believe is written a way that would be amenable to those who prefer to hum equations than solve them.)
The rate at which a population adapts to its environment, as measured by the increase in its mean fitness is directly proportional to the rate of recombination and the relative magnitude to which the five loci contribute fitness, as measured by the parameter 1−c (Figure 3). If there is more variation in the population at the start of a simulation, as there is when we start with 10 or 50 randomly selected lineages (Figure 4), the rate of evolution is faster than when the population is initially monomorphic as in Figure 3. This is of course anticipated from Fisher's fundamental theorem of natural selection [29] as well as simple logic.
To provide an overview of the relative contributions of mutation and recombination to the rate of evolution in this model, we did fifty simulations with different mutation and recombination rates. We started these runs with either a single genotype of intermediate fitness, 2,2,2,2,2 genotype, or 10 randomly chosen clones (Figures 5A and 5B, respectively). In these figures we plot the mean and standard error of the time required for the mean fitness of the population to reach 0.001 less the maximum fitness, . As anticipated from the results presented in Figures 3 and 4, the time to reach maximum fitness decreases with the rates of recombination and mutation. Saying this another way the rate of adaptive evolution increases with the rate of recombination and mutation. Although the mutation process is biased towards generating lower fitness alleles, with only five loci and the fitness and other parameters employed, the effects of generating less fit mutations on the average fitness are imperceptible. This is the case even with a mutation rate of 10−5, all of the variation in fitness determined by these five loci, c = 0, and populations initiated with the highest fitness genotype 3,3,3,3,3 (data not shown). Although lower fitness mutants are produced, they are purged by “purifying” selection and do not accumulate.
As noted in Figure 3, the extent to which recombination increases the rate of evolution is proportional to the intensity of selection at the loci subject to recombination, the selection differential. To explore this relationship a bit more and begin to consider the contribution of the form of the fitness function, we have performed simulations with c = 0.5 and exponents e = 1, e = 2 and e = 3 (see Figure 2B). As e increases, the contribution of the higher number alleles becomes proportionally greater and the time to reach maximum fitness is reduced. The results of these “experiments” are presented in Figure 6.
To provide a more detailed view of the contribution of the initial variability to the effects of recombination on the rate of evolution, we made 50 runs with each set of parameters. Each run would terminate when the mean fitness was nearly its maximum or 5000 generations passed. The results of these simulation experiments are presented in Table 1. To better evaluate the relative contributions of the initial variability and the rate of recombination, we performed a two-way analysis of variance (ANOVA) on the first three rows (2 df) and four columns (3 df). Increasing the amount genetic variability in the population at the start of each run and the rate of recombination significantly increases the rate of evolution, (p<10−16). There is also a significant interaction p<0.005 for the combined effects of initial variability and rate of recombination.
To explore the conditions under which the capacity for homologous gene recombination will provide an advantage to a population, we consider mixtures of two genetically distinct populations, one of which does not recombine (in which variation is only generated by mutation), or recombines at a lower rate than the other. For each population, mutation, recombination and selection occur as described for the single population simulations. Although the total density of the two-population community remains constant, the densities and relative frequencies the two competitors change at a rate that depends on their respective mean fitness. Unless otherwise stated, recombination only occurs within a population. In these simulations, the rate parameter of recombination of the #1 population exceeds that of the #2 population, χ1>χ2 and in most cases χ2 is 0.
In Figure 7, we follow the changes in the ratio of the two populations for different situations with the #1 and #2 populations initially monomorphic for the intermediate fitness genotype, 2,2,2,2,2. If there is no cost to recombination and initially the #1 and #2 populations are equally frequent, the recombining #1 population has an advantage over the one that is not recombining, #2 (Figure 7A), i.e. in 9/10 runs the recombining populations prevailed. When there is 1% fitness cost associated with recombination and initially equal frequencies of the #1s and #2 populations, in the majority of runs the non-recombining population has an advantage (Figure 7B).
Although in the absence of an intrinsic fitness cost, HGR provides a clear advantage when the recombining population is common, this is not necessarily the case when the recombining population is initially rare (Figure 7C). On the other hand, the capacity for HGR can prevent the establishment of an initially rare, higher fitness, non-recombining population (Figure 7D).
In the simulations described above (Figure 7) the population are initially monomorphic and recombination does not come into play until sufficient variation builds up through recurrent mutation (see Figure 3). Qualitatively, the results obtained with runs stated with 10 randomly selected clones are similar to those initiated with no variability (compare Figures 7 and 8), but there are quantitative differences. The most conspicuous of the quantitative differences between the results presented in Figure 7 and Figure 8 is that when the populations are initially variable, the outcome of competition is more likely to end in a stalemate than the loss of the #1 or #2 population. This was particularly true for the runs initiated with a rare recombining population (Figure 8C). The reason for this stalemate is that the two populations both reach the maximum fitness before the run terminates and selection ceases.
To provide a broader and more quantitative perspective of the effects of recombination on the outcome of competition, including invasion and prevention of invasion, we made 50 independent runs with different rates of recombination, different initial frequencies of the #1 and #2 populations and different fitness costs. As controls for these “experiments” we used 100 simulations with the two populations having the same recombination rates. These simulations were run until the density of one population fell below 105 (the total density remained constant at 2×108) or 2000 generations passed. The results of these experimental and control simulations are presented in Table 2. In these simulations a population “won” the competition when the density of the other population fell below 105 or when it had the highest relative frequency after 2000 generations passed.
For the simulations initiated with a monomorphic, all genotype 2,2,2,2,2 populations, in the absence of a cost and initially equal frequencies, the recombining population “won” in all 50 simulations. When the initial frequency of the recombining population was 0.05, the recombining population won about 20% of the time. When the initial frequency of the #1 was 0.005, the non-recombining #2 won in all 50 simulations. In the parallel no-fitness-cost runs with the initially variable population and initially equal densities in the majority (but not all) of the runs the #1 populations “won” or was “winning” when the runs terminated at 2000 generations. These runs illustrate how the advantages of recombination are reduced when the initial frequency of the recombining population is lower than that of the non-recombining population. The largest quantitative difference between the populations with and without initial variability is in the time required for this outcome (winning) to obtain, which, as suggested by the single population runs, is longer in the initially monomorphic populations. That recombination was responsible for the winning # 1 population in these no-fitness-cost runs with the recombining population initially rare, can be seen from the controls where there was no recombination in the # 1 population.
With a 1% or 2% fitness cost associated with recombination, even when the recombining and non-recombining populations are initially equally frequent, the non-recombining population almost invariably prevails when the competitors are initially monomorphic. A very different situation obtains when at the start of the competition there is genetic variability (10 randomly selected runs). Under these conditions even with the lowest rate of recombination examined, χ = 5×10−15, a substantial fraction of the recombining populations wins even in the face of a 2% cost in intrinsic fitness. Moreover, the time before the recombining population wins is significantly shorter than that in the runs where the non-recombining, #2, population wins. This effect of initial variability also obtains in situations where the non-recombining population is initially common. With an initially variable population, recombination provides a substantial advantage in competition with a rare but intrinsically fitter population. This is less so when the population is initially monomorphic. But even then with a sufficiently high rate of recombination the #1 population can prevail in competition with a high fitness non-recombining population.
To obtain more information about the contribution of the intensity and form of the fitness function to the conditions under which within-host selection would favor recombination, we performed simulated competition experiments with different values of the exponent e. As noted in Figure 5A, the intensity of selection due to these five loci and the contribution of the highest fitness 3 allele to that increase is directly proportional to e. The results of these simulations are presented in Table 3.
The effects of having a greater range of fitness values and a greater contribution of the highest fitness allele, 3, can be seen by comparing the simulation runs in Table 3 with the parallel runs in Table 2. Most importantly, with a greater fitness range associated with these 5 loci and proportionally greater contribution of the #3 allele, when the competing populations are of roughly equal frequency, the increase in the rate of adaptation due to recombination is more likely to overcome the fitness burden associated with recombination it would with a more modest fitness range. This is true not only for the simulations initiated with genetically variable populations but also for those initiated with monomorphic, 2,2,2,2,2 populations (compare the outcomes of the e = 1 runs in Table 2 with the corresponding e>1 simulations in Table 3).
On the other hand, even in the absence of a fitness cost associated with recombination and a greater contribution of these 5 loci and the number 3 allele to fitness, this capacity for gene shuffling does not enable the recombining population to invade when its frequency is low, 0.005. Although a greater contribution of the number 3 allele to fitness and a greater fitness differential associated with these five loci augments the likelihood of the recombining population winning over an initially rare but higher fitness, non-recombining competitor.
In the preceding, the recombining and non-recombining populations were at the start of each run genetically identical, either monomorphic for the same genotype, 2,2,2,2,2 or had the same set of genetically variable clones. To begin to explore the more realistic situation where the competing populations with and without sexual proclivity are initially different genetically, we performed simulations with 10 initially different random collections of genotypes for #1 and #2 populations. The results of these simulations are presented in Table 4.
In the absence of selection against recombination and initially equal frequencies of the #1 and #2 populations, as measured by the relative numbers of winners and losers the recombining population has an advantage over the non-recombining despite the initial genetic differences between these populations. This can be seen, by comparing the simulations for the recombining “experimental” populations (rows, 2, 3 and 4) and the non-recombining control (row 1). Notably, the rate of recombination seems to have no effect on the frequency of winning. Under these conditions, the initial fitness of the competing populations plays a more prominent role in determining the outcome of competition than the increase in fitness occurring during course of competition. By this same winning and losing criteria, in the absence of an intrinsic fitness cost, recombination increases the likelihood of the #1 population ascending to dominance when it is initially relatively rare (0.05). On the other hand, with initial genetic differences in the recombining and non-recombining populations and a cost associated with capacity for recombination, HGR does not provide a statistically significant advantage for the #1 population. Moreover, with initial differences in the genetic composition of the recombining and non-recombining populations, the time to winning by the non-recombining population is less than that of the recombining.
In all of the preceding runs, we assumed that recombination only occurs within a population. It may well be that both populations can contribute as donors even when they both cannot serve as recipients, e.g. when recombination is through the uptake of exogenous DNA, transformation. To explore this situation, we used a version of the simulation where the donors for recombination can be chosen from both populations, with the choice dependent solely on their relative frequencies.
In Table 5 we compare the outcomes of simulations where only members of the recombining population #1 serve as donors as well as recipients with corresponding situation where members of both populations can serve as donors. In the simulation results presented in this table, the recombining population has a 2% intrinsic fitness cost. In the runs where the initial frequency of the recombining populations was 0.05, the acquisition of genes from the non-recombining population increased the likelihood of invasion. Although with a 1∶1 ratio there was a significantly higher frequency of recombining populations winning when both populations served as donors in the initially monomorphic runs, this was not the case for the simulations initiated with 10 randomly selected clones. On the other hand, when the initial frequency of the recombining population was 5% for both the initially monomorphic and polymorphic populations, the recombining population was more likely to win when both populations served as the source of genes for recombination. When both populations served as donors and both were initially polymorphic, winning by the recombining population took less time then it did for the winning non-recombining populations. It should be noted, however, that for any value of χ when both populations serve as donors because of the greater density of the population,the frequency of recombination was greater than when only one competitor served as the donor.
The recombining and non-recombining populations considered in the preceding two population simulations are the extremes. It may well be that both competing populations are capable of recombination but do so at different rates. Based on the single population and the preceding mixed population results we would anticipate that if this were the case and all else were equal, the population with the higher rate of recombination would prevail. This is indeed confirmed by our simulation experiments. For example, for 100 simulations with initially monomorphic, intermediate fitness 2,2,2,2,2 populations in equal frequency, (μ = 10−7,c = 0.5, e = 3) and recombination rate parameters χ1 = 5×10−13 and χ2 = 5×10−15, in 95 of the runs the #1 population won, or was winning at 2000 generation in the remaining 5 runs. On the other hand, with these starting conditions, when the initial densities of the #1 and #2 populations were respectively 107 vs. 1.9×108, the population with the lower recombination rate won in 99 out of 100 runs and the population with the higher rate of recombination won in only one run. The situation is different when both populations can serve as donors as well as recipients. In this 107 vs. 1.9×108 contest between the populations with high (#1) and low (#2) rates of recombination, the score for 100 runs were #1 won 45 times, #2 won 41 times, and the numbers of # 1s and #2s winning at the 2000 generation termination were, respectively 11 and 3.
We interpret the results of this computer simulation study as support for the proposition that that there are realistic conditions where homologous gene recombination (HGR) will increase the rate at which bacterial populations adapt to their environment. These results are also consistent with the hypotheses that by increasing rates of adaptive evolution, HGR can provide a population a selective advantage when competing with otherwise identical or even somewhat more fit populations that are unable to shuffle homologous genes or do so at lower rates. Our mixed population simulations, however, also illustrate a major caveat to the hypothesis that homologous gene recombination in bacteria evolved in response to selection for increasing rates of adaptive evolution. Even in the absence of a fitness cost, the recombining population will only have an advantage over a non-recombining population when the recombining population is relatively common; HGR will not be favored when it is rare.
The validity and generality of these predictions are, of course, empirical questions. They are however, questions that can be addressed experimentally. And, as noted in our Introduction, there have been at least four experimental studies testing the hypothesis that recombination increases the rate at which bacterial populations adapt to culture conditions. The results of two of these experiments are consistent with this hypothesis, Cooper's study with F-plasmid-mediated recombination in E. coli B [24] and Baltrus and colleagues study of transformation-mediated recombination in Helicobacter pylori [25]. The results of the other two reports, Souza and colleague's study of Hfr-mediated recombination in E. coli [26] and Bacher and colleagues study of transformation-mediated recombination in Acinetobacter baylyi [27] are interpreted to be inconsistent.
How well do the results of this simulation study account for the outcomes of these recombination – rates of adaptive evolution experiments? We believe that at least at a qualitative level, the results of the three of these studies for which this model is a reasonable analog [24],[25],[27] are consistent with the predictions of these simulations. The format of the experiments by Souza and colleagues [26] were different from that of this model and therefore we do not believe these simulations are appropriate for interpreting their results. In their experiments, two genetically different E. coli strains were used; a Hfr strain of E. coli K-12 and a F- strain of E. coli B. Although the Hfr strain donated genes to the E. coli B, under the conditions of their experiments this donor did not replicate and it was not present throughout the course of the experiment as assumed in our model.
Although the details of the [24],[25],[27] experiments were different from those specified by this simple model, their basic structure was similar to that of the single population simulations initiated with monoclonal (2,2,2,2,2) populations. In these experiments, which were initiated with single clones of either recombining (Rec+ or Com+) or non-recombining (Rec− or Com−) populations, the bacteria were growing in liquid media and reached densities of 5×107 per ml or greater. Although the rate constants of recombination χ were not estimated in these experimental studies, it was clear that recombination was occurring at a substantial rate. The frequency of gene replacement by recombination in the Rec+ E. coli B and Com+ H. pylori experiments exceed that expected by mutation, and in the Cooper study the rate of gene replacement by recombination greater is greater than that of the elevated rate of mutation of a mutS strain. For recombination mediated by HFR, F', F+ plasmid in E. coli, χ, it seems reasonable to conclude that in the Cooper experiments c>10−13 (Cornejo and Levin, In Preparation- but available, see www.eclf.net ). We would also expect χ>10−13 for the H. pylori experiments and possibly in the Acinetobacter baylyi study as well. This is certainly the case for the only two experimentally obtained estimates χ we know of for transforming bacteria, H. influenzae [30] and B. subtilis [31], both of which are on the order of χ∼10−12.
With population densities, mutation and recombination rate constants in the ranges of these experiments, our simulations show that recombining populations evolved more rapidly than those that did not have this capacity for shuffling homologous genes. For any given mutation and recombination rate parameters, the rate and magnitude of increase in mean fitness depended on the fitness function. Cooper's observation that recombination increased the rate of adaptation to culture conditions with a higher mutation rate ∼3 times greater than it did with a lower rate [24] is also consistent with the predictions of this model; mutation and recombination act synergistically to increase rates of adaptive evolution. Although Bacher and colleagues [27] interpret the results of their experiments with A. baylyi to be inconsistent with the hypothesis that HGR increases rates of adaptive evolution, that is not the case for all the results they report. In their higher density experiments not only does the fitness of the population increase to a greater extent than in their low density experiments, but this increase in fitness was considerably as well as significantly greater (p = 0.00012 for a two tailed t-test) for the transformation competent population than the non-competent controls.
While we are unaware of direct experimental evidence for an affirmative answer to this question from natural population studies, based on the predictions of the model we would anticipate a positive answer. Retrospective, multi-locus sequence studies suggest that the rates of gene of replacements by homologous recombination in species like Streptococcus pneumoniae exceed that by mutation by a factor of 10 or so [32],[33],[34], and are even greater for some species, like H. pylori [35],[36].
To put these retrospective estimates of recombination rates into the context of our model and its parameters, consider the following intuitive argument. Assume a 1-hour generation time, a habitat of 1 ml, a population of 108 bacteria and a mutation rate of 10−8 per cell per generation. In the course of an hour in that population, for any given locus, an average of 1 mutant would be produced. If gene replacements by recombination occur at 10 times that rate, there would be 10 recombinants at that locus for a value of χ = 10/(108×108) = 10−15. As noted in our simulations, even at this low rate and an initially monoclonal population, recombination can increase the rate of adaptive evolution over that which would be anticipated by mutation alone. Moreover, natural populations of many bacteria are likely to be composed of multiple lineages and would be genetically variable at many loci. In accord with our simulations the pace at which recombination increases the rate of adaptive evolution would on average increase with the extent of genetic variability of the population, see Figure 5.
Processes, like homologous gene recombination, that increase rates of adaptive evolution would be to the advantage of a population and augment its prognosis for surviving the vicissitudes of an ever-changing environment. This is, of course, the most common explanation for ubiquity of HGR among extant species of eukaryotes. Indeed, the presumed lack of recombination, sex to be more provocative, in ancient groups of seemingly successful organisms like the bdelloid rotifers make them intriguing objects for study [37]. Whether accelerating rates of adaptive evolution is the selective force responsible for the evolution and maintenance of recombination in eukaryotes is a subject of some controversy [20],[38], a subject that we are pleased to say is beyond the scope of this report. The population and evolutionary dynamics of recombination in bacteria are fundamentally different from that of sexually reproducing eukaryotes. In the bacteria, recombination depends on density and is not a part of the reproductive process. If they wish to procreate, sexually reproducing eukaryotes have no choice but to find mates and generate recombinant progeny, independently of density of their populations.
Here, we postulate that once the mechanisms for HGR are established in a bacterial population, the advantage accrued by a more rapid rate of adaptation to environmental conditions can promote their maintenance, even if they engender a modest cost in fitness. The necessary condition for this to obtain is that the adaptive process is continuous. This may be the case when a population enters a new environment and/or is confronted by either physical or biological factors that reduce the rates of survival or reproduction (the fitness) of its members. As long as the population is continually confronted with situations where selection favors new genotypes, as was postulated for evolution of mutator genes [39], recombination could continue to be favored and be maintained. This would not be the case if recombination engenders a fitness cost and the population is confronted with extensive periods of adaptive stasis. Under these conditions, the frequency of the recombining population will continue to decline. And, because of the frequency- and density- dependent nature of selection for recombination, the recombining population may not be able to recover. In this interpretation, the maintaining mechanisms of horizontal gene transfer by HGR by increasing rates of adaptive evolution is not an equilibrium outcome; on “equilibrium day” [40], recombination will be lost. Moreover, because HGR accelerating rates would not provide an advantage to a recombining population when it is initially rare, it is even less likely to have been a selective force for the original evolution of mechanisms for HGT than it is for maintaining those mechanisms once they evolved.
Models can be used to generate hypotheses and, in a quantitative way, evaluate their plausibility. They cannot be used to test those hypotheses! We are unaware of published empirical studies testing the hypotheses that selection for HGR is frequency- and density- dependent. These are, however, hypotheses that can be tested with experiments similar to the single clone studies testing the hypothesis that HGR increases rates of adaptive evolution [24],[25],[27]. The idea would be to follow the changes in frequency of Com+ or Rec+ in competition with Com− or Rec− clones with different initial frequencies of these competitors and in populations of different densities. We postulate that under conditions where they accelerate rates of adaptive evolution in single clone culture and adjusting for intrinsic fitness differences: (1) when introduced at roughly equal frequencies, the recombining population will have an advantage over a non-recombining competitor and, (2) the recombining population will not have that advantage when it is initially rare (in our simulations much less than 1%.). We also postulate that because of a lower rate of production of mutants as well as the lower frequency of recombination (which would be proportional to the square of the density of the recombining population); (3) the rate of adaptive evolution would be less in recombining populations of low density than otherwise identical populations of higher density and, (4) the minimum frequency for a recombining population to have a selective advantage in competition with one that cannot recombine would be inversely proportional to the total density of the recombining population.
Using the long-term evolved strains of E. coli B developed by Richard Lenski and colleagues [41],[42],[43] it should be possible to experimentally test the hypothesis that HGR will only be favored when there is relatively intense selection for adaptation to culture conditions. Although those experimental E. coli B populations continued to evolve in different ways as time proceeded the largest increase in mean fitness relative to the ancestral occurred within the first 5,000 or so generations. We postulate that if in an experiment similar to that in [24] the F'lac constructs were made with E. coli B taken from later generations, say >20,000, the recombining population will not evolve more rapidly than one that is not recombining.
Two of the three major mechanisms responsible for HGT and HGR in bacteria, conjugation and transduction, are not properties of the bacteria but rather that of their parasites, primarily conjugative plasmids and bacteriophage. One needn't postulate that these processes evolved and are maintained by selection favoring bacteria with the capacity for HGT. The most parsimonious hypothesis for recombination mediated by plasmids and phage is as a coincidental byproduct of the infectious transfer of these elements and the host's recombination repair system [44],[45]. This would also be the case for recombination resulting from cell fusion [11] or transformation mediated by natural electroporulation or cold shocks. In this interpretation, accelerating the rate of adaptive evolution by HGR mediated by these processes are a lagniappe rather than a product of adaptive evolution. To be sure we can make up and probably construct mathematical models illustrating ways by which bacteria evolve mechanisms to be more receptive to plasmids and phage carrying genes on their behalf, but we see no need to stretch our imaginations in that direction.
The third main mechanism for HGT and HGR in bacteria, the uptake and incorporation of exogenous DNA, i.e. competence and transformation, are intrinsic properties of bacteria rather that of their parasites. We postulate that under some conditions HGR accelerating rates of adaptive evolution will promote the maintenance of competence and transformation. HGR accelerating rates of adaptive evolution is, however, only one of at least three non-exclusive mechanisms that operate synergistically to maintain competence for the uptake of exogenous DNA. The other three are; (1) the acquisition of templates for the repair of double stranded breaks in DNA [46],[47]; the uptake of nutrients and nucleotides [48],[49],[50],[51], and (3) episodic selection favoring transiently non-growing subpopulations of competent cells and rare transformants [31]. In accord with these three hypotheses, transformation (recombination) is a coincidental byproduct of competence.
As is the case with meiotic recombination in eukaryotes, accounting for the selective pressures responsible for original evolution of competence and transformation is more problematic than explaining their maintenance once they have evolved. Competence is a complex character that requires the coordinated activity of a large number of genes [15],[52],[53],[54],[55]. What are the selective pressures responsible for the evolution of these genes and coordinating their activity? Because recombination will only be favored when it is common, we postulate that HGR accelerating rates of adaptive evolution cannot account for the original evolution of natural competence and transformation. For the same reasons, we postulate that this is also the case for the episodic selection for competence. The DNA repair and food hypotheses have the virtues of selection operating at the level of an individual bacterium rather than populations and thereby allowing competence to be favored when it is rare, rather than only when it is common. On the other hand, these two hypotheses raise other issues about whether they can account for the original evolution of competence. For a recent critical consideration of these “other issues” we refer the reader to the Discussion in [31]. At this juncture, we accept the selection pressures responsible for the origins of competence and transformation in bacteria as a delicious, but yet-to-be solved evolutionary problem.
In our simulations we have restricted the theater of evolution to single populations. A long-standing argument for the evolution of recombination is that higher rates of adaptive evolution provide an advantage to the collective, the group, rather than individuals [19],[29]. Populations that evolve more rapidly are more likely to prevail and survive longer than those with lower rates of adaptive evolution. In theory there are conditions where group- or interpopulation- level selection can lead to the evolution of characters that are at a disadvantage within populations [56],[57],[58],[59]. And, mechanisms of this type have been postulated to play a role in the evolution of recombination in bacteria [60]. While we prefer individual-level selection operating within populations on the grounds of parsimony, we can't rule out the possibility that competence and transformation evolved and is maintained by some form of group- level selection.
|
10.1371/journal.ppat.1005277 | Dengue Virus Non-structural Protein 1 Modulates Infectious Particle Production via Interaction with the Structural Proteins | Non-structural protein 1 (NS1) is one of the most enigmatic proteins of the Dengue virus (DENV), playing distinct functions in immune evasion, pathogenesis and viral replication. The recently reported crystal structure of DENV NS1 revealed its peculiar three-dimensional fold; however, detailed information on NS1 function at different steps of the viral replication cycle is still missing. By using the recently reported crystal structure, as well as amino acid sequence conservation, as a guide for a comprehensive site-directed mutagenesis study, we discovered that in addition to being essential for RNA replication, DENV NS1 is also critically required for the production of infectious virus particles. Taking advantage of a trans-complementation approach based on fully functional epitope-tagged NS1 variants, we identified previously unreported interactions between NS1 and the structural proteins Envelope (E) and precursor Membrane (prM). Interestingly, coimmunoprecipitation revealed an additional association with capsid, arguing that NS1 interacts via the structural glycoproteins with DENV particles. Results obtained with mutations residing either in the NS1 Wing domain or in the β-ladder domain suggest that NS1 might have two distinct functions in the assembly of DENV particles. By using a trans-complementation approach with a C-terminally KDEL-tagged ER-resident NS1, we demonstrate that the secretion of NS1 is dispensable for both RNA replication and infectious particle production. In conclusion, our results provide an extensive genetic map of NS1 determinants essential for viral RNA replication and identify a novel role of NS1 in virion production that is mediated via interaction with the structural proteins. These studies extend the list of NS1 functions and argue for a central role in coordinating replication and assembly/release of infectious DENV particles.
| Dengue virus (DENV) is a major arthropod-borne human pathogen, infecting more than 400 million individuals annually worldwide; however, neither a therapeutic drug nor a prophylactic vaccine is currently available. Amongst the DENV proteins, non-structural protein 1 (NS1) is one of the most enigmatic, being required for RNA replication, but also secreted from infected cells to counteract antiviral immune response, thus contributing to pathogenesis. Despite its essential role at early stages of the viral replication cycle, the molecular determinants governing NS1 functions are unknown. Here, we used a combination of genetic, high-resolution imaging and biochemical approaches and found that NS1 additionally plays an important role for the production of infectious virus particles. By using a novel trans-complementation system with fully functional epitope-tagged NS1, we show that NS1 interacts with the structural proteins residing in the envelope of the virus particle. An NS1 variant retained in the endoplasmic reticulum still supported efficient DENV particle production, demonstrating that secretion of NS1 is dispensable for virion production. This study expands the list of functions exerted by NS1 for the DENV replication cycle. Given this multi-functional nature, NS1 appears to be an attractive target for antiviral therapy.
| Dengue is the most prevalent arthropod-borne viral disease affecting around 400 million people worldwide and causing around 25,000 deaths per year [1]. Dengue virus (DENV) infections can lead to a wide range of clinical manifestations, ranging from asymptomatic to life-threatening dengue hemorrhagic fever and shock syndrome. However, in spite of its high medical relevance, no prophylactic vaccines or antiviral therapies are currently available and therefore a better understanding of the flavivirus life cycle is essential to promote the development of effective therapeutic regimens.
DENV has a single stranded RNA genome of positive polarity, encoding for a polyprotein that is co- and post-translationally processed into three structural proteins (capsid, prM, and envelope) and seven nonstructural proteins (NS1-NS2A-NS2B-NS3-NS4A-NS4B-NS5) [2]. After viral entry and release of the genomic RNA into the cytoplasm of infected cells, newly synthesized viral proteins induce massive remodeling of intracellular membranes, creating distinct intracellular structures where viral RNA replication and virion assembly take place [3,4]. Nucleocapsid formation, thought to occur in close proximity to replication sites, is likely accompanied by acquisition of a lipid envelope via budding into endoplasmic reticulum (ER) membranes enriched in the envelope protein E and prM [5,6], through as yet undefined mechanisms. Assembled virions, stored within ER stacks in highly ordered arrays, are then released from the cell via the conventional secretory pathway, where cleavage of the prM protein by furin, a protease residing in the trans-Golgi network (TGN), renders the viral particles infectious.
Flavivirus NS1 is a multifunctional 48-kDa glycoprotein that is translocated into the ER lumen co-translationally. Within the ER, NS1 promptly dimerizes upon addition of high-mannose carbohydrates [7], and is targeted to three destinations: the viral replication sites, the plasma membrane and the extracellular compartment. The majority of secreted NS1 is a soluble, proteolipid particle forming an open-barrel hexameric shell with a central channel occupied by lipids [8]. The three-dimensional high-resolution structure of the DENV NS1 dimer was recently solved by X-ray crystallography [9], providing valuable insights into the complex NS1 fold (Fig 1). The dimer contains three domains: first, a small β-roll domain formed by two intertwined β-hairpins; second, a Wing domain, composed of an α/β subdomain and a discontinuous connector that sits against the β-roll; third, a β-ladder domain, formed by 18 antiparallel β-strands (9 contributed by each monomer) assembled in a continuous β-sheet that runs along the whole length of the dimer (Fig 1A, left panel). The protrusion created by the β-roll and the connector subdomain renders one side of the dimer hydrophobic, and has been proposed to face the ER membrane and to interact with other transmembrane viral proteins [9,10]. Conversely, within the NS1 hexamer, the β-roll faces the interior of the lipoparticle, where it associates with the central lipid core (Fig 1A, right panel). On the opposite side of the β-roll, both in the dimeric and the hexameric form, the distal tips of the β-ladder and the Wing domain loops point outward, and are therefore exposed to the solvent.
Secreted NS1 as well as NS1 residing on the plasma membrane and within cells, plays important roles in immune evasion via binding to complement proteins and modifying or antagonizing their functions [11–14]. Besides its immune evasive functions, NS1 modulates early events in viral RNA replication, was shown to co-localize with double strand RNA (dsRNA) and to interact with NS4B [10,15–17]. Indeed, deletion of NS1 from the viral genome completely abrogates replication, but ectopic expression of NS1 in trans can efficiently rescue NS1-deleted (ΔNS1) viruses [18–21].
Because of its essential role early in RNA replication, genetic studies have thus far provided limited information on the molecular determinants of NS1 responsible for the viral replication cycle and did not investigate possible functions of the protein for assembly and release of infectious virus particles. By using a combination of genetic, high-resolution imaging and biochemical approaches we discovered a novel role of NS1 for the production of infectious DENV particles that is linked to NS1 interaction with the structural proteins, but independent from NS1 secretion.
Sequence analysis and visual inspection of the recently solved three-dimensional crystal structure [9] of NS1 were performed to assess the degree of conservation of amino acid residues and to identify the most relevant positions to be targeted by site-directed mutagenesis (Fig 1B). Based on their distribution within the NS1 dimer and their relative conservation across the Flavivirus genus, we selected 46 residues for alanine scanning mutagenesis, including five invariant cysteine residues (C4, C55, C179, C291, C312), recently shown to be engaged in disulfide bonds and playing an essential role in stabilizing the protein fold [9,22]. To dissect the impact of each individual mutation on the different steps of the viral replication cycle, we assessed viral RNA replication and virus spread by taking advantage of a DVR2A luciferase reporter virus genome (Fig 2A). VeroE6 cells were electroporated with in vitro transcripts of wild-type (WT) or a given NS1 mutant and viral replication was assessed by luciferase activity 24, 48, 72, 96 and 120 h later (Fig 2B). Additionally, a replication-deficient NS5 mutant (GND) with a lethal mutation affecting the RNA-dependent RNA polymerase activity was included as negative control. Based on the replication phenotypes, half of the NS1 mutants displayed only minor defects or replicated comparably to WT (Table 1). Conversely, 23 mutations, including those affecting cysteine residues engaged in disulfide bonds, severely or completely blocked viral RNA replication (Fig 2B and Table 1, underlined mutants). Most of these mutations clustered on the core of the Wing and β-ladder domain, affecting residues that point towards the β-roll (S1 Fig), which has been proposed to face the ER membrane [9]. Interestingly, in close proximity to the previously reported di-amino acid motif (N10-K11) suggested to mediate interaction with NS4B and association with ER membranes [10], a mutation targeting W8 within the NS1 β-roll domain completely abrogated viral RNA replication. Similarly, mutations within the greasy finger loop of the β-ladder, namely Y158A and G161A, resulted in a lethal phenotype as already shown for alanine substitutions at residues G159 and F160 [9]. Altogether, these results provide a comprehensive map of molecular determinants within NS1 essential for viral RNA replication, highlighting an important role for selected residues of the β-roll and β-ladder domains, clustering on hydrophobic protrusions within the NS1 dimer structure (S1 Fig).
To determine the impact of each mutation on the production of infectious virus particles, culture supernatants of transfected cells (Fig 2B) were harvested 72 h after transfection and used to infect naïve VeroE6 cells. Virus production was determined by luciferase assay 48 h later (Fig 3A). This experiment revealed a group of mutations (S114A, W115A, D180A, T301A) with minor effects on RNA replication, but massive impairment of virus production (up to ~2.5 Log10 reduction compared to WT) (Fig 3B). Noteworthy, a mutation targeting T117 within the unresolved stretch of the Wing domain and in close proximity to S114 and W115, slightly enhanced particle production. Altogether, these results suggest a previously undiscovered role of NS1 for the production of infectious DENV particles.
Next we wanted to corroborate this observation and rule out that impaired virus production was an indirect consequence of diminished replication fitness rather than a specific defect in assembly or release of viral progeny. To this end, we assessed the impact of these mutations on RNA replication in the context of a sub-genomic reporter replicon (sgDVR2A) that does not support virus production, thus measuring replication independent from a possible contribution of virus spread (Fig 4A). In this and all subsequent analyses we focused on mutants with selective alterations of virus production (S114A, W115A, D180A, T301A; Fig 3B) in order to avoid possible indirect effects resulting from impaired replication fitness (Fig 3B). Therefore, NS1 mutants with strong replication defects were excluded (Table 1). Moreover, since several NS1 mutants with a defect in virus production were slightly impaired in RNA replication we included as control the NS1 mutation R314A that caused minimal defect in replication, but did not affect virus production (Table 1). Furthermore, mutant T117A was included in the analysis because of its increased capacity to produce infectious DENV particles and its close proximity to some of the sites where mutations caused a selective reduction of virus production. VeroE6 cells were transfected with in vitro transcripts of the replicon constructs and replication was measured 24, 48 and 72 h later (Fig 4A). While at early times post transfection we observed a moderate reduction in luciferase activity for S114A, W115A, D180A and T301A compared to WT (2 to 5-fold; 24 h.p.t.), all mutants replicated comparably at 48 and 72 h.p.t., arguing for a minor contribution of RNA replication to the observed reduction in infectious particle production. As already observed in the context of the full-length reporter virus, the T117A mutant exhibited a replication profile comparable to WT also within the subgenomic sgDVR2A replicon whereas R314A exhibited an overall decrease in replication fitness, with a 50% to 75% reduction at any time point. Collectively, these results demonstrate a selective defect for a sub-group of NS1 mutants in assembly and/or release of virus particles.
To investigate further the phenotype of the NS1 mutants with respect to the production of infectious intra- and extracellular virus particles, and to corroborate these observations in human cells, we determined the infectivity profiles of each mutant in the context of a full-length DENV genome transfected into human hepatoma Huh7 cells. Three days post-transfection, titers of infectious virus released into the culture media or contained in cells were determined by limiting dilution assay with naïve cells. The results shown in Fig 4B demonstrate that the amounts of intra- and extracellular infectivity were altered in all mutants, albeit to very different degrees. In agreement with the results obtained with the reporter virus genome, we found that alanine substitutions at residues S114, W115, D180 and T301 reduced extracellular infectivity titers up to 100-fold, confirming an essential role of these amino acid residues in NS1 for particle production. Interestingly, the amounts of intracellular virus particles were also reduced 5- to 10-fold. While this impairment argued for a defect of the NS1 mutants in virus assembly or maturation, the higher reduction of extracellular virus titers suggested an additional effect on particle release as inferred from the ratio of intra- to extracellular virus titers and comparison with the WT (Fig 4C). The R314A mutation reduced the virus titer only ~3.5-fold and did not affect the ratio of intra- to extracellular infectivity, consistent with a subtle effect on assembly or virus maturation (Fig 4C). Interestingly, the T117A mutant produced 12-fold more intracellular virus than WT, concomitant with a ~5-fold higher titer of extracellular virus particles, indicating accelerated assembly or infectivity maturation and reduced virus particle release. In conclusion, these results suggest that alanine substitutions at position S114, W115, T117, D180 or T301 of NS1 alter the production of infectious virus, supporting the notion that NS1 is a critical determinant for assembly or release of infectious virus particles.
NS1 accumulates in extracellular fluids as a homo-hexamer with a lipidic core [8,23,24] and besides its immune evasive functions [11–13] was shown to enhance virus attachment upon entry [25]. In addition, some studies hypothesized a link between NS1 secretion and virus assembly or release [26]; however the lack of genetic tools in those days allowing the selective block of NS1 secretion in the context of a complete replication cycle precluded any functional investigation.
Prior studies have shown that YFV, KUNV and WNV mutants lacking NS1 do not replicate, but can be rescued when NS1 is complemented in trans by ectopic expression of the full-length protein [17,18,21,27]. To elucidate the possible role(s) of NS1 secretion in the DENV replication cycle, we utilized a similar approach and generated a DENV genome containing a 97 amino acids in-frame deletion within the NS1 gene (DVR2AΔNS1). This mutant retained the N-terminal 156 and the C-terminal 99 residues, respectively (Fig 5A, left panel). In parallel, we engineered a set of helper VeroE6 cell lines, constitutively expressing different NS1 variants after lentiviral transduction of expression vectors containing the complete NS1 coding region (NS1WT) or the empty pWPI vector (CTRL) that served as positive and negative controls, respectively. Furthermore, we engineered C-terminally tagged variants carrying a HA-affinity epitope (NS1HA) or the well-described KDEL motif (NS1KDEL), responsible for retrieval of ER luminal proteins from the Golgi apparatus by retrograde transport (Fig 5A, right panel). Correct protein expression and secretion of each NS1 variant was confirmed by western-blotting (Fig 5B). As expected, NS1HA had a slower electrophoretic mobility than the WT. Furthermore, both NS1WT and NS1HA were readily detected in the culture media, while NS1KDEL was effectively retained in the ER.
Next, we assessed rescue of viral RNA replication and particle production by the NS1 variants by using transfection of the DVR2AΔNS1 genome into each helper cell line. As shown in Fig 5C, the ΔNS1 genome was able to replicate in NS1WT, NS1HA and NS1KDEL helper cells, while no luciferase activity could be detected in CTRL cells lacking NS1. Of note, when each naïve helper cell line was infected with virus-containing culture fluids harvested 72 h.p.t, comparable levels of luciferase activity were detected in all conditions, indicating that C-terminally HA-tagged NS1 is fully functional (Fig 5D). Importantly, the rescue of particle production by NS1KDEL shows that secretion of NS1 is dispensable for infectious DENV particle production.
To address the relative efficiency of trans-complementation upon virus infection, NS1WT-, NS1HA- and NS1KDEL-expressing VeroE6 cells were infected with trans-complemented DVR2AΔNS1 particles (ΔNS1TCP), produced in VeroE6_NS1WT cells. Culture fluids were harvested 24, 48 and 72 h later and the amounts of produced particles were determined by focus-forming unit (FFU) assay on NS1WT cells (Fig 6A). Consistent with the luciferase assay data, ΔNS1TCP did not produce infectious virus on CTRL cells that do not express NS1, confirming that this mutant fails to replicate in the absence of NS1. Titers of infectious ΔNS1TCP particles released from VeroE6_NS1WT, NS1HA or NS1KDEL cells were higher than DVR2A wild-type infection on CTRL cells at early times post-infection (24 and 48 h p.i.), with no appreciable differences observed at later time points (72 h p.i.) (Fig 6B). Interestingly, while none of the trans-complemented TCPs gave rise to clearly visible plaques as previously reported [17], DENV-containing foci detected by immunostaining were larger than those produced by the full-length wild-type virus (Fig 6B, lower panel).
Additionally, lysates and culture supernatants of infected cells were harvested 24, 48 and 72 h p.i. to evaluate protein expression and secretion upon ΔNS1TCP infection. Under these conditions, virus replication in the various helper cell lines was comparable as judged by the intracellular protein levels of NS5 (Fig 6C; “Intra“). Furthermore, secretion profiles of HA- WT- and KDEL-tagged NS1 resembled those of uninfected cells, with the latter being efficiently retained intracellularly also upon DENV infection (Fig 6C; “Extra“).
To unequivocally confirm that the ΔNS1TCP system fully recapitulates wild-type DENV infections, we additionally investigated the ultrastructural morphology of NS1HA helper cells upon infection with DVR2AΔNS1 TCPs by using transmission electron microscopy. In agreement with the replication and infectious particle production data, NS1HA cells infected with DVR2AΔNS1 TCPs contained the characteristic membrane invaginations which have been proposed to represent viral replication factories (vRFs) [5,6] and electron-dense virus particles forming regular arrays within the ER (Fig 6D). Moreover, a large number of virus particles were observed on the plasma membrane or accumulating within the extracellular space between adjacent cells (Fig 6E). These structures were absent in both uninfected NS1HA cells and ΔNS1TCP-infected CTRL cells (S2 Fig).
In conclusion, these results demonstrate full functionality of HA-tagged and ER-retained NS1 supporting both DENV RNA replication and production of infectious virus particles.
Since NS1 secretion appeared functionally unlinked to infectious particle production, we next hypothesized that NS1 function(s) required for the late steps of the viral replication cycle might involve interactions between NS1 and the structural DENV proteins. To address this hypothesis we took advantage of our ΔNS1TCP system using HA-tagged NS1 for trans-complementation. DVR2AΔNS1 TCP stocks produced and titered in NS1WT helper cells were used to infect NS1HA target cells at an MOI of 1. Forty-eight hours later, cell lysates were subjected to HA-affinity capture using anti-HA agarose beads, and purified NS1 protein complexes or whole cell lysates were analyzed by western-blot using C-, prM-, E- and NS5-specific antibodies. Specificity of western-blot and immunoprecipitation analysis was monitored by including VeroE6_CTRL and VeroE6_NS1WT cells, respectively. As shown in Fig 7A, upon infection of NS1 helper cells with ΔNS1TCP, comparable amounts of structural proteins accumulated in both NS1WT and NS1HA cells, confirming that DVR2AΔNS1 replicates efficiently in both cell lines. Most interestingly, upon HA-immunoprecipitation, all three structural proteins (E, prM and C) specifically co-precipitated with NS1HA arguing for an interaction between NS1 and DENV virions and possibly also subviral particles. In spite of comparable protein amounts in the cell lysates, no specific signal was detected in case of the non-tagged NS1WT or for the NS5 protein, confirming specificity of the NS1HA-immunoprecipitation.
To corroborate the interaction between NS1 and the structural proteins, we performed analogous pull-down experiments using cells that had been infected with wild-type DENV. VeroE6 cells were mock-infected or infected with DENV-2 at an MOI of 1 and 48 hours later cell lysates were subjected to immunoprecipitation using a rabbit pre-immune serum (PIS) or an NS1-specific polyclonal antiserum (Fig 7B). Interestingly, also under these conditions E and prM were specifically co-immunoprecipitated with NS1 whereas C was not detected. The absence of C might be due to the overall lower efficiency of this immunocapture approach, to suboptimal conditions for antibody binding or to an altered ratio of subviral particles to infectious virions in the TCP system as compared to wild-type virus-infected cells (see discussion). Nevertheless, the interaction between E and NS1 was confirmed in a reciprocal approach by which the NS1 protein in DENV-2 infected cells could be specifically co-immunoprecipitated with envelope (S3 Fig). Altogether, these results support the notion that NS1 interacts with the viral envelope glycoproteins.
Based on the results described above, we hypothesized that the selective defect in infectious particle production observed for some of the NS1 point mutants was due to an altered association with the envelope glycoproteins. To investigate this hypothesis, we analyzed the association of selected NS1 mutants with the structural proteins by co-immunoprecipitation in the DVR2AΔNS1 TCP system, because it allowed highly efficient pull-down of NS1. We engineered stable VeroE6 cell lines expressing HA-tagged forms of the NS1 mutants S114A, W115A, T117A, D180A, T301A and R314A and infected these cell lines with ΔNS1TCP at an MOI of 1. Samples harvested 48 h.p.i., together with positive and negative controls, were subjected to HA-specific pull-down and cell lysates, immunocomplexes or culture supernatants were analyzed as described above. All cell lines expressed HA-tagged NS1 variants and despite small variations in the expression levels rescued DVR2AΔNS1 replication to similar extents as judged by the abundance of E, prM and C (Fig 8A) and the luciferase activity in the cell lysates (S5 Fig). Interestingly, analyses of NS1HA-immunocomplexes revealed marked differences in the interaction profiles of these mutants with the structural proteins. In case of NS1 mutants S114A and W115A, a significant reduction in E and prM co-precipitation, concomitant with a loss of the C-specific signal was observed, arguing for impaired association of these NS1 variants with DENV virions (Fig 8B; reduction of Envelope: ~7.5- and 3.8-fold, respectively). In contrast, mutants D180A and T301A exhibited a selective loss of co-precipitated C while retaining high prM and E interaction, suggesting that NS1 contributes to virus production in an additional manner that is independent from interaction with the envelope glycoproteins. Consistent with their higher competence in supporting infectious particle production, T117A and R314A were still able to bind all three structural proteins, although the C-specific signal was lower as compared to wild-type.
These results were further corroborated by co-localization analyses of HA-tagged NS1 mutants with the envelope glycoproteins revealing reduced co-localization in case of the two mutants that were impaired in virus production (S114A and W115A; S6 Fig), but no significant change in case of all the other mutants. Importantly, no major differences could be observed with respect to NS1 secretion into the culture supernatants (Fig 8C), strengthening the notion that NS1 secretion is functionally unlinked to its role in the production of infectious extracellular virus particles.
To further investigate the interaction between NS1 and the structural proteins, we used immunofluorescence to visualize the sub-cellular colocalization of NS1 with the structural proteins. To allow detection of putative DENV assembly sites and/or assembled virus particles, we aimed to perform simultaneous immunostaining of capsid, envelope and NS1. Since we were limited by the availability of antibodies allowing for triple staining, we engineered a C-terminally mCherry-tagged NS1 variant (NS1mCherry) to be visualized without requirement for antibodies (Fig 9A). In the initial set of experiments, we confirmed correct NS1mCherry expression and sub-cellular distribution. Importantly, mCherry-tagged NS1 supported efficient DENV RNA replication and infectious particle production upon infection with ΔNS1TCP demonstrating full functionality of the fluorescently tagged NS1 (Fig 9B and 9C). Taking advantage of this approach, we analyzed infected cells by confocal microscopy and observed several discrete structures where Envelope, Capsid and NS1 colocalized (Fig 9D). Although the nature of these structures that might correspond to putative assembly sites or assembled virus particles is not clear, this colocalization provides additional evidence for a previously unreported association between NS1 and the structural DENV proteins.
To overcome the inherent limitations of fluorescence microscopy in allocating specific proteins to distinct subcellular structures and to elucidate the nature of these NS1-positive structures we performed correlative light-electron microscopy (CLEM) of VeroE6_NS1mCherry cells infected with ΔNS1TCP (Fig 10). MCherry-fluorescent structures were allocated by confocal microscopy of fixed cells grown on photo-etched gridded coverslips (Fig 10A and 10B), which were subsequently processed for EM (Fig 10C). After alignment of confocal and electron microscopy images we were able to identify NS1-enriched structures. As observed with NS1HA, NS1mCherry-infected cells also contained the characteristic DENV-induced membrane rearrangements. Indeed, we found that highly fluorescent mCherry-positive areas corresponded to ER, vesicle packets (VPs) and ER-associated bags containing arrays of DENV particles (Fig 9D and 9E). These results are consistent with our previous studies describing the association of NS1 with VPs as determined by immuno-EM [5]. Most importantly, the present data further support our assumption that NS1 associates with assembled virus particles (Fig 10Db and 10Ed), in agreement with our results from immunoprecipitation and confocal microscopy experiments.
Flavivirus NS1 has emerged as one of the most enigmatic proteins forming distinct intra- and extracellular complexes and contributing to pathogenesis as well as viral replication. While soluble and cell-surface–associated NS1 was shown to modulate complement activation pathways through interactions with host proteins, such as the regulatory protein factor H, complement factor C4 or clusterin [11–13,28] and to induce cross-reacting antibodies to human proteins [29–32], the role of NS1 in the viral replication cycle has so far been elusive. NS1 was initially thought to be involved in virus assembly or maturation, given its subcellular localization within the ER lumen and its secretion profile, largely mirroring that of the structural proteins prM and E [33,34]. However, this hypothesis was challenged by the co-localization of NS1 with dsRNA, a marker for RNA replication sites and by biochemical and genetic evidences supporting an essential role of NS1 in viral RNA replication [10,35–37]. Additionally, NS1 has been reported to interact with NS4B and/or NS4A, which are assumed to relay NS1 signals to other components of the viral replicase through their ER luminal segments.
In the present study, we used a combination of genetic, biochemical, and imaging approaches to investigate the functional role of intra- and extracellular NS1 in the DENV replication cycle. In addition to the previously reported mutations within the N-linked glycosylation sites [35,38] and the cysteine residues engaged in disulfide bonds [39], we identified 18 additional residues that completely or severely reduced DENV replication. Interestingly, most of these mutations clustered on residues located towards the proposed ER binding site (S1 Fig). Among these, a conserved tryptophan residue at amino acid position 8 (W8A) was found to be essential for RNA replication. This residue is located in close proximity to the previously reported di-amino acid motif (N10K11) within the β-roll domain of WNV NS1 that is also critical for efficient RNA replication [10,40]. These residues are located in an exposed region of the NS1 dimer suggested to face the ER membrane and possibly mediating the interaction with NS4B. Moreover, within the NS1 hexamer, these residues point towards the inner cavity of the barrel and might contribute to the association of NS1 with its lipid cargo (S1B Fig) [9]. While these results are consistent with the proposed way how the NS1 dimer associates with intracellular membranes and provide a detailed genetic map of NS1 residues essential for viral RNA replication, further studies are required to decipher the impact of these mutations on protein-protein interactions and induction of ultra-structural membrane rearrangements.
Most interestingly, our mutagenesis approach identified a group of NS1 mutants with selective defects in infectious particle production. These mutants (S114A, W115A, D180A and T301A) had only minor or negligible defects in NS1 protein stability and viral RNA replication as judged by their replication kinetics in the context of a sub-genomic replicon (Fig 4A), but released up to 100-fold lower amounts of infectious DENV particles than the wild-type (Figs 3A and 4B). Of note, amounts of intracellular infectivity were reduced ~5- to 10-fold, arguing that the NS1 mutations affected both assembly (as evidenced from intracellular virus titers) and release of infectious DENV particles (indicated by titer reduction in cell culture supernatants). Although we cannot precisely comment on the individual contribution of NS1 to virus assembly and particle release, the identification of an additional NS1 mutant (T117A) producing ~12-fold higher amounts of intracellular infectious virus strongly hints towards a primary role of NS1 in the assembly of infectious DENV particles.
Based on previous studies reporting the ability of ectopically-expressed NS1 to trans-complement YFV or WNV NS1 deletion mutants [18,21], we established a DENV-based ΔNS1TCP system to characterize NS1 functions. Taking advantage of this method, we probed the capacity of an intracellularly retained KDEL-tagged NS1 (NS1KDEL) to support production and release of viral particles and demonstrate that secretion of NS1 is dispensable for these functions. Of note, KDEL-tagged proteins are shuttling between the ER and the Golgi apparatus and thus, NS1KDEL could still assist early vesicular traffic of viral particles, i.e. from virion budding sites (ER) to early secretory compartments (ERGIC) (Fig 11). Recently it has been reported that depletion of the KDEL receptor or a subset of class II Arf proteins reduces secretion of non-infectious sub-viral particles and that prM—KDEL receptor interaction plays a role in virus secretion, arguing that flavivirus release from infected cells is assisted by specific sorting mechanisms [41,42]. However, additional viral or host factors appear to be required to coordinate these processes, since knock-down of KDEL receptor or Arf4+5 expression reduced YFV and DENV1-3 secretion less than 10-fold and had no effect on DENV4 and WNV particle production. Further studies will be needed to elucidate the possible contribution of NS1 trafficking from the ER to early or intermediate secretory compartments for DENV particle release.
By using HA-tagged NS1 variants (NS1HA) and ΔNS1 virus-infected cells, we identified a previously unreported interaction between NS1 and the envelope glycoproteins E and prM. Although more than two decades ago E—NS1 complexes were identified in insect cells infected with a selected group of flaviviruses, the interaction was suggested to result from unspecific protein aggregation, lacking functional significance and not reproducible in DENV-2- and YFV-infected cells [26]. In contrast, several lines of evidence suggest that NS1 interaction with these structural proteins is specific and of importance for the production of infectious DENV particles: (i) the co-immunoprecipitation of NS1 with prM and E and, possibly indirectly, with C; (ii) the identification of triple-positive structures enriched in NS1, E and C by confocal microscopy; (iii) the detection of NS1 in compartments containing assembled virions by CLEM. Importantly, immunoprecipitation experiments using lysates of wild-type DENV-2-infected cells and an NS1-specific antibody confirmed the interaction of NS1 with both prM and E glycoproteins. Of note, while in this experimental set-up no specific signal could be detected for the capsid protein, it was well co-precipitated in the DVR2AΔNS1 TCP system. This discrepancy might be due to the overall lower NS1 capture efficiency in case of wild-type virus-infected cells. Alternatively, it is possible that in wild-type virus-infected cells subviral particles (SVPs) are produced in higher abundance relative to virus particles, whereas in the DVR2AΔNS1 TCP system production of virus particles might be favored relative to SVPs. Consistent with this assumption we observed a much faster and more efficient production of infectious virus particles in our TCP system than with wild-type virus-infected cells (Fig 6B). Assuming that NS1 can also interact with prM/E present on the surface of SVPs, in case of their excess production we would still observe coprecipitation of NS1 with the envelope glycoproteins, whereas C would no longer be co-precipitated. In contrast, when virus particles are produced in excess over SVPs, we would observe NS1 co-precipitation with prM/E and C, i.e. virions. This is consistent with the observed colocalization of NS1 with C and E (Fig 9) and the enrichment of NS1 in areas containing fully assembled DENV particles (Fig 10). Moreover, the conclusion is consistent with the membrane topology of E, prM and NS1 that are located in the lumen of the ER, while capsid resides on the cytoplasmic side of ER membranes (Fig 11). Since no tangible interactions between capsid and the glycoproteins were reported to date, the most likely explanation for the apparent NS1-C co-precipitation is an interaction between NS1 and assembled virions. While the precise mechanism remains to be determined, the specificity and functional relevance of these interactions was corroborated by the interaction profiles of NS1 mutants having defects in infectious particle production. Mutations affecting highly conserved residues of the flexible solvent-exposed loop in the NS1 Wing domain (S114A and W115A; Fig 8D) simultaneously abrogated C, prM and E association arguing that this NS1 loop is engaged in interaction with DENV virions. In contrast, mutations affecting residues in the β-ladder domain (D180A and T301A) preserved glycoprotein binding, but prevented association with capsid arguing for a second function of NS1 in the assembly of DENV particles that is independent from envelope protein interaction. While the exact mechanism remains to be established, it is tantalizing to speculate that NS1 might, directly or indirectly via these residues, interact with other viral or cellular factor(s) required for the formation of infectious DENV particles. Alternatively, NS1 might assist membrane budding or conformational changes in prM/E required for the envelopment of nucleocapsids. This hypothesis is supported by the localization of “capsid non-binder” mutants in the NS1 dimer structure. In fact, D180 resides at the intersection of the Wing and β-ladder domain and T301 points towards the ER membrane and is solvent exposed (Fig 8D). These mutations might affect the NS1 dimer fold, its affinity for membranes or its membrane-bending ability, while preserving prM/E-association through the distal tips of the Wing domain that points towards the ER lumen. Alternatively, these mutations might induce conformational constraints in prM/E complexes, reducing their plasticity and capability to envelope budding nucleocapsids, affect the recruitment of prM/E to assembly sites or only indirectly affect association with capsid, as a consequence of altered interactions with other viral [43,44] or host factors playing critical roles in virion morphogenesis. However, ultrastructural analysis of NS1 mutants failed to identify assembly intermediates, the only striking difference to the wild-type being a marked reduction in the overall amount of electron-dense particles found in proximity to or directly at the plasma membrane (S2 Fig). These results suggest that nucleocapsid formation and envelopment are coupled or that naked nucleocapsids have an extremely short half-life. Further experiments will be required to dissect the exact role of NS1 in the assembly process of infectious virus particles.
Assembly of flavivirus particles is a poorly understood process (reviewed in [45]). High-resolution imaging approaches have shed some light on the topological arrangement of viral RNA replication and virus particle assembly. The viral replicase machinery is assumed to reside in highly organized membranous structures, designated as vesicle packets (VPs). These structures are formed by ER invaginations containing pores that would allow the release of newly synthesized viral RNA to be used for packaging into virions [5,6]. While genetic evidence suggests the involvement of several non-structural proteins and host factors in flavivirus assembly [41,43,44,46–48], the underlying molecular mechanisms are not known. Besides the lack of tangible interactions between C and the prM/E complex, major limitations are posed by the difficulty to visualize assembly intermediates and the seemingly unspecific incorporation of genomic RNA into nucleocapsids [49,50]. Based on the results presented here, it is tempting to speculate that NS1 might assist virion morphogenesis via its lipid-remodeling activity [8,9], its affinity for membranes [9] and its ability to interact with both non-structural [10,15] and structural proteins. In this respect NS1 might provide essential lipids or recruit essential host factor required for the biogenesis of the replication complex, while coordinating the recruitment of E and prM to assembly sites juxtaposed to the viral replicase [5] (Fig 11).
In conclusion, the present study provides a comprehensive genetic map of NS1 determinants important for viral RNA replication and identifies a novel role of NS1 for the production of infectious DENV particles. We demonstrate that NS1 interacts with the envelope glycoproteins presumably on the surface of virions and these interactions are required for efficient production of infectious virus particles. Given its multiple roles in counteracting host defense, promoting RNA replication and enhancing production of virus particles, NS1 exemplifies the genetic economy of flaviviruses and emerges as attractive target for antiviral drugs.
The mouse monoclonal antibody recognizing human GAPDH (sc-47724/0411) was purchased from Santa Cruz Biotechnology (Santa Cruz, CA). The mouse anti-Envelope monoclonal antibody (3H5-1) was purchased from ATCC. The mouse monoclonal antibody 6F3.1 reacting with the capsid protein was a kind gift of Dr. John G. Aaskov (Queensland University of Technology, Australia); rabbit polyclonal serum anti-capsid was a kind gift of Dr. Andrea Gamarnik (Fundación Instituto Leloir, Argentina). The rabbit polyclonal antibodies recognizing Envelope, NS1, NS5 and prM were previously described [5]. Rabbit anti-HA antibody (Ab9110) and mouse anti-NS1 antibody (ab41623) were purchased from Abcam. The mouse monoclonal anti-HA (H3663) and anti-Flag (F1804) antibodies, agarose anti-HA conjugated beads and secondary anti-mouse and anti-rabbit horse-radish peroxidase-conjugated antibodies were purchased from Sigma-Aldrich (Sigma-Aldrich, Saint Louis, MO).
Huh7 [51], HeLa [52], VeroE6 (ATCC #CRL-1586), and BHK-21 (ATCC #CCL-10) cells were maintained in Dulbecco's modified Eagle medium (DMEM; Invitrogen, Karlsruhe, Germany) supplemented with 2 mM l-glutamine, nonessential amino acids, 100 U/ml penicillin, 100 μg/ml streptomycin and 10% fetal calf serum. VeroE6 stably expressing NS1WT or tagged derivatives were cultured in the presence of 10 μg/ml of puromycin.
Samples were denatured in 2x protein sample buffer (200 mM Tris [pH 8.8], 5 mM EDTA, 0.1% Bromophenolblue, 10% sucrose, 3% SDS, 1 mM DTT) and incubated for 5 min at 95°C. Proteins were separated by SDS-polyacrylamide gel electrophoresis (PAGE) and transferred onto polyvinylidene difluorid membranes by using a MINI-SDS-PAGE wet-blotting apparatus (Bio-Rad, Munich, Germany). Membranes were blocked with 5% non-fat dry milk in PBS/0.5% Tween-20 (PBST) and incubated with primary antibodies (capsid 1:50; GAPDH 1:1,000; HA 1:1,000; E 1:1,000; prM 1:500; NS1 1:500) by over-night incubation at 4°C or for 1 h at room temperature. After 3 washes with PBST, membranes were incubated with secondary horse radish peroxidase-conjugated antibodies, developed with the Western Lightning Plus-ECL reagent (Perkin Elmer; Waltham, MA) and bands were imaged using an Intas ChemoCam Imager 3.2 (Intas, Göttingen).
VeroE6 helper cells expressing different forms of NS1 were infected with ΔNS1TCP at an MOI of 1. Two days later, cells were fixed with 4% PFA for 10 min at room temperature, permeabilized with 0.5% (vol/vol) Triton X-100 in PBS and aspecific biding sites blocked with PBS containing 5% FBS for 30 min at RT.
For staining of NS1mCherry cells, rabbit anti-NS1 (S3 Fig), or rabbit anti-C and mouse anti-Envelope antibodies (Fig 8) were used in combination with goat anti-mouse Alexa 647-conjugated and donkey anti-rabbit Alexa488-conjugated secondary antibodies. For staining of NS1HA cells (wild-type and mutants) (S4 Fig), rabbit anti-HA and mouse anti-Envelope antibodies were used in combination with goat anti-mouse Alexa 568-conjugated and donkey anti-rabbit Alexa488-conjugated secondary antibodies. Nuclear DNA was stained with 4′,6-diamidino-2-phenylindole (DAPI) (Molecular Probes, Karlsruhe, Germany). Coverslips were mounted in Fluoromount-G mounting medium (Southern Biotechnology Associates, Birmingham, AL). For 3D visualization of NS1mCherry, E and C, samples were imaged with an Ultraview ERS spinning disk (PerkinElmer Life Sciences) on a Nikon TE2000-E inverted confocal microscope using a Plan-Apochromat VC 100× objective (numeric aperture [NA], 1.4). Optical sections of 0.13 μm were acquired separately for each channel. Z-stacks were deconvolved with a theoretical point-spread function, and chromatic shifts between green and far-red dyes were corrected using Autoquant X3 software. 3D reconstructed images were created using the Imaris 8 software package. For colocalization analyses of HA-tagged NS1 mutants and envelope fluorescence signals, Pearson's correlation coefficient was calculated on single plane images, by using the integrated function in Fiji (ImageJ).
For determination of virus titers by limiting dilution assay, Huh7 target cells were seeded into 96-well plates (104 cells/well) the day before infection. Cells were inoculated with serial dilutions of virus-containing supernatants that had been filtered through a 0.45-μm-pore-size filter. Infected cells were detected by immune staining of the E protein using the mouse anti-E antibody (3H5-1; diluted 1:500) and secondary horseradish peroxidase-conjugated antibody (1:200). Virus titers (expressed as 50% tissue culture infective dose [TCID50]/ml) were calculated as previously reported [53]. For determination of virus titers by Focus forming unit (FFU) assay, VeroE6 or VeroE6_NS1WT cells (2x105 cells/well) seeded into 24-well plates, were infected with serial dilutions of 0.45-μm-filtered supernatants and incubated in the presence of 0.8% methylcellulose for 5 days. Monolayers were rinsed twice in PBS, fixed with 5% PFA and permeabilized with 0.2% (v/v) TritonX-100 in PBS for 15 min. Infected foci were detected by immune staining of the E protein using the mouse anti-E antibody (3H5-1; diluted 1:1,000 in PBS) and secondary horseradish peroxidase-conjugated antibody (1:200). Alternatively, virus titers were determined by standard plaque assay (PFU) on target VeroE6 cells as previously described [54]. To determine intracellular infectivity titers, transfected cells were disrupted by several freeze–thaw cycles as described earlier [55]. In brief, transfected Huh7 cells were extensively washed with PBS, scraped off the plate into PBS and centrifuged for 5 min at 700 × g. Cell pellets were resuspended in complete DMEM (containing 15 mM HEPES, pH 7.2–7.5) and subjected to three cycles of freezing and thawing by using liquid nitrogen and a thermo block set to 37°C. Cell debris was removed by centrifugation at 20,000 × g for 10 min at 4°C. Virus-containing culture supernatants from transfected cells were treated in the same way and infectivity was determined in parallel by limiting dilution assay as described above.
VeroE6 cells were seeded into 15-cm2 dishes (7.5x106 cells/dish). Twenty-four hours later, cell monolayers were either mock-infected or infected with DENV-2 (MOI = 1). Forty-eight hours later, cell monolayers were scraped into 1 ml lysis buffer (50 mM Tris-HCl [pH 8.0], 0.5% NP-40, 150 mM NaCl and protease inhibitor cocktail (cOmplete, Roche)). After 30 min incubation on ice, cell debris was removed by 15 min centrifugation at 13,800xg and 400 μl of clarified cell lysate was incubated with rabbit pre-immune serum (PIS) or rabbit anti-NS1 antiserum for 6 hours at 4°C in a head-to-head shaker. Samples were incubated with 40 μl of protein A beads slurry (Sigma Aldrich, St. Louis, USA) for 1 hour at 4°C, washed three times with 1 ml of lysis buffer and protein A-bound complexes were transferred into a fresh tube. After a final wash with 1 ml of lysis buffer for 15 min at 4°C, protein complexes were eluted at room-temperature by two consecutive steps with 75 μl of 0.1 M glycine [pH 2.5] for 5 min. Collected supernatants were immediately neutralized by adding 7.5 μl 1 M Tris-HCl [pH 8] and denatured for 5 min at 95°C in the presence of 33 μl of 6X SDS sample buffer. Alternatively, mouse anti-HA or mouse anti-E antibodies were used in combination with protein G beads slurry (Sigma Aldrich, St. Louis, USA) and the immunoprecipitation was carried exactly as described above. For ΔNS1TCP experiments, VeroE6 cells stably expressing an empty pWPI vector (CTRL), or wild-type NS1 (NS1WT), or HA-tagged NS1 (NS1HA) or derivatives thereof were seeded into 10-cm2 dishes (3x106 cells/dish). Twenty-four hours later, cell monolayers were infected with DVR2AΔNS1 (MOI = 1) for 4 h at 37°C. Forty-eight hours post-infection, cell monolayers were scraped into 1 ml lysis buffer (50 mM Tris-HCl [pH 8.0], 0.5% NP-40, 150 mM NaCl and protease inhibitor cocktail (cOmplete, Roche) as recommended by the manufacturer). After 30 min incubation on ice, cell debris was removed by 15 min centrifugation at 13,800xg. For HA-specific affinity capture, samples were incubated with HA-specific agarose beads (Sigma-Aldrich, St.Louis, USA) for 5 h by continuously inverting the tubes at 4°C. Beads were washed three times for 20 min with large volumes of lysis buffer at 4°C and samples were eluted at room-temperature in two consecutive steps with 3% SDS in PBS for 5 min and PBS for 5 min, respectively. The two eluates were pooled and precipitated over-night at -20°C with 4 volumes of ice-cold acetone. Samples were centrifuged for 30 min at 20,000xg, air-dried, resuspended in 2x SDS sample buffer and boiled for 5 min at 95°C. Alternatively, 4 h before ΔNS1TCP infection, VeroE6 cells stably expressing wild-type NS1 (NS1WT) or HA-tagged NS1 (NS1HA) were transfected with 10 μg of pcDNA 3.1(+) empty vector or pCMV_NS4B-FLAG (encoding a C-terminally Flag-tagged NS4B protein of Hepatitis C virus) by using the TransIT-LT1 transfection reagent (MirusBio LLC, Madison, WI, USA) as recommended by the manufacturer. Infection and HA-specific immunoprecipitation were carried out exactly as described above. Eluted proteins were further analyzed by western blot as specified in the results section. For analysis of secreted NS1, supernatants were clarified through 0.45 μm filters and incubated with 40 μl mouse anti-HA slurry beads over-night at 4°C, in a head-to-head shaker. Immunoprecipitates were washed three times with lysis buffer and eluted as described above. Alternatively cell culture supernatants containing NS1 were used undiluted for SDS-PAGE.
In vitro transcripts were generated as previously described [56]. For RNA transfection, single-cell suspensions were prepared by trypsinization, washed with PBS, and resuspended at a concentration of 1x107 cells (Huh7) or 1.5x107 cells (VeroE6) per ml in Cytomix, supplemented with 2 mM ATP and 5 mM glutathione. Five to 10 μg of subgenomic or genomic in vitro transcript was mixed with 400 μl of the cell suspension and transfected by electroporation using a Gene Pulser system (Bio-Rad) and a cuvette with a gap width of 0.4 cm (Bio-Rad) at 975 μF and 270 V. Cells were immediately diluted into 20 ml of DMEM cplt and seeded in the appropriate format (1ml/well in 24-well plates; 2 ml/well in 12-well plates; 15 ml/dish in 15 cm-diameter dishes).
Human immunodeficiency virus (HIV)-based particles that were pseudotyped with the vesicular stomatitis virus glycoprotein (VSV-G) were generated by transfection of 293T cells as described previously [53]. For production of transducing lentiviral particles, 293T cells were co-transfected with a transfer vector encoding the gene of interest and a puromycin resistance gene (pWPI_Puro), the HIV-1 packaging plasmid (pCMV) and a VSV-G expression vector (pMD.G) (ratio 3:3:1). Cells were transfected using the CalPhos mammalian transfection kit as recommended by the manufacturer (Becton Dickinson). After 48 and 72 h, supernatants were harvested, clarified through 0.45 μm pore size filters, pooled and stored in aliquots at -20°C until use. Titers of lentiviral particles were estimated by colony-forming unit (CFU) assay using HeLa cells and serial dilutions of each lentiviral stock. Inoculated cells were subjected to selection using the appropriate antibiotic for 5–7 days and surviving cell colonies were stained for 15min with a 1% crystal violet solution. Colonies were counted under a bright-field inverted microscope and lentivirus titers were calculated as CFU/ml.
Huh7 or VeroE6 cells transfected with full-length or subgenomic DVsR2A in vitro transcripts were seeded as specified in the results section (typically 12- or 24-wells plates). Replication was determined by measuring luciferase activity in cell lysates 4, 24, 48 and 72 h after transfection. For determination of luciferase activity, cells were washed once with PBS and lysed by adding 200 μl of luciferase lysis buffer as previously described [57]. Cells were frozen immediately at −70°C and after thawing, lysates were resuspended by gentle pipetting. For each well 20 μl lysate, mixed with 400 μl assay buffer (25 mM glycylglycine, 15 mM MgSO4, 4 mM EGTA, 1 mM DTT, 2 mM ATP, 15 mM K2PO4 [pH 7.8], 1.42 μM coelenterazine H), were measured for 10 sec in a tube luminometer (Lumat LB9507, Berthold, Freiburg, Germany). In some cases (24-well plates, 100 μl lysis buffer per well) a plate luminometer was used (Mithras LB940, Berthold, Freiburg, Germany). Each well was measured in duplicate. To determine the amount of infectious virus particles released into culture supernatants 72 h after electroporation, naïve VeroE6 cells were inoculated with culture supernatants of transfected cells and 48 h later luciferase activity was determined. Kinetics of virus replication were calculated by normalizing the relative light units (RLU) measured at a given time point to the respective 4 h value.
The plasmid containing a synthetic version of the full-length DENV-2 strain 16681 (pFK-DVs) and subgenomic constructs derived therefrom without or with luciferase reporter were previously described [56]. For NS1-targeted site-directed mutagenesis external primers NS1_BamHI_f (5’-CTG GGA TTT TGG ATC CTT GGG AGG AG-3’) and NS1_KasI_r (5’-TCC GTC ATA GTG GCG CCT ACC ATA AC-3’) were used in combination with mutagenic forward and reverse primers (the full list of primers is available upon request). Amplicons containing the desired point mutation in the NS1 coding region were inserted into the full-length DVsR2A constructs via the BamHI-KasI restriction sites. Full-length non-reporter constructs containing selected NS1 mutations were generated by insertion of a DNA fragment that was excised via BamHI and KasI from pFK-DVsR2A into pFK-DVs; subgenomic luciferase reporter constructs were generated by DNA fragment exchange using MluI and KasI and insertion into pFK-sgDVsR2A. The pCMV_NS4B-FLAG expressing C-terminally Flag-tagged NS4B protein of Hepatitis C virus was previously described [58]. The pWPIpuro-based NS1 constructs used for production of lentiviral vectors were generated by PCR-based amplification of the NS1 encoding sequence plus the last 24 codons of the envelope coding region, by using full-length genomic constructs as template and the following primers: pWPIpuro_BamHI_NS1_HA_frw (5’-GCT GGG ATC C ACC ATG AGC ACC TCA CTG TCT GTG ACA CTA GTA TTG GTG-3’) and pWPIpuro_NS1_Stop_NheI_rev (5’-AGA TAG CTA GCC TAA GCT GTG ACC AAG GAG TTG ACC AAA TTC-3’). Amplicons were inserted into pWPI-Puro via BamHI and SpeI restriction sites. For variants encoding the C-terminal HA epitope or the KDEL ER retrieval sequence, the primer pWPIpuro_BamHI_NS1_HA_frw was used in combination with NheI_NS1-HA_stop_rev (5’-ATA GCT AGC CTA AGC GTA ATC TGG AAC ATC GTA TGG GTA TGA TCC AGC TGT GAC CAA GGA GTT GAC CAA ATT CTC TTC TTT CT-3’) or pWPIpuro_NS1_Stop_NheI_KDEL_rev (5’-AGA TAG CTA GCC TAT AGC TCG TCC TTA GCT GTG ACC AAG GAG TTG ACC-3’), respectively.
The pWPIpuro-based NS1mCherry expression construct was generated by amplifying the mCherry sequence contained in pFKI389neoNS3-3′δg_JFH-1_NS5A-aa2359_mCherry [59] with primers pWPIpuro_SpeI_mCherry_frw (5´-TCA ACT CCT TGG TCA CAG CTA CCG GTG GAT CGA TGG TGA GCA AGG GCG AGG A-3´) and pWPIpuro_SpeI_mCherry_rev (5´-AAA ACT AGT CTA CTT GTA CAG CTC GTC CAT GC-3´) and the NS1 coding sequence contained in pWPIpuro_NS1 with primers pWPIpuro_SpeI_NS1_frw (5´-GAC ACT AGT ATT GGT GGG AAT TGT GAC AC-3´) and pWPIpuro_SpeI_NS1_rev (5´-TCC TCG CCC TTG CTC ACC ATC GAT CCA CCG GTA GCT GTG ACC AAG GAG TTG A-3´). The two PCR fragments were used to generate an intermediate amplicon using primers pWPIpuro_SpeI_NS1_frw and pWPIpuro_SpeI_mCherry_rev, which was inserted into pWPIpuro via SpeI. Finally, the envelope leader peptide sequence was excised from pWPIpuro_NS1 wild-type and inserted upstream of the NS1mCherry sequence via BamHI-MluI.
The ΔNS1 full-length DVsR2A genome used for trans-complementation studies in NS1 helper cell lines, was created by insertion of a 97 codon in-frame deletion into the NS1 open reading frame using NS1_BamHI_f and NS1_KasI_rev as external primers and NS1_156_frw (5’-GAA TTC GTT GGA AGT TGA ACA CAA CTA TAG ACC AGG CTA-3’) and NS1_156_rev (5’-TAG CCT GGT CTA TAG TTG TGT TCA ACT TCC AAC GAA TTC-3’) as internal primers. Thus, in the final construct, the first 156 codons and the last 99 codons of NS1 were retained, respectively.
Sequence alignment of NS1 open-reading frames were performed using the ClustalW algorithm available in the JalView Desktop software and a ClustalW scoring algorithm, with the following isolates (UniprotKB/Swiss-Prot accession numbers are given): DV-1 (Brazil/97-11/1997) P27909; DV-2 (Thailand/16681-PDK53) P29991; DV-3 (Martinique/1243/1999) Q6YMS3; DV-4 (Thailand/0348/1991) Q2YHF0; West Nile virus P06935; Yellow Fever (Ivory Coast/1999) Q6J3P1; Japanese Encephalitis virus (SA-14) P27395; Kunjin virus (MRM61C) P14335; St. Louis Encephalitis virus (MS1-7) P09732. Molecular graphics were performed with the UCSF Chimera package developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco [60] on the DENV-2 NS1 crystal structure (Protein Data Bank [PDB] accession no. 4O6B).
VeroE6-based helper cell lines expressing different NS1HA mutants were seeded onto glass coverslips (5x104 cells/well) and 16 h later, infected with 1 MOI of DVR2AΔNS1. After a 48 h incubation period, cells were fixed and prepared for transmission electron microscopy as described previously [61]. For correlative light-electron microscopy, VeroE6-based helper cell lines expressing NS1mCherry were seeded into glass-bottom culture dishes containing photo-etched gridded coverslips (MatTek Corporation, Ashland, MA) and infected with 1 MOI of DVR2AΔNS1. After 48 hours, cells were fixed with 4% PFA and 0.2% glutaraldehyde in PBS for 30 min at room temperature, washed three times with PBS, stained with DAPI and analyzed by fluorescence microscopy to acquire optical sections of 0.13 μm as described above. NS1mCherry-positive cells were imaged and their position on the gridded coverslip was recorded. Cells were then processed for analysis as described previously [61]. The DAPI signal was used for correlation purposes and images were adapted by using the Image J (version 1.46r) and Adobe Photoshop (version 12.1.1) software packages.
Statistical analyses were performed by applying the two-tailed, unpaired Student’s t-test available within the GraphPad Prism (ver. 5.0) software.
|
10.1371/journal.pntd.0000959 | Levofloxacin Cures Experimental Pneumonic Plague in African Green Monkeys | Yersinia pestis, the agent of plague, is considered a potential bioweapon due to rapid lethality when delivered as an aerosol. Levofloxacin was tested for primary pneumonic plague treatment in a nonhuman primate model mimicking human disease.
Twenty-four African Green monkeys (AGMs, Chlorocebus aethiops) were challenged via head-only aerosol inhalation with 3–145 (mean = 65) 50% lethal (LD50) doses of Y. pestis strain CO92. Telemetered body temperature >39°C initiated intravenous infusions to seven 5% dextrose controls or 17 levofloxacin treated animals. Levofloxacin was administered as a “humanized” dose regimen of alternating 8 mg/kg and 2 mg/kg 30-min infusions every 24-h, continuing until animal death or 20 total infusions, followed by 14 days of observation. Fever appeared at 53–165 h and radiographs found multilobar pneumonia in all exposed animals. All control animals died of severe pneumonic plague within five days of aerosol exposure. All 16 animals infused with levofloxacin for 10 days survived. Levofloxacin treatment abolished bacteremia within 24 h in animals with confirmed pre-infusion bacteremia, and reduced tachypnea and leukocytosis but not fever during the first 2 days of infusions.
Levofloxacin cures established pneumonic plague when treatment is initiated after the onset of fever in the lethal aerosol-challenged AGM nonhuman primate model, and can be considered for treatment of other forms of plague. Levofloxacin may also be considered for primary presumptive-use, multi-agent antibiotic in bioterrorism events prior to identification of the pathogen.
| Yersinia pestis is the causative agent of bubonic plague as well as a rare severe form known as primary pneumonic plague resulting from the inhalation of contaminated aerosols. The relative ease of aerosol preparation and high virulence makes Y. pestis a dangerous bioweapon. The current study describes the treatment of established pneumonic plague with the widely available, broad-spectrum fluoroquinolone antibiotic levofloxacin in a nonhuman primate model. African green monkeys inhaled a target dose of 100 lethal doses for 50% of animals (LD50) and were monitored for fever and vital signs by telemetry. Fever was the first sign of illness, correlating with bacteremia but preceding radiographic pneumonia, and initiated intravenous levofloxacin treatment in doses designed to mimic antibiotic levels achieved in humans. All animals treated with saline died and all animals completing 10 days of treatment survived, with resolution of high fever within 24–48 hours. We conclude that levofloxacin may be an appropriate broad-spectrum antibiotic for presumptive therapy in an aerosolized bioweapons attack and should be studied for treatment of bubonic plague.
| Yersinia pestis is the causative agent of bubonic plague, initiated by the bite of an infected flea, and primary pneumonic plague (PPP, often called inhalational plague) resulting from the inhalation of aerosolized contaminated environmental dusts [1], [2]. Y. pestis is also one of the most dangerous bioweapons due to the relative ease of lethal aerosol preparation, high virulence, the rapidity of onset of symptoms and death by PPP, and its history of use as a bioweapon [3]. Naturally acquired PPP is relatively rare with few outbreaks in the developing world [4], [5], the risk of person-to-person dissemination is low [6], and few detailed reports describe the evolution of the human disease [7], [8], [9].
Development of vaccines and therapeutics for plague can not utilize human trials either in natural or intentional infections, and must rely on animal models. The U.S. Food and Drug Administration (FDA) “animal rule” (21 CFR Part 314) permits approval of therapeutics and vaccines based on testing in appropriate animal models. The mouse model [10],[11],[12] and rat model [13] of pneumonic plague bears multiple similarities to the human disease, including a brief anti-inflammatory incubation period followed by the rapid evolution of the pro-inflammatory fulminant disseminated disease. The molecular arsenal secreted by Y. pestis is well characterized [14] and appears to mediate the anti-inflammatory phase in the lung [15], [16]. Potential antibiotic therapies have been screened in the mouse model [17], but valid extrapolation of efficacy from mice to humans is not yet established.
Nonhuman primates have been known for decades to be highly susceptible to the pneumonic form of plague [18], [19], [20] and have been considered extensively for their value as animal models of pneumonic plague [21]. The African Green monkey (AGM) is also highly susceptible to Y. pestis infection by the aerosol route [20]. The AGM may have several advantages over the cynomolgus macaques in that telemetered fever above 39°C is the first and uniform sign of systemic disease following aerosolized Y. pestis challenge, and there appears to be less individual variation in innate resistance among the AGMs [22].
Identifying the optimal antibiotic for treatment of pneumonic plague faces several challenges. First the disease presents with non-specific symptoms of fever and pneumonia until late stages when hemoptysis suggests a diagnosis other than community-acquired pneumonia (CAP). Syndromic surveillance may not detect an outbreak within a few days of a bioterrorism event, and laboratory diagnosis may be delayed [23], [24], [25]. Second, the antibiotic must be widely available in all hospitals and established as a drug-of-choice not only for bacterial pneumonia but also for multiple biothreat agents including anthrax [26], [27]. Third, the antibiotic should ideally have excellent oral bioavailability, so that use of oral antibiotic could be used if a massive bioterrorism event taxed hospital facilities. The currently recommended antibiotic to treat plague, gentamicin, is not widely used in CAP unless Pseudomonas species are suspected. However, levofloxacin satisfies the second and third criteria since it is efficacious and widely used for CAP [28], [29].
The objective of this study was to evaluate the efficacy of intravenous treatment with levofloxacin following a lethal aerosol challenge to Y. pestis CO92 in the AGM model. Treatment was initiated after the onset of fever in order to use a readily available clinical marker of established disease rather than rely on a marker for bacteremia or other indicator of disseminated disease. Even though targeted levels of plasma antibiotic concentrations were not uniformly met, all treated AGMs survived established pneumonic plague, and resolution of systemic signs was apparent after only 2–3 days of treatment.
Wild-caught African Green monkeys (Chlorocebus aethiops) (Alpha-Genesis, Inc.) weighed 3–8 kg and were at least 2 yrs old. All procedures were conducted under protocols approved by the Institutional Animal Care and Use Committee in Lovelace Respiratory Research Institute facilities accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC). Animals were individually housed in stainless steel cages with wire mesh bottoms, in rooms on a 15/9-h light/dark cycle at temperature between 20–25°C and relative humidity between 11–73%. Diet was Harlan Teklad Certified 20% Monkey Diet 2050C (Harlan-Teklad) twice daily, supplemented with treats, with ad libitum municipal tap water.
Animals were conditioned to a restraint collar, poles, restraint chairs, and limb restraints. Radiotelemeters (Model T30F, Konigsberg, Inc.) were implanted subcutaneously in the left abdominal wall for continuous monitoring of body temperature, intrathoracic pressure, respiratory rate, heart rate, and electrocardiographic traces. Venous access catheters (Broviac, Cohorts 1 and 2; or Hickman dual-port, Cohort 3) were inserted in the right femoral vein, tunneled through the right flank and back, emerging through the skin of the upper mid-back and protected by a jacket [30]. No study animals had received systemic antibiotics within 28 days prior to aerosol exposure with Y. pestis strain CO92 nor topical mupirocin ointment within 14 days of aerosol exposure. Animals were moved into the Animal Biosafety Laboratory-3 at least 1 week prior to aerosol challenge with Y. pestis to permit acclimatization and obtain baseline values for telemetry measurements.
Twenty-six AGMs in three cohorts were randomized into treatment groups. Two animals were removed from Cohort 2, one prior to infectious challenge due to health reasons and one after challenge due to initiation of treatment prior to becoming febrile. The subsequent analysis included the remaining 24 animals. Animals were randomized into test groups using a computerized data acquisition system (Path-Tox 4.2.2; Xybion) based on body weights and randomized into exposure order using Microsoft Excel's random number generator.
Y. pestis strain CO92 was originally isolated in 1992 from a person with a fatal case of pneumonic plague [8] and was supplied by C.R. Lyons at the University of New Mexico. All work done was performed under Biosafety Laboratory-3 conditions. For each cohort exposure, one working stock cryovial of Y. pestis was removed from frozen storage, thawed, and used to inoculate five tryptose blood agar base (TBAB)+yeast extract slants. After incubation at 28±2°C for 72±8 h the slants were washed with 1% peptone, combined and centrifuged at 4100 rpm at 5±3°C for 25±5 min. The cell pellet was suspended in 1% peptone and the optical density at 600 nm (OD600) was determined. The bioaerosol sprays were prepared in brain heart infusion broth (BHIB) from the suspended centrifuged culture based on the OD600 and a previously prepared concentration/OD curve. The suspension was adjusted to achieve the target aerosol exposure dose of 100±50 LD50 doses or approximately 35,000 cfu of Y. pestis [Pitt MLM, DN Dyer, EK Leffel, et al. Ciprofloxacin treatment for established pneumonic plague in the African Green Monkey. Abstract B-576, 46thICAAC meeting, San Francisco, CA, September 28, 2006].
After fasting overnight the animals were anesthetized with 2–6 mg/kg Telazol 15 min prior to aerosol exposure and baseline radiographs (Study Day 0). The exposure system consisted of a head-only exposure unit contained in a Class 3 biosafety glovebox [31] and previously described in our laboratories [32], [33]. Real-time plethysmography (Buxco) measuring respiratory frequency, tidal volume, and minute volume targeted an inhaled volume of 5 L, with actual exposure times ranging from 10–15 min. Suspensions of Y. pestis strain CO92 were nebulized in a Collison nebulizer (MRE-3 jet, BGI, Inc.), and delivered to the freely breathing anesthetized AGMs. The bacteria-containing aerosol was sampled directly into an all glass impinger (AGI; Ace Glass, Inc.) drawn from the head-only exposure apparatus downstream from the primate's nares and bacteria concentrations were confirmed by quantitative bacterial culture and purity was assessed by colony morphology. The target particle size was 1–3 µm, determined using an Aerodynamic Particle Sizer Spectrometer (Model 3321, TSI, Inc.; Cohorts 1 and 2) or a GRIMM Portable Aerosol Spectrometer Model 1.109 (Cohort 3) for 0.5–20-µm particles. The mass mean diameter of the aerosols was determined to be 1.9–2.4 µm (1.32–2.80 geometric standard deviation). Aerosolized pathogen dose was calculated using the following formula: Dose = (C×V), where C is the concentration of viable pathogen in the exposure atmosphere, and V is the volume inhaled.
Levofloxacin (Levaquin Injection Premix in Single-Use Flexible Containers as 5 mg levofloxacin/mL 5% dextrose; Ortho-McNeil, Inc.) or the control solution (5% dextrose in water, D5W) was infused into the femoral vein catheter by syringe pump over 30±5 min. Infusions were initiated within 6 h of the appearance of telemetered fever defined as a mean temperature ≥39°C for more than 1 h. Because previous studies demonstrated a clearance of levofloxacin approximately three times more rapid in rhesus macaques than in humans [10], daily levofloxacin infusions were dosed at 8 mg/kg body weight, followed by 2 mg/kg administered 12.0±0.5 h later. Infusions of levofloxacin or D5W were continued until death, moribund euthanasia, or 20 infusions had been completed.
Clinical observations were made twice daily cage-side noting activity, posture, nasal discharge, sneezing, coughing, respiratory characteristics, ocular discharge, inappetance/anorexia, stool characteristics, seizures, neurologic signs, or other abnormalities. Body weights were measured within 1 week prior to aerosol exposure, on the day of aerosol challenge, and at necropsy.
Implanted T30F telemetry devices continuously monitored body temperature, respiratory rate, heart rate, and electrocardiogram. Temperature, respiration signal, and ECG was recorded by CA Recorder software (D.I.S.S., LLC) every 5 min and averaged for hourly values by VR2 software (D.I.S.S., LLC). Respiratory rate was recorded by intrapleural pressure changes. Heart rate was recorded by the software counting R waves per minute.
The decision for euthanasia was based on the development of at least two moribund criteria: >60 respirations/min or deep labored breathing; abnormal repolarization signals (persistently inverted T waves or depressed ST segment); seizures; falling off perch; unresponsive to stimulation; refusal to eat offered food. The Principal Investigator or staff veterinarian making decisions regarding euthanasia was blinded to the animal's treatment group.
A digital chest X-ray was taken at the time of anesthesia prior to aerosol exposure on Day 0, on Day 5 for animals necropsied for moribund disease or surviving to that point, and on Day 28 for animals surviving in the first two cohorts. Radiographs were qualitatively reviewed by a veterinary radiologist (Veterinary Imaging Center of South Texas) who was blinded to treatment group and stage of disease.
For levofloxacin plasma concentrations a sample of venous blood was drawn 10–30 min before the onset of infusions number 3, 6, and 19 (trough level) and 5–15 min (peak level) after the termination of infusions number 1, 3, 6, and 19. The plasma was centrifuge-filtered through a 0.2-µm Nanosep MF centrifugal filter (Pall Corp.) at 13,000×g for 40 min and extracted in 4 ml dichloromethane containing 250 µL KH2PO4 70 mM∶NaHPO47H2O 80mM, 2∶3 v∶v. After phase separation the dichloromethane was evaporated, and the residue was reconstituted with 100 µL acetonitrile∶pure water, 1∶1 v∶v with 0.1% formic acid. Chromatographic separation was conducted with an Agilent 1100 HPLC with a Discovery HS F5 (Supelco #56700-U) column. The fluorescence detector was set at excitation wavelength of 296 nm and emission wavelength of 504 nm. Data was processed using Varian Galaxie Chromatography Data System software version 1.8.505.5. The lower limit of quantification was established at 30 ng/ml and the method met pre-determined performance criteria for selectivity, accuracy, precision, recovery, calibration curve, and dilutional linearity.
For quantitative bacteriology daily on Days 2–6, and on Days 14 and 28, venous blood (target volume of 0.5–1.5 mL) was collected percutaneously from the femoral vein through a site washed three times with povidone iodine, transferred to an EDTA tube, and three log10 dilutions plated. To increase sensitivity, an undiluted 1-mL aliquot was inoculated into heart infusion broth w/ 1% xylose, incubated for up to 72 h and if growth noted, plated for confirmation of Y. pestis.
Serum for clinical chemistry was collected before exposure and on Days 2, 6, and 28 post challenge and analyzed using a Hitachi 911 Clinical Chemistry Analyzer (Roche Diagnostics) or a PMod Clinical Chemistry Analyzer. Whole blood for hematology was collected percutaneously from the femoral vein and transferred to a tube containing EDTA. One drop was smeared onto a glass slide for manual differential count and analyses were made using an Advia 120 (Bayer Corporation, Diagnostic Division).
Moribund or end-of-study animals were anesthetized with intramuscular 10 mg ketamine/kg body weight and euthanized by intravenous Euthasol. Tissues, including lung, tracheobronchial lymph nodes, liver, spleen, and brain, were collected for quantitative bacteriology and histopathology. Lung lobes were gently inflated with 10% neutral-buffered formalin (NBF) to approximate normal volume prior to immersion fixation. Tissues sections were fixed in NBF, cut 4–6 µm thick, mounted on slides, and stained with hematoxylin and eosin.
Summary statistics (e.g., means, standard deviation, charts, graphs, etc.) were calculated for quantitative parameters (BioSTAT Consultants, Portage, MI). Survival was the primary endpoint and was examined by Fisher's exact test. Analyses of secondary endpoints were performed as repeated measures ANOVA (SAS). For all analyses, a P value of ≤0.05 was considered to be a significant difference.
AGM were exposed to aerosolized Y. pestis CO92 in three separate cohorts. The estimated group mean (±SD) -delivered LD50 doses were 74 (31.0) in the first cohort, 124 (10.5) in the second cohort, and 22 (23.1) in the third cohort, with a range over all animals from 3–145 LD50 doses (Figure 1). All seven control animals succumbed to the challenge with Y. pestis CO92 and were moribund euthanized or died before euthanasia could be performed on Days 4 or 5 following challenge. All 16 AGMs treated with levofloxacin for 10 days survived until planned euthanasia on Day 28. One animal (Y160) began treatment on day 3 pe but was euthanized on Day 9 was after one day of vomiting and inability to retain food. Blood culture was positive for Y. pestis at initiation of treatment on Day 3 but subsequent blood cultures on Days 4–7 and tissues collected at necropsy were negative. Histopathology of the stomach revealed necrosis of the gastric epithelium and no evidence of active Y. pestis infection in other organs. While there was no evidence for treatment failure in this animal, in an intention-to-treat analysis (including Y160) the difference in survival was statistically significant at p<0.001, (Wilcoxon).
In cohort 3 seven of the eight treated animals were delivered doses of aerosolized Y. pestis less than that received by the two controls. If the animals in Cohort 3 are removed from analysis, the difference in survival between treatment and control remains significant (p<0.001).
Among the 7 controls only 5 were bacteremic prior to onset of fever and infusions of D5W with bacterial loads ranging from 1.2–4.9 log10 CFU/mL (Figure 2). All 5 controls tested more than 12 h after onset of infusons were bacteremic with bacterial loads of 2.5–5.5 log10 CFU/mL. Among the levofloxacin-treated animals, 13 of 17 (76%) had Y. pestis bacteremia detected prior to or at the onset of fever and infusions with levels ranging from 1.8 to 4.9 log10 CFU/mL. No bacteremia was detected after onset of levofloxacin infusions up through day 7 post-exposure, or a total of 60 post-treatment samples.
Lung tissues from levofloxacin-treated animals lacked detectable Y. pestis at necropsy 11–17 days after termination of antibiotic infusions. Tissue pathogen loads in untreated animals ranged from 1.5×104 to ≥3.0×105 cfu/mL and tissue load was highest in the lungs and tracheobronchial lymph nodes up to 1.5×1010 cfu/g. There was no significant relationship between onset of fever and detection of bacteremia by once-daily sampling of whole blood, nor were there significant relationships between inhalation challenge dose and onset of fever or onset of detected bacteremia (Figure 1).
The raw data were calculated as hour averages for temperature, heart rate and respiratory rate as illustrated in Figure 3. For statistical comparisons the data for each animal was calculated as change from the baseline (delta value) for that animal at that time of day. The means (standard deviations) of each treatment group for each vital sign recorded is illustrated in Figure 3 D, E and F. The delta values at 12:00 and 00:00 (midnight) for each day are displayed to account for diurnal variation. Increases above baseline values for each animal in body temperature, heart rate, and respiratory rate were apparent during the Day 2 to Day 3 interval but for both the control and treated groups of animals significant increases (delta values) were not seen until Day 3 (approximately 72 h pe). By Day 4.5 the group mean increase in all three vital signs of the control group was greater than the group mean increases over baseline in the treated group. The apparent return to baseline body temperature in the control group at day 5.5 pe was due to moribund hypothermic animals at this time. The resolution of fever, tachycardia and tachypnea in the treated group was gradual over the next 4 days until day 9 when vital signs had returned to baseline (Figure 3 G, H, I). After the end of the 10-day treatment, the African Green monkeys were observed for an additional 11 to 16 days and telemetered temperature, respiratory rate, and heart rate revealed no evidence of clinical relapse.
All pre-challenge radiographs showed no underlying disease. In all three untreated animals studied, large multilobar infiltrates correlated in location with palpable consolidation noted during necropsy. In general, however, the radiographs underestimated the extent of pneumonia found by gross pathology. In all nine animals treated with levofloxacin in Cohorts 1 and 2, chest radiographs on Day 5 revealed pulmonary infiltrates in one to four lobes. These radiographs were obtained approximately 120 h after aerosol challenge and up to 67 h after fever onset and antibiotic initiation, indicating that pneumonia was established in all treated animals (Figure 4). Pulmonary radiographs 28 days after challenge (15–17 days after termination of therapy) in five treated animals in Cohort 1 were normal without any apparent residual of previous pulmonary infection.
The levofloxacin levels measured in the first four samples taken in eight treated animals demonstrate intragroup consistency (Table 1). The peak (maximum concentration, Cmax) level after the first and second doses of 8 mg/kg body weight ranged from 2.4–4.6 µg/mL. Only 3 of the 16 infusions attained a level within two standard deviations of the target concentration of 6.2±1.0 µg/mL, yet all animals treated for 10 days survived the lethal aerosol challenge. All trough levels exceeded the MIC of the Y. pestis CO92 strain (0.03 µg/mL) [34].
Serum alanine aminotransferase (ALT), aspartate aminotransferase (AST), lactate dehydrogenase (LDH), and total bilirubin increased in control animals at euthanasia and treated animals on day 6, compared to pre-study values (data not shown, p<0.01) but there was no statistical difference in these values between the control and treated groups. Creatinine was elevated only in the control group. Treated animal values returned to pre-study values by the end of the study.
Total white blood cell count increased in the control group from 11.2±2.5×103/dL(mean ± SD) to 41.0±32.9×103/dL at euthanasia on day 5, compared to no increase in the treated group (baseline 9.5±2.4 compared to 12.0±4.4 at Day 6). Differences in neutrophil and monocyte countsbetween the two groups are consistent with a decrease in inflammation after approximately two days of antibiotic administration. The hematocrit increased in the control group (Day 5, 55±6.1) but not in the treated group (Day 6, 39.7±5.2) compared to baseline values for both groups (47.5±5.5).
In the moribund control animals multiple lobes contained extensive parenchymal hemorrhage and marked fibrinosuppurative pneumonia. Findings included enlarged tracheobronchial lymph nodes, discolored liver and spleen without enlargement, an enlarged heart in two animals, and fluid on the brain of one animal. Septal histiolymphocytic infiltrates within the pulmonary parenchyma seen in over half of the levofloxacin-treated animals 28 days after challenge are consistent with resolution of pneumonia. No evidence of abscesses or neutrophilic alveolitis was found in any levofloxacin-treated animal 28 days after challenge.
This study demonstrated the efficacy of intravenous levofloxacin treatment to prevent death from lethal pneumonic plague in AGMs. All seven animals receiving intravenous D5W died while all 16 animals completing the 20-infusion course of levofloxacin survived until Day 28 post-exposure. Demonstration of efficacy depended on three features of the study design: selection of the AGM model, initiation of treatment for severe disease but prior to irreversible disease, and selection of the appropriate antibiotic and dosing schedule.
The AGM model mimics human disease in most respects including precipitous course of disease following a brief asymptomatic anti-inflammatory phase and establishment of the primary pneumonia [22]. In spite of the absence of hemoptysis and coagulopathy, the AGM model is suitable to test the efficacy of treatment for a bioterrorism-associated disease. In this study untreated AGMs exhibited multifocal pneumonia, high-grade bacteremia, dissemination to liver and spleen, and 100% lethality from 40 to145 aerosol-LD50 doses, confirming previous results [22]. The course of disease was rapid, with telemetry-documented fever onset occurring within 53–93 h of exposure in most animals. Three AGMs dosed below 40 LD50 doses had later onsets of fever of 124, 125, and 165 h post exposure, but in the study population as a whole time to onset of fever was not significantly related to inhaled dose of pathogen. Five of the 16 survivors randomized to treatment received an aerosol challenge less than 40 LD50 doses, but even if these 5 were removed from analysis, protection was still significant. Cynomolgus macaques are also highly susceptible to inhaled Y. pestis [35], and exhibit the two-phase disease course [33]. The macaques, however, may have the drawbacks of modest febrile response in some animals [33], and variable levels of innate resistance and unexpected survival following lethal doses [36] (K E Van Zandt et al, 2010, Efficacy of cethromycin against lethal Yersinia pestis inhalation challenge in cynomolgus macaques, Abstract B-057, 4th Biodefense Research, Am Soc Microbiol, Baltimore, MD).
Treatment infusions were initiated by the appearance of fever, and thus similar to the ‘late treatment’ studied in the mouse model [17]. The increase in body temperature detected by continuously monitored telemetry and defined as >39°C in all animals was in retrospect >2°C above the diurnal background (Figure 3). Bacteremia was found in most animals at the time of onset of fever. Chest radiographs taken 1–2 days after the onset of fever documented the presence of detectable pneumonia in all animals tested (Cohorts 1 and 2), consistent with the treatment of established pneumonia. Interestingly, one animal was removed from the study post-challenge due to premature treatment approximately 8 h prior to becoming febrile. Nonetheless, a single 8-mg/kg dose of levofloxacin which did not prevent fever resulted in survival for the 28-day observation period. Nonetheless, the important question remains unanswered how late in the progression of disease will levofloxacin remain efficacious in reversing the rapid progression of pneumonic plague.
Few antibiotics, including streptomycin and tetracyclines, are approved for the treatment of plague, but these antibiotics are toxic and have limited availability. Several antibiotics for plague, including doxycycline and gentamicin, have been supported by clinical experience [37], [38]. In the AGM model of pneumonic plague, however, oral doxycycline initiated within 6 h of the onset of fever resulted in only 40% survival (H Lockman et al. Efficacy of oral doxycycline against pneumonic plague in African Green monkeys. Abstract G-098, 4th Biodefense Research Meeting, Am Soc Microbiol, Feb 22, 2010, Baltimore MD). An oral ketolide cethromycin has in vitro antimicrobial activity similar to gentamicin but in a cynomolgus macaque model of pneumonic plague the highest dose resulted in 90% survival when given only 24 h after inhalation challenge (K E Van Zandt et al, 2010, Efficacy of cethromycin against lethal Yersinia pestis inhalation challenge in cynomolgus macaques, Abstract B-057, 4th Biodefense Research, Am Soc Microbiol, Baltimore, MD). Many beta-lactam antibiotics and fluoroquinolones have significant in vitro activity against Y. pestis [39], but beta-lactam antibiotics did not have significant in vivo treatment efficacy whith late treatment in the mouse model [17]. Ciprofloxacin has been proposed as the primary treatment for mass casualties in a bioterrorism event involving Y. pestis [3]. In a post-exposure prophylaxis study following lethal aerosol challenge in mice, ciprofloxacin given 44h post-exposure was 90% effective in preventing death [40]. In the AGM ciprofloxacin showed efficacy for treatment of inhalational plague [Pitt MLM, DN Dyer, EK Leffel, et al. Ciprofloxacin treatment for established pneumonic plague in the African Green Monkey. Abstract B-576, 46thICAAC meeting, San Francisco, CA, September 28, 2006]. In a post-exposure prophylaxis study for anthrax in rhesus macaques, however, ciprofloxacin did not prevent mortality in 56% of aerosol-challenged monkeys [41], leaving open the question of multi-biothreat agent efficacy of ciprofloxacin.
This study evaluated the efficacy of levofloxacin, a fluoroquinolone antibiotic with FDA approval for a wide range of Gram positive and Gram negative infections, including severe Gram negative pneumonias. This antibiotic has broad efficacy against many select agents including Y. pestis [42] and is used in most inpatient health care facilities in the United States, making it an appropriate candidate for rapid availability in the event of a bioterrorism event. Levofloxacin should continue as a candidate for such an event even in light of recently demonstrated toxicities [43], [44], which were not evaluated in this NHP model, due to the high and rapid morbidity of primary pneumonic plague. An evaluation has occured in an in vitro pharmacodynamic infection model levofloxacin sterilized the culture without resistance selection [45]. In a mouse study of pneumonic plague, levofloxacin treatment conferred 100% survival when treatment began 24 h after aerosol exposure to 20 LD50 doses [34]. Levofloxacin achieves high concentrations in human lung tissue and alveolar macrophages with levels 2–4 times that in plasma [46], [47].
The human dose for Gram negative pneumonia is 500 mg intravenously every 24 h. The peak level in plasma following this dose is 6.2±1.0 µg/mL and an area under the curve of 48.3±5.4 µg·h/mL (Levaquin package insert), so these were our targeted levels. A previous study in rhesus macaques developed a “humanized” dosing regimen for levofloxacin, as the elimination of levofloxacin is 3-fold more rapid in nonhuman primates than humans [27]. Using data from an earlier study of levofloxacin pharmacokinetics in AGMs, the daily dose schedule used in this study was calculated to be 8 mg/kg followed by 2 mg/kg 12 h later (Blaire Osborn, unpublished data). Levofloxacin levels (Table 1) demonstrated that achieved levels were only half of the targeted Cmax, yet the dose administered was successful in curing established plague pneumonia. Nonetheless the interpretation of efficacy in this study is dependent on dosing sufficient to achieve a peak plasma dose of 2.4 µg/mL or greater, and continuous plasma levels above the minimum inhibitory concentration of the organism.
The standard of care for treatment of suspected bubonic plague instructs the inclusion of gentamicin among other broad-spectrum antibiotics [3], [37], [38]. Our results in the AGM model of pneumonic plague, the most severe of plague syndromes, suggests that levofloxacin would likely be efficacious in bubonic plague. There is no established model for bubonic plague in nonhuman primates but a clinical trial of levofloxacin could be safely undertaken for the treatment of bubonic plague.
|
10.1371/journal.pbio.2000330 | Transient Duplication-Dependent Divergence and Horizontal Transfer Underlie the Evolutionary Dynamics of Bacterial Cell–Cell Signaling | Evolutionary expansion of signaling pathway families often underlies the evolution of regulatory complexity. Expansion requires the acquisition of a novel homologous pathway and the diversification of pathway specificity. Acquisition can occur either vertically, by duplication, or through horizontal transfer, while divergence of specificity is thought to occur through a promiscuous protein intermediate. The way by which these mechanisms shape the evolution of rapidly diverging signaling families is unclear. Here, we examine this question using the highly diversified Rap-Phr cell–cell signaling system, which has undergone massive expansion in the genus Bacillus. To this end, genomic sequence analysis of >300 Bacilli genomes was combined with experimental analysis of the interaction of Rap receptors with Phr autoinducers and downstream targets. Rap-Phr expansion is shown to have occurred independently in multiple Bacillus lineages, with >80 different putative rap-phr alleles evolving in the Bacillius subtilis group alone. The specificity of many rap-phr alleles and the rapid gain and loss of Rap targets are experimentally demonstrated. Strikingly, both horizontal and vertical processes were shown to participate in this expansion, each with a distinct role. Horizontal gene transfer governs the acquisition of already diverged rap-phr alleles, while intralocus duplication and divergence of the phr gene create the promiscuous intermediate required for the divergence of Rap-Phr specificity. Our results suggest a novel role for transient gene duplication and divergence during evolutionary shifts in specificity.
| Many molecular pathways are found multiple times in a given organism, where they are often reutilized for different functions. Such expansion of a family of pathways requires two main evolutionary processes—acquisition of additional copies of the pathway's genes and divergence of interaction specificity to prevent cross-talk between pathways while preserving interactions within each copy of the pathway. In bacteria, acquisition can occur horizontally, by transfer between different lineages, or vertically, by duplication within the lineage. Interaction specificity is thought to diverge through a promiscuous intermediate component that prevents loss of interaction during the process. In this work, we study the mechanisms underlying the extreme expansion of the Rap-Phr cell–cell signaling family in the Bacillus genus. Specificity of Rap-Phr interaction is critical for guiding preferential action towards kin. We find that horizontal transfer and not duplication guides the acquisition of an already divergent Rap-Phr variant. Surprisingly, duplication still has a key role during expansion, as duplication and subsequent divergence of the signaling molecule gene provide the promiscuous intermediate state needed for divergence of specificity. We therefore identify two complementary roles for horizontal and vertical processes in the evolution of social bacterial pathways.
| The evolution of signaling complexity often occurs by diversification and repeated utilization of signal transduction pathways [1–6]. This generally requires two processes: the acquisition of homologous copies of the pathway's components and the co-diversification of interacting components to ensure specificity of interaction within a pathway while avoiding cross-talk between pathways (Fig 1A) [7,8]. Bacteria have a multitude of signal transduction pathways, which have undergone evolutionary expansion and divergence of specificity, such as two-component systems [6,8–10], antisigma-sigma factors [11], and toxin-antitoxin systems [12,13]. The large number of available bacterial genomes allows for high-resolution analysis of evolutionary expansion, rendering bacterial signal transduction a favorable model system for studying diversification.
While eukaryotes can only acquire paralogous genes through duplications, bacteria can acquire them either by gene duplication (Fig 1C, right) or by horizontal transfer (Fig 1C, left) [14]. Previous works on the prevalence of these two processes in the acquisition of bacterial two-component signal transduction pathways have indicated that it is dominated by gene duplication, but it is also affected by horizontal transfer [15–17]. However, the coarse-grained resolution of these studies prevents the distinction between vertical acquisition and horizontal transfer between closely related strains [18,19].
The second requirement for paralogous expansion is the divergence of interaction specificity between pathways (Fig 1B). This is generally thought to evolve using a promiscuous form of one of the interacting components, which can interact with both variants of its partner (Fig 1B, top) [2,4,12,20–22]. The promiscuous form can be the ancestral state, subsequently evolving into two states of different specificity [4,20], or it can be an evolutionary intermediate between the ancestral specific state to a novel state [22,23]. The ability to distinguish between these two diversification scenarios typically depends on our capacity to infer and analyze the ancestral state from phylogenetic data [2,20,21]. A recent work used deep mutational scanning to show the abundance of promiscuous bacterial intermediates in the evolution of a bacterial toxin-antitoxin family [12]. This approach, however, cannot distinguish whether the promiscuous form is ancestral or intermediate or determine the evolutionary relevance of the identified diversifying trajectories. The modes by which rapidly diversifying signaling families expand are therefore still unclear.
The Rap-Phr cell–cell signaling system of Bacilli can serve as a model system to study bacterial modes of expansion and diversification [24–26]. The cytoplasmic Rap receptor can bind, and sometimes dephosphorylate, its target, leading to inhibition of target activity [27,28]. The cognate phr gene codes for a pre-polypeptide, which undergoes multiple cleavage events during its secretion, resulting in the release of a mature penta- or hexa-peptide Phr autoinducer [25,29–31]. The mature Phr peptide is transported into the cytoplasm through the oligopeptide permease system, where it can interact with Rap receptors [26], subsequently leading to major conformational changes in the Rap protein and preventing Rap from repressing its target [27,28,32,33].
Rap-Phr systems have mostly been studied in the B. subtilis 168 lab strain. This strain encodes for eight paralogous rap-phr loci, each coding for a different Phr autoinducer. In addition, it encodes for three orphan rap genes that lack a cognate phr locus [33–35]. Despite the genomic expansion of paralogous Rap-Phr systems, they all have the same overall structural organization and most have a redundant function in repressing either Spo0F or ComA, two key response regulators of the Bacillus stress response network [31,36–40]. We recently demonstrated how social selection can explain the acquisition of additional Rap-Phr systems, despite their redundant regulation of the same target [41]. Some rap-phr loci are encoded by mobile genetic elements [31,39,42–47], and while many mobile-element-associated Rap systems maintain their repressive effect on Spo0F or ComA, some also play a direct role in controlling the mobility of their associated mobile genetic elements [44,47].
To study the expansion and diversification of the Rap-Phr family, we combined computational mining of available Bacillus genomes and experimental characterization of the target and autoinducer specificity of multiple Rap-Phr systems. We found that at the organismal level, acquisition of a novel Rap-Phr paralogous system occurred by horizontal gene transfer (Fig 1C, left). At the locus level, diversification of Rap-Phr specificity was facilitated by phr gene duplication or intragenic duplication of the Phr autoinducer coding sequence, followed by diversification of the autoinducer sequence. We show that the diverged duplicated phr form can serve as a promiscuous intermediate between two states of specificity (Fig 1B, bottom). Therefore, the extreme diversity of the Rap-Phr system results from a combination of horizontal and vertical processes operating at two different levels of genetic organization.
To understand the extent to which Rap-Phr systems are prevalent in the Bacillus genus, we downloaded 413 whole-genome sequences of strains from this genus from the National Center for Biotechnology Information (NCBI) (S1 Data). The species association of these genomes is heavily biased towards genomes from the B. subtilis (127 strains) and B. cereus (216 strains) groups of species (S1 Fig). The conserved Rap structure, which includes a three-helix N-terminal and tetra-tricopeptide repeat C-terminal domain [24,32,33], allowed us to search for Rap homologs in all the genomes using the basic local alignment search tool for translated DNA (BLAST tblastn program). Following identification of Rap homologs, we searched for candidate phr genes, relying on the known organization of annotated phr genes (S2 Fig)—short open reading frames with a secretion signal sequence, located immediately downstream of the Rap gene in the same direction (see Methods) [24].
Our final database contained ~2,700 functional rap genes (S1 and S2 Data). rap genes were identified in all strains of the B. subtilis and B. cereus groups as well as in two evolutionarily distinct Bacillus species—B. halodurans and B. clausii (S1 Fig). Notably, all strains harboring at least one rap gene encoded for multiple rap paralogs (Fig 2A, S5 Data). The number of rap genes differed between groups, averaging 11 ± 2 (mean ± standard deviation [st.dev.]) in the B. subtilis group and 6 ± 3 in the B. cereus group. In B. subtilis 168, there are three orphan rap genes not accompanied by an adjoining phr gene, a scenario typical for B. subtilis strains, which have an average 2.7 ± 1 orphan raps per strain (mean ± st.dev.). In contrast, almost all (>95%) of the rap genes from the B. cereus group had an adjoining putative phr gene.
We next performed a phylogenetic analysis of the ensemble of rap genes (see Methods for details of the phylogenetic analysis and S3 and S4 Data). Although the overall divergence of Rap homologs was large, the family was clearly divided into two groups, corresponding to the division between the B. cereus and B. subtilis groups of species (Fig 2B). This suggests that the divergence of rap genes in each of these groups occurred after their evolutionary separation, with no horizontal transfer between groups. The phylogeny of the B. clausii and B. halodurans Rap proteins suggests that they acquired their rap genes by one or two horizontal gene transfer events, respectively, from B. subtilis group isolates, followed by intraspecific diversification and accumulation, as seen in the two major groups (Fig 2B). No evidence of divergence of the Rap protein through recombination of different Rap homologs was found (Methods). These observations suggest that diversification and paralogous expansion occurred independently multiple times during the evolution of the Rap-Phr system.
To gain further insight into the population genetics underlying the diversification of the Rap-Phr family, we focused on the B. subtilis group, in which multiple Rap-Phr systems have been previously characterized. The known Phr autoinducer sequences and the patterns of phr sequence conservation along the Rap phylogenetic tree were used to identify putative penta or hexa-peptide autoinducers and to cluster the Rap proteins (Fig 2C). All together, we defined 102 clusters with 81 unique Phr autoinducer peptides. To the best of our knowledge, this extreme autoinducer diversity is much greater than that observed in any other family of quorum-sensing systems.
In order to identify the mode of acquisition of a novel Rap-Phr system into a genome, we analyzed the level of horizontal gene transfer of Rap-Phr systems. We used two independent measures to estimate this trait—guanine-cytosine (GC)-content analysis and abundance analysis. First, because mobile element-related genes in B. subtilis typically have a significantly lower GC-content as compared to the rest of the genome [48], we characterized Rap-Phr as mobile if their GC-content was significantly lower than the average GC-content of their respective strain (Methods, S3 Fig). We found that 75% of Rap-Phr clusters were mobile (Fig 2C, bottom right). In parallel, mobile (or accessory) genes can be identified by their intermittent appearance within strains of a given species. Thus, we constructed an association matrix, in whcih each Rap cluster was marked as either present or absent in each of the genomes of B. subtilis group isolates (S4 Fig). With few exceptions, raps identified as core genes by their GC-content appeared in the great majority of isolates from a given species, whereas mobile Rap systems appeared in only a few strains (S5 Fig, S5 Data), demonstrating a good correlation between the two measures of mobility. These results suggest that the acquisition of a novel Rap-Phr into a genome is dominated by horizontal gene transfer. To determine whether duplication, as the alternative mode of acquisition, occurred as well, we searched for cases in which two Rap-Phr systems from the same cluster were coded in the same genome (S4 Fig). Only five such cases were identified, three of which occurred in clusters that were categorized as mobile by both criteria above and may result from a recent duplicated introduction of a single mobile element. Therefore, there is little evidence for direct duplication events in Rap-Phr that belong to the “core” genome.
We noted that different species in the B. subtilis group have different core Rap paralogs (S4 Fig). The observed diversity pattern fits a slow ongoing process of fixation of Rap-Phr systems, in which some Rap variants (e.g., RapA and RapC) are fixed in multiple related species, while others are fixed only in a single species within the group. Interestingly, all orphan Rap systems belonged to the core group, by both modes of assessment, with RapD present in all but one species (B. licheniformis). Although mobile Rap-Phr systems dominated the population diversity, most Rap paralogs in any given strain (60% ± 10% mean ± st.dev.) belonged to the core group (Fig 2C, top right). Despite the abundance of available genomic data, we do not observe a saturation of Rap-Phr diversity with strain number (S6 Fig). The large diversity is also evident from the fact that several rap-phr clusters were found only on separately sequenced plasmids [49–52].
Current data indicate that despite the large sequence and autoinducer diversity, many of the Rap receptors target either Spo0F, ComA, or both. The evolutionary rate of target specificity shift is unclear. To experimentally determine the specificity of multiple Rap systems, we used two isolates (marked in S4 Fig), B. amyloliquefaciens FZB42 and B. licheniformis ATCC 14580, as templates for the cloning of ten novel rap genes, most of them core genes of their respective species (Fig 3 and S4 Fig, S1 Table). These newly cloned Rap genes belong to previously unexplored branches of the Rap phylogeny. Correspondingly, these Rap homologs almost double the phylogenetic diversity [53] of characterized Rap proteins (from ~6% to ~10% of the total phylogenetic diversity of Rap proteins in the B. subtilis group). In addition, seven B. subtilis-associated rap genes (rapC, F, I, P, J, B, and D) were cloned under the control of the inducible hyper-spank promoter [54]. All genes were introduced into the genome of strain PY79. To prevent interference by the endogenous Phr product, the rapC, F genes were introduced into a strain with a deletion of these two systems (S7 Fig, S5 Data). Finally, an available phrA deletion was used to assay the effect of RapA.
The different Rap proteins were assayed for their effects on the Spo0A and ComA pathways by measuring their impact on sporulation efficiency and expression of the ComA-regulated srfA promoter using a yellow fluorescent protein (YFP) reporter [43], respectively (Fig 3A, S5 Data, Methods). Because ComA activity is indirectly affected by the Spo0A pathway [56], we introduced a spo0A deletion into each of the YFP-reporting strains.
Five out of the ten novel Rap proteins affected both sporulation efficiency and srfA expression, while RapBL4 affected only sporulation. Four of the novel rap overexpression constructs did not strongly affect either pathway. The eight B. subtilis-associated Rap proteins had the expected, previously characterized effect, with RapA, I, B, and J affecting Spo0A [26,33], RapF, C, and D affecting ComA [34,37], and RapP affecting both pathways [43].
These data allow us to better estimate the rate at which target choices change along the evolution of the Rap lineage. Based on our results and those reported by others, we assembled a phylogenetic tree of 25 Rap variants whose targets have been at least partially characterized (Fig 3B). We used the GLOOME program [55] to estimate the rate of gains or losses of regulation of the Spo0A and ComA pathways. We found the most parsimonious switching model to include 12 gain and loss events (Fig 3B), starting with an ancestral strain that regulated spo0A activity. This ancestral strain acquired the ability to control ComA in multiple independent events (see Discussion).
The interactions between Rap proteins with the aforementioned targets were recently analyzed at the structural level, allowing for the identification of specific Rap residues that directly interact with each target [27,28]. Upon analysis of the conservation of these residues in the characterized Rap proteins according to their functional targets (S8 Fig), we found that the amino acid residues, where RapF interacts with its target ComA, were not conserved in many other ComA-interacting Rap proteins. In contrast, the RapH amino acid residues at the interface with Spo0F were highly conserved in characterized Rap proteins, irrespective of whether they regulate the Spo0A pathway or not. The high level of conservation of Spo0F-interacting residues and low level of conservation of ComA-interacting residues were also demonstrated upon analysis of all B. subtilis group-associated Rap proteins using the ConSurf program (S9 Fig) [57]. These results further support the ancestral origin of the interaction between Rap proteins and Spo0F and the independent gain of ComA interaction by multiple sub-lineages of Rap proteins, as suggested by the parsimony analysis (Fig 3B).
Anecdotal experiments in strain 168 have indicated that divergent Rap-Phr pairs are orthogonal—a receptor from one Rap-Phr strain will predominantly respond only to its cognate autoinducer [33,35,38,41]. This notion has not been studied systematically. It is also unclear whether divergent Rap-Phr systems encoded on different chromosomes would maintain orthogonality. We took advantage of the large collection of inducible Rap systems to thoroughly analyze these points. Fourteen custom-made putative autoinducer peptides were assayed for their ability to restore gene expression in the presence of different inducible Raps (Fig 3C). We used a peptide concentration of 10 μM, a level that exceeds both the measured affinity of Phr peptides to cognate Raps and the physiological levels of Phr (Methods) [25,30,43]. The interactions between peptides and Rap proteins were monitored using either the PsrfA-YFP or PspoIIG-YFP reporter constructs, depending on whether the Rap targets the ComA or Spo0A pathways, respectively (Fig 3C, S5 Data). Rap proteins that affect both pathways were assayed only once.
We found that the repressive effect of all Rap proteins on gene expression was alleviated by addition of saturating amount of their respective cognate Phr peptide. One exception to this rule was RapBL4, which did not interact with its putative Phr pentapeptide (with amino-acid sequence GRAIF). We also found that the orphan RapB, J proteins were not affected by any Phr but that a 10-fold higher concentration of PhrC did activate them, in accordance with previous works that suggested this weak interaction (Fig 3C) [33,35].
We observed strict maintenance of orthogonality of Rap-Phr systems residing on the same genome. In two cases, cross-talk between two systems encoded by different strains was detected, with RapBL5 responding to both its cognate PhrBL5 autoinducer and to the related PhrBA1 autoinducer, and RapBA2 responding to its cognate PhrBA2 and more weakly to PhrBL6. Notably, the RapBL5 and RapBA1 proteins were only weakly divergent (Fig 3B), but RapBA2 and RapBL6 were unrelated. Our data therefore support the notion of strong orthogonality of Rap-Phr systems encoded in the same genome and some orthogonality between divergent Rap-Phr systems encoded by different genomes.
Thus far, our results suggest that the Phr peptides coevolve with their cognate Rap receptors to maintain the specificity of interaction, while Rap receptor affinity to its main two targets can change. To understand the evolution of the Phr peptides, we studied the diversity of the phr sequences in further detail. All Rap-Phr systems analyzed to date have a single phr gene, coding for a single Phr autoinducer penta- or hexa-peptide. In contrast, we found multiple cases in our database where peptide autoinducer coding regions were duplicated (Fig 4, S10 Fig). In some cases, all putative autoinducer repeats were identical (Fig 4A and S10E Fig), while in others, a single phr gene coded for multiple similar, but nonidentical, putative peptide autoinducers (Fig 4B and S10A–S10D Fig). For example, the Phr prepeptides of a group of closely homologous Rap proteins (related to RapH) contain multiple varying repeats of the motif [S/I][D/I/N/Y]RNT[T/I] (S10B Fig).
We also identified two subclusters of the B. subtilis Rap-Phr systems, in which the entire phr gene had undergone a duplication event (S10C and S10D Fig). The putative peptide autoinducers of the two phr genes had also diverged. We also observed sequence duplications (either intragenic or full-gene) events in rap-phr loci of other Bacilli (S11A–S11C Fig). A similar analysis of the related NprR-NprX quorum-sensing family [58,59] showed duplications in some nprX genes as well (S11D Fig, Methods). These results indicate that autoinducer duplications are abundant and that putative autoinducer sequences diverge after duplication.
Autoinducer duplication may facilitate the coevolution of Rap-Phr pairs by allowing a duplicated and diverged phr gene to serve as a transient promiscuous intermediate (Fig 1B). To experimentally examine this possibility, we analyzed a subset of Rap-Phr systems of the RapK/RapG/RapBL3 cluster (Fig 4B) [37]. The phr gene of three closely related systems in this cluster encodes for a pre-peptide with two putative autoinducer peptides. One system, designated RapK2-RR, encodes twice for the putative penta- or hexapeptide ERPVG(T). The second system (RapK2-KR) encodes for the putative autoinducers ERPVG(T) and EKPVG(T), while the third (RapK2-KK) encodes twice for the putative autoinducer EKPVG(T). RapK2 variants (S1 Table) were cloned under the control of the hyper-spank promoter and monitored for the effect of their overexpression on a PspoIIG-YFP reporter, as described above (Fig 4C, S5 Data). spoIIG promoter activity was repressed to background levels in all three overexpression strains. spoIIG promoter activity was restored in strains overexpressing either RapK2-RR or RapK2-KK upon addition of their cognate hexapeptides (ERPVGT and EKPVGT, respectively), but not when their non-cognate hexapeptide was added (Fig 4C). RapK2-KR, whose cognate Phr encodes both type of peptides, responded only to the addition of ERPVGT. None of the strains responded to the addition of the relevant pentapeptides. The specificity shift between the different RapK2 variants may therefore provide an example for the role of duplication in such an event.
The diversity and functional orthogonality of diverging signal transduction paralogs in bacteria are well characterized [9,11,12]. However, the way by which new paralogs are acquired and diversified and, specifically, the roles of horizontal and vertical events in this process are not clear. In this work, we showed that both horizontal and vertical processes are crucial for the expansion of the Rap-Phr quorum-sensing system but operate at different levels of organization. Acquisition of a novel Rap-Phr system was shown to be facilitated by horizontal gene transfer, while diversification was facilitated by phr duplications within the diversifying locus (Fig 5).
A key novelty of our finding is exposure of the role of duplications in the divergence of Rap-Phr system specificity. The prevalence of intragenic and whole-gene phr duplications and the experimental analysis of the RapK2-PhrK2 variants (Fig 4 and S10 and S11 Figs) suggest that transient Phr duplication and divergence play a role during evolutionary shifts in Rap-Phr specificity (Fig 4D). The ancestral copy of Phr interacts with the ancestral Rap form, while the duplicated and diverged autoinducer copy has the potential to interact with a coevolved receptor. Importantly, this mechanism differs in two aspects from the common views of pathway diversification. First, divergence of a signaling pathway is typically linked with a promiscuous protein that can interact with the two forms of its partner [2,12]. Here, the promiscuous form is the duplicated phr and not a single Phr peptide, which interacts with both Rap variants. Second, duplications are typically only considered important if both diverged duplicates survive over evolutionary timescales [3]. In contrast, Phr duplication and divergence are evolutionarily crucial for a specificity shift, but to complete the shift it has to be transient—with either divergence or duplication itself being lost. In the specific case we examined, the duplication persisted, while duplicate diversity was transient (Fig 4D).
We found that Rap acquisition and divergence occurred independently in multiple evolutionary lineages, indicating that it is an intrinsic feature of the function of this system. In the B. subtilis group alone, we identified dozens of putative Phr autoinducer peptides, rendering it the largest known quorum-sensing family. While the high orthogonality of a significant number of pairs was experimentally verified (Fig 3B), further experimental work will be required to explore the level of orthogonality between all clusters. In fact, some cases of cross-interactions were detected (Fig 3C). Notably, strong cross-interactions between Rap-Phr pairs encoded in the same genome were not observed, while weak interactions, as seen between PhrC and the orphans RapB and J (Fig 3C) [33,35], or RapF [41], were noted. In general, the functional importance of nonspecific interactions is unclear. More specifically, the interaction between orphan Raps and non-cognate Phrs may be physiologically irrelevant, given the low affinity of these interactions (~100 μM) compared with the physiological concentration of Phrs (~100 nM) [25]. Whether orphan Raps interact with other signaling molecules remains to be determined. Notably, formation of orphan Raps is rare, and all major B. subtilis orphan Raps are anciently fixed in their genome (Fig 2C and S4 Fig).
Rap proteins also diverge with regards to the targets they regulate. Our data indicate that the distinction between targets is not the result of ancient diversification but rather is the result of an ongoing process of multiple events of gain and loss of target regulation (Fig 3A and 3B). The GLOOME parsimony analysis indicated that the ancestral Rap receptor regulated the Spo0A pathway and that the regulation of ComA by Rap proteins has been gained multiple times within the B. subtilis group. This is in agreement with the low sequence conservation between different ComA-regulating Rap variants (S7 and S8 Figs) and the absence of ComA homologs in the majority of B. cereus strains. In addition, the mechanism of ComA regulation differs across Raps. RapF blocks comA binding to DNA through direct competition for the DNA binding domain [27]. In contrast, Rap60 does not block ComA DNA binding but prevents the DNA-bound ComA from activating transcription [46].
Phr diversification is accompanied by coevolution of the Rap receptor. A key future challenge is to identify the amino-acid positions of Rap that are crucial for its coevolution and the underlying structural principles of peptide-receptor specificity. Our results suggest that even key conserved features of this interaction can be lost during diversification. Specifically, arginine at the second position of the Phr autoinducer is highly conserved due to a salt-bridge with a conserved negatively charged residue on the Rap protein (Fig 2) [32,33]. While substitution to lysine (as in PhrG or PhrK2-KK) does not dramatically interfere with this interaction, this residue is substituted with the uncharged leucine in PhrBL3. Correspondingly, the conserved Rap aspartate residue is substituted by the non-charged glutamine residue in RapBL3, suggesting that the electrostatic interaction has been replaced by another type.
The mobile nature of the majority of Rap-Phr systems (Fig 2C and S3 and S4 Figs) and their functional adaptive role in the transfer of mobile elements [47,52] and in social interactions [41] indicate that they act as “kind-discrimination” systems [62]. Such systems mediate discriminative interactions between bacteria or between their genetic parasites. Kind-discrimination can operate through various mechanisms such as cell–cell signaling [60], intracellular toxin-antitoxin [63], aggregation [64,65], surface exclusion [66], bacteriocin-immunity [67], or contact-mediated toxins [68,69]. Most kind-discrimination systems are two-gene systems with a high divergence of specificity between interacting pairs. Like the Rap-Phr system, some of these systems tend to accumulate in large numbers within bacteria [70]. Notably, the social nature of kind-discrimination implies that alleles may strongly interact even if they are not encoded in the same bacterium. This presumably increases selection pressure for specificity (Fig 5).
Intralocus duplication events can yield a transient promiscuous intermediate in other kind-discrimination systems. A recent analysis showed that toxin-antitoxin systems can shift specificity through promiscuous toxin intermediates, which can mediate interactions with two different antitoxins [12]. However, this does not rule out the possibility of duplication as an alternative mechanism. Further analysis of natural variation will be required to further assess these phenomena. One difference between Rap-Phr and other kind-discrimination systems is the short length of both the phr gene and the mature autoinducer peptide. This may promote duplication and neofunctionalization in peptide-based quorum-sensing systems in comparison to other systems where both interacting partners are larger globular proteins. Interestingly, fungi mating pheromones show a striking similarity to the Phr diversity, with both genic and intragenic duplications of the pheromone peptide coding sequences as well as some cases where there is sequence variability between duplicates [71]. Duplications may arise by sexual selection [72], and it has been suggested, but not proven, that they may facilitate diversification.
Taken together, the data from this work and previous works [41,47] show how duplication and rapid horizontal transfer can work together to rapidly expand a signaling pathway family operating at multiple levels of selection. Further work will be needed to determine the generality of this phenomenon.
|
10.1371/journal.pntd.0006121 | Increased activated memory B-cells in the peripheral blood of patients with erythema nodosum leprosum reactions | B-cells, in addition to antibody secretion, have emerged increasingly as effector and immunoregulatory cells in several chronic inflammatory diseases. Although Erythema Nodosum Leprosum (ENL) is an inflammatory complication of leprosy, the role of B- cell subsets has never been studied in this patient group. Therefore, it would be interesting to examine the contribution of B-cells in the pathogenesis of ENL. A case-control study design was used to recruit 30 untreated patients with ENL and 30 non-reactional lepromatous leprosy (LL) patient controls at ALERT Hospital, Ethiopia. Peripheral blood samples were obtained before, during and after treatment from each patient. Peripheral blood mononuclear cells (PBMCs) were isolated and used for immunophenotyping of B- cell subsets by flow cytometry. The kinetics of B-cells in patients with ENL before, during and after Prednisolone treatment of ENL was compared with LL patient controls as well as within ENL group. Total B-cells, mature B-cells and resting memory B-cells were not significantly different between patients with ENL reactions and LL controls before treatment. Interestingly, while the percentage of naive B-cells was significantly lower in untreated ENL patients than in LL patient controls, the percentage of activated memory B-cells was significantly higher in these untreated ENL patients than in LL controls. On the other hand, the percentage of tissue-like memory B-cells was considerably low in untreated ENL patients compared to LL controls. It appears that the lower frequency of tissue-like memory B-cells in untreated ENL could promote the B-cell/T-cell interaction in these patients through downregulation of inhibitory molecules unlike in LL patients. Conversely, the increased production of activated memory B-cells in ENL patients could imply the scale up of immune activation through antigen presentation to T-cells. However, the generation and differential function of these memory B-cells need further investigation. The finding of increased percentage of activated memory B-cells in untreated patients with ENL reactions suggests the association of these cells with the ENL pathology. The mechanism by which inflammatory reactions like ENL affecting these memory cells and contributing to the disease pathology is an interesting area to be explored for and could lead to the development of novel and highly efficacious drug for ENL treatment.
| Some leprosy patients develop reactions which cause a significant morbidity and mortality in leprosy patients. There are two types of leprosy reactions, type 1 and type 2 reactions. Type 2 or Erythema nodosum leprosum (ENL) is an immune-mediated inflammatory complication of leprosy which occurs in lepromatous and borderline lepromatous leprosy patients. The exact cause of ENL is unknown. Immune-complexes and T-cells are suggested as the aetiology of ENL. However, the contribution of B-cells in ENL reactions has never been addressed. In the present study we described the role of B-cell subsets in ENL reaction and compared with non reactional LL patient controls before, during and after corticosteroids treatment. We found increased antigen experienced and activated B-cells in untreated ENL patients compared to those without the reaction (LL patients). This implies that B-cells are associated with ENL pathology. Therefore, the finding provides a ground for future research targeting B-cells to develop effective drug for ENL treatment.
| B-cells enable the antigen-specific humoral immunity by forming highly specific antibodies during primary immune response. B-cells within the lymphoid tissue of the body such as bone marrow, spleen and lymph nodes, are stimulated by antigenic substances to proliferate and transform into plasma cells and the plasma cells in turn produce immunoglobulins which bind to cognate antigen [1]. Although B-cells are traditionally known as precursors for antibody-secreting plasma cells, they may also act as antigen-presenting cells (APC) and play an important role in the initiation and regulation of T and B cell responses [1, 2]. However, B-cells may also involve in disease pathology especially in autoimmune disorders. The pathogenic roles of B-cells in autoimmune diseases occur through several mechanistic pathways that include autoantibodies, immune-complexes, dendritic and T-cell activation, cytokine synthesis, chemokine-mediated functions, and ectopic neolymphogenesis [2].
Memory B-cells are B-cell sub-types that are formed within the germinal centres following primary infection and are important in generating an accelerated and more robust antibody-mediated immune response in the case of re-infection also known as a secondary immune response. Recent advances in tracking antigen-experienced memory B-cells have shown the existence of different classes of memory B-cells that have considerable functional differences. Currently there are three types of memory B-cells: resting, activated and tissue like memory B-cells, [3]. Activated memory B-cells have been shown to function as effective antigen presenting cells (APCs) to naive T-cells [4]. Tissue-like memory B-cells (TLM) expressed patterns of homing and inhibitory receptors similar to those described for antigen-specific T-cell exhaustion. Tissue like memory B-cells proliferate poorly in response to B-cell stimuli, which is consistent with high-level expression of multiple inhibitory receptors. Higher percentage of TLM has been reported in immunosuppressive diseases such as HIV [5, 6].
Leprosy is a spectrum disease with the polar tuberculiod (TT) and lepromatous (LL) forms and the three borderlines forms including borderline tuberculoid (BT), mid borderline (BB) and borderline lepromatous (BL) [7]. TT characterized by strong cell-mediated immune response which restricts the spread of M. leprae while the LL forms are characterized by lack of cell mediated immune response which allows the growth and spread of M. leprae in these patients [8]. Studies have shown that circulating high levels of antibodies to M.leprae specific antigens in LL patients although these antibodies are unable to control the growth and the spread of M.leprae [9].
The study of the humoral immunity in leprosy has largely been restricted to antibodies. Patients towards lepromatous leprosy (LL) pole of the spectrum have higher antibody concentration as compared with the tuberculoid (TT) pole. Elevation of the polyclonal isotypes of these classes of antibody types with the highest concentration has reported in patients with LL forms compared to the other clinical types of the spectrum [10–13]. However, the role of these antibodies in the pathogenesis of leprosy is poorly understood.
Although the in-situ presences of plasma cells and B-cells have been reported in leprosy, the role of these cells in the pathology of leprosy lesions is unclear. Both Plasma cells and B-cells have been detected in tuberculoid and lepromatous leprosy lesions [14]. It was speculated that these lesional B-cells could influence T-cell responses and /or play a role in maintaining the inflammatory reaction in leprosy partly through the local secretion of antibodies. However, data supporting such hypothesis are lacking. It is generally thought that antibodies against M. leprae components do not play a significant role in protection against leprosy. However, antibodies may play a role in the uptake of M. leprae by mononuclear phagocytes and hence the pathogenesis of the diseases [15].
There are two types of leprosy reaction, type one and erythema nodosum leprosum (ENL) reactions. ENL is an immune-mediated inflammatory complication affecting about 50% of patients with lepromatous leprosy (LL) and 10% of borderline lepromatous (BL) patients [16–18]. ENL can occur before, during or after successful completion of multi-drug therapy (MDT). The onset of ENL is acute, but it may pass into a chronic phase and can be recurrent [19].
B -cells are the least studied immune cells in leprosy in general and leprosy reactions in particular. An increased percentage and absolute count of B-cells in the sera from patients with ENL has been reported [20], but normal numbers of circulating B-cells have also been reported [21]. A study looking at T-cell phenotypes in ENL lesions showed a normal proportion of B-cells in these patients [22]. In a prospective cohort study of 13 untreated patients with acute ENL reaction, polyclonal IgG1 antibody synthesis was elevated compared to patients with stable lepromatous leprosy and decreased after the disease had subsided. However, the concentration of polyclonal IgG2 had revealed the reverse trend: decreased before treatment and increased after treatment [23]. These authors also investigated the frequency of antibody secreting B-cells in the blood compartment of these patients with the Enzyme-Linked ImmunoSpot (ELISPOT) and found that the decrease in M. leprae specific IgG1 antibody was not related to the down-regulation of B-cell responses.
In addition to antibody secretion, B-cells have emerged increasingly as both effector and immunoregulatory cells in several chronic inflammatory diseases [24]. The role of B-cells in the pathogenesis of autoimmune disorders such as rheumatoid arthritis (RA) and systemic lupus erythematosus (SLE) is now being re-examined [1]. It would therefore be interesting to examine the contribution of B-cells in the pathogenesis of ENL.
Informed written consent for blood and skin biopsies were obtained from patients following approval of the study by the Institutional Ethical Committee of London School of Hygiene and Tropical Medicine, UK, (#6391), AHRI/ALERT Ethics Review Committee, Ethiopia (P032/12) and the National Research Ethics Review Committee, Ethiopia (#310/450/06). Under 18 years old patients were excluded from the study. Vulnerable and minor groups were also excluded from the study. All patient data analyzed and reported anonymously.
A case-control study with follow-up after the initiation of prednisolone treatment was used to recruit 30 untreated patients with ENL reactions and 30 non-reactional LL patient controls between December 2014 and January 2016 at ALERT Hospital, Ethiopia.
All patients recruited into this study were attending the ALERT Hospital, Addis Ababa, Ethiopia. The patients were classified clinically and histologically on the leprosy spectrum based on the Ridley-Jopling (RG) classification schemes [7]. ENL was clinically diagnosed when a patient with BL or LL leprosy had painful crops of tender cutaneous erythematous skin lesions [17]. New ENL was defined as the occurrence of ENL for the first time in a patient with LL or BL. Lepromatous leprosy was clinically diagnosed when a patient had widely disseminated nodular lesions with ill-defined borders and BI above 2 [19]. Patients with ENL were treated according to the World Health Organization (WHO) treatment guideline with steroids that initially consisted of 40mg oral prednisolone daily and the dose was tapered by 5mg every fortnight for 24 weeks. All patients were received WHO-recommended leprosy multidrug treatment (MDT).
Twenty micro-litter of venous blood was collected into sterile BD heparinised vacutainer tubes (BD, Franklin, Lakes, NJ, USA) before treatment, during treatment on week 12 and after treatment on week 24 from each patient and used for PBMC isolation. PBMCs were separated by density gradient centrifugation at 800g for 25 min on Ficoll-Hypaque (Histopaque, Sigma Aldrich, UK) as described earlier [25]. Cells were washed three times in sterile 1x phosphate buffered saline (PBS, Sigma Aldrich, UK) and re-suspended with 1mL of Roswell Park Memorial Institute (RPMI medium 1640 (1x) + GlutaMAX+ Pen-Strip GBICO, Life technologies, UK). Cell viability was determined by 0.4% sterile Trypan Blue solution (Sigma Aldrich, UK) ranged from 94–98%. PBMC freezing was performed using a cold freshly prepared freezing medium composed of 20% Foetal Bovine Serum (FBS, heat inactivated, endotoxin tested ≤5 EU/ml, GIBCO Life technologies, UK), 20% dimethyl sulphoxide (DMSO) in RPMI medium 1640 (1x). Cells were kept at -80°C for 2–3 days and transferred to liquid nitrogen until use. Cell thawing was done as described [26]. The procedure is briefly described as: cells were incubated in a water bath (37°C) for 30 to 40 seconds until thawed half way and re-suspended in 10% FBS in RPMI medium 1640 (1x) (37°C) containing 1/10,000 benzonase until completely thawed, washed 2 times (5 minutes each) and counted. The percentage viability obtained was above 90%.
Cells were harvested, transferred to round bottomed FACS tubes (Falcon, BD, UK) and washed twice at 400g for 5 minutes at room temperature. The cells were resuspended in 50μl of PBS and incubated in 1ml of 10% human AB serum (Sigma Aldrich, UK) for 10 minutes in the dark at room temperature to block nonspecific Fc-mediated interactions, and centrifuged at 400g for 5 minutes. After resuspending cells in 50μL PBS buffer, Life/dead staining was performed at a concentration of 1μl /1mL live/dead stain (V500 Aqua, Invitrogen, Life technologies, UK) for 15 minutes at 4°C in the dark. Cells were washed once and stained for surface markers directed against CD10 (FITC), CD19 (PerCp-Cy5.5), CD27 (V500), CD21 (V450) and Isotype control (IgG1) (all from BD Biosciences, UK), Live/dead (eFluoro 780, Invitrogen, Life technologies, UK). A single-stained OneComb eBeads (affymetrix, eBioscience, UK) for all fluorescence compensation except for the live dead stain were used. For the viability dye, cells rather than beads were stained and used for fluorescence compensation.
Forward scatter height (FSC-H) versus Forward scatter area (FSC-A) plots were used to select singlets, and FSC-A versus dead cell marker plots identified viable cells. Side scatter area (SSC-A) versus FSC-A plots were used to discriminate lymphocytes from monocytes and residual granulocytes. The threshold for FSC was set to 5,000. For each sample, 500,000–1,000,000 cells were acquired. The percentage of B-lymphocytes (CD19+), Mature B-cells (CD19+CD10-), Nave B-cells (CD19+CD10-C27-CD21+), resting memory B- cells (CD19+CD10-CD27+CD21+), activated memory B cells(CD19+CD10-CD27+CD21-) and tissue like memory B-cells (CD19+CD10-CD27-CD21-) was defined relative to the parent population with FlowJo version 10 (Tree Star, USA) using logicle (bi-exponential) transformation as recommended [27, 28] (Fig 1). Data were exported to Excel spreadsheet for each sample and compiled for further analysis.
Differences in percentage of B-cell subsets were analyzed with either the two-tailed Mann-Whitney U test or the Wilcoxon signed rank non-parametric tests using STATA 14 version 2 (San Diego California USA). Graphs were produced by GraphPad Prism version 5.01 for Windows (GraphPad Software, San Diego California USA). The median and Hodges-Lehmann estimator were used for result presentation. Hodges–Lehmann is used to measure the effect size for non-parametric data [29]. P-values were corrected for multiple comparisons. The statistical significance level was set at p≤0.05.
The median percentage of B-lymphocytes (CD19+) in patients with ENL (9.5%) and LL controls (11.6%) was not statistically significantly different at recruitment. During treatment, the percentage of B-cells slightly increased to 10% in patients with ENL and to 14% in LL patient controls but did not show a statistically significant difference. However, after treatment, the median percentage of B-cells appreciably decreased to 5.7% in patients with ENL while it was slightly decreased in LL patient controls to 12.0% and the difference between the two groups was statistically significantly different (P≤ 0.001; ΔHL = 6.02%) (Fig 2A).
The kinetic analysis of B-lymphocytes within ENL group at different treatment time points has shown that the median percentage of B-lymphocytes before and during treatment of patients with ENL was 9.5% and 9.9% respectively (P>0.05). However, after treatment, the percentage of these cells was significantly decreased to 5.7% compared with before and during treatment (P ≤ 0.05) (Fig 2D).
The median percentage of mature (CD19+CD10-) and immature (CD19+CD10+) B-cells in patients with ENL and LL controls before and after treatment were also measured. A significant difference was not observed with regard to the median percentage of mature and immature B-cells in patients with ENL reactions and non-reactional LL patient controls before and after treatment (Fig 2B). Similarly, the median percentage of mature B-cells before and after treatment was not statistically significantly different within ENL group (Fig 2E).
The median percentage of naive B-cells (CD19+CD10-CD27-CD21+) in untreated ENL patients was significantly lower (76.0%) than in LL patient controls (86.4%) (Fig 2C). This implies that more number of B-cells in untreated ENL patients are antigen experienced than those from LL patient controls. On the other hand, the percentage of naive B-cells within ENL group was not significantly changed before and after treatment (Fig 2F).
The median percentage of memory B-cell subtypes (resting, activated and tissue-like memory B-cells) in patients with ENL reactions and non-reactional LL controls as well as within ENL group was analysed in the unstimulated PBMCs before, during and after treatment.
The median percentage of activated memory B-cells (CD19+ CD10-CD27+CD21-) was significantly higher in patients with ENL (2.6%) than in LL patient controls (1.4%) before treatment (P≤ 0.005). During treatment, the percentage of activated memory B-cells (AM) in patients with ENL and LL controls increased to 3.8% and 4.4% respectively and the difference was not statistically significantly different (P> 0.05). After treatment, the percentage of these memory cells did not change (Fig 3B).
A comparison within ENL has shown that the median percentage of activated memory B-cells in untreated ENL patients was higher (2.6%) than after treatment (1.3%) and the difference was statistically significantly different (P≤ 0.005) (Fig 3E). Hence, it seems that activated memories B-cell is associated with ENL reactions.
The median percentage of resting memory B-cells (CD19+ CD10-CD27+CD21+) was 5.8% in patients with ENL and 4.8% in LL patient controls before treatment and the result was not statistically significantly different. However, during treatment it was increased to 15.2% in patients with ENL and was higher than in LL patient controls (8.6%) (P≤0.05). After treatment, the proportion of these memory cells was decreased to below 5% in both groups and was not statistically significantly different (Fig 3A).
Analysis within ENL group has shown that the proportion of resting memory B-cells (RM) was considerably lower (5.8%) before treatment than during treatment (15.3%) (P≤0.001). After treatment, the proportion of these resting memory B-cells was decreased to 3.7% and it was significantly lower than before and during treatment (P≤ 0.05) (Fig 3D).
Interestingly, untreated ENL patients had significantly lower (5.2%) median percentage of tissue-like memory B-cells (TLM) (CD19+CD10-CD27-CD21-) than the corresponding LL patient controls (10.7%) (P≤0.05). However, after treatment a significant difference was not observed between these two groups (Fig 3C). Similarly, comparison within ENL group has shown that the median percentage of TLM is significantly lower in untreated ENL patients than after treatment (P≤ 0.05) (Fig 3F).
Memory B-cells are subtypes of B-cells that are formed within the germinal centres following infection. They proliferate and differentiate into antibody producing plasma cells also called effector B-cells in response to re-infection. Memory B-cells rapidly differentiate into plasmablasts that produce class-switched antibodies which are capable of clearing the infection far more quickly than naive B-cells [3]. The different classes of memory B-cells have been studied in various chronic viral infections such as hepatitis and HIV and in several autoimmune diseases [30, 31]. The role of B-cells in the pathogenesis of ENL has been speculated in several studies but has never been studied. For the first time, we studied B-cells and the memory B-cell sub-types in patients with ENL and LL controls at different time points (before, during and after treatment) to investigate the dynamics of these cells during the course of prednisolone treatment.
The percentage of total B-cells was not significantly different in the two groups before treatment. However, after treatment, the proportion of B-cells was significantly reduced from 9.5% to 5.7% in patients with ENL. The reduction of B-cells after prednisolone treatment of patients with ENL could be either transitory or associated with the subsiding of the ENL reaction which needs further investigation. The success of Rituximab to deplete B-cells for the treatment of rheumatoid arthritis has stimulated investigation of its effects in several other immune disorders, and considerable interest in the potential of drugs that can modulate B-cell function for the treatment of such diseases [32]. Thus, the finding of reduced B-cells after ENL subsides poses the question whether depleting B-cells could be effective in the treatment of ENL.
Patients with ENL had a significantly lower naïve B-cells (76.0%) than LL controls (84.6%) before treatment (P≤0.05; L-H = 6.75). It implies that more B-cells are antigen experienced in untreated ENL patients compared to in LL patient controls. However, the proportion of naïve B-cells was still unusually high in spite of the presence of abundant M. leprae antigens in these patients.
A significant difference was not observed with regard to the frequency of resting memory B cells (RM) in the two groups before treatment. However, the median percentage of activated memory B-cells (AM) was significantly higher in patients with ENL (2.6%) than in LL controls (1.4%) before treatment. Several studies have shown that activated memory B-cells are increased in patients with disease flares in systemic lupus erythematosus (SLE) [33] and rheumatoid Arthritis [34]. However, the biology of ENL and autoimmune diseases is different and whether activated memory B-cells are undesirable or not in the pathogenesis of ENL should be further investigated to arrive at a conclusive evidence. Nevertheless, activated B-cells may be primed to plasma B-cells which in turn produce immunoglobulins [30]. These immunoglobulins could interact with the M.leprae antigen and thereby form excess immune-complexes beyond clearance or activated B-cells may serve as antigen presenting cells i.e. presenting M.leprae antigens to T-cells. Depending on the magnitude of antigen presentation, the T-cell response could be excessive and may cause tissue damage.
Patients with ENL had lower percentage of tissue-like memory (TLM) B-cells (5.2%) than LL controls (10.7%) before treatment. Several studies have indicated that TLM B-cells represent the exhausted state of B-cells since they express several inhibitory receptors, including the immunoreceptor tyrosine-based inhibitory motif (ITIM)-containing inhibitory receptor Fc receptor-like protein 4 (FcRL4) [35]. TLM B-cells show a reduced tendency to proliferate in response to cognate antigen [36]. The expression of FcRL4 on human B-cell lines disrupts immune synapse formation and blocks antigen induced B-cell receptor (BCR) signalling [37]. They also express not only FcRL4 but also a number of other inhibitory and chemokine receptors that would reduce the likelihood of B and T cell interaction [5]. It has been shown that a specific siRNA knockdown of FcRL4 and other inhibitory receptors may lead to a rescue of Ig secretion and proliferation in these tissue-like memory B-cells [38]. In contrary to our findings, increased proportion of TLM have been reported in chronic infections such as in hepatitis C virus and malaria infections, and in certain autoimmune diseases [39, 40]. The decreasing tendency of TLM and increased secretion of activated memory B-cells may indicate the activation of B-cells when LL patients develop ENL reactions. The finding of increased percentage of TLM in LL patients in this study may partly explain why B-cells are unable to control M.leprae multiplication in spite of their abundance in LL patients. The T-cell unresponsiveness in LL may also be associated with the increased production of TLM B-cells. It appears that the higher frequency of TLM B-cells in LL could alter the B-cell/T-cell interaction through blocking B-cell receptors and this hypothesis could be a fertile area for future investigation. Once, this hypothesis is proved, the search for the B-cell immunomodulators that safely overcome this exhaustion phenotype may be necessary in order to develop proper immune response to LL patients. On the other hand, the significant reduction of TLM during ENL reaction suggests the down regulation of inhibitory molecules and thereby increases immune activation in LL patients leading to the onset ENL reaction. Thus, our finding implies that TLM B-cells could have a role in the initiation of ENL reactions which is a new exciting area for further investigation.
|
10.1371/journal.pbio.2005345 | Inter-subunit interactions drive divergent dynamics in mammalian and Plasmodium actin filaments | Cell motility is essential for protozoan and metazoan organisms and typically relies on the dynamic turnover of actin filaments. In metazoans, monomeric actin polymerises into usually long and stable filaments, while some protozoans form only short and highly dynamic actin filaments. These different dynamics are partly due to the different sets of actin regulatory proteins and partly due to the sequence of actin itself. Here we probe the interactions of actin subunits within divergent actin filaments using a comparative dynamic molecular model and explore their functions using Plasmodium, the protozoan causing malaria, and mouse melanoma derived B16-F1 cells as model systems. Parasite actin tagged to a fluorescent protein (FP) did not incorporate into mammalian actin filaments, and rabbit actin-FP did not incorporate into parasite actin filaments. However, exchanging the most divergent region of actin subdomain 3 allowed such reciprocal incorporation. The exchange of a single amino acid residue in subdomain 2 (N41H) of Plasmodium actin markedly improved incorporation into mammalian filaments. In the parasite, modification of most subunit–subunit interaction sites was lethal, whereas changes in actin subdomains 1 and 4 reduced efficient parasite motility and hence mosquito organ penetration. The strong penetration defects could be rescued by overexpression of the actin filament regulator coronin. Through these comparative approaches we identified an essential and common contributor, subdomain 3, which drives the differential dynamic behaviour of two highly divergent eukaryotic actins in motile cells.
| Actin is one of the most abundant and conserved proteins across eukaryotes. Its ability to assemble from individual monomers into dynamic polymers is essential for many cellular functions, including division and motility. In most cells, actin is able to form long and stable filaments. However, an actin of the malaria-causing parasite Plasmodium, while having a very similar monomer structure to actins from other eukaryotes, forms only short and unstable filaments. These short and dynamic filaments are crucial in allowing the parasite to move very rapidly in tissue. Here we investigated the basis of these differences. We used molecular dynamics simulations of actin filaments to investigate the actin–actin interfaces in filaments from Plasmodium and rabbit. We next engineered parasites to express chimeric actins that contained different parts of rabbit and parasite actin and thereby identified actin residues important for parasite viability and progression across the life cycle. We could rescue the most prominent defect specifically with overexpression of the actin binding protein coronin. This suggests that the more stable actin harms the parasite and that coronin helps in recycling filaments. By screening the effects of actin chimeras in mammalian cells, we also identified regions that allow these different actins to efficiently interact with each other. Taken together, our results improve our understanding of the interactions required for actin to incorporate into filaments across divergent eukaryotes.
| Actin is a highly conserved cytoskeletal protein with essential roles in cell division, contraction, and motility. Cell motility is an important process in biological development, cancer metastasis, and tissue penetration by both immune cells and pathogens. Many eukaryotic cell types have the ability to move continuously in a substrate-dependent ameboid manner in both 2D and 3D environments, typically by deforming their cellular shape and protruding the cell’s leading edge [1–3]. Gliding motility is an alternative mode of locomotion displayed by some bacteria and protists that is independent of cell shape changes [4–6]. The malaria-causing parasite Plasmodium employs gliding motility in several phases of its life cycle. Gliding is essential for successful penetration of host organs such as the mosquito midgut and salivary glands as well as the mammalian liver [7]. Prior to liver infection, high motility speeds of 1–3 μm/s allow the parasite to escape from the dermis, where it is deposited by the biting mosquito, and thereby evade infiltrating neutrophils, the fastest migrating human cells, which move much slower (1–5 μm/min) [8,9]. Despite these striking differences in locomotion modes and speeds, the fundamental requirement for these cells is the dynamic turnover of actin filaments [10–12].
The actin monomer has a highly conserved structure that consists of four subdomains and a central nucleotide (adenosine triphosphate [ATP], adenosine diphosphate with inorganic phosphate [ADP + Pi], or adenosine diphosphate alone [ADP]) binding cleft (Fig 1A) [13]. Actin possesses the ability to self-assemble from monomers (G-actin) to form filaments (F-actin), which in turn can form higher order filamentous structures. Particular regions in the actin subdomains, such as the hydrophobic plug (H-plug) of subdomain 3 and the highly flexible DNAse I-binding loop (D-loop) of subdomain 2, as well as the nucleotide state, have been implicated as major contributors to the formation and stability of filaments [14–16]. Actin isoforms of most eukaryotes inherently form long and stable filament structures (greater than 1 μm in length), display high sequence conservation across species (>90% similarity from yeast to humans), and are regulated by a large set of actin binding proteins (ABPs) [17]. These ABPs can affect the monomer-filament ratio and fulfil a variety of regulatory roles, including but not limited to nucleation and elongation (e.g., formin), monomer binding (e.g., profilin), filament binding (e.g., coronin), and filament severing (e.g., cofilin/actin depolymerising factor [ADF] family). In contrast, protozoan parasites express divergent actins (60%–80% identity with vertebrate actins) that typically differ in their ability to form actin filaments. Giardia, trypanosomid, and apicomplexan actins are refractory to actin polymerisation modulating compounds, such as latruculin, and their structures are difficult to visualise with phalloidin [18–21]. Similarly, actin 1 of Plasmodium has fundamental shifts in its functional properties: despite a similar monomer tertiary structure [22] to mammalian actin, Plasmodium actin only forms short filaments of approximately 100 nm in length, has a noncanonical filament structure that is dynamically unstable [16,22–28], displays slow polymerisation yet rapid depolymerisation rates [29], and is regulated by a highly reduced set of predicted ABPs [30]. Such altered properties are crucial for intracellular parasite growth [31–33] and efficient parasite motility [12,31,33–35]. These fundamental differences make Plasmodium actin a useful model for the comparative assessment of key amino acid residues that result in altered filament properties. Here, we combined multiscale molecular dynamics (MD) simulations with three newly developed molecular genetic screens to identify the contribution of particular amino acid residues to the altered properties of canonical rabbit actin and the evolutionarily distant Plasmodium actin 1. The results reveal a distinct region in subdomain 3 as the primary contributor to divergent actin behaviour and filament incorporation.
Plasmodium actin 1 is one of the most divergent eukaryotic actins known (80% sequence identity compared to rabbit alpha actin, S1 Fig). To assess the structural and dynamic differences between actin monomers from parasites, as represented by Plasmodium actin 1, and canonical mammalian actin monomers, as represented by rabbit actin, we first employed all-atom MD simulations. Apart from a more flexible region in subdomain 4, comparison of the breathing dynamics of the actin monomers revealed no marked differences in behaviour between the actins from the two species (S2 Fig). To assess actin filament dynamics, we constructed 15-mer Plasmodium and rabbit actin filaments based on electron microscopy (EM) data for the rabbit alpha actin filament [36] and performed five coarse-grained (CG) MD simulations of 10-μs duration for each filament. These simulations recapitulated the modified filament architecture observed in previous EM studies, in terms of alpha angle and subunits per crossover [16,22,27] (S3 Fig). Moreover, our models allowed for the comparative assessment of the dynamics of the molecular interactions both within and between subunits in the filaments. Comparison of the filament models for the two species revealed that both filaments displayed essential common interaction hot spots in regions involved in intermolecular contacts (Fig 1B–1E). Two interaction regions in subdomains 1 and 3 (designated S1b and S3a) are highly conserved across species (over 90% sequence identity of Plasmodium compared to rabbit alpha actin). Other interaction regions corresponded to sequences in subdomains 2, 3b, and 4 (Fig 1E, overall sequence identities of 78%, 72%, and 77%, respectively). Interestingly, within these regions there are residue differences between Plasmodium and rabbit actin filaments and thus altered contacts between filament subunits (Fig 1E). Thus, whereas the locations of the interfaces are conserved, the residues are, on average, more divergent at the interfaces than elsewhere (S1 Table).
We sought to assess the contribution of these divergent regions to the altered behaviour of actin species in their cellular context. As a first unbiased screen, chimeras containing exchanges between Plasmodium and mammalian actin were generated and tested for their ability to replace the endogenous actin 1 gene in the parasite (Fig 2A and 2B). To do so, we established a two-step genetic methodology that allowed for replacement of the endogenous copy while leaving the desired locus selection marker free and with minimal flanking changes in the parasite genome (S4 Fig). For rabbit actin-based chimeras, our screening approach revealed that any changes to this actin were insufficient to obtain viable parasite lines, strongly indicating that no one region is capable of full restoration of parasite-like actin function from a canonical backbone. Even for exchanges introduced into the parasite actin, the majority of the parasite actin chimeras were unable to functionally replace the endogenous gene. Importantly, changes to subdomains 2 and 3 of the parasite actin resulted in parasite death, indicating that these regions provide essential features required for parasite blood stage growth (the stage at which the genetic manipulations are performed, Fig 2A). Further, mutants with single point mutations in these subdomains (P42Q and K270M) were also lethal, strongly implicating the essentiality of these regions for normal parasite function in the blood stages of infection. However, two of the most divergent regions could be exchanged with the canonical mammalian actin equivalent regions: an N-terminal domain swap (PbS1aOc, residues 1–33, 67% sequence identity with rabbit alpha actin with one additional acidic residue on its N-terminus) and a subdomain 4 replacement (PbS4Oc, residues 182–263, 77% sequence identity with rabbit alpha actin). Interestingly, these transgenic parasites displayed asexual growth rates in the blood that were comparable to the wild-type replacement control (Fig 2C), suggesting that modification of these regions does not significantly affect intracellular growth and erythrocyte invasion.
The malaria parasite needs to propagate and disseminate in a variety of tissue environments and different temperatures, suggesting that the required actin dynamics of these other stages could be different [37,38]. We thus tested if the PbS1aOc and PbS4Oc chimeric lines were effective in infection of Anopheles mosquitoes. After transmission to the mosquito midgut, the parasite develops into midgut-penetrating ookinetes, which can traverse the epithelia to establish oocysts on the basal lamina (Fig 2A). These oocysts produce thousands of sporozoites that are subsequently released into circulation and actively invade the insect’s salivary glands [7]. PbS1aOc produced slower moving ookinetes and reduced oocyst numbers (Fig 3A and 3B). Furthermore, while the numbers of sporozoites in the mosquito circulation were within the usual range, suggesting normal oocyst development, a marked reduction in salivary gland occupancy and sporozoite motility was observed. PbS1aOc displayed a 50% reduction in parasite numbers in the salivary glands (Fig 3C). While this line was capable of infecting naive mice at similar rates to wild-type controls (Fig 3D and 3E) and showed a similar motile population compared to wild-type controls (Fig 3F), these parasites moved at a lower average speed (Fig 3G). Thus, a modification of the N-terminal region of actin, with a concomitant increase in the number of acidic residues, resulted in a decreased motility of two different parasite stages, which affected their ability to penetrate the organs of the mosquito.
Alteration of the amino acid residue composition of subdomain 4 (PbS4Oc) resulted in a much more pronounced defect in salivary gland invasion (Fig 3C). Again, this line was still able to cause infection in mice by natural mosquito transmission, although the decreased numbers of parasites in these glands resulted in a concomitant delay in infection (Fig 3D and 3E). The motile population of salivary gland–resident parasites was reduced and these moved in a discontinuous fashion (Fig 3F–3H). Interestingly, these parasites often paused during motility, a phenomenon not yet described for any Plasmodium mutant (Fig 3H). Intriguingly, such events were often accompanied by short reversals in direction and detachment at the rear of the parasite (Fig 3I and 3J, S1 Movie and S2 Movie). While pausing did not change the average speed compared to wild-type, there was an increased range of speeds obtained by individual parasites (Fig 3G).
Changing the subdomain 4 region involved the alteration of 20 amino acid residues to canonical equivalents. Strikingly, the pausing phenotype was also observed by combined modification of only three residues in this region (three combined parasite to mammalian residue mutants H195T, G200S, and E232A: mutant HGE/TSA) (Fig 3 and S5 Fig). CG MD simulations with this triple mutant indicate that a change of these three residues results in a shift of filament parameters toward rabbit actin (S3 Fig) and an atypical interaction profile that is not observed in either species: residues F54 and Y189 display enhanced contacts with Y170 and H174, respectively, which could have consequences for actin filament turnover (S6 Fig) [16,36]. Taken together, altering actin dynamics by mutation of nonessential actin regions results in parasites that, while behaving normally in the mammalian host, have prominent inabilities to effectively colonise the mosquito vector.
Plasmodium actin cannot be visualised using typical imaging approaches [39], yet expressing the F-actin binding Plasmodium coronin fused with mCherry enabled visualisation of F-actin in motile sporozoites [37]. Coronin localises at the periphery of nonmotile sporozoites and relocalises to the rear in motile sporozoites in an actin filament–dependent fashion [37]. To investigate a possible change in localisation, we transfected PbS4Oc-expressing parasites with mCherry-tagged coronin under the control of a sporozoite-specific promoter (Fig 4A, S7 Fig). Curiously, coronin localised to the periphery in an actin-independent manner in these sporozoites resembling the localisation in nonmotile wild-type sporozoites or after treatment with the actin filament modulator cytochalasin D (Fig 4B). This indicates that the majority of the tagged coronin was unable to bind the modified actin, similar to observations with a coronin actin binding mutant and when parasites are treated with filament stabiliser, jasplakinolide [37]. Intriguingly, overexpression of coronin, but not overexpression of profilin and ADF2, rescued efficient motility and salivary gland penetration of PbS4Oc sporozoites (Fig 4C–4E). Overexpression of coronin in WT sporozoites also increased their capacity to move [37], suggesting that in PbS4Oc sporozoites, coronin overexpression is compensating for its reduced binding ability to a mutated actin. To further probe the domain contributions to this rescue, we tested whether mutations to coronin would improve motility and invasion of the PbS4Oc parasite line. Mutations in the N-terminal actin binding domain (coronin mutant R349E, K350E), as expected, could not rescue the PbS4Oc phenotype. We also tested if the actin binding N-terminal domain of coronin (residues 1–388, lacking the unique region and coiled-coil domain), which was shown to bind and bundle actin [40], affected motility and invasion. This also could not improve the invasion and motility of the PbS4Oc parasite, suggesting a cooperative involvement of both N- and C-terminal regions of coronin in mediating rescue (Fig 4C–4E).
Above, we identified subdomains 2 and 3 as key contributors to essential processes in blood stage parasites (Fig 2B). In order to assess the contribution of these regions to altered actin dynamics, we performed a second screen by expressing these variants (PbS2Oc, OcS2Pb [residues 34–71], PbS3bOc, and OcS3bPb [residues 264–338]) as mCherry-tagged additional copies under the control of a sporozoite stage-specific promoter (Fig 5A, S8 Fig and S9 Fig). This allowed for careful phenotypic characterisation in the parasite without the lethality observed in the initial screen. Tagged actins revealed interesting differences: an additional copy of parasite actin was present throughout the parasite cell, as expected [41]. Yet, mammalian actin was only cytoplasmic with a notable absence in the nuclear region (Fig 5B). Furthermore, the two ‘control’ parasite lines (expressing either parasite or canonical actin) differed in their response to jasplakinolide: parasite mCherry-actin accumulated at both the front and rear of the cell (Fig 5B), sites implicated for F-actin formation (in the case of formin 1, which is located at the front; S8 Fig) and turnover (rear) [37,40,41,42]. In contrast, mammalian mCherry-actin displayed no change in localisation (Fig 5B). This suggests that tagged mammalian actin is not readily incorporated into jasplakinolide-stabilised parasite F-actin in the sporozoite. We next assessed localisation and jasplakinolide responsiveness with the actin chimeras. Unexpectedly [22], the chimeras containing exchanges in subdomain 2 (a highly flexible and divergent region between species) were similar in behaviour to their corresponding controls (Fig 5C). In contrast, exchanging subdomain 3 resulted in reciprocal effects: a mammalian actin containing this corresponding parasite region (residues 264–338 with 21 changes; 72% sequence identity) resulted in a changed localisation under jasplakinolide, thus suggesting more efficient incorporation to endogenous filaments (Fig 5D). Changing the same region in the parasite actin resulted in a cellular distribution and jasplakinolide response similar to mammalian actin, suggesting that subdomain 3 contains the contacts necessary and sufficient for incorporation into divergent filaments. Importantly, the effect appears to be independent of the H-plug, as changing the two divergent amino acid residues on the loop and the two most divergent residues outside this loop back to residues in the canonical actin (K270M, A272S, E308P, and T315Q) still rendered an additional copy of actin that behaved similarly to Plasmodium actin (Fig 5D) [16]. Parasites expressing rabbit actin displayed a 50% reduction in parasite average speed (Fig 5E). Interestingly, all chimeras of Plasmodium actin consistently moved with a slightly higher average speed compared to their corresponding mammalian chimeras. OcS3bPb, the only mammalian actin chimera that responded to jasplakinolide treatment, moved at a similar speed to the wild-type Plasmodium actin control (Fig 5E).
To test whether the specific regions that we identified above are common contributors to actin dynamics in other cell types, we performed a third screen by transfecting the panel of chimeras into different mammalian cells as additional copies. Higher eukaryotic cells did not readily incorporate parasite actin into stable filamentous structures (Fig 6A and 6B, S10 Fig). Furthermore, cells containing parasite actin moved slightly faster in a random migration assay (Fig 6C). These observations indicate that malaria parasite actin can be used as a template to identify the minimum determinants required for filament incorporation in higher eukaryotic cells. Strikingly, changing subdomain 2 or 3 into the corresponding mammalian equivalent resulted in the parasite GFP-actin more readily incorporating into filament networks (Fig 6A and 6B). In contrast, exchanging individual regions of mammalian actin did not result in decreased incorporation, suggesting that more than one region can provide the required minimal contacts necessary for incorporation.
To identify key amino acid residues contributing to filament incorporation, multiple mutations (whereby parasite actin residues were converted to conserved mammalian counterparts) in both subdomain 2 and subdomain 3 were generated and analysed (Fig 6A and 6B). Consistent with the results of the parasite lines, single and multiple mutations of both the H-plug and regions typically on the surface of the actin filament in subdomain 3 did not result in improved filament incorporation, suggesting an extensive interface in the subdomain 3 region required for incorporation. Remarkably, a single residue change (N41H) in subdomain 2 of Plasmodium actin was sufficient to rescue filament incorporation (Fig 6A and 6B). This indicates that the presence of an imidazole group alone provides sufficient interaction for more efficient filament incorporation (Fig 6A and 6B). Thus, crucial differences between actin species, including the ability for a monomer to incorporate into divergent filaments, are due to the changes in particular divergent regions in specific subdomains only. Furthermore, subdomain 3 is the key common contributor to filament incorporation in two very diverse cellular systems.
While, structurally, the Plasmodium actin 1 monomer is similar to that of canonical mammalian actins [22], Plasmodium actin filaments are shorter and more dynamic [22,26,27]. Here, we have shown that Plasmodium actin is unable to efficiently incorporate into actin filaments of mammalian cells and that tagged mammalian and Plasmodium actin localise differently in parasites. Using MD simulations as well as chimeric and mutagenesis genetic screens, we identified the fundamental contributors that underlie these different characteristics. Our data independently confirm that these two actins occupy a similar structural space and identify divergent amino acid residues responsible for the differences in filament dynamics. Plasmodium was unable to tolerate allelic changes in subdomains 2 and 3, suggesting that these regions have crucial features required for parasite biology and are, particularly P42 and the H-plug, key contributors to contact sites. The data suggest that modifications of these two regions result in filaments that are too stable and thus lethal for the parasite, similar to the effects observed with jasplakinolide treatment or genetic ablation of the actin regulator, ADF1, and the alpha capping protein [31,32,43,44].
Unlike exchanges in subdomains 2 and 3, the parasite readily tolerated the exchange of highly divergent regions of subdomain 4 and the N-terminal residues (Fig 2). These parasite lines grew like wild-type parasites in the blood stage and were only affected in their progression in the mosquito host (Fig 3). Importantly, this indicates that red blood cell invasion, while dependent on actin [33], is more tolerant of minor changes to its sequence. This observation is consistent with ablating ABP expression having little effect on the parasite in the mammalian host but having a strong effect once the parasite infects the insect vector [37,44–48]. While other studies looked indirectly with actin regulators, our study investigated the core of the machinery itself across the life cycle, revealing that modest changes to dynamics have important consequences in mosquito colonisation. This indicates that the greatest selection pressure on the parasite actin sequence could be during active organ penetration of the mosquito.
Changes to the N-terminal sequence resulted in a consistent decrease in speed between the two parasite stages that employ gliding motility in the mosquito (Fig 3B and 3G). Given the well-known interaction between the actin N-terminus and myosin [49–56], it is reasonable to suggest that the decrease in cell speed is due to a reduced interaction between these two proteins. We thus propose that parasites containing this change in acidic residue content have reduced force transmission of actomyosin, which results in reduced average speeds. This hypothesis could be probed in vitro using classical myosin sliding filament assays with the respective Plasmodium machinery components [57].
Previous studies in yeast have indicated a relatively low threshold in opisthokonts to accommodate changes to its actin subdomain 1 sequence, in which the addition of a single acidic residue to the N-terminus rendered viable yet sick yeast cells, while two additional acidic residues were lethal for the cell [58]. In comparison, the addition of another acidic residue to parasite actin was surprisingly well tolerated (Figs 2 and 3): these parasites could complete the life cycle and move sufficiently well to cause infection (albeit at a lower overall efficiency in the mosquito). Interestingly, yeast cells with changes to subdomain 1 and the N-terminus were sensitive to changes in growth temperature [58–61]. Given the marked differences in body temperatures between mosquito and mammal, it is possible that changes in parasite actin sequence affect the well-tuned actin dynamics required at different temperatures. In vitro polymerisation assays with these mutants at both 37°C and 21°C would provide important insights into differences in the polymerisation kinetics at different temperatures. These detailed kinetic studies are especially interesting given that hybrid actins between rabbit muscle and yeast equivalents can produce unexpected changes to certain biochemical properties [58].
Interestingly, modifications of three amino acid residues in subdomain 4 rendered a parasite that frequently paused and reversed direction during migration (Fig 3H–3J). We propose that this change in motility is due to more stable actin filaments, which might be misoriented and hence allow partial rearward motility. Similar movements can be observed if Toxoplasma parasites are treated with high concentrations of jasplakinolide, a filament-stabilising drug [62]. We showed previously that coronin can bind to actin filaments in motile sporozoites but not to those treated with jasplakinolide [37]. Intriguingly, in the PbS4Oc sporozoites, coronin did not bind actin filaments and hence this parasite to some extent phenocopies jasplakinolide treatment. These parasites could also not enter salivary glands efficiently, for which well- tuned actin dynamics are important [37,45]. Yet, overexpression of coronin rescued both motility and invasion phenotypes (Fig 4). This indicates a central role for coronin in rapid filament recycling in the highly motile parasite. We propose that the discontinuous movements in subdomain 4 mutants (PbS4Oc and HGE/TSA) result from reduced actin filament recycling due to altered architecture and interactions at the filament interface, particularly by enhanced interactions of F54 and Y189. By alteration of subdomain 4 residues, the resulting contacts by conserved amino acid residues render the parasite actin filament less prone to disassembly as choreographed by coronin (S11 Fig).
Reduced invasion and aberrant motility were only rescued by full-length coronin, while mutated or truncated coronins were insufficient to rescue these phenotypes. This may suggest a cooperative requirement of the coronin domains for actin turnover (Fig 4C–4E). This result was somewhat surprising, because a recent publication on the biochemical properties of P. falciparum coronin suggested that much of the classical coronin function could be carried out by the N-terminal actin binding domain alone [40]. Our work indicates that, in the context of the P. berghei actin mutated sporozoite, both regions of coronin are required for full functionality. The mechanism by which coronin could be mediating Plasmodium filament recycling could be similar to that in higher eukaryotes. In these, coronin mediates a spatial-temporal recruitment of other filament regulators [63–65]. Coronin binds first and, depending on the nucleotide state of the actin [63,66], can serve to shield or allow access of ADFs to the filament. Binding of these additional factors together results in destabilisation of the filament. Given the close proximity of coronin and ADF on canonical filaments, it was also suggested that coronin could interact directly with ADF in the filament [64–67]. It is tempting to speculate that the rescue mediated by coronin could be due to a stepwise recruitment by coronin, with depolymerisation factors on the highly dynamic Plasmodium actin filament, which thus enhances severing. Our rescue experiments thus hint at a possibly conserved spatial-temporal coronin mechanism for filament regulation beyond what has currently been demonstrated for opisthokonts.
The C-terminal region of coronin appears important for this rescue, which suggests that this domain could allow for an interaction with other ABPs [68]. Indeed, it has been suggested for coronin of the related parasite T. gondii that the C-terminal coiled-coil domain could function as a molecular recruitment hub [69]. Alternatively, the C-terminal region simply provides an improved interaction with the filament, resulting in efficient turnover in the parasite. Together, our observations suggest interplay between both halves of Plasmodium coronin to mediate efficient rescue in motile sporozoites.
A recent high-resolution cryo-EM structure of the jasplakinolide-stabilised Plasmodium actin filament [16] is in good agreement with our models and generally fits with our biological observations. However, some of our in vivo observations could not be directly inferred from the static structure. For example, the combined mutation of H195 and G200 (to Thr and Ser, respectively) at a lateral interface in Plasmodium actin had no significant effect on the parasite across the entire complex life cycle. A third residue, E232, which is not at an interface, with the exception of one published structure [70], needed to be mutated in order for an effect on parasite biology to be observed, suggesting a dynamic interplay across several residues in this region (Fig 3C–3H). It is possible that E232 could provide a compensatory interaction when H195 and G200 are mutated. Our systematic approach, which spans from molecular models to cells, has thus identified novel cooperative interactions within actin filaments (Fig 7).
In this study, we made use of additional copy expression systems in highly divergent cells. It is important to note that these assays can only provide insights into monomer incorporation into filaments. Thus, we can make no direct inferences regarding whether each actin mutant, if present as the primary untagged actin in the cell, can fully complement wild-type function. For example, PbS2Oc was lethal as a replacement yet behaved similarly to wild-type Plasmodium actin as an additional copy in the parasite (Fig 5). Newly available genetic tools, such as inducible knockout systems [33], might be adapted to probe these otherwise lethal mutants for their functional consequences in cells.
Our observation that parasite actin does not efficiently incorporate into the mammalian cell actin network provided a useful tool in understanding the minimal requirements for network incorporation. Interestingly, subdomain 4 conversion of parasite actin did not improve network incorporation efficiency, while changing subdomain 2 or 3 resulted in a considerable incorporation into filamentous structures. Subdomain 2, with its highly dynamic extendibility, has been implicated to act as the primary interaction arm, after initial contact by subdomain 4, to allow the incoming monomer to bind to the barbed end [36]. The N41H mutant in Plasmodium actin subdomain 2 alone shows improved incorporation into mammalian actin filaments. Histidine has the capacity to have more contacts than asparagine and thus this position might be the key prominent interaction to allow optimal contact with the filament [16] (Fig 1E). Indeed, oxidation of H40 results in an actin monomer that is unable to polymerise [71,72]. Here, we extend this understanding to include this contact as important for an incoming monomer to dock onto a filament. We, with others, have shown that subdomain 3 provides important contact sites across the strand once the monomer has been incorporated and also for the incoming monomer to be included in the filament [15,36,73] (Fig 1 and Fig 7). Furthermore, we cannot exclude that subdomain 3 could be providing the required contact interface to allow for increased ABP-mediated incorporation [74]. We propose that having at least one of these important subdomain sites (2 or 3b) is sufficient for effective incorporation in higher eukaryotic cells. Identification of altered ABP binding between actin species could shed light on the primary factors responsible for monomer incorporation. Notably, there is a consistent feature between divergent systems: alteration of subdomain 3 allowed for a mammalian equivalent to be more efficiently incorporated into the highly divergent parasite cytoskeleton, indicating that this region could be a common ‘site of recognition’ across differently behaving actin species. This could be tested with other highly divergent actins to assess whether this region acts as a common recognition point across species.
Through our comparative approaches, we have identified essential contributors to the differential behaviour of two highly divergent actin species (Fig 7). We have identified a region of subdomain 3 outside the H-plug as providing important contacts to allow an otherwise divergent actin to be more efficiently incorporated into canonical actin filaments, thus enhancing our molecular understanding of the dynamic space required for filament incorporation. We show that a deeper understanding of Plasmodium actin dynamics sheds light on the fundamental properties of the actin polymer across eukaryotes. Such findings provide insights into the structural space required for actin subunits to assemble in order to meet the diverse needs for filament dynamics of different cells, ranging from protozoan parasites through to classical mammalian systems.
All animal experiments were performed according to the German Animal Welfare Act (Tierschutzgesetz) and were approved by the responsible German authorities (Regierungspräsidium Karlsruhe, numbers G3/11, G-134/14, G-283/14).
Monomer MD simulation
|
10.1371/journal.pgen.1000008 | Data and Theory Point to Mainly Additive Genetic Variance for Complex Traits | The relative proportion of additive and non-additive variation for complex traits is important in evolutionary biology, medicine, and agriculture. We address a long-standing controversy and paradox about the contribution of non-additive genetic variation, namely that knowledge about biological pathways and gene networks imply that epistasis is important. Yet empirical data across a range of traits and species imply that most genetic variance is additive. We evaluate the evidence from empirical studies of genetic variance components and find that additive variance typically accounts for over half, and often close to 100%, of the total genetic variance. We present new theoretical results, based upon the distribution of allele frequencies under neutral and other population genetic models, that show why this is the case even if there are non-additive effects at the level of gene action. We conclude that interactions at the level of genes are not likely to generate much interaction at the level of variance.
| Genetic variation in quantitative or complex traits can be partitioned into many components due to additive, dominance, and interaction effects of genes. The most important is the additive genetic variance because it determines most of the correlation of relatives and the opportunities for genetic change by natural or artificial selection. From reviews of the literature and presentation of a summary analysis of human twin data, we show that a high proportion, typically over half, of the total genetic variance is additive. This is surprising as there are many potential interactions of gene effects within and between loci, some revealed in recent QTL analyses. We demonstrate that under the standard model of neutral mutation, which leads to a U-shaped distribution of gene frequencies with most near 0 or 1, a high proportion of additive variance would be expected regardless of the amount of dominance or epistasis at the individual loci. We also show that the model is compatible with observations in populations undergoing selection and results of QTL analyses on F2 populations.
| Complex phenotypes, including quantitative traits and common diseases, are controlled by many genes and by environmental factors. How do these genes combine to determine the phenotype of an individual? The simplest model is to assume that genes act additively with each other both within and between loci, but of course they may interact to show dominance or epistasis, respectively. A long standing controversy has existed concerning the importance of these non-additive effects, involving both Fisher [1] and Wright [2]. Estimates of genetic variance components within populations have indicated that most of the variance is additive [3],[4]. Increasing knowledge about biological pathways and gene networks implies, however, that gene-gene interactions (epistasis) are important, and some have argued recently that much genetic variance in populations is due to such interactions [5],[6],[7],[8]. It is important to distinguish between the observations of dominance or epistasis at the level of gene action at individual loci, exemplified by a table of genotypic values, and the observations of variance due to these components in analysis of data from a population. For example, at a completely dominant locus almost all the variance contributed is additive if the recessive gene is at high frequency [3],[4].
An understanding of the nature of complex trait variation is important in evolutionary biology, medicine and agriculture and has gained new relevance with the ability to map genes for complex traits, as demonstrated by the recent burst of papers that report genome-wide association studies between complex traits and thousands of single nucleotide polymorphisms (SNPs) [9],[10],[11],[12],[13]. Here we attempt to resolve the alternative sources of evidence on the importance of non-additive genetic variation. We evaluate the evidence from empirical studies of genetic variance components and indeed find that additive variance typically accounts for over half and often close to 100% of the total genetic variance. We then present new theory and results that show why this is the case even if there are non-additive effects at the level of gene action.
In view of the apparent conflict between the observations of high proportions of additive genetic variance (often half or more of the phenotypic variance, and even more of the total genetic variance) and the recent reports of epistasis at quantitative trait loci (QTL) [8], we consider explanations beyond that of simple sampling errors and bias of estimates. We focus particularly on the role that the distribution of gene frequencies may play in the relation between the genetic model and the observed genetic variance components.
Genetic variance components depend on the mean value of each genotype and the allele frequencies at the genes affecting the trait [3],[4],[17]. Unfortunately the allele frequencies at most genes affecting complex traits are not known, but the distribution of allele frequencies can be predicted under a range of assumptions. This distribution depends on the magnitude of the evolutionary forces that create and maintain variance, including mutation, selection, drift and migration. As the effects on fitness of genes at many of the loci influencing most quantitative traits are likely to be small, we can invoke theory for neutral alleles to serve as a reference point. An important such reference is the frequency distribution under a balance between mutation and random genetic drift due to finite population size in the absence of selection. If mutations are rare, the distribution of the frequency (p) of the mutant allele is f(p)∝1/p, i.e. approximately L-shaped [2],[35],[36], with the high frequency at the tail being due to mutations arising recently. The allele which increases the value of the trait may be the mutant or ancestral allele, so its frequency has a U-shaped distribution f(p)∝1/p+1/(1−p) = 1/[p(1−p)]. As we shall use it often, we define the ‘U’ distribution explicitly by this formula. For loci at which the mutants are generally deleterious, the frequency distribution will tend to be more concentrated near p = 0 or 1 than for this neutral reference point. As another simple reference point we use the uniform distribution, f(p)∝1, 1/(2N) ≤ p ≤ 1−1/(2N), with N the population size. This approximates the steady state distribution of a neutral mutant gene which has been segregating for a very long time [2], and also has much more density at intermediate gene frequencies than the ‘U’ distribution. Our third reference point is at p = 0.5, as in populations derived from inbred crosses, and is the extreme case of central tendency of gene frequency.
These analyses assume a gene frequency distribution which is relevant to no selection. For a more limited range of examples we consider the impact of selection on the partition of variance. We consider a limited range of genetic models, some simple classical ones and others based on published models of metabolic pathways or results of QTL mapping experiments.
Uniform: f(p) = 1, assuming N is sufficiently large that the discreteness of the distribution and any non-uniformity as p approaches 1 or 0 can be ignored, i.e. integrated over 0 to 1. This and the ‘U’ gene frequency distributions are, for simplicity, assumed to be continuous.
Neutral mutation model (‘U’): f(p)∝1/[p(1−p)]. To standardise the distribution, with population size N assumed to be large, note that
Thus , where K∼ln(2N).
Genetic variance components are obtained by integration of expressions for the variance as a function of p for a specific model of the gene frequency distribution. For multiple locus models the distribution of all loci is assumed to be identical and there is no linkage disequilibrium. We focus on the contribution of additive genetic variance (VA) to genotypic variance (VG).
Many general points are illustrated by two simple examples, the single locus model with dominance and the two locus model with AA interaction, so we consider these in more detail. For the single locus model with genotypic values for CC, Cc and cc of +a, d and −a, respectively, VA = 2p(1−p)[a+d(1−2p)]2 and VD = 4p2(1−p)2d2. For d = a, i.e. complete dominance of C, VA = 8p(1−p)3a2 and VD = 4p2(1−p)2a2 and thus: at p = 0.5, VA = (2/3)VG; if the dominant allele is rare (i.e. p → 0), VG → 8p and VA/VG → 1, and if it is common, VG → 4p2 and VA/VG → 0. Note, however, that VG and VA are much higher when the dominant allele is at low frequency, e.g. 0.1, than are VG and VD when the recessive is at low frequency, e.g. p = 0.9. Even for an overdominant locus (a = 0), all genetic variance becomes additive at extreme gene frequencies. Considering now expectations (E) over the frequency distributions, let η2 = E(VA)/E(VG), an equivalent to narrow sense heritability if VE = 0. For the ‘U’ distribution, η2 = 1−d2/(3a2+2d2) and for the uniform distribution, η2 = 1−2d2/(5a2+3d2). Hence, for a completely dominant locus, η2 = 0.8 and η2 = 0.75 respectively; whereas VA/VG = 0.67 for p = 0.5. In summary, the fraction of the genetic variance that is additive genetic decreases as the proportion of genes at extreme frequencies decreases (Table 2).
The genotypic values (see Theory section) for the simple AA model for double homozygotes BBCC and bbcc are +2a and for bbCC and BBcc are 0, and all single or double heterozygotes are intermediate (+a). With linkage equilibrium, VA/VG = 1−HpHq/[Hp+Hq−3HpHq], where the heterozygosities are Hp and Hq at loci B and C. Thus VA/VG → 1 if either locus is at extreme frequency (i.e. p or q → 0 or 1), and equals 0 when p = q = 0.5. If p = q, for gene frequencies 0.1, 0.2, 0.3 and 0.4, VA/VG = 0.88, 0.69, 0.43 and 0.14. For the uniform distribution η2 = 2/3, and for the ‘U’ distribution, the variances are a function of the population size, because more extreme frequencies are possible at larger population sizes. Thus η2 = (2−4/K)/(2−3/K), where K = ln(2N), so η2 → 1 for large K. Any residue is VAA.
These two examples, the single locus and A × A model, illustrate what turns out to be the fundamental point in considering the impact of the gene frequency distribution. When an allele (say C) is rare, so most individuals have genotype Cc or cc, the allelic substitution or average effect of C vs. c accounts for essentially all the differences found in genotypic values; or in other words the linear regression of genotypic value on number of C genes accounts for the genotypic differences (see [3], p 117). Hence almost all VG is accounted for by VA.
With the ‘U’ distribution, most genes have one rare allele and so most variance is additive. Further examples (Table 2) illustrate this point, including the duplicate factor and complementary models where there is substantial dominance and epistasis. These models show mostly VA for the ‘U’ distribution for a few loci but the proportion of the variance which is additive genetic declines as the number increases. With many loci, however, such extreme models do not explain the covariance of sibs (i.e. any heritability) or the approximate linearity of inbreeding depression with inbreeding coefficient, F, found in experiments [3],[4],[40],[41],[42], or the linearity in response to artificial selection [43].
We also analysed a well-studied systems biology model of flux in metabolic pathways [38],[39],[44] and found again that the expected proportion of VG that is accounted for by VA is large (Table 3).
A number of QTL analyses using crosses between populations (some inbred, some selected) have been published in which particular pairs (or more) of loci have been identified to have substantial epistatic effects [8]. We consider examples of the more extreme cases of epistasis found, obtaining variance components by numerical integration. Results are shown in Table 4, for examples from [8] deliberately chosen as extreme. Even so, the proportion of the genetic variance that is additive is high with the ‘U’ distribution, except in the dominance × dominance example. Further, as these examples were selected by Carlborg and Haley and us as cases of extreme epistasis, it is not unreasonable to assume that the real epistatic effects are smaller than their estimates.
A test of the hypothesis that the lack of non-additive variance observed in populations of humans or animals is because gene frequencies near 0.5 are much less common than those more extreme, not because non-additive effects are absent, is to compare variance components among populations with different gene frequency profiles. For crops such as maize and for laboratory animals, estimates can be got both from outbreds and from populations with gene frequencies of one-half derived from crosses of inbred lines. There are a limited number of possible contrasts and linkage confounds comparisons of variation in F2 and later inter se generations, however, so it is difficult to partition variation between single locus and epistatic components (e.g. [17] ch. 7).
The most extensive data are on yield traits in maize. The magnitudes of heritability and of dominance relative to additive variance estimated for different kinds of populations in a substantial number of studies (including 24 on F2 and 27 on open-pollinated, i.e. outbreds) have been summarised [59]. Average estimates of h2 were 0.19 for open-pollinated populations, 0.23 for synthetics from recombination of many lines, 0.24 for F2 populations, 0.13 for variety crosses and 0.14 for composites. Estimates of VA/VG (from tabulated values of VD/VA [59]) were 0.57, 0.55, 0.50, 0.42 and 0.43, respectively, which are inconclusive but indicate relatively more dominance variance at frequencies of 0.5. Analyses of the magnitude of epistasis at the level of effects, rather than variance, do not provide consistent patterns. For example, in two recent analyses of substantial data sets of F2 populations of maize, one found substantial epistasis [60] and the other almost none [61]. In an analysis of a range of traits in recombinant inbred lines, F2 and triple test crosses [62] in Arabidopsis thaliana, there was substantial additive genetic and dominance variance for all traits, with most estimates of VD/VA in the range 0.3 to 0.5, essentially no significant additive × additive epistatic effects, but several cases of epistasis involving dominance [63].
Although there does appear to be more dominance variance in populations with gene frequencies of one-half than with dispersed frequencies, from these results we cannot reject or accept the hypothesis that there is relatively much more epistatic variance in such populations. One explanation is indeed that there is not a vast amount of epistatic variance in populations at whatever frequency, although another is that maize has unusually small amounts of epistasis. Many additive QTL were identified in an analysis of a line derived from the F2 of highly divergent high and low oil content lines from the long term Illinois maize selection experiment, but with almost no evidence of epistasis or indeed dominance effects [64]. In contrast, an F2 of divergent lines of long-term selected poultry and an F2 from inbred lines of mice showed evidence of highly epistatic QTL effects for body weight [65],[66]. We do not claim to understand these different results, but as has been pointed out [67],[68], QTL with significant epistatic interaction effects might not represent the majority of QTL with small effects contributing to gene networks.
We have summarised empirical evidence for the existence of non-additive genetic variation across a range of species, including that presented here from twin data in humans, and shown that most genetic variance appears to be additive genetic. There are two primary explanations, first that there is indeed little real dominant or epistatic gene action, or second that it is mainly because allele frequencies are distributed towards extreme values, as for example in the neutral mutation model. Complete or partial dominance of genes is common, at least for those of large effect; and epistatic gene action has been reported in some QTL experiments [8],[69]. Detailed analyses in Drosophila melanogaster, using molecular and genetic tools available for it, identify substantial amounts of epistasis, including behavioural traits [70] and abdominal bristle number [71], yet most genetic variation in segregating populations for bristle number appears to be additive (as noted above). But many QTL studies of epistatic gene action suffer from a high degree of multiple testing, increasingly so the more loci and orders of interaction are included, such that they may be exaggerating the amount of epistasis reported. On the assumption that many of the effects are indeed real, we have turned our attention to the second explanation.
The theoretical models we have investigated predict high proportions of additive genetic variance even in the presence of non-additive gene action, basically because most alleles are likely to be at extreme frequencies. If the spectrum of allele frequencies is independent of which are the dominant or epistatic alleles, VA/VG is large for almost any pattern of dominance and epistasis because VA/VG is low only at allele frequencies where VG is low, and so contributes little to the total VG. The distribution of allele frequencies is expected to be independent of which are the dominant or epistatic alleles for neutral polymorphisms; but under natural selection the favourable allele is expected to be common and lead to high or low VA/VG depending on whether it is dominant (low VA) or recessive (high VA). The equivalent case for epistasis is that all genotype combinations except one is favourable (low VA) vs. only one genotype combination is favourable (high VA).
If genetic variation in traits associated with fitness is due almost entirely to low frequency, deleterious recessive genes which are unresponsive to natural selection, these traits would show low VA/VG. However, neither the empirical evidence nor the theory supports this expectation. There seems to be substantial additive genetic variance for fitness associated traits [21] and fitness itself [30],[31],[72]. Although heritabilities for such traits may be low, they show high additive genetic coefficient of variation (evolvability) [29], and the correlation of repeat records is typically little higher than the heritability (e.g., litter size in pigs), indicating that VA/VG is one-half or more. In agreement with this, when the life history of deleterious, recessive mutants was modelled, VA/VG was found to be 0.44 (Table 6), basically because rare recessives contribute so little variance, albeit most is VD, in non-inbred populations.
We believe we have a plausible gene frequency model to explain the minimal amounts of non-additive genetic and particularly epistatic variance. What consequences do our findings have? For animal and plant breeding, maintaining emphasis on utilising additive variation by straightforward selection remains the best strategy. For gene mapping, our results imply that VA is important so we should be able to detect and identify alleles with a significant gene substitution effect within a population. Such variants have been reported from genome-wide association studies in human population [9],[10],[11],[12],[13]. Although there may well be large non-additive gene effects, the power to detect gene-gene interactions in outbred populations is a function of the proportion of variance they explain, so it will be difficult to detect such interactions unless the effects are large and the genes have intermediate frequency. Thus we expect that the success in replicating reported epistatic effects will be even lower than it is for additive or dominance effects, both because multi-locus interactions will be estimated less accurately than main effects and because they explain a lower proportion of the variance. Finally, if epistatic effects are real, gene substitution effects may vary widely between populations which differ in allele frequency, so that significant effects in one population may not replicate in others. |
10.1371/journal.pgen.1005484 | RAB-10-Dependent Membrane Transport Is Required for Dendrite Arborization | Formation of elaborately branched dendrites is necessary for the proper input and connectivity of many sensory neurons. Previous studies have revealed that dendritic growth relies heavily on ER-to-Golgi transport, Golgi outposts and endocytic recycling. How new membrane and associated cargo is delivered from the secretory and endosomal compartments to sites of active dendritic growth, however, remains unknown. Using a candidate-based genetic screen in C. elegans, we have identified the small GTPase RAB-10 as a key regulator of membrane trafficking during dendrite morphogenesis. Loss of rab-10 severely reduced proximal dendritic arborization in the multi-dendritic PVD neuron. RAB-10 acts cell-autonomously in the PVD neuron and localizes to the Golgi and early endosomes. Loss of function mutations of the exocyst complex components exoc-8 and sec-8, which regulate tethering, docking and fusion of transport vesicles at the plasma membrane, also caused proximal dendritic arborization defects and led to the accumulation of intracellular RAB-10 vesicles. In rab-10 and exoc-8 mutants, the trans-membrane proteins DMA-1 and HPO-30, which promote PVD dendrite stabilization and branching, no longer localized strongly to the proximal dendritic membranes and instead were sequestered within intracellular vesicles. Together these results suggest a crucial role for the Rab10 GTPase and the exocyst complex in controlling membrane transport from the secretory and/or endosomal compartments that is required for dendritic growth.
| Dendrites are cellular extensions from neurons that gather information from other neurons or cues from the external environment to convey to the nervous system of an organism. Dendrites are often extensively branched, raising the question of how neurons supply plasma membrane and dendrite specific proteins from the source of synthesis inside the cell to developing dendrites. We have examined membrane trafficking in the PVD neuron in the nematode worm C. elegans to investigate how new membrane and dendrite proteins are trafficked. The PVD neuron is easy to visualize and has remarkably long and widely branched dendrites positioned along the skin of the worm, which transmits information about harsh touch and cold temperature to the nervous system. We have discovered that a key organizer of vesicle trafficking, the RAB-10 protein, localizes to membrane vesicles and is required to traffic these vesicles that contain plasma membrane and dendrite proteins to the growing PVD dendrite. Further, our work revealed that a complex of proteins, termed the exocyst, that helps fuse membrane vesicles at the plasma membrane, localizes with RAB-10 and is required for dendrite branching. Together, our work has revealed a novel mechanism for how neurons build dendrites that could be used to help repair damaged neurons in human diseases and during aging.
| Dendrites and axons are two distinct functional and morphological domains of neurons. Due to the complexity and heterogeneity in the morphology of dendrites, it has been challenging to study the development of dendrites in comparison with the more uniform, simply structured thread-like axons. Previous studies have shown that dendritic growth relies heavily on the secretory pathway and endosomal function[1,2]. In Drosophila, loss-of-function mutations in genes encoding the small GTPases Rab1 and Sar1, which are key regulators of ER-to-Golgi vesicular transport [3], severely reduce the growth of dendrites. Notably, loss of Rab1 and Sar1 do not diminish axon outgrowth, suggesting that the mechanisms underlying the extension of dendrites and axon are distinct [1]. Furthermore, Golgi outposts, which are primarily found in dendrites but not axons, play an important role in supplying membranes for dendritic branching and growth [1]. These experiments suggest that membrane components generated in the ER and trafficked to the Golgi are essential for dendritic growth. In addition, Rab5 and Rab11-dependent endocytic membrane trafficking has also been implicated in dendrite morphogenesis [2,4]. The molecular mechanisms that deliver membranes from the Golgi, Golgi-outposts and endosomes to the dendritic plasma membrane, however, are unclear.
To identify the membrane trafficking mechanisms that support dendrite branching and growth, we use the C. elegans multi-dendritic PVD neurons as a model. The PVD neurons exist as a pair, PVDL and PVDR, and they function to detect harsh mechanical forces and cold temperatures [5–7]. Each PVD neuron sits on one side of the animal and has a single axon that extends to the ventral nerve cord, as well as a highly branched dendritic arbor that covers most of the body, except for the neck and head [8]. Recently, the transmembrane leucine-rich repeat protein DMA-1 was identified as a PVD dendritic receptor. DMA-1 recognizes skin-derived pre-patterned cues that promote dendrite stabilization and branching [9–11]. In addition, the claudin-like transmembrane protein HPO-30 also promotes dendrite stabilization [12]. Both DMA-1 and HPO-30 are dendrite specific proteins that are rarely observed in axons. The mechanisms that regulate their sorting and trafficking to the dendritic membranes are still unknown [9,12].
The small GTPase Rab10, which is an ortholog to the yeast Sec4p protein that controls post-Golgi vesicle trafficking, has been shown to mediate polarized membrane addition during axonal growth in mammals. In axons Rab10 is activated by the mammalian ortholog of the Drosophila gene Lethal giant larvae, Lgl1. The Lgl1 protein dissociates the Rab10-GDI complex [13]. Activated Rab10 then interacts with multiple effector proteins to direct distinct steps of axonal membrane addition. These include an initial interaction with myosin Vb (MYO5B), which controls the biogenesis of post-Golgi Rab10 carriers [14]. Rab10 then binds with c-Jun N-terminal kinase-interacting protein 1 (JIP1) to facilitate anterograde transport of Rab10 cargos [15]. Finally, Rab10 binds myristoylated alanine-rich C-kinase substrate (MARCKS). The Rab10-MARCKS interaction allows the docking and fusion of Rab10 vesicles with the axonal plasma membrane [16]. Unlike extensive studies on the role of Rab10 in axonal growth, it is unclear whether Rab10 is required for dendrite arborization during development.
Several studies have shown association between the conserved exocyst complex and Rab10 GTPases [17]. The exocyst complex is composed of eight subunits. In C. elegans these are encoded by the genes sec-3, sec-5, sec-6, sec-8, sec-10, sec-15, exoc-7 and exoc-8 [18]. The exocyst complex functions as the effector of the yeast Rab10 ortholog Sec4p, and facilitates tethering, docking and fusion of secretory vesicles during bud formation [17]. The exocyst complex also associates with Rab10 in renal epithelial cells and may mediate membrane transport to the primary cilium [19]. Both Rab10 and the exocyst complex are further required for the exocytic transport of the glucose transporter Glut4 [20,21]. However, whether Rab10 and the exocyst complex function together during other processes, such as dendritic growth and branching, is not clear.
Here, we report that loss of the C. elegans rab-10 gene reduces dendritic arborization of the PVD neuron. We show that RAB-10 functions cell-autonomously, and localizes to the Golgi and the early endosomes in the PVD neurons. Further, we find that deficiencies in rab-10 and the exocyst subunits cause accumulation of the dendritic membrane proteins DMA-1 and HPO-30 within intracellular vesicles. We also show that Rab10 and the exocyst complex are required for dendrite arborization in Drosophila, and dendritic spine formation in mammalian neurons. Together, these data suggest that Rab10 and the exocyst complex play a conserved role in controlling Golgi-to-plasma membrane and/or endosome-to plasma membrane trafficking required for dendrite morphogenesis.
To identify potential regulators of membrane trafficking during dendritic branching and growth of the multi-dendritic PVD neuron, a candidate-based genetic screen was performed. We crossed animals harboring mutations in genes encoding proteins important for various membrane trafficking pathways with a PVD neuron-specific fluorescent marker (F49H12.4>gfp or ser2prom3>gfp). The morphology of PVD dendritic arbors was then examined. We found that two putative null alleles of rab-10 presented severely abnormal PVD dendrite morphology (Table 1 and Fig 1A–1E). In the proximal region (including the middle and tail areas), rab-10 (ok1494) and rab-10 (dx2) mutant animals contained far fewer dendritic branches compared to wild-type. For example, a count of secondary dendrites within a 100μm long region along the primary dendrite anterior to the PVD cell body in rab-10 (ok1494) and rab-10 (dx2) animals revealed an average of 1.1±0.3 and 2.8±0.7 secondary dendrites. In comparison, wild-type animals within this region contained 11.1±0.5 secondary dendrites (Fig 1F and 1G). Tertiary and quaternary branches were even more affected and were essentially absent in the proximal region of rab-10 (ok1494) and rab-10 (dx2) animals (Fig 1F, 1H and 1I). Interestingly, in the distal area of the PVD, the dendritic branching and growth were minimally affected by loss of rab-10, indicating that a rab-10-independent mechanism mediates distal dendritic branching (Fig 1A–1E, 1G–1I). The growth of the primary dendrite and axon of the PVD appeared normal in rab-10 deficient animals (Fig 1A–1E, and S1 Fig). rab-10 was also required for the dendritic arborization of the FLP neuron, which covers the head region and has a similar morphology and function as the PVD neuron (S2 Fig) [8]. However, rab-10 was dispensable for the growth of unbranched dendrites of OLL, AWB, and AWC neurons, suggesting a specific role of rab-10 in mediating the growth of dendritic branches in multi-dendritic neurons (S2 Fig). Together, these data indicate that RAB-10 is required for the elaboration of branched dendrites in C. elegans.
To determine if RAB-10 activity is required within the PVD, we tagged full length RAB-10 at its N-terminus with GFP and expressed this construct under a PVD-specific promoter (ser2prom3) in rab-10(ok194) animals. GFP::RAB-10 expressed from multicopy extrachromosomal arrays in the PVD neuron fully rescued the morphology of the PVD dendritic arbor in most animals in two independent lines (Fig 2A–2C), indicating that RAB-10 functions within the PVD to promote proximal dendritic arborization. As a GTPase, RAB-10 cycles between the GDP-bound inactive form and GTP-bound active form. To test whether its function in promoting dendrite branching and growth requires GTPase activity, we expressed both GDP-locked (T23N) and GTP-locked (Q68L) forms of RAB-10 and examined their rescuing ability. Dominant-negative RAB-10 (T23N) not only failed to rescue the proximal PVD defects in rab-10 (ok1494) mutants, but also disrupted the distal dendrite arbor in wild-type animals (Fig 2C, S3 Fig). In contrast, constitutively active RAB-10 (Q68L) fully rescued the PVD dendrite morphogenesis defects in rab-10 (ok1494) mutant animals (Fig 2C). Over-expressing constitutively active RAB-10 (Q68L) did not cause over-growth of dendrites in PVD neuron (S4 Fig), suggesting that other factors limit dendrite growth and patterning. We conclude that RAB-10 functions cell autonomously in the PVD neuron and requires GTPase activity to promote PVD dendrite morphogenesis.
RAB-10 orthologs are known to regulate Golgi to plasma membrane vesicle trafficking as well as endocytic recycling events [22–26]. To determine if RAB-10 localizes with either Golgi or endosomes in the PVD neuron we tagged full length RAB-10 with GFPnovo2, a mutant form of GFP that is brighter. We generated a single copy insertion line using miniMos method to minimize potential ectopic localization of RAB-10 protein induced by over-expression [27,28]. Similar to the multicopy line, we crossed this line into the rab-10 null mutant and found that it fully rescued the PVD dendrite arborization defect (S5 Fig). Many intracellular vesicles were labeled by GFP::RAB-10 in the PVD dendrites. Notably, there was a strong correlation between GFP::RAB-10 localization and mCherry::FAPP1-PH (Golgi reporter) and a strong correlation of GFP::RAB-10 and mCherry::RAB-5 (early endosome reporter) (Fig 2D and 2E) [29]. These observations are consistent with previous studies showing RAB-10 localization to the Golgi and early endosomes in the C. elegans intestinal cells [24]. Supporting a role in mediating vesicular trafficking, time-lapse recordings revealed that GFP::RAB-10 labeled vesicles moved bi-directionally along dendrites, consistent with localization to transport vesicles (S1 Movie). Together, these data suggest that RAB-10 might mediate PVD dendrite outgrowth by regulating Golgi-to-membrane trafficking or endosomal membrane recycling events.
To attempt to determine whether the PVD dendrite morphogenesis defect in rab-10 mutants might be due to a role for RAB-10 in Golgi-to-plasma membrane trafficking, endocytic recycling, or both, we perturbed each pathway and tested whether it altered PVD dendritic arborization. Consistent with previous studies in Drosophila and rat hippocampus neurons implicating ER-to-Golgi trafficking in dendritic growth, over-expressing dominant-negative RAB-1 (a key regulator of ER-to-Golgi trafficking) caused dramatically reduced branching in the PVD (Table 1, and S6 Fig) [1]. In contrast, genetic loss of rme-1 or chat-1, and over-expression of dominant-negative RAB-5 or RAB-11.1 (all key regulators of endocytic recycling) [30–33], had no effect of PVD dendrite morphology (Table 1). Together, these data suggest a role for RAB-10 in mediating Golgi-to-plasma membrane trafficking that is important for dendrite morphogenesis. Notably, however, we cannot rule the possibility that RAB-10 regulates endosomal recycling independent of rme-1, rab-5 or rab-11.1. Thus, RAB-10 might also regulate an endosome-to-plasma membrane transport pathway independent of RME-1, RAB-5, RAB-11.1 function in the PVD neuron that is required for dendritic arborization.
Rab GTPases are molecular switches that exert their functions by recruiting and releasing specific effectors [34]. We next sought to identify possible effectors that function with RAB-10 in mediating the branching and growth of PVD dendrites. We first examined EHBP-1, an Eps 15 domain binding protein that acts as an effector of RAB-10 in endocytic recycling and secretory pathways [26]. Mutant animals of ehbp-1, however, had normal PVD dendritic arbors (Table 1). Importantly, we cannot fully exclude the possibility that the PVD dendrite development was rescued by maternally loaded ehbp-1 mRNA or EHBP-1 protein, since we can only examine ehbp-1 homozygous animals derived from heterozygous mothers. We next tested members of the exocyst complex, an established effector of yeast Sec4p with which RAB-10 shares high homology [35]. Notably, loss of the exocyst subunit exoc-8 in viable null mutant animals or loss of sec-8 in the mutant progeny of heterozygous sec-8 mutant mothers, caused a PVD dendrite morphogenesis defect in the proximal but not in the distal region (Table 1, and Fig 3A–3F)[36]. Further, the growth of the primary dendrite and the axon was normal (Fig 3A–3C, S1 Fig). This phenotype was similar to rab-10 mutants, although the growth of secondary dendrites was only mildly affected in exoc-8 and sec-8 mutant worms (Fig 3A–3F). Expression of exoc-8 cDNA under the PVD-specific promoter (ser2prom3) fully rescued the dendritic arborization defect in exoc-8 (ok2523) mutant worms, indicating that the exocyst complex functions in the PVD to promote dendritic arborization (Fig 3G).
Homozygous mutants of two other exocyst components, sec-5 and sec-10, derived from the heterozygous mothers did not show any obvious PVD dendrite defect (Table 1). To test whether this was due to maternal rescue, we used a newly developed targeted protein degradation system to specifically remove the SEC-5 protein from the PVD [37]. This system takes advantage of cell type specific expression of the E3 ubiquitin ligase substrate-recognition subunit ZIF-1, which recognizes proteins tagged with the 36 amino acid ZF1 zinc-finger domain. To selectively remove SEC-5 in the PVD neuron, we drove ZIF-1 in PVD using the ser2prom3 promoter in a sec-5::zf1::yfp knock-in strain (sec-5(xn51))—a strain where both copies of the endogenous sec-5 genes are tagged with zf1 (Fig 4A) [37]. Confirming a role for the exocyst complex in promoting dendritic arborization, depleting ZF1::YFP tagged SEC-5 in PVD resulted in PVD dendrite arborization defects (Fig 4C–4E). In some of these animals, both distal and proximal regions lacked menorah structures, suggesting that sec-5 might be important for the growth of both proximal and distal dendrites (Fig 4D). Control animals expressing ser2prom3>ZIF-1 transgene alone and control animals expressing ZF1::YFP tagged SEC-5 alone had normal PVD dendrite morphology (Fig 4B and 4E). We conclude that the exocyst complex plays a cell-autonomous role in promoting dendritic arborization and that maternal contributions of some of exocyst components contribute to this function. Further, based on the more severe PVD phenotype after ZF-1 tag-directed loss of sec-5 versus the exoc-8 null mutant, our results also suggest that exocyst components have different requirements (perhaps reflecting their differential necessity for exocyst activity) during PVD dendrite morphogenesis.
Yeast Sec4p and the exocyst complex function together to promote docking and possible fusion of post-Golgi vesicles [17]. Thus, we hypothesized that the PVD dendritic arborization defects in rab-10, exoc-8 and sec-8 mutants might in part be due to the failure of docking of post-Golgi vesicles. To test this, we built fluorescent reporters for the dendritic membrane proteins, DMA-1 and HPO-30, transmembrane proteins expressed in PVD neurons that function to mediate dendritic branching and stabilization [9–12]. Consistent with previous studies, DMA-1::GFP and HPO-30::GFP were localized in the dendritic membranes, and in some intracellular vesicles in wild-type animals (Fig 5A–5C, and S7 Fig) [9,12]. These two transgene reporters likely represent endogenous protein localization, as the transgenes rescued the PVD dendrite arborization defects of dma-1(tm5159) and hpo-30(ok2147) mutants (S5 Fig).
Consistent with a role in docking and fusion of vesicles, loss of rab-10 and exoc-8 caused severe accumulation of DMA-1::GFP containing vesicles within the growing PVD dendrites (Fig 5D–5I). We quantified the number of vesicular units (which might be a single vesicle or a vesicle cluster) in three areas—the distal, middle and proximal regions. We found that wild-type animals contained on average 29.6±1.9 vesicular units in a 76.8μm x 76.8 μm area from the middle region of the PVD. In contrast rab-10 (ok1494) and exoc-8 (ok2523) mutant worms contained a dramatic two-to-three fold increase in vesicular units—65.6±2.8 and 87.8±6.8, respectively (Fig 5J). Furthermore, in wild-type animals the DMA-1::GFP containing vesicles mainly localized to the primary dendrites, and rarely appeared in the higher-order dendrites (including the secondary, tertiary and quaternary dendrites; Fig 5A–5C). In contrast, in rab-10 (ok1494) and exoc-8 (ok2523) mutant worms, numerous vesicles appeared in the higher-order dendrites (Fig 5D–5I and S8 Fig).
The presence of numerous DMA-1::GFP intracellular vesicles in rab-10 and exoc-8 mutants suggested that delivery of DMA-1 to the membrane might be reduced. To test this idea, the fluorescence intensity of DMA-1::GFP at the surface (which we presume is predominantly plasma membrane localization) of the primary dendrites was determined. In the distal region of the dendritic arbor, the intensity of DMA-1::GFP at the primary dendrite in rab-10 (ok1494) and exoc-8 (ok2523) mutant worms was decreased by 63.9% and 47.8% compared with wild-type animals (Fig 5K). In the rab-10 mutants, reduction of dma-1 by RNAi-mediated knock-down further suppressed distal dendritic arborization, suggesting that the reduced levels of DMA-1 on the surface of distal dendrites is sufficient to promote dendritic stabilization and branching (S9 Fig). In the proximal region, the decrease in DMA-1 was more dramatic. The intensity of DMA-1::GFP at the surface of the primary dendrite in rab-10 (ok1494) and exoc-8 (ok2523) mutant worms was decreased by 89.5% and 90.3% in the middle regions, and 94.4% and 94.8% in the tail regions compared to wild-type animals (Fig 5K). HPO-30::GFP showed similar vesicular accumulation and decreased dendrite surface localization in rab-10 (ok1494) and exoc-8 (ok2523) mutant animals (S7 Fig). These results offer compelling evidence that rab-10 and exocyst activity are required for the vesicular delivery of DMA-1 and HPO-3 to the dendritic membrane.
The similar phenotypes after loss of exocyst components and rab-10, as well as functions of these molecules in yeast and vertebrates in delivering vesicles to the plasma membrane [17,19,35,38,39] led us to examine whether RAB-10 and exocyst components coexist in vesicles in the PVD dendrites. EXOC-8::GFP and mCherry::RAB-10 strongly colocalized on intracellular vesicles (Fig 6A), indicating that the exocyst might function together with rab-10 to mediate vesicle delivery. Next, we examined the subcellular localization of RAB-10 in animals harboring mutations in the exocyst components exoc-8 and sec-8. Loss of these exocyst components caused a dramatic accumulation of RAB-10 labeled vesicles in the PVD dendrites (Fig 6B). Wild-type animals contained 39.8±2.6 GFP::RAB-10 vesicular units in a 88.1μm x 88.1 μm area from the middle region of the PVD neuron. Worms with mutations in exoc-8 (ok2523) and sec-8 (ok2187) contained over three fold more GFP::RAB-10 vesicular unites—129.7±7.5 and 127.6±2.7, respectively (Fig 6C). To determine whether the accumulated vesicles containing GFP::RAB-10 were the same population observed in exoc-8(ok2523) mutants carrying DMA-1::GFP or HPO-30::GFP, we expressed DMA-1::GFP or HPO-30::GFP and mCherry::RAB-10 in exoc-8(ok2523) mutant animals. Confirming these are the same vesicle population, most of the mCherry::RAB-10 labeled vesicles also contained DMA-1::GFP or HPO-30::GFP (Fig 6D and 6E). To test whether RAB-10 functions to recruit EXOC-8 onto vesicles, the EXOC-8::GFP reporter was crossed into the rab-10 loss-of-function mutant. EXOC-8::GFP, however still localized to vesicles in PVD neurons in rab-10 mutants (S10 Fig). Taken together, these results suggest the exocyst complex promotes fusion of RAB-10 carriers within the PVD neuron to facilitate dendritic growth and stabilization, but that the exocyst complex is recruited to these vesicles in a RAB-10 independent manner.
To investigate whether Rab10 and exocyst complex-mediated dendritic membrane transport is required for the dendrite morphogenesis in other organisms, we examined Drosophila class IV dendritic arborization neurons and cultured rat hippocampal neurons after loss of rab10 and exocyst components. In Drosophila, RNAi mediated knock-down of rab10 and exo84 significantly reduced dendrite branching and growth (S11 Fig). Compared to animals treated with control RNAi, rab10 and exo84 RNAi targeted loss resulted in a 20% decrease in both the number of total end points and the total dendritic arbor (S11 Fig). ShRNA mediated knock-down of rab10, sec8 and exoc84 did not alter the total dendrite length in cultured rat embryonic hippocampal neurons, but did lead to a dramatic 67% reduction in the density of dendritic spines at 21 days in vitro (DIV) compared to control shRNA treated neurons (S12 Fig). Collectively, we conclude that rab10 and the exocyst complex have important and likely conserved roles during dendrite morphogenesis.
Since some higher-order dendrites in the PVD still formed in rab-10 null alleles, we hypothesized that another Rab protein might regulate dendritic branching. To test this idea, we examined the function of the RAB-10 related GTPase RAB-8 [40]. Notably, homozygous viable rab-8(tm2526) null mutant animals showed normal PVD morphology (Fig 7A). This suggested that if RAB-8 functions in the PVD, it might act redundantly with RAB-10. To test this idea, we first attempted to create animals with null mutations in both rab-8 and rab-10. The rab-8 and rab-10 double mutant animals, however, were sterile, which made it challenging to determine PVD morphology [26]. Interestingly, we found that expression of a dominant-negative RAB-8 (T22N) in the PVD neuron caused a severe dendrite morphogenesis defect (S3 Fig). In 15% (n = 40) and 12% (n = 50) of transgenic animals (two independent lines) harboring the transgene of dominant negative RAB-8, the dendritic growth in both distal and proximal regions was dramatically reduced (S3 Fig). These results suggest that the dominant negative RAB-8 might act to block the function of both the RAB-8 and RAB-10 proteins (perhaps through inhibition of a common guanine nucleotide exchange factor) and that RAB-8 may function redundantly with RAB-10 to promote PVD morphogenesis.
To more directly test the idea that RAB-10 and RAB-8 function together in regulating PVD morphogenesis, we took advantage of a newly developed CRIPSR/Cas9-mediated conditional knock-out method and specifically disrupted the function of the rab-10 gene in the PVD neuron and other descendants of the seam cell lineage by restricting Cas9 endonuclease expression using nhr-81 promoter (Pnhr-81>Cas9)[41,42]. In three separate lines, approximately 5–10% of animals (11.6% (35/303), 5.1% (4/78) and 4.9% (3/61)) targeted with conditional PVD knock-out of rab-10 (which we refer to as rab-10(cKO)) generated a PVD phenotype. Animals displaying a PVD dendrite arborization defect, showed a similar phenotype as that of rab-10 null mutants (Fig 7B). These results suggest that the CRIPSR/Cas9-mediated conditional knock-out only disrupts the rab-10 gene in the PVD lineage in a small percentage of transgenic animals, but that when it does target rab-10, it completely perturbs rab-10 function. We crossed the most penetrant rab-10(cKO) line into the rab-8 deletion mutant. We hypothesized that if rab-8 functions redundantly with rab-10, it should enhance the rab-10(cKO) PVD phenotype. From 302 animals that carried the Pnhr-81>Cas9 and PU6>rab-10-sgRNAs transgene, 27 animals (8.9%) showed severely defective PVD dendritic arbors. We measured the total number of secondary dendrites, and found that the rab-10(cKO) rab-8(tm2526) animals had fewer secondary dendrites than rab-10(cKO) alone (Fig 7E). Further, we observed that 55.6% (15/27) of rab-10(cKO) rab-8(tm2526) animals had truncated posterior primary dendrites, which was rarely observed in rab-10(cKO) or rab-8(tm2526) strains (2.8% (1/35) and 0% (0/25), respectively) (Fig 7F). Taken together these results suggest that the related GTPases RAB-8 and RAB-10 function redundantly to promote dendritic morphogenesis in the PVD neuron.
In this study, we used the two multi-dendritic PVD neurons (PVDL and PVDR) in C. elegans, as a model system for studying dendritic arborization. We show that the small GTPase RAB-10 is required for the growth and branching of higher-order dendrites in PVD neurons. RAB-10 localizes to Golgi and early endosomes, and its loss resulted in severe dendrite arborization defects in the proximal region of PVD neurons. Furthermore, we found that mutations in several exocyst complex components, resulted in a similar dendrite morphogenesis defect. We propose that the exocyst complex is an effector of RAB-10 and promotes docking and fusion of secretory vesicles and/or recycling endosomes, which is crucial for dendritic growth and branching (Fig 8).
Previous work in Drosophila dendritic arborization neurons identified rab1, sec23 and sar1, three genes that mediate ER-to-Golgi traffic, as essential regulators of membrane addition during dendritic growth [1]. This study also demonstrated that laser ablation of Golgi-outposts within the dendrites reduced dendrite arborization. Together this work strongly implicated ER-to-Golgi trafficking as well as Golgi outposts as essential for dendrite morphogenesis. These findings left open the question, however, of how membrane trafficking from the Golgi to the dendritic plasma membrane is regulated. Through a candidate-based genetic screen and analysis of the C. elegans multidendritic PVD neuron we have identified the small GTPase Rab10 as a post-Golgi regulator of vesicle trafficking required for dendrite morphogenesis.
Rab10 is known to regulate both Golgi-to-plasma membrane vesicle trafficking and endocytic recycling [13,22,24–26,43–45]. Our data suggest that the C. elegans RAB-10 protein primarily promotes post-Golgi trafficking in the PVD neuron to facilitate dendritic growth and branching. We did not observe any obvious PVD dendrite morphogenesis defects after inhibiting numerous important mediators of endocytic recycling. This includes rme-1 and chat-1 mutant animals and animals carrying transgenes for dominant-negative rab-5 or rab-11.1 GTPases [30,31]. In contrast, animals expressing a dominant-negative rab-1, which regulates ER-to-Golgi trafficking, showed dramatically reduced PVD dendrite branching and growth. Furthermore, animals harboring exoc-8 and sec-8 mutations, which encode two subunits of the exocyst complex that function to dock secretory vesicles onto fusion sites on the plasma membrane, led to PVD dendrite morphogenesis defects similar to rab-10 [17]. These data strongly implicate a function for the RAB-10 protein in mediating membrane trafficking from the secretory pathway that is crucial for dendritic branching in the PVD. Since we cannot rule out the possibility that RAB-10 might also regulate recycling of endosomes in a RME-1/RAB-5/ RAB-11.1-independent manner, we suggest it is possible that RAB-10 might also contribute to PVD morphogenesis by regulating endocytic trafficking (see Fig 8).
Although loss of rab-10 resulted in a dramatic reduction in dendrite arborization in the PVD and FLP neurons, the absence of rab-10 function did not affect the growth of the PVD primary dendrites or the unbranched dendrites of the OLL, AWB and AWC neuron. This might reflect a heavy reliance on post-Golgi trafficking in the PVD and FLP neurons to supply the dramatic expansion in membrane required to form highly branched dendrites. In support of this idea, a recent study in Drosophila revealed that Rab10 and exocyst-mediated membrane trafficking is crucial for the elaborate branching of tracheal terminal cells [44]. It is also possible that rab-10 may transport cargoes that are necessary for the branching and growth of higher-order dendrites. Consistent with this notion, loss of rab-10 severely affected the dendritic transport of DMA-1 and HPO-30, two dendritic specific transmembrane proteins that are required for the branching and stabilization of higher-order dendrites.
The exocyst complex is a well-characterized effector of yeast Sec4p, which shares high sequence homology with RAB-10 [26]. In yeast the exocyst complex functions to target Sec4p secretory vesicles to sites of exocytosis [35]. Suggesting a similar mechanism in the PVD dendrite, we found that loss of the exocyst components exoc-8 and sec-8 caused a dendrite morphogenesis defect that was similar to rab-10 mutants. In animals carrying mutations in these genes, GFP tagged DMA-1 and HPO-30 were sequestered in intracellular vesicles and the dendrite surface expression was greatly reduced. The similarity of the phenotypes suggests that the exocyst complex functions as an effector of RAB-10 to mediate trafficking of secretory vesicles to the dendritic plasma membrane. This idea is further supported by the colocalization of RAB-10 and EXOC-8 on intracellular vesicles in the PVD dendrites. Notably, compared to the PVD dendrite morphogenesis defects observed in the rab-10 mutants, the dendrite phenotypes in exoc-8 and sec-8 mutants were not as severe. This could indicate that another mechanism mediates tethering, docking or fusion of RAB-10 cargo vesicles to the dendritic plasma membrane. However, it is likely that the exoc-8 and sec-8 mutant animals examined were not nulls for exocyst function. While the deletion allele of exoc-8(ok2523) is thought to completely remove exoc-8 function [36], it does not lead to lethality, which is associated with loss of all other exocyst subunits. These observations suggest that loss of exoc-8 does not completely eliminate exocyst function [46]. Further, the sec-8(ok2187) homozygotes were obtained from heterozygous mothers that likely provided maternally loaded sec-8 mRNA or SEC-8 protein. Consistent with this idea, the removal of the exocyst component SEC-5 through ZF-1 tag-mediated degradation often resulted in a complete loss of dendritic branching. We thus suggest that the exocyst complex is the primary effector of RAB-10. Further, given that loss of SEC-5 lead to a more severe dendrite arborization defect than loss of RAB-10, it is likely the exocyst complex has functions outside of regulating RAB-10 vesicle trafficking (possibly RAB-8 vesicle trafficking, see below).
In rab-10 mutants, the growth and branching of more distal dendrites was minimally affected, suggesting that a rab-10 independent mechanism exists. Notably, expressing dominant negative RAB-10 in PVD caused both distal and proximal dendrite morphogenesis defects. A possible explanation is that dominant negative RAB-10 competes for RAB activators, such as a guanine nucleotide exchange factor (GEF), that interferes with another RAB protein that is essential for distal dendrites. A strong candidate for this other RAB protein is RAB-8. Although loss of rab-8 had no obvious effect on PVD dendrite morphogenesis, expression of a dominant negative RAB-8 in the PVD caused similar dendritic phenotype to the dominant negative RAB-10. rab-8 is related to rab-10, and rab-8 and rab-10 function redundantly in mediating secretion in C. elegans germ cells [26]. Using a newly developed CRIPSR/Cas9-mediated conditional knock-out method, we found that conditional knockout of rab-10 in a rab-8 null mutant showed stronger dendrite arborization defect than loss of rab-10 alone. These observations offer compelling evidence that RAB-10 and RAB-8 function redundantly to regulate PVD dendritic arborization.
RAB-10 and the exocyst complex are evolutionarily conserved from C. elegans to humans. We observed that knock-down of rab10 or exo84 caused a significant reduction of the dendrite arbor in the Drosophila class IV dendritic arborization neuron. Knock-down of rab10, exoc84 or sec8 had no effect on the total length of the dendrites in rat cultured hippocampal neurons, but the number of dendritic spines was greatly reduced. Compared to the PVD dendrite phenotypes of C. elegans rab-10 mutants, knock-down of rab10 in Drosophila and rat neurons showed weaker dendrite morphogenesis defects. This might be due to the use of null alleles of rab-10 in our studies of C. elegans compared to the RNAi strategy that led to reductions but not complete loss of rab10 function in Drosophila and rat. Consistent with this, knock-out of sec5 in Drosophila neurons causes more severe neurite outgrowth defect [47]. Nevertheless, our results imply that rab-10 and exocyst-dependent membrane transport is a conserved mechanism used to build dendritic arbors and dendritic spines during neural development.
Dendrite morphogenesis defects are associated with many neurological and neurodevelopment disorders, including Autism spectrum disorders, Alzheimer’s disease and Parkinson’s disease [48,49]. Recovery from these diseases will rely on efficient dendrite regrowth and branching to form functional neuronal circuits. Here we identified RAB-10 and the exocyst complex as critical and conserved players during dendrite morphogenesis. Our findings reveal an evolutionarily conserved membrane transport mechanism that can efficiently supply membranes and newly synthesized transmembrane proteins to support rapid dendrite growth and morphogenesis. These findings may help in the development of new therapeutic strategies to help repair damaged neurons in human diseases and during aging.
All animals were euthanized in accordance with the recommendations of the American Veterinary Medical Association and the UCSF Institutional Animal Care and Use Committee. Rats were euthanized with CO2 prior to dissection. After treatment with CO2, bilateral thoracotomy was performed to ensure the death of the animal. All experimental procedures used comply with regulations adopted by UCSF authorities with support from the Guidelines on Euthanasia of the American Veterinary Medical Association.
C. elegans was grown on OP50 Escherichia coli-seeded nematode growth medium agar plates at 20℃ unless otherwise noted. Alleles used are as follows: Linkage group I (LGI): rab-10 (ok1494), rab-10 (dx2), exoc-8 (ok2523), sec-8 (ok2187), exoc-7 (ok2006), rab-8 (tm2526) and dma-1(tm5159). LGII: sec-5 (tm1413), sec-5(xn51), and rrf-3(pk1426). LGIV: sec-10 (tm3437), and chat-1 (ok1681). LGV: ehbp-1 (ok2140), rme-1 (b1045) and hpo-30(ok2047). Primer sequences for genotyping are available on request. RNAi experiments were performed as previously described [50], except that the rrf-3(pk1246) mutation was used to enhance the RNAi efficacy in the PVD neuron. The dma-1 RNAi clone was obtained from Vidal RNAi feeding library and verified by DNA sequencing [51].
Standard germ line transformation by gonadal micro-injection was used to generate transgenic lines. Plasmid DNAs and fusion PCR products were used at 1–20 ng/ μl, and co-injection marker unc-119 (+) or Pmyo-2>mcherry or Pmyo-3>mcherry or Punc-122>rfp or Podr-1>gfp was used at 1–30 ng/ μl. Chromosome integrated stable lines were generated by following gamma ray irradiation protocol [52]. Single copy transgene was generated by following a miniMos-based protocol [28]. Transgenes used are listed in S1 Table.
All the plasmids were constructed using standard molecular cloning methods. pPD49.26 and pPD95.75 were used as vectors. Coding regions of rab-1, rab-8 and rab-11.1 with dominant-negative mutations were amplified from a RAB toolkit [53]. Fusion PCR was performed as previously described [54]. Plasmids and fusion PCR product used are listed in S2 Table. Primers used are listed in S3 Table.
pWZ243 Pnhr-81>Cas9 (20ng/ul), pWZ170 PU6>rab-10-sgRNA #1 (target DNA sequence was 5’GAAGAGCATGTCATACGGT3’) (20ng/ul), pWZ171 PU6>rab-10-sgRNA #2 (target DNA sequence was 5’GCAATTTGAAGAGCATGTCATA3’) (20ng/ul), Pmyo-2>mcherry (1ng/ul) and Pmyo-3>mcherry (5ng/ul) were injected into wyIs592 (ser2prom3>myr-gfp) worms. Transgenic lines were identified and maintained based on fluorescence of the mCherry co-injection markers expressed in the pharyngeal muscles and body wall muscles. L4 and young adult stage transgenic animals were quantified for PVD dendrite arborization defects.
To quantify the percentage of animals with abnormal PVD dendritic morphology (Table 1), mid-L4 to young adult stage hermaphrodite animals were anesthetized using 1mg/ml levamisole in M9 buffer, mounted on 2% agar pads and examined for PVD morphology using a compound fluorescence microscope (Carl Zeiss) with a 63X/1.4NA objective lens. Wild-type animals develop many menorah-like structures with higher-order dendrites. An animal was considered abnormal if it lacked more than four menorahs (either by lacking quaternary dendrites, tertiary dendrites or secondary dendrites) or if it had truncated primary dendrites.
To quantify PVD dendrite and axon morphology, z-stack images of PVD neurons were taken of mid- or late L4 stage animals using either a spinning disk confocal microscope (Axio Imager; Carl Zeiss) with a 40x objective lens (1.4 NA) equipped with an EM charge-coupled device (CCD) camera (Hamamatsu Photonics) and a spinning disc confocal scan head (CSU-10; Yokogawa Electric Corporation) controlled by Micro-Manager software with 488 and 561 laser lines (for images showed in Figs 1A–1C, 2A–2B, 3A–3B, and S2A–S2D and S3A–S3B), or using a Zeiss LSM710 confocal microscope (Carl Zeiss) with a Plan-Apochromat 40X/1.3NA objective (for images showed in Figs 4B–4D, 7A–7D, S2E–S2F, S3C–S3D, S4A–S4B, S5A–S5F, S6A–S6C and S9A–S9B). The numbers of secondary, tertiary and quaternary dendrites were quantified using maximum intensity projections generated from z-stack images using ImageJ.
Mid-L4 stage hermaphrodite animals were anesthetized using 1mg/ml levamisole in M9 buffer, mounted on 2% agar pads, and then single focus plane images for GFP::RAB-10 labeled vesicles in the primary dendrites were captured using a spinning disk confocal microscope with a 100x objective lens (1.4 NA). Images were taken 1 frame per 0.8 second for 1.6 minutes (120 frames). Movies were made using ImageJ.
Mid-L4 stage transgenic animals were anesthetized using 1mg/ml levamisole in M9 buffer, mounted on 2% agar pads and dual color images were collected using a spinning disk confocal microscope (Axio Imager; Carl Zeiss) with either a 63x (for images showed in Figs 2D–2E and 6D–6E) or 100x (for images showed in Fig 6A) objective lens (1.4 NA). Colocalization between RAB-10 and RAB-5, FAPP1-PH and EXOC-8 was quantified using Coloc 2, a Fiji’s plugin for colocalization analysis (http://fiji.sc/Coloc_2). The Pearson correlation coefficient index is shown for each group.
Z-stacks of ser2prom3>DMA-1::GFP and ser2prom3>HPO-30::GFP fluorescence in distal, middle and tail region of mid-L4 stage worms were taken using a spinning disk confocal microscope (Axio Imager; Carl Zeiss) with a 100x Plan-Aprochromat objective lens (1.4 NA). Fluorescence across the primary dendrites was quantified using the “measure” function of ImageJ software from single focal plane images. Background subtraction levels were determined from regions outside of the worms, lacking any GFP signal.
Z-stacks of ser2prom3>DMA-1::GFP and ser2prom3>HPO-30::GFP fluorescence in distal, middle and tail region of mid-L4 stage worms were taken using a spinning disk confocal microscope (Axio Imager; Carl Zeiss) with a 100x Plan-Aprochromat objective lens (1.4 NA). Number of vesicular units (either single vesicles or vesicle clusters) was quantified from images of maximum intensity projections generated from z-stack images using ImageJ. The size of each image was 76.8μm x 76.8 μm.
Z-stacks of ser2prom3>GFP::RAB-10 fluorescence in middle region of mid-L4 stage worms were taken using a spinning disk confocal microscope (Axio Imager; Carl Zeiss) with a 63x Plan-Aprochromat objective lens (1.4 NA). Number of vesicular units (either single vesicles or vesicle clusters) was quantified from maximum intensity projections generated from z-stacks using ImageJ. The size of each image was 88.1μm x 88.1 μm.
The UAS-Gal4 system was used to express RNAi in the class IV neurons of Drosophila larvae. UAS-RNAi lines are available from Bloomington Stock Center: y[1] v[1]; P{y[+t7.7] v[+t1.8] = TRiP.JF02058}attP2 (BL 26289), y[1] v[1]; P{y[+t7.7] v[+t1.8] = TRiP.JF03139}attP2 (BL 28712) and from the Vienna Drosophila Resource Center: w[1118]; P{GD16778}v46791/CyO; (v46791), w[1118]; P{GD11816}v30112/TM3; (v30112). UAS- RNAi lines were crossed to ppktdGFP; ppkgal4, UAS-dcr2 virgins. Larvae resulting from this cross were imaged at third larval instar. The dendritic arbor was quantified as described [55].
Hippocampal neurons cultured from E19 Long-Evans rats (Charles River Lab) were plated at a density of 200,000 neurons per 18 mm acid treated glass coverslips (Fisher) coated with 0.06 mg/ml poly-D-lysine (Sigma) and 2.5 ug/ml Laminin (Sigma). Neurons were plated using plating media (Modified Eagle Medium + 10% Fetal Bovine Serum (Hyclone), 0.45% dextrose, 0.11 mg/ml sodium pyruvate, 2mM glutamine, Reagents from UCSF cell culture facility). Cultures were transferred to maintenance media (Neurobasal Media, Invitrogen+ 0.5mM Glutamine+1X B27, Invitrogen and Penicillin/Streptomycin) 4 hours post plating. Half of the media was replaced with fresh media every 4 days.
All shRNA constructs were made in the pLentilox3.7 vector backbone using TTCAAGAGA as the loop sequence. Sequences used for constructing the shRNAs are: 1) Rab10 (NM_017359.2 Start position: 2373) “TTGACTCTATCATTGTTTA” 2) Exoc84 (NM_139043.1 Start position: 310) “CGCAGAACCTGAAGCGCAA” 3) Sec8 (NM_053875.1 Start position: 887) “CCGTTAAAGCCATTAAAGA”. ShRNA transfections were performed using Lipofectamine-2000 (Invitrogen) following manufacturer’s guidelines. Transfections were done at DIV7 and DIV14 and fixed 48 hours post transfection at DIV9 and DIV16, respectively using 4% PFA+4% sucrose at room temperature for 15 min. We did not observe toxicity or neuronal death in any of the shRNA treatment conditions. Neurons were blocked in blocking buffer (10% normal donkey serum+ 0.2M glycine+ 0.1% triton-x100 in Phosphate buffer saline) for an hour followed by overnight incubation with primary antibodies (mouse anti GFP, from Roche and rabbit anti-MAP2 from Chemicon). Coverslips were washed thrice in PBS and incubated for 2 hours with secondary antibodies (mouse Alexa 488 and Rabbit Alexa 568, from Jackson ImmunoResearch) followed by three washes in PBS before mounting onto slides for analysis using Fluromont mounting media from EMS).
Imaging of rat hippocampal neurons was performed on Leica Sp5 scanning confocal microscope with Laser lines 488 and 561 using a 40X NA 1.25 objective with zoom 0 for dendrite length and zoom 3.0 for dendritic spine imaging. Confocal image stacks were acquired at 1024X1024 pixels with 0.4micron z spacing between two frames such that the entire depth of the neuron/dendrite was imaged. Analysis was performed on maximum projected image stacks using ImageJ. ShRNA transfected cells were identified by GFP expression. Neurite length was calculated using the ‘Measure’ function of ImageJ and dendrite density was calculated by manually counting the number of spines per 100 microns of dendrite length.
One-way ANOVA followed by post-hoc comparisons using the Dunnett’s test was used for all the figures, except S11 Fig (one-way ANOVA followed by post-hoc comparisons using the Holm-Sidak test was used in this figure). Error bar means standard error of mean (SEM) for all the figures except S11 Fig (error bar means standard deviation in this figure).
|
10.1371/journal.ppat.1006784 | The role of host DNA ligases in hepadnavirus covalently closed circular DNA formation | Hepadnavirus covalently closed circular (ccc) DNA is the bona fide viral transcription template, which plays a pivotal role in viral infection and persistence. Upon infection, the non-replicative cccDNA is converted from the incoming and de novo synthesized viral genomic relaxed circular (rc) DNA, presumably through employment of the host cell’s DNA repair mechanisms in the nucleus. The conversion of rcDNA into cccDNA requires preparation of the extremities at the nick/gap regions of rcDNA for strand ligation. After screening 107 cellular DNA repair genes, we herein report that the cellular DNA ligase (LIG) 1 and 3 play a critical role in cccDNA formation. Ligase inhibitors or functional knock down/out of LIG1/3 significantly reduced cccDNA production in an in vitro cccDNA formation assay, and in cccDNA-producing cells without direct effect on viral core DNA replication. In addition, transcomplementation of LIG1/3 in the corresponding knock-out or knock-down cells was able to restore cccDNA formation. Furthermore, LIG4, a component in non-homologous end joining DNA repair apparatus, was found to be responsible for cccDNA formation from the viral double stranded linear (dsl) DNA, but not rcDNA. In conclusion, we demonstrate that hepadnaviruses utilize the whole spectrum of host DNA ligases for cccDNA formation, which sheds light on a coherent molecular pathway of cccDNA biosynthesis, as well as the development of novel antiviral strategies for treatment of hepatitis B.
| Hepadnavirus cccDNA is the persistent form of viral genome, and in terms of human hepatitis B virus (HBV), cccDNA is the basis for viral rebound after the cessation of therapy, as well as the elusiveness of a cure with current medications. Therefore, the elucidation of molecular mechanism of cccDNA formation will aid HBV research at both basic and medical levels. In this study, we screened a total of 107 cellular DNA repair genes and identified DNA ligase 1 and 3 as key factors for cccDNA formation from viral relaxed (open) circular DNA. In addition, we found that the cellular DNA ligase 4 is responsible for converting viral double-stranded linear DNA into cccDNA. Our study further confirmed the involvement of host DNA repair machinery in cccDNA formation, and may reveal new antiviral targets for treatment of hepatitis B in future.
| Hepadnavirus specifies a group of hepatotropic viruses that carry a single copy of the partially double stranded relaxed circular (rc) viral DNA genome in the enveloped virion particle [1]. Hepadnavirus infects mammalian and avian hosts with strict species-specific tropism, including human hepatitis B virus (HBV) and duck hepatitis B virus (DHBV) [2]. It is estimated that HBV has infected 2 billion people globally, resulting in more than 250 million chronically infected individuals who are under the risk of cirrhosis and hepatocellular carcinoma (HCC) [3, 4].
Upon infection of an hepatocyte, the hepadnaviral rcDNA genome is delivered into the nucleus and converted into an episomal covalently closed circular (ccc) DNA, which exists as a minichromosome and serves as viral mRNA transcription template [5, 6]. One mRNA species, termed pregenomic (pg) RNA, is packaged into the cytoplasmic nucleocapsid, where the viral polymerase reverse transcribes pgRNA into viral minus strand DNA, followed by asymmetric plus strand DNA synthesis to yield the major rcDNA genome or a minor double stranded linear (dsl) DNA form [7]. The mature nucleocapsid either acquires viral envelope proteins for virion egress, or recycles the viral DNA to the nucleus to replenish the cccDNA reservoir [8]. Therefore, cccDNA is an essential component of the hepadnavirus life cycle for establishing a persistent infection, and cccDNA elimination is an undisputed ultimate goal for a cure of hepatitis B [9]. However, the available drugs for treatment of chronic hepatitis B are rarely curative due to their failure to eliminate cccDNA [10]. Therefore there is an urgent unmet need to fully understand HBV cccDNA biology and develop novel effective treatments to directly target cccDNA formation and maintenance [11, 12].
Unlike the episomal circular genomes of other DNA viruses, such as papillomaviruses and polyomaviruses [13, 14], HBV cccDNA does not undergo semiconservative replication, but is mainly converted from rcDNA [1]. The molecular mechanism by which rcDNA is converted into cccDNA remains obscure. Comparing the major differences between rcDNA and cccDNA (Fig 1), a series of well-orchestrated biological reactions are required to cope with the terminal molecular peculiarities of rcDNA during cccDNA formation, including: 1) completion of viral plus strand DNA synthesis; 2) removal of the 5’-capped RNA primer at the 5’ terminus of plus strand DNA; 3) removal of viral polymerase covalently attached to the 5’ end of minus strand DNA; 4) removal of one copy of the terminal redundancy on minus strand DNA; 5) ligation of both strands to generate the wildtype cccDNA [5, 9]. Previous studies by us and others have identified a rcDNA species without the viral polymerase attachment on the minus strand DNA, namely deproteinized rcDNA (DP-rcDNA) or protein-free rcDNA (PF-rcDNA), which is a putative functional precursor for cccDNA [15, 16]. The molecular mechanism underlying rcDNA deproteinization is largely unknown. Previous studies have demonstrated that the host tyrosol-DNA phosphodiesterase 2 (TDP2) is able to unlink the covalent bond between viral polymerase and rcDNA in vitro, but its role in cccDNA formation remains controversial [17, 18]. We have further demonstrated that DP-rcDNA is produced in cytoplasmic viral capsid and transported into nucleus by cellular karyopherins [19]. Upon nuclear delivery of DP-rcDNA, it is hypothesized that the host functions, most likely the DNA repair machinery, recognize the gaps/nicks in rcDNA as DNA breaks (damages), and repair it into the perfect circular cccDNA [5, 9, 20]. Such an assumption has only been theoretically conceivable and has not been experimentally confirmed until a recent study reported that host DNA polymerase κ (POLK) is involved in HBV cccDNA formation during de novo infection [21]. In addition, hepadnaviral dslDNA can also be converted into cccDNA through illegitimate self-circularization after preparing the termini for intramolecular ligation, however, dslDNA-derived cccDNA normally carries insertion-deletion mutations (indels) at the junction region, a typical phenotype of host non-homologous end joining (NHEJ) DNA repair activity [22–24]. In line with this, we have previously demonstrated that an NHEJ component Ku80 is responsible for cccDNA formation from DHBV dslDNA [25], thus further confirming the involvement of DNA repair machinery in cccDNA formation.
In order to systematically identify the host factors involved in cccDNA formation, we screened 107 human DNA repair genes for their effects on HBV cccDNA formation through shRNA knock-down in HepDES19 cells, and selected gene candidates (screen hits) for validation by functional inhibitions in multiple cell systems and assays. We report herein that the cellular DNA ligase (LIG) 1 and 3 are responsible for hepadnavirus cccDNA formation from rcDNA, and the conversion of dslDNA to cccDNA requires LIG4. Such findings will shed light on the molecular mechanism of cccDNA biosynthesis and suggest novel antiviral targets for treatment of chronic hepatitis B.
We screened a total of 107 cellular DNA repair genes for their effects on HBV cccDNA production by shRNA knock-down (see Materials and methods for detailed screening procedures), and the primary screening result is summarized in S1 Fig. The screened genes were grouped based on their primarily associated DNA repair pathways. Knock-down of 8 genes showed more than 50% reduction of cccDNA compared to control knock-down, including LIG1 and LIG3. The low hit rate of the shRNA screen may be due to the incomplete depletion of the targeted genes and/or functional redundancy of the cellular DNA repair genes/pathways. Surprisingly, a large number of gene knock-down resulted in cccDNA upregulation. Such unanticipated observation indicates that certain DNA repair genes may negatively regulate cccDNA formation through rcDNA sequestration or even degradation, which awaits further investigations.
We prioritized DNA ligases for functional validation in cccDNA formation for two reasons. Firstly, although the redundant activities of host DNA repair factors may be involved in cccDNA formation, it is likely that the DNA ligation is an essential and final step to seal the nicks/gaps on rcDNA. Secondly, there are only three known DNA ligases in mammalian cells, including LIG1, LIG3, and LIG4 [26], and we have previously ruled out the involvement of NHEJ repair apparatus, which contains LIG4, in rcDNA to cccDNA conversion [25].
In addition to the above cell-based screening assay, considering that the conversion of rcDNA into cccDNA occurs in the cell nucleus where the DNA repair machinery functions, we established an in vitro cell-free cccDNA formation system with nuclear protein extract. With this assay, we have tested the ability of cccDNA formation from the purified DHBV virion rcDNA. As shown in Fig 2A, although cccDNA could not be directly detected by Southern blot, we were able to detect cccDNA signal by a sensitive PCR assay with primers targeting the sequences outside of the gap region in DHBV rcDNA (S2A Fig), a similar principle has been used for the quantitative detection of HBV cccDNA by real-time PCR [27]. The assay specificity is sufficient enough to distinguish 0.3 pg of cccDNA from 30 pg of input rcDNA (S2B and S2C Fig). Sequence analysis of the cccDNA amplicons demonstrated a perfect repair of rcDNA gaps in nuclear extract. It is worth noting that the observed cccDNA PCR signal from the above in vitro cccDNA formation assay might be derived from rcDNA-like template with one strand being repaired into a closed circular DNA. In line with this, a recent study has identified a HBV rcDNA species with a covalently closed minus strand but an open plus strand through digesting the Hirt DNA samples from cell cultures with 3’→5’ exonuclease I and III (ExoI/III), which is possibly an intermediate during rcDNA to cccDNA conversion [28]. We thus attempted to search for such rcDNA-like species that might be generated during the in vitro cccDNA formation reaction by the similar approach. However, neither the minus-strand nor the plus-strand closed circular ssDNA was detected by Southern blot (S3 Fig), indicating that the rcDNA species with covalently closed minus or plus strand in this in vitro cccDNA formation assay, if any, is below the detection limit of Southern blot. Further optimization is needed to improve the assay robustness. Nonetheless, the current in vitro cell-free cccDNA formation assay may still be able to partly recapitulate the nuclear events during rcDNA to cccDNA conversion, which provides a simple and convenient system to assist the study of DNA repair functions in hepadnavirus cccDNA formation.
To assess the role of host DNA ligases in hepadnavirus cccDNA formation, LIG1/3 inhibitor L1 and L25, and pan ligase inhibitor L189 that inhibit the interaction between human DNA ligases and nicked DNA [29], were added into the in vitro cccDNA formation reaction. As shown in Fig 2B, all the ligase inhibitors tested completely blocked the rcDNA-to-cccDNA conversion in nuclear extract.
Taking advantage of the fast kinetics and high productivity of DHBV cccDNA formation, we established a tetracycline (tet)-inducible DHBV stable cell line in the background of HepG2 cells, namely HepDG10, to study hepadnavirus cccDNA formation in human cells. As shown in Fig 3A, DHBV pgRNA, core DNA and cccDNA were rapidly and robustly produced in HepDG10 cells upon withdrawal of tet, cccDNA could be detected by Southern blot as early as day 4 post induction and it accumulated to more significant levels after longer induction. The authenticity of the cccDNA band shown on Southern blot was confirmed by heat denaturation and EcoRI linearization (S4 Fig). Considering that HBV cccDNA formation in cell cultures is extremely time-consuming [15], the HepDG10 cell line thus provides a robust and convenient cell-based system for assessing human gene functions in hepadnavirus cccDNA formation through loss-of-function approaches (e.g. chemical inhibitors, gene knock down/out), which the relative short assay window may avoid or reduce the potential cytotoxic effects and/or functional redundancy. Consistent with the results from the in vitro nuclear extract-based cccDNA formation assay (Fig 2B), pan-ligase inhibitor L189 also inhibited DHBV cccDNA accumulation in HepDG10 cells without affecting core DNA replication (Fig 3B), further indicating a role of ligase(s) in cccDNA formation. The incomplete inhibition of cccDNA formation by L189 might be due to partial inhibition of cellular ligases under the nontoxic concentrations tested in HepDG10 cells.
To further assess the role of DNA ligases in cccDNA formation, the expression of LIG1 and LIG3 in HepDG10 cells was completely blocked through gene knock-out by CRISPR/Cas9 (Fig 4A, top panel). The indel mutation sequencing result confirmed the disruption of LIG1 and LIG3 gene at genomic DNA level in the established HepDG10-LIG1 K.O. and HepDG10-LIG3 K.O. cells, respectively (S5A, S5B, S6A and S6B Figs). Depletion of LIG1 or LIG3 did not affect the rcDNA production prominently but resulted in a significant reduction of cccDNA in HepDG10 cells (Fig 4A), suggesting that both LIG1 and LIG3 are involved in cccDNA formation. Similar result was observed with another clone of HepDG10-LIG1 K.O. and HepDG10-LIG3 K.O. cells. The incomplete inhibition of cccDNA formation by knocking out either LIG1 or LIG3 might be due to a redundant function between these two DNA ligases. Consistent with this, Sanger sequencing of cccDNA from the control and LIG1/3 knock-out cells showed wild type DHBV sequence between DR1 and DR2.
We have also attempted to CRISPR-out LIG1 and LIG3 simultaneously in HepDG10 cells, but were unable to obtain clones with complete knock-out of both genes, which might be due to that at least one ligase gene is required for cell viability. However, in a single colony-derived cell line with LIG1/3 double knock-down, which the knock-down efficiency was confirmed by both Western blot (Fig 4B, top panel) and T7E1 indel assay (S7 Fig), DHBV cccDNA formation was more profoundly inhibited without affecting core DNA replication (Fig 4B), further confirming the requirement of LIG1/3 in cccDNA synthesis. In addition, the similar phenomena was observed in HepDG10 cells with LIG1/3 single or double knock-down by lentiviral shRNA (S8 Fig).
It is of note that the protein-free rcDNA on Hirt DNA Southern blot was also reduced under LIG1 or LIG3 knock-out (Fig 4, bottom panels). We reason that the concurrent decrease of protein-free rcDNA might be a consequence of cccDNA reduction because the Hirt DNA sample from HepDG10 cells contained a large quantity of nicked cccDNA, which has the indistinguishable electrophoretic mobility as true rcDNA but cannot be mild heat denatured into dslDNA form (S9 Fig). The presence of nicked cccDNA in Hirt extraction has been described in previous studies [16, 30].
To validate the role of LIG1/3 in HBV cccDNA formation, the individual ligase was knocked out in HBV stable cell line HepDES19 cells by CRISPR/Cas9 (Figs 5A, S5A–S5C and S6A–S6C). Upon tet induction, the cytoplasmic rcDNA remained unchanged in LIG1 or LIG3 knock-out cells compared to control knock-out cells (Fig 5B). cccDNA qPCR assay demonstrated a significant reduction of cccDNA in LIG1/3 knock-out HepDES19 cells (Fig 5C). In order to avoid the contamination of rcDNA in cccDNA quantitation, we developed a method to eliminate rcDNA in Hirt DNA samples before qPCR, which involves a 85°C heat denaturation step that selectively denatures rcDNA into single stranded (ss) DNA, followed by Plasmid-safe ATP-dependent DNase (PSAD) treatment to remove non-cccDNA templates (Materials and methods) (S10 Fig).
During natural infection, hepadnavirus cccDNA is formed from both the invading rcDNA in virion and the newly synthesized rcDNA [1]. Because the DHBV or HBV stable cell line used in above studies only makes cccDNA through the rcDNA recycling pathway, we, thus, further assessed the role of LIG1/3 in HBV infection system. To do so, LIG1/3 expression was stably suppressed in HepG2-NTCP12 cells by lentiviral shRNA (Fig 6A). The control and LIG1/3 knock-down HepG2-NTCP12 cells were infected with HBV particles in the presence of nucleoside analogue 3TC which is known to block de novo HBV DNA replication but not the initial cccDNA formation [21, 31], making the system suitable for studying the first round cccDNA formation from the input viruses. Upon infection, the levels of HBV cccDNA and core protein were markedly lower in HepG2-NTCP cells with LIG1 or LIG3 knock-down compared to the control knock-down cells (Fig 6B, 6C and 6D), suggesting that LIG1 and LIG3 are also required for the first round HBV cccDNA formation during de novo infection.
In order to further validate the role of DNA ligases in hepadnavirus cccDNA formation and rule out the potential off-target effects caused by CRISPR knock-out, we reconstituted LIG1 and LIG3 expression in their corresponding HepDG10 knock-out cells by transfecting plasmid expressing sgRNA-resistant LIG1 and LIG3 gene, respectively (Fig 7A and 7B, top panels). Upon tet induction, DHBV cccDNA formation was successfully rescued in LIG1 and LIG3 knock-out cells by restoring the expression of each ligase (Fig 7A and 7B, bottom panels), which further confirmed that both LIG1 and LIG3 play a critical and specific role in hepadnavirus cccDNA formation.
Furthermore, LIG1 or LIG3 was ectopically expressed into their corresponding HepG2-NTCP12 shRNA knock-down cells through transfection, followed by HBV infection in the presence of 3TC. As shown in S11 Fig, restoration of LIG1 or LIG3 expression significantly enhanced HBV infection, as visualized by core immunostaining.
With rcDNA serving as the major viral genome DNA form and cccDNA precursor, hepadnavirus replication produces a minor double stranded linear (dsl) DNA species, which can also be converted into cccDNA format [1]. We previously reported that Ku80 protein in cellular NHEJ DNA repair pathway is involved in DHBV cccDNA formation from such dslDNA form [25]. In this study, we set out to assess the role of LIG4, the end effector of NHEJ pathway, in hepadnavirus cccDNA formation through gene knock-out approach. Because the classical CRSIPR/Cas9-based knock-out system requires NHEJ apparatus to introduce indel mutations at the DNA cleavage site, we made use of an alternative microhomology-mediated end-joining (MMEJ) DNA repair based CRISPR PITCh system to obtain the LIG4-null cell line [32] (Materials and methods) (S12 Fig). As shown in Fig 8A, LIG4 expression was completely depleted in the established LIG4 knock-out HEK293T cells. Next, we transfected the control and LIG4 K.O. cells with either the wildtype DHBV-1S construct or a DSL-DHBV plasmid supporting dslDNA-only replication (Fig 8B). The results demonstrated that the cccDNA formation in the context of wildtype DHBV replication was not affected by knocking out LIG4 (Fig 8B, comparing lane 2 to lane 1), while the cccDNA formation from dslDNA was completely abolished in the absence of LIG4 (Fig 8B, lane 4 vs lane 3). Furthermore, restoration of LIG4 expression was able to rescue cccDNA formation in DSL-DHBV transfected LIG4 K.O. cells (Fig 8C). Collectively, the above data strongly supports a conclusion that LIG4 is specifically required for generating cccDNA from the hepadnaviral dslDNA genome.
The establishment and persistence of hepadnavirus infection is dependent upon the viral cccDNA, which is a non-replicating episomal viral genome deposited in the nucleus of infected cell after conformational conversion from viral rcDNA [9]. Due to the limited gene-coding capacity of hepadnavirus genome, the virus needs to borrow host functions to complete its lifecycle [8]. The cellular DNA repair is a well-conserved surveillance and restoration system to detect and heal the damage in chromosomal DNA, by which maintains the stability and integrity of the host genome for replication and transcription [33, 34]. It is plausible that hepadnaviruses hijack the cellular DNA repair apparatus for cccDNA formation by disguising the rcDNA as a “damaged” DNA [9, 20]. The two gaps on rcDNA would be recognized as lesions for DNA repair by the host, and the DNA termini and their associated modifications are expected to undergo trimming, elongation, and ligation, during cccDNA formation (Fig 1). However, the host DNA repair pathway responsible for cccDNA formation remain largely unknown, and thus far, only a few host DNA repair enzymes have been reported to be involved in cccDNA formation, including the tyrosol-DNA phosphodiesterase 2 (TDP2) [17], polymerase κ (POLK) and λ (POLL) [21]. In this study, we screened 107 host DNA repair factors to assess their individual effect on HBV cccDNA formation by lentiviral shRNA knock-down, and identified and validated the host DNA LIG1 and LIG3 as key factors for hepadnavirus cccDNA formation by using a battery of in vitro and cell-based assays. Such work hence provides new insights into the mechanisms underlying hepadnavirus cccDNA formation in hepatocyte nucleus.
As part of the cellular DNA replication and repair machineries, DNA ligases complete joining of DNA strands by catalyzing the phosphodiester bond formation. Specifically, LIG1 ligates the Okazaki fragments during chromosomal DNA synthesis, and it is involved in the ligation steps of homologous recombination repair (HRR), long-patch base-excision repair (BER) and nucleotide excision repair (NER); LIG3 is responsible for sealing single strand DNA breaks during the process of short-patch BER and NER [26]. The involvement of DNA ligases in cccDNA formation indicated that the process of rcDNA termini generates ends that can be ligated and DNA ligases are the end-effectors for sealing the breaks of rcDNA. It is worth noting that previous studies have shown that LIG3, but not LIG4, is essential for nuclear DNA replication in the absence of LIG1 [35, 36]; and LIG1 is a backup enzyme for LIG3 in BER and NER DNA repair pathways [37, 38]. Such functional redundancy between LIG1 and LIG3 may explain the unaffected cell viability and the incomplete inhibition of cccDNA formation by knocking down/out LIG1 or LIG3 only (Figs 4A, 5–7 and S8A) or knocking down both (Figs 4B and S8B). In previous studies, the redundant functions in cccDNA formation have also been observed between POLK and POLL [21], and perhaps between TDP2 and an unknown TDP2-like protein(s) [17]. It is of note that the potential role of TDP2 in cccDNA formation remains controversial. While one study demonstrated that knock-down of TDP2 inhibited, or at least delayed, DHBV cccDNA formation [17]; another study suggested that TDP2 might even serve as a negative regulator of HBV cccDNA formation rather than a facilitator [18], and a recent study showed that TDP2 chemical inhibitors did not inhibit HBV infection in cell cultures [39]. In our shRNA screen, POLK or TDP2 lentiviral shRNA did not significantly reduce cccDNA formation in HepDES19 cells (S1 Fig), indicating that cellular functional redundancy for each enzyme might also exist in our experimental system. Further validation and mechanistic studies are required to reconcile these results. On the other hand, it is also possible that the first round cccDNA formation from virion DNA during infection and the intracellular cccDNA amplification pathway may have preference for different DNA repair enzymes, or there is hepatic cell line- or clone-specific requirement of host DNA repair factors/pathways for cccDNA formation. We had attempted to CRIPSR out both ligases in HepDG10 cells but only achieved partial double knock-down (Fig 4B), suggesting that at least one of LIG1 and LIG3 is required by the cells, and perhaps by hepadnaviruses as well. However, our data does not completely rule out a possibility that a LIG1/3-independent ligation mechanism might be involved in cccDNA formation, such as DNA topoisomerase I (TOP1) which has been suggested to play a role in rcDNA circularization through its DNA endonuclease and strand transferase activities [40]. Based on the previous and current data, it can be also inferred that hepadnaviruses have evolved to take advantage of the functional redundancy of host DNA repair machinery for a successful cccDNA formation.
The mechanism underlying the different efficiency of cccDNA formation between HBV and DHBV remains largely unknown, but likely in a virus-specific but not host-specific manner [16, 25, 30]. Based on that, we created the cell-free and the human hepatoma cell-based DHBV system to facilitate the identification and validation of host and viral regulators of cccDNA formation (Figs 2 and 3). In addition to the possible determining factors for cccDNA formation in the steps of rcDNA maturation, deproteinization, nuclear importation and uncoating, whether the cellular DNA repair system differentially recognizes and repairs nuclear HBV and DHBV rcDNA into cccDNA remains obscure. In this study, we found that both viruses employ LIG1 and LIG3 for cccDNA formation (Figs 4–7, S8 and S11), suggesting that the different repair process of HBV and DHBV rcDNA, if any, should be at the steps upstream of rcDNA end joining.
Though LIG1 and LIG3 have overlapping functions, knocking out/down of LIG3 resulted in relatively lower level of cccDNA than LIG1 knock-out/down (Figs 4–7 and S8), which suggests that LIG3 may play a more important role in cccDNA formation. In line with this notion, the two separated nicks/gaps in rcDNA are reminiscent of single strand breaks, which are preferable substrates for LIG3 in BER- and NER-mediated single strand break repair (Fig 1). Moreover, during the primary screen, two other BER components, APEX1 and POLB [41], emerged as candidates for positive regulator of cccDNA formation (S1 Fig), further suggesting the potential involvement of short-patch BER in cccDNA formation, which awaits further systematic investigations.
With the protein and RNA attachments at the 5’ end of minus- and plus-strand, respectively, hepadnavirus rcDNA is not a typical DNA break substrate for the major known repair pathways, and it is unknown whether the two gaps in rcDNA are repaired simultaneously or separately, including the final ligation step. A recent study revealed a nuclear rcDNA species with a covalently closed minus strand but an open plus strand, indicating that the nick on minus strand may be sealed first during cccDNA formation [28]. However, we did not observe an increased accumulation of protein-free rcDNA after blocking cccDNA synthesis in LIG1 or LIG3 knock-out cells (Figs 4, 6, 7B and S8). This phenomena may be due to a fact that the processed rcDNA ready for ligation is unstable. Previous studies have shown that the Hirt DNA samples from DHBV replicating cells contain high levels of nicked cccDNA which might be generated intracellularly or during the Hirt extraction [16, 30]. We also found that the protein-free rcDNA in Hirt extraction from HepDG10 cells were largely nicked cccDNA (S9 Fig), indicating that the observed reduction of protein-free rcDNA in LIG1/3 knock-out cells might be a consequence of cccDNA reduction (Figs 4 and 7). Nonetheless, further characterization of the nuclear rcDNA in LIG1/3 knock-out cells will provide further information for understanding the biological processes of rcDNA termini prior to the final ligation step during cccDNA formation.
In parallel with the bona fide rcDNA-to-cccDNA formation during hepadnavirus infection, the viral dslDNA byproduct is also repaired into cccDNA with indel mutations at the joint region [16, 23]. Although the dslDNA-derived cccDNA is generally defective of initiating a new round of viral DNA replication, it remains functional to express HBsAg and thus may play a role in viral pathogenesis. Based on the linear format of dslDNA and the indel mutations of its cccDNA derivative, it is hypothesized that dslDNA is a substrate for host error-prone NHEJ DNA repair system, and we have previously reported that another NHEJ component Ku80 is required for DHBV cccDNA formation from the dslDNA but not rcDNA [25]. LIG4 is the DNA ligase responsible for performing the last step of double strand DNA end joining in the NHEJ pathway [24]. In this study, we have demonstrated that LIG4 plays an essential role in cccDNA formation from DHBV dslDNA, and no functional redundancy was observed between LIG4 and other ligases (Fig 8). In addition, it has been reported that the chromosome DNA double strand breaks are targets for DHBV DNA integration [42], which indicates that the NHEJ machinery, including LIG4, is also responsible for the integration of hepadnavirus dslDNA into host genome.
Altogether, our study revealed a critical role of cellular DNA ligases in hepadnavirus cccDNA biosynthesis. Another possible function of DNA ligases in hepadnavirus life cycle can be to maintain the integrity of cccDNA, provided the preexisting cccDNA undergoes DNA damage and the host cell is able to repair it. Based on our observations, the DNA ligase inhibitors, which are currently under development for anti-cancer therapy [43], may be developed into host-targeting antiviral means to treat chronic hepatitis B by blocking cccDNA formation and/or repair.
HepG2 and 293T cells were purchased from ATCC and cultured in DMEM/F12 medium (Gibco) supplemented with 10% fetal bovine serum, 100 U/ml penicillin and 100 μg/ml streptomycin. The tetracycline-inducible HBV (Genbank accession number: U95551) stable cell line HepDE19 and HepDES19 were established previously [15], and maintained in the same way as HepG2, but with the addition of 1 μg/ml tetracycline (tet) and 400 μg/ml G418. When required, the culture medium was switched to tet-free to initiate HBV replication in HepDE19 and HepDES19 cells. HBV infectious particles were collected from the supernatant of HepDE19 cells, and the infection of HepG2-NTCP12 cells and HBV core protein (HBc) immunofluorescence microscopy were conducted according to a previously published protocol [44]. DHBV virions were purified from the serum of virally infected ducks as previously described [19]. DNA ligase 1/3 inhibitors L1 (5-(methylthio)thiophene-2-carboxylic acid) and L25 (2,3-dioxoindoline-7-carboxylic acid), and the pan ligase inhibitor L189 (6-Amino-2,3-dihydro-5-[(phenylmethylene)amino]-2-4(1H)-pyrimidineone) were purchased from Tocris Biosciences. Lamivudine (3TC) was kindly provided by Dr. William Mason (Fox Chase Center Center).
DHBV total RNA in cell cultures was extracted by TRIzol (Invitrogen) and detected by Northern blot [25]. HBV and DHBV cytoplasmic core DNA and whole cell Hirt DNA were extracted and analyzed by Southern blot as previously described [15, 45, 46]. HBV total DNA and cccDNA qPCR were performed according to the literature with modifications [47–49]. Firstly, HBV total DNA in the total Hirt DNA sample, including protein-free rcDNA and cccDNA, was first quantified by qPCR with 0.8 μM of forward primer (5’-CCGTCTGTGCCTTCTCATCTG-3’ (nt 1551–1571)), 0.8 μM of reverse primer (5’-AGTCCAAGAGTYCTCTTATGYAAGACCTT-3’ (nt 1674–1646)), and 0.2 μM of TaqMan probe (5’-FAM- CCGTGTGCACTTCGCTTCACCTCTGC-TAMRA-3’ (nt 1577–1602)). In the meantime, the relative cellular mitochondrial DNA (COX3 gene) level in each Hirt DNA samples was quantified by SYBR green qPCR with forward primer 5’-CCCTCTCGGCCCTCCTAATAACCT-3’ and reverse primer 5’-GCCTTCTCGTATAACATCGCGTCA-3’. Next, to reduce the contamination of HBV rcDNA in the qPCR detection of cccDNA, the Hirt DNA sample was first heated at 85°C for 5 min to denature rcDNA into single-stranded DNA, followed by Plasmid-safe ATP-dependent DNase (PSAD) (Epicentre) treatment at 37°C for 16 h. The PSAD reaction was then stopped by heat inactivation at 70°C for 30 min, and the samples were further purified by DNA clean-up spin column (Zymo Research). Real-time PCR amplification of 2 μl cleaned cccDNA sample was performed in a 20 μl reaction containing 0.9 μM forward primer (5’-ATGGAGACCACCGTGAACGCCC-3’(nt 1610–1631)), 0.9 μM reverse primer (5’-TCCCGATACAGAGCTGAGGCGG-3’(nt 2021–2000)), and 0.2 μM TaqMan probe (5’-FAM-TTCAAGCCTCCAAGCTGTGCCTTGGGTGGC-TAMRA-3’; nt 1865–1894). The cccDNA qPCR primers were designed to target the HBV DNA sequences outside of the gap region in rcDNA and to avoid PCR amplification of the integrated HBV genome in HepDES19 cells. The FastStart Essential DNA Probes Master (Roche) and FastStart Universal SYBR Green Master (Roche) were used to assemble TaqMan and SYBR Green qPCR reactions, respectively. The qPCR was run by Roche LightCycler 96 under the following thermal cycling conditions: 10 min at 95°C, followed by 15 sec at 95°C and 1 min at 61°C for 50 cycles. The cccDNA qPCR data was normalized by the cellular mitochondrial DNA quantitation.
The supercoiled cccDNA from DHBV transfected 293T cells were extracted by a previously developed alkaline lysis method with minor modifications [22]. Briefly, cells in a 6-well-plate were lysed in 200 μl lysis buffer containing 10 mM Tris-HCl (pH7.4), 1 mM EDTA, and 0.2% NP40, at room temperature for 10 min. The lysate was mixed gently with 200 μl alkaline lysis buffer (0.1 M NaOH, 6% SDS) and incubated at 37°C for 30 min, followed by adding 100 μl 3 M KAc (pH5.0) and mixing gently. After incubating on ice for 10 min and centrifuging at 12,000 rpm for 5 min, supernatant was collected and extracted by phenol twice. DNA was precipitated by ethanol and subjected to Southern blot assay.
The customized Mission lentiviral shRNA DNA repair gene family set was purchased from Sigma-Aldrich. The library contains 586 lentiviral shRNA targeting 140 DNA repair genes of all the known DNA repair pathways except for NHEJ, with an average of 3–5 different shRNA sequence against each target gene. The virus stocks were aliquoted in 96-well-plate with viral titer ranging from 1×107 to 3×107 g.e/ml. The TDP2 lentiviral shRNA was purchased from Santa Cruz Biotechnology and added into the library.
To establish stable DNA repair gene knock-down cell lines. HepDES19 cells were infected with the pooled lentiviral shRNA targeting the same DNA repair gene or control lentiviral shRNA in the presence of tet, 2 days later, the cells were selected by puromycin (3 μg/ml) for 1 week and the antibiotics-resistant cells were pooled and expanded into cell lines. The cells transduced by different lentiviral shRNA exhibited variable growth rate during puromycin selection but all were viable. A total of 107 DNA repair gene lenti-shRNA transduced cell lines were obtained. To assess the effect of RNAi on cccDNA production, the knock-down cells under confluent condition (1×106 cells per 35mm-dish) were cultured in the absence of tet for 10 days, total Hirt DNA was extracted and subjected to Southern blot or HBV total DNA and cccDNA qPCR. It is known that the HBV Hirt DNA level is positively related to HBV DNA replication level, especially the cytoplasmic rcDNA level [15, 16, 19]. To assess the efficiency of cccDNA formation under gene knock-down, the relative cccDNA levels in DNA repair gene knock-down cells compared to control knock-down cells were normalized by total HBV Hirt DNA signals, which indicates a relative rcDNA-to-cccDNA conversion rate.
Plasmid pTREHBVDE, the vector delivered the HBV transgene in HepDE19 cells, has been described previously [15]. DHBV-1S is a plasmid supporting DHBV (Genbank Accession No.: K01834) DNA replication upon transfection into cell cultures [50]. Plasmid 1Sdsl-3 (renamed to DSL-DHBV in this study) was a derivative of DHBV-1S with an artificial point mutation of G2552C in viral genome that supports double stranded linear (dsl) DNA replication but fails to make rcDNA [25, 51]. Plasmid pcLIG1-FLAG expressing human DNA ligase 1 (LIG1, GenBank Accession No.: NM_000234) with C-terminal FLAG-tag was purchased from Genescript (clone ID: OHu14319). To construct ligase 3 (LIG3) expression plasmid pcLIG3, the ORF of LIG3 was PCR amplified from MGC human LIG3 sequence-verified cDNA (GE Healthcare Dharmacon, clone ID: 6092747) by forward primer 5’-5’-CGGGATCCATGTCTTTGGCTTTCAAGATCTTCTT-3’ (BamHI site is underlined) and reverse primer 5’-GCTCTAGACTAGCAGGGAGCTACCAGTCTCCGTTT-3’ (XbaI site is underlined), and cloned into the BamHI/XbaI restricted pcDNA3.1 vector (Invitrogen). Plasmid pcLIG4-FLAG expressing the C-terminal FLAG-tagged human Ligase 4 (LIG4) was purchased from Genescript (clone ID: OHu13291).
DHBV virion DNA which contain predominantly rcDNA and a minor portion of dslDNA were purified from serum derived virions and quantified by Southern blot using DHBV DNA marker as standard according to published literature [52]. DHBV cccDNA was extracted from Dstet5 cells and gel purified as previously described [15]. The nuclear extract was prepared from HepG2 cells and stored in aliquots following a published protocol [53]. To assemble the DNA repair reaction, the purified DHBV virion DNA was mixed with 4 μl nuclear extract in 200 μl reaction buffer containing 20 mM HEPES, 80 mM KCl, 10 mM MgCl2, 1 mM ATP, 1 mM DTT, and 50 μM dNTPs, and incubated at 37°C for 30 min. To stop the reaction, 10 μl of 1% SDS, 20 μl of 0.5 M EDTA and 10 μl of 10 mg/ml pronase were added and incubated at 37°C for an additional 30 min. Next, the mixture was subjected to phenol and phenol:chloroform extraction, and viral DNA was precipitated down by ethanol and dissolved in 10 μl nuclease-free H2O. The obtained viral DNA was analyzed by Southern blot or cccDNA-specific PCR. The PCR reaction was assembled by mixing 0.5 μl DNA sample, 12.5 μl 2× PCR buffer (Clontech), 1 μl of 20 μM forward primer (5’-GCCAAGATAATGATTAAACCACG-3’), 1 μl of 20 μM reverse primer (5’-TCATACACATTGGCTAAGGCTC-3’), 0.5 μl Terra polymerase (Clontech), and 9.5 μl H2O. The DNA was amplified by 22 cycles of heat denaturation at 95°C for 30 sec, annealing at 55°C for 30 sec, and extension at 72°C for 30 sec. The PCR product was subjected to agarose gel electrophoresis and stained by ethidium bromide.
To detect the possible closed minus strand or plus strand after in vitro cccDNA formation in nuclear extract, exonucleases Exo I and III (ExoI/III) were used to degrade DNA strands with a free 3’ end and preserve closed circular DNA in either single-stranded (SS) or double-stranded (DS) form as described previously [28]. Briefly, 5 ng of DHBV rcDNA were subjected to the in vitro cccDNA formation reaction as described above. After reaction, the recovered DNA were dissolved in 20 μl water and treated with 0.25 μl each of Exo I and Exo III at 37°C for 2 h in 1×NEB Cutsmart buffer. 5 ng of DHBV rcDNA without going through in vitro cccDNA formation reaction served as positive control for ExoI/III digestion. The digestion products were directly subjected to electrophoresis and Southern blotting for hybridization with p32-labeled DHBV minus- or plus-strand specific riboprobe.
Plasmid pTRE-GFP-DHBV, which bidirectionally supports the tet-inducible expression of GFP and DHBV pgRNA, was constructed as follows. Firstly, a DNA fragment, which, in the 5’ to 3’ orientation, contained the partial sequence (nt 12–430) of pBI vector (Clontech, GenBank Accession No.: U89932) and the reverse complementary sequence of GFP ORF followed by SV40 polyadenylation signal, with unique PstI and KpnI restriction site at the 5’ and 3’ end, respectively, was chemically synthesized (Genescript). Then the PstI/KpnI restricted fragment was cloned into the same endonuclease treated pTREHBVDE plasmid to generate pTRE-GFP-HBV. Next, another DNA fragment containing nt 425–468 sequence from pBI, a spacer sequence (5’-GCAGAGCTCGTTTGATC-3’), and DHBV sequence (nt 2524-3021/1), with unique KpnI and EcoRI site at 5’ and 3’ end, respectively, was chemically synthesized (Genescript), and the fragment was inserted into the KpnI/EcoRI sites of pTRE-GFP-HBV to generate pTRE-GFP-DHBV-EcoRI-HBV. The HindIII site in the backbone sequence downstream of the remaining HBV sequence in pTRE-GFP-DHBV-EcoRI-HBV was further mutated to SalI site by using QuikChange II Site-Directed Mutagenesis Kit (Agilent Technologies) to obtain pTRE-GFP-DHBV-EcoRI-HBV-SalI. One unit length of DHBV genome was amplified from DHBV-1S by PCR with forward primer (5’- CGGCTAGAATTCATGCTCATTTGAAAGCTT-3’, nt 3011-3021/1-19, DHBV EcoRI site is underlined) and reverse primer (5’-AATTAAGTCGACAATTCTAGCCGTAATCGGATA-3’, nt 3021–3001, non-DHBV SalI site is underlined) and digested by EcoRI and SalI and cloned into the same sites in pTRE-GFP-DHBV-EcoRI-HBV-SalI to generate the final product pTRE-GFP-DHBV.
To establish tetracycline-inducible DHBV stable cell lines, HepG2 cells were cotransfected by pTRE-GFP-DHBV and pTet-off which expresses tet-responsive transcriptional activator (tTA) (Clontech) with 7:1 molar ratio. The transfected HepG2 cells were selected with 500 μg/ml G418 in the presence of 1 μg/ml tet. G418-resistant colonies were picked and expanded into cell lines. To determine DHBV positive cell lines, the candidate clones were cultured in 96-well-plate with tet-free medium for 6 days and subjected to fluorescence microscopy to select GFP-positive clones. Then, the GFP-positive cells were lysed in 1% NP40 and the cytoplasmic lysate was subjected to dot blotting as described previously [47]. The dot blot was hybridized by α-32P-UTP (800 Ci/mmol, Perkin Elmer) labeled minus strand specific full-length DHBV riboprobe, and the obtained DHBV positive cell lines were further assessed for their tet-inducible DHBV core DNA replication and cccDNA production by Southern blot. A DHBV cccDNA highly producing cell line clone was named HepDG10. The maintenance and induction of HepDG10 cells were performed in the same way as HepDES19 cells.
Lentiviral particles expressing U6 promoter-driven shRNA for knocking down human LIG1 or LIG3 were purchased from Sigma-Aldrich. The shRNA coding sequences for knocking down LIG1 and LIG3 are listed in S1 Table. HepDG10 and HepG2-NTCP12 cells were transduced by above lentiviral LIG1 or LIG3 shRNA or control shRNA per manufacturer’s instruction. The transduced cells were selected with 3 μg/ml puromycin and the antibiotics-resistant cells were pooled and expanded into cell lines. The knock-down levels of LIG1 and LIG3 in knock-down cell lines were assessed by Western blot by using antibodies against LIG1 (sc-271678, clone C5, Santa Cruz Biotechnology) and LIG3 (sc-390922, clone E7, Santa Cruz Biotechnology), respectively, and compared to control knock-down cells. β-actin served as loading control for Western blot by using anti-actin antibody (MAB1501, clone C4, Millipore).
To construct a lenti-vector expressing both LIG1 and LIG3 shRNA, a chemically synthesized DNA fragment containing the H1 promoter sequence and downstream LIG3 shRNA sequence (S1 Table) was cloned into the unique EcoRI site of lenti-shRNA plasmid DNA pLKO.1-LIG1 (TRCN0000048494, Sigma), giving rise to the bicistronic lenti-shRNA plasmids with two orientations of the H1-shLIG3 cassette right downstream of the original U6-shLIG1 cassette. The head-to-tail and tail-to-tail dimer clones were named pLKO.1-LIG1/3C1 and pLKO.1-LIG1/3C2, respectively, and pLKO.1-LIG1/3C2 was used to prepare lentiviral shRNA particles with MISSION Lentiviral Packaging Mix (Sigma). HepDG10 cells were transduced with lentiviral shLIG1/3 and the puromycin-resistant cells were pooled and expanded into stable cell line, namely HepDG10-shLIG1/3. The double knock-down of LIG1 and LIG3 were determined by Western blot comparing to the aforementioned control knock-down cells.
LIG1 and LIG3 knock-out cell lines were generated through CRISPR-mediated genome editing of LIG1 and LIG3 gene loci. The single guide (sg) RNAs targeting two different sites of human LIG1 and LIG3 gene were designed at http://www.e-crisp.org/E-CRISP and shown in S5 and S6 Figs. In addition to the general criteria for sgRNA design [54], the sgRNAs were designed to target either the 5’ end of the ORF or the functional domain coding sequences of LIG1 or LIG3. Furthermore, the designed sgRNA sequences do not possess any possible CRISPR sites in DHBV or HBV sequences. The synthetic sgRNA oligo pairs (S2 Table) were annealed and cloned into BbsI-digested lentiCRISPRv2 control vector (Addgene # 52961, gift from Dr. Feng Zhang). Lentivirus preparations were performed according to the protocols from Dr. Feng Zhang’s Lab (genome-engineering.org). Briefly, each lentivector was co-transfected with packaging plasmids psPAX2 and pMD2.G (Addgene# 12260 and 12259, respectively, gift from Dr. Didier Trono) in molar ratio of 4:3:1 into 293T cells by Lipofectamine 2000 (Invitrogene), and 48 h later, media was collected, centrifuged at 1,000 × g for 10 min, filtered through a 0.45um filter, and virus titers were determined by lentiviral titration kit. Lentiviral transduction of HepDG10 or HepDES19 and antibiotics-selection were performed as above described to generate control and LIG1/LIG3 stable knock-out cell lines, specifically HepDG10 LIG1 K.O., HepDG10 LIG3 K.O., HepDES19 LIG1 K.O., and HepDES19 LIG3 K.O. cells. The LIG1 or LIG3 knock-out phenotype was confirmed by Western blot and indel sequencing assay. The corresponding coding sequences for epitopes of LIG1 and LIG3 antibodies have no overlap with the gene targeting sites of the designed sgRNAs.
To knock out both LIG1 and LIG 3 in HepDG10 cells, the cells were transduced with lentiviruses encoding CRISPR/Cas9-LIG1-sgRNA1 and CRISPR/Cas9-LIG3-sgRNA1 together (1:1 ratio). The puromycin selection and clone screening was performed as described above. The indel mutations of sgRNA targeting sites in LIG1 and LIG3 gene loci was detected by T7E1 assay.
For indel sequencing analysis of LIG1 and LIG3 genes, total genomic DNA from the control and LIG1 or LIG3 knock-out cells were extracted using DNeasy blood and tissue kit (Qiagen) according to the manufacturer’s protocol. The genomic sequence region covering the CRISPR target site was amplified by PCR using the indel detection primers (S3 Table) and cloned into T vector pMD19 (Clontech) for Sanger sequencing. The LIG1/3 DNA sequence from control and knock-out cells was aligned to determine the CRISPR-induced mutations.
For T7E1-based indel assay, primers used to amplify DNA fragments containing LIG1-sgRNA1 and LIG3-sgRNA1 targeting region were listed in S3 Table, indel mutations were detected by Guide-it Mutation Detection Kit (Clontech) according to manufacturer’s manual.
Because LIG4 is an essential component in NHEJ DNA repair pathway in eukaryotes [55, 56], the conventional NHEJ-based CRISPR/Cas9 knock-out system is not able to generate LIG4 knock-out cell lines. Therefore a newly developed microhomology-mediated end-joining (MMEJ) repair based CRISPR/Cas9 knock-in system [32] was used to establish LIG4 knock-out cells. Briefly, annealed sgRNA targeting the last exon of LIG4 gene (S12 Fig and S4 Table) was inserted into pX330A-1×2 (Addgene# 58766, gift from Dr. Takashi Yamamoto) to obtain pX330A-1×2-LIG4-gRNA, and after Golden Gate assembly using BsaI (New England Biolabs), the cassette of PITCh-gRNA from pX330S-2-PITCh (Addgene# 63670, gift from Dr. Takashi Yamamoto) was inserted into pX330A-1×2-LIG4-gRNA to generate the All-in-One pCRISPR/Cas9-PITCh-LIG4-gRNA vector containing both LIG4 gRNA and PITCh-gRNA. Then, pCRIS-PITChv2-LIG4 was constructed based on pCRIS-PITChv2-FBL (Addgene# 63672, gift from Dr. Takashi Yamamoto) by performing two separate PCR, one to amplify the vector backbone and one to amplify LIG4-specific microhomology arm containing EGFP-2A-Puro knock-in cassette. Generic 5′-reverse and 3′-forward primers were used for vector backbone amplification, and LIG4-specifc primers containing the desired microhomologies were used to amplify the insertion (S4 Table). The above two purified PCR fragments were conjugated by using the In-Fusion HD cloning kit (Clontech) to generate plasmid pCRIS-PITChv2-LIG4. To generate LIG4 knock-out cell line, 293T cells were transfected with pCRIS-PITChv2-LIG4 and the All-in-One plasmid pCRISPR/Cas9-PITCh-LIG4-gRNA with a molar ratio of 1:2, followed by puromycin (1 μg/ml) selection. The puromycin-resistant colonies were pooled together and subjected to fluorescence microscopy for determining EGFP-positive cells with successful knock-in, and then the knock-out efficiency of LIG4 was determined by Western blot using antibodies against LIG4 (sc-28232, clone H-300, Santa Cruz Biotechnology). Control knock-in 293T cells were made by transfecting pCRIS-PITChv2-FBL and pX330A-FBL/PITCh (Addgene# 63671, gift from Dr. Takashi Yamamoto) that target human fibrillarin (FBL) gene.
The “NGG” protospacer adjacent motif (PAM) sequence of LIG1 sgRNA1 locates just ahead of the targeted exon (S5 Fig), therefore plasmid pcLIG1-FLAG was directly used in the function rescue experiment by transfecting the HepDG10-LIG1 K.O. cells. To avoid the integrated lentiviral CRISPR/Cas9-LIG3 sgRNA system targets LIG3 ectopic expression plasmid, the LIG3 sgRNA1 corresponding PAM motif “CGG” in pcLIG3 was mutated to “CGA” by Q5 Site-Directed Mutagenesis Kit (New England Biolabs) with primers (forward: 5’-AACTAGAGCGaGCCCGGGCCA-3’, reverse: 5’- TCTCAAACATGCATTTAATGTGGTACCAC-3’) but without changing the amino acid sequence of LIG3. The sgRNA-resistant pcLIG3 was used to transfect HepDG10-LIG3 K.O. cells in the function rescue experiment. Because the LIG4 knock-out 293T cells were made by transient transfection-mediated knock-out, pcLIG4-FLAG was used directly in rescue experiment by transfection.
|
10.1371/journal.pgen.1001244 | GC-Rich Sequence Elements Recruit PRC2 in Mammalian ES Cells | Polycomb proteins are epigenetic regulators that localize to developmental loci in the early embryo where they mediate lineage-specific gene repression. In Drosophila, these repressors are recruited to sequence elements by DNA binding proteins associated with Polycomb repressive complex 2 (PRC2). However, the sequences that recruit PRC2 in mammalian cells have remained obscure. To address this, we integrated a series of engineered bacterial artificial chromosomes into embryonic stem (ES) cells and examined their chromatin. We found that a 44 kb region corresponding to the Zfpm2 locus initiates de novo recruitment of PRC2. We then pinpointed a CpG island within this locus as both necessary and sufficient for PRC2 recruitment. Based on this causal demonstration and prior genomic analyses, we hypothesized that large GC-rich elements depleted of activating transcription factor motifs mediate PRC2 recruitment in mammals. We validated this model in two ways. First, we showed that a constitutively active CpG island is able to recruit PRC2 after excision of a cluster of activating motifs. Second, we showed that two 1 kb sequence intervals from the Escherichia coli genome with GC-contents comparable to a mammalian CpG island are both capable of recruiting PRC2 when integrated into the ES cell genome. Our findings demonstrate a causal role for GC-rich sequences in PRC2 recruitment and implicate a specific subset of CpG islands depleted of activating motifs as instrumental for the initial localization of this key regulator in mammalian genomes.
| Key developmental genes are precisely turned on or off during development, thus creating a complex, multi-tissue embryo. The mechanism that keeps genes off, or repressed, is crucial to proper development. In embryonic stem cells, Polycomb repressive complex 2 (PRC2) is recruited to the promoters of these developmental genes and helps to maintain repression in the appropriate tissues through development. How PRC2 is initially recruited to these genes in the early embryo remains elusive. Here we experimentally demonstrate that stretches of GC-rich DNA, termed CpG islands, can initiate recruitment of PRC2 in embryonic stem cells when they are transcriptionally-inactive. Surprisingly, we find that GC-rich DNA from bacterial genomes can also initiate recruitment of PRC2 in embryonic stem cells. This supports a model where inactive GC-rich DNA can itself suffice to recruit PRC2 even in the absence of more complex DNA sequence motifs.
| Polycomb proteins are epigenetic regulators required for proper gene expression patterning in metazoans. The proteins reside in two main complexes, termed Polycomb repressive complex 1 and 2 (PRC1 and PRC2). PRC2 catalyzes histone H3 lysine 27 tri-methylation (K27me3), while PRC1 catalyzes histone H2A ubiquitination and mediates chromatin compaction [1], [2]. PRC1 and PRC2 are initially recruited to target loci in the early embryo where they subsequently mediate lineage-specific gene repression. In embryonic stem (ES) cells, the complexes localize to thousands of genomic sites, including many developmental loci [3]–[5]. These target loci are not yet stably repressed, but instead maintain a “bivalent” chromatin state, with their chromatin enriched for the activating histone mark, H3 lysine 4 tri-methylation (K4me3), together with the repressive K27me3 [6], [7]. In the absence of transcriptional induction, PRC1 and PRC2 remain at target loci and mediate repression through differentiation. The mechanisms that underlie stable association of the complexes remain poorly understood, but likely involve interactions with the modified histones [8]–[12].
Proper localization of PRC1 and PRC2 in the pluripotent genome is central to the complex developmental regulation orchestrated by these factors. However, the sequence determinants that underlie this initial landscape remain obscure. Polycomb recruitment is best understood in Drosophila, where sequence elements termed Polycomb response elements (PREs) are able to direct these repressors to exogenous locations [13]. PREs contain clusters of motifs recognized by DNA binding proteins such as Pho, Zeste and GAGA, which in turn recruit PRC2 [14]–[17]. Despite extensive study, neither PRE sequence motifs nor binding profiles of PRC2-associated DNA binding proteins are sufficient to fully predict PRC2 localization in the Drosophila genome [1], [16], [18], [19].
While protein homologs of PRC1 and PRC2 are conserved in mammals, DNA sequence homologs of Drosophila PREs appear to be lacking in mammalian genomes [13]. Moreover, it remains controversial whether the DNA binding proteins associated with PRC2 in Drosophila have functional homologs in mammals. The most compelling candidate has been YY1, a Pho homolog that rescues gene silencing when introduced into Pho-deficient Drosophila embryos [20]. YY1 has been implicated in PRC2-dependent silencing of tumor suppressor genes in human cancer cells [21]. However, this transcription factor has also been linked to numerous other functions, including imprinting, DNA methylation, B-cell development and ribosomal protein gene transcription [22]–[26].
Recently, researchers identified two DNA sequence elements able to confer Polycomb repression in mammalian cells. Sing and colleagues identified a murine PRE-like element that regulates the MafB gene during neural development [27]. These investigators defined a critical 1.5 kb sequence element that is able to recruit PRC1, but not PRC2 in a transgenic cell assay. Woo and colleagues identified a 1.8 kb region of the human HoxD cluster that recruits both PRC1 and PRC2 and represses a reporter construct in mesenchymal tissues [28]. Both groups note that their respective PRE regions contain YY1 motifs. Mutation of the YY1 sites in the HoxD PRE resulted in loss of PRC1 binding and partial loss of repression, while comparatively, deletion of a separate highly conserved region from this element completely abrogated PRC1 and PRC2 binding as well as repression [28].
In addition to these locus-specific investigations, genomic studies have sought to define PRC2 targets and determinants in a systematic fashion. The Ezh2 and Suz12 subunits have been mapped in mouse and human ES cells by chromatin immunoprecipitation and microarrays (ChIP-chip) or high-throughput sequencing (ChIP-Seq) [3]–[5],[29]. Such studies have highlighted global correlations between PRC2 targets and CpG islands [5], [30] as well as highly-conserved genomic loci [4], [7], [31]. Recently, Jarid2 has been shown to associate with PRC2 and to be required for proper genome-wide localization of the complex [32]–[35]. Intriguingly, Jarid2 contains an ARID and a Zinc-finger DNA-binding domain. However, it is unclear how Jarid2 could account for PRC2 targeting given the lack of sequence specificity and the low affinity of its DNA binding domains [33], [36]. In summary, a variety of sequence elements including CpG islands, conserved elements and YY1 motifs have been implicated in Polycomb targeting in mammalian cells. Causality has only been demonstrated in two specific instances and a unifying view of the determinants of Polycomb recruitment remains elusive.
Here we present the identification of multiple sequence elements capable of recruiting PRC2 in mammalian ES cells. This was achieved through an experimental approach in which engineered bacterial artificial chromosomes (BACs) were stably integrated into the ES cell genome. Evaluation of a series of modified BACs specifically identified a 1.7 kb DNA fragment that is both necessary and sufficient for PRC2 recruitment. The fragment does not share sequence characteristics of Drosophila PREs and lacks YY1 binding sites, but rather corresponds to an annotated CpG island. Based on this result and a genome-wide analysis of PRC2 target sequences we hypothesized that large GC-rich sequence elements lacking transcriptional activation signals represent general PRC2 recruitment elements. We tested this model by assaying the following DNA sequences: (i) a ‘housekeeping’ CpG island which was re-engineered by removal of a cluster of activating motifs; and (ii) two large GC-rich intervals from the E. coli genome that satisfy the criteria of mammalian CpG islands. We found that all three GC-rich elements robustly recruit PRC2 in ES cells. We propose that a class of CpG islands distinguished by a lack of activating motifs play causal roles in the initial localization of PRC2 and the subsequent coordination of epigenetic controls during mammalian development.
To identify DNA sequences capable of recruiting Polycomb repressors in mammalian cells, we engineered human BACs that correspond to genomic regions bound by these proteins in human ES cells.
We initially targeted a region of the human Zfpm2 (hZfpm2) locus, which encodes a developmental transcription factor involved in heart and gonad development [37]. In ES cells, the endogenous locus recruits PRC1 and PRC2, and is enriched for the bivalent histone modifications, K4me3 and K27me3 (Figure 1A). We used recombineering to engineer a 44 kb BAC containing this locus and a neomycin selection marker. The modified BAC was electroporated into mouse ES cells, and individual transgenic ES cell colonies containing the full length BAC were expanded (Figure S1). Fluorescent in situ hybridization (FISH) confirmed integration at a single genomic location (Figure S2).
We used ChIP and quantitative PCR (ChIP-qPCR) with human specific primers to examine the chromatin state of the newly incorporated hZfpm2 locus. This analysis revealed strong enrichment for K27me3 and K4me3 (Figure 1B). In addition, we explicitly tested for direct binding of the Polycomb repressive complexes using antibody against the PRC1 subunit, Ring1B, or the PRC2 subunit, Ezh2. We detected robust enrichment for both complexes in the vicinity of the hZfpm2 gene promoter (Figure 1B). To confirm this result and eliminate the possibility of integration site effects, we tested two additional transgenic hZfpm2 ES cell clones with unique integration sites as well as a fourth transgenic ES cell line containing a distinct Polycomb target locus, Pax5. In each case, we observed a bivalent chromatin state analogous to the endogenous loci (Figure S3). Similar to endogenous bivalent CpG islands, we found the Zfpm2 CpG island was DNA hypomethylated (Figure S4). These results suggest that DNA sequence is sufficient to initiate de novo recruitment of Polycomb in ES cells.
A key function of Polycomb repressors is to maintain a repressive chromatin state through cellular differentiation. To determine if the integrated BAC is capable of maintaining K27me3, the hZfpm2 transgenic ES cells were differentiated to neural progenitor (NP) cells in vitro [38]. ChIP-qPCR analysis revealed continued enrichment of K27me3 but loss of K4me3 (Figure 1C), a pattern frequently observed at endogenous loci that are not activated during differentiation [39].This indicates that DNA sequence at the hZfpm2 locus is sufficient to initiate K27me3 chromatin modifications in ES cells, and maintain the repressive chromatin state through neural differentiation.
We next sought to define the sequences within the hZfpm2 BAC required for recruitment of Polycomb repressors. First, we re-engineered the 44 kb hZfpm2 BAC to remove 20 kb of flanking sequences that contained distal non-coding conserved sequence elements (Figure 1A). When we integrated the resulting 22 kb construct into ES cells we found that it robustly enriches for PRC1, PRC2, K4me3 and K27me3 (Figure 1B). Hence, these particular distal elements do not appear to be required for the recruitment of the complexes. Next, we considered the necessity of the CpG island which corresponds to the peak of Ezh2 enrichment in ChIP-Seq profiles (Figure 1A). We excised a 1.7 kb fragment containing the CpG island, and integrated the resulting BAC (ΔCGI) into ES cells. The ΔCGI BAC failed to recruit PRC1 or PRC2, and showed significantly reduced K27me3 levels relative to the other constructs (Figure 1B). This suggests that the CpG island is essential for recruitment of Polycomb proteins to the hZfpm2 locus.
We next asked whether the hZfpm2 CpG island is sufficient to recruit Polycomb repressors to an exogenous locus. To test this, we selected an unremarkable gene desert region on human chromosome 1 that shows no enrichment for PRC1, PRC2 or K27me3 in ES cells (Figure 2A). We also verified that the gene desert BAC alone does not show any enrichment for K27me3 or Ezh2 when integrated into ES cells (Figure 2B). Using recombineering, we inserted the 1.7 kb sequence that corresponds to the hZfpm2 CpG island into the gene desert BAC. The resulting construct was integrated into mouse ES cells and three independent clones were evaluated. ChIP-qPCR analysis revealed strong enrichment for K27me3, K4me3 and PRC2 over the inserted CpG island (Figure 2C, Figure S5). In contrast, we observed relatively little enrichment for the PRC1 subunit Ring1B (Figure 2C). We confirmed the specificity of these enrichments with primers that span the boundary between the insertion and adjacent gene desert sequence. Notably, K27me3 enrichment was detected across the gene desert locus up to 2.5 kb from the inserted CpG island (Figure 2C). This indicates that the localized CpG island can initiate K27me3 that then spreads into adjacent sequence. Lastly we found no YY1 enrichment across the CpG island by ChIP-qPCR (Figure S5). Together, these data suggest that the hZfpm2 CpG island contains the necessary signals for PRC2 recruitment but is insufficient to confer robust PRC1 association.
The functionality of a CpG island in PRC2 recruitment is consistent with prior observations that a majority of PRC2 sites in ES cells correspond to CpG islands [4], [5] and with the striking correlation between intensity of PRC2 binding and the GC-richness of the underlying sequence (Figure 2D). We therefore considered whether specific signals within the Zfpm2 CpG island might underlie its capacity to recruit PRC2.
First, we searched for sequence motifs analogous to the PREs that recruit PRC2 in Drosophila. We focused on motifs recognized by YY1, the nearest mammalian homolog of the Drosophila recruitment proteins. Notably, both of the recently described mammalian PREs contain YY1 motifs [27], [28]. The 44 kb hZfpm2 BAC contains 11 instances of the consensus YY1 motif. However, none of these reside within the CpG island (Figure S6) (see Methods). We also examined YY1 binding directly in ES cells and NS cells using ChIP-Seq. Consistent with prior reports, YY1 binding is evident at the 5′ ends of many highly expressed genes, including those encoding ribosomal proteins, and is also seen at the imprinted Peg3 locus (Figure 2E, Table S1) [26]. However, no YY1 enrichment is evident at the Zfpm2 locus. Moreover, at a global level, YY1 shows almost no overlap with PRC2 or PRC1, but instead co-localizes with genomic sites marked exclusively by K4me3 (Figure 2F, Figure S6, and Table S1). Thus, although YY1 may contribute to Polycomb-mediated repression through distal interactions or in trans, it does not appear to be directly involved in PRC2 recruitment in ES cells.
We previously reported that CpG islands bound by PRC2 in ES cells could be predicted based on a relative absence of activating transcription factor motifs (AMs) in their DNA sequence [5]. We reasoned that transcriptional inactivity afforded by this absence of AMs is a requisite for PRC2 association [40], [41]. This could explain why PRC2 is absent from a majority of CpG islands, many of which are found at highly active promoters. Consistent with this model, when we examined a recently published RNA-Seq dataset for poly-adenylated transcripts in ES cells, we found that virtually all of the high-CpG promoters (HCPs) lacking Ezh2 are detectably transcribed (Figure S7). The small proportion of HCPs that are neither Ezh2-bound nor transcribed may reflect false-negatives in the ChIP-Seq or RNA-Seq data. Alternatively, these HCPs tend to correspond to CpG islands with relatively low GC-contents and lengths and may therefore have insufficient GC-richness to promote PRC2 binding (Figure S7). Thus, correlative analyses implicate large GC-rich elements that lack transcriptional activation signals as general PRC2 recruitment elements in mammals.
To obtain direct experimental support for the general sufficiency of large GC-rich elements lacking AMs in PRC2 recruitment, we carried out the following experiments. First, we tested whether a K4me3-only CpG island could be turned into a PRC2 recruitment element by removing activating motifs. We targeted a 1.3 kb CpG island that overlaps the promoters of two ubiquitously expressed genes – Arl3 and Sfxn2. Neither gene carries K27me3 in ES cells, or in any other cell type tested (Figure S8, and data not shown). This CpG island was selected as it has many conserved AMs clustered in one half of the island (Figure 3A). We hypothesized that the portion of the Arl3/Sfxn2 CpG island lacking AMs would, in isolation, lack active transcription and recruit PRC2. In contrast, we predicted that the half containing multiple AMs would lack Polycomb. To test this, we generated two additional BAC constructs containing the respective portions of the Arl3/Sfxn2 CpG island positioned within the gene desert, and integrated these constructs into ES cells (Figure 3A). ChIP-qPCR shows that the portion of the CpG island lacking AMs is able to recruit PRC2 and becomes enriched for K27me3 (Figure 3B). In contrast, the AM-containing portion shows no enrichment for K27me3 or Ezh2, but is instead marked exclusively by K4me3, similar to the endogenous human locus (Figure 3C, Figure S8). Thus, a GC-rich sequence element with no known requirement for Polycomb regulation can recruit PRC2 when isolated from activating sequence features.
Next, we tested whether even more generic GC-rich elements might also be capable of recruiting PRC2 in ES cells. Here, we focused on sequences derived from the genome of E. coli, reasoning that there would be no selection for PRC2 recruiting elements in this prokaryote given the complete lack of chromatin regulators. We arbitrarily selected three 1 kb segments of the E. coli genome. Two with GC contents above the threshold for a mammalian CpG island but that each contained few AMs, and one AT rich segment as a control (Table S3). We recombined each segment into the gene desert BAC and integrated the resulting constructs into ES cells. ChIP-qPCR confirmed that both GC-rich E. Coli segments recruit Ezh2 and form a bivalent chromatin state (Figure 4A, 4B, Figure S9). Notably, the GC-rich segment also enriches for Jarid2, a PRC2 component with DNA binding activity (Figure S10). In contrast, the AT-rich segment did not recruit Ezh2 or enrich for either K4me3 or K27me3 (Figure 4C, Figure S9). Together, our findings suggest that GC-rich sequence elements that lack signals for transcriptional activation have an innate capacity to recruit PRC2 in mammalian ES cells.
Several lines of evidence suggest that the initial landscape of Polycomb complex binding is critical for proper patterning of gene expression in metazoan development [1], [2], [13]. Failure of these factors to engage their target loci in embryogenesis has been linked to a loss of epigenetic repression at later stages. Accordingly, the determinants that localize Polycomb complexes at the pluripotent stage are almost certainly essential to the global functions of these repressors through development.
We find that DNA sequence is sufficient for proper localization of Polycomb repressive complexes in ES cells, and specifically identify a CpG island within the Zfpm2 locus as being critical for recruitment. We provide evidence that GC-rich elements lacking activating signals suffice in general to recruit PRC2. This includes demonstrations (i) that a motif devoid segment of an active ‘housekeeping’ CpG island can recruit PRC2; and (ii) that arbitrarily selected GC-rich elements from the E. coli genome can themselves mediate PRC2 recruitment when integrated into the ES cell genome.
Several possible mechanistic models could explain the causality of GC-rich DNA elements in PRC2 recruitment (Figure 5). First, we note that CpG islands have been shown to destabilize nucleosomes in mammalian cells [42]. At transcriptionally inactive loci, this property could increase their accessibility to PRC2-associated proteins with DNA affinity but low sequence specificity, such as Jarid2 or AEBP2 [32]–[35], [43] (Figure S10). Although this association would be abrogated by transcriptional activity at most CpG islands, those lacking activation signals would remain permissive to PRC2 association (Figure 5). In support of this model, PRC2 targets in ES cells are also enriched for H2A.Z and H3.3, histone variants linked to nucleosome exchange dynamics [44], [45]. Alternatively or in addition, targeting could be supported by DNA binding proteins with affinity for low complexity GC-rich motifs or CpG dinucleotides, such as CXXC domain proteins [46]. Localization may also be promoted or stabilized by long and short non-coding RNAs [47]–[50] as well as by the demonstrated affinity of PRC2 for its product, H3K27me3 [11], [12]. Notably, PRC2 recruitment in ES cells appears distinct from that in Drosophila, as we do not find evidence for involvement of PRE-like sequence motifs or mammalian homologues such as YY1.
It should be emphasized that PRC2 localization does not necessarily equate with epigenetic repression. Indeed virtually all PRC2 bound sites in ES cells, and all CpG islands tested here, are also enriched for K4me3, and presumably poised for activation upon differentiation. Epigenetic repression during differentiation may require PRC1 and thus depend on additional binding determinants. YY1 remains an intriguing candidate in this regard, given prior evidence for physical and genetic interactions with PRC1 [51], [52]. YY1 consensus motifs are present in the Polycomb-dependent silencing elements recently identified in the MafB and HoxD loci. Interestingly, the HoxD element combines a CpG island with a cluster of conserved YY1 motifs. Mutation of the motifs abrogated PRC1 binding but left PRC2 binding intact. Still, the fact that only a small fraction of documented PRC2 and PRC1 sites have YY1 motifs or binding suggests that this transcription factor may act indirectly and/or explain only a subset of cases. Nonetheless, it is likely that a fully functional epigenetic silencer would require a combination of features, including a GC-rich PRC2 element as well as appropriate elements to recruit PRC1. Further study is needed to expand the rules for PRC2 binding to include a global definition of PRC1 determinants and ultimately, to understand how the initial landscape facilitates the maintenance of gene expression programs in the developing organism.
BAC constructs CTD331719L (‘Zfpm2 44’), CTD-2535J16 (‘Pax5’) and CTD-3219L19 (‘Gene Desert’) were obtained from Open Biosystems. Recombineering was done using the RedET system (Open Biosystems) in DH10B cells. Homology arms 200–500 bp in length were PCR amplified and cloned into a PGK; Neomycin cassette (Gene Bridges). This cassette was used to recombineer all BACs to enable selection in mammalian cells. The 22 kb hZfpm2 BAC was created by restricting the hZfpm2 BAC at two sites using ClaI, and re-ligating the BAC lacking the intervening sequence. The CpG island was excised from the 22 kb hZfpm2 BAC by amplification of flanking homology arms, and cloned into a construct containing an adjacent ampicillin cassette (Frt-amp-Frt; Gene Bridges). After recombination, the ampicillin cassette was removed using Flp-recombinase and selection for clones that lost ampicillin resistance (Flp-706; Gene Bridges). PCR across the region confirmed excision of the CpG island. For the Gene Desert BACs, the Zfpm2, Arl3, Sfxn2 and E. coli CpG islands were amplified with primers containing XhoI sites and cloned into the Frt-amp-Frt vector that contains homology arms from the Gene Desert region. The final constructs were confirmed by sequencing across recombination junctions. All primers used for CpG islands and recombineering homology arms are listed in Table S2.
ES cells (V6.5) were maintained in ES cell medium (DMEM; Dulbecco's modified Eagle's medium) supplemented with 15% fetal calf serum (Hyclone), 0.1 mM ß-mercaptoethanol (Sigma), 2 mM Glutamax, 0.1 mM non-essential amino acid (NEAA; Gibco) and 1000U/ml recombinant leukemia inhibitory factor (ESGRO; Chemicon). Roughly 50 µg of linearized BAC was nucleofected using the mouse ES cell nucleofector kit (Lonza) into 106 mouse ES cells, and selected 7–10 days with 150 µg/ml Geneticin (Invitrogen) on Neomycin resistant MEFs (Millipore). Individual resistant colonies were picked, expanded and tested for integration of the full length BAC by PCR. Differentiation of hZfpm2 ES cell clone 1 into a population of neural progenitor (NP) cells was done as previously described [53]. FISH analysis was done as described previously [54]. DNA methylation analysis was done as previously described [55] and primers used to amplify bisulfite treated DNA are listed in Table S2.
For each construct, between one and three ES cell clones were expanded and subjected to ChIP using antibody against K4me3 (Abcam ab8580 or Upstate/Millipore 07-473), K27me3 (Upstate/Millipore 07-449), Ezh2 (Active Motif 39103 or 39639), or Ring1B (MBL International d139-3) as described previously [5], [7], [39]. ChIP DNA was quantified by Quant-iT Picogreen dsDNA Assay Kit (Invitrogen). ChIP enrichments were assessed by quantitative PCR analysis on an ABI 7500 with 0.25 ng ChIP DNA and an equal mass of un-enriched input DNA. Enrichments were calculated from 2 or 3 biologically independent ChIP experiments. For K27me3, and Ezh2 enrichment, background was subtracted by normalizing over a negative genomic control. Error bars represent standard error of the mean (SEM). We confirmed that the human specific primers do not non-specifically amplify mouse genomic DNA. Primers used for qPCR are listed in Table S2.
Genomewide maps of YY1 binding sites were determined by ChIP-Seq as described previously [39]. Briefly, ChIP was carried out on 6×107 cells using antibody against YY1 (Santa Cruz Biotechnology sc-1703). ChIP DNA was used to prepare libraries which were sequenced on the Illumina Genome Analyzer. Density profiles were generated as described [39]. Promoters (RefSeq; http://genome.ucsc.edu) were classified as positive for YY1, H3K4me3 or H3K27me3 if the read density was significantly enriched (p<10−3) over a background distribution based on randomized reads generated separately for each dataset to account for the varying degrees of sequencing depth. ChIP-Seq data for YY1 are deposited to the NCBI GEO database under the following accession number GSE25197 (http://www.ncbi.nlm.nih.gov/projects/geo/query/acc.cgi?acc=GSE25197). Sites of Ezh2 enrichment (p<10−3) were calculated genomewide using sliding 1 kb windows, and enriched windows within 1 kb were merged. DNA methylation levels were calculated using previously published Reduced Representation Bisulphite Sequenced (RRBS) libraries [55]. Composite plots represent the mean methylation level in sliding 200 bp windows in the the 10 kb surrounding the TSSs of the indicated gene sets.
YY1 motifs were identified using the MAST algorithm [56] where a match to the consensus motif was defined at significance level 5×10−5. Candidate CpG islands for TF motif analysis were identified by scanning annotated CpG islands (http://genome.ucsc.edu) for asymmetric clustering of motifs related to transcriptional activation in ES cells [5]. Motifs shown in Figure 3A and Figure S6 are from UCSCs TFBS conserved track. GC-rich elements from the E. coli K12 genome were selected by calculating %GC and CpG O/E in sliding 1 kb windows. Sequences matching the criteria for mammalian CpG islands while simultaneously being depleted of motifs related to transcriptional activation [5] were chosen for insertion into mouse ES cells. Transcriptionally inactive HCPs were selected based on a lack of transcript enrichment by both expression arrays [39] and RNA-Seq data [57]. In the case of RNA-Seq, each gene was assigned the maximum read density within any 1 kb window of exonic sequence. To ease analysis of promoter CpG island statistics, only HCPs containing a single CpG island were considered.
|
10.1371/journal.pgen.1008379 | The monothiol glutaredoxin GrxD is essential for sensing iron starvation in Aspergillus fumigatus | Efficient adaptation to iron starvation is an essential virulence determinant of the most common human mold pathogen, Aspergillus fumigatus. Here, we demonstrate that the cytosolic monothiol glutaredoxin GrxD plays an essential role in iron sensing in this fungus. Our studies revealed that (i) GrxD is essential for growth; (ii) expression of the encoding gene, grxD, is repressed by the transcription factor SreA in iron replete conditions and upregulated during iron starvation; (iii) during iron starvation but not iron sufficiency, GrxD displays predominant nuclear localization; (iv) downregulation of grxD expression results in de-repression of genes involved in iron-dependent pathways and repression of genes involved in iron acquisition during iron starvation, but did not significantly affect these genes during iron sufficiency; (v) GrxD displays protein-protein interaction with components of the cytosolic iron-sulfur cluster biosynthetic machinery, indicating a role in this process, and with the transcription factors SreA and HapX, which mediate iron regulation of iron acquisition and iron-dependent pathways; (vi) UV-Vis spectra of recombinant HapX or the complex of HapX and GrxD indicate coordination of iron-sulfur clusters; (vii) the cysteine required for iron-sulfur cluster coordination in GrxD is in vitro dispensable for interaction with HapX; and (viii) there is a GrxD-independent mechanism for sensing iron sufficiency by HapX; (ix) inactivation of SreA suppresses the lethal effect caused by GrxD inactivation. Taken together, this study demonstrates that GrxD is crucial for iron homeostasis in A. fumigatus.
| Aspergillus fumigatus is a ubiquitous saprophytic mold and the major causative pathogen causing life-threatening aspergillosis. To improve therapy, there is an urgent need for a better understanding of the fungal physiology. We have previously shown that adaptation to iron starvation is an essential virulence attribute of A. fumigatus. In the present study, we characterized the mechanism employed by A. fumigatus to sense the cellular iron status, which is essential for iron homeostasis. We demonstrate that the transcription factors SreA and HapX, which coordinate iron acquisition, iron consumption and iron detoxification require physical interaction with the monothiol glutaredoxin GrxD to sense iron starvation. Moreover, we show that there is a GrxD-independent mechanism for sensing excess of iron.
| Iron is an essential trace element for almost all organisms in all kingdoms of life. On the other hand, iron excess is toxic. Therefore, to maintain cell homeostasis, the balance between iron uptake and iron consumption has to be tightly regulated.
Previous studies have shown that iron homeostasis in the pathogenic mold Aspergillus fumigatus is mainly regulated by two transcription factors, SreA, the repressor of siderophore biosynthesis and reductive iron assimilation [1], and HapX, which is a repressor of iron-consuming pathways and activator of iron acquisition [2]. Moreover, HapX is essential for adaptation to iron excess. When iron concentrations increase, HapX changes its function from a repressor to an activator of iron-consuming and detoxifying pathways to avoid iron toxicity. Consequently, HapX is crucial for adaptation to both iron starvation (-Fe) and high iron concentrations (hFe), i.e. lack of this regulator causes growth defects under -Fe as well as hFe [3]. Notably, both the -Fe and hFe functions of HapX require the HapB/HapC/HapE CCAAT-binding complex (CBC) as a DNA binding platform [4].
SreA and HapX are interconnected in a feedback-loop [5]: Expression of sreA is repressed by HapX during -Fe [2] and, in turn, hapX expression is repressed by SreA under iron sufficiency/excess [1]. Moreover, HapX induces sreA expression in response to iron.
Fungal iron sensing has been studied most intensively so far in the yeasts Saccharomyces cerevisiae and Schizosaccharomyces pombe [6,7]. Remarkably, there is little similarity with respect to transcriptional iron regulation between S. cerevisiae and A. fumigatus. Despite the fact that both, HapX and SreA are conserved in most ascomycetes, S. cerevisiae lacks classical homologs of SreA and HapX. In this yeast, adaptation to iron starvation is mainly mediated by two paralogous transcription factors, termed Aft1 and Aft2 [8–10]. Adaptation to hFe by transcriptional activation iron detoxification is mediated by the bZIP transcription factor Yap5 [11]. Nevertheless, S. cerevisiae Yap5 and HapX show similarities. Both transcription factors are essential for iron detoxification by activation of vacuolar iron deposition. Moreover, they share a highly conserved cysteine-rich region (CRR) that is crucial for this function and which has been shown to coordinate a [2Fe-2S] cluster in Yap5 [3,12]. In contrast to HapX, however, Yap5 has no function during iron starvation. S. pombe employs a homolog of SreA, termed Fep1 [13] and a regulator displaying similarity with HapX, termed Php4 [14]. Similar to HapX, Php4 acts as repressor of iron-consuming functions during iron starvation, but in contrast to HapX it is not involved in activation of iron detoxification. Taken together, S. cerevisiae, S. pombe and A. fumigatus show significant differences with regard to the employed iron-regulatory transcription factors and the molecular mechanisms of iron sensing in A. fumigatus are largely uncharacterized.
In both S. cerevisiae and S. pombe, the cytosolic monothiol glutaredoxins Grx3/4 respectively Grx4 have been shown to be involved in iron sensing [15,16] and coordination and transport of [2Fe-2S] clusters. These proteins contain a thioredoxin (Trx)-like domain, for which a canonical reductase activity has been excluded [17], and a glutaredoxin (Grx) domain comprising a highly conserved CGFS motif. Coordination of [2Fe-2S] clusters is performed via the cysteine residue of the CGFS motif and two glutathione residues, which leads to dimerization of these monothiol glutaredoxins [18–20].
In the current study, we characterized the role of the cytosolic monothiol glutaredoxin of A. fumigatus (Afu2g14960), designated GrxD. We demonstrate that GrxD is essential for iron sensing by the iron-responsive transcription factors HapX and SreA, particularly for signaling iron starvation conditions. The study revealed both similarities and differences to iron sensing in other fungal species.
Protein BLAST searches identified the A. fumigatus homolog, termed GrxD, of S. cerevisiae Grx3/4 and S. pombe Grx4, respectively. Alignment of GrxD homologs demonstrated high conservation, even between distantly related species (Figs 1 and S1). Compared to the Trx-like domain, the Grx domain shows significantly higher conservation including the [2Fe-2S] cluster coordinating CGFS motif.
To investigate GrxD function in A. fumigatus, we aimed to delete the grxD gene via replacement by a hygromycin resistance-conferring cassette (hph) (S2 Fig). Several attempts were unsuccessful, indicating that grxD is an essential gene, which we proved by heterokaryon rescue [21]. In short, this technique is based on the fact that A. fumigatus cells contain multiple nuclei. The fungal transformation procedure usually targets only the genome of one nucleus leading to heterokaryosity, in our case grxD+hph- (wt; containing grxD but lacking hph) nuclei and grxD-hph+ (ΔgrxD; lacking grxD but containing hph conferring hygromycin resistance) nuclei, which was proven by Southern blot analysis (Fig 2A). During conidiation, nuclei are separated since conidia contain only a single nucleus. Conidia of eight heterokaryotic transformants were able to grow under non-selective conditions but not in the presence on hygromycin (Fig 2A), demonstrating the inability of ΔgrxD (grxD-hph+) conidia to grow; i.e. grxD is an essential gene.
Due to the lethality of grxD deletion, we generated strains, in which grxD is under the control of the xylose-inducible xylP promoter (PxylP, [22]). These strains were generated without and with C-terminal tagging of GrxD with the yellow fluorescent protein derivative Venus, yielding strains PxylP:grxD and PxylP:grxDvenus, respectively (Fig 2B). PxylP displays xylose concentration-dependent activation. Without xylose supplementation, activity of this promoter is very low, i.e. expression of essential genes under this promoter in A. fumigatus led to the inability to grow [23]. Although we proved that grxD is essential (Fig 2A), strains PxylP:grxD and PxylP:grxDvenus were able to grow without xylose-induction on solid minimal medium (Fig 2C). This indicates that very low expression is sufficient to support growth. Nevertheless, we observed growth deficiencies under iron starvation, which were ameliorated with increasing iron concentrations (Fig 2C), which indicates a role of GrxD in iron homeostasis. Overexpression of grxD with and without venus-tagging decreased growth under excess iron, but not under iron starvation or iron replete conditions (Fig 2C), indicating that a surplus of GrxD impedes adaptation to high iron conditions.
To further analyze GrxD function, we generated A. fumigatus mutants producing PxylP-driven GrxD variants lacking either the 19 N-terminal amino acids (strain PxylP:grxDΔ19) or the whole Trx domain (PxylP:grxDvenusΔtrx, Fig 1), whereby in the latter strain GrxD was C-terminally tagged with Venus (Fig 2B). Under non-inducing conditions (without xylose), truncation of 19 N-terminal amino acids or truncation of the complete Trx domain, respectively, blocked growth during iron starvation and iron sufficiency (Fig 2D). Growth of both mutant strains was rescued by xylose supplementation, whereby the strain expressing the Trx domain lacking GrxD required higher xylose supplementation indicating lower activity. Important to note, C-terminal tagging with Venus did not affect function of GrxD, at least judged by growth ability (Fig 2E).
The fact that, in contrast to strains PxylP:grxD and PxylP:grxDvenus, strains PxylP:grxDΔ19 and PxylP:grxDvenusΔtrx were unable to grow in -Fe conditions under non-induced conditions indicates that truncation of the N-terminal 19 amino acids or, even more pronounced, the truncation of the Trx domain decreases activity of GrxD. This might be due to decreased protein stability or hampered function. Nevertheless, under xylose-inducing conditions, all strains were able to grow under all conditions, which indicates that in contrast to the whole GrxD protein, the Trx domain is not essential for growth, at least when overexpressed. Consequently, the Grx domain is likely essential for growth.
As shown above, N-terminal truncated GrxD versions (PxylP:grxDΔ19 and PxylP:grxDvenusΔtrx) were not able to grow at non-inducing conditions during iron starvation or iron sufficiency (Fig 2D). However, high iron supplementation partially rescued the growth of these strains at non-inducing conditions (Fig 2D). These data indicate that GrxD is involved in iron homeostasis with an important role especially during iron starvation. This is in agreement with decreased growth of strains with down-regulated GrxD, without and with C-terminal Venus-tagging, under iron starvation but not iron sufficiency and iron excess (Fig 2C).
Occasionally, cultivation of PxylP:grxDΔ19 conidia on plates resulted in suppressor mutants. We characterized one of these mutant strains, termed PxylP:grxDΔ19sup, in more detail. In contrast to PxylP:grxDΔ19, PxylP:grxDΔ19sup was able to grow without xylose-induction under iron starvation and iron sufficiency (Fig 2D). Under 0.1% xylose-inducing conditions, PxylP:grxDΔ19sup displayed a similar radial growth under iron starvation but decreased growth under iron sufficiency and high iron conditions compared to PxylP:grxDΔ19 (Fig 2D). These results indicated that the suppressor mutation present in this strain leads to a defect in adaptation to iron excess.
Northern analysis revealed an additional sreA transcript as well as de-repression of hapX and mirB (encoding a siderophore transporter) during iron sufficiency in strain PxylP:grxDΔ19sup compared to PxylP:grxDΔ19 (S3A Fig). These results suggested that the suppressor mutation affects the function of SreA, which has previously been shown to repress transcription of these two genes [1]. PCR amplification analyses of the sreA locus (S3B and S3F Fig) followed by rapid amplification of cDNA ends (3´-RACE) and nucleotide sequencing (S3C and S3D Fig) revealed that the suppressor mutation caused a chromosomal rearrangement (S3E Fig), which results in truncation of SreA within the DNA-binding region.
To independently confirm the genetic interaction between grxD and sreA, the sreA gene was deleted in a PxylP:grxDΔ19 background. This mutant, PxylP:grxDΔ19/Δ sreA, displayed the same growth pattern as PxylP:grxDΔ19sup (Fig 2D), which affirms that sreA loss-of-function rescues the growth defect caused by down-regulation of grxD during iron starvation and sufficiency.
To analyze whether inactivation of SreA rescues growth only in response to downregulation of GrxD (PxylP:grxDΔ19/ΔsreA at non-inducing conditions) or also complete lack of GrxD, we aimed to delete the grxD gene in a ΔsreA background. In contrast to wt background (see above), this approach was successful. Compared to wt, the ΔgrxD/ΔsreA strain displayed severely decreased radial growth under iron starvation, iron sufficiency and iron excess, but it was viable (Fig 2F).
SreA is the repressor of iron uptake and SreA inactivation results in increased iron acquisition [1]. Consequently, the identified genetic interaction between grxD and sreA, together with the rescue of growth of the PxylP:grxDΔ19 strain under non-inducing conditions by high iron supplementation (Fig 2D), indicate that lack of GrxD results in iron shortage, possibly caused by the requirement of GrxD for sensing iron starvation.
To monitor endogenous and PxylP-controlled grxD expression, we performed Northern analysis. In wt grxD transcript levels decreased with increasing iron supplementation (Fig 3A). In PxylP:grxDΔ19, grxD expression was highly induced under xylose-induced conditions and decreased below detection limit upon xylose withdrawal demonstrating functionality of PxylP-mediated conditional grxD expression (Fig 3A).
In agreement with wt grxD transcript levels, Western blot analysis demonstrated that Venus-tagged GrxD (GrxDVenus) protein levels decreased with increasing iron availability when grxD was expressed from the endogenous promoter (strain grxDvenus; Fig 3B). Under control of the xylP promoter, the protein level of Venus-tagged full-length GrxD (GrxDVenus, strain PxylP:grxDvenus) was highly decreased under non-inducing compared to inducing conditions (Fig 3C). Interestingly, hFe conditions slightly decreased the GrxDVenus protein level under xylose-inducing conditions, which indicates an influence of iron on xylP promoter activity or on the grxD transcript stability.
To analyze protein levels of the GrxD variant lacking the 19 N-terminal amino acids (GrxDΔ19), we generated a strain in which C-terminally Venus-tagged GrxDΔ19 is under the control of the xylP promoter (strain PxylP:grxDvenusΔ19). This strain showed identical growth compared to the untagged version PxylP:grxDΔ19 (S4 Fig). Compared to GrxDVenus, the protein levels of the Venus-tagged GrxD variants lacking the 19 N-terminal amino acids (GrxDVenusΔ19, strain PxylP:grxDvenusΔ19) or the Trx domain (GrxDVenusΔTrx, strain PxylP:grxDvenusΔtrx) were slightly decreased under inducing conditions. Remarkably, under steady-state, non-inducing, high iron conditions (Fig 3C), truncation of the 19 N-terminal amino acid residues (GrxVenusΔ19) decreased the protein level compared to GrxDVenus although not as much as truncation of the entire Trx domain (GrxVenusΔTrx). Due to the use of the same promoter, these data indicate higher protein stability of GrxDVenus compared to the truncated versions. These results most likely provide the explanation for the lack of growth of strains PxylP:grxDvenusΔ19 and PxylP:grxDvenusΔtrx during iron starvation and sufficiency under non-inducing conditions (Figs 2D and S4) in contrast to strain PxylP:grxDvenus (Fig 2C).
Subcellular localization of Venus-tagged GrxD was determined by fluorescence. To visualize the nucleus, we expressed a gene encoding histone H2A tagged with monomeric red fluorescence protein (H2AmRFP) in recipient strains PxylP:grxDvenus and PxylP:grxDvenusΔtrx (yielding strains PxylP:grxDvenus/H2AmRFP and PxylP:grxDvenusΔtrx/H2AmRFP).
Fluorescence microscopy with these strains revealed that GrxDVenus and GrxDVenusΔTrx displayed predominant nuclear localization during iron starvation but not iron sufficiency (Fig 4). During iron sufficiency, we did not observe organelle-specific accumulation of GrxDVenus. The nuclear localization indicates a regulatory role of GrxD at least during iron starvation. Noteworthy, it has been demonstrated previously that HapX also accumulates in the nucleus during iron starvation [3].
To identify GrxD-interacting proteins, A. fumigatus strains wt, PxylP:grxDvenus and PxylP:grxDvenusΔtrx were cultivated under iron starvation (-Fe), sufficiency (0.03 mM Fe) and excess (5 mM Fe) and the corresponding crude cell extracts were subjected to GFP-Trap affinity purification [24]. Here, wt served as a negative control to distinguish specifically interacting proteins from false positive bound ones. Effective enrichment of GrxDVenus and GrxDVenusΔTrx proteins was validated by SDS-PAGE and silver staining as well as Western blot analysis (S5 Fig). Eluates from three independent biological GFP-Trap replicates were subsequently analyzed by nLC-MS/MS. For visualization of the specific enrichment of GrxD-interacting proteins, label-free quantification (LFQ) abundances of the most enriched proteins identified in PxylP:grxDvenus and PxylP:grxDvenusΔtrx GFP-Trap eluates were plotted against their LFQ abundances in wt control eluates (Fig 5 and Tables 1 and S1).
We identified HapX as one of the most highly enriched proteins by GrxDVenus GFP-Trap under iron limitation (Fig 5A). HapX was also detected in iron sufficient and high-iron conditions (Fig 5B and 5C), however, with lower abundance, most likely due to its low protein level under these conditions [3]. Inversely, SreA was preferentially co-purified under iron sufficiency and excess (Fig 5B and 5C), again reflecting the expression pattern of SreA [1]. These data indicate that GrxD constitutively interacts with HapX irrespective of the cellular iron status and at least under iron sufficiency and iron excess also with SreA; possibly, GrxD interacts also constitutively with SreA—the missing detection of the interaction during iron starvation might be due to the low expression of sreA during this condition [1].
In addition, proteins that are part of the cytosolic iron-sulfur protein assembly (CIA) machinery, namely Nbp35 (Afu2g15960), Dre2 (AFUB_008090) and Mms19 (Afu8g05370), were enriched with high abundance under standard and excess iron levels (Fig 5B and 5C). The CIA machinery was investigated extensively in the model organism S. cerevisiae. These studies showed that the monothiol glutaredoxins Grx3 and Grx4 play an indispensable role for cytosolic iron-sulfur (FeS) cluster biogenesis. An early step in cytosolic [4Fe-4S] cluster assembly involves Nbp35 forming a hetero-tetrameric scaffold complex with Cfd1 on which a [4Fe-4S] cluster is bound transiently [25,26]. Dre2 belongs to the CIA electron transfer complex and is needed for formation of the [4Fe-4S] cluster on Nbp35 [27,28]. Mms19 is part of the CIA targeting complex consisting of Cia1, Cia2 and Mms19, which, together with Nar1 transfers the [4Fe-4S] cluster to target apoproteins [29,30].
The precise site of requirement of monothiol glutaredoxins in the cytosolic FeS protein biogenesis has not been determined yet. In yeast, Grx3/4 is required for FeS cluster assembly on Dre2 and Nar1 [19]. How GrxD is exactly involved in the CIA of A. fumigatus remains to be elucidated. Nevertheless, these data underline the specificity of the approach.
Grx4 protein interaction studies in S. pombe demonstrated that the Trx domain is essential for a stable protein interaction with both the iron regulators Fep1 (SreA ortholog) [31] and Php4 (HapX ortholog) [32]. Therefore, we were interested whether the GrxD Trx domain is necessary for all of the detected GrxD protein interactions in A. fumigatus. To address this topic, we analyzed our quantitative GrxDVenus and GrxDVenusΔTrx GFP-Trap co-purification data for selected interaction partners in detail (Fig 6). The GrxD Trx domain appeared to be dispensable for GrxD-HapX complex formation irrespective of the iron supplementation (Fig 6B). In contrast, the GrxD Trx domain was essential for GrxD-SreA protein interaction (Fig 6C) indicated by a severely decreased SreA LFQ abundance in the absence of the Trx domain. Likewise, GrxDVenusΔTrx pull-down enrichment of the CIA proteins Dre2, Nbp35 and Mms19 was less effective (Fig 6D–6F).
Unexpectedly, we identified the putative copper metallothionein CmtA (encoded by Afu4g04318) as an interaction partner of GrxD, preferably under iron excess conditions (Fig 6G). A recent study regarding cmtA regulation and CmtA protein function in A. fumigatus [33] revealed that cmtA expression is not regulated by copper availability and that CmtA is not required for copper detoxification. Consistently, the cmtA ortholog in A. nidulans (AN7011), termed MtlA, was found to be dispensable for copper ion tolerance [34]. Our GFP-Trap pull-down results may suggest that a GrxD-CmtA complex is involved in iron detoxification and/or transport, however this hypothesis has to be verified by future experiments.
Furthermore, our data suggested an interaction of GrxD with two putative BolA family proteins, Bol1 (Afu7g01520) and Bol3 (Afu6g12490). The Trx domain was dispensable for GrxD-Bol1 interaction, but GrxD-Bol3 interaction was dependent on its presence (Fig 6H and 6I). However, both A. fumigatus proteins contain an N-terminal mitochondrial targeting sequence, suggesting that these proteins are localized in mitochondria. In support, homologs of A. fumigatus Bol1 and Bol3 from other Aspergillus species also contain N-terminal mitochondrial targeting sequences. In agreement, fluorescence microscopy of a strain (PgpdA:bol1venus) expressing Bol1 C-terminally tagged with Venus (Bol1Venus) suggested that Bol1 is mainly localized in mitochondria (S6 Fig). It has been demonstrated previously that the homologous S. cerevisiae BolA proteins Bol1 and Bol3 form complexes with mitochondrial Grx5, which lacks a Trx domain [35]. As GrxD is localized in the cytosol and nucleus, the interaction with both mitochondrial Bol1 and Bol3 proteins in vivo appears unlikely. One possible explanation for their detected GrxD interaction is the artificial mixture of the proteins when cellular compartments are disrupted during sample preparation. A similar phenomenon has been observed in S. cerevisiae for interaction of Grx3/4 with Bol1, respectively Bol3 [36], which are both localized in mitochondria [35]. Nevertheless, we can neither exclude that a minor fraction of Bol1 is localized in the cytosol nor that Bol3 is exclusively or partially localized in the cytosol and that GrxD indeed interacts with these BolA-like proteins in vivo as described in other organisms [15,31,37,38].
To exemplary confirm GFP trap affinity purification results, we performed co-immunoprecipitation (co-IP) with subsequent Western blot detection (S7 Fig). HapX or SreA, respectively, was immunoprecipitated and purified from PxylP:grxDvenus and PxylP:grxDvenusΔtrx whole cell lysates using rabbit α-HapX, or rabbit α-SreA antibodies covalently linked to Protein-A-Sepharose. Western blot analysis demonstrated co-IP of GrxDVenus with both HapX and SreA (S7 Fig). These experiments confirmed that GrxDVenus interacts with both HapX and SreA, while truncation of the Trx domain GrxDVenusΔTrx blocks interaction with SreA but not with HapX.
For in vitro co-purification experiments, A. fumigatus GrxD was fused with a C-terminal His-tag (GrxDHis6) and bicistronically co-expressed in Escherichia coli with a polypeptide representing the A. fumigatus HapX C-terminus (HapX161-491) that contains all four cysteine-rich regions (CRR; Fig 7A). To investigate the interaction between both proteins, GrxDHis6 was enriched from crude cell extract via its His-tag using a Ni-Sepharose column. Consequently, co-purification of HapX161-491 requires binding to GrxDHis6. After initial Ni-chelate chromatography, we observed that GrxDHis6 and HapX161-491 were co-enriched (Fig 7B). The GrxD His-tag was subsequently removed by tobacco etch virus (TEV) protease treatment and the GrxD-HapX161-491 complex stability was further analyzed by preparative size exclusion chromatography (SEC). Two major peaks appeared during SEC and their apparent molecular masses were estimated based on the elution volumes of protein calibration standards. For peak 1, a molecular mass of 152.9 kDa (Fig 7B) approximately corresponding to a heterotetrameric complex consisting of two HapX161-491 and two GrxD subunits (theoretical mass: 130.4 kDa) was calculated. For peak 2, a molecular mass of 27.7 kDa corresponding to a theoretical molecular mass of a GrxD monomer (29.75 kDa) was determined. Additionally, UV-Vis spectra (250–550 nm) were recorded for peak 1 and 2 (Fig 7C). The reddish-brown color of the GrxD-HapX161-491 complex (peak 1) as well as the absorption maxima at 322 and 415 nm indicated the incorporation of a [2Fe-2S] ligand, as spectra of [2Fe-2S] proteins are typically more complex than those of [4Fe-4S] proteins, which display only one characteristic peak around 400–420 nm [39]. In contrast, GrxD separated in excess from the GrxD-HapX161-491 complex (peak 2) appeared colorless and displayed no absorption at 322 and 415 nm (Fig 7C). We hypothesized that the reddish-brown color of the GrxD-HapX161-491 complex is mainly derived from binding of an FeS ligand by HapX161-491 CRR. This was supported by SEC purification of HapX161-491 in the absence of GrxD, which yielded a reddish-brown colored SEC fraction displaying a UV-Vis spectrum almost identical to that of the GrxD-HapX161-491 complex (Fig 7D and 7E). These data strongly indicate that HapX is able to coordinate FeS cluster(s) without GrxD.
To analyze the in vitro GrxD-HapX161-491 protein-protein interaction in more detail, two GrxDHis6 mutants were constructed, co-produced with HapX161-491 and purified from E. coli crude cell extracts. Based on the results of the in vivo co-purification experiments, the Trx domain was deleted first. Consistent with our in vivo data, removal of the GrxD Trx domain had no impact on GrxDΔTrx-HapX161-491 protein interaction in vitro (Fig 7F). In a second step, GrxD cysteine (C) residue 191 was mutated to alanine (A). GrxD C191 is part of the CGFS active site motif in the Grx domain, which is highly conserved and known to be important for iron sensing through binding of a [2Fe-2S] cluster in S. cerevisiae [19,40] and S. pombe [32,41]. In S. pombe, the CGFS site’s cysteine is required for iron-dependent Grx4-Php4 complex formation [32]. In this study, mutation of the GrxD C191 to A did not influence binding to the HapX161-491 CRR in E. coli (Fig 7G).
HapX harbors four CRR, which might participate in iron sensing. As reported previously [3], CRR-A and B (Fig 7A) are crucial for adaptation to iron excess. In particular, the mutation of C 203 to A in CRR-A or exchange of C277 to A in CRR-B rendered A. fumigatus more susceptible to iron overload. C277 is part of the CRR-B C277GFCSDGTPCIC motif, which is reminiscent to the CGFCNDNTTCVC [2Fe-2S] cluster binding site in S. cerevisiae Yap5 [12]. To elucidate the impact of both C203 and C277 on GrxD-HapX161-491 complex formation, we targeted C203 and C277 by site-directed mutagenesis and replaced them by alanine. Neither HapX161-491 C203A exchange nor C277A substitution affected binding of the respective HapX versions to GrxD (Fig 7H and 7I). In summary, we conclude that the Trx domain and residue C191 of GrxD as well as residues C203 and C277 in HapX are not required for in vitro complex formation between GrxD and HapX.
As gene deletion was not possible in wt cells, we developed a protein depletion strategy to investigate the effects of GrxD deficiency. We avoided to use strain ΔgrxD/ΔsreA as it was not possible to measure effects of GrxD deficiency on SreA in this strain and as growth of this mutant was severely impaired. To study the effects of GrxD depletion on iron regulation, we employed PxylP:grxDΔ19, which allowed to decrease grxD expression to a lethal amount without xylose induction, while growth was fully rescued with a moderate (0.1%) concentration of xylose (see above, Fig 2).
To analyze the effect of GrxD depletion on iron regulation, we performed Northern analysis of iron regulated genes during iron starvation and sufficiency. For GrxD depletion, PxylP:grxDΔ19 was grown under inducing conditions for 20 h at 25°C and subsequently grown for another 20h at 37°C without xylose to repress grxD expression. This method was used previously to investigate essential genes [23]. During iron starvation, GrxD depletion decreased transcript levels of hapX and mirB, which were upregulated during iron starvation in wt (Fig 8A and 8B). On the other hand, GrxD depletion increased transcript levels of sreA (Fig 8A and 8B) and cccA (Fig 8B), which are downregulated during iron starvation in wt. During iron sufficiency, GrxD depletion did not significantly affect transcript levels of these genes. These data emphasize that GrxD is involved in iron regulation and is important for adaptation to iron starvation rather than iron sufficiency.
Repression of sreA and cccA during iron starvation has previously been shown to depend exclusively on HapX [2,3]. Therefore, the de-repression of these genes found upon GrxD depletion indicates that GrxD is required for signaling iron starvation to HapX. To test whether the effects on mirB are linked to SreA or HapX, we also depleted GrxD in strains lacking SreA (strain PxylP:grxDΔ19/ΔsreA). It has been shown previously that sreA is downregulated in wt during iron starvation and lack of SreA results in de-repression of iron-uptake genes (mirB, hapX) during iron sufficiency [1]. Deletion of sreA in PxylP:grxDΔ19 increased expression of mirB upon GrxD depletion, albeit not to wt level. This indicated that GrxD is required to inactivate the repressing function of SreA under iron starvation. The absence of full induction in GrxD depleted PxylP:grxDΔ19/ΔsreA compared to the appropriate reference (ΔsreA) indicates that GrxD is not only required to inactivate SreA-mediated repression of mirB, but also for the induction of mirB expression, likely via activation of HapX inducing function.
Interestingly, grxD was also de-repressed during iron sufficiency in a SreA deficient strain (Fig 8A), suggesting that SreA is a repressor of grxD transcription during iron sufficiency. In agreement, MEME analysis [42] of grxD promoter regions of 20 different Aspergillus species identified the highly conserved motif 5´-ATCWGATAA-3´ (S8 Fig), which was previously shown to be the consensus motif for DNA-binding by SreA [1]. This regulatory pattern is similar to that in S. pombe, since grx4 transcript levels are about 2-fold elevated in iron-starved cells [43], but contrasts the situation in S. cerevisiae because grx4 is here under control of Yap5, which activates grx4 gene expression in iron excess conditions [44].
Previously, HapX was shown to be essential for transcriptional short-term induction of iron-consuming genes [3]. To investigate whether this induction depends on GrxD, we shifted GrxD-depleted cells from iron starvation to iron sufficiency (Fig 8B). Such a shift causes extensive transcriptional rearrangements including repression of iron uptake (mainly via SreA, [1]) and induction of iron-consuming genes (mainly via HapX, [3]). Remarkably, GrxD depletion did not completely block induction of sreA and cccA in this set-up indicating independence of GrxD.
To prove that this induction is not mediated by remaining GrxD protein levels upon GrxD depletion, Northern blot analysis was performed using strain ΔgrxD/ΔsreA, which lacks GrxD and SreA. The shift from iron starvation to iron sufficiency still induced cccA in this mutant, although the response was decreased compared to wt (Fig 8C). cccA is exclusively regulated by HapX [3] and therefore its induction during sFe proves that GrxD is, at least partially, dispensable for HapX function during iron excess. The most likely explanation for the decreased response is the transcriptional downregulation of iron acquisition mechanisms during iron starvation in GrxD-lacking cells (see above), which decreases iron uptake in the iron shift.
In summary, these data indicate that GrxD is required during iron starvation conditions to activate HapX iron starvation function (i.e. repression of iron-consuming genes and induction of iron uptake) and to inactivate SreA function (i.e. repression of iron uptake), but not for iron sensing by HapX under iron excess.
FeS clusters in GrxD homologs are coordinated by C191 in the CGFS motif located in the Grx domain (Fig 1). To analyze the function of this cysteine residue in A. fumigatus iron-regulation, we overexpressed C-terminal venus-tagged grxD-variants (targeted to the pksP locus and expressed under control of the strong constitutive PgpdA promoter of glyceraldehyde-3-phosphate dehydrogenase encoding gene) using PxylP:grxDΔ19 as recipient strain (Fig 9A). This strategy allowed for growth during induction with xylose regardless of the functionality of the pksP-targeted grxD-variant due to grxDΔ19 expression of the endogenous PxylP-controlled grxD gene. Without xylose induction, only the pksP-located version is expressed allowing phenotypical characterization of the pksP-targeted grxD variant.
Overexpression of grxDC191A was unable to rescue the growth defect caused by lack of GrxD (non-inducing conditions) during iron starvation and iron sufficiency, demonstrating that replacement of cysteine residue 191 by alanine blocks GrxD function (Fig 9B). In contrast, expression of grxDvenusC191S was able to rescue the lack of GrxD during iron sufficiency but not iron starvation (Fig 9B). Similarly, serine can partially compensate for the function of this cysteine residue in the S. cerevisiae GrxD homolog [19,45]. Endogenous (wt) GrxD protein levels are highest under iron starvation (Fig 3B), indicating a higher GrxD requirement under iron starvation, which might explain the lack of compensation by GrxDVenusC191S under this condition. Alternatively, C191 might be particularly important for adaptation to iron starvation. Interestingly, under xylose-inducing conditions (leading to expression of grxDΔ19) overexpression of grxDC191A or grxDC191S decreased growth particularly during iron starvation indicating a dominant negative effect of these GrxD variants.
As overexpression of grxDvenusC191S was partially able to compensate downregulation of grxDΔ19, we generated a mutant strain expressing exclusively PxylP-driven grxDvenusC191S (Fig 9A). Indeed, overexpression (xylose-induction) of grxDvenusC191S also enabled growth in this set-up in an iron supply-dependent manner: wt-like (or even better than wt) growth during high iron conditions, decreased growth during iron sufficiency but only poor growth during iron starvation (Fig 9C), as observed above in PxylP:grxDΔ19/PgpdA:grxDvenusC191S (Fig 9B).
Northern analysis revealed that overexpression of either grxDvenus or grxDvenusC191S increased expression of hapX during iron starvation (Fig 9D). As hapX expression is mainly regulated by SreA repression, these data indicate that overexpression of either grxDvenus or grxDvenusC191S inactivates SreA. In agreement, GrxD deficiency constitutively activated SreA (Fig 8A). Remarkably, overexpression of grxDvenusC191S but not grxDvenus decreased expression of mirB during iron starvation (Fig 9D). This result resembles GrxD deficiency (Fig 8B) and indicates that the residual function of GrxDVenusC191S is not sufficient to maintain the iron-regulatory function under iron starvation. As mirB expression requires not only inactivation of SreA (and SreA is highly inactivated as judged by the hapX expression) but also induction by HapX, these findings indicate that GrxDVenusC191S fails to activate HapX in contrast to GrxDVenus. In contrast to iron starvation, overexpression of grxDvenusC191S or grxDvenus had no significant effect on these genes during iron sufficiency (Fig 9D). Taken together, these data underline the importance of GrxD for sensing of iron starvation.
As shown previously [23] and above (Fig 8B), a short-term shift from iron starvation to iron sufficiency upregulates sreA and cccA. This response was previously shown to be mediated by HapX [3] and does not require GrxD as shown here (Fig 8B and 8C). Remarkably, however, this regulation was blocked by overexpression of GrxDVenus but not GrxDVenusC191S (Fig 9D). As GrxD dimers are capable of [2Fe-2S] cluster coordination, these data might indicate that GrxDVenus but not GrxDVenusC191S competes with HapX for [2Fe-2S] and thereby blocks activation of the high-iron function of HapX. In agreement, a grxDvenus overexpressing strain displayed severe growth deficiencies at excess iron conditions (Fig 2C). The observed difference between GrxDVenus and GrxDVenusC191S in these experiments is most likely based on the decreased [2Fe-2S] binding affinity of GrxDVenusC191S compared to GrxDVenus.
Recently, we have shown that iron sensing in A. fumigatus depends on a signal from mitochondrial (ISC) but not on cytosolic (CIA) iron-sulfur cluster biosynthesis and on glutathione biosynthesis [23]. Here we demonstrate that A. fumigatus monothiol glutaredoxin GrxD is required to activate HapX-mediated adaptation to iron starvation as well as for inactivation of SreA during iron starvation. Thereby GrxD acts as sensor for iron starvation, most likely by modulating the signal for iron availability, which is generated by ISC.
GrxD homologs have previously been shown to be involved in iron sensing in the ascomycetous yeast species S. cerevisiae, S. pombe and the basidiomycetous yeast species Cryptococcus neoformans [15,19,46,47]. Yet, these fungal species and the filamentous ascomycete A. fumigatus display significant differences with respect to transcriptional iron regulators and the role of the GrxD homologs.
S. cerevisiae employs two paralogs, Grx3/4, which are essential for growth dependent on the genetic background [19]; in S. pombe, mutants lacking Grx4 are viable only under microaerophilic conditions [43,46]; in C. neoformans, deletion of the entire Grx4 gene but not truncation of the Grx domain is lethal [47]. Here we demonstrate that in A. fumigatus GrxD is essential for growth, whereby the cysteine residue in the Grx domain plays a crucial role, while the Trx domain is dispensable for growth, at least when the Grx domain is overexpressed. As shown for Grx4 in S. cerevisiae [19], GrxD has most likely also a dual function in A. fumigatus: a regulatory role in iron sensing as well as in transport of [2Fe-2S] clusters in cellular metabolism. Moreover, Grx3/4 have been suggested to be involved in stress resistance in S. cerevisiae via affecting actin dynamics and Sir2 glutathionylation [48,49].
In agreement, co-IP approaches revealed physical interaction of GrxD not only with the iron regulators SreA and HapX, but also with CIA components. Likewise, physical interaction of Arabidopsis thaliana GrxD homolog GRXS17 and CIA components has been observed previously [50]. Lethality of lack of GrxD might be a synergistic effect of its dual roles. The fact that we found that lack of SreA suppresses the lethal effect of lack of GrxD and that high iron supplementation suppresses the growth defect caused by GrxD downregulation indicates however that the role in iron sensing is the major reason for its essentiality under standard conditions. Our in vivo approaches indicated that the Trx domain of GrxD is required for interaction with SreA but not HapX. In agreement, in vitro studies with recombinant proteins revealed that neither the Trx domain nor the cysteine residue in CGFS motif in the Grx domain, which is essential for the [2Fe-2S] cluster coordination, are required for physical interaction of GrxD with HapX. Moreover, cysteine residues, which have previously been shown to be essential for in vivo function of HapX under high-iron conditions [3], were found to be dispensable for physical interaction of GrxD with HapX.
The paralogous S. cerevisiae transcription factors mediating adaptation to iron starvation, Aft1/2, are conserved exclusively in closely related Saccharomycotina and do not display any similarity to HapX or SreA. In S. cerevisiae, lack of Grx3/4 results in constitutive activation of Aft1/2 irrespective of the iron status. Thus, Grx3/4 is required for inactivation of Aft1/2 during iron sufficiency [15], i.e. sensing of iron sufficiency. The S. cerevisiae transcription factor mediating adaption to iron excess, Yap5 shows similarities to HapX, but has no function during iron starvation [11]. This indicates that HapX homologs have evolved in a modular manner, whereby A. fumigatus HapX combines protein modules and respective functions for adaption to iron excess from S. cerevisiae Yap5 and functions for adaption to iron starvation from S. pombe Php4 (see below). Similar to Yap5, HapX contains two cysteine-rich regions (CRR), which are crucial for high iron functions [3], whereby one of these contains a perfectly conserved CGFC motif, which was shown to be essential for Yap5 function and [2Fe-2S] cluster coordination [12]. We found in the current study that recombinant HapX displays a reddish-brown color and a UV-Vis spectrum indicative of [2Fe-2S] coordination. Together with our previous observation that activation of the HapX high-iron function depends on ISC but not CIA, our data indicate that HapX senses high iron conditions via [2Fe-2S] coordination similar to Yap5. Remarkably, [2Fe-2S] coordination by Yap5 was shown to be independent of Grx3/4 [12]. Similarly, we also observed that GrxD is dispensable for the activation of the HapX high-iron function in A. fumigatus (Figs 8B and 8C and 10).
The transcription factors maintaining iron homeostasis in S. pombe are termed Fep1 and Php4 [13,51]. Fep1 is a homolog of SreA and shares the same function. The HapX homolog Php4 lacks a bZIP-type DNA-binding region but, similar to HapX, interacts with the Php2/Php3/Php5 CBC via its N-terminal CBC-binding domain resulting in repression of iron-consuming pathways under iron starvation [51]. However, in contrast to HapX, Php4 appears to lack a function in activation of iron acquisition during iron starvation and is not involved in adaptation to iron excess. In agreement, the CRR that are conserved and essential for high-iron functions in S. cerevisiae Yap5 and A. fumigatus HapX are not conserved in Php4. In S. pombe, lack of Grx4 caused constitutive activation of the repressing functions of both Php4 and Fep1 [46], i.e. it caused repression of iron acquisition during iron starvation via Fep1 and repression of iron-consuming pathways during iron sufficiency via Php4 and, therefore, deleterious effects during both iron starvation and sufficiency. This finding contrasts the situation in A. fumigatus, in which lack of GrxD caused regulatory defects only during iron starvation. Thus, GrxD appears to modulate the activity of SreA in A. fumigatus in a similar way as Grx4 affects Fep1 in S. pombe (Fig 10). In contrast to Php4 in S. pombe, however, lack of GrxD did not trigger constitutive HapX iron starvation functions. On the contrary, GrxD depletion impaired HapX mediated adaptation to iron starvation (Fig 10), which indicates significant mechanistic differences in the mode of action of the monothiol glutaredoxin in regulation of S. pombe Php4 and A. fumigatus HapX. In S. pombe, Php4 and Grx4 form a heterodimer, irrespective of the cellular iron status via the Trx domain of Grx4 [32]. During iron sufficiency Php4 and Grx4 are suggested to coordinate a [2Fe-2S] cluster with GSH as additional ligand [16], which causes export from the nucleus to block Php4 activity. In contrast to Php4, HapX appears to coordinate [2Fe-2S] clusters also without GrxD, similar to S. cerevisiae Yap5 (see above). Unlike S. pombe Php4, A. fumigatus HapX also has a function in high-iron conditions and therefore it is unlikely that inactivation of HapX iron-starvation functions (repression of iron-consuming pathways, activation of iron acquisition) involves export of HapX from the nucleus, which could explain evolution of mechanistic differences in modulation of activity of Php4 and HapX.
C. neoformans employs homologs of A. fumigatus SreA and HapX, termed Cir1 and HapX, respectively [52,53]. In contrast to A. fumigatus SreA, however, Cir1 is also involved in adaptation to iron starvation, e.g. activation of iron acquisition. Recently, the GrxD homologue Grx4 was demonstrated to be essential for activation of Cir1 functions via physical interaction, i.e. lack of GrxD phenocopied lack of Cir1 [47]. This differs from the situation in A. fumigatus, in which lack of GrxD renders SreA constitutively active.
Taken together, the role of GrxD homologs in iron sensing has been demonstrated in different fungal species. In all these species, GrxD homologs display physical interaction with the employed iron regulators. However, these transcription factors show in part significant differences in protein domains and mode of action. These differences are most likely the reason for the different regulatory consequences of lack of GrxD in the analyzed species. Moreover, GrxD homologs show different regulatory patterns in different fungal species. In S. cerevisiae, expression of the Grx3/4-encoding genes is upregulated during iron sufficiency compared to iron starvation, which is mediated by Yap5 [44]. In contrast, in S. pombe and C. neoformans, Grx4 is upregulated during iron starvation compared to iron sufficiency [43,47]. In these species, Grx4 is preferentially located in the nucleus. C. neoformans, Grx4, however, shows increased nuclear localization under iron starvation compared to iron sufficiency [46,47]. For A. fumigatus GrxD we found a similar expression and localization pattern as in C. neoformans. Moreover, we discovered a negative feedback-loop between GrxD and SreA: GrxD is required to repress the function of SreA during iron starvation, while SreA transcriptionally represses expression of the GrxD encoding gene during iron sufficiency (Fig 10).
Iron sensing by S. cerevisiae Aft1/2 and S. pombe Fep1 has been shown to involve not only a GrxD homolog but also a cytosolic BolA2-like protein, termed Fra2. In both organisms Fra2 deficiency resembles Grx3/4 or Grx4 deficiency, i.e. a constitutive increase of iron uptake in S. cerevisiae and constitutive repression of iron uptake in S. pombe [54,55]. Similar to S. cerevisiae and S. pombe [56], the genome of A. fumigatus and other Aspergillus species encodes two BolA-like proteins containing mitochondrial targeting sequences. However, in contrast to S. cerevisiae and S. pombe, Aspergillus spp. appear to lack a cytosolic BolA2-like protein (although dual localization cannot be excluded) indicating another possible difference in the iron sensing apparatus in these molds.
An intriguing question is of course how GrxD mechanistically modulates the function of SreA and HapX. For S. pombe it has been suggested that GrxD signals iron starvation to Fep1 by removing iron, not [2Fe-2S], bound by Fep1 [46]. Later on, it was shown that Fep1 coordinates a [2Fe-2S] cluster, not iron, by a highly conserved CRR [57]. Nevertheless, GrxD-mediated removal of [2Fe-2S] clusters bound by SreA and HapX appears to be a conceivable mode of action for signaling iron starvation. Such a model is supported by the fact that overexpression of grxD impaired adaptation to iron sufficiency, i.e it blocked short-term induction of cccA expression, which depends exclusively on HapX [3]. This effect was not seen when the [2Fe-2S] cluster coordinating cysteine residue in the CGFS motif of GrxD was replaced by a serine residue, which decreases the affinity for the [2Fe-2S] cluster [19]. These data might suggest that in this set-up GrxD competes for [2Fe-2S] clusters with HapX, which impairs iron sensing by HapX. Moreover, this cysteine to serine exchange also impaired transcriptional adaptation to iron starvation, i.e. high-affinity [2Fe-2S] binding by GrxD is crucial for sensing iron starvation. The severe growth defect of downregulation of GrxD in A. fumigatus is likely a combination of deficiencies in iron sensing and [2Fe-2S] transport. Alternative to GrxD-mediated removal of [2Fe-2S] clusters bound by SreA and HapX, GrxD might signal iron starvation in complexes with HapX and SreA by inducing conformational changes upon [2Fe-2S] cluster coordination. Thus, the cytosolic monothiol glutaredoxin GrxD is involved in iron sensing in A. fumigatus as shown previously for other fungal species. However, our studies revealed significant differences in the mode of action of GrxD and the consequences of the lack of GrxD in this mold, which underlines a remarkable plasticity in iron sensing in fungi.
The virulence defect of A. fumigatus mutants lacking siderophore biosynthesis [58–60] or HapX [2], as well as the transcriptional upregulation of iron acquisition pathways [61] in murine infection models indicate that A. fumigatus faces iron starvation in vivo. Moreover, plasma was recently shown to inhibit growth of A. fumigatus as long as transferrin was not iron saturated, i.e., in the absence of”non-transferrin bound iron” [62]. In line with A. fumigatus facing iron starvation during growth in plasma we found that GrxD localizes to the nucleus during growth in plasma (S9 Fig) similar to growth during iron starvation in minimal medium (Fig 4). In contrast, supplementation of plasma with high amounts of iron blocked the predominant nuclear localization (S9 Fig) similar to growth under iron sufficiency in minimal medium (Fig 4). Taken together, these data implicate that GrxD plays a role in adaptation to iron starvation during infection. In this regard noteworthy, lack of the Grx domain in the GrxD ortholog renders of C. neoformans avirulent in a murine infection model [47]. Moreover, the essential role of GrxD for viability of A. fumigatus underlines the importance of iron metabolism and homeostasis.
Strains used in this study are listed in S2 Table. Oligonucleotides used in this study are listed in S3 Table. Growth assays were performed in Aspergillus minimal medium (1% (w/v) glucose, 20 mM glutamine, salt solution and iron-free trace elements according to [63] and Aspergillus complex medium (2% (w/v) glucose, 0.2% (w/v) peptone, 0.1% (w/v) yeast extract, 0.1% (w/v) casamino acids, salt solution and iron-free trace elements according to [63]. Iron (FeSO4) was added separately as indicated in the respective figures. However, -Fe, +Fe and sFe stands for iron starvation (no iron), 0.03 mM iron, and shift to 0.03 mM iron after precedent iron starvation, respectively. PxylP-driven genes are repressed unless xylose (w/v) is added to the medium, which is indicated in the respective Figures. For solid growth, the medium was solidified with 1.8% (w/v) agarose.
In phase one, 108 spores of strains of interest were shaken in 50 ml minimal medium +Fe at 25°C with 0.1% (w/v) xylose (inducing conditions to enable GrxDΔ19 production and thereby growth) for 20 h. Germlings were centrifuged and washed once with water to remove iron and xylose before being re-suspended in 100 ml minimal medium containing no xylose. To deplete already produced GrxDΔ19 in phase two, growth was continued for 20 h at 37°C. During phase two, the growth conditions were -Fe, +Fe or sFe. Controls were treated the same way.
For microscopy in minimal medium, strains were grown in well chamber slides (Ibidi) with 2 x 104 spores/well (final concentration 105/ml) for 18h at 37°C with 0.05% (w/v) xylose under iron starvation (-Fe) or iron sufficiency (+Fe). Growth in these chamber slides was hardly sufficient to generate iron starvation after 18 h. To increase iron starvation, -Fe media contained 0.5 mM of the ferrous-iron chelator bathophenanthroline disulfonic acid (BPS). For growth in human blood plasma, spores were inoculated in plasma without or with spiking with 0.1 mM iron to override iron starvation. Spore inoculation and incubation was identical to microscopy with minimal medium. Human plasma was obtained from the bloodbank of Medical University Innsbruck and treated as described previously [62].
Mycelia were examined with a spinning-disc confocal microscopic system (Ultra VIEW VoX; PerkinElmer, Waltham, MA) that was connected to a Zeiss AxioObserver Z1 inverted microscope (Zeiss, Oberkochen, Germany). Images were acquired with Volocity software (PerkinElmer) with a 63x oil immersion objective with a numerical aperture of 1.4. The laser wavelengths used for excitation of Venus and mRFP were 488 and 561 nm, respectively
A schematic overview for the generation of all mutant strains is given in S2 Fig.
To simultaneously exchange the endogenous promoter of grxD and include a Venus-tag, a plasmid containing grxD 5’-region, hph, PxylP, grxD (including 3’-region) and pUC19 backbone was generated. Parts of this plasmid were amplified with primers oKM11-16 and pMMHL15 [23] or A. fumigatus wt gDNA as template and finally assembled with NEBuilder (New England Biolabs) in pUC19 yielding plasmid pKM1. Subsequently pKM1 was linearized with oKM26 and oKM27 to integrate the venus-tag (amplified with oKM28 and oKM29 from phapXVENUS-hph [3]) via seamless cloning (NEBuilder; New England Biolabs) yielding pKM1+venus. The insert of pKM1+venus was amplified with primers oKM11 and oKM16 and transformed into a wt recipient strain via homologous recombination. Thereby endogenous grxD was exchanged. As two possibilities for homologous recombination at the grxD locus were available (S2 Fig), we received two types of transformants, PxylP:grxD and PxylP:grxDvenus.
Site-directed mutagenesis (Q5 Site-Directed Mutagenesis Kit; New England Biolabs) was performed with pKM1+venus (see above) and primers oMM182 and oMM184 yielding pMMHL43. The insert was amplified with oKM11 and oKM16 and transformed into a wt recipient strain yielding strain PxylP:grxDvenusΔtrx via homologous recombination at the grxD locus. Thereby endogenous grxD was exchanged.
To integrate mRFP-tagged histone H2A driven by constitutive gpdA promoter, a plasmid containing fragment PgpdA:mRFP:H2A, a phleomycin resistance cassette (ble), a pksP homologous site and pUC19 backbone was generated. Subunits of this plasmid were amplified with primers oMM189-194 and plasmid pME3173 [64], A. fumigatus wt gDNA or pAN8-1 [65], respectively, as template and finally assembled with NEBuilder (New England Biolabs) in pUC19 yielding plasmid pMMHL44. The plasmid was linearized with BamHI and integrated into the pksP locus of recipient strains (PxylP:grxDvenus or PxylP:grxDvenusΔtrx) via homologous recombination at the pksP locus. This gene encodes for a polyketide synthase, which is involved in conidial pigmentation [66]. Disruption of pksP allows for fast screening of positive integrations, as ΔpksP strains produce white conidia.
To delete sreA, a plasmid containing sreA 5’-region, a pyrithiamine resistance cassette (ptrA), sreA 3’-region and pUC19 backbone was generated. Subunits of this plasmid were amplified with primers oMM164-169 and A. fumigatus wt gDNA or pSK275 (syn. pME3024 [67]) as template and finally assembled with NEBuilder (New England Biolabs) in pUC19 yielding plasmid pMMHL38. The insert of pMMHL38 was amplified with oMM164 and oMM169 and transformed into a wt recipient strain. Thereby sreA was deleted via homologous recombination.
venus-tagging of grxD was performed by employing CRISPR technology as described previously [68]. We used the hygromycin resistance-mediating AMA-plasmid pFC332 and grxD targeting sequence AGGCTCCTGCCAGCGCTTGA as protospacer sequence, yielding pMMHL49. A repair template was amplified with oKM15 and oKM16 from pKM1+venus (see above). The repair template and pMMHL49 were together transformed into a wt recipient strain. This procedure caused cleavage at the grxD locus by CRISPR and integration of the repair template via homologous recombination. By subsequent growth on non-selective media the CRISPR plasmid was lost yielding grxDvenus, a marker-free strain, in which endogenous grxD is tagged with venus without further manipulation of the grxD locus.
The 5’-region of grxD was amplified with primers oAfgrx4-oe1 and oAfgrx4-oe2 and digested with AvrII (fragment A). Truncated grxD was amplified with primers oAfgrx4-oe4 and oAfgrx4-oe5 and digested with NcoI. The PxylP sequence was liberated from plasmid pxylPp [69] by digestion with NotI and NcoI. Both, truncated grxD and PxylP were ligated via their NotI overhang, the fragment was amplified with primers oAfgrx4-oe6 and oAfgrx4-oe7 and digested with XbaI (fragment B). The hygromycin resistance cassette was released from plasmid pAN7-1 by digestion with XbaI and AvrII (fragment C). Fragments A, B and C were ligated via AvrII and XbaI overhangs. The resulting fragment was amplified with primers oAfgrx4-oe3 and Afgrx4-oe8 and integrated into a wt recipient strain via homologous recombination at the grxD locus yielding PxylP:grxDΔ19. Thereby endogenous grxD was exchanged. As grxD is essential (see Results) growth under non-inducing conditions (no xylose) was inhibited. However, streaking out >108 spores on non-inducing agar plates yielded colonies. At least one of these, designated as PxylP:grxDΔ19sup, harbored a mutation suppressing the lethal effect caused by grxD deficiency.
A construct containing grxD 5’-region, hph, PxylP and the 19 aa truncated version of grxD as 3’-homologous region was amplified from strain PxylP:grxDΔ19 gDNA with primers oAfgrx4-1 and oAfgrx4-oe5 and transformed into grxDvenus as recipient strain via homologous recombination.
To inactivate sreA or hapX in a PxylP:grxDΔ19 background, the knockout constructs were amplified from ΔsreA or ΔhapX gDNA with primers oMM164 and oMM169 or oAfhapX-1 and oAfhapX-2, respectively, and transformed into a PxylP:grxDΔ19 recipient strain via homologous recombination yielding strains PxylP:grxDΔ19/ΔsreA and PxylP:grxDΔ19/ΔhapX.
To inactivate grxD, a plasmid containing grxD 5’-region, hph, grxD 3’-region and pUC19 backbone was generated. Subunits of this plasmid were amplified with primers oMM301-306 and A. fumigatus wt gDNA or pAN7-1 [70] as template and finally assembled with NEBuilder (New England Biolabs) in pUC19 yielding plasmid pMMHL61. The insert of pMMHL61 was amplified with oMM301 and oMM306 and transformed into wt as recipient strain. This procedure yielded heterokaryotic transformants, containing two different nuclei (grxD+hph-; wt; containing grxD but lacking hph and grxD-hph+; ΔgrxD; lacking grxD but containing hph) as described in Results. The amplified cassette was also transformed into ΔsreA as recipient strain. Thereby grxD was deleted via homologous recombination.
To rescue grxD deficiency, a plasmid was generated containing pksP and PgpdA:grxD:venus in backbone PgpdA-lacZ-trpCT-pJET1.2 [71]. The pksP fragment was amplified with oAf-pksP1-f and oAf-pksP2-r and integrated into the HindIII site of PgpdA-lacZ-trpCT-pJET1.2 yielding pMMHL6. Subsequently, pMMHL6 was partially amplified with oMM156_HL6fwd and oMM157_HL6rev and assembled with grxD:venus amplified from pKM1+venus with primers oMM158_grxDfwd and oMM159_venus_rev using NEBuilder (New England Biolabs). The resulting plasmid pMMHL37 was used for site-directed mutagenesis (Q5 Site-Directed Mutagenesis Kit; New England Biolabs) with primers oMM313 and oMM314 or oMM314 and oMM315 to generate pMMHL63 or pMMHL64, respectively. pMMHL37, pMMHL63 and pMMHL64 were linearized with FseI and transformed into PxylP:grxDΔ19 as recipient strain to obtain strains PxylP:grxDΔ19/PgpdA:grxDvenus, PxylP:grxDΔ19/PgpdA:grxDvenusC191A and PxylP:grxDΔ19/PgpdA:grxDvenusC191S via homologous recombination in locus pksP.
To exchange endogenous grxD by a PxylP-driven grxD version in which cysteine 191 is replaced by serine, pKM1+venus was used for site directed mutagenesis (Q5 Site-Directed Mutagenesis Kit; New England Biolabs) with primers oMM313 and oMM314 yielding pMMHL65. The insert was amplified with oKM11 and oKM16 and transformed into a wt recipient strain yielding strain PxylP:grxDvenusC191S via homologous recombination in locus grxD. Thereby endogenous grxD was exchanged.
To constitutively express venus tagged bol1 from the pksP locus, a plasmid was generated consisting of PgpdA-driven bol1 followed by venus assembled in pMMHL37 as backbone. Therefore, bol1 was amplified with primers oMM358 and oMM359 from A. fumigatus wt gDNA and assembled with linearized pMMHL37 (linearized with primers oMM356 and oMM357) in a NEBuilder (New England Biolabs) reaction yielding the final plasmid pMMHL83. This plasmid was subsequently linearized with FseI and integrated into locus pksP via homologous recombination.
RNA was isolated using TRI Reagents (Sigma) according to the manufacturer’s manual. 10 μg of RNA was used for electrophoresis on 2.2 M formaldehyde agarose gels and subsequently blotted onto Amersham Hybond-N Membranes (ThermoFisher). Transcripts of interest were detected with DIG-labeled probes amplified by PCR.
DNA was isolated by PCI extraction and isopropanol precipitation. To confirm the gene-specific restriction pattern of the genetic manipulations, DNA was digested with restriction enzymes specific for the respective gene. The resulting restriction fragments were separated on an agarose gel and transferred to Amersham Hybond-N Membranes (ThermoFisher) by capillary blotting with NaOH. Signals for correct integration were detected with DIG-labeled probes amplified by PCR.
Rabbits were immunized with polypeptides corresponding to the amino acid residues of HapX161-491 and SreA308-546. Sequences were PCR-amplified as NdeI-NotI fragments from cDNA and inserted into a pET-21b(+) vector (Novagen) to obtain polypeptides with a C-terminal 6x-His tag. The resulting plasmids were introduced into E. coli Rosetta BL21 cells (Novagen), designed to enhance the expression of eukaryotic proteins that contain codons rarely used in E. coli. Expression was induced for 4 h at 37°C with 0.1 mM isopropyl-β-D-thiogalactopyranoside (IPTG). Proteins were purified from cleared lysates by incubation, 2 h at 4°C, with 0.5 ml of Ni-NTA Agarose Resin (Qiagen). Beads were washed repeatedly with phosphate buffer saline (PBS) containing 75 mM imidazole followed by PBS with 100 mM imidazole before proteins were eluted with 500 mM imidazole. Imidazole was removed by extensive dialysis against PBS. Protein material was lyophilized and used to immunize rabbits (by Davids Biotechnologie GmbH, Regensburg, Germany). The specificity of the obtained antibodies was tested by Western analysis (S10 Fig).
Proteins were extracted using a reported procedure [72] involving solubilization from lyophilized mycelial biomass with NaOH, followed by their precipitation with trichloroacetic acid (TCA). Aliquots were resolved in 10–12% (w/v) SDS-polycrylamide gels and transferred to nitrocellulose membranes. Western blots were reacted with rabbit α-HapX or rabbit α-SreA antiserum (1:20,000), mouse α-GFP antibody (1:10,000; Roche, 11814460001) mouse α‐Tub antibody (1:10,000; Sigma, T6199) as primary antibodies and with peroxidase coupled antibodies as secondary antibodies (1:10,000; anti-Rabbit; Sigma, A1949 or 1:10,000; anti-Mouse; Sigma, A4416). Proteins were detected using Amersham Biosciences ECL.
Covalent coupling of rabbit α-HapX respectively rabbit α-SreA antibodies (antiserum) to Protein-A-Sepharose (GE Healthcare) was performed according to [73]. For the negative control IgGs contained in preserum were covalently linked to Protein-A-Sepharose. In short: 1 ml of Protein-A-Sepharose slurry (50%) was mixed with 0.5 ml (anti)serum and treated with 20 mM dimethylpimelidate in 0.2 M Na-tetraborate. The reaction was stopped with 0.2 M ethanolamine. Immunoprecipitation assays were performed according to [74]. Mycelia were grown for 16 h in minimal medium containing 0.1% xylose and no iron supplementation for HapX immunoprecipitation, or 0.03 mM iron for SreA immunoprecipitation. For protein extracts, 40 mg of mycelia were grinded and dissolved in 1 ml protein extraction buffer containing 20 mM Tris-HCl pH 8, 110 mM KCl, 10% (v/v) glycerol, 0.1% (v/v) Triton X-100, 1μl BitNuclease (Biotool) and protease inhibitor (cOmplete ULTRA EDTA-free, Roche). Extracts were mixed with 50 μl of covalently linked rabbit α-HapX, rabbit α-SreA or rabbit preserum beads and incubated for 3h at 4°C in a rotating wheel. Subsequently the beads were washed three times (10 min at 4°C in a rotating wheel) with chilled protein extraction buffer and increasing salt concentrations (110 mM, 500 mM and 750 mM KCl). Bound proteins were eluted in 40 μl of Laemmli sample buffer at 95°C. Twenty microliters of aliquots were resolved in 10% SDS-polyacrylamide gels and transferred to nitrocellulose for Venus detection. Venus tagged GrxD or GrxDΔTrx was detected with mouse α-GFP (1:10,000; Roche, 11814460001) and α‐Tubulin was detected with mouse α‐Tub (1:10,000; Sigma, T6199) as primary antibody and with a peroxidase-coupled secondary antibody (1:10,000; anti-Mouse IgG; Sigma, A4416). HapX respectively SreA were detected with rabbit α-HapX or rabbit α-SreA antisera (1:20,000). To avoid the detection of rabbit IgGs, which were used for the co-IP, a conformation specific anti-Rabbit IgG antibody (1:1000; Cell Signaling Technology, L27A9) was used in combination with a peroxidase-coupled anti-Mouse IgG secondary antibody (1:10,000; Sigma, A4416). For the detection Amersham Biosciences ECL was used.
3’ RACE was performed using FirstChoice RLM-RACE Kit (ThermoFisher). Total RNA from PxylP:grxDΔ19sup was reverse transcribed with the oligo-dT containing primer 3' RACE Adapter. The resulting cDNA was used for Touchdown PCR with sreA (5’-UTR)-specific forward primer oKM31 and adapter-specific reverse primer 3’ RACE Outer Primer. To increase specificity, the resulting PCR product(s) were amplified in a second PCR with nested primers oKM30 and 3’ RACE Inner Primer. This procedure yielded a fragment (~900bp) which was isolated and sequenced (S3 Fig).
A. fumigatus mycelia were harvested in Stop buffer [75] at 4°C after growth for 22 h and freeze-dried. Protein extraction was performed according to a modified procedure from [75] using HK buffer for total protein extraction. All steps were carried out at 4°C in the cold room. In short, 100 mg of mycelium powder was dissolved in 1 ml HK buffer, centrifuged twice at 20,187 x g for 15 min and 500 μl of the supernatant was incubated with GFP-Trap agarose beads (ChromoTek) for 1 h. The beads were washed twice in HK buffer without IGPAL, twice in wash buffer (25 mM Tris/HCl pH 7.5, 500 mM NaCl, 5 mM EDTA and 15 mM EGTA) and once in ultrapure water. Proteins were eluted in 10% (v/v) acetonitrile and 5% (v/v) acetic acid and used for nLC-MS/MS measurement, Western blot detection and silver staining.
LC-MS/MS analysis was carried out on an Ultimate 3000 nano (n) RSLC system coupled to a QExactive Plus mass spectrometer (both Thermo Fisher Scientific, Waltham, MA, USA). Peptides were trapped for 5 min on an Acclaim Pep Map 100 column (2 cm x 75 μm, 3 μm) at 5 μl/min followed by gradient elution separation on an Acclaim Pep Map RSLC column (50 cm x 75 μm, 2μm). Eluent A (0.1% (v/v) formic acid in water) was mixed with eluent B (0.1% (v/v) formic acid in 90/10 acetonitrile/water) as follows: 0 min at 4% B, 6 min at 6% B, 14 min at 10% B, 20 min at 14% B, 35 min at 20% B, 42 min at 26% B, 46 min at 32% B, 52 min at 42% B, 55 min at 50% B, 58min at 65% B, 60–64.9 min at 96% B, 65–90 min at 4% B. Positively charged ions were generated at 2.2 kV using a stainless steel emitter and a Nanospray Flex Ion Source (Thermo Fisher Scientific). The QExactive Plus was operated in Full MS / data-dependent MS2 (Top10) mode. Precursor ions were monitored at m/z 300–1500 at a resolution of 70,000 FWHM (full width at half maximum) using a maximum injection time (ITmax) of 120 ms and an AGC (automatic gain control) target of 1e6. Precursor ions with a charge state of z = 2–5 were filtered at an isolation width of m/z 1.6 amu for HCD fragmentation at 30% normalized collision energy (NCE). MS2 ions were scanned at 17,500 FWHM (ITmax = 120 ms, AGC = 2e5). Dynamic exclusion of precursor ions was set to 20 s. The LC-MS/MS instrument was controlled by QExactive Plus Tune 2.9 and Xcalibur 3.0 with DCMS Link.
Tandem mass spectra were searched against the Aspergillus Genome Database (AspGD) of Aspergillus fumigatus Af293 (http://www.aspergillusgenome.org/download/sequence/A_fumigatus_ Af293/current/A_fumigatus_Af293_current_orf_trans_all.fasta.gz; 2018/09/18) and the protein sequence of Dre2 (AFUB_008090; the Dre2 ortholog is not present in the Af293 gene annotation) as well as further modified protein sequences (e.g. Venus-tag) using Proteome Discoverer (PD) 2.2 (Thermo) and the algorithms of Sequest HT (version of PD2.2) and MS Amanda 2.0. Two missed cleavages were allowed for the tryptic digestion. The precursor mass tolerance was set to 10 ppm and the fragment mass tolerance was set to 0.02 Da. Modifications were defined as dynamic oxidation of Met, acetylation of Ser, phosphorylation of Ser, Thr, and Tyr and ubiquitination (GG) of Lys as well as static Cys carbamidomethylation. At least two peptides per protein and a strict false discovery rate (FDR) < 1% were required for positive protein hits. The Percolator node of PD2.2 and a reverse decoy database was used for q-value validation of spectral matches. Only rank 1 proteins and peptides of the top scored proteins were counted. The Minora algorithm of PD2.2 was applied for relative label-free quantification. GFP-Trap eluates from wt A. fumigatus mycelial extracts were used for quantification of nonspecifically co-purified proteins.
Proteins were separated by SDS-PAGE using NuPAGE 4–12% (w/v) Bis-Tris gradient gels (Invitrogen). Silver staining was performed using the SilverQuest Silver Staining Kit (Invitrogen) according to the manufacturer’s protocol. For Western detection, proteins were transferred onto a PVDF membrane using the iBlot 2 dry blotting system (Invitrogen). The membrane was blocked in 3% (w/v) bovine serum albumin (BSA) dissolved in 1x PBST (137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 2 mM KH2PO4, 0.05% (v/v) Tween 20). As primary antibody rabbit α-GFP (abcam, ab290) was used, followed by secondary antibody HRP-conjugated anti-Rabbit IgG (ICL) incubation. The membrane was developed using the 1-Step Ultra TMB-Blotting chromogenic substrate (Thermo Scientific).
For individual expression and protein purification, synthetic genes coding for full-length GrxD and HapX amino acids 161–491 (cysteine-rich C-terminus) were cloned into the NdeI and BamHI sites of the pET-MCN vector pnEA/vH [76] producing C-terminally His6-tagged GrxD (pnEA/vH-GrxD) and HapX161-491 (pnEA/vH-HapX161-491) fused to a TEV cleavage site. For co-expression, the synthetic gene coding for HapX161-491 was initially cloned into the NdeI and BamHI sites of the pET-MCN vector pnCS producing untagged HapX161-491 (pnCS-HapX161-491). Subsequently, the BglII/XbaI fragment from pnCS-HapX161-491 was subcloned into the BglII and SpeI sites of pnEA/vH-GrxD generating a bicistronic expression cassette. Site-directed mutagenesis was performed with the QuikChange II site-directed mutagenesis kit (Agilent) according to the manufacturer’s protocol. Primers used for mutagenesis are listed in S3 Table.
E. coli BL21(DE3) cells (New England Biolabs) were transformed with the respective plasmid for autoinduction in Overnight Express Instant TB medium (Novagen). Wet biomass was harvested by centrifugation (10,543 x g) and the cell paste was stored at -80°C. Frozen bacterial cells were resuspended in lysis buffer (50 mM HEPES pH 8.0, 300 mM NaCl, 2 mM glutathione, 10 mM imidazole, 1 mM AEBSF) and disrupted at 1000 bar using a high-pressure homogenizer (Avestin Emulsiflex C5). Cell debris were removed by centrifugation (48,384 x g), the pH was adjusted to 8.0 and the supernatant clarified by filtration through a 1.2 μm membrane. His6-tagged proteins were then purified by Ni-chelate affinity chromatography using a 20 ml Ni-Sepharose FF column (GE Healthcare) and proteins were eluted with 500 mM imidazole. Fractions containing either HapX161-491-His6 or the HapX161-491/GrxD-His6 complex were digested with TEV protease for 4 h at room temperature and loaded onto a Superdex 200 prep grade 26/60 size exclusion chromatography column (GE Healthcare) that was equilibrated with 25 mM HEPES pH 7.5, 150 mM NaCl, 2 mM glutathione. UV-Vis absorption spectra were recorded in the range from 250 to 550 nm with a JASCO V-630 spectrophotometer.
|
10.1371/journal.ppat.1003574 | Sexuality Generates Diversity in the Aflatoxin Gene Cluster: Evidence on a Global Scale | Aflatoxins are produced by Aspergillus flavus and A. parasiticus in oil-rich seed and grain crops and are a serious problem in agriculture, with aflatoxin B1 being the most carcinogenic natural compound known. Sexual reproduction in these species occurs between individuals belonging to different vegetative compatibility groups (VCGs). We examined natural genetic variation in 758 isolates of A. flavus, A. parasiticus and A. minisclerotigenes sampled from single peanut fields in the United States (Georgia), Africa (Benin), Argentina (Córdoba), Australia (Queensland) and India (Karnataka). Analysis of DNA sequence variation across multiple intergenic regions in the aflatoxin gene clusters of A. flavus, A. parasiticus and A. minisclerotigenes revealed significant linkage disequilibrium (LD) organized into distinct blocks that are conserved across different localities, suggesting that genetic recombination is nonrandom and a global occurrence. To assess the contributions of asexual and sexual reproduction to fixation and maintenance of toxin chemotype diversity in populations from each locality/species, we tested the null hypothesis of an equal number of MAT1-1 and MAT1-2 mating-type individuals, which is indicative of a sexually recombining population. All samples were clone-corrected using multi-locus sequence typing which associates closely with VCG. For both A. flavus and A. parasiticus, when the proportions of MAT1-1 and MAT1-2 were significantly different, there was more extensive LD in the aflatoxin cluster and populations were fixed for specific toxin chemotype classes, either the non-aflatoxigenic class in A. flavus or the B1-dominant and G1-dominant classes in A. parasiticus. A mating type ratio close to 1∶1 in A. flavus, A. parasiticus and A. minisclerotigenes was associated with higher recombination rates in the aflatoxin cluster and less pronounced chemotype differences in populations. This work shows that the reproductive nature of the population (more sexual versus more asexual) is predictive of aflatoxin chemotype diversity in these agriculturally important fungi.
| Fungal pathogen populations have mixed proportions of vegetative propagation and sexual reproduction ranging from predominantly clonal to varying levels of sexuality. Aflatoxins are the most potent naturally occurring carcinogens known and aflatoxin-producing Aspergillus flavus and A. parasiticus show extensive genetic and mycotoxin diversity. Population genetic studies and experimental matings in the laboratory have revealed the underlying genetic mechanisms and adaptive processes that create and maintain aflatoxin diversity. These studies provided unequivocal evidence of meiosis, crossing over, and aflatoxin heritability, but whether these processes directly influence genetic diversity in nature with respect to aflatoxin formation is not clear. Here, our work with A. flavus, A. parasiticus and A. minisclerotigenes from fields in different continents shows that populations with higher mean recombination rates exhibit less variability in aflatoxin profiles due to genetic intermixing, whereas populations with lower recombination rates have greater variability in aflatoxin profiles due to increased fixation of specific toxin chemotypes. Therefore, sexuality generates novel toxin chemotypes but tends to equalize toxin differences in populations. Our study highlights how an understanding of variation in mating-type frequency, fertility and recombination in these fungi is crucial for the selection of nontoxigenic biocontrol strains for long-term reduction of aflatoxins in target regions.
| Aspergillus flavus and A. parasiticus are important fungal colonizers of food crops as well as pathogens of animals and produce the carcinogenic aflatoxins of which aflatoxin B1 is the most carcinogenic natural compound known [1], [2]. The two species occur in soil and drought stress in plant hosts enhances their pathogenic success [2], [3]. A. flavus has two recognized morphotypes that are differentiated based on sclerotial size. The L- (large) strain of A. flavus forms sclerotia greater than 400 µm in diameter and the S- (small) strain produces sclerotia less than 400 µm [4]. Both strains may produce B1+B2 aflatoxins (AFs) and the toxic indol-tetramic acid, cyclopiazonic acid (CPA) [5]. Aflatoxins and CPA often co-contaminate agricultural products [6]. Another species, A. minisclerotigenes, has the S-strain morphotype and produces both B and G aflatoxins in addition to CPA [7]. The majority of A. parasiticus strains also produce B and G aflatoxins but not CPA [5]; non-aflatoxigenic strains have been reported and typically accumulate O-methylsterigmatocystin (OMST) and dihydro-O-methylsterigmatocystin (DHOMST), the immediate precursors to B aflatoxins [8], [9], [10]. The loss of G aflatoxin production in A. flavus has been attributed to defects in, or complete absence of, the cypA gene that encodes cytochrome P-450 [11]. Moreover, a single point mutation can make the difference between AF+ and AF− strains [12] and partial or complete deletion of genes in AF and CPA clusters are known to exist in A. flavus such that strains may be AF+/CPA+, AF−/CPA−, AF+/CPA−, and AF−/CPA+ [13], [14].
Sexual reproduction in A. flavus L and A. parasiticus is heterothallic and occurs between strains of opposite mating type, either MAT1-1 or MAT1-2 [15], [16], [17], [18], [19]. Much of the observed heterogeneity in AF chemotype diversity in A. flavus and A. parasiticus can be attributed to intra-specific genetic exchange and recombination [18]. Genetic exchange is possible through independent assortment and crossing over during sexual reproduction or through parasexuality in heterokaryons, which are formed by the fusion of vegetatively compatible strains [20], [21]. Vegetative incompatibility among strains gives rise to vegetative compatibility groups (VCGs) that limit genetic exchange through the parasexual cycle and may eventually lead to isolation and homogeneity in toxin phenotype [22]. Aflatoxin production and morphology (sclerotium size and number; conidial color) are highly consistent within a given VCG [23]. In contrast, sexual reproduction in A. flavus and A. parasiticus occurs between individuals that belong to different VCGs and often differ in their toxigenicity [14], [16]. Experimental populations, derived from crossing sexually compatible strains in the laboratory, show high heritability of aflatoxin production in progeny strains as well as patterns of recombination in the aflatoxin cluster that mirror linkage disequilibrium (LD) in field populations [18].
In population genetic studies of a single field population in the United States, we showed that DNA sequence variation is partitioned into several distinct LD blocks across 21 intergenic regions in the aflatoxin gene clusters of A. flavus and A. parasiticus [14], [24]. Moreover, genealogical analysis of non-recombining cluster regions in A. flavus and A. parasiticus revealed trans-species polymorphisms and balancing selection acting on the non-aflatoxigenic trait in A. flavus [14] and on G1 dominant chemotypes in A. parasiticus [24]. In these studies, our ability to detect and estimate more frequent (or recent) recombination events in the aflatoxin cluster relied on the frequency of two or more distinct chemotype allelic classes in a population. In A. flavus L, DNA sequence polymorphisms in the aflatoxin gene cluster were shown to delimit two distinct evolutionary lineages named IB and IC [14], [25]. Lineage IB includes strains with partial or complete deletions of the aflatoxin cluster or full-cluster strains with many fixed polymorphisms when compared to lineage IC, which includes aflatoxigenic isolates and those that are non-aflatoxigenic due to loss-of-function mutations [14]. Lineages IB and IC are phylogenetically distinct based on DNA sequence variation across the entire aflatoxin cluster [14] and genome-wide using oligonucleotide-based array comparative genome hybridization [26]. In A. parasiticus, sequence variation was found to be associated with G1-dominant strains, which share a distinct evolutionary lineage with A. flavus L [24]. Recombination between divergent alleles with many fixed polymorphisms yields distinct LD blocks, whereas reduced recombination activity may be the result of a selective sweep for an advantageous chemotype or a population bottleneck that greatly reduces genetic variation [24].
The correlation between toxin chemotype profile and VCG suggests that asexual reproduction fixes diverse toxin chemotypes in populations whereas sexuality creates new VCGs with different toxin profiles. Although we expect the frequency of mating types to be close to a 1∶1 ratio in heterothallic fungi, a significant skew in the ratio does not imply a decrease in the size of the population undergoing sexual reproduction; this effective population size is also a function of the number of hermaphrodites and female sterile strains [27]. Here, we explore the contributions of asexual and sexual reproduction to mycotoxin diversity in global populations of A. flavus, A. parasiticus and A. minisclerotigenes. This knowledge is integral for improving biocontrol strategies worldwide and providing long-term mitigation of aflatoxin contamination in target regions.
Aspergillus flavus L and S strains, A. parasiticus and A. minisclerotigenes were sampled from peanut field soils collected in different geographic regions representing five continents: United States (North America), Argentina (South America), Queensland (Australia), India (Asia), and Benin (Africa). Ecological data such as climate, peanut cultivar, and soil type were compiled for each region (Table 1). Climate data were based on compilations of monthly measurements taken 1950–2000 at weather stations closest to sampling sites (http://www.worldclim.org/). Twenty equidistant soil samples were collected along a diagonal line spanning each field. Population densities for A. flavus L and S, A. parasiticus and several other species in Aspergillus section Flavi (Table 2) were determined by dilution plating soil samples on modified dichloran-rose bengal medium and counting the number of colony-forming units (CFUs) according to Horn & Dorner [28]. Approximately four isolates each of A. flavus L and S and A. parasiticus per soil sample (when available) were single-spored by dilution plating conidia onto malt extract agar and incubating approximately 20 h at 30 C. Germlings arising from single conidia, as viewed under the light microscope at 200–400×, were then transferred to Czapek agar slants. Sample sizes for populations are shown in Table 3. Concentrations of B and G aflatoxins were determined by growing isolates on yeast extract-sucrose broth and analyzing using high performance liquid chromatography [29]. Strains were grown in 4-mL vials containing 1 mL of yeast extract sucrose broth (sucrose, 150 g; yeast extract [Difco], 20 g; soytone [Difco], 10 g; distilled water, 1 L; pH adjusted to 6.0 with HCl) for 7 d at 30°C in darkness. Vials were inoculated with approximately 1000 dry conidia and incubated under stationary conditions. Mycelial weights were not measured; replicates were incubated at the same time. A. flavus L and S, A. parasiticus and putative A. minisclerotigenes were then grouped into their distinct chemotype classes. The molecular evidence for distinguishing A. flavus S from A. minisclerotigenes is provided below. A. flavus L isolates were categorized as either aflatoxigenic (B1+B2) or non-aflatoxigenic, with non-aflatoxigenic isolates belonging to lineage IB or IC. For the S strain morphotype, chemotype classes were tentatively identified as A. flavus (B only) and A. minisclerotigenes (B+G). For A. parasiticus the three classes were B1 dominant (G1/B1≤0.5), equivalent (0.5<G1/B1<2.0) and G1 dominant (G1/B1≥2). We use the term “chemotype” in a broader sense to include proportionalities. Frequency distributions for distinct toxin chemotype classes were generated for A. flavus L and S (B1+B2), A. parasiticus (G1/B1), and A. minisclerotigenes (G1/B1) isolates from each geographic location. For A. flavus L, we determined the aflatoxin midpoint concentration from frequency distribution plots and the proportion of high B-producing strains (B1+B2>100 µg/mL). We graphically portrayed differences in aflatoxin concentrations for species and morphotypes from each locality using a cumulative distribution function and tested for significant differences between toxin distributions using a Kolmogorov-Smirnov test, as implemented in Matlab (MathWorks Inc., Natick, MA, USA).
Fungal isolates were grown on potato dextrose broth for 3–5 days at 30°C in darkness. Mycelial pellets for each isolate were harvested and freeze dried, and DNA was isolated from a single pellet as previously described [24]. PCR amplification and DNA sequencing of target loci were performed using oligonucleotide primers and thermal cycling conditions, also described previously [24]. Mating types MAT1-1 and MAT1-2 were determined for all isolates using multiplex-PCR [19]. All isolates were clone-corrected using DNA sequence variation at two intergenic cluster regions, aflM/aflN and aflW/aflX, and at two non-cluster loci, acetamidase (amdS) and tryptophan synthase (trpC). This MLST uniquely types approximately 84% and 59% of the VCG diversity in A. flavus and A. parasiticus, respectively [16], [18], [30]. When clone-correcting multilocus haplotypes that contained both mating types, the haplotype was counted twice as a MAT1-1 and a MAT1-2. Phylogenetic reconstructions of DNA sequence variation in aflM/aflN, aflW/aflX, MAT1-1, MAT1-2, amdS and trpC were previously [31] shown to differentiate, into distinct clades, the sympatric A. flavus S strains that produce B aflatoxins from the S-strain morphotype isolates that produce both B and G aflatoxins in Argentina, Australia and Benin (Table 3); moreover, the SBG isolates in the present study are broadly monophyletic with ex type A. minisclerotigenes CBS 117635 [7] and distinct from the small sclerotial A. nomius, A. parvisclerotigenus or an unnamed taxon based on variation in beta-tubulin and calmodulin (data not shown).
Clone correction was performed to eliminate accidentally sampling the same individual multiple times or detecting epidemiological effects that do not contribute to long-term population processes. To do this, the null hypothesis of no significant difference in the frequency of MAT1-1 and MAT1-2 isolates for each species and geographic region was tested using a two-tailed binomial test. The test was performed on two genetic scales: the uncorrected sample and the clone-corrected sample as determined by MLST. A significant difference in mating-type frequency in the uncorrected sample but no significant difference after clone-correction or no significant difference for both uncorrected and clone-corrected samples was interpreted as primarily sexual, whereas a significant difference in mating-type frequency before and after clone-correction was interpreted as primarily asexual [27]. We used Fisher's exact test implemented in Matlab to test the relationship between 1) mating type (MAT1-1 and MAT1-2) and aflatoxin chemotype class (B1+B2>0 and B1+B2 = 0 in A. flavus L; G1/B1≤0.5, 0.5<G1/B1<2.0, G1/B1≥2 in A. parasiticus), and 2) the relationship between the relative proportion of reproduction (asexual>sexual and sexual>asexual) and aflatoxin chemotype class. For A. parasiticus we also performed the tests assuming two broad chemotype classes (G1/B1≈1 and G1/B1≠1).
The influence of asexual and sexual reproduction on recombination in the aflatoxin cluster and overall toxin diversity was examined by reconstructing patterns of LD in the aflatoxin cluster for a subset of isolates representing distinct MLSTs in A. flavus L and S, A. parasiticus and A. minisclerotigenes. Previous population genetic studies showed that multilocus cluster haplotypes are identical within a VCG and that recombination in the aflatoxin cluster is detected only between VCGs [14], [18], [24]. The subset for LD analysis was therefore selected to maximize VCG (MLST) and toxin diversity. Moreover, recombination is nonrandom and species-specific such that LD blocks and recombination hotspots are conserved among geographically separated strains [31]. We therefore determined the LD block structure and rate of recombination in the aflatoxin cluster by focusing on the intergenic regions separating LD blocks identified in the United States populations of A. flavus and A. parasiticus [14], [24]. For A. flavus L and S and A. minisclerotigenes, the regions sequenced were aflE/aflM, aflM/aflN (hypE), aflN/aflG, aflG/aflL, aflL/aflI, and aflI/aflO, which define six distinct LD blocks [14]. For A. parasiticus, we sequenced aflB/aflR, aflS/aflH, aflH/aflJ, aflJ/aflE, aflE/aflM, aflG/aflL, and aflK/aflV, which define five LD blocks [24]. Figure 1 shows a schematic representation of the aflatoxin gene cluster and the regions that were sequenced for LD analysis. LD was examined by 1) combining all sequenced loci for each locality, species and morphotype using SNAP Combine [32] into a single concatenated sequence alignment, 2) collapsing the alignment to infer multi-locus haplotypes using SNAP Map [32] with the options of recoding indels (insertions/deletions) as binary characters and excluding infinite sites violations, and 3) generating an LD plot for all variable positions using the Clade and Matrix [33] programs implemented in SNAP Workbench [34]. LD was quantified using the coefficient of determination (r2) between the allelic states at pairs of sites and a two-sided Fisher's Exact test, as implemented in Tassel version 1.1.0 [35]. LD blocks were based on the number of contiguous pairs of sites that were both strongly correlated (0.8<r2<1) and significantly linked (P<0.01). Because highly divergent haplotypes sampled once or at a low frequency could be potential targets of balancing selection in aflatoxin gene clusters [14], [24], they were not excluded in the LD analyses and the strength of LD was assessed using both r2 and 2×2 contingency tests. All sequences have been deposited in GenBank under Accession numbers HM353147–HM355445 and HM745560–HM745901.
For each population, we estimated the minimum number of recombination events (Rh) using the RecMin program [36] and the population recombination rate per base pair using Hey and Wakeley's γ estimator, implemented in SITES version 1.1 [37]. Because cluster sequences may comprise a heterogeneous mix of highly divergent alleles [14], [24], we used the composite likelihood method and the programs convert, lkgen, interval and stat in the LDhat Version 2.2 package [38] to calculate population mean recombination rates in the aflatoxin clusters of A. flavus L and S, A. parasiticus and A. minisclerotigenes. The convert program was used for calculating summary statistics that included the number of segregating sites (s) and the average pairwise difference between sequences (π); Watterson's θ [39]. Tajima's D [40] and Fu and Li's D* [41] were used as tests of neutrality and population size constancy. The Bayesian reversible-jump Markov chain Monte Carlo (rjMCMC) scheme implemented in interval was used to estimate population mean recombination rates under a crossing-over model [18]. Before using interval, a lookup table file was created using the lkgen program for each population sample from a pre-computed table (http://ldhat.sourceforge.net/instructions.shtml) and Watterson's estimate of theta per site. The interval parameters were 1,000,000 iterations for the rjMCMC procedure; 3,500 iterations between successive samples from the chain, as recommended in the user's manual (http://ldhat.sourceforge.net/manual.pdf); and a block penalty of 0. The stat program was used to summarize the interval output for each population in terms of the upper and lower 95% confidence interval bounds on the average recombination rate across the cluster.
We examined a total of 758 isolates that included A. flavus L and S, A. parasiticus and A. minisclerotigenes sampled from five continents. Aspergillus flavus L was found in all regions, but A. flavus S, A. parasiticus and A. minisclerotigenes were not present in all sampled regions (Tables 2, 3). Across all A. flavus L population samples, the total concentrations of B aflatoxins generally ranged from zero to approximately 200 µg/mL, with only a few outliers in Benin, the United States and Australia having concentrations greater than 200 µg/mL (Figure 2). According to the cumulative distribution function, the percentage of A. flavus L isolates having a high concentration of B aflatoxins (>100 µg/mL) was skewed among localities, with Australia harboring the most toxigenic isolates with 36% (29/80) followed by the United States with 35% (27/79), Benin with 21% (17/80), India with 9% (7/80), and Argentina with 6% (5/80) (Figure 2; Table 4; Tables S1, S2, S3, S4, S5). The percentage of isolates having a low concentration of B aflatoxins (<50 µg/mL) was 83, 70, 55, 48, and 33% for Argentina, India, Benin, the United States and Australia, respectively (Figure 2). Cumulative toxin distribution functions of A. flavus L were not significantly different between the United States and Australia samples using a Kolmogorov-Smirnov test (P = 0.2201) but the United States and Australia were each significantly different from Argentina and India (P<0.001). The cumulative toxin distribution for Argentina also was significantly different from that of India (P<0.001) and Benin (P<0.001); however, India and Benin were not significantly different from each other (P = 0.0708). The Benin toxin distribution was significantly different from that of Australia (P = 0.002) but not significantly different from the United States (P = 0.1061). Total aflatoxin midpoint concentrations from the United States and Australia were 60 and 80 µg/mL, respectively, whereas in Argentina, India and Benin midpoint concentrations were only 0, 30 and 40 µg/mL, respectively (Table 4). Approximately 60% (48/80) of the A. flavus L isolates sampled in Argentina were non-aflatoxigenic (Figure 2, Table S1), and 35 out of the 48 non-aflatoxigenic isolates (73%) belonged to lineage IB. By comparison, the India and Benin population samples contained four and two A. flavus L isolates, respectively, in lineage IB and only singletons from this lineage were found in the United States and Australia (Table 4).
In A. parasiticus populations, the frequencies of the three chemotype classes B1 dominant, G1/B1 equivalent and G1 dominant differed significantly among localities (Figure 2). The Argentina sample (n = 80) had more G1- and B1-dominant isolates (41 and 27, respectively) than G1/B1 equivalent isolates (12) (Table 5) and the distribution of G1/B1 was approximately partitioned into three chemotype classes (Figure 2). By contrast, the United States sample showed significantly more G1/B1 equivalent isolates (n = 59) than G1- and B1-dominant isolates (4 and 9, respectively) (Table 5). In Australia, the G1-dominant and G1/B1 equivalent chemotype classes (43 and 32, respectively) predominated over B1-dominant isolates (n = 4) (Table 5). The G1/B1 ratio for Australia and the United States showed a unimodal distribution (Figure 2), and cumulative toxin distribution functions for Argentina, the United States and Australia (Figure 2, Tables S6, S7, S8) were significantly different from each other using a Kolmogorov-Smirnov test (P<0.0001).
Populations sampled from Australia (A. flavus S and A. minisclerotigenes) and Benin (A. minisclerotigenes) were partitioned into their respective chemotype classes B and B+G (Table 6). The cumulative toxin frequency distribution for A. flavus S from Australia showed that approximately 6% (3/50) of the isolates had a high concentration of B aflatoxins (>100 µg/mL), which was significantly different (P<0.0001) from the 36% (29/80) of L strains from Australia with high B aflatoxins (Figure 2). The A. minisclerotigenes toxin distributions (Figure 2; Table 6) were not significantly different between Australia and Benin (P = 0.076). The toxin profiles for all A. flavus S and A. minisclerotigenes strains are found in Tables S9, S10, S11.
There was a significant disparity in the number of MAT1-1 and MAT1-2 isolates of A. flavus L in the Argentina, India and Benin populations, with MAT1-1 being the dominant mating type for both the uncorrected and MLST-corrected samples (Table 4). In the United States and Australia, MAT1-2 was more abundant than MAT1-1 in the uncorrected samples, but this difference was not significant after clone correction (Table 4). Mating-type ratios were also skewed in favor of MAT1-1 in A. parasiticus populations sampled from Argentina and the United States, whereas MAT1-2 predominated in Australia (Table 5). In Argentina, 98% (78/80) of the isolates were MAT1-1; clone correction of 63 MAT1-1 strains yielded 15 multilocus haplotypes with 10 haplotypes represented only once. The largest MAT1-1 multilocus haplotype comprised 29 strains. In contrast, clone correction of the United States and Australia samples of A. parasiticus showed that differences in MAT1-1 and MAT1-2 were not significant (P = 0.0963 and 0.4582, respectively). Similarly, mating-type ratios showed no significant deviation from 1.0 in clone-corrected samples of A. flavus S and A. minisclerotigenes from Australia and Benin (Table 6).
In A. flavus L, there was a significant association between mating types (MAT1-1 and MAT1-2) and the two aflatoxin chemotype classes (B1+B2 = 0 and B1+B2>0) in Argentina (P<0.001; Table S1) using a Fisher's exact test; however, there was no significant association between mating type and chemotype classes in India (P = 1.0; Table S2), Benin (P = 0.4327; Table S3), United States (P = 0.4725; Table S4), and Australia (P = 0.4246; Table S5). In A. parasiticus, there were not enough data to observe an association between mating type and the three chemotype classes (G1/B1≤0.5, 0.5<G1/B1<2.0, G1/B1≥2) in Argentina (Table S6) and there was no relationship between mating type and chemotype in the United States (Table S7) and Australia (Table S8). By comparison, in A. flavus L populations (Tables S1, S2, S3, S4, S5), the two aflatoxin chemotype classes (B1+B2 = 0 and B1+B2>0) were significantly (P<0.0001) associated with the relative proportion of reproduction (asexual>sexual and sexual>asexual). Similarly, in A. parasiticus populations (Table S6, S7, S8), the three aflatoxin chemotype classes (G1/B1≤0.5, 0.5<G1/B1<2.0, G1/B1≥2) were significantly (P<0.0001) associated with the relative proportion of reproduction (asexual>sexual and sexual>asexual; Table 5); there was also a significant (P<0.0001) association of the latter with two broad chemotype classes (G1/B1≈1 and G1/B1≠1).
Sympatric populations of A. flavus L and S, A. parasiticus and A. minisclerotigenes were sampled only from Australia and Argentina (Table 3). In A. flavus L, patterns of LD in the aflatoxin gene cluster were conserved across all populations but there were differences in the size of LD blocks and recombination parameters (Figure 3; Table 7). While the six distinct blocks observed in the United States can also be discerned in Australia, Argentina and India, blocks 4, 5 and 6 were merged into a single LD block in Benin (Figure 3). The Benin A. flavus L population with three distinct blocks showed the most extensive LD in the cluster (Figure 3), also evidenced by the lowest population mean recombination rate (2Ner; ρ = 0.0006), the lowest recombination rate per base pair (γ = 0.0002) and smallest minimum number of inferred recombination events (Rh = 1) (Table 7). The minimum number of recombination events and rates were similar in the other two predominantly clonal A. flavus L populations in India (ρ = 0.0069, γ = 0.0016, Rh = 5) and Argentina (ρ = 0.0026, γ = 0.0024, Rh = 7). The predominantly sexual A. flavus L populations in the United States and Australia harbored an almost identical LD block structure (Figure 3), and isolates from both locations were similar in their aflatoxin concentrations (Table 4), recombination rate estimates (γ = 0.0011 and 0.0010, respectively) and minimum number of recombination events (Rh = 6 and 5, respectively) (Table 7). The positive and non-significant values of Tajima's D and Fu & Li D* tests indicated the presence of divergent alleles and balancing selection on aflatoxin production and non-production in A. flavus L aflatoxin clusters (Figure S1) [14]. Estimates of π and θ were very similar across all A. flavus L populations, which indicate no significant underlying differences in mutation rates and population genetic structure.
In A. parasiticus, the five LD blocks identified in the United States were not as distinct in Australia and only blocks 4 and 5 were detected; the largest LD block in the United States (block 2) was further split into two blocks in Australia (Figure 3). The population mean recombination rate in the aflatoxin cluster was six-fold higher in Australia than in the United States (ρ = 0.0285 and 0.0049, respectively) and a similar trend was observed in overall estimates of recombination rate per base pair (γ = 0.0099 and 0.0016, respectively) and minimum number of recombination events (Rh = 8 and 4, respectively) (Table 7). No recombination was detected in the Argentina A. parasiticus population (Figure 3, Table 7). In all cases, populations of A. parasiticus with higher recombination rates had more segregating sites in the cluster (Table 7). The negative values of Tajima's D and Fu & Li D* indicated a reduction of genetic variation across the entire cluster (Figure S2). This was most pronounced in the A. parasiticus population from Argentina (π = 0.0013, θ = 0.0036), which is highly clonal based on mating-type distributions (Table 5). Population parameter estimates and neutrality tests for A. flavus S in Australia (π = 0.0120, θ = 0.0157, s = 121) were very similar to those for sympatric A. parasiticus (π = 0.0106, θ = 0.0109, s = 126) (Table 7). By contrast, A. minisclerotigenes cluster population parameters in Benin (π = 0.0554, θ = 0.052, s = 397) were approximately double those of sympatric A. flavus L (π = 0.0365, θ = 0.0283, s = 222), with a population mean recombination rate (ρ = 0.0108) in A. minisclerotigenes that was several orders of magnitude larger than that of sympatric A. flavus L (ρ = 0.0006) and with resolution of only a single LD block comprising A. flavus L blocks 4, 5 and 6 (Figures 3 and S3).
In heterothallic and hermaphroditic fungal species, mating type segregates as a single Mendelian locus such that a 1∶1 ratio is expected in a sexually reproducing population [27]. The results from this study indicate that the proportion of clone-corrected MAT1-1 and MAT1-2 in populations of A. flavus L and A. parasiticus is a useful indicator and predictor of whether populations are more clonal or sexual in reproduction. Moreover, the reproductive nature of the population (more sexual versus more asexual) is predictive of aflatoxin chemotype, in that predominantly asexual populations show a larger proportion of non-aflatoxigenic A. flavus L and an excess of G1- and B1-dominant A. parasiticus clones. There were too few data points (one per field per species) to directly test whether mating type frequency correlates with aflatoxin chemotypes; however, we were able to test the relationship between the relative proportion of sexual versus asexual reproduction and chemotype diversity. Overall, sexuality generates novel toxin chemotypes but tends to equalize toxin differences in populations. Sexual populations of A. flavus, A. parasiticus and A. minisclerotigenes from fields in different continents showed less variability in aflatoxin profiles due to genetic intermixing, whereas asexual populations exhibited greater variability in aflatoxin profiles due to increased fixation of specific toxin chemotypes.
In A. flavus L, a significant skew in the mating-type ratio was associated with higher recombination rates in the aflatoxin gene cluster and less pronounced chemotype differences. Predominantly asexual A. flavus L populations had lower mean recombination rates in the aflatoxin gene cluster, a larger proportion of non-aflatoxigenic clones and larger LD blocks. Although the size of LD blocks varied in asexual populations, block boundaries were conserved among different localities, suggesting a nonrandom distribution of recombination hotspots, as reported in other fungi [42]; infrequent recombination would initially give rise to larger LD blocks and as recombination rates increase there would be a gradual erosion of LD and more blocks that coincide with recombination hotspots. For example, overall estimates of population mean recombination rates in A. flavus L were 12-fold (0.0069/0.0006) larger in India and 4-fold (0.0026/0.0006) larger in Argentina than in Benin, which had only three LD blocks spanning the same physical distance (Figure 3; Table 7). Although A. flavus L is predominantly clonal in India, Argentina and Benin (Table 4), the ratio of asexual∶sexual reproduction is highest in Benin. By contrast, mean recombination rates in predominantly sexual A. flavus L populations (United States, Australia) were on average 23-fold (0.07/0.003) larger than in asexual populations (Argentina, India, Benin). Low recombination rates were also associated with distinct aflatoxin chemotype classes that included a relatively high frequency of non-aflatoxigenic clones (Figure 2). Approximately 60% (48/80) of the A. flavus L strains in Argentina were non-aflatoxigenic, followed by 26% (21/80) in Benin, 18% (14/80) in India, 15% (12/79) in the United States, and 14% (11/80) in Australia. Overall, A. flavus L populations with a mating type ratio closer to 1∶1 had higher population mean recombination rates, which translated into more recombination between non-aflatoxigenic and predominantly aflatoxigenic strains, thereby equalizing chemotype differences, as observed in laboratory crosses [18].
In Argentina, a broad sampling of A. flavus L from peanut seeds and soil revealed approximately 49% were non-aflatoxigenic with 13% harboring deletions of aflatoxin cluster genes (S. N. Chulze, personal communication), which suggests that lineage IB may be more prevalent than lineage IC. In this case, clonal proliferation as a result of directional selection on non-aflatoxigenicity may preserve lineage IB whereas sex between lineages IB and IC will increase the proportion of new genotypes that are aflatoxigenic, as demonstrated in A. flavus L populations derived from experimental matings [18]. Similarly, the lower recombination rate of A. flavus L in Benin may not necessarily be the result of lower recombination rates per se, but instead a paucity of sexually fertile lineage IB strains that would allow us to track recombination events when they occur. As seen in Tables 4 and 7, when the number of A. flavus L isolates in lineage IB increases from two in Benin to 35 in Argentina (n = 80), there is a four-fold increase in the rate of recombination (ρ) and a seven-fold increase in the minimum number of recombination events (Rh) in the cluster. Despite differences in population mean recombination rates, nucleotide diversity (π) and population mutation rate parameter (θ) were similar in magnitude, which suggests that divergent IB and IC alleles exist in all populations, but limited recombination results in extensive LD in the aflatoxin cluster (Figs. 3, 1S). For example, even though A. flavus L in Argentina and India showed an LD block structure similar to that observed in the United States and Australia, contingency testing revealed stronger LD in Argentina (see upper diagonal matrix in Figure 3) than in India. This suggests mating type ratio alone is not a good predictor of LD patterns in the aflatoxin cluster. In the absence of sex, non-aflatoxigenic strains may have an advantage over aflatoxigenic strains during vegetative growth or clonal populations in more temperate latitudes may be disproportionate for lineage IB isolates and therefore favor non-aflatoxigenicity. There may also be an ecological cost to aflatoxin production in certain environments depending on the level of competition or stress, such that alleviating competition favors non-aflatoxigenicity.
In A. parasiticus, a significant skew in the mating-type ratio was also correlated with both qualitative and quantitative differences in aflatoxin production that included a relatively high frequency of isolates in B1-dominant and G1-dominant classes. For example, A. parasiticus in Argentina was predominantly clonal based on mating-type frequencies; moreover, there was no detectable recombination in the aflatoxin cluster and the G1/B1 toxin distribution showed an excess of G1- and B1-dominant isolates (Figure 2), possibly the result of disruptive selection for B1- and G1-dominant traits. The lack of recombination in the A. parasiticus population from Argentina may have driven the fixation of both B1- and G1-dominant chemotypes. Alternatively, there may have been a recent selective sweep of the MAT1-1 mating type acting on B1 and G1 dominant chemotypes. In contrast, the predominantly sexual A. parasiticus populations in the United States and Australia showed higher recombination rates, distinct LD blocks in the cluster and a greater proportion of the equivalent chemotype class (0.5<G1/B1<2.0). The equivalent G1/B1 ratios in sexual populations suggest mating between parents that are high and low producers, resulting in progeny strains with intermediate toxicities of parental strains, as observed in experimental crosses [26]. Moreover, strains of A. parasiticus accumulating O-methylsterigmatocystin (OMST) were only found in sexual populations, suggesting that another outcome of sex in A. parasiticus may be to increase chemotype diversity. Because OMST accumulation results from the substitution of a single amino acid residue in aflQ [10], which is immediately adjacent to block 5 in A. parasiticus (Figure 1), it is plausible that more sexual reproduction will increase the probability of transferring this mutation to other strains via crossing over in the aflatoxin cluster. Alternatively, there may have been trans-species evolution as previously reported [24] such that A. flavus L and A. parasiticus OMST-accumulating and G1-dominant strains share a recent common ancestor, which may also be indicative of hybridization. In A. flavus L and A. parasiticus, fertile crosses comprise parents belonging to different VCGs [15], [17] and it is possible that inter-specific barriers to hyphal fusion may also be suppressed during inter-specific mating. This supports an earlier observation that A. flavus and A. parasiticus show a high degree of genome similarity that is comparable to other inter-fertile species [43] and points to the possibility of hybridization in nature, which has been shown to be experimentally feasible [44]. Because A. minisclerotigenes strains are more similar to A. parasiticus than A. flavus L in terms of B and G aflatoxin production and the existence of G1-dominant strains, we hypothesize that A. minisclerotigenes and A. parasiticus aflatoxin clusters are under similar evolutionary constraints; for example, both have an intact aflF/aflU intergenic region necessary for G aflatoxin production [45]. In this paper chemotypes are phenotypic groupings. It is possible that B+G toxin groups may be associated with genetic differences in the aflatoxin cluster that do not necessarily include the specific genes (e.g., aflU) directly responsible for mycotoxin profiles.
A skew in the mating-type ratio may be indicative of other processes such as genetic drift due to female sterility that can shift populations toward clonality; if the frequency of sex in populations is low, then the signature of clonality should be detectable. For the sympatric A. parasiticus and A. flavus populations in the United States, the uncorrected mating-type distributions are significantly skewed in opposite directions such that A. parasiticus has a higher frequency of MAT1-1 and A. flavus has a higher frequency of MAT1-2, although these differences are not significant after clone correction. This differential skew in the uncorrected samples in the United States may be driven by species-specific differences in fertility such that a greater proportion of the fertile females are MAT1-2 in A. flavus and MAT1-1 in A. parasiticus, but this cannot be ascertained without further mating studies. Alternatively, a higher frequency of one mating type may be the result of increased fitness on a function other than mating. The mating-type genes MAT1-1 and MAT1-2 encode putative transcription factors regulating pheromone and pheromone receptor genes as well as other genes not involved directly in the mating process [27]. The dominance of MAT1-2 in A. flavus L sexual populations in the United States and Australia suggest that populations can have an overriding clonal component despite undergoing sex [46]. There was also evidence of sex in clonal populations of A. flavus L from Argentina, India and Benin. Clonal populations of A. flavus L overall were predominantly MAT1-1 even though these fungi were sampled from diverse soil ecologies and exposed to different environmental conditions (Table 1). Sampling more fields in different geographical regions will be necessary to fully understand the role of different ecological and environmental factors on aflatoxin production.
Understanding the underlying genetic processes that generate diversity in A. flavus and A. parasiticus populations has direct implications in biological control in which competitive non-aflatoxigenic strains of A. flavus are applied to crops to reduce aflatoxin contamination [47]. Our observation that aflatoxin chemotype diversity in a population is associated with the reproductive nature of the population (more sexual versus more asexual) can be useful in fine-tuning biocontrol to the underlying population dynamics of a specific field. We expect that more sexual populations will exhibit higher mean rates of recombination in the aflatoxin cluster and display a more unimodal distribution of toxin concentrations. For example, Argentina is a mostly clonal population for both A. flavus and A. parasiticus, and MAT1-1 greatly outnumbers MAT1-2 even after clone correction. An indigenous non-aflatoxigenic isolate that is MAT1-1 might be recommended as a biocontrol agent in such a field, since the potential to recombine with indigenous MAT1-2 toxin producers is relatively low; however, the degree of fertility of the introduced strain may also be an important consideration and in this case, the number of distinct VCGs in the field and their fertility as deduced from laboratory crosses, may be more informative for biocontrol. In contrast, the frequency of MAT1-1 and MAT1-2 isolates for A. flavus and A. parasiticus in the Australia field was approximately 1∶1 even after clone correction. Under such circumstances, the potential of a biocontrol strain for recombining with a toxin producer is greater and approaches that focus on other biological traits, such as female sterility, may be more effective.
|
10.1371/journal.pntd.0003631 | Discovery of Novel Rhabdoviruses in the Blood of Healthy Individuals from West Africa | Next-generation sequencing (NGS) has the potential to transform the discovery of viruses causing unexplained acute febrile illness (UAFI) because it does not depend on culturing the pathogen or a priori knowledge of the pathogen’s nucleic acid sequence. More generally, it has the potential to elucidate the complete human virome, including viruses that cause no overt symptoms of disease, but may have unrecognized immunological or developmental consequences. We have used NGS to identify RNA viruses in the blood of 195 patients with UAFI and compared them with those found in 328 apparently healthy (i.e., no overt signs of illness) control individuals, all from communities in southeastern Nigeria. Among UAFI patients, we identified the presence of nucleic acids from several well-characterized pathogenic viruses, such as HIV-1, hepatitis, and Lassa virus. In our cohort of healthy individuals, however, we detected the nucleic acids of two novel rhabdoviruses. These viruses, which we call Ekpoma virus-1 (EKV-1) and Ekpoma virus-2 (EKV-2), are highly divergent, with little identity to each other or other known viruses. The most closely related rhabdoviruses are members of the genus Tibrovirus and Bas-Congo virus (BASV), which was recently identified in an individual with symptoms resembling hemorrhagic fever. Furthermore, by conducting a serosurvey of our study cohort, we find evidence for remarkably high exposure rates to the identified rhabdoviruses. The recent discoveries of novel rhabdoviruses by multiple research groups suggest that human infection with rhabdoviruses might be common. While the prevalence and clinical significance of these viruses are currently unknown, these viruses could have previously unrecognized impacts on human health; further research to understand the immunological and developmental impact of these viruses should be explored. More generally, the identification of similar novel viruses in individuals with and without overt symptoms of disease highlights the need for a broader understanding of the human virome as efforts for viral detection and discovery advance.
| Next-generation sequencing, a high-throughput method for sequencing DNA and RNA, has the potential to transform virus discovery because it does not depend on culturing the pathogen or a priori knowledge of the pathogen’s nucleic acid sequence. We used next-generation sequencing to identify RNA viruses present in the blood of patients with unexplained fever, as well as apparently healthy individuals in a peri-urban community in Nigeria. We found several well-characterized viruses in the blood of the febrile patients, including HIV-1, hepatitis B and C, as well as Lassa virus. We also discovered two novel rhabdoviruses in the blood of two apparently healthy (afebrile) females, which we named Ekpoma virus-1 and Ekpoma virus-2. Rhabdoviruses are distributed globally and include several human pathogens from the genera lyssavirus and vesiculovirus (e.g., rabies, Chandipura and vesicular stomatitis virus). The novel rhabdoviruses identified in this study are most similar to Bas-Congo virus, which was recently identified in an individual with an acute febrile illness. Furthermore, we demonstrate evidence of high levels of previous exposure to the two rhabdoviruses among our larger study population. Our results suggest that such rhabdovirus infections could be common, and may not necessarily cause overt disease. The identification of viral nucleic acid sequences in apparently healthy individuals highlights the need for a broader understanding of all viruses infecting humans as we increase efforts to identify viruses causing human disease.
| Viral discovery is rapidly advancing, driven by the advent of high-throughput technologies like next-generation sequencing (NGS) [1]. Applying NGS as a diagnostic tool holds the promise for vastly expanding our understanding of the spectrum of microbes infecting humans, as it does not require a priori knowledge of the pathogens present. It also has the potential to elucidate the spectrum of disease-causing viruses in patients with undiagnosed acute febrile illness (UAFI), a common occurrence in health clinics around the world [2]. NGS can also serve to increase the power of surveillance systems to detect infrequent zoonotic transmissions that have the potential to become pandemics [3].
NGS has already been used successfully as both a diagnostic tool and a means to discover novel viruses associated with human disease [4–8]. Examples of these discoveries include novel arenaviruses [5], phleboviruses [4], and coronaviruses [8]. Recently a novel rhabdovirus, now referred to as Bas-Congo virus (BASV), was identified in the blood of a patient from central Africa who was suspected of suffering from viral hemorrhagic fever [9].
However, a better understanding of the spectrum of viruses infecting humans is needed to fully realize the potential of NGS and differentiate between pathogenic and non-pathogenic viruses. This global problem is particularly acute in tropical regions throughout the world, where the burden of infectious disease remains high and the bloodstream virome of large numbers of apparently healthy individuals has not been characterized. Most studies of UAFI lack comparisons with apparently healthy individuals and rely on small-scale associations (in some cases even a single patient sample) without any statistical support or the ability to determine causality [7,9].
In this study we use high-throughput NGS to elucidate the spectrum of RNA viruses present in the blood of patients with UAFI in a population from southeastern Nigeria, using apparently healthy members of the same community for comparison. While we detected only known and common viral nucleic acid sequences in the UAFI patients, we were able to assemble full-length genomes of two novel, highly divergent rhabdoviruses from two apparently healthy individuals. We found that these viruses were similar to BASV and to viruses of the genus Tibrovirus. By conducting a serosurvey of our study cohort, we found that exposure to these novel viruses was unexpectedly high. Our findings suggest that human infection with certain types of rhabdoviruses may be common, and highlight the need for a broader understanding of the human virome as the use of NGS for microbial discovery advances.
Our study population consisted of men and women from all age groups and socioeconomic backgrounds living in and around Irrua, a modest-sized peri-urban village in southeastern Nigeria (for further descriptions of the study population see S1 Table). As part of a partnership with the Irrua Specialist Teaching Hospital (ISTH) to study Lassa fever, we collected blood samples from suspected Lassa fever patients that tested negative for Lassa virus by reverse transcription PCR (RT-PCR) and subjected them to NGS (S1 Table). We hypothesized that UAFI patients with symptoms resembling viral hemorrhagic fever could be infected with other pathogens that cause severe illness. We additionally collected samples from apparently healthy individuals (i.e., individuals whose temperature was in the normal range and did not have any overt symptoms of illness) from the surrounding populations as part of the 1000 Genomes Project, and as part of a control population for our studies of Lassa fever.
We performed collections of febrile cases and apparently healthy controls under approved IRB protocols in Nigeria (Oyo State Ministry of Health, ISTH) and the US (Tulane University, Harvard University, Harvard School of Pubic Health, and the Broad Institute). All adult subjects provided informed consent, and a parent or guardian of any child participant (aged under 18 years) provided informed consent on their behalf. All children 7 and older additionally provided assent. Individuals provided written informed consent. If an individual was unable to read, a study staff read the document to the participant or guardian. The individual then provided a thumbprint, and the consent form was cosigned by the study staff as well as a witness. The use of thumbprints was specifically approved by the IRB granting institutions.
We collected approximately 5–10 mL of venous blood in EDTA vacutainer tubes, centrifuged them to obtain the plasma from cellular fractions, and inactivated the plasma by adding buffer AVL (Qiagen). We added carrier RNA to some of the samples as indicated in S2 Table. In the case of the apparently healthy controls, we collected an additional aliquot of ‘unadulterated’ plasma that was not inactivated with buffer AVL.
We constructed RNA-seq libraries as previously described [10]. We prepared some of the libraries from extracted RNA for either single individuals (referred to as singletons) or from RNA pooled from several individuals (referred to as pools) (S2 Table). We treated all samples with DNase. We primed RNA using random hexamers, or modified hexamers (5’-NNNNNNV-3’ from Integrated DNA Technologies) if carrier RNA was present in the sample. We amplified the resulting libraries by PCR, pooled, and sequenced on an Illumina HiSeq 2500 according to the manufacturer’s specifications. Primers used for Sanger sequencing are listed in S3 Table. The raw data has been deposited to SRA under BioProject ID PRJNA271229.
We processed individual afebrile controls as described for UAFI samples; however, the method of pooling differed. We pooled and filtered unadulterated plasma (without AVL) samples and centrifuged them at 104,000 x g for 2 hours at 4°C. We resuspended the viral pellet in buffer and used it to construct libraries for sequencing. AVL denatures viral particles, thus preventing centrifugation of the particles. We have observed comparable results between samples inactivated by AVL and those that are not.
We trimmed raw Illumina sequences consisting of 100 bp paired-end reads to remove bases from the ends of the reads with low quality scores, and discarded all reads shorter than 70 bp after quality trimming. We removed human and other contaminating reads using BMTagger (NCBI), and removed duplicate reads and low complexity reads using PRINSEQ [11]. We assembled reads de novo using MetaVelvet [12] followed by Trinity [13]. We used contigs of at least 200 bp for BLASTn or BLASTx queries of the GenBank nucleotide (NT) or protein (NR) databases (E-score cutoffs of 10-6 and 102, respectively). In a parallel pipeline, we used individual reads for BLASTn or BLASTx queries of GenBank with the same E-score cutoff values. We performed taxonomic classification of assembled contigs and individual reads and visualized them using MEGAN 4 [14]. We considered samples to have a virus present if MEGAN 4 ‘min support’ was ≥5 and ‘min score’ was ≥50. We assessed statistical significant differences in the distributions of viruses between UAFI samples and apparently healthy individuals using a two-tailed Fisher’s exact test with α<0.05 considered significant.
We used quantitative real-time PCR to measure the number of Ekpoma viral RNA copies per milliliter of blood using the RNA-to-CT 1-Step Kit (Applied Biosystems). The primers, which amplify an ~100bp region in the polymerase (L) gene, have the following sequences:: EKV-1: 5’-AAGAGTTGTTGGGATGGTCAGA-3’ (forward) and 5’- TGATTCTTGCTTCTCGCTCGAT-3’ (reverse); and EKV-2 primers: 5’-TGGCCAATTCCTTGGCTATCCCCT-3’ (forward) and 5’-TCCCGCCGGAGACATACATCTT-3’ (reverse). We amplified PCR reactions on the ABI 7900 sequence detection system using the following cycling parameters: 30 minutes at 48°C, 10 minutes at 95°C, and 40 cycles of 15 seconds at 95°C and 1 minute at 60°C. A serial dilution of a synthetic DNA amplicon, which corresponds to the amplified region of the polymerase gene, was used to quantify the number of viral cDNA copies in the reaction. Human K562 RNA and RNA purified from the plasma of an afebrile individual (244M), were used as negative controls.
We performed multiple sequence alignments of rhabdovirus nucleoprotein (N), glycoprotein (G), matrix (M), phospoprotein (P) and polymerase (L) amino acid sequences using MAFFT v6.902b18 [15] with the following parameters:—localpair—maxiterate 1000—reorder—ep 0.123 before being trimmed using trimAl v1.419 [16] with the maximum likelihood specific parameter:-automated1. We used PROTTEST [17] to identify rtREV+I+G [18] as the best evolutionary model and made maximum likelihood phylogenies with RAxML v7.3.0 [19]. Trees were bootstrapped using 500 pseudo-replicates. We also created trees using MrBayes v3.2 [20]. We first built trees using 46 rhabdovirus sequences and included parainfluenza virus-1 as an outgroup, to find the novirhabdoviruses as the likely root of the rhabdovirus tree, which has been previously described [21]. We then excluded parainfluenza virus-1 and built a tree using the 46 rhabdovirus sequences (S6A Fig), which allowed us to select VSV as a likely outgroup for the tibroviruses and ephemeroviruses. Subsequent alignments and trees were then created using only the tibroviruses and ephemeroviruses, including EKV-1, EKV-2, and BASV, as well as VSV. We found that using parainfluenza virus-1 or the novirhabdoviruses as the root, gave the same tree topology. Relevant accession numbers can be found in S4 Table.
We cloned His-tagged N genes from EKV-1 and EKV-2 into pET45B(+) and expressed them in E. coli. We lysed the cells in the presence of protease inhibitors and purified the proteins with HisPur Ni-NTA Spin Columns (Thermo Scientific). We confirmed the purity of the proteins by Western Blot. We created ELISA plates by coating the EKV-1 and EKV-2 N proteins onto 96-well plates at 2μg/mL in carbonate-bicarbonate buffer overnight at 4°C. Human IgG specific to EKV-1 or EKV-2 was detected by ELISA as previously described [22]. We calculated cut-off values based on the mean of the US controls (N = 137) plus three or five standard deviations.
We selected blood samples from 195 UAFI and 328 afebrile controls for RNA sequencing by Illumina NGS (S1 Fig). We collected a number of demographic and clinical parameters (S1 Table) for each individual in our study.
We successfully constructed 120 RNA-seq libraries from UAFI samples (94 singletons and 26 pools) comprising a total of 195 individuals, and 58 RNA-seq libraries from afebrile apparently healthy control samples (34 singletons and 24 pools) comprising a total of 328 individuals (S5 Table). Illumina sequencing generated a total of 3.71 billion 100 base pair (bp) paired-end reads. We analyzed these samples using a bioinformatics and computational pipeline developed in our laboratory (S2A Fig). After filtering out low-quality sequences, duplicates, human reads and common contaminants, less than 0.5% of the reads typically remained in each library (S2B–D Fig).
We examined the overall composition of reads identified in 94 singleton UAFI samples and in 34 apparently healthy singleton controls (Fig. 1). We found ~25% of the filtered reads returned no BLAST hit or were unable to be unequivocally assigned to the eukarotya, prokaryota or viral kingdoms. The majority of filtered reads in both UAFI and afebrile libraries were bacterial.
The overall percentage of viral reads was similar between UAFI patients and afebrile controls (3.3% and 2.4%, respectively). The majority of viral reads were derived from three sources: human adenovirus C, phages, or GB virus C (S6 Table and S1 Text). GB virus C, a non-pathogenic RNA virus [23], was identified in 18% of UAFI singleton libraries and 12% of singleton healthy controls (Fig. 1B and S3 Fig); a higher percentage of pooled healthy controls contained GB virus C, possibly because each pool contained a greater number of individual samples compared to the UAFI pools. We identified several well-characterized pathogenic RNA viruses, including LASV, HIV-1, hepatitis C and dengue virus in the UAFI patients (Fig. 1B and S6 Table). We did not find any evidence for the presence of Ebola virus. LASV was the most frequent pathogenic virus observed in UAFI cases and the only virus statistically enriched in the UAFI as compared to the apparently healthy controls (P-value = 0.002, Fisher’s test; S3 Fig). Although samples were DNAse treated, we also detected several DNA viruses, including hepatitis B virus, herpesvirus 4 (Epstein-Barr virus), herpesvirus 5 (human cytomegalovirus), and herpesvirus 8 (Kaposi’s sarcoma virus) (Fig. 1B and S6 Table).
In two pools of RNA from afebrile individuals, we identified reads with distant relationships to previously identified rhabdoviruses. A PCR assay developed to identify the infected individual within each pool revealed two infected females aged 45 (sample 13M) and 19 (sample 49C). We named the two viruses Ekpoma virus-1 (EKV-1; from 13M) and Ekpoma virus-2 (EKV-2; from 49C) because both individuals lived in Ekpoma, a village located about ten kilometers from ISTH.
We assembled several long contiguous overlapping rhabdovirus sequences (contigs) (Fig. 2A). From these contigs we synthesized virus-specific primers for EKV-1 and EKV-2 and used Sanger sequencing to confirm the results of Illumina sequencing and fill in missing parts of the genomes (Fig. 2B). The combined sequencing produced two genomes of 12,659 bp (EKV-1) and 12,674 bp (EKV-2) (GenBank accession numbers KP324827 and KP324828).
The coverage of EKV-1 ranged from 1–71x (median 9x) and the coverage of EKV-2 ranged from 1–29x (median 8x; Fig. 2C). We did not find any additional samples that contained reads from these two novel rhabdoviruses.
The Rhabdoviridae family includes at least eleven genera [24]. We found that the genomic organization of EKV-1 and EKV-2, like BASV, is the same as members of the genus Tibrovirus (S4 Fig). The viral genomes consist of the prototypical five open reading frames (ORFs) found in most rhabdoviruses (N, P, M, G, and L) as well as at least three additional ORFs of unknown function (U1 to U3) [25] (Fig. 2B). The latter three ORFs are also seen in other members of the genus Tibrovirus and their presence clearly distinguishes these viruses from the closely related genus Ephemerovirus.
We found that the sequence identity among the Ekpoma viruses was low, ranging from 33.2–39.4% for the different ORFs at the protein level (S4 Fig). The nucleoprotein and polymerase were the most highly conserved proteins (S5 Fig), while U1–U3 were the most divergent. Overall, EKV-2 was more similar at the amino acid level to BASV (39.4% identity) than it was to EKV-1 (35.1% identity).
To determine the place of the Ekpoma viruses within the rhabdovirus phylogeny we constructed maximum likelihood and Bayesian trees for the major viral proteins. We found that EKV-1 and EKV-2 clustered with BASV, TIBV, and Coastal Plains virus (Figs. 3A and S6). We further found that EKV-1 is a closer evolutionary relative to TIBV than to EKV-2 or BASV. EKV-2, in contrast, formed another branch with BASV (Fig. 3A, B). Though these viruses were discovered in geographically distant locations, phylogenetic analyses suggest the presence of a distinct group of viruses in the Tibrovirus genus capable of human infection. Based on phylogenetic relationships, host range and genomic architecture, we propose that BASV, EKV-1 and EKV-2 should all be included within the genus Tibrovirus.
To assess the level of human exposure to the novel rhabdoviruses, we developed enzyme-linked immunosorbent assays (ELISAs) to detect antibodies that recognized the N proteins of EKV-1 and EKV-2. We performed a serosurvey for EKV-1 and EKV-2 on 457 samples and found that significantly more Nigerian individuals (n = 320) had EKV-1- and EKV-2-specific antibodies than apparently healthy US controls (n = 137; Fig. 3C; P-value < 0.0001, Mann-Whitney test). Using conservative positivity cut-off values, we found that ~10% of Nigerian individuals show evidence of previous exposure to EKV-1 (Table 1 and Fig. 3C). The seropositivity to EKV-2 was much higher, with ~50% of Nigerians showing evidence of previous exposure (Table 1 and Fig. 3C). We did not observe any significant difference in the sex or age-range of the individuals with antibody titers to EKV-1 or EKV-2 (S7 Fig). We cannot rule out that our assays do not cross-react with other similar rhabdoviruses, which could inflate the overall seroprevalence observed for the Ekpoma viruses; however, it should be noted that limited cross-reactivity was observed between EKV-1 and EKV-2 (S8A Fig). While we found strong cross-reactivity between our assays for EKV-1 and rabies virus (S8B Fig), the correlation between EKV-2 and rabies virus was much less pronounced (S8C Fig). Importantly, when testing general cross-reactivity in our assays by comparing the ELISA results from the rhabdoviruses to that of LASV, we did not find any correlations (S8D–F Fig).
Acute infection with RNA viruses often produces high viral loads. To assess the level of viremia, we used quantitative real-time PCR to measure EKV-1 and EKV-2 viral copy number. We detected 4.5 million viral genome copies per milliliter of plasma in the individual infected with EKV-1 and 46,000 viral genome copies per milliliter of plasma in the individual infected with EKV-2 (S9 Fig). These numbers, while informative, should be interpreted with caution, as sample degradation may have affected the number of viral copies detected.
After the discovery of the two Ekpoma viruses, we sought to further determine the health of the infected individuals. Nearly two years after their initial blood draw, we conducted oral interviews with both individuals and collected convalescent serum samples. Both individuals tested negative for the two Ekpoma viruses by PCR upon testing of their convalescent samples (S10 Fig); however, using our ELISA assays, we found that they both had antibodies reacting with EKV-1 or EKV-2, as expected (S11 Fig). Notably, while both individuals had antibody titers at the time of infection and in the follow-up samples, the woman infected with EKV-2 showed lower titer in her follow-up sample, as compared to the original blood draw (S11B Fig).
The woman infected with EKV-1 could not recall any episode of febrile illness in the weeks or months following the collection of her initial blood sample. The woman infected with EKV-2 revealed that she suffered an episode of febrile illness two weeks after we collected her blood sample. She was admitted to the hospital where her illness was clinically diagnosed as malaria. While the individual’s illness resolved after anti-malarial treatment, we cannot confirm whether a malaria parasite was the causal agent.
We attempted to isolate EKV-1 and EKV-2 by using plasma from the infected individuals to inoculate cultures of Vero E6, BHK, C6/36 mosquito, LLC-MK2, SW13 and biting midge (Culicoides variipennis) cell lines. We did not observe any evidence of viral cytopathic effects in these cultures, nor could we detect any virus by qPCR or electron microscopy. We also attempted to isolate the viruses by intracranial inoculation of newborn mice; however, we did not observe any signs of illness over 14 days. It is possible that the viruses may not be able to infect any of the tested cells or animals, however, potential sample degradation may have compromised the infectivity of viral particles.
We used high-throughput NGS to elucidate the spectrum of RNA viruses present in the blood of patients with UAFI in a population from southeastern Nigeria, using apparently healthy members of the same community for comparison. NGS has the advantage of being able to identify pathogens without culturing or a priori knowledge of the pathogen’s nucleic acid sequence.
Despite the advantages of NGS, there are certain biases in our approach. First, the selection of blood limited our investigation to a single anatomical compartment. Many viruses cannot be detected in the blood (e.g., rabies virus which is strictly neurotropic). A complete understanding of a febrile or healthy person’s virome necessitates sequencing of all tissues in the body, which for practical reasons, is not possible. The ability to identify novel viruses is also limited to sequences that have some homology existing sequences in a public database. Highly divergent and truly novel pathogens may be missed by conventional BLAST searches. In our study, ~25% of filtered reads returned no BLAST hit or were unable to be unequivocally assigned to the eukaryotya, prokaryota or viral kingdoms. Despite these limitations however, we were able to identify EKV-1 and EKV-2, both of which have only about 35% amino acid similarity to already known viruses.
In our study we made an unexpected discovery of nucleic acid sequences suggestive of novel rhabdoviruses in our apparently healthy controls. The identified viruses, EKV-1 and EKV-2, most closely resemble members of the genus Tibrovirus, and in particular BASV, based on genomic structure and phylogenic analyses. BASV was recently identified in an individual from central Africa displaying symptoms suggestive of viral hemorrhagic fever [9]. Despite detection in an apparently healthy individual, EKV-2 is the most closely related virus to BASV identified to date.
Tibroviruses, including Tibrogargan, Coastal plains and Bivens Arm viruses, are transmitted by culicoidies insects and are known to cause subclinical infections in cattle and other ruminants [26]. Their amino acid sequence similarity to Tibrogargan and Coastal Plains viruses raises the possibility that they might be vector-borne [26–29]. If true, infection could be common in environments where biting insects are ubiquitous, like central and western Africa. Many rhabdoviruses have already been discovered in sub-Saharan Africa using conventional methods—mostly in insects and vertebrates (Fig. 4). Our results suggest many more remain to be discovered, and that a number of these may infect humans.
Consistent with the potential for widespread and subclinical infection by rhabdoviruses, our serosurvey uncovered evidence for very high exposure to EKV-1 or EKV-2, with nearly 50% of our apparently healthy cohort showing evidence of EKV-2 exposure. Despite this high rate, we did not detect any EKV-1 or EKV-2 nucleic acids in the UAFI patients. These results suggest that members of the genus Tibrovirus are unlikely to be common causes of viral hemorrhagic fever as has been suggested for BASV [9].
We attempted to isolate EKV-1 and EKV-2, but were unsuccessful in our efforts. We speculate that sample handling may have caused degradation of viral particles. Alternatively, these novel viruses may not infect the common cell types we selected for culturing. Historically, isolating a virus from an infected individual is a necessary step for demonstrating the existence of the novel virus and that the patient was infected. However, as NGS becomes more common, it is likely that many new viruses will be identified that cannot easily be cultured. That does not mean these viruses cannot be studied biochemically or “recreated” in the laboratory. Parts of the virus can be synthesized de novo and incorporated into existing viral vectors. In some cases, the entire nucleic acid sequence of the virus can be synthesized de novo, introduced into cells, and potentially cultured.
The recent discovery of three related rhabdoviruses—two in apparently healthy individuals (this study) and one in an acutely ill patient [9]—highlights the challenges of determining the true cause of unexplained illness. Many factors determine whether a particular virus will produce disease in the infected host, including genetic variation in the virus and the host, nutritional and immune status, and the presence of co-infections that may increase susceptibility to otherwise benign agents. Identifying the cause of disease becomes even more challenging since multiple microbes are present in a sample, including commensal bacteria and viruses.
Proving disease causality is a centuries-old problem and identifying a potential pathogen is merely the first step in a long process. Researchers have recently proposed revisions to Koch’s postulates—the first framework for assessing causality—in light of advancing modern molecular techniques [30,31] to add rigor to the pursuit. Yet there are still a number of limitations to current studies. For many studies, investigators were only able to study a single patient sample [9]. Without sufficient numbers of samples from infected patients and matched apparently healthy individuals, it is impossible to interpret the clinical significance of a single virus detection. It remains possible that BASV produced an asymptomatic infection, like the control subjects infected with EKV-1 and -2 in our study, while the acute illness was actually due to another agent, like the rotavirus (which the authors propose was a laboratory contaminant), or one of the many bacteria also present in the sample [9]. Of course, the true source of the infection could have been none of the microbes identified in the blood. Sampling of other tissues would be needed to rule out localized infections as the cause of disease.
Regardless of whether infection with particular rhabdoviruses is symptomatic or not, the discovery of novel rhabdoviruses could be of importance to human health. Members of the Rhabdoviridae, such as lyssaviruses and vesiculoviruses, produce serious neurotropic disease in humans [32,33]. Others, such as vesicular stomatitis virus (VSV), produce subtle neurotropic infections with few acute disease symptoms. BASV, like VSV, appears to have broad tissue tropism [34] and may infect similar cell types. Further studies are needed to determine if the novel rhabdoviruses discovered in this study produce neurotropic outcomes in humans similar to those of lyssaviruses and vesiculoviruses [35–37].
How should future studies using NGS tackle the issue of disease causality in these and other newly discovered microbes? The most obvious approach involves finding a statistical association with the microbe in disease and non-disease states, similarly to what we show for LASV in this study (S3 Fig). This requires collecting matched controls from either the patient or members of the community who do not have the disease. This approach faces its own challenges. If viral or host factors play a substantial role in disease outcome, it might necessitate large sample collections. Isolation of the pathogen and propagation in an animal model or tissue culture can provide valuable insights into its pathogenicity and effect on the host’s response to infection.
The recent advent of NGS has the potential to transform the centuries-old pursuit of finding disease-causing pathogens and to elucidate the complete human virome. But in the process, it will be important to be cautious. As the vast majority of viruses studied over the past century have been those that cause disease, the large-scale sequencing of samples from vertebrates and insects will likely be biased towards identifying novel benign viruses rather than pathogenic ones. Although many newly discovered viruses may not cause overt symptoms of disease, they may have immunological and developmental consequences—perhaps by increasing susceptibility to other pathogens or affecting other aspects of human development.
Pathogen discovery tools are evolving rapidly. Investigations that harness these new tools will likely identify a plethora of new viruses in humans, animals, and insects. Developing systems to assess causality, especially through the thorough sampling of non-disease-affected controls, will be critical to realizing the potential of NGS as a routine diagnostic tool.
|
10.1371/journal.pgen.1001361 | The Exocyst Protein Sec10 Interacts with Polycystin-2 and Knockdown
Causes PKD-Phenotypes | Autosomal dominant polycystic kidney disease (ADPKD) is characterized by
formation of renal cysts that destroy the kidney. Mutations in PKD1 and PKD2,
encoding polycystins-1 and -2, cause ADPKD. Polycystins are thought to function
in primary cilia, but it is not well understood how these and other proteins are
targeted to cilia. Here, we provide the first genetic and biochemical link
between polycystins and the exocyst, a highly-conserved eight-protein membrane
trafficking complex. We show that knockdown of exocyst component Sec10 yields
cellular phenotypes associated with ADPKD, including loss of flow-generated
calcium increases, hyperproliferation, and abnormal activation of MAPK. Sec10
knockdown in zebrafish phenocopies many aspects of polycystin-2
knockdown—including curly tail up, left-right patterning defects,
glomerular expansion, and MAPK activation—suggesting that the exocyst is
required for pkd2 function in vivo. We observe
a synergistic genetic interaction between zebrafish sec10 and
pkd2 for many of these cilia-related phenotypes.
Importantly, we demonstrate a biochemical interaction between Sec10 and the
ciliary proteins polycystin-2, IFT88, and IFT20 and co-localization of the
exocyst and polycystin-2 at the primary cilium. Our work supports a model in
which the exocyst is required for the ciliary localization of polycystin-2, thus
allowing for polycystin-2 function in cellular processes.
| ADPKD, the most common potentially lethal monogenetic disorder, is caused by
mutations in PKD1 and PKD2. We are beginning to appreciate the important roles
these gene products, and others, play in cilia, which are thin rod-like
organelles projecting from the cell surface. Defects in cilia function are
associated with a variety of human diseases, including all variants of
polycystic kidney disease. Despite intense study of cilia and how they influence
disease, it is not understood how proteins are targeted and delivered to cilia.
Our work provides the first link between the exocyst, a conserved eight-protein
complex involved in protein localization, and a disease gene, PKD2. Knockdown of
the exocyst protein Sec10 results in a number of cellular- and cilia-related
phenotypes that are also seen upon pkd2 loss—both in
kidney cells and zebrafish. We then demonstrate specific genetic and biochemical
interactions between sec10 and pkd2. We
further show that Sec10 interacts with other ciliary proteins, such as IFT20 and
IFT88. From this work, we propose that the exocyst is involved in bringing
multiple types of ciliary proteins to the cilium. Given that the exocyst is
required for cilia structure and function, the exocyst may play a role in
cilia-related human diseases.
| ADPKD is the most common potentially lethal monogenetic disorder, affecting 12
million people worldwide [1]. ADPKD is characterized by the development of numerous
renal cysts, which greatly increase kidney size, perturb kidney function, and
eventually lead to kidney failure. While we know that mutations in PKD1 and PKD2
cause ADPKD [2], [3], we are only beginning to understand how the
proteins—polycystin-1 and polycystin-2—regulate the cellular phenotypes
associated with cystogenesis.
Interactions between polycystin-2, a calcium-permeable cation channel [4], [5], and
polycystin-1 may act to regulate calcium signaling in normal kidney cells [6]. Consistent
with calcium regulation being relevant to cystogenesis, ADPKD cells show a lower
basal intracellular calcium concentration [7]. Furthermore, altered calcium
regulation has been linked, through cyclic AMP (cAMP) signaling, to phenotypes
observed during cystogenesis, such as increased cell proliferation and abnormal
fluid secretion. Addition of cAMP agonists cause ADPKD cells, but not normal kidney
cells, to stimulate proliferation via the MAPK pathway [6], [8], [9].
Growing evidence suggests that the cilium is an important site of polycystin
function. Kidney tubular epithelial cells have a single non-motile primary cilium
that acts as a mechanosensor, triggering a rise in intracellular calcium in response
to fluid flow [10], [11]. Polycystins-1 and -2 localize to the primary cilium of
kidney cells [12],
[13], and the
calcium response to fluid flow requires polycystin function [14]. Consistent with the idea that
mechanosensation is relevant to cystogenesis, ADPKD cells are unresponsive to fluid
flow [15].
Research in animal models suggests that cilia play important roles, not only during
adult kidney function, but also throughout early embryonic development (reviewed in
[16]).
Zebrafish has been increasingly used as a model organism to expand our understanding
of the in vivo function of ciliary proteins through studies
utilizing mutants that affect cilia, and morpholino antisense knockdown of ciliary
proteins. Loss of intraflagellar transport proteins (reviewed in [17]), which are
required for cilia assembly, results in body axis curvatures (“curly
tails”), left-right defects, pronephric cysts, edema, and small eye phenotypes
[18]-[21]. Other mutants that show disrupted cilia length or
motility similarly show curly tails, left-right defects, and pronephric cysts [20], [22]-[26].
These phenotypes, which comprise the range of cilia-related phenotypes in zebrafish,
suggest that proper cilia formation and/or function is required for multiple
developmental processes. The mechanistic relationship connecting cilia to each
phenotype is understood to differing degrees depending on the specific phenotype.
The connection is well understood for left-right patterning and pronephric
development. Left-right patterning governs the stereotypical positioning of organs,
which is preceded and directed by left-sided expression of the Nodal signaling
pathway (reviewed in [27]). The asymmetric expression of the nodal
genes spaw, lefty1, and lefty2 in
zebrafish is itself thought to be established by cilia-dependent fluid flow in
Kupffer's vesicle [18], [28]. Indeed, mutants that show disrupted cilia length or flow
in Kupffer's vesicle subsequently show randomized nodal gene
expression and left-right defects [18], [28]. Cilia in the pronephric
tubules are similarly thought to be important for pronephric development such that
perturbations in motility result in tubule dilations and cystogenesis [25], [26].
Research into pkd2 function in zebrafish has further strengthened
the idea that polycystin-2 functions in the cilium. Knockdown of
pkd2 by morpholino [29]-[31] or in mutants [20], [31] produces
phenotypes that are consistent with a role in cilia function: curly tails,
left-right defects, pronephric cysts, and edema. Indeed, polycystin-2 is expressed
in Kupffer's vesicle, and mutations in pkd2 lead to defects in
left-right patterning in zebrafish and mice [29]-[32]. However,
pkd2 is unique in zebrafish for a number of reasons. First, it
is the only reported mutant to consistently display a curly tail up phenotype [20], [29]-[31], as
opposed to the typical curly tail down phenotype of other cilia mutants. Secondly,
pkd2 knockdown does not produce observable defects in cilia
structure [29]-[31] or motility [25], [30]. Therefore,
pkd2 is likely to be important for cilia function in a way that
is distinct from a role in cilia formation, maintenance, or motility. For example,
it has been proposed that pkd2 may play a specific mechanosensory
role related to calcium regulation during left-right patterning in mice [33].
While we are beginning to identify the roles ciliary proteins play in diverse
biological processes, there is little known about how these proteins are transported
to the cilium [34].
The exocyst, originally identified in S. cerevisiae
[35], is a highly
conserved 750kD eight-protein complex known for the targeting and docking of
vesicles carrying membrane proteins [36]. It is comprised of Sec3,
Sec5, Sec6, Sec8, Sec10, Sec15, Exo70, and Exo84 [37]. Notably, in addition to being
found near the tight junction, we localized exocyst proteins to the primary cilium
in kidney cells [38], [39]. Sec10 and Sec15 are the most vesicle-proximal of the
exocyst components. Sec10 has been shown to directly bind to Sec15, which, in turn,
directly binds Sec4, a Rab GTPase on the surface of transport vesicles. Sec10 then
acts as a “linker”, by binding the other exocyst components through Sec5
[40]. Our
previous studies suggested that the exocyst would no longer be able to bind Sec15
and target/dock transport vesicles without Sec10, and would, instead, disintegrate
and be degraded. Importantly, we showed that knockdown of exocyst Sec10 in
Madin-Darby canine kidney (MDCK) cells abrogated ciliogenesis, while Sec10
overexpression enhanced ciliogenesis. Furthermore, Sec10 knockdown caused abnormal
cystogenesis when the cells were grown in a collagen matrix, and decreased the
levels of other exocyst components and the intraflagellar transport protein 88
(IFT88). This was in contrast to knockdown of exocyst components Sec8 and Exo70,
which had no effect on ciliogenesis, cystogenesis, or levels of other exocyst
components [39].
These data uncovered a role for the exocyst, and especially the Sec10 component, in
building the primary cilium. Given its known role in trafficking proteins to the
plasma membrane [41]-[44], we have proposed that Sec10 and the exocyst may be
required in the cilium to target and dock vesicles carrying proteins important for
ciliogenesis.
Here we show that Sec10 knockdown, in vitro in MDCK cells and
in vivo in zebrafish, results in phenotypes associated with
loss of polycystins and ADPKD. We specifically demonstrate a genetic and biochemical
interaction between Sec10 and polycystin-2, as well as show co-localization at the
primary cilium, providing further evidence that the exocyst is important for
polycystin-2 function. Furthermore, we show biochemical interactions between Sec10
and the ciliogenesis proteins IFT88 and IFT20. Our results demonstrate that the
exocyst is required for pkd2 function in the cell. Together with
our previous results, these data suggest that the exocyst is important for
maintaining both cilia structure and function. Exocyst dysfunction may therefore
contribute to ciliopathies including ADPKD, and Sec10 may represent a novel target
for the development of effective treatments.
Given that loss of polycystin-2 leads to ADPKD, we first determined whether
exocyst Sec10 knockdown or overexpression in MDCK cells produced ADPKD-like
phenotypes. Primary cultures of ADPKD cells fail to show the expected rise in
calcium levels in response to a shear flow [7], [15]. After growing these MDCK
cell lines to confluent monolayers on Transwell filters, conditions that we have
previously shown results in ciliation [39], we measured steady state
levels of intracellular calcium using the Fura-2 indicator. We then tested
whether cells exposed to a constant 5 ml/minute flow rate over the apical
surface responded appropriately with an increase in calcium levels. Sec10
knockdown cells showed a significantly lower basal calcium level than both
control MDCK and Sec10-overexpressing cells, and calcium levels in the Sec10
knockdown cells also failed to increase in response to fluid flow (0.2%
increase in Sec10 knockdown versus 5.8% in control and 26.2% in
Sec10-overexpressing cells) (Figure
1A). Thus, Sec10 knockdown cells do indeed produce phenotypes similar
to that observed in ADPKD cells [7], [15]. The reason for the limited
increase in calcium in response to fluid flow in control T23 MDCK cells, that
constitutively express the tetracycline transactivator that drives the Sec10
shRNA, is likely due to the fact that ciliogenesis, for unknown reasons, is
sporadic—with the literature describing from ∼30% of T23 MDCK
cells being ciliated (as we see [39]), to as few as 14.6% being ciliated [45]. The loss
of mechanosensation in Sec10 knockdown cells is consistent with the loss of
cilia in these cells [39], [46]. Similarly, since Sec10-overexpressing cells display
longer cilia [39], the increased calcium response may reflect a heightened
mechanosensory capability of those cilia.
Cellular hyperproliferation is another major feature of ADPKD cells [47], so we
investigated whether the abnormal calcium level observed in Sec10 knockdown
cells was associated with hyperproliferation. Using a luminescence-based
viability assay, Sec10 knockdown cells showed an increased rate of proliferation
after 48 hours (Figure 1B).
Sec10-overexpressing cells, by contrast, showed a relatively normal rate of
proliferation.
The mitogen activated protein kinase (MAPK) pathway is activated during cell
proliferation and kidney development [48], [49]. It has been shown in some
mouse models of ADPKD that the MAPK pathway is activated, and that blockage of
extracellular-signal regulated kinase (ERK) slows the development of polycystic
kidney disease [50]. It has been theorized that the decreased
intracellular calcium in primary ADPKD cells, resulting from the dysfunctional
primary cilia, leads to increased cyclic AMP (cAMP) and, therefore, protein
kinase A (PKA) activity, with downstream MAPK pathway hyperactivation [51]-[53]. We
therefore tested whether the low intracellular calcium and hyperproliferative
phenotypes of Sec10 knockdown cells were accompanied by activation of the MAPK
pathway. The MAPK pathway involves the phosphorylation cascade of Raf, MEK, and
ERK. By Western blot analysis, phosphorylated ERK (pERK)—a measure of MAPK
activation—was increased in Sec10 knockdown cells relative to normal cells
by 4.6-fold (Figure 2A and
2B). This increase in pERK was blocked completely with a 1-hour
treatment of a MEK inhibitor (U0126) and a src-family inhibitor (PP2). Treatment
with a cAMP-activated PKA inhibitor (H-89) partially blocked activated ERK,
restoring levels of pERK to approximately that of control MDCK cells. Treatment
with other kinase inhibitors, including a PKC inhibitor (BIM) and an mTOR
inhibitor (Rapamycin), were ineffective in blocking overactivation of pERK in
Sec10 knockdown cells (Figure 2C
and 2D), suggesting that the pERK increase was specific for the MAPK
pathway. These data support the idea that the increased pERK observed in Sec10
knockdown cells is due to the combined upstream activities of PKA, Src, and
MEK.
Together, our in vitro results show that Sec10 knockdown cells
display many cellular phenotypes shared with ADPKD cells: from abnormal calcium
regulation associated with an insensitivity to fluid flow, to increased
proliferation associated with MAPK activity.
To determine how sec10 affects cilia and cilia-related processes
in vivo, we utilized morpholinos (MOs) to knockdown
zebrafish Sec10 (zfSec10) levels. Our first start-site morpholino was
ineffective at knocking down zfSec10 levels (see Materials and Methods). Since exocyst Sec8 knockout mice display
very early embryonic lethality, well before kidney development occurs [54], we did
not focus on developing working start-site morpholinos because these would
affect both maternal and zygotic transcripts and could cause early phenotypes
that would preclude any analysis of specific phenotypes. Thus, we utilized
splice-site morpholinos, which would bypass any early general requirement for
the exocyst and allow us to focus on later tissue-specific effects of Sec10
knockdown.
Two splice-site morpholinos (MOs) against zebrafish sec10 were
designed and injected either alone, 15 ng sec10e2i2-MO1, or as a combined dose
of 8 ng sec10e2i2-MO1 +8 ng sec10e3i3-MO2. Hereafter, we will use the
following shorthand for such morpholino-injected embryos: “15ng sec10MO
embryos” and “8+8ng sec10MO embryos”, respectively.
Aberrant splicing was verified by sequencing transcripts from 24 hours post
fertilization (hpf) cDNA libraries made from sec10MO embryos. Multiple splicing
variants were not observed. Sequencing of the transcript from 15ng sec10MO
embryos revealed a 25 bp deletion in exon 2. The same 25 bp deletion and an
additional 27 bp deletion in exons 3 and 4 were observed in the 8+8ng
sec10MO injected embryos. Both cases result in a truncated 33 amino acid protein
product. Furthermore, the level of Sec10 knockdown was assayed directly by
Western blot with antibody against human Sec10 (hSec10) [39] (Figure 3A). While some zfSec10 remained at 1
day post fertilization (dpf) in sec10MO embryos, levels were significantly
reduced soon thereafter. Therefore, these sec10 splice-site morpholinos can be
used to effectively knockdown Sec10.
Given the ciliogenesis defects observed with Sec10 knockdown in
vitro
[39], we
predicted that pronephric cilia would be shorter in sec10MO embryos.
Surprisingly, pronephric cilia length at 1 dpf appeared normal by
immunofluorescence (Figure
3B-3C′). We then assayed whether cilia motility was affected.
Whereas mammalian kidney cilia are non-motile, pronephric cilia in the zebrafish
are motile [18], [25]. We assayed cilia
motility at 2 dpf and, surprisingly, found that it was intact in sec10MO embryos
(Videos S1 and S2).
The discrepancy in ciliogenesis phenotypes between Sec10 knockdown in
vitro and in vivo may be explained by incomplete
knockdown of Sec10 protein in zebrafish. Since we utilized a splice-site
morpholino, it is likely that maternally-deposited RNA and/or protein was
sufficient to allow for establishment of the cilia at 1 dpf. Indeed, our Western
blot analysis detected Sec10 protein at this time (Figure 3A). Therefore, pronephric cilia may
only require Sec10 for initial ciliogenesis, but not for later maintenance of
the structure. This may also be true for cilia motility.
Unexpectedly, sec10MO embryos showed defects in pronephric development despite
the absence of pronephric cilia structure and motility defects. At 1 dpf, the
cilia were disordered specifically within the medial pronephros (uninjected:
n = 0/3 disorganized; 15ng sec10MO:
n = 1/3 disorganized; 8+8ng sec10MO:
n = 3/5 disorganized, compare Figure 3B and 3C), where we have observed
dilations and pronephric cysts in other zebrafish cilia mutants [20], [25].
Since the disorganization suggested pronephric tubule dilation, we performed
histological analysis to look directly for pronephric defects. At 3 dpf, sec10MO
embryos did not show obvious dilations in the pronephros; however, the
morphology of the glomerulus was abnormal. Instead of a normal compact U-shaped
glomerulus, sec10MO embryos showed disorganization, which may be due to
increased cell number (uninjected: n = 0/1 disorganized;
15ng sec10MO: n = 3/5 disorganized; 8+8ng sec10MO:
n = 1/2 disorganized, compare Figure 3D′ and 3E′).
Therefore, while in vivo Sec10 knockdown did not affect
pronephric cilia structure or motility, we still observed defects in pronephric
development.
Though sec10 knockdown did not affect ciliogenesis in
vivo, sec10 knockdown may still perturb other
aspects of cilia function, even if cilia structure and motility are intact.
Consistent with this possibility, sec10MO embryos displayed a range of gross
phenotypes that have been observed in other zebrafish cilia mutants, including
smaller body size, small eyes, edema, and curly tail up [18]-[21], [25], [26], [31].
These phenotypes were variably penetrant (Table 1), despite similar levels of protein
knockdown (Figure 3A, data
not shown). These phenotypes suggest that while a residual level of maternal
Sec10 protein was adequate to maintain cilia structure in sec10MO embryos,
higher levels are required for full wild-type cilia function.
One way in which sec10 could be important for cilia function is
through regulating pkd2 function. Notably,
pkd2 knockdown in zebrafish does not produce defects in
cilia structure [29]-[31] or motility [25],
[30]—similar to what we observed with Sec10 knockdown by
splice-site morpholinos. Since pkd2 is specifically known to be
important for cilia function, and our in vitro analysis
revealed ADPKD-like behaviour in Sec10 knockdown cells, we wanted to determine
whether Sec10 knockdown would share phenotypes associated with
pkd2 knockdown in vivo as well.
Importantly, we noticed that the curly tail up phenotype of sec10MO embryos was
reminiscent of the unique curly tail up observed from loss of
pkd2 in zebrafish (uninjected: 0% curly tail up,
n = 42; compared to 15ng sec10MO: 51%,
n = 97; and 8+8ng sec10MO: 6%,
n = 32; Figure
4A and 4C) [20], [29]-[31].
To determine whether sec10MO embryos shared other phenotypes with pkd2MO embryos,
we investigated whether sec10MO embryos displayed left-right defects. Indeed,
like pkd2MO embryos, sec10MO embryos show defects in left-right patterning with
respect to the positioning of the visceral organs (Table 2). Additionally, we observed defects
in asymmetric nodal gene expression in sec10MO embryos (Table 2), as would be
expected if cilia function was disrupted. Ciliary function in Kupffer's
vesicle is known to be upstream of asymmetric nodal expression
[28].
Thus, loss of sec10 may affect left-right patterning indirectly
through its effects on cilia and/or polycystin-2 function.
Also similar to pkd2MO embryos [29], sec10MO embryos showed glomerular expansion in the
pronephros by in situ hybridization with the glomerular marker,
Wilm's tumor 1a (wt1a) at 3 dpf
(Figure 4G-4I′).
Wild-type embryos showed a condensed glomerular stain (100% condensed,
n = 30). By contrast, only 50% of 15ng sec10MO
embryos showed a condensed stain (50% condensed, 31% moderate
enlargement, 19% severe enlargement, n = 16),
similar to 4ng pkd2MO embryos (35% condensed, 59% moderate
enlargement, 6% severe enlargement, n = 32). Given
the glomerular disorganization we observed by histology (Figure 3E′), the expansion in the
wt1a stain in sec10MO embryos may be due to increased cell
proliferation.
In spite of the expanded glomerulus, sec10MO embryos did not display obvious
glomerular dilations by histology, nor did they show pronephric cysts. While we
note that the lack of dilation in sec10MO embryos contrasts with the glomerular
dilation observed in pkd2MO embryos [20], [25], [30], zygotic
pkd2 mutants do not show glomerular dilation either [20], [31]. It
is likely that maternal contribution of polycystin-2 and Sec10 explains why we
do not observe glomerular dilation or pronephric cysts in pkd2
mutants and sec10MO embryos, respectively.
To determine whether these in vivo pronephric defects were
accompanied by the cellular phenotypes we observed with Sec10 knockdown
in vitro, we assayed pERK levels by Western blot to
determine the level of MAPK activation. Consistent with our in
vitro results, pkd2MO and sec10MO embryos showed abnormally
increased pERK levels at 5 dpf, relative to uninjected embryos, at 5 dpf, (Figure 4J). MAPK activation
was more pronounced in sec10MO than in pkd2MO embryos.
Therefore, sec10MO embryos share curly tail up, left-right, pronephric, and
cellular phenotypes with pkd2MO embryos. Consistent with our in
vitro analysis, we observed that loss of sec10
partially phenocopies loss of pkd2, supporting the idea that
exocyst function is required for polycystin-2 function. Furthermore, our
observation of these cilia-related phenotypes is consistent with a role for
sec10 in cilia function, even though overt defects in cilia
length and motility were not observed upon knockdown.
Our in vitro and in vivo analyses together
support a link between exocyst sec10 and the ADPKD gene
pkd2. While the shared curly tail up phenotype is more
specific to pkd2, the left-right defects and
wt1a expansion phenotypes shared between sec10MO and pkd2MO
embryos have also been observed upon knockdown of other ciliary proteins [18],
[24]. We
therefore wanted to directly test for a specific genetic interaction between
these two genes. We titrated both sec10 and pkd2 morpholinos to find suboptimal
doses that did not result in strong gross phenotypes on their own.
Interestingly, when we co-injected both morpholinos at these reduced doses we
observed a striking synergistic effect on the curly tail up phenotype (Figure 4D-4F, Figure S1A,
S1A′). Co-injection of 0.25ng pkd2MO and 7.5ng sec10MO yielded
curly tail up phenotypes when each morpholino alone produced completely
wild-type tails. Likewise, co-injection with a slightly higher dose of pkd2MO
(2ng pkd2MO) shifted almost all embryos from a range of curly up phenotypes into
a severe curly up phenotype. We also observed effects upon the left-right defect
(Figure S1B
and S1B′) and the wt1a glomerular expansion
(Figure S1C
and S1C′) phenotypes. We had to use different suboptimal doses
of pkd2 morpholino for the phenotypes because left-right defects, curly tail,
and pronephric phenotypes are extremely dose-sensitive to pkd2
levels [31]. The genetic interaction we observed between
sec10 and pkd2 morpholinos suggests that
sec10 may play a role in pkd2 function in
multiple cilia-related processes.
The observed genetic interaction can be interpreted in two ways:
pkd2 and sec10 may act in parallel
pathways, or in the same pathway. In the former case, redundancy between two
parallel pathways—one requiring pkd2 and one requiring
sec10—explains the lack of phenotypes in the single
suboptimal dose conditions; a slight reduction of both pathways upon
co-injection of both morpholinos leads to a failure to complement. In the latter
case, where pkd2 and sec10 act in the same
pathway, reduction of function at two steps prevents wild-type function in a
dosage-sensitive manner. This latter interpretation is supported by a similar
synergistic effect that was observed between the two ciliary proteins Seahorse
and Inversin [23]; these proteins were then shown to biochemically
interact, supporting the idea that they are likely to act in the same pathway
[23],
[24].
Previously, we proposed that Sec10 and the exocyst are important for
transporting proteins important for ciliary structure [39]. We, therefore, suggest a
similar model for transporting proteins important for ciliary function, like
polycystin-2. If Sec10 is similarly required to transport polycystin-2, then
morpholino co-injection would further impair ciliary polycystin-2 levels beyond
that seen following direct knockdown of polycystin-2 by a sub-optimal dose of
morpholino, because a reduced amount of Sec10 would be present to effectively
transport the remaining polycystin-2.
In light of the genetic interaction we observed between sec10
and pkd2, we wanted to determine whether we could detect a
biochemical interaction between Sec10 and polycystin-2. Using lipofectamine, we
transfected a cDNA encoding human polycystin-2-myc into human embryonic kidney
293 (HEK293) cells. Western blotting of the transfected HEK293 cell lysates with
antibodies against both polycystin-2 (not shown) and the myc epitope tag (Figure 5A), identified a band
at approximately twice the molecular weight of polycystin-2, suggesting it was
in polymeric form. Purified Sepharose-immobilized Sec10-GST was then used as an
affinity resin to pull down specific binding proteins from polycystin-2-myc
transfected HEK293 cell lysates. Western blotting, using antibodies directed
against both polycystin-2 and the myc epitope tag, identified polycystin-2 as a
Sec10-GST binding protein that was not recovered on the GST resin alone (Figure 5B). Using another
technique, we showed that exocyst Sec8 co-immunoprecipitated with polycystin-2,
but not the isotype control, from intracellular vesicles isolated from mouse
kidney lysate (Figure 5C).
Having shown that the exocyst and polycystin-2 can interact in cell lysates, we
next investigated if native polycystin-2 would co-localize with the exocyst at
the primary cilium. We previously showed by immunofluorescence and electron gold
microscopy that the exocyst localized to the primary cilium in MDCK cells [39]. Using a
polyclonal polycystin-2 antibody that recognizes canine polycystin-2 [55], we first
demonstrated that polycystin-2 co-localizes with acetylated alpha tubulin at the
primary cilium in MDCK cells (Figure S2). We then showed that exocyst Sec8
co-localizes along the length of the primary cilium with polycystin-2 (Figure 5D).
Consistent with our phenotypic and genetic analyses in zebrafish, we observe that
Sec10 and polycystin-2 biochemically interact and co-localize in the cilia of
cultured renal tubule epithelial cells.
Since we observed a biochemical interaction between the exocyst and polycystin 2,
we wanted to determine whether the exocyst biochemically interacts with other
ciliary proteins. We focused on ciliary proteins such as IFT88 and IFT20,
because they have both been found to interact with polycystin-2 [56], [57].
Previously we demonstrated that levels of IFT88 were reduced upon Sec10 knockdown
in vitro
[39]. IFT88 is
required for cilia structure [58] and is mutated in the orpk mouse
model of PKD [59]. Interestingly, IFT88 has been shown to be in a
complex with polycystin-2 and possibly trafficked together with it in the cell
[57].
Consistent with this interpretation, IFT88 is not itself required for
polycystin-2 localization to primary cilia in cultured cells [56]. We used
GST-pulldown assays to test for biochemical interactions, as relatively larger
amounts of binding proteins can be obtained from the affinity column. A
Sec10-GST fusion protein was purified on glutathione Sepharose and used as an
affinity matrix for the purification of specific binding proteins from HEK293
cell lysates. IFT88 interacted with Sec10 and was found in the pulldown fraction
(Figure 6A). To
demonstrate specificity, we probed for GAPDH, a protein not known to interact
with the exocyst. In Figure
6A (bottom), GAPDH is seen in the cell lysate, but not in the
Sec10-GST pulldown fractions. The glutathione-Sepharose immobilized Sec10 also
pulled down exocyst Sec8 from the cell lysate, serving as a positive control for
exocyst binding (Figure
6A).
IFT20 is a ciliary protein known to be trafficked to cilia on vesicles, and
knockout of IFT20 leads to polycystic kidney disease in mice [56], [60].
Interestingly, IFT20 has been shown to be important for polycystin-2
localization to the primary cilium [56]. IFT20 was also found in
the pulldown fraction, and is, therefore, a Sec10 binding partner (Figure 6A). As a negative
control, in all the pulldown experiments bead-immobilized GST alone was used and
no proteins were detected in the pulldown fractions.
Using another technique, we showed that Sec8, IFT88, and IFT20 all
co-immunoprecipitated with Sec10-myc from MDCK cell lysates, but not the isotype
control (Figure 6B).
Therefore, Sec10 biochemically interacts not only with proteins important in
cilia function (like polycystin-2), but also with proteins implicated in cilia
formation and trafficking, such as IFT88 and IFT20.
Given that IFT88 has been shown to interact with polycystin-2 [57], we next
investigated if IFT88 might act as a direct bridging protein between Sec10 and
polycystin-2. An immortalized cell line, 94D, derived from the cortical
collecting duct cells of an Oak Ridge Polycystic Kidney mutant mouse
(orpk), and deficient in IFT88, was generated by Yoder and
colleagues [61]. An immortalized IFT88 “rescue” cell line,
BAP2, was also generated with endogenous levels of IFT88. To determine if IFT88
is necessary for the interaction between polycystin-2 and Sec10, Sec10-GST
pulldowns using lysate from both the 94D and BAP2 cell lines were performed.
There was no difference in the amount of polycystin-2 pulled down from the
IFT88-deficient or -replete cell lines (Figure 6C). Therefore, IFT88 is not required
for the Sec10 interaction with polycystin-2.
We then performed in vitro translation of Sec8, polycystin-2,
IFT88, and p53, followed by pulldown with Sec10-GST. We demonstrated an
interaction between Sec10 and Sec8 (our positive control), but not p53 (our
negative control), polycystin-2, or IFT88 (Figure 6D). Therefore, while the exocyst
biochemically interacts with polycystin-2 and IFT88, these interactions are not
direct, indicating there are remaining proteins yet to be identified in these
complexes.
As determined by Western blot, equal amounts of polycystin-2 are seen in control,
Sec10-overexpressing, and Sec10 knockdown MDCK cells (Figure 6E). If Sec10 is required for
polycystin-2 localization, we would expect to see a change in polycystin-2
localization in Sec10 knockdown MDCK cells. Indeed, Sec10 knockdown cells show a
loss of native polycystin-2 localization at the primary cilium by
immunofluorescence staining (Figure
6F). However, it should be noted that this result is not surprising
given that Sec10 knockdown cells, as we previously reported, have no, or few,
cilia [39].
Therefore, the loss of polycystin-2 localization could well be an indirect
effect of the ciliogenesis defect.
Here, we describe the first genetic and biochemical link between the exocyst and a
human disease gene, PKD2. Phenotypic analyses of sec10MO embryos support a role for
sec10 in multiple cilia-related processes, which is surprising
given the absence of obvious defects in cilia structure or motility. Thus, while our
previous work indicated a role for Sec10 in ciliogenesis [39], the results presented here
suggest that Sec10 is important for cilia function as well. Furthermore, we
demonstrate a specific genetic interaction with pkd2, a gene that
influences cilia function without affecting cilia structure. Knockdown of Sec10
in vitro and in vivo partially phenocopies
knockdown of the ADPKD protein polycystin-2. Consistent with these results, we
observe that Sec10 and polycystin-2 co-localize in the primary cilium in
vitro. We further report biochemical interactions between multiple
exocyst proteins and ciliary proteins—polycystin-2, IFT88, and IFT20. Together
with our previous work [39], our studies demonstrate that the exocyst protein Sec10
is likely to be important for both cilia formation and function.
We demonstrate that Sec10 knockdown also leads to phenotypes associated with ADPKD.
In vitro, we show that knockdown of Sec10 leads to decreased
basal intracellular calcium levels and lack of a calcium response to fluid flow.
Conversely, Sec10-overexpressing cells showed a significantly increased calcium
response to fluid flow. These results were not unexpected given the defects in
ciliogenesis in Sec10 knockdown cells, and the increased ciliogenesis seen in
Sec10-overexpressing cells [39]. We also showed increased cell proliferation in Sec10
knockdown cells. Increased cell proliferation is a well-known characteristic of
ADPKD cells and plays a major role in the formation of the cysts that destroy the
kidney, leading some to refer to ADPKD as “neoplasia in disguise” [47].
Our analysis of the role sec10 plays in zebrafish development then
revealed that Sec10 may also be required for cilia function, separate from a role in
cilia formation. When we used morpholinos to knockdown Sec10 levels in
vivo in zebrafish, we did not observe a gross defect in pronephros
cilia morphology or motility. While this was initially surprising because in
vitro knockdown results in severe ciliogenesis defects [39], we believe that
maternal Sec10—which is unaffected by our splice-site morpholinos—is
sufficient to allow for cilia assembly. However, we believe this residual maternal
protein was unable to restore complete cilia function during zebrafish development
because sec10MO embryos still showed cilia-related phenotypes that have been
observed in other cilia mutants in zebrafish—such as left-right patterning
defects and glomerular expansion. Therefore, we believe the partial knockdown of
Sec10 levels in vivo with splice-site morpholinos allowed us to
uncover a role for exocyst Sec10 in cilia function.
If true, we would expect that future studies knocking down both maternal and zygotic
Sec10 with a start-site morpholino might recapitulate a ciliogenesis defect like
that observed in vitro. We tested one start-site morpholino but it
did not effectively knockdown Sec10 (see Materials and
Methods). It should also be noted that even translation blocking
morpholinos do not always produce maternal and zygotic losses seen in actual
maternal zygotic mutants. For example, translation blocking MOs against IFT88 in
fish show almost complete loss of protein by Western blot [62], and yet still do not display
phenotypes observed in the maternal-zygotic ift88/oval zebrafish
mutant [63]. Thus,
a maternal-zygotic sec10 mutant would be required to definitively
say whether or not Sec10 is required for ciliogenesis in zebrafish, as it is in MDCK
cells.
Finally, we provide phenotypic and genetic evidence that sec10 may
be important specifically for pkd2 function in these cilia-related
processes. This is interesting because pkd2 is one of the causative
genes for ADPKD and could explain the ADPKD-like phenotypes we observed upon Sec10
knockdown in vitro. Multiple lines of evidence support our
interpretation that sec10 is important for pkd2
function in vivo: 1) pkd2 knockdown has been
implicated in multiple cilia-related processes and is known to affect cilia
function, but not structure or motility [14], [29]-[31], [33]; 2) Similar to
pkd2 knockdown, sec10MO embryos share several cilia-related
phenotypes—including the curly tail up and MAPK activation phenotypes, even
though they similarly do not show defects in cilia length or motility; 3) We
observed specific genetic interactions between sec10 and
pkd2 for multiple cilia-related phenotypes—including
curly tail up, left-right defects, and aberrant wt1a glomerular
expansion.
It will be important to tease apart the extent to which sec10
phenotypes are explained solely by inhibition of pkd2 function. We
have provided evidence supporting the idea that many of the sec10
phenotypes are likely due to its regulation of pkd2. But it
plausible that sec10 may be important for the function of other
ciliary proteins as well. This is supported by the fact that sec10MO embryos possess
cilia-related phenotypes, such as small eyes, that are not observed upon loss of
pkd2.
While we favour a model where sec10 is required for ciliary
function, we recognize that Sec10 may play a different role since some proteins
implicated in cilia function also have cilia-independent functions. Many ciliary
proteins are not exclusively localized to the cilium, and it has recently been
argued that polycystin function in the endoplasmic reticulum is more relevant for
the observed curly tail phenotype [64]. Indeed, multiple IFT proteins have recently been shown
to have important functions in non-ciliated cells [65].
We observed activation of ERK (MAPK) following knockdown of Sec10 both in
vitro and in vivo. Similar to polycystin-2 knockout in
mouse [66], we
report that polycystin-2 knockdown in zebrafish results in elevated pERK levels.
Since we propose that polycystin-2 function requires the exocyst, it is not
surprising that we also observed increased pERK levels upon Sec10 knockdown. It
remains to be determined how MAPK activation relates to cystogenesis, since sec10MO
embryos did not show pronephric cysts. We did, however, observe glomerular
disorganization that could have been the result of unchecked proliferation. So it is
possible that while MAPK activity may regulate proliferation associated with
cystogenesis, MAPK hyperactivation alone is not sufficient to cause cystogenesis. If
true, this may explain conflicting results others have observed using MAPK
inhibitors to abrogate cystogenesis. On the one hand, inhibition of ERK activation
with the oral MAP/ERK kinase inhibitor, PD184352, largely prevented cystogenesis in
the pcy mouse model of polycystic kidney disease [50]. On the other
hand, inhibition of ERK activation with the MEK inhibitor, U0126, failed to prevent
cystogenesis in a Pkd1 mouse model of ADPKD [66]. The effectiveness of these
inhibitors may depend on the specific genetic background and the role MAPK
activation plays in cystogenesis in that background. While we observed MAPK
hyper-activation in both sec10MO and pkd2MO embryos, only pkd2MO embryos show cysts
[20], [25], [30].
In light of its known role in trafficking basolateral proteins to the plasma membrane
[41]-[44], we propose
that the exocyst is required for the delivery of ciliary proteins to the cilium,
including polycystin-2. This model would explain why Sec10 knockdown results in
defects in both cilia formation and function.
If our model were true, identification of the IFT B proteins IFT88 and IFT20 in
complex with exocyst proteins could mean one of several possibilities: 1) IFT B and
polycystin-2 proteins are both being trafficked by the exocyst; 2) The IFT B complex
is directly responsible for trafficking polycystin-2 and the exocyst is only
directly responsible for trafficking IFT B. We favour the first model (Figure 7) for the following
reasons: 1) The biochemical interaction between Sec10 and polycystin-2 that we
report did not require IFT88; 2) If the exocyst was required in an indirect fashion
for pkd2 function, through the action of IFT proteins, then it
should not have been possible to observe cilia-related phenotypes in
vivo in the absence of pronephric ciliary structural defects.
Relatedly, IFT20 may play a special role in bridging these complexes since it has
been implicated in Golgi-to-cilium vesicular traffic [56]. Unlike other IFT proteins,
which are only localized to the cilium, IFT20 is also observed in the Golgi. Pazour
and colleagues have proposed that IFT20 may mark vesicles carrying proteins destined
for the cilium [56]. IFT20 has been specifically implicated in localizing
polycystin-2 to the cilium [56], and polycystin-2 may be trafficked as part of a larger
complex containing IFT88 and IFT57 [57]. Our detection of an interaction between Sec10 and IFT20
suggests that IFT20-marked vesicles may utilize the exocyst to dock at the cilium,
or that the exocyst may chaperone IFT20-positive vesicles from the Golgi to the
primary cilium (Figure 7).
Interestingly, IFT20 knockdown by splice-site morpholinos in zebrafish does not
significantly disrupt pronephric cilia structure, even though it results in ciliary
loss in the otic vesicle [50]. Therefore, loss of IFT20 may result in milder
ciliogenesis defects then are observed following knockdown of other IFT proteins in
zebrafish.
However, future studies are needed to clarify the interactions among the exocyst, IFT
proteins, and the polycystins. Detailed immunofluorescence studies are needed to
demonstrate that the exocyst is directly involved in ciliary trafficking. If our
model were true, we would expect to see a reduction in the ciliary localization of
polycystin-2 upon Sec10 knockdown. Unfortunately, we are unable to test this
in vitro, which is a limitation of our study, because of the
ciliogenesis defects upon Sec10 knockdown in MDCK (Figure 6F) and ARPE-19 cells (data not shown).
In vivo, we were unable to consistently detect sub-cellular
localization of polycystin-2 in the cilium by immunofluorescence utilizing multiple
polycystin-2 antibodies (data not shown), thus we were unable to look at
polycystin-2 localization in the cilia that remain in sec10MO embryos. Therefore, it
remains to be seen how the biochemical interactions we observe between the exocyst
and ciliary proteins relate to their trafficking to the cilium.
In summary, we have shown that Sec10, a conserved and crucial component of the
exocyst complex, is likely to promote pkd2 function in many
cilia-related phenotypes. These data support an additional role for the exocyst, not
only in ciliogenesis [39], but also for cilia function. We report a novel genetic
interaction between sec10 and pkd2, which supports
the ADPKD-like and pkd2-like phenotypes we observed upon Sec10
knockdown in vitro and in vivo. Our data support a
model in which the exocyst is required for the trafficking of proteins essential for
ciliary function and structure, possibly in conjunction with IFT20. Future work will
reveal the extent to which the exocyst participates in ciliary trafficking, and
whether it can be utilized as a novel target for therapeutic intervention in ADPKD,
a disease for which there are currently no approved treatments beyond supportive
care.
All zebrafish experiments were approved by the Institutional Animal Care and Use
Committee at Princeton University.
All MDCK cell lines used were derived from low passage type II MDCK cells that
were obtained from Dr. K. Mostov (UCSF, San Francisco, CA), and which were
originally cloned by Dr. D. Louvard (European Molecular Biology Laboratory,
Heidelberg, Germany). The monoclonal MDCK type II cell line with silenced
expression of Sec10 through stable transfection of shRNA designed against the
canine Sec10 gene, as well as the Sec10 overexpression MDCK type II cell line,
have been previously described [39]. MDCK cells were cultured on 0.4 µm Transwell
filters in modified Eagle's minimal essential medium (MEM) containing
Earl's balanced salt solution and L-glutamine, with 5% fetal bovine
serum, 100 U/ml penicillin, and 100 µg/ml streptomycin.
The IFT88-deficient and –rescued cell lines, 94D and BAP2, respectively, a
generous gift of Dr. Brad Yoder, were derived from the cortical collecting duct
cells of the orpk mouse, and have been previously described
[61]. The
orpk mutant mouse has an insertional mutation in the Ift88/Tg737/polaris gene,
which results in a hypomorphic allele, with extremely low Ift88 protein levels.
This well-studied mutant mouse was crossed with the ImmortoMouse (Charles River
Laboratories) and an immortalized cortical collecting duct cell line was
established. These cells were transfected with a wild-type Ift88 gene to restore
Ift88 expression (BAP2 line), or with an empty vector to retain hypomorphic
mutant Ift88 cells (94D line). To keep these cell lines undifferentiated and
immortalized, cells were grown at 33°C with 10 U/ml interferon-γ in
collecting duct media (DMEM/F-12, 10% FBS, 1.3 µg/l sodium
selenite, 1.3 µg/l T3, 5 mg/l insulin, 5 mg/l transferrin, 2.5 mM
L-glutamine, 5 µM dexamethasone, 100 U/ml penicillin, and 100 µg/ml
streptomycin). To differentiate these cells and inactivate SV40 large-T antigen,
cells were grown at 37°C in the absence of interferon-γ for 3 days
before performing Sec10-GST pulldowns.
Small molecule inhibitors, which were incubated with cells at the manufacturer
recommended concentrations for one hour, include: the MEK inhibitor U0126 at 10
µM (Promega, Madison, WI), PKA inhibitor H89 at 10 µM (Calbiochem,
San Diego, CA), Src family inhibitor PP2 at 10 µM (Calbiochem), PKC
inhibitor bisindolylmaleimide I (BIM) at 1 µM (Calbiochem), and mTOR
inhibitor Rapamycin (Rap) at 20 nM (LC Laboratories, Woburn, MA).
Live cell fluorescent imaging and intracellular calcium concentration
measurements were done as previously described [46]. Briefly, cells were grown
to confluent monolayers on 0.4 µm clear polyester permeable supports in a
perfusion chamber, which allows precisely controlled shear fluid flow across the
apical surface of the epithelium. After 5-7 days in confluent monolayers, cells
were loaded with fura 2 (10 µM fura 2/AM; Teflabs, Austin, TX), and
transferred to the incubation chamber on the microscope, where a constant
temperature of 37°C was maintained, with a 5% CO2
concentration. Cells were then perfused with Ringer's solution with 1 mM
probenecid. We performed dual-excitation wavelength fluorescence microscopy
(Photon Technologies, Birmingham, NJ) with a Nikon microscope, x20 S Fluor
long-working distance objective, and a cooled SenSys charged-coupled camera
(Photometrics, Tucson, AZ). Fura 2 was excited at wavelengths of 340 and 380 nm
and emitted fluorescence was measured at 510 nm. Data were obtained in each
experiment from a grid of 20 regions of interest each containing 8–10
cells. Cells were maintained in the absence of apical flow for 10 minutes
followed by an abrupt increase to a flow rate of 5 ml/min. The basolateral flow
rate remained constant at 1.5 ml/min.
For cell proliferation measurements, MDCK cells were seeded in parallel 96-well
plates in triplicates at 4000 cells/well, and were allowed to attach for 24 h at
37°C. The medium was replenished on all plates, and the cells on one plate
were counted using the CellTiter-Glo viability assay (Promega) and considered
cell population at t = 0 h. Cells on other plates were
grown for 24 h and 48 h at 37°C, and then counted using the same method.
Luminescence was measured using a Becton Dickson microplate reader, and
proliferation rates were calculated by dividing cell number at a given timepoint
by the initial number of seeded cells (t = 0).
For MDCK cell lysates, immunoblotting was performed as previously described [39]. To make
zebrafish protein lysates, embryos were washed in E3 buffer and re-suspended in
Ringer's solution. To remove the yolk protein, the sample was vortexed five
times in 30 seconds bursts and the supernatant was removed following a gentle
centrifugation at 300g for 1 minute at 4°C. The pellet was re-suspended in
lysis buffer [67] supplemented with a protease inhibitor cocktail
(P1860, Sigma) and 200 µM PMSF. The protein lysate supernatant was removed
following centrifugation at 14,000g for 10 minutes at 4°C. Protein samples
were mixed with Laemmli buffer and boiled for 5 minutes. Immunoblotting was
performed using standard protocols.
The antibodies used in this study are: rabbit polyclonal anti-Sec10 which was
described previously [39], rabbit polyclonal anti-phospho-ERK1/2 (#9101, Cell
Signaling, Danvers, MA), rabbit polyclonal anti-total ERK1(/2) (sc-94, Santa
Cruz Biotechnology, Inc., Santa Cruz, CA), mouse monoclonal anti-GAPDH (G8795,
Sigma, St. Louis, MO), rabbit polyclonal anti-γ tubulin (T5192, Sigma),
mouse monoclonal anti-acetylated α-tubulin (T6793, Sigma), rabbit polyclonal
anti-IFT88 ([68]), mouse monoclonal anti-Sec8 (Assay Designs, Ann
Arbor, MI), mouse monoclonal anti-myc (#2276, Cell Signaling), rabbit polyclonal
anti-IFT20 ([56], a generous gift from Dr. Greg Pazour), mouse
monoclonal anti-polycystin-2 (D-3, Santa Cruz Biotechnology, Inc.), rabbit
polyclonal anti-TRPM4 (Santa Cruz Biotechnology), rabbit polyclonal
anti-polycystin-2 (a gift from the Johns Hopkins Research and Clinical Core
Center), normal isotype mouse IgG control (Santa Cruz), and normal isotype
rabbit IgG control (Santa Cruz).
Embryos were injected at the one-to-8-cell stage, and morpholinos were diluted
with phenol red tracer at 5 μg/μL phenol and injected at 500 picoLiter
or 1 nanoLiter/embryo. Splice-site morpholinos designed against zebrafish
sec10 were purchased from Gene Tools, LLC. (Philomath, OR):
sec10e2i2-MO1 (5′-
AATATTCTGTAACTCACTTCTTAGG -3′), sec10e3i3-MO2
(5′-CAAATGTAAAGACGACTGACTTGGT-3′), and
sec10AUG-MO1 (5′-
CGAATAATTGAGCTGTCGTAGCCAT-3′).
Sec10e2i2-MO1 was designed to target the exon 2-intron 2 boundary (hence
“e2i2”). Sec10e3i3-MO2 was targeted against the exon 3-intron 3
boundary (hence “e3i3”). Splice-site morpholinos were injected
either as a single dose of 15 ng sec10e2i2-MO1 (designated in the text as
“15ng sec10MO”) or a combined dose of 8 ng sec10e2i2-MO1 +8 ng
sec10e3i3-MO2 (designated in the text as “8+8ng sec10MO”) per
embryo. Developmental delay was more noticeable with 8+8ng sec10MO.
Sec10e3i3-MO2 did not show gross phenotypes when injected alone. sec10AUG-MO1
was designed to target the ATG (hence “AUG”, complementary sequence
underlined above). sec10AUG-MO1 was tested but did not knockdown zfSec10 by
Western blot analysis even at doses of 15 ng per embryo (data not shown); since
some MOs fail to work for inexplicable reasons, this was not pursued further.
pkd2AUG MO (5′-AGGACGAACGCGACTGGAGCTCATC-3′), start site
[20], was
injected at 4 ng per embryo.
Immunostaining of MDCK cells grown on Transwell filters was performed as
previously described [39], except the cells were fixed with 4%
paraformaldehyde for 15 minutes at 37°C. Immunofluorescence for pronephric
cilia and histology in zebrafish was performed similar to previously published
protocols [25]. The antibodies used in this study for
immunofluorescence studies: mouse monoclonal anti-acetylated α-tubulin
(T6793, Sigma), rabbit polyclonal anti-polycystin-2 antibody (a gift from the
Johns Hopkins Research and Clinical Core Center, [55]), goat-anti-mouse
IgG2b/g2b chain specific-FITC (Southern Biotech#1090-02, Birmingham, AL).
In situ hybridization in zebrafish was performed using
standard protocols [69]. Probes used: wt1a, myl7,
foxa3, ins, spaw, lefty1,
and lefty2.
All images were captured in TIF format and processed in Adobe Photoshop CS4. For
immunofluorescence, zebrafish embryos were imaged on a Zeiss LSM 510 Confocal
Microscope with a 40x water objective and captured at 2x zoom, and the LSM Image
Browser application; TIFs were captured at 150 ppi. For histology and in
situ hybridization, samples were imaged using a Leica DM RA2
Microscope, a Leica DFC490 digital camera, and the Leica Application Suite v
3.1.0 software; TIFs were captured at 96 ppi. For live images, embryos were
imaged on a Leica MZ FLIII Stereo-Fluorescence Microscope using a Jenoptik
LaserOptikSysteme ProgResC14 digital camera and Picture Frame v2.3 software;
TIFs were captured at 150 ppi. For video microscopy of pronephric cilia, embryos
were imaged as previously described [25], using an Olympus
BX51 Upright Microscope with a 60x water immersion objective, an Andor
Technology Luca EMCCD digital camera, and Matlab software.
Full-length human Sec10 cDNA was cloned in frame into the plasmid pGEX-4T-1
(Amersham Biosciences, Piscataway, NJ), and transformed into the DE3 strain of
Escherichia coli (Stratagene, La Jolla, CA). GST fusion
protein expression was induced by adding
isopropyl-1-thio-β-D-galactopyranoside to growing cultures and shaking for
an additional 3 h at 37°C. Recombinant proteins were purified with
glutathione-Sepharose (Amersham Biosciences) following bacterial cell lysis. For
pull-down experiments, lysates from wild-type HEK293 cells, HEK293 cells
transfected with PKD2-myc (a generous gift from Dr. S. Solmo), or differentiated
94D and BAP2 cell lines, were incubated overnight with Sec10-GST, or GST only,
bound to glutathione-Sepharose. Pull-downs were washed extensively, and then
resuspended in Laemmli buffer and boiled, and equal amounts were electrophoresed
by SDS-PAGE. Bound IFT88, IFT20, GAPDH, Sec8, and polycystin-2-myc were detected
by Western blot analysis.
Co-immunoprecipitations for polycystin-2 were performed from intracellular
vesicle fractions of mouse kidney lysates, isolated as described [70] . Isolated
vesicles were incubated overnight with polycystin-2 antibody, or equal amounts
of control mouse IgG, and protein complexes were precipitated with ProteinG
Dynabeads (Invitrogen). After washing five times in vesicle isolation buffer,
the precipitated protein complexes were analyzed with SDS-PAGE and Western
blotting. Co-immunoprecipitations for Sec10 were performed from confluent MDCK
cells overexpressing Sec10 containing a myc epitope tag, since our Sec10
antibody cannot be used for immunoprecipitation. Cells were grown 5-7 days past
confluency, washed with PBS, and incubated with 1 mM of the membrane-permeable
chemical crosslinker dithiobis(succinimidylpropionate) (DSP) (Thermo Scientific)
for 30 minutes at room temperature. The cells were quenched with TBS for 15 min,
and then lysed in Co-IP lysis buffer (20 mM HEPES pH 7.4, 120 mM NaCl, 1 mM
EDTA, 1% IGEPAL CA-630). Soluble proteins from the lysates were incubated
overnight with 2 µg of various antibodies in parallel with equal amounts
of isotype control IgG. Protein-G agarose (Invitrogen) was used to precipitate
the protein complexes, and after five washes with the Co-IP buffer, the agarose
resin was resuspended in Laemmli buffer. Equal amounts of samples were
electrophoresis by SDS-PAGE, and co-immunoprecipitated proteins were detected by
Western blot analysis utilizing the Trueblot-HRP secondary antibodies
(eBioscience).
One-way ANOVAs, with post-hoc Tukey test for statistical significance, were
performed to compare band intensities from Western blots and cell proliferation
rates using the Prism statistical software (Graphpad, San Diego, CA).
Fisher's exact tests for statistical significance were performed to compare
phenotypic analyses in zebrafish using R statistical software.
|
10.1371/journal.pcbi.1004897 | Optimal Current Transfer in Dendrites | Integration of synaptic currents across an extensive dendritic tree is a prerequisite for computation in the brain. Dendritic tapering away from the soma has been suggested to both equalise contributions from synapses at different locations and maximise the current transfer to the soma. To find out how this is achieved precisely, an analytical solution for the current transfer in dendrites with arbitrary taper is required. We derive here an asymptotic approximation that accurately matches results from numerical simulations. From this we then determine the diameter profile that maximises the current transfer to the soma. We find a simple quadratic form that matches diameters obtained experimentally, indicating a fundamental architectural principle of the brain that links dendritic diameters to signal transmission.
| Neurons take a great variety of shapes that allow them to perform their different computational roles across the brain. The most distinctive visible feature of many neurons is the extensively branched network of cable-like projections that make up their dendritic tree. A neuron receives current-inducing synaptic contacts from other cells across its dendritic tree. As in the case of botanical trees, dendritic trees are strongly tapered towards their tips. This tapering has previously been shown to offer a number of advantages over a constant width, both in terms of reduced energy requirements and the robust integration of inputs at different locations. However, in order to predict the computations that neurons perform, analytical solutions for the flow of input currents tend to assume constant dendritic diameters. Here we introduce an asymptotic approximation that accurately models the current transfer in dendritic trees with arbitrary, continuously changing, diameters. When we then determine the diameter profiles that maximise current transfer towards the cell body we find diameters similar to those observed in real neurons. We conclude that the tapering in dendritic trees to optimise signal transmission is a fundamental architectural principle of the brain.
| Integration of synaptic inputs relies on the propagation of currents arising from sources across the dendritic tree. Whilst active processes strongly contribute to current flow in most neurons [1–3], understanding the passive backbone to transmission is key to an intuitive grasp of dendritic function; the results of Wilfrid Rall in highlighting the properties of cylindrical dendrites [4–6] are of foundational importance in compartmental modelling and computational neuroscience. Dendrites are, however, not generally cylindrical. The distal taper seen in the majority of all cases appears to both increase passive current flow towards the soma [7–9], thus reducing the energy requirements of active compensatory processes, and to contribute to the phenomenon of dendritic democracy, where somatic voltage amplitudes are equalised between different synaptic sites [10–12].
Common numerical approaches to modelling taper treat a dendritic cable as a series of cylinders or linearly tapering frusta [5,13–18]. Whilst these techniques are accurate and powerful, there is much to be gained from an analytical solution to the voltage in terms of intuition and computational speed. A number of solutions for the voltage in non-uniform cables exist [19–21], but these involve either the more tractable cases of varying electrotonic properties with constant radius or are limited to a few forms of radius taper.
We present an asymptotic approximation to the voltage in dendrites with any given taper profile using the insight that voltage attenuation is substantially faster than radius change in realistic morphologies. A particularly appealing prospect for such an approach is that the optimal taper profile to transfer distal synaptic currents to the soma can then be derived using variational calculus. The optimal taper profile is shown to match the results of numerical optimisation and predict radii measured experimentally from a number of different cell classes.
A length of passive dendrite tapers with radius at distance x given by r(x). The leak conductance per unit area is denoted gl, the axial resistance ra, and the membrane time constant τ. Then the voltage above equilibrium v(x, t) at location x and time t obeys the generalised cable equation
τ∂v∂t=−v+12raglr(x)11+(r′(x))2∂∂x[r2(x)∂v∂x]
(1)
The rate of voltage attenuation is generally significantly steeper than the rate of change of dendritic radius, allowing use of the method of multiple scales [22] to accurately approximate the voltage evolution. We introduce X = ϵx as the ‘slow’ taper variable and treat it as independent of x. Large regions of most dendritic trees admit small values of ϵ (~0.01, S1 Fig).
Expanding in ϵ, gives the first-order steady-state solution (see Methods)
v(x)=λ(x)[Ae∫x′x1λ(s)ds+Be−∫x′x1λ(s)ds]
(2)
for λ(x) = r(x)2ragl the location-dependent electrotonic length, x′ a site of current injection, and constants A and B determined by the boundary constraints.
To demonstrate the validity of this approximation, we generated a series of artificial dendritic cables and compared the first-order approximation to the numerical solution (Fig 1). The artificial cables have periodically changing diameters with a random amplitude for each period. Increasing the period and reducing the amplitude smooths the artificial cable, reducing ϵ and improving the approximation. The multiple-scales solution provides an accurate approximation to the voltage in realistic dendritic cables.
The simple form seen here allows for the usual features of cable theory to be reconstructed. In particular, standard analytic results for voltage propagation in complex dendritic structures and time-dependence have easy analogies in tapering cables. Greater accuracy can also be achieved, up to a point, by taking higher-order terms in ϵ. These results are shown in the Supporting Information.
An analytical expression for the voltage at leading order allows for study of the optimal dendritic radius profile to propagate synaptic currents towards the soma. Previous work in this direction lacked a continuous representation of the voltage profile and used numerical methods to explore optimality [9]. Calculus of variations provides a framework in which to define the optimal profile (for the leading-order component of the voltage) continuously.
Given a dendritic cable of length L with volume V and distal (minimal) radius rL, the goal is to maximise the voltage at the proximal end of a dendritic cable for synaptic currents arising at all points along the cable. This means maximising the functional
J=∫0L1λ72(x′)e−∫0x′1λ(s)dsdx′
(3)
where the effect of ‘reflected’ current at the distal end has been neglected due to the relatively fast time course of excitatory potentials.
The maximisation gives an optimal radius profile of (see Methods)
r(x)=α(L−x)2+rL
(4)
where α is fitted to match the volume of the cable V. This profile matches the results of numerical optimisation (Fig 2).
Having found the optimal single cable for voltage propagation, it remains to be shown how far real dendritic trees correspond to this optimality. Wilfrid Rall [4] showed that if the diameters of cylindrical sections at dendritic branch points satisfied the relationship dp3/2 = dc13/2+dc23/2, matching the conductance across the branch, then the entire dendritic tree could be collapsed to a single cylinder. Rall’s relationship is rarely satisfied in real dendrites [20,23,24]. Using a Rallian diameter ratio at a branch, however, allows us to ensure that the transition between parent and daughter branches obeys the quadratic optimality condition. This makes it possible to map quadratic radii onto complex dendritic morphologies by constraining dendrites to locally obey optimality (see Methods). The resulting predicted morphologies show how far dendritic trees are globally optimised to transmit and equalise current transfer.
We have selected a number of neuronal classes with a broad array of functions to examine the validity of our predictions (Fig 3A). It should be noted here that obtaining reliable measurements of dendritic radius is experimentally very challenging and this makes exact comparisons difficult. Different cell types satisfy the equivalent quadratic criterion to different degrees. Of the cell classes studied, the best agreement was for fly neurons, which might be considered genetically more hardwired [25,26]. In terms of mammalian neurons, the best agreement was found for dentate gyrus granule cells. These cells are known to both obey Wilfrid Rall’s branching criterion [27] and undergo continuous replacement throughout life [28]. These results suggest that our model might best match cells with a stereotypic morphology and therefore an initially optimal passive backbone.
The diameter profiles of apical and basal dendrites in cortical pyramidal cells match optimality to different degrees. The apical tree appears well described in terms of quadratic equivalent taper, despite differences at the trunk of the apical dendrite. As the apical dendrite might be more strongly specialised in propagating dendritic spikes, deviations might not be surprising. The predicted diameter profile for the basal dendrites was less accurate. Here there appear to be sections of the reconstruction that are much more voluminous than their length relative to other branches would suggest. This might imply that the relationship between nearby cells exerts a stronger influence than is seen elsewhere and that local cortical microcircuits display preferential connections in some directions.
No agreement was found for cerebellar Purkinje cells, where the general taper profile is much shallower than would be expected and dendrites often exhibit alternate bulges and narrower regions. The distinctive layered structure of the cerebellum means that excitatory synaptic inputs arrive in distinct locations, strong synapses from climbing fibres proximally and individually weaker, but much more numerous, synapses from parallel fibres distally. These two types of inputs are implicated in different spiking patterns, complex and simple spikes respectively, and the functional relationship between the two is beyond the scope of our general optimality principle.
Structurally, the agreement between ideal and observed morphologies therefore varies with specific function, but the model provides a good fit to large regions of many dendritic trees. We can, however, show how well the quadratic taper performs for all classes studied (Fig 3B). Plotting the current transfer from all nodes to the soma illustrates the advantages of quadratic taper against a constant diameter across the tree and provides a slight advantage over observed morphologies. Our results highlight the importance of a specific form of taper in maximising current transfer and equalising synaptic inputs.
Interestingly, for the dendrites where current transfer loss was largest because of either the size (the apical dendrite of the pyramidal cell) or because of a high membrane conductivity (as was the case in the fly neurons), the diameters tended to be better predicted by optimal current transfer. Where cells deviate substantially from passive optimality, for example specifically along the trunk of the apical dendrite of a pyramidal cell or across a Purkinje cell, there is evidence that these sections of dendrite favour functions other than the unidirectional propagation of excitatory synaptic currents towards the soma.
The fact that voltages in dendrites typically decay much more quickly than radii allows us to make a simple and accurate approximation to the propagation of currents across real dendritic trees. The compact form of the voltage approximation allows for a straightforward reproduction of the standard results of cable theory [4–6]. Further, this result allows the continuous optimum taper profile for transmitting synaptic currents to the soma to be deduced. The optimal radius profile tallies with notions of both dendritic democracy [11,12,29] and energy optimisation [9] and provides a close match to reconstructed dendritic morphologies across a range of cell classes.
Dendrites perform an array of non-linear computations involving active processes and local inhibition; the general principle of global passive optimality does not explain every facet of dendritic function, but does provide an important new intuition. The simple forms of both voltage and optimal radius link signal transmission and dendritic diameters, allowing a clearer intuitive understanding of the function of dendritic trees.
Consider the homogenous steady-state voltage equation for a cable with arbitrary radius r(x)
∂∂x[r2(x)∂v∂x]−2raglr(x)1+(r′(x))2v=0
(5)
with boundary conditions
dvdx|x=L=0 limx→−∞v=0
(6)
r typically changes more slowly as a function of x than v does, specifically r(x) = ρ(ϵx) for ϵ ≪ 1. S1 Fig shows typical values of ϵ for a range of reconstructed morphologies. It is possible to treat the ‘fast' voltage length variable x and the ‘slow' radius length variable ϵx as independent using the method of multiple scales. Then drdx = ϵdρdx and the steady-state voltage equation becomes
0=ρ2d2vdx2+2ϵρρ′dvdx−2raglρ1+(ϵρ′)2v
(7)
Introducing the new variable w such that w = ρϵv allows us to write the voltage equation as
0=d2wdx2−(2ragl1+(ϵρ′)2ρ+ϵ2(ρ2ρ‴−3ρρ′ρ″+2(ρ′)3)2ρ3+(ϵρ′)24ρ2)w0=d2wdx2−f(ϵx)w
(8)
Note that 1+(ϵρ')2≈1+(ϵρ')22 and that −f, the coefficient of w, will always be negative making the solution appropriately non-oscillatory. We seek solutions of the form
w(x)=μ(ϵx)e∫xσ(ϵs)ds
(9)
for μ and σ real. Substituting this into the above equation gives at first order
w(x)=ρ(x)1/4(2ragl)1/4[Ae∫x2raglρ(s)ds+Be−∫x2raglρ(s)ds]
(10)
for some constants A and B. Writing λ(x) = ρ(x)2ragl as the distance-dependent electrotonic length gives the leading-order form
v(x)≈λ(x)[Ae∫x1λ(s)ds+Be−∫x1λ(s)ds]
(11)
To determine the response to a current injection of magnitude Iapp at site x′, note that the Green's function g(x, x′) solves the equation
∂∂x[r2(x)∂g∂x]−2raglr(x)1+(r′(x))2g=δ(x−x′)
(12)
subject to a given set of boundary conditions. Away from x′, the solution is given by the homogenous voltage above, namely for x < x′
gx<x′(x,x′)=λ(x)B1e−∫x1λ(s)ds
(13)
using the fact that voltages are required to decay towards the soma. For x > x′
gx>x′(x)=λ(x)[A2e∫x1λ(s)ds+B2e−∫x1λ(s)ds]
(14)
Here, the sealed-end condition gives the relationship between the constants as
B2=A2(2+λ′(L)2−λ′(L))e2∫x′L1λ(s)ds
(15)
Continuity of voltage at x′ ensures
B1=A2(1+k)
(16)
for k the ratio between A2 and B2 given by the sealed end condition. Conservation of current at the point of injection relates all three constants
g′x<x′(x′)+raπρ2(x′)Iapp=g′x<x′(x′)B1(λ′(x′)−2)+2raλ(x′)πρ2(x′)=A2(λ′(x′)+2+k(λ′(x′)−2)))
(17)
giving the coefficients in terms of the initial parameters as
B1=raλ(x′)2πρ2(x′)[1+(2−λ′(L)2+λ′(L))e−2∫x′L1λ(s)ds]IappA2=raλ(x′)2πρ2(x′)[2−λ′(L)2+λ′(L)]e−2∫x′L1λ(s)dsIappB2=raλ(x′)2πρ2(x′)Iapp
(18)
Note that B1(x′) is the input resistance at site x′.
As we are primarily interested in voltage at the proximal terminal of the dendrite, we focus on the solution in the region x < x′ and evaluate the voltage at x = 0. The first-order approximation holds for a region of size ϵ−1 away from the site of current injection. Section 4 of the S1 Text describes how to extend this approximation to account for higher-order terms, which can allow for greater accuracy (S2 Fig), as well as voltage transients and voltage propagation in branched structures (S3 Fig).
It is possible to use calculus of variations to study the functions r(x) that give extremal values of a functional J[x, r, r′]. We seek to define the radius profile that maximises current transfer. In this case we seek to maximise the total current transfer to the proximal end x = 0, from all injection sites x′ = 0 to x′ = L, under constraints of fixed terminal radii or total cable volume. Writing the voltage at 0 due to current injection at x′ as v(0, x′) such that
v(0,x′)=raλ(x′)2πρ2(x′)[1+(2−λ′(L)2+λ′(L))e−2∫x′L1λ(s)ds]Iappλ(0)e−∫0x′1λ(s)ds
(19)
We seek to maximise the functional
J=∫0Lv(0,x′)dx′=∫0LKdx′
(20)
J is a functional of the functions λ(x) and ∫x1λ(s)ds. It is convenient to write Λ(x) = ∫x1λ(s)ds so that Λ'(x) = 1λ(x). For J to take a maximal or minimal value, it is necessary for the integrand K to satisfy the Euler-Lagrange equation
0=∂K∂Λ−ddx∂K∂Λ′
(21)
with boundary conditions following from the original constraints. Introducing the constants C1 = raλ(0)2π(2ragl)2 and C2 = 2-λ'(L)2+λ'(L)e-2∫0L1λ(s)ds allows us to write
K[Λ(x),Λ′(x)]=C1Λ′(x)72[e−Λ(x)+C2eΛ(x)]
(22)
The Euler-Lagrange equations give that J will not be maximised unless Λ satisfies
0=97[C2eΛ(x)−e−Λ(x)]−92Λ″(x)(Λ′(x))2[C2eΛ(x)+e−Λ(x)]
(23)
To solve this in terms of elementary functions we introduce a further assumption that current is injected sufficiently far from the distal end for the contribution of ‘reflected' current to the input resistance to be negligible (this applies more generally when considering responses to transient current injection). This assumption is equivalent to making C2eΛ(x) vanishingly small, giving the equation
Λ″(x)(Λ′(x))2=−27ddx[−1Λ′(x)]=−27
(24)
Using the definitions of Λ(x) and λ(x), and the boundary conditions gives (for a constant C3)
r(x)2ragl=±27x+C3r(x)=α(L−x)2+rL
(25)
where rL is the distal (minimal) radius and α is determined by matching volumes or proximal radii as required.
It should be noted that whilst the current transfer functional described here is one of a number of possible functionals to optimise, it provides a straightforward and robust description of dendritic function. Further, with temporally active conductance-based synapses, there will be a potential further attenuation of more distal inputs that is beyond the scope of this study.
The final comparison of optimal dendritic taper to real morphologies requires an algorithm for mapping a quadratic taper onto complex branched structures. In particular it requires a principled consideration of the way to distribute dendritic radius at branch points. We seek to equalise conductance at branch points using Rall's 3/2 power relationship; that for a parent radius r0, and daughter radii r1 and r2, then r03/2 = r13/2+r23/2. The ratio between r1 and r2 is defined by the lengths l1 and l2 of the two daughter branches such that r1/l13/2 = r2/l23/2. The two daughter branches appear to the parent branch to be a single branch with length l0 = (l13/2+l23/2)2/3. The algorithm for applying these principles to a real dendritic morphology with complex branching structure is described below.
Five cell classes are discussed in the paper, covering an array of functions and species. All morphologies are publicly available. Blowfly calliphora vicina HS (25 examples) and VS (30 examples) neuron morphologies are published with the TREES toolbox [18]. The passive parameters used are axial resistance ra = 60Ωcm and membrane conductance gl = 5 × 10−4S cm−2 for both. Mouse dentate gyrus granule cells (3 examples) are published on ModelDB (Accession no. 95960)[30]. The passive parameters used are ra = 210Ωcm and gl = 4 × 10−5S cm−2. Rat Purkinje cells (2 examples) are published on NeuroMorpho (IDs NMO_00891 and NMO_00892)[31], with ra = 150Ωcm and gl = 5 × 10−5S cm−2. Rat Layer V pyramidal cells (3 examples) are published on ModelDB (Accession no. 139653)[32], with ra = 150Ωcm and gl = 5 × 10−5S cm−2 for both basal and apical dendrites.
Simulations are carried out in MATLAB using the TREES toolbox package [18]. The numerical simulations in Figs 1, 3, S1 and S2 use standard functions described in the toolbox. The non-parametric numerical optimisation in Fig 2 follows an algorithm adapted from an earlier study [9]. The algorithm assigns radii to seven segments of a cable modelled using the TREES toolbox and uses the MATLAB function ‘fminsearch' to maximise the current transfer to the proximal end. This is repeated 50 times to produce a maximum over all trials. The radii of the six distal segments are fitted to a continuous quadratic equation ax2+bx+c (as described in [9]) to produce the numerical results of Fig 2.
A function to map an optimal radius profile onto an arbitrary dendritic morphology will be published in the TREES toolbox to accompany this paper.
|
10.1371/journal.ppat.1000297 | The Macrophage Scavenger Receptor A Is Host-Protective in Experimental Meningococcal Septicaemia | Macrophage Scavenger Receptor A (SR-A) is a major non-opsonic receptor for Neisseria meningitidis on mononuclear phagocytes in vitro, and the surface proteins NMB0278, NMB0667, and NMB1220 have been identified as ligands for SR-A. In this study we ascertain the in vivo role of SR-A in the recognition of N. meningitidis MC58 (serogroup B) in a murine model of meningococcal septicaemia. We infected wild-type and SR-A−/− animals intraperitoneally with N. meningitidis MC58 and monitored their health over a period of 50 hours. We also determined the levels of bacteraemia in the blood and spleen, and measured levels of the pro-inflammatory cytokine interleukin-6 (IL-6). The health of SR-A−/− animals deteriorated more rapidly, and they showed a 33% reduction in survival compared to wild-type animals. SR-A−/− animals consistently exhibited higher levels of bacteraemia and increased levels of IL-6, compared to wild-type animals. Subsequently, we constructed a bacterial mutant (MC58-278-1220) lacking two of the SR-A ligands, NMB0278 and NMB1220. Mutation of NMB0667 proved to be lethal. When mice were infected with the mutant bacteria MC58-278-1220, no significant differences could be observed in the health, survival, bacteraemia, and cytokine production between wild-type and SR-A−/− animals. Overall, mutant bacteria appeared to cause less severe symptoms of septicaemia, and a competitive index assay showed that higher levels of wild-type bacteria were recovered when animals were infected with a 1∶1 ratio of wild-type MC58 and mutant MC58-278-1220 bacteria. These data represent the first report of the protective role of SR-A, a macrophage-restricted, non-opsonic receptor, in meningococcal septicaemia in vivo, and the importance of the recognition of bacterial protein ligands, rather than lipopolysaccharide.
| Macrophages are innate immune cells that provide a first defence against infection. Several receptors on the surface of macrophages mediate recognition of invading pathogens, and one of these is the Macrophage Scavenger Receptor A (SR-A). SR-A recognises Neisseria meningitidis, a bacterium that causes meningitis and septic shock, via proteins on the surface of the bacterium. In this study we investigated the interaction of SR-A with N. meningitidis in a mouse model for septic shock, by infecting mice with N. meningitidis and comparing a mouse strain expressing SR-A with one that does not. The health of mice not expressing SR-A deteriorated more rapidly and fewer animals survived compared to those expressing SR-A. Mice lacking SR-A had higher numbers of bacteria in their blood and also produced more cytokines that can cause septic shock. We also infected mice with bacteria that did not express two of the proteins recognised by SR-A. In this case, no differences in survival, levels of bacteria, or cytokines were detected between animals that expressed SR-A and those that did not. Therefore, we show that the macrophage receptor SR-A is protective in the development of septic shock induced by N. meningitidis.
| The innate immune system is a first line of defence against invading pathogens and macrophages play an integral role in the innate immune defence against bacterial infection. This is based on the recognition of conserved microbial structures, termed pathogen-associated molecular patterns (PAMPs), by a range of pattern recognition receptors (PRRs). Macrophages (Mφ) express different classes of innate PRRs with diverse functions, including the phagocytic Scavenger receptors (SRs) [1]. One of the receptors belonging to this family is the class A scavenger receptor (SR-A), which has been shown to recognise a range of polyanionic molecules [2]. SR-A is a trimeric type II transmembrane glycoprotein consisting of a cytoplasmic tail, transmembrane domain, spacer region, α-helical coiled coil domain, collagenous domain and C-terminal cysteine-rich domain. SR-A expression is mostly restricted to macrophages and is not found on polymorphonuclear neutrophils or monocytes [3]. Recently, the expression of SR-A was shown on mast cells [4] and specific sub-populations of bone marrow-derived dendritic cells (DC) and splenic DC [5]. Selected endothelial cells and smooth muscle cells within atherosclerotic plaques also express SR-A [6].
The Scavenger receptors play an important role in microbial recognition and clearance, and SR-A has been shown to bind both Gram-positive and Gram-negative bacteria [7]. Hampton and colleagues first proposed that SR-A might be involved in antimicrobial host defence, based on their observation that SR-A could bind lipid A, an integral part of lipopolysaccharide (LPS) [8]. Subsequent studies showed that SR-A also recognises the Gram-positive cell-wall component lipoteichoic acid (LTA) [9]. Furthermore, SR-A binds to different LTA structures with varying specificity depending on their exposed negative surface charge. Unmethylated bacterial CpG DNA, another major immunostimulatory microbial product, is also recognised by SR-A [10]. Using a range of Gram-positive organisms, Dunne and co-workers confirmed that both soluble and cell-associated forms of SR-A are not only able to bind bacterial components, but can also recognise intact live organisms [11]. SR-A has been shown to play a role in both infection and inflammation. In vivo studies with three Gram-positive organisms have shown that SR-A−/− mice are more susceptible to infection. SR-A−/− animals exhibited deficient clearance of bacteria from the liver and spleen in experimental Listeria monocytogenes infection [12]. SR-A−/− mice also showed increased susceptibility to Staphylococcus aureus and Streptococcus pneumoniae infection [13],[14]. A possible anti-inflammatory host-protective role of SR-A was proposed by Haworth et al., who observed that SR-A−/− mice formed normal granulomas in response to BCG (Bacille Calmette-Guérin) priming [15]. However, these animals were more susceptible to endotoxic shock as a result of increased pro-inflammatory cytokine secretion in response to additional lipopolysaccharide (LPS) challenge. In addition, SR-A has been shown to modulate chemokine levels in specific acute inflammatory conditions to ensure an inflammatory response of the appropriate magnitude [16].
Neisseria meningitidis is a Gram-negative obligate commensal bacterium that colonises the human nasopharynx, however when the bacterium crosses this barrier, it causes meningitis and rapid septicaemia, particularly in young children and teenagers. We have shown previously that uptake of N. meningitidis by macrophages is mediated almost exclusively via SR-A [17]. Interestingly, experiments employing an N. meningitidis lpxA mutant revealed that recognition of N. meningitidis by SR-A was independent of lipopolysaccharide, and we identified three bacterial surface protein ligands for SR-A, namely NMB0278, NMB0667 and NMB1220 [18]. In this study we investigated the in vivo role of SR-A in inflammation in a murine meningococcal septicaemia model. We also ascertained the contribution of the identified surface protein ligands by constructing bacterial mutants in the SR-A ligands and examining the effects of a double mutant in vivo. We show that SR-A knock-out mice are more susceptible to septicaemia induced by N. meningitidis than wild-type mice, and that for double knock-out bacteria lacking two SR-A ligands, these effects are abrogated.
Unless otherwise stated, all chemicals were from Sigma (Poole, United Kingdom). Acetylated low density lipoprotein (AcLDL) and Rhodamine Green X (RdGnX) were obtained from Molecular Probes (Eugene, OR, USA). The TMB substrate reagent set was purchased from BD Biosciences Pharmingen (San Diego, CA). All culture media were from Gibco (Paisley, United Kingdom). The rat monoclonal anti-CD68 monoclonal antibody FA-11 was obtained from AbD Serotec (Kidlington, UK). The rat monoclonal antibody against the 7/4 murine differentiation antigen was generated in this laboratory [19]. M5114, the rat monoclonal antibody recognising murine MHC-II was obtained from R&D Systems (Abingdon, UK).
The Neisseria meningitidis strains used in this study are listed in Table 1. All strains were grown overnight at 37°C on brain-heart infusion (BHI) medium (Oxoid), supplemented with Levinthal's reagent (10% vol/vol) and solidified with agar (1% [wt/vol]; Bioconnections), in an atmosphere of 5% CO2. For selection of strains following transformation, kanamycin (75 µg ml−1) or erythromycin (6 µg ml−1) was added to the culture medium. Escherichia coli strain DH5α was used to propagate recombinant DNA constructs and was grown at 37°C on Luria-Bertani (LB) medium supplemented with kanamycin (50 µg ml−1), erythromycin (300 µg ml−1) or ampicillin (50 µg ml−1) where appropriate. For fluorescent labelling, N. meningitidis were fixed with 70% ethanol and labelled with RdGnX according to the manufacturer's instructions.
Recombinant DNA techniques were performed as described by Sambrook et al. [20]. Restriction endonuclease and DNA modifying enzymes were obtained from Boehringer Mannheim or New England Biolabs and used according to the manufacturers' instructions. Oligonucleotide primers were synthesised by Sigma-Genosys. Standard polymerase chain reaction (PCR) amplifications were performed in 50 µl reaction volumes (final concentrations: 20 mM Tris-HCl, pH 8.4; 50 mM KCl; 2.5 mM MgCl2; 0.4 µM forward primer; 0.4 µM reverse primer; 0.4 mM dNTPs) with 1.25 U Taq recombinant polymerase (Invitrogen, Paisley, UK) in a Master-Cycler (Eppendorf) gradient thermal cycler. Thirty cycles of PCR were performed, each consisting of 1 min denaturation at 94°C, 1 min annealing at typically 5°C below Tm and 1 min extension at 72°C, with a final prolonged extension of 10 min at 72°C. Chromosomal DNA was prepared from N. meningitidis strains as described previously [21]. A list of oligonucleotide primers used in this study is given in Table 2.
Outer membrane vesicles were prepared from N. meningitidis cells as described by Heckels and Williams et al [22],[23]. Briefly, cells were harvested from confluent growth of bacteria on 40 BHI plates into 0.2 M lithium acetate (40 ml, pH 5.8) and extracted at 45°C in the presence of 2 mm glass beads (20 ml). Live bacteria were removed by centrifugation at 13,00 g, 20 min, the supernatant was then subjected to a repeat of this step. The outer membranes were recovered by centrifugation at 11,000 g, 4°C, 2 h and the resuspended in 200 µl dH2O. The protein concentration of the outer membranes was determined by performing a Lowry MicroAssay (Sigma) according to the manufacturer's instructions. 15 µg of total protein was diluted in dH2O and loaded in each well. Whole cell lysates were prepared from N. meningitidis strains grown overnight, by harvesting and resuspending cells in PBS and then adding the equivalent amount of dissociation buffer (125 mM Tris, pH 6.8, 20% [v/v] glycerol, 3.9% [w/v] SDS, 10% β-mercaptoethanol, 0.04% [w/v] bromphenol blue). All samples were boiled at 100°C for 5 min. Samples were separated by tricine-sodium dodecyl sulphate-polyacrylamide gel electrophoresis (T-SDS-PAGE) using 16.5% gels run at 30 mA at 4°C for 18 hr [24], and were visualised by staining with silver nitrate according to the manufacturer's instructions (Amersham Biosciences).
To mutate the N. meningitidis NMB0278 gene, the gene was first amplified by PCR from strain H44/76 chromosomal DNA with oligonucleotide primers 278-f/278-r and cloned into the plasmid pT7Blue (Novagen). A kanamycin resistance (kanR) cassette was excised from pUC4-kan by digestion with HincII and inserted into the HincII site within the cloned NMB0278 gene. The resulting construct, pT7-278-kan was used to transform N. meningitidis strain MC58 as described previously [25]. Screening of transformants was performed by PCR using primers designed to bind within the kanR cassette and in the neighbouring gene to identify transformants containing a single, disrupted copy of NMB0278.
To mutate the N. meningitidis NMB0667 gene, the gene was first amplified by PCR from strain H44/76 chromosomal DNA with oligonucleotide primers 667-f/667-r and cloned into the plasmid pT7Blue (Novagen). An erythromycin resistance cassette (ermC) was excised from pER2 [26], by digestion with HincII and inserted into the SmaI site within the cloned NMB0667 gene. The resulting construct, pT7-667-ery, was used to transform N. meningitidis strain MC58. Transformants were screened by PCR using primers designed to bind within the ermC cassette and in the neighbouring gene to identify a transformant containing a single, disrupted copy of NMB0667.
To mutate the gene NMB1220, the 5′ and 3′ regions of the gene were first amplified by PCR from N. meningitidis strain MC58 chromosomal DNA, with oligonucleotide primers 1220-3f/1220-3r and 1220-5f/1220-5r, respectively, before cloning each product separately into pT7Blue, resulting in the plasmids pT7-1220-3 and pT7-1220-5. The ermC cassette was amplified from pER2 [26] with oligonucleotide primers ery-eag-f and ery-eag-r, digested with EagI and inserted into the EagI site present within the cloned 3′ region of the NMB1220 gene in pT7-1220-3 to give the construct pT7-1220-3-ery. The cloned 3′ region of NMB1220, interrupted with ermC, was excised from pT7-1220-3-ery with XbaI and BglII and inserted into the corresponding site within pT7-1220-5. The resulting construct, pT7-1220-ery, was used to transform N. meningitidis strain MC58. Transformants were screened by PCR using oligonucleotide primers 1220-f/1220-r designed to bind within NMB1220 to identify a transformant containing a single, disrupted copy of NMB1220.
Bacteria cultured on BHI plates were assayed in pooled human serum as described previously [27].
Bone marrow-derived macrophages (BMMφ) were prepared as described previously [7]. Mφ were routinely cultured in RPMI supplemented with 100 U/ml penicillin, 100 µg/ml streptomycin and 2 mM L-glutamine (PSG), 10% foetal calf serum (FCS) and 15% L-cell conditioned medium.
Bone marrow-derived macrophages were plated in 6-well bacteriological plastic dishes at a density of 1×106 Mφ per well 24 hours before use. Mφ were washed twice in PBS and then incubated in Opti-MEM medium (Invitrogen, Paisley, UK) containing fluorescently-labelled bacteria as specified. After incubation with bacteria, the culture medium was removed and the cells washed three times with PBS. Cells for flow cytometry were harvested with PBS containing 10 mM EDTA and 4 mg/ml lidocaine-HCl and fixed with 4% (v/v) formaldehyde in PBS. Fluorescence was analysed on a FACScan (Becton Dickinson, Mountain View, CA) using the FL-1 or FL-2 photomultiplier where appropriate and the results analysed with CellQuest software. Results are representative of at least 3 independent experiments.
A murine intraperitoneal challenge model for bacterial clearance was employed [28]. C57BL/6J wild-type mice and a corresponding SR-A−/− knock-out mouse strain were used. SR-A−/− animals were developed and bred onto C57BL/6J background using standard molecular biology techniques [12]. All animals were bred and housed under specific pathogen-free conditions.
Meningococcal strains were grown overnight at 37°C in 5% CO2 on BHI plates as described above. Muller Hinton broth (8 ml) supplemented with 0.25% (w/v) glucose in a 50 ml tissue culture flask was inoculated with 1.2×109 cfu from the overnight growth resulting in an initial OD620 of ∼0.1. The flask was incubated horizontally on a gently rocking platform at 37°C in 5% CO2 and bacteria were cultured to mid-logarithmic growth phase, defined as OD620 of ∼0.5 (approximately 2.5 h). The bacteria were transferred to 1.5 ml tubes and harvested by centrifugation at 1900 g for 5 min and then resuspended in PBS. The bacterial suspensions were adjusted to the required concentration for inoculation in BHI broth.
Bacterial doses of 1×105 cfu/mouse were injected intraperitoneally (i.p.) with human holo-transferrin (Sigma, 10 mg/mouse) in a total volume of 500 µl to groups of 6–8-week-old wild-type C57BL/6J and corresponding SR-A−/− knock-out inbred female mice. At the time of infection, the actual dose delivered to each group of mice was determined by serial dilution and replicate colony plating. At 18 h after the initial infection, mice were boosted i.p. with a further dose of human holo-transferrin (10 mg/mouse in 200 µl PBS). The health of the animals was monitored and scored at regular time points according to the symptoms presented as follows: Healthy = 5, ruffled fur = 4, sticky eyes = 3, ruffled fur and sticky eyes = 2, immobile = 1. As soon as immobile mice were detected they were humanely killed. Scores were then collated and averaged for each group at the various time points [29]; (A. Gorringe, personal communication). Survival curves were also plotted and statistical significance determined using the Log-rank (Mantel Cox) test. Blood samples from the tail vein (5 µl) were taken at 20 h post-infection and serial dilutions plated to determine bacteraemia. All dilutions were made with PBS. At termination, blood was collected by cardiac puncture and spleens removed. Half of each spleen was fixed in 2% paraformaldehyde in HEPES-buffered isotonic saline for immunohistochemical analysis. The remaining spleen was homogenized and serial dilutions of spleen and blood were plated to determine bacteraemia. The remaining blood was separated by centrifugation and the plasma collected and frozen at −80°C for later use. All procedures involving animals were conducted according to the requirements of the United Kingdom Home Office Animals (Scientific Procedures) Acts, 1986.
Fixed tissues were transferred to a solution of 0.1 M sodium phosphate buffer containing 20% sucrose, placed in Tissue-Tek OCT compound (VWR International Ltd., Lutterworth, UK) and snap-frozen in isopentane cooled by dry ice. Frozen sections were cut on a Leica cryostat (5 µm thick), collected onto 1.5% gelatinized slides, air dried for 1 hour and stored at −20°C. Sections were washed in 0.1% Triton X-100 and endogenous peroxidase activity was quenched by incubation in PBS containing 0.01 M glucose, 0.001 M sodium azide and 40 U glucose oxidase for 15 min at 37°C. 5% Normal rabbit serum was used as blocking agent for non-specific binding and avidin/biotin blocking agents (Vector Laboratories Ltd., Peterborough, UK) were employed according to the manufacturer's instructions. Sections were incubated for 60 min in the respective primary antibodies or isotype-matched controls, washed and incubated for 30 min with the respective affinity purified biotinylated secondary antibodies. Finally, sections were washed and incubated with the avidin-biotin peroxidase complex (ABC elite, Vector Laboratories Ltd., Peterborough, UK) for 30 min and staining visualised by incubation with 0.5 mg/ml diaminobenzidine (Polysciences Inc., Northampton, UK) and hydrogen peroxide in 10 mM imidazole. Sections were counterstained with 0.1% methyl green (Vector Laboratories Ltd. Peterborough, UK) and mounted in DPX (VWR International Ltd., Lutterworth, UK).
The concentration of interleukin-6 (IL-6) in the plasma of infected mice was determined using an OptEIA Mouse IL-6 ELISA set (BD Biosciences, San Diego) according to the manufacturer's instructions.
Groups of five 6–8-week-old female wild-type C57BL/6J and corresponding SR-A−/− animals were infected with 1×106 N. meningitidis MC58 (wild-type)+1×106 N. meningitidis MC58-278-1220 (mutant) i.p, along with 10 mg human holo-transferrin. The two bacterial strains were individually grown overnight as before. Serial dilutions of the inoculum were also plated onto both BHI medium and BHI medium supplemented with kanamycin (which selects for MC58-278-1220 mutant bacteria), in order to verify the dose and ratio of wild-type to mutant bacteria. A second dose of 10 mg human holo-transferrin was injected i.p. at 18 h. The health of the animals was monitored as before. At termination, blood was collected by cardiac puncture and spleens removed. Spleens were homogenized and serial dilutions of blood and spleen samples were plated on BHI medium and BHI medium supplemented with kanamycin. Enumeration of wild-type bacteria and mutant bacteria allowed for the determination of the CI ratio between wild-type and mutant bacteria using the following formula: CI = (wild-type output/mutant output)/(wild-type input/mutant input). The statistical significance of the results was determined using the paired student's t-test.
To investigate the importance of NMB0278, NMB0667 and NMB1220 for recognition of N. meningitidis by SR-A, plasmids were constructed containing each gene interrupted by insertion of a kanamycin or erythromycin resistance cassette. The constructs pT7-278-kan and pT7-667-ery resulted from a single PCR product from NMB0278 and NMB0667, respectively, cloned into pT7Blue and then interrupted with an antibiotic resistance cassette. It is interesting to note that when a similar approach was utilised with NMB1220, the initial pT7-1220 construct proved to be highly unstable and consequently a two step approach of cloning the 5′ and 3′ regions of the gene together with some flanking DNA was undertaken, which proved to be successful. The plasmids pT7-278-kan, pT7-667-ery and pT7-12220-ery were transformed into N. meningitidis strain MC58.
The serogroup B genome sequence contains two homologues of NMB0278, namely NMB0294 and NMB0407. To ensure that only NMB0278 had been disrupted, specific oligonucleotide primers were designed to bind within the kanR cassette and in the neighbouring gene to identify transformants containing a single, disrupted copy of NMB0278. This transformant was designated MC58-278.
The gene NMB0667 shows low levels of homology to a degenerate DNA methylase found elsewhere in the serogroup B genome (NMB1223), so primers designed to bind within the ermC cassette and in the neighbouring gene were utilised to identify transformants containing a single, disrupted copy of NMB0667. A number of transformants were screened using different combinations of oligonucleotide primers. For each transformant containing an interrupted copy of NMB0667, an intact copy of the gene was present in tandem, therefore we concluded that NMB0667 is an essential gene, and consequently we were unable to obtain a mutant neisserial strain deficient in this protein.
Double mutants where both NMB1220 and NMB0278 had been interrupted were constructed by transforming MC58-1220 with chromosomal DNA from MC58-278. The disruption of NMB0278 and NMB1220 was confirmed by the use of specific oligonucleotide primers as for the single mutants. This transformant was designated MC58-278-1220.
The growth of the MC58-278, MC58-1220 and MC58-278-1220 in vitro was compared to that of the parental strain MC58. The bacteria were grown in Muller-Hinton broth in the same manner as for the preparation of inoculum for the mouse infection studies, except that growth was followed over an 8 h period and the OD620 was measured throughout (Figure 1A). All mutants exhibited growth curve patterns indistinguishable from the parent strain. Outer membranes were prepared from the mutants and parental strain following growth on BHI plates overnight and separated by T-SDS-PAGE (Figure 1B). The resulting profiles show that no significant differences were observed in the protein and LPS profiles. A comparison of the parental strain MC58 and the mutant strains showed no difference in the killing effect in pooled normal human serum (data not shown).
In order to determine the distribution of these genes, a PCR screen was undertaken using primers specific not just for the gene under consideration, but also for its genomic location due to the homologous reading frames present for NMB0278 and NMB0667. 107 strains were screened using the following sets of primers; 278-out-f/278-out-r, 667-f/667-r, 667-f2/667-r and 1220-f/1220-r. This collection of strains was highly diverse and included representatives of disease and carriage isolates, along with well characterised reference strains. With the exception of one invasive disease strain, all gave a PCR product of the expected size for each of the three genes (refer to Table 3). The one anomalous strain demonstrated a PCR product for NMB0667 and NMB1220 but not for NMB0278. All three genes were determined to be present by this method in 7 N. gonorrhoeae isolates, however the distribution was found to vary considerably when other commensal species of Neisseria were analysed (data not shown).
To ascertain the in vitro macrophage uptake of wild-type bacteria compared to that of mutant bacteria, wild-type and SR-A−/− bone marrow-derived macrophages were incubated at 37°C for 2 h with ethanol-fixed fluorescently labelled N. meningitidis MC58 or the mutant bacteria with deletions in either NMB0278, NMB1220 or both (MC58-278, MC58-1220 and MC58-278-1220), respectively, at an MOI of 20∶1 (Figure 2). Association of bacteria with macrophages was measured by flow cytometry. All the bacterial strains were taken up by wild-type macrophages, whereas uptake was reduced by at least 70% in macrophages lacking SR-A. However, no differences could be detected in the association of the mutant bacteria with wild-type macrophages. This could be attributed to the fact that there are multiple ligands for SR-A on N. meningitidis [18] and that the gene encoding at least one known ligand, NMB0667, could not be deleted. In addition, the strains still contained homologues for NMB0278.
Since no differences could be detected between wild-type and mutant bacteria in their interaction with SR-A in vitro, we set up an in vivo murine septicaemia model to investigate the role of SR-A in the clearance of N. meningitidis. First, we tested several bacterial doses in wild-type and SR-A−/− animals to determine the intraperitoneal (i.p.) dose required to establish bacteraemia in the blood and spleen, without causing rapid death of the animals, so that they could be monitored over time. From these experiments we selected a dose of 1×105 cfu/mouse (data not shown). Groups of six age-matched female C57BL/6J wild-type and corresponding SR-A−/− knock-out mice were injected i.p. with 1×105 N. meningitidis MC58 cfu/mouse and 10 mg human holo-transferrin as a bacterial iron source. N. meningitidis requires iron for growth and is unable to sequester iron from murine transferrin [30],[31]. A second dose of human holo-transferrin was administered i.p. at 18 h to maintain available iron levels in the blood. Animals were monitored regularly over a period of 48 h for symptoms of septicaemia. Each individual was assigned a health score at each time point, according to the severity of symptoms (animals with a score of 5 were healthy, and a score of 1 was assigned when they were immobile, refer to materials and methods). Scores were then collated and averaged for the group, and the results plotted to provide a health curve (Figure 3A).
Although animals in both groups exhibited symptoms of septicaemia, wild-type animals remained healthier, particularly after the second injection of iron, which would support bacterial proliferation. Survival curves (Figure 3B) also show that all the SR-A−/− animals had died at 32 h, while 33% of the wild-type animals survived beyond 48 h and recovered to full health. Overall, SR-A−/− animals showed more rapid deterioration of health and died more quickly than did wild-type animals. This could be correlated with the observation that SR-A−/− animals had higher levels of bacteraemia in their blood at both 20 h and 48 h than did wild-type animals (Figure 3C). Although there seemed to be a trend towards higher bacterial levels in the spleens of SR-A−/− animals, the differences in bacteraemia between the two strains were not statistically significant. We also measured levels of interleukin-6 (IL-6) in plasma from blood collected at 48 h. SR-A−/− animals consistently showed significantly higher levels of the pro-inflammatory cytokine IL-6 than wild-type animals.
Spleen sections were also analysed by immunohistochemistry with three antibodies, FA-11 (an intracellular macrophage marker, staining macrosialin [CD68]), 7/4 Ag (a neutrophil and monocyte marker) and MHC-II (an antibody recognising the major histocompatibility class II molecule, a marker expressed on resident dendritic cells and activated macrophages). FA-11 (macrosialin) is a pan-macrophage marker, indicating the macrophage infiltration into the spleen. Staining with MHC-II shows that the macrophages are activated. The 7/4 Ag staining shows the infiltration of activated monocytes and polymorphonuclear neutrophils (Figure 4). Therefore all the splenic samples showed a high number of infiltrating activated macrophages and neutrophils.
Next, we tested the mutant bacteria lacking both SR-A ligands, NMB0278 and NMB1220, for their ability to establish septicaemia in wild-type and SR-A−/− animals. Groups of eleven age-matched female C57BL/6J wild-type and corresponding SR-A−/− knock-out mice were injected i.p. with 1×105 N. meningitidis MC58-278-1220 cfu/mouse and 10 mg human holo-transferrin as before. Animals were monitored regularly, assigned health scores and their survival plotted (Figure 5A and 5B). Animals in both groups exhibited symptoms of septicaemia, however differences observed between wild-type and SR-A−/− animals when infected with wild-type bacteria, were absent. Overall both mouse strains showed fewer symptoms of septicaemia and had a higher survival rate when infected with mutant MC58-278-1220 bacteria (Figure 5A and 5B). No statistically significant differences were observed in the levels of bacteraemia in either the blood or the spleen between wild-type and SR-A−/− animals (Figure 5C). IL-6 levels were also not statistically significantly different between wild-type and SR-A−/− animals and were lower overall than in animals infected with wild-type bacteria (Figure 5D). Therefore the SR-A-mediated effects observed when mice were infected with wild-type bacteria were abrogated for mutant bacteria lacking two SR-A ligands.
When MC58-278-1220 mutant bacteria were injected into wild-type and SR-A−/− mice, we observed not only the abrogation of the SR-A-mediated effect, but also that mutant bacteria seemed to cause less severe symptoms of septicaemia and animals had a higher survival rate. To test whether this was a bona fide observation, we employed a competition assay to determine whether wild-type bacteria would out-compete mutant bacteria in vivo. To obtain a bacterial count for wild-type and mutant bacteria in each case, we injected 1×106 wild-type MC58 and mutant MC58-278-1220 bacteria i.p. at a ratio of 1∶1. The respective bacterial numbers in blood and spleen were determined by replicate plating of serial dilutions on BHI medium and BHI medium supplemented with kanamycin, which would select for mutant bacteria carrying the kanamycin resistance cassette used to disrupt the gene encoding NMB0278. Table 4 shows the competitive indices in wild-type and SR-A−/− animals. A competitive index of 6.891 and 5.090 was obtained for bacterial counts from blood and spleen from wild-type animals, respectively. This indicates that more wild-type bacteria remained than did mutant bacteria. In SR-A−/− animals, a competitive index of 2.781 and 1.660 was obtained from blood and spleen, respectively.
In this study we evaluated the in vivo role of SR-A in meningococcal septicaemia induced by N. meningitidis MC58 (serogroup B), and ascertained the role of the N. meningitidis surface proteins, previously identified to be ligands for SR-A [18]. The neisserial surface protein ligands were NMB0278, NMB0667 and NMB1220. NMB1220 has been shown to be surface expressed and similar experiments showed the surface expression of NMB0278 and NMB0667 [32] and personal communication). The function of these proteins is unknown, however they show some homology to proteins identified in other bacterial species. Sequence analysis suggests that NMB0278 has homology to E. coli DsbA, which functions in disulphide bond formation [33]. Interestingly, the DsbA protein of Haemophilus influenzae, another obligate human pathogen colonising the nasopharynx, has recently been shown to be a virulence factor [34]. The C-terminus of NMB0667 has 20% homology with the ZipA protein from E. coli, which is involved in septum formation during cell division [35]. NMB1220 belongs to the stomatin/Mec-2 protein family, which are oligomeric lipid raft-associated integral proteins that regulate the function of ion channels and transporters. We constructed deletion mutants in N. meningitidis MC58 for all three proteins, respectively, and also generated a double mutant lacking NMB0278 and NMB1220. We employed several approaches to delete NMB0667, however this mutation proved lethal, which may be linked to the possible role of NMB0667 in septum formation during cell division. All the mutant strains showed no deficiency in growth characteristics or LPS and protein profile. Through PCR analysis, we showed that the genes encoding these proteins are present in a wide variety of neisserial strains, including invasive disease and carriage strains. Closely related homologues are also present in other bacteria, making them ideal PAMPs (pathogen-associated molecular patterns) and targets for PRRs. It should however be noted that since the genes encoding these proteins are also present in non-pathogenic bacteria and commensal strains, the term “PAMPs”, though commonly used in this context, could be considered a misnomer.
In vitro uptake of the mutant bacteria by bone marrow-derived macrophages from wild-type and SR-A−/− animals did not show any differences when compared to wild-type bacteria. This could be attributed to the fact that at least one other SR-A ligand, NMB0667, was present, along with two NMB0278 homologues, which could mediate recognition and uptake. Therefore, we studied the in vivo role of SR-A in a murine model for meningococcal septicaemia. We injected N. meningitidis MC58 into wild-type C57BL/6J and corresponding SR-A−/− mice and monitored the manifestation of symptoms of septicaemia (ruffled fur, sticky eyes), as well as the survival of the animals. The health of SR-A−/− animals consistently deteriorated more rapidly and there was a 33% difference in survival when compared to wild-type animals. Analysis of blood samples taken at 20 h post-infection and at termination showed that SR-A−/− animals had higher levels of bacteraemia and the pro-inflammatory cytokine, IL-6. Induction of IL-6 is commonly associated with meningococcal septicaemia in humans [36]. Although not statistically significant, spleen samples also indicated higher bacteraemia for SR-A−/− animals, and splenic sections revealed infiltration of activated macrophages, monocytes and neutrophils. Therefore, in this model, SR-A played an important role in clearance of N. meningitidis MC58. This is the first report of the in vivo importance of SR-A in a model using a Gram-negative pathogenic organism.
Subsequently, we injected the mutant bacteria lacking both NMB0278 and NMB1220 at the same dose i.p. into wild-type and SR-A−/− mice and monitored them as before. Interestingly, the differences observed between the mouse strains when infecting with wild-type bacteria were abrogated for mutant bacteria lacking the two SR-A ligands, and overall less severe symptoms were observed. Furthermore, the levels of bacteraemia and IL-6 between wild-type and SR-A−/− animals did not differ significantly. Therefore the surface proteins NMB0278 and NMB1220 are at least partially implicated in the SR-A-mediated effects observed for wild-type bacteria. We previously showed that NMB0278 and NMB1220 are also Toll-like receptor (TLR) agonists, and that SR-A was required for full activation of TLR pathways (Plüddemann et al. JII, in press). Although SR-A ligation does not directly mediate cytokine induction, the overall lower cytokine levels in animals infected with mutant bacteria could be linked to the absence of NMB0278 and NMB1220. The competitive index assay confirmed that mutant bacteria were cleared more readily, proving that this observation was not due to inter-experimental variations. These data indicate a role for SR-A in clearance of N. meningitidis and development of symptoms of septicaemia. Thus far, the development and progression of septicaemia has mainly been linked to LPS, however these data indicate an additional role of neisserial surface protein recognition in this process.
It is clear that the i.p. mouse challenge model described here does not represent the natural pathogenesis of neisserial disease, however it does model the overwhelming septicaemia that is characteristic of invasive meningococcal disease. Since SR-A is not expressed on polymorphonuclear neutrophils, our results signify an important role for macrophages and SR-A in the development and progression of meningococcal septicaemia.
|
10.1371/journal.pntd.0005334 | Unrecognized Emergence of Chikungunya Virus during a Zika Virus Outbreak in Salvador, Brazil | Chikungunya virus (CHIKV) entered Brazil in 2014, causing a large outbreak in Feira de Santana, state of Bahia. Although cases have been recorded in Salvador, the capital of Bahia, located ~100 km of Feira de Santana, CHIKV transmission has not been perceived to occur epidemically, largely contrasting with the Zika virus (ZIKV) outbreak and ensuing complications reaching the city in 2015.
This study aimed to determine the intensity of CHIKV transmission in Salvador between November 2014 and April 2016. Results of all the CHIKV laboratory tests performed in the public sector were obtained and the frequency of positivity was analyzed by epidemiological week. Of the 2,736 tests analyzed, 456 (16.7%) were positive. An increasing in the positivity rate was observed, starting in January/2015, and peaking at 68% in August, shortly after the exanthematous illness outbreak attributed to ZIKV.
Public health authorities and health professionals did not immediately detect the increase in CHIKV cases, likely because all the attention was directed to the ZIKV outbreak and ensuing complications. It is important that regions in the world that harbor arbovirus vectors and did not experience intense ZIKV and CHIKV transmission be prepared for the potential co-emergence of these two viruses.
| Since 2014, Brazil has experienced simultaneous transmission of dengue (DENV), chikungunya (CHIKV) and Zika virus (ZIKV), hampering clinical differentiation of infections due to the close manifestations caused by these viruses. Salvador, Brazil’s third largest city, was one of the cities most affected by those arboviral diseases. While a large acute exantematous illness (AEI) outbreak attributed to ZIKV has been recognized in the city, transmission of CHIKV appeared to have been much less intense. To obtain a comprehensible overview of the recent epidemiological situation, we investigated the proportion of chikungunya positive results among Salvador residents that were tested for CHIKV infections from November 2014 to April 2016. We demonstrated a high rate of CHIKV detection in serum samples during June-November 2015, shortly after an AEI outbreak attributed to ZIKV. The increase in CHIKV cases was not promptly detected by public health authorities and health professionals, likely because all the attention was directed to the ZIKV outbreak and its ensuing complications. It is important for regions that harbor Ae. aegypti (and potentially other vectors), but have not yet been subject to much transmission of ZIKV, CHIKV, DENV (and to potential other new and emerging arbovirus), to be prepared for the potential co-circulation of these viruses and associated diagnostic challenges.
| Chikungunya virus (CHIKV), an arbovirus transmitted by Aedes mosquitoes, can cause a clinical disease that resembles dengue and other arboviral infections [1]. Acute-phase signs and symptoms include fever, myalgia and rash, but severe arthralgia that may become chronic is the main hallmark of the disease [1]. An increase in Guillain-Barré Syndrome (GBS) was also observed during a CHIKV outbreak in French Polynesia, indicating that this arbovirus may be associated with severe neurological outcomes [2]. Clinical diagnosis is difficult, especially where co-circulation of other arbovirus such as dengue (DENV) and Zika (ZIKV) viruses occurs. In addition, as sequential arboviral infections and even co-infections could play a role in severe clinical manifestations, there is a need of a better understanding of multiple arboviruses transmission dynamics [3].
Brazil has been in the spotlight for arbovirus transmission, especially since epidemics of Zika virus (ZIKV) in early 2015 [4–6] were followed by outbreaks of GBS in adults and microcephaly in newborns [7,8]. During January-September 2016 (up to epidemiological week 37), Brazil recorded 200,465 ZIKV cases, 236,287 CHIKV cases and 1,438,624 DENV cases [9]. Oropouche and Mayaro virus infections have been identified sporadically in the country, restricted so far to the North and Midwest region [10–14], whereas Yellow Fever occurs endemically in the Amazon region with occasional transmission in the Midwest, South and Southeast regions, with 322 reported cases from July 2014 to June 2015 in Brazil [15]. Salvador, the largest city in the northeastern region of Brazil, and the capital of Bahia State, was one of the cities most affected by ZIKV. While there was widespread occurrence of ZIKV cases in Salvador, transmission of CHIKV appeared to have been much less intense, to the extent that an outbreak was not detected by local health authorities, as during the period an outbreak of acute exantematous illness (AEI) attributed to ZIKV occurred over 14,000 AEI cases were reported in contrast to 58 CHIKV cases reported in Salvador [6]. This is intriguing, since CHIKV has caused large outbreaks in most places where it is introduced [16–19], and CHIKV cases were first detected in Brazil in May 2014, in Feira de Santana, a city located approximately 100 km north of Salvador [20]. In Feira de Santana, CHIKV reached outbreak levels, with 4,088 reported cases in 2015 (an incidence of 668.0 cases/100,000 inhabitants) [21]. In contrast, in Salvador, 1,240 CHIKV cases were reported in 2015 (an incidence of 42.7 cases/100,000 inhabitants, more than an order of magnitude lower) [21].
However, CHIKV transmission in Salvador may have been masked by the ZIKV, GBS and microcephaly outbreaks, which prevailed in the media and got much of the attention of health professionals. Here we describe the results of an investigation aiming to determine the intensity of transmission of CHIKV in Salvador, Brazil, during the period of occurrence of ZIKV, GBS and microcephaly outbreaks.
In collaboration with the Salvador Secretary of Health and the State’s Central Laboratory of Public Health (LACEN-BA), we retrospectively analyzed all the serum sample results from Salvador patients that were tested for CHIKV at LACEN-BA between November 4, 2014 and April 19, 2016. Other than research and private laboratories, LACEN-BA is the sole public health laboratory in the State of Bahia capable of performing arbovirus diagnosis. It receives samples from patients suspected of arboviral disease from public health units throughout the state. The decision as to which patients’ serum samples are collected and sent for CHIKV testing is made by the attending physician, and follows the Brazilian guidelines for CHIKV suspicion, which is defined by sudden onset of fever (>38.5°C) and arthralgia or intense arthritis in residents or visitors of endemic or epidemic areas [22].
Serum samples collected through the fifth day of onset of symptoms were tested by CHIKV IgM ELISA [23] or by CHIKV RT-PCR [24], while those collected more than 5 days after onset were only tested using CHIKV IgM ELISA. CHIKV IgM ELISA testing was performed during the whole study period, whereas CHIKV RT-PCR was performed for samples from November 2014 to December 2015. Additionally, during the AEI outbreak attributed to ZIKV [6], samples sent to LACEN-BA due to AEI symptoms were also tested for CHIKV. We considered a sample positive for CHIKV if it tested positive by either ELISA or RT-PCR.
We constructed an epidemiological curve by epidemiological week of the date of serum collection, plotting the absolute and relative frequency of CHIKV detection. The percentage of positive samples was smoothed using a 5-week moving average, wherein the count of events for a given week was averaged with the values in the 2 previous and 2 following weeks to reduce week-to-week variation.
This investigation was performed using de-identified secondary laboratory data obtained by routine activities of the Epidemiological Surveillance Office/Municipal Secretariat of Health from Salvador, Bahia, Brazil. The study was approved by the Salvador Secretariat of Health and the Oswaldo Cruz Foundation Ethics Committee.
Of 3,042 serum samples from Salvador patients received by LACEN-BA for CHIKV testing during the study period, 2,656 (87%) were tested only using ELISA, 49 (2%) were tested only by RT-PCR, 31 (1%) were tested by both methods, and for 306 (10%) no result was available (either due to an insufficient sample or because it has not yet been tested). Samples that were not tested by at least one method (306) were excluded from analysis, resulting in a final count of 2,736 analyzed serum samples.
In total, 456 (16.7%) of the 2,736 samples analyzed during the study period were positive for CHIKV. Of the samples tested by RT-PCR 45% (36/80) were positive and of those tested by ELISA, 15.7% (422/2,687) were positive. The first laboratory-confirmed chikungunya case detected through this study occurred in week 3 (January 23) of 2015, and the positivity rate increased steadily until week 8 (February 22–28) (Fig 1). This trend was then interrupted and reversed, coinciding with the vast increase in number of samples tested, reaching more than 350 samples during week 18 (May 3–9) of 2015. The positivity rate then increased again, starting in week 20 (May 17), and peaking during weeks 26–47 (June 28-November 28) of 2015, when more than 25% (up to 68% at week 36, September 6–12, 2015) of the tested samples were positive for CHIKV. The percentage of CHIKV positive samples from Salvador remained at levels of ~10–20% through the rest of 2015 and the first weeks of 2016.
Additionally, during the study period a total of 3,328 samples were tested for DENV (from December 2014 to April 2016) using IgM or NS1 ELISA (Panbio Diagnostics, Brisbane, Australia), of which 599 were positive. Of the 2,736 samples tested for CHIKV, 1,126 were also tested for DENV (from December/2014 to April/2016) and 158 (14.0%) were positive. Concomitant positivity for DENV and CHIKV was found for 37 samples (36 CHIKV positivity by ELISA and 1 by RT-PCR). A total of 129 samples were tested for ZIKV (from April/2015 to April/2016) by real-time RT-PCR [25], four of which were positive. Laboratory testing for both CHIKV and ZIKV was performed for 58 samples (collected between the end of July 2015 and April 2016), none of which were positive for ZIKV (three of the 58 were positive for CHIKV by IgM ELISA).
The increase in the frequency of CHIKV positive laboratory results among Salvador patients during 2015 suggest that the intensity of CHIKV transmission in the city followed the same temporal pattern observed for the laboratory exams, with CHIKV transmission likely peaking in August, shortly after the exanthematous illness outbreak attributed by excess to ZIKV only [6]. Although Salvador established a surveillance for CHIKV detection following Feira de Santana’s outbreak in 2014 [26], the virus’ introduction and subsequent spread in the city was not promptly noticed by the health authorities, because their main focus was on the AEI outbreak attributed to ZIKV, which affected about 14,000 people over a two month period [7]. Additionally, due to the overwhelming demand, especially during the AEI outbreak, laboratory testing was not performed in a timely manner. Therefore, the health authorities were only informed of the increase of CHIKV cases retrospectively.
The interruption of the ascending trend in CHIKV positivity in weeks 8–20 corresponds to the peak of the AEI outbreak attributed to ZIKV in Salvador (April 19–May 23, 2015) [7]. This probably accounts for the increase in the number of samples tested, given that suspected AEI cases were also tested for CHIKV. Despite the high number of samples tested, the frequency of CHIKV detection was relatively low (<10.0%), supporting the hypothesis that the AEI outbreak was associated mostly with ZIKV infections [6]. Thus, the low frequency of CHIKV detection during the AEI outbreak period should be interpreted with caution, since testing likely included primarily patients with AEI, rather than CHIKV suspected cases. Unfortunately, we cannot distinguish samples from this period that were sent to be tested because of a clinical suspicion of CHIKV from samples that were tested because of AEI. It is likely that the increase in CHIKV cases in Salvador in 2015 started before the first cases of AEI were detected, but the epidemiological curve lagged because testing targeted mostly AEI patients during this period. This possibility is supported by the fact that the first laboratory confirmed CHIKV cases were from January 2015.
This finding also suggests that the CHIKV transmission in Salvador was less explosive than the 2015 ZIKV outbreak (over 17,000 reported cases in nearly 10 weeks) (7), but, in contrast, was of longer duration and may have resulted in established endemic transmission, given that the percentage of CHIKV positive samples from Salvador remained at levels of ~10–20% through the rest of 2015 and the first weeks of 2016. It is also possible that the CHIKV outbreak reported here is under-estimated, while the ZIKV outbreak is over-estimated (i.e., all severe manifestations observed in Salvador were attributed almost entirely to ZIKV circulation). Even though the number of people infected by both viruses was certainly under-estimated given how surveillance of cases was assembled, health-seeking behavior and the general perception that AEI was a self-limited mild disease.
Salvador was one of the epicenters of ZIKV, GBS and microcephaly outbreaks in Brazil during 2015. A causal relation between ZIKV and the congenital disorders outbreaks has been established [27,28]. In French Polynesia, a case-control investigation also pointed to a link between prior ZIKV infection and GBS development [29], supporting a relation between the outbreaks of ZIKV and GBS in Brazil. However, CHIKV has also been previously related with GBS in both French Polynesia and Réunion Island [2,30]. Thus, our findings of intense CHIKV transmission in Salvador between June and November co-occurring with the period of the GBS outbreak in the city [7], support a possible connection between CHIKV infections and GBS development in Salvador.
Some limitations need to be acknowledged. First, the majority of the samples were tested only using IgM-based serology, and thus cross-reaction to other alphavirus has to be considered. However, although the occurrence of other alphavirus such as Mayaro have been described in the North and Midwest regions of Brazil [10,12], there is no evidence for their circulation in Salvador. Second, our epidemiological curve is based on the time when a sample was taken from the patient, which might not necessarily represent the time of infection, especially as CHIKV infections may result in chronic clinical manifestations and serum samples may have been collected for diagnosis a long time after disease onset. In this case, IgM antibodies would no longer be present and IgG-ELISA would be more appropriate. Also, although RT-PCR for acute-phase samples and IgM detection in paired samples would provide a more accurate diagnosis [31], a different algorithm for CHIKV testing was adopted due to limited resources. Third, this study included only cases that sought healthcare and whose attending physician requested laboratory testing for either CHIKV or differential diagnosis of an AEI, thus underestimating CHIKV cases. Additionally, with syndromic surveillance it is not possible to define accurately the etiology of cases, therefore laboratory testing is essential. In this study, we tried to address this limitation by analyzing the available laboratory results for all patients tested for CHIKV in Salvador, and our results are supported by previously published data showing that the AEI outbreak in Salvador that peaked in May was mainly due to ZIKV [6]. Yet, community-based studies using serological tests are needed to help better ascertain the intensity of CHIKV transmission, and there is an urgent need for ZIKV serological tests to accurately assess the intensity of ZIKV transmission. Fourth, CHIKV outbreaks in Feira de Santana appear to have occurred in two waves, the first in June-December 2014 and the second starting at January-2015 [32]. Since the first patients from Salvador to be tested for CHIKV infection were not tested until November 2014, we might have missed any earlier CHIKV transmission in Salvador during the first wave of transmission in Feira de Santana. Lastly, both viral isolation and genome sequencing are not routinely performed by LACEN-BA; thus, detailed information on strains responsible for this outbreak was not available. However, other studies have identified CHIKV infections in Bahia associated with the East-Central-South African (ECSA) strain [32,33].
Our findings reinforce the need for a better understanding of the co-circulation of these arboviruses. In such a setting, with high intensity of transmission of more than one arbovirus, co-infections may be common, as these viruses have the same vector, future studies are needed to better understand the role of sequential and co-infections in the severity of clinical manifestations. Failing of detect the co-circulation of other arbovirus in a timely fashion hampers the ability to implement actions to prevent and treat severe or chronic manifestations that may elapse, such as incapacitating chronic arthralgia in the case of CHIV infections, for example. In addition, the unrecognized co-circulation of other arboviruses could partially explain the occurrence of other severe outcomes in the region, such as GBS. Even with regard to microcephaly, Brazil authorities are now set to explore the country's peculiar distribution of Zika-related microcephaly [34]. The concentration of such cases in the Northeast, where all three arboviruses have been co-circulating needs to be considered (together with other risk factors). As co-circulation of arboviruses is likely occurring in several other tropical cities, researchers, physicians, and public health professionals must consider CHIKV as a differential diagnosis together with DENV and ZIKV when studying arbovirus transmission and disease, while examining suspected case patients and when performing surveillance in Brazil and elsewhere.
Clinical differential diagnosis between these arboviruses is difficult, as observed by Salvador’s experience during the AEI outbreak when ZIKV, DENV and CHIKV were co-circulating [6], are three capable of causing AEI. In such a scenario, syndromic surveillance can provide a rough estimate of disease transmission, but a syndromic laboratory assessment approach testing all patients with non-specific arboviral disease symptoms for DENV, ZIKV and CHIKV (such as multiplex testing), should be considered, if possible.
|
10.1371/journal.pcbi.1002737 | Next-Generation Sequencing of Human Mitochondrial Reference Genomes Uncovers High Heteroplasmy Frequency | We describe methods for rapid sequencing of the entire human mitochondrial genome (mtgenome), which involve long-range PCR for specific amplification of the mtgenome, pyrosequencing, quantitative mapping of sequence reads to identify sequence variants and heteroplasmy, as well as de novo sequence assembly. These methods have been used to study 40 publicly available HapMap samples of European (CEU) and African (YRI) ancestry to demonstrate a sequencing error rate <5.63×10−4, nucleotide diversity of 1.6×10−3 for CEU and 3.7×10−3 for YRI, patterns of sequence variation consistent with earlier studies, but a higher rate of heteroplasmy varying between 10% and 50%. These results demonstrate that next-generation sequencing technologies allow interrogation of the mitochondrial genome in greater depth than previously possible which may be of value in biology and medicine.
| This manuscript details a novel algorithm to evaluate high-throughput DNA sequence data from whole mitochondrial genomes purified from genomic DNA, which also contains multiple fragmented nuclear copies of mtgenomes (numts). 40 samples were selected from 2 distinct reference (HapMap) populations of African (YRI) and European (CEU) origin. While previous technologies did not allow the assessment of individual mitochondrial molecules, next-generation sequencing technology is an excellent tool for obtaining the mtgenome sequence and its heteroplasmic sites rapidly and accurately through deep coverage of the genome. The computational techniques presented optimize reference-based alignments and introduce a new de novo assembly method. An important contribution of our study was obtaining high accuracy of the resulting called bases that we accomplished by quantitative filtering of reads that were error prone. In addition, several sites were experimentally validated and our method has a strong correlation (R2 = 0.96) with the NIST standard reference sample for heteroplasmy. Overall, our findings indicate that one can now confidently genotype mtDNA variants using next-generation sequencing data and reveal low levels of heteroplasmy (>10%). Beyond enriching our understanding and pathology of certain diseases, this development could be considered as a prelude to sequence-based individualized medicine for the mtgenome.
| The first complete human ‘genome’ sequenced was that of the mitochondrion in 1981 [1]. Since then, over 8,250 complete human and 3,220 complete non-human vertebrate mitochondrial genomes have been sequenced (http://www.ncbi.nlm.nih.gov). These contributions have come from numerous laboratories, where obtaining the complete sequence of even the ∼16.5 kb circular mitochondrial genome has been labor intensive and expensive. As an exemplar, it would be desirable to obtain this sequence on tens of thousands of samples in a simple, inexpensive, yet accurate manner. Beyond enriching many aspects of human biology, this development could be considered as a prelude, or even as a prerequisite, to sequence-based individualized medicine. Indeed, the mitochondrial genome, despite its unique structure and function, is an excellent ‘model system’ to identify and solve the technical, biological and medical problems that genomic medicine will encounter.
The mitochondrial genome (mtgenome) has multiple attractive structural and functional features. First, it is small at 16,569 bp (revised Cambridge Reference Sequence, rCRS) [2]. Second, it is divided into a small (6.8%) non-coding displacement loop (D-loop) or control region which provides the origin for mtDNA replication, and a large (93.2%) coding region compactly housing 37 genes (22 tRNAs, 13 proteins and 2 rRNAs) that encode proteins critical to the electron transport chain [1]. The unique biochemical functions of the mitochondria and its high functional content suggest that a higher fraction of mitochondrial, as compared to nuclear, mutations is likely to be functionally deleterious and have distinct phenotypes. Consequently, we have an enhanced possibility of understanding the logic of how sequence variation affects biochemical functions and organismal phenotypes. Third, depending on cell type, each cell contains hundreds or more of mitochondria, each mitochondrion harboring 2–10 genomes. Thus, the functional consequences of mtgenome variation acutely depend on the tissue, and are thus a model for all genes.
Genetic variation in the mtgenome has been critical to demonstrating its unique features of matrilineal inheritance [3], [4], lack of recombination [5], higher variability than the nuclear genome [6], [7] and hypervariability within the D-loop as compared to the rest of the mtgenome [8], [9]. These features have allowed delineation of mitochondrial haplotypes and haplogroups along maternal lines of descent in different human populations, and greatly contributed to our current understanding of human population structure and evolution. In turn, mitochondrial haplogroups have become a marker of an individual's ancestry. A surprising aspect of the mitochondrial genome has been its unusually large impact on human disease given its small size, owing to its high coding ratio and high mutation rate. The impact of mutations in the mtgenome on tissues with high-energy needs, such as muscle, has long been recognized in genetic disorders such as myoclonus epilepsy with ragged red fibers (MERRF) and Leber's hereditary optic neuropathy (LHON) [10], [11]. More broadly, mutations in the mtgenome have been identified in, or associated with, many complex disorders such as cancer, cardiovascular disease, neurodegeneration, diabetes and hearing loss [10], [12]–[16]; accumulation of mutations in the mitochondrial genome is a natural part of aging [17], [18] and the development of tumors as well [12]. Therefore, improved methods to sequence the mtgenome are of value to both biology and medicine.
The 100–1,000-fold higher mutation rate in mitochondria, as compared to the nuclear genome, is owing to the lack of a DNA repair system within the organelle [19]. Thus, alterations in the mtgenome sequence occur frequently, visualized as two or more mitochondrial genomes of different sequence within a single human. Such ‘heteroplasmy’ has long been considered rare but it is one major explanation for the variation in phenotypes between maternally related individuals with a deleterious mitochondrial mutation since different individuals within the same maternal lineage may harbor different proportions of wildtype to mutant mitochondria. However, strictly on theoretical grounds, heteroplasmy must be common since each oocyte has multiple mitochondria, as compared to the single nuclear genome. Therefore, any new mutation has a significant probability of being lost through mitochondrial segregation in the daughter cells after fertilization (mitochondrial “drift”) and needs to be balanced by additional mutations to allow variation. This may be a second reason for the higher mitochondrial mutation rate observed through heteroplasmy in all tissues. Indeed, some have proposed that, under the “mutation-drift-selection” scenario, heteroplasmy should be the default state for mtDNA in all tissues of the body from mitochondrial segregation of inherited variation or from somatic mutation [20]. Indeed, all extant mitochondrial polymorphisms must have gone through a heteroplasmic state after their origin by mutation.
A number of studies have demonstrated heteroplasmy, but its mechanism and incidence in the general population remains unknown since the detection of heteroplasmy has been hindered by the resolution of available sequencing technologies. While Sanger sequencing allows for complete coverage of the mtgenome, it is limited by the lack of deep coverage and low sensitivity for heteroplasmy detection when it is much less than 50% [21]. The Affymetrix Mitochip Array 2.0 containing the full mtDNA sense and antisense sequences tiled on an array has been successfully used in our laboratory for full mtgenome sequencing with slightly improved heteroplasmy detection [22], [23]. However, neither of these technologies allows the assessment of individual mitochondrial molecules. In contrast, next generation sequencing technology is an excellent tool for obtaining the mtgenome sequence and its heteroplasmic sites rapidly and accurately since it allows deep coverage of the genome through multiple independent sequence reads. In fact, two recent studies demonstrate that the degree of heteroplasmy can vary across an order of magnitude (typically <5% but occasionally >50%) [24] and multiple sites with the mtgenome have heteroplasmy rates >10% [25].
In this study, we present the complete mitochondrial genomic sequence and heteroplasmic status of 40 samples from the International HapMap Project [26] using the next-generation 454 GS FLX pyrosequencing platform. The samples include 20 individuals from the CEU (European ancestry) and 20 individuals from the YRI (African ancestry) reference panels; these are mtgenome sequences isolated without any contamination from nuclear embedded numts (see results) and from publicly available reference samples. The availability of such reference samples is critical as the samples could serve as a basis for reproducing and benchmarking new sequencing technologies. To enable analyses, we developed novel sequence processing and analysis algorithms, both for mapping against the reference sequence and for de novo assembly, for confident determination of the mitochondrial sequence. Our analyses demonstrate sequence accuracy of near 100%, nucleotide diversity of 1.6×10−3 for CEU and 3.7×10−3 for YRI, patterns of sequence variation consistent with earlier studies, but a high rate of heteroplasmy varying between 10% and 50%.
Twenty-two unique CEU (European ancestry) and twenty-two unique YRI (African ancestry) samples from the International HapMap Project [26], including two sets of duplicates for each population (CEU: NA10851, NA10856; YRI: NA18500 & NA18503), were sequenced. The DNA used was enriched for mitochondrial sequences by long range PCR (LPCR) of three ∼5–6 kb segments using mtgenome-specific primers. Although mitochondrial sequencing using total cellular DNA is possible and easy, and is being routinely performed with heteroplasmy detection [27][28], we avoided this approach because the human nuclear genome has >1,200 non-functional mtgenome fragments (numts) [29] and mitochondrial pseudogenes that complicate mtgenome sequence assembly and introduces numerous polymorphism and heteroplasmic artifacts. Thus, despite its simplicity it is quite erroneous, as we will demonstrate. LPCR reduced this possibility greatly since <5% of insertion sites are >5 kb. Additionally, our primers are designed to avoid nuclear genome amplification; each primer set is specific for the mtgenome as verified by BLAST (refer to methods). We completed sequencing using the 454 GS FLX system by pooling 12 individually tagged samples into each lane of a 4-region gasket PicoTiterPlate (PTP). Two YRI samples (NA19209 and NA19116) were discarded from analysis as both samples showed only one of three amplicons with an unusually high number of sites containing two different nucleotides at high frequencies; this could have arisen from a sample mixture. In addition, two CEU samples (NA12750 and NA12872) were removed due to suspected mislabeling. The results presented are from the remaining 40 samples. On average, each sample had 10,554 reads with a standard deviation of 2,652 reads. The read length distributions were similar and consistent across all samples; the distribution across all 44 samples (including duplicates) show read lengths across a wide range but 93.7% of them are between 200–300 bp. The average read length was 250 bp with a standard deviation of 36 bp (Supporting Figure S1) so that the yield per sequencing run was ∼2.6 megabases (mb).
Our approach for obtaining the mtgenome sequence was to map quality filtered reads against the reference sequence (rCRS) to identify homoplasmic and heteroplasmic variant sites. We also introduce a novel method for de novo assembly of the reads into a circular genome. An important consideration in our study was to obtain high accuracy of the resulting called bases. We accomplished this by quantitative filtering of reads that were error prone. We finally estimated an accuracy of the resulting sequence and an analysis of its genetic features.
The overall quality of the data is summarized in Figure 1. It portrays normalized coverage and the 0-centered ratio of forward/reverse reads at each position of the mtgenome. The average coverage across all 40 samples in YRI and CEU was ∼120-fold. However, the total number of reads varied per sample so that we normalized coverage by a sample's total number of reads. Second, we assessed the directionality bias in the reads by computing ρ = (r−1)/(r+1) where r is the ratio of forward to reverse reads at a position. We present data on normalized coverage and read ratio as an average across the 20 samples for each population, YRI and CEU respectively. This is displayed along the mitochondrial genome (Figure 1) as a function of local GC-content, calculated using a sliding window of length 51 bp (25 bp before and after each position) across the circular genome. The figure also illustrates where the D-Loop and amplicons lie along the mitochondrial genome. As can be seen, the average coverage falls and the read ratio spikes prior to the PCR amplicon overlap regions in both populations. However, ρ fluctuations are not due to variations in GC content.
The human mitochondrial genome can be sequenced at very high accuracy and rapidly using next generation sequencing technology as we, in this study, and other recent studies [24], [25], have shown. All of these studies have in common that they have uncovered patterns of sequence variation as has been described before but quantified the novel finding of a high rate of heteroplasmy in multiple individuals and across the mtgenome. Our study, however, has made three additional and important contributions. First, we have sequenced widely and publicly available biological samples so that our experiments can be replicated and provide a basis for future benchmarking and technology comparisons. Second, our methodology for variant and heteroplasmy detection is quantitative and parametric so that the method can be further optimized with additional experiments and new data. Third, we have developed a method for de novo sequence assembly of the mitochondrial circular genome with an internal test of sequence accuracy (identity of antegrade and retrograde assembly along a circular genome).
Each of the above developments is significant for understanding mitochondrial biology and medicine. First, DNA sequencing technology is advancing and new platforms that include single-molecule sequencing are on the horizon [35]. The availability of multiple sequencing methods on publicly available biological samples, such as those we have used, is the only certain way for comparing different technologies and their relative advantages and disadvantages. Second, we believe that the parameters we have used for identifying variants and heteroplasmy will need to be varied depending on the specific technology used and its features such as directional bias, read accuracy, difficulty in reading through homopolymeric tracts and coverage. Consequently, our approach is general and generalizable. Third, mapping reads against a reference suffers from the disadvantage of not being able to confidently identify insertions or inversions. The de novo methods we have introduced can rectify this deficiency particularly since our preliminary exploration of 40 sequences suggests that it produces high-quality assemblies.
The problems associated with recovery of target mitochondrial DNA from a biological sample, its DNA sequencing using short reads, the assembly of these reads into an mtgenome and its interpretation of variation and heteroplasmy are invariably confounded. We chose to recover the mtgenome in each individual by three distinct long-range PCR segments, analogous to Li et al. (2010) and in contrast to He et al. (2010). Our primers are designed to specifically target mtDNA and avoid introducing any artifacts from the numerous mitochondrial fragments (numts) in the nuclear human genome. Even if there is indeed some contamination from numts, this effect is expected to be small since it is assumed that there are many more copies of the entire mtgenome than two numts copies per the >1,200 autosomal insertion sites. However, specific fragments are present in >100 copies and can, and do, get amplified [29]. We expect that single molecule sequencing will reduce or eliminate this potential technical artifact. It is currently popular to extract and assemble the mitochondrial genome from whole genome sequencing of total cellular DNA [27]; Picardi and Pesole (2012) have recently done so from off-target exome sequencing data. But, these latter authors also show that ∼1% of all reads map to the mtgenome and not to known numts! Consequently, extensive filtering may be necessary to derive the mtgenome but this might also lose the genome-specific features including heteroplasmic sites. In other words, comparison of our data with those of others needs to consider how the mt DNA was isolated in the first place.
In this study, we have made no attempt to estimate the cost of sequencing a single mtgenome in any accurate way. In any case, we have demonstrated that we can obtain such sequence rapidly and with an error rate <5.63×10−4. Our crude estimate is that each sequence can be obtained for ∼$50 at high throughput much of this cost being the cost of mt DNA recovery. If so, studies of an entire cohort of individuals who have been measured for numerous medically relevant traits and are being followed for disease outcomes would be an ideal pilot experiment for individualized medicine.
Forty-four reference DNA samples of unrelated individuals from the International HapMap project were studied using 454 pyrosequencing technology. The samples included 22 Yoruba samples from Nigeria (YRI: NA18500, NA18503, NA18506, NA18516, NA18523, NA18852, NA18855, NA18858, NA18861, NA18870, NA18912, NA19092, NA19101, NA19116, NA19137, NA19140, NA19152, NA19159, NA19171, NA19200, NA19203 & NA19209) and 22 Utah residents of European ancestry (CEU: NA06993, NA06994, NA07019, NA10851, NA10854, NA10856, NA10863, NA11831, NA11881, NA11882, NA11995, NA12004, NA12005, NA12144, NA12145, NA12146, NA12156, NA12248, NA12750, NA12760, NA12872 & NA12891), four of which were studied in duplicate (NA18500 and NA18503 from YRI; NA10851 and NA10856 from CEU). Additionally, four of these samples were sequenced using Sanger sequencing and the Affymetrix Mitochip Array 2.0 (NA06994, NA12146, NA18516, and NA18523) for comparison. We also evaluated the Standard Reference Material (SRM) 2394 developed by the National Institute of Standards and Technology (NIST). These are a set of eight mixtures (mass percentages of 1%, 2.5%, 5%, 10%, 20%, 30%, 40%, and 50%) of two 285 bp mitochondrial amplicons that differ in sequence by only one nucleotide and is obtained from two different human cell lines. After QC checks that detected sample contamination, data from NA19209, NA19116, NA12750 and NA12872 were dropped from further analysis.
For pyrosequencing, we enriched for the mitochondrial genomic DNA by long range PCR (∼5–6 Kb) for three overlapping amplicons using high-fidelity TaKaRa LA Taq (TaKaRa Biomedicals) in 50 µl reactions (50 ng gDNA, 1× LA PCR buffer, 0.3 µM of each primer, 400 µM dNTPs, 2.5 U LA Taq). The primer sequences used were those described in Maitra et al (2004). Each primer set was blasted against the entire human genome to verify that there was no nuclear genome amplification. In silico PCR also confirmed no nuclear genome targets amplification by any of the three distinct primer sets. The success of the amplification reaction was checked by gel electrophoresis. The PCR products were then cleaned using the QIAquick PCR purification kit (QIAGEN) following the column purification protocol and the DNA was eluted in 30 µl of Elution Buffer to obtain a higher concentration. The actual concentration was determined using the Quant-iT PicoGreen dsDNA kit (Invitrogen). To obtain a uniform representation of the entire mtgenome, the amplicons were pooled in equimolar amounts (amount per amplicon [ng] = fraction of total x total amount needed). Since the pyrosequencing protocol required more than 5 µg of total DNA at a concentration of 300 ng/µl we performed at least two PCR reactions per amplicon. After pooling the three amplicons per reference sample in equimolar amounts, the samples were run through a QIAquick purification column to concentrate the pool to the desired 300 ng/µl concentration.
For Sanger sequencing, the mtgenome was amplified in 24 overlapping PCR fragments (800–900 bp) as described in Rieder et al 1998. For easy detection during sequencing, M13 tags were added to all forward and reverse primer sets. PCR reactions and cycling conditions were optimized across all primer sets and used 1× PCR Buffer, 200 µM dNTP, 0.5 U Taq2000, 10 ng DNA, and 0.5 µM of each primer. Confirmation of the reactions' specificity was assessed by 2% agarose gel electrophoresis. The final concentration of each amplicon was determined using the Quant-iT PicoGreen dsDNA kit (Invitrogen).
All sequencing using Sanger chemistry were performed by a commercial entity (Agencourt) for each individual PCR product on an automated ABI3730xl platform using a concentration of 15–25 ng/µl in 30 µl of TE buffer; individual sequence traces were provided. The Sanger sequence for each sample was assembled and analyzed in the SeqManII program from the DNASTAR Lasergene® v.7.0 analysis software suite. All sequencing reads for an individual sample were imported and assembled into one contiguous consensus sequence by aligning them to the revised Cambridge Reference Mitochondrial Sequence (rCRS). The variant bases for each sample were determined and used as the genotype for that sample for further analysis. Peak intensities for each sequence variant identified by the program were manually reviewed.
For pyrosequencing of the 48 samples, including duplicates, we pooled the pooled long range PCR products per sample in four batches of 12 each using 454's Multiplex Identifiers (MID) that are molecular barcodes that serve as unique tags to identify each sample post-sequencing. These mitochondrial DNA pools were sequenced on a 4-gasket PicoTiterPlate (PTP) using the GS FLX sequencing system. Standard emPCR and sample preparation were followed as recommended by the manufacturer (Roche Inc.)
As an additional precaution against misalignments, we developed an improved version of the BLAST algorithm. BLASTN uses an affine gap costs model and allows control of gap opening, gap extension and mismatch penalties and are particularly problematic for homopolymer stretches due to undercalls and overcalls. To accurately align these reads against a reference sequence, we needed an aligner that adjusts the gap penalties depending on the presence and length of the homopolymer sequence. The standard Smith-Waterman algorithm for aligning two sequences can be extended to handle these situations as follows. Let c(n,m) be the penalty for a n-length homopolymeric stretch of the reference appearing as an m-length stretch in the read. Then, the dynamic programming algorithm was modified to consult the c matrix also when computing the optimal alignment of the sequences. The entries of the c(n,m) matrix needed to be defined heuristically. In the current study, we set c(n,m) such that in homopolymer stretches of length ≥5, two gaps were ignored and the remaining penalized using the standard affine gap penalty model of BLASTN. In homopolymeric stretches of length 4, one gap was ignored. Since the largest homopolymeric stretch in the mitochondrial sequence is only 8 bases long, these values in the c(n,m) table were sufficient to yield good results. For performance reasons, we carried out alignment first using BLAST. Portions of the resultant alignment that were likely to benefit from our homopolymer-aware aligner were identified and refined using a Perl implementation of the model described above.
We developed an independent de novo assembly of each mtgenome. In our approach, we initially populate a database comprising all unique n-mers (n = 27 here) and the frequency of each n-mer in the raw read data. To populate the database we slide a window, n bases long, along each read and record the sequence within the window as the read is traversed. Starting at the first base position, the n-mer comprising the first base and the subsequent n-1 bases is recorded. The window position is then incremented 1 base at a time until all n-mers from the read have been entered into the database. If an n-mer sequence already exists in the database, the number of occurrences (multiplicity, m) is incremented by 1. As an example, the distribution of m over all 454 reads for sample NA06993 is shown in Supporting Figure S9. The distribution is multimodal. The peak at multiplicity m = 1 comprises all n-mers that contain one or more 454 sequencing errors and that are not repeated as a group in any other read of the particular region of the genome. The peak near m = 50 is the mode of the local, n-mer matched, consensus coverage of the genome. The high multiplicities in the tail of the distribution are due to genomic regions where the long PCR segments overlap.
To de novo assemble the mtgenome using the n-mer database data, we make the following minimal set of assumptions: 1) there are no duplicated n-mers within the genome; 2) there are no palindromic n-mers, i.e., an n-mer on the L-strand of the mtGenome is not found in reverse complement form on the H-strand and vice versa, and 3) for a short n-mer drawn from the genome, the sequence read of this n-mer is more likely to be correct than contain an error. The third assumption depends on the sequence-context-dependent error rate of the 454 platform. If we consider as a characteristic value, λ = 0.005, for the average 454 error rate per base, then for any n-mer, the probability that the n-mer is error free is given by p = (1−λ)n. If we choose n = 27, this gives p = 0.87. This means that a majority of database n-mers are correct given that most of the mtgenome sequences are “average in content”. This calculation assumes errors are uncorrelated along the n-mer, which is not the case for the sequence context of long homopolymeric runs (see below). Our choice of n = 27 is a compromise value that seeks to ensure the validity of assumptions 1–3: the shorter an n-mer is, the more likely it is to be repeated in the mtgenome or be a palindrome; on the other hand, if the n-mer is chosen to be too long, the majority of n-mers derived from reads at a given genome position will contain an error somewhere within the n-mer. The satisfaction of assumption 3 allows us to apply a “majority base wins” criterion as our basis for selecting sequences in our de novo consensus assembly.
The de novo assembly initially proceeds by searching the database for the n-mer matching at position 1 on the L-strand of the rCRS and ensuring the multiplicity m for L- strand and H- strand sequences at this position exceed 10. This starting condition was satisfied for all mtgenomes assembled (i.e., no genome contained a polymorphism with respect to the rCRS in this portion of the genome, otherwise successive positions along the rCRS could readily be probed until this condition was satisfied). First, de novo assembly proceeds in the antegrade direction (with increasing rCRS position). We form the four possible candidates for the successive n-mer in the sequence and their respective reverse complements by dropping the first base of the n-mer at rCRS position 1 and adding A, T, C, or G to the end. The database is then searched for each candidate n-mer and its reverse complement, and the sum of the respective forward and reverse n-mer multiplicities is recorded for each candidate n-mer. According to assumption 3), the appropriate choice of the subsequent n-mer is the one that is the most abundant in the database. The selected new base is then added to the de novo assembly and the process is repeated until the starting n-mer sequence at rCRS position 1 is again encountered (exploiting the circular nature of the mtgenome). The antegrade de novo assembly is then complete. To assign consensus coverage at each base position we form the n-mer from the antegrade assembly in which the position in question is at the center, with (n_mer-1)/2 bases on either side. The database is then searched for this n-mer and the sum of the L-strand and H- strand multiplicites, m, is recorded as the consensus coverage.
As a check on the antegrade de novo consensus assembly, the entire assembly process above is repeated by sequencing from rCRS position 1 in the retrograde direction using the same database. Here, the base at the end of the L-strand n-mer is dropped and the candidate n-mers for the next position in the retrograde direction are formed by adding A, T, C, or G to the beginning of the n-mer. The alternative assemblies in the antegrade and retrograde directions are subsequently compared to identify discrepancies for further investigation and curation.
Substitution heteroplasmy candidates, and their respective fractions with respect to the consensus sequence, can then be determined by replacing the central base at each position with the other three possible bases, and then summing the L-side and H-side multiplicities of the n-mers in the database. Indel heteroplasmys with respect to the consensus can also be determined using a method aligning the unused n-mers in the database against the consensus.
|
10.1371/journal.pntd.0001984 | Modeling the Impact and Costs of Semiannual Mass Drug Administration for Accelerated Elimination of Lymphatic Filariasis | The Global Program to Eliminate Lymphatic Filariasis (LF) has a target date of 2020. This program is progressing well in many countries. However, progress has been slow in some countries, and others have not yet started their mass drug administration (MDA) programs. Acceleration is needed. We studied how increasing MDA frequency from once to twice per year would affect program duration and costs by using computer simulation modeling and cost projections. We used the LYMFASIM simulation model to estimate how many annual or semiannual MDA rounds would be required to eliminate LF for Indian and West African scenarios with varied pre-control endemicity and coverage levels. Results were used to estimate total program costs assuming a target population of 100,000 eligibles, a 3% discount rate, and not counting the costs of donated drugs. A sensitivity analysis was done to investigate the robustness of these results with varied assumptions for key parameters. Model predictions suggested that semiannual MDA will require the same number of MDA rounds to achieve LF elimination as annual MDA in most scenarios. Thus semiannual MDA programs should achieve this goal in half of the time required for annual programs. Due to efficiency gains, total program costs for semiannual MDA programs are projected to be lower than those for annual MDA programs in most scenarios. A sensitivity analysis showed that this conclusion is robust. Semiannual MDA is likely to shorten the time and lower the cost required for LF elimination in countries where it can be implemented. This strategy may improve prospects for global elimination of LF by the target year 2020.
| The Global Program to Eliminate Lymphatic Filariasis (LF) employs annual mass drug administration (MDA) of antifilarial drugs to reduce infection rates in populations and interrupt transmission. While this program is working well in many countries, progress has been slow in others, and some countries have not yet started MDA programs. We used computer simulation modeling and cost projections to study how increasing MDA frequency from once to twice per year would affect program duration and costs. Our results suggest that semiannual MDA is likely to reduce the time required to eliminate LF by 50% and reduce total program costs (excluding the cost of donated drugs) in most situations. For these and other reasons, we expect semiannual MDA to be superior to annual MDA in most endemic settings. Semiannual MDA should be considered as a means of accelerating LF elimination in areas where it can be implemented, because this may improve prospects for global elimination of LF by the target year 2020.
| The Global Program to Eliminate Lymphatic Filariasis (GPELF) was launched in 2000 with the aim of eliminating lymphatic filariasis (LF) as a public health problem by 2020 [1]. The recommended strategy is to treat entire at-risk populations annually with a single dose of ivermectin and albendazole (IVM+ALB) in sub-Sahara Africa or with diethylcarbamazine and albendazole (DEC+ALB) in other regions for a minimum of 5 years [2]. Mapping studies suggest that mass drug administration (MDA) is needed in 72 endemic countries [3].
As indicated in the GPELF 2010 progress report, progress toward LF elimination varies widely between countries [3]. Some countries started their MDA programs early and may have already interrupted LF transmission, while other countries lag behind [3]. Nineteen countries had not yet started MDA, and geographical coverage was incomplete in 24 others. Reasons cited for slow progress in some areas included major logistic challenges, political instability, conflict, and co-endemicity with Loa loa [4]. Also, results from ongoing MDA programs have sometimes been disappointing. Sentinel site data collected after 5 years of annual MDA show that microfilaria (mf) prevalence had dropped to 0% in about two-thirds of sentinel sites sampled. However, mf rates had decreased by less than 50% in 10% of the sites sampled [4].
With the goal of LF elimination by 2020 in mind, it is now important and timely to study whether elimination programs can be accelerated. A straightforward option would be to increase the frequency of MDA from once per year (annual) to twice per year (semiannual). While increasing MDA frequency might be expected to shorten the time required for elimination, the magnitude of this effect is uncertain. Only one study directly compared the impact of annual and semiannual MDA and this was for brugian filariasis: semiannual MDA with DEC alone caused a more rapid decline in mf prevalence than annual treatment. However, the duration of this study was too short to support conclusions regarding elimination [5]. Results from other studies of semiannual MDA are difficult to interpret, because they did not provide results from a control area with annual MDA [6], [7], [8].
For decision-making, it is also important to consider how costs per year and overall costs for LF elimination programs are likely to change if MDA frequency is increased. Of course, costs per year will increase, but they will not necessarily double, and the cumulative cost for the entire program may even decline. The costs of twice yearly MDA have not been formally studied for LF or other neglected tropical diseases. However, they can be projected from detailed cost data by activity and cost item that are available for yearly MDA for LF and soil-transmitted helminthiasis [9], [10], [11], [12].
We have used the well-established simulation model LYMFASIM to estimate the number of treatment rounds and duration of MDA programs that would be needed to eliminate LF with annual and semiannual MDA in different settings. Simulations were performed for typical endemic areas in West Africa (with IVM+ALB treatment and Anopheles transmission) and in India (with DEC+ALB treatment and Culex transmission) with different pre-control endemicity levels and MDA coverage rates. In addition, we have compared projected costs of annual or semiannual MDA, both per year and for the total required duration of LF elimination programs.
We estimated the costs of MDA for LF programs with annual and semiannual treatment from the perspective of the endemic country government. The cost analysis covers financial and economic costs. The financial costs are the costs of all inputs purchased in cash for MDA, including purchased MDA drugs, materials and supplies, ministry of health personnel salaries, and per diem payments for community drug distributors [11]. Economic costs also include the costs of donated drugs for MDA (India: albendazole; Burkina Faso: ivermectin, albendazole). Costs were calculated for a target population of 100,000 eligible persons in three steps.
We studied the extent to which key assumptions affect conclusions regarding the relative cost of the two MDA schedules (once or twice yearly MDA) in a univariate sensitivity analysis. On the cost side, we assessed the effect of changing the discount rate to 0% or 6% instead of 3%, the effect of including the cost of donated drugs, and we considered the scenario where drugs are only bought for people who are actually treated instead of for all eligibles (with the idea that any remaining drugs would be stored and used in a later round). These factors do not influence the number of rounds required, but they may affect the total costs of annual and semiannual treatment programs and influence policy decisions.
With respect to the simulations, we examined the impact of changing assumptions regarding the efficacy of drugs on adult worms. This may affect the total number of treatment rounds (and total costs) required for LF elimination programs with annual or semiannual MDA. The fraction of worms assumed to be killed or permanently sterilized after each treatment was varied with low, medium (baseline) and high values (50%, 65%, and 80% for DEC+ALB, and 20%, 35% and 50% for IVM+ALB). Further, we studied the impact of including variability in this parameter, so that the fraction of worms killed or sterilized varies randomly between individuals in each treatment cycle and within individuals in different treatment cycles. The variation is described by a beta distribution with the mean equal to the baseline fraction of worms killed/sterilized and standard deviation equal to 0.3.
Figure 1 shows an example of model-predicted trends in mf prevalence. The presented trends are for a West African area with a pre-control mf prevalence of 20%. Six rounds of annual MDA with IVM+ALB were provided starting at time 0. Coverage was 70% and drug efficacy was quantified according to our baseline assumptions. The figure displays the trend of 25 runs that were all conducted with the same input assumptions. Variation in the outcomes is due to stochasticity. In this example, 1 out of 25 runs showed recrudescence after stopping MDA; one other run seemed to be moving to elimination, but this was not yet achieved. The probability of elimination in this case was 23/25 (92%).
Figure 2 illustrates how the model-predicted probability of LF elimination increases with the number of MDA rounds provided. Results are shown for the India and West Africa model variants, for annual and semiannual MDA, and for different coverage levels.
Table 3 shows the expected costs per treatment round by activity and cost item for annual and semiannual MDA in both regions. Providing semiannual instead of annual MDA reduces the cost per MDA round. This cost reduction is 11% for India and 18% for West Africa, if costs of donated drugs are excluded from the analysis. The cost reduction is smaller if donated drug costs are included (7% for India and only 1% for West Africa).
Table 4 shows the number of treatment rounds for achieving a 99% probability of elimination, under our baseline assumptions. This number is highly dependent on treatment coverage and pre-treatment mf prevalence rates, but it does not depend much on the frequency of treatment (annual or semiannual). In most circumstances, therefore, the total duration of semiannual MDA is about half of that for annual MDA. In very unfavorable circumstances (areas with high baseline infection rates and very low MDA coverage), one extra MDA round may be required to reach elimination with semiannual MDA.
The total costs of MDA programs depend on the cost per round, the required number of MDA rounds, and thereby also on local circumstances and coverage rates. Table 4 shows estimated total costs for LF elimination programs, assuming an annual discount rate of 3% for future costs. In this analysis, which does not count the cost of any donated drugs, projected costs of semiannual MDA are almost always lower than costs of annual MDA. In West Africa, this is even true when semiannual MDA requires one more MDA round, because of the large reduction in cost per round. In the single India scenario where semiannual MDA required one more round than annual MDA, the projected total program costs were comparable for annual and semiannual MDA.
Table 5 and Table 6 summarize the results of the sensitivity analyses for India and West Africa. The tables show the ratio of total program costs for semiannual MDA over annual MDA under varied assumptions. This ratio shows which approach is less expensive (with values <1 indicating that semiannual MDA is cheaper and vice versa), and it provides an indication of the relative cost difference. Changing the discount rate (0% or 6%) had little impact on the projected total costs of semiannual and annual MDA programs and their ratios. Its effect increased with the duration of MDA, and a higher discount rate tends to favor the slower annual MDA programs. But the total program cost of semiannual MDA was lower or comparable to the cost of annual MDA in all scenarios.
Including the costs of donated drugs changed the outcome of the cost analysis significantly. The costs per treatment round increased by a large amount (by an amount that was the same for annual and semiannual treatment), and the relative difference was reduced. While semiannual MDA remained cheaper in most Indian scenarios, it became slightly more expensive in the West African scenarios. The highest increase (17%) was seen in the West African scenario with the highest endemicity (pre-control mf prevalence of 27%), because here semiannual MDA would require one more round than annual MDA. Whether drugs are purchased for all eligibles in every round or for the percentage of people treated only (assuming that previously unused drugs were not wasted/expired), hardly affected the ratio of total program cost of semiannual over annual MDA.
Model assumptions about the percentages of adult worms killed (or permanently sterilized) by a single treatment affected the total number of treatment rounds needed to achieve elimination and therefore the estimated total program costs. However, this did not have a major impact on ratios of total program cost for semiannual vs. annual MDA programs (Table 6). Adding random variation in the percentage of adult worms killed (or permanently sterilized) sometimes led to an extra treatment round in either annual or semiannual MDA, but nevertheless semiannual MDA was still favored in this analysis.
Our simulations and cost calculations suggest that semiannual MDA will achieve LF elimination in about half of the time that would be required with annual MDA. Estimated total program costs were strongly driven by the required number of treatment rounds, and this in turn depended on pre-treatment endemicity levels and MDA coverage rates. However, total program costs for endemic countries (i.e. excluding the cost of donated drugs) were always lower for the semiannual MDA program or comparable.
Cost calculations were based on observed data from 1996 and 2002 [9], [11], which were then adjusted to reflect current day practices and prices. The absolute cost estimates are subject to various assumptions. For the current purpose, though, the main interest is in the relative cost of semiannual vs. annual MDA, which is much less dependent on the assumptions. Key assumptions in the cost projections did not affect the conclusion that the cost of LF elimination with semiannual MDA is lower than or similar to the cost of programs with annual MDA. A high discount rate (reflecting a strong preference to delay cost to the future) favors annual MDA programs, in which the expenses are spread over a longer period and postponed further into the future. However, the efficiency gains of semiannual MDA mostly compensate for this. If the high costs of donated drugs are included in the cost estimates, the relative difference in cost per round diminishes and becomes negligible in West Africa. In West Africa, therefore, the efficiency gain no longer compensates for the effect of discounting or the need for an extra treatment round in semiannual MDA. But this situation only occurs when many MDA rounds are required because of unfavorable transmission conditions (as in our high endemic West African scenario). Slightly increased program costs may be justified in such situations, because here the positive impact of increasing MDA frequency on total program duration is very strong. We did not test the impact of future inflation with different annual inflation rates, but this would work in favor of shorter duration semiannual MDA programs, and it would tend to strengthen our conclusions.
Estimates of the required duration of MDA in different settings were obtained by computer simulation, because empirical evidence from LF elimination programs is still limited. Many countries have made great strides, and some have stopped MDA, but no country that had ongoing transmission of LF in 2000 has been verified to have interrupted transmission of the infection using MDA [4]. Modeling is a powerful tool for decision making in infectious disease control [30], but predictions are subject to uncertainty [31]. An important uncertainty in our study concerns the efficacy of drugs. The sensitivity analysis showed that more treatment rounds would be required if the employed drugs are less effective than assumed and vice versa, while adding random variability in the percentage of worms killed by treatment did not influence the predicted outcomes. In any case, these assumptions equally affected predictions for semiannual and annual MDA programs and did not significantly affect the relative cost difference between the two strategies.
Field studies are needed to confirm projected cost reductions that can be achieved with semiannual MDA in both regions and to assess any indirect effects that might affect the relative efficiency of annual vs. semiannual MDA. For example, the likelihood that unused medication is stored and used in subsequent rounds may be higher in semiannual than in annual MDA programs. Also, it is possible that increased treatment frequency will increase coverage rates (e.g. due to higher population awareness) and reduce systematic non-compliance (e.g. due to the fact that MDA does not always take place in the same season). Such changes could reduce the number of MDA rounds needed for elimination and further increase the efficiency of semiannual vs. annual MDA programs. But the opposite could also occur if insufficient effort is made to maintain high coverage rates.
The efficiency gain in cost per treatment round achieved by shifting from annual to semiannual MDA was somewhat different for India and West Africa. This reflects differences in program organization and costing structure in the two regions [9], [11]. For example, the West African cost estimates included central administrative costs, laboratory costs, and adverse reaction monitoring, while these costs were not counted in the estimates for India. In general, the efficiency gain achieved is dependent on strategic choices (e.g. on activities to repeat and available budgets), health systems, program organization, and the local cost of different inputs. Results could be somewhat different in other settings. In the supporting information text S1, we show how the relative difference in total program costs depends on the relative difference in cost per treatment round, the required number of treatment rounds and applied discount rate.
The duration of MDA varies between regions because of differences in exposure patterns to mosquitoes, characteristics of the vector, timing of MDA, immigration of people, etc. Simulation results are therefore not directly generalizable to other areas, but this is not pertinent to the comparison of annual and semiannual MDA durations. This becomes clear when one compares results projected in this study for LF elimination programs in India and West Africa; although there are important differences between these models that result in very different estimates for the number of MDA rounds needed for elimination (generally higher in Africa), the basic conclusion that doubling MDA frequency halves the required duration of LF elimination programs and reduces total program costs is valid for both of these regions and it should also apply to other regions.
Besides the total program costs, there are other important factors to consider in deciding whether MDA frequency should be increased. Increasing treatment frequency leads to a faster decline in the incidence of LF infection. This should increase the likelihood of achieving LF elimination by the target year of 2020, which is very relevant for countries that have not yet started their MDA programs. Incidence of clinical manifestations will also decline faster, which results in larger population health gain in terms of the total number of DALYs averted and results in increased productivity. Quantification of these extra benefits was beyond the purpose of this study. Increasing the treatment frequency and reducing program duration may also be beneficial for other reasons. E.g., shorter programs may be more politically attractive to health officials, and they would also be expected to have reduced risks of interruption due to natural disasters, political instability, or wars. Shorter programs may also reduce the risk of emergence of resistance to anthelmintics during LF elimination programs. Since albendazole and ivermectin also affect other diseases than LF, increasing the treatment frequency would increase their impact on diseases like soil-transmitted helminths and other NTD's – albeit for a shorter period.
Potential barriers for increasing the frequency of MDA are the increased cost per year and practical difficulties that may be associated with semiannual MDA. Increased annual drug requirements may exceed supplies of donated drugs. Also, more frequent MDA might overwhelm countries' capacities for delivering MDA to endemic populations, in view of already heavily burdened health systems and many competing health priorities [32]. Semiannual MDA may not be feasible in all areas due to weather, seasonal migration of populations, or logistical considerations. Other factors may play a role when LF elimination is integrated with programs for control of other neglected tropical diseases (NTDs). That is to say, how would accelerated LF elimination affect control programs for other NTDs?
Poor-performing programs, with very low treatment coverage, require relatively many treatment rounds. Increasing the treatment frequency from annually to semiannually would reduce the total program duration by about half, but not the number of treatment rounds. However, investments or strategies that increase coverage rates will improve results of annual or semiannual MDA, thereby reducing the number of treatment rounds required and the total costs (see Table 4).
In summary, computer simulations suggest that the frequency of MDA – annual vs semiannual – does not strongly influence the total number of treatment rounds required to achieve LF elimination. The costs per year are higher with semiannual MDA, but total program costs (excluding donated drugs) are projected to be lower or about the same when semiannual MDA is used. The few situations where the total program costs of semiannual MDA are slightly higher are also challenging situations for LF elimination where semiannual MDA may improve the odds of success. Therefore, we expect semiannual MDA to be superior to annual MDA in most endemic settings. Considering the GPELF goal of LF elimination by 2020, semiannual MDA should be considered as a means of accelerating LF elimination in areas where it can be implemented.
|
10.1371/journal.pntd.0004382 | Multiple Exposures to Ascaris suum Induce Tissue Injury and Mixed Th2/Th17 Immune Response in Mice | Ascaris spp. infection affects 800 million people worldwide, and half of the world population is currently at risk of infection. Recurrent reinfection in humans is mostly due to the simplicity of the parasite life cycle, but the impact of multiple exposures to the biology of the infection and the consequences to the host’s homeostasis are poorly understood. In this context, single and multiple exposures in mice were performed in order to characterize the parasitological, histopathological, tissue functional and immunological aspects of experimental larval ascariasis. The most important findings revealed that reinfected mice presented a significant reduction of parasite burden in the lung and an increase in the cellularity in the bronchoalveolar lavage (BAL) associated with a robust granulocytic pulmonary inflammation, leading to a severe impairment of respiratory function. Moreover, the multiple exposures to Ascaris elicited an increased number of circulating inflammatory cells as well as production of higher levels of systemic cytokines, mainly IL-4, IL-5, IL-6, IL-10, IL-17A and TNF-α when compared to single-infected animals. Taken together, our results suggest the intense pulmonary inflammation associated with a polarized systemic Th2/Th17 immune response are crucial to control larval migration after multiple exposures to Ascaris.
| Human ascariasis caused by the helminths Ascaris lumbricoides and Ascaris suum, is the most prevalent neglected tropical disease in the world, affecting more than 800 million people and mainly school-aged children. The parasite life cycle may be divided in two distinct phases after the initial infection: (i) migration of parasitic larval stages through several tissues (intestinal mucosa, blood circulation, liver and lung/airways), namely larval ascariasis; and (ii) establishment of adult worms in the lumen of the small intestine, causing the despoilment of nutrients and secretion/excretion of parasitic products, which down modulate the immune response of the host and characterizing the chronic infection. Over the past decades special focus has been dedicated to chronic Ascaris infection with the assessment of immunopathological aspects of the disease in chronically infected individuals from endemic areas. However, knowledge about larval ascariasis still remains scarcedue the limitations of diagnostic techniques and the need for an experimental model that would mimic natural infection. Indeed, many aspects of the immunobiology of the early Ascaris infection are still poorly understood. In this context, single and multiple infections in mice were performed in order to characterize the parasitological, histopathological, tissue functional and immunological aspects of larval ascariasis.
| New information from the Global Burden of Disease Study 2010 (GBD 2010) indicates that more than 800 million people are infected with Ascaris spp. (A. lumbricoides and A. suum), which ranks ascariasis as the most common affliction of people living in poverty [1]. In the past, A. suum has been implicated as an anthropozoonotic species based on epidemiological evidence from the field [2], molecular similarity to A. lumbricoides [3, 4] and also by experimental infection in humans [5]. Ascariasis is frequently associated with high rates of reinfection in endemic areas due to a constant exposure to the infective form of the parasite [6, 7].
Human ascariasis is characterized by a Th2 and regulatory immune response [8, 9], although innate production of IL-5, IL-6 and TNF-α seems to play crucial role in the pathogenesis of experimental larval ascariasis [10]. Despite the lack of evidence on Ascaris infection, new studies have been proposed that an IL-6-dependent, Th17 response might play an important role into the pathogenesis of helminth infections [11] and allergic manifestations [12], resulting in modulation of the Th2 response and possible susceptibility of the host to the parasitic infection. The role of IL-17 in the pathogenesis of helminth infection was highlighted in the development of hepatointestinalperioval granulomas caused by Schistosoma mansoni infection [13].
Larval ascariasis (established by larval migration through the host’s organs) is characterized by intense pulmonary injury and inflammatory infiltration, which is initially comprised of neutrophils during the peak of larval migration and followed by later infiltration of eosinophils and mononuclear cells [10]. The robust inflammatory response elicited by parasitic migration seems to be protective to the host [10] and might represent the establishment of concomitant immunity to new helminthic infections. Of note, epidemiological studies have demonstrated that children are more susceptible to a higher prevalence and intensity of Ascaris infection than adults [1, 14], implying that partial protection against the parasite is acquired over the years. However, the mechanisms underlying the susceptibility/resistance to ascariasis still remain unknown and need to be elucidated.
Therefore, the use of a murine experimental model for Ascaris infection is currently crucial and may provide detailed information on the biology of early Ascaris spp. infection. In the current study, inbred BALB/c mice were employed due to its susceptibility to Ascaris spp. infection [10, 15] and also to allow comparison with previous immunopathological studies [10, 15], particularly those involving subsequent challenge infection with A. suum [16, 17]. Here, we evaluated the parasitological and immunological aspects of multiple exposure to Ascaris infection in mice, focusing on the immunopathological mechanisms that underlie protection against larval ascariasis.
For this study, 30 BALB/c mice (male, 8 weeks old) were obtained from the Central Animal Facility from the Federal University of Minas Gerais, Brazil. Animals were subcutaneously treated (0.2% / 20 mg of live weight) with Ivermectin (Ouro Fino, Brazil), and stool examinations were regularly performed to confirm the absence of any parasitic infection.
Animals were divided into three groups: the non-infection group (NI), which received PBS only; the single-infection group (SI), which received two doses of PBS and 2,500 fully embryonated A. suum eggs at the last time point; and the reinfection group (RE), where all animals received three doses of 2,500 fully embryonated A. suum eggs every two weeks (Fig 1). Tissue harvesting for parasitological and immunological evaluation was performed as previously described following the peak migration of larvae from the liver, lungs and intestine, which are observed on the 4th, 8th and 12th day of infection, respectively [10].
Adult A. suum worms were harvested from pigs at a Brazilian slaughterhouse (Belo Horizonte, Minas Gerais, Brazil). Eggs were isolated from uteri of female worms by gentle mechanical maceration and further purified by use of cell strainers (70 μm). Isolated eggs were incubated with 0.2 M H2SO4 for embryonation, as described by Boes and colleagues [18]. After the 100th day of culture, which corresponds to the peak of larval infectivity [10], fully embryonated eggs were used in experimental infections.
Using a gavage needle, all animals received either 200 μl of PBS or 2,500 embryonated eggs in 200 μl PBS, according to the group or timepoint. Parasite burden was evaluated by recovery of larvae from liver (n = 6/group), the lungs (n = 6/group), and the small intestine (n = 6/group). Tissues were collected, sliced with scissors and placed in a Baermann apparatus for 4 hours in the presence of PBS at 37°C. The recovered larvae were then fixed (1% formaldehyde in PBS) and counted under an optical microscope.
The haematological profiles of infected and reinfected mice were evaluated during the different stages of the larvae migration. Briefly, 500 μL of blood was collected from a superficial vein of 6 BALB/c mice per group using capillary Pasteur pipettes primed with anticoagulant EDTA. The total leukocytes were counted using an automated hematological analyzer (Bio-2900 Vet, Bioeasy, USA) and percentages and absolute numbers of lymphocytes, monocytes, eosinophils and neutrophils were further determined by optical microscopy in blood smears stained with Giemsa.
In order to determine the cytokine profile in the serum, 500 μl of blood was collected from each mouse at all experimental timepoints. Blood was collected from the retro-orbital sinus using a capillary Pasteur pipette without anticoagulant. Collected blood was transferred to Eppendorf tubes for coagulation, followed by centrifugation and serum collection. The production of IL-2, IL-4, IL-6, IL-10, IL-17A, IFN-γ and TNF-α was assessed by flow cytometry (Th1/Th2/Th17 Cytometric Bead Array, BD Biosciences, USA) using a FACScan (BD Biosciences, USA) according to the manufacturer's recommendation. Serum levels of IL-5 were measured using a sandwich ELISA kit (R&D Systems, USA) according to the manufacturer's instructions. The absorbance was determined by a VersaMax ELISA microplate reader (Molecular Devices, USA) at a wavelength of 492 nm, and the cytokine concentration (pg/mL) for each sample was calculated by interpolation from a standard curve.
The activities of eosinophil peroxidase (EPO) and neutrophil myeloperoxidase (MPO) in the lung homogenates were measured according to a method described by Strath and modified by Silveira [19, 20]. After tissue homogenization (Power Gen 125 –Fisher Scientific Pennsylvania, USA), the homogenate was centrifuged at 8,000 g for 10 min at 4°C and the remaining pellet was examined to determine the activity of EPO and MPO. For the EPO assay, the pellet was homogenized in 950 μL PBS and 0.5% hexadecyltrimethylammonium bromide (Sigma Chemical Co, St. Louis, MO, USA) and then frozen/thawed three times using liquid nitrogen. The lysate was then centrifuged (1,500 g, 4°C, 10 min) and the supernatant was distributed (75 μL/well) in a 96-well microplate (Corning, USA) followed by the addition of 75 μL of substrate (1.5 mM OPD and 6.6 mM H2O2 in 0.05 M Tris-HCl, pH 8.0). After incubation for 30 minutes at room temperature, the reaction was stopped by the addition of 50 μL of 1 M H2SO4 and the absorbance was determined at 492 nm.
For the MPO assay, the pellet was homogenized in 200 μL of buffer 1 solution (0.1 M NaCl, 0.02 M Na3PO4, 0.015 M Na2EDTA, pH 4.7) followed by centrifugation (1,500 g, 4°C, 10 min). 800 μL of buffer 2 solution (0.05 M NaPO4, 0.5% hexadecyltrimethylammonium bromide) was added to the pellet and the mixture was homogenized and then frozen/thawed three times using liquid nitrogen. The lysate was centrifuged (1,500 g, 4°C, 10 min), and the supernatant was used for the enzymatic assay. 25 μL/well were distributed on to 96-well microplates (Corning, USA) followed by the addition of 25 μL of substrate TMB (3.3'-5.5;- tetramethylbenzine + 1.6 mM dimethylsulfoxide) and 100 μL of 0.5 M H2O2. After incubation for five minutes at room temperature, the reaction was stopped by the addition of 100 μL of sulphuric acid (1 M H2SO4). Absorbance was determined by a VersaMax ELISA microplate reader (Molecular Devices, USA) at a wavelength of 450 nm.
For the analysis of the cellularity and blood loss in the bronchoalveolar lavage (BAL), six single- and reinfected BALB/c mice were euthanized eight days after infection. Basically, a 1.7 mm catheter was inserted into the trachea of the animals, and 1 mL of PBS was used twice for perfusion and aspiration in order to assess the leukocyte infiltration in the bronchoalveolar compartment. The bronchoalveolar lavage was filtered on cell strainers 70 μm (BD, USA) to recover and quantify the A. suum larvae presented in the BAL. The material was centrifuged at 3,000 g for 10 minutes and the pellet was used to determine the total number of leukocytes and differentiation of macrophages, lymphocytes, eosinophils and neutrophils using optical microscopy. The supernatant was used to quantify the amount of total protein and haemoglobin content. Samples from six non-infected BALB/c mice were used as controls.
The quantification of total protein was determined by BCA Protein Assay kit (Thermo Scientific, USA) and was performed on BAL to measure possible protein leakage into the airways, as previously described [21]. The results were expressed as μg of total protein per mL of BAL. The extent the alveolar haemorrhage was assessed by the amount of hemoglobin (Hb) detected in BAL supernatant using the Drabkin method, as previously described [22]. The hemoglobin concentration in the samples was determined spectrophotometrically by measuring absorbance at 540 nm and interpolation from a standard hemoglobin curve, starting at 1 mg/mL. Hemoglobin content was expressed as μg of Hb per mL of BAL.
For the liver and lung histological analysis, organs were removed on the fourth and eighth days after infection and were fixed in a 10% solution of formaldehyde (Synth, Brazil) in PBS for 72 hours. After processing in alcohol and xylol, tissue fragments were embedded in paraffin, and 4 μm thick sections were obtained and stained with hematoxylin and eosin (H&E). The tissues were analyzed using KS300 software coupled to a Carl Zeiss image analyzer (Oberkochen, Germany).
Severity of liver injury was assessed by calculation of all areas of inflammation and necrosis by morphometric analysis. Hepatic lesions were assessed through a 10x objective microscope Axiolab Carl Zeiss and the images were captured using a JVC TK-1270/RGB microcamera (Tokyo, Japan). The area of lesions was measured in μm2 using KS300 software coupled to Carl Zeiss image analyzer (Oberkochen, Germany). All slides were digitized by scanner Canon Lide 110 at 300 dpi resolution. The pixels of each histological section were fully screened, with subsequent creation of a binary image and calculating the total area of the cut. The area of the lower cutoff was used as a minimum standard of tissue to be statistically analyzed [23].
To evaluate the intensity of pulmonary inflammation and hemorrhage, the degree of thickening of interalveolar septa was calculated. Thirty random images were captured at 40x objective, comprising an area of 1,6x106 mm2. Through the KS300 software, all pixels of the lung tissue in the real image were selected to create a binary image, digital processing and calculating the area in mm2 of the interalveolar septum [24].
Mice were anesthetized with a subcutaneous injection of ketamine and xylazine (8.5 mg/kg xylazine and 130 mg/kg ketamine) to maintain spontaneous breathing under anesthesia. Mice were tracheostomized, placed in a body plethysmograph and connected to a computer-controlled ventilator (Forced Pulmonary Maneuver System, Buxco Research Systems, Wilmington, North Carolina USA). This laboratory set-up, specifically designed for use on mice, has only a canula volume (death space) of 0.8 mL and provides semi-automatically three different maneuvers: Boyle’s Law FRC, quasi-static pressure-volume and fast-flow volume maneuver. First, an average breathing frequency of 160 breaths/min was imposed to the anesthetized animal by pressure-controlled ventilation until a regular breathing pattern and complete expiration at each breathing cycle was obtained. Under mechanical respiration the Dynamic Compliance (Cdyn) and Lung Resistance (Rl) were determined by Resistance and Compliance RC test. To measure the Forced Vital Capacity (FVC) and Inspiratory Capacity (IC), the quasi-static pressure-volume maneuver was performed, which inflates the lungs to a standard pressure of +30 cm H2O and then slowly exhales until a negative pressure of -30 cm H2O is reached. The quasi-static chord compliance (from 0- to +10 cm H2O) was calculated with this maneuver considering the volume/pressure of the expiration. With the fast flow volume maneuver, lungs were first inflated to +30 cm H2O and immediately connected to a highly negative pressure in order to enforce expiration until -30 cm H2O. Forced Expiratory Volume (forced expiratory volume at 100 milliseconds, FEV100) was recorded during this maneuver. Suboptimal maneuvers were rejected and for each test in every single mouse at least three acceptable maneuvers were conducted to obtain a reliable mean for all numeric parameters.
Statistical analyses were performed using the software GraphPad Prism 6 (GraphPad Inc., USA). Grubb's test was used to detect possible outliers in the samples. For comparison of parasitic burden (Fig 2) and areas of lesion in the liver and lungs (Figs 3G and 4G), the Mann-Whitney test was used. Data from EPO (Fig 4H) and MPO (Fig 4I) assays and also from pulmonary mechanics (Fig 5), haemoglobin, protein levels and BAL cellularity (Fig 6) were analysed by Kruskal-Wallis test followed by Dunn's test. Finally, Two-way ANOVA with multiple comparisons test was performed to assess differences between the groups in function of time (Figs 7 and 8). All tests were considered significant when the p value was equal or less than 0.05.
The maintenance and use of animals were in strict accordance with the recommendations of the guidelines of the Brazilian College of Animal Experimentation (COBEA). The protocol was approved by the Ethics Committee for Animal Experimentation (CETEA) of the Universidade Federal de Minas Gerais, Brazil (Protocol# 45/2012). All efforts were made to minimize animal suffering.
Based on a previous study from our group, we sought to determine whether multiple exposures to A. suum influenced the number of larvae recovered from liver, lung and small intestine. While no differences in the number of larvae recovered in the liver from single or multiple infection was observed (Fig 2A), the reinfected group showed a significant reduction in the number of larvae recovered from the lungs (p = 0.004) and BAL (p = 0.02) when compared to single-infected group at the 8th day post-infection (Fig 2B and 2C). Following the life cycle, no significant differences in the larvae recovered in the intestine from both infected and reinfected groups were observed at 12 days post-infection (Fig 2D).
After we confirmed that multiple exposures to A. suum induced significant protection indicated by a reduction of parasite burden in the lungs, we further evaluated the pattern of cellular response by histopathological analysis in order to try to explain the mechanisms of this protection.
Although there were no differences in the numbers of larvae recovered from the liver on the 4th day post-infection between single- and multiple-infected animals, the latter group showed a larger lesion area caused by larvae migration when compared to the single-infected group (p = 0.002) (Fig 3G). In the microscopical analysis of liver parenchyma, areas with hepatocyte necrosis and polymorphonuclear inflammatory infiltrate–composed primarily of eosinophils and neutrophils–were observed in the single-infected group (Fig 3C and 3D). These findings were even more pronounced in the reinfection group, in which granulomas were also present (Fig 3E and 3F).
The lungs of animals from both single- and reinfection groups showed microscopic lesions in the lung parenchyma that were characterized by the presence of a polymorphonuclear inflammatory infiltrate consisting primarily consisted of eosinophils and also thickening of the interalveolar septa when compared to controls (Fig 4A–4F). Despite the significant reduction of parasite burden in the reinfected group, the area of pulmonary lesion was considerably higher in this group when compared to the single-infected animals (p = 0.002) (Fig 4G).
Of note, the thickening of the septa was less pronounced in the single-infection group (Fig 4C and 4D) compared with the reinfection group (Fig 4E and 4F), suggesting that multiple parasitic exposures lead to chronic lung injury associated with tissue remodeling. The chronic activation of eosinophils and neutrophils in lung tissue was evident in the results of assessment of eosinophil peroxidase and myeloperoxidase activities at 8 days post-infection in the lung, which was significantly higher during reinfection when compared to single exposure to the parasite (p = 0.0004 and p = 0.02, respectively) (Fig 4H and 4I). Collectively, these data indicate that multiple exposures to Ascaris spp. induced a chronic and robust immune response in the lungs of reinfected group, ultimately related to increased tissue damage and protection against progression of the parasitic cycle.
The analysis of pulmonary mechanics during inflammation was performed by forced spirometry technique to further investigate the physiological modifications caused in lung functions after 8 days of single or multiple Ascaris infection (Fig 5). The pulmonary test detects different types of physiologic parameters in mouse lungs: (i) lung volumes to determine the effects of tissue damage by evaluation of lung volume loss, mostly caused by aedema and airway thickness, as presented by Functional Vital Capacity (Fig 5A) and Inspiratory Capacity (Fig 5B); (ii) evaluation of elastic properties of lung tissue is a measure of the lung's ability to stretch and expand by measuring the compliance (Compliance = ΔVolume/ΔPressure), which is assessed as Static lung compliance (Cchord) (change in volume for any given applied to pressure point of curve from 0 to +10 Cm H2O) (Fig 5C) and Dynamic lung compliance (Cdyn), which is the compliance of the lung at any given time during actual movement of air (Fig 5D); (iii) Forced Expiratory Volume at 100 msec (FEV100), which is the volume exhaled during the first 100 milliseconds of a forced expiratory maneuver started from the level of Total Lung Capacity (it is the standard index for assessing and quantifying airflow movement into the lungs); and (iv) Lung resistance, which is the resistance of the respiratory tract to the airflow movement during normal inspiration and expiration, where Rl = [(Atmospheric Pressure–Alveolar Pressure)/V] (Fig 5F).
By using Forced Pulmonary Maneuvers, we observed that single or multiple Ascaris infections in mice caused loss of respiratory area induced by aedema and lung septa thickening, as indicated by significant reduction in Functional Vital Capacity (Fig 5A) and Inspiratory Capacity (Fig 5B) values when compared to respective controls. Concerning the assessment of pulmonary elasticity by lung compliance and resistance analysis, we detected that infected animals displayed reduced Chord Compliance (Fig 5C) and Dynamic Compliance (Fig 5D) indicating modification in pulmonary extracellular matrix components. Moreover, infected mice presented alterations in respiratory airway flow 8 days after single or multiple infections with decreased Forced Expiratory Volume at 100 milliseconds (FEV100) when compared to controls (Fig 5E), suggesting that lung injury induced by Ascaris reduced the airflow into airways. A progressive elevation in Lung Resistance according to the number of infections (single or three infections) was detected (Fig 5F), indicating loss of pulmonary elasticity in both groups, but more pronounced in reinfected mice. Together, our data shows that infected animals showed alteration in airway flow (Fig 5C), loss of respiratory area (Fig 5A and 5B) and reduction of tissue elasticity (Fig 5D–5F) induced by worm migration into airways and subsequent tissue damage. Ultimately, our data suggests that altered lung functions may occur by induction of chronic lung injury and immune responses against Ascaris spp.
The analyses of leukocytes in the BAL of single- and reinfected mice on the 8th day post-infection indicated airway haemorrhage compared with the non-infected animals (Fig 6). The presence of bleeding was higher in the single exposured animals than in reinfected group, which is consistent with the significantly higher haemoglobin levels observed in the BAL of single-infection group animals (Fig 6A), also associated with worm influx into airways (Figs 2C and 6B). Moreover, the levels of total protein in BAL were higher in the single-infection group (Fig 6C), which might also be related to increased worm transmigration and haemorrhage in BAL. In contrast, BAL of reinfected animals presented a substantial increase in the number of total leukocytes (Fig 6D).
The evaluation of differential cell counting in the BAL reinforced the notion that reinfection induces a chronic lung inflammation as observed by increased number of granulocytes in airways, and also significant augmentation of cells from adaptive immunity as phagocytes and lymphocytes, as depicted by cytospin preparations, when compared to the remaining groups (Fig 6E–6H). Our data suggest that single-infection induces acute tissue damage, followed by haemorrhage and exudation related to increased worm transmigration into airways, while reinfection might elicit a pulmonary immune response against Ascaris spp. resulting in a decreased number of worms present in the airways.
To elucidate the immunopathological mechanisms involved in protection against reinfection, peripheral blood from both single- and reinfected animals was collected on days zero, 4, 8 and 12 post-infection and the systemic immune responses were evaluated. Differences in the time of infection (0, 4, 8 and 12 days post-infection), the type of infection (single or multiple infection) and the interaction between these two factors were evaluated in all groups. Both factors (time and type of infection) contributed to the observed differences in the count of circulating monocyte and eosinophils, which were significantly augmented in the reinfected groups according to the progression of infection until the 8th day of infection (where the peak of the parasitism in the lungs is reached).
As observed in the BAL (Fig 6), an increased number of leukocytes were detected in the blood (Fig 7). Remarkably, significant differences were observed in the number of circulating lymphocytes on the 4th day of reinfection (Fig 7A) and monocytes on the 4th and 8th days of reinfection (Fig 7B), which were higher in the reinfected animals when compared to the single infection and control animals. While the number of circulating neutrophils was significantly higher in the reinfected animals on the 4th day of infection (Fig 7C), the eosinophil counts were increased in the 12th day post-infection in the single infected group and on all evaluated days to the animals that received three experimental infections (Fig 7D).
Concerning the systemic cytokine profile during reinfection (Fig 8), production of Th2 cytokines were detected only after a third exposure to the parasite, with significant production of IL-4 on day 0 and IL-5 on days 4th and 8th of the study (Fig 8A and 8B, respectively). Higher levels of IL-10 were detected on day -14 (after second exposure to A. suum infection) and were sustained until the end of the study (Fig 8C). Production of inflammatory cytokines was also observed, with detection of significantly higher production of IL-6 from the 4th day after the third infection (Fig 8D). Of note, the most striking finding demonstrated that reinfected mice presented higher levels of IL-17A when compared with single-infected and control animals (Fig 8E).
Finally, while production of IFN-γ was significantly higher in the reinfected group only at the late stage of the third infection (12th days) when compared to single-infected and control animals (Fig 8F), TNF-α production followed the IL-17A profile, where reinfected animals presented higher cytokine production from the 4th day after the multiple exposure to the parasite (Fig 8G). IL-2 production was also higher in the reinfected group compared to the single- and non-infected group at the -14 and 0 day of time points (Fig 8H).
Recently, Gazzinelli-Guimarães et al. [10] using the BALB/c mouse strain have characterized the full pattern of A. suum larval migration and highlighted the immunopathological changes in lung tissue triggered by the larvae during a primary larval ascariasis. In the present work, the parasitological and immunological aspects of Ascaris spp. infection in mice were evaluated, comparing single and multiple infections and focusing on the possible mechanisms that control the protection against larval ascariasis. Indeed, several animal studies have shown that prior exposure to helminths can induce protection against reinfection to A. suum [25], Strongyloides ratti [26], Neodiplostomum seoulensis [27, 28], Clonorchis sinensis [29] and S. stercoralis [30], amongst other helminths. The current study also provides strong evidence that multiple exposures to A. suum induces partial protection during larval migration through lung tissue and particularly corroborate previous findings in calves where a decrease in the number of larvae was observed after a second exposure to the parasite [25].
Larval ascariasis starts precisely after Ascaris sp. larvale hatch in the host’s small intestine and it is characterized by progressive larval migration through the large intestine mucosa, blood circulation, liver, lungs and airways and finally back to the small intestine where they mature as adult worms, leading to a chronic and a long-term infection. Recently, an important anti-helminthic role played by eosinophils in intestine mucosal defense against invading A. suum larvae was demonstrated [31]. However, the mechanisms of protection developed by the host during hepato-tracheal migration of A. suum larvae were not fully elucidated. In this study, it was shown that pulmonary immune responses represent a crucial role to prevent reinfection, and the number of migrating larvae was reduced consistently only when the larvae reach the lungs and airways.
Associated with the control of larval migration, the multiple exposures to A. suum elicited an increased cellularity in the BAL determined by adaptive immune cells (lymphocytes and macrophages) and an intense eosinophilic and neutrophilic pulmonary inflammation that might ultimately lead to a severe impairment of the respiratory function. Moreover, the multiple exposures to Ascaris, apart from the expansion in the number of circulating inflammatory cells (mainly eosinophils and neutrophils), also induced a polarized Th2/Th17 lymphocyte response defined by higher levels of systemic cytokine production of IL-4, IL-5, IL-10, IL-6, TNF-α and IL-17A in comparison to single-infected animals.
Animals with multiple exposures to the parasite exhibited peripheral eosinophilia at all of the evaluated (0, 4, 8 and 12) time points, a significant increase of eosinophils in bronchoalveolar lavage fluid and higher EPO activity in the lung on the eighth day after the last infection. Moreover, the inflammatory infiltrate in the lungs was composed primarily of eosinophils and they were more prominent in the reinfected group. While an increased activity of eosinophils in mice reinfected with S. stercoralis was not directly associated with destruction of parasitic larvae [30], in our study we observed that a remarkable rise in count and activity of eosinophils was followed by a substantial reduction in the parasitic burden, suggesting the importance of eosinophils in the clearance of Ascaris infection [32]. On the other hand, the presence of eosinophils would be implicated in tissue repair and remodeling due to the extensive lung injury and hemorrhage associated with migration of larvae into airways, which lead to alveolar edema and robust inflammation and to further changes in pulmonary functions detected by spirometry. Of note, the reinfection induced larger areas of lung when compared with animals from the single-infection group, suggesting also that multiple exposures could lead to repeated tissue injury and chronic inflammation, causing progressive loss of respiratory function. The paradoxical patterns of immune response in lungs of reinfected animals, where significant inflammatory responses and decrease of parasitic burden are observed, suggest the possible beneficial effect of inflammation for helminth protection during pulmonary migration. The controversial contribution of the inflammatory response to the control of parasitic burden has been demonstrated in animal models of susceptibility or resistance to a single exposure to the parasite [33, 34]. Severe pulmonary inflammation was considered not important to the control of A. suum infection but directly associated to the tissue repairing induced by larval migration [33]. Of note, inflammatory responses in the liver were associated with control of parasitic infection as observed in animals resistant to A. suum infection [34]. Some cytokines released by eosinophils such as TNF (a hallmark of inflammatory response) could also be implicated, among other important physiological functions, in tissue remodelling [35]. Interestingly, some previous studies have demonstrated that TNF plays a key role in the expulsion of helminth T. muris while working synergistically with a Th2 response, especially IL-13 [36, 37].
Models of resistance and susceptibility have been studied for gastrointestinal helminths. Usually, protection against reinfection is mediated by the Th2-type response, where as susceptibility is associated with a pro-inflammatory response [37, 38]. Our data demonstrated that reinfected mice showed an increased production of IL-17A, IL-6, TNF-α, IL-4 and IL-5, suggesting a pattern of mixed Th2/Th17 responsiveness, which was previously observed in studies with helminths [8–10, 39, 40] and also allergic disorders, such as rhinitis and asthma [41]. The production of IL-10 in the reinfected group as previously demonstrated during T. muris and S. mansoni infections [42], however, might be possibly related to immunomodulation of Th1 or Th17 inflammatory responses [42–44]. Indeed, the increased levels of IL-17A in multiple exposures to A. suum might reflect the intense and chronic inflammation in the tissue remodeling, as is well demonstrated in a model of pulmonary fibrosis [45]. Among the pleiotropic mechanisms of action, this cytokine acts on neutrophil recruitment and helps in elimination of bacteria associated with the surface of the parasite and also in phagocytosis of cellular debris present in the areas of necrosis [46]. Moreover, IL-17A contributes to granulomatous inflammatory and fibrosing reactions in animals infected with S. japonicum [47] or animals continuous exposured to Aspergillus fumigatus [48].
It is noteworthy that reinfected animals exhibited significantly higher levels of IL-4 until the beginning stage (fourth day) of the last infection. As the increased expression of receptors for IL-4 induces the expression of IL-10 [46], the combination of these two cytokines is crucial for the control of wound damage caused by migration of Nippostrongylus braziliensis larvae through the organs of the host. Thus, IL-10 might control inflammation as IL-4 may mediate tissue healing by promoting the response from macrophages and eosinophils [42, 43, 46, 49]. Moreover, while previous studies demonstrated that IL-4 and IL-10 have inhibitory effects on IL-17A [30, 40], the presence of systemic production of these cytokines in animals that were repeatedly exposured to A. suum was not yet clear.
The coexistence of a mixed Th2/Th17 in reinfected mice with increased levels of IL-4 and IL-17A after each infection might also be associated to tissue injury and remodeling. Indeed, it has been shown that these cytokines have a role in development of lung fibrosis in response to chronic tissue injury [45, 50]. Morever, the robust Th2/Th17 response may reduce larval burden in multiple exposures, thus playing a dual role by acting direct on leukocytes to induce aspecific immune response against Ascaris and by inducing tissue thickening and fibrosis that frustrate worm migration. As observed in the histopathological analysis, the repeated exposure to the parasite could cause large areas of damage in the liver and lungs with concurrent increased host inflammatory response by eosinophis and neutrophils and hence a need for regulation of immune responses. Interestingly, the histophatological analysis showed that single-infected mice presented higher levels of hemoglobin and proteins in the BAL, even when the same animals had less damage to areas in pulmonary tissue. Possibly, during the first parasitic infection, when the protective response observed after multiple exposures are not yet effective, a higher number of Ascaris larvae migrates through the lungs and airwarys aggravating the destruction of capillary vessels and causing tissue damage, exudation and extravasation of proteins in the BAL.
Associated with increased tissue damage and exudation caused by larval migration, substantial changes in pulmonary physiology such as loss of pulmonary volume, airway flow and elasticity were observed as a consequence of intense parenchymal injury and aedema in single-infected mice. After multiple exposures to Ascaris, the persistence of physiological modulation and the chronic and repetitive lung parenchymal injury with intense eosinophilic immune responses are consistent with human larval ascariasis associated with Loeffler's syndrome/eosinophilic pneumonitis [51, 52]. Our data suggests the development of pulmonary fibrosisas a cumulative effect of larval ascariasis and eosinophil persistence in tissue, proceding to restrictive and non-reversible lung disease. Moreover, the injury in pulmonary parenchyma and tissue remodeling by fibrogenesis might lead to a progressive increase in pulmonary resistance, detected either in single or multiple infections, as a consequence of tissue scarring induced by multiple larval migrations into airways and anti-helminthic Th2/Th17 and pro-fibrogenic immune responses.
Taken together, our data indicate that, after multiple larval ascariasis, the host is able to mount a protective response against reinfection. Such a finding may explain the worldwide epidemiological distribution of Ascaris given the majority of infected individuals in areas of high endemicity suffer only low parasitic burden. Furthermore, this study suggests the intense systemic and pulmonary inflammatory responses that occur after repeated exposures might be fundamental to induction of protection. However, in the same scenario, the intense airway and lung inflammation that is triggered to control larval migration may be responsible for respiratory malfunction, possibly even asthma, in a host that has been multiply exposed to the parasite. This hypothesis might explain instances of seasonal eosinophilic pneumonitis and asthma among Saudi Arabs exposed to Ascaris sp. larvae [53]. While data from our study might explain the differences in indicators of inflammation in putatively immune (probably comprising multiply exposed individuals) versus susceptible (likely suffering the first exposure) children [54], further studies are required to provide more evidence about the biology of the interaction between Ascaris and the host, focusing mainly focused on elucidation of the immune mechanisms and pathways of protection that are triggered to control larval ascariasis burdens and tissue damage. This could enable the development of new strategies to prevent or treat Ascaris infection.
|
10.1371/journal.pbio.1002440 | Kif13b Regulates PNS and CNS Myelination through the Dlg1 Scaffold | Microtubule-based kinesin motors have many cellular functions, including the transport of a variety of cargos. However, unconventional roles have recently emerged, and kinesins have also been reported to act as scaffolding proteins and signaling molecules. In this work, we further extend the notion of unconventional functions for kinesin motor proteins, and we propose that Kif13b kinesin acts as a signaling molecule regulating peripheral nervous system (PNS) and central nervous system (CNS) myelination. In this process, positive and negative signals must be tightly coordinated in time and space to orchestrate myelin biogenesis. Here, we report that in Schwann cells Kif13b positively regulates myelination by promoting p38γ mitogen-activated protein kinase (MAPK)-mediated phosphorylation and ubiquitination of Discs large 1 (Dlg1), a known brake on myelination, which downregulates the phosphatidylinositol 3-kinase (PI3K)/v-AKT murine thymoma viral oncogene homolog (AKT) pathway. Interestingly, Kif13b also negatively regulates Dlg1 stability in oligodendrocytes, in which Dlg1, in contrast to Schwann cells, enhances AKT activation and promotes myelination. Thus, our data indicate that Kif13b is a negative regulator of CNS myelination. In summary, we propose a novel function for the Kif13b kinesin in glial cells as a key component of the PI3K/AKT signaling pathway, which controls myelination in both PNS and CNS.
| Myelin is a multilayered extension of the Schwann and oligodendrocyte cell membranes, which wraps around neuronal axons to facilitate propagation of electric signals and to support axonal metabolism. However, the signals regulating myelin formation and how they are integrated and controlled to achieve homeostasis are still poorly understood.
In Schwann cells, the Discs large 1 (Dlg1) protein is a known brake of myelination, which negatively regulates the amount of myelin produced so that myelin thickness is proportional to axonal diameter. In this paper, we report that in Schwann cells Dlg1 itself is tightly regulated to ensure proper myelination. We propose that Dlg1 function is further controlled by the Kif13b kinesin motor protein, which acts as a "brake of the brake" by downregulating Dlg1 activity. Surprisingly, we found that in oligodendrocytes Dlg1 is a positive and not a negative regulator of myelination. Thus, Kif13b-mediated negative regulation of Dlg1 ensures appropriate myelin production and thickness in the central nervous system. Our data further extend recently emerged unconventional roles for kinesins, which are usually implicated in cargo transport rather than in the modulation of signaling pathways. The elucidation of mechanisms regulating myelination may help to design specific approaches to favor re-myelination in demyelinating disorders in which this process is severely impaired.
| Myelination is a multistep process that includes axon recognition and contact, ensheathment, and myelin biogenesis. In this process, discrete sets of proteins and lipids are specifically assembled to generate and maintain distinct structural and functional domains necessary for nerve function [1–5]. During myelination, positive and negative regulators must be tightly controlled so that myelin thickness is strictly proportional to axonal diameters. However, the molecular mechanisms that promote and regulate myelination as well as the molecular machineries responsible for the transport and targeting of vesicles during myelin biogenesis are largely unknown. For example, Kif1b is the only motor protein identified thus far implicated in central nervous system (CNS) myelination in Danio rerio (zebrafish) [6].
We previously reported that in Schwann cells the Kif13b motor protein (also known as guanylate kinase-associated kinesin [GAKIN] in humans) is part of a complex that titrates membrane formation during Schwann cell myelination [7]. We found that Kif13b interacts with the Discs large 1 (Dlg1) scaffold in Schwann cells and that the downregulation of either Kif13b or Dlg1 expression in Schwann cell/dorsal root ganglia (DRG) neuron co-cultures decreases myelination in vitro [7]. Another study independently reported that Dlg1-silenced Schwann cells in vitro showed migration defects and reduced expression of the polarity protein Par3 [8]. Occasionally, silenced cells overcame their migration defect and myelinated, but the resulting myelin segments were thicker than those of controls, which indicated Dlg1 as a negative regulator of myelin sheath thickness [8]. This role was further assessed in vivo, as we and others subsequently reported that mouse nerves lacking Dlg1 expression specifically in Schwann cells have hypermyelination, myelin outfoldings, and demyelination as a consequence of myelin instability [8,9]. Dlg1 is thought to act in complex with phosphatase and tensin homolog (PTEN) to reduce AKT (v-AKT murine thymoma viral oncogene homolog) activation; thus, it is a brake on myelination [8].
Kif13b kinesin is a plus end motor protein that mediates the transport of several cargos in polarized cells [10–16]. In PC12 cells, Kif13b negatively regulates centaurin-α1/PIP3BP (phosphatidylinositol-3,4,5-trisphosphate binding protein), a GTPase activating protein (GAP) for Arf6 (ADP-ribosylation factor) GTPase and promotes Arf6 plasma membrane activation [16]. In neurons, Kif13b transports centaurin-α1/PIP3BP and PIP3 to the tip of neurites to promote neuronal polarity [11].
To further investigate the function of the Kif13b/Dlg1 complex in myelination in vivo, we generated a novel Kif13b floxed allele and conditional knock-out mouse models with specific ablation of Kif13b or Dlg1 in either Schwann cells or oligodendrocytes. Here, we report that Kif13b has opposite roles in the control of myelination in the peripheral nervous system (PNS) and CNS. Our data indicate that in Schwann cells, Kif13b interacts with p38γ mitogen-activated protein kinase (MAPK) to promote phosphorylation and ubiquitination of Dlg1. Consistent with this observation, loss of Kif13b results in reduced levels of p38γ MAPK, increased Dlg1 expression, and reduced myelin thickness.
Finally, we report that Kif13b also controls Dlg1 function in oligodendrocytes by promoting its negative regulation. However, our data indicate that, in contrast to Schwann cells, Dlg1 does not reduce but rather enhances AKT activation in oligodendrocytes. Thus, Kif13b is a novel negative regulator of CNS myelination.
We previously reported that, in the peripheral nerve, Kif13b is mainly detected in cytosolic compartments of myelin-forming and non-myelin-forming Schwann cells [7]. To investigate the role of Kif13b in Schwann cells in vivo, we generated a Kif13bFloxed (hereafter, Kif13bFl) allele in which exon 6 was flanked by lox-P sites. Using the Cre/loxP technology, excision of exon 6 produces a frameshift leading to the introduction of a premature stop codon and to nonsense-mediated mRNA decay (Fig 1A–1C). To ablate Kif13b specifically in Schwann cells, we generated Kif13bFl/Fl P0-Cre mice, in which the myelin protein zero (MPZ) promoter drives Cre recombinase expression specifically in Schwann cells, starting from E13.5 [17,18]. Deletion of exon 6 was documented by PCR analysis on DNA from the sciatic nerve, where a recombination band of 378 bp was specifically detected (Fig 1D). Kif13b protein expression was ablated in sciatic nerve lysates from Kif13bFl/Fl P0-Cre mice, thus also confirming that Kif13b is mainly expressed by Schwann cells in the nerve (Fig 1E).
We then analyzed Kif13bFl/Fl P0-Cre sciatic nerves starting at P30 by performing semithin section and ultrastructural analyses. In mutant nerves, we noted a higher number of fibers displaying Schwann cell nuclei and the surrounding cytoplasm, suggesting a shorter internodal length (Fig 2A). Consistent with this, we found that Kif13bFl/Fl P0-Cre quadriceps nerves had indeed a higher percentage of fibers with shorter internodes, particularly in the range between 500 and 600 μm (Fig 2B). Cajal bands are cytoplasmic channels located at the abaxonal surface of myelinated fibers and are involved in the biosynthesis and assembly of myelin [3]. Ablation of the Schwann cell protein Periaxin disrupts Cajal bands and is also associated with reduced longitudinal growth of Schwann cells [19]. However, subsequent work from the same group has shown that loss of Cajal bands as a result of Drp2 ablation causes focal hypermyelination and concomitant demyelination [20]. We analyzed Kif13bFl/Fl P0-Cre quadriceps nerves, but we did not detect major differences in Cajal band structures between mutant and control nerves (Fig 2C). Our findings are consistent with the view that the longitudinal growth of Schwann cells does not correlate with Cajal band integrity [20]. As myelin thickness is proportional to axonal diameter and internodal length [3], we evaluated myelin thickness in Kif13bFl/Fl P0-Cre nerves by performing ultrastructural analysis. By measuring the g-ratio—the ratio between axonal diameter and fiber diameter—we observed reduced myelin thickness in mutant quadriceps nerves at P30, which displayed increased g-ratio values as compared to controls (ultrastructural analysis, Kif13bFl/Fl P0-Cre, 0.75 ± 0.008, 575 fibers; Kif13bFl/+, 0.70 ± 0.009, 516 fibers, n = 3 animals per genotype, p = 0.03). At P20, myelin thickness was normal in Kif13bFl/Fl P0-Cre nerves, suggesting that myelination is not delayed in this mutant (ultrastructural analysis, g-ratio values: Kif13bFl/Fl P0-Cre, 0.715 ± 0.007, 400 fibers; Kif13bFl/+, 0.71 ± 0.004, 403 fibers, n = 3 animals per genotype, p = 0.49). Reduced myelin thickness was still present in nerves of older mice at 8 mo, as g-ratio values were increased in mutant nerves (Fig 2D).
Finally, following crush nerve injury, remyelinating Kif13bFl/Fl P0-Cre nerves also displayed thinner myelin (S1 Fig).
In conclusion, our data indicate that loss of Kif13b specifically in Schwann cells affects longitudinal and radial myelin growth during development and remyelination after injury. Of note, myelination is not delayed in Kif13b mutant nerves at early stages of development, suggesting that Kif13b-mediated regulation occurs only during active myelination after P20.
To investigate the molecular basis of the observed myelin phenotype, we looked at the expression level of Dlg1, a known interactor of Kif13b and a negative regulator of Schwann cell myelination in vivo [7–9]. Interestingly, we found that Dlg1 expression level was increased in Kif13bFl/Fl P0-Cre nerves at P20 (Fig 3A). Note that the increase is particularly evident in the lower band (Fig 3B), which corresponds to a hypo-phosphorylated isoform of Dlg1 [21,22]. In contrast, Dlg1 mRNA levels were downregulated in mutant nerves (Fig 3C), which suggested that Dlg1 protein was more stable in the absence of Kif13b. To assess whether other negative regulators could contribute to the observed effect on myelination, we also looked at Ddit4/REDD1 expression levels. Ddit4/REDD1 is a known negative regulator of myelination, which downregulates the mechanistic target of rapamycin (mTOR) pathway by activating the tuberous sclerosis complex TSC1/2 [9]. We found that Ddit4 was similarly expressed between wild-type and mutant nerves at P10 and P20 (Fig 3D).
Dlg1 interacts with Kif13b in Schwann cells and is known to potentiate PTEN phosphatase activity on PIP3, thus downregulating AKT activation [7–9]. Consistent with this, phosphorylation of AKT at S473 was decreased in Kif13bFl/Fl P0-Cre nerves as compared to controls at P20, when AKT phosphorylation starts to decline during postnatal nerve development (Fig 3E) [9]. On the contrary, in Kif13bFl/Fl P0-Cre nerves, phosphorylation of AKT at T308 was not significantly different from controls (Fig 3F). This finding may indicate activation of the feedback loop involving mTOR and molecules upstream of PI3K, as also already observed in other mutants [9,23–26].
Finally, we found normal expression levels of NRG1 type III (and the phosphorylation of its receptor ErbB2), Krox20, and Oct6, known regulators of myelin initiation, further supporting that reduced myelin thickness of Kif13bFl/Fl P0-Cre nerves is associated with enhanced negative regulation of postnatal myelination and not with a delay in myelin program initiation (S2 Fig).
Phosphorylation is known to modulate protein–protein interactions necessary for the cytoskeletal localization of Dlg1 [27,28]. In particular, serine phosphorylation correlates with Dlg1 inactivation, and hyperphosphorylated Dlg1 interacts with ubiquitin ligases, which mediate its ubiquitination and degradation [8,21,22,27–30]. Thus, we hypothesized that increased Dlg1 protein levels in Kif13bFl/Fl P0-Cre nerves could result from reduced serine phosphorylation and/or ubiquitination. By immunoprecipitating Dlg1 from sciatic nerve lysates at P20, we observed a decrease of Dlg1-serine phosphorylation in Kif13bFl/Fl P0-Cre nerves compared to controls (Fig 4A). As expected, in Dlg1Fl/Fl P0-Cre sciatic nerve lysates, the phosphorylated band was not detected (Fig 4B). Then, to evaluate whether the decrease in serine-phosphorylation correlated with increased stability, we determined the pattern of Dlg1 ubiquitination. Consistent with our hypothesis, by immunoprecipitating Dlg1 from Kif13bFl/Fl P0-Cre nerves at P4 and P10, we found that Dlg1 was less ubiquitinated in mutant nerve lysates when compared to controls (Fig 4C).
Our data suggest that the hypomyelination in Kif13bFl/Fl P0-Cre nerves results from increased Dlg1 stability/activity and enhanced negative regulation of AKT. Hence, we hypothesized that 50% reduction of Dlg1 gene expression in the Kif13bFl/Fl P0-Cre background might rebalance Dlg1 levels and rescue the phenotype. Thus, we generated Kif13bFl/Fl//Dlg1Fl/+; P0-Cre mice, and we compared these mutants with Kif13bFl/Fl//Dlg1+/+; P0-Cre mouse nerves. By performing western blot analysis, we observed that Dlg1 expression and AKT phosphorylation levels in Kif13bFl/Fl//Dlg1Fl/+; P0-Cre sciatic nerve lysates were rescued at a level similar to controls (Fig 4E and 4F). Accordingly, myelin thickness in Kif13bFl/Fl//Dlg1Fl/+; P0-Cre nerves was also restored (Fig 4D). Overall, our data suggest that Kif13b negatively regulates Dlg1 stability and activity in Schwann cells. Thus, in kif13bFl/Fl P0-Cre nerves, increased Dlg1 activity reduces AKT signaling and myelination.
Since Kif13b regulates PNS myelination, we sought to assess whether Kif13b has a similar role in the CNS. First, we confirmed Kif13b mRNA expression in optic nerves and in myelinated tracts of the corpus callosum (Fig 5A and 5B). Then, we generated a Kif13bFl/- CNP-Cre mouse with conditional inactivation of Kif13b in newly generated oligodendrocytes [31]. To achieve maximum efficiency of CNP-Cre mediated recombination, we generated a compound heterozygous mouse for a Kif13bFl allele and a Kif13b- (null) allele. We first assessed downregulation of Kif13b mRNA expression in Kif13b Fl/- CNP-Cre optic nerves by performing quantitative RT-PCR analysis (Fig 5A). Western blot analysis confirmed reduction of Kif13b protein expression in lysates from corpus callosum of Kif13b Fl/- CNP-Cre mice (Fig 5B). We then performed morphological analysis of optic nerves and spinal cords at P30. Surprisingly, we observed increased myelin thickness with decreased g-ratios in both Kif13bFl/- CNP-Cre optic nerves and spinal cords as compared to either Kif13bFl/+ or Kif13b-/+ controls (Fig 5C and 5D). However, at P90, myelin thickness in either Kif13bFl/- CNP-Cre optic nerves or spinal cords was normal, suggesting a transient effect of Kif13b loss (S3 Fig).
At the molecular level, AKT phosphorylation at S473 was enhanced in both Kif13bFl/- CNP-Cre optic nerves and spinal cords at P30 (Fig 5E), consistent with the observed hypermyelination and the role of AKT in promoting CNS myelination [32]. We then explored whether, as in the PNS, Kif13b regulates myelination by controlling Dlg1 expression levels. First, we assessed whether Kif13b interacts with Dlg1 in oligodendrocytes in vivo. By performing GST pull down assays from rat optic nerve lysates using GST-Kif13b/MBS as a bait, we identified Dlg1, suggesting the existence of a Kif13b/Dlg1 complex (Fig 5F). Interestingly, we noted that in spinal cord and optic nerve lysates Dlg1 isoforms were expressed in the range 140–150 KDa, as already observed in sciatic nerves (Fig 3A), where Dlg1 is not expressed in the axon [7,8]. This finding suggests that the Kif13b/Dlg1 interaction likely occurs in oligodendrocytes and not in axons/neurons, where the main Dlg1/SAP97 isoform runs at a different molecular weight (97KDa). Next, we evaluated Dlg1 protein expression in Kif13bFl/- CNP-Cre mice and we found increased Dlg1 levels in both Kif13bFl/- CNP-Cre optic nerves and spinal cords at P30 (Fig 5G). This result is consistent with the hypothesis that Kif13b negatively regulates Dlg1 expression also in the CNS.
Overall, our findings indicate that Kif13b is a transient negative regulator of myelination in the CNS as its downregulation in oligodendrocytes increases myelin thickness and enhances AKT activation. Moreover, we suggest that also in the CNS Kif13b interacts with Dlg1 and negatively regulates its stability.
In Kif13bFl/- CNP-Cre mice, increased myelin thickness is associated with enhanced Dlg1 expression. However, if Dlg1 acts as a negative regulator of myelination in oligodendrocytes as well, we would have expected to observe hypomyelination and not hypermyelination. Thus, we hypothesized that in oligodendrocytes Dlg1 might have the opposite role in the control of myelination, being a promoter rather than an inhibitor. To test this hypothesis, we generated Dlg1Fl/FlCNP-Cre conditional knockout mice in which Dlg1 was ablated in oligodendrocytes. We first demonstrated a reduction of Dlg1 protein expression in Dlg1Fl/Fl CNP-Cre optic nerves at P30 (Fig 6A). Then, we performed morphological analyses of optic nerves and spinal cords starting at P30. Consistent with our hypothesis, mutant optic nerves and spinal cords displayed reduced myelin thickness and increased g-ratios (Fig 6C and 6D). Hypomyelination was also evidenced by decreased myelin basic protein (MBP) expression levels in spinal cord lysates from Dlg1Fl/Fl CNP-Cre mice (Fig 6B). As in the case of Kif13bFl/- CNP-Cre mutants, myelin thickness of Dlg1Fl/Fl CNP-Cre optic nerves and spinal cords was normal at P90, suggesting a transient role of Dlg1 in the control of myelination (S3 Fig).
To investigate the mechanism by which Dlg1 promotes myelination in oligodendrocytes, we examined the phosphorylation state of AKT in lysates from optic nerves and corpus callosum of Dlg1Fl/Fl CNP-Cre mutants. We found that AKT phosphorylation at both S473 and T308 was reduced in both Dlg1Fl/Fl CNP-Cre optic nerves and corpus callosum as compared to controls, consistent with the decreased myelination (Fig 6E and 6F).
Since (1) AKT phosphorylation depends on PIP3 levels and on the activity of the PI3K class I and (2) Dlg1 has been described to interact with the regulatory subunit of PI3K class I, p85, in epithelial cells [29], we hypothesized that also in oligodendrocytes Dlg1 may interact with p85, influencing PI3K activity upstream of AKT.
To address this point, we first explored p85 expression levels in optic nerves and spinal cords at P30 and found that p85 protein levels were reduced in Dlg1Fl/Fl CNP-Cre mice (Fig 6H–6I'). Next, GST pull down experiments from P11 rat optic nerve lysates demonstrated that Dlg1 and p85 are interactors of GST-Kif13b/MBS (Fig 6G), thus providing evidence for the existence of a complex involving Kif13b, Dlg1, and p85. Interestingly, by performing co-immunoprecipitation and pull down experiments, we did not observe interaction between p85 and the Kif13b/Dlg1 complex in the PNS in sciatic nerves. Consistent with this, p85 was similarly expressed in Kif13bFl/Fl P0-Cre and Dlg1Fl/Fl P0-Cre mutant sciatic nerves as compared to controls (S4 Fig). These findings suggest that in the PNS, in contrast to the CNS, the Kif13b/Dlg1 complex does not involve p85.
In conclusion, similarly to Schwann cells, downregulation of Kif13b expression in oligodendrocytes is associated with increased Dlg1 levels. However, in the CNS, Dlg1 promotes myelination. Thus, downregulation of Kif13b expression in oligodendrocytes causes hypermyelination.
Finally, we asked how downregulation of Kif13b expression results in increased Dlg1 stability in both PNS and CNS. In previous yeast two-hybrid screening analyses, we had found that the PDZ2+3 domain of Dlg1 directly interacts with the p38γ MAPK isoform [7,33], as also previously reported for HEK293 cells [28]. Since p38γ can phosphorylate and negatively regulate the interaction of Dlg1 with cytoskeletal protein partners, we further investigated the interaction of Kif13b, p38γ, and Dlg1 in the nerve in vivo. We first confirmed Dlg1 and p38γ interaction by performing co-immunoprecipitation experiments from sciatic nerve lysates (Fig 7A). Next, we observed that Kif13b/MBS-GST was able to pull down both Dlg1 and p38γ from nerve lysates, suggesting that Kif13b, p38γ, and Dlg1 may be part of the same complex (Fig 7B). To provide further evidence for this hypothesis, we investigated p38γ expression levels in mutants with conditional ablation of either Kif13b or Dlg1 in Schwann cells. Interestingly, p38γ expression levels were decreased in Kif13bFl/Fl P0-Cre sciatic nerves at both P20 and 9 mo (Fig 7C and 7D) but not in Dlg1Fl/Fl P0-Cre nerves (Fig 7E and 7F), suggesting that p38γ acts downstream of Kif13b and upstream of Dlg1. To confirm these results, we analyzed the sciatic nerves of p38γ knock-out mutants. As expected, nerves from p38γ-null mice were hypomyelinated (Fig 7G–7I), supporting the hypothesis that p38γ is a novel promoter of Schwann cell myelination.
Finally, since our data suggest that Kif13b may similarly regulate Dlg1 also in the CNS, we assessed whether a Kif13b, p38γ, and Dlg1 complex could be detected in oligodendrocytes. As expected, Dlg1 and p38γ co-immunoprecipitate from optic nerve lysates (Fig 8A) and Kif13b/MBS-GST is able to pull down both Dlg1 and p38γ (Fig 8B). Even if not as striking as in Schwann cells, p38γ expression levels were decreased in Kif13bFl/- CNP-Cre optic nerve lysates but not in Dlg1Fl/Fl CNP-Cre, suggesting that p38γ acts downstream of Kif13b and upstream of Dlg1 (Fig 8C–8E).
As p38α is the MAPK isoform known to regulate myelination in both PNS and CNS [34–41], we assessed whether Kif13b/Dlg1 may also form a complex with p38α. Interestingly, by performing pull down experiments, we found that Dlg1 does not interact with p38α in either optic or sciatic nerve lysates. Consistent with this, expression levels of p38α in either sciatic nerves or spinal cords of Kif13b conditional knock-out mutants were similar to controls (S5 Fig).
Overall, these findings suggest a similar mechanism of Kif13b and p38γ-mediated regulation of Dlg1 in both PNS and CNS, with opposite outcomes on the control of myelination, as Dlg1 is a brake on myelination in the PNS and a positive regulator in the CNS.
Microtubule-based kinesin motors have many cellular functions, including the transport of a variety of cargos to different parts of the cell [42]. Motors can also be used to place cargos on a long distance, such as signaling complexes or developmental determinants in neurons or embryos, respectively. However, unconventional functions have recently emerged, and kinesins have also been reported to act as scaffolding proteins and signaling molecules [43]. In particular, Kif13b has been recently shown in hepatocytes to work as a scaffold and to enhance caveolin-1 dependent internalization of LRP11 receptor [12]. In T cells, Kif13b acts as a signaling molecule that controls CARD11 scaffold localization at the synapse and downregulates TCR signaling [13]. In this work, we further extend these findings on unconventional roles of kinesins and propose a novel mechanism by which the Kif13b motor protein regulates Dlg1 scaffold activity and titrates the PI3K/AKT signaling with two opposite outcomes in PNS and CNS myelination (Fig 9).
Here we report that downregulation of Kif13b expression in Schwann cells is associated with reduced myelin thickness, decreased AKT activation, and increased levels of Dlg1, a known brake on PNS myelination acting on the PIP3-AKT-mTOR pathway [8,9]. As Kif13b and Dlg1 interact in Schwann cells [7], we hypothesized that Kif13b may control myelination through the Dlg1 scaffold itself, by regulating its stability and function. Indeed, in support of our hypothesis, in Kif13bFl/Fl//Dlg1Fl/+; P0-Cre double mutant sciatic nerves, Dlg1 expression levels and myelin thickness are similar to wild type.
Interestingly, we report here that in Kif13bFl/Fl P0-Cre nerves, in which Dlg1 expression levels are increased, myelination is not delayed at very early stages of postnatal nerve development, and reduced myelin thickness is evident when AKT activation starts to physiologically decline, after P20 [8,9]. This observation is consistent with the phenotype of mutant mice lacking Dlg1, specifically in Schwann cells [9]. We previously reported a transient increase in myelin thickness and occasional myelin outfoldings in Dlg1Fl/Fl P0-Cre nerves starting from P10 [9]. However, even if enhanced, myelination was not accelerated in Dlg1Fl/Fl P0-Cre nerves, and, at very early stages of postnatal nerve development, the number of myelinated fibers and myelin thickness were similar to control nerves. Thus, Dlg1 may act as a brake on myelination to downregulate AKT activation at the peak of myelination, when AKT phosphorylation starts to decline. In support to this hypothesis, myelin outfoldings, a focal form of hypermyelination that is thought to be linked to AKT overactivation and loss of Dlg1-mediated negative control on myelination, are observed in the nerve after 3 w of postnatal development [44].
Dlg1 stability is controlled by phosphorylation and ubiquitination [8,21,22,27,28,30,45,46]. In Drosophila, the PAR1 kinase directly phosphorylates Dlg1 at conserved sites and negatively regulates its mobility and targeting at postsynaptic membranes of neuromuscular junctions [27]. Osmotic stress-induced serine phosphorylation of Dlg1 by p38γ MAP kinase can induce Dlg1 dissociation from the glucokinase-associated dual specificity phosphatase (GKAP) and the cytoskeleton, negatively regulating Dlg1 [28]. Finally, phosphorylated DLG1 interacts with the β-TrCP ubiquitin ligase receptor, which mediates ubiquitination of the protein [30]. Thus, we investigated whether enhanced Dlg1 protein expression levels in Kif13bFl/Fl P0-Cre nerves correlated with a decrease in serine phosphorylation and/or ubiquitination. Consistent with our hypothesis, we found that in Kif13bFl/Fl P0-Cre nerves Dlg1 is less phosphorylated and less ubiquitinated, suggesting that Kif13b promotes radial myelin growth by directly or indirectly influencing Dlg1 stability and expression.
We also suggest that p38γ MAPK could be the kinase that, downstream of Kif13b, phosphorylates Dlg1 to regulate its activity. Indeed, p38γ MAPK is known to interact with and to phosphorylate serine residues of Dlg1 in other cells [28]. We identified p38γ in a yeast two-hybrid screening analysis using a nerve cDNA library and Dlg1 as a bait [7,33]. Moreover, we show that p38γ, Dlg1, and Kif13b form a complex in the nerve. More importantly, sciatic nerves of p38γ-null mice are hypomyelinated, thus confirming the hypothesis that p38γ, by phosphorylating and negatively regulating Dlg1, acts as a promoter of myelination downstream of Kif13b. Unfortunately, antibodies that can specifically recognise phosphorylated p38γ are not available to assess whether activated p38γ could interact with Kif13b and Dlg1.
Interestingly, the role of p38γ MAPK in the regulation of PNS myelination has not yet been assessed. Previous studies suggested that p38 MAPK mediates laminin signaling in vitro to promote Schwann cell elongation and alignment at the very first stages of differentiation [34]. Hossain et al., suggested that p38 directs Schwann cell differentiation by regulating Krox-20 expression, thus further supporting the role of p38 MAPK as a positive regulator of PNS myelination [35]. However, on the basis of the MAPK inhibitors used, the observed effect was likely to be mediated by the p38α or p38β [35]. A more recent study reported that in vitro p38 MAPK promotes the de-differentiated state of Schwann cells during Wallerian degeneration, by inducing c-Jun expression and by inhibiting myelin gene expression, and also suggested that p38 MAPK is a negative regulator of Schwann cell differentiation and myelination during development [36]. On the basis of the antibodies used recognizing the phosphorylated state of MAPK as well as the MAPK inhibitor used (SB203580), other isoforms rather than p38γ are more likely to mediate this function [36].
How can both Kif13b and p38γ control Dlg1 phosphorylation, ubiquitination, and stability? Kif13b could transport and localize the kinase at membranes where Dlg1 is enriched to downregulate, in complex with PTEN, PIP3 levels, and AKT activation [47]. Indeed, in Kif13b-null but not in Dlg1-null nerves p38γ expression levels are reduced, thus suggesting that p38γ is downstream of Kif13b and upstream of Dlg1. Alternatively, the binding of Kif13b with Dlg1, which is mediated by the membrane-associated guanylate kinase homologue binding stalk (MBS) and guanylate kinase homologue (GUK) domains, respectively, may relieve intramolecular inhibition in either Kif13b or Dlg1, as already reported [48]. For example, following Kif13b binding, a conformational change in Dlg1 (open state) can be induced so that target residues for serine phosphorylation can be exposed and accessible to p38γ kinase-mediated phosphorylation. Unfortunately, p38γ-specific inhibitors are not available to further investigate these mechanisms.
Our data convey a novel function for Kif13b/p38γ as negative regulators of Dlg1 in the PI3K/AKT signalling pathway. Interestingly, Kif13b has already been proposed as a negative regulator in other studies. For example, in PC12 cells, KIF13B negatively regulates centaurin-α1/PIP3BP (PIP3 binding protein), a GAP for Arf6, thus promoting Arf6 GTPase plasma membrane activation [16]. Further, in T cells, KIF13B negatively regulates TCR signaling to NF-kB, by redistributing the CARD11 scaffold from the center of the synapse to a more distal region [13].
We also show that Kif13b is a negative regulator of CNS myelination. Indeed, we observed that downregulation of Kif13b expression in oligodendrocytes results in increased myelin thickness and AKT activation, consistently with the role of AKT in promoting CNS myelination [32]. Similar to PNS, we found that Kif13b interacts with Dlg1 and that loss of Kif13b is associated with increased Dlg1 levels, thus suggesting a negative regulation mediated by Kif13b on Dlg1. Given these similarities, we investigated whether the increased Dlg1 level and stability in oligodendrocytes could also result from a decrease in p38γ-mediated phosphorylation. Indeed, we found that Kif13b, Dlg1, and p38γ MAPK interact in optic nerves and that p38γ expression is decreased in Kif13b but not in Dlg1 mutants, as already observed in the PNS. These findings suggest that p38γ may act downstream of Kif13b and upstream of Dlg1 to negatively regulate Dlg1 activity.
The role of the p38γ isoform in the regulation of CNS myelination has not been yet assessed. As for PNS, only p38α has been investigated in the CNS. Inhibition of p38α activity or expression in vitro in a co-culture system has been reported to prevent oligodendrocyte progenitor differentiation and myelination [37–39]. Another study suggested that p38α MAPK supports myelin gene expression in the brain through several mechanisms acting on both positive and negative regulators of differentiation [40]. More recently, myelination was found to be impaired in mice with conditional inactivation of p38α MAPK in oligodendrocyte progenitor cells [41]. Interestingly, the same authors observed an opposite effect of p38α MAPK in remyelination, as mutant mice exhibited a more efficient remyelination as compared to controls following demyelination [41]. These studies further support the notion that the regulation of myelination is a very complicated process, in which different signals arising from the extracellular matrix, axons, and astrocytes in the CNS must be correctly integrated in time and space within the same cell to achieve homeostasis [49–56].
If Dlg1 is a brake on myelination in the CNS as in the PNS, how can loss of Kif13b and elevation of Dlg1 result in increased CNS myelin thickness? Surprisingly, our data indicate that in oligodendrocytes Dlg1 is a positive and not a negative regulator of myelination, as its loss is associated with reduced myelin thickness and AKT activation. Interestingly, in addition to Dlg1, other molecules have been found to control myelination with opposite roles in PNS and CNS [57–60]. For example, myosin light chain II phosphorylation promotes myelination in the PNS and inhibits myelination in the CNS [57].
To determine the mechanism by which Dlg1 could promote CNS myelination acting on the PI3K-AKT pathway, we sought to investigate the regulatory subunit of PI3K class I, p85, a known interactor of Dlg1 in epithelial cells [29]. Consistent with this, we found that Dlg1 interacts with p85 in the optic nerve, likely to modulate PI3K class I activity, PIP3 levels, and ultimately AKT activation. Interestingly, phosphorylation of DLG1 on serine and threonine is known to prevent DLG1 interaction with SH2 domains of p85/PI3K [29]. Thus, we could speculate that Dlg1, when hypophosphorylated, may display a higher affinity for the SH2 domains of p85, whose activation is necessary for PI3K activity regulation [61].
Whether in oligodendrocytes Dlg1 also promotes myelination by other mechanisms, which can converge on AKT activation, remains to be determined.
The following primary antibodies were used: mouse anti-KIF13B (provided by Dr. A. Chishti); mouse anti-Dlg1 (Stressgen); mouse anti-phosphoserine (Alexis Biochemicals); rabbit anti-ubiquitin (Santa Cruz Biotechnology); rabbit anti-DRP2 (provided by Dr. D. Sherman); rabbit anti-phospho-Akt (Ser473) (Cell Signaling); rabbit anti-phospho-Akt (Thr308) (Cell Signaling); rabbit anti-Akt (pan) (Cell Signaling); rabbit anti-phospho-p44/42 MAPK (Erk1/2) (Thr202/Tyr204) (Cell Signaling); rabbit anti-p44/42 MAP Kinase (Cell Signaling); rabbit anti-Neuregulin-1α/β1/2 (C20) (Santa Cruz Biotechnology); rabbit anti-p-Neu (Tyr 1248)-R (i.e., p-ErbB-2) (Santa Cruz Biotechnology); rabbit anti-Neu (C-18) (i.e., ErbB-2) (Santa Cruz Biotechnology); rabbit anti-PI3 Kinase p85 (Cell Signaling); rat anti-MBP (Millipore); rabbit anti-p38α (Santa Cruz); rabbit anti-p38γ (R&D Systems); rabbit anti-calnexin (Sigma-Aldrich); mouse anti-β-tubulin (Sigma-Aldrich); rabbit anti-actin (Sigma-Aldrich). For immunofluorescence, secondary antibodies included fluorescein (FITC)-conjugated and rhodamine (TRITC)-conjugated donkey anti-mouse or rabbit IgG (Jackson ImmunoResearch). For western blotting, secondary antibodies included horseradish peroxidase (HRP)-conjugated goat anti-rabbit and rabbit anti-mouse immunoglobulins (Dako), and IRDye 800- and 680-conjugated goat anti-mouse, goat anti-rabbit, and goat anti-rat IgG (Li-Cor Biosciences). As negative control in immunoprecipitation experiments, ChromPure mouse IgG whole molecules were used (Jackson ImmunoResearch).
All experiments involving animals were performed in accordance with Italian national regulations and covered by experimental protocols reviewed by local Institutional Animal Care and Use Committees.
The pFlrt-1 vector, including lox-P sites, FRT-flanked neomycin resistance gene (neo), and PGK-TK, was used to target the Kif13b gene. The selected Kif13b mouse genomic regions to be inserted in the targeting vector were amplified from a BAC clone spanning the Kif13b gene and obtained from The Center for Applied Genomics (The Hospital for Sick Children, Ontario, Canada).
To generate the targeting vector for homologous recombination, a 503 bp BamHI fragment including exon 6 and flanking intronic regions was first inserted between lox-P sites in pFlrt-1. In a second step, a 4,606 bp fragment containing exon 7 was inserted into the BstBI site downstream of the PGK-neo cassette to constitute the long arm for homologous recombination. Finally, a fragment of 2,000 bp containing exon 5 was cloned into the SalI site upstream to the first lox-P and represented the short arm for homologous recombination.
After electroporation of TBV2 embyonic stem cells (129S2/SvPas), recombinant clones were screened by Southern blot analysis. Digestion with KpnI and hybridization with two probes designed on exon 6 (inside the recombination) and upstream of exon 5 (outside the 5′ end of the recombination) revealed two bands of 7,671 bp (wild type) and of 9,671 bp (containing the neo cassette).
Similarly, SmaI digestion of genomic DNA and hybridization using a probe designed at the 3′ end of the targeted region, outside the recombination boundaries, detected two bands of 7,694 bp (the targeted allele, since one SmaI restriction site is present within the neo cassette) and of 9,151 bp (the wild-type allele).
Two different correctly targeted clones were injected into C57BL6 blastocysts (Core Facility for Conditional Mutagenesis San Raffaele/Telethon Transgenic Service) to obtain transmission of the Floxed allele through the germline.
The neo cassette was removed in vivo by crossing heterozygous Kif13bFl (neo)/+ with Flpe transgenic mice. Heterozygous Kif13bFl/+ animals were crossed with P0-Cre [17,18] transgenic mice to excise exon 6 specifically in Schwann cells. To generate Kif13bFl/Fl P0-Cre conditional knockout mice, Kif13bFl/+ P0-Cre animals were crossed with homozygous Kif13bFl/Fl. Kif13bFl/Fl mouse nerves had normal myelin thickness and mean g-ratio values similar to wild-type mice, thus suggesting that Kif13bFl/+ does not represent a hypomorphic allele.
To obtain Kif13bFl/- CNP-Cre [31] mice with conditional inactivation of Kif13b in oligodendrocytes, Kif13bFl/+ mice were first crossed with CMV-Cre transgenic mice. Then, after germline segregation of the CMV-Cre transgene, Kif13b -/+ (without CMV-Cre) were crossed with Kif13bFl/+ CNP-Cre mice to obtain Kif13bFl/- CNP-Cre conditional null. In this way, we increased CNP-Cre mediated recombination efficiency on the Floxed allele in the Kif13bFl/- CNP-Cre genotype.
The Dlg1Fl (C57/BL6 strain) allele has been already reported (Zhou et al., 2008). To generate Dlg1 conditional knockout mice in oligodendrocytes, homozygous Dlg1Fl/Fl mice were crossed with heterozygous Dlg1Fl/+ mice carrying the CNP-Cre transgene.
To obtain 50% reduction of Dlg1 specifically in Schwann cells in a Kif13bFl/Fl P0-Cre background, Kif13bFl/Fl P0-Cre mice were first crossed with Dlg1Fl/Fl P0-Cre mice. Then, Kif13bFl/+//Dlg1Fl/+; P0-Cre double heterozygous mice were crossed to obtain Kif13bFl/Fl//Dlg1Fl/+; P0-Cre mice. These latter were compared with Kif13bFl/Fl//Dlg1+/+; P0-Cre mice and controls (only floxed alleles without Cre) within the same litters.
The generation of p38γ-null mice has been already reported [28].
For all the experiments involving animals, n ≥ 5 animals per genotype of either sex were analysed.
Genotype analysis on Kif13b mutant mice was carried out on tail genomic DNA using primer pairs A plus B (415 bp floxed band and 342 bp wild type band) or A plus C (966 bp floxed band, 834 bp wild type band, and 378 bp recombined band). Genotype analysis of the Dlg1 floxed allele and of the p38γ-null locus has already been reported [9,28].
RT-PCR was performed as described previously [7,9]. Designed probes were used to amplify mouse Kif13b and the endogenous reference transcript calnexin. The comparative Ct method was used. As calibrator, a control sample ΔCt was chosen for each selected transcript. The ΔΔCt (ΔCt of each normalized selected transcript minus ΔCt of the calibrator) was calculated. Expression levels of Kif13b mRNA are indicated as 2-ΔΔCt values. For statistical analysis, SD was calculated for triplicate samples of each reaction and SEM is indicated on the average of the determinations from different animals.
Three to five animals per genotype for each time point were analysed. Semithin analysis of quadriceps and sciatic nerves and ultrastructural analysis of optic nerves and spinal cords were performed as described previously [62].
To perform morphometric analysis, digitalized images of fiber cross sections were obtained from corresponding levels of the quadriceps or sciatic nerves with a 100x objective and Leica DFC300F digital camera (Milan, Italy). Five images per animal were analysed with the Leica QWin software (Leica Microsystem) and the g-ratio calculated as the ratio between the mean diameter of an axon (without myelin) and the mean diameter of the same axon including the myelin sheath.
For morphometric analysis on ultrastructural sections, 20 images per animal were taken at 4000x (LEO 912AB Transmission Electron Microscope, Milan, Italy) and the g-ratio values determined by measuring axon and fiber diameters.
Internodal lengths were measured as described using Openlab (PerkinElmer) [19], and 100 internodes of two quadriceps nerves were evaluated for each animal (n = 3).
Adult mice were anesthetized with avertin (trichloroethanol, 0.02 ml/g of body weight), and crush injury was performed as previously described [63]. After skin incision, the sciatic nerve was exposed and crushed distal to the sciatic notch for 20 s with fine forceps previously cooled in dry ice. To identify the site of injury, forceps were previously dropped into vital carbon. The nerve was replaced under the muscle and the incision sutured.
Protein lysates from mouse sciatic nerves, corpus callosum, optic nerves, and spinal cords for western blot analysis were prepared using a lysis buffer containing 2% SDS, 50 mM Tris buffer pH 8.0, 150 mM NaCl, 10 mM NaF, 1 mM NaVO3, and complete protease and phosphatase inhibitors (Roche). For the detection of phosphorylated antigens, samples were lysed with a buffer containing 1%TX-100. Protein quantification was performed using BCA assay (Pierce, Thermo Scientific).
Mouse and rat sciatic nerves were lysed in a buffer containing 1% NP-40, 150 mM NaCl, 50 mM Tris buffer pH 8.0, 10 mM NaF, 1 mM NaVO3, and complete protease and phosphatase inhibitors (Roche). Following centrifugation at 13,000 rpm for 15 min at 4°C, equal amounts of protein lysates were incubated with 6–8 ug of mouse anti-Dlg1 antibody (Stressgen) or mouse IgG for control (Jackson ImmunoResearch). After 3 h of incubation with the antibody at 4°C, 35 μl of protein G agarose (settled) (Sigma-Aldrich) was added to immunocomplexes within the lysates and incubated for 1 h and 30 min at 4°C. The agarose beads were washed two times with cold PBS-Tween 0.1% and once with cold PBS. The immunoprecipitated product was denatured in Laemmly buffer (Biorad) with β-mercaptoethanol and resolved by SDS-PAGE.
Kif13b/MBS cDNA was cloned into pGEX-4T2 expression vector and expressed together with GST alone in Escherichia coli BL21(DE3) cells [7]. Recombinant proteins were purified directly from bacterial extract on glutathione-Sepharose 4 Fast Flow beads.
Rat sciatic and optic nerves were lysed in a buffer containing 1% NP-40, 50 mM Tris buffer pH 7.4, 10% glycerol, 100 mM NaCl, 10 mM NaF, and 1 mM NaVO3. Equal amounts of protein lysates were incubated for 4 h at 4°C with immobilized GST-Kif13b/MBS proteins and GST as control. After three washes with a buffer containing 0.5% NP-40, 50 mM Tris buffer pH 7.4, 10% glycerol, 100 mM NaCl, 10 mM NaF, and 1 mM NaVO3, the bead pellets were dissolved in Laemmly buffer with β-mercaptoethanol, resolved by SDS-PAGE, and analyzed by immunoblotting. To show the relative amount used of GST-Kif13B/MBS and GST, beads were dissolved again in Laemmly buffer with β-mercaptoethanol, resolved by SDS-PAGE, and the gels stained with Coomassie.
SDS-PAGE gels were transferred to PVDF membranes (Millipore) or to nitrocellulose (Millipore) at 4°C in 20% methanol blotting buffer. Filters were blocked in 5% dry milk in PBS-0.1% Tween 20 overnight at 4°C and immunoblotted with primary antibodies diluted in 3% dry milk in PBS-0.1% Tween. For phosphorylated antigens, an additional blocking was performed for 30 min at RT in 3% bovine serum albumin (BSA) (Sigma-Aldrich), 0.5% gelatin, 0.1% Tween, 1 mM EDTA pH 8.0, 0.15 M NaCl, 10 mM Tris buffer pH 7.5, followed by incubation with primary antibodies diluted in the same blocking solution. Secondary antibodies, either horseradish peroxidase-conjugated (Dako) or IRDye 800- and 680-conjugated (Li-Cor Biosciences), were used and immunoblots revealed by using either ECL/ECL-prime developing systems and films for chemiluminescent detection (Amersham) or by Odyssey CLx Infrared Imaging System (Li-Cor Biosciences).
Statistical analysis was performed using the Student t test, two tails, unequal variants, and α = 0.005 were considered. All results are shown as mean ± SEM.
Figures were prepared using Adobe Photoshop version 11.0 (Adobe Systems).
|
10.1371/journal.pntd.0005867 | Increased hepatotoxicity among HIV-infected adults co-infected with Schistosoma mansoni in Tanzania: A cross-sectional study | Little is known about hepatotoxicity in patients with schistosome and HIV co-infections. Several studies have reported increased liver enzymes and bilirubin levels associated with schistosome infection. We investigated whether HIV-infected adults on antiretroviral therapy who had S. mansoni co-infection had a higher prevalence of hepatotoxicity than those without.
We determined the presence and grade of hepatotoxicity among 305 HIV-infected outpatients who had been on medium-term (3–6 months) and long-term (>36 months) antiretroviral therapy in a region of northwest Tanzania where S. mansoni is hyperendemic. We used the AIDS Clinical Trial Group definition to define mild to moderate hepatotoxicity as alanine aminotransferase, alanine aminotransferase, and/or bilirubin elevations of grade 1 or 2, and severe hepatotoxicity as any elevation of grade 3 or 4. We determined schistosome infection status using the serum circulating cathodic antigen rapid test and used logistic regression to determine factors associated with hepatotoxicity. The prevalence of mild-moderate and severe hepatotoxicity was 29.6% (45/152) and 2.0% (3/152) in patients on medium-term antiretroviral therapy and 19.6% (30/153) and 3.3% (5/153) in the patients on long-term antiretroviral therapy. S. mansoni infection was significantly associated with hepatotoxicity on univariable analysis and after controlling for other factors associated with hepatotoxicity including hepatitis B or C and anti-tuberculosis medication use (adjusted odds ratio = 3.0 [1.6–5.8], p = 0.001).
Our work demonstrates a strong association between S. mansoni infection and hepatotoxicity among HIV-infected patients on antiretroviral therapy. Our study highlights the importance of schistosome screening and treatment for patients starting antiretroviral therapy in schistosome-endemic settings. Additional studies to determine the effects of schistosome-HIV co-infections are warranted.
| Schistosoma sp. are parasitic worms that infect at least 218 million people worldwide. Over 90% of these individuals live in Africa, where HIV infection is also endemic. Schistosome worms lay eggs that damage the gastrointestinal and genitourinary tracts, causing extensive morbidity and mortality. Patients who have HIV and Schistosoma mansoni co-infections are at risk for damage to the liver due to both the effects of the schistosome parasite and the side-effects of antiretroviral therapy. However, little is known about the additional liver effects of schistosome infection in patients already taking antiretroviral therapy. Therefore, we conducted a study in northwest Tanzania, where our prior work has shown that approximately one-third of HIV-infected patients also have schistosome infections, to investigate the effect of co-infection with Schistosoma mansoni on liver damage in patients taking antiretroviral therapy. We studied 305 HIV-infected outpatients on medium and long-term antiretroviral therapy and determined both liver damage and S.mansoni infection in those patients. We found that among patients on antiretroviral therapy, those with HIV-schistosome co-infection were 3 times more likely to have liver damage than those with HIV infection alone. Our work shows the importance of screening and treating for Schistosoma mansoni to decrease the risk of liver damage in patients infected with HIV.
| Hepatotoxicity increases mortality and morbidity among HIV-infected patients on antiretroviral therapy (ART), especially those with high CD4+ T-lymphocyte (CD4) counts at initiation of ART [1]. The prevalence of hepatotoxicity due to ART is expected to increase in sub-Saharan Africa as ART becomes more widely available and as patients initiate treatment at higher CD4 counts [2]. Most studies on hepatotoxicity in HIV-infected patients on ART have been conducted in the first 3 months of ART, when the majority of hepatotoxic reactions are believed to occur and when many patients are also receiving anti-tuberculosis treatment [3]. Only a few studies have assessed hepatotoxicity in HIV-infected patients on ART for more than 3 months. Two small studies have shown that both hepatitis B and hepatitis C are associated with hepatotoxicity in HIV-infected patients on ART for more than one year [4–5]. Beyond this, little is known about hepatotoxicity in HIV-infected patients on long-term ART.
Schistosoma sp. are parasitic worms that infect at least 218 million people worldwide [6]. Schistosoma mansoni alone infects more than 83 million people, primarily in Africa, South America and the Caribbean [7]. In regions of Tanzania in which S. mansoni is highly endemic, an estimated 30–50% of HIV-infected patients have S. mansoni co-infection [8]. Classic teaching on S. mansoni is that its eggs cause Symmer’s pipestem fibrosis with resultant portal hypertension, while preserving hepatocellular function [9–10]. However, several studies have reported that schistosome infection may additionally lead to elevated liver enzymes and bilirubin levels [11–13]. To our knowledge, no studies have investigated the effects of S. mansoni infection on liver function in HIV-infected patients on ART.
Therefore, our objectives were: (1) to assess the prevalence of hepatotoxicity in HIV patients on medium-term and long-term ART and (2) to determine whether HIV-infected adults on ART who had S. mansoni co-infection had a higher prevalence of hepatotoxicity than HIV-infected adults on ART without schistosome co-infection.
Bugando Medical Centre (BMC) is a referral hospital located in Mwanza city, on the southern shore of Lake Victoria, Tanzania. In this region, the prevalence of HIV is 5% [14]. We conducted a cross-sectional study in outpatients seeking care at BMC’s HIV clinic, which has registered approximately 15,000 patients since its opening in 2004. On average, 45 patients per month are started on ART. Our team's clinical experience working at the BMC HIV clinic over the past five years demonstrates that approximately 1/3 of HIV-infected patients have S. mansoni infection, and < 1% have S. haematobium.
To assess for hepatotoxicity in patients using ART in the medium and longer term, we focused on two groups: patients who had been on ART for 3–6 months (GROUP 1), and patients who had been on ART for >36 months (GROUP 2). We calculated that enrolling 152 adults in each group would provide 80% power to detect a difference in hepatotoxicity among those with versus without schistosome infection. Calculations were based on the prediction that 5% of HIV-infected adults without schistosome infection would have hepatotoxicity, compared to 20% of those with HIV-schistosome co-infection [13].
All adults >18 years old who were attending BMC HIV clinic between August and December 2014, who provided written informed consent and had been on ART for 3–6 months or >36 months were eligible for enrolment. We excluded patients with known poor adherence to ART, defined according to the Tanzanian national HIV Care and Treatment Program as missing more than 3 doses or more than 2 days of ART, as assessed monthly by nurses and physicians at the HIV clinic. Assessments were made both by patient report and by pill counts.
Patients were recruited consecutively until the predetermined sample size was reached. Baseline information for each patient was obtained from the HIV clinic database and patients’ files. We interviewed patients to obtain history and demographic information, and performed physical examinations that included weight, height, and liver span measurements. Liver ultrasound was performed in all patients to determine the portal vein diameter. Urine was tested for Schistosoma mansoni using a point-of-care urine Circulating Cathodic Antigen dipstick test (Rapid Medical Diagnostics, Pretoria, South Africa). Results of the CCA test were recorded as negative, 1 for a faintly visible test line, 2 for a test line that was equal in intensity to the control line, and 3 for a test line that was higher intensity than the control line, according to the manufacturer’s statement that the intensity of the test line is qualitatively related to the intensity of the schistosome infection. All participants provided serum and plasma samples to the BMC clinical laboratory for determination of Alanine Aminotransferase (ALT), Aspartate Aminotransferase (AST) bilirubin levels, CD4 counts, and the presence of hepatitis B surface antigens and hepatitis C antibodies.
Liver enzymes and bilirubin levels were quantitated using the COBAS Integra 400 Plus machine (Roche, Basel, Switzerland). Hepatitis B was tested by surface Antigen rapid test (Laborex IVD Italiano SRL, Milan, Italy) and hepatitis C by antibody rapid test (Guangzhou Wondfo Biotech Co, Ltd, Guangzhou, China). CD4 levels were determined using the BD Tritest CD3/CD4/CD45 (BD Biosciences, San Jose, California, USA).
Analysis was done using Microsoft Excel and Stata/IC version 13 (College Station, Texas). We used the AIDS Clinical Trial Group definition of hepatotoxicity to define mild to moderate hepatotoxicity as AST, ALT, and/or bilirubin elevations of grade 1 or 2, and severe hepatotoxicity as AST, ALT, and/or bilirubin elevation of grade 3 or 4. For AST and ALT these groups are 1.25–5 and >5 times the upper limit of normal, and for bilirubin these groups are 1.1–2.9 and >2.9 times the upper limit of normal. Categorical variables were described as proportions and were compared between GROUP 1 and GROUP 2 using the chi-squared or Fisher’s Exact test while continuous variables were summarized by medians and interquartile ranges and compared using the Wilcoxon rank-sum test.
When an outcome had a value of less than 5 on the chi-squared analysis, we determined the strength of the association using Fisher’s Exact test. All variables that showed an association (p<0.10) with any hepatotoxicity on univariable analysis were subjected to Firth multivariable logistic regression analysis to identify the factors independently associated with hepatotoxicity among HIV-infected adults using ART due to the presence of several factors that were significant and had a small number of outcomes [15].
Ethical approval to conduct this study was obtained from the Joint Research and Publication Committee of the Catholic University of Health and Allied Sciences and BMC (CREC/067/2015). All results were communicated to patients’ primary care physicians. Patients found to have hepatitis B co-infection were given tenofovir-based regimens and those with S. mansoni were given praziquantel.
At the time of the study, the clinic had 3,971 patients receiving ART. Of these, 2,486 had been taking ART for 3–6 months or for more than 36 months, of which 17 were noted to have poor adherence to medications and were excluded. We approached 309 consecutive patients who were being seen at the clinic during the study period. Of these, four were not willing to participate and the remaining 305 were enrolled in the study (152 in GROUP 1 and 153 in GROUP 2). Patients in GROUP 1 had been on ART for a median of 4.4 [3.6–5.4] months, and those in GROUP 2 had been on ART for 76.4 [51.2–91.3] months.
There were significantly more females in GROUP 2 than in GROUP 1 (75.8% (116/153) versus 62.5% (95/152), p = 0.01). GROUP 2 patients were also significantly older (41 [37–47] versus 38 [34–48] years, p = 0.001). Significantly fewer patients in GROUP 2 reported their occupation as “peasant/farmer” compared to GROUP 1 (14/153, 9.2% versus 31/152, 20.4%, p = 0.006). There was a significantly higher use of anti-TB medication in GROUP 1 (11.2% (17/152) versus 3.3% (5/153), p = 0.008, Table 1).
Mild-moderate hepatotoxicity affected 29.6% (45/152) of adults in GROUP 1 and 19.6% (30/153) in GROUP 2 (p = 0.04, Table 2). Three patients in GROUP 1 and five patients in GROUP 2 had severe hepatotoxicity. Two of these, both in GROUP 2, were using nevirapine, and the other six were on efavirenz-based regimens.
In total, 8 patients had severe hepatotoxicity. Among these patients, 1 had both hepatitis B and C, 2 were using anti-tuberculosis medications and were also schistosome antigen positive, 2 had hepatitis B alone, and 1 reported herbal medication use. The remaining 2 patients with severe hepatotoxicity were negative for S. mansoni, hepatitis B, hepatitis C, alcohol use, and anti-tuberculosis medications. These two patients had both been on ART for 3–6 months, and both reported a history of lake water contact and rice cultivation.
Factors associated with hepatotoxicity are presented in Table 3. On multivariable analysis by Firth logistic regression, factors that remained significantly associated with hepatotoxicity included: hepatitis B surface antigen positivity (OR = 122 [7–2121], p = 0.001), use of anti-tuberculosis medication (OR = 42 [2–803], p = 0.014), hepatitis C antibody positivity, (OR = 6.2 [2.5–15.8], p<0.001) and schistosome antigen positivity (OR = 3.0 [1.6–5.8], p = 0.001). In addition to the factors listed in Table 3, contact with lake water (OR = 1.9 [1.1–3.2], p = 0.013) and history of cultivating rice (OR = 2.3 [1.3–3.8], p = 0.002) were also associated with hepatotoxicity by univariable analysis.
In addition, S. mansoni infection was associated with higher transaminases: ALT in those with S. mansoni was 32 [19–57] U/L versus 26 [16–37] in those without, and AST was 39 [24–68] versus 35 [25–47] (p = 0.018 and p = 0.088, respectively). Portal vein diameter was also larger in those with schistosome infection: 11.0 [10.0–12.3] versus 10.4 [9.8–11.2] cm (p = 0.0098). Liver span was larger as well (10 [9–11] versus 9 [9–10] cm, p = 0.0024).
The association of S. mansoni infection with hepatotoxicity was also assessed separately in the two groups. S. mansoni infection was significantly associated with hepatotoxicity in Group 1 (OR 3.0 [1.4–6.4], p = 0.003) and had results in a similar direction that did not reach significance for Group 2 (OR 2.1 [0.7–6.2], p = 0.18).
Our study identifies S. mansoni infection as a novel risk factor for hepatotoxicity among HIV-infected patients on antiretroviral therapy living in regions in which this parasitic infection is endemic. S. mansoni infection was associated with a three-fold increased odds of hepatotoxicity among HIV-infected adults even after adjusting for other known risk factors for hepatotoxicity. The robustness of our data is attested by the concordance of our findings with past literature; hepatitis B, hepatitis C, and anti-tuberculosis medication use were strongly associated with hepatotoxicity [1]. It is also supported by the ultrasound findings demonstrating the expected increased portal vein diameter, and not only elevated transaminases, in those with S. mansoni infection. The high rates of hepatotoxicity that we identified in this study, which are notably higher than those reported from other regions of Tanzania, may be due to higher rates of schistosomiasis in this population [16].
This is the first study to our knowledge to show an association between hepatotoxicity and S. mansoni infection in HIV-infected adults using ART, and one of few to associate hepatocellular dysfunction with this parasitic infection. Two Brazilian studies also showed increased levels of AST, ALT, gammaglutamyltransferase (γ-GT), alkaline phosphatase and total bilirubin in those with S. mansoni infections [12–13]. Taken together with these Brazilian studies, our work draws attention to the important consequences of S. mansoni infection not only on portal pressure but also on hepatocellular function. Additional work to explore this impact, particularly in HIV-infected patients, is urgently needed. Only two other studies were conducted in co-infected patients not on ART: one found no increased impact of co-infection on the liver [17], while the other found an association between the intensity of S. mansoni infection and liver and spleen size but did not observe increased periportal fibrosis in co-infected patients [18]. We are now following up the patients who participated in this study to determine whether treatment of schistosomiasis leads to improvement of hepatotoxicity in this at-risk population.
In our study, hepatitis B surface antigen positivity was significantly associated with hepatotoxicity, as previously documented [1]. Hepatitis B-associated hepatotoxicity has also been shown to be more severe in the setting of schistosome co-infection [19–20]. A recent review concluded that subjects with schistosome and hepatitis B co-infections have a prolonged carriage state, resulting more often in chronic hepatitis with severe cirrhosis and higher mortality compared to patients with hepatitis B infection alone [20]. Our finding of hepatocellular toxicity in S. mansoni-infected patients affirms this idea, with the more severe cirrhosis being fuelled by the hepatotoxic effects of both diseases in concert.
A limitation of our study is that, because of the use of the CCA test, we were unable to differentiate between schistosome species. Although multiple studies demonstrate that S. mansoni is by far the prevalent schistosome species in our CTC clinic population, it is possible that several cases of S. haematobium could have existed in this population and have been included in the analysis. In addition, because our study was cross-sectional and we could not determine how long people had been infected with S. mansoni, we cannot conclude on a causality link nor do we know the impact of length of co-infection.
In conclusion, our study furthers the importance of prior recommendations that patients starting ART in schistosome-endemic areas should be screened and treated for schistosomiasis [8]. We further urge follow-up liver function testing 3–6 months after ART initiation, particularly in those with schistosomiasis, hepatitis, or anti-tuberculosis treatment. We postulate that treatment of schistosome infections may prevent hepatotoxicity in patients, both during the crucial window of initiating and stabilizing ART regimens and much later as well.
|
10.1371/journal.ppat.1006385 | Platelet proteome reveals novel pathways of platelet activation and platelet-mediated immunoregulation in dengue | Dengue is the most prevalent human arbovirus disease worldwide. Dengue virus (DENV) infection causes syndromes varying from self-limiting febrile illness to severe dengue. Although dengue pathophysiology is not completely understood, it is widely accepted that increased inflammation plays important roles in dengue pathogenesis. Platelets are blood cells classically known as effectors of hemostasis which have been increasingly recognized to have major immune and inflammatory activities. Nevertheless, the phenotype and effector functions of platelets in dengue pathogenesis are not completely understood. Here we used quantitative proteomics to investigate the protein content of platelets in clinical samples from patients with dengue compared to platelets from healthy donors. Our assays revealed a set of 252 differentially abundant proteins. In silico analyses associated these proteins with key molecular events including platelet activation and inflammatory responses, and with events not previously attributed to platelets during dengue infection including antigen processing and presentation, proteasome activity, and expression of histones. From these results, we conducted functional assays using samples from a larger cohort of patients and demonstrated evidence for platelet activation indicated by P-selectin (CD62P) translocation and secretion of granule-stored chemokines by platelets. In addition, we found evidence that DENV infection triggers HLA class I synthesis and surface expression by a mechanism depending on functional proteasome activity. Furthermore, we demonstrate that cell-free histone H2A released during dengue infection binds to platelets, increasing platelet activation. These findings are consistent with functional importance of HLA class I, proteasome subunits, and histones that we found exclusively in proteome analysis of platelets in samples from dengue patients. Our study provides the first in-depth characterization of the platelet proteome in dengue, and sheds light on new mechanisms of platelet activation and platelet-mediated immune and inflammatory responses.
| Dengue is the most frequent hemorrhagic viral disease and re-emergent infection in the world. Recent decades were marked by a progressive global expansion of the infection including a higher frequency of severe dengue. Currently there is no effective vaccinal coverage or specific therapies, while efforts aimed at vector control have failed to stop the progression of epidemics and expansion of the disease. An increased understanding of the molecular physiology is of paramount importance for the establishment of new therapeutic targets and better clinical management. Dengue fever is characterized by thrombocytopenia and vascular leak. Although thrombocytopenia is a hallmark of dengue, the molecular phenotype and activities of platelets in the pathogenesis of dengue is not well elucidated. This work characterizes the proteome of platelets isolated from patients with dengue and includes validation of functionally-linked protein networks that we identified, using samples from a larger cohort of dengue patients. Moreover, in vitro experiments revealed activities of platelets that have recognized importance to dengue pathogenesis, including chemokine release, antigen presentation, and proteasome activity. Finally, our results identify circulating histones as a novel mechanism of platelet activation in dengue. These findings provide new evidence for platelet immune activities in dengue illness, and mark an advance in the understanding of this disease.
| Dengue is an infectious disease caused by four antigenically-related serotypes of dengue virus (DENV-1 to -4). It is the most frequent hemorrhagic viral disease and re-emergent infection in the world, with over 2.5 billion people living in high-risk transmission areas and more than 90 million apparent infections registered annually [1,2,3]. DENV-infection may present distinct clinical manifestations varying from mild self-limited dengue fever to life-threatening severe dengue, a syndrome associated with increased vascular permeability, hypovolemia, hypotension and eventually organ dysfunctions and shock [1,3,4]. Thrombocytopenia is a common feature in dengue syndromes and the drop in platelet counts is temporally associated with hemodynamic instability and progression to severity [5,6,7,8]. Nevertheless, the roles played by platelets in dengue pathogenesis remain poorly understood.
Platelets are highly specialized effector cells in hemostasis. Besides well-known hemostatic activities, newly-recognized platelet functions mediate both immune protective activities, including pathogen sensing and host responses, and inflammatory and immune injury to the host [9,10,11]. It is increasingly recognized that activated platelets have a repertoire of mechanisms for immune effector activity including release of cytokines and interaction with leukocytes [9,12,13]. As an example, it was recently shown that platelets are able to process and present antigens derived from exogenous plasmodial proteins in a fashion involving major histocompatibility complex (MHC) (also termed human leukocyte antigen–HLA) class I [14]. New discoveries of platelet biology of this type suggest that knowledge of global changes in platelet proteome, phenotype and function in dengue infection may contribute to a broader understanding of the pathobiology of dengue disease, as in other infections (9–11).
DENV has been detected in circulating platelets from infected patients [15,16]. In vitro studies demonstrated DENV binding mechanisms and viral protein synthesis by platelets [17,18]. We have recently shown that platelet activation contributes to altered vascular barrier integrity and innate immune activation in dengue [19,20]. Nevertheless, the mechanisms underlying platelet activation and function in dengue patients remain incompletely understood. Here we describe a shotgun proteomic approach intended to identify and quantify changes in platelet protein abundance in patients with dengue in comparison to that in platelets from healthy volunteers using a label-free mass spectrometry (MS)-based quantification. We found 252 differentially abundant proteins among dengue and control platelets. After an in silico biological process characterization, we observed high significance in proteins belonging to antigen processing and presentation, platelet activation, and immune and inflammatory responses activities. In parallel studies, platelet activation and secretion of stored chemokines was verified in an expanded cohort of dengue patients. DENV infection of platelets from healthy volunteers in vitro also induced platelet activation and chemokines secretion. In addition, DENV infection enhanced platelet expression of HLA class I and its surface display through mechanisms depending on proteasome activity. Interestingly, our proteome approach detected histones, a group of proteins with diverse biologic activities, exclusively in platelets from dengue-infected patients. Our findings indicate that platelets sequester circulating histones released during dengue infection, contributing to platelet activation. Taken together, our results indicate that the platelet proteome is altered in a functionally-significant fashion in dengue, identify new pathways involved in platelet activation in dengue infection, and provide new insights into dengue pathogenesis.
In order to investigate in-depth global changes in the platelet proteome during dengue infection, platelets (isolated with depletion of CD45+ leukocytes) from eight patients with clinical characteristics of having dengue were lysed in RapiGest SF (Waters) and prepared for proteomic analysis as described in the methods. After serological and molecular diagnostic confirmation through detection of IgM antibodies against dengue E protein and viral genome in patient plasma, two dengue-negative patients were excluded and samples from six dengue-confirmed patients (whose characteristics are presented in Table 1) were applied to a shotgun proteomic approach as follows.
Platelets from six patients with dengue (designated the dengue condition) and five healthy volunteers (designated the control condition) were digested with trypsin and fractionated with the isoelectric focalization of peptides (OFFGEL) system, generating 12 fractions. It was previously reported that OFFGEL fractionation, prior to MS analysis, enables identification of more peptides per protein particularly in low abundant molecules, and provides reliable results in both qualitative and quantitative levels [21,22]. Afterwards, a shotgun proteomic approach was applied (liquid chromatography tandem mass spectrometry–LC-MS/MS), where each OFFGEL fraction was analyzed on a high resolution mass spectrometer (Orbitrap XL) in technical triplicates. Quantification reproducibility was obtained according to normalized spectral abundance factors provided by PatternLab for Proteomics software. The MS raw files are available at http://max.ioc.fiocruz.br/supplementaryfiles/trugilho2016/ and readable by proteome analysis’ programs including the open source software PatternLab, Proteowizard or the Xcalibur from Thermo Fischer.
Through these approaches, we were able to identify with high confidence (FDR < 1%) a total of 13,362 and 15,792 peptides in control and dengue samples, respectively, which infers up to 3,336 protein entries from Nextprot databank in both conditions (Fig 1A and S1A and S1B Table). There were no peptides from the DENV databank reliably detected in both conditions. Approximately 58% of proteins (1,956) were inferred from more than 4 peptides, and about 37% (1,236) had at least one proteotypic peptide observation (S2B Table). A simplified list of 1,777 proteins, according to the maximum parsimony criterion, is available in S1D Table. Dengue and control biological conditions shared 2,557 protein identifications; 440 and 339 proteins were uniquely detected in dengue and control platelet samples, respectively (Fig 1A and S2A and S2B Table).
Differentially abundant proteins were reported by Pattern Lab’s T Fold module (Fig 1B). One hundred and sixty-seven proteins showed statistically significant differences in their abundance between dengue and control platelets (S2C Table). As shown in Fig 1B, 86 proteins were significantly up-regulated while 81 proteins were significantly down-regulated in platelets from dengue patients compared to healthy volunteers.
Our differential abundance analysis considered proteins exclusively identified in each condition and at least in three replicates. This filtering process decreased the list of 440 and 339 proteins uniquely identified in dengue and control to 116 and 61, respectively (S2A and S2B Table). Although there are no p-values assigned to each protein, we argue this stringency criterion (i.e., being identified in more than one sample), plus the fact of only being identified in one biological condition, strongly suggests a differential abundance. As such, our final result shortlists the original 344 protein entries (167 shared proteins, 116 from dengue and 61 from control), down to 252 non-redundant entries (S2D Table).
We used the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) to generate a protein interaction map and categorize the differentially abundant proteins according to biological process classification in the Gene Ontology (GO) database. We reassembled the STRING network in the PINV software for improved visualization, facilitating data interpretation. The analysis of the 252 differentially expressed proteins revealed 905 possible interactions between 224 proteins. After generating the network graph (Fig 2A) we performed a GO analysis for biological processes. Although proteins in different biological processes overlapped, the most statistically enriched ones were assigned as “antigen processing and presentation” along with “platelet activation” (S3 Table). These enriched pathways, whose general GO identifiers are GO:0019882 (p-value = 2,73−10) and GO:0030168 (p-value = 8.36−9), respectively, were identified by the presence of at least fifteen exclusive differentially expressed proteins each. Interestingly, most of the proteins in the “antigen processing and presentation” pathway belong to HLA class I genes (colored in red, Fig 2). Other important biological processes highlighted were “protein polyubiquitination” (GO:0000209 with 14 related proteins and p-value = 5.88−9) and the closely related “proteasomal protein catabolic process” (GO:0010498 with 10 related proteins and p-value = 4.69−5). The proteins reported in these processes, named together as “proteasome activity”, directly interact with those in “platelet activation” (colors yellow and blue respectively, Fig 2). In addition, we found the GO terms “inflammatory response” and “defense response” with high significance and to be equally important. Finally, a cluster of histones was also distinguished (colored in green, Fig 2A). Importantly, most of the histones reported were histone H2A. A second protein interactions graph highlights these aforementioned relevant pathways (Fig 2B). The final list of differentially abundant proteins (S2D Table) together with the analysis of protein interaction maps obtained by STRING guided us to subsequent validation experiments.
To investigate the relevance of the pathways identified in platelet proteome analysis to the pathophysiology of dengue, we validated the proteomic data in a cohort comprised by 36 dengue-infected patients with mild to severe dengue syndromes (Table 1 and Table 2). As shown in Table 2, dengue with warning signs and severe dengue patients had lower platelet counts, lower plasma albumin levels and higher frequencies of clinical signs of increased vascular permeability when compared to patients with mild dengue. We previously reported that platelets from dengue patients are activated [18]. Similarly, platelets from dengue patients included in this work were also activated as evidenced by P-selectin (CD62P) surface expression (Fig 3A). Importantly, platelet activation was higher in patients presenting dengue with warning signs and severe dengue syndromes compared to mild dengue patients (Fig 3A).
The most representative protein entries in “inflammatory response” pathway were granule-stored chemokines (four out of eight protein entries, orange at Fig 2). We then analyzed the content of Platelet Factor 4 (PF4/CXCL4), a chemokine expressed exclusively by platelets and megakaryocytes that is stored in platelet alpha granules and secreted upon platelet activation [9]. PF4V1/CXCL4L1, a variant of PF4/CXCL4, was found to be down-regulated in platelets from dengue patients compared to control (Fig 2B and S2D Table). Likewise, western blot analysis of PF4/CXCL4 revealed lower PF4 quantities in platelets from patients with dengue compared to healthy volunteers (Fig 3B). Importantly, activated platelets from patients with dengue released higher levels of PF4/CXCL4 ex vivo when compared to control platelets (Fig 3C), despite reduced PF4 content (Fig 3B). These data suggest that platelets activated in vivo during dengue infection increasingly release granule-stored PF4/CXCL4, which may lead to its reduced intracellular content. Consistent with this, plasma levels of PF4/CXCL4 were increased in dengue-infected patients compared to healthy volunteers (Fig 3D). Similar results were observed for the granule-stored chemokine RANTES/CCL5 (Fig 3E and 3F).
Next, we investigated whether DENV infection directly induces platelet secretion of granule stored chemokines. Platelets from healthy volunteers were stimulated with DENV or mock culture medium in vitro, and thrombin-activated platelets were used as positive control (Fig 3G). DENV infection significantly increased the proportion of P-selectin-expressing platelets after six hours when compared to Mock-stimulated platelets (55.3 ± 21.7% for DENV versus 27.9 ± 16.2% for Mock, p = 0.002 paired t test) (Fig 3G). DENV infection also increased platelet secretion of the granule-stored chemokines PF4/CXCL4 and RANTES/CCL5 (Fig 3H and 3I). These data demonstrate DENV-triggered platelet translocation of granule stored factors and are in agreement with platelet degranulation in vivo during dengue infection, as suggested above.
“Antigen processing and presentation” and “proteasome activity” were major activities identified based on differentially abundant proteins, with upregulation of HLA class I and proteasome subunits in platelets from dengue patients compared to healthy volunteers (Fig 2B; S2D Table and S3 Table). In agreement, HLA class I expression was increased in platelets from dengue patients compared to healthy volunteers when detected by western blot (Fig 4A). It has been shown by in-depth RNA sequencing that the mRNA for HLA class I subunits including b2-microglobulin and HLA are highly expressed in platelets [23]. Thus, we investigated if DENV infection in vitro increases HLA class I expression by platelets isolated from healthy uninfected donors. Platelets from healthy volunteers were stimulated with thrombin, mock or DENV, and HLA class I expression was evaluated by western blot. As shown in Fig 4B, DENV infection increased the expression of HLA class I in platelets. In contrast, platelet activation by thrombin did not affect HLA class I protein synthesis (Fig 4B). We next determined whether DENV infection enhances HLA class I trafficking and surface display. We observed a population of platelets that expressed significantly (p<0.05) higher levels of HLA class I on surface (HLA class I High) at 3 and 6 hours post DENV infection compared to Mock (Fig 4C). Importantly, infection of platelets with DENV in the presence of the translational inhibitor cyclohexamide (10 μM), which inhibited HLA class I synthesis by platelets (Fig 4D), prevented platelets to increase HLA class I surface display in response to DENV (Fig 4E), suggesting that increased HLA class I synthesis is necessary before HLA class I trafficking to surface. To gain insights into the role played by proteasome protein processing in generating peptides for HLA class I presentation on platelets, platelets were treated with the proteasome inhibitor bortezomib (1 μM) for 30 min prior to DENV infection. Inhibition of proteasome activity prevented platelets to enhance HLA class I surface expression at 3 and 6 hours post DENV infection (16.8 ± 8.4% HLA class IHigh for bortezomib versus 45.0 ± 6.9% for vehicle, p<0.005 paired t test, 6 hours post infection) (Fig 4F), suggesting that HLA class I loading by platelets depends on proteasome-generated peptides. Nevertheless, if the peptides presented derive from proteasome degradation of platelet or viral proteins remains unknown and should be further evaluated in the future.
To determine whether proteasome inhibition impaired HLA class I expression in a selective way, we evaluated the effects of bortezomib on thrombin- and DENV-triggered platelet activation. As shown in Fig 4G and 4H, treatment with bortezomib did not affect platelet P-selectin surface expression following thrombin-stimulation or DENV-infection, suggesting that inhibiting proteasome activity inhibits HLA class I surface display in a specific fashion.
In our proteomic analysis histones were detected with statistical confidence exclusively in platelets from dengue-infected patients (Fig 2 and S2D Table). In agreement, histone H2A was detected by western blot in platelets from dengue patients but not in platelets from healthy volunteers (Fig 5A). In contrast, histones H2B and H3 were not detected in platelets from patients with dengue or in control platelets (S1 Fig). It was previously demonstrated by in-depth RNA sequencing that platelets from healthy volunteers have message RNA for all core histone subunits [23]. We then determined if DENV infection directly induces histone H2A synthesis by platelets. Platelets from healthy volunteers were stimulated with thrombin, mock or DENV in vitro, and histone H2A expression was evaluated. As shown in Fig 5B, DENV infection did not induce histone H2A expression by platelets, suggesting that histone H2A in platelets from dengue patients derives from its synthesis by infected megakaryocytes or from sequestration of free histones by platelets in the peripheral circulation.
We then measured the levels of histone H2A in plasma from patients with dengue and healthy volunteers and observed increased levels of circulating histone H2A during DENV infection (Fig 5C). In addition, higher levels of histone H2A were observed in plasma from dengue with warning signs and severe dengue patients compared to mild dengue (Fig 5C). To further explore the possibility that platelets are able to sequester histones circulating in plasma in dengue-infected patients, platelets from healthy volunteers were incubated (37°C at a 5% CO2 atmosphere) with 20% plasma from dengue-infected patients or from healthy volunteers. Platelet exposure to plasma from patients with dengue significantly increased their content of histone H2A when detected by western blot (Fig 5D). Inhibition of platelet protein translation by cyclohexamide (10μM) did not reduce platelet accumulation of histone H2A in response to dengue plasma whereas the cytoskeleton assembly inhibitor Cytochalasin B (1 μM), reduced the content of histone H2A protein (Fig 5E). These results demonstrate that histone H2A in platelets from dengue-infected patients may at least in part derive from the sequestration of circulating free histones by platelets.
Next, we evaluated whether platelet binding of circulating cell free histone H2A was a common feature among dengue and the related arbovirus diseases zika and chikungunya fever (S4 Table). Interestingly, we observed increased levels of circulating histone H2A in plasma from patients with dengue compared to zika or chikungunya patients (Fig 5F). Consistent with this observation, when platelets from healthy uninfected donors were incubated with plasma from patients with dengue, zika or chikungunya, higher content of histone H2A accumulated in platelets exposed to dengue plasma compared to zika or chikungunya (Fig 5G).
It has been previously shown that free histones bind to and activate platelets in vitro and in vivo [24,25,26,27]. However, the role played by circulating free histones in platelet activation during dengue infection remains unclear. Patients with dengue have increased levels of histone H2A in circulation (Fig 5C). We then investigated whether cell-free histone H2A is able to activate platelets in vitro. Treatment of platelets from healthy uninfected donors with recombinant human histone H2A significantly increased P-selectin translocation to the surface (Fig 6A) and secretion of PF4/CXCL4 into the supernatant (Fig 6B). Unfractionated histones have been shown to activate cellular responses through mechanisms involving toll like receptor (TLR) binding and calcium-mediated signaling [24,25,26,28,29]. Treatment of platelets with the calcium chelator BAPTA-AM (20 μM) significantly impaired P-selectin surface expression and PF4 secretion by histone H2A-activated platelets (Fig 6C and 6D). In addition, blocking of TLR4 significantly reduced platelet secretion of PF-4 and trended to reduce P-selectin translocation to platelet surface (Fig 6E and 6F), indicating that platelet activation by histone H2A partially depends on TLR4 binding. To investigate whether histone H2A in plasma from dengue patients is able to activate platelets, we incubated platelets from healthy volunteers with plasma from dengue-infected patients for 1, 2 and 4 hours, and observed increased platelet P-selectin surface expression in response to dengue plasma (Fig 6G). Next, we treated plasma from dengue infected patients and control plasma with rabbit IgG or anti-histone H2A (20 μg/mL) for 30 min prior to platelet stimulation. As shown in Fig 6H, blocking histone H2A prevented dengue plasma from inducing P-selectin translocation to the platelet surface.
Thrombocytopenia is a hallmark of dengue disease. Platelet count decline is temporally coincident with the critical phase of infection and correlates with the extension of hemodynamic instability and plasma leakage [6,7,30]. Although dengue pathophysiology is not completely elucidated, it has been shown that platelet activation plays a major role in inflammatory amplification and thrombocytopenia during dengue infection [18,19,20]. Considering this, our proteomic approach aimed to identify differentially abundant proteins in platelets from patients with dengue and matched healthy volunteers attempting to elucidate platelet activities during dengue illness. Our results reveal differentially expressed platelet proteins that point to immunoinflammatory platelet reprogramming in dengue-infected patients compared to healthy subjects. To gain insights into the platelet phenotype in dengue, we performed in silico analysis of protein interactions and gene ontologies and identified five main biological processes or components: “platelet activation”, “inflammatory response”, “antigen processing and presentation”, “proteasome activity” and “histones”. Finally, phenotypical and functional changes related to each of these processes were validated using platelet samples from a larger cohort of patients and by complementary mechanistic and functional assays.
Activation of circulating platelets in patients with dengue has been previously reported [18,20,31]. In our proteome analysis, proteins related to platelet activating signaling including PAR4 (F2RL3), G protein subunits (GNA12 and GNA14) and p38 MAPK (MAPK14) were increased in platelets from patients with dengue (blue dots in Fig 2B and S2D Table), potentially contributing to increased platelet activation during dengue infection. Consistent with these observations, dengue patients in the present study had increased P-selectin surface expression on platelets. In addition, P-selectin surface expression was increased in patients presenting dengue with warning signs and severe dengue syndromes compared to mild dengue (Fig 3A). P-selectin is a glycoprotein stored in platelet α-granules that is translocated to the surface and released in suspension during platelet activation [9]. It is the main adhesion molecule responsible for platelet interaction with monocytes [9,13,20,32], and circulating platelet-monocyte aggregates have been detected in dengue-infected patients [20,33]. Despite platelets from patients with dengue have increased P-selectin expression at baseline, it was recently shown that P-selectin trafficking to surface in response to thrombin receptor activating peptide stimulation ex vivo was lower in platelets from dengue patients compared to control [31], which is consistent with platelet exhaustion of α-granules proteins. Beyond P-selectin, platelet α-granules store numerous cytokines, chemokines and growth factors [9]. In agreement, we also observed exhaustion of the granule-stored chemokine PF4/CXCL4 in platelets from patients with dengue. Platelet exhaustion of PF4/CXCL4 content occurred in parallel with increased PF4/CXCL4 in plasma from dengue patients and PF4/CXCL4 secretion by platelets that were activated by DENV infection in vitro (Fig 3B–3D). A recent study reported that patients with severe dengue have lower levels of PF4/CXCL4 in plasma when compared to mild dengue patients [34]. This may result from lower platelet counts in patients with severe dengue, or from enhanced platelet exhaustion of PF4/CXCL4 content in severe dengue patients. More studies are still necessary to address the role played by platelet exhaustion in severity of dengue. Because of its roles in endothelial dissociation and angiogenesis [35], PF4 is potentially involved in vasculopathy of dengue syndromes.
Although increased P-selectin surface expression has been shown in platelets from dengue patients [18,31], the mechanisms underlying platelet activation in dengue are not completely understood. As shown here and in previous publications [18,36], platelets can be directly activated by DENV infection in vitro (Fig 3G). Platelet activation by DENV presents delayed kinetics of P-selectin expression compared to thrombin stimulation (Fig 3G) [18]. While “traditional” platelet activation by G-protein coupled receptors are rapid, it is now known that platelet activation in responses to infectious and immune stimuli, including LPS, can be delayed and sustained [9,37]. DENV activation of platelets requires infective DENV binding to DC-SIGN [18], a surface receptor involved in DENV binding and replication by platelets [17]. However, platelet activation in patients with dengue peaks at the critical phase of infection, when DENV particles are no longer circulating [18,31]. This indicates that other mechanisms are involved in platelet activation during nonviremic phases of dengue illness. Here we provide evidence that circulating histone H2A contributes to increased platelet activation in dengue (Fig 6). Histone H2A was detected exclusively in platelets from dengue infected patients, and higher levels of histone H2A were observed in plasma from dengue patients with warning signs and severe dengue. Histones can be released from necrotic cells and tissues or by neutrophils during the formation of neutrophil extracellular traps (NET), composed by released chromatin components (DNA and histones) and granule proteins [38,39,40]. Regarding this, DENV was recently shown to induce neutrophils to extend NETs in vitro [41]. Despite the fact that circulating histones have not been previously shown in dengue, a report showed that higher levels of circulating DNA associate with shock outcome in dengue-infected patients [42]. After exposing platelets to plasma from patients with dengue, we observed that histone H2A binds to and activates platelets (Figs 5 and 6). Similarly, when whole blood is exposed to histones in vitro, histones bind to platelets leading to platelet aggregation [26]. Accordingly, injection of histones into mice leads to histone accumulation in sites of thrombosis and to thrombocytopenia [24,26,27]. In experimental sepsis in mice and in patients with sepsis, a syndrome with many parallels with severe dengue, circulating histones are major mediators of vascular damage and disseminated intravascular coagulation (DIC) [27,29,43]. In addition to platelet activation, circulating free histones also activate endothelial cells amplifying the activation of inflammation and coagulation through endothelium expression of tissue factor and extrusion of Weibel-Paled bodies [24,28]. While the roles played by cell-free histones in vasculopathy, shock and organ dysfunction in dengue remain to be precisely determined, our observations strongly suggest that histone-mediated platelet activation may contribute to dengue pathogenesis.
In protein interaction analysis, the “antigen processing and presentation” pathway was closely associated with “proteasome activity” (red and yellow in Fig 2, respectively). The proteasome is a protein complex responsible for protein degradation in nucleated cells that has been previously reported to be present in platelets [44]. HLA class I proteins bind to and display on cell surface peptides from physiologic protein degradation by proteasome [45]. Through self-peptides presentation in HLA class I, nucleated cells and platelets survive cytotoxic T cell or NK cell immunosurveillance [45,46]. During viral or parasite infections, however, proteins from pathogens are also digested and presented by HLA class I, allowing cytotoxic T cell to eliminate infected cells [45]. A specific proteasome complex termed the immunoproteasome is constitutively expressed in immune cells and accelerate peptide generation for MHC class I antigen presentation, including during viral infections [45,47,48]. Recently, immunoproteasome subunits were reported to be present and functional in platelets [49]. Here, we showed that proteasome activity is required for increased HLA class I surface display in platelets following DENV infection (Fig 4). The ability of platelets to present exogenous antigens in HLA class I was recently demonstrated by Chapman and colleagues in vitro and in experimental model of cerebral malaria in vivo [14]. Platelets were able to activate T cell-mediated responses through HLA class I-mediated presentation of pathogen-derived antigens in that study [14]. Platelet-mediated T cell activation has demonstrated roles in immune activation and cytotoxic T lymphocyte-mediated platelet destruction [14,50,51,52,53] potentially contributing to cytokine storm and thrombocytopenia, both important pathogenic mechanisms of severe dengue [5,6,7,30,54]. Nonetheless, whether platelets process and present DENV-derived antigens in MHC class I and whether it impacts T cell activation and thrombocytopenia in dengue requires further investigation.
Several mechanisms can be involved in platelet proteome changes during natural DENV infection in humans. Even though platelets do not have nucleus, they have stored RNA molecules and diverse mechanisms for post transcriptional processing of RNA using specialized pathways to change their proteome, phenotype and function [9,23,55]. In addition, changes in platelet proteome observed in our study may result from alterations in megakaryocyte biology during dengue disease. It was previously demonstrated in nonhuman primates and in ex vivo infection of human marrow aspirates that megakaryocytes are the main target for DENV in marrow [56,57]. Even though we demonstrate that DENV infection increases HLA class I protein synthesis and surface display by platelets and that platelets sequester cell-free histones from dengue plasma, we recognize that alterations in platelet cargo during thrombopoiesis may contribute to these changes in the platelet proteome in dengue. In this regard, the CXC motif chemokines GRO1/CXCL1, MIP-2α/CXCL2 and GRO3/CXCL3 were detected exclusively in platelets from dengue patients by proteome (orange in Fig 2B), suggesting that dengue infection may change platelet granule’s protein content through a more inflammatory profile. While we took measures to deplete leukocytes from our platelet preparations, we were not able to completely eliminate leukocyte contamination. Nevertheless, our proteome analysis revealed that the leukocyte marker CD45 (PTPRC) was detected only in control samples (purple in Fig 2A, S2 Table), excluding leukocyte contamination as a determinant for increased HLA class I, histones or chemokines expression in platelets from dengue patients.
Emerging evidence identifies platelets as dynamic cells that represent a link between inflammation and pro-thrombotic responses in many vascular and inflammatory processes [9,10,11,58]. Consistent with this notion, our findings provide novel biological evidence that platelets undergo dynamic changes in dengue resulting in phenotypic changes implicated in immune and inflammatory processes that are of recognized relevance to dengue pathophysiology (Summarized in Fig 7). Platelets may be activated during dengue illness by parallel or sequential mechanisms, which may include direct DENV infection of platelets as well as indirect activation resulting from platelet signaling by sequestered circulating histones. This infection-driven reprograming of platelets in dengue alters the regulation of HLA class I expression on platelets and the secretion of cytokines and chemokines. Thus, platelets can affect the immune and inflammatory milieu of dengue illness, with potential consequences to disease progression and severity. These functional changes demonstrated in platelets from patients and in vitro experiments in this report, and others that may be discovered from our analysis of the platelet proteome in dengue patients, will contribute to a better understanding of platelet activities in dengue pathogenesis.
Peripheral vein blood samples were obtained from thirty-six serologically/molecularly confirmed DENV-infected patients examined at the Instituto Nacional de Infectologia Evandro Chagas (INI)–Fundação Oswaldo Cruz, Rio de Janeiro, Brazil, during the dengue outbreaks of 2011–2013. Clinical and laboratorial characteristics of dengue-infected patients are presented in Tables 1 and 2. Samples were collected on an average of 4.4±1.8 days after onset of illness and first symptom presentation. Peripheral vein blood was also collected from ten patients with zika virus infection and ten patients with chikungunya fever examined at the Quinta D’or hospital in 2016 (S4 Table); and from twenty-two age-matched healthy subjects.
Dengue-infected patients were classified according to the World Health Organization (WHO) dengue case definition guideline [3] as having mild dengue (47.2%), dengue with warning signs (44.4%) or severe dengue (8.3%). Diagnosis of dengue patients (which were included before zika and chikungunya virus introduction in Brazil) was confirmed as clinical symptoms and signs consistent with dengue with positive plasma detection of DENV genome, IgM antibodies against DENV E protein and/or DENV NS1 antigen. All zika and chikungunya infected patients had the diagnostic confirmed by detection of zika virus (ZIKV) or chikungunya virus (CHIKV) genome, respectively. For viral RNA detection and typing, viral RNA was extracted (QIAamp Viral RNA mini kit, Quiagen) from plasma samples and processed as previously described [59,60]. Levels of IgM and IgG specific for DENV E protein were measured using standard capture ELISA Kit according to the manufacturer’s instructions (PanBio). DENV NS1 protein was detected in patient plasma using the NS1 detection Kit (BioRad). Primary and secondary infections were distinguished using IgM/IgG antibody ratio as previously described [61,62,63]. Five of six included patients were diagnosed with a secondary dengue infection (Tables 1 and 2).
The study protocol was approved by the Institutional Review Board (INI # 016/2010 and IOC/FIOCRUZ # 42999214.1.1001.5248) and the experiments were performed in compliance with this protocol. Written informed consent was obtained from all volunteers prior to any study-related procedure in accordance with the Declaration of Helsinki.
Peripheral blood samples were drawn into acid-citrate-dextrose (ACD) and centrifuged at 200 x g for 20 minutes to obtain platelet-rich plasma (PRP). Platelets were isolated from PRP and CD45+ leukocytes were depleted from platelet preparations as previously described [19,20]. Briefly, PRP was centrifuged at 500 xg for 20 min in the presence of 100 nM Prostaglandin E1 (PGE1) (Cayman Chemicals). The supernatant was discarded, and the platelet pellet was resuspended in 2.5 mL of phosphate-buffered saline containing 2 mM EDTA, 0.5% human serum albumin and 100 nM PGE1 and incubated with anti-CD45 tetrameric antibody complexes (1:25) for 10 minutes and with dextran-coated magnetic beads (1:50) for additional 15 minutes before purification in a magnet (Human CD45 depletion kit; StemCell, Easy Sep Technology). Recovered platelets were resuspended in 25 mL of PSG (PIPES-saline-glucose: 5 mM C8H18N2O6S2, 145 mM NaCl, 4 mM KCl, 50 mM Na2HPO4, 1 mM MgCl2-6H2O, and 5.5 mM glucose) containing 100 nM of PGE1. The platelet suspension was centrifuged at 500 xg for 20 minutes. The supernatant was discarded and the pellet resuspended in medium 199 (Lonza). The purity of the platelet preparations (>99% CD41+) was confirmed by flow cytometry.
Isolated platelets from each dengue patient (n = 6) or control subject (n = 5) were individually assessed by proteome analysis as follows. Platelets (1 x 108) were centrifuged at 700 x g for 10 min and resuspended in 50 mM NH4HCO3 containing 0.2% of RapiGest SF (Waters) for cell lysis. The protein concentration was determined in each individual sample using the bicinchoninic acid assay (BCA) according to the manufacturer's instructions (Sigma-Aldrich). Each sample (100 μg of protein), was reduced with dithiothreitol to a final concentration of 5 mM for 3 h at 37°C. After reaching room temperature, the samples were alkylated with iodoacetamide to a final concentration of 15 mM for 15 min while protected from light. Trypsin (Promega, USA) was added in a 1:50 (p/p) enzyme/substrate ratio. The digestion was performed for approximately 24 h at 37°C, and reaction was stopped with 1% formic acid. Aliquots of 50 μg of the initial digests were quantitated by Nanodrop spectrophotometry at 280 nm (Thermo Fisher Scientific), desalted with POROS R2 resin (Applied Biosystems), packaged in micropipette tips (Millipore, Bedford, USA) and equilibrated in TFA 1%. After washing with TFA 0.1%, the peptides were eluted in TFA 0.1% containing acetonitrile 70% and completely dried in vacuum centrifuge.
Peptides were solubilized with peptide OFFGEL solution and separated using an Agilent 3100 OFFGEL Fractionator with OFFGEL High Res Kit, pH 3–10 immobilized pH gradient (IPG) DryStrips according to manufacturer’s instructions (Agilent Technologies, Germany). Twelve well fractionations were focused for 20 kVh with a maximum current of 50 mA and power of 200 mW. Each fraction was separately desalted, as described in the previous section, and suspended in 40 μL of 1% formic acid.
The desalted peptides from each OFFGEL fraction were loaded onto a 10 cm reversed phase (RP) column and separated on-line to the mass spectrometer by using Easy nLC II (Thermo Scientific). Four microliters were initially applied to a 2-cm long (100 μm internal diameter) trap column packed with 5 μm, 200 A Magic C18 AQ matrix (Michrom Bioresources, USA) followed by separation on a 10-cm long (75 μm internal diameter) separation column. Samples were loaded onto the trap column at 2000 μL/min while chromatographic separation occurred at 200 nL/min. Mobile phase A consisted of 0.1% (v/v) formic acid in water while mobile phase B consisted of 0.1% (v/v) formic acid in acetonitrile. Peptides were eluted with a gradient of 2 to 40% of B over 32 min followed by up to 80% B in 4 min, maintaining at this concentration for 2 min more, before column reequilibration. The HPLC system was coupled to the LTQ-Orbitrap XL via a nanoscale LC interface (Thermo Scientific). Source voltage was set to 1.9 kV, and the temperature of the heated capillary was set to 200°C and tube lens voltage to 100 V. MS1 spectra were acquired on the Orbitrap analyzer (300 to 1,700 m/z) at a 60,000 resolution (FWHM at m/z 400). FTMS full AGC target was set to 500,000 and ion trap MSn AGC target was set to 30,000. For each survey scan, the 10 most intense ions were submitted to CID fragmentation (minimum signal required of 10,000; isolation width of 2.5 m/z; normalized collision energy of 35.0; activation Q of 0.25 and activation time of 30 s) followed by MS2 acquisition on the linear trap analyzer. Dynamic exclusion option was enabled and set with the following values for each parameter: repeat count = 1; repeat duration = 30 s; exclusion list size = 500; exclusion duration = 45 s and exclusion mass width = 10 ppm. Data were acquired in technical triplicates using the Xcalibur software (version 2.0.7).
The raw data files were processed and quantified using PatternLab for Proteomics software (http://www.patternlabforproteomics.org/) [64]. Peptide sequence matching (PSM) was performed using the Comet algorithm [65] against the protein-centric human database NeXtProt [66] (manually annotated and recommended by HUPO—Human Proteome Organization) plus a FASTA file containing DENV sequences retrieved from the NCBI database (GeneBank taxon number 14,164). A target-reverse strategy was employed. The search considered tryptic and semi-tryptic peptide candidates. Cysteine carbamidomethylation and oxidation of methionine were considered as fixed and variable modifications, respectively. The Comet search engine considered a precursor mass tolerance of 40 ppm and bins of 1.0005 for the MS/MS. The validity of the peptide spectrum matches were assessed using PatternLab’s Search Engine Processor (SEPro) module [67]. Identifications were grouped by charge state (+2 and > +3) and then by tryptic status (semi-tryptic), resulting in four distinct subgroups. For each result, the XCorr, DeltaCN and ZScore values were used to generate a Bayesian discriminator. SEPro then automatically established a cutoff score to accept a false-discovery rate (FDR) of 1% based on the number of decoys, independently performed on each data subset, resulting in a false-positive rate that was independent of tryptic status or charge state. Additionally, a minimum sequence length of 6 amino acid residues was required. Similar proteins, which represent an identical sequence and consist of a fragment of another sequence, were eliminated. Then, only PSMs with less than 5 ppm were considered to compose a final list of mapped proteins supported by at least three independent characteristics (e.g., identification of a peptide in charge states, modified and non-modified version of the same peptide, or different peptides). All identification results are reported with less than 1% FDR both in peptide and protein levels. Spectral counting were used as a surrogate for semi-quantitation according to the normalized spectral abundance factor (NSAF) [68]. Differentially abundant proteins were pinpointed using PatternLab’s TFold module with a Benjamini–Hochberg q-value of 0.05 [69]. The approximately area-proportional Venn diagram module displayed all mapped proteins in each condition.
Protein interaction networks for differentially abundant proteins were developed using the STRING database (http://string-db.org/) [70]. Enrichment analysis for biological processes annotation was performed using the Gene Ontology (GO) databank available as a tool inside STRING. The generated networks were edited according to the GO terms classification and submitted to the Protein Interaction Network Visualizer—PINV (http://biosual.cbio.uct.ac.za/pinv.html) [71].
DENV serotype 2 strain 16881 was propagated in C6/36 Aedes albopictus mosquito cells and titrated by plaque assay on BHK cells [72]. The quantity of infectious particles was expressed as plaque forming units (PFU)/mL. Platelets from healthy uninfected donors were incubated (37°C in a 5% CO2 atmosphere) with DENV-2 at a multiplicity of infection of 1 PFU/ platelet, with thrombin (Sigma, T1063) (0,5 U/mL) or with recombinant human histone H2A (BioLabs, M2502S) for the indicated times. Supernatants from uninfected C6/36 cell cultures (mock) were produced using the same conditions and used as a control for platelet stimulation by DENV. To characterize the mechanisms involved in platelet surface expression of HLA class I, we pre-incubated platelets with the proteasome inhibitor bortezomib (LC Laboratories, MA) (1 μM) or the translational inhibitor cyclohexamide (10μM) for 30 min prior to DENV infection.
Platelets from healthy volunteers were incubated (37°C in a 5% CO2 atmosphere) with plasma from dengue-infected patients or heterologous healthy volunteers for the indicated times. To characterize the role played by circulating histones in platelet activation, plasma samples were treated with anti-histone H2A (Santa Cruz sc-10807) (20 μg/mL) for 30 min prior platelet stimulation. To characterize the mechanisms involved in platelet activation by cell free histone H2A, platelets were pretreated with the calcium chelator BAPTA-AM (Sigma) (20 μM) or anti-TLR4 neutralizing antibodies (eBioscience 169917–82) (20 μg/mL) for 30 min prior stimulation with histone H2A.
Platelets (1–5 x 106) were incubated with FITC-conjugated anti-CD41 (BD Phamingen, CA) (1:20), PE-conjugated anti-CD62-P (BD Pharmingen, CA) (1:20) and APC-conjugated anti-HLA-A, B, C (Biolegend, CA) (1:50) for 30 min at 37°C. Isotype-matched antibodies were used to control nonspecific binding of antibodies. Platelets were distinguished by specific binding of anti-CD41 and characteristic forward and side scattering. A minimum of 10,000 gated events were acquired using a FACScalibur flow cytometer (BD Bioscience, CA).
In vitro stimulated platelets or freshly isolated platelets from dengue patients and healthy volunteers were lysed (0.15 M NaCl, 10mM Tris pH 8.0, 0.1 mM EDTA, 10% Glicerol and 0.5% triton X-100) in the presence of protease inhibitors (Roche, Indianapolis, IN). Platelet proteins (20 μg) were separated by 15% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS–PAGE) and transferred into nitrocellulose membranes. Membranes were blocked in Tris-buffered saline (TBS) supplemented with 0.1% Tween 20 (TBS-T) plus 5% milk for 1 h before incubation overnight with primary mouse anti-human PF4 (R&D Systems) (1:500) or rabbit-anti-human Histone H2A (Cell Signaling) (1:1000) or biotinylated-mouse-anti-human HLA class I (eBioscience) (1:1000), or for 1 h with mouse anti-human β-actin (Sigma Aldrich) (1:20,000) antibodies. After washing five times in TBS-T, membranes were revealed using fluorescent dye-conjugated or peroxidase-conjugated secondary antibodies (Vector) (1:10,000) or streptavidin (R&D) (1:200).
Washed platelets (109 per mL) isolated from eight healthy volunteers or eight dengue-infected patients were incubated at 37°C in a 5% CO2 atmosphere. After 4 hours of incubation, the platelets were pelleted, the supernatants were harvested and the secreted levels of PF-4/CXCL4 and RANTES/CCL5 were measured using standard ELISA protocol according to manufacturer’s instructions (R&D systems). PF4/CXCL4 and RANTES/CCL5 were also measured in supernatants of platelets obtained from the same patients at the recovery phase (average 17.5±6.5 days after the onset of illness).
Plasma samples were collected from ACD-anticoagulated blood from patients and healthy volunteers and frozen in liquid nitrogen until use. PF4/CXCL4 plasma levels were quantified in 1:200 diluted samples. Circulating histone H2A was measured using a standard ELISA protocol according to manufacturer’s instructions (LSBio, LS-F238).
Complementary statistics was performed using GraphPad Prism, version 5.0 (GraphPad, San Diego, CA). The numerical demographic and clinical variables are expressed as median and interquartile range (25–75 percentile) or as number and percentage (%). All numerical variables were tested for normal distribution using the Kolmogorov-Smirnov test. For comparisons among three groups we used Oneway ANOVA to determine differences and Bonferroni’s multiple comparison test to locate the differences among groups. For comparisons between two groups we compared the continuous variables using the t test for parametric distribution or the Mann–Whitney U test for nonparametric distribution. The paired two-tailed t-test was used to compare in vitro stimulated platelets with unstimulated platelets from the same donor. Qualitative variables were compared by the two tailed Fisher test using Epi-Info software version 7.0 (CDC).
|
10.1371/journal.pgen.1006859 | An autonomous metabolic role for Spen | Preventing obesity requires a precise balance between deposition into and mobilization from fat stores, but regulatory mechanisms are incompletely understood. Drosophila Split ends (Spen) is the founding member of a conserved family of RNA-binding proteins involved in transcriptional regulation and frequently mutated in human cancers. We find that manipulating Spen expression alters larval fat levels in a cell-autonomous manner. Spen-depleted larvae had defects in energy liberation from stores, including starvation sensitivity and major changes in the levels of metabolic enzymes and metabolites, particularly those involved in β-oxidation. Spenito, a small Spen family member, counteracted Spen function in fat regulation. Finally, mouse Spen and Spenito transcript levels scaled directly with body fat in vivo, suggesting a conserved role in fat liberation and catabolism. This study demonstrates that Spen is a key regulator of energy balance and provides a molecular context to understand the metabolic defects that arise from Spen dysfunction.
| All animals need energy to fuel development and survive as adults. Excess energy stored as fat provides a means to endure periods when external energy is unavailable, but there is a delicate balance between accumulating sufficient fat stores and becoming obese. While the enzymes that mediate energy deposition into and mobilization from fat stores are well studied, the complex upstream regulatory pathways have not been fully worked out. We report here that two members of a conserved family of RNA-binding proteins, Spen and Nito, operate in fat storage cells in fruit fly larvae to control the expression of genes that mediate energy liberation from fat stores. Manipulating Spen or Spenito function grossly perturbs larval energy metabolism, including imbalances in the amounts of stored fats, key metabolites, and metabolic enzymes, and resulting in defects in survival under starvation conditions. Interestingly, Nito opposes Spen functions, indicative of a regulatory mechanism that helps keep energy balance in check. We find that the mouse homologs of Spen and Nito, which were known to regulate gene expression in other pathways, respond similarly to changes in body fat induced by a high-fat diet, suggesting that the balancing effect of these two proteins also prevents mammalian obesity.
| Organisms strive to achieve homeostasis between energy intake and utilization, but also must maintain energy stores to survive when utilization exceeds intake. Demands for utilizable energy trigger hydrolysis of triglycerides stored in adipose cells to produce free fatty acids that are released into the circulatory system. Once within the energy-requiring cells, fatty acids must be conjugated first to coenzyme A and then to carnitine for transport across the inner mitochondrial membrane. During fasting, fat used for fuel is primarily derived from adipose tissue triglycerides, and the mobilization of fatty acids from triglyceride stores is a key regulatory step.
Obesity is caused by excess energy stored in the form of triglycerides (TAGs) [1]. Genetic factors dictate 40–70% of the variance in body mass index (BMI) and obesity predisposition [2–9], but understanding individual gene function in obesity is complicated by the multigenic and multi-systemic nature of the disease. Drosophila provides a powerful model to investigate mechanisms of energy storage and utilization [10–17]. The fat body (FB) corresponds to mammalian liver and white adipose tissue (WAT) and stores glycogen and TAGs [18]. Assessment of energy regulation during the larval stage is particularly informative, since energy is balanced between utilization (to fuel foraging behaviors and larval growth) and storage (to fuel later growth during the pupal stage) [10, 15, 19]. We previously identified 66 genes required to prevent excess fat accumulation in larvae, including many homologs of mammalian genes with established roles in energy balance [16]. In addition, we identified a class of genes for which mammalian homologs have not yet been implicated in fat regulation. These genes represent potential new directions in obesity research.
The Drosophila split ends (spen) gene is essential for viability and encodes an extremely large (>5,500 amino acids) RNA-binding protein known to regulate the transcription of key effectors of a number of signaling pathways. Spen promotes Wingless (Wg) signaling in flies and the orthologous Wnt signaling pathway in mammals [20, 21], and suppresses Notch signaling in flies and mammals [22–26]. Spen contains three RNA recognition motifs (RRMs) near its N terminus and, near the C terminus, the archetype Spen paralog and ortholog C-terminal (SPOC) domain [27]. Spenito (Nito), a much smaller (793 amino acids) Spen family member with RRMs and a SPOC domain, acts redundantly with Spen to promote Wg signaling [28], whereas during eye development it acts antagonistically to Spen [29]. Nito has additional roles in sex determination [30, 31] and neuronal function [31]. Importantly, the mammalian homologs of both Spen (SPEN/MINT/SHARP, hereafter mSpen) and Nito (Rbm15/OTT1, hereafter mNito) were recently found to be regulators of X chromosome inactivation via RRM-mediated interactions with the long, noncoding RNA (lncRNA) Xist [32–35]. In addition to activation or repression of transcription, Spen family proteins influence alternative splicing [30, 31, 36–39] and nuclear export of RNAs [36, 40, 41], and are commonly mutated in cancers [20, 42], but mechanistic details are lacking. Identification of spen hypomorphs in our unbiased screen for fat mutant larvae [16] represented the first evidence that Spen family proteins have a role in organismal adiposity. spen was independently identified in a subsequent genome-wide RNAi-based screen for increased adiposity in adult flies [43]. Mutation of the Spen homolog in C. elegans, Din-1, strongly increased stored fat, indicative of a conserved role in the regulation of fat storage [44]. However, these studies did not determine in which tissue Spen or its homologs act to control fat storage, or what defects in metabolism resulted in (or were reflected by) the accumulation of stored fat.
Here we analyze Spen and Nito function in the regulation of body fat in Drosophila larvae using a combination of genetic, cell biological, and biochemical approaches. We further monitor adipose tissue expression of mSpen and mNito in response to diet-induced obesity. Our results suggest a conserved RRM-mediated role for Spen homologs in the control of energy metabolism in fat storage tissues.
Third instar (L3) larvae homozygous for a hypomorphic P-element insertion allele in the spen locus float in a sucrose solution in which control larvae sink, indicative of lower overall density and consistent with elevated body fat [16, 45]. Most larval fat is stored in the FB [10, 15, 19]. To test if Spen is required specifically in the FB to prevent excess fat accumulation, we measured larval density upon FB-restricted (via a dcg>GAL4 driver [46, 47]) expression of one of five distinct spen-targeting RNAi constructs. In all five cases, FB knockdown of Spen (dcg>iSpen, hereafter referred to as Spen KD) resulted in lower density compared to both a knockdown control (dcg>iw) and genetic background controls (iSpen/+, iw/+, and dcg/+) (Fig 1A and S1 Fig), recapitulating the whole animal mutant phenotype. Buoyancy/density correlates strongly with adiposity as assessed directly via gas chromatography coupled with mass spectrometry (GC/MS) to measure levels of neutral lipids [16]. We calculated percentage body fat in this way for the same animals tested by the buoyancy assay. KD of Spen in the FB increased body fat by ~18% (Fig 1B, mean ± SEM 8.0% ± 0.2% for Spen RNAi compared to 6.8% ± 0.1% for w RNAi control; P < 0.01 by ANOVA). Notably, although females of all genotypes stored more fat than males, for both sexes the increase in buoyancy resulting from Spen depletion was similar (mean fold change for all sucrose concentrations ± SEM, 7.9 ± 1.5 for females and 6.0 ± 1.3 for males, P = 0.34 by unpaired t test) (S1B and S1C Fig). Additionally, a trans-heterozygous combination of hypomorphic spen alleles [48] resulted in a similar density phenotype (S2A Fig). A smaller but significant decrease in density was also observed in larvae heterozygous for a null and a wildtype (WT) allele [49] (S2B Fig). Levels of glycogen, the other major form of energy stored in the FB, were also decreased in larvae when Spen was depleted (Fig 1C). FB-restricted Spen overexpression (dcg>Spen) was sufficient to drive fat depletion (Fig 1D). Our findings thus support a FB role for Spen in control of fat storage.
Both food intake and energy expenditure can influence levels of stored fat [50, 51]. To ask if changes in feeding and foraging behaviors contributed to the increase of fat levels in the Spen KD larvae, we assessed food consumption and locomotion. Spen KD in the FB increased food intake in early L3 larvae compared to controls (Fig 1E). Furthermore, pre-wandering L3 larvae showed decreased locomotor activity (Fig 1F). Both behavioral changes align with the increased stored fat in these animals. By contrast, no change in food intake or locomotion accompanied the lean phenotype resulting from FB-restricted Spen overexpression (S3A and S3B Fig), indicating that behavioral changes did not contribute to the decrease in energy stored as fat. We conclude that overexpressed Spen acts autonomously in the FB to produce these effects.
Changes in levels of stored fat can result from changes in FB cell size or number [52, 53] or lipid droplet (LD) morphology or density [54]. To better understand the effects of Spen manipulation, we generated by flp-mediated recombination [55, 56] clones of FB cells in which Spen was either knocked down or overexpressed, surrounded by WT FB cells. GFP was co-expressed in both conditions to mark construct-expressing cells, and LDs were labeled with the lipophilic Nile Red [57] (Fig 1G–1J). Spen depletion caused significantly larger and more intensely stained LDs compared to controls (Fig 1G and 1H and S4A and S4B Fig), although FB cell size and number were unaffected (S4C and S4D Fig). FB cells overexpressing Spen were smaller, with LDs of normal size and staining intensity (Fig 1I and 1J and S4E–S4G Fig). As with Spen depletion, the number of FB cells per clone was unaffected by Spen overexpression (S4H Fig). spen FB mutant clones resulting from flp-mediated mitotic recombination in a heterozygous background produced significantly larger and more brightly stained LDs compared to WT clones (S2C–S2H Fig). We conclude that Spen functions autonomously in FB cells to regulate the amount of fat stored in LDs.
Despite their propensity to accumulate extra fat, larvae in which Spen was depleted from the FB died more rapidly than controls when reared from hatching in a sucrose solution, i.e., deprived of fats and amino acids (Fig 2A). Spen overexpression had no effect (S3C Fig). If energy stores can be accessed normally, excess energy in the form of fat can provide a crucial advantage during starvation [16, 58]. On the other hand, the advantage is lost regardless of the abundance of stored energy if the mutant animals are unable to mobilize it. Larvae lacking Spen in the FB thus appeared to be defective in accessing energy stores and/or in extracting energy from a limited diet, and may be in a state of “perceived starvation”. Either defect could drive the overfeeding and lethargy that we observed with a regular diet (Fig 1E and 1F). Indeed, FB-specific Spen depletion also caused a one-day developmental delay (19.8 hours ± 1.3 hours as compared to dcg>iw), consistent with a dearth of available energy, although we cannot exclude other causes. These results point to a role for Spen in regulating the liberation of energy stored as fat in the FB.
Spen and its homologs influence other pathways via control of transcription [59–61]. Accordingly, we suspected that the transcript levels of key metabolic enzymes would be affected by Spen manipulation in the FB, and tested this prediction using RNA sequencing (RNAseq). 440 of the 516 genes whose levels significantly changed when Spen was KD in the FB were classified by the PANTHER system [62]. 173 (39.3%) of the classified genes were categorized as being involved in a “metabolic process”, representing the largest functional “biological process” category (followed by “cellular process”, 30.7%). We observed striking changes in transcripts encoding proteins involved in β-oxidation, a process by which fatty acids are broken down to provide acetyl-CoA for the TCA cycle. Though redundant enzymes participate in β-oxidation reactions, three key enzymes involved in this pathway were significantly downregulated in Spen KD larvae (Fig 2B–2D), namely acyl-CoA dehydrogenase, enoyl-CoA hydratase, and 3-hydroxyacyl-CoA dehydrogenase. These enzymes participate in the release of a two-carbon chain from the fatty acid. Furthermore, significant downregulation of trehalase (Fig 2E) pointed to a potential blockage in disaccharide catabolism and, as a consequence, glycolysis.
With regard to the high-fat phenotype of Spen-depleted larvae, three lipases, potentially necessary for liberating stored fat, were downregulated (Fig 2F). Additionally, PEPCK (phosphoenolpyruvate carboxykinase) was highly induced (Fig 2G), a hallmark of the starvation response [63] that fits with the predicted state of “perceived starvation” resulting from an inability to access stored fats or dietary energy. Furthermore, 39 of the 126 genes significantly upregulated in Spen-depleted FBs are induced by fasting/starvation (S1 Table and [63]), and 69 of the 390 genes significantly downregulated in Spen-depleted FBs, are downregulated upon fasting/starvation (S2 Table and [63]), providing additional evidence of the similarities between Spen depletion and starvation. These findings provide strong support for the role of Spen in modulating substrate utilization for catabolism and energy production.
To define at a molecular level the metabolic defects accompanying Spen manipulation, we performed Ultra-High Pressure Liquid Chromatography (UHPLC)-MS-based metabolomic analysis on larvae in which Spen was knocked down or overexpressed in the FB, along with appropriate controls for each. We monitored 178 metabolites, and found that nearly every metabolite involved in glycolysis was significantly decreased in Spen-depleted larvae (Fig 3), consistent with a depletion in these animals of key sources of usable energy, and an accumulation of molecules in which energy is stored. As in most insects, trehalose is the primary circulating sugar in Drosophila, and is broken down to glucose to fuel cellular processes [64, 65]. Consistent with the reduction of trehalase observed by RNAseq (1.9-fold decrease, P = 0.0009, Fig 2E), trehalose levels were significantly elevated in Spen KD larvae (Fig 3), indicating impaired conversion into glucose and thus decreased glycolytic intermediates.
In addition to defective mobilization of carbohydrate sources for energy production, we found clear defects in β-oxidation. Acyl-carnitines are key intermediates of β-oxidation that permit fatty acid transport into mitochondria [66], which is the rate-limiting step of β-oxidation. Spen KD larvae were significantly depleted of free carnitine as well as nearly every medium- and long-chain fatty acyl-carnitine (Fig 3 and S5A Fig), consistent with the observed decreases in β-oxidation enzymes (Fig 2B–2D) and suggestive of a defect in β-oxidation. Finally, the levels of many free amino acids decreased in Spen KD larvae (S5B Fig), while markers of protein catabolism n-acetylmethionine and hydroxyproline were increased in Spen KD larvae and decreased in Spen-overexpressing ones (Fig 3), consistent with increased proteolysis in response to Spen KD. Among the transcripts that increased significantly in Spen KD larvae are three predicted trypsin-family proteases (CG11529, CG31326, and CG8299, the latter increased ~80-fold) that may be good candidates to mediate elevated protein catabolism. These metabolic changes provide direct evidence of a defect in energy mobilization via catabolism of carbohydrate and lipid energy sources, and may indicate the use of amino acids as an energy source.
Importantly, FB overexpression of Spen had effects opposite to that of Spen depletion with regards to β-oxidation, including increased levels of carnitine (Fig 3). Spen overexpression did not significantly alter glycolytic metabolites or acyl-carnitine levels, although the steady state of some TCA cycle intermediates and a few amino acids were slightly elevated (Fig 3 and S5C and S5D Fig). These findings further indicate that Spen regulates fat catabolism.
Despite the extreme size of the Spen protein, only the RRMs and SPOC domain have been functionally characterized in the context of other pathways. We obtained two Spen truncation alleles, one that lacks the C-terminal region including the SPOC domain but retains the RRMs (ΔSPOC), and one that retains only the C-terminal region and lacks the RRMs (SPOConly) (Fig 4A). In other contexts, each allele can behave in a dominant-negative fashion. For example, expression of ΔSPOC in midline glial cells results in completely penetrant lethality [67]. Expression of SPOConly with an engrailed driver reduces or eliminates Senseless expression, suggesting an absolute requirement for this domain of Spen in its regulation of Wg signaling [21]. To test for dominant negative effects in Spen regulation of fat storage, each of these alleles was overexpressed in the FB. If the truncation had no effect, we predicted that overexpression would cause a lean phenotype, as observed with Spen overexpression using two independent constructs (Figs 1D and 4F), whereas a dominant-negative effect would result in a similar phenotype to Spen depletion and elevate fat (Fig 1A).
ΔSPOC-overexpressing larvae were unable to survive at 25°C or 18°C, arresting at the L2 stage. At 16°C, where Gal4 is less active and levels of overexpression are lower [68], development was delayed by 11–13 days compared to controls (12 days 2.28 hours ± 23.3 hours as compared to dcg>GFP) and only 5–10% of larvae survived to L3. Although we cannot exclude a neomorphic effect, we favor the interpretation that this developmental delay is an extreme version of the one-day delay observed upon Spen depletion, and thus is a manifestation of “perceived starvation” resulting from dominant inhibition of Spen function in catabolism. L3 larvae obtained at 16°C were tested by buoyancy and compared to larvae overexpressing GFP, SPOConly, or a full-length Spen construct (Spen-FL). Whereas Spen overexpression decreased larval buoyancy (Figs 1D and 4F), expression of ΔSPOC strongly increased larval buoyancy, and expression of GFP or SPOConly had no effect (Fig 4F and S6A Fig). Analysis of feeding and activity showed no significant changes (S6B and S6C Fig). By staining isolated tissues of ΔSPOC larvae with Nile Red to label neutral lipids [57], we noticed a striking phenotype resulting from ΔSPOC overexpression. FBs in these larvae were almost non-existent (Fig 4B and 4C), but the FB tissue that remained stained much more brightly and contained very large LDs, some of which appeared to have “leaked” out of FB cells (Fig 4D–4E). Unlike tissues from control animals, brighter staining was also observed in the brains, imaginal discs, and guts of larvae overexpressing Spen ΔSPOC in the FB (S7 Fig). The appearance of fat deposits in tissues where fat does not normally accumulate is consistent with the elevated body fat phenotype, and is reminiscent of similar effects documented in the Drosophila Seipin mutant lipodystrophy model [69, 70].
Analysis of clones of FB cells expressing Spen-FL, ΔSPOC, or SPOConly along with GFP revealed that, as with other full-length Spen overexpression constructs (Fig 1I and 1J and S4G Fig), Spen-FL overexpression resulted in smaller FB cells (S8A, S8B and S8I Fig). While LD intensity was unchanged (S8F Fig), this particular Spen-FL transgene also decreased LD size (S8E Fig), a stronger phenotype than observed with the Spen-OEX transgene (Fig 1I and 1J and S4E Fig). SPOConly overexpression resulted in normally-sized FB cells with no significant changes in LD or cell size or morphology (S8A, S8C and S8G–S8J Fig), indicating that the SPOC domain is required for the ability of Spen to deplete stored fat when overexpressed. ΔSPOC overexpression, on the other hand, caused a striking phenotype suggestive of catastrophic defects in metabolism. Specifically, many of the clones consisted of a few extremely small cells containing nuclei (marked with strong GFP signal) and little else (Fig 4J and 4K). ΔSPOC overexpression may cause pleotropic defects, including cell death. However, considering that similar effects have been previously documented for FB cells during starvation [71–73], we favor a model in which the SPOC domain is required for normal Spen function in fat regulation and RRMs alone sequester crucial factors in a non-functional manner. Hence, overexpressing a version of Spen harboring the RRMs but lacking the SPOC domain perturbs the ability of full-length Spen to interact with such factors and carry out its normal function(s).
FB overexpression of full-length Spen had no effect on survival during starvation (Fig 4H). Both ΔSPOC and SPOConly were significantly more sensitive to starvation than controls (Fig 4H), very similar to Spen KD, although the ΔSPOC effect was far stronger. The ability of SPOConly overexpression to dominantly curtail survival during starvation contrasts with the lack of observed effects on buoyancy or LD appearance in FB cells, and suggests that the roles for Spen in fat storage and the starvation response are not strictly coupled. Phenotypes of all Spen truncation lines are summarized in Fig 4M.
In other pathways, Spen and Nito function either redundantly (e.g. Wg signaling [28]) or antagonistically (e.g. EGFR pathway during eye development [29]). To determine the relationship between the two Spen family members in fat regulation, we first depleted Nito from the FB and tested buoyancy. Nito depletion caused a lean phenotype (Fig 5A), similar to Spen overexpression. Introducing one copy of a Nito null allele [30] caused a very slight lean phenotype (S2A Fig) that was lost with further outcrossing (S2B Fig), hence in the absence of unknown background modifiers Nito is haplosufficient to promote normal fat storage. FB clones in which Nito was depleted had modestly smaller cells and lipid droplets, consistent with the observed lean phenotype (Fig 5C and 5D and S9A and S9C Fig). Cell number and LD intensity were not affected (S9B and S9D Fig). To ask if excess Nito inhibits Spen, we overexpressed full-length Nito in the FB. Larvae were unable to complete development even when reared at 16°C, a phenotype reminiscent of the developmental delays observed upon Spen depletion or overexpression of Spen-ΔSPOC. Full-length Nito overexpression produced clones that consisted of tiny cells in which only the nucleus was discernable (Fig 5E and 5F), similar to the effects of Spen-ΔSPOC. While we cannot at this time rule out effects on cell survival factors unrelated to metabolism, this phenotype is consistent with FB cell death due to starvation [71–73]. Finally, we asked if Nito depletion or overexpression affected sensitivity to starvation. Nito KD larvae died slightly earlier than controls (Fig 5B), as would be expected for lean animals with fewer fat stores to draw upon. Overexpression of full-length Nito caused premature death under sucrose-only conditions (Fig 4I), consistent with defects in utilization of energy from stores and/or imbalanced diets.
The N-terminal RRMs were required for these effects of Nito overexpression, as larvae overexpressing an N-terminally truncated version (Nito-ΔN) developed normally at all temperatures and were indistinguishable from controls with regard to buoyancy or other metabolic behaviors (Fig 4A, 4G and 4M and S6D–S6F Fig). On the other hand, overexpression of Nito-ΔC, which retains the RRMs but lacks the SPOC domain, caused L2 arrest regardless of temperature. Nito-ΔN overexpression did not affect lipid storage, cell size, or cell number (S8A, S8D and S8K–S8N Fig). In striking contrast, expression of a Nito-ΔC construct lacking the SPOC domain phenocopied overexpression of full-length Nito, with the majority of clones containing tiny cells (Figs 4L, 4M and 5F). Importantly, the lack of phenotypes resulting from Nito-ΔN did not reflect a failure to localize to the nucleus, as both Nito truncations localize appropriately [29]. Nito-ΔC-overexpressing larvae were sensitive to starvation, similar to the effects of full-length Nito (Fig 4I and 4M). Overexpression of Nito-ΔN caused starvation sensitivity that was milder than what we observed for full-length Nito or Nito-ΔC (Fig 4I), analogous to the effects of Spen-SPOConly overexpression (Fig 4H, 4M). Taken together, these data support a model wherein Nito antagonizes Spen function in catabolism of stored energy in a mechanism that requires both the RRMs and SPOC domain, with SPOC-less Nito RRMs able to act in a potent dominant-negative manner.
If mSpen and/or mNito function is important in preventing excess fat accumulation in mammals, we predicted that driving fat accumulation via a high-fat diet (HFD) might trigger changes in the expression of these genes in mice. For individual mice fed either normal chow or a HFD for 30 weeks, we measured both body fat percentage (mass of isolated white adipose tissue (WAT) divided by body mass) and, via RT-qPCR, mSpen or mNito transcript levels in the isolated uterine WAT. The HFD increased body fat by ~2.6-fold on average (54.2 ± 1.8%, n = 7 for HFD compared to 20.9 ± 2.9%, n = 5 for normal chow, unpaired t test P < 0.0001). Strikingly, both mSpen and mNito transcript levels (normalized to levels of 4 housekeeping genes) correlated strongly with body fat percentage (Fig 6, R = 0.65, P < 0.05 for mSpen and R = 0.74, P < 0.01 for mNito by unpaired two-tailed t test). While further studies will be required to determine how changes in mSpen and mNito expression in animals made obese by a HFD reflect the normal functions of these proteins, we take these data as evidence that Spen and Nito functions in fat storage are conserved from flies to mammals.
Our work provides the first detailed investigation of a fat regulatory role for Spen in any organism, and the first evidence that Nito also functions in this process. Spen depletion in the FB drastically increased stored fat (Fig 1A and 1B). Spen has been implicated in multiple pathways involved in endocrine signaling, including Notch [49, 74], Wingless [21], and nuclear receptor signaling [44, 61, 75]. We find it unlikely that nuclear receptor pathways are relevant to the fat regulatory role we define, because we did not observe upon Spen depletion or overexpression consistent changes in the expression of genes that are targets of those pathways. Furthermore, the lack of phenotypes involving fat storage per se upon overexpression of Spen-SPOConly (Fig 4F) argues against a role for Wg signaling, in which the same construct has potent dominant negative effects [21]. Conversely, whereas a C-terminally truncated version of mSpen has little effect on Notch signaling [23], the strong fat phenotypes resulting from Spen-ΔSPOC overexpression suggest that Spen does not regulate fat via the Notch pathway.
Notably, Spen KD larvae also exhibited behavioral changes (increased food intake, decreased locomotion) that may have contributed to the fat increase (Fig 1E and 1F). Thus, in addition to direct roles in fat accumulation within fat storage cells, Spen may be involved in a cross-talk pathway between the FB and the brain. However, we strongly support a model wherein increased food intake is instead an attempt to compensate for a condition of “perceived starvation” resulting from an inability to access energy stores. Similarly, a lack of available energy may restrict locomotion. This hypothesis is further strengthened by the observation that Spen overexpression was sufficient to deplete stored fat (Fig 1D) but did not cause opposing behavioral phenotypes (S3A and S3B Fig).
Mosaic analysis confirmed an autonomous role for Spen in FB cells. Spen KD in clones throughout the FB showed a striking increase in LD size (Fig 1H and S4A Fig). Larger LDs normally have lower surface tension, and the stored fat is easier to access [76]. LD remodeling in WT animals is a highly regulated process involving specific factors, some of which were identified in a genome-wide RNAi screen in cultured Drosophila S2 cells [14, 77, 78]. Notably, our RNAseq data revealed that the products of several LD-regulating genes were significantly altered by Spen depletion, including l(2)01289 (~7-fold decreased, P < 0.0001 by unpaired two-tailed t test), CG3887 (1.3-fold decreased, P = 0.001), and eIF3-S9 (1.5-fold increased, P = 0.0008). While it is unclear if these changes are direct effects of Spen depletion, they may explain why LDs in Spen KD larvae are large yet apparently inaccessible, resulting in starvation sensitivity.
Consistent with the observed changes in FB cell and LD morphology and starvation sensitivity, changes in metabolites and gene expression in Spen KD larvae pointed to a drastic defect in lipid catabolism. Defects in β-oxidation were the most obvious, in part because the opposite effects were observed upon FB-restricted Spen overexpression. Spen depletion led to a decrease in the levels of free and acyl-conjugated carnitine, as well as of transcripts of three of the four enzymes necessary to break down acyl-carnitines into free fatty acids (Fig 2B–2D and Fig 3). Three lipases were also downregulated (Fig 2F), which likely further contributes to an inability to convert energy stored as TAGs into usable forms. While an apparent upregulation of gluconeogenesis is evident, as supported by alterations in aspartate (Fig 3) and PEPCK expression (Fig 2G), these processes may be unable to completely compensate for decreased trehalose utilization, and these defects may contribute to the lethargy phenotype resulting from Spen KD. Consequently, surviving the loss of Spen may require breakdown of protein into free amino acids in order to anaplerotically replenish the TCA cycle, consistent with changes in expression of proteases, the observed decrease in many free amino acids (S5B Fig), as well as increases in protein catabolism and collagen turnover markers (N-acetylmethionine and hydroxyproline) (Fig 3). Of note, sustained proteolysis is a marker of aging and inflammation, a phenotype that has been previously associated with decreased locomotion in human and mouse models of physical activity, suggesting potential future ramifications of Spen’s role in metabolism with respect to aging/inflammation research [79]. Finally, the observed decrease in glycogen levels upon Spen KD (Fig 1C) supports a model wherein glycogen is used as a carbohydrate source (in lieu of decreased levels of trehalose (Fig 2E and Fig 3)) to fuel glycolysis. The overall metabolic defects we describe are distinctly different from what has been observed upon manipulation of other fat regulators (e.g. Sir2 [16]), suggesting that Spen operates in a previously undescribed pathway.
Our results with Spen and Nito truncations provide additional mechanistic insight into how these proteins function in fat regulation. Overexpressing Spen-ΔSPOC reversed the phenotype of full-length Spen overexpression, and instead resulted in similar phenotypes to Spen depletion. Nito-ΔC overexpression had the same effects: larvae arrested development and FB clones mimicked starvation even when dietary nutrients were abundant. Overexpression of the Spen-SPOConly construct had no effect on FB cells, as was the case for Nito-ΔN. Thus only Spen harboring the RRMs and the SPOC domain was able to promote fat depletion when overexpressed. Conversely, only truncated forms of Spen or Nito that retain the RRMs dominantly perturbed both FB cell viability and organismal resistance to starvation.
Recent studies of X chromosome inactivation found that mSpen RRMs mediate binding to the lncRNA Xist [32–35]. Rbm15 (mNito) also binds Xist [32, 33], and is required for N6-methyladenosine (m6A) modification of that lncRNA, which is in turn required for its ability to repress X chromosome transcription [80]. Nito is a subunit of the Drosophila m6A methyltransferase complex and is required for RNA binding by that complex; Nito knockdown severely decreases global m6A modification of mRNA [31]. Interestingly, the m6A demethylase FTO/ALKBH9 was the first human obesity susceptibility gene identified by genome-wide association studies [81–83], but the relevant nucleic acid target(s) remain unknown. Our work provides the first hint that an RNA bound by Spen and/or Nito may be a key FTO substrate.
These findings lead us to propose a model for Spen and Nito function in the regulation of fat storage (Fig 7). Spen binds via its RRMs to one or more RNAs and, via recruitment of other factors, promotes the expression of enzymes key for mobilization of energy stored as fat (e.g. lipases). The mechanism of activation may be direct or indirect, and via alternative splicing, activation/repression of transcription, or effects on RNA stability and/or translation. Moreover, RNA binding partners may be mRNA or non-coding RNA. Future work will be required to make these distinctions. We propose that the Spen SPOC domain is critical for this function, but undefined domains in between the N-terminal RRMs and C-terminal SPOC domain are also important, and these are not shared with Nito. We propose that Nito binds via its RRMs the same or a largely overlapping set of RNAs, and also recruits additional factors via its SPOC domains, but–either because it fails to recruit specific factors recruited by Spen, or because it recruits other factors not recruited by Spen—Nito ultimately inhibits/represses the same energy-storage-mobilizing enzymes that are activated by Spen (Fig 7). Overexpressed Spen or Nito fragments containing RRMs sequester target RNAs away from endogenous full-length Spen and the other effectors of fat storage control. Finally, our findings in mouse adipose tissue that mSpen and mNito both increase in expression when a HFD drives fat accumulation lead us to believe that in WT animals Nito acts as a counterbalance to Spen in order to fine-tune fat storage. Future studies delving into more mechanistic details may lead to treatments for obesity and related metabolic disorders that result from perturbation of the pathway that we elucidate here.
W1118 (3605), w; dcg>Gal4 (7011), y1 sc v1; +; UAS-Spen RNAi (33398), y1 v1; UAS-Spen RNAi (50529), y1 sc v1; +; UAS-Nito RNAi (34848), y1 v1; +; UAS-w RNAi (28980), y1 w; UAS-Spen (20756), w; UAS-GFP (9331), w; +; UAS-GFP (9330), Spen14O1 (5808), Spen16H1 (5809), Spen3 (8735), and Spen5 (8734) were obtained from the Bloomington stock center. Spen14O1 is an hypomorphic allele while Spen16H1 is a null [48]. Spen3 and Spen5 are null alleles, caused by small deletions in the Spen locus leading to truncations of the protein [49]. w; +; UAS Spen RNAi (48848), w; UAS Spen RNAi (45943), w; UAS Spen RNAi (108828), and w; UAS w RNAi (30033) we obtained from the Vienna Drosophila Resource Center (predicted off-targets in Table 1). w; UAS-SPOConly, w; UAS-ΔSPOC, w; UAS-Spen-FL, w; UAS-Nito-ΔN, w; UAS-Nito-ΔC, and w; UAS Nito-FL were generous gifts from Ilaria Rebay and Ken Cadigan. ΔSPOC contains all but the last ~1500 amino acids of Spen [67] while SPOConly contains only the last 936 amino acids of Spen as well as a nuclear localization signal [21]. Nito-ΔC contains the first 593 amino acids of Nito while Nito-ΔN contains only the last 322 amino acids of Nito [29]. Nito1 is a null mutant that was a generous gift from Norbert Perrimon [30]. w; act>cd2>Gal4 UAS-GFP was obtained elsewhere [55]. y1 sc v1; +; UAS-Spen RNAi (33398) was used for all Spen KD experiments excluding those explicitly stated otherwise. y1 w; UAS-Spen (20756) is an EP overexpression line used for overexpression experiments including the initial density assay, RNA sequencing, and metabolomics analysis. w; UAS-Spen-FL was a generous gift from Bertrand Mollereau [84] and is a full-length Spen insertion used for subsequent overexpression experiments including truncation density and starvation assays and clonal analysis. Similar results were obtained with both the Spen-EP and Spen-FL lines. Animals were reared at 25°C unless otherwise specified and fed a modified Bloomington media (with malt) containing 35g yeast per liter. Food was made fresh each week and used within the week. Eggs were collected on grape plates at 25°C and 50 first-instar larvae were transferred 22–24 hours later into a vial of food.
Density assays were performed as previously described [16, 45] with 50 larvae per sample (n = 8 samples per genotype). For sex-specific density assays, two samples of 50 larvae each were collected, pooled, and sorted for sex prior to performing the assay. ANOVA was used to calculate statistical significance with Prism 6 software.
Ten larvae from the group tested in the buoyancy assay (including both floaters and sinkers) were collected, frozen in liquid nitrogen, and weighed as a group. Larvae were homogenized and neutral lipids were extracted and analyzed as previously described [16, 85] using a Thermo Fisher Trace 1300-ISQ GC/MS system. n = 8. ANOVA was used to calculate statistical significance with Prism 6 software.
Ten wandering third instar larvae were collected and frozen in liquid nitrogen. Larval samples were prepared using the Hexokinase (HK) Assay Kit (Sigma, St. Louis, MO) as described [86]. Briefly, animals were homogenized and heat treated. Sample was divided into two sets, one which was treated with amyloglucosidase to digest glycogen and one that was treated with PBS. These samples along with glycogen and glucose standards treated similarly were incubated for 1 hour at 37°. 100 μL HK reagent was added to each standard and sample and measured for absorbance at 340 nm in 96-well plates using a Cytation 3 plate reader (BioTek, Winooski, VT). Glycogen levels were determined by subtracting the absorbance measured for the untreated samples (basal glucose level) from the amyloglucosidase treated samples. n = 17. ANOVA was used to calculate statistical significance with Prism 6 software.
Thirty early L3 larvae were collected, placed on a spot of yeast paste containing 0.5% food dye FD&C Red #40 on an agar plate at 25°C and the larvae allowed to eat for 30 minutes and processed as previously described [16]. n = 4. ANOVA was used to calculate statistical significance with Prism 6 software.
Pre-wandering L3 larvae were collected and tracked for movement as previously described [87]. n = 4. Two-tailed unpaired t test was used to calculate statistical significance with Prism 6 software.
Wandering third instar larvae of the genotypes hs flp; act>cd2>gal4 UAS-GFP, hs flp; act>cd2>gal4 UAS-GFP UAS-Spen RNAi, hs flp; act>cd2>gal4 UAS-GFP UAS-Nito RNAi, hs flp; act>cd2>gal4 UAS-GFP UAS-w RNAi, hs flp; act>cd2>gal4 UAS-GFP UAS-Spen, hs flp; act>cd2>gal4 UAS-GFP UAS-Spen-ΔSPOC, hs flp; act>cd2>gal4 UAS-GFP UAS-Spen-SPOConly, hs flp; act>cd2>gal4 UAS-GFP UAS-Spen-FL, hs flp; act>cd2>gal4 UAS-GFP UAS-Nito-ΔC, hs flp; act>cd2>gal4 UAS-GFP UAS-Nito-ΔN, and hs flp; act>cd2>gal4 UAS-GFP UAS-Nito-FL were dissected, fixed and stained with Nile Red (Invitrogen), as described in more detail elsewhere [16]. Stained tissues were imaged on a Leica TCS SP5 laser-scanning confocal microscope with LASAF software.
Mitotic clonal analysis was performed using larvae of the genotype hs flp; FRT40A ubi>GFP / FRT40A and hs flp; FRT40A ubi>GFP / FRT40A Spen5. These animals were heat shocked directly after egg deposition for 3 hours at 37°. Larvae were collected at wandering stage and dissected, fixed, and stained with Nile Red as above.
Clones were analyzed for LD size and intensity using an algorithm written for ImageJ. Briefly, all clones were outlined and region location recorded. The FB tissue boundary was selected based on threshold. Once clone and tissue boundaries were defined, LDs were automatically outlined based on intensity threshold of the LD and measured for size and average pixel intensity. LDs of each clone were then compared to the LDs from surrounding non-manipulated cells as well as to KD or OEX control clones. FB cell size was analyzed by manually outlining each cell within the clones and measuring for area. Two-tailed unpaired t tests were used to calculate statistical significance with Prism 6 software. FB cell number was calculated by manually counting the number of cells within each clone. ANOVA was used to calculate statistical significance with Prism 6 software.
Fifty larvae were placed in 20% sucrose/PBS and analyzed daily for survival. Dead larvae were removed immediately after scoring and the sucrose was changed daily. Log-rank test was used to calculate statistical significance with Prism 6 software.
Forty larval fat bodies were dissected for each genotype and total RNA was extracted using Trizol (Life Technologies) reagent following manufacturer's instructions. A total of 200–500 ng of total RNA was used to prepare the Illumina HiSeq libraries according to manufacturer’s instructions for the TruSeq Stranded mRNA Library Prep Kit. The mRNA template libraries were sequenced on the Illumina HiSeq4000 platform at the University of Colorado’s Genomics and Sequencing Core Facility using a 1x50bp format. Derived sequences were analyzed by applying a custom computational pipeline consisting of the open-source gSNAP [88], Cufflinks, and R for sequence alignment and ascertainment of differential gene expression [89–93]. Briefly, reads generated were mapped to the Drosophila genome by gSNAP [88], expression (FPKM) derived by Cufflinks [94], and differential expression analyzed with ANOVA in R. GO annotations were predicted using Panther 11.1 [62], Gene Ontology versions 1.2, annotation 2017-04-24.
Briefly, individual larvae (10 per sample, n = 3 samples per genotype) were suspended in 1 ml of methanol/acetonitrile/water (5:3:2, v/v) pre-chilled to -20°C and vortexed continuously for 30 min at 4°C. Insoluble material was removed by centrifugation at 10,000xg for 10 min at 4°C and supernatants were isolated for metabolomics analysis by UHPLC-MS. Analyses were performed as previously described [95–97] using a Vanquish UHPLC system coupled to a Q Exactive mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA. Graphs, heat maps and statistical analyses (either t-Test or ANOVA) were performed with GraphPad Prism 5.0 (GraphPad Software, Inc, La Jolla, CA).
All procedures involving animals were performed in accordance with published National Institutes of Health Guidelines. The University of Colorado Anschutz Medical Campus Institutional Animal Care and Use Committee approved this study and all procedures and housing conditions used to complete it. Mice were housed in facilities at the Anschutz Medical Campus’s Center for Comparative Medicine with free access to food and water for the study’s duration (22–24°C; 14:10 h light-dark cycle). Female C57BL/6 mice were bred in house. At 8 weeks of age mice either continued on chow diet (Harlan 2920xi) or were placed on a defined high fat diet (60% kcal fat; Research Diets D12492i) for 30 weeks. Body weights were collected weekly. Upon completion of the feeding experiments, determination of the body composition of each animal was performed by quantitative magnetic resonance EchoMRI-900 whole-body composition analyzer (Echo Medical Systems; Houston, TX). At termination, mice were euthanized with CO2, followed by heart puncture. Tissues were collected and immediately frozen in liquid N2. Total RNA was extracted from ~50 mg uterine adipose tissue for each mouse sample using Trizol (Life Technologies) reagent following manufacturer's instructions. RT was performed using Oligo d(T) 23 and M-MuLV Reverse transcriptase (NEB) per manufacturer's instructions. qPCR was performed using PowerUpTM SYBR® Green Master Mix (Applied Biosystems). Reactions were run in an Applied Biosystems Step One Plus qPCR machine. Primer sequences: mSpen: 5’-ggctctggttctctacagcg-3’ and 5’-ctccatgcagtgataaaatgcc-3’ mNito: 5’-gcactggccaaatctgaagaag-3’ and 5’-tccatcagaggcccatgtaaac-3’. mNito results were also confirmed with an independent primer pair. Two technical repeats were performed on 5–7 biological replicates (standard chow n = 5, HFD n = 7). Percent body fat was calculated by dividing the fat mass by total body weight. P value and correlation coefficient was obtained by unpaired two-tailed t test from the average of the technical repeats using Prism 6 software.
|
10.1371/journal.pgen.1007134 | Optical silencing of body wall muscles induces pumping inhibition in Caenorhabditis elegans | Feeding, a vital behavior in animals, is modulated depending on internal and external factors. In the nematode Caenorhabditis elegans, the feeding organ called the pharynx ingests food by pumping driven by the pharyngeal muscles. Here we report that optical silencing of the body wall muscles, which drive the locomotory movement of worms, affects pumping. In worms expressing the Arch proton pump or the ACR2 anion channel in the body wall muscle cells, the pumping rate decreases after activation of Arch or ACR2 with light illumination, and recovers gradually after terminating illumination. Pumping was similarly inhibited by illumination in locomotion-defective mutants carrying Arch, suggesting that perturbation of locomotory movement is not critical for pumping inhibition. Analysis of mutants and cell ablation experiments showed that the signals mediating the pumping inhibition response triggered by activation of Arch with weak light are transferred mainly through two pathways: one involving gap junction-dependent mechanisms through pharyngeal I1 neurons, which mediate fast signals, and the other involving dense-core vesicle-dependent mechanisms, which mediate slow signals. Activation of Arch with strong light inhibited pumping strongly in a manner that does not rely on either gap junction-dependent or dense-core vesicle-dependent mechanisms. Our study revealed a new aspect of the neural and neuroendocrine controls of pumping initiated from the body wall muscles.
| Since feeding is an essential behavior for the survival of animals, it is modulated by a variety of neural and neuroendocrine signals that are generated depending on internal and external conditions. To elucidate the cellular and molecular mechanisms underlying the regulation of feeding, the nematode Caenorhabditis elegans, which is composed of a small number of identifiable cells, provides a unique system. In C. elegans, the pumping movement of a feeding organ called the pharynx has been subjected to intensive genetic studies. Compared to the factors promoting pumping, however, the inhibitory mechanisms of pumping are less well understood. In this paper, we report that optogenetic silencing of the body wall muscles, which drive the locomotory movement of worms, inhibits pumping in the pharynx. Signals emanating from muscles are likely to trigger pumping inhibition, raising an interesting possibility that the proprioceptive sense detecting the relaxation of body wall muscles might be involved. When the Arch proton pump was activated with weak light, signals for pumping inhibition are transferred into the pharynx mainly through two pathways: one involving gap junction-dependent mechanisms through pharyngeal I1 neurons, which mediate fast signals, and the other involving dense-core vesicle-dependent mechanisms, which mediate slow signals. Strong activation of Arch inhibits pumping very strongly via other mechanisms. Thus, we have revealed a new link between pumping and the body wall muscles, and confirmed the important cooperation of neural and neuroendocrine circuits in the regulation of feeding behaviors.
| Feeding is essential for the survival, growth, and proliferation of animals. Accordingly, feeding behaviors change depending on both the internal physiological state and external environmental conditions [1]. For studying the neural and neuroendocrine mechanisms underlying the control of feeding behaviors, the nematode Caenorhabditis elegans, which is amenable to a variety of experimental manipulations, provides a unique system.
C. elegans takes food in through a hollow organ called the pharynx that is located in the head [2]. Morphologically, the pharynx consists of three parts: the corpus, isthmus, and terminal bulb (TB), in anterior to posterior order. The pharyngeal muscles in the corpus and the TB undergo cycles of synchronized contraction and relaxation [3]. This rhythmic movement, called pumping, allows the worm to swallow and filter foods such as Escherichia Coli in liquid and expel excess fluid. The pharynx also contains the pharyngeal nervous system that comprises 20 neurons of 14 types. Electron microscopy (EM) study showed that the pharynx is isolated from the rest of the body by a basement membrane, and the synaptic connection between the pharyngeal and the extrapharyngeal nervous systems is limited to a gap junction between the I1 neuron in the pharynx and the RIP neuron outside the pharynx [4].
The pharyngeal nervous system participates in the regulation of pumping. Remarkably, the pharynx can continue pumping even when all of the pharyngeal neurons are ablated, although the frequency of pumping decreases profoundly [5]. MC pharyngeal neurons are mainly responsible for high-frequency pumping; ablation of the other pharyngeal neurons did not affect pumping frequency significantly [6]. These results indicate that pumping of the pharynx is basically myogenic, similar to pumping of the vertebrate heart [7], with the pharyngeal nervous system playing mainly modulatory roles. The pharynx, however, stops pumping under certain conditions. Pumping ceases during lethargus, the period before molting [8]. Various types of stress, such as heat shock [9][10], physical stimulation to the body [11][12], and blue light [13][14], induce pumping quiescence. Thus, neural mechanisms that repress the spontaneous activity of the pharyngeal muscles must exist. The I1-RIP connection is, in fact, required for pumping inhibition caused by tail tap [15], though the details remain unknown.
Previous studies have also shown that the neuroendocrine system, including biogenic amines such as serotonin [16][17][18] and peptides [10] secreted from dense-core vesicles, is involved in the regulation of the pumping rate. Although the pumping rate decreases without food in wild-type (WT) animals, the rate remains high in unc-31 mutants [19], which are defective for post-docking calcium-regulated dense-core vesicle fusion. Pumping inhibition by exposure to a high level of CO2 is also impaired in unc-31 mutants [20]. A neuropeptide FLP-13 expressed in the extrapharyngeal ALA neuron mediates pumping inhibition after heat shock [10]. Although these findings underscore the importance of the neuroendocrine control of pumping, the manner in which peptides secreted by extrapharyngeal cells affect the pumping remains poorly understood.
Optogenetic tools have been widely used for controlling the activity of neurons and muscle cells by using light illumination in vivo. Archaerhodopsin-3(Arch) is a light-driven outward proton pump [21], frequently used in C. elegans for silencing muscle activity [22][23][24]. Here we report that optical silencing of the body wall muscles induces reduction in the pumping rate. Our genetic studies revealed that two independent pathways, one employing neural transmission and the other employing neurohumoral signals, are involved in the transmission of signals for the pumping inhibition response.
In a previous report, we expressed Archaerhodopsin-3(Arch), a green light-driven proton pump, in the body wall muscle cells of worms using the myo-3 promoter carried by an extrachromosomal array, ncEx3031. The transgenic animal, ST300, ceased forward and backward movement immediately after starting illumination with green light (550 nm) [22] (Fig 1A and 1B). At the same time, the body of the animal became straighter and longer. The arrest of the locomotory movement and body elongation were sustained throughout the 1 min period of illumination and, after illumination was stopped, the animal immediately recovered normal locomotory movement and body length. In addition, we noticed that illumination affected pharyngeal pumping. This unexpected finding prompted us to further examine pumping both during and after illumination.
We recorded pumping over a period of 3 min, which spans before (1 min), during (1 min), and after (1 min) illumination (Fig 1C). Green light illumination at 22 mW/mm2 gradually decreased the pumping rate, and pumping stopped completely 30 s after illumination was initiated. The pumping rate in animals that were grown without all-trans-retinal (ATR) as a control did not change at all upon illumination (Fig 1C). The pumping remained inhibited for a while after the light was turned off, in sharp contrast to the quick recovery of the locomotory movement and body length. The time lag between the response of the body to illumination and that of pumping suggests that the pumping inhibition is not the result of the silencing of the pharyngeal muscles by green light, but is instead a consequence of the silencing of the body wall muscles.
Previous studies have shown that the myo-3 gene encodes a myosin heavy chain expressed specifically in the body wall muscles and vulva muscles, but not in the pharyngeal muscles whereas the myo-2 gene, encoding another myosin heavy chain protein, is expressed specifically in the pharyngeal muscles. To confirm that Arch::GFP driven by the myo-3 promoter is not expressed in the pharyngeal muscles, we generated a transgenic line, ST326, carrying myo-2p::mCherry together with myo-3p::Arch::gfp. (S1A and S1B Fig)[25][26]. A fluorescence micrograph of ST326 showed that the GFP and mCherry signals did not merge in the pharyngeal muscles (S1C Fig). This indicates that green light did not silence the pharyngeal muscles directly in animals carrying myo-3p::Arch::gfp.
These results suggested that silencing the body wall muscles is the key event that triggers pumping inhibition. It is, however, possible that light activation of the Arch proton pump might have effects other than silencing cells. For example, it may lower the extracellular pH, which can consequently activate neighboring cells expressing acid-sensing ion channels (ASICs) [27]. It is also possible that hyperpolarization beyond the physiological level might have unknown pathological effects. To circumvent these problems, we attempted to silence the body wall muscles using ACR2, a natural light-gated anion channel [28][29]. In the transgenic line ST371 expressing ACR2::GFP by the myo-3 promoter, we found that blue light illumination at 1.5 mW/mm2, which immediately inhibited the locomotion of the animals, also caused pumping inhibition in a manner similar to that caused by Arch (Fig 1D).
The light illumination induced two apparent changes in the body: body elongation and arrest of locomotion. To test whether unperturbed forward movement is necessary for maintaining pumping at a high frequency, we examined pumping in mutants with locomotion defects. UNC-54 is a myosin heavy chain expressed in the body wall muscles, but not in the pharyngeal muscles. The unc-54 (e190) mutants rarely moved [30], but the pumping rate was not significantly different from that in WT animals. We found that illumination with green light in unc-54 mutants carrying myo-3p::Arch::gfp decreased the pumping rate (Fig 1E). The unc-15 (e73) mutants, whose paramyosin gene is affected [31], exhibit a locomotion-defective phenotype similar to that of the unc-54 mutants. We found that light activation of myo-3p::Arch in the unc-15 mutants also led to pumping inhibition (Fig 1F). Thus, hindered locomotion itself is not required for inducing pumping inhibition. Moreover, illumination induced elongation of the body in the unc-54 mutant background; the body lengthened by 1.5 ± 0.6% 0.5 s after illumination with green light was initiated (n = 4 worms). These observations imply that elongation of the body is correlated with pumping inhibition.
To reveal the neural mechanisms underlying the Arch-mediated pumping inhibition response, we examined various mutants defective in neural transmission (Fig 2). The number of pumps during the 10 s period immediately after starting illumination (Fig 2P) and immediately after stopping illumination (Fig 2Q) was counted in animals carrying myo-3p::Arch::gfp as an extrachromosomal array.
We first examined the involvement of major neurotransmitters, such as acetylcholine, glutamate, and GABA. Green light illumination induced pumping inhibition in mutants for genes encoding a vesicular acetylcholine transporter UNC-17/VAChT, a vesicular glutamate transporter EAT-4/vGluT, and a GABA-synthesizing enzyme UNC-25/GAD. Although the sample size is small, this indicated that none of the major classical neurotransmitters is sufficient on their own to fully convey information for the pumping inhibition.
It has been reported that the release of acetylcholine from MC neurons is important for the maintenance of high pumping rates [6]. As reported previously, in snt-1/synaptotagmin mutants defective for neurotransmitter release, the pumping rate is very low (Fig 2B) [32]. The rate decreased further during illumination, however. The unc-17/VAChT mutants defective for ACh release responded similarly to illumination (Fig 2D). These results suggest that the pumping inhibition is not completely caused by the reduction in acetylcholine release from MC neurons.
Although none of the mutations singly abolished the pumping inhibition response completely, two of them, unc-7 and unc-31, which partly suppressed the pumping inhibition response in different manners, attracted our attention. UNC-7 is one of the innexin proteins forming gap junctions in invertebrates [33]. In unc-7 mutants, the pumping rate decreased slowly during illumination, and also recovered slowly after illumination was stopped (Fig 2H and 2P). UNC-31 is C. elegans CADPS/CAPS required for the calcium-dependent secretion of dense-core vesicles [34]. In unc-31 mutants, the pattern of the induction of pumping inhibition did not significantly differ from that in WT animals, but inhibition was incomplete during the 1 min period of illumination. After the green light was turned off, the pumping rate quickly returned to the normal level (Fig 2J and 2Q). These results indicated the distinct roles of UNC-7-dependent gap junctions and UNC-31-dependent secretion of dense-core vesicles in conveying signals mediating the pumping inhibition response triggered by Arch-mediated body wall muscle silencing.
For further examination of the involvement of UNC-7 and UNC-31, we used ncIs53 worms carrying chromosomally integrated myo-3::Arch::gfp transgenes, which showed a pumping inhibition response similar to, but more constant than, that with the extrachromosomal transgene ncEx3031 used in the preceding sections. First, we examined the response to green light with an intensity of 10 mW/mm2 (Figs 3 and S3). Although almost all of the ncIs53 worms stopped pumping 40 s after starting illumination (Fig 3A); a fraction of the unc-7 (e5) mutants failed to stop pumping during the 1 min period of illumination (Figs 3B and 4E). The unc-7 mutants also exhibited a retarded response to turning on/off the light compared to the WT. Namely, they continued high-frequency pumping for a while after starting green light illumination, and took a longer time to recover from inhibition after the light was turned off (Figs 3B and 4B, 4H and 4K). Another unc-7 allele, e139, showed a similar phenotype (S4A Fig). In contrast, in the unc-31 (e928) mutants, pumping inhibition was rapidly induced, similar to that in WT animals, and post-illumination recovery to the normal pumping rate was instantaneous (Figs 3C and 4B and 4K). Another unc-31 allele, e169, showed a similar phenotype (S4B Fig). From these results, we speculated that the UNC-7-dependent pathway is responsible for triggering the rapid response whereas the UNC-31-dependent pathway mediates the slow and lingering response.
In order to examine the relationship between the UNC-7-dependent pathway and the UNC-31-dependent pathway, we generated unc-31(e928);unc-7(e5) double mutants. After 1 min illumination, the average reduction of pumping in the double mutants was limited by 40% (Figs 3D and 4E). This indicates that the UNC-7/innexin-dependent pathway and the UNC-31/CADPS/CAPS-dependent pathway play a major role in conveying signals for the pumping inhibition response caused by illumination of light with this intensity.
We then examined the dependence of the pumping inhibition response on the intensity of light. At a lower intensity of 2.5 mW/mm2, the pumping was also inhibited in WT, unc-7(e5), and unc-31(e928) animals, albeit to a lesser extent compared with that at 10 mW/mm2 (Figs 3A, 3B and 3C and 4D): pumping was reduced to 24%, 74%, and 46% on average, respectively, 1 min after initiating illumination. unc-7(e5) and unc-31(e928) animals exhibited a slow and rapid response to turning the light on/off, respectively (Figs 3B, 3C and 4A, 4J), similar to that with illumination at 10 mW/mm2. In contrast, pumping in unc-31(e928);unc-7(e5) double mutants was minimally affected (Figs 3D and 4D): the average reduction after illumination for 1 min was limited to 95%. This indicates that pumping-inhibiting signals caused by illumination with this intensity of light are mediated almost exclusively by the UNC-7-dependent and UNC-31-dependent pathways.
At a higher intensity of 40 mW/mm2, WT animals stopped pumping almost instantly when illumination was initiated, while the pumping rate in animals raised without ATR was not affected upon illumination (Fig 3A). unc-7(e5), unc-31(e928), and unc-31(e928); unc-7(e5) animals also exhibited a rapid reduction in pumping and stopped pumping almost completely 1 min after illumination was initiated (Figs 3B, 3C and 3D and 4F). This indicates that pumping inhibition with this intensity of light is mostly independent of both UNC-7 and UNC-31. Post-illumination recovery in unc-31(e928) and unc-7(e5):unc-31(e928) animals was rapid (Figs 3C, 3D and 4L) whereas pumping in WT and unc-7(e5) animals recovered only partly even 1 min after illumination was terminated (Figs 3A, 3B and 4I).
Taken together, these results support the notion that the UNC-7-dependent pathway mediates rapid signals, and the UNC-31-dependent pathway mediates slow signals, and that they function in parallel, at least in part, to inhibit pumping. At higher light intensity, there is a mechanism mediating a predominant inhibitory effect on pumping, which is independent of both UNC-7 and UNC-31.
The gap junction formed between I1 and RIP is the sole neural connection linking the pharyngeal and extrapharyngeal nervous systems [4]. It was reported that UNC-7 is expressed in the pharyngeal I1 neurons [33]. To test the possibility that the UNC-7-dependent pathway requires I1 neurons for transmitting inhibition signals from the extrapharyngeal nervous system, we ablated the I1 neurons. We performed the experiment with green light with an intensity of 5.5 mW/mm, where the average reduction of pumping in unc-7(e5):unc-31(e928) animals after illumination for 1 min was approximately 20% (Figs 5C and S6). The pumping in the I1(-) worm showed retarded responses to turning on the green light, similar to those in the unc-7 mutants compared to WT animals (p < 0.1. WT: n = 3; I1 (-): n = 4) (Fig 5A and 5B). In addition, we found that ablation of I1s in unc-31(e928) mutants exhibited a reduced pumping inhibition response: reduction in pumping after illumination for 1 min was, on average, 30% (Figs 5C and S6), which was comparable to that in the unc-31:unc-7 double mutants, and differed from the control unc-31 mutants (p < 0.05. unc-31: n = 3; unc-31 I1 (-): n = 5). These results indicate that ablation of I1s and defects of UNC-7 have similar effects on the pumping inhibition response, suggesting that the UNC-7-dependent pathway transmits the signals via I1 neurons. These results also indicate that the UNC-31-dependent pathway functions independently, at least partly, from transmission via RIP-I1, which is the only neural connection between the intra- and extra-pharyngeal nervous systems.
Various stimuli are known to induce pumping inhibition in C. elegans. Here we have reported for the first time that the activation of optogenetic silencers in the body wall muscles affects pumping. This newly found phenomenon provides a unique opportunity to study the regulatory mechanisms of pumping that operate under free-moving conditions.
Our genetic screen revealed two major pathways mediating signals for pumping inhibition caused by Arch-mediated silencing of body wall muscles: the UNC-7/innexin-dependent pathway and the UNC-31/CADPS/CAPS-dependent pathway. This confirmed the important cooperation of neural and neuroendocrine circuits in the regulation of feeding behaviors.
How does the activation of optogenetic silencers in the body wall muscle cells trigger the pumping inhibition response? We found that pumping was inhibited not only by activation of the Arch proton pump, but also by activation of the ACR2 anion channel, suggesting strongly that relaxation of the body wall muscle cells can in itself trigger the response. Our analyses using locomotion-defective mutants also showed that arrested movement caused by relaxation of the body wall muscle cells is not critical for the induction of pumping inhibition.
The cellular and molecular mechanisms underlying the presumed perception of the relaxation of the body wall muscle cells remain unclear. Muscle cells may emit some unknown retrograde signals in response to forced relaxation. Alternatively, changes in body posture caused by forced relaxation of the body wall muscles may play an important role. Interestingly, we found that pumping inhibition upon optical silencing of the body wall muscles, even in a locomotion-defective unc-54 mutant, concurred with the body elongation. We speculate that there is a possible involvement of the proprioceptive sense detecting the status of muscle cells. Muscles in animals have various types of proprioceptive organs that detect changes in the tension and length of the muscles [35]. In C. elegans, DVA neurons and B-type motor neurons have been reported to sense body curvature in order to control the bending of the body [36][37]. Certain mechanoreceptors, which are usually involved in sensing touch stimuli, may also play a role in detecting changes in body posture. Whether these neurons are involved in the perception of muscle relaxation that triggers the pumping inhibition response remains to be determined in future studies.
Although our result using the ACR2 anion channel strongly suggests that body wall muscle relaxation causes pumping inhibition, it does not exclude the possibility that other factors are involved in pumping inhibition caused by activation of the Arch proton pump. We found that activation of Arch with illumination at a high intensity (40 mW/mm2) has a strong and quick inhibitory effect on pumping, which does not rely on the UNC-7/innexin-dependent and UNC-31/CADPS/CAPS-dependent pathways. Strong activation of Arch might lead to changes in certain physiological conditions in the body, such as pH, which could directly affect the pharynx. The possible involvement of factors other than body wall muscle relaxation in the Arch-mediated pumping inhibition remains to be examined in future studies. Further analyses of pumping inhibition using the ACR2 anion channel would also be helpful to clarify the issue.
Our genetic experiments revealed that pathways mediating the pumping inhibition response partly depend on UNC-7/ innexins. UNC-7 is expressed in pharyngeal I1 and extrapharyngeal RIP neurons [33] to form gap junctions between them [4]. Since electrical synapses between I1-RIP are the only neural connection linking intra- and extra-pharyngeal neurons, it is likely that direct synaptic inputs to the pharynx are compromised in unc-7 mutants. Our finding that I1-ablation affected the pumping inhibition in a manner similar to the unc-7 mutation supports the notion that UNC-7 functions in I1s to transmit the pumping inhibition signal from RIPs into the pharynx.
A critical role of the electrical coupling of I1s and RIPs in the pumping inhibition response was previously suggested: Ivermectin, which is an agonist of the glutamate-gated chloride channel in C. elegans, kills worms through inhibition of feeding whereas unc-7 mutants are resistant to this drug [38]. It has also been reported that pumping inhibition caused by a tail tap was partly repressed in unc-7 mutants [12]. Recent studies have shown that optogenetic activation and silencing of I1 neurons induces an increase and decrease in the pumping rate, respectively [13][39]. Based on these facts, we speculate that relaxation of the body wall muscles is likely to result in hyperpolarization of I1s, which in turn induces the pumping inhibition response. It can also be speculated that the presumed hyperpolarization of I1s is transmitted from RIPs via gap junctions. The quick response of pumping to turning on/off the light in unc-31 mutants is consistent with the idea that the UNC-7-dependent pathway transfers fast signals in which ordinary chemical and electrical synapses are engaged.
Our genetic analysis using unc-31 mutants indicates that pathways mediating the pumping inhibitory signal partly rely on a dense-core vesicle-dependent mechanism. In unc-7 mutants, the pumping responded relatively slowly to turning the light on/off. These observations imply that the UNC-31-dependent pathway mediates humoral signals, which have slow and lasting effects; this is consistent with the known function of UNC-31/CAPS for secretion of neuropeptides and biological amines [34][40].
Previous studies reported the involvement of UNC-31 in pumping inhibition caused by different types of stimuli [19][41][42]. We found that the pumping inhibition by green light illumination occurs even when I1 neurons, the sole connectors between the extra- and intra-pharyngeal nervous systems, were ablated. Thus, in the UNC-31-dependent pathway, it is highly likely that factors secreted by extrapharyngeal tissues act directly on the pharynx. In fact, several neuropeptides that are secreted from extrapharyngeal neurons are known to affect pumping [43][44]. Notably, FLP-13 secreted by extrapharyngeal ALA neurons promotes quiescence, including repression of pumping following heat shock [10]. In our experiment, however, flp-13 mutations failed to affect the pumping inhibition response significantly, suggesting the participation of other inhibitory factors.
Although C. elegans continues pumping almost throughout its lifetime, the pumping rate is modulated under certain circumstances that influence the posture of worms. For example, the swimming movement of worms in liquid consists of body movements that are quite distinct from crawling on solid agar. When crawling worms start swimming in water, they stop pumping [45]. When the body is stabilized physically, worms also stop pumping [39]. Both heat shock and lethargus arrest the body movement concomitantly with reducing the pumping rate [8][46][47]. Therefore, it seems that pumping and body posture are somehow linked. Although the physiological relevance of pumping inhibition caused by relaxation of the body wall muscle cells remains to be examined in the future, it may play a role in coordinating the pumping with the body posture through feedback from the latter to the former.
C. elegans strains were grown on a bacterial lawn of E. coli OP50 on nematode growth medium (NGM) agar [48]. Animals were maintained at 20°C. Plates with all-trans-retinal (ATR) (Sigma-Aldrich, St. Louis, USA) were kept in the dark.
A destination vector containing the rig-3 promoter, which drives gene expression in a subset of neurons including I1 neurons, pDEST-rig-3p, was constructed by inserting the polymerase chain reaction (PCR)-amplified genomic fragment into the SphI site of pDEST-PL (a gift from Hidehito Kuroyanagi) using the following primers: 5’aaGCATGCggaaaaatgtgagatcttcgctgaaa3’ and 5’aaGCATGCgaatgaagttcttctgcaaggaatga3’. pGW-rig-3p::mCherry was generated by recombination between pDEST-rig-3p and pENTR-mCherry using the Gateway system (Invitrogen, San Diego, USA). pMT001: myo-3p::ACR2::gfp was generated by recombination between pDEST-myo-3p and pENTR-ACR2::gfp [29]. pOKA049: myo-3p::Arch::gfp was described previously [22].
Transgenic animals were generated by microinjection of DNA into the gonad of N2 hermaphrodites [49]. pOKA049 (myo-3p::Arch::gfp, 100 ng/μl) and pGW-rig-3p::mCherry (50 ng/μl) were injected together into N2 worms to create strain ST357 carrying ncEx9112. pOKA049 (75 ng/μl), pRF4 (rol-6d, 125 ng/μl), and pCFJ90 (myo-2p::mCherry, 10 ng/μl) were injected together to create strain ST326 carrying ncEx9198. Strain ST371 carrying ncEx3941 was generated by the injection of pMT001 (myo-3p::ACR2::gfp, 300 ng/μl) and pRF4 (rol-6d, 100 ng/μl) into N2 worms. Strain ST300 carrying ncEx3031(myo-3p::Arch::gfp; rol-6d) [22] was crossed with mutants to create the following strains with different mutant backgrounds:
ST302 unc-31(e928) IV;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST303 snt-1(n2665) II;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST304 egl-3(n729) V;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST305 tph-1(n4622) II;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST306 unc-29(e193) I;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST308 unc-49(e382) III;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST309 unc-25(e156) III;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST310 unc-17(e245) IV;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST311 unc-54(e190) I;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST312 eat-4(ad572) III;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST313 cat-1(e1111) X;cnEx3031(myo-3p::Arch::gfp, rol-6d), ST314 unc-7(e5) X;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST315 inx-4(e1128) V;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST319 tdc-1(n3420) II;ncEx3031(myo-3p::Arch::gfp, rol-6d), ST320 flp-13(tm2427) IV;ncEx3031(myo-3p::Arch::gfp, rol-6d).
For chromosomal integration, ST357 L4 larvae were gamma-irradiated with cobalt 60 [50]. The integrated transgene strain ST322 carrying ncIs53(myo-3p::Arch::gfp, rig-3p::mCherry) was outcrossed seven times, and then crossed with unc-7 and unc-31 mutants to generate the following strains:
ST323 unc-31(e928) IV; ncIs53(myo-3p::Arch::gfp, rig-3p::mCherry), ST324 unc-7(e5) X; ncIs53(myo-3p::Arch::gfp, rig-3p::mCherry), ST325 unc-31(e928) IV; unc-7(e5) X; ncIs53(myo-3p::Arch::gfp, rig-3p::mCherry), ST361 unc-31(e169) IV; ncIs53(myo-3p::Arch::gfp, rig-3p::mCherry), ST362 unc-7(e139) X; ncIs53(myo-3p::Arch::gfp, rig-3p::mCherry), ST363 unc-15 (e73) I; ncIs53(myo-3p::Arch::gfp, rig-3p::mCherry).
In the strains carrying the Ex3031(myo-3p:Arch::gfp, rol-6d) transgene, animals strongly expressing Arch::GFP were used for behavioral assays. Animals were moved onto experimental plates seeded with a solution of OP50 containing 500 μM ATR immediately before transfer of the animals, and incubated for at least 8 h before the assay was started.
The behavioral analyses were performed under an upright stereo-microscope (MVX10, Olympus) with an objective lens (MV PLAPO 2xC, Olympus). Movies were acquired at a rate of 30 frames/s with a digital video camera (GZ-HM570B, JVC) mounted on the microscope. For activation of Arch, animals were illuminated with a fluorescent light source (U-HGLGPS, Olympus) through a filter set (U-MRFPHQ/XL, OLYMPUS). Animals carrying the myo-3p:ACR2::gfp transgene were illuminated with a blue light (460–480 nm) through a filter set (U-MWIB2, OLYMPUS) at the intensity of 1.5 mW/mm2. At this intensity, the blue light by itself did not affect the pumping rate, as described previously [14]. The duration of light illumination was controlled with a shutter controller (SSH-C2B, Sigma Koki) using software (SSH-C2B_Demo_Software, Sigma Koki). Light intensity was measured as described previously [22].
Animals that pumped constantly were chosen for experiments. The pharyngeal pumping was counted manually by visual inspection of movies played at slow speed. One movement sequence consisting of an opening and a closing of the TB was counted as one pumping movement.
The basal pumping rate sometimes differed among mutants. In order to compare the responses to green light for different mutant backgrounds, the number of pumps during the 10 s period before green light illumination of the respective mutants was set as a standard for normalization of the pumping rate.
In assays of N2, unc-31, unc-7, and unc-31;unc-7 double mutant animals, we excluded trials in which the pumping rate was less than 25 pumps per 10 s during the 30 s before light stimulation was initiated.
Laser ablations were performed using two-photon microscopy (ZEISS, LSM880; control soft: ZEN2 black edition; two-photon laser generation: Chameleon, COHERENT; objective lens: ZEISS Plan-APOCHROMAT 63× oil immersion lens). Animals were mounted on 4% agar pads and anesthetized with 10 mM levamisole. I1 neurons were identified in transgenic L1/L2 larvae expressing mCherry under the control of the rig-3 promoter. The cell body of I1s was irradiated with an 860 nm laser (Bleaching Mode; Pixel Dwell: 8.24 μs; Iterations: 20; laser power: 100%). Ablation of I1s was confirmed by the absence of fluorescence in the cell body position. Behavioral assays were conducted on adult animals following a recovery period of 1 to 3 days after ablation. After the behavioral assay, the pharyngeal neurons in the animal were observed using confocal microscopy (FV300, Olympus) to ensure cell ablation.
The method used to measure animal body length was described previously [51]. The body length of ST311 unc-54; ncEx3031(myo-3p::Arch::gfp, rol-6d) was measured using the ImageJ public domain software and compared with that at 0.5 s before and after applying light illumination.
The pumping rate in a 10 s period after illumination relative to that before starting illumination was calculated, and the scores were used for statistical analysis. The Steel's many-one rank sum test, Kruskal-Wallis test with post-hoc Steel-Dwass multiple comparison test and Mann-Whitney U Test was performed for data shown in Figs 2, 4, and 5, respectively. The statistical results are compiled in S1, S2 and S3 Tables, respectively.
|
10.1371/journal.ppat.1001275 | Epstein-Barr Virus Nuclear Antigen 3C Facilitates G1-S Transition by Stabilizing and Enhancing the Function of Cyclin D1 | EBNA3C, one of the Epstein-Barr virus (EBV)-encoded latent antigens, is essential for primary B-cell transformation. Cyclin D1, a key regulator of G1 to S phase progression, is tightly associated and aberrantly expressed in numerous human cancers. Previously, EBNA3C was shown to bind to Cyclin D1 in vitro along with Cyclin A and Cyclin E. In the present study, we provide evidence which demonstrates that EBNA3C forms a complex with Cyclin D1 in human cells. Detailed mapping experiments show that a small N-terminal region which lies between amino acids 130–160 of EBNA3C binds to two different sites of Cyclin D1- the N-terminal pRb binding domain (residues 1–50), and C-terminal domain (residues 171–240), known to regulate Cyclin D1 stability. Cyclin D1 is short-lived and ubiquitin-mediated proteasomal degradation has been targeted as a means of therapeutic intervention. Here, we show that EBNA3C stabilizes Cyclin D1 through inhibition of its poly-ubiquitination, and also increases its nuclear localization by blocking GSK3β activity. We further show that EBNA3C enhances the kinase activity of Cyclin D1/CDK6 which enables subsequent ubiquitination and degradation of pRb. EBNA3C together with Cyclin D1-CDK6 complex also efficiently nullifies the inhibitory effect of pRb on cell growth. Moreover, an sh-RNA based strategy for knock-down of both cyclin D1 and EBNA3C genes in EBV transformed lymphoblastoid cell lines (LCLs) shows a significant reduction in cell-growth. Based on these results, we propose that EBNA3C can stabilize as well as enhance the functional activity of Cyclin D1 thereby facilitating the G1-S transition in EBV transformed lymphoblastoid cell lines.
| Epstein-Barr virus (EBV), a ubiquitous human herpesvirus, is linked to the development of multiple cancers, including lymphomas and epithelial carcinomas. EBNA3C, one of its essential latent antigens encoded by EBV, is expressed in EBV-associated lymphomas and contributes to aberrant cell growth after EBV infection. Cyclin D1 over-expression is associated with numerous cancers and is crucial for the transition from G1 to S phase in the mammalian cell-cycle. This study demonstrates that EBNA3C can enhance the functional activity of the Cyclin D1/CDK6 complex which in turn facilitates the G1 to S transition by neutralizing the growth inhibitory effects of pRb. Thus, manipulation of Cyclin D1 functions by EBNA3C provides a favorable environment to promote malignant transformation of EBV infected B-cells.
| Epstein–Barr virus (EBV) is a B-lymphotropic human herpes virus that persists indefinitely in latently infected B-cells. EBV infection occurs early in life for most people and is associated with a broad spectrum of benign and malignant diseases including Burkitt's lymphoma (BL), nasopharyngeal carcinoma (NPC), Hodgkin's disease (HD) and lymphomas associated with immuno-compromised individuals, including AIDS patients and post-transplant patients receiving immune-suppressive therapy [1]. EBV infection in B-cell leads to aberrant cell division and under favorable conditions the infected B-cells will continue to proliferate indefinitely, resulting in development of immortalized lymphoblastoid cell lines (LCLs) [1], [2].
One of the most noteworthy EBV-host cell interactions is the establishment of viral latency. There are three major types of latency, each having its own distinct viral-gene expression pattern [1], [2]. Type I latency is usually noticed in BL tumors with predominant expression of EBV encoded nuclear antigen 1 (EBNA-1) [1], [2]. Type II latency is demonstrated in NPC and HD, where EBNA-1, latent membrane protein 1 (LMP-1), LMP-2A and -2B proteins are significantly detected [1], [2]. Type III latency, also termed as ‘growth program’ [1], [2] is typically seen in LCLs expressing six latent nuclear proteins (EBNA-1, -2, -3A, -3B, -3C, and -LP), three latent membrane proteins (LMP-1, -2A, and -2B), and the viral RNAs which includes the EBERs and BARTs (33, 62).
Molecular genetics analyses have demonstrated that at least six EBV latent genes (EBNA-1, -2, -3A, -3C, -LP, and LMP-1) are essential for in vitro immortalization [1], [2], indicating that a complex cascade of molecular events is required to surpass normal growth controls. One scenario which accounts for EBV-mediated B-cell immortalization is modulation of critical positive and/or negative regulators of cell-cycle progression, such as cyclins, cyclin-dependent kinases (CDKs), cyclin-dependent kinase inhibitor proteins (CDKIs), tumor-suppressors and apoptosis related proteins which includes p53 and pRb [3].
EBNA3C, one of the essential EBV latent antigens, has been shown to function both as a transcriptional activator and a repressor [4], [5], [6]. It has also been shown to interact with numerous transcription modifiers, including c-Myc [4], prothymosin α [7], histone deacetylases [8], CtBP [9], NM23-H1 [10], DP103 [11], SCFSkp2 [12], p300 [13] and p53 [14] which contributes to EBV induced transformation mediated by EBNA3C. In addition, a large body of evidence indicates that EBNA3C can also deregulate the cell-cycle machinery through direct protein-protein interaction and post-translational modification of important cell-cycle regulatory proteins, including Cyclin A [15], [16], pRb [17], p53 [14], Mdm2 [18], and Chk2 [19].
So far, studies probing EBNA3C functions provide perhaps the best link between latent EBV infection and the pRb regulated checkpoint which controls the G1-S phase transition [20], [21]. EBNA3C was previously shown to indirectly target pRb regulated pathways [15], [20]. EBNA3C also activates E2F-dependent promoters and can induce foci formation in colony formation assays [20]. Additionally, EBNA3C overcomes the ability of the CDK inhibitor - p16INK4A to block transformation and noticeably drives serum-starved cells through the G1-S restriction point [20], [21]. More recently, we have shown that EBNA3C directly targets pRb and may indirectly target the pRb regulated checkpoint by associating with Cyclin A as well as Cyclin D1 known to be important in phophorylating pRb [15], [16]. Despite this body of evidence, a clear molecular link between these molecules responsible for disrupting the G1-S phase blockage and EBNA3C is yet to be demonstrated.
Cell-cycle progression is dependent on the activity of cyclins, a family of proteins whose levels oscillate in synchrony with cell-cycle progression, and its functional partner CDKs [22]. Cyclin D (D1, D2 and D3) is expressed in the mid-G1 phase in the mammalian cell-cycle [23]. Among the D-type cyclins, Cyclin D1 is the most ubiquitous and is frequently over-expressed in numerous human malignancies [24], [25]. Cyclin D1 over-expression is often associated with increased gene expression due to gene amplification or post-translational modification [26]. Accumulation of Cyclin D1 in cancer can result in overcoming ubiquitin-mediated degradation through several distinct mechanisms [26].
Cyclin D1, together with its catalytic partners CDK4 or CDK6, promotes G1-S-phase transition via phosphorylation of pRb and disrupting the pRb-E2F1 repressor complex [23]. These functions of Cyclin D1 ensure efficient initiation of S phase [26], [27]. During late G1 and S phases, Cyclin D1 is phosphorylated on Thr-286 by GSK3β, which triggers nuclear export and proteasomal degradation through E3 ubiquitin ligase, SCFFBX4-αB crystallin [26]. Thus, subversion of either of these functions may result in unrestrained cell proliferation and oncogenesis.
The cyclin D1 gene is located on chromosome 11q13, close to the bcl-1 locus, and is considered to be a proto-oncogene with evidence indicating that its derangement contributes to the development of tumors [28]. Mantle cell lymphomas have been reported to over-express Cyclin D1 due to a characteristic genetic translocation [28]. In addition, patients with tumors over-expressing Cyclin D1 have been shown to have a particularly poor prognosis [25], [29]; however, over-expression of Cyclin D1 has been demonstrated for a vast series of human malignancies including breast cancers, esophageal cancers and pancreatic cancers [25], [30]. Over-expression of Cyclin D1, regardless of its gene alteration, caused abnormal cell proliferation, resulting in oncogenesis [22], [23], [31]. cyclin D2, considered also as a proto-oncogene, is located on chromosome 12p13, and unlike Cyclin D1, Cyclin D2 has been reported to be expressed normally in B-lymphocytes [32]. Interestingly, it has been observed that ectopic over-expression of Cyclin D2 efficiently blocks cell-cycle progression [33], suggesting an alternate role for Cyclin D2 in promoting exit from the cell-cycle and maintaining cells in a non-proliferative state. These observations suggest that D-type cyclins may have different roles depending on their levels of expression and cell type, which may also be independent of CDK activity. Reports have shown that immortalization of primary B-lymphocytes by EBV is accompanied by transcriptional activation of cyclin D2 gene but not cyclin D1 [32], [34]. However, Cyclin D1 protein has been shown to be significantly expressed in a number of EBV positive LCLs [35], [36] or EBV positive SCID mice lymphomas [37]. Surprisingly, these studies did not directly set out to explore the contribution of Cyclin D1 in EBV-mediated B-cell oncogenesis.
A previous study from our lab showed an in vitro interaction between the EBV encoded antigen EBNA3C and Cyclin D1 [16]. The experiments described in this current study explore the consequences of this interaction in terms of EBV mediated transformation of primary B-cells as well as growth maintenance of LCLs. We now show that EBNA3C stabilizes as well as enhances the kinase activity of the Cyclin D1/CDK6 complex, and the nuclear localization of Cyclin D1 to bypass the G1 restriction point. Importantly, this study provides the first evidence to show that the essential EBV latent antigen EBNA3C targets Cyclin D1, which is different from previous reports, and describes a potential fundamental mechanism by which EBV deregulates the mammalian cell-cycle in EBV-associated human cancers by facilitating the G1-S transition.
Myc, flag, GFP and GST tagged EBNA3C vectors have been described previously [14], [18]. pcDNA3-HA-Ub was kindly provided by George Mosialos (Aristotle University of Thessaloniki, Thessaloniki, Greece). Vectors pcDNA3-Cyclin D1, pcDNA3-1x flag-Cyclin D2 and pcDNA3-1x flag-Cyclin D3 were provided by Alan Diehl (University of Pennsylvania School of Medicine, Philadelphia) and used to generate pA3F-Cyclin D by cloning PCR amplified DNA into pA3F vector [4]. GST Cyclin D1 vectors were cloned by inserting PCR amplified DNA into pGEX-2TK vector (GE Healthcare Biosciences, Pittsburgh, PA). pGEX-Cyclin D1 (286A) was generated by PCR using pA3F-Cyclin D1 as template. Sh-RNA vector, pGIPZ (Open Biosystems, Inc. Huntsville, AL) and lentiviral packaging vectors were described [38]. CDK6 cDNA cloned into pA3F vector was derived from HEK 293 cell RNA that was purified with TRIzol reagent and reverse transcribed with Superscript II (Invitrogen, Inc., Carlsbad, CA). Mouse antibodies to Cyclin D1 (DSC-6) and Sp1 (1C6), and rabbit antibody to Ub (FL-76) were from Santa Cruz Biotechnology, Inc (Santa Cruz, CA). Rabbit antibodies to Cyclin D2 and D3 were kindly provided by Alan Diehl (University of Pennsylvania School of Medicine, Philadelphia). Mouse antibodies to flag-epitope (M2) was from Sigma-Aldrich Corp. (St. Louis, MO) and to GAPDH was from US-Biological Corp. (Swampscott, MA). Antibodies to HA-epitope (12CA5) or Myc-epitope (9E10) were prepared from cell culture supernatants as described [14], [18]. Mouse (A10) or rabbit antibody to EBNA3C were described [14], [18].
HEK 293, 293T and Saos-2 (p53-/- pRb-/-) cells were obtained from Jon Aster (Brigham and Women's Hospital, Boston, MA, USA). Saos-2 and U2OS are human osteosarcoma cell line [39]. HEK 293, HEK 293T, U2OS, and Saos-2 cells were grown in Dulbecco's modified Eagle's medium (DMEM; HyClone, Logan, UT) supplemented with 10% fetal bovine serum (FBS; HyClone, Logan, UT), 50 U/ml penicillin (HyClone, Logan, UT), 50 µg/ml streptomycin (HyClone, Logan, UT) and 2 mM L-glutamine (HyClone, Logan, UT). BL lines BJAB, Ramos, BL41 and B95.8 infected BL41 (BL41/B95.8) were kindly provided by Elliott Kieff (Harvard Medical School, Boston, MA). MutuI, MutuIII were provided by Yan Yuan (School of Dental Medicine, University of Pennsylvania, Philadelphia, PA). These BL lines and LCL1 and LCL2 were maintained in RPMI 1640 (HyClone, Logan, UT) supplemented as described above. EBNA3C expressing BJAB lines were described [14], [18]. Unless otherwise stated all cultures were incubated at 37°C in a humidified environment supplemented with 5% CO2.
Adherent cells were transfected by electroporation with a Bio-Rad Gene Pulser II electroporator as described [14], [18].
Peripheral blood mononuclear cells (PBMC) from healthy donors were obtained from University of Pennsylvania Immunology Core. As described [40], approximately 10 million PBMC were mixed with virus supernatant in 1 ml of RPMI 1640 with 10% FBS for 4 hr at 37°C. Cells were centrifuged for 5 min at500 g, discarded the supernatant, pelleted cells and resuspended in 2 ml of complete RPMI 1640 medium in 6 well plates. EBV GFP expression visualized by fluorescence microscopy was used to quantify infection. The protein and mRNA level of the infected cells was detected after 3 days of post-infection.
Transfected cells were harvested, washed with ice cold PBS and lysed in 0.5 ml ice cold RIPA buffer [1% Nonidet P-40 (NP-40), 10 mM Tris pH 8.0, 2 mM EDTA, 150 mM NaCl, supplemented with protease inhibitors (1 mM phenylmethylsulphonyl fluoride (PMSF), 1 µg/ml each aprotinin, pepstatin and leupeptin]. Lysates were precleared with normal mouse serum plus 30 µL of Protein A/G Sepharose (1 h, 4°C). 5% of the precleared lysate was saved for input control and the protein of interest was captured by rotating the remaining lysate with 1 µg of specific antibody overnight at 4°C. Immuno-complexes were captured with 30 µl of a 1∶1 mixture of Protein-A and Protein-G Sepharose beads, pelleted and washed 5X with ice cold RIPA buffer.
For western blots, input lysates and IP complexes were boiled in laemmli buffer [41], fractionated by SDS-PAGE and transferred to a 0.45 µm nitrocellulose membrane. The membranes were then probed with specific antibodies followed by incubation with appropriate infrared-tagged secondary antibodies and viewed on an Odyssey imager. Image analysis and quantification measurements were performed using the Odyssey Infrared Imaging System application software (LiCor Inc., Lincoln, NE).
Escherichia coli BL21 cells were transformed with plasmids for each glutathione S-transferase (GST) fusion protein and protein complexes containing the tagged proteins were purified essentially as described before [14], [18].
For in vitro binding experiments, GST fusion proteins were incubated with cell lystaes or 35S-labeled in vitro-translated protein in binding buffer (1x phosphate-buffered saline [PBS], 0.1% NP-40, 0.5 mM dithiothreitol [DTT], 10% glycerol, supplemented with protease inhibitors). In vitro translation was done with the TNT T7 Quick Coupled Transcription/Translation System (Promega Inc., Madison, WI) according to the manufacturer's instructions.
Cells were immuno-stained as described [18] with few modifications. Briefly, U2OS cells plated on coverslips were transfected with expression vectors as indicated, using Lipofectamine 2000 (Invitrogen, Carlsbad, CA) according to manufacturer's protocol. After 36 h of transfection cells were fixed. B-cells were air-dried and subsequently fixed. Transiently expressed flag-tagged Cyclin D1 was detected using M2-antibody, and GFP-EBNA3C was detected by GFP fluorescence. In B-cells, endogenously expressed Cyclin D1 and EBNA3C proteins were detected using specific antibody. The slides were examined with a Fluoview FV300 confocal microscope (Olympus Inc., Melville, NY).
Total RNA was isolated by using TRIzol reagent according to the instructions of the manufacturer (Invitrogen, Inc., Carlsbad, CA). cDNA was made by using a Superscript II reverse transcriptase kit (Invitrogen, Inc., Carlsbad, CA) according to the instructions of the manufacturer. The primers were for cyclin D1, 5′-TGCCCTCTGTGCCACAGATG-3′, and 5′-TCTGGAGAGGAAGCGTGTGA-3′, for cyclin D2 5′-TGCTCTGTGTGCCACCGACTT-3′, and 5′-CAGCTCAGTCAGGGCATCACAA-3′, for cyclin D3 5′-TTTGCCATGTACCCGCCATCCA-3′ and 5′-CCCGCAGGCAGTCCACTTCA-3′, and for GAPDH 5′-TGCACCACCAACTGCTTAG-3′ and 5′-GATGCAGGGATGATGTTC-3′. Quantitative real-time PCR analysis was done as described [18] in triplicate.
15×106 HEK 293T cells were transfected by electroporation with DNA vectors expressing a specific protein. Cells were incubated for 36 h and pretreated for an additional 6 h with 20 µM MG132 (Enzo Life Sciences International, Inc., Plymouth Meeting, PA) before harvesting. Proteins were immunoprecipitated with specific antibodies and resolved by SDS-PAGE. The extent of ubiquitination of immunoprecipitated complexes were detected by HA-specific antibody (12CA5) against HA-Ub tagged proteins.
15×106 HEK 293 cells were transfected with expression plasmids. After 36 h cells were PBS washed and resuspended into hypotonic buffer [5 mM Pipes (KOH) pH 8.0, 85 mM KCl, 0.5% NP-40 supplemented with protease inhibitors). After 10-min incubation on ice, cells were homogenized with 20 strokes in a Dounce homogenizer, nuclei were pelleted (2300 g for 5 min) and the cytosolic material was collected. Nuclear pellets were PBS washed, resuspended in nuclear lysis buffer (50 mM Tris, pH 8.0, 2 mM EDTA, 150 mM NaCl, 1% NP-40, and protease inhibitors), lysed by vortexing periodically for 1 h. Soluble nuclear fraction was separated by centrifugation at 21000 g for 10 min. Total protein was measured by Bradford protein assay and 50 µg of total protein was resolved by SDS-PAGE. The efficiency of nuclear and cytoplasmic fractionation was confirmed by western blot against nuclear transcription factor Sp1 and cytoplasmic protein GAPDH.
15×106 HEK 293T cells were transfected with plasmids expressing flag-Cyclin D1 (5 µg), flag-CDK6 (5 µg) and increasing amount of myc-EBNA3C (0, 5, 10, 20 µg). For GSK-3β kinase assay cells were transfected with DNA vectors that express myc-tagged GSK-3β (10 µg) and flag-tagged EBNA3C (20 µg). Cells were harvested and protein complexes were immunoprecipitated (IP) using either M2 (for cyclin D1) or 9E10 ascites fluid (for GSK-3β). IP complexes were then washed with buffer A (25 mM Tris [pH 7.5], 70 mM NaCl, 10 mM MgCl2, 1 mM EGTA, 1 mM DTT, plus protease and phosphatase inhibitors) and incubated in 30 µl of kinase buffer B (buffer A plus 10 mM cold ATP, and 0.2 µCi of [γ-32P]-ATP/µl) supplemented with either 4 µg of histone H1 (Upstate Biotechnology, Inc., Lake Placid, N.Y.) or bacterially purified GST-pRb (residues 792-928) for 30 min at 30°C. The reaction was stopped by adding 2X laemmli buffer [41] and heating to 95°C for 10 min. Labeled proteins were resolved by 12% SDS-PAGE. Band quantitation was performed using the ImageQuant software (GE Healthcare Biosciences, Pittsburgh, PA).
Cells were transiently transfected using electroporation with plasmids as indicated in the text. After 36 hours transfection, cells were treated with 40 µg/ml cyclohexamide (CalBiochem, Gibbstown, NJ) and lysates were subjected to immunoblot analyses. Band intensities were quantitated using Odyssey 3.0 software provided by Odyssey imager (LiCor Inc., Lincoln, NE).
Short-hairpin oligonucleotides directed against EBNA3C were designed (Dharmacon Research, Chicago, IL). The sense strand of the EBNA3C-shRNA sequence is 5′-tcgagtgctgttgacagtgagcgaCCATATACCGCAAGGAATAtagtgaagccacagatgtaTATTCCTTGCGGTATATGGgtgcctactgcctcggaa-3′. The sense strand of cyclin D1 sh-RNA sequence is 5′-tcgagtgctgttgacagtgagcgaCAAACAGATCATCCGCAAAtagtgaagccacagatgtaTTTGCGGATGATCTGTTTgtgcctactgcctcggaa-3′ [42]. Upper-case letters indicate 19-nucleotide (nt) either EBNA3C or cyclin D1 target sequences respectively and lowercase letters indicate hairpin and sequences necessary for the directional cloning into pGIPZ (Open Biosystems, Inc. Huntsville, AL). Single-stranded EBNA3C and cyclin D1 oligonucleotides were first annealed and then cloned into the Xho I and Mlu I restriction sites of pGIPZ vector. The fidelity of cloned double-strand DNA was confirmed by DNA sequencing. In parallel, a commonly available control shRNA sequence (Dharmacon Research, Chicago, IL): (5′-TCTCGCTTGGGCGAGAGTAAG-3′) that lacks complementary sequences in the human genome was also cloned into pGIPZ vector.
Lentivirus production and transduction of EBV-transformed B-cells (LCLs) were essentially carried out as previously described [38].
Saos-2 (p53-/- pRb-/-) were transfected using Ca3(PO4)2 method as described [38]. After 24 h transfection, cells were selected using DMEM supplemented with 1000 µg/ml G418; Invitrogen). After a 2-week selection, 5×106 cells were harvested, lysed in RIPA buffer and subjected for immunoblot analyses. Approximately 0.1×106 cells from each set of samples were plated into each well of the 6-well plates and cultured for 6 days. Viable cells from each well were counted by trypan blue exclusion method daily using a Bio-Rad TC10 Automated cell counter.
For LCLs, approximately 1×106 cells were plated into each well of the 6-well plates and cultured at 37°C in complete RPMI medium. Cells were counted similarly for 20 days. Both experiments were performed in duplicate and were repeated two times.
5×106 Saos-2 (pRb-/-) cells were transfected as described [38] and cultured in DMEM supplemented with 1 mg/ml G418 (Invitrogen, Inc., Carlsbad, CA). After a 2-week selection, cells were fixed on the plates with 4% formaldehyde and stained with 0.1% crystal violet (Sigma-Aldrich Corp., St. Louis, MO). The area of the colonies (pixels) in each dish was calculated by Image J software (Adobe Inc., San Jose, CA). The data are shown as the average of three independent experiments.
For serum starvation experiments, the culture medium was replaced with RPMI 1640 and 0.1% FBS for 12 h. Cells were PBS washed, fixed in cold 70% ethanol for 30 min at 20°C, PBS washed and stained 2 h in buffer containing 50 mg/ml propidium iodide, 10 mM Tris pH 7.5, and 500 U/ml RNAseA in dark. PBS washed cells were analyzed for cell-cycle profile by FACS Calibur system and Cellquest software (Becton-Dickinson Inc., San Jose, CA).
In order to determine whether EBV infection alters Cyclin D expression, approximately 10×106 human resting peripheral blood mononuclear cells (PBMC) were infected by BAC GFP-EBV as previously described [40] for 4 h and western blot analysis was performed on samples collected 3 days after infection. The results showed that EBV infection leads to a significant induction of all three Cyclin D protein levels 3 days post-infection, with no preference for any particular D-type cyclins (Fig. 1A). Similarly, western blot results of Burkitt's lymphoma (BL) cell line BL41 and BL41 infected with wild-type EBV strain B95.8 (BL41/B95.8) also showed elevated levels of Cyclin Ds with Cyclin D1 expression more dramatically changed compared to other Cyclin Ds (Fig. 1B). Since Cyclin D1 expression was induced significantly after EBV infection in both PBMC and BL cell line, we next wanted to determine if the induction was related to a specific EBV latent protein expressed during type III latency. The results showed that the levels of both Cyclin D1 and Cyclin D2 proteins were induced in type III latency BL cell line MutuIII compared to latency I expressing MutuI BL cell line (Fig. 1C). These results differ with previously published observations which suggested that B-cells infected with EBV do not express Cyclin D1 [43], [44], [45]. However, in agreement with previously published results [32], our real-time PCR data showed that EBV infection led to a significant increase of cyclin D2 mRNA level in LCLs (LCL1 and LCL2) when compared to EBV negative BL cells (BJAB and Ramos) whereas, there was little or no detectable change for cyclin D1 mRNA (Fig. 1F). Real-time PCR data obtained from two other matched sets of cell lines BL41 – BL41/B95.8 and MutuI – MutuIII also showed similar results as above (Fig. 1G and 1H, respectively). These results suggest that D-type cyclins are regulated through distinctly different mechanisms in EBV infected B-cells. EBV effects on Cyclin D2 are at the level of its transcript stability whereas the effects on Cyclin D1 or D3 seem to be post-translational.
To elucidate the effects of the EBV encoded essential nuclear antigen, EBNA3C on Cyclin D1, BL lines BJAB and E3C #7, a BJAB stably expressing EBNA3C were analyzed. The western blot results showed a significant increase in Cyclin D1 protein expression among D-type Cyclins in E3C #7 cells compared to the BJAB control cells and smaller changes in Cyclin D2 and D3 (Fig. 1D). The effect of EBNA3C on Cyclin D1 steady-state levels was not due to changes in the transcription as EBNA3C expression did not alter the level of cyclin D1 mRNAs in these cells as seen above (Fig. 1I). To further verify the role of EBNA3C on Cyclin D1 protein accumulation, we determined the levels of Cyclin Ds in a lymphoblastoid cell line with the EBNA3C mRNA specifically targeted by short-hairpin RNA (Sh-E3C). The western blot data showed that the expression level of Cyclin D1 in the LCLs stably knocked-down for EBNA3C (Sh-E3C) was significantly diminished as compared to the control cell line (Sh-Control) (Fig. 1E), however the expression levels of other Cyclin Ds was not altered (Fig. 1E). These results indicate that EBNA3C can contribute to Cyclin D1 accumulation in latently infected EBV positive cells.
To demonstrate that EBNA3C can stabilize Cyclin D1 protein levels, HEK 293 cells were transfected with an increasing amount of an expression construct expressing EBNA3C and tested for endogenous Cyclin D1 protein level. The results showed that EBNA3C stabilizes Cyclin D1 protein expression in a dose dependent manner (Fig. 1J).
We earlier determined that EBNA3C plays a critical role in modulating the ubiquitin (Ub)-proteasome machinery [12], [17], [18]. Therefore, to investigate whether the increase of Cyclin D1 levels was because of the inhibition of Ub-proteasome mediated destabilization by EBNA3C, transiently co-transfected cells were treated with the proteasome inhibitor, MG132. The results showed that both the treatment with MG132, and presence of EBNA3C led to a significant accumulation (six fold) of Cyclin D1 when compared to mock treatment or vector control (Fig. 1K). Therefore the increased levels of Cyclin D1 observed in the presence of EBNA3C and MG132 is a result of stabilization of Cyclin D1 likely by EBNA3C inhibition of the Ub-proteasome degradation system. Importantly, both CDK6 and EBNA3C levels were not altered by MG132 (Fig. 1K).
To directly determine EBNA3C stabilization of Cyclin D1, HEK 293 cells were transfected with flag-Cyclin D1, flag-CDK6, and EBNA3C expression vectors. Thirty-six hours later, cells were treated with protein synthesis inhibitor cycloheximide, and samples were collected at 0, 1, and 2 hours. Western blots probed with flag antibody showed that the stability of Cyclin D1 protein was significantly enhanced by EBNA3C co-expression, whereas in the absence of EBNA3C, Cyclin D1 was degraded to near completion by 2-h after addition of CHX (Fig. 1L, grey bar). Cyclin D1 half life was determined to be 2 h in EBNA3C expressing cells; however, it shortened noticeably to less than 1 h when Cyclin D1 was expressed alone (Fig. 1L, bar diagram). The results also indicated that both EBNA3C and CDK6 were notably stable throughout the experimental period of time and had no sign of protein degradation (Fig. 1L, CDK6 indicated as black bar). Overall, the results of these experiments suggest EBNA3C can stabilize Cyclin D1 by regulating its targeted degradation likely through the Ub-proteasome degradation system.
Recently we have shown that ectopic expression of EBNA3C leads to stabilization of an important cellular oncoprotein, Mdm2 by inhibiting its poly-ubiquitination [18]. The increased stability of Cyclin D1 in the presence of EBNA3C, prompted us to examine whether EBNA3C similarly inhibits poly-ubiquitination of Cyclin D1 and so enhances its stability. To explore this possibility, three cell lines were selected, the EBV negative cell line BJAB, BJAB stably expressing EBNA3C (E3C #7) and an EBV positive lymphoblastoid cell line (LCL2). Immnuprecipitation using specific antibody against Cyclin D1 resulted in formation of high molecular weight species of Cyclin D1 migrating at a slower rate in BJAB cells while in BJAB cells stably expressing EBNA3C or in LCL2 significantly less of these high molecular weight bands were observed (Fig. 2A). Re-probing of the same membrane with Ub specific antibody showed a similar pattern (Fig. 2A). This result indicates that the activity responsible for the change in Cyclin D1 bands is present in EBV positive cells (LCL2) and EBNA3C expressing cell line (E3C #7) when compared to the EBV negative BJAB cells.
To directly address this phenomenon, an ubiquitination experiment was set up, where HEK 293T cells were transiently co-transfected with expression constructs for HA-Ub, flag-Cyclin D1 and myc-EBNA3C and the ubiquitination of the Cyclin D1 was assessed by immunoprecipitation followed by Western blotting (Fig. 2B). The result demonstrated a significant and reproducible reduction in Cyclin D1 poly-ubiquitination level in EBNA3C expressing cells (Fig. 2B). Similar experiments were performed separately using two different cyclins, Cyclin A and Cyclin E to determine if this effect was specific for Cyclin D1. However, neither Cyclin A nor Cyclin E poly-ubiquitination levels were reduced in the presence of EBNA3C (Fig. 2C and 2D). To determine whether the poly-ubiquitination level of the other D-type cyclins was also affected in the presence of EBNA3C, we tested flag-tagged Cyclin D2 and D3 for ubiquitination in the absence and presence of EBNA3C. Importantly, poly-ubiquitination of both Cyclin D2 and D3 was efficiently inhibited in the presence of EBNA3C (Fig. 2E). This result indicates that EBNA3C can profoundly affect the poly-ubiquitination of all Cyclin Ds and thus enhance their stability.
We have shown earlier that EBNA3C interacts with Cyclin D1 in vitro along with other cyclins including Cyclin A and Cyclin E [16]. In order to determine whether EBNA3C forms a complex with Cyclin D1 in cells to enhance its stability, we performed binding assays using co-IP experiments. HEK 293T cells were co-transfected with expression constructs for myc-EBNA3C and flag-Cyclin D1. The results showed that ectopically expressed EBNA3C associated with Cyclin D1 in cells (Fig. 3A and 3B). To further determine whether this binding occurred under endogenous settings, Cyclin D1 was immunoprecipitated from EBV negative cell line, BJAB and two EBV transformed lymphoblastoid cell lines, LCL1 and LCL2 expressing EBNA3C. EBNA3C was detected by Western blot analysis using A10, an EBNA3C specific monoclonal antibody and showed efficient co-immunoprecipitation (Fig. 3C). In a separate experimental setting, Cyclin D1 was immunoprecipitated from BJAB cells and BJAB cells stably expressing EBNA3C (E3C#10). Similarly co-IP of EBNA3C was demonstrated using the A10 antibody (Fig. 3D). To further corroborate the association in human cells, a GST-pulldown experiment was conducted; where bacterially expressed GST-Cyclin D1 was incubated with cell lysates prepared from either BJAB cells or BJAB cells stably expressing EBNA3C (E3C#7 and E3C#10). EBNA3C was seen to strongly associate with GST-Cyclin D1 but not with the GST control (Fig. 3E). Coomassie staining of a parallel gel showed the amount of GST and GST-Cyclin D1 proteins used in the binding assay (Fig. 3E, right panel). Analysis of the data from the ectopic expression system as well as cell lines endogenously expressing Cyclin D1 and EBNA3C at physiological levels strongly demonstrated an association between Cyclin D1 and EBNA3C in human cells.
We have previously shown that a small N-terminal region of EBNA3C (residues 130-160) binds to Cyclin D1 in vitro [16]. To map the domain of EBNA3C that interacts with Cyclin D1, HEK 293T cells were transfected with expression constructs for flag-Cyclin D1 and either full-length EBNA3C (residues 1-992), EBNA3C residues 1-365, EBNA3C residues 366–620, or EBNA3C residues 621-992. All EBNA3C expression constructs were fused in frame with a myc epitope tag at the C-terminus of the protein. As expected, the results showed that Cyclin D1 co-immunoprecipitated with full-length EBNA3C as well as with the N-terminal domain of EBNA3C (residues 1–365) (Fig 4A, left-middle panel, lanes 2 and 3, respectively) whereas no co-IP was detected with vector control or other truncated versions of EBNA3C (Fig 4A, left-middle panel, lanes 1, 4 and 5). To further corroborate the binding data, an in vitro GST-pulldown experiment was performed using in vitro translated 35S-radiolabeled fragments of EBNA3C (residues 1–100, 1–129, 1–159 and 1–200) within the N-terminal domain. In vitro precipitation experiments with bacterially expressed GST-Cyclin D1 showed strong association with residues 1–159 and 1-200 of EBNA3C (Fig. 4B, bottom panel, lanes 3 and 4, respectively), but not with EBNA3C residues 1–100 or 1–129 (Fig. 4B, bottom panel, lanes 1 and 2, respectively). All fragments of EBNA3C failed to interact with the GST control, indicating that the observed binding was specific for Cyclin D1 (Fig. 4B, middle panel, lanes 1 to 4).
In an attempt to gain insights into the functionality of the association between Cyclin D1 and EBNA3C, a series of N- and C-terminal deletion mutants of Cyclin D1 (residues 1–50, 40–170, 171–260 and 241–295) were designed according to their domain distribution [46], [47] and tested for their ability to bind EBNA3C using in vitro binding experiments. The results of the GST-pulldown assay clearly showed that full-length Cyclin D1, the N-terminal pRb binding region (residues 1-50) and the C-terminal domain which is known to regulate Cyclin D1 stability (residues 171–260) strongly associated with EBNA3C (Fig. 4C, top panel, lanes 3, 4 and 6, respectively). However, no binding was detected with the other truncated versions of Cyclin D1 (the CDK4/6 binding domain, residues 40–170 and the PEST domain, residues 241–295) or with the GST control (Fig. 4C, top panel, lanes 2, 5, and 7). Importantly, the C-terminal domain of Cyclin D1 (residues 171–260) bound to EBNA3C with relatively higher affinity than the full-length or the N-terminal site (Fig 4C).
In order to determine the specificity of EBNA3C and Cyclin D1 interaction, we next performed a co-immunoprecipitation assay using all three flag-tagged D-type Cyclins. Interestingly, the results showed that EBNA3C forms complexes with all three D-type Cyclins in cells, suggesting that EBNA3C has specificity for interaction with Cyclin D1, D2 and D3 (Fig. 4D).
Increased expression of Cyclin D1 has been seen in a number of cancers [25], [30]; however, this enhanced expression is usually not sufficient to drive the oncogenic process. Emerging evidence suggests that nuclear accumulation of Cyclin D1 resulting from altered nuclear trafficking and proteolysis is critical for its oncogenic phenotype [31]. In order to determine the effect of EBNA3C on the sub-cellular localization of Cyclin D1, asynchronously growing U2OS cells were transfected with expression vectors encoding flag-tagged Cyclin D1 and GFP-tagged EBNA3C. Localization of Cyclin D1 was determined by indirect immunofluorescence using a monoclonal antibody against the flag epitope (Fig. 5A, panels f, h, j, l). While Cyclin D1 mostly localized to the cytoplasm in the absence of EBNA3C (Fig. 5A, panels f, h), it was predominantly localized to the nucleus in the presence of EBNA3C (Fig. 5A, panels j, l). To quantitatively compare the Cyclin D1 signals in the nuclear and cytoplasmic compartments, 10 different fields of the stained slides were examined and the bar diagram represents the mean of three independent experiments which showed that nuclear localization was increased by 20% (Fig. 5A, bar diagram). To further corroborate these results showing that EBNA3C promotes nuclear localization of Cyclin D1, the sub-cellular localization of endogenous Cyclin D1 was determined in three different cell lines – EBV negative BL cell line BJAB, BJAB cells stably expressing EBNA3C (E3C# 7) and an EBV transformed B-cell line LCL2, using a specific antibody against cyclin D1. As anticipated, the results showed that cyclin D1 was predominantly localized in the nucleus of both EBNA3C positive BJAB cells (Fig. 5B, panels f, g) and EBV positive cells LCL2 (Fig. 5B, panels j, k), but was almost exclusively cytoplasmic in the EBV negative BJAB cells with no EBNA3C expressed (Fig. 5B, panels b, c).
Based on immuno-fluorescence studies, we observed that Cyclin D1 localization was mainly restricted to the cytoplasmic fraction of asynchronously growing cells. However, on expression of EBNA3C the localization of Cyclin D1 was predominantly nuclear. To further support these data, transiently transfected HEK 293 cells were subjected to sub-cellular fractionation and fractionated proteins were analyzed by immunoblot analysis. The result showed that flag-tagged Cyclin D1 alone was detected approximately 50% in both cytoplasmic and nuclear fractions, respectively (Fig. 6A, compare lanes 1 and 4). However, when co-transfected with EBNA3C, flag-Cyclin D1 was detected predominantly within the nuclear fraction (Fig. 6A, compare lanes 3 and 6), with an approximately 50% increase compared to flag-Cyclin D1 alone (Fig. 6A, compare lanes 1 and 3). EBNA3C was detected completely within nuclear fraction (Fig. 6A, lanes 2 and 3). The efficiency of cytoplasmic and nuclear fractionation was confirmed by localization of nuclear transcription factor Sp1 and cytoplasmic protein GAPDH (Fig. 6A). These observations strongly suggested that the apparent nuclear trans-localization of Cyclin D1 mediated by EBNA3C, as determined by indirect immuno-fluorescence microscopy or sub-cellular fractionation assay may be due to deregulation of the critical regulatory kinase GSK-3β, a negative regulator of Cyclin D1 nuclear retention and protein stability [31]. We thus decided to examine whether EBNA3C can nullify the effect of GSK-3β on Cyclin D1 function.
GSK-3β can direct the nuclear export of Cyclin D1 via a CRM1-dependent pathway [31]. To examine whether EBNA3C can block Cyclin D1 nuclear export, we tested the ability of EBNA3C to override GSK-3β triggered Cyclin D1 nuclear export. To test this possibility, HEK 293 cells were transfected with expression vectors encoding flag-tagged Cyclin D1, with or without GSK-3β and myc-tagged EBNA3C. Fractionated cell lysates were analyzed by western blot to clarify flag-tagged Cyclin D1 localization. As expected, Cyclin D1 was primarily present in the cytoplasmic fraction both in the absence and presence of GSK-3β (Fig. 6B, lanes 1 and 4). In contrast, Cyclin D1 was largely detected within the nuclear fraction when co-expressed with EBNA3C (Fig. 6B, lane 3). Interestingly, even in the presence of GSK-3β nuclear fractionation of Cyclin D1 was greatly increased when co-expressed with EBNA3C compared with the vector control (Fig. 6B, compare lanes 1 and 3).
GSK-3β has been shown to phosphorylate Cyclin D1 on Thr-286 in vitro [31], and is postulated to be a major regulator of protein levels and intracellular distribution of Cyclin D1 [31]. To establish a plausible explanation for the inhibitory effects of EBNA3C on GSK-3β dependent Cyclin D1 subcellular localization, we first asked whether EBNA3C can form a complex with GSK-3β to negatively modulate its activity and to also determine whether the kinase activity of GSK-3β is inhibited in the presence of EBNA3C. To this end, we co-expressed myc-tagged GSK-3β and flag-tagged EBNA3C and assessed their interaction through co-immunoprecipitation experiment. The results showed that indeed EBNA3C can form a complex with GSK-3β (Fig. 6C, compare lanes 3 and 4). Next, an in vitro kinase assay was conducted where GSK-3β was immuno-precipitated in the absence and presence of EBNA3C, and tested for its ability to phosphorylate recombinant GST-Cyclin D1 proteins (wild-type and T286A mutant Cyclin D1). The results showed that the phosphorylation level of wild-type GST-Cyclin D1 by GSK-3β was reduced by more than 4 fold in the presence of EBNA3C (Fig. 6D, compare lanes 1 and 2). As expected, no phosphorylation bands were observed in case of mutant GST-Cyclin D1 (T286A) indicating the specificity of this experiment (Fig. 6D, lanes 3 and 4). Parallel blots showed the protein expression levels in whole cell-lysate (Fig. 6D), and the amount of purified GST-Cyclin D1 used in this experiment (Fig. 6D). These results indicated that EBNA3C may regulate Cyclin D1 sub-cellular localization probably by blocking the function of GSK-3β.
To address the functional consequences as a result of the association of Cyclin D1 and EBNA3C, we tested the activity of Cyclin D1/CDK6 complexes for the ability to phosphorylate histone H1 or recombinant GST-pRb (residues 792-928). HEK 293T cells were transiently transfected with increasing amounts of a myc-tagged EBNA3C expression construct. Flag-tagged Cyclin D1/CDK6 immunoprecipitated complexes were assayed for in vitro kinase activity as determined by histone H1 or GST-pRb phosphorylation (Fig. 7A and B, respectively). The results showed that Cyclin D1-dependent kinase activity increased in a dose-responsive manner with increased expression of EBNA3C (Fig. 7A and B). Phosphorimager analysis revealed 1.6-times more P32-Histone H1 and 2.3-times more P32-GST-pRb (Fig. 7A and B). Parallel blots showed the expressed protein levels (Fig. 7A and B, top two panels) and the amount of substrates (histone H1 or GST-pRb) used in this study (Fig. 7A and B).
Cyclin D1/CDK4/6 complexes are rate-limiting for G1 progression by contributing to the sequential phosphorylation of pRb, and thereby canceling the growth-suppressive function of pRb, thus facilitating entry into S-phase [26], [27]. Previously we have shown that EBNA3C facilitates pRb degradation by enhancing its poly-ubiquitination through recruitment of the SCFSkp2 E3 ligase activity [17]. To test whether EBNA3C coupled with Cyclin D1/CDK6 complex regulates pRb stabilization, a stability assay was performed using cycloheximide (CHX) treated Saos-2 (pRb-/- p53-/-) cells co-transfected with plasmids expressing myc-tagged pRb, flag-tagged Cyclin D1, flag-tagged CDK6, and EBNA3C (Fig. 7C). The results clearly showed that independent expression of either Cyclin D1/CDK6 complex or EBNA3C reduced pRb expression levels (Fig. 7C [upper panel], compare lanes 1-9). Surprisingly, when both EBNA3C and Cyclin D1/CDK6 complex were expressed together, little or no pRb was detected (Fig. 7C [lower panel], lanes 1-3), indicating that EBNA3C can also facilitate pRb degradation in cooperation with Cyclin D1/CDK6 either through stabilization of Cyclin D1 (Fig. 7C [lower panel], compare lanes 4–9) or increasing kinase activity of Cyclin D1/CDK6 complex.
In order to probe whether EBNA3C enhances pRb poly-ubiquitination in a Cyclin D1-dependent manner for degradation, we performed an in vivo ubiquitination assay. HEK 293T cells were co-transfected with expression constructs for myc-tagged pRb, HA-tagged Ub, flag-tagged Cyclin D1 and CDK6 and untagged EBNA3C as indicated (Fig. 7D). pRb was immunoprecipitated with myc antibody, and ubiquitinated-pRb was detected by probing blots with HA antibody. In agreement with the previous result, poly-ubiquitination of pRb was significantly enhanced in the presence of EBNA3C alone (Fig. 7D, compare lanes 3 and 4) and slightly further elevated in the presence of both EBNA3C and Cyclin D1/CDK6 complex (Fig. 7D, compare lanes 4 and 6) indicating that EBNA3C together with Cyclin D1/CDK6 is important for inducing pRb poly-ubiquitination and its subsequent degradation.
To determine the effect of EBNA3C and Cyclin D1/CDK6 complex on pRb mediated cell growth suppression, an osteosarcoma cell line, Saos2, was transfected with the expression plasmids for myc-tagged pRb, flag-tagged Cyclin D1, flag-tagged CDK6 and EBNA3C as indicated in the figure (Fig. 8A–D). Cells were additionally transfected with a GFP expression vector. The cells were selected with G418 for 2 weeks and the proliferation rate of the selected cells was measured by an automated cell counter for 6 days (Fig. 8). Dead cells were excluded using Trypan Blue staining. The rationale for choosing Saos2 as recipient cells was that cell growth suppression and morphological changes can be observed in Saos2 cells that express pRb de novo [48]. The results showed that EBNA3C together with Cyclin D1/CDK6 complex effectively reduced the growth suppressive effect of pRb. The cell-proliferation rate of cells expressing pRb either with EBNA3C or Cyclin D1/CDK6 complex was 1.5-2 fold higher than pRb alone (Fig. 8A). However, interestingly EBNA3C together with Cyclin D1/CDK6 complex significantly enhanced the cell proliferation rate, which was approximately either 6 fold higher than pRb alone or 3 fold higher than pRb when co-expressed with either EBNA3C or Cyclin D1/CDK6 complex (Fig. 8A). To check the expression levels of these proteins, the selected cells were subjected to western blot analysis (Fig. 8B). The results showed that the pRb expression levels were significantly reduced in EBNA3C or Cyclin D1/CDK6 expressing samples, whereas no changes of expression were observed for other proteins (Fig. 8B). GAPDH was used as an internal loading control and expression of GFP indicated equivalent amount of total protein lysate prepared from selected cells (Fig. 8B).
In order to corroborate the previous experiment, we next performed a colony formation assay, where cells were similarly transfected with different combinations of expression constructs as stated above. After selection of the transfected cells with G418 similarly as stated above for 2 weeks, the number of antibiotic-resistant colonies was counted (Fig. 8C–D). In agreement with the previous experiment, the results showed that co-expression of both EBNA3C and Cyclin D1/CDK6 proteins with pRb in Saos-2 cells resulted in an increase in the number of colonies compared to pRb alone (Fig. 8C, compare panels 1–3 and Fig. 8D, bar diagram). However, interestingly EBNA3C together with Cyclin D1/CDK6 complex markedly increased the antibiotic-resistant colonies (Fig. 8C, compare panels 1-4 and Fig. 8D, bar diagram). Overall, these results indicate that EBNA3C can utilize the function of Cyclin D1/CDK6 to neutralize the growth inhibitory effect of pRb.
In the context of the above-described results, we hypothesized that EBNA3C exploits Cyclin D1/CDK6 to promote LCL proliferation by inactivating pRb. To address this, LCLs were stably transduced with lentiviruses that express short hairpin RNA to silence either EBNA3C (Sh-E3C) or cyclin D1 (Sh-CyD1). The Sh-Control RNA is not complementary to human genome sequences. Stable transduction was verified by GFP expression (Fig. 9A). The expression levels of knocked down genes among these cells were then detected by Western blot analysis (Fig. 9B). The results showed that the level of EBNA3C or Cyclin D1 was knocked down by sh-RNA whereas LCL1 transduced with sh-Control had levels similar to LCL1 (Fig. 9B). The results also showed that pRb expression levels were slightly increased in both EBNA3C and Cyclin D1 knocked down samples, whereas there were no alterations observed for other Cyclin D expression levels (Fig. 9B). In order to determine whether both EBNA3C and Cyclin D1 are critical to maintain the proliferation of EBV transformed cells, a proliferation analysis was done (Fig. 9C). The results showed that the proliferation rate of both wild-type LCL1 and LCL1 infected with the lentivirus control sh-RNA (Sh-Control) expressing physiological level of both EBNA3C and Cyclin D1 was significantly higher than that of LCLs with Sh-E3C and Sh-CyD1 (Fig. 9C). In agreement with the previously published results [20], [49], we also observed that the proliferation rate of LCLs containing Sh-E3C with reduced levels of EBNA3C expression was approximately 3 fold slower than that of control cell-lines (Fig. 9C). Interestingly, the proliferation rate of LCLs with Sh-CyD1 was 50% higher than LCLs with Sh-E3C and only about 1.5 fold lower than that of control. This suggests that other D-type cyclins might be involved in LCL growth, particularly Cyclin D2 which was shown earlier to be associated in EBV mediated lymphomagenesis and probably transcriptionally up-regulated by one of the other major EBV latent antigen LMP1 [45]. However, it is clear from repeated analyses that cyclin D1 knock-down correlates with an increase in doubling time. The results support the notion that EBNA3C and cyclin D1 are critical for driving the growth of EBV transformed cells.
It has been shown earlier that both EBV positive cells and cells stably expressing EBNA3C can bypass G1/S phase checkpoint caused by serum starvation [20], [35]. Cell-cycle profiles of cells cultured in medium with 0.1% FBS were analyzed by flow cytometry (Fig. 10). Analyses of serum-starved, EBV negative cell lines BJAB and DG75 and LCLs sh-E3C and sh-CyD1 revealed an increased percentage of cells at the G0/G1 phase of the cell cycle (Fig. 10A, B) and decreased percentage of cells in the G2/M phases (Fig.10A, C). Fig. 10B and 10C represents the difference in both G0/G1 and G2/M phases due to serum starvation, respectively. However, under the same culture conditions, the EBV-positive LCLs - LCL1, LCL2, LCL1-with Sh-control and BJAB-cells stably expressing EBNA3C (E3C# 7 and E3C# 10) continued through the cell-cycle without being arrested at any particular phase (Fig. 10A histograms, B and C). Furthermore, the results also indicated that upon knockdown of both EBNA3C and Cyclin D1, LCLs underwent a substantial degree of apoptosis (Sub G0) in response to serum starvation, similar to EBV negative cell lines, BJAB and DG75 (Fig. 10A). However, there was no sign of apoptosis observed either in BJAB cells stably expressing EBNA3C or wild-type LCLs (Fig. 10A). Altogether, this experiment demonstrated that EBNA3C and Cyclin D1 positively contribute to cell growth in EBV transformed cells and are critical for overriding the G1 block as a result of serum starvation.
The cyclin D1 gene amplification has been observed in cancers of the breast, head and neck or larynx [50], [51], [52]. Chromosomal rearrangement is another cause of Cyclin D1 over-expression associated with centrocytic lymphomas [53], parathyroid adenomas [54] and mantle cell lymphoma [28]. The obvious association of Cyclin D1 with cancer has led the investigators to uncover its oncogenic properties. In fact, Cyclin D1 was shown to cooperate with the Ras oncoprotein for cell transformation [55].
Earlier reports have suggested that immortalization of primary B-lymphocytes by EBV is accompanied by transcriptional activation of the cyclin D2 gene but not cyclin D1 [43], [44], [45]. However, a number of studies showed noticeable changes in Cyclin D1 protein levels in both EBV positive LCLs [35] and EBV positive SCID mice lymphomas [37]. Despite the controversy regarding the Cyclin D1 expression in EBV positive B-lymphoma cells, it is clear that in order to deregulate the entire mammalian cell-cycle, EBNA3C manipulates G1 restriction point through disruption of Cyclin/CDK-pRb-E2F pathway [20].
Cyclin D1 is over-expressed in a variety of human cancers that do not exhibit cyclin D1 gene amplification or structural abnormalities of the cyclin D1 locus, which suggests that increased Cyclin D1 stability is a potential mechanism. Mutations of cyclin D1 at T286 and P287 have been found in human tumors [24] and alter Cyclin D1 nuclear localization as well as stability. Our data showed that both EBV infection in primary B-cells and EBV persistence in cancer cell lines increased protein stability. However the cyclin D1 mRNA level was unchanged. Similar to virus infection, EBNA3C gene expression increased Cyclin D1 levels without altering mRNA levels. It is important to determine if these effects also occur in vivo.
The results presented here also demonstrated that the expression of Cyclin D2 and D3 were up-regulated in quiescent cells infected with EBV probably through distinctly different mechanisms. EBV infection or its transforming protein latent membrane protein 1 (LMP1) up-regulates Cyclin D2 expression in primary B-lymphocytes and Burkitt's lymphoma cells [45]. None of the studies have shown an important role for Cyclin D3 in EBV-mediated cell transformation. Studies have suggested that the D-type cyclins may have non-overlapping functions at specific steps in B-cell differentiation [32], and that the expression of different D-type cyclins may be influenced by EBV infection through distinctive pathways. Thus, a potential mechanism which involves the contribution of Cyclin D1 in EBV-mediated B-cell transformation is the constitutive induction of these key cell-cycle regulators which leads to pRb hyper-phosphorylation and uncontrolled cell proliferation.
Several lines of evidence suggest that Cyclin D1 is targeted by the E3 ligase, SCFFBX4-αB crystallin for degradation [26]. Elevated expression of FBX4 and αB crystallin is also found to trigger the destruction of wild-type Cyclin D1, but not the phosphorylation-deficient Cyclin D1 mutant, D1T286A [26]. Thus, impairment of SCFFBX4-αB crystallin function may also account for Cyclin D1 overexpression. Data from the ubiquitination assay showed that EBNA3C efficiently inhibits Cyclin D1 poly-ubiquitination, which led us to speculate that EBNA3C may interact with this particular E3 ligase and inhibit its ability to ubiquitinate Cyclin D1. The SCFSkp2 ligase has also been shown to be involved in the degradation of Cyclin D1 [56], [57], [58], and knockdown of Skp2 molecule promoted marked accumulation of Cyclin D1 [57]. EBNA3C interacts with SCF components to regulate the stability of p27KIP1 and pRb [12], [17]. It is likely EBNA3C inhibition of SCFSkp2 reduces Cyclin D1 ubiquitination. EBNA3C may be a deubiquitinase or associate with one to regulate the stability of Mdm2 [18] and likely Cyclin D1. Since the expression level of Cyclin D1 is related to the potential for malignancy and the prognosis of a variety of cancers [30], [31], revealing the mechanisms governing the ubiquitin-proteasome mediated degradation of Cyclin D1 is of importance in designing therapeutic interventions. Conceivably, this approach could amplify the therapeutic window using Cyclin D1 as a target and enhance the efficacy of conventional drugs against EBV mediated oncogenesis.
We have shown earlier that EBNA3C can interact with Cyclin D1 using an in vitro GST-pulldown experiment [16]. Here, we examined the molecular association between EBNA3C and Cyclin D1 complexes to obtain a more in-depth understanding of the different domains of EBNA3C that modulate the activity of Cyclin D1 which will lead to further understanding the basic mechanism by which EBV regulates the mammalian cell-cycle. EBNA3C associates with Cyclin D1 via the same N-terminal domain, residues 130-190, that has been shown to bind many critical cell-cycle regulators [18] including other Cyclins - A and E [16]. In addition, a recent genetic study using recombinant EBV expressing conditionally active EBNA3C showed that deletion of this particular domain could not support cell proliferation of EBV transformed LCLs, signifying the importance of this domain within EBNA3C [49]. The association of EBNA3C with different Cyclins suggests is perhaps cell-cycle dependent and one may replace another depending on the stage in the cell-cycle, which ultimately leads to aberrant cell proliferation in EBV transformed cells. The previously published data and the data herein were generated using asynchronously growing cells; therefore it would be interesting to further elucidate these interactions in a cell-cycle dependent manner. However, using chemical synchronization is likely to distort the true activities underlying EBNA3C function with Cyclin complexes. Nevertheless, we will be undertaking this line of experimentation in the near future.
To promote G1-S phase transition, nuclear localization of Cyclin D1 is critical and it occurs either via decreased proteolysis in cytoplasm which facilitates nuclear import or through inhibition of GSK-3β function which triggers nuclear export via phosphorylation at T286 [27], [59]. Immunofluorescent studies showed that EBNA3C expression enforces nuclear localization of Cyclin D1. Increased stability and nuclear accumulation of Cyclin D1 in the presence of EBNA3C prompted us to examine whether EBNA3C can also negatively regulate GSK-3β function linked to the stability of Cyclin D1. Indeed, our data show that EBNA3C forms a complex with GSK-3β and significantly reduces its kinase activity toward Cyclin D1, thereby enhancing the nuclear retention of Cyclin D1. Altogether, these data suggest that either by increasing nuclear import by blocking the poly-ubiquitination level of Cyclin D1 or inhibiting nuclear export of Cyclin D1 via inhibiting the kinase activity of its negative regulator GSK-3β, EBNA3C would likely ensure the efficient nuclear accumulation of Cyclin D1 during G1-phase. However, we cannot eliminate the possibility that EBNA3C may also facilitate Cyclin D1 nuclear accumulation through additional strategies.
Cyclin D1 is central to the coordination of the cell-cycle progression at the G1 to S phase transition by integrating the control of pRb phosphorylation with the transcriptional activity of E2F [60]. Cyclin D1 in association with its binding partner, CDK4 or 6 phosphorylates pRb to facilitate S phase entry [60]. Previously we have shown that EBNA3C enhances the kinase activity of Cyclin A/CDK2 complex [15] and recruits an E3 ligase SCFSkp2 to degrade pRb [17]. Similarly, here we show that by an in vitro kinase assay EBNA3C increases the activity of Cyclin D1/CDK6 complex toward both Histone H1 and a truncated mutant of pRb. Moreover, EBNA3C in conjunction with Cyclin D1/CDK6 complex increases pRb poly-ubiquitination and thereby enhances its degradation process. In addition, we show EBNA3C coupled with Cyclin D1/CDK6 complex significantly abolishes the growth suppressive function of pRb in Saos-2 cells.
Studies using serum starved conditions have shown that both EBV and its essential nuclear antigen, EBNA3C can bypass G1 restriction point probably through disruption of Cyclin/CDK-pRb-E2F pathway [21], [36]. LMP1 has also been shown to be associated with resistance to G1 arrest during serum starvation [36]. Taking advantage of these findings, together with the use of sh-RNA mediated gene knockdown strategies, we have generated knockdown lymphoblastoid cell-lines targeting both EBNA3C and cyclin D1 transcripts and assayed for cellular proliferation to carefully determine the plausible role of both of these viral and cellular oncoproteins. Indeed, our results show that both EBNA3C and Cyclin D1 are critical for efficient proliferation of EBV infected B-cells. Moreover, the results point out that upon knockdown of these gene products, cells undergo significant apoptosis, probably through induction of the activities of the tumor suppressors – p53 and pRb. Earlier results [14] and the data herein adequately show that EBNA3C critically regulates the growth suppressive properties of both p53 and pRb.
Overall, we have shown in this report that the essential EBV latent antigen, EBNA3C physically interacts with and stabilizes Cyclin D1 by blocking nuclear export or inhibiting the poly-ubiquitination. In addition, EBNA3C alters pRb phosphorylation as well as stability by enhancing Cyclin D1/CDK6 kinase activity, thereby nullifying pRb mediated growth suppressive activity (Fig. 11). Furthermore, knockdown of both EBNA3C and Cyclin D1 expression by lentivirus-delivered sh-RNA demonstrated that both EBNA3C and Cyclin D1 play a critical role in cell proliferation in EBV transformed cells. Thus, the present study provides an insight into the mechanisms linked to the development of EBV-associated B-cell lymphomas through the enhancement of a major cell-cycle component, Cyclin D1, which is known to orchestrate the activities of a vast range of cellular networks that are important in the development of human cancers.
|
10.1371/journal.pgen.1001373 | The History of African Gene Flow into Southern Europeans, Levantines, and Jews | Previous genetic studies have suggested a history of sub-Saharan African gene flow into some West Eurasian populations after the initial dispersal out of Africa that occurred at least 45,000 years ago. However, there has been no accurate characterization of the proportion of mixture, or of its date. We analyze genome-wide polymorphism data from about 40 West Eurasian groups to show that almost all Southern Europeans have inherited 1%–3% African ancestry with an average mixture date of around 55 generations ago, consistent with North African gene flow at the end of the Roman Empire and subsequent Arab migrations. Levantine groups harbor 4%–15% African ancestry with an average mixture date of about 32 generations ago, consistent with close political, economic, and cultural links with Egypt in the late middle ages. We also detect 3%–5% sub-Saharan African ancestry in all eight of the diverse Jewish populations that we analyzed. For the Jewish admixture, we obtain an average estimated date of about 72 generations. This may reflect descent of these groups from a common ancestral population that already had some African ancestry prior to the Jewish Diasporas.
| Southern Europeans and Middle Eastern populations are known to have inherited a small percentage of their genetic material from recent sub-Saharan African migrations, but there has been no estimate of the exact proportion of this gene flow, or of its date. Here, we apply genomic methods to show that the proportion of African ancestry in many Southern European groups is 1%–3%, in Middle Eastern groups is 4%–15%, and in Jewish groups is 3%–5%. To estimate the dates when the mixture occurred, we develop a novel method that estimates the size of chromosomal segments of distinct ancestry in individuals of mixed ancestry. We verify using computer simulations that the method produces useful estimates of population mixture dates up to 300 generations in the past. By applying the method to West Eurasians, we show that the dates in Southern Europeans are consistent with events during the Roman Empire and subsequent Arab migrations. The dates in the Jewish groups are older, consistent with events in classical or biblical times that may have occurred in the shared history of Jewish populations.
| The history of human migrations from Africa into West Eurasia is only partially understood. Archaeological and genetic evidence indicate that anatomically modern humans arrived in Europe from an African source at least 45,000 years ago, following the initial dispersal out of Africa [1], [2]. However, it is known that Southern Europeans and Levantines (people from modern day Palestine, Israel, Syria and Jordan) have also inherited genetic material of African origin due to subsequent migrations. One line of evidence comes from Y-chromosome [3] and mitochondrial DNA analyses [4]–[6]. These have identified haplogroups that are characteristic of sub-Saharan Africans in Southern Europeans and Levantines but not in Northern Europeans [7]. Auton et al. [8] presented nuclear genome-based evidence for sharing of sub-Saharan African ancestry in some West Eurasians, by identifying a North-South gradient of haplotype sharing between Europeans and sub-Saharan Africans, with the highest proportion of haplotype sharing observed in south/southwestern Europe. However, none of these studies used genome-wide data to estimate the proportion of African ancestry in West Eurasians, or the date(s) of mixture. Throughout this report, we use “African mixture” to refer to gene flow into West Eurasians since the divergence of the latter from East Asians; thus, we are not referring to the much older dispersal out of Africa ∼45,000 years ago but instead to migrations that have occurred since that time.
We assembled data on 6,529 individuals drawn from 107 populations genotyped at hundreds of thousands of single nucleotide polymorphisms (SNPs) (Table S1). This included 3,845 individuals from 37 European populations in the Population Reference Sample (POPRES) [9], [10], 940 individuals from 51 populations in the Human Genome Diversity Cell Line Panel (HGDP-CEPH) [11], [12], 1,115 individuals from 11 populations in the third phase of the International Haplotype Map Project (HapMap3) [13], 392 individuals who self reported as having Ashkenazi Jewish ancestry from the InTraGen Population Genetics Database (IBD) [14] and 237 individuals from 7 populations in the Jewish HapMap Project [15]. For most analyses, we used HapMap3 Utah European Americans (CEU) to represent Northern Europeans and HapMap3 Yoruba Nigerians (YRI) to represent sub-Saharan Africans, although we also verified the robustness of our inferences using alternative populations.
We curated these data using Principal Components Analysis (PCA) [16] (Table S2), with the most important steps being: (i) Removal of 140 individuals as outliers who did not cluster with the bulk of samples of the same group, (ii) Removal of all 8 Greek samples as they separated into sub-clusters in PCA so that it was not clear which of these clusters was most representative, (iii) Splitting the Bedouins into two genetically discontinuous groups, and (iv) Reclassifying the 5 Italian groups into three ancestry clusters (Sardinian, Northern-Italy, and Southern-Italy) (see details in Text S1, Figure S1). A comparison of results before and after this curation is presented in Table S3, where we show that this data curation does not affect our qualitative inferences.
To study the signal of African gene flow into West Eurasian populations, we began by computing principal components (PCs) using San Bushmen (HGDP-CEPH- San) and East Eurasians (HapMap3 Han Chinese- CHB), and plotted the mean values of the samples from each West Eurasian population onto the first PC, a procedure called “PCA projection” [17], [18]. The choice of San and CHB, which are both diverged from the West Eurasian ancestral populations [19], [20], ensures that the patterns in PCA are not affected by genetic drift in West Eurasians that has occurred since their common divergence from East Eurasians and South Africans. We observe that many Levantine, Southern European and Jewish populations are shifted towards San compared to Northern Europeans, consistent with African mixture, and motivating formal testing for the presence of African ancestry (Figure 1, Figure S2).
To formally test for the presence of African mixture, we first performed the 4 Population Test (Figure S3). This test is based on the insight that if populations A and B form sister groups relative to C and D, the allele frequency differences (pA-pB) and (pC-pD) should be uncorrelated as they represent independent periods of random genetic drift [21]. Applying the 4 Population Test to the proposed relationship (YRI,(Papuan,(CEU,X))) where X is a range of West Eurasian populations, we find significant violations for all Southern European, Jewish and Levantine populations but not for Northern Europeans (Table 1). The results remain unchanged even when we use alternate topologies replacing YRI with other African populations (Text S2, Table S4). We further verified these inferences with the 3 Population Test [21], which capitalizes on the insight that for any 3 populations (X; A, B), the product of the allele frequency differences (pX-pA) and (pX-pB) is expected to be negative only if population X descends from a mixture of populations related to populations A and B [21] (Figure S3). We verified that this method is robust to SNP ascertainment bias by carrying out simulations showing that the 3 Population Test detects real admixture even if all SNPs used in the analysis are discovered in population A, population B, or in both populations A and B (Text S3; Table S5; Figure S4). Application of the test to each West Eurasian population (using A = YRI and B = CEU) finds little or no evidence of mixture in North Europeans but highly significant evidence in many Southern European, Levantine and Jewish groups (Table 1).
To estimate the proportion of sub-Saharan African ancestry in the various West Eurasian populations that showed significant evidence of mixture, we used f4 Ancestry Estimation [21], a method which produces accurate estimates of ancestry proportions, even in the absence of data from the true ancestral populations. This method estimates mixture proportions by fitting a model of mixture between two ancestral populations, followed by (possibly large) population-specific genetic drift. Briefly, we calculate a statistic that is proportional to the correlation in the allele frequency difference between West Eurasians and sub-Saharan Africans, and divide it by the same statistic for a population of sub-Saharan African ancestry, like YRI (Figure 2). This method has been shown through simulation to be robust to ascertainment bias on the SNP arrays and deviations from the assumed model of mixture (e.g. date and number of mixture events) [21].
Application of f4 Ancestry Estimation suggests that the highest proportion of African ancestry in Europe is in Iberia (Portugal 3.2±0.3% and Spain 2.4±0.3%), consistent with inferences based on mitochondrial DNA [6] and Y chromosomes [7] and the observation by Auton et al. [8] that within Europe, the Southwestern Europeans have the highest haplotype-sharing with Africans. The proportion decreases to the north and we find no evidence for mixture in Russia, Sweden and Scotland (Table 2, Figure S5). We also detect about 3-5% sub-African ancestry in all the Jewish populations, a finding that is novel as far as we are aware, and certainly has not been unambiguously demonstrated or quantified. For Levantines, the proportions are often higher: 9.3%±0.4% in Palestinians and >10% in the Bedouins (standard errors were calculated using a Block Jackknife as described in Materials and Methods). Table 2 presents the ancestry estimates that we obtain for all West Eurasian populations with significant evidence of mixture by the 4 Population Test (Z-score < -3). To test if our inferences are dependent on the sub-Saharan African population that was used as the reference group, we also repeated analyses with other sub-Saharan African populations replacing YRI. This analysis shows that our estimates of mixture proportions do not change significantly based on the ancestral population used (Text S2c, Table S6). We obtained similar estimates when we applied STRUCTURE 2.2 [22] to estimate the mixture proportions using ∼13,900 independent markers (that were not in linkage disequilibrium (LD) with each other) (Table 2, Figure S6).
The finding of sub-Saharan African ancestry in West Eurasians predicts that there will be a signature of admixture LD in the populations that experienced this mixture. That is, there will be LD between all markers that are highly differentiated between the two ancestral populations and the allele will be strongly correlated to the local ancestry [23]. Hence, there will be chromosomal segments of African ancestry with lengths that reflect the number of recombination events that have occurred since mixture, and thus can be used to estimate an admixture date. Figure 3 shows that this expected pattern is observed empirically in the decay of LD in four example West Eurasian populations, where we enhance the effects of admixture LD by weighting the SNP comparisons by frequency difference between the ancestral Africans (YRI) and ancestral West Eurasians (CEU). In the Southern European, Jewish and Levantine populations, this procedure produces clear evidence of admixture LD (Figure 3). However, Northern Europeans (Russians in Figure 3) do not show any evidence of African gene flow, consistent with the 4 Population and 3 Population Test results and Figure 1. Similar results are seen for other West Eurasian and Jewish populations that show evidence of mixture in the 4 Population Test.
To estimate a date for the mixture event, we developed a novel method ROLLOFF that computes the time since mixture using the rate of exponential decline of admixture LD in plots such as Figure 3. ROLLOFF computes the correlation between a (signed) statistic for LD between a pair of markers and a weight that reflects their allele frequency differentiation in the ancestral populations. By examining the correlation between pairs of markers as they become separated by increasing genetic distance and fitting an exponential distribution to this rolloff by least squares, we obtain an estimate of the date (see Materials and Methods and Text S4). ROLLOFF also computes an approximately normally distributed standard error by carrying out Weighted Jackknife analysis [24], where we drop one chromosome in each run and study the fluctuation of the statistic in order to assess the stability of the estimate.
To verify the accuracy and sensitivity of ROLLOFF, we carried out extensive simulations by constructing the genomes of individuals of mixed ancestry by sampling haplotypes from North Europeans (CEU) and West Africans (YRI) (see Materials and Methods). We verified that ROLLOFF produces accurate estimates of the date of mixture, even in the case of old admixture (up to 300 generations – Figure 4) and is robust to substantially inaccurate ancestral populations as well as fine scale errors in the genetic map (Text S4; Figure S7; Figure S8; Table S7; Table S8). In addition, to test the robustness of our inferences, we applied all the methods to African Americans and obtained consistent results for the proportion of mixture (79.4±0.3%) and date of mixture (6±1), which is in agreement with previous reports [25], [26]. However, in the case of low mixture proportion and old admixture dates, we observed that there is a slight bias in the estimated date (Text S4d, Table S9). This effect is related to the weakness of the signal: it attenuates as the sample size or admixture proportion becomes larger (Text S4d, Table S10, Table S11).
An important concern was how ROLLOFF would perform when the true history of admixture involved multiple pulses of gene exchange, rather than the single pulse of gene exchange that we modeled. To explore this, we first simulated two distinct gene flow events, and then estimated the date using a single exponential distribution. The simulations show that ROLLOFF's estimate of the date tends to correspond reasonably well to the more recent admixture event, with a slight upward bias towards the older date. Second, we performed simulations under a continuous gene flow model and found that the estimated dates are intermediate between the start and end of the gene flow, as expected (Figure S9; Figure S10; Table S12). To explore if we could obtain a better inference of the range of dates, we tried fitting sum of multiple exponential distributions, but this did not work reliably, which may be related to the well-known difficulty of fitting a sum of exponentials to data with even a small amount of noise [27] (Text S4). Pool and Nielsen recently showed that multi-marker haplotype data could be useful for distinguishing a single pulse of gene exchange from changing migration rates over time [28]. However, a complication with applying this approach to relatively old dates is that haplotype-based methods need to model background LD. In the case of old mixture events (dozens or hundreds of generations), inaccurate modeling of background LD can bias estimates [26], [29]. We are not aware of any published method that can produce accurate date estimates while modeling background LD correctly for mixture dates as old as those that have been explored by ROLLOFF in Figure 4.
We applied ROLLOFF to all the West Eurasian populations that gave significant signals of mixture by the 4 Population Test, fitting a single exponential decay in each case. We estimate that the date of sub-Saharan African mixture in Portugal is 45±5 generations and in Spain is 55±3 generations. We estimate a more recent date of 34±3 for Bedouin-g1, 33±2 for Bedouin-g2, and 34±2 generations for Palestinians. We estimate older dates of ∼70–150 generations in the various Jewish populations, with wide and in most cases overlapping confidence intervals (Table 2; Figure S11). Averaging the mixture dates over all populations from each region (weighted by the inverse of the squared standard error), we obtain an average of 55 generations for Southern Europeans, 34 for Levantines and 89 for Jews.
As described above, in our simulations to explore the behavior of ROLLOFF we detect an upward bias in the date estimates that grew worse with older mixture dates, small mixture proportions, and small sample sizes (but does not appear to be affected by use of inaccurate ancestral populations). To assess the degree to which this bias might be affecting our date estimates, we performed simulations for each population in Table 2 separately, in which we set the number of samples, mixture proportion and time since mixture to match the parameters estimated from the real data. We repeated our simulations 100 times for each parameter setting and estimated the bias of our estimated date from the true (simulated) date. The bias is very small for the most of the Southern European and Levantine samples, which generally had large sample sizes, recent dates, and high mixture proportions. However, the bias is larger for the Jewish groups (Table 2, Table S13). Correcting for the bias inferred in our simulation of Table S12, we obtain corrected estimates of the average date of 55 generations for Southern Europeans, 32 for Levantines, and 72 for Jews. A caveat about these regional date estimates is that they reflect weighted averages across the populations in each region. However, the admixture events detected within each region may not reflect the same historical events; for example, it is plausible that the sub-Saharan African admixture in Spain and Italy have different historical origins.
The finding of African ancestry in Southern Europe dating to ∼55 generations ago, or ∼1,600 years ago assuming 29 years per generation [30], needs to be placed in historical context. The historical record documents multiple interactions of African and European populations over this period. One potential opportunity for African gene flow was during the period of Roman occupation of North Africa that lasted until the early 5th century AD, and indeed tomb inscriptions and literary references suggest that trade relations continued even after that time [31], [32]. North Africa was also a supplier of goods and products such as wine and olive oil to Italy, Spain and Gaul from 200–600 AD, and Morocco was a major manufacturer of the processed fish sauce condiment, garum, which was imported by Romans [33]. In addition, there was slave trading across the western Sahara during Roman times [7], [34]. Another potential source of some of the African ancestry, especially in Spain and Portugal, is the invasion of Iberia by Moorish armies after 711 AD [35], [36]. If the Moors already had some African ancestry when they arrived in Southern Europe, and then admixed with Iberians, we would expect the admixture date to be older than the date of the invasion, as we observe.
The signal of African mixture that we detect in Levantines (Bedouins, Palestinians and Druze) – an average of 32 generations or ∼1000 years ago – is more recent than the signal in Europeans, which might be related to the migrations between North Africa and Middle East that have occurred over the last thousand years, and the proximity of Levantine groups geographically to Africa. Syria and Palestine were under Egyptian political control until the 16th century AD when they were conquered by the Ottoman Empire. This is in concordance with our proposed dates. In addition, the Arab slave trade is responsible for the movement of large numbers of people from Africa across the Red Sea to Arabia from 650 to 1900 AD and probably even prior to the Islamic times [7], [37]. We caution that our sampling of the Middle East is sparse, and it will be of interest to study African ancestry in additional groups from this region.
A striking finding from our study is the consistent detection of 3–5% sub-Saharan African ancestry in the 8 diverse Jewish groups we studied, Ashkenazis (from northern Europe), Sephardis (from Italy, Turkey and Greece), and Mizrahis (from Syria, Iran and Iraq). This pattern has not been detected in previous analyses of mitochondrial DNA and Y chromosome data [7], and although it can be seen when re-examining published results of STRUCTURE-like analyses of autosomal data, it was not highlighted in those studies, or shown to unambiguously reflect sub-Saharan African admixture [15], [38]. We estimate that the average date of the mixture of 72 generations (∼2,000 years assuming 29 years per generation [30]) is older than that in Southern Europeans or other Levantines. The point estimates over all 8 populations are between 1,600–3,400 years ago, but with largely overlapping confidence intervals. It is intriguing that the Mizrahi Irani and Iraqi Jews—who are thought to descend at least in part from Jews who were exiled to Babylon about 2,600 years ago [39], [40]—share the signal of African admixture. (An important caveat is that there is significant heterogeneity in the dates of African mixture in various Jewish populations.) A parsimonious explanation for these observations is that they reflect a history in which many of the Jewish groups descend from a common ancestral population which was itself admixed with Africans, prior to the beginning of the Jewish diaspora that occurred in 8th to 6th century BC [41]. The dates that emerge from our ROLLOFF analysis in the non-Mizrahi Jews could also reflect events in the Greek and Roman periods, when there were large communities of Jews in North Africa, particularly Alexandria [34], [42]. We detect a similar African mixture proportion in the non-Jewish Druze (4.4±0.4%) although the date is more recent (54±7 generations; 44±7 after the bias correction). Algorithms such as PCA and STRUCTURE show that various Jewish populations cluster with Druze [15], which coupled with the similarity in mixture proportions, is consistent with descent from a common ancestral population. Importantly, the other Levantine populations (Bedouins and Palestinians) do not share this similarity in the African mixture pattern with Jews and Druze, making them distinct in their admixture history.
A caveat to these results is that we estimated dates assuming instantaneous mixture, but in fact we have not distinguished between the patterns expected for instantaneous admixture and continuous gene flow over a long period. In Text S4f, we report simulations showing that for continuous gene flow, the dates from ROLLOFF reflect the average of mixture dates over a range of times, and so the date should be interpreted only as an average number.
A potential issue that could in theory influence our findings is that the exact population contributing to African ancestry in West Eurasians is unknown. To gain insight into the African source populations, we carried out PCA analyses, which suggested that the African ancestry in West Eurasians is at least as closely related to East Africans (e.g. Hapmap3 Luhya (LWK)) as to West Africans (e.g. Nigerian Yoruba (YRI)) (the same analyses show that there is no evidence of relatedness to Chadic populations like Bulala) (Text S5 and Figure S12). We also used the 4 Population Test to assess whether the tree ((LWK, YRI),(West Eurasian, CEU)) is consistent with the data, and found no evidence for a violation, which is consistent with a mixture of either West African or East African ancestors or both contributing to the African ancestry in West Eurasians (Table S14; Figure S13). Historically, a mixture of West and East African ancestry is plausible, since African gene flow into West Eurasia is documented from both West Africa during Roman times [34] and from East Africa during migrations from Egypt [7]. It is important to point out, however, that the difficulty of pinpointing the exact African source population is not expected to bias our inferences about the total proportion and date of mixture. The f4 Ancestry Estimation method is unbiased even when we use a poor surrogates for the true ancestral African population (as long as the phylogeny is correct), as we confirmed by repeating analyses replacing YRI with LWK, and obtaining similar results (Table S15). Our ROLLOFF admixture date estimates are also similar whether we use LWK or YRI to represent ancestral African population (Table S15), as predicted by the theory.
In summary, we have documented a contribution of sub-Saharan African genetic material to many West Eurasian populations in the last few thousand years. A priority for future work should be to identify the source populations for this admixture.
We analyzed individuals of West Eurasian ancestry from several sources: The Population Reference Sample (POPRES) [9]–[10] (n = 3,845 samples from 37 populations genotyped on an Affymetrix 500K array), the Human Genome Diversity Cell Line Panel (HGDP-CEPH) [12] (n = 940 samples from 51 populations genotyped on an Illumina 650K array), The International Haplotype Map (HapMap) Phase 3 [13] (n = 1,115 samples from 11 populations genotyped on an Illumina 1M array), the InTraGen Population Genetics Database (IBD) [14] (n = 392 Ashkenazi Jews genotyped on an Illumina 300K array) and the Jewish HapMap Project [15] (n = 237 from 7 Jewish populations genotyped on an Affymetrix 6.0 array). We created a merged dataset containing 6,529 individuals -out of which 3,614 individuals of West Eurasian, African and Eastern Eurasian ancestry were used for the final analysis. Detailed information about the number of individuals and markers included in each analysis is provided in Table S1. We used NCBI Build 35 to determine physical position and the Oxford LD-based map genetic to determine genetic positions of all SNPs [43].
Source code and executables for the ROLLOFF software are available on request from NP.
|
10.1371/journal.pgen.1003982 | Tay Bridge Is a Negative Regulator of EGFR Signalling and Interacts with Erk and Mkp3 in the Drosophila melanogaster Wing | The regulation of Extracellular regulated kinase (Erk) activity is a key aspect of signalling by pathways activated by extracellular ligands acting through tyrosine kinase transmembrane receptors. In this process, participate proteins with kinase activity that phosphorylate and activate Erk, as well as different phosphatases that inactivate Erk by de-phosphorylation. The state of Erk phosphorylation affects not only its activity, but also its subcellular localization, defining the repertoire of Erk target proteins, and consequently, the cellular response to Erk. In this work, we characterise Tay bridge as a novel component of the EGFR/Erk signalling pathway. Tay bridge is a large nuclear protein with a domain of homology with human AUTS2, and was previously identified due to the neuronal phenotypes displayed by loss-of-function mutations. We show that Tay bridge antagonizes EGFR signalling in the Drosophila melanogaster wing disc and other tissues, and that the protein interacts with both Erk and Mkp3. We suggest that Tay bridge constitutes a novel element involved in the regulation of Erk activity, acting as a nuclear docking for Erk that retains this protein in an inactive form in the nucleus.
| Extracellular regulated kinases (Erk) mediate signalling by pathways activated by tyrosine kinase transmembrane receptors. The level of activated Erk depends on a highly regulated balance between cytoplasmic kinases and nuclear/cytoplasmic phosphatases, which determine the state of Erk phosphorylation. This affects Erk activity and its subcellular localization, defining the repertoire of Erk targets, and consequently, the cellular response to Erk. In this work, we use a genetic approach to characterise the gene tay bridge as a novel component of the EGFR/Erk signalling pathway. Tay bridge has a domain of homology with human AUTS2, and was previously identified due to the neuronal phenotypes displayed by loss-of-function mutations. We show that Tay bridge antagonizes EGFR signalling in the Drosophila melanogaster wing disc and other tissues, and that the protein interacts with both Erk and Mkp3. We suggest that Tay bridge constitutes a novel element involved in the regulation of Erk activity, acting as a nuclear docking for Erk that retains this protein in an inactive form in the nucleus. These results could provide important insights into the clinical consequences of AUTS2 mutations in humans, which are related to behavioural perturbations including autism, mental retardation, Attention Deficit Hyperactivity Disorder and alcohol drinking behaviour.
| The Epidermal Growth Factor Receptor (EGFR) signalling pathway is a conserved module that plays multiple roles during development and tissue homeostasis in eukaryotic organisms [1]–[3]. The best-characterized functions of the pathway involve the EGFR downstream proteins Sos, Ras, Raf, Mek and Erk, the MAPK that is encoded by rolled in Drosophila melanogaster [4]. The activity of these core components is required in multiple developmental contexts, influencing cell proliferation, migration, apoptosis, epithelial integrity and cell fate acquisition [1], [5]. A key node in the regulation of EGFR signalling occurs at the level of Erk phosphorylation and de-phosphorylation by Mek and dual-specificity phosphatases, respectively [6]–[8]. In general, upon activation by Mek, the Erk serine/threonine kinase is transported into the nucleus, where it can phosphorylate specific transcription factors, regulating their activity and consequently gene expression. Erk is de-phosphorylated and inactivated by dual-specificity phosphatases, which promote Erk accumulation in an inactive state in the cytoplasm [2], [9].
The nucleus-cytoplasm compartmentalization of Erk is also regulated by several proteins acting as scaffolds, which influence the kinetics of Erk activation by favouring its association with upstream components, or that target Erk to different substrates by regulating its subcellular localization [10]–[11]. Thus, Kinase suppressor of Ras (Ksr) and MEK partner 1 (MP-1) facilitate the phosphorylation of Erk by Mek [11]–[16], whereas β-arrestin and Sef (Similar Expression to FGF genes) serve as scaffolds directing Erk activity toward different subcellular localizations and sets of target proteins [17]–[18]. In fact, because Erk lacks nuclear localization or export sequences, it appears that its subcellular compartmentalization is mostly determined by binding to scaffolds, anchors and substrates [8], [10], [19]. In the absence of active export, Erk tends to accumulate inside the nucleus, and it has been suggested that imported Erk binds to nuclear anchoring proteins that difficult its free diffusion to the cytoplasm [6].
The EGFR signalling system has been extensively characterised in Drosophila, an organism that has been instrumental to identify the intricacies of signalling regulation in vivo [1], [20]–[22]. Furthermore, the exquisite sensitivity of several developmental processes to variations in levels of EGFR signalling has driven the search and identification of many components of the pathway through genetic screens, expression profiling and cell culture experiments [22]–[25]. The wing disc, the epithelial tissue that gives rise to the adult wing and part of the thorax, is particularly sensitive to changes in the levels of EGFR signalling [26]–[27]. The function of EGFR in this tissue is required for cell proliferation and viability [28], for the specification of the wing disc and its territorial subdivision [26], [29]–[32], and also in cell fate choices affecting sensory organs and veins [33]–[34]. In this last process, the function of the pathway is needed to promote the formation of the veins, longitudinal stripes of cells that differentiate a cuticle thicker and more pigmented than the cuticle of inter-vein cells [35]–[36].
We conducted a gain-of-function screen aimed to identify genes regulating wing vein differentiation, expecting that some of these genes would encode novel components of the signalling pathways driving the formation of these structures [37]. In this screen, we identified a P-UAS insertion in the gene tay bridge (tay) that in combination with a vein-specific Gal4 driver causes the elimination of the longitudinal veins, a phenotype reminiscent of loss of EGFR activity in the developing veins [27], [37]. Tay encodes a large protein of 2486 amino acids expressed predominantly in the central nervous system [38]. Mutant tay flies present a constriction in the protocerebral bridge, and display reduced walking speed, reduced sensitivity to the effects of alcohol and defective compensation of rotatory stimuli during walking [38]–[39]. The Carboxi-terminal part of Drosophila Tay presents homology with mammalian AUTS2, a neuronal nuclear protein that is related to autism [40]–[41], mental retardation [42], [43], Attention Deficit Hyperactivity Disorder [44], and alcohol drinking behaviour [39]. Auts2 expression is maximal in maturating neurons and declines as these cells become mature, suggesting that its function is required for neuronal differentiation [41], [45].
Here we report a genetic and developmental analysis of tay in the wing disc, and show that the function of Tay here is primarily related to the regulation of EGFR signalling. Thus, excess and loss of tay results in opposite phenotypes of loss- and extra veins, respectively, that are caused by changes in the levels of Erk activity. In addition, Tay level of expression modifies the phenotypic outcomes of altered EGFR signalling. We identify molecular interactions between Tay and Erk that might underline both the effects of Tay on Erk phosphorylation and the effects of Erk on Tay nuclear accumulation. All together, our results suggest that Tay is a novel component of the EGFR/Erk signalling pathway that regulates the nucleus/cytoplasm distribution of Erk.
EP-866 is a P-GS element inserted in the first intron of tay, and was selected in a gain-of-function screen designed to identify genes that, when over-expressed, affect the differentiation of the wing veins [37]. The combination of EP-866 with a variety of Gal4 lines reduces the size of the wing and causes the partial loss of longitudinal veins (Fig. 1A–D; Fig. S1H–J). The most extreme phenotypes are observed in combinations of EP-866 with Gal4 drivers expressed in the entire wing blade and hinge (nub-Gal4/EP-866; Fig. 1B). A weaker version of this phenotype is detected in combinations with a Gal4 driver expressed only in the central region of the wing blade (salEPv-Gal4/EP-866; Fig. 1C). The reduction in wing size and loss of veins occurs in a compartment-specific manner, as they are also observed in combinations with the hh-Gal4 and ap-Gal4 drivers (Fig. S1J and data not shown). In all cases, the drastic reduction in wing size is associated with a reduction of cell proliferation, and not to the induction of cell death. Thus, wing discs of combinations between EP-866 and Gal4 drivers show a very low number of mitotic cells and no activation of Caspase3 (Fig. S1A–G′). When the gene affected by the EP-866 insertion is over-expressed during pupal development, the size of the wing is normal, but the veins fail to differentiate (Fig. 1D). EP-866/Gal4 combinations also display phenotypes in other adult structures, including fusion of tarsal joints in the legs (dll-Gal4/EP-866; Fig. S1A, C), a significant reduction in the size of the eye (ey-Gal4/EP-866; data not shown) and loss of sensory organs in the thorax (ap-Gal4/EP-866; Fig. S1B, D). The strength of the EP-866/Gal4 phenotype increases with the number of copies of both the Gal4 and the EP-866 insertion (Fig. S1K–M).
The most likely candidate to cause the over-expression phenotype of EP-866/Gal4 combinations is the gene tay (Fig. 1E). Nonetheless, the genes CG15916 (5 Kb) and shibire (7 Kb) are close to the EP-866 insertion, and adjacent to tay is located CG9066, which is oriented in the 3′ to 5′ direction of transcription regarding the UAS sequences of the P-GS insertion. We know that tay, CG15916 and shi are over-expressed when EP-866 is combined with the salEPv-Gal4 [37]. However, the phenotypes of wing size reduction and loss of veins observed in EP-866/salEPv-Gal4 and EP-866/shv-Gal4 flies are suppressed when we introduced a UAS-tay-RNAi construct in these combinations (Fig. 1F–G, compare with 1C and D, respectively). In addition, the over-expression of Tay results in identical phenotypes of variable vein loss and wing size reduction (see below), indicating that tay causes the over-expression phenotypes of EP-866/Gal4 combinations.
tay encodes a protein of 2486 amino-acids which most remarkable characteristic is a 30% of identity in the 1764–2019 amino acid region with a 486–782 stretch of the 1295 amino acid long human protein AUTS2 (Autism Susceptibility Candidate 2) (see below). The expression of tay occurs ubiquitously in all imaginal discs (Fig. 1I and data not shown), although we can also observe higher levels of expression in cells adjacent to the veins during pupal development (Fig. 1J). Tay is also expressed at other developmental stages, and during embryonic development its mRNA and protein are detected prominently in the central nervous system (Fig. S3E–G and data not shown). To visualize the accumulation of the Tay protein, we generated a specific polyclonal antibody (Fig. S2B), and found that the protein is present in the nucleus of all imaginal discs and salivary gland cells (Fig. 1K–N). The accumulation of Tay is very much reduced or lost in dorsal wing compartments expressing a tay RNA interference (Fig. 1H–H′). We also confirmed the specificity of this antibody by staining cells homozygous for a tay deficiency, where we found that the signal is completely lost (Fig. S2C–C′). The subcellular localization of the protein in wing discs over-expressing Tay is mostly nuclear, although some cytoplasmic staining is detected at higher level of over-expression (Fig. 1O–O′). These observations suggest that the adult phenotypes associated to Tay over-expression are caused by the accumulation of Tay at higher than normal levels in the nuclei of imaginal cells that normally express the gene. Interestingly, we also detected Tay in the cytoplasm of a subset of motoneurons in the central nervous system (CM and JFdC, data not shown), indicating that the protein subcellular localization is regulated in a cell-type specific manner.
To identify the normal requirement of Tay during wing development, we reduced the levels of tay mRNA by expressing its RNA interference (tay-RNAi) in different domains of the wing disc. When tay-RNAi is expressed in the wing blade (638-Gal4/UAS-tay-i) the wings are reduced in size (32% smaller than wild type wings without changes in cellular size), display ectopic veins and show some defects in the most distal region of the wing margin (Fig. 2B). These phenotypes are caused by the reduction of tay, because they are enhanced in a genetic background with only one copy of the gene (Fig. 2C–D; 638-Gal4/+; Df(1)tay/UAS-tay-i). To generate stronger loss-of-function conditions, we made two small deficiencies by transposition (EP-866Rev34 and EP-866Rev40; see Fig. S2A), and a deficiency that eliminates tay and the adjacent gene CG16952 (Df(1)tay; Fig. S2A). These alleles are embryonic lethal in homozygous flies, and consequently they were analysed in mitotic recombination clones. The results obtained in Df(1)tay, EP-866rev40 and EP-866rev34 clones were identical, with cells deficient for tay forming clones that differentiate ectopic veins in inter-vein territories (Fig. 2E–G and data not shown). Interestingly, only a fraction of the mutant cells in each clone differentiate as ectopic veins of normal thickness (Fig. 2E–G). These phenotypes were very similar to those observed in wings expressing the tay-RNAi (compare with Fig. 2A–D).
The over-expression of tay in the wing imaginal disc prevents vein differentiation, macrochaetae formation and wing growth. Conversely, loss of tay function causes the formation of veins in inter-vein regions. These phenotypes are reminiscent to those caused by alterations in the levels of EGFR signalling, because loss of EGFR function impedes vein differentiation, and the increase in EGFR activity causes the formation of extra veins [27], [46]. To study the possible interactions between Tay and EGFR signalling, we made genetic combinations in which tay gain or loss of expression conditions were introduced in genetic backgrounds with modified EGFR activity. We find that the reduction of tay expression enhances the extra-vein phenotype caused by increased EGFR signalling. Thus, knock-down of tay enhances vein differentiation in RasV12 (Fig. 3A–C) and ectopic rhomboid (Fig. 3D–F) backgrounds. These observations suggest that Tay function is necessary either to attenuate EGFR signalling or to reduce the response to particular levels of EGFR signalling. Compatible with these possibilities, Tay over-expression enhances the loss-of-vein phenotype caused by reduced activity of the pathway, for example in a situation when the expression of EGFR is reduced (Fig. 3G–I). Interestingly, the reduction of tay expression does not modify the complete loss of vein phenotype caused by strong reductions in EGFR signalling (Fig. 3J–L), indicating that Tay function is mostly required to modulate the levels of EGFR signalling once the pathway has been activated.
To analyse whether changes in the expression of tay directly affect EGFR signalling, we monitored the levels of di-Phosphorylated Erk (dP-Erk) and the expression of the EGFR transcriptional targets Delta and argos in tay over-expression conditions. The accumulation of dP-Erk in wild type wing discs is maximal in the developing L3 and L4 longitudinal veins and in the marginal veins [34]; Fig. 4B). dP-Erk accumulation is strongly reduced in these territories when Tay is over-expressed in the wing blade (Fig. 4F, compare with 4B). The expression of Delta (Dl), which is regulated by EGFR signalling during imaginal development [47], is maximal in the veins L3, L4 and L5 and in the marginal veins in wild type wing discs (Fig. 4C). Over-expression of tay in the central region of the wing blade causes a reduction of Dl expression in the veins L3 and L4 (Fig. 4G, compare with 4C). The vein L5 is not affected, because it is located outside the domain of salEPv-Gal4 expression (Fig. 4F–G). Therefore, this vein serves as an internal control in these experiments. We also observed changes in the transcription of argos, which expression is also regulated by the EGFR pathway and is maximal in the veins L3, L4 and L5 and in the marginal veins in wild type wing discs [48]; Fig. 4D). Over-expression of tay reduces argos-LacZ expression (Fig. 4H, compare with 4D). In all cases, the changes in Erk phosphorylation and in Dl/argos gene expression caused by Tay over-expression were consistently stronger than the loss of vein phenotype observed in the corresponding adult wings, as these wings still differentiate some stretches of the L3 and L4 veins (Fig. 4E).
We also checked the effects of loss of Tay in the accumulation of dP-Erk. For this experiment we expressed tay-RNAi in the dorsal compartment of the wing (ap-Gal4/UAS-tay-i). In these discs the ventral compartment serves as an internal control. We observed that the reduction of tay expression increases dP-Erk accumulation in dorsal compartments compared with the ventral ones (Fig. 4J–J′). In addition, the expression of tay-RNAi in the entire wing blade (638-Gal4/UAS-tay-i) causes ectopic argos-lacZ expression (Fig. 4K, compare with 4D). Finally, we check whether excess of Tay can modulate dP-Erk accumulation under strong conditions of constitutive pathway activation. We find that Tay over-expression reduces the levels of dP-Erk induced by RasV12 in the central region of the wing disc (Fig. 4L′, M′), and also the phenotype of ectopic veins caused by RasV12 (Fig. 4L, M), suggesting that the negative effect of Tay on the activity of the EGFR pathway occurs downstream of Ras activation and affects the accumulation of dP-Erk. The effects of Tay loss and gain on dP-Erk accumulation were also detected in other imaginal discs, such as the eye disc (not shown) and the leg disc (Fig. S3C–D′), and also in embryos mutant for tay (Fig. S3A–B′), suggesting that Tay functions as a general modulator of Erk phosphorylation.
The preferential nuclear localization of Tay and its effects on EGFR signalling and Erk phosphorylation prompted us to study the interactions between Tay and EGFR pathway components which subcellular localization shifts between the nucleus and the cytoplasm. We focussed this analysis on Erk and its specific phosphatase Mkp3. These proteins can interact with each other in the cytoplasm, where Mkp3 retains ERK and prevents its phosphorylation, and also in the nucleus, where Mkp3 de-phosphorylates and inactivates Erk [8], [49]. In addition, the phenotypes caused by the loss of Erk or Mkp3 are very similar to those cause by tay over-expression or loss of function, respectively. To study the genetic interactions between Tay and Erk we over-expressed wild type Erk or its mutant form sevenmaker (Erksem), which bears a single amino acid substitution preventing Erk interactions with Mkp3 [50]–[51]. The use of Erksem allows the analysis of Erk over-expression conditions in the absence of its interaction with Mkp3. The formation of ectopic veins caused by a reduction in Tay levels is only weakly increased when the normal form of Erk is over-expressed (Fig. 5D–F). In contrast, loss of tay in a background of Erksem over-expression causes a strong increase in the differentiation of extra-vein tissue (Fig. 5H), compared with loss of only tay (Fig. 5D) or with Erksem over-expression (Fig. 5G). Interestingly, Tay over-expression reduces, but does not suppress, the ectopic veins caused by Erksem (Fig. 5A–B). These results suggest that Erksem is much more effective when Tay levels are reduced, and, conversely, that Tay is less effective antagonizing Erk when this protein cannot interact with Mkp3.
In the case of Mkp3, the loss of veins caused by its over-expression (Fig. 5I) is not modified by loss (not shown) or excess of tay (Fig. 5J), confirming that Tay levels are not relevant upon a strong loss of Erk activation. In contrast, the formation of extra veins observed in tay loss-of-function conditions (Fig. 5K, M) depends on the gene dosage of Mkp3, becoming stronger in Mkp3M76-R2b heterozygous flies (Fig. 5N, compare with M) or upon expression of Mkp3-RNAi (Fig. 5L, compare with K). One possible explanation for these interactions is that Tay participates in the regulation of Erk inactivation, perhaps by promoting its de-phosphorylation. This possibility is compatible with the strong reduction of Erk phosphorylation caused by Tay over-expression, and implies that Tay over-expression phenotypes should be dependent on the presence and activity of Erk phosphatases such as Mkp3. However, we notice that the phenotype of Tay over-expression is not modified in Mkp3 null mutant backgrounds (Fig. 5O–Q). Thus, although we cannot exclude a role of Mkp3 in Tay function, this result indicates that the effects of Tay over-expression are not mediated exclusively by the activity of Mkp3.
Next, we wanted to visualize the activation of Erk in genetic backgrounds where the level of Erk and Tay expression is changed and the activity of the EGFR pathway is increased. To this end, we made tagged forms of Tay (Tay-Flag), Erk (Erk-HA) and Erksem (Erksem-HA) and studied the accumulation of dP-Erk in wing discs of different genotypes. The expression of Erk-HA and Erksem-HA causes very weak (Erk-HA; Fig. 5E) or moderate (Erksem-HA; Fig. 5G and 6I) extra veins. In none of these over-expression backgrounds we were able to detect changes in the pattern or level of dP-Erk accumulation (Fig. 6A and 6E). The reduction of Erk phosphorylation caused by Tay over-expression (Fig. 4) is still observed when either Erk-HA (Fig. 6B) or Erksem-HA (Fig. 6F) is expressed in combination with Tay. The strong activation of the pathway caused by RasV12 is also observed in backgrounds of Erk-HA or Erksem-HA expression (Fig. 6C and G, respectively). The introduction of Tay in these backgrounds causes a moderate reduction in dP-Erk accumulation (Fig. 6D and H, compare with 6C and G), although the resulting phenotype of ectopic vein differentiation is not reduced (Fig. 6K–L). From these observations we conclude that Tay is still effective in promoting the de-phosphorylation of Erk under conditions of Erk and Erksem over-expression, but less so in backgrounds of strong pathway activation.
The subcellular localization of Mkp3 and Erk is dynamic, shifting between the nucleus and the cytoplasm [8], [49]. We wanted to analyse whether Tay influences the accumulation of these proteins in wing imaginal cells in over-expression conditions. First, we confirmed that Mkp3-Myc is preferentially localised in the cytoplasm (Fig. S4A–A′″), and that both Erk-HA and Erksem-HA are detected in the nucleus and in the cytoplasm, with Erk-HA distributed at higher levels in the cytoplasm (Fig. 7A and E, respectively and Fig. S4C–C″ and E–E′″). The co-expression of Mkp3-Myc and Tay-Flag does not modify the preferential cytoplasmic (Mkp3) or nuclear (Tay) accumulation of these proteins (Fig. S4B–B′″). The co-expression of Mkp3-Myc and Erk-HA results in a clear cytoplasmic retention of Erk-HA (Fig. 7B, compare with A). In contrast, Mkp3-Myc does not modify the homogeneous nucleus-cytoplasm distribution of Erksem-HA (Fig. 7F, compare with E). Neither the localization of Tay-Flag or Erk-HA changes when both are co-expressed in the same cells of the central region of the wing disc (Fig. 7C and Fig. S4D–D′″). In addition, the expression of RasV12 does not affect the localization of Erk-HA, which is still localised in the nucleus and cytoplasm (Fig. 7D, compare with A, and Fig. S5A–A″).
In contrast, both Erksem-HA and Tay-Flag display a heterogeneous distribution when co-expressed (Fig. 7G, J–J″ and Fig. S4F–F′″). We took higher magnification pictures of sections taken from the most anterior region of the salEPv-Gal4 domain of expression, because in these cells the level of over-expression are lower and Tay retains its nuclear localization (Fig. S7). We observed that the nuclear level of Erksem-HA and Tay in each cell are not correlated (r2 = 0.09; n = 60). A similar heterogeneous distribution of ERKsem was observed in a RasV12 background (Fig. 7H and Fig. S5C–C″), and also when both Tay-Flag and Erksem-HA were co-expressed in a Rasv12 background (Fig. S5D–D″). We do not understand the molecular bases for these changes in Erksem and Tay accumulation in the presence of each other or upon strong pathway activation, but they might be related to a dynamic regulation of protein turnover when Tay and Erk are co-expressed at higher levels. To get a quantitative view of Erksem nuclear-cytoplasmic localization, we took serial sections of the wing disc, quantified the levels of Erksem in the cytoplasm (apical in the epithelium; Fig. 7O) and nucleus (medial in the epithelium; Fig. 7O), and calculated the average cytoplasm/nucleus ratio of Erksem signal in different genetic backgrounds (Fig. 7K–O). These measures show that Erksem is mostly localised apically in the cell (cytoplasm), and that both the presence of Rasv12 (Fig. 7K) or Tay-Flag (Fig. 7L) strongly reduce the amount of cytoplasmic Erksem and weakly increase the level of nuclear Erksem (Fig. 7N). In this manner, the expression of either RasV12 or Tay changes Erksem localization in a similar manner, but although both Tay and RasV12 reduce the cytoplasm/nucleus ratio of Erksem accumulation, Erk activation, as visualised by the presence of dP-Erk (see Fig. 6), only occurs in RasV12 conditions.
We next considered the possibility that Tay might be directly interacting with Erk or Mkp3 in co-immunoprecipitation and pull-down experiments. Co-immunoprecipitation experiments were carried out from protein extracts obtained from embryos expressing combinations of Tay.FL-Flag, Mkp3-Myc, Erksem-HA and Erk-HA (see Fig. S2D–E). Tay.FL-Flag was never detected in western blots, perhaps because the size of the protein prevents its transference to the membrane. However, when Tay-Flag is co-expressed with Mkp3-Myc or Erksem-HA, we detected co-immunoprecipitation when the IP was made using anti-Flag and the western blot revealed using anti-Myc (Fig. 8A, line T+M from IP lanes) or anti-HA (Fig. 8B, line T+E from IP lanes). In protein extracts from embryos expressing only Mkp3-Myc or Erksem-HA and IP with anti-Flag, we never detected Myc or HA (Fig. 8A–B, lines M and E, from IP lanes, respectively). The interaction between Tay and Erk and between Tay and Mkp3 might be direct, because they were also observed in pull-down experiments using in vitro translated Tay incubated with Erk-GST and Mkp3-GST fusion proteins (Fig. 8C).
To identify the region of Tay involved in these interactions, we made several truncated forms of the protein (Fig. 8E and data not shown), and expressed them in the wing disc. We found that the 1292 amino acid N-terminal fragment of Tay (Tay.1) is located exclusively in the cytoplasm (Fig. 8D′), and its over-expression does not affect the differentiation of veins (Fig. 8D′). In contrast, the 1030 amino acid C-terminal fragment (Tay.2) is accumulated preferentially in the nucleus (Fig. 8D″), similar to the full-length Tay-Flag protein (Fig. 8D). Interestingly, the expression of Tay.2 consistently results in stronger phenotypes of vein loss and reduced wing size than those caused by the over-expression of the full-length protein (Fig. 8D″ compare with D). This C-terminal fragment includes the domain of homology detected between Tay and human AUTS2. The distribution of Erksem is not modified in the presence of the N-terminal portion of Tay (data not shown). In contrast, Tay.2 results in the same changes in the cytoplasm/nucleus ratio of Erksem accumulation as Tay.FL (Fig. 7N–O).
The C-terminal 1030 amino acid Tay fragment (Tay.2) contains all the information necessary to regulate the subcellular localization of the protein, and also all the domains necessary to reproduce the effects of the full-length protein (see above). We repeated the immunoprecipitation experiments using this fragment, and found that Tay.2 retains its interaction with Erksem (Fig. 8G, line T+E, IP lanes), but loses its ability to interact with Mkp3 (Fig. 8F, line T+M, IP lanes). The failure of Tay.2 to interact with Mkp3 might increase the titration of ERK by Tay.2, explaining why Tay.2 interferes with EGFR signalling more efficiently than the full-length protein.
We also found that the levels of Tay accumulation in the nucleus are much higher than normal in cells over-expressing Erk or Erksem (Fig. 8H–H′ and Fig. S7A–D). As Erk or Erksem over-expression do not change the expression of tay (not show), these observations indicate that Erk increases the stability of Tay in the nucleus. This effect is independent of EGFR signalling, as neither RasV12 nor Mkp3 over-expression modified the accumulation of endogenous (Fig. 8I–I′) or over-expressed Tay (Fig. S6A–B″). We conclude from these data that Tay can interact with Erk in the nucleus and that Erk protects Tay from degradation.
Drosophila Tay and human AUTS2 are very different proteins in sequence and length, but they share a small 250 amino acid stretch with significant homology (Fig. 9A). Our deletion analysis of Tay indicates that this region is included in the smaller fragment of Tay that we found has biological activity and nuclear localization (C.M. and J.F.dC., unpublished results). We wanted to check whether AUTS2 expressed in flies was able to reproduce some of the effects observed in Tay over-expression conditions. A Flag-tagged form of AUTS2 expressed in the wing disc is localised exclusively in the nuclei (Fig. 9B–C), the same as Tay. Interestingly, the expression of AUTS2 in the wing leads to a phenotype of ectopic vein formation reminiscent to the consequence of Tay loss (Fig. 9D). The extra veins that develop in AUTS2 over-expression conditions depend on EGFR signalling, because they are eliminated when the expression of Erk is reduced (Fig. 9E–F). AUTS2 also enhances the formation of extra veins caused by the expression of Erksem (Fig. 9H–I), and causes an increase in the levels of activated Erk (ap-Gal4/UAS-hAUTS2-Flag; Fig. 9J–J′). These data suggest that AUTS2 is able to interact with some, but not all, targets of Tay, and raise the possibility that AUTS2 normal function in humans is related to the regulation of the Erk signalling pathway, albeit in an opposite manner as Tay.
Signalling by Erk in response to growth factors regulates growth, differentiation and survival of cells in a variety of developmental contexts [7], [52]–[53]. The extent and level of Erk activation relies on its phosphorylation state, which in turns regulates Erk subcellular localization and interactions with downstream effectors and other proteins [7], [10], [54]–[55]. Erk activation is transient, and failures in the mechanisms responsible for its inactivation can drive developmental defects and oncogenic transformations [10]. In this work we identified Tay as a novel nuclear component that interacts with Erk and is involved in the maintenance of appropriate levels of Erk activity.
We have addressed the requirements and function of tay mostly in the wing disc, a convenient developmental system to analyse the contribution of signalling pathways to the regulation of organ size and pattern formation [56]. Tay was previously described as a protein that regulates locomotion and other neural aspects [38]–[39]. We have observed that changes in the level of EGFR signalling in the nervous system also cause locomotion defects (Molnar and de Celis, in preparation), which is indicative of a role of Tay in the regulation of EGFR signalling also in the nervous system. In the context of wing development and vein differentiation, the loss of tay results in the differentiation of extra veins in inter-vein territories. This phenotype is very similar to those obtained in conditions of excess of EGFR signalling, suggesting that Tay negatively regulates the activity or the response to this pathway. In addition, loss of tay also causes a reduction in the size of the wing blade, a phenotype that is not expected in a situation of excess of EGFR/ERK activity. This last result suggests that Tay might also have functions independent of its role in the regulation of EGFR signalling. The consequences of gain of Tay expression mostly indicate that the role of Tay is related to the modulation of EGFR signalling. Thus, excess of Tay expression in different imaginal discs results in phenotypes that can be attributed to loss of EGFR signalling, such as loss of veins and bristles [33], wing size reduction and failures in tarsal joint formation [57] and ommatidial differentiation (data not shown).
We further explore the relationships between Tay and EGFR signalling in genetic combinations in which the activity of the pathway is altered in backgrounds with modified levels of Tay expression. In all cases, we observed synergistic interactions between loss of tay and excess of EGFR, and between excess of tay and loss of EGFR activity. Furthermore, we notice that the extra veins differentiating in tay mutants require EGFR function, suggesting that Tay modulates EGFR signalling during vein formation. All together, the results of genetic combinations indicate that cells with lower levels of Tay become more sensitive to an increase in EGFR signalling, and that Tay over-expression prevents cells to acquire the level of EGFR signalling required for vein formation.
The negative effect of Tay on EGFR signalling is more directly visualised by considering the effects of Tay in Erk phosphorylation and in the expression of the EGFR/Erk targets genes Dl and argos. Thus, Tay over-expression strongly suppresses Erk phosphorylation and prevents the expression of Dl and argos in the developing veins. Conversely, in loss of tay conditions we detect an increase in the levels of phosphorylated Erk, which is accompanied by a moderate ectopic expression of argos. The extra-vein phenotype of loss of tay is not as extreme as the massive vein differentiation that occurs upon strong and constitutive activation of the EGFR pathway. In fact, tay mutant wings differentiate a similar pattern of extra veins as moderate increases in EGFR signalling caused by, for example, mutations in the Mkp3 gene [58]. This suggest us that Tay primary function is to prevent increases in EGFR/Erk signalling in places where the pathway must be active but only at low levels. Thus, high levels of EGFR activity and dP-Erk accumulation are restricted to the presumptive veins in wild type third instar wing discs, but the pathway is also active at lower levels in the inter-veins, where it promotes cell proliferation and survival [28]. In tay or Mkp3 mutant backgrounds, a fraction of these cells initiates the vein differentiation program, escaping the negative feed-back loops that maintain low dP-Erk levels and entering the positive feed-back loops that normally operate in vein territories through the regulation of rhomboid expression [59]. In this model, Tay would participate in a mechanism that favours Erk de-phosphorylation and its nuclear retention in an inactive form. This mechanism of Tay action is compatible with the effects of its over-expression, which essentially cause a failure to accumulate dP-Erk in vein territories, and consequently a loss of vein differentiation.
Signalling by Erk proteins in the nucleus is in part regulated by the rate of Erk nucleus/cytoplasm shuttling [60]. In the nucleus, signal termination involves Erk de-phosphorylation by nuclear phosphatases and also its sequestration away from cytoplasmic Erk kinases [61]. Because Erk does not contain nuclear localization nor export sequences, its subcellular localization relies on proteins acting as anchors [8]. We observed direct interactions between Tay and Erk and between Tay and Mkp3, and these interactions were also detected in immunoprecipitation experiments from embryo protein extracts. These data suggests that Tay could form part of protein complexes including both Erk and Mkp3 in the nucleus.
A direct interaction between Tay and Erk is also compatible with several observations regarding Tay stability and Erk subcellular localization. First, Erk and Erksem increase the accumulation of Tay in the nucleus, and do so independently of EGFR signalling, as neither RasV12 nor Mkp3 over-expression modified Tay accumulation. Second, Tay over-expression prevents the accumulation of dP-Erk, whereas loss of Tay has the converse effect. Finally, Tay over-expression modifies Erksem subcellular localization, increasing the nucleus/cytoplasm ratio of Erksem accumulation. In this regard, it is worth noting that the expression of RasV12 has the same effects on Erksem subcellular localization as the over-expression of Tay, as both Tay and RasV12 increase the nuclear/cytoplasm ratio of Erksem accumulation. We notice that the effects of Tay on Erk localization are only manifest when we used the Erksem form. Because we also see that Erksem is not retained in the cytoplasm by Mkp3, we reason that Erksem, liberated of cytoplasmic anchorage by Mkp3, is more sensitive to pathway activation and to the presence of other anchoring proteins, and that Tay might play this role in the nucleus.
We also observed a direct interaction between Tay and Mkp3. Mkp3 is a dual-specificity phosphatase that is predominantly localised in the cytoplasm, but it shuttles between the nucleus and cytoplasm and could play a role in translocating inactive Erk from the nucleus to the cytoplasm [8]. It is possible that Tay could promote the nuclear function of Mkp3, but in addition, Tay should also act independently of Mkp3 to promote Erk inactivation and retention, because Tay is able to down-regulate Erk activity in Mkp3 mutant backgrounds.
Most of the Tay interacting region with Erk is localised to the C-terminal part of Tay, a 1000 amino acid long region that includes the domain of homology between Tay and human AUTS2. This fragment of Tay fails to interact with Mkp3, and is even more efficient than the full-length protein in its effects on Erk subcellular localization and in its antagonism on Erk signalling. Intriguingly, AUTS2 expressed in the wing disc also interferes with EGFR signalling, but it does so in an opposite manner to Tay or to the Tay C-terminal domain. We cannot extract many conclusions from the consequences of AUTS2 expression in the wing disc, but speculate that this protein retains some of its interactions with Drosophila Erk that might protect this protein from inactivation by nuclear phosphatases. Similarly, the effects of AUTS2 on Drosophila EGFR signalling are compatible with a role for this protein in the regulation of Erk activity in humans, and that this effects might underline the effects of zebrafish, murine and human mutations in the onset of neurological disorders.
From the analysis in the wing disc we conclude that Tay interacts with Erk in the nucleus, affecting its phosphorylation and promoting its nuclear retention. In this context, it is interesting to note that the free diffusion of human ERK2 is impeded within the nucleus, and that this limitation in mobility increases after ERK2 stimulation [6]. This has lead to postulate that ERK2 retention in the nucleus involves high-affinity interactions with unidentified low-mobility sites that are constitutively expressed [6]. We suggest that Tay could play such a role in vivo, acting as a nuclear anchor for Erk that facilitates its inactivation by nuclear phosphatases and its retention in an inactive state.
We used the Mkp3 allele Mkp3M76-R2b [58], and the deficiencies EP-866rev34, EP-866rev40 and Df(1)tay (see below). We used the following Gal4 lines: shv3kpn-Gal4 [62], 638-Gal4, nub-Gal4, salEPv-Gal4 [63], ap-Gal4, hh-Gal4, bs-Gal4, 1348-Gal4, dll-Gal4, eye-Gal4 and da-Gal4 [64]. We also used the UAS lines: UAS-RasV12 [65], UAS-Rafact [66], UAS-Erksem [67], UAS-Erk-HA [55], UAS-Erksem-HA [55], UAS-rhomboid [47], UAS-EGFR, UAS-EGFRDN [68] and UAS-GFP [69] and the P-GS lines EP-M76 and EP-866 [37]. We generated the following UAS lines: UAS-tay-i, UAS-tay.FL-Flag, UAS-tay.1-Flag, UAS-tay.2-Flag, UAS-hAUTS2-Flag, and UAS-Mkp3-Myc. We also used the RNA interference lines UAS-Mkp3-i (23911), UAS-tay-i (29021) and UAS-rolled-i (35641) from the VDRC Stock Center, and the lines UAS-EGFR-i (10079R-2) and UAS-ras-i (9375R-1) from NIG-Fly.
Df(1)tay: We used the insertions e03798 and d06351 [70], which are separated by 15 Kb of DNA including tay and part of CG16952. Flipase (FLP)-induced recombination was induced by a daily 1 h heat shock at 37°C to the progeny of e03798/d06351; hsFLP/+ females and FM7 males. Ten putative e03798-d06351/FM7 offspring females were individually crossed to FM7 males, and after 3 days, were used to extract genomic DNA to determinate by PCR the existence of FLP recombination. The position of the flanking insertions e03798 and d06351 and the extent of the tay deficiency are described in Suppl. Fig. S2A.
EP-866rev40 and EP-866rev34: We used Δ2–3 as a source of transposase to mobilize the EP-866 P-GS element. Males carrying both EP-866 and Δ2–3 were crossed with N55e11/FM7c females. The offspring EP-866 males with white phenotype were selected to make individual stocks. A complementation test was done to analyse the behaviour of these new alleles.
Fifty wild type (control) and homozygous EP-866rev40 and EP-866rev34 embryos were used to extract genomic DNA to identify by PCR the genomic region excised by the mobilization of the P-GS. We used the following primers: 5′GCCGTGGAAATGGACTCTG3′ and 5′TTGCTGCTGCTGGTGAAAT3′. The size of the amplified fragments was 3629pb in wild type embryos, 2373pb in EP-866rev34 embryos and 932pb in EP-866rev40 embryos. The size of the generated deficiencies was confirmed by sequencing the PCR fragments sub-cloned in the pGEM-T-Easy vector (Promega) confirming an excision of 1276pb in EP-866rev34 and of 2717pb in EP-866rev40.
Homozygous Df(1)tay, EP-866rev40 and EP-866rev34 clones were induced in larvae of the following genotypes: Df(1)tay f36a FRT18A/FRT18A UbiGFP; hsFLP/+; EP-866rev40 f36a FRT18A/FRT18A UbiGFP; hsFLP/+ and EP-866rev34 f36a FRT18A/FRT18A UbiGFP; hsFLP/+, respectively. Homozygous tay mutant cells were recognized in the adult wing by the cellular marker forked (f) and in the wing disc by the absence of GFP.
We used the rabbit antibodies: anti-phospho-Histone3, anti-activated Cas3 and anti-diphosphorylated ERK1&2 (Cell Signalling). We also use the mouse monoclonal antibodies: anti-c-Myc 9E10 (Santa Cruz Biotechnology), anti-HA 12CA5 (Sigma), anti-FlagM2 (Sigma), anti-βGal (Promega), and anti-FasIII, anti-Dl and anti-Arm from the Hybridoma Bank at University of Iowa (Iowa City, IA). Alexa Fluor secondary antibodies (used at 1∶200 dilution) were from Invitrogen. To stain the nuclei we used To-Pro and to stain F-actin we used Alexa Fluor Phalloidin, from Invitrogen. Imaginal wing discs were dissected, fixed, and stained as described in [72]. Confocal images were taken in a LSM510 confocal microscope (Zeiss). In situ hybridization with the tay probe were carried out as described [72]. We used the cDNA LD22609 as template to synthesize the tay probe. The quantification of Erksem nuclear and cytoplasmic staining was carried out in Z-sections taken from 6 proximo-distal planes of 6 discs of each genotype along the length of the epithelium with the program ImageJ.
The fusion proteins Mkp3-GST and Erk-GST and the GST protein (negative control) were expressed in E. coli BL21 (DE3), using the constructs pGEX2TK-DMkp3 and pGEX4T1-DErk [73] and the vector pGEX2T, respectively, and were purified using Glutathione Sepharose 4B (GE Healthcare). The complete Tay protein was generated from the cDNA LD22609 using the TNT T7 Coupled Reticulocyte Lysate System (Promega) and radiolabeled with S35-Met. The pull-down assay was performed incubating over-night at 4°C the same amount of GST or GST fusion proteins bound to Glutathione Sepharose4B with in vitro translated Tay. After centrifugation and washes the proteins were resolved by 6% SDS/PAGE and the existence of pull-down proteins was analysed by autoradiography. The pulldown experiments were repeated five times with the same results.
|
10.1371/journal.pntd.0002563 | Burden of Mycobacterium ulcerans Disease (Buruli Ulcer) and the Underreporting Ratio in the Territory of Songololo, Democratic Republic of Congo | Cutaneous infection by Mycobacterium ulcerans, also known as Buruli ulcer (BU), represents the third most common mycobacterial disease in the world after tuberculosis and leprosy. Data on the burden of BU disease in the Democratic Republic of Congo are scanty. This study aimed to estimate the prevalence rate and the distribution of BU in the Songololo Territory, and to assess the coverage of the existing hospital-based reporting system.
We conducted a cross-sectional survey (July–August 2008) using the door-to-door method simultaneously in the two rural health zones (RHZ) of the Songololo Territory (RHZ of Kimpese and Nsona-Mpangu), each containing twenty health areas. Cases were defined clinically as active BU and inactive BU in accordance with WHO-case definitions.
We detected 775 BU patients (259 active and 516 inactive) in a total population of 237,418 inhabitants. The overall prevalence of BU in Songololo Territory was 3.3/1000 inhabitants, varying from 0 to 27.5/1000 between health areas. Of the 259 patients with active BU, 18 (7%) had been reported in the hospital-based reporting system at Kimpese in the 6–8 months prior to the survey.
The survey demonstrated a huge variation of prevalence between health areas in Songololo Territory and gross underreporting of BU cases in the hospital-based reporting system. Data obtained may contribute to better targeted and improved BU control interventions, and serve as a baseline for future assessments of the control program.
| Buruli ulcer (BU) is a necrotizing bacterial infection of skin, subcutaneous tissue and bone, caused by an environmental pathogen, Mycobacterium ulcerans. BU is considered as one of the Neglected Tropical Diseases with a poorly known global prevalence, and mainly affects remote rural African communities. Data on the burden of BU disease in the Democratic Republic of Congo (DRC) are scanty. The present study is the first exhaustive survey in DRC on the frequency of BU in the community. The survey demonstrated large variations in prevalence between health areas in Songololo Territory. Moreover, our data showed that the BU cases in the hospital-based reporting system reflect only the tip of the iceberg of the true active BU prevalence. Indeed, only one in thirteen active BU cases was notified at the hospital at Kimpese in the 6–8 months prior to the survey. The present data will serve as a baseline assessment for the evaluation of control interventions in the study area, and, more generally, this study aims to raise awareness about the issue of underdetection of BU and the importance of increasing access to diagnosis and care. As such, we hope the study will contribute to improved control of BU.
| Cutaneous infection by Mycobacterium ulcerans, also known as Buruli ulcer (BU), represents the third most common mycobacterial disease in the world after tuberculosis and leprosy [1]. In Africa, children under 15 years old have the highest incidence, but healthy persons of all ages, races, and socioeconomic classes are susceptible [2], [3]. Rates of infection among males and females are equal [3]. BU most affects the extremities [2], [4], and is diagnosed in the majority of patients at the ulcerative stage [5]. The disease has a scattered focal distribution within endemic regions, which impedes accurate estimation of disease burden [5], [6].
BU is considered as one of the Neglected Tropical Diseases (NTDs) with a poorly known global prevalence [7], and mainly affects remote rural African communities [8]. A recent review on prevalence [9] reported that, of the estimated 7,000 cases of BU reported annually worldwide, more than 4,000 cases occur in Sub-Saharan Africa. The largest numbers of reported BU cases are from the West African countries of Côte d'Ivoire (about 2,000 cases annually), Benin and Ghana, each reporting about 1,000 cases a year [3]. Various prevalence rates (Table 1) have been reported from different endemic regions in Sub-Saharan Africa [6], [10]–[13].
In the Democratic Republic of Congo (DRC), more than 500 BU cases had been reported before 1980 [14]. The first BU case reports in the Province of Bas-Congo were published in the 1960s and 1970s [15]–[17]. However, in-depth interviews of former patients conducted in the Bas-Congo by Meyers et al. strongly supported the concept that BU was an ancient disease in that region [14]. After 1980, there was a silent period of 20 years without any cases reported in the scientific literature [14]. A national hospital-based survey conducted in 2004 identified 487 clinically suspected cases of BU from six provinces [18]. Between 2002–2004, an apparent resurgence of BU was reported in Songololo Territory [4], known to be the main focus of BU in the country [17]. Since the end of 2004, the General Reference Hospital (GRH) of the Institut Médical Evangélique (IME) of Kimpese launched a specialized BU program sponsored by American Leprosy Missions, offering in-patient treatment free-of-charge and supplementary aid. A recent study has shown a strong increase in the number of admitted BU cases at the IME Hospital after the start of the BU Control Project [19]. Although the number of BU cases admitted in the hospital was rising, data on the exact prevalence and the extent of the disease in the region was lacking. We set up a study to obtain relevant information for planning subsequent control activities, and to provide baseline data for future control program assessments. This study aimed (i) to assess the prevalence and the geographic distribution of BU, (ii) to determine the epidemiologic characteristics of BU, and (iii) to determine the project coverage in Songololo Territory, the target endemic region of the project.
The Congolese Ministry of Health granted approval to conduct the survey. We obtained ethical clearance for this study from the Institutional Review Board of IME (N° IME/CS/01/2008). All patients, or their guardian in the case of minors, provided written informed consent for all diagnostic and treatment procedures and publication of any or all images derived from the management of the patient, including clinical photographs that might reveal patient identity. After informed consent had been given, data were recorded on a Community BU Form recommended by WHO. Patient care was free of charge.
The case search covered two rural health zones (RHZ), Kimpese and Nsona-Mpangu, both located in Songololo Territory (Figure 1), one of ten territories of Bas-Congo Province. It is situated in the District of Cataractes and covers an area of 8,190 Km2, approximately 15.2% of the total surface of the province, with a population of 237,418 inhabitants in 2008 (enumeration conducted on December 2007 by the Central Offices of the 2 RHZ). An average of 6 persons per household was used as a regional estimate, giving a total of 39,569 households to be visited by 80 community health workers (CHW).
Songololo Territory is limited in the north by the Congo River, in the west by Sekebanza Territory, in the east by Mbanza-Ngungu Territory and in the south by the northern border of Angola. Each RHZ is subdivided into 20 health areas (Table S1 & Table S2). The primary level of health care facilities includes the Rural Health Posts (HP), Health Centres (HC) and Reference Health Centres (RHC), and the secondary level is represented by the GRH.
We conducted a cross-sectional survey (July–August 2008) using the door-to-door method simultaneously in the two RHZ of the Songololo Territory (i.e., Kimpese and Nsona-Mpangu), each containing twenty health areas. Cases were defined clinically as active BU and inactive (healed) BU in accordance with WHO-case definitions [20]. We defined functional limitation as any reduction in the range of motion of one or more joints, and assessed it by clinical observation. Lesions were considered as mixed forms when the simultaneous presence of different forms of disease, including bone and joint involvement, in the same patient was noted. In addition, we defined as simple ulcerative forms (SUF) the ulcerative lesions not associated with other clinical lesions such as papule, nodule, plaque, edema or osteomyelitis at the same site. Lesions were categorized as follows: A single lesion <5 cm (Category I); a single lesion 5–15 cm (Category II); a single lesion >15 cm, multiple lesions, and lesions at critical sites (face, breast and genitalia) or osteomyelitis (Category III). The status of relapse was assessed by questioning the patients, or their guardian in the case of minors, on the history of the lesion, and defined as the reappearance of an ulcer or another form of the disease at the original site of the lesion or elsewhere during the 12 months that followed the end of the previous treatment (antibiotics and/or surgery).
This study was conducted in two phases: a preparatory phase and an investigation phase. During the preparatory four-week phase (June 2008), the purpose of the study was explained to the local political and health authorities, and their approval was obtained. Then, 80 CHW, i.e., 40 per RHZ, were trained in the use of the survey tools (BU community form, pictorial document to recognize BU) and in the identification of suspected BU cases in their communities. We also trained six physicians (working in the RHC of both RHZ), two nurse-supervisors of the leprosy and tuberculosis program (LT), and 40 head nurses (in charge of peripheral health areas), in active case-finding of BU cases and in the use of the survey tools.
For the survey, each RHZ was provided with 1 motor bike, 1 Global Position System device, 4 digital photo cameras, 30 bicycles (at least 1 for each health area), 25 megaphones (at least 1 for each health area), drugs and required medical and laboratory consumables.
The investigation phase was divided in two periods. The first period (two to three weeks depending on health area) consisted of making an inventory of all BU-like cases by the CHW, using the door-to-door approach in all villages and in each section of two cities in Songololo Territory (Songololo city and Kimpese city). The recommendation to CHW was to visit 40 households per day. A pictorial document, showing different clinical manifestations of BU, was presented to the head of the household or his/her representative asking if any household members presented similar lesions. If the head of the household was not present, the household was revisited once. The second period (6 weeks) included the clinical validation of suspected BU cases by trained health professionals. The eight validation teams were each composed of two people: firstly, a team member of the BU Project (physician or nurse), or another physician, or a LT supervisor, and secondly one of the head nurses.
The diagnostic confirmation process of suspected cases involved the collection of swabs from ulcerative lesions and fine needle aspirates from non-ulcerative lesions, followed by laboratory analyses (bacteriology and/or molecular biology) according to WHO recommendations [20]. The initial direct smear examinations for acid-fast bacilli were made at the IME/Kimpese laboratory, followed by in vitro culture for M. ulcerans. Samples were sent in tubes to the “Institut National de Recherche Biomédicale” in Kinshasa, DRC, where PCR for the detection of M. ulcerans DNA was performed, according to WHO recommendations [20]. The external quality control was conducted by the Unit of Mycobacteriology of the Institute of Tropical Medicine in Antwerp, Belgium.
The study was carried out simultaneously in the different health areas of both RHZ. Data were recorded on a standardized Case Registry Form elaborated by WHO (BU02), entered into an Excel database (Microsoft Corporation, Redmond, WA) and analyzed with Epi-Info version 3.3.2 (Centers for Diseases Control and Prevention, Atlanta, GA). The Pearson chi-square test was used to compare proportions with a significance level set at 5%, and the Fisher's exact test when an expected cell value was less than 5. Coverage was calculated as the number of active cases detected who had visited the BU reference center in IME Hospital. We produced the distribution maps of BU in Songololo Territory using ArcGIS 9.2 (ESRI, Redlands, CA, USA).
The CHW visited a total of 39,044 households distributed across 9 sections of two cities (Kimpese and Songololo), 46 hamlets and camps, and 547 villages of the Songololo Territory. The estimated coverage of the study was 98.6%. During the household visits, the CHW inventoried 2,516 persons with BU-like lesions, among which 775 (30.8%) were validated in a second step as probable cases of BU, all forms included (i.e., 259 with active and 516 with inactive lesions). A total of 72 out of 241 (30%) patients with active lesions in whom a sample could be taken were confirmed by at least one positive laboratory test for M. ulcerans. The overall prevalence of BU (active and inactive) in Songololo Territory was 3.3/1000 inhabitants, varying from 0 to 27.5/1000 between health areas, while the prevalence of active BU was 1.1/1000 inhabitants with the minimum of 0.3/1000 when only active, laboratory confirmed BU, was considered. Table 2 shows the prevalence of different BU forms in both RHZ of Songololo Territory, and the distribution per health area is presented in Figures 2, 3, S1 and S2. The overall prevalence for the RHZ of Kimpese was 2.6 per 1000 inhabitants and could vary between health areas from 0.1 (Kimbanguiste) to 24.4 (Mukimbungu). The prevalence of BU active forms was 1 per 1000 inhabitants, varying between health areas from 0.1 (Kimbanguiste) to 5.7 (Mukimbungu). The health areas of Mukimbungu and Kasi, located in the North of the RHZ of Kimpese, are the most endemic, representing together 60% of the identified patients during the survey (Table S1).
Sixty percent of the identified patients in the RHZ of Nsona-Mpangu were from 3 health areas, Kisonga, Nkamuna, and Songololo (Table S2). The overall prevalence in this RHZ was 4.4 per 1000 inhabitants, varying from 0 (health areas Nduizi, Nkenge and Pala Bala) to 27.5 (Kisonga). The prevalence of active forms of BU was 1.3 per 1000 inhabitants, varying between health areas from 0 (Nduizi, Nkenge and Pala Bala) to 3.8 (Kisonga).
The age distribution of all cases ranged from 2 to 94 years (Median 27, Interquartile range (IQR) 14–44) with no significant differences between active and inactive cases. The supplementary tables provide the detailed information.
We observed a predominance of female gender (60%) among the recorded cases. Among the 259 patients with active lesions, no sex difference was observed, as 130 (50.2%) were female. The proportion of new cases was far higher (94%) than the relapses. The ages ranged from 2 to 94 years (Median 27 years; IQR 11–47 years), and the distributions in the two RHZ were similar. Among these 259 patients, 192 (74%) had ulcerative lesions and 62 (23.9%) were diagnosed with functional joint limitations. Lesions on the limbs were predominant, representing 90% of the sites of lesions. Regarding the patients' categorization, 48.8% were in category I, 31.5% category II, and 19.7% category III. The proportion of patients with ulcerative lesions was higher (p<0.001) in the RHZ of Kimpese (83%) compared to the RHZ Nsona-Mpangu (63.6%). Less than half of the patients of the RHZ of Kimpese (41.2%) and more than half (57.6%) in the RHZ of Nsona-Mpangu were in category I (p = 0.031) (Table S3).
Female patients predominated amongst active confirmed cases compared to unconfirmed cases; on the other hand, male patients were more frequent in active unconfirmed patients (p = 0.029). No differences in the age distribution were observed between active confirmed and unconfirmed patients. The lower limb locations were significantly more frequent amongst active unconfirmed patients (p<0.001). Upper limb sites predominated (p<0.001) amongst active confirmed patients (Table 3).
Features of active cases in the two RHZ were quite similar, with a few exceptions. The ulcerated forms (p<0.001) and functional limitations on diagnosis (p<0.001) predominated in the RHZ of Kimpese. Features of inactive cases in the two RHZ were similar but functional limitations were more often observed in the RHZ of Kimpese (p = 0.005) (Table S4).
Only 25 BU patients were admitted and notified at the General Hospital IME/Kimpese between January and August 2008, amongst which 18 were still under treatment for active BU during the survey. Thus, 93% of all active BU patients at the time of the community survey were not captured by the hospital-based reporting system, corresponding to a ratio of 1 reported case for approximately 13 unreported cases.
The present study is the first exhaustive population-based survey in DRC aiming to assess the prevalence and distribution of BU in a well-circumscribed endemic region. The survey demonstrated a huge variation in prevalence between health areas and gross underreporting of BU cases in Songololo Territory, compared with the ongoing hospital-based reporting system.
Case-definition during the survey was essentially clinical. Case validation was performed by physicians from the BU project and physicians working in the area, well-trained in BU diagnosis, assisted by either a nurse from the BU project or a LT-supervisor, with the nurse responsible for the health area. We are aware of the limitations of clinical diagnosis, which is dependent on the range of experience of health professionals. This may account for certain non-BU cases included in this study. In endemic regions, depending on the clinical stage of the disease, BU may be confused with many other conditions such as nodular onchocerciasis, cyst, lipoma, lymphadenitis, phagedenic tropical ulcer, pyomyositis, necrotizing fasciitis [20], [21], to name a few. Our study showed that 72 out of the 241 (30%) patients who were tested, were confirmed in the laboratory. The low confirmation rate is mostly due to the relatively high number (almost half) of the ulcers being in an advanced stage of healing. Likewise, the technical problems encountered by peripheral health professionals when sampling non-ulcerated lesions and wounds, where mixtures of traditional herbs had been applied, may have played a role. Nevertheless, lesions due to another etiology misclassified as BU cannot be excluded, as lower limb locations were significantly more frequent among active unconfirmed patients. Indeed, among 92 clinically suspected patients recruited from the RHZ of Nsona Mpangu, Kibadi et al. found 31 (33.7%) PCR negative patients and among them, 25 with histopathological features not compatible with BU (chronic inflammation and bacterial infections due to gram positive cocci) [22].
Despite these limitations, we suggest that our results reflect the endemicity of BU in Songololo Territory reasonably well. In fact, the areas previously established as most endemic were corroborated through this survey, as were the non- or hypoendemic areas [15]–[17], [4].
When considering only active lesions, no sex difference was observed, similar to findings in other studies [2], [11], [12], [23], [24], although our study showed a predominance of females among all cases detected (active and inactive), because among inactive cases, 64.9% (335/516) were females and only 35.1% (181/516) were males. Females predominated also among active confirmed BU cases. This preponderance may be due to time itself, or the fact that the population was predominantly female. When referring to the national census figures (July 2008 estimates), for a total population of 66,514,504 inhabitants, 50.3% were female and 49.7% male.
Among the 259 patients with active lesions, the majority (66%) were over age 15, similar to previous findings in the same area [19]. Ages observed in this survey were higher than found in other disease-endemic countries [2], [10], [12], [25]. The median ages for both RHZ were similar with the median age of 25 years found in Ghana [11], and relatively high when compared to the 15.5 years observed in Cameroon [13]. The predominant clinical presentation was an ulcerative lesion in 192 cases (74%). This is consistent with studies in Côte d'Ivoire [10] and Cameroon [12], [13], while the percentage of ulcerative lesions was lower in some other studies, for example, 48.5% in Ghana [11], approximately 50% in Benin from 1997 to 2001 [26] and 57.5% in 2004 in the same country [6]. Of the 259 active cases, 62 (23.9%) were diagnosed with joint functional limitations, similar to previous findings in the same area [19], and in other African endemic regions [6], [12]. The general finding of limbs being most affected was confirmed in this study [2], [11]–[13], [23], [24].
The results presented in Table 3 shows that nearly 50% of the BU patients had category I lesions. A similar observation was made in the District of Akonolinga, Cameroon [13]. Ambulatory treatment, based on antibiotic therapy in the primary health care facility, is indicated for this category of patients. Indeed, most category I and some category II lesions may heal completely with antibiotic treatment alone [3], [27]. The introduction of antibiotic therapy [28] has shifted the balance between surgical treatment, mainly limited to reference centers, and antibiotics administered at the most peripheral level of the health system [3].
The clinical presentation of BU was different in the two health zones (Table S3). The degree of functional limitation was significantly higher in patients in Kimpese and they had more often ulcerated lesions. We speculate that this difference is most likely due to differences in health seeking behavior, with higher patient delays in Kimpese, notwithstanding the fact that they were living at shorter distance from the IME hospital. In recent years, an influential religious sect has been a factor in the reluctance to seek medical care in the Kimpese area.
Although the number of BU patients admitted at the hospital has increased in recent years, the survey results have demonstrated that the coverage of the population at risk was still insufficient. Of the 259 patients with active BU, 18 (7%) had been reported in the hospital-based reporting system. Porten et al. reported a coverage of 16%, limited to the area close to the Akonolinga hospital in Cameroon, where Médecins Sans Frontières (MSF) opened a BU programme in 2002. The need for improved access to care in the high prevalence areas was emphasized [13]. In the same area, Grietens et al. found that despite the significant reduction in costs for medical care, hospital treatment for BU often remained financially and socially unaffordable for patients and their households, leading to the abandonment of biomedical treatment, or avoiding it altogether. They concluded in their study that from a socio-economic perspective, a decentralized treatment system may limit the impoverishment of households caused by a long hospitalization period [29]. We agree with this opinion because bringing treatment as close as possible to the communities will have a significant mitigating impact on the socio-economic repercussions of BU.
The survey demonstrated large variations in prevalence between health areas within an endemic health zone consistent with previous studies in other African BU-endemic regions [6], [12], [13].
Tables S1 and S2 show that in both RHZ, 60% of patients were respectively identified from 2 out of 20 health areas (Mukimbungu, Kasi) in the RHZ of Kimpese and 3 out of 20 health areas (Kisonga, Nkamuna, Songololo) in the RHZ of Nsona-Mpangu.
Therefore, priority in case detection should be given to the most endemic health areas. A close collaboration with the provincial Leprosy & Tuberculosis control officers may facilitate the integration of BU activities at the primary health care centers. In fact, the use of the same case-confirmation network or the organization of integrated supervisions would help to reduce the BU intervention costs.
Data obtained in this survey may contribute to better targeted and improved BU control interventions, and serve as a baseline for future assessments of the control program.
|
10.1371/journal.pgen.1000244 | Modifier Effects between Regulatory and Protein-Coding Variation | Genome-wide associations have shown a lot of promise in dissecting the genetics of complex traits in humans with single variants, yet a large fraction of the genetic effects is still unaccounted for. Analyzing genetic interactions between variants (epistasis) is one of the potential ways forward. We investigated the abundance and functional impact of a specific type of epistasis, namely the interaction between regulatory and protein-coding variants. Using genotype and gene expression data from the 210 unrelated individuals of the original four HapMap populations, we have explored the combined effects of regulatory and protein-coding single nucleotide polymorphisms (SNPs). We predict that about 18% (1,502 out of 8,233 nsSNPs) of protein-coding variants are differentially expressed among individuals and demonstrate that regulatory variants can modify the functional effect of a coding variant in cis. Furthermore, we show that such interactions in cis can affect the expression of downstream targets of the gene containing the protein-coding SNP. In this way, a cis interaction between regulatory and protein-coding variants has a trans impact on gene expression. Given the abundance of both types of variants in human populations, we propose that joint consideration of regulatory and protein-coding variants may reveal additional genetic effects underlying complex traits and disease and may shed light on causes of differential penetrance of known disease variants.
| The ultimate goal of genome-wide association studies (GWAS) is to explain the proportion of variation in a phenotypic trait that can be attributed to genetic factors. The past two years have seen a plethora of successes in this field, yet, for most traits, a large fraction of variation remains unexplained. Epistasis, or interaction between genetic variants, is a largely under-explored factor, which may shed some light in this area. We use the HapMap populations to investigate interactions between regulatory and protein-coding variants and their impact on gene expression. We show that if a specific protein-coding variant has a functional impact, this can be modified by a co-segregating regulatory variant (cis interaction). Furthermore, the authors demonstrate that such modification effects between variants at one locus may affect the expression of other genes in the cell in a trans manner. The aim of this article is to present a framework though which variation can be considered in the context of GWAS. Viewing variation from this underappreciated angle may, in some cases, provide an explanation for differential penetrance of complex disease traits, but also for non-replication of GWAS results that may arise as a consequence of such interactions.
| Most disease association studies to date attempt to link single genetic variants to a specific phenotype [1],[2],[3],[4]. The genetic interaction between variants, also called epistasis, results in a phenotypic effect that is conditional on the combined presence of two or more variants [5],[6]. The prevalence and biological significance of epistasis has always been an area of interest in the field of human genetics and quantitative genetics, but its contribution to phenotypic variation has remained obscure, largely because genetic interactions have proven difficult to test [7]. This difficulty arises primarily because it is unclear which variant combinations should be tested and under which model of epistasis. To date, most strategies that address the effects of epistasis in humans involve millions of agnostic pairwise tests and fall into two broad categories: exhaustive testing of interactions between all pairs of variants across the genome [8], or testing of interactions between all pairs of those variants that each have an independent main effect on the phenotype of interest [8],[9],[10]. It is not entirely clear whether improvements in statistical methods will be sufficient to address the problem of epistasis. Therefore, the development of realistic biological models of epistatic interactions may reduce the statistical cost of dealing with many comparisons and facilitate the development of such methodologies.
To date, such an approach has been most feasible in model organisms and for specific genes or biological pathways that have been well-characterised. One classic example is the Adh locus in Drosophila where a series of regulatory SNPs in complex linkage disequilibrium (LD) modify the effects of a protein-coding variant [11],[12]. The protein-coding variant affects the catalytic efficiency of the ADH protein, whereas the regulatory variants have an impact on protein concentration. Catalytic efficiency and protein levels affect the overall activity of ADH showing that large effects attributed to a single locus may arise as a consequence of multiple associated variants. More recent studies in Drosophila reveal epistatic effects between genes affecting traits such as ovariole number [13] and olfactory avoidance [14]. In cases where little is known about the genes sculpting a phenotype, addressing the possibility of epistasis becomes more challenging. The value of assessing the impact of genetic interactions is highlighted in a recent study interrogating cardiac dysfunction in Drosophila [15]. A major susceptibility locus for this trait has been detected, but the importance of examining the phenotype in different genetic backgrounds was highlighted as a means to detect variants contributing to the phenotype through interactions with the prime susceptibility locus. The extent of epistasis in a more global way has been demonstrated in yeast where experiments on gene expression revealed that interacting locus pairs are involved in the inheritance of over half of all transcripts[5]. Furthermore, a large proportion of the eQTLs attributable to interaction effects were not detected by single locus tests. This suggests that analysis of interaction effects in other systems is likely to uncover additional associations.
In humans, most documented cases of epistasis have been detected in instances where there are biological clues as to which genes should be tested for interaction. Epistasis between two multiple sclerosis (MS) associated HLA-DR alleles was demonstrated by Gregerson et al. [16] who showed that one allele modifies the T-cell response activated by a second allele, through activation-induced apoptosis contributing to a milder form of MS-like disease. Similarly, Oprea et al. [17] demonstrated that a specific modifier effect is protective against spinal muscular atrophy (SMA). SMA arises from a homozygous deletion of SMN1, but some deletion homozygotes escape the disease phenotype due to the modulating effects of expression of PLS2. With the explosion of successful genome-wide association studies over the past two years, the need to address epistasis in a systematic, genome-wide approach is becoming increasingly pressing. The case of MS clearly illustrates this: as with most complex disorders, MS has a polygenic heritable component characterised by underlying complex genetic architecture [18]. Association studies to date have met with modest success in identifying MS-causing genes, and a large proportion of phenotypic variation remains unexplained. The expectation is that this residual variation arises at least in part, as a consequence of gene-gene interactions. Finally, epistasis may mask and prevent replication of otherwise real genetic effects due to differential fixation of variants that modulate the primary disease variant and affect the degree of penetrance of certain disease alleles.
In this study we propose a biological framework that could be useful for global survey of epistatic (modifier) effects in humans, which avoids exhaustive testing of agnostic pairs and involves prioritisation of variants to be tested. Two types of functional variants are common throughout the human genome and are present at appreciable frequencies in populations: regulatory variants with an impact on the expression patterns and levels of genes [19],[20],[21],[22],[23] and protein-coding nucleotide variants affecting protein sequence [19],[24]. To date, the effects of these variants have been considered independently of each other. In this study we perform an evaluation of the joint effects of regulatory and protein-coding variants to genome-wide expression phenotypes in humans in order to highlight an underappreciated angle of functional variation.
Our model brings together quantitative and qualitative variation. A gene that has an identified cis regulatory variant is differentially expressed among individuals of a population where that variant is segregating [20],[23]. If this gene also contains coding variation, then, assuming that mRNA levels are indicative of mature protein levels, the resulting protein products will not only differ in quantity (expression level) but also in quality or type (amino acid sequence) among individuals. Furthermore, depending on the historical rate of recombination between the regulatory and the coding variants, different allelic combinations (haplotypes) can arise on the two homologous chromosomes in a population. Phasing, the arrangement of the alleles at each variant with respect to one another, can differ between individuals in the population (Figure 1) [25]. If this is the case, the epistasic effect arising from these two variant types can be explored under a specific and testable biological model.
Using this model as a main principle, we explored the degree to which protein-coding variants can be modulated by cis-acting regulatory variants in human populations. In a previous study [20] we identified a set of SNPs (minor allele frequency (MAF)≥0.05) implicated in regulation of activity of genes in EBV-transformed lymphoblastoid cell lines (LCLs) of the unrelated individuals of the HapMap populations [26],[27] (60 Caucasians of Northern and Western European origin (CEU), 45 unrelated Chinese individuals from Beijing University (CHB), 45 unrelated Japanese individuals from Tokyo (JPT), and 60 Yoruba from Ibadan, Nigeria (YRI)). LCLs represent one particular cell type and even though there may be some effect arising from EBV transformation, it has been demonstrated that genetic effects on gene expression , such as the ones we describe below, are readily identifiable and mappable, and replicate in independent population samples. We henceforth refer to genetic variants associated with gene expression levels as candidate regulatory SNPs (rSNPs) and regard them as proxies for the linked functional variants that drive differential expression levels of nearby genes. The protein-coding variants considered under this model are non-synonymous SNPs (nsSNPs), i.e. variants that give rise to an amino acid substitution in the protein product. nsSNPs harboured in genes with varying expression levels are hereon termed differentially expressed (DE).
Two strategies were applied to detect DE nsSNPs in the HapMap populations. The first strategy involved scanning genes with known rSNPs [20] for nsSNPs. The aim was to identify nsSNPs that are predicted to be DE as a consequence of a nearby regulatory variant tagged by an identified rSNP. From the 606, 634, 679 and 742 genes with rSNPs previously identified [20] (0.01 permutation threshold and estimated false discovery rate (FDR) = 20%) in the CEU, CHB, JPT and YRI respectively, we found 159, 168, 180 and 202 of these genes (union of 484) containing 286, 304, 311 and 393 nsSNPs respectively (union of 909) (Table S1). We infer that these nsSNPs are DE as they reside in genes with experimentally-derived varying expression levels. This means that there are allelic effects on gene expression such that, depending on the genotypes of the rSNP and nsSNP and on the phasing of their alleles, one can make predictions about the relative abundance of the two alleles of a transcript in the cell.
The second strategy for DE nsSNP discovery involved direct association testing between nsSNP genotype and expression levels of the gene in which the nsSNP resides. We performed the test for each expressed gene harbouring at least one nsSNP. This strategy aimed to identify DE nsSNPs that are in LD with a regulatory variant. Depending on the strength of the regulatory effect, such variants may or may have not been detected in our initial scan for rSNPs. Relative distances between rSNPs and nsSNPs can vary, but in the special case where this distance is short in genetic terms, the two variants may be in LD [25]. Under these circumstances we expect that the nsSNP itself will demonstrate some degree of association with expression levels of the gene in which it resides. We used standard methodologies described in Stranger et al. 2007 [20] (see Methods) to test for genotype-expression associations in each population and in three multiple-population sample panels: (a) all four HapMap populations, (b) three populations (CEU-CHB-JPT), and (c) two populations (CHB-JPT). The choice of these panels represents a pooling strategy by which we sequentially remove individuals of the most genetically distant population sample.
For the single-populations analysis, with significance evaluated at the 0.01 significance threshold as determined by 10,000 permutations, we expect 56 nsSNPs and 34 genes to have at least one significant association by chance. We detected 242, 276, 267 and 255 nsSNPs (union of 703; estimated FDR ∼21%) with significant associations to expression levels of the gene in which they reside for the CEU, CHB, JPT and YRI populations respectively. These associated nsSNPs correspond to 196, 226, 210 and 211 genes (union of 560; estimated FDR ∼16%) (Table 1). For the multiple-population analysis at the same significance threshold (using conditional permutations that account for population differentiation–see Methods), we detected 345, 362 and 417 nsSNPs (estimated FDR ∼15%) for the four, three and two population groups respectively, corresponding to 284, 296 and 320 significant genes (estimated FDR ∼11%) (Table 1). Overall, the multiple-population analysis yielded a total of 587 nsSNPs with significant associations, corresponding to 461 genes. Taken together, the association analyses indicate that 884 nsSNPs (688 genes) are associated with expression levels of the genes they are in, suggesting that they are in LD with regulatory variants driving their expression. In this specific case of association, the nsSNP itself serves as a proxy to the regulatory variant. Therefore, knowledge of associated nsSNP genotype for an individual enables us to make a prediction about the relative abundance of the two alleles of a transcript containing the nsSNP.
To summarize, two classes of DE nsSNPs were discovered: (a) nsSNPs mapping in genes with a previously-identified rSNP (909 nsSNPs, considering nsSNPs of all frequencies) and (b) nsSNPs showing a significant association with expression levels of the gene they are in (884 nsSNPs, considering nsSNPs with MAF≥0.05) (Figure 2a & b). From a non-redundant total of 8233 nsSNPs tested, we predict that 1502 of these (∼18.2%) are DE. If mature protein levels mirror on average transcript levels, which is a reasonable biological hypothesis, then this high fraction has important implications for the levels of protein diversity in the cell.
Of the 884 DE nsSNPs discovered through association testing (set b above), only 291 also possess a previously identified rSNP. This suggests that rSNP detection in our previous study [20] was conservative and that nsSNPs can act as tags of (markers for) nearby, undiscovered regulatory variants. With this in mind, we expect that LD between rSNP- nsSNP pairs in which the nsSNP had a significant association (0.01 permutation threshold) with gene expression, will be greater than LD between rSNP- nsSNP pairs in which the nsSNP was not associated. To explore this, we used data from the single population analysis, and compared the distribution of r-squared values (a measure of LD) between the two rSNP-nsSNP pair types. As expected, we found much higher LD between rSNP-nsSNP pairs where the nsSNP showed a significant association (Mann-Whitney p<0.0001) (Figure 2c). This confirms that in most cases, association of the nsSNP with gene expression of its own gene is due to a regulatory variant that acts as proxy to the identified rSNP.
So far we have used genotypic associations, not direct allele-specific quantification (allele-specific expression; ASE), to derive relative abundance estimates for transcripts of genes containing nsSNPs. To experimentally verify the predictions of the association tests (i.e. that the alleles of associated nsSNP are DE), we tested a subset of nsSNPs for allele-specific expression [22],[28] in heterozygote CEU and YRI individuals. The initial experiment included a total of 141 nsSNPs from category (b) of DE nsSNPs. nsSNPs of this category provide a prediction of the relative abundance of the two alleles as transcripts in the cell. The assay performed was new and proved noisy. As a result it was possible to confirm and analyze signals for 28 nsSNPs. For individuals heterozygous for each nsSNP, we assigned relative expression of the two alleles. We then compared the experimentally derived relative abundance, by ASE, with the predictions of relative abundance from the genotypic association test. We found that predicted and experimentally-quantified relative expression of alleles of nsSNPs were in agreement for 89% (16 out of 18) and 90% (9 out of 10) of nsSNPs tested in the CEU and the YRI populations, respectively. This is in agreement with the FDR estimated above. This strongly suggests that the relative abundance of alternative coding transcripts can be inferred reliably by genotypic associations.
To assess the potential biological impact of DE nsSNPs we compared three functional attributes of those amino acid substitutions arising from DE nsSNPs and those arising from non-DE nsSNPs (nsSNP MAF≥0.05, to assess common nsSNP consequences). We investigated: (1) the relative position of substitution on the peptide, as different effects may arise depending on whether the nsSNP is at the beginning or the end of the peptide), (2) the resulting change in peptide hydrophobicity which may alter the interactions of a protein [29], and (3) the resulting change in Pfam score (a measure of amino acid profile in each position of a protein domain)[30], which assesses the integrity of protein domains that are evolutionary conserved and likely to harbour important functions. In all cases the properties of DE nsSNPs were not different from those of nonDE nsSNPs. Though indirect and not comprehensive, this signal suggests that DE nsSNPs may be a random subset of nsSNPs (Figure S1 a–c).
To assess how many DE nsSNPs have a known function, we explored the OMIM database [31] and found that 71 (out of 1502) DE nsSNPs have an OMIM entry (Table S2). DE nsSNPs were found to map in genes with a role in cancer susceptibility (BRAC1 (+113705), BARD1 (+113705)), asthma and obesity (ADRB2 (+109690)), Crohn's disease (DLG5 (*604090)), myokymia (KCNA1 (*176260)), diabetes (OAS1 (*164350)), chronic lymphatic leukaemia (P2RX7 (*602566)) emphysema and liver disease (P I(+107400)), severe keratoderma (DSP (+125647)), and familial hypercholesterolemia (ABCA1 (+600046)). In some cases the functional role of the nsSNP remains unclear and the noise in reported functional effects in OMIM is well-known and very difficult to assess in a study such as the present, but there are examples where specific effects have been attributed to nsSNPs. For example, rs28931610 in DSP is predicted to change disulphide bonding patterns and alter the peptide tertiary structure; rs28933383 in KCNA1 causes a substitution in a highly conserved position of the potassium channel and is predicted to impair neuronal repolarization; rs28937574 in P2RX7 is a loss of function mutation associated with chronic lymphatic leukaemia; rs28931572 in PI entails a replacement of a polar for a non-polar amino acid and is predicted to disrupt tertiary structure of the protein, and rs2230806 in ABCA1 is associated with protection against coronary heart disease in familial hypercholesterolemia. The modulation of such strong effects by cis regulatory variation may increase the complexity and severity of these biological effects.
Thus far we have presented indirect evidence for an interaction in cis where the effect of an nsSNP is modulated by a co-segregating regulatory variant tagged by an rSNP. Under such circumstances, and if the gene containing the nsSNP has downstream targets, then it is likely that the expression of downstream genes may also be affected. In other words, apart from the modification effect observed in cis, we wanted to test for the genome-wide effects of this interaction directly, in a statistical framework. To do this we carried out ANOVA by testing the main effects of rSNPs and nsSNPs and their interaction term (rSNP×nsSNP) on genome-wide gene expression (trans effects). The rationale behind this approach is that if an rSNP-nsSNP interaction is biologically relevant, its effect may influence the expression of downstream targets of the gene harbouring the rSNP-nsSNP pair. The power to detect an interaction is maximized when all combinations of genotypes are present, each at appreciable frequencies in the population. To increase power of interaction detection, we pooled rare homozygotes with heterozygotes into a single genotypic category, creating a 2×2 table of genotypes. This does not bias our statistic as shown by permutations below. We performed this analysis in the CEU population sample as CHB and JPT population samples were small (45 individuals) and the YRI sample has generally shown low levels of trans effects in previous analyses [20]. We tested 22 rSNP-nsSNP pairs (SNP pairs) with low LD (D′≤0.5) and a MAF≥0.1 for both SNPs, against genome-wide expression. At the 0.001 nominal p-value threshold, we expect roughly 331 significant associations (assuming a uniform distribution of p-values) for the interaction term. We observe 412, which corresponds to an estimated FDR of 80%. This is overall a weak signal (see also Figure 3a), but signals at the tail of the distribution appear to be real given the limited power of this analysis (Figure 3c, d).
To test for potential biases in the statistic used, we carried out the same tests using permuted gene expression values (a single permutation was done by maintaining the correlation structure of gene expression data–see Methods) relative to the rSNP-nsSNP genotypes. We explored the p-value distribution of the rSNP-nsSNP interaction for observed and permuted data (Figure 3b) and found an abundance of low p-values in the observed data. There appears to be some degree of p-value inflation in the observed data relative to the permuted data which is most likely due to correlations in gene expression data. This however does not affect the enrichment of p-values seen in the tails of the observed distribution relative to expected distributions (uniform distribution of p-values) or the p-value distributions derived from permuted gene expression data. The permutation was not done in order to assess significance thresholds but rather in order to assess the enrichment of tests with low p-values in the observed data and is in agreement with the enrichment derived from the enrichment under a uniform distribution of p-values. To further evaluate the robustness of the interactions, we repeated the analysis for the top 10 rSNP-nsSNP significant pairs against their corresponding trans-associated gene expression phenotype, after permuting rSNP genotypes relative to nsSNP genotypes and gene expression values. As expected, the significance of the interaction vanishes in the permuted data. The conditional effects of alleles at the rSNP and nsSNP loci can therefore have a very different impact on the expression of other genes in the cell. This conditional effect on gene expression is illustrated in Figures 3 c and d which show two examples of an rSNP-nsSNP interaction (p = 4.5×10−11 and p = 2.2×10−5), and in Table 2 where the summary statistics and specific information of SNPs and genes are illustrated for the 10 most significant interaction effects. These plots illustrate the effect on gene expression of different rSNP-nsSNP genotypic combinations. In Figure 3c for example, SNP rs3009034 has an effect on gene expression of gene NDN only if the genotype of SNP rs13093220 is homozygous for the common allele. The phenotypic effect of such interactions is even more prominent in Figure 3d where we observe opposite directions of the effect of SNP rs1704196 on gene RLF depending on the genotype on SNP rs6776417.
We have presented a biological framework to interrogate functional genetic variation by focusing on a specific case of epistasis between regulatory and protein-coding variants. We have shown that regulatory variants may have an impact on the protein diversity of cells by differentially modulating the expression of protein-coding variants. In cis, regulatory variants can amplify or mask the functional effects of protein-coding variants, which might consequently result in a milder or more severe phenotype to the one expected if only the protein-coding variant were present. We have shown that such interactions can affect the expression of other genes in the cell (trans effect), in a manner that can only be revealed if the interaction term of the two variants is considered.
The conditional effects of alleles of functional variants may therefore have important consequences for complex phenotypic traits. The extent to which epistasis affects phenotypes remains an under-explored area, but the critical importance of such interactions is starting to emerge [17]. We provide a biological framework for considering and conditioning existing disease associations on known regulatory and protein-coding variants, in an approach that also provides a potential explanation for the differential penetrance of known disease variants. The abundance of cis regulatory and protein-coding variants in human populations and the generic nature of this type of epistatic interaction (no assumptions about specific biological pathways) makes it very likely that such interactions are common genetic factors underlying complex traits and their consideration is likely to reveal important associations that have not been detected thus far. Furthermore, this consideration is particularly important for studies that fail to replicate the primary disease associations in newly tested populations, since it is hypothesized that some of the failures are due to differential frequency of modifier alleles between the first and second population. The consideration of the interactions described above may assist in better interpretation of non-replicated signals.
Total RNA was extracted from lymphoblastoid cell lines of the 210 unrelated individuals of the HapMap [26],[27] (Coriell, Camden, New Jersey, United States). Gene expression (mRNA levels) was quantified using Illumina's commercial whole genome expression array, Sentrix Human-6 Expression BeadChip version 1 (∼48,000 transcripts interrogated; Illumina, San Diego, California, United States) [32]. Hybridization intensity values were normalized on a log2 scale using a quantile normalization method [33] across all replicates of a single individual followed by a median normalization method across all 210 individuals. A subset of 14,456 probes (13,643 unique autosomal genes) that were highly variable within and between populations was selected from the 47,294 probes on the array, and were used for the analysis. A detailed description can be found in Stranger et al.[20].
HapMap nsSNPs (version 21, NCBI Build 35) were mapped onto Refseq genes using nsSNP and gene coordinate information. rSNPs are defined as those phase II HapMap SNPs (version 21, NCBI Build 35, minor allele frequency (MAF)≥0.05) with a cis significant association at the 0.01 permutation threshold, as described in Stranger et al. [20]. The genomic location of rSNPs is within 1 Mb from the probe genomic midpoint.
We tested nsSNP genotype for association with expression levels of the gene it is in using an additive linear regression model [3],[20],[34] applied to each population separately. Our association analysis employed: 1) nsSNP genotypes for the unrelated individuals of each HapMap population (MAF≥0.05) from the HapMap phase II map for each population (version 21, NCBI Build 35) and 2) normalized log2 quantitative gene expression measurements for the 210 unrelated individuals of each of the original four HapMap populations (60 Caucasians of Northern and Western European origin (CEU), 45 unrelated Chinese individuals from Beijing University (CHB), 45 unrelated Japanese individuals from Tokyo (JPT), and 60 Yoruba from Ibadan, Nigeria (YRI)).
To assess significance of association between nsSNP genotype and expression variation of the gene harbouring the nsSNP, we performed 10,000 permutations of each expression phenotype relative to the genotypes [20]. An association to gene expression was considered significant if the nominal p-value from the linear regression test was lower than the 0.01 tail of the distribution of the minimal p-values (among all comparisons for a given gene) from each of the 10,000 permutations of the expression phenotypes. For genes containing more than one nsSNP the most stringent permuted p-value was retained.
To increase the power of the nsSNP association analysis we combined data (nsSNP genotypes and normalized expression values) for unrelated individuals of multiple populations [20]. We compiled three different multiple population comparison panels: 1) CEU-CHB-JPT-YRI, 2) CEU-CHB-JPT, 3) CHB-JPT. Association tests were carried out for each population panel separately using linear regression. Conditional permutations (randomization of data within each population as described in Stranger et al. [20]were performed to assess significance of the nominal p-values. This approach accounts for the population differentiation and prevents detection of spurious associations [20]. For each of the 14,456 probes in each multiple population panel, expression values were permuted among individuals of a single population followed by regression analysis of the grouped multi-population expression data against the grouped multi-population permuted nsSNP genotypes. Associations were considered significant if the nominal p-value was lower than the threshold of the 0.01 tail of the distribution of the minimal p-values from the 10,000 permutations of the expression phenotypes. For genes containing more than one nsSNP the most stringent permuted p-value was retained. It is important to note that in all cases permutations maintained the correlated structure of gene expression values (i.e. all gene expression values were randomized as a block for each individual).
Genomic DNA (gDNA) and total RNA were extracted from lymphoblastoid cell lines of the unrelated CEU and YRI HapMap individuals (Coriell, Camden, New Jersey, United States) using Qiagen's AllPrep kit. RNA was treated with Turbo DNA-free (Ambion) to minimize gDNA contamination. The RNA was concentrated and further cleaned with RNeasy MinElute columns (Qiagen). Total RNA and gDNA were quantified using Nanodrop Spectrophotometer and either Quant-iT RNA or DNA reagents (Invitrogen). Double stranded (ds) cDNA was synthesised from 250 ng of cleaned RNA. The first strand was synthesised with Superscript III (Invitrogen) and random hexamers. The second strand was synthesised with DNA polymerase I (Invitrogen), ribonuclease H (Invitrogen) and dNTPs. The 96-well plate containing the ds cDNA samples was cleaned using Multiscreen PCR plate (Millipore).
The Oligo Pool All (OPA) was custom made by illumina and is based on the Golden Gate assay. Only exonic SNPs≥45 bp from both exon edges were chosen for submission to illumina for assay design to ensure that the assay would work equally well for genomic and cDNA. SNPs that failed according to illumina's design scores were discarded. Paired ds cDNA and gDNA were dried down in 96-well plates and re-suspended in 5 µl of HPLC purified water. Golden Gate assays were then run for all samples using the manufacturer's standard protocol for gDNA (i.e. ds cDNA was treated exactly the same way as gDNA). Reactions were hybridised to 8×12 Sentrix Array Matrix (SAM) Universal Probe Sets so that 96 arrays could be run in parallel. Each bead type (probe) is present on a single array on average 30 times. All reactions were run in duplicate, so that each cell line had two ds cDNA replicate and two gDNA replicate hybridizations. SAMs were scanned with a Bead Station (illumina). A total of 1536 assays were interrogated on the array but only 141 were nsSNPs from this study and only 28 were selected based on data quality for further analysis.
Data from each array was summarised by calculating the per bead type average of 4 quantities after outlier removal: the log2(Cy3) and log2(Cy5) intensities, average log-intensities (1/2log2(Cy5.Cy3)) and log-ratios (log2(Cy5/Cy3)). Outliers were beads with values more than 3 median absolute deviations from the median. Arrays with low dynamic range (determined using an inter-quartile range cut-off of less than 1 for either the log2(Cy3) or log2(Cy5) summary intensities) were discarded. The summarised data was normalized by median centering the log-ratios. All analysis was carried out in R using the beadarray package [35]. Direction of expression (high/low) was assigned to alleles for nsSNPs fulfilling threshold criteria from the association study (adjusted r2 value≥0.27; i.e. the nsSNP explained at least 27% of the variance in gene expression of the gene it is found in so the effect is expected to be large) and the ASE assay (average cDNA log-intensity≥12 within a population).
In each population an nsSNP is defined to be DE if: 1) it maps within a gene that also has an independently identified cis rSNP or 2) it shows a significant association with its own gene's expression levels. We tested those rSNPs with the strongest association per gene with nsSNPs of all frequencies. The total number of nsSNPs that are predicted to be DE is the non-redundant union of 1) and 2).
LD values (r-squared and D′) were calculated by a pairwise estimation between rSNPs and nsSNPs genotyped in the same individuals and within a 100 kb window (Ensembl version 46). LD estimates for rSNP-nsSNP pairs with and without an associated nsSNP (0.01 permutation threshold) were compared using a Mann-Whitney U test.
Given that nsSNPs are likely to be functional we explored three aspects of the resulting amino acid substitution: 1) Relative position of substitution on the peptide, as a percent of peptide total length. 2) Hydrophobicity change in peptide resulting from the amino acid substitution. For each pair of variant sequences the hydrophobicity at the position of the variant amino acid was calculated using the Kyte-Doolittle algorithm using a window size of 7 (centred on the variant amino acid). The difference between hydrophobicty scores was then taken for each of the variant pairs in the dataset. 3) Pfam score change in peptide sequence resulting from the amino acid substitution. All sequences were searched against the profile-HMM library provided by the Pfam database (release 22.0) using hmmpfam from the HMMer software package (version 2.3.2, http://hmmer.janelia.org/) and a default cut off E-value of 10. Only the HMM_ls library was used so that domain assignments to a pair of variant sequences were comparable. The set of Pfam domain assignments were then filtered such that only the domains that overlapped with the SNP position and that at least one of the domain assignments from a pair of variant sequences scored above the Pfam defined gathering threshold, were considered in the subsequent analysis. The difference between the two E-values was then taken for each of the variant pairs in the dataset.
Our aim was to test the interaction effects of rSNP with nsSNP on expression phenotypes in trans in the CEU population Our strategy involved pooling the minor allele homozygote and the heterozygote into a single genotypic category and then coding genotypes as 0 (major allele homozygote) or 1 (heterozygote and minor allele homozygote) for both SNP types. As a result four possible rSNP-nsSNP genotypic combinations were possible: 0-0, 1-0, 0-1, 1-1. We performed ANOVA to test the effects of the rSNP, the nsSNP, and the interaction term (rSNP×nsSNP) in the same model against gene expression phenotypes in trans (in each case excluding the gene from which the rSNP-nsSNP pair originates). Tests were carried out for 22 SNP pairs with low LD (D′≤0.5) between the rSNP and the nsSNP and a MAF≥0.1 for both variants (to avoid outlier effects).
To assess significance of the interaction p-values we generated a single permuted dataset of expression values relative to the combined genotypes and compared the p-value distribution of the interaction term for the observed and the permuted data. To further evaluate the robustness of the observed interactions we permuted the rSNP genotypes relative to nsSNP genotypes and gene expression phenotypes, and re-ran the ANOVA association test for the top 10 most significant interactions.
|
10.1371/journal.pgen.1005502 | Discovering Genetic Interactions in Large-Scale Association Studies by Stage-wise Likelihood Ratio Tests | Despite the success of genome-wide association studies in medical genetics, the underlying genetics of many complex diseases remains enigmatic. One plausible reason for this could be the failure to account for the presence of genetic interactions in current analyses. Exhaustive investigations of interactions are typically infeasible because the vast number of possible interactions impose hard statistical and computational challenges. There is, therefore, a need for computationally efficient methods that build on models appropriately capturing interaction. We introduce a new methodology where we augment the interaction hypothesis with a set of simpler hypotheses that are tested, in order of their complexity, against a saturated alternative hypothesis representing interaction. This sequential testing provides an efficient way to reduce the number of non-interacting variant pairs before the final interaction test. We devise two different methods, one that relies on a priori estimated numbers of marginally associated variants to correct for multiple tests, and a second that does this adaptively. We show that our methodology in general has an improved statistical power in comparison to seven other methods, and, using the idea of closed testing, that it controls the family-wise error rate. We apply our methodology to genetic data from the PROCARDIS coronary artery disease case/control cohort and discover three distinct interactions. While analyses on simulated data suggest that the statistical power may suffice for an exhaustive search of all variant pairs in ideal cases, we explore strategies for a priori selecting subsets of variant pairs to test. Our new methodology facilitates identification of new disease-relevant interactions from existing and future genome-wide association data, which may involve genes with previously unknown association to the disease. Moreover, it enables construction of interaction networks that provide a systems biology view of complex diseases, serving as a basis for more comprehensive understanding of disease pathophysiology and its clinical consequences.
| Many of our common diseases are driven by complex interactions between multiple genetic factors. Disease-specific, genome-wide association studies have been the prominent tool for cataloging such factors, by studying the genetic variation of a gene in a population and its association with the disease. However, these studies often fail to capture interactions between genes despite their importance. Interactions are notoriously difficult to investigate, because testing the large number of possible interactions using contemporary statistical methods requires very large sample sizes and computational resources. We have taken a step forward by developing a new statistical methodology that significantly reduces these requirements, making the study of interactions more feasible. We show that our methodology makes it possible to study interactions on a large scale without compromising the statistical accuracy. We further demonstrate the utility of our methodology on data relating to coronary artery disease and discover three distinct interactions that might provides new clues to the pathophysiology. The study of genetic interactions have the potential to link disease genes together into disease networks that provide a more detailed description of the interaction between genes and how it drives the disease.
| Cardiovascular disease, cancers, diabetes and chronic obstructive pulmonary disease, accounting for almost 60% of the causes of death 2013, globally [1], are all examples of complex diseases. A complex disease is characterized by an intricate system of interactions between genetic, epigenetic, other intrinsic factors, and environmental factors, that constitutes its pathophysiology. The genetic architecture of many common complex diseases is poorly understood. For example, the 46 robustly associated variants that have been found for coronary artery disease (CAD) only explain 10.6% of the estimated heritability; this was shown in a recent meta-analysis of almost 200,000 individuals [2]. The same pattern of unexplained, or missing, heritability is found in most common complex diseases [3]. Assuming that the estimated heritability is correct, the possible explanations for the high ratio of missing heritability include 1) a large number of causal genetic variants, each with a small effect, 2) sequence variation that is commonly excluded from analysis, e.g. copy number variation or rare variants, 3) other commonly unmeasured heritable components, e.g. heritable epigenetic modifications, and 4) interaction effects between common variants. Moreover, any combination of these explanations is plausible. In this paper we focus on the inference of interaction in genetic association studies; this is sometimes called epistasis, epistatic interaction or genetic interaction; here we will refer to it as genetic interaction or simply interaction.
Genetic interactions are characterized by two or more variants producing an unexpected phenotype that is not easily explained by the marginal effects of the individual variants. Extensive studies in model organisms have shown that genetic interactions are common phenomena [4]. The field was pioneered by Bateson [5], who studied genetic interactions in plants and chicken. More recently, synthetic lethal interactions (in which the simultaneous occurrence of two mutations, by themselves without effect, lead to cell death) have been studied extensively in yeast and Caenorhabditis [6–8] and interactions between quantitative trait loci have been studied in mouse, Drosophila and Caenorhabditis [9–11]. Since interactions are widespread in other organisms, it seems unlikely that such effects would not exist in humans. Furthermore, genes are linked in metabolic, regulatory and signaling pathways and it is likely that this will be reflected as interactions between variants, as has been shown for transcriptional regulation in Drosophila [12]. Therefore, studies of genetic interactions have a strong potential to provide important insights about disease biology—specifically, interactions reflect dependencies in pathophysiology and may allow predictions of effects (and side effects) that are relevant for disease prognosis and treatment.
Several approaches have been developed to study genetic interactions (see [13–15] for three excellent methodology reviews). In medical genetics, the prevalent tool for modeling single variant association in unrelated individuals has been generalized linear models (GLM). The advantages of GLMs are flexibility in modeling the phenotype, easy interpretation and straightforward adjustment for confounders. Although the GLM framework can model both discrete and continuous outcomes; we will, in this work, concentrate on the case-control outcome. Studies of interactions are, however, not without issues. Firstly, the identification of interactions depends on the scale relating the genotypes to phenotypes. Secondly, because the GLMs are fitted by iterative procedures, the computational burden is high. Thirdly, straightforward multiple testing correction leads to low statistical power. We now elaborate briefly on these three issues.
The dependency of GLMs on a scale sometimes causes confusion [16, 17]. The scale is determined by a link function that maps the phenotype to the linear predictors. For example, for two predictors a and b, the phenotype y can be determined by an additive (y = a + b) or by a multiplicative (y = ea+b) model. A commonly used link function in case/control studies is the logit, which is used in logistic regression. This displays a combination of mathematically favorable properties: it models the case/control selection bias, the parameters have minimal sufficient statistics, and it is the maximum entropy null model [18]. However, the choice of scale is to a large extent a modeling issue and should not be based on mathematical convenience alone. For example, when, for a set of variants, the presence of a risk allele in any single variant is sufficient to cause the disease, the log-complement link function yields an appropriate model [17, 19]. Ultimately, the best choice of scale depends on the unknown biological model that has generated the data. The choice of scale is very problematic because, even if the true model underlying the data displays interaction, it is often possible to select a scale that diminishes the interaction effect [20]. Conversely, if the true model does not display interaction, then there is another scale that, in the asymptotic case, will display interaction [17]. In response to this, Knol and VanderWeele [21] suggest that the p-value of an interaction should be reported on a set of reasonable scales to show whether the interaction seems invariant of scale. We follow this suggestion and, furthermore, extend it by constructing a test for interaction that is invariant over a set of link functions.
A different approach builds on the rationale that if, for interacting variants, certain combinations of alleles affect disease risk, then this would be reflected in differential enrichment for these allele combinations between cases and controls, and therefore a difference in their linkage coefficients (LDcases and LDcontrols). The LD-contrast test [22] compares the normalized difference of LDcases and LDcontrols as a χ2-distributed statistic for interaction. A recently been proposed version of the LD-contrast test [23] uses genotypes recoded to a pair of binary variables (according to model of inheritance).
A third approach, the multi-factor dimensionality reduction (MDR) [24], uses dimensionality reduction techniques to recode the 3 × 3 penetrance matrix into a binary variable that optimally classifies cases and controls. This is then evaluated by cross-validation and a permutation procedure is used to estimate significance. Several variants of MDR have been developed [25, 26].
One common approach to improve the computational complexity, has been to introduce the naive assumption that it is impossible for two variants to be simultaneously associated with the phenotype unless they interact. Examples include the method of [27], using a log-linear GLM, and many of the variations of the MDR method, including the original one [24, 25]. Under this naive assumption it is, in the GLM setting, sufficient to compare three models: two single variant association models and the saturated model, which will represent interaction. The parameters of these models can be efficiently estimated since they all have closed form solutions. Unfortunately, this simplification allows interactions to be incorrectly inferred between two variants that both are associated with a main effect, but there is no interaction (we will refer to this as double main association). As a consequence, genuine genetic interactions may be obscured by these double main associations [28]. In this work, we will focus on inference of genuine interactions.
Finally, the reduction in statistical power implied by correction for multiple tests constitutes a major limitation for performing interaction studies on a larger scale. For an investigation of interaction between all pairs of a set of n = 500,000 variants, a Bonferroni correction for n(n − 1)/2 tests gives a significance threshold of 4.0 ⋅ 10−13, which is considerably lower than the corresponding significance level 1.0 ⋅ 10−7 for a standard single variant analysis. The burden of multiple tests grows exponentially with the number variants involved in the tested interactions, and, henceforth, we will limit ourselves to the case of pair-wise interactions. Various screening strategies have been applied in attempts to improve power. These may use prior information that identify a smaller set of candidate variant pairs (we will investigate two such approaches in our analysis of biological data) or they may be based on the data at hand. An example of the latter include the screening test of Marchini [29], which removes variant pairs lacking a marginal effect for one or both of the participating variants. Millstein et al. [30], using a reasoning similar to the LD-contrast test above, suggested a LDcases screening test for significant linkage enrichment among cases. However, observing that this induced a bias in the subsequent main test, they also proposed a LDcohort screening test. The latter test relies on the linkage enrichment in cases also showing as a linkage enrichment of the pooled cases and controls, but formally does not use any prior information about disease state. Various combinations of screening and main tests have been proposed: marginal screen with logistic GLM main test [29], LDcohort screen with logistic GLM main test [30, 31], and LDcases screen with LD-contrast main test [23].
In this work we introduce a stage-wise multiple testing methodology that exhaustively tests all variant pairs. In this methodology, a sequence of hypotheses is considered in order of increasing complexity. Only variant pairs that cannot be explained better by a simpler hypothesis compared to the most complex hypothesis (representing interaction) are tested at subsequent stages. This is conceptually different from the screening approach by Marchini [29], which instead requires that a variant pair fits an intermediate screening hypothesis (of single marginal association) better than the simplest hypothesis (of no association) for it to be tested at the subsequent stage. Because the hypotheses considered are closed under intersection, we show, in two situations, that the family-wise error rate is controlled. Furthermore, since the models under the simpler hypotheses can be estimated efficiently, our methodology allows the use of full GLMs. The multiple testing correction is alleviated and results in a substantial increase of power compared to the Bonferroni correction. We also construct a scale-invariant test for interaction using several link functions. Furthermore, we assess a set of statistical methods for inferring genetic interactions on synthetic data and show that our methodology improves on these. Lastly, we discover three distinct interactions that are associated with CAD, of which one includes a novel locus.
In this section, we describe our multiple testing methodology, which is aimed at large-scale pairwise interaction testing. We show that it gains additional power by separating a complex hypothesis into stages of simpler null models. We have derived two methods that rely on different assumptions, having different effects on the bounds of the family-wise error rate (FWER). We start by briefly reviewing general linear models (GLM), which we use to express our model of interaction, as well as the simple null models. Frequently, when GLMs are applied in pairwise interaction testing, FWER is bounded using Bonferroni.
In this section, we will give an account of three investigations of statistical power that all indicate the utility of our stage-wise methodology. The generation of simulated data used in these investigations are described in Material and methods section Generation of synthetic data for estimation of statistical power.
The intuitive idea behind the stage-wise methodology is that we aim to (1) reduce the number of tests in later stages compared to earlier, while (2) asserting that actual interactions advance to later stages. We show in the Results section Analysis of biological data, below, that the number of tests in the last stage is in fact substantially reduced, suggesting that aim (1) is unlikely to be a problem. Here, we have investigated aim (2 by comparing the power of the tests in the first and last stage. That is, for data generated from HA, we compare the power of the likelihood ratio test of H1 against HA to that of the test of H4 against HA. Indeed, the results in Fig 2 (using data generated from a double dominant interaction model) suggests that the test in the first stage, at least under these conditions, have substantially greater power than that in the last stage. However, the test in the first stage can obviously not be used as a test for interaction by itself, since it measures any kind of association to the phenotype, including, for example, pairs for which only one of the variants is associated.
We further investigated the distribution of statistical power of seven methods using simulated data generated from the spectrum of all possible interaction models, following the ideas of [35] (see Material and methods section Generation of synthetic data for estimation of statistical power for details). The first of these methods is our static method, and the remaining methods include four methods based on a logit-link GLM with different screening strategies, Logistic (without screening), Marginal+logistic [29], CSS+logistic [30] and R2+logistic [31]) and two methods based on the LD-contrast test with different screening strategies, LD-contrast (without screening), and Sixpac [23] (a LDcases+LD-contrast method), for details, see Material and methods section Comparison of statistical methods. It should be noted that none of the latter six methods are scale-invariant—one may expect that this property would enhance their power. For simplicity of simulations, we only evaluated the static method here; however, since the adaptive method is more powerful than the static, this can also be viewed as a conservative estimate of the power of the adaptive method. As can be seen in Fig 3, the static method consistently has greater power than the other approaches. The marginal+logistic method performs best of the remaining methods, while the the LD-contrast method have the worst performance. In S1–S4 Figs, we also report the result of a more computationally intensive power comparison, including the above methods, as well as our adaptive stage-wise scale-invariance method and the Model-based MDR (MB-MDR) method [26] (see S2 Text for details). These results corroborate those above, that is, for most models our stage-wise methods performs better than the other methods (see further discussion in S2 Text).
Intuitively, when more variants are associated with the phenotype in our stage-wise methodology, the multiple testing correction in the intermediate stages becomes larger, and therefore statistical power is reduced. For this reason, we investigated how the statistical power depends on the number of associated variants using data simulated from the double-dominant interaction model (see Material and methods section Generation of synthetic data for estimation of statistical power). As shown in Fig 4, the power decreases as the number of associated variants increases. Because of the additional penalty of the weight, the static method can have lower power than directly testing interaction using a Bonferroni correction, precisely when M(M − 1) > w4 N(N − 1) (where N is the total number of variants and M is the number of associated variants). It can be noted that for our biological data, M(M − 1) = 306 ≪ w4 N(N − 1) = 346,035,421.8 (based on the N = 33,963 tested variants and the M = 18 robustly associated CAD variants present on the IBC-chip, cf. S1 Table).
Both the static and adaptive stage-wise methods are based on the likelihood ratio test, which is asymptotically correct. As we show in S1 Text, the adaptive method controls the FWER asymptotically. For the static method on the other hand, we can even show that the FWER is controlled for any data size. Consequently, it is interesting to investigate the behavior of both methods on finite data and compare it with that of the same seven methods as in the power comparison (Logistic, Marginal+logistic, CSS+logistic, R2+logistic, LD-contrast, Sixpac, and MB-MDR). We considered two cases, one close to the assumptions of our methods, and one designed to be challenging. We investigated these two cases using simulated data (see further Material and methods section Generation of synthetic data for estimation of statistical power).
In the first case, we used each of our null models, H1: no association, H2/H3: single main association, and five models for double main association (H4) with the identity, log, log-complement, odds and a logit link functions, respectively, to generate the phenotype based on a single pair of variants. The first seven rows in Table 2 show that both the static and adaptive methods accurately control the FWER under these circumstances (i.e., FWER is below or close to the expected value of 0.05). All seven other methods control the FWER for the no association and single main models. However, for the double main models, they control FWER only on the multiplicative scale (i.e., with the log, logit and log complement link functions). For the remaining models (double main:identity and double main:odds) these methods fail to control FWER, with the exception of the R2+logistic that controls FWER for the double main:identity model.
In the second case, where we attempted to construct instances that challenge the additional asymptotic assumptions made in the adaptive method. The phenotype was here determined by an multivariate additive GLM with logit link function on a set of L ∈ {10, 20, 30} markers. The parameter distributions were chosen with the intention to let only a small and difficult subset of the variant pairs to reach the stage they belong to. The last three rows in Table 2 show that the static method controls the FWER, but suggest that for the adaptive method, FWER is inflated by approximately a factor 3 compared to the desired rate. The remaining methods controls FWER in this setting, possibly an effect of the data being generated on a multiplicative scale.
We applied our stage-wise methodology on genome-wide CAD case-control data from the PROCARDIS study, using our five link functions. To enhance such a large-scale analysis, we explore three strategies for selecting subsets of variant pairs to test. The first strategy represents a genome-wide approach, while the latter two strategies were designed to a priori enrich for pairs likely to exhibit interaction. For the same reason, our main focus will be on the more powerful adaptive method, combined with validation of any significant discoveries in a separate cohort.
In the first strategy all 229,050,992 pairs, for which the product of the minor allele frequencies > 0.04, were selected. The stage-wise methodology subsequently reduced the number of pairs to 15269, 7712 and 93. This analysis resulted in seven variant pairs that were significant for at least one link function, see Table 3. We used genomic proximity to coarsely estimate genes corresponding to these variant pairs. One variant pair, indicating an interaction between IL1R1 and CDNK2B-AS1, was significant on the additive odds scale, only. The p-values for the other scales were quite far from significance, indicating that this association is not scale-invariant. In other words, this interaction should be interpreted with care, as we cannot exclude the possibility that this is the effect of double main association, e.g., on the logit scale, without interaction. The remaining six variant pairs indicated an interaction between the genes MIA3 and CDNK2B-AS1. None of these passed the scale-invariance test (i.e., was significant for all link functions). However, for most of these variant pairs, the p-values for all scales are of the same magnitude and reasonably close to the significance level of 0.05 (see, e.g., the rs4846770–rs518394 variant pair), perhaps suggesting that this could be an effect of insufficient power rather than scale dependency.
In the second strategy 314,445 pairs were selected based on loci previously associated with CAD. This is based on the common hypothesis that some robust CAD associations may be the marginal effects of interacting variant pairs. Candidate pairs were formed by taking each of the previously associated CAD variants, see S1 Table, and combining it with each other variant. Interestingly, similar to the results from the all-vs-all strategy above, one variant pair indicating an interaction between MIA3 and CDNK2B-AS1 was significant for several link functions, but again, just, failed the scale-invariance test, see Table 3. Somewhat unexpectedly, this variant pair did not coincide with any of those in the all-vs-all analysis. However, it turns out that, while variants for both MIA3 and CDNK2B-AS1 have previously been robustly associated to CAD (see S1 Table), these variants did not include any member of the top-scoring variant pairs in the all-vs-all analysis. This enrichment strategy might therefore have been suboptimal.
In the third strategy, we used prior information from HumanNet [36], a probabilistic functional gene network that links genes for which significant evidence of interaction have been provided in one or more omics experiments; this resulted in 2,319,906 variant pairs. We found two variant pairs that were significant for all five link functions, thereby passing the scale-invariance test, see Table 3. For each of these two pairs, genomic proximity suggests that an interaction between PSRC1 and CXCL6 is associated to CAD, and, thus, may play a role in its pathophysiology. The exact mechanism of the interaction is, however, unknown, and the evidence for it in HumanNet was merely reported as co-expression between human genes.
The maximal effect size for each discovered interaction range from 0.4718 to 1.379 (S2 Table); as a reference, the effect sizes for robustly CAD-associated single variants are commonly around 0.285 [37]. While, after adjustment for age, sex, smoking, and population stratification, most effect sizes were reduced, this was not the case for the CXCL6-PSRC1-related interactions (see S3 Table). The penetrance pattern of one of the CXCL6-PSRC1 variant pairs, rs4694178 and rs602633, is shown in S5 Fig. Of note, it shows a marked directional change in risk for individuals carrying the major rs4694178 homozygote and the minor rs602633 homozygote.
We then investigated the reproducibility of the CXCL6-PSRC1-related interaction on a non-overlapping sub-cohort of PROCARDIS. This sub-cohort consists of 1797 cases and 2677 controls, which were genotyped on the Illumina Human1M Quad chip. The exact variants of the significant pair were not genotyped, and was therefore imputed (and hard-called) using the 1000 Genomes phase 3 reference panel. We tested interaction directly using a GLM combined with our link functions. This resulted in the p-values 0.174, 0.241, 0.103, 0.056, 0.156, for the identity, log, log-complement, odds and logit link functions respectively. We note that, while the p-value for the odds scale is close to significance, the replication clearly did not pass the scale-invariance test. Despite this, the penetrance patterns for different allele combinations were very consistent between discovery and replication analyses, compare S5 and S6 Figs. Of note, is that the minor allele frequency for rs602633 is relatively different between the two cohorts, see S4 Table. We, furthermore, expanded the search to the ten closest variants on both sides of both significant variants. The best variant pair, rs11730560 and rs1277930, reached nominal significance, and the p-values were 0.023, 0.0311, 0.014, 0.0084, 0.0209, again for the identity, log, log-complement, odds and logit link functions respectively. It did, however, not pass multiple testing accounting for all the 380 tested variant pairs.
We also performed analyses using the static method assuming 100 marginally associated variants with the same search strategies, but no variant pair was significant on any scale for any of the strategies. This may be a consequence of the expected lower power of the static method.
We have introduced a new stage-wise methodology that is statistically and computationally efficient for large-scale inference of genetic interactions. We have derived two separate methods: The first is the static method that uses a priori estimated multiple testing correction factors; here we have used the number of published robustly associated CAD SNPs to obtain such an estimate. The second adaptive method does not rely on the assumption of known correction factors, but uses the number of associated variant pairs at each stage to compute the multiple testing factors. To the best of our knowledge, this is the first method that uses the idea of a closed set of hypotheses to perform an exhaustive pairwise scan of interactions. We have shown that this stage-wise method performs better on a large number of interaction models compared to other statistical methods. The basic idea is that instead of directly testing all possible variant pairs for interaction, we use a sequence of more general association tests as a filter to reduce the number of pairs until only potential interactions remains. This shifts much of the multiple testing burden from the final interaction test to the preceding general tests. Because the tests leading up to the interaction test in general are more powerful (i.e., interactions will not be discarded), this results in higher overall power.
Our simulation results show that our new methods in general have higher statistical power than other common interaction inference methods. For certain specific models and low MAFs, the Sixpac method [23] perform relatively well, but its performance over the spectrum of all possible interaction models is low. The simulations suggest that, in ideal cases, it may be possible to infer interactions using our stage-wise methodology even when correcting for 1012 pairs, since each stage greatly reduces the number of tested interactions. However, we conjecture that, in practice, it will be important to take advantage of prior information in order to reduce the number of tested interactions; for example, we used information from the HumanNet database to select candidate interactions. Moreover, the methodology presented in this paper can also be combined with screening procedures such as LDcohort [30, 31] or the efficient probable approximate complete search algorithm of [23]. This may give even further gains in power and computational speed.
Deciding which scale to work on (i.e., which link function to use, see Table 1) can be troublesome and many researchers advocate a favorite scale for statistical or biological reasons. Testing on a single scale will improve the statistical power for interactions that fit that scale compared to testing multiple scales. However, if pairs of variants are additive on another scale, this approach will lead to an increased number of false positives, in the sense that there exists simpler models that explain the data. In our framework we offer a compromise: we display all pairs that are significant on at least one scale, but also provide a test that require significance on multiple scales. In this way, a researcher can interpret the significance of an individual scale in the context of the other scales. From our analyses of biological data, no particular scale appear to consistently be the critical one for the scale-invariance test.
We note that the scale-invariance test provides an advantage in terms of FWER control. While most other methods failed to control FWER for data generated with a link function that was sufficiently dissimilar from that underlying the method, the scale-invariance test allowed our methods to control FWER for data generated with any tested link function. Although the static method could be derived using closed testing, the derivation of the adaptive method relied on additional assumptions that may be difficult to satisfy in practice. We observed that this could cause inflation of the FWER under a specifically designed additive model with multiple weakly associated variants. We note that, while analytically straight-forward to work with, the FWER is known to be a conservative control of the experimental error at the expense of power [38]. One future direction could therefore be to investigate other error control measures, for example the false discovery rate (FDR) [38]. Moreover, there are several cases where the advantages, in terms of computational efficiency and statistical power, of the adaptive method may compensate for a relatively modest inflation in the FWER. Specifically, as validation is conventionally required in genetics studies, the adaptive method can be used as a powerful tool in the discovery phase of large-scale studies.
Our biological analysis identifies the well known CDKN2B-AS1 locus, or ANRIL, which encodes an anti-sense RNA [39]. The region contains several variants that are robustly associated with CAD but the pathophysiology of ANRIL is unknown. Interestingly, we detect an interaction between CDKN2B-AS1 and MIA3, another established CAD locus [2], potentially indicating a new lead on CAD pathophysiology. Variants in the CELSR2-PSRC1-SORT1 gene cluster have previously been shown to be associated to CAD and lipid traits [2], although the exact causal relation of the genes is unclear. Our results suggest that HumanNet’s co-expression-based connection between CXCL6 and PSRC1 in fact mirrors a genetic interaction in CAD, supporting a role of PSRC1 in CAD (in line with recent results [40]). Moreover, inflammation has long been seen as an important component of the pathology of atherosclerosis, but few inflammation genes have been implicated by genome-wide association studies [41], and only in meta-analyses. It is therefore interesting that in the two sets of variant pairs unbiased with respect to CAD, we find interactions involving genes clearly implicated in regulation of inflammation, i.e., the interleukin- and chemokine-related genes IL1R1 and CXCL6 (IL8). Of course, follow-up functional investigations are required to fully understand the potential pathophysiological consequences of these interactions.
Complex diseases are multi-gene and multi-factorial diseases characterized by complex interactions between genetic, regulatory, metabolic and environmental factors. The majority of complex disease genome-wide association studies have employed traditional association analyses of single genetic markers, which only have been able to explain a small fraction of the disease heritability. A perhaps more conclusive approach would be to reconstruct the complex dependencies between factors as an interaction network reflecting the disease pathophysiology. This approach, however, has so far been hampered by the lack of efficient methods for inference of interactions associated to disease. The static and adaptive methods are two effective ways to discover genetic interactions, and the flexibility of GLMs allows them to be applied to a wide range of different phenotypes. Genetic interactions, and in particular the construction of interaction networks explaining the pathophysiology of the disease, have a potential for clinical relevance, both in terms of prognosis, treatment and drug development. The ideas of stage-wise testing is furthermore applicable outside medical genetics, whenever a large number of complex hypotheses are tested.
The PROCARDIS study was carried out in accordance with the Helsinki Declaration and approved by the Institutional Review Board (IRB) at each one of the 4 recruiting centers: the Regional Ethics Review Board at Karolinska Institutet, Stockholm in Sweden (approval number 98-482 and 03-491), the IRB at the University of Munster, Munster, in Germany, the IRB at the Mario Negri Institute, Milano in Italy and the IRB at the University of Oxford, Oxford, United Kingdom. All study participants provided their written informed consent to participate in the study, a procedure approved by each one of the local ethical committees.
A subset of the PROCARDIS cohort has previously [42] been genotyped with the Illumina IBC chip, a iSelect Custom Genotyping BeadChip [43]. This chip contains 48,742 variants in approximately 2,100 candidate genes that are believed to be involved in vascular disease processes. The subset of PROCARDIS used in this study are 3,162 cases and 3,353 controls of which 3,865 are males and 2,650 are females. The disease phenotype is CAD (including myocardial infarction). Multidimensional factor analysis indicated no significant population structure.
The following quality control was performed. We removed variants with a minor allele frequency < 0.05, with significant deviation from Hardy-Weinberg equilibrium p < 10−6, and removed the variants from the X-chromosome to avoid confounding with gender, leaving us with 33,963 variants.
We performed simulations of all possible weight combinations with a precision of 0.1, the results can be seen in S5 Table. The choice seems to have little impact, and the best weight combination was 0.1, 0.3, 0.3 and 0.3, which is the one we used on biological data.
We used two different simulation strategies for the power estimation. The first of these was used to compare the stage-wise scale-invariance method to other methods, see further next section. Models were constructed by enumerating all possible penetrance matrices displaying interaction for a single variant pair [35], as follows: The models were initially restricted to complete penetrance, that is, the penetrance is either 0 or 1, which allowed us to enumerate all 29 = 512 penetrance matrices. Only models considered to interact were included, here a model was defined as an interaction if the penetrance matrix could not be decomposed according to Risch’s [19] definition of genetic heterogeneity. That is, formally, let P be 3 × 3 binary penetrance matrix. Then P is not an interaction if and only if there exists two 3 × 3 binary matrices, R with identical rows, and C with identical columns, and P cannot be written as the logical OR between R and C. The genetic heterogeneity definition was chosen because it excludes most marginal effect-only models, thereby reducing noise in the power estimation, and because it can easily be evaluated for complete penetrance matrices. The penetrance matrix was then reduced to continuous values by changing the 0’s to a specified base risk of β0 and the 1’s to β0 + β1. To enhance comparison of models, we used heritability, H2, as a summary measure of all genetic effects in a model, where heritability was defined as
H 2 = ∑ i , j ( Pr ( Y = 1 ) - Pr ( Y = 1 ∣ X 1 = i , X 2 = j ) ) 2 Pr ( X 1 = i , X 2 = j ) Pr ( Y = 1 ) ( 1 - Pr ( Y = 1 ) ) .
For each model, the parameter β1 was adjusted to obtain heritabilities of 0.005, 0.010 and 0.015. Using this enumeration we obtain a set of models, each defined by a matrix of penetrances for each genotype combination Pr(Y = 1∣X1 = i, X2 = j) (cf. S7, S8 and S9 Figs and S2 Text). The genotypes for cases and controls were then generated using Bayes’ theorem
Pr ( X 1 = i , X 2 = j ∣ Y = 1 ) = Pr ( Y = 1 ∣ X 1 = i , X 2 = j ) Pr ( X 1 = i ) Pr ( X 2 = j ) ∑ k , l Pr ( Y = 1 ∣ X 1 = k , X 2 = l ) Pr ( X 1 = k ) Pr ( X 2 = l )
to get the multinomial distribution over genotypes. We generated 1,000 data sets from each of these models. We assumed a balanced design (i.e., same number of cases and controls), the sample size for each group was varied over 2,000, 3,000, and 4,000, the heritability was varied over 0.015, 0.020 and 0.025, and the minor allele frequency was fixed to 0.3. Each data set comprised a single interacting variant pair, and to model multiple testing, we assumed that there were 106 variants and 1012 variant pairs tested. For each model and each parameter combination, the power of a method to detect interaction was estimated over the 1,000 replicates. The method’s power over the spectrum of tested interaction models were then summarized in an exceedence plot.
We performed two additional power analyses using a second simulation strategy, where we used data simulated from a specific interaction model, the double dominant model (described in S2 Text.1), in which α = β1 = β2 = γ1 = γ2 = 0 and δ11 = δ12 = δ21 = δ22 = x, and a logit link function was used. The value x was then varied to get the heritability 0.01, 0.02 and 0.03. This analysis used a fixed sample size of 3,000 cases and 3,000 controls. The minor allele frequency was set to 0.3 at both loci.
In the first of these two power analyses, we investigated how the relative power in detecting an associated variant pair generated under an interaction model varies over the different individual stages in the stage-wise approach, specifically we compared the power in the first and the last stages (i.e., using the null models H1 and H4). The parameters of the double dominant model can be seen in S6 Table.
In the second power analysis, where we studied the power of the static method to detect an interacting pair as a function of the estimated number of marginally associated variants, we set the total number of variants tested, N = 1,000,000, and the number of marginally associated variants was varied, M ∈ {10, 20, 100}. For the static method the corrections for each stage, in order, then was set to N(N − 1)/2, M ⋅ N, M ⋅ N and M ⋅ (M − 1)/2. The parameters of the double dominant model can be seen in S7 Table.
The FWER estimation was based on simulated data. We generated data from ten different null models representing two different cases: The first case corresponds to the null models in our stage-wise methodology: no association, single main association and five null models with double main effects corresponding to the link functions in Table 1; the second case represents a more challenging scenario and comprise three null models with multiple main effects.
For each null model we generated 200 data set replicates that contained 500 − L unassociated and L associated variants, where L depends on the null model, L = 0 for the no association, L = 1 for the single association, L = 2 for the double main models, and L = 10, 20, 30 for the multivariate model. For each data set there was therefore L associated variants according to the null model. For each replicate this resulted in 124,750 pairs. The minor allele frequency was sampled uniformly between 0.2 and 0.4. We sampled individuals until we obtained 4000 cases and 4000 controls. The parameters used in each null model can be seen in S8 Table.
For the null models with multiple main effects, we used an additive logistic regression model to generate the phenotype. Let L ∈ {10, 20, 30} be the number of variants to include from the chromosome, then the model was defined
log ( p 1 - p ) = β 0 + ∑ i = 1 L β i x i
where xi ∈ {0, 1, 2} and βi ∼ N(0.15, 0.01). The intercept was set to −9.0.
We evaluated the power of our static stage-wise scale-invariant method in comparison to six other statistical methods. In our main, large-scale analysis using data generated from an enumeration of all possible interaction models (see Material and methods section Generation of synthetic data for estimation of statistical power), we restricted ourselves to statistical methods that could efficiently compute a p-value with enough precision to test how they performed in realistic scenarios: Four methods based on a direct interaction test (i.e., in our framework description above, testing hypothesis H4 against HA) with a logistic link function GLM, but employing different screening strategies: Logistic—no screening. Marginal+logistic—the marginal screening method described by [29], which uses a GLM that tests the marginal effect of each variant at an optimistic significance level 0.1 for screening. The screening approaches used in CSS+logistic [30] and R2+logistic [31] are both LDcohort-based, but differ in the definition of the χ2-based statistic, and the choice of significance threshold used for the screening: χ2 ≥ 3 (corresponding to p ≤ 0.39) and p ≤ 10−4, respectively. Thirdly, we test two methods based on the LD-contrast test with different filtering strategies: LD-contrast—no screening. Sixpac—the method of [23], which recodes variant genotypes into two binary variables (according to dominant and recessive coding) and then combines LDcases screening with a LD-contrast main test. The significance level was set to 0.05. We assumed that there were 1012 variant pairs present on the chip and that there existed one interacting pair. For the methods without screening (Logistic and LD-contrast), as well as for the Sixpac method, we corrected for 1012 pairs. For the remaining screening methods, we corrected for the expected number of null variant pairs passing the screening, by taking the product of the p-value threshold and the total number of pairs (i.e., Marginal+logistic:(0.1⋅1062)≈5⋅109, CSS+logistic: 0.39 ⋅ 1012 = 3.9 ⋅ 1011, and R2+logistic: 10−4 ⋅ 1012 = 108). A pair was declared significant if it passed the significance level of both the screening and the main test. For all these methods we used the Holm-Bonferroni correction for multiple testing, which is more powerful than the classic Bonferroni correction. For the Static stage-wise method we corrected for 1012 pairs, 100 ⋅ 106 pairs, 100 ⋅ 106 pairs and 4950 pairs in each of the four stages respectively, to simulate the situation with 100 associated variants.
We also performed a second, smaller-scaled, but computationally more demanding, power comparison using data generated from specific interaction models and null models (described in detail in S2 Text). In addition to the seven methods enumerated above, this comparison also included our adaptive stage-wise, scale-invariant method and the Model-Based MDR (MB-MDR) method [26], which is a parametric extension of the MDR method that addresses some shortcomings of the original MDR method, in particular adjustment for main effects (these methods require the generation of data sets complete with both null and interaction pairs and could not be evaluated in the main power comparison above).
Lastly, we also used the same nine methods in a FWER comparison using simulated data generated as described in Material and methods section Generation of synthetic data for FWER estimation.
A C++ implementation of all methods and source code for all experiments is available at: https://github.com/mfranberg/besiq.
|
10.1371/journal.pntd.0003370 | Mycobacterium africanum Is Associated with Patient Ethnicity in Ghana | Mycobacterium africanum is a member of the Mycobacterium tuberculosis complex (MTBC) and an important cause of human tuberculosis in West Africa that is rarely observed elsewhere. Here we genotyped 613 MTBC clinical isolates from Ghana, and searched for associations between the different phylogenetic lineages of MTBC and patient variables. We found that 17.1% (105/613) of the MTBC isolates belonged to M. africanum, with the remaining belonging to M. tuberculosis sensu stricto. No M. bovis was identified in this sample. M. africanum was significantly more common in tuberculosis patients belonging to the Ewe ethnic group (adjusted odds ratio: 3.02; 95% confidence interval: 1.67–5.47, p<0.001). Stratifying our analysis by the two phylogenetic lineages of M. africanum (i.e. MTBC Lineages 5 and 6) revealed that this association was mainly driven by Lineage 5 (also known as M. africanum West Africa 1). Our findings suggest interactions between the genetic diversity of MTBC and human diversity, and offer a possible explanation for the geographical restriction of M. africanum to parts of West Africa.
| Tuberculosis remains one of the main global public health problems. Human tuberculosis is caused by bacteria known as the Mycobacterium tuberculosis complex (MTBC). The MTBC includes a variant called Mycobacterium africanum, which causes up to half of all tuberculosis cases in West Africa. For reasons unknown, M. africanum does not occur in other parts of the world. To explore the possible reasons for this geographic restriction of M. africanum, we analysed a large collection of bacterial strains isolated from tuberculosis patients in Ghana. We genetically characterized these bacterial isolates and collected relevant socio-demographic and epidemiological data. We found tuberculosis patients infected with M. africanum were more likely to belong to the Ewe ethnic group, compared to patients carrying other MTBC bacteria. The Ewes are indigenous inhabitants of coastal regions in West Africa that have previously been shown to harbour a high prevalence of M. africanum. Our findings support the hypothesis that different variants of MTBC have adapted to different human populations, and offer a possible explanation for the geographical restriction of M. africanum to West Africa.
| Tuberculosis (TB) remains the leading cause of adult death by a single infectious disease world-wide [1]. Despite the high mortality caused by TB, only 5% to 10% of infected immunocompetent individuals progress from initial infection to active disease [1]. In 2013, an estimated 9.0 million new cases and 1.5 million deaths due to TB occurred; with 30% of the global burden of TB occurring in Africa, an indication of the strong association with HIV/AIDS [1].
TB is caused by a group of closely related bacteria referred to as the Mycobacterium tuberculosis complex (MTBC). MTBC comprises M. tuberculosis sensu stricto and M. africanum which are the main agents of TB in humans, and several variants adapted to various domestic and wild mammal species, including M. bovis, M. caprae, M. microti and M. pinnipedii [2]. MTBC relevant to human disease has been classified into seven main phylogenetic lineages [3]–[4]: Lineages 1 to 4 together with Lineage 7 are collectively known as M. tuberculosis sensu stricto, whereas Lineage 5 and 6 are also known as M. africanum West Africa I and II, respectively [5].
Because MTBC harbours limited genetic diversity compared to other bacteria [6], for a long time the assumption was that host and environmental factors were the only relevant determinants driving the course of TB infection. However, recent studies have challenged this dogma. Indeed, experimental infection models have shown that MTBC strains differ in virulence and immunogenicity [7], and epidemiological studies have demonstrated that in addition to host and environmental factors, strain diversity contributes to the variable outcome of TB infection and disease in clinical settings [8].
The MTBC lineages adapted to humans exhibit a strong phylogeographic population structure [4]. This together with the finding that the MTBC most likely originated in Africa and accompanied human migrations over millennia [9] has led to the proposal that the different lineages of human-associated MTBC might have locally adapted to different human populations [10]. Support for this notion comes from the observation that in metropolitan settings, MTBC lineages tend to transmit preferentially among sympatric (as opposed to allopatric) host populations [11], and that this sympatric host-pathogen association is perturbed by HIV co-infection [12].
Previous work showed that in Ghana, human TB is caused by six out of the seven MTBC lineages, with 20% of all cases attributed to Lineages 5 and 6 [13] (i.e. M. africanum West-Africa I and West-Africa II, respectively). M. africanum is highly restricted to West-Africa for reasons unknown [5], [10]. One possibility could be that M. africanum has adapted to particular human populations in that region of the world. To address this possibility, we performed a retrospective molecular epidemiological study of MTBC in Southern Ghana. We combined bacterial genotyping with detailed demographic and epidemiological patient data and sought for associations between host factors and the main MTBC lineages prevailing in Ghana.
All study protocols including oral and written consent format used for this study were approved by the Institutional Review Board (IRB) of the Noguchi Memorial Institute for Medical Research, Legon-Ghana (NMIMR; Federal wide Assurance number FWA00001824) and from the Ethikkommission Beider Basel (EKBB) in Basel, Switzerland. The standard procedure for sampling as outlined by the National Tuberculosis Program (NTP) for the routine management of TB in Ghana was used in the study. Written (in the case of literate participants) or oral (for illiterates) informed consent was sought from all participants before inclusion in the study. For minors (below sixteen years of age) consent was sought from their parents/guardians before enrolment into the study. In the case of minors between sixteen and eighteen years, in addition to parental consent, assent was sought from them before enrolment into the study. As per the guidelines of the IRB of NMIMR, information confidentiality was strictly adhered to. In addition, objectives and benefits of the study were explained to all participants.
The study was conducted from July 2007 to August 2011. All patients were diagnosed as sputum Acid-Fast-Bacilli-positive pulmonary TB cases by microscopy. The patients were recruited before treatment initiation from five main health facilities; Korle-Bu Teaching Hospital in the Greater Accra region, Agona Swedru Government Municipal Hospital, Winneba Government Hospital, St Gregory Catholic Clinic from the Central Region and Effia-Nkwanta Regional Hospital from Western Region of Ghana. Information on age, sex, nationality, ethnicity, employment status, previous history of TB, crowding, substance abuse and duration of symptoms were obtained from the patients with a structured questionnaire. Patients with missing information or culture-negative status were excluded from analysis. Ethnicity was classified in line with Ghana demographic data 2010 [14]. Patient origin was defined by place of residence.
Sputum samples obtained were decontaminated using 5% oxalic acid [15] and inoculated on two pairs of Lowenstein Jensen (LJ) slants; one supplemented with 0.4% sodium pyruvate to enhance the isolation of M. africanum and M. bovis, and the other with glycerol for the growth of M. tuberculosis sensu stricto. The cultures were incubated at 37°C and were read weekly for growth for a maximal duration of 16 weeks. MTBC strains were identified by detection of insertion sequence IS6110 as previously described [16]. Classification into the main phylogenetic lineages of MTBC was achieved by large sequence polymorphism typing identifying regions of difference (RD) [2] in a stepwise manner. Firstly, all isolates were screened for RD9. RD9-undeleted strains were further sub-typed for the “Cameroon” strain family (known to be most dominant Lineage 4 sub-lineage in Ghana) by screening for deletion of RD726 [11]. Isolates identified as RD9-deleted were subsequently sub-typed for Lineage 5 and 6 using RD711 and RD702 flanking primers, respectively. All lineages identified were confirmed by TaqMan real time PCR (TaqMan, Applied Bio systems, USA) using probes targeting lineage-specific SNPs as reported [17].
All MTBC isolates were typed by spoligotyping [18]. This was performed according to the manufacturer's instructions, using commercially available kits (Isogen Bioscience BV Maarssen, The Netherlands). Spoligotyping patterns were defined according to SITVITWEB database [19] (http://www.pasteur-guadeloupe.fr:8081/SITVIT_ONLINE). SITVITWEB assigned shared types numbers were used whenever a spoligotyping pattern was found in the database while families and subfamilies were assigned based on the MIRU-VNTRplus database (http://www. miru-vntrplus.org) [20]. Shared types were defined as patterns common to at least two or more isolates. All patterns that could not be assigned were considered orphan spoligotypes.
Information from the structured questionnaire was double entered using Microsoft Access and validated to remove duplicates and data entry inconsistencies. Multivariable logistic regression models were used to compare patient characteristics associated with M. africanum compared to M. tuberculosis sensu stricto. All statistical analyses were performed in STATA 10.1 (Stata Corp., College Station, TX, USA).
A total of 622 TB patients were included in this study. Age of patients ranged from 8 to 77 years with a median age of 35 years (Table 1). Overall, 208/622 (33.4%) were females with median age of 33 years; the remaining 414 (66.6%) were males with a median age of 36. Twenty-nine out of the 622 patients (4.6%) were children (age<16 years). Most patients originated from Greater Accra Region (325 cases, 52.3%), followed by 268 cases (43.1%) from Central Region with the remaining twenty-nine patients (4.6%) from Western Region of Ghana. Out of the 622 patients, 596 (95.8%) were Ghanaians, 21 (3.3%) were Liberians, 2 Togolese (0.3%) and 1 (0.2%) each of Nigerian, Ivorian and Gambian origin, respectively. Most of the patients were of Akan ethnicity (N = 427, 68.7%), followed by Ga (N = 104, 16.7%), Ewe (N = 71, 11.4%) with the remaining ethnicities accounting for 3.2% (N = 20). In terms of education, 436 patients (70.1%) were illiterates, 44 (7.1%) primary education, 132 (21.2%) had up to secondary education, and the remaining 10 (1.6%) tertiary education. More than half of the study population (N = 324, 52%) consumed alcohol on a regular basis, while only 44 (7%) smoked.
MTBC isolates were obtained from all 622 TB patients. Upon genotyping, 9 of these (1.4%) produced ambiguous results and were thus excluded from further analysis. Hence, a total of 613 isolates were used for further analysis. Based on LSP and SNP typing, we identified six out of the seven human-associated MTBC lineages in our study sample (Table 2). The dominant lineages were Lineage 4 with 483 cases (78.8%), Lineage 5 (N = 86, 14.0%) and Lineage 6 (N = 19, 3.1%). Eleven isolates (1.8%) belonged to Lineage 1, 10 to Lineage 2 (includes Beijing; 1.6%), and the remaining 4 isolates to Lineage 3 (0.7%). Among the 483 Lineage 4 isolates, 313/483 (65.0%) belonged to the sub-lineage of Lineage 4 known as the ‘Cameroon family’. No M. bovis was identified in our sample.
All isolates were further sub-typed using spoligotyping (Table 2). We detected a total of 117 spoligotypes, 485/613 isolates (79%) had previously defined shared type number (SIT). The remaining 128 isolates could not be defined by the SITVIT database and thus were defined as ‘orphan’. In addition to Cameroon sub-lineage, seven additional sub-lineages were identified among Lineage 4 based on spoligotyping; Ghana (N = 75, 15.5%), Haarlem (N = 37, 7.7%), Uganda I (N = 15, 3.1%), Uganda II (N = 7, 1.4%), LAM (N = 5, 1.0%), S (N = 4 (0.8%), and X (N = 2, 0.4%).
Table 3 illustrates the association of socio demographic and behavioural factors with the main MTBC lineages present in our study sample. Using multivariable logistic regression model analysis, we found that individuals of Ewe ethnicity were significantly more likely to present with TB caused by M. africanum as opposed to M. tuberculosis sensu stricto irrespective of their place of residence (adjusted odds ratio (adjOR) = 3.02; 95% confidence interval (CI): 1.67–5.47, P<0.001) (Table 3, S1 Fig.). This association was independent from other risk factors. Moreover, we found TB caused by M. africanum to be associated with smoking (adjOR = 2.02; 95% CI: 0.95–4.27) when compared to M. tuberculosis sensu stricto. However, this association was only borderline statistically significant (P = 0.07). No significant associations between MTBC lineages and other patient variables were found. Because M. africanum comprises two phylogenetic distinct lineages (i.e. MTBC Lineages 5 and 6), we performed a stratified analysis by lineage. Using multivariate logistic regression model analysis, we observed a significant association between Ewe ethnicity and Lineage 5 (adjOR) = 2.79; 95% CI: 1.47–5.29, P<0.001). This association was independent from other risk factors (Table 4). Interestingly, based on univariate analysis, we also saw an association between Ewe ethnicity and Lineage 6 (adjOR = 4.03; 95% CI: 1.33–10.85). However, because of the limited number of Lineage 6 isolates (n = 18) multivariate analyses could not be performed to confirm the independence of this association.
Our retrospective molecular epidemiological investigation of MTBC clinical isolates from Southern Ghana revealed that i) the Cameroon sub-lineage of Lineage 4 is the dominant cause of human TB in this region, ii) 17.1% of human TB is caused by M. africanum, iii) TB patients infected with M. africanum were more likely to smoke, and iv) to belong to the Ewe ethnic group.
Our finding that the Cameroon sub-lineage causes 65% of human TB in Ghana confirms our previous report from Ghana [13], and is in agreement with findings from neighbouring countries. In particular, the Cameroon sub-lineage was previously found to cause 40% of human TB in Cameroon [21], 45% in Nigeria [22] and 33% in Chad [23]. The reasons for the success of this sub-lineage in this region of Africa are unclear but could be due to a founder effect and/or particularly high fitness in the corresponding patient populations. Similarly, other successful sub-lineages of Lineage 4 have been observed in other regions of Africa, including Uganda [24] and Zimbabwe [25].
We found that in Ghana, M. africanum still accounts for 17.1% of all human TB, which is similar to the prevalence we reported several years ago [13]. This is in contrast to a study in Cameroon [21] that indicated a sharp decrease in TB caused by M. africanum during the last decades. A potential explanation for the decline of M. africanum in some West African countries includes possible out-competition by M. tuberculosis, as M. africanum has been associated with reduced virulence in animal models [26]–[27], and a longer latency and a slower rate of progression to active disease in humans [28]. Of note, our finding that smoking was associated with infection by M. africanum as opposed to M. tuberculosis sensu stricto is consistent with the notion that M. africanum might be less virulent in immunocompetent hosts [7]. This notion is also supported by a previous study in the Gambia reporting a significant association between M. africanum West Africa II and HIV co-infection [29]. However, no such association was found between M. africanum West Africa I and II in Ghana [30]. Because information on HIV status was not available for the present study, we could not explore this question here. Taken together, there is a need for further investigation to ascertain why M. africanum is declining in some regions of West Africa, but not in Ghana, and whether this phenomenon can be attributed to differences in virulence and/or other factors.
One reason for why the prevalence of M. africanum might be more stable in Ghana than in some other countries is that this bacterial lineage might be particularly well adapted to (some) human populations in Ghana. Our finding that M. africanum was independently associated with Ewe ethnicity supports this possibility. Moreover, this association was largely driven by Lineage 5, and not the result of a single outbreak as the spoligotyping patterns among M. africanum isolates from Ewe patients were diverse (Table 5). From available data, we know that M. africanum, in particular Lineage 5 is prevalent in countries around the Gulf of Guinea [13], [31], and particularly frequent in Benin and Ghana [13], [32], two countries with large Ewe populations [33]. The Ewe speaking ethnic group traditionally forms part of the Gbe language family which includes the Fons of Benin, the Aja of Togo and the Phla-phera of western Nigeria [33], [34]. Although the Ewe, Fons, Aja and phla-phera are different dialects of the same Gbe language family, members of theses individual groups are interrelated [33], [34]. Together they constitute the indigenous inhabitants of coastal West Africa.
Associations between particular MTBC lineages and human ethnicities have been observed before. For example, in San Francisco, Lineage 1, 2 and 4 were strongly associated with Filipino, Chinese, and “white” ethnicities, respectively [11]. More recently, Hui ethnicity was found to be associated with the Beijing family of MTBC in China [35]. While social “cohesion” is likely to restrict intermingling between individuals belonging to different ethnic groups and thus transmission of MTBC between these groups, biological factors could also play a role in the association between different MTBC genotypes and human populations. Self-defined ethnicity has been shown to be a reliable proxy for human ancestry [36], and human genetic diversity has been linked to an increased or reduced susceptibility to TB [37]. Importantly, recent studies indicate that human genetic susceptibility to TB is further influenced by the MTBC genotype [10]. In particular, studies have reported human genetic polymorphisms that influence the susceptibility to TB caused by M. africanum but not M. tuberculosis sensu stricto or vice versa [38]. For example, a study performed in Ghana reported a human polymorphism in 5-lipoxygenase (ALOX5) associated with increased TB risk [39]. Stratification by MTBC lineage revealed that this association was mainly driven by M. africanum indicating that this human polymorphism increases the risk of TB in a MTBC lineage-specific matter. ALOX5 is involved in the synthesis of leukotrienes and lipoxins, which are important mediators of the inflammatory response [39]. Conversely, a human polymorphism reported recently in the Mannose Binding Lectin (MBL) was associated with protection against TB caused by M. africanum but not M. tuberculosis sensu stricto [40]. Moreover, this latter study also found that M. africanum bound human recombinant MBL more efficiently, perhaps leading to an improved uptake of M. africanum by macrophages and selection of deficient MBL variants among human populations exposed to M. africanum [40].
Our study has several limitations. First, data on HIV co-infection was not available. This might have influenced our results on the patient characteristics associated with M. africanum. Secondly, this study was not population-based as patients were recruited only at three government hospitals. Hence, some degree of selection bias cannot be excluded.
In conclusion, our study provides novel insights into the interaction between environmental, host and pathogen variability in human TB. In particular, the observed association between M. africanum and Ewe patient ethnicity suggests a possible explanation for the geographical restriction of M. africanum to parts of West Africa. Our findings also highlight the need to consider this variability in the development of new tools and strategies to control TB.
|
10.1371/journal.ppat.1005154 | Heterosexual Transmission of Subtype C HIV-1 Selects Consensus-Like Variants without Increased Replicative Capacity or Interferon-α Resistance | Heterosexual transmission of HIV-1 is characterized by a genetic bottleneck that selects a single viral variant, the transmitted/founder (TF), during most transmission events. To assess viral characteristics influencing HIV-1 transmission, we sequenced 167 near full-length viral genomes and generated 40 infectious molecular clones (IMC) including TF variants and multiple non-transmitted (NT) HIV-1 subtype C variants from six linked heterosexual transmission pairs near the time of transmission. Consensus-like genomes sensitive to donor antibodies were selected for during transmission in these six transmission pairs. However, TF variants did not demonstrate increased viral fitness in terms of particle infectivity or viral replicative capacity in activated peripheral blood mononuclear cells (PBMC) and monocyte-derived dendritic cells (MDDC). In addition, resistance of the TF variant to the antiviral effects of interferon-α (IFN-α) was not significantly different from that of non-transmitted variants from the same transmission pair. Thus neither in vitro viral replicative capacity nor IFN-α resistance discriminated the transmission potential of viruses in the quasispecies of these chronically infected individuals. However, our findings support the hypothesis that within-host evolution of HIV-1 in response to adaptive immune responses reduces viral transmission potential.
| Despite the available HIV-1 diversity present in a chronically infected individual, single viral variants are transmitted in 80–90% of heterosexual transmission events. These breakthrough viruses may have unique properties that confer a higher capacity to transmit. Determining these properties could help inform the rational design of vaccines and enhance our understanding of viral transmission. We isolated the transmitted variant and a set of related non-transmitted variants from the transmitting partner near the estimated date of transmission from six epidemiologically linked transmission pairs to investigate viral correlates of transmission. The simplest explanation that transmitted variants are inherently more infectious or faster replicators in vitro did not hold true. In addition, transmitted variants did not replicate more efficiently than their non-transmitted counterparts in dendritic cells or in the presence of interferon-alpha in vitro, suggesting that they are not uniquely adapted to these components of the innate immune system. More ancestral genomes that were relatively sensitive to antibody neutralization tended to transmit, supporting previous reports that mutational escape away from the adaptive immune response likely reduces the ability to transmit. Our investigation into the traits of transmitted HIV-1 variants adds to the understanding of viral determinants of transmission.
| HIV-1 transmission is characterized by an extreme genetic bottleneck, the basis of which is unclear. Studies of both the highly diverse envelope glycoprotein [1–3] and full HIV-1 genomes [4] demonstrated that 80–90% of heterosexual transmissions are initiated by a single virus variant selected from the diverse viral quasispecies present in the chronically infected transmitting partner. These variants, which are different in each transmission event, have been named transmitted/founder (TF) viruses. Studying TF viruses could enhance our understanding of viral transmission and inform HIV prevention strategies.
The TF is rarely the dominant variant in the plasma or genital tract of the transmitting partner [5,6], which suggests that transmission is not entirely stochastic and may involve selection. A number of prior studies have identified distinctive properties of TF variants [4,7–19], particularly in analyses of the TF viral envelope (Env) glycoprotein. Reported characteristics of TF virus Envs include a selection for CCR5-tropism [2,20], a predominance of shorter and less glycosylated Env proteins [1,11,15,18,19], a preference for binding α4β7 [10,21] and a selection for more ancestral variants [8,22]. Although these studies observed selection of viral traits, others found that acute and chronic variants had similar characteristics. By generating infectious molecular clones (IMC) with the env genes from linked recipients and transmitting partners in a common viral backbone, acute and chronic donor viruses displayed similar CD4 and CCR5 requirements for cell entry, low macrophage tropism, and no preferential usage of alternative coreceptors [23,24]. Furthermore, studies of env only clones from acute infection compared with chronic control viruses have shown similar CD4 T cell subset tropism, low macrophage tropism, and a lack of effect of blocking α4β7 on infection [25].
Selection of viral traits outside of the env gene has also been observed during heterosexual transmission. We recently described a selection bias during transmission for more consensus-like HIV-1 variants, in gag, pol and nef genes, from a cohort of 137 subtype C infected epidemiologically-linked transmission pairs [7]. This study suggested that in vivo fitness of consensus-like HIV-1 variants increased their likelihood of transmission [7]. Studies of full-length infectious molecular clones of TF viruses, in comparison to control viruses derived from chronic infection, have also demonstrated increased particle infectivity, as well as an enhanced resistance to interferon-α (IFN-α) in TF viruses [13,17].
While informative, conclusions of these previous studies are limited in that only individual genes were examined, or corresponding non-transmitted (NT) variants from the transmitting partner were unavailable as controls. HIV-1 IMC with the full complement of HIV-1 proteins have not been generated from both partners of transmission pairs nor evaluated for genetic and phenotypic signatures during transmission. Characterizing TF variants in comparison to NT variants from epidemiologically-linked partners could provide further insight into the viral requirements of HIV-1 transmission, potentially leading to new targets for intervention.
Here, we describe genetic and phenotypic comparisons of full-length genome TF and NT variants from six subtype C epidemiologically-linked heterosexual transmission pairs. We amplified and sequenced near full-length HIV-1 genomes by single genome amplification (SGA) to assess genetic selection during transmission. In addition, we cloned the complete TF genome along with a representative panel of NT variants. These clones were used to assess the relative in vitro fitness of TF variants as measured by particle infectivity, neutralizing antibody resistance, replicative capacity in PBMC and dendritic cells, as well as IFN-α resistance. We found a strong selection bias toward consensus sites across the entire genome, at both the amino acid and nucleotide level, in all six pairs. The TF variants were also more sensitive to neutralization by donor antibodies than NT variants. However, no evidence was found for TF variants exhibiting increased particle infectivity, replicative capacity, or IFN-α resistance when compared to the transmitting partner’s NT variants. Thus, in these six subtype C transmission pairs the transmission potential of TF variants is not discriminated by inherent in vitro replicative capacity or interferon resistance, and may be determined by alternate phenotypes difficult to dissect in these in vitro systems.
Full-length genome HIV-1 variants derived from linked transmission pairs have yet to be evaluated for characteristics associated with transmission. To define whether TF variants exhibit distinct properties, we compared them to their NT counterparts in six heterosexual epidemiologically-linked transmission pairs. We selected five female-to-male and one male-to-female therapy-naïve subtype C epidemiologically-linked transmission pairs from the Zambia-Emory HIV Research Project (ZEHRP) based on the availability of plasma samples at the nearest time points following transmission (average 28 days post estimated date of infection) (Table 1). We PCR amplified, using a high-fidelity polymerase, and sequenced a total of 167 HIV-1 near full-length single genome amplicons as described previously [26]. All six linked recipients were in Fiebig Stage II of infection, and were infected with a single variant from the donor quasispecies, as demonstrated by star-like phylogeny in a median of 8 near full-length genome amplicons per sample [2,4]. This allowed us to infer an unambiguous consensus TF sequence from the genetically homogeneous population of sequences in each linked recipient.
For phylogenetic analyses, we aligned full-length nucleotide sequences as well as concatenated full proteome amino acid sequences of 115 HIV-1 single genomes (each TF virus represented by a single consensus sequence), with the HIV-1 consensus/ancestral alignment from the Los Alamos National Laboratory (LANL) HIV database. We generated maximum likelihood trees of the full-length genome and proteome alignments for all six transmission pairs, and confirmed that all pairs were epidemiologically linked, since each TF variant fell clearly within the branches of the linked donor virus variants. Each transmission pair clustered independently on the phylogenetic tree with bootstrap values of 100 (Fig 1). All six linked donor partners were chronically infected and demonstrated viral diversity in their plasma near the time of transmission (Fig 1).
We previously demonstrated a consistent transmission bias for variants with consensus-like amino acid residues across the Gag, Pol and Nef proteins by population sequencing in a cohort of 137 epidemiologically-linked subtype C transmission pairs [7]. Although this finding has been shown for the gag and env genes independently, it has not been confirmed by full-length genome SGA from the transmitting partner's quasispecies [7,8]. We examined the selection bias for more consensus-like viruses by measuring the pairwise distance (branch length), of each viral variant to the LANL subtype C consensus node on the full-length nucleotide and amino acid phylogenetic trees (Fig 1). TF variants had a significantly shorter pairwise distance to the subtype C consensus node than the median of their corresponding NT variants for both nucleotide (Fig 2A; p = 0.0156) and amino acid (Fig 2B; p = 0.0469) sequences. These transmission pairs confirm, as previously described, a selection bias for consensus-like amino acid and nucleotide sites across the viral genome during transmission.
In a previous study, TF virions exhibited enhanced infectivity in comparison to chronic control viruses on TZM-bl cells [13]. To test particle infectivity within transmission pairs, we generated full-length IMC for 40 viral variants, including the 6 TF variants and 3–8 NT variants from each chronically infected transmitting partner, as described previously [26]. We selected variants to represent the genetic diversity present in the donor near the time of transmission (Fig 1), and confirmed that the IMC and amplicon sequences were identical by whole genome sequencing. We also excluded the rare sequences that contained gross genetic defects, such as large deletions and frameshift mutations in gene coding regions. For each IMC, we generated virus stocks by transfection of 293T cells.
We defined particle infectivity as the ratio of infectious units, as measured by the virus titer on TZM-bl cells, a standard reporter cell line whose permissivity correlates with that of PBMC [27], over total amount of virions, measured by reverse transcriptase activity of the virus stock. We confirmed that the particle infectivity of a subset of virus stocks generated from 293T cells and harvested 48 hours after transfection (for consistency, as particle infectivity decreased over time post-transfection, S1A Fig) correlated with the particle infectivity of virus stocks generated from PBMC 8 days following infection (S1B Fig, p < 0.0001, r = 0.9455). Analysis of the particle infectivity of virus stocks produced from all of the infectious molecular clones showed that the particle infectivities of all viruses tested ranged from 7x10-5 to 1x10-2, and that there was also a wide range of particle infectivities within each transmitting partner’s quasispecies (Fig 3). In pair 3678, the TF variant was the most infectious virus compared to the rest of the transmitting partner’s variants, while the TF from pair 3576 was the least infectious (Fig 3). TF variants spanned the thousand-fold range of particle infectivities measured for all the viruses tested, as can be seen by the TF from pairs 3618 and 4473, which are found on extreme ends of the particle infectivity spectrum. Across all six transmission pairs, we observed no significant selection for infectivity when comparing the TF to the median of the transmitting partner’s quasispecies (Fig 3; p = 0.6875). In these subtype C transmission pairs particle infectivity did not constitute a dominant determinant of transmission fitness.
We previously reported that Env glycoproteins derived from early viruses in acutely infected linked recipients were on average more sensitive to neutralization by plasma from the transmitting partner, compared to autologous Envs directly derived from the transmitting partner [1]. Antibody neutralization of SGA-derived genome length TF and NT variants, derived from the first month of infection, from heterosexual epidemiologically-linked transmission pairs, has not been examined to date. Using a previously described TZM-bl neutralization assay [1,28], we evaluated neutralization of full-length TF and autologous NT IMC by plasma from the transmitting partner near the time of transmission. Donor plasma (diluted 1:100) demonstrated relatively weak neutralization against the majority of viruses tested in each panel, with a median of 18% neutralization. The highest level of neutralization was seen in pair 4473 against the TF (51%) (Fig 4A). Overall, TF variants were more efficiently neutralized compared to the medians of the transmitting partner’s NT variants (Fig 4A; p = 0.031). Additionally, greater neutralization negatively correlated with distance to the amino acid subtype C consensus (Fig 4B; p = 0.011, r = -0.4995), suggesting a link between these two measurements. Consistent with our previous findings, transmission did not select for TF variants with greater neutralizing antibody resistance to donor plasma.
Selection for consensus-like TF variants in these six transmission pairs may indicate a selection for viruses with greater in vivo fitness, as hypothesized from a study of 137 linked transmission pairs [7]. To determine whether this translated into a similar fitness advantage in activated CD4 T cells, we measured the replicative capacity (RC) of viruses in vitro. TF and NT IMC were tested for in vitro replication by infection of stimulated peripheral blood mononuclear cells (PBMC), at equal multiplicities of infection (MOI). Since the number of infectious particles to total particles varied greatly between all virus stocks tested, we based the amount of virus used for each replication experiment on a consistent MOI (0.01), rather than equal amounts of virus particles, in order to normalize for initial infectivity. We measured virus growth by reverse transcriptase activity of cell culture supernatants every 48 hours for ten days (Fig 5A). RC scores were generated for each virus based on the area under the curve of virus growth, as described in the methods. TF viruses exhibited a wide range of RC among all the viruses tested, and the relative RC of TF as compared to NT viruses from the same donor also varied substantially (Fig 5B). For instance, the TF from pair 3576 had the lowest RC when compared to the transmitting partner’s quasispecies. Alternatively, pairs 3618 and 3678 had TF viruses with relatively high RC, although they were not the highest replicators from their transmitting partner’s quasispecies (Fig 5B). In total, we saw no significant selection for TF viruses having higher in vitro RC than the median RC of the NT viruses tested (Fig 5C; p = 0.219). Similar to particle infectivity, which correlated with RC over all the viruses tested (S2 Fig; p = 0.0005, r = 0.5712), there was no evidence for a distinct replicative capacity profile associated with transmission. In addition, viruses closer to consensus typically had lower in vitro replicative capacities, since the pairwise distance to subtype C consensus correlated with in vitro replicative capacity (Fig 5D; p = 0.0158, r = 0.4168).
Since dendritic cells have also been implicated as an initial target cell for establishment of HIV-1 infection in the genital mucosa [29], we examined the ability of the 6 TF and a limited set of 6 NT variants with similar in vitro RC scores, to productively infect and replicate in immature monocyte derived dendritic cells (MDDC) in vitro. We cultured MDDC by isolation and differentiation of blood-derived CD14+ monocytes from healthy donors and infected them with virus at a high MOI of 1. We assayed virus production by measuring the reverse transcriptase activity present in cell culture supernatants every 48 hours for twelve days. We found that the TF and NT variants studied did not significantly differ in their ability to replicate in MDDC (S3 Fig; p = 0.87). Of the twelve TF & NT variants, six had detectable replication in MDDC (3 TF & 3 NT), suggesting that productive infection of MDDC is limited, even at a high MOI, and is not a requirement for transmission. Overall, these data suggest that HIV-1 transmission is permissive to TF variants with a wide range of in vitro replicative capacities relative to the transmitting partner’s quasispecies.
By conducting in vitro replication assays in cells pre-treated with exogenous interferon-α (IFN-α), previous studies found that subtype B and subtype C TF variants were relatively resistant to IFN-α compared to a panel of chronic viruses [13] or later variants from the same individual [17]. These studies suggested a selection during the HIV-1 transmission bottleneck for variants adept at escaping innate immunity, specifically the antiviral effects of IFN-α. However, these studies were not done in epidemiologically-linked transmission pairs, and thus were unable to directly compare TF viruses to related NT variants in the donor quasispecies near the time of transmission. To test whether the subtype C TF viruses investigated here exhibited relative resistance to IFN-α, as compared to NT variants derived from the transmitting partner's quasispecies, we assayed in vitro virus replication in PBMC in the presence and absence of IFN-α.
We assayed viral replication in activated CD8-depleted PBMC in the presence and absence of 5,000 U/ml of IFN-α, which was added 24 hours prior to infection in order to maximally inhibit viral replication, as described previously [17]. Supernatant HIV-1 p24 antigen levels were measured every 48 hours for 10 days to assess the kinetics of viral replication. In the initial 21 variants tested, growth of virus in the presence of IFN-α was tightly correlated with in vitro RC scores in the absence of IFN-α (Fig 6A; p < 0.0001, r = 0.8844), suggesting that in vitro growth in the presence of IFN-α was largely determined by viral replicative capacity.
In light of this, we attempted to delineate subtle differences in IFN-α resistance by performing further experiments with selected NT variants that exhibited relatively similar replication kinetics to the TF in each pair (to minimize the impact of replication differences). The replication of these selected viruses was assessed in activated PBMC in the presence and absence of 1,000 U/ml of IFN-α (added 24 hours prior to infection), monitoring virus replication by reverse transcriptase activity in the supernatant. An example of such an assay for transmission pair 331 is shown in Fig 6B. When compared to the tested NT variants, TF viruses did not differ significantly in resistance to IFN-α (assessed as the ratio of the RC score in the presence and absence of IFN-α) (Fig 6C, p = 0.219). In pair 331 and 4473, the TF appeared to be more IFN resistant than the NT viruses from the same donor (Fig 6C). In pair 3618 the TF was near the median of the NT variants, while in three pairs (3576, 3678, and 4248), the TF was the most sensitive to IFN-α. Overall, the IFN-α resistance of the TF viruses did not differ significantly from the median of the NT variants (Fig 6C).
Because the TF viruses were not found to be more IFN-resistant than donor NT viruses, we validated the method used for analysis of IFN resistance with 3 subtype B TF and 6-month consensus virus pairs that had previously been demonstrated to differ in their IFN-α resistance [17]. As shown in Fig 6C, the 3 TF viruses were each confirmed to be more IFN-resistant than the matched 6-month virus from the same subject, verifying the ability of the methods used here to detect previously documented differences in viral IFN resistance. In the six subtype C epidemiologically-linked transmission pairs studied we also observed that IFN-α resistance correlated with the virus' ability to replicate (S4A Fig). Although the RC and IFN-α resistance of the six subtype B TF and 6-month viruses was not statistically correlated, these subtype B TF viruses did have higher RC scores than their matched 6-month variants (S5 Fig). Overall, these data suggest that a component of IFN-α resistance is the ability of TF and NT HIV-1 variants to replicate.
To confirm that this finding was independent of the amount of IFN used to inhibit viral growth, we measured replication at day 7 for four viruses with a representative range of RC scores using a range of IFN-α concentrations (0.5 U/ml–10,000 U/ml). The relative sensitivity of these viruses was consistent across the range of IFN-α concentrations tested (S4B Fig). Additionally, we tested a limited subset of TF and NT variants for their ability to induce IFN-α, which may have influenced IFN-α resistance measurements, and found that IFN-α levels above background were not detectable at day 8 in either PBMC or MDDC infected cultures (S4C Fig). Hence HIV-1 transmission from these donors was not mediated by TF viruses that exhibited higher levels of interferon resistance than NT viruses, indicating that heterosexual HIV-1 transmission is permissive to viruses anywhere within the range of in vitro interferon resistance profiles observed in the donors studied here, and factors other than IFN-α resistance constituted the dominant determinants of transmission fitness in these pairs.
The rapid within-host diversification of HIV-1 observed during chronic infection, which represents a primary obstacle to effective HIV prevention strategies, contrasts starkly with the viral homogeneity evident following transmission. The stringent genetic bottleneck is most pronounced in heterosexual transmission, where a vast majority of new infections are established by single viral variants. Correlates of transmission may become evident by studying the properties of these transmitted/founder (TF) variants, which, in turn, could help inform effective HIV-1 vaccine design. Studies of early and transmitted variants have found genetic and phenotypic signatures associated with transmission; however, none have examined full-length TF variants and corresponding non-transmitted (NT) variants present near the time of transmission from epidemiologically-linked transmission pairs.
In this study, we applied new molecular techniques to investigate the requirements of HIV-1 transmission in six subtype C transmission pairs. We amplified, sequenced, and generated infectious molecular clones (IMC) of matched full-length TF viruses very early after infection (Fiebig stage II) and near full-length NT variants from 22–45 days following the estimated date of transmission. Technical limitations associated with amplifying full-length virus from genital tract samples required us to amplify from patient plasma. Despite this limitation, we previously showed in eight epidemiologically-linked transmission pairs that the TF was most highly related to NT variants that were absent from the predominant genital tract subpopulations, and were found in both blood and genital tract of the donor partner [5]. Consistent with this, in pair 331 we observed a NT variant in the plasma of the transmitting partner with only three amino acid differences from the TF across the entire proteome (Fig 1B).
In this study, we generated IMCs from the diverse donor quasispecies with great sequence accuracy and selected variants in an unbiased fashion. Five of the six pairs were female-to-male, the route by which the most stringent bottleneck occurs [7,30]. Since high donor viral load and the presence of genital ulcers and inflammation (GUI) in the recipient can, to a certain degree, mitigate selection bias in the bottleneck, it is important to note that these six pairs include three donors with viral loads >100,000 RNA copies/ml, as well as one recipient with a reported GUI in the twelve months prior to seroconversion [3,7]. Despite these caveats, single variant transmission was observed in all six pairs.
Consistent with our previous findings [7], we observed selection during transmission for variants with more consensus-like amino acid and nucleotide DNA sequences from the available quasispecies present in the donor at the time of transmission, across the full viral proteome and genome, respectively (Fig 2). It has been shown that HIV-1 within-host diversity during chronic infection is greater than between-host diversity, suggesting conservation of certain genetic elements during transmission [31]. In conjunction, studies of subtype A and D heterosexual transmission pairs demonstrated transmission of more ancestral viral variants, by measuring distances of each variant to their most recent common ancestor (MRCA) on a phylogenetic tree of Env sequences [8,22]. In the current study, the LANL subtype C consensus node falls near the subtype C MRCA highlighting the equivalence of these two measurements (Fig 1). Thus, HIV-1 transmission consistently selects for variants that more closely resemble ancestral and consensus-like viruses, indicating that evolution in the host decreases transmission potential.
The viral diversification observed during chronic infection due to adaptive immune pressure targeted specifically against HIV-1 is likely driving viral evolution away from consensus [1,7,32,33]. We have previously shown that acquisition of resistance to antibody neutralization comes with a transmission fitness cost [1]. We similarly found that TF viruses were more sensitive to neutralization by donor plasma acquired near the time of transmission when compared to the corresponding NT variants. It should be noted, however, that a limitation to this finding is that NT variants were cloned and tested with plasma from approximately four weeks after the estimated date of infection, although the two pairs with the largest time gap between transmission and sampling did not show the greatest neutralization of the TF. These data also reaffirm that TF variants are generally not resistant to antibody neutralization [1,34]. As expected, donor plasma tested against contemporaneous viruses (TF or NT) demonstrated limited neutralization capacity. Moreover, neutralization sensitivity correlated with the distance to consensus over all the viruses tested. Considering these observations, it is reasonable to propose that selection of antibody sensitive variants during transmission is a side effect of the transmission cost associated with non-consensus adaptations in general, and not an underlying mechanism of transmission itself.
In order to address the role of viral fitness in transmission, we measured the in vitro fitness of a subset of viruses from six transmission pairs. Although a previous study found that TF viruses were more infectious than chronic control viruses [13], we found no bias towards increased infectivity when comparing the TF to the corresponding NT variants. Particle infectivity in TZM-bl cells correlated with replicative capacity in PBMC, suggesting that entry into TZM-bl cells is representative of a component of viral replication in primary cells. Viral replicative capacity in activated PBMC, a fundamental measure of in vitro fitness, was also not higher for TF variants in comparison to the corresponding NT variants, and none of the TF variants exhibited the highest replicative capacity from among the tested NT variants. We found that more consensus-like variants, which are typically those that transmit, had lower in vitro replicative capacities over all the variants tested, indicating that higher in vitro replicative capacity is not linked to transmission. TF variants were also not observed to have enhanced replicative capacity in monocyte-derived dendritic cells, an in vitro model for dendritic cells, which may act as an initial target cell for establishment of HIV-1 infection. These findings argue against the original hypothesis that consensus-like variants would have higher in vitro replicative capacities. Thus, in vitro RC in activated PBMC or MDDC may not reflect in vivo transmission fitness, potentially because replication in stimulated PBMC may recapitulate the inflammatory environment that occurs some time after transmission and during chronic infection rather than conditions initially encountered at initial sites of virus replication. We cannot rule out the possibility that replication assays in cell types more representative of mucosal transmission, such as tissue resident CD4+ T cells or Langerhans cells, may yield different results. However, consistent with our observations, previous studies found a significant negative correlation between similarity to consensus and in vitro RC in a larger number of patients in differing cohorts using gag-chimeras [35,36]. Transmission of low in vitro fitness variants may seem counterintuitive; however, full-length TF IMC as well as over 200 transmitted Gag chimeras have been shown to exhibit a wide range of in vitro replicative capacities [35,37,38], as we found for our six TF viruses. A recent theoretical model of HIV transmission predicted that variants with lower replicative capacity via increased latency would exhibit a greater transmission potential in vivo [39], and it is therefore possible that modestly lower in vitro replicative capacity is an advantage during transmission.
A potential selection factor during mucosal transmission is the early innate immune response to HIV-1. Innate antiviral cytokines including IFN-α are induced at initial sites of HIV-1 replication in the mucosa and draining lymph nodes [40,41], hence HIV-1 variants that are more resistant to the antiviral effects of IFN-α may have an advantage during transmission. Indeed, cross-species transmission of Simian Immunodeficiency Virus (SIV) to humans required escape from the interferon stimulated APOBEC3 restriction factors by enhanced Vif antagonism [42]. A recent in-depth study using the rhesus macaque model also found that IFN-α treatment prior to intrarectal SIVMAC251 inoculation reduced the number of transmitted variants and increased the number of challenges necessary to initiate infection [43]. Consistent with the hypothesis that type 1 IFNs contribute to the transmission bottleneck, previous studies using HIV-1 found that TF variants are generally more resistant to IFN-α in vitro than viruses present during early chronic infection [13,17]. Fenton-May et. al. [17] found that TF viruses from both subtype B and C infected subjects were more resistant to IFN-α when compared to matched variants generated from the same individual six months post-infection or during early chronic infection. Parrish et. al. [13] found that TF viruses are more resistant to IFN-α than viruses from unmatched chronic controls, though this was true only for the subtype B and not for the subtype C variants they studied. In six subtype C transmission pairs studied here we did not observe that TF viruses exhibited enhanced resistance to IFN-α compared to NT viruses. TF variants did not replicate to higher levels in the presence of IFN-α, nor did they have higher ratios of replication in the presence versus the absence of IFN-α.
These differing results could be due to differences in experimental protocols, as well as difficulties in separating inherent replicative capacity from interferon resistance. We therefore tested the IFN-α resistance of previously studied TF and 6-month viruses and confirmed that these TF variants were more resistant to the effects of IFN-α, consistent with previous observations. In addition, we found that the TF variants had higher replicative capacities than the 6-month consensus variants, although for this group of viruses IFN-α resistance did not directly correlate with viral replicative capacity. The influence of viral replicative capacity on measures of interferon resistance is not fully understood. The impact of multiplicity of infection on measured interferon resistance has been noted previously [44], so in the current studies we utilized a low multiplicity to ensure adequate target cell availability even for the higher replicating viruses. We chose a MOI of 0.01 for our assays since it represented an input virus dose at which we were able to measure both replication differences between viruses, as well as IFN-α resistance differences (S4D Fig). For the viruses tested from the six Zambian transmission pairs, we found that in vitro replication in the presence of interferon correlated with replication in the absence of IFN-α, such that the rank order of virus replicative capacities from lowest to highest was similar in the presence or absence of interferon. Even when a subset of viruses with more closely-matched levels of replication were studied so that we were better able to observe IFN-α resistance differences, we found that TF variants were not IFN-α resistant compared to the matched NT variants.
The lack of difference in the IFN resistance of TF and NT viruses in these transmission pairs may be due to the length of time for which the chronically infected viral donors had been infected prior to viral transmission to their partners and derivation of the viruses studied. Fenton-May et. al. showed that while IFN-α resistance decreased over the first 6 months following infection, it subsequently increased in different subjects at timepoints from 2–7 years post-infection [17]. Edlin et. al. and Kunzi et. al. further showed that viruses isolated from individuals who had progressed to AIDS were more IFN resistant than viruses from asymptomatic chronically-infected individuals [45,46]. Likewise, Parrish et. al. proposed that their observation of differences in IFN resistance between TF IMCs and IMCs from unmatched chronically-infected subjects in a subtype B-infected cohort, but not in a subtype C-infected cohort, may have been due to the subtype C-infected donors being sampled at later timepoints in chronic infection [13]. However, it should be noted that we did observe a range of interferon sensitivities across the six transmission pairs, with greater than a 100-fold difference being observed between TF viruses. In future, it would be of interest to determine whether chronically-infected donors in the Zambian discordant couples cohort who failed to transmit infection to their partners harbor more IFN-sensitive viruses than those present in the virus-transmitting donors studied here. However on the basis of the current results it seems likely that IFN-α does not make a major contribution to the HIV-1 transmission bottleneck, or may do so only in some transmission scenarios.
Transmission selection for consensus-like and more neutralization-sensitive TF variants suggests that within-host evolution of HIV-1 in response to human adaptive immune responses may cause a loss of fitness required for the establishment of infection in a naive host following transmission. We show that relatively high in vitro replicative capacity and preferential IFN-α resistance were not selected for during transmission of subtype C HIV-1 in the six pairs studied here. Thus, the in vitro assays of HIV-1 replication employed here may not be measuring some of the key determinants of transmission fitness, and other models of HIV transmission, such as low dose intravaginal challenges of humanized mice, or human genital explant cultures, may be needed to determine the phenotypic requirements of HIV-1 transmission that genetic differences are pointing to.
The six HIV-1 subtype C transmission pairs investigated in this study were enrolled in the heterosexual discordant couple cohort at the Zambia-Emory HIV Research Project (ZEHRP) in Lusaka, Zambia. Human subjects protocols were approved by both the University of Zambia Research Ethics Committee and the Emory University Institutional Review Board. HIV-1 serodiscordant couples in this cohort were provided counseling and testing on a monthly basis prior to the negative partner becoming HIV-1 positive. The recipients were enrolled in the International AIDS Vaccine Initiative (IAVI) Protocol C early-infection cohort. Epidemiological linkage was defined by phylogenetic analyses of HIV-1 gp41 sequences from both partners [47]. All individuals in this study were ART naive during the time of sampling.
Viral RNA extraction and near full-length genome single genome amplification were performed as described in Deymier et al. 2014 [26]. Briefly, viral RNA was extracted from 140μl of plasma using the QIAamp Viral RNA mini kit (Qiagen) and was used for cDNA synthesis carried out with Superscript III (Life Technologies) and an anchored Oligo(dT)18 primer. The cDNA was used immediately for PCR amplification. Near full-length single genome PCR amplification was performed by serially diluting cDNA, followed by two rounds of PCR amplification, so that ~30% of wells became positive. Both rounds of PCR were performed in 1x Q5 Reaction Buffer, 1x Q5 High GC Enhancer, 0.35 mM of each dNTP, 0.5 μM of primers and 0.02 U/μl of Q5 Hot Star High-Fidelity DNA Polymerase (NEB) in a total reaction volume of 25 μl. First round primers were, 1U5Cc and 1.3’3’PlCb, and second round primers were 2U5Cd and 2.3’3’plCb [48]. Cycling conditions for both reactions are 98°C for 30s, followed by 30 cycles of 98°C for 10s, 72°C for 7.5min, with a final extension at 72°C for 10min. PCR reactions were run on a 1% agarose lithium acetate gel at 300 V for 25 min in order to determine the presence of a 9 kb band.
Positive ~9kb single genome amplicons were gel-extracted using the Wizard SV Gel and PCR Clean-Up System (Promega). Purified ~9 kb PCR amplicons were sent for sequencing to the University of Alabama Birmingham (UAB) sequencing core for Sanger sequencing.
In conjunction, multiple amplicons from recipient 3576 were sequenced by single-molecule nucleic acid sequencing (Pacific Biosciences), to confirm the TF [49]. Briefly, SMRTbell libraries were constructed according to the manufacturer's instructions for 10kb amplicons. PCR reactions of DNA amplicons were purified using Wizard SV Gel and PCR Clean-Up System (Promega) and mixed at equal concentrations to a total of 3ug DNA. Library preparation quality was assessed on a Bioanalyzer and SMRT sequencing on the PacBio RSII was performed following primer annealing and P4 polymerase binding to the library preparations. The consensus of the reads, aligned to the HXB2 reference sequence, were then taken to form a TF sequence, which matched the Sanger sequence.
All 9kb viral sequences were aligned in Geneious bioinformatics software (Biomatters, Aukland, NZ) using MUSCLE [50], followed by hand aligning. The Los Alamos National Database HIV Consensus/Ancestral Sequence Alignments were used as reference sequences (http://www.hiv.lanl.gov/content/sequence/NEWALIGN/align.html). Phylogenetic trees were generated using the DIVEIN web server (http://indra.mullins.microbiol.washington.edu/DIVEIN/) [51]. Phylogenetic analyses were performed by maximum likelihood parsimony under Phylogeny/Divergence/Diversity. For nucleotide sequence analysis a general time reversible model was used, with a fixed gamma distribution parameter of 1, and performed with 100 bootstraps. Amino acid phylogenetic analysis was performed using the HIVw model of evolution, with 100 bootstraps [52]. Pairwise distances from each branch node to the subtype C consensus node were extracted from the distance matrices of the phylogenetic trees.
HIV full-length genome infectious molecular clones were generated as described in Deymier et al. 2014 [26]. Briefly, linked recipient specific primers were generated in order to amplify the full long terminal repeat (LTR) from the linked recipient white cell pellet DNA. This LTR was cloned into a pBluescript vector, and the TF sequence of the LTR sequence was inferred as the consensus sequence from multiple clones. Subsequently, a three-piece DNA HD In-Fusion HD cloning (Clontech) ligation reaction using a reamplified clonable near-full length amplicon and two LTR pieces generated by PCR from the linked recipient LTR generated the full-length IMC. TF IMC were correct for the entire genome, whereas NT variants were chimeric for only for the R region of both 5’ and 3’ LTR, which was taken from the TF of that transmission pair. IMC were sequenced in order to confirm a match to the sequence of the single genome amplicon from which it was derived.
293T (American Type Culture Collection) cells were transfected with 1.5μg of plasmid DNA, using the Fugene HD transfection reagent (Roche) according the manufacturer’s protocol. Viral stocks were collected 48 hours post transfection and clarified by centrifugation. These virus stocks were then titered for infectivity on TZM-bl cell, as described previously [53]. The virus stocks were also measured for reverse transcriptase (RT) activity using a radiolabeled reverse transcriptase assay [53]. Particle infectivity of each virus was determined as the ratio of titer (infectious units/μl) over RT signal (RT/μl) for 3 independent experiments. Particle infectivity over time was measured by sampling 8ul (0.4%) per time point over a 3 day period.
Frozen peripheral blood mononuclear cells (PBMC) from buffy coats were thawed and stimulated with 20 U/ml of interleukin-2 (IL-2) and 3ug/ml of phytohemagglutinin (PHA) in R10 (Roswell Park Memorial Institute (RPMI) 1640 Medium supplemented with 10% defined fetal bovine serum (FBS), 1 U/ml penicillin, 1ug/ml streptomycin, 300ug/ml L-glutamine) for 72 hours at 37C. After 48 hours, 1,000 IU/ml of interferon-α2a (Sigma Aldrich, Product # SRP4594) was added to a portion of cells 24 hours prior to infection. 1×106 cells were then infected in 15ml conical tubes by 2 hour spinoculation at 2,200 rpm with an MOI of 0.01 based on the TZM-bl titer in triplicate. Cells were then washed twice in 13ml RPMI, resuspended in 500ul of R10 media and plated in a 48 well plate in triplicate. 50ul of supernatant was then sampled every 48 hours starting with a day zero time point taken 2–3 hours after plating to get a baseline reverse transcriptase activity for each infection well using the radiolabeled reverse transcriptase assay.
Where noted in the text, an alternative strategy for another independent experiment with CD8-depleted PBMC was used with a few differences: anti-CD3 (R&D Systems clone UCHT1; 50ng/ml working concentration) and anti-CD28 (eBioscience clone CD28.2; 100ng/ml working concentration) antibodies were used to stimulate MACS microbead (from Miltenyi plus the MACS LD columns) CD8-depleted PBMC from three separate donors in a mixed lymphocyte reaction and then infected at an MOI of 0.1 based on TZM-bl titer. 2×105 cells were then infected in the presence and absence of 5,000 IU/ml of interferon-α2a (Peprotech) and cells were washed three times with 10ml of RPMI and supernatant tested by a modified ELISA assay using the AlphaLISA HIV p24 (high sensitivity) kit (Product # AL291C PerkinElmer) per protocol instructions, using the same media for the standard as in the sample and loading 5ul per well.
The replication score (RC score) for each variant was calculated using a normalized area under the curve. The median of the replicates were background subtracted using the day 2 time point, adjusted for sampling by a measured exponential decay correction, and area under the curves (AUC) were divided by the AUC for a standard lab adapted subtype C virus, MJ4, to compare across transmission pairs analyzed on different days. Interferon-α2a resistance was measured in a similar fashion, followed by calculating the ratio of the RC score in the presence of interferon divided by the RC score in the absence of interferon.
Monocyte derived dendritic cells (MDDC) were isolated from two healthy blood donors by CD14 positive bead isolation (Miltenyi Biotec), followed by culture at 37°C in R10, supplemented with 40 ng/ml IL-4 (Peprotech) and 20ng/ml GM-CSF (Peprotech) for 7 days. MDDC differentiation was confirmed by flow cytometry using the following antibodies and stains: α-CD14 PB (clone M5E2), α-CD11c APC (clone S-HCL-3), α-HLA-DR V500 (clone G46-6) (BD Biosciences), and the LIVE/DEAD Fixable Near-IR Dead Cell Stain Kit (Life Technologies). The phenotype of MDDC after 7 days was CD14 low, CD11c high, and HLA-DR high, as expected. Cells were harvested, and 3x105 MDDC were seeded in a flat-bottom 96-well plate. MDDC were infected in a volume of 250μl of R10 with virus at an MOI of 1 for 4 hours. Cells were then washed three times with RPMI, and cultured for 12 days in R10 supplemented with 40 ng/ml IL-4 and 20ng/ml GM-CSF. 50μl of culture supernatant was collected every two days and replaced with fresh media. The supernatant was then analyzed for virus production by the radiolabelled RT assay [53].
IFN-α levels were measured by the VeriKine Human IFN Alpha ELISA Kit from supernatants 8 days after PBMC and MDDC infections with a subset of viruses from pairs 331 and 3678. The negative controls included media from PBMC and MDDC uninfected cultures. The positive controls included IFN-α spiked media equal to the initial amount of IFN-α utilized in these infections, along with supernatant from an infection carried out in the presence of IFN-α.
IMC derived virus and plasma taken from the same time point in the transmitting partner (donor) along with the TF from the recipient, were used to test antibody neutralization of variants circulating near the time of transmission. The TZM-bl neutralization assay was adapted for use with IMCs, in a similar fashion to what has been published previously for IMC [54] and pseudoviruses [1,28]. Briefly, heat inactivated plasma was serially diluted 5-fold starting at 1:100, and each dilution was then mixed with 20 IU/ul of virus at a 1:1 ratio. After incubation at 37°C for 1 hour, the plasma and virus mixtures were used to infect previously seeded TZM-bl cells (24 hours prior to infection at 6x103 cells per well in a 96-well plate). After a 40 hour incubation, the Promega Reporter Buffer was used to lyse cells according to manufacturer instructions and, following two freeze-thaw cycles, luciferase was measured with the Luciferase Assay System from Promega (Catalog # E1501) in the supernatants on a luminometer using the Gen5 2.00 software. Maximal percent inhibition (compared to the no plasma control) was calculated at a dilution of 1:00 after background subtraction and removal of variants with a signal less than three times background for cell only control wells. The data is averaged from each virus run in duplicate from two independent experiments.
|
10.1371/journal.pgen.1002841 | Rapid Turnover of Long Noncoding RNAs and the Evolution of Gene Expression | A large proportion of functional sequence within mammalian genomes falls outside protein-coding exons and can be transcribed into long RNAs. However, the roles in mammalian biology of long noncoding RNA (lncRNA) are not well understood. Few lncRNAs have experimentally determined roles, with some of these being lineage-specific. Determining the extent by which transcription of lncRNA loci is retained or lost across multiple evolutionary lineages is essential if we are to understand their contribution to mammalian biology and to lineage-specific traits. Here, we experimentally investigated the conservation of lncRNA expression among closely related rodent species, allowing the evolution of DNA sequence to be uncoupled from evolution of transcript expression. We generated total RNA (RNAseq) and H3K4me3-bound (ChIPseq) DNA data, and combined both to construct catalogues of transcripts expressed in the adult liver of Mus musculus domesticus (C57BL/6J), Mus musculus castaneus, and Rattus norvegicus. We estimated the rate of transcriptional turnover of lncRNAs and investigated the effects of their lineage-specific birth or death. LncRNA transcription showed considerably greater gain and loss during rodent evolution, compared with protein-coding genes. Nucleotide substitution rates were found to mirror the in vivo transcriptional conservation of intergenic lncRNAs between rodents: only the sequences of noncoding loci with conserved transcription were constrained. Finally, we found that lineage-specific intergenic lncRNAs appear to be associated with modestly elevated expression of genomically neighbouring protein-coding genes. Our findings show that nearly half of intergenic lncRNA loci have been gained or lost since the last common ancestor of mouse and rat, and they predict that such rapid transcriptional turnover contributes to the evolution of tissue- and lineage-specific gene expression.
| The best-understood portion of mammalian genomes contains genes transcribed into RNAs, which are subsequently translated into proteins. These genes are generally under high selective pressure and deeply conserved between species. Recent publications have revealed novel classes of genes, which are also transcribed into RNA but are not subsequently translated into proteins. One such novel class are long noncoding RNA (lncRNA). LncRNA loci are controlled in a similar manner to protein-coding genes, yet are more often expressed tissue-specifically, and their conservation and function(s) are mostly unknown. Previous reports suggest that lncRNAs can affect the expression of nearby protein-coding genes or act at a distance to control broader biological processes. Also, lncRNA sequence is poorly conserved between mammals compared with protein-coding genes, but how rapidly their transcription evolves, particularly between closely related species, remains unknown. By comparing lncRNA expression between homologous tissues in two species of mouse and in rat, we discovered that lncRNA genes are “born” or “die” more rapidly than protein-coding genes and that this rapid evolution impacts the expression levels of nearby coding genes. This local regulation of gene expression reveals a functional role for the rapid evolution of lncRNAs, which may contribute to biological differences between species.
| The mammalian transcriptome has recently been shown to be surprisingly diverse in its extent and encoded functions [1]–[3], much of which are noncoding RNAs (ncRNAs) as they are not translated into proteins. The ability to sequence the entire transcriptome in an unbiased manner has not only allowed more complete characterization of well described and highly abundant noncoding RNAs with known function, such as transfer RNAs, small nuclear RNAs, small nucleolar RNAs and ribosomal RNAs, but have also revealed additional ncRNA species. For example, a number of long ncRNAs (lncRNAs) larger than 200 nucleotides (nt) have been discovered [2], [4], [5]. Many lncRNA loci are intergenic, when transcription occurs wholly within the genomic intervals between two adjacent protein-coding genes [6]. Some lncRNAs can be transcribed divergently from a neighbouring protein-coding transcript using identical or almost identical transcriptional initiation complexes [6]. In addition, lncRNAs overlapping with protein-coding genes can be transcribed from either strand [6]–[8].
Although the precise roles of many lncRNAs remain unknown, in general they are thought to act in transcriptional regulation [6], [9], [10]. LncRNAs can regulate gene expression programs through a variety of mechanisms, including interactions with chromatin remodelling complexes or transcription factors [11]. Consistent with a cis-regulatory role, co-expression of intergenic lncRNA loci with their neighbouring protein-coding genes has been observed [12], [13] and a number of intergenic lncRNAs have demonstrated roles in regulating the expression of genes in their genomic vicinity [9]. Some intergenic lncRNAs appear to regulate the expression of both neighbouring and distal genes [14], [15]. Indeed, many intergenic lncRNAs have been experimentally demonstrated to have roles in regulating transcription of distally located targets, in trans [16]. Nevertheless, the exact proportion and the distinguishing features of cis- and trans-acting intergenic lncRNAs remain unknown.
If lncRNAs' functional roles are conserved it is expected that their loci should be evolutionarily preserved. Indeed, the transcripts and promoters of mammalian intergenic lncRNAs exhibit signatures of selective constraint: their promoters are highly conserved across vertebrates [2] and they have accumulated fewer substitutions than neighbouring putative neutral sequence [17], [18]. However little is yet known of the evolutionary persistence of lncRNA transcription. Generally the loss and gain of functional noncoding sequence can occur rapidly, with approximately half of all functional ancestral nucleotides predicted to have been gained or lost in mouse or rat since their common ancestor [19]. Other noncoding RNAs, in particular tRNAs, have been shown to exhibit rapid turnover of their transcribed loci, despite conservation of their function [20]. Turnover of regulatory elements underlies species-specific transcriptional evolution and may be associated with phenotypic changes [21].
Only a small minority of intergenic lncRNAs in mouse or human were found to have transcribed orthologous sequences in the other species [22], [23]. This might reflect turnover of transcribed loci, or it might imply that intergenic lncRNAs, which are often lowly expressed and tissue specific [6], [9], [18], [23], have transcribed orthologous sequences that remain undetected. Indeed, analysis of the transcription of three intergenic lncRNA loci across homologous regions of the mammalian and avian brain revealed that some intergenic lncRNAs can have conserved expression patterns [24].
To resolve the extent of lncRNA transcriptional turnover it is important to undertake a careful comparison of lncRNA transcription in homogeneous and homologous tissues. Achieving this in closely related species also allows the distinction of transcriptional turnover from DNA sequence turnover and furthermore might reveal otherwise unexpected mechanisms of regulatory divergence. Here we experimentally and computationally explored the genetic structure and function of lncRNA loci in matched tissues from three closely related rodent species, Mus musculus domesticus (C57BL/6J), Mus musculus castaneus and Rattus norvegicus.
We identified transcripts expressed in the liver of three young adult male Mus musculus domesticus (inbred strain C57BL/6J termed hereafter Mmus) individuals by directional, stranded ribosomal RNA (rRNA)-depleted transcriptome sequencing (total RNAseq) (Figure 1A) (see Materials and Methods). Data from three independent biological replicates were pooled. About 80% of sequencing reads were mapped [25] to the reference Mmus (mm9) genome and liver gene expression was detectable for 61% of all UTRs and coding exons annotated in the mouse genome (coverage: 66%). We found that a substantial fraction of sequencing reads map to unannotated, likely noncoding, loci consistent with previous results [26].
Using our total transcriptome sequencing data we assembled de novo 56917 transcripts [27] expressed in the Mmus liver (Figure 1A). As a consequence of the short-read single end nature of our data, our transcripts can be fragmented due to incomplete coverage of the full-length cDNA. To identify independent transcripts, we performed genome-wide chromatin immunoprecipitation followed by sequencing (ChIPseq) against trimethylation of lysine 4 of histone H3 (H3K4me3), which marks the beginning of actively transcribed genes [28] and identified enriched regions [29] (Figure 1A) (see Materials and Methods).
We intersected the genomic locations of 18303 H3K4me3 enriched regions with the predicted 5′ end of our RNAseq-defined Mmus transcripts longer than 200 bases in length, thereby predicting 8915 distinct transcription start sites (TSSs) (Figure 1A). As found in previous studies, we identified a limited number of protein-coding genes that exhibited evidence of bidirectional transcription at their TSS (Figure S1, Table S10) [30]. Most of these transcribed regions are likely noncoding and are not further addressed in our study except when supported by a de novo assembled noncoding transcript [31].
Similarly, we identified transcripts that were either intergenic (n = 388) or intragenic (n = 8527) based on their overlap with Mmus protein-coding gene annotations (Figure 1A) (see Materials and Methods). Intergenic transcripts lacking protein-coding potential [32] were annotated as long intergenic ncRNAs (intergenic lncRNAs) (n = 316, Table S3). Next we defined transcribed loci as clusters of one or more transcripts with overlapping exonic or intronic nucleotides. From 293 of these loci only intergenic lncRNA transcripts were expressed (Table S3 and S4). The vast majority (n = 233) of these intergenic lncRNA loci have no overlap with intergenic lncRNAs annotated in the mouse genome by Ensembl (build 64), demonstrating that current mouse intergenic lncRNA catalogues are largely incomplete [18]. Mmus liver intergenic lncRNAs transcripts were significantly (two-tailed Mann-Whitney test, typically p<1×10−4) found to be: (i) more lowly expressed, (ii) shorter and (iii) to have fewer exons than their protein-coding transcript counterparts (Table S2) consistent with previous reports [23], [33].
The second group of 7289 intragenic loci comprises 8527 transcripts overlapping protein-coding genes (Ensembl build 60, Table S3 and S4). Forty-nine loci have overlapping antisense RNAs transcribed from the opposite strand and marked by separate H3K4me3 enriched regions indicating independent transcriptional initiation (Table S9). Examples in this category include the constitutively expressed noncoding RNA Kcnq1ot1 [34].
Most protein-coding genes are expressed in multiple tissues [35]. In contrast, lncRNA expression tends to be spatially and temporally restricted [6], [18], [23], [36]. We validated the expression of 15 randomly selected liver expressed intergenic lncRNA transcripts by quantitative PCR (RT-qPCR) in seven Mmus adult tissues (Figure 1B) and nine intragenic antisense lncRNA transcripts by strand specific RT-qPCR [8] in four adult tissues (Figure S2D). These tissues were chosen because they show different degrees of cell type complexity and biological functionality. We found that the large majority of the tested intergenic and intragenic antisense lncRNA transcripts are predominately expressed in liver.
Large changes in gene expression are observed during tissue development [37]. In order to identify whether the intergenic lncRNAs we identified are developmentally regulated during hepatocyte differentiation, we measured the abundance of representative lncRNAs by RT-qPCR at embryonic stages E10, E12, E14 and E18 and adult stage P62. Our data showed that lncRNAs are also extremely specific to the adult developmental stage of liver. In summary, the intergenic lncRNAs we identify are specifically expressed in nutritionally unstressed adult liver (Figure 1C).
Sequence comparison of mouse intergenic lncRNAs and their human and rat orthologous sequence have shown that these transcripts tend to be constrained, an evolutionary hallmark of functionality, albeit at much lower levels than protein-coding genes [17], [18]. However little is yet known about transcriptional turnover of lncRNA during evolution. To address the transcriptional turnover of lncRNAs, we explored their transcription across three rodents. In addition to Mmus, we studied transcript expression in the adult liver of a closely related mouse Mus musculus castaneus (CAST/EiJ termed Mcas) and in the rat (Rattus norvegicus, termed Rnor) (Figure 2). The two mouse subspecies, Mmus and Mcas, diverged about one million years ago (MYA) and last shared a common ancestor with Rnor about 13 to 19 MYA [38] (Figure 2A). These differences in species separation across evolutionary time allowed us to take two snapshots of transcriptional turnover during rodent evolution, using the closest wild-derived mouse species (Mcas) to Mmus that is commercially available and Rnor as the evolutionary nearest rodent species with a well-annotated genome. Similar to the characterization of transcripts in Mmus liver, we performed RNAseq and H3K4me3 ChIPseq experiments in Mcas and Rnor, and identified 158 and 605 intergenic lncRNAs respectively (Tables S1, S5, S6, S7, S8). The observed difference between the numbers of annotated intergenic lncRNA loci across the three rodents (293, 158 and 605 for Mmus, Mcas and Rnor, respectively) can be either due to experimental bias or underlying biology. To test the contribution of the difference in read number of each species RNAseq library (Table S1), we reassembled transcripts in Mmus and Rnor after randomly selecting from Mmus and Rnor libraries the same number of reads as Mcas, our smallest library (Table S1). For each species we repeated this procedure 10 times. By comparing the numbers of intergenic lncRNAs in Mmus or Rnor that overlapped a transcript from these recreated libraries, we found that the differences in numbers of lncRNAs between mice (Mmus and Mcas) species are mostly due to the depth of sequencing. After adjusting the read number of the Mmus RNAseq library to the Mcas RNAseq library, we identified a mean of 154 intergenic lncRNA loci (standard deviation = 3.4) for Mmus, a similar number to the one assembled in Mcas (n = 158), suggesting that the difference in the number of lncRNA loci is due to an experimental bias. In contrast, in Rnor, using the same number of sequencing reads the reduction approach afforded a mean of 284 intergenic lncRNA loci (standard deviation = 5.9). This number corresponds to a 80% rise over the 158 Mcas intergenic lncRNA loci and indicates that there is an increase of liver lncRNA loci in the rat lineage.
We next considered if during rodent evolution lncRNA loci were conserved in their transcription in a similar manner to protein-coding genes. We defined transcriptional turnover as instances of genomic loci for which syntenic sequence is conserved between two or more species yet transcription of this conserved sequence is not. To determine conservation of transcribed loci, we combined H3K4me3 peaks with RNA sequencing reads overlapping (by more than 1 bp) the syntenic regions to create a stringent set of conserved loci (see Materials and Methods). These loci show evidence of both transcriptional initiation and transcript formation. Owing to the availability of its larger number of publicly available genome wide resources, such as spatial and temporal expression patterns [39], we anchored our analysis on Mmus. To allow differentiation between sequence and transcriptional turnover we only considered Mmus loci that have aligned orthologous sequence in the rat genome [intergenic lncRNA loci n = 268 (91.5%), protein-coding loci n = 6723 (92.2%)].
We then classified mouse loci according to their transcriptional conservation into three classes: those specific to Mmus, if evidence of expression was found only in Mmus; those conserved in Mus genus, when evidence of transcription was found in Mmus and Mcas but not in Rnor; and, those conserved across these rodents, when expression evidence was found in Mmus, Mcas and Rnor (Figure 2A, Table S4). Our definition does not explicitly take into account conservation of exon-intron structure. Globally, H3K4me3 and RNAseq signals were grouped according to our classification (Figure 2B–2C).
In order to confirm that the observed differences were not solely a consequence of biases introduced by sequencing depth, we validated our interspecies comparisons by semi-quantitative RT-PCR in independent biological replicates from adult livers of Mmus, Mcas and Rnor for 24 intergenic lncRNA transcripts from four categories (rodent conserved, Mus genus conserved, Mmus-specific, and Rnor-specific, Figure S3). These RT-PCR results confirmed that our global approach accurately identifies species- and lineage-specific intergenic lncRNAs.
Turnover of transcription is considerably more frequent for intergenic lncRNA loci than for protein-coding genes in the rodent liver (Figure 2D). A significantly smaller fraction of intergenic lncRNA than protein-coding loci exhibit conserved transcription across rodents [intergenic lncRNA loci n = 160 (59.7%), protein-coding loci n = 6169 (91.7%), two-tailed Fisher's exact test, p<10−3]. Conversely, a significantly higher proportion of intergenic lncRNA than protein-coding loci are specific to the Mmus lineage [intergenic lncRNA loci n = 30 (11.2%), protein-coding loci n = 75 (1.1%), two-tailed Fisher's exact test, p<10−3].
The difference in sequencing depth between the three species influenced the number of annotated intergenic lncRNAs. To account for this effect and provide a more conservative estimate of transcriptional conservation we considered the set of intragenic and lncRNA loci that were assembled after adjusting the Mmus and Rnor RNAseq library sizes to that of Mcas (see Materials and Methods). Intragenic and intergenic lncRNA loci were annotated as previously. We considered a Mmus locus to have conserved expression if it had an overlapping H3K4me3 peak and an overlapping transcript (>1 bp). As previously, we found protein-coding gene loci to be more often conserved in rodents (1326/2415, 55%) than intergenic lncRNA loci (31/110, 28%, two-tailed Fisher's exact test, p<10−3).
Next we aimed to gain initial insights into the conservation of exon-intron structures of Mmus intergenic lncRNAs. For mouse intergenic lncRNAs and protein-coding loci whose transcription was conserved in rat (160 and 6169 loci, respectively) we compared the coverage by RNAseq reads of mouse exonic nucleotides in the rat orthologous regions. We found that rodent conserved protein-coding transcripts have a significantly higher coverage (median 78%) than intergenic lncRNA (median 47%, two-tailed Mann-Whitney test, p<2×10−16, Figure S4). This observation can be a consequence of lower coverage of low abundance transcripts and/or lower conservation of exon-intron structure for intergenic lncRNAs.
Similarly, we observed that the transcriptional conservation of noncoding transcripts that overlap protein-coding genes in antisense orientation also showed a rapid decay across rodent evolution. Only 36% of the Mus conserved intragenic antisense transcripts are expressed in Rnor (Figure S2). These results indicate that the large majority of ncRNAs are conserved in the Mus genus but not in the evolutionarily further distant species Rnor. The apparent low conservation of intragenic antisense transcription is consistent with previous conservation analysis [33].
To investigate transcriptional turnover of intergenic lncRNAs beyond the rodent lineage, we used publicly available polyA+ transcriptome sequencing data for the adult human liver (Human BodyMap 2.0 RNAseq data). Rodents and human shared a common ancestor over 90 MYA [40]. We considered in this analysis only Mmus transcripts whose expression was supported by at least one overlapping polyA+ sequencing read [41]. We found that the majority of mouse intergenic lncRNA loci overlap polyA+ reads (273/293 loci), suggesting that few intergenic lncRNA loci assembled here transcribe only non-polyadenylated transcripts. We discarded 1368 (18.8%) protein-coding and 159 (58.2%) intergenic lncRNA loci in Mmus that lack an apparent orthologous sequence in the human or rat genome [42]. As observed for the rodent lineage, a significantly smaller fraction of Mmus intergenic lncRNA than protein-coding genes orthologous in humans are expressed in the liver [intergenic lncRNA loci (n = 76, 56.7%), protein-coding loci (n = 5689, 96.1%), two-tailed Fisher's exact test, p<10−3) (Figure S5). Our data indicate that the fraction of liver transcribed mouse intergenic lncRNAs expressed in the orthologous region of the human genome is two-fold higher (two-tailed Fisher's exact test, p<10−3) than prior estimates [22], which supports the use of homologous tissue types to investigate levels of transcriptional conservation of tissue specific transcripts, such as intergenic lncRNAs. We conclude that rapid turnover of intergenic lncRNAs is not restricted to the rodent lineage, but is widespread among eutherian mammals.
Next we examined how sequence constraint reflects transcriptional conservation of intergenic lncRNA and protein-coding loci. For each transcript we considered its most 5′ nucleotide to correspond to the transcriptional start site and defined its promoter as the 400 nucleotides upstream of this site. We compared the mouse-rat nucleotide substitution rate for intergenic lncRNA loci (dloci) and promoters (dpromoter), to rates for genomically neighbouring and non-overlapping ancestral repeats [ARs (dAR)] with matched G+C content [18], [43]. ARs are transposable element-derived sequences that were present in the last common ancestor of human and mouse; most of these sequences have been observed to evolve neutrally and hence provide reliable proxies for local neutral mutation rates [44]. We first confirmed that Mmus liver-expressed intergenic lncRNA loci accumulated mutations at a significantly slower rate than adjacent neutral sequence (Figure S6A) (dloci = 0.148, dAR = 0.164, two-tailed Mann-Whitney test, p<3×10−7). In line with this observation, long sequence segments that have preferentially purged insertions or deletions in Mmus and Rnor lineages were 1.6-fold enriched in intergenic lncRNA transcription over expected levels (permutation test, p<10−3) [44]. As previously reported [12], [18] the sequences of intergenic lncRNA loci evolve more rapidly than those of full-length protein-coding loci (Figure S6B) (dloci/dAR = 0.902; protein-coding dloci/dAR = 0.857; two-tailed Mann-Whitney test, p<2×10−3). Additionally, the putative core promoters of intergenic lncRNAs accumulated significantly more substitutions than those of protein-coding genes (Figure S6C) (intergenic lncRNA dpromoter/dAR = 0.843; protein-coding dpromoter/dAR = 0.746, two-tailed Mann-Whitney test, p<2×10−5). The discrepancy between this result and published findings [2] is likely due to the incompleteness of lncRNA transcripts' 5′ ends and thus to incomplete delineation of lncRNA promoter sequences.
To determine whether loss of transcription is associated with loss of sequence constraint, we compared Mmus to Rnor nucleotide substitution rates between two groups of intergenic lncRNAs: those specific to the Mus genus (Mmus and Mcas) and those conserved among these rodents (Mmus, Mcas and Rnor). Rodent conserved intergenic lncRNA loci show evidence for purifying selection on both transcribed (two-tailed Mann-Whitney test, p<4×10−10) (Figure 3A) and putative promoter sequences (two-tailed Mann-Whitney test, p<3×10−12) (Figure 3B). Intergenic lncRNA loci transcribed in the Mus genus but not in Rnor, exhibit no constraint in transcribed regions (two-tailed Mann-Whitney test, p>0.2) (Figure 3A). Mus genus-conserved putative core promoters accumulated significantly fewer substitutions than neighbouring putatively neutral sequence (median dprom = 0.151 and dAR = 0.165, two-tailed Mann-Whitney test, p<5×10−3) suggesting they evolved under purifying selection (Figure 3B). Negative selective pressure was significantly higher on the promoters of loci with rodent conserved transcription than on promoter sequence with Mus genus-specific transcription (rodent conserved median dprom/dAR = 0.783, Mus genus-specific median dprom/dAR = 0.901, two-tailed Mann-Whitney test, p<7×10−3).
We asked whether the observed low degree of sequence constraint on intergenic lncRNA loci, relative to protein-coding genes, was due to rapid transcriptional turnover of a subset of intergenic lncRNAs. To test this, we compared Mmus to Rnor nucleotide substitution rates for the transcribed sequences (including exons and introns) between the subset of intergenic lncRNA loci exhibiting conserved expression in the rodent liver (n = 160) with the corresponding set of protein-coding genes (n = 6641) and found no significant difference (intergenic lncRNA dloci/dAR = 0.827, protein-coding dloci/dAR = 0.842 two-tailed Mann-Whitney test, p>0.58) (Figure S7A). For loci conserved in rodents, nucleotide substitution rates of intronic and exonic sequence were compared between Mmus and Rnor. Introns (dintron) of protein-coding genes and intergenic lncRNAs evolved at comparable rates (intergenic lncRNA dintron/dAR = 0.959, protein-coding dintron/dAR = 0.986, two-tailed Mann-Whitney test, p>0.28) (Figure S7C). In contrast, protein-coding gene exons evolve under strong purifying selection (intergenic lncRNA dexon/dAR = 0.805, protein-coding dexon/dAR = 0.484, two-tailed Mann-Whitney test, p<10−15) (Figure S7B) likely to ensure the maintenance of their coding potential during evolution.
Our results therefore indicate that intergenic lncRNA loci that were gained or lost in recent Mus evolution evolved neutrally between mouse and rat. Conversely, rodent conserved intergenic lncRNAs have accumulated fewer substitutions than neighbouring neutral sequence indicating that conservation of transcription is reflected in sequence constraint.
Mammalian intergenic lncRNA loci and their genomically adjacent protein-coding genes show a significant tendency to exhibit similar spatiotemporal expression profiles [12], [13], [15], [23], [45]. We found intergenic lncRNA transcription in liver occurs significantly more frequently near to protein-coding genes that are expressed in the liver [39] than expected by chance (see Materials and Methods; 1.6-fold; permutation test, p<5×10−3). Complementary results were obtained using Database for Annotation, Visualization, and Integrated Discovery (DAVID) tissue annotation categories (Figure S8) [46]. About 30% of the protein-coding genes closer to intergenic lncRNA loci were classified as liver expressed (p<3×10−5).
We considered whether lineage-specific transcription of intergenic lncRNAs might associate with the expression level of genomically adjacent protein-coding genes (see Materials and Methods). If intergenic lncRNAs have no effect on nearby protein-coding gene expression, then lineage-specific differences in gene expression of genes should be unaffected by whether a neighbouring intergenic lncRNA locus is transcribed.
The existence of relatively large numbers of lineage-specific intergenic lncRNAs in mouse and rat permitted this hypothesis to be tested using Mmus and Rnor. Two additional reasons that we specifically analysed the intergenic lncRNAs identified in these two species were (i) the high quality of the genome annotations, relative to Mcas, and (ii) the existence of other published datasets that permitted further validation of our results [20].
First, we normalised gene expression for Mmus and Rnor RNAseq data (see Materials and Methods, Figure S9A) and validated the fold-difference on 17 selected protein-coding mRNA by RT-qPCR (Figure S9C and S9D). In order to obtain a baseline for transcriptional variation between species from this normalised set, we first estimated the fold difference in liver expression between 230 Mmus housekeeping protein-coding genes [47] and their one-to-one orthologous genes in Rnor (median fold-difference in expression = 0.020, see Materials and Methods). Next, we identified the closest protein-coding gene for each conserved or lineage-specific Mmus or Rnor intergenic lncRNA. We selected the intergenic lncRNA loci whose neighbouring protein-coding genes had annotated [48] one-to-one orthologs in the second species (Table S9).
We found that the expression levels of the genes whose nearest intergenic lncRNA locus showed conserved expression between rodents (n = 148) were similar to housekeeping gene levels (median fold-difference = −0.035, two-tailed Mann-Whitney test, p>0.36) (Figure 4, Table S12).
We then asked whether gene expression levels alter when a nearby intergenic lncRNA is gained or lost in one species. In contrast to the conserved situation above, we found that those protein-coding genes A nearest to lineage-specific intergenic lncRNA loci (n = 137) tended to be expressed at a higher level, with a median increase in gene expression of approximately 25% (median fold-difference = 0.212, two-tailed Mann-Whitney test, p<0.005) (Figure 4, Table S12). We repeated this analysis and confirmed this result using an independent dataset [20]. We found that the median expression levels of protein-coding gene loci adjacent to lineage-specific intergenic lncRNA loci were significantly higher than those of protein-coding gene loci near conserved intergenic lncRNA loci (two-tailed Mann-Whitney test, p<7×10−5) (Figure S9B, Figure S10). Transcription increased for half (50%) of those protein-coding genes lying adjacent to lineage-specific intergenic lncRNA loci, when assessed using either total RNA or mRNA expression; in contrast, less than a third (29%) of protein-coding genes near conserved intergenic lncRNA loci show consistent increased expression in both datasets (two tailed Fisher's exact test, p<0.05, Figure S11), suggesting that in some cases gain or loss of intergenic lncRNAs may influence the expression levels of neighbouring genes. We next investigated if some relative orientations of lineage-specific lncRNA transcription were more frequently associated with increased expression of the most proximal protein-coding gene. We divided lineage-specific intergenic lncRNA and protein-coding gene pairs into three classes (Figure S12A): tandem (48 gene pairs) if transcription occurred in the same orientation, divergent (71 gene pairs), or convergent (17 gene pairs) if transcription occurred in opposite directions either diverging or converging, respectively. All three relative genomic arrangements are associated with increased expression of the closest protein-coding genes. Both tandem and convergent orientations are associated with significantly increased expression at the 5% level while divergent orientation is significant at the 10% level (p<0.08, Figure S12B).
We considered a number of possible interpretations for this apparent association of lineage-specific intergenic lncRNAs with increased transcription of nearby protein-coding genes. The increased gene expression could be either (i) due to regional modifications to the genome that co-ordinately influence all coding and noncoding loci [49] or (ii) correlated with the transcription of the proximal intergenic lncRNA locus [13], [15]. A key distinguishing feature between these two mechanisms is whether lineage-specific expression of intergenic lncRNAs is associated with regional increases in transcription.
To test this, we identified the next most proximal protein-coding gene B, beyond its closest protein-coding gene A (Figure 4A). Genes duplicated in tandem often share regulatory elements and, as a consequence, exhibit similar expression patterns [50]. To account for this evolutionary bias, we excluded 17 protein-coding genes B that were annotated [48] as protein-coding gene A paralogs (see Materials and Methods). In contrast to the observed lineage-specific effects on protein-coding genes A, the expression levels of protein-coding genes B were not significantly affected (two-tailed Mann-Whitney test, p>0.7) by either conserved (median fold-difference = 0.078) or lineage-specific (median fold-difference = −0.088) intergenic lncRNA transcription (Figure 4, Table S13).
We next tested whether similar results might be obtained for lineage-specific protein-coding genes. We used the previously identified set of Mus-genus lineage-specific expressed protein-coding genes. We identified genes A′ as the closest protein-coding genes to these loci, protein-coding A′ (Figure S13). We excluded paralogous protein-coding gene pairs and considered only protein-coding genes A′ with a one-to-one ortholog in rat (89 genes). Transcription levels of nearby genes appear unaffected by the presence of lineage-specific protein-coding gene transcription in the genomic vicinity (median fold-difference = 0.052, two-tailed Mann-Whitney test, p>0.4) (Figure S13). As an additional control, we compared the densities of chromatin boundary elements (CCCTC-binding factor [CTCF]-bound sites) and DNase I hypersensitivity sites in the intergenic regions between (i) the lineage-specifically expressed intergenic lncRNA locus and its neighboring protein-coding gene A and (ii) protein-coding genes B, using data from previous studies [51], [52]. We found no significant differences between these densities (permutation test p>0.2). The association between lineage-specific lncRNA transcription and increased expression levels of neighbouring protein-coding genes might depend on the distance between their transcriptional start sites (TSSs). The median distance of the TSS of a lineage-specifically expressed intergenic lncRNA with its closest protein-coding gene is 22 kb. However, no significant correlation was observed between this distance and the median fold difference in expression for protein-coding genes measured between mouse and rat (Pearson correlation, R = −0.03, p = 0.76, Figure S14).
Our comparison of matched tissues in two species thus revealed that birth or death of intergenic lncRNAs is associated with changes in transcription of proximal protein-coding genes.
To investigate the evolution of lncRNAs, we identified the highest confidence set of lncRNAs in matched, nutritionally unstressed, adult livers of three closely related rodent species: Mmus, Mcas and Rnor, by combining genome-wide interrogation of chromatin signatures and total RNA expression. This highly conservative set of lncRNAs confirmed a number of prior observations. First, many intergenic and antisense lncRNA loci are expressed in a cell/tissue- or time-specific manner: we found that the intergenic lncRNAs present in adult liver are not only absent from other adult tissues, but are perhaps surprisingly even absent in developing mouse liver. These temporally- and spatially- restricted expression patterns, together with their relatively low expression levels, likely explain why our intergenic lncRNA set shows limited overlap with previously reported sets [18]. From our analysis, two major results emerged: first, that intergenic and antisense lncRNA transcription can evolve extremely rapidly between closely related mammals; second, that this rapid evolution seems to occur simultaneously with increased expression of neighbouring protein-coding genes.
Previous studies have indicated that 12 to 15% of lncRNAs are conserved between human and mouse, based on comparison of EST and cDNA datasets from disparate experimental designs [22], [23]. Our matched interspecies data are perhaps better suited to establish experimentally the rate of lncRNA turnover. The use of mouse and rat, being closely related species, minimises the effects of genomic sequence divergence, thus better uncoupling sequence and transcriptional changes. Transcription of noncoding loci is more frequently gained or lost than transcription of protein-coding genes; between 28% and 61% of intergenic and antisense lncRNAs, respectively are specific to the Mus genus. We expect similar turnover will be found in most cell types of various developmental stages given that liver is a typical somatic tissue [53]. The transience of intergenic lncRNA transcription is mirrored by changes to selective pressures acting on their sequences. Our results are consistent with purifying selection acting on transcribed intergenic lncRNA loci, and with no selection acting on untranscribed orthologous sequence in other species. This coupling of transcriptional conservation with sequence constraint suggests that conserved intergenic lncRNA loci are biologically significant in rodents.
The expression levels of intergenic lncRNAs and their genomically neighbouring protein-coding genes have previously been shown to be positively correlated [12], [13]. We find that species-specific transcription of intergenic lncRNAs correlates with elevated expression of neighbouring protein-coding genes. The increased transcription observed among neighbouring genes is unique to intergenic lncRNAs, and seems unlikely to be due to local changes in chromatin environment. If the intergenic lncRNAs in other tissues and species behave similarly, intergenic lncRNAs could contribute substantially to lineage-specific and tissue-specific evolution of gene expression.
The rapid turnover we observed in lncRNA transcription strongly resembles what was recently reported for transcription factor binding events [54]–[56], tRNA transcription [20] and functional regulatory sequences in general [19]. For instance, between 10 to 20% of transcription factor binding events overlap between human and mouse liver [56], which is similar in scale to what we now find for intergenic lncRNAs. These parallels suggest that rapid evolution is a general feature of noncoding regulatory mechanisms.
It was recently proposed that intergenic lncRNAs have minimal impact on the transcriptional regulation of their neighbouring protein-coding genes [16], [23]. By exploiting the rapid birth and death of noncoding RNAs, we revealed that intergenic lncRNAs could contribute to lineage-specific changes in the expression levels of neighbouring protein-coding genes. Our data do not preclude distal regulatory roles, which might be lineage-specific, for some or all intergenic lncRNAs we investigate. It will now be crucial to understand how intergenic lncRNAs evolve and to unravel the molecular mechanisms underlying lineage-specific gene expression changes associated with intergenic lncRNAs.
ChIPseq, RNAseq, and RT-PCR experiments were performed on liver material isolated from three rodents: Mus musculus domesticus (Mmus), Mus musculus castaneus (Mcas), and Rattus norvegicus (Rnor). Each ChIPseq and RNAseq experiments were performed on at least two independent biological replicates from different animals. Mmus and Mcas (male adults, 10 weeks old) were obtained from the Cambridge Research Institute. Rnor (male adults, 9 weeks old) were obtained from Charles River. All tissues were either treated post-mortem with 1% formaldehyde for ChIP experiments or flash-frozen in liquid N2 for RNA experiments. The investigation was approved by the ethics committee and followed the Cambridge Research Institute guidelines for the use of animals in experimental studies under Home Office license PPL 80/2197.
ChIP sequencing experiments were performed as described previously [57] using H3K4me3 antibody (CMA304) [58]. In brief, the immunoprecipitated DNA was end-repaired, A-tailed, ligated to the sequencing adapters, amplified by 18 cycles of PCR and size selected (200–300 bp). For RNA-sequencing library preparation, total RNA was extracted using Qiazol reagents (Qiagen) and DNase-treated (Turbo DNase, Ambion). Ribosomal RNA was depleted from total RNA using RiboMinus (Invitrogen). RNA was reversed transcribed and converted into double-stranded cDNA (SuperScript cDNA synthesis kit, Invitrogen), sheared by sonication followed by paired end adapter (Illumina) ligation and prior to PCR amplification cDNA was UNG-treated to maintain strand-specificity [59]. After passing quality control on a Bioanalyzer 1000 DNA chip (Agilent) libraries were sequenced on the Illumina Genome Analyzer II (single-ended) and post-processed using the standard GA pipeline software v1.4 (Illumina).
H3K4me3 ChIPseq and associated input DNA control ChIPseq reads were aligned to the corresponding reference genomes (mm9 for Mus musculus domesticus and Mus musculus castaneus; Rn4 for Rattus norvegicus) using MAQ version 0.7.1 (default parameters) [60]. Reads mapping to multiple genomic locations were discarded. Genomic regions enriched over matching input DNA control were defined using MACS version 1.3.7.1 using the default parameters [61]. Comparative analysis was carried out using the Galaxy web tool [62]. Total RNA sequencing reads were mapped with Tophat (version 1.3.0) [25], using default parameters. A file containing the mapped coordinates of mouse and rat ESTs and mRNA mapped coordinates (downloaded from UCSC on the 11th March 2011) was provided to facilitate total RNA read mapping across splice junction for Mmus and Mcas, and Rnor respectively. Reads mapping to rRNA, tRNA and mtRNA were masked and the remainder were used to assemble transcripts de novo using Cufflinks (version 1.3.0) [27].
We filtered out transcripts smaller than 200 nucleotides (nt) and without an H3K4me3 peak overlapping their predicted transcriptional start site (TSS). Transcripts overlapping protein-coding gene annotations (by one or more base pair) from RefSeq, Ensembl (build 60) [48] and UCSC were annotated as intragenic. To discriminate between unannotated protein-coding and putatively noncoding transcripts we estimated the coding potential of all intergenic transcripts using the coding potential calculator (CPC) [32]. We annotated all transcripts with a coding potential less than 0 as intergenic long noncoding RNAs (intergenic lncRNAs). The 400 nt region upstream of the 5′ end (TSS) of each intergenic lncRNA or protein-coding transcript was annotated as a putative promoter. Transcribed loci were defined as non-overlapping regions with one or more transcripts that can contain overlapping exonic or intronic nucleotides. Loci containing only transcripts predicted to be intergenic lncRNAs were annotated as intergenic lncRNA loci. The remainder were annotated as protein-coding loci.
For the identification of antisense transcripts from the Cufflinks output file (n = 56917), we first identified 2383 transcripts overlapping protein-coding genes in antisense orientation in Mmus. This number included four types of ambiguous cases that were systematically removed: (i) annotated protein-coding transcripts (removing 1816 transcripts), (ii) antisense transcripts lacking an H3K4me3 peak independent from the TSS of overlapping protein-coding gene (removing 324 transcripts), (iii) transcripts lacking H3K4me3 marks at their 5′ end, and (iv) mapping assembly artefacts, revealed by visual inspection (collectively removing 90 transcripts). Taking all of these cases into consideration, 49 loci (or 153 antisense transcripts) were annotated in Mmus. A similar procedure was conducted in Mcas and Rnor, revealing 66 loci in total.
To identify lncRNAs deriving from bidirectional transcription at TSSs of protein-coding genes, we subtracted divergently transcribed protein-coding genes from our list of actively transcribed protein-coding genes. The TSSs of gene loci are spanned by one H3K4me3 peak and the evidence of divergent transcription is represented by RNAseq reads mapping in opposite directions. We identified divergent reads within an 1 kb window of a protein-coding gene's annotated TSS (Ensembl, build 60) [30].
Heatmaps and transcription start site aggregation plots were constructed using seqMINER [63].
To account for the difference in RNAseq library size between the three rodent species (Table S1) Mmus and Rnor transcripts were assembled using the same number of reads in Mcas library, the smallest RNAseq library. Reads were randomly selected without replacement and transcripts reassembles using Cufflinks and annotated as described above.
RT-PCR analysis of lncRNAs was performed by reverse transcription of 10 µg of DNase-treated total RNA according to the manufacturer's protocols using 200 U SuperScript-II Reverse Transcriptase (Invitrogen Corporation), 0.5 µg oligo(dT) and 0.5 µg random primers or 1 µg gene-specific primers (see Table S11). Negative controls were included in RT reactions. The cDNAs were then treated with RNase H at 37°C for 1 hour. Each PCR reaction typically contained 25 ng of cDNA, 5 pmol of the gene-specific primers (Table S11), 10 µL PCR Master Mix (Bioline), and 2 µL of the diluted cDNAs in a total volume of 20 µL. Reactions were carried out in triplicate in ABI 7900HT Fast Real-Time PCR system at the optimal temperature, as defined by provider instructions.
The significance of genome-wide associations between intergenic lncRNAs and their neighbouring protein-coding genes was assessed using Genome Association Tool (GAT) (Heger et al., in preparation). GAT compares the observed number of overlapping nucleotides between a set of segments with particular annotations to what would be expected from random placement of these segments. Expected densities are obtained using a randomisation procedure that accounts for G+C content and chromosome specific biases. A previous version of GAT was used in [9], [18]. This tool infers associations between intergenic lncRNA loci (segments) across the following annotation sets: (I) mouse-to-rat indel purified segments [44] and (II) liver-expressed protein-coding gene territories (Average Difference values >200) [39]. A protein-coding gene territory is defined as the genomic region containing all nucleotides that are closer to the gene than they are to its most proximal up- and downstream protein-coding genes, as described elsewhere [9], [18]. As a second tool, we used the gene functional classification tool Database for Annotation, Visualization, and Integrated Discovery (DAVID) (default parameters: count = 2 and ease = 0.1) [46] to explore the enrichment of tissue gene expression.
Regions of the mouse and rat genome that are enriched in CTFC binding were obtained from [51]. DNase hypersensity sites (DHS) in the mouse adult liver were obtained from [52]. Only male and sex independent DHS peaks that were either annotated as being robust and standard were considered in this analysis. GAT (Heger et al., in preparation) was used to test the observed density of these two class of regulatory elements in the intergenic region between lineage-specific intergenic lncRNAs and protein coding gene A (Figure 4) to what would be expected based on their distribution across the intergenic regions between lineage-specific intergenic lncRNA and protein-coding gene B (Figure 4).
Orthologous regions between Mmus and Rnor were identified using whole genome pairwise alignments [42]. An intergenic lncRNA locus was considered to be expressed in another species when its orthologous (between Mus species and Rnor) or equivalent (between Mmus and Mcas) position had an overlapping (>1 bp) H3K4me3 peak and one or more overlapping RNAseq reads. Due to the lack of H3K4me3 data for human, overlap (>1 bp) by one or more RNAseq reads in the orthologous human location was considered sufficient evidence for transcriptional conservation of an Mmus locus in human sequence. Only Mmus loci whose transcription was supported by one or more polyA+ selected sequencing read [41] were considered in this analysis. Identical criteria were used to determine the conservation of antisense lncRNA loci. An antisense lncRNA locus was judged to be expressed in another species when its orthologous position had an overlapping (>1 bp) H3K4me3 peak and one or more overlapping RNAseq reads in opposite orientations. We visually inspected these calls on 66 loci across the three rodent species.
Nucleotide constraint between Mmus and Rnor locus, exon, intron or putative promoter was estimated as described previously [18]. Pairwise substitution rates between Mmus and Rnor genomic regions were estimated using BASEML from the PAML package with the REV substitution model [64]. The substitution rate of the region of interest was compared to the rate observed for non-overlapping adjacent (<500 kb) ancestral repeats (inserted before the primate and rodent split) with similar G+C content [18].
Mmus and Rnor protein-coding transcript annotations were downloaded from Ensembl (build 60, http://www.ensembl.org/index.html) and used to define a set of constitutive exons for each gene. To account for differences in size of constitutively expressed portions of Mmus and Rnor genes, the total number of overlapping reads per nucleotide in Rnor was adjusted to what would be expected if the sequence in Rnor had the same length as that observed in Mmus. The expression of a gene in Rnor or Mmus is proportional to the sum of reads mapped to their exons divided by their combined length. To allow comparison of gene expression between species, read counts were normalized using TMM (edgeR package) [65]. Briefly, to estimate the normalised library size for each species, it was assumed that 60% of expressed genes were transcribed at similar levels in the two species. Other cut-offs (50% and 70%) yielded similar results. The normalised Mmus and Rnor library size was used to calculate the expression level (as total number of fragments per kb of sequence per million reads mapped, FPKM) of each gene in each species.
Each intergenic lncRNA locus was paired with its genomically closest protein-coding gene. Only pairs whose protein-coding genes had one-to-one orthologs between Mmus and Rnor were considered. The fold difference in expression levels of protein-coding genes associated with lineage-specific (Mus-genus or Rnor-specific) or rodent conserved expression was estimated between [6] the same direction. To calculate the fold difference in expression for each housekeeping gene between Mmus and Rnor species X and Y were randomly assigned. Fold expression differences for protein-coding genes B or A′ (Figure 4, Figure S12) were calculated in a similar manner.
Apart from permutation tests all other statistical analysis were performed using the R package [66].
RNAseq and H3K4me3 ChIPseq sequencing data are available from ArrayExpress under accession number E-MTAB-867. Additional mRNAseq data used was E-MTAB-424.
|
10.1371/journal.pbio.1002051 | A Novel TGFβ Modulator that Uncouples R-Smad/I-Smad-Mediated Negative Feedback from R-Smad/Ligand-Driven Positive Feedback | As some of the most widely utilised intercellular signalling molecules, transforming growth factor β (TGFβ) superfamily members play critical roles in normal development and become disrupted in human disease. Establishing appropriate levels of TGFβ signalling involves positive and negative feedback, which are coupled and driven by the same signal transduction components (R-Smad transcription factor complexes), but whether and how the regulation of the two can be distinguished are unknown. Genome-wide comparison of published ChIP-seq datasets suggests that LIM domain binding proteins (Ldbs) co-localise with R-Smads at a substantial subset of R-Smad target genes including the locus of inhibitory Smad7 (I-Smad7), which mediates negative feedback for TGFβ signalling. We present evidence suggesting that zebrafish Ldb2a binds and directly activates the I-Smad7 gene, whereas it binds and represses the ligand gene, Squint (Sqt), which drives positive feedback. Thus, the fine tuning of TGFβ signalling derives from positive and negative control by Ldb2a. Expression of ldb2a is itself activated by TGFβ signals, suggesting potential feed-forward loops that might delay the negative input of Ldb2a to the positive feedback, as well as the positive input of Ldb2a to the negative feedback. In this way, precise gene expression control by Ldb2a enables an initial build-up of signalling via a fully active positive feedback in the absence of buffering by the negative feedback. In Ldb2a-deficient zebrafish embryos, homeostasis of TGFβ signalling is perturbed and signalling is stably enhanced, giving rise to excess mesoderm and endoderm, an effect that can be rescued by reducing signalling by the TGFβ family members, Nodal and BMP. Thus, Ldb2a is critical to the homeostatic control of TGFβ signalling and thereby embryonic patterning.
| Cells depend on signals from their microenvironment to carry out their normal functions and coordinate responses. Once initiated, such signals often self-amplify via positive feedback to reach a sufficient level, when negative feedback can then be employed to dampen excess signalling. These feedback loops dynamically add or remove signalling components to maintain homeostasis. Their activation is often driven by the same signal transduction components, making it difficult to understand how signalling builds up in the first place. Here we find that the transcription co-factor Ldb2a enables differential response dynamics of negative and positive feedback upon the induction of TGFβ signalling. We show that Ldb2a directly activates expression of a TGFβ inhibitor that mediates negative feedback, while also repressing expression of TGFβ ligands that drive positive feedback. Moreover, expression of Ldb2a is itself activated by TGFβ signals. Thus, when Ldb2a levels are initially low, TGFβ signalling can self-amply and build up signal via positive feedback without being countered by negative feedback. We show that this regulatory mechanism is active in developing zebrafish embryos, where a loss of Ldb2a results in the over production of mesodermal and endodermal tissue types as a consequence of elevated TGFβ family signalling.
| In vertebrates, the transforming growth factor β (TGFβ) superfamily comprises a large number of ligands, including TGFβs, Nodal, Activin, and bone morphogenetic proteins (BMPs), each of which can direct lineage-specific transcriptional responses that regulate biological processes as diverse as cell proliferation, differentiation, apoptosis, and severe diseases caused by their mis-regulation [1]. In response to extracellular ligand binding, trans-membrane receptors phosphorylate receptor-activated Smads (R-Smads) in the cytoplasm. Different ligand-stimulated pathways converge and signal through two main R-Smad pathways, with Nodal/TGFβ/Activin mediated by R-Smad2/3 and BMP by R-Smad1/5/8 [2]. Activated R-Smads interact with the common partner Smad4 (Co-Smad4) to carry the signals into the nucleus, where the Smad complexes associate with additional transcription factors (TFs) and co-factors, as well as co-activators or co-repressors, to regulate downstream target genes [3].
The level of TGFβ signalling is established by homeostatic regulation, which dynamically adds or removes signalling components to maintain a sufficient and constant level of activity. For example, TGFβ signals activate expression of their own ligands [4–9]. After secretion from the cell, these ligands bind transmembrane TGFβ receptors, implementing positive feedback to self-amplify and sustain signals at a sufficient level and to propagate the signals into neighbouring cells. The inhibitors of TGFβ signalling, such as Leftys and inhibitory Smad6 and Smad7 (I-Smad6/7), can also be induced by TGFβ family signals, thereby generating negative feedback to dampen excess signalling [8–12]. These positive and negative feedbacks are coupled, as the TGFβ-responsive induction of both is by direct binding of R-Smads and Co-Smad4 to ligand or inhibitor genes [2,6,8,9,13–17]. Activation of TGFβ family signalling pathways results in rapid recruitment of transcriptional co-activators to ligand and I-Smad genes, leading to their up-regulation in vivo [8,9]. In zebrafish, the expression of Nodal ligand genes and Smad7 can be induced by R-Smad3 expression [12]. It has been demonstrated that coupled positive and negative feedback confers flexibility on signal switches and enables precise modulation of signal responses [18–20]. However, whether and how the activation of negative and positive feedbacks can be uncoupled is not known.
LIM domain binding proteins (Ldbs) are multi-functional non-DNA binding adaptor proteins that assemble TF complexes on target genes [21–25]. Components of such Ldb complexes, Lmo4 and Gata1/2 for example, have been shown to recruit R-Smad complexes onto TGFβ target genes [9,26,27]. By comparing published chromatin immunoprecipitation (ChIP)-seq datasets of genome-wide protein-DNA binding profiles for R-Smad1/3 and Ldb1 [8,9,21], we have obtained evidence that Ldb1 co-localises with R-Smad1/3 at a substantial subset of R-Smad target sites across the genome, suggesting that Ldb1 might function together with R-Smads to implement transcriptional responses to TGFβ family signalling. In vertebrates, a paralogue, Ldb2, shares a high percentage of amino acid sequence identity and structural similarity with Ldb1 [28], but its functions are largely unknown. In this study, we present in vivo functional and phenotypic data showing that Ldb2 regulates Nodal/BMP signalling and is required for early embryogenesis. Furthermore, we identify I-Smad7 and a Nodal ligand, Sqt, as direct target genes activated or suppressed respectively by Ldb2a, and show that the fine tuning of TGFβ family signalling requires both positive and negative control by Ldb2a complexes.
We compared published ChIP-seq datasets of Ldb1, the BMP effector, R-Smad1, and the Nodal/Activin/TGFβ effector, R-Smad3 [8,9,21,29]. We found that the binding of Ldb1 overlaps R-Smad1 or R-Smad3 binding at a substantial subset of R-Smad targets across the genome (Fig. 1A and 1B), including at the known TGFβ target genes, I-Smad6 and I-Smad7 (Fig. 1C and 1D). Ldb1 binding at these loci was validated in murine cells by ChIP-quantitative PCR (qPCR) (Fig. 1E). The ChIP-seq of Ldb1 had been performed in murine bone marrow cells or day 4 embryoid body (EB)-derived Flk1+ haemato-endothelial precursor cells [21,29], whereas the ChIP-seq of R-Smad1 and R-Smad3 had been carried out in murine G1ER erythroid progenitor cells and murine pro-B cells, respectively [8,9]. Nevertheless, the widespread co-localisation of Ldb1 and R-Smads, albeit in different cell types, suggests the potential for functional cooperation between these factors.
Ldb1 does not bind DNA directly but has been shown to assemble complexes containing Scl (Tal1) and Gata1/2 on DNA via motifs including Ebox, GATA, and Ets [23,24,29]. Genome-wide comparison of ChIP-seq datasets suggests that Scl and Gata1/2 co-occupy a substantial subset of Ldb1-binding sites with R-Smad1 or R-Smad3 (S1 Fig.). Indeed, the most enriched motifs identified in genomic sequences bound by R-Smads also include GATA, Ebox and Ets [8,9]. Taken together, these observations identify Ldb proteins as potential modulators of TGFβ superfamily signalling, possibly by associating with R-Smads to regulate TGFβ targets.
To analyse the role of Ldbs in TGFβ signalling in vivo, we first monitored their expression during early embryonic development when TGFβ family members are known to be critical. Throughout early zebrafish development, ldb2a shows greater specificity than the ubiquitous ldb1a, ldb1b, or ldb2b (S2 Fig. and data retrieved from the Zebrafish Information Network (ZFIN) [30]). At 15 hours post fertilisation (hpf), ldb2a is present in the notochord and the lateral mesoderm, which gives rise to haematopoietic, endothelial, and pronephric derivatives (S2A Fig.). At 26 hpf, ldb2a expression continues in and around the blood vessels (S2B Fig.). Maternal/zygotic ldb2a is expressed ubiquitously throughout cleavage and blastula stage (0–4.7 hpf) embryos (S2C–S2F Fig.), but immediately before and during gastrulation (4.7–10 hpf), ldb2a becomes more specific in the yolk syncytial layer (YSL) (Figs. 2A and S2F, white arrowheads), an important source of Nodal signalling crucial for the specification of gastrula germ layers. This suggests a possible role for Ldb2a in signalling by this TGFβ superfamily member, we therefore initially focussed our studies on the function of Ldb2a in Nodal signalling during gastrula embryonic development.
To determine if Ldb2a functions in Nodal signal transduction (illustrated in Fig. 2B–2E), we knocked it down using two antisense morpholinos (MOs), a splice MO targeting the boundary of intron3 and exon4, and a MO targeting the ATG site (S3A–S3E Fig.). Both MOs cause similar defects (S3F–S3K Fig.), and co-injection of ldb2a mRNA with the splice MO was able to rescue ldb2a morphant phenotypes (S3L–S3T Fig.). Moreover, we injected NLS-Cas9 protein together with a small guide RNA targeting the ATG of ldb2a, and observed that a significant proportion of resultant mosaic F0 mutants phenocopy the morphants (S3U–S3W Fig.). Altogether, these data confirm the specificity of the ldb2a MOs. Upon ldb2a knockdown, we saw an increase in the level of the phosphorylated Nodal effector, p-Smad2, by the shield stage (6 hpf), while the level of total Smad2/3 was comparable to the wild-type control (Fig. 2B). We also observed up-regulated activity of a TGFβ reporter (SBE-luciferase [31]) (Fig. 2C). Thus, ldb2a knockdown up-regulates Nodal signalling, suggesting that Ldb2a normally acts to suppress Nodal signalling.
Another TGFβ superfamily member, BMP, plays critical roles during gastrulation and signals through R-Smad1, which also co-occupies the genome with Ldb1 (Figs. 1A and S1A). We therefore examined the BMP signal transduction pathway in ldb2a morphants. The activity was unaffected at the shield stage (S4A and S4B Fig.) but significantly increased by the end of gastrulation (the tailbud stage, 10 hpf), as shown by the level of p-Smad1/5/8 and the activity of a BMP-specific reporter (Id1-BRE2-luciferase [32]) (Fig. 2D and 2E). Thus, ldb2a loss-of-function promotes BMP signal transduction, suggesting that Ldb2a normally acts to suppress BMP signalling.
The consequences of the excessive Nodal signalling in ldb2a morphants included up-regulation of the Nodal-induced genes, cyclops (cyc) and squint (sqt) (Fig. 2F–2I’). Expression of bmp4 was also increased by the tailbud stage (Fig. 2J–2K’) and remained up-regulated during somitogenesis (S4C and S4D Fig.). These genes code for ligands that implement positive feedback to sustain and propagate signalling. Taken together, ldb2a knockdown enhances expression of Nodal and BMP ligands, suggesting a negative effect of Ldb2a on positive feedback for Nodal and BMP signalling.
In addition to the expression of ligands, readout of Nodal signalling also includes expression of various germ layer genes, as Nodal induces the mesendoderm while restricting the ectoderm [33–35]. Consistent with the excessive Nodal signalling observed in ldb2a morphants, expression of ntl, a mesendoderm marker, was expanded towards the presumptive ectoderm (Fig. 3A and 3B), while expression of gata2, a non-neural ectoderm marker, and otx2, a neural ectoderm marker, was reduced (Figs. 3C, 3D, S5A–S5B’). In addition, another Nodal target, mixer/bon, expressed in the mesendoderm at the onset of gastrulation and becoming restricted to the endoderm during late gastrulation [36,37], and critical for proper endoderm specification in a Nodal-dependent manner [38], displayed increased expression in ldb2a morphants at the shield and 80% epiboly stages, suggesting a critical role for Ldb2a in the specification of endoderm (S5C–S5F’ Fig.). Taken together, these data suggest that some of the ectoderm is converted to mesoderm and endoderm in ldb2a morphants, consistent with the excessive Nodal signalling observed in these embryos.
To monitor the stability of the patterning effects of Ldb2a via Nodal, we examined genes expressed in mesendoderm-derived tissues of ldb2a morphants at later stages. At the 13-somite stage (∼15 hpf), markers of the mesendoderm-derived lateral mesoderm, such as a lateral mesodermal gene, hand2, a pronephric duct gene, pax2.1, and a haemangioblast gene, scl, displayed up-regulated expression in ldb2a morphants (Fig. 3F and 3G). We also observed up-regulation of other lateral mesodermal genes, including the haemangioblast genes lmo2, gata2, and fli1, erythroid genes gata1 and draculin, a myeloid gene pu.1, and the pronephric duct genes pax8 and lim1 (Figs. 3H, 3I, and S6A–S6N). To quantify expression of genes in the lateral mesoderm, we performed quantitative real-time PCR (qPCR) analyses and observed an increased level of fli1 RNA in ldb2a morphants at the 12-somite stage (S6O Fig.). In addition, Tg(gata1a:GFP)la781 zebrafish embryos injected with the ldb2a MO showed a clear up-regulation of GFP expression (S6P and S6Q Fig.), indicating an increase in the protein level of Gata1, but also in the number of Gata1 positive cells.
Consistent with the unchanged BMP activity at the beginning of gastrulation, dorsoventral patterning of ldb2a morphants remained balanced, shown by increased expression of both a ventral mesendoderm marker, eve1, and a dorsal mesendoderm marker, gsc (S5G–S5J’ Fig.). However, the activity of BMP signalling and expression of bmp4 became up-regulated in ldb2a morphants during late gastrulation (Fig. 2D, 2E, 2J, and 2K’), when high level BMP continues to specify ventral and posterior mesodermal tissues. After gastrulation, we indeed observed increased expression of genes marking the lateral mesoderm, derived from the ventro-posterior mesoderm (Figs. 3F–3I and S6).
To further investigate the effects of Ldb2a activity via a combination of Nodal and BMP after gastrulation, we examined expression of paraxial and dorsal mesodermal genes in ldb2a morphants. They were indeed up-regulated (by excessive Nodal) but less severely compared to the ventrally expressed genes (influenced by both Nodal and BMP), as shown by increased expression of shh (notochord) and myoD (somite) in the 10%–30% most affected ldb2a morphants (Fig. 3J–3M). Furthermore, the effect of ldb2a knockdown in the ventro-lateral mesendoderm-derived tissues remained evident until 24 hpf, when we observed up-regulated expression of flk1, tie1, dll4, and deltaC in endothelial cells of ldb2a morphants (Figs. 3N, 3O, and S7A–S7F). Taken together, our findings indicate that ldb2a loss-of-function induces mesodermal and endodermal while restricting ectodermal fates, especially in the ventro-lateral regions, and that this fate change is stable (Fig. 3P).
To confirm that the ectopic mesendoderm formation in ldb2a morphants is due to the up-regulation of Nodal and BMP signalling, we tried to reverse the effects by reducing these signals. When treated with an Alk4/5/7 (Nodal/Activin/TGFβ receptors) inhibitor, SB431542, ldb2a morphants were rescued with respect to ectopic expression of cyc (Fig. 4A–4C) and of scl and pax2.1 (Fig. 4D–4F). Moreover, bmp4 knockdown by MO injection also rescued the increased expression of scl and pax2.1 in ldb2a morphants (Fig. 4G–4I). These observations suggest that Ldb2a functions through Nodal signalling to restrict the specification of mesendoderm and through BMP signalling to restrict the specification of ventro-lateral mesendoderm.
Under normal circumstances, once Nodal signalling is up-regulated, negative feedback dampens excess signalling. However, the fact that a stable Nodal-dependent effect of ldb2a knockdown was seen suggests that negative feedback might not be fully active. Such feedbacks for both Nodal and BMP can be mediated by their common inhibitor, I-Smad7 [10–12,39]. Smad7 antagonises Nodal and BMP signal transduction via multiple mechanisms, dampening the phosphorylation of R-Smads, the formation of R-Smad/Co-Smad4 complexes, or the binding of R-Smad/Co-Smad4 to DNA [40–45]. By causing disruption of these mechanisms, altered Smad7 levels can eventually lead to changes in expression of Nodal targets, including ligand and mesendodermal genes. We first confirmed the role of Smad7 as a Nodal inhibitor in zebrafish embryos, showing that cyc expression was increased by smad7 MO knockdown (S8 Fig.) [12], but decreased by smad7 overexpression (Fig. 5A–5C). Loss-of-smad7 also increased expression of the Nodal target, mixer, in the mesendoderm (Fig. 5D and 5E). We then showed that indeed Smad7-mediated negative feedback is defective in ldb2a morphants, as shown by decreased levels of Smad7 mRNA and protein (Fig. 5F and 5G). Importantly, the increased cyc expression in ldb2a morphants was further up-regulated by co-injection of a level of smad7 MO that did not give a phenotype on its own. This synergistic effect between ldb2a and smad7 MOs implies that they function in the same pathway.
Leftys also mediate auto-regulatory negative feedback for Nodal signalling [4]. However, as a direct target induced by Nodal, expression of lefty1 was increased, as opposed to decreased like smad7, in ldb2a morphants (Fig. 5L and 5M), consistent with the excessive Nodal signalling in these embryos. Moreover, Ldb2a and Smad7 are synergistic on lefty1 expression (Fig. 5N and 5O), as seen for mixer. Therefore, Ldb2a is required for the negative feedback driven by Smad7 but not by Lefty1.
Upon ldb2a knockdown, expression of Nodal ligands and I-smad7 was affected immediately after the mid-blastula transition (MBT) (Fig. 5F, 5I, and 5M), suggesting that the regulation of these genes by Ldb2a may be direct. Indeed, ChIP of zebrafish shield-stage embryos followed by qPCR analysis showed an enrichment of Ldb2a at the promoter of smad7 and upstream of the Sqt ATG site (Fig. 6A, with primers shown in 6B). For ChIP-qPCR analysis in zebrafish, we adapted the in vivo biotinylation method described by de Boer and colleagues [46] for the zebrafish system. We injected low-level Avi (biotin acceptor peptide)-tagged ldb2a mRNA that does not cause any defect on its own (S9 Fig.), together with NLS-BirA (bacterial biotin ligase), in order to biotinylate Ldb2a in vivo; we then precipitated Biotin-Ldb2a-chromatin using streptavidin beads for subsequent analyses. We previously showed that the loss of Ldb2a exerted opposite effects on expression of different sets of genes induced by the same R-Smad pathways (i.e., down-regulation of I-Smad7 and up-regulation of Nodal/BMP ligands). Altogether these data suggest that Ldb2a directly activates expression of Smad7 but suppresses that of TGFβ family ligand genes, uncoupling the negative and positive feedbacks that are otherwise induced by the same R-Smad signalling.
To further explore how Ldb2a regulates expression of these genes, we mined published protein partner and DNA binding site datasets for Ldbs. Most of our current knowledge of the Ldb family is from studies of Ldb1. Since Ldb1 and Ldb2 share highly conserved protein sequence and structure, they likely function through similar mechanisms. In haematopoietic lineages, Ldb1 functions as a bridging molecule, with Lmo2/4, to assemble TF complexes that bind DNA through SBE, E-box, GATA, and Ets motifs [21,23,24]. LMO4 interacts with R-SMADs to mediate the TGFβ inputs in human epithelial cells [27]. Other components of Ldb1 complexes, such as Gata1/2, have also been shown to modulate TGFβ family signalling by assembling and recruiting Smad complexes onto TGFβ target genes [9,26]. The Smad7 and Sqt genes contain conserved SBE, E-box, and GATA motifs (Fig. 6B) [47], which are known to be enriched at Smad and/or Ldb binding sites [8,9,21]. As ChIP-seq data comparison suggests that Ldb1 co-localises with R-Smad3 at the I-Smad7 gene (Fig. 1D), Ldb1/2 might assemble TF complexes to recruit R-Smads to the Smad7 locus. As previously shown, direct binding of Ldb1 at I-Smad7 was confirmed by ChIP-qPCR in either murine EBs or Flk1+ cells (Fig. 1E), supporting our observations. Taken together, we provide evidence that Ldb2a acts together with R-Smads to bind Smad7 at the SBE/E-box and directly activates TGFβ-induced expression of Smad7. On the other hand, Ldb2a suppresses Sqt expression, possibly via forming a repressor complex binding to the Sqt locus. Thus the homeostatic mechanism regulating Nodal/BMP levels of signalling requires both positive and negative control by Ldb2a complexes. Deficiency of ldb2a caused dysregulation of I-smad7 expression, which subsequently disrupted the negative auto-regulating circuit, contributing to excessive activation of Nodal/BMP signalling via unrestricted positive feedback.
We conclude that Ldb2a plays critical roles in controlling both negative and positive feedback on TGFβ signalling in vivo, discriminating the responses of the I-Smad7-mediated negative feedback from the ligand-driven positive feedback. Disruption of this apparatus makes a substantial impact on embryonic development.
To gain further insights into the role of Ldb2a in TGFβ signalling, we studied the regulation of ldb2a expression by Nodal signalling. ChIP-seq datasets show an enrichment of R-Smad3 at the Ldb2 locus in various cell types and this enrichment can be stimulated by Activin/Nodal signalling (Fig. 6C) [8,9,48]. To study whether Ldb2a is regulated by TGFβ signals, we treated zebrafish embryos with the Nodal inhibitor SB431542, from the MBT stage. We examined expression of ldb2a and other Nodal targets at 0.5 and 3 hours after treatment, and showed that their expression was decreased by the blockade of Nodal signalling (Fig. 6D). Thus, an Ldb2a-mediated coherent feed-forward loop delays the activation of Smad7 expression and the suppression of ligand gene expression. As a consequence, Ldb2a discriminates the response speed of the positive and negative feedback circuits during signal propagation, allowing the accumulation of signalling through unrestricted positive feedback before negative feedback becomes fully established.
Altogether our data suggest the following model: during the initiation and propagation of TGFβ signalling, expression of ligands is immediately up-regulated, whereas I-Smad7 transcription is delayed by its requirement for Ldb2a, which gradually accumulates in response to the same signal (Fig. 6E). This mechanism allows signalling to self-amplify until adequate levels of Ldb2a enable the fully active Smad7-driven negative feedback, together with the direct restriction of positive feedback, to dampen excess signalling. Thus, the coherent feed-forward loop involving Ldb2a serves to delay the activation of negative feedback and the suppression of positive feedback. Despite the maternal expression of Ldb2a, this mechanism is likely to be specifically active during zygotic transcription, as phenotypes shown here were mainly caused by a splice MO that only knocks down zygotic Ldb2a. In agreement with this hypothesis, the level of maternal ldb2a RNA drops around the MBT stage, just before its zygotic expression increases (S2C Fig.).
We conclude that Ldb2a plays critical roles in stabilising TF complexes that control both negative and positive feedback on TGFβ signalling in vivo. It utilises a feed-forward circuit that discriminates the responses of the Smad7-mediated negative feedback from ligand-driven positive feedback. Disruption of this apparatus makes a substantial impact on embryonic development.
We have compared published ChIP-seq datasets of R-Smads and Ldb1 complex components, and shown that they co-occupy a significant proportion of the genome in different cell types, which suggests potential roles for Ldbs in TGFβ signalling. This was validated by in vivo studies showing that Ldb2a does indeed modulate R-Smad/TGFβ family signalling during zebrafish development.
Ldbs are non-DNA binding adaptor proteins, mediating the formation of TF complexes containing partners that are also crucial for TGFβ pathways. For example, LMO4, another non-DNA binding protein in Ldb complexes, interacts with R-SMAD1, 2, 5, 8, and Co-Smad4, in response to TGFβ signalling in human epithelial cells [27]. GATA1, a TF in Ldb complexes, has been shown to assemble with SMAD1 on BMP response elements (BREs) in human HepG2 (liver hepatocellular) cells and is required for strong activation of a BRE in the first intron of Smad7 [26]. In addition, another TF in Ldb1 complexes, Gata2 [21], also co-occupies genomic sites with Smad1 in murine erythroid progenitors [9]. Gata1 was also shown to direct Smad1 binding to erythroid-specific genes during erythroid differentiation. Altogether these observations suggest that Ldbs may nucleate R-Smad complexes to modulate TGFβ family signalling.
Known DNA binding motifs for the Smad and Ldb complexes were found in the Smad7 locus, including GATA, Ets, SBE, and Ebox, some of which having already been identified as active regulatory elements and required for TGFβ inducibility of I-Smads in human cells [14,26,49]. We have also shown that Ldb2a co-binds the conserved R-Smad binding site in the I-Smad7 promoter and directly activates I-Smad7 expression. On the other hand, Ldb2a also binds the Nodal ligand gene, Sqt, but represses its expression. This effect is also likely to be direct, because the expression of Nodal ligands increased immediately after the MBT when the ldb2a splice MO could only just have begun to have an effect. It has been shown that the first intron of Sqt, the promoter/proximal upstream region, and a distal upstream sequence together drive expression of the reporter gene in axial mesoderm, which does not reflect endogenous sqt expression [47], suggesting the existence of an element responsible for repressing sqt expression beyond the genomic regions used. Our ChIP-qPCR analyses showed that Ldb2a binds the Sqt locus, and expression of Nodal ligands/targets in the axial mesoderm was indeed increased by ldb2a knockdown. Thus, our findings and evidence from the literature suggest that Ldb2a represses sqt expression by binding to an unknown regulatory element.
The ChIP assay of Ldb2a in zebrafish has been a great challenge because Ldb proteins do not directly bind DNA. Moreover, few antibodies work for ChIP assays in zebrafish, including the zLdb2a antibody we generated. We therefore injected ldb2a mRNA tagged by HA or biotin at low enough doses to not cause any morphological or phenotypic disruption. The biotin-ChIP succeeded in detecting the direct binding of Ldb2a at I-Smad7 and Sqt. The ChIP assays were performed during early gastrulation when, like the injected RNA, Ldb2a is active in most cells of the embryo. Thus, our observations are likely to reflect physiological interactions.
The loss of ldb2a in zebrafish embryos increased the phosphorylation of R-Smads and the activity of TGFβ-responsive cis-regulatory elements, as well as the expression of TGFβ target genes. These observations suggest that Ldb2a normally restricts Nodal/BMP signal transduction and our subsequent experiments show that both an increase in ligand expression and a loss of smad7 expression contribute to the signalling perturbation seen in ldb2a morphants.
Knockdown of ldb2a led to excessive specification of mesendoderm and derivatives during development. Chemically restricting Nodal activity rescued the ectopic mesendoderm induction caused by ldb2a knockdown, while bmp4 loss-of-function rescued the extra increase in lateral mesoderm specification. Therefore, Ldb2a functions in embryonic patterning through Nodal and BMP signalling. Reflecting the elevation of both signalling pathways, the effect of ldb2a depletion on the ventro-lateral and posterior mesendoderm fates (e.g., blood, vasculature, pronephric, and tail mesodermal tissues) was more significant than on other mesodermal lineages (e.g., trunk somites, notochord, heart, and head mesodermal tissues), as the ventro-lateral and posterior mesendoderm is formed by exposure to a combination of Nodal and BMP morphogens during gastrulation [33–35]. We have therefore shown that disruption of the Ldb2a-controlled responses to TGFβ signals makes a substantial impact on embryonic development.
Insight into the biological significance of the discrimination among R-Smad targets by Ldb2a was provided by the discovery that the Ldb2a gene might itself be bound by R-Smads and transcribed in response to TGFβ family signalling. Thus, an Ldb2a-mediated coherent feed-forward loop slows down the transcriptional response of I-Smad7. As a consequence, Ldb2a discriminates the response speeds of the positive and negative feedback circuits during signal propagation, allowing the accumulation of signalling through positive feedback before the negative feedback is fully established. Recent publications [20,50] have provided mathematical simulations and experimental investigations suggesting that coupled positive and negative feedback circuits enable cellular systems to produce optimised responses to stimuli with respect to signal duration and amplitude. Here for the first time, we have shown that the two feedback pathways can be uncoupled.
All animal experiments were performed under a Home Office Licence according to the Animals Scientific Procedures Act 1986, UK, and approved by local ethics committees.
The ChIP-seq datasets of each protein (Smad1, Smad3, Ldb1, Scl/Tal1, Gata2, and Gata1) were downloaded from the NCBI gene expression omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo). For Smad1 (ChIP-seq in murine G1ER cells), Smad3 (murine pro-B cells), and Ldb1 (murine bone marrow cells), their mapped reads on the MM8 genome (bed format) were used for peak calling analysis using MACS (version 1.4.2), while IgG was used as the negative control. Genome-wide comparison of ChIP-seq datasets was performed as previously described [8]. Briefly, the location of Smad1/3 binding (query datasets, shown in x-axes) (Fig. 1A and 1B) in relation to Smad1/3- or Ldb1-enriched sites (base datasets, y-axes) was visualised by Java Treeview with the average reads density calculated in 100-bp bins ±2.5 kb around each Smad peak position suggested by MACS. These plots show the overlaps between Ldb1 binding regions and the enriched sites of Smad1/3 genome-wide. The location of Scl, Gata2, Smad1, Gata1, and Smad3 binding in relation to Ldb1-enriched sites was also visualised (S1 Fig.). These plots show the overlaps between Ldb1 binding sites and the enriched regions of the other five proteins genome-wide.
Wild-type and Tg(gata1a:GFP)la781 [51] embryos and adult fish were bred and maintained as described [52].
MO oligonucleotides (S2 Table, GeneTools) were dissolved in Milli-Q water to 25 ng/μl and stored at room temperature. Micro-injections were performed with 1 nl of each MO injected into the yolk cell of 1–2-cell stage embryos, at concentrations shown in S2 Table.
To generate GFP-tagged ldb2a mRNA for injection, the entire ldb2a reading frame was first cloned into the Gateway vector pDONR™221. Full-length ldb2a PCR fragments were generated via superscript III one-step RT-PCR system (Invitrogen) using total RNA extracted from 24 hpf embryos, with gLdb2 FWD1 and gLdb2 REV1 primers (S1 Table). Gateway cloning technology (Invitrogen) generated an ldb2a entry vector in pDONR221 back bone, which was sequenced and recombined with pCSGFP2 [53] to create a full length ldb2a-GFP plasmid, in which the ldb2a gene was placed immediately upstream of the GFP coding sequence.
Untagged or HA-tagged ldb2a fragments were amplified from 24 hpf cDNA with Ldb2-F3/Ldb2-R4, Ldb2-F2/Ldb2-R4, or Ldb2-F2/Ldb2-R5 primer pairs (S1 Table), cloned into pGEM-T easy vectors (Promega) and sequenced. To generate HA-tagged ldb2a mRNA for injection, ldb2a fragments were excised from ldb2a-pGEM-T entry vectors and cloned into the pCS2+ vector. To generate Avi-tagged ldb2a mRNA for injection, the Flag-ldb2a-2A fragment was amplified from the ldb2a-pGEM-T entry vector with 5′-Flag-ldb2a/3′-2A-ldb2a primers (S1 Table) by 2-step PCR using Phusion DNA Polymerase (NEB). The Flag-ldb2a-2A fragment was then annealed with an Avi-Tev-Flag oligo (S1 Table), followed by amplification of the Avi-Tev-Flag-ldb2a-2A fragment with 5′-Avi-fusionF/3′-2A-fusionR primers (S1 Table) and cloning into the pMTB2-eGFP vector using In-Fusion HD Cloning kit (Clontech).
Capped mRNA for micro-injection was in vitro transcribed from 1 μg linearised DNA template, using the Ambion mMESSAGE mMACHINE kits, and purified by QIAGEN RNeasy Micro kit, according to manufacturers’ instructions. Murine or zebrafish Smad7 mRNAs were synthesised from published Flag-pcDNA3-mSmad7 vectors [10] or a PCS2-zSmad7 construct [12], respectively. Synthesised mRNA was aliquoted and stored at −80°C, and injected to 1-cell stage zebrafish embryos.
Wild-type and ldb2a morphant embryos were treated with 25 μM or 100 μM SB431542 [54] from the 8-cell stage until collection at the sphere, shield, tailbud, or somitogenesis stages. Control embryos were treated with an equal volume of DMSO added to fish water.
Whole mount in situ hybridisation on zebrafish embryos was carried out as described [55]. Digoxigenin (DIG) or fluorescein labelled antisense RNA probes were transcribed from linearised templates using T3, T7, or Sp6 RNA polymerases (Roche). DIG and fluorescein antibodies were detected using BM-purple (Roche) or Fast Red [56], respectively.
Protein extracts were prepared according to Link and colleagues [57]. Primary antibodies were used at 1:500–1:2,000 dilutions. Antibodies used included: Phospho-Smad1/5 (Ser463/465) (41D10) (Cell Signaling number 9516); Phospho-Smad2 (Ser465/467) (Cell Signaling number 3101); Smad2/3 (N-19): sc-6032 (Santa Cruz); Smad6/7 (N-19): sc-7004 (Santa Cruz).
50 pg SBE-luciferase [31] or Id1-BRE2-luciferase [32] constructs were co-injected with ldb2a MO into the streaming yolk or the yolk-free cell of 1-cell stage zebrafish embryos. 50 pg pCMV-LacZ plasmids were co-injected to normalize injection efficiency. Gastrula stage embryos were collected and washed with PBS. 20–50 embryos were homogenised in 200 μl lysis buffer (provided in the Roche Luciferase Reporter Gene Assay kit) by aspirating through 23G syringes and incubated on ice for 10 minutes, followed by a brief centrifugation. Supernatants were separated into duplicates for each assay. 50 μl and 25 μl of the supernatant were used to measure the activity of luciferase and β-galactosidase, respectively, as described [58].
Total RNA was isolated with the RNAeasy Microkit (QIAGEN). Quantitative PCR was performed with SybrGreen (Applied Biosystem). Data were collected with the ABI-PRISM 7000 or 7500 Sequence Detection system. β-actin1/2, EF1α, and GAPDH were used as internal controls. The relative abundance for each sample was computed by the comparative method (∆∆Ct). Statistical analysis was by the two-sample equal variance t-test. Error bars indicate the standard deviation. Primers are listed in S1 Table. Previously published primers as described [59].
The sgRNA sequence targets the sense strand near the ATG of ldb2a. The template DNA of sgRNA was generated by PCR with Phusion polymerase (NEB) in HF buffer with a unique oligonucleotide encoding a T7 polymerase-binding site and the sgRNA target sequence (zLdb2a-ATG sgRNA F) and a reverse oligonucleotide encoding the remainder of the sgRNA sequence (sgRNA-R). In vitro transcription was performed with 100 ng purified DNA template using the Megascript T7 kit (Ambion), and sgRNA purified by phenol chloroform extraction and isopropanol precipitation. sgRNA was stored in aliquots at −80 °C. To generate ldb2a mutants, 1 ng NLS-Cas9 protein and 500 pg sgRNA were injected into the cell of 1-cell stage embryos. The control group was injected with 1 ng NLS-Cas9 alone.
Genomic DNA was extracted by homogenizing single zebrafish embryos in 20 μl of 50 mM NaOH, followed by incubation at 95°C for 8 minutes (gastrula embryos, older embryos require longer incubation), cooling to 4°C, and addition of 2 μl (10%) of 1 mM Tris-HCl (pH = 8) to neutralize the solution [60]. A 178-bp fragment spanning the sgRNA target site was amplified from control or mutant gDNA using the LC-Green Plus (BioFire Inc), HotShot Diamond PCR Master mix (Clent Lifescience), with ldb2a HRMA F1/ldb2a HRMA R1 primers. Details of the qPCR followed by HRMA were described previously [61]. PCR products from HRMA were cloned into pGEM-T vectors (Promega) and 16 colonies from each embryo were sequenced with T7 and SP6 primers.
ChIP-seq and ChIP-qPCR of endogenous Ldb1 (using anti-Ldb1 antibody N-18, Santa Cruz) on murine Flk1+ BL-CFCs isolated from day 4 EBs was performed as described [29]. 36-bp raw reads were mapped against NCBI build 37.1 of the mouse genome with ELAND (Illumina). Uniquely mapped reads were extended to 200 bp and then transformed into the genome-wide reads density (coverage) with the ShortRead Bioconductor package [62]. The coverage from ChIP and IgG control was visualized on a mirror of the UCSC genome browser.
ChIP-qPCR analyses of Ldb2a in zebrafish gastrula embryos were performed as described [63], using two different methods for the IP: (a) inject low-level (50 pg) HA-Ldb2a mRNA that does not cause any defects on its own, and then precipitate HA-Ldb2a using HA antibody-coupled dynabeads (Anti-HA tag antibody: ChIP Grade, abcam ab9110; Dynabeads Protein A for Immunoprecipitation, Novex); (b) inject 50 pg Avi-Ldb2a (Avi: biotin acceptor peptide) together with NLS-BirA (bacterial biotin ligase), and then precipitate Biotin-Ldb2a using Streptavidin-coupled Dynabeads (Dynabeads MyOne Streptavidin T1, Invitrogen). For (b), we adapted the in vivo biotinylation method described previously [46] for the zebrafish system.
The following previously published datasets were used: Ldb1, Scl, and Gata2 in murine bone marrow cells [21], Ldb1 in murine day 4 EB-derived Flk1+ cells [29], Smad1 and Gata1 in murine G1ER erythroid progenitors cells [9], Smad3 in murine pro-B cells [8].
|
10.1371/journal.pgen.1007902 | Intronic CNVs and gene expression variation in human populations | Introns can be extraordinarily large and they account for the majority of the DNA sequence in human genes. However, little is known about their population patterns of structural variation and their functional implication. By combining the most extensive maps of CNVs in human populations, we have found that intronic losses are the most frequent copy number variants (CNVs) in protein-coding genes in human, with 12,986 intronic deletions, affecting 4,147 genes (including 1,154 essential genes and 1,638 disease-related genes). This intronic length variation results in dozens of genes showing extreme population variability in size, with 40 genes with 10 or more different sizes and up to 150 allelic sizes. Intronic losses are frequent in evolutionarily ancient genes that are highly conserved at the protein sequence level. This result contrasts with losses overlapping exons, which are observed less often than expected by chance and almost exclusively affect primate-specific genes. An integrated analysis of CNVs and RNA-seq data showed that intronic loss can be associated with significant differences in gene expression levels in the population (CNV-eQTLs). These intronic CNV-eQTLs regions are enriched for intronic enhancers and can be associated with expression differences of other genes showing long distance intron-promoter 3D interactions. Our data suggests that intronic structural variation of protein-coding genes makes an important contribution to the variability of gene expression and splicing in human populations.
| Most human genes have introns that have to be removed after a gene is transcribed from DNA to RNA because they not encode information to translate RNA into proteins. As mutations in introns do not affect protein sequences, they are usually ignored when looking for normal or pathogenic genomic variation. However, introns comprise about half of the human non-coding genome and they can have important regulatory roles. We show that deletions of intronic regions appear more frequent than previously expected in the healthy population, with a significant proportion of genes with evolutionary ancient and essential functions carrying them. This finding was very surprising, as ancient genes tend to have high conservation of their coding sequence. However, we show that deletions of their non-coding intronic sequence can produce considerable changes in their locus length. We found that a significant number of these intronic deletions are associated with under- or over-expression of the affected genes or distant genes interacting in 3D. Our data suggests that the frequent gene length variation in protein-coding genes resulting from intronic CNVs might influence their regulation in different individuals.
| Most eukaryotic protein coding genes contain introns that are removed from the messenger RNA during the process of splicing. In humans, up to 35% of the sequenced genome corresponds to intronic sequence, while exons cover around the 2.8% of the genome (based on the genome version and gene set used for this study). Human introns can have very different lengths, contrarily to exons. This difference in intron length leads to substantial differences in size among human genes, which cause differences in the time taken to transcribe a gene from seconds to over 24 hours [1]. Indeed, intron size is highly conserved in genes associated with developmental patterning [2], suggesting that genes that require a precise time coordination of their transcription are reliant on a consistent transcript length. It has been suggested that selection could be acting to reduce the costs of transcription by keeping short introns in highly expressed genes [3], which are enriched in housekeeping essential functions [4]. Genes transcribed early in development [5–7] and genes involved in rapid biological responses [8] also conserve intron-poor structures. Interestingly, Keane and Seoighe [9] recently found that intron lengths of some genes tend to coevolve (their relative sizes co-vary across species) possibly because a precise temporal regulation of the expression of these genes is required. In fact, these genes tend to be coexpressed or participating in the same protein complexes [9].
It is well known that introns contribute to the control of gene expression by their inclusion of regulatory regions and non-coding functional RNA genes or directly by their length [10–12]. Despite the importance of introns in regulating transcription levels, transcription timing and splicing, little attention has been payed to their potential role in human population variability studies. A recent analysis of the literature has revealed a substantial amount of pathogenic variants located “deep” within introns (more than 100bp from exon-intron boundaries) which suggests that the sequence analysis of full introns may help to identify causal mutations for many undiagnosed clinical cases [13]. Given that direct associations between intronic mutations and certain diseases have been reported [13–16], we need to characterise the normal genetic variability in introns so we can better distinguish normal from pathogenic variations.
We studied the effect of structural intronic variants on protein coding gene loci in healthy humans using five copy number variant (CNV) maps of high resolution [17–21]. Most of these CNVs were detected using whole genome sequencing (WGS) data, which allows to determine the exact genomic boundaries of these variants. CNVs may have neutral, advantageous or deleterious consequences [22] and can be classified in gains (regions that are found duplicated when compared with expected number from the reference genome, which is 2 for autosomes), losses (homozygously or heterozygously deleted regions) and gain/loss CNVs (regions that are found duplicated in some individuals—or alleles—and deleted in others). Each of the maps in our study was derived from a different number of individuals, from different populations and using different techniques and algorithms for CNV detection (S1 Fig and S1 Fig). Due to these differences, each dataset provided us with a different set of CNVs (S1 Fig), which we analysed independently, excluding sex chromosomes and private variants.
CNVs affect genes in different ways depending on the degree of overlap with them. Some CNVs cover entire genes (from now on whole gene CNVs), other CNVs overlap with part of the coding sequence but not the whole gene (exonic CNVs) and other CNVs are found within purely intronic regions (intronic CNVs, not overlapping with any exon from any annotated isoform, Fig 1A). The latter group is the most common, with 63% of all CNVs falling within intronic regions, but remains the least studied. More than the 95% of these 12,986 intronic CNVs are losses (12,334) or gain/loss CNVs (652) (S1 Appendix [23]). The prevalence of losses in introns is in stark contrast with whole gene CNVs (1,412), which tend to be exclusively gains (55% of the cases) or gain/loss CNVs (25% of the cases) (Fig 1B–1D).
Surprisingly, purely intronic losses are not only the most prevalent form of CNV, but also they are observed more often than expected by chance in most datasets (S2 Table). We compared the observed values with expected distributions calculated using permutations in local and global background models (see Methods and S2 Fig). We find significantly more deletions (4.14–9.3%) falling in introns than expected in 3 out of the 5 maps (4.14% in Sudmant-Nature, P = 0.0002, global permutation test). For the sake of clarity, the P-values in the main text correspond to the results in Sudmant-Nature’s map [20] using the global background model (unless otherwise indicated). The results obtained with the alternative background model and with both models in additional maps are shown in the supplementary tables and figures.
In contrast with intronic deletions, there are 51.2% fewer coding deletions (overlapping with exons) than would be expected by chance (P < 1e-04, Sudmant Nature, global permutation test). These patterns are consistent using the two different background models (S2 Fig and S2 Table) and the enrichment is not limited to any specific range of intron sizes (S3 Fig).
The enrichment of deletions in introns might seem contradictory to what was originally reported by the 1000 Genomes (1KG) Project [20], as they stated that introns had less CNVs than expected by chance. However, we would like to note that they did not separate purely intronic from intron-exon overlapping deletions, while we are talking about strictly intronic deletions (see Methods for details). Indeed, if we group all purely intronic and intron-exon overlapping deletions together, we also observe a significant depletion (S2 Table, S2 Fig).
The enrichment of intronic deletions could be explained as a consequence of the negative selection of CNVs in exonic regions. To better understand the selective pressure on introns, we decided to compare the prevalence of deletions in intronic regions and in intergenic regions of similar size. Compared to intergenic regions, introns are less enriched with deletions (FC = 0.81, P = 2.23e-308, paired Student’s t-test). In addition to this, the deletions we find within intergenic regions are, on average, larger in intergenic fragments (FC = 1.14, P-value 6.23e-35, S4 Fig). In conclusion, intronic regions are less enriched in deletions than intergenic regions of similar size. These results suggest that the sequence and distance between exons are more conserved in intronic regions than in intergenic regions.
The percentage of each intron that can be lost in the population due to CNV losses is highly variable, from 0.03% to 98.1% (51bp to 293kb), representing a loss of the 0.01% to 77.5% of the total genic size (51bp to 893.4kb, Fig 2A–2C). Some examples of genes with a notable change in size after a single intronic deletion in one individual are the neuronal glutamate transporter SLC1A1 (Solute Carrier Family 1 Member 1), with a loss of the 37% of its genic size (Fig 2D) and the LINGO2 (Leucine Rich Repeat And Ig Domain Containing 2, alias LERN3 or LRRN6C) gene with a loss of the 34% of its size.
The combination of different intronic deletions within a gene can give place to alleles of several different sizes (Fig 2E). Following with the same two examples, in the dataset from the final phase of the 1KG Project [20], we found 5 different intronic deletions in SLC1A1. These deletions result in 8 different sizes of genes in the population, with individual losses ranging from 1.1kb to 48kb. In LINGO2, the 20 different deletions give place to 36 different gene lengths in the 1KG population, with losses of 51bp to 233kb. The gene with more different allele sizes in the 1KG population [20] is CSMD1 (CUB And Sushi Multiple Domains 1), with a total of 66 common intronic annotated deletions that, combined, produce 150 alleles of different sizes. Strikingly, CSMD1 is highly conserved at the protein level and is amongst the most intolerant genes to functional variation. According to the ranking of the RVIS (Residual Variation Intolerance Score) gene scores [24], which is based on the amount of genetic variation of each gene at an exome level, only 0.169% genes in the human genome are more intolerant to variation in their coding sequence than CSMD1. In summary, intolerance to variation in the coding sequence seems to be compatible with extreme variation in the intronic sequence. These losses might affect their regulation without affecting their protein structure.
A total of 1,638 OMIM genes carry intronic deletions in the population. Diseases associated to SLC1A1 (OMIM: 133550) include Dicarboxylic Aminoaciduria and susceptibility to Schizophrenia, while LINGO2 (OMIM: 609793) has variants associated with essential tremor and Parkinson disease and also has an intronic SNP associated with body mass [25]. CSMD1 has been associated to diseases such as Benign Adult Familial Myoclonic Epilepsy (Malacards [26], MCID: BNG079) and Smallpox (MCID: SML019). Interestingly, rare intronic deletions in this gene have been recently reported to be associated to both male and female infertility [27]. To better understand possible epistatic effects between protein-coding and intronic mutations, it will be useful to incorporate information about gene length variation in future studies of these disease genes.
Structural variants in the germline DNA constitute an important source of genetic variability that serves as the substrate for evolution. Therefore, dating the evolutionary age of genes allows the study of structural variants that were fixed millions of years ago. Whole gene CNVs are known to differentially affect genes depending of their evolutionary age, mainly involving evolutionary young genes [28]. Genes of younger ages are generally cell-type specific, while ancient genes tend to be more conserved, ubiquitously expressed and enriched in cellular essential functions. Intrigued to see many cases where intronic CNVs were affecting highly conserved protein-coding loci, we compared the distribution of coding (including exonic and whole gene) and intronic deletions across different gene ages (Fig 3). These and subsequent analyses were done using 3 maps: Sudmant-Nature’s [20], Zarrei’s [19] and Abyzov’s [17] maps. Handsaker’s [18] and Sudmant-Science’s [21] maps were discarded because they had very few intronic deletions (less than 1,000, S1 Fig).
We observed that most ancient genes are depleted of deletions that affect their coding regions, while primate-specific genes are enriched with coding CNVs (Fig 3A), meaning that the coding region of recent genes has a higher tendency to be lost or disrupted. This pattern was also observed when CNV gains were included (S5 Fig). The generation of random background models revealed that ancient genes (present in the Sarcopterygii ancestor) were significantly depleted of coding region losses (both exonic and whole gene, P < 1e-04, global permutation test), while these were enriched in young genes (from Hominoidea to Homo sapiens, P < 1e-04, global permutation test; see Fig 3A and S2 Table).
In contrast with coding deletions, the number of intronic deletions have a uniform distribution across gene ages, being slightly enriched in ancient genes in Sudmant-Nature’s map (P = 2e-04, global permutation test, Fig 3B and S2 Table). A similar pattern was also observed when taking only genes with big introns (larger than 1,500 bp, see S2 Table) and by calculating the enrichment within big introns independently from genes (S7 Fig and S2 Table). Remarkably, larger sizes of introns in ancient genes lead to a higher proportion of these genes being affected by intron deletions (S6 Fig). Therefore, while coding deletions are less frequent in ancient genes than in younger genes, intronic deletions are more frequent in the older ones (Fig 3C and S8 Fig).
We would expect that essential genes, which tend to be ancient [29], could be an exception to the enrichment of deletions. Essential genes have on average shorter introns than the rest of the genes [30,31] and relative to the genes of the same evolutionary age (S9A and S9B Fig). Up to 1,154 essential genes carry intronic deletions if we take into account all five CNV maps. In Sudmant Nature, 907 essential genes have intronic deletions, a higher number than expected by chance (P = 0.034, global permutation test, S2 Table).
We investigated if intron variability in genes was associated with any biological function. Genes with more or less intronic deletions than expected by chance (S3 Table, see Methods) were not associated to any particular function using DAVID [32]. Nevertheless, genes with less intronic deletions than expected show more protein-protein interactions among them than expected by chance (P = 2.43e-10, calculated with STRING [33]). These results are compatible with previous evolutionary studies that showed high levels of conservation of intron length in genes associated with development protein complexes in mammals [2], presumably to facilitate a more precise temporal regulation of expression [9].
Population stratification of CNVs has previously been suggested to be indicative of loci under adaptive selection [20,21]. We identified 352 highly stratified variants (HSVs, maximum Vst>0.2, see Methods) from Sudmant-Nature’s map overlapping with protein-coding genes: 282 are intronic, 53 exonic and 17 whole gene. We classified deleted regions according to the age of the genes and the type of gene structure affected and calculated the percentage of each group that is highly stratified (Fig 3D). Interestingly, the contribution of intronic HSVs is higher for younger genes, a pattern coherent with the expected higher functional impact of HSVs in older genes. Remarkably, the percentage of intronic HSVs is similar or higher than that of whole-gene and exonic HSVs in all age groups (and always higher than partial exonic deletions). These signatures of potential positive selection in purely-intronic CNVs suggest that a fraction of them might contribute to human adaptation.
Multiallelic CNVs affecting whole genes have been shown to correlate with gene expression: generally, the higher the number of copies of a gene, the higher its expression levels [18,20]. We hypothesized that intronic size variation may also impact the expression of the affected genes (without affecting the actual number of copies of the gene). Therefore, we looked into the possible effect of intronic hemizygous deletions on gene expression variation at the population level, comparing the effects with hemizygous deletions in coding (whole gene and exonic) and intergenic non-coding deletions (Fig 4). We used available RNA-seq data from Geuvadis [34] that was derived from lymphoblastoid cell lines for 445 individuals for whom we have the matching CNV data from the 1KG Project [20].
In order to look for differences in gene expression we selected variants for which we had at least 2 hemizygous individuals (individuals with copy number = 1) and at least 2 wild-type individuals (copy number = 2) and we compared the expression levels among these two groups to identify deleted CNV regions associated with expression quantitative trait loci (eQTL, Fig 4F). We will refer to the deleted regions associated with expression changes as DEL-eQTLs, and the genes associated with as eGenes. For comparative purposes, we first looked at the effect of hemizygous deletions in coding regions (whole gene and exonic DEL-eQTLs). We found that 7 eGenes out of 50 genes with whole gene deletion CNVs resulted in significant downregulation of gene expression in lymphoblastoid cell lines (14%, a higher number eGenes than expected by chance, P < 5e-4, permutation test, Fig 4F). In addition, we found 35 eGenes out of 437 genes with partial exonic deletions that were differentially expressed (8%, a number higher than expected by chance, P < 5e-4, permutation test, Fig 4F). The majority of these eGenes (32/35) where down-regulated in the individuals carrying the deletion.
Although intronic deletions do not affect the coding sequence of genes, we observed significant differences in gene expression in 53 eGenes out of the 1,505 genes with intronic deletions, a number of intronic-eGenes that is also higher than expected by chance (P < 51e-4, permutation test) (Fig 4F). Given the higher abundance of intronic deletions in the population, the absolute number of intronic-eGenes (53 genes) was similar to the total of coding-eGenes (39 genes, Fig 4F and S4 Table). Of the intronic-eGenes, 62% were downregulated and the other 38% upregulated, suggesting that intronic deletions might result both in enhancing or repressing gene expression (while coding losses mostly associate to gene down-regulation). Regulatory regions are known to be preferentially located in first introns [35]. From all 56 intronic eDeletions that are associated to changes of gene expression in our study, 17 (30.4%) are found within first introns. However, this percentage is not significantly higher than in non eDeletions (26%, P = 0.54, Fisher’s test). Finally, we identified that four of the intronic cis-eDeletions in lymphoblastoid cells are HSVs, suggesting adaptive potential of these expression differences. These intronic HSVs are located in four ancient genes (Sarcopterygii or older): EXOC2, SKAP2, PTGR1 and PHYHD1 (S10 Fig). EXOC2 is an essential gene encoding one of the proteins of the exocyst complex and is among the top 5% most conserved genes in human (RVIS = 3.34).
Since intron length can impact the inclusion of alternative exons [36], we hypothesised that there might be genes with differentially expressed transcripts (eTranscripts) in any gene containing an intronic deletion. In addition to the 53 intronic-eGenes, we found 217 intronic-eTranscripts in a total of 185 genes (this is more than expected by chance, P = 0.018, permutation test, Fig 4F and S5 Table). These results suggest that deletions within introns may cause the inclusion or exclusion of exons and thus influencing the relative proportion of alternative transcripts in many genes.
Changes in GC content as the result of intronic deletions might also contribute to these splicing differences, as in genes with long introns, the recognition of introns and exons by splicing machinery is based on their differential GC content [37,38] and the lower GC content in introns facilitates their recognition. We found that, in general, the deleted sequences have a significantly higher GC content to that of the introns where they are located (P = 1.8e-28, paired Student’s t-test), and the loss of these sequences causes a significant decrease of the overall GC content of the introns (P = 2.23e-16, paired Student’s t-test) (S11 Fig and S12 Fig). This drop of GC content is more pronounced in introns with deletions originated through transposable element insertion (TEI, P = 2.01e-9, paired Student’s t-test). The 84% of TEI deletions overlap almost completely with Alu elements (S13 Fig), which are known to be GC rich. The GC drops happening in introns with deletions associated to non-allelic homologous repair (NAHR) are less significant (P = 0.0063), while the difference is not significant in deletions caused by non-homologous repair (NH) (P = 0.7676). The drop in intronic GC content associated to most TEI and many NAHR deletions would increase the difference of GC content between introns and their flanking exons, what could facilitate exon definition during splicing and might contribute to the observed differential expression of some transcripts. It has been recently shown that human enhancers are associated to high GC [39] and that Alu elements can act as enhancers [40], suggesting that deletions could not only alter splicing but also influence regulatory features located within introns.
Introns in human are particularly enriched in regulatory regions and frequently interact with gene promoters of other genes via chromatin looping (Fig 4D). Therefore, deletions in introns that show long-range interactions with promoters of other genes could potentially affect their expression (trans effects). We used promoter-capture Hi-C published data for B-lymphocytes [41] to link intronic regions and gene promoters. We identified 322 deletions in intronic regions that interact with gene promoters of other genes (672 in total). Taking all combinations of genes and the trans-intronic regions with deletions, we searched for intronic trans-DEL-eQTLs: intronic regions that, when deleted, are associated with changes in expression of a different gene. Twelve of these genes were found to be significantly differentially expressed in the individuals presenting an intronic deletion in another gene (trans-intron-eGenes, Fig 4F and 4G). For example, PRSS36 (Protease, Serine 36) is downregulated in individuals with an intronic deletion in SETD1A (SET Domain Containing 1A) gene, while LIAS (Lipoic Acid Synthetase) gene is upregulated in individuals with a intronic deletion in PDS5A (PDS5 Cohesin Associated Factor A) (Fig 4G). In addition, 81 transcripts from 65 genes were also differentially expressed (trans-intron-eTranscripts) in the individuals with a trans-DEL-eQTLs. The loss of intergenic fragments in 3D contact with a gene were associated to a similar number of DEGs than the DEGs associated to intronic trans-DEL-eQTLs (16 trans-eGenes, 123 eTranscripts associated to intergenic deletions, Fig 4F and S5 Table).
We analysed the age of different types of eGenes and observed that whole-gene and exonic eGenes are enriched in young age classes (Fig 5A). This pattern is very different in intronic and intergenic eGenes: intronic cis-eGenes are enriched in old ages, while intronic trans-eGenes and intergenic-eGenes do not seem to be associated with gene age. If we compare the RVIS of the different types of eGenes, we find that whole gene and exonic eGenes are actually among the most tolerant genes to point mutations in their coding sequence (Fig 5B). In contrast, we found that a significant proportion of intronic cis-eGenes with low RVIS percentiles, indicating that protein-coding genes that are intolerant to point mutations at the protein level can have intronic deletions associated to gene expression changes. Strikingly, trans-eGenes show the lowest RVIS percentiles, indicating that intronic variation might impact the gene expression of interacting genes that are quite intolerant to coding mutations (Fig 5B).
To further study the potential impact of intronic deletions in regulatory regions, we analyzed the co-occurrence of these events with enhancers. In eGenes or eTranscripts, 15 intronic DEL-eQTLs overlap with enhancers (an overlap that is higher than expected by chance, P = 0.023, odds ratio = 2.04, Fisher’s test). These 15 deletions represent the 24% of the tested intronic deletions overlapping with enhancers in this cell type. We need to consider that many intronic deletions were not investigated because they fall within genes that are expressed in other cell types. Based on our observations in lymphoblastoid cells, we estimate that there might be 105 additional eDels of the 422 that overlap with enhancers. Regarding the deletions not overlapping with enhancers, we found that the distance between the DEL-eQTL and the closest enhancer is shorter than the distance of the deletions not associated with expression changes (P = 9.2e-04, Student’s t-test). These results suggest that intronic DEL-eQTLs could also be affecting interactions between promoter and intronic enhancers without directly disrupting the enhancer sequence.
Motivated by these findings, we investigated if there is a global tendency (independently of gene expression) for intronic deletions to affect or not affect enhancers. First, we observed that enhancers are enriched in introns (P < 1e-04, global permutation test) agreeing with previous findings in plants [42,43]. Strikingly, we find that intronic deletions and intronic enhancers co-occur in the same intron more often than expected by chance (P < 2.2e-16, Fisher’s test), possibly because most intronic deletions and intronic regulatory features are found in very long introns. However, by randomly relocating each intronic deletion within the same intron, we observed that the direct overlap of the deletions with enhancers is significantly lower than expected (P = 0.0304, global permutation test, S6 Table). A possible functional interpretation of these results is that there might be some degree of plasticity on the distance between intronic enhancer and promoters, but many intronic enhancers might be essential and cannot be lost. Interestingly, as we saw above, the loss of non-essential intronic enhancers can be associated to changes of gene expression.
Intronic CNVs constitute the most abundant form of CNV in protein-coding genes (Fig 1) and might have a previously unsuspected role in human evolution and disease. This variation in intronic length in healthy human populations implies that the actual size of many genes is different among individuals and, therefore, it might change in populations over time. However, little attention has been given to this variability even if gene length has been shown to be important in many genes.
We have shown that intronic deletions are occurring more often than expected by chance in three different CNV maps (using two different background models). Other studies have previously reported that CNVs are impoverished [20,44] or neither impoverished nor overrepresented within introns [45]. To explain this apparent controversy, we have to carefully review the different definitions of “intronic CNVs”. Here, we looked at deleted regions located completely within constitutive intronic regions (excluding intronic regions that contain alternative exons). Mu et al. [45] showed that purely intronic CNVs in general are not either enriched or impoverished in their dataset, but they observed that the subset of events associated to NAHR are found more often than expected by chance using the Pilot Phase dataset of the 1KG Project. We obtained similar results using the CNV map by Abyzov et al. [17] (Phase 1 of the 1KG) where intronic deletions were neither significantly enriched nor impoverished but the subset of NAHR deletions was significantly enriched (FC = 1.17, P = 0.0002). These results illustrate the importance of a clear definition of intronic CNVs and the danger of generalising the results of one particular study. Each study is normally biased to detect different mechanisms, different sizes or types of CNVs or events observed at different frequency in distinct populations. Finally, it should be noted that deletions annotated in CNV maps are not based on the ancestral human genome but on the reference genome [46]. In consequence, a fraction of the so-called deletions could in fact be inserted regions that are present in the reference. However, an additional comparative genomic analysis based on recent high quality primate assemblies [23] show that most of them correspond to actual deletion events in humans (S1 Appendix).
Our results suggest that copy number variation is shaping gene evolution in different ways depending on the age of genes, duplicating or deleting young genes and contributing to fine-tuning the regulation in both young and old genes (Fig 5C). Although we expect stronger functional effects for CNVs affecting the coding sequence, we have shown that intronic sequences are more conserved than intergenic regions of similar characteristics and that some purely intronic CNVs also show signatures of potential positive selection. Interestingly, the proportion of highly segregating intronic CNVs is similar or higher than for coding CNVs. Popadin et al. showed that primate-specific genes in human are enriched in single nucleotide variants correlated with gene expression (cis-eQTLs) with their associated SNPs tending to be closer to the TSS than in older genes [47]. These data highlight the need of dissecting the different types of genetic variation in order to understand the complex relationships between SNPs, CNVs, gene expression and gene age. While point mutations near the TSS [47] and coding CNVs seem to have a higher effect in young genes, intronic CNVs are frequently associated with gene expression variation in genes of any age. Finally, it is important to highlight that an unknown proportion of these strong statistical associations could actually be the result of other unexplored variants linked with certain CNV alleles.
Previously published studies on the effect of genetic variants on gene expression have proven the effect of CNVs on expression variability [48–50]. Chiang and co-workers identified 789 SVs associated to changes in gene expression, most of them (88.3%) not overlapping with exons from the eGene [50]. DeBoever and co-workers observed that a large proportion of common CNVs associated with gene expression levels is located in intergenic regulatory regions [49]. However, research on the subject has been mostly restricted to SVs found within 1Mb from the gene and previous works did not analyse intronic regions in detail. In contrast, we relied on Hi-C data to define deletions affecting regions in 3D contact with a gene. In this way, we do not require the CNV to be located within any particular distance to the TSS position of a gene. We tested intergenic eCNVs that can be located at any distance from 864bp to 82Mb from the nearest gene.
The differences that we observe in gene expression could be the result of intronic CNVs affecting the rate of transcription, the splicing process, the stability of RNA or a combination of them. For example, intronic deletions interfering with splicing recognition might trigger the nonsense-mediated decay (NMD) pathway that would degrade the transcript. Recently, it has been shown that the balance of unspliced and spliced mRNA (RNA velocity) is a cell type-specific signature that can be used to predict the future increases or decreases of gene expression in single cells [51]. As the amount of unspliced transcript detected will depend on the length of introns—which can be highly variable in some genes—we would expect that the RNA velocity of intron-varying genes will be also varying in human populations.
Despite the clear trends shown, our results are likely to underestimate the extent of the impact of intron losses in gene expression. On one hand, we only investigated the effect on gene expression in lymphoblastoid cell lines. On the other hand, the regulatory data currently available is also limited. The interaction maps change in different cell types [41,52] and many enhancers are tissue-specific [53]. Therefore, the loss of intronic sequence could affect the expression of genes in other cell types. In addition, the 3D contacts involving frequently deleted regions in the population will be underrepresented in the interaction map used in our study, as they are less likely to be present in the assayed samples. The availability of CNV, personal gene expression and genome interactomes from multiple tissues will allow to evaluate more accurately what is the impact of coding and non-coding deletions in the whole organism.
Whole genome CNV maps were downloaded from 5 different publications [17–21]. For our analysis we selected autosomal and not private CNVs. Some extra filters were applied to some maps: In Handsaker et al. we removed CNVs marked as low quality and all the variants from two of the individuals (NA07346 and NA11918) because they were not included in the phased map. From Zarrei’s maps we used the stringent map that considered CNVs that appeared in at least 2 individuals and in 2 studies. The complete list of CNVs analysed is available in S7 Table.
Autosomal gene structures and sequences were retrieved from Ensembl [54] (http://www.ensembl.org; version 75) and principal isoforms were determined according to the APPRIS database [55], Ensembl version 74. In order to avoid duplicate identification of introns, intronic regions were defined as regions within introns that aren’t coding in any transcript of any gene. When analyzing real introns, in order to avoid duplicate identification of introns, the principal isoform with a higher exonic content was taken. The complete list of genes affected by different types of CNVs is available in S8 Table.
The list of essential genes was obtained by aggregating lists of genes reported as essential after CRISPR-based genomic targeting [56,57], gene-trap insertional mutagenesis methodology [58], and shRNA [59–61].
An age was assigned to all duplicated genes as described before [28]. In the case of singletons gene ages were assigned from the last common ancestor to all the genes in their family according to the gene trees retrieved from Ensembl. Singleton’s ages can be noisy for genes suffering important alterations as gene fusion/fission events or divergence shifts. As a consequence, these ages should not be interpreted as the age of the oldest region of the gene, but as a restrictive definition of gene age considering a similar gene structure and gene product.
The ages (from ancient to recent) and number of genes per age are as follows: FungiMetazoa: 1119, Bilateria: 2892, Chordata: 1152, Euteleostomi: 8230, Sarcopterygii: 182, Tetrapoda: 154, Amniota: 408, Mammalia: 375, Theria: 515, Eutheria: 848, Simiiformes: 233, Catarrhini: 170, Hominoidea: 106, Hominidae: 64, HomoPanGorilla: 204, HomoSapiens: 500. For some analyses, Primates age groups (Simiiformes to HomoSapiens) were collapsed. For other analyses, we only grouped the 16 ages in three, “ancient” (collapsing groups from FungiMetazoa to Sarcopterygii), “middle” (from Tetrapoda to Eutheria) and “young” genes (Primates).
Intronic regions were assigned the evolutionary age of the gene they belonged to. In the cases when an intron could be assigned to more than one gene, the most recent age was assigned to them.
To estimate statistical significance of our results we performed permutation tests. In order to compare the number of overlaps of CNVs with genic functional elements we compared our observed values to a background model. A global background was obtained by relocating all the CNVs in the whole genome 10,000 times, avoiding low-mappability regions in R package “BSgenome.Hsapiens.UCSC.hg19.masked”). Genome coordinates and low mappability regions were downloaded using RegioneR package [62]. A local background was obtained by segmenting the genome in 278 windows of at least 10Mb and randomly shuffling the CNVs within their original window 10,000 times, also avoiding low-mappability regions. P-values were computed using a function derived from the permTest function from package RegioneR version 1.6.2 [62]. Code is available in https://github.com/orgs/IntronicCNVs.
We compared the location of the CNVs in our datasets and compared with their distribution in the random models in order to calculate enrichments or depletions depending on the intron size and gene age and essentiality.
To compare the content of deletions between intronic and intergenic DNA, we randomly selected a subset of 500 intronic regions and assigned an intergenic region with the most similar size to each of the introns. We then calculated the total number of deletions in the intronic and the intergenic compartments, as well as their sizes and the percentage of region that is lost. We repeated the sampling 10,000 times and compared (with a paired Student’s t-test) the distribution of deletion number and size in intronic versus intergenic regions.
We downloaded a genome-wide set of regions that are likely to be involved in gene regulation from the Ensembl Regulatory Build [63], assembled from IHEC epigenomic data [64]. We checked if introns are enriched in these regulatory features (promoters, enhancers, promoter flanking regions or insulators) by comparing to a random background model generated by relocating 10,000 times all regulatory features in the genome. P-values are the fraction of random values superior or inferior to the observed values.
In order to check for the significance of the overlaps between intronic deletions and regulatory features we relocated 10,000 times each intronic deletion within their host intronic region, avoiding overlaps with exons. Then, we compared the observed and the expected overlap with regulatory features. Introns that overlapped with low-mappability regions were previously removed.
Genomic sequences were obtained from the primary GRCh37/hg19 assembly, and were used for calculating the GC content of introns and intronic CNVs. Differences in GC content between a CNV and the intron where it is located were calculated with paired Student’s t-tests taking as statistical unit the CNV. The same was done for changes in intronic GC content before and after a deletion.
Alu element genomic coordinates were extracted from the RepeatMasker tracks from UCSC, build GRCh37.
The analysis of intronic deletions generated through different mechanisms was done using the dataset from Abyzov’s [20] study.
We used available RNA-seq data at Geuvadis [34] that was derived from lymphoblastoid cell lines for 445 individuals who were sequenced by the 1KG Project and for whom we have the intronic deletions in the largest CNV map [20]. We focused our analyses on the 763 genes that have only one intronic deletion in the population with at least two individuals affected in the Geuvadis dataset. For each of these genes we classified the PEER normalized gene expression levels [65] in two groups: 1) gene expression of individuals homozygous for the reference genotype and 2) gene expression of individuals with one allele with the deletion and the other with the reference genotype. We then performed Student’s t-tests to compare the expression of the two different genotypes. We corrected for multiple testing with p.adjust R function (Benjamini-Hochberg method). In addition, in order to see if the number of significant differentially expressed genes is higher than expected by chance, for each intronic deletion, we the shuffled 10,000 times the genotypes of the individuals and performed t-tests with the expression of the random groups of wild-type and heterozygous individuals. For example, if a deletion is found in heterozygosis in 50 individuals and the rest are wild-type, we will test if there is differential expression when comparing the expression of 50 randomly selected individuals versus the rest. By repeating this shuffling 10,000 times for every tested deletion we can calculate the expected percentages of significantly differentially expressed genes.
The number and size of expected intronic deletions per gene was calculated in two different ways: 1) relocating 10,000 times all deletions in the whole genome (except for low mappability regions) and 2) relocating 1,000 times all intronic deletions within the intronic regions. In both cases, a score was generated to determine what genes have more or less intronic deletions than expected. This score was calculated taking into account 1) the ranked position of the number of intronic deletions per gene divided by their median expected value, 2) the ranked position of the observed divided by the median expected size of the deletions, 3) the ranked position of the percentage of intronic content that is lost, 4) the ranked inverse of the expected intronic loss and 5) the ranked frequency of the deletion in the 1KG Project populations, if available. Because the frequency of the event depends on the reference genome, we find that a deletion present in, for example, all except for two individuals, should probably be considered as a rare gain and the deletion should be the reference. For this reason, the values were normalized in a way that 0.5 would be the maximum frequency and 0.9 and 0.1 would be given the same position in the ranking. Once all rankings were calculated and normalized from 0 to 1, a score was assigned to each gene by averaging their five ranks. The final set of 458 genes with less deletions than expected is the intersection of the top 500 genes of the two randomizations, and the set of 484 genes with more deletions than expected, the intersection of the bottom 500 genes.
Functional enrichment analysis of the genes with a lower scores and higher scores was performed with DAVID [32] and STRING [33]. Enrichment of essential genes in our datasets was performed with a Fisher test using our list of essential genes (see the “Essential genes” section in Materials and Methods).
For the study of population stratification of deletions, Vst statistics were extracted from Sudmant Nature [20]. As in Sudmant Nature, a cutoff of 0.2 was selected to indicate high population stratification of a locus.
|
10.1371/journal.ppat.1006137 | Naturally Acquired Human Immunity to Pneumococcus Is Dependent on Antibody to Protein Antigens | Naturally acquired immunity against invasive pneumococcal disease (IPD) is thought to be dependent on anti-capsular antibody. However nasopharyngeal colonisation by Streptococcus pneumoniae also induces antibody to protein antigens that could be protective. We have used human intravenous immunoglobulin preparation (IVIG), representing natural IgG responses to S. pneumoniae, to identify the classes of antigens that are functionally relevant for immunity to IPD. IgG in IVIG recognised capsular antigen and multiple S. pneumoniae protein antigens, with highly conserved patterns between different geographical sources of pooled human IgG. Incubation of S. pneumoniae in IVIG resulted in IgG binding to the bacteria, formation of bacterial aggregates, and enhanced phagocytosis even for unencapsulated S. pneumoniae strains, demonstrating the capsule was unlikely to be the dominant protective antigen. IgG binding to S. pneumoniae incubated in IVIG was reduced after partial chemical or genetic removal of bacterial surface proteins, and increased against a Streptococcus mitis strain expressing the S. pneumoniae protein PspC. In contrast, depletion of type-specific capsular antibody from IVIG did not affect IgG binding, opsonophagocytosis, or protection by passive vaccination against IPD in murine models. These results demonstrate that naturally acquired protection against IPD largely depends on antibody to protein antigens rather than the capsule.
| Streptococcus pneumoniae is a major global killer. Invasive pneumococcal disease (IPD) is the most severe form of infection. Surprisingly, the natural mechanisms of immunity to IPD in healthy individuals are unclear. The success of vaccines stimulating anti-capsular antibodies have led to the belief that the same mechanism lies behind natural protection. Using studies with pooled human immunoglobulin, we demonstrate that this is not the case and instead IgG recognising the bacterial surface proteins appears to have the dominant functional role. This finding supports efforts towards protein antigen-based vaccines, and opens the possibility of stratifying potential risk for individuals of IPD.
| Streptococcus pneumoniae is a leading cause of infectious disease related death, responsible annually for up to a million child deaths worldwide [1]. Pneumonia represents the greatest burden of disease caused by S. pneumoniae [2], and despite current vaccination strategies the burden of pneumococcal pneumonia remains high. Invasive pneumococcal disease (IPD) is the most severe form of S. pneumoniae infection and mainly affects very young children and older adults. This is attributed to an underdeveloped adaptive immune system in infants, and to waning natural immunity combined with co-morbidities in the older adult. A clear understanding of the mechanisms of natural-acquired adaptive immunity to S. pneumoniae is essential to characterise why both the young and elderly are at high risk of disease and for the development of effective preventative strategies. Vaccines based on the polysaccharide capsule of S. pneumoniae are highly protective against the capsular serotypes included in the vaccine preparation [3–5], and protection correlates with the level of anti-capsular antibody responses. It has generally been assumed that the type-specific anti-capsular antibodies that can develop in response to colonisation or episodes of infection are also the main mechanism of natural adaptive immunity against IPD [6, 7]. However, there is little good evidence supporting the concept that levels of anti-capsular antibodies predict risk of IPD in unvaccinated individuals.
As well as causing symptomatic disease, S. pneumoniae asymptomatically colonises the nasopharynx, affecting at least fifty percent of infants and approximately ten percent of adults [8]. Colonisation is an immunising event. In humans, it leads to antibody responses to capsular polysaccharide [9], but also induces both antibody [10–14] and cellular immune responses to protein antigens [15, 16]. Serum levels of antibody to multiple pneumococcal surface proteins rise in the first few years of life [13], and have been show to fall in older age for a limited number of antigens [17]. Similar adaptive immune responses are observed in mouse models of nasopharyngeal colonisation [11, 18–25]. In animal models, these anti-protein responses alone can be protective, with T-cell mediated immunity preventing re-colonisation and non-invasive pneumonia[15, 24, 25] and anti-protein antibody responses protecting against IPD [19, 20, 22, 24]. Recent human data suggests that Th17-cell mediated responses to protein antigens also play an important role in protection against colonisation in humans [26] with implications for vaccine design [27]. There are several converging lines of evidence from human studies which support the concept that naturally-acquired anti-protein antibodies can also protect against S. pneumoniae infections. Lower serum IgG levels to a range of pneumococcal proteins correlate with susceptibility to acute otitis media [28, 29] and respiratory tract infections in children [30]. Passive transfer of human serum from experimentally challenged human volunteers protected mice against invasive challenge with a different capsular serotype of pneumococcus [20], providing proof of concept that ‘natural’ antibodies against bacterial proteins induced through nasopharyngeal exposure can protect against IPD. Furthermore, the incidence of IPD falls after infancy for all serotypes of S. pneumoniae, irrespective of how commonly the serotype is carried in the nasopharynx [31] suggesting that naturally-induced adaptive immune mechanisms are serotype-independent. If the protection against IPD that develops naturally through colonisation requires anti-protein antibody responses rather than serotype-specific anti-capsular antibody, this would represent an important readjustment in our understanding of immunity to S. pneumoniae. It would have major implications for identifying subjects with an increased risk of infection, understanding mechanisms of immunosenescence that increase susceptibility to S. pneumoniae with age, and for guiding future vaccine design.
Passive transfer of pooled human immune globulin (IVIG) is an established treatment to prevent infections in individuals with primary antibody deficiency [32, 33], in whom S. pneumoniae is a leading cause of disease [34]. Previous investigations in mice have indicated that IVIG may protect against experimental IPD [35, 36]. Commercially-manufactured IVIG is pooled immunoglobulin G (IgG) from >1000 different donors [37], and therefore represents the pooled antibody responses acquired through natural exposure across a population. We have used IVIG to determine the targets of natural acquired immunity to S. pneumoniae and the relative functional importance of anti-capsular and anti-protein responses for prevention of IPD.
ELISAs using the whole S. pneumoniae cell as the antigenic target confirmed that IVIG contained significant titres of IgG that recognised S. pneumoniae (Table 1). Polysaccharide-specific ELISAs demonstrated that IVIG contained IgG that recognised common S. pneumoniae capsular serotypes and cell wall polysaccharide (CWPS) (Table 1). To assess whether IVIG contained IgG that bound S. pneumoniae proteins, immunoblots were performed against lysates of several S. pneumoniae strains of differing capsular serotypes. Multiple protein antigens were recognised by IVIG with a largely similar pattern of bands for all strains, suggesting the major protein targets of IVIG are generally conserved between capsular serotypes of S. pneumoniae (Fig 1A). Competitive inhibition was used to assess which antigens contributed significantly towards the whole cell ELISA titres for the TIGR4 strain. Pre-incubation of IVIG with a soluble bacterial lysate reduced whole cell ELISA IgG titres in a dose-dependent manner, which was partially reversed by pre-treating the soluble lysate with the protease trypsin (Fig 1B). In contrast, neither purified capsular polysaccharide nor CWPS affected whole cell ELISA IgG titres (Fig 1C). The whole cell ELISA assays were repeated for four different S. pneumoniae serotypes with competitive inhibition by encapsulated and unencapsulated bacterial lysates (Fig 2A–2D). The results demonstrated that for two of four strains lysates of encapsulated and unencapsulated bacteria equally reduced the IgG binding titre in the whole cell ELISAs. For the D39 (serotype 2) and serotype 3 strain whole cell ELISA titres were inhibited to a greater extent by lysates of encapsulated bacteria compared to unencapsulated bacteria. Further whole cell ELISAs for these two strains demonstrated that the unencapsulated mutants blocked IgG binding to unencapsulated mutants (Fig 2E and 2F), indicating the reduced inhibition in the whole cell ELISAs against the wild-type strain is likely to be due to the effects of anti-capsular antibody. These data show that IVIG contains antibodies to both capsular, CWP and protein antigens, but which class of antigens made the dominant contribution to IVIG recognition varied to an extent between S. pneumoniae strains when assessed using whole cell ELISAs.
To identify protein targets for IgG in IVIG, lysates of S. pneumoniae mutants lacking specific surface proteins were probed with IVIG. The results showed that IgG in IVIG recognised the cell wall proteins PspA, PspC and PhtD and at least two lipoproteins (shown using the lipoprotein deficient strain Δlgt), including PiaA (Fig 3A). Immunoblotting of recombinant proteins confirmed that IVIG contains IgG that recognises multiple (but not all) S. pneumoniae protein antigens tested (Fig 3B). To assess whether protein targets for naturally acquired IgG to S. pneumoniae were conserved between donors from different geographical regions we performed immunoblots against S. pneumoniae lysates with a further commercially available IVIG preparation (Vigam) obtained from the USA, and with sera pooled from 20 Malawian subjects. The results showing an almost identical band pattern for each source of IgG (Fig 3C), suggesting a high degree of consistency for the major protein antigen targets for IgG obtained from different geographical regions. A Luminex assay of antibody binding to 19 different S. pneumoniae surface proteins conjugated to xMAP beads was used to semi-quantify responses from different sources of pooled human IgG to specific protein antigens. The Luminex assay confirmed that IgG in IVIG recognised multiple protein antigens including PsaA, PpmA, PhtD, PhtE, PspA, pneumolysin (Ply) and PspC (Fig 3D). Overall, the strength of IgG binding to individual S. pneumoniae protein antigens between the different sources of antibody correlated strongly, with PhtD and PspC as the dominant antigens in all three sources of pooled human IgG (Fig 3D, and for correlation of Vigram versus Intratech R2 = 0.966).
To assess whether there is significant variation between individuals in which S. pneumoniae antigens are recognised by naturally acquired IgG, whole cell ELISAs to four S. pneumoniae serotypes, immunoblots against S. pneumoniae lysates, the Luminex assay of protein antigen responses, and capsular serotype antibody ELISAs were repeated using sera from six young adult HIV negative Malawian individuals (mean age 29 years, range 21 to 36, 3 male, 3 female). The results showed all the individuals investigated have significant anti-protein antibody responses (Fig 4). However, there were variations between individuals in whole cell ELISA titres to different S. pneumoniae strains (Fig 4A) and the levels of antibodies to some protein antigens as shown by variations in band strengths in the immunoblot (Fig 4B) and in the results for the Luminex bead assay (Fig 4C). For all the strains tested whole cell ELISA titres from individuals correlated with the mean anti-protein antigen responses, whereas there was no correlation to anti-capsule antibody levels except for the serotype 1 strain (S1 Fig). These data support the hypothesis that anti-protein responses dominate IgG recognition of S. pneumoniae in human sera.
To investigate whether anti-protein antigen responses could be affected by age, an electrochemiluminescence-based multiplex assay based on MesoScale Discovery (MSD, Rockville, MD, USA) technology [13] was used to measure responses to 27 protein antigens in sera from 10 individuals aged over 62 years (mean 67.2 years) and 10 young adult individuals (mean age 31.2 years). In general, mean anti-protein antigen responses were slightly lower for the aged subjects (Fig 3D), with the most marked differences being for PspC (Fig 3E) and PcpA (Fig 3F). The difference between older and younger sera reached statistical significance for PspC.
Functionally important IgG responses to S. pneumoniae were assessed using a flow cytometry assay to measure total IgG binding to intact live bacteria from different S. pneumoniae strains. Incubation in IVIG resulted in significant IgG binding to four different strains of S. pneumoniae. The level of IgG binding was either increased or unaffected when the assay was repeated using otherwise isogenic unencapsulated mutant derivatives of each strain, indicating that most of the IgG was binding to non-capsular antigens (Fig 5A and 5B). Conversely, pre-treatment with Pronase to degrade surface protein antigens (Fig 5C), using D39 mutant strains with reduced expression of dominant surface proteins (Δlgt, missing all lipoproteins, and ΔpspA/pspC missing the corresponding choline binding proteins) (Fig 5D), or pre-incubation of IVIG with an unencapsulated TIGR4 strain (Fig 5E), reduced the amount of IgG binding to the TIGR4 strain suggesting proteins were the target antigens. To further demonstrate that capsular polysaccharide was not the target for IgG binding, the assay was repeated using Streptococcus mitis strains genetically manipulated to express the serotype 4 S. pneumoniae capsule [37]. There was some binding of IgG in IVIG to the surface of the S. mitis strain indicating the presence of antibodies to surface antigens. However, there was no increase in IgG binding to the S. mitis strain expressing the S. pneumoniae serotype 4 capsule compared to wild-type S. mitis (Fig 5F). Conversely, expression by S. mitis pspC, one of the dominant S. pneumoniae protein antigens recognised by IgG in IVIG (Figs 3B, 3D, 4C and 4D), resulted in a large increase in IgG binding (Fig 5G). These results indicate that protein antigens (including lipoproteins, PspA and PspC) rather than capsular polysaccharide are the major surface targets for IgG binding to live S. pneumoniae.
To further assess whether immune recognition of live S. pneumoniae is dependent on IgG recognition of protein antigens, IgG from IVIG was selectively enriched for responses to S. pneumoniae protein antigens using antibody affinity purification columns coated with unencapsulated S. pneumoniae lysates. The enriched IVIG (eIVIG) preparation made using either the TIGR4 or D39 unencapsulated strains had a markedly higher whole cell ELISA titres to both the TIGR4 and D39 encapsulated S. pneumoniae strains compared to untreated IVIG (Fig 6A–6D). Despite the eIVIG preparation IgG concentration being only 30 μg/ml, approximately 1/150 the concentration in IVIG, incubation in eIVIG still resulted in IgG binding to S. pneumoniae in the flow cytometry assay (Fig 6E–6F). These data confirm that IgG targeting S. pneumoniae protein antigens can mediate IVIG immune recognition of S. pneumoniae.
IgG binding to S. pneumoniae can cross-link bacteria to form bacterial aggregates that are more susceptible to complement opsonisation [38]. Microscopy showed addition of IVIG to S. pneumoniae TIGR4 resulted in the formation of bacterial aggregates (Fig 7A), the relative size of which could be measured by flow cytometry using increases in forward scatter (Fig 7B). Both encapsulated and unencapsulated TIGR4 S. pneumoniae formed bacterial aggregates in IVIG, indicating these did not require recognition of capsular antigen (Fig 7A and 7B). Furthermore addition of IVIG restricted the increase in OD580 over time for different S. pneumoniae strains cultured in THY broth, and this effect was particularly noticeable for unencapsulated strains (Fig 7C–7F). Vigorous pipetting raised the OD580 to similar levels for both encapsulated and unencapsulated TIGR4 strains (Fig 7G), with no significant differences in numbers of bacterial CFU between the strains (log10 CFU / ml for the TIGR4 strain 7.60 SD 0.14, for the TIGR4Δcps 7.85 SD 0.11 after 6 h incubation in THY plus 10% IVIG). These results indicated that the reduction in the increase in OD580 over time in THY containing IVIG was due to formation of bacterial aggregates. When IVIG was pre-treated with papain to yield monovalent Fab fragments, the majority of the inhibitory effect of increase in OD580 effect was lost, confirming that bacterial aggregation was caused by cross-linking of bacterial cells via the Fab portions of IVIG (Fig 7H). Overall, the aggregation data demonstrate that the dominant target antigen for functionally important IgG binding to S. pneumoniae incubated in IVIG was not the polysaccharide capsule.
In vitro assays were used to assess the effects of IVIG on interactions of encapsulated and unencapsulated S. pneumoniae TIGR4 with phagocytes. Opsonisation with IVIG enhanced the association of S. pneumoniae with a murine macrophage cell lines (Fig 8A) and with fresh purified human neutrophils (Fig 8B), and enhanced neutrophil killing of S. pneumoniae (Fig 8C). For all three assays, IVIG had a proportionally greater effect on the unencapsulated strain than the encapsulated strain. These data support the hypothesis that anti-capsular IgG is not important in mediating the opsonophagocytic effects of natural human IgG present in IVIG.
Passive vaccination was used to investigate the protective efficacy of IVIG in different murine models of S. pneumoniae TIGR4 infection. Mice were given a total of 12.8mg of IVIG (Intratect, Germany, 40 g/L) in two separate i.p. injections 3 h and immediately before challenge with S. pneumoniae. This IVIG dose was selected as it is equivalent to the doses used in replacement therapy in primary immunodeficiency. In a test dose experiment, human IgG was readily detectable in the sera of IVIG-treated mice three h following the second intraperitoneal injection at approximately 1.5 g/L, within the same order of magnitude of circulating IgG levels in humans (7+ g/L) (Fig 9A). Human IgG was not detectable in mouse bronchoalveolar lavage fluid (BALF) in uninfected mice. Following S. pneumoniae lung infection by i.n. inoculation of 5x106 CFU of TIGR4, human IgG concentrations increased in BALF over time (Fig 9B) and correlated with BALF murine albumin levels, a marker of serum leak into alveolar spaces (Fig 9C). IVIG treatment had no effect on the inflammatory response to S. pneumoniae pneumonia, both in terms of inflammatory cell numbers (S2 Fig) or levels of the pro-inflammatory cytokine TNF-α in BALF post-infection (control group 2828 SEM 670 versus IVIG group 2665 SEM 506 pg/ml) in the lavage fluid following infection). IVIG treatment also had no effect on bacterial CFU in lavage fluid 2.5 hours after low dose inoculation with TIGR4 (Fig 9D), a time point and inoculum dose when alveolar macrophages are the main effector cell [38]. However, at 24 h following challenge, infected mice that had been pre-treated with IVIG were strongly protected against the development of bacteraemia (present in 100% of controls but only 17% of IVIG treated mice) and partially protected against lung infection, with 2 log10 fewer S. pneumoniae CFU in lung tissue compared to controls (Fig 9E). Pre-treatment with IVIG also protected mice against developing bacteraemia 4 h following direct i.v. bacterial challenge (Fig 9F). Protection against bacteraemia required macrophages, since their depletion by pre-treatment with liposomal clodronate (Fig 9G) reduced IVIG-dependent S. pneumoniae clearance from the blood (Fig 9H). The partial protection provided by IVIG within the lungs was lost when mice were depleted of neutrophils before infection by treatment with anti-Ly6G antibody (Fig 9I). Mice depleted of neutrophils failed to develop bacteraemia even without passive vaccination with IVIG. These data confirm that passive vaccination with IVIG strongly protects mice against IPD, and that protection was dependent on phagocytes.
To directly demonstrate that the protection afforded by IVIG is not mediated via anti-capsular antibody, IVIG was pre-treated to deplete anti-capsular antibody prior to testing its protective effects against IPD in vivo. Selective depletion of capsular serotype 4 specific antibody depletion was achieved by incubating IVIG with the S. mitis strain expressing the S. pneumoniae serotype 4 capsule. This process had no effect on the pattern and level of IgG binding to protein antigens in immunoblot and in ELISA for at least two specific proteins (Fig 10A). Whilst the depletion process almost completely removed serotype 4 anti-capsular IgG from the IVIG (Fig 10B), it had no effect on total IgG binding to the surface of S. pneumoniae when assessed by flow cytometry (Fig 10C). Passive transfer of IVIG depleted of type 4 serotype specific antibody to mice still protected against bacteraemia developing after i.n. inoculation of TIGR4 S. pneumoniae (Fig 10D), and after i.v. inoculation of TIGR4 restricted blood CFU to similar levels seen in mice given untreated IVIG (Fig 10E). These data confirm that IVIG does not require IgG to capsular polysaccharide to protect against invasive infection due to S. pneumoniae.
The bimodal distribution of S. pneumoniae infections in the very young and elderly suggests there is a significant degree of naturally-acquired immunity that evolves in early life and then wanes in later life. This naturally-acquired immunity is probably acquired through multiple episodes of nasopharyngeal colonisation with S. pneumoniae that repeatedly affect all humans rather than solely after disease episodes [16, 18–20, 24, 31]. Human epidemiological and experimental evidence from mouse models of infection suggest naturally-acquired immunity has a serotype-independent component [20, 28, 29, 31], yet the assumption remains that antibody to capsular antigen is the dominant mechanism of protection against IPD [6, 7]. As a consequence, clinical assessment of susceptibility to IPD is dependent on measuring anti-capsular IgG levels.
IVIG is a source of pooled IgG that contains naturally-acquired antibody to S. pneumoniae. We have used in vitro and in vivo experiments to compare the relative functional importance of the anti-capsular and anti-protein antigen IgG in mediating protection against S. pneumoniae. Overall, the data show greater importance for anti-protein rather than anti-capsular IgG, summarised as follows: (1) For both surface binding of IgG measured by flow cytometry and in vitro aggregation capsular antigen was not the main target for the four serotypes investigated. Data from the whole cell ELISAs were more mixed, with evidence of some contribution of anti-capsular IgG for two of the four strains assessed. However, IgG surface binding to live bacteria has been show to be a better surrogate for protection than ELISA titre [18]. (2) Enzymatic degradation of surface proteins, absorbtion of anti-protein antibody by incubation with unencapsulated TIGR4 strain, or reduced expression of some classes of surface proteins due to mutation all reduced total IgG binding to S. pneumoniae. (3) Expression by S. mitis of an immunodominant protein antigen but not the serotype 4 capsule increased IgG recognition when incubated in IVIG. (4) A low concentration of an IVIG derivative enriched for anti-protein responses to S. pneumoniae recognised heterologous S. pneumoniae strains in whole cell ELISA and flow cytometry IgG binding assays. (5) Loss of the capsule did not impair the protective effects of IgG in functional assays of neutrophil and macrophage phagocytosis of the TIGR4 S. pneumoniae strain. (6) Specific depletion of serotype 4 anti-capsular antibodies from IVIG had no effect on IgG binding to intact bacteria and did not abrogate the ability of IVIG to protect against IPD when tested in mouse models of infection. These data form the first evidence to our knowledge demonstrating the redundancy of naturally-acquired human IgG against capsular antigens in protection against IPD, with protection afforded by anti-protein antibody instead. By necessity, the four strains investigated represent only a proportion of the 97 S. pneumoniae capsular serotypes currently known [39], and we have only been able to deplete anti-capsular antibody for the serotype 4 strain as this is the only available S. pneumoniae capsular serotype expressed in S. mitis. Hence, although the in vitro aggregation and IgG binding data suggest capsule antigen is not functionally relevant for the four serotypes investigated, it remains possible that for selected serotypes anti-capsular antibody has a greater role in mediating protection against IPD than we have identified here. In addition, as we have not been able to make a sufficient quantity of an IVIG derivative effectively depleted of anti-protein antigen responses, we have not been able to explicitly demonstrate in the mouse model of infection that protection is dependent on anti-protein responses rather than to other potential non-capsule non-protein antigens. Our data also do not preclude an important role for naturally-acquired antibody to capsular antigens at other body sites, for example for prevention of nasopharyngeal colonisation [40]. Despite these caveats, the different strands of data we have presented here provide strong support for the hypothesis that the protection in humans against IPD mediated by naturally acquired IgG is not dependent on capsular antibodies. Instead protection seems to require recognition of bacterial surface proteins.
Protection against S. pneumoniae infection depends on phagocytes, with different cell types having dominant roles at different anatomical sites and at different time points. Alveolar macrophages are important for bacterial clearance during early lung function [38], whereas recruited neutrophils are important for controlling bacterial numbers in the lung at later time points [41]. In mice at least, protection against S. pneumoniae bacteraemia and therefore IPD is highly dependent on splenic and reticuloendothelial macrophages [42]. In the mouse model of S. pneumoniae lung infection, passive vaccination with IVIG did not reduce BALF CFU, even at early time points after low dose infection. These results suggest that alveolar macrophages did not mediate the protective effect, although this has not been formally confirmed by infections in mice depleted of alveolar macrophages. Depletion of neutrophils prevented the protective efficacy of IVIG within the lung, whereas systemic depletion of macrophages prevented its protective efficacy in the blood. Unexpectedly, depletion of neutrophils prevented septicaemia in the mouse model of pneumonia, preventing this model from being used to assess whether there is a role for neutrophils in IVIG-mediated protection against bacteraemia. Investigating this would require using neutrophil depletion in the systemic infection model, which we have not assessed. IVIG therapy has been used for immunomodulation, but in our model did not affect cellular recruitment to lavage fluid or TNFα responses. These results suggest that IVIG had no major effects on the inflammatory response to S. pneumoniae, although they do not exclude potentially beneficial effects on other aspects of the inflammatory response.
We have demonstrated that IgG in IVIG recognises a large number of S. pneumoniae protein antigens, several of which were identified using immunoblots and a Luminex assay and these include current protein vaccine candidate antigens [43]. There was a striking similarity between which protein targets were quantitatively dominant in binding IgG in IVIG from different geographical sources, with PspA, PhtD, PsaA and PpmA having the strongest antibody recognition in all IgG sources investigated. These similarities suggest that the immunodominance of certain protein antigens is largely independent of human genetic variation. Our protein target identification was biased towards existing well-described antigens, and further non-biased assessment is needed to identify all the antigens recognised by naturally acquired antibody. Several of the immunodominant surface proteins such as PspC and PspA are antigenically variable, and as only a single variant was represented on the Luminex assay it is unclear whether antibody recognition of these antigens is specific to certain alleles. Expression of PspC did increase IgG binding to live S. mitis, and for the D39 strain deletion of surface lipoproteins or both PspA and PspC both reduced IgG binding. These data suggest that PspC, PspA and lipoproteins may contribute towards IgG recognition of S. pneumoniae, but further investigation is necessary to identify which protein antigens are required for the protective IgG responses. This will be technically challenging as it is highly likely there is functional redundancy for IgG binding to S. pneumoniae surface proteins, and using mutants lacking specific protein antigens to identify functionally important targets for IgG in mouse infection models will be confounded by the importance for virulence of many of the potential protein antigens (e.g. PspA, PspC, Ply, PhtD). We also demonstrated IgG binding to S. mitis itself, which may be due to cross-recognition of S. mitis and S. pneumoniae surface antigens, or specific responses to S. mitis induced by natural oropharyngeal colonisation.
These data demonstrating that antibodies to S. pneumoniae capsular polysaccharide are not the major target of protective naturally acquired IgG have several important clinical implications. Firstly, measuring levels of anti-capsular antibody may not identify those patients at risk of IPD. Instead, measurement of antibodies to a range of protein targets or to whole S. pneumoniae by flow cytometry may be more relevant. Secondly, it may explain why individuals with specific-deficiencies in anti-polysaccharide antibody production, who are at increased risk of sino-pulmonary infection do not have the same high risk for invasive IPD as subjects with complete agammaglobulinaemia [44, 45]. Thirdly, the exponential rise in the incidence of S. pneumoniae infection with increasing age is thought to be related to immunosenescence. Antigen responses to a small number of protein antigens have been shown to be lower in the elderly [17], and we have also shown reduced responses to PspC in a small number of older subjects. These data suggest that one reason for the increased incidence of S. pneumoniae with age could be waning anti-protein antibody levels. Further investigation of the effects of age on anti-protein antigen responses and the functional consequences of any changes is needed to establish whether this hypothesis is correct. Fourthly, if there is reduction in S. pneumoniae colonisation in infants as a result of future vaccines with greater serotype coverage, this could potentially reduce anti-protein mediated natural immunity and perhaps lead to a paradoxical increase in adult disease, as has been postulated for the effects of Bordetella pertussis vaccination [46]. Finally, by identifying the mechanisms of naturally acquired immunity to S. pneumoniae, we can design vaccination strategies to improve these. For example, a multivalent protein vaccine using the dominant protein antigens should provide effective protection against IPD.
To conclude, we present multiple lines of supporting evidence that the protective benefits of human naturally acquired IgG against IPD is not, as previously thought, largely dependent on antibody to capsular polysaccharide antigen. Instead, natural human IgG-mediated protection against IPD seems to be dependent on IgG against protein antigens that are highly conserved between different geographical sources of IgG. These findings have important implications for identifying patients at risk of IPD, understanding relevant mechanisms of immunosenescence, and for novel vaccine development.
Wild-type S. pneumoniae serotype 4 strain TIGR4 and its unencapsulated mutant were kind gifts of J. Weiser (Univ. Pennsylvania). D39 and its unencapsulated mutant D39-DΔ were kind gifts of J. Paton (Univ. Adelaide). The ΔpspC, ΔpspA, ΔppmA, Δlgt, ΔphtD, ΔpiaA, and Δply mutant strains have been previously described [47–51]. Serotype 19F strain EF3030 was a kind gift of D. Briles (Univ. Alabama), and the serotype 6B stain ST6B, serotype 14F strain ST14, and serotype 23F strains were kind gifts from B. Spratt (Imperial College). The unencapsulated mutant strains of 0100993 and ST23F were made by replacing the cps locus (Sp_0346 to Sp_0360) with the Janus cassette [52]. The S. mitis strain expressing the S. pneumoniae TIGR4 serotype 4 capsule has been previously reported [53]. To construct the S. mitis pspC+ mutant strain, the TIGR4 pspC gene was amplified by PCR and integrated between S. mitis flanking DNA using PCR ligation before transformation into the S. mitis strain, similar to the mutagenesis strategy as described [22]. Bacteria were cultured overnight at 37°C in 5% CO2 on Columbia agar (Oxoid) supplemented with 5% horse blood (TCS Biosciences). Working stocks were made by transferring one colony of S. pneumoniae to Todd-Hewitt broth supplemented with 0.5% yeast extract (THY), grown to an OD of 0.4 (approximately 108 CFU/ml) and stored at -80°C in 10% glycerol as single use aliquots. CFU were confirmed by colony counting of log10 serial dilutions of bacteria cultured overnight on 5% Columbia blood agar. To partially digest surface proteins, bacteria were suspended in 500μl PBS with or without 100μg Pronase (Roche), incubated for 20min at 37°C shaking at 150rpm, followed by addition of 20μl of 25X Complete Mini-Protease Inhibitor (Roche). Bacteria were then washed twice in PBS and re-suspended in PBS+10% glycerol. Bacterial lysates were prepared as described previously [48]. When required, 20μl of lysate (1500 μg/ml) was treated with 10μl trypsin (2.5mg/ml, Gibco, Invitrogen) or PBS (control lysates) and incubated overnight, before the addition of 10μl 25X Complete Protease Inhibitor (Roche).
Intratect was a kind gift of Biotest Pharma GmbH, Dreieich, Germany. Vigam (Bioproducts Laboratories Ltd, Elstree, UK) was obtained commercially. Both contain 5% pooled human intravenous immunoglobulin. Dilutions of IVIG described for experimental data refer to dilutions of the 5% product rather than the resulting IgG concentration. Individual sera were collected from HIV-negative healthy adults in Malawi (age range 19 to 49 years, mean 29 years, 16 male and 4 female) who had not been immunised against S. pneumoniae. Serum from elderly subjects (age range 62 to 78 years, 6 males, 4 females) and young adult controls subjects (age range 24 to 33 years, 4 males, 6 females) was a kind gift from Dr Elizabeth Sapey, University of Birmingham. Specific antibody was depleted from IVIG by bacterial surface absorption with either unencapsulated TIGR4 or S. mitis expressing the serotype 4 capsule [53]. Bacteria were grown to OD580 0.4, washed and re-suspended to OD 1.0 using PBS, and 4mls were pelleted by centrifugation before re-suspension in 1.8mls of IVIG (Intratect). The suspension was incubated for 1hr at 37°C, shaking at 100rpm. The antigen-depleted IVIG was recovered by centrifugation and the process repeated. Mock absorbed IVIG was prepared by following the same process but without addition of bacteria. IVIG was pre-treated with papain to yield monovalent Fab fragments using the Pierce Fab Preparation Kit according to the manufacturer’s instructions and confirmed by immunoblot. Enriched (e)IVIG was prepared by affinity chromatography as previously described [54]. For the affinity resin, unencapsulated TIGR4 or D39 cultures were grown for 16h, pelleted and resuspended in 1 volume of coupling buffer (0.1 M Sodium bicarbonate, 0.5 M Sodium chloride; pH 8.3). Cells were pressure lysed at 200 MPa using a pressure cell homogeniser (Stansted) and the resulting lysates were 0.2 μm filtered and dialysed against 5L of coupling buffer for 4 h at RT. Lysates were concentrated using Vivaspin 20 centrifugal concentrators with a molecular weight cut of 10 kDa (GE healthcare) and coupled to cyanogen bromide activated agarose (Sigma-Aldrich) at a concentration of approximately 1 mg/ml according to the manufacturer’s instructions.
Whole cell, or specific antigen (individual proteins, capsular polysaccharide or cell wall polysaccharide) ELISAs were performed as previously [18, 55–57]. Recombinant PhtD was a kind gift of C. Durmort [58] and PsaA was a kind gift from J. Paton [59]. IgG binding to a panel of bacterial proteins and multiple capsular serotypes were assessed using Luminex assays [55] and electrochemiluminescence-based multiplex assay based on MesoScale Discovery (MSD, Rockville, MD, USA) technology as previously described [13, 55, 60]. For immunoblotting, bacterial lysates were separated by SDS-PAGE and transferred on to nitrocellulose membranes as previously described [36]. Membranes were probed with IVIG (Intratect) or pooled human sera (1:1000). To assess IgG binding to the bacterial surface, flow cytometry was performed as previously described [57, 61, 62].
Effects of IVIG on bacterial aggregation during growth were assessed by inoculating THY with 1x106 of S. pneumoniae and measuing the OD580 over an 8 h period in the presence of 10% IVIG (Intratect, 40mg/ml IgG) or PBS. Following 8 h growth, cultures were fixed onto polylysine slides (VWR), stained with rapid Romanowsky staining (Diff-Quick) and imaged under light microscopy (Olympus, BX40) at 100X using Q capture pro software. Bacterial aggregation was directly assessed by incubating bacteria diluted in PBS to 1X106 CFU/ml at 37°C in 5%CO2 for 1 hr with 0%, 1%, 5%, 10%, IVIG (Intratect 40mg/ml IgG). After fixation in 50μl 10% NBF, particle size was asessed by flow cytometry using a FACSCalibur with Cellquest and Flowjo software (BD Bioscience, UK) as a change in forward-scatter (FSC). Bacterial phagocytosis was measured as previously described as the association of FAM-SE labelled bacteria with either RAW 264.7 macrophages (MOI 10) [38, 53] or freshly isolated human neutrophils [57]. Briefly, RAW 264.7 murine cells were grown in RPMI supplemented with 10% heat-inactivated foetal calf serum. After washing, they were infected with FAM-SE labelled bacteria at an MOI of 10 which had been pre-incubated with IVIG or PBS for 30 mins at 37 C. After 45 min, cells were harvested with trypsin, fixed with paraformaldehyde (PFA) and fluorescence assessed using a FACS Calibur flow cytometer with Cellquest and Flowjo software (BD Bioscience, UK). For neutrophil phagocytosis, similarly opsonised labelled bacteria were incubated with freshly isolated human granulocytes for 30 min at MOI 20, after which they were fixed with PFA and assessed by flow cytometry. To assess bacterial killing by human neutrophils, pre-opsonised bacteria were incubated with freshly isolated granulocytes for 45 min after which they were serially diluted, plated and incubated overnight prior to colony counting.
For passive immunisation experiments with IVIG, 6 to 8 week old age-matched outbred CD1 mice (Charles River, UK) received two i.p. injections of IVIG totalling 12.8mg IgG or the equivalent volume of PBS 3 h prior and immediately before S. pneumoniae TIGR4. Challenges were given either i.n. with 50μl of PBS containing 1x107 CFU or i.v. with 100μl of PBS containing 5x105 CFU. To ensure aspiration of the IN inoculum, mice were anaesthetised using 4% halothane (Vet-Tech). At the designated time points after inoculation, mice were culled and BALF, lung homogenates, and blood obtained for plating to calculate bacterial CFU as described previously [19, 48]. BALF was collected by instilling the lungs with 1ml PBS via an incision in the trachea. This was recovered by aspiration repeated three times. Splenic macrophages were depleted by i.v. administration of 100ul of 5mg/ml liposomal clodronate (controls were given PBS liposomes) [38, 63]. Macrophage depletion was confirmed by a 50% reduction in F4/80+ splenocytes by flow cytometry using anti-F4/80-phycoerythrin (Caltag). To deplete Ly6G+ neutrophils, 600 μg anti-Ly6G monoclonal antibody (1A8m, Bioxcell) was administered by i.p. injection 24 hours prior to infection challenge depletion, as previously [24], resulting in a 94.8% decrease in neutrophils recruited to lavage fluid 24 hours after infection. Murine albumin was measured by ELISA using a commercially available kit following manufacturer’s instructions (Bethyl Laboratories). Murine TNF-alpha was measured by ELISA and BALF cell counts in cytospins as previously described [24]. Human IgG was measured in murine samples using a commercially available ELISA kit following manufacturer’s instructions (Cambridge Bioscience).
Data are presented as group means with error bars representing standard deviations (SDs). Student’s unpaired T-test was used to compare the mean of two groups or analysis of variance (ANOVA) for comparisons between multiple groups, using Bonferroni post-test comparisons. F tests were used to assess if the slope of linear regressions were statistically different to 0. Statistical tests were performed using Graph Pad Prism software, and P values < 0.05 were considered significant.
Experiments were approved by the UCL Biological Services Ethical Committee and the UK Home Office (Project Licence PPL70/6510). Experiments were performed according to UK national guidelines for animal use and care, under UK Home Office licence. Blood samples were taken from human volunteers in Malawi with approval of the University of Malawi College of Medicine Research and Ethics Committee and the Liverpool School of Tropical Medicine Research Ethics Committee (Ref: 00.54).
|
10.1371/journal.pntd.0000935 | From Re-Emergence to Hyperendemicity: The Natural History of the Dengue Epidemic in Brazil | Dengue virus (DENV) was reintroduced into Brazil in 1986 and by 1995 it had spread throughout the country. In 2007 the number of dengue hemorrhagic fever (DHF) cases more than doubled and a shift in the age distribution was reported. While previously the majority of DHF cases occurred among adults, in 2007 53% of cases occurred in children under 15 years old. The reasons for this shift have not been determined.
Age stratified cross-sectional seroepidemiologic survey conducted in Recife, Brazil in 2006. Serostatus was determined by ELISA based detection of Dengue IgG. We estimated time-constant and time-varying forces of infection of DENV between 1986 and 2006. We used discrete-time simulation to estimate the accumulation of monotypic and multitypic immunity over time in a population previously completely susceptible to DENV. We projected the age distribution of population immunity to dengue assuming similar hazards of infection in future years. The overall prevalence of DENV IgG was 0.80 (n = 1427). The time-constant force of infection for the period was estimated to be 0.052 (95% CI 0.041, 0.063), corresponding to 5.2% of susceptible individuals becoming infected each year by each serotype. Simulations show that as time since re-emergence of dengue goes by, multitypic immunity accumulates in adults while an increasing proportion of susceptible individuals and those with monotypic immunity are among young age groups. The median age of those monotypically immune can be expected to shift from 24 years, 10 years after introduction, to 13 years, 50 years after introduction. Of those monotypically immune, the proportion under 15 years old shifts from 27% to 58%. These results are consistent with the dengue notification records from the same region since 1995.
Assuming that persons who have been monotypically exposed are at highest risk for severe dengue, the shift towards younger patient ages observed in Brazil can be partially explained by the accumulation of multitypic immunity against DENV-1, 2, and 3 in older age groups, 22 years after the re-introduction of these viruses. Serotype specific seroepidemiologic studies are necessary to accurately estimate the serotype specific forces of infection.
| The spread of dengue virus is a major public health problem. Though the burden of dengue has historically been concentrated in Southeast Asian countries, Brazil has become the country that reports the largest number of cases in the world. While prior to 2007 the disease affected mostly adults, during the 2007 epidemic the number of dengue hemorrhagic fever cases more than doubled, and over 53% of cases were in children under 15 years of age. In this paper, we propose that the conditions for the shift were being set gradually since the re-introduction of dengue in 1986 and that they represent the transition from re-emergence to hyperendemicity. Using data from an age stratified seroprevalence study conducted in Recife, we estimated the force of infection (a measure of transmission intensity) between 1986–2006 and used these estimates to simulate the accumulation of immunity since the re-emergence. As the length of time that dengue has circulated increases, adults have a lower probability of remaining susceptible to primary or secondary infection and thus, cases become on average younger. If in fact the shift represents the transition from re-emergence to hyperendemicity, similar shifts are likely to be observed in the rest of Brazil, the American continent and other regions where transmission emerges.
| Dengue infection constitutes a major threat for urban populations of Latin America and Asia [1],[2]. Important differences in the clinical and epidemiological profile of dengue between the countries of Latin America and Southeast (SE) Asian have been observed. While in SE Asian countries dengue hemorrhagic fever (DHF) is common and morbidity and mortality has traditionally concentrated in children under 15 years of age, in American countries the disease affects mostly adult populations and manifests primarily as dengue fever (DF) [3], [4].
Several hypotheses have been proposed to explain these differences. It has been shown that children from Central America, Venezuela, and Colombia may not develop vascular permeability as readily as children from SE Asia after secondary dengue infection,[5]–[6] and that there may be a high prevalence of dengue resistance genes among black populations of Brazil and the Caribbean [7], [8]. An additional explanation for the low numbers of DHF in American countries may be underreporting of cases that do occur, due to technical difficulties or a limited capacity to perform diagnosis that meet the criteria of the WHO case definition [3]. None of these explanations are fully satisfactory in explaining the differences between the two regions.
Dengue was reintroduced in Brazil in 1986, after an absence of at least 20 years (except for an epidemic in Roraima in 1981 and sporadic cases). Since then, Brazil has become the country that reports the largest number of cases to the WHO, accounting for over 70% of cases reported in the Americas [9], [10]. Three serotypes currently circulate throughout the country; DENV 1 was reintroduced to Rio de Janeiro in 1986, DENV 2 in 1990, and DENV 3 in 2002, and from Rio they spread to the rest of the country[11]. While prior to 2007 the majority of DHF cases in Brazil occurred among adults aged 20–40 years of age, in 2007 the annual number of DHF cases more than doubled over previous years and a shift in the age distribution was reported [12]. In 2007, 53% of cases occurred in children under 15 years old. The shift was most noticeable in the Northeast region, where children accounted for 65% of the total number of DHF cases, while other regions such as the Central-West and North did not experience a significant shift and most of the DHF cases continued to occur among adults [12].
Although the cause of this shift is likely to be multifactorial, we propose that the conditions for it were being set gradually since the re-emergence of DENV in 1986 and that the current epidemiological profile represents the transition from re-emergence to hyperendemicity. In a setting where transmission is constant, people who are exposed for a longer time have a greater cumulative probability of infection. In Brazil, circulation of DENV virus for over 20 years has resulted in the accumulation of immunity in older individuals, driving the average age of primary and secondary infection towards younger age groups.
Using data from a serological study performed in Recife, in Northeast Brazil, we estimate the force of infection and basic reproductive number of dengue in three areas of distinct socio-economic status for the period 1986–2006, in order to better understand transmission intensity over this period. We then use these estimates to simulate the accumulation of monotypic and multitypic immunity in a population previously susceptible to dengue virus, and to predict the expected age distribution of DHF cases in the future.
This study was reviewed and approved by the ethics committee of the CPqAM-Fiocruz/Brazilian Ministry of Health (No. 49/04). Written consent to participate in the study was obtained from each person (or their guardian) after a full explanation of the study was provided. All personal identifiers were removed prior to secondary data analysis at Johns Hopkins University.
This study was based on a serological sample of households in Recife, Pernambuco, Brazil, conducted between August and September 2006. The first dengue outbreak in the state of Pernambuco occurred in 1987 (DENV 1). No additional autochthonous cases were reported until 1995, when DENV 2 was introduced causing a new epidemic. Since 1995 cases have been reported every year. DENV 3 was first isolated in Pernambuco in 2002 [13].
The study population and methods have been described in detail by Braga et al. [14] Briefly, Recife has 1.5 million inhabitants. The climate is humid, with an average temperature of 25°C and rainfall of approximately 2000 mm per year. Three neighborhoods were selected to represent low, medium and high socio-economic areas. A systematic age stratified sample was obtained, using the Census 2000 data that provides the total population size, number of households and age distribution in the three areas [15]. Residents aged between 5 and 64 years were eligible for the survey. Serum samples were screened for IgG antibodies against DENV with an enzyme-linked immunoassay commercial kit (Dengue IgG-ELISA, PanBio, Ltd., Brisbane, Australia). Tests were performed in duplicate according to the manufacturer's instructions. This test does not determine the presence of immunity to specific dengue serotypes, but the presence of immunity to any dengue serotype.
The force of infection (λ) is a measure used to characterize the intensity of transmission in a given setting and estimates the per capita rate of acquisition of infection by susceptible individuals. Age stratified serological surveys can provide information about the force of infection over a period of time, λ(t), as described elsewhere [16]. Assuming that the risk of infection does not vary with age, the difference in seroprevalence between subjects a and a+1 years of age can be attributed to the transmission intensity between a and a+1 years ago. To estimate λ(t), for the period 1986–2006, we used a model based upon one described by Ferguson et al.[17] We estimated constant and time-varying forces of infection. Detailed information regarding the methods used can be found in Text S1 in Supporting Information S1.
R0 is the number of secondary infections generated by a primary case in a completely susceptible population. R0 gives insight into the level of control that is required to reduce incidence and eventually block transmission. Detailed information regarding the methods used to estimate R0 can be found in Text S1 in Supporting Information S1.
To estimate the accumulation of monotypic and multitypic immunity in a population previously susceptible to dengue, we performed a discrete-time simulation by applying the forces of infection estimated from the seroprevalence data onto a simulated immunologically naive population structured by-age like the one of Recife. We used independent data on the years in which the different serotypes were introduced into Brazil/Pernambuco to apply the estimated hazards only in those years when particular serotypes were known to have circulated [11], [13]. The age profile of the population was obtained from the 2000 census data. We conducted simulations until age distributions of immunity reached equilibrium and used both constant and time-varying hazards.
Since we did not have seroprevalence data to estimate the force of infection beyond 2006, we assumed that λ(t) after 2006 has been constant and equal to the average hazard over the period 1986–2006.
All statistical analyses were performed using R statistical package (version 2.10.1).
The dataset contained data on 1427 subjects aged 5 to 20 years, 593 (41.6%) from area 1, 480 (33.6%) from area 2 and 342 (24.0%) from area 3. Figure 1 shows the age-specific seroprevalences of each of the areas (black dots). Area 1, the neighborhood of low socioeconomic status showed a significantly higher seroprevalence when compared to Area 3, the high socioeconomic stratum neighborhood (0.85 (95%CI 0.82–0.88) vs. 0.70 (95%CI 0.65–0.75), p<0.0001). The middle class neighborhood (Area 2) also showed a significantly higher seroprevalence when compared to Area 3 (0.82 vs. 0.70, p = 0.0002). Based on the seroprevalence and that approximately 87000 cases were notified in Recife during this period, it is clear that less then 10% of the infections were reported [13].
The estimated average time-constant force of infection for the period 1986–2006 was 0.052 (95% CI 0.041–0.063). On average, each serotype infected 5.2% of susceptible individuals each year.
Time constant s for the three areas were 0.068 (95%CI 0.045, 0.091), 0.056 (95%CI 0.035, 0.077) and 0.035 (95%CI 0.019, 0.051). Though the difference between these forces of infection is not statistically significant, a trend is seen towards higher hazards of infection in settings of lower socioeconomic status.
As can be expected, the fit of the model improved significantly when we allowed for time-varying forces of infection (likelihood ratio test, p = 0.006). According to this model (Figure 2), the average yearly force of infection ranged between 0 and 0.057 between 1986 and 1998, and then peaked at 0.26 in 1999. As has been reported elsewhere the correlation between incidence and estimated force of infection is poor (r = 0.21) [18].
Given that it has been reported that between 1987 and 1995 there were no autochthonous dengue cases in the state of Pernambuco, we also fit a model constraining the force of infection for these years to be 0 [13]. The fit of this 8-parameter model was not significantly different from the fit of the model that did not constrain these hazards to be zero (LR test, p = 0.99) or from the saturated model (LR test, p = 0.34).
Figure 1 shows the fit of 1) constant (red lines) and 2) time-varying models (green lines) to the age-specific seroprevalence data in the three areas and overall areas.
The correlation between the annual hazards estimated in areas 1 and 3 (r = 0.79) is high, while the correlation between 1 and 2 and between 2 and 3 is poor (r = 0.06 and 0.18, respectively).
Using the time constant and time varying λs we estimated an overall R0 of dengue in Recife of 2.7 (95%CI 2.45, 3.11). For the three areas the R0 estimates were 3.3 (95%CI 2.45, 4.18), 2.8 (95%CI 2.09, 3.64) and 2.1 (95%CI 1.56, 2.66), respectively.
Figure 3 shows the age distribution of susceptible, monotypically immune and multitypically immune at different time-points after the introduction of DENV 1, 2 and 3 into a previously susceptible population, assuming a constant risk of infection of 0.052/year/serotype. As the number of years of DENV circulation increases, multitypic immunity accumulates among adults, and susceptibles and monotypically immune become increasingly concentrated in younger age groups.
Assuming that cases of DHF occur primarily among people who experience secondary infection, the age distribution of people who are at risk of secondary infection (i.e. of people who have been exposed to a single dengue serotype) should approximate the age distribution of DHF cases [19]. Hence, our results suggest that as years after re-emergence go by, the mean, median and modal ages of cases will decrease. For = 0.052, the model estimates that while 10 years after re-emergence the median, mean and modal age of cases (monotypically immune) would be 24, 29.0 and 14 years respectively, these numbers would decrease to 13, 15.2 and 11 years 50 years after re-emergence. Similarly, while it is expected that only up to 27% of DHF cases would occur in children under 15 years of age 10 years after the re-emergence, 50 years after re-emergence this proportion would increase to 58%. Figure 4 shows the age distribution of hospitalized dengue cases in Pernambuco in 2007, based on official notification records, and the estimated distribution according to our model (20 years after re-emergence) [20].
The strength and the speed of the shift in the age distribution of immunity depend on the underlying force of infection (Figure 5 and Table S1 in Supporting Information S1). The estimated median and modal ages of monotypically immune for = 0.03 are 19 and 15 years respectively, 50 years after re-emergence, while these ages drop to 11 and 6 years for = 0.07.
Results were similar if time-varying, instead of constant forces of infection were applied, or if for each serotype was weighted taking into account the serotype predominance reported for the different years in the state of Pernambuco [13].
A dramatic increase in the number of DHF cases and a shift in age group predominance of DHF were observed during the 2007 dengue epidemic in Brazil, the first re-emergence of the DENV-2 serotype predominance since 1990. Our results suggest that this shift can be partly explained by the accumulation of multitypic immunity in the adult population over time after the re-emergence of DENV-1 in 1986, DENV-2 in 1990 and DENV-3 in 2002. As the length of time of co-circulation of multiple serotypes of dengue in Brazil increases, adults have a lower probability of remaining susceptible to infection. As a result, cases become on average younger as completely susceptible individuals and monotypically immune individuals are more likely to be from younger age groups.
If the accumulation of multitypic immunity in adult population is in part responsible for the observed shift in age group predominance of severe dengue cases, we would expect similar shifts to have occurred in central and northern South America, where several DENV serotypes have been known to circulate since the 1970's. In Mexico, Venezuela, Nicaragua and Colombia most of the severe cases occur among children <15 years of age and a similar trend is being observed in Honduras.[21], [22],[23]. In contrast, such a trend has not been observed in countries where multiple serotypes only started circulating in the 90's. If the central/west regions of Brazil continue to experience high DENV forces of infection and multiple circulating serotypes we expect a similar shift in age group predominance to occur in the coming years.
Since DHF is more likely to occur in children, a decrease in the mean age of secondary infection might also be expected to lead to an increase in the proportion of dengue infections that lead to severe symptoms or DHF cases [24], [25]. In the 1990's, after DENV-2 was introduced, 0.06% of reported dengue cases in Brazil resulted in DHF/DSS. This percentage increased to 0.21% in 2007 [9]. This observed increase in DHF may also have been a result of changes in virulence of particular dengue viruses that were circulating or due to the fact that the overall force of infection has increased as has been proposed.
According to our model, the speed of the shift is proportional to the magnitude of the average force of infection. Higher average forces of infection lead to a more rapid shift of the age distributions of immunity and to a younger median and modal age of monotypically immune. Thus, the shift can be expected to be slower in regions that have been exposed to weaker forces of infection or where the re-emergence of multiple serotypes was delayed. This may explain why the shift has only been observed in major cities and certain regions of Brazil. The Northeast region of Brazil, where Recife is located, has the highest proportion of children among DHF cases, and it has also been traditionally the region with the highest incidence rates of dengue fever since 1986 [9], [12]. The Central-West region, where the shift is not yet apparent has shown high incidence rates of dengue fever only during the last 7 years [9].
Our estimate of the average λ and R0 in Recife is lower than those estimated for Thailand for the period 1980–2005 (λ = 0.1, R0 = 5.2) [26]. Our model predicts that average forces of infection of 0.1 would be associated with a mean age of severe or DHF cases of 8 years, and this is consistent with what has been traditionally observed in SE Asian countries. Both Thailand and Singapore have experienced significant decreases in transmission intensity over the last few years that have been accompanied by an increase in the average age of cases [26],[27],[28]. If the force of infection in Recife continues to be as high as it has been over the last 20 years, or higher, it is likely that within the next decade the age distribution of DHF in Recife (and other American regions with high forces of infection) will resemble the age distribution observed in SE Asia, with most cases concentrated in the adolescent population. However, our projections are meant to be qualitative rather than quantitative. The actual seroprevalences observed in the future in Recife may differ from our projections depending on secular trends in the transmission intensity of dengue and population demographics.
There are several limitations to this study. Even though our results present an explanation for why DHF may have shifted towards children over the years since introduction, the mechanism that we propose is gradual and does not explain the sudden change observed in 2007–2008. The recirculation of DENV-2 into certain cities in 2007, after almost 7 years of DENV 3 predominance and the resultant increase in secondary cases may have determined the observation of an age shift in 2007 and not before, even though it had been gradually taking place [9]. As reported by the Ministry of Health, during 1998–2006 the percentage of severe dengue cases in children increased from 9.5% (in 1998) to 22.6% (in 2001). Although our results suggest that the major driver of the shift is the accumulation of immunity in older age groups, fluctuations in serotype specific transmission intensity, serotype predominance, characteristics of the virus or serotype predominance may have also played a role in determining the visibility of the shift.
Our model predicts that after 20 years of exposure to a constant force of infection of 0.05 per year, children 15 years old or younger should only account for 31% of DHF cases while the data shows that in 2007, 70% of cases in Recife occurred among children of this age group. This discrepancy may arise due to the fact that the model does not take into account age-dependence of infection or clinical presentation. If children are more likely to develop severe disease, then the observed distribution of cases is likely to be skewed towards lower age groups.
The fact that the available serological study does not contain serotype specific information limits our ability to estimate serotype specific forces of infection, interactions (enhancement/inhibition) and basic reproductive numbers. Similarly, the cross-sectional nature of this dataset does not allow us to control for potential confounding by age dependent transmission intensity. Longitudinal data and data from seroprevalence studies using serotype specific methods such as the PRNT are essential in order to properly reconstruct the transmission intensity over the last 20 years.
This analysis has important public health implications on planning public health responses to dengue for the next decade. Dengue is the most rapidly spreading vector borne viral disease. If the age shift in fact represents the transition from re-emergence to hyperendemicity, similar shifts in age are likely to be observed in the rest of Brazil, the American continent and other regions where dengue has emerged more recently.
|
10.1371/journal.ppat.1002369 | Metagenomic Analysis of Fever, Thrombocytopenia and Leukopenia Syndrome (FTLS) in Henan Province, China: Discovery of a New Bunyavirus | Since 2007, many cases of fever, thrombocytopenia and leukopenia syndrome (FTLS) have emerged in Henan Province, China. Patient reports of tick bites suggested that infection could contribute to FTLS. Many tick-transmitted microbial pathogens were tested for by PCR/RT-PCR and/or indirect immunofluorescence assay (IFA). However, only 8% (24/285) of samples collected from 2007 to 2010 tested positive for human granulocytic anaplasmosis (HGA), suggesting that other pathogens could be involved. Here, we used an unbiased metagenomic approach to screen and survey for microbes possibly associated with FTLS. BLASTx analysis of deduced protein sequences revealed that a novel bunyavirus (36% identity to Tehran virus, accession: HQ412604) was present only in sera from FTLS patients. A phylogenetic analysis further showed that, although closely related to Uukuniemi virus of the Phlebovirus genus, this virus was distinct. The candidate virus was examined for association with FTLS among samples collected from Henan province during 2007–2010. RT-PCR, viral cultures, and a seroepidemiologic survey were undertaken. RT-PCR results showed that 223 of 285 (78.24%) acute-phase serum samples contained viral RNA. Of 95 patients for whom paired acute and convalescent sera were available, 73 had serologic evidence of infection, with 52 seroconversions and 21 exhibiting a 4-fold increase in antibody titer to the virus. The new virus was isolated from patient acute-phase serum samples and named Henan Fever Virus (HNF virus). Whole-genome sequencing confirmed that the virus was a novel bunyavirus with genetic similarity to known bunyaviruses, and was most closely related to the Uukuniemi virus (34%, 24%, and 29% of maximum identity, respectively, for segment L, M, S at maximum query coverage). After the release of the GenBank sequences of SFTSV, we found that they were nearly identical (>99% identity). These results show that the novel bunyavirus (HNF virus) is strongly correlated with FTLS.
| Initially in 2007, and again between 2008 and 2010, cases of a life-threatening disease with sudden fever, thrombocytopenia, and leukopenia were reported in Henan Province, China. Patient reports of tick bites suggested that this disease could be infectious or tick-transmitted. Many patients were provisionally diagnosed with human granulocytic anaplasmosis (HGA). However, only 24 of 285 (8%) had objective evidence of HGA, suggesting that other pathogens likely contributed to fever, thrombocytopenia and leukopenia syndrome (FTLS). Illumina sequencing was used for direct detection in clinical samples of pathogens possibly associated with FTLS. A novel bunyavirus was found only in samples from FTLS patients. Further epidemiologic and laboratory investigation confirmed that the novel bunyavirus was associated with FTLS. The results illustrate that metagenomic analysis is a powerful method for the discovery of novel pathogenic agents. Combined with epidemiologic investigation, it could assist in rapid diagnosis of unknown diseases and distinguish them from other diseases with similar symptoms caused by known pathogens.
| In May 2007, a county hospital in Xinyang City, Henan Province treated three patients with fever, abdominal pain, bloating, nausea, vomiting, gastrointestinal bleeding, and elevated aminotransferases. The local hospital diagnosed the disease as acute gastroenteritis. A family member of one patient reported the disease to the Henan Center for Disease Control and Prevention (CDC), which sent a team to investigate. The investigation revealed that the disease had the following characteristics: (1) acute onset with fever; (2) low white blood cell and platelet counts; (3) high levels of alanine and aspartate transaminases; (4) positive urine protein. On the basis of these features, the Henan CDC excluded the possibility of gastrointestinal disorders. In order to identify the disease etiology, the Henan CDC team used the above clinical characteristics as the case definition to search for similar cases in local hospitals in this and neighboring counties, while establishing a disease surveillance system that required all medical institutions to report cases that met the above case definition. Altogether, 79 cases were found in 2007 in Henan, with 10 fatalities (case fatality rate, 12.7%). All patients were farmers and resided in mountainous or hilly villages, and many had reported tick bites 7–9 days before illness, further suggesting an infectious etiology. In recent years, patients with similar clinical symptoms were reported with human granulocytic anaplasmosis (HGA; Anaplasma phagocytophilum) in neighboring Anhui province [1]. In 2005, there was an epidemic of Tsutsugamushi (scrub typhus/Orientia tsutsugamushi) in this area [2]. Clinical investigations, epidemiological analyses, and laboratory testing prompted consideration of rickettsial diseases as possible causes, including HGA, human monocytic ehrlichiosis (HME; Ehrlichia chaffeensis), and Tsutsugamushi disease. Specific methods such as polymerase chain reaction (PCR) and immunofluorescence assays (IFAs) for these pathogens were then used to determine if these cases were attributable to HGA or HME [3], [4]. However, only 18 of 79 (22.7%) patients were positive for A. phagocytophilum based on serology and DNA testing. Thus, the disease was initially considered at least partly caused by A. phagocytophilum, and cases were provisionally diagnosed as suspected HGA based on clinical and epidemiological data [1]–[3], [5]–[7]. In the 3 years since 2007, 206 suspected cases have been discovered in Henan, but there was only a very low positive rate of A. phagocytophilum confirmation (6 of 206 patients) and no pathogen was isolated. Similar cases were also reported in the mountainous and hilly areas of nearby Shandong, Jiangsu, Hubei and Anhui provinces, indicating that the disease already existed for some time and was widely distributed [7]. We decided to address the possible causative pathogen underlying this infection.
On the basis of epidemiological and clinical characteristics, we considered two types of diseases to be possible: rickettsial and arthropod-borne viral disease. Because of the low rates of A. phagocytophilum (rickettsial disease) detection, the research team intensified its virus search to take into account arthropod-borne viruses, including Flaviviridae (Dengue viruses [DENV], Japanese encephalitis virus [JEV]), Togaviridae (Chikungunya virus, Eastern equine encephalitis virus [EEEV], Western equine encephalitis virus), and Bunyaviridae (Crimean-Congo hemorrhagic fever virus, Hantaan virus, and Rift valley fever virus) [8]. Specific PCR assays for these viruses were used [9]–[16]. However, none of the patients from 2007 to 2010 was positive for these viruses, suggesting a new infectious agent, possibly a virus, remained to be discovered. Thus, the syndrome was considered an emerging infectious disease and was named fever, thrombocytopenia and leukopenia syndrome (FTLS). To identify the etiology, the research team adopted the following strategy: 1) sequencing of randomly amplified cDNA/DNA from FTLS patient samples using high-throughput Illumina sequencing to specifically explore viral communities present in patients suffering from FTLS, 2) PCR detection of target DNA directly from clinical specimens, 3) viral culture, 4) immunodetection methods, and 5) electron microscopic study of the morphology of the cultured virus.
Culture followed by serological and molecular tests is a standard approach for identifying an unknown virus. However, culture of an unknown virus is time-consuming, even taking several years to confirm a novel infection like HIV [17]. Otherwise, virus culture often fails because of the lack of cell lines capable of supporting propagation of viruses (e.g., hepatitis B and C virus). Methods for cloning nucleic acids of microbial pathogens directly from clinical samples offer opportunities for pathogen discovery, thereby laying the foundation for future studies aimed at assessing whether novel or unexpected viruses play a role in disease etiology. Random PCR and subtractive cloning sequencing have identified previously unknown pathogens as etiological agents of several acute and chronic infectious diseases [18], [19]. Recently, high-throughput sequencing approaches have been used for pathogen detection and discovery in clinical samples [20]–[22]. We also developed a method for exploring viruses, both known and novel, using high-throughput Illumina sequencing. In this study, high-throughput Illumina sequencing was applied to specifically explore the viral communities in patients with FTLS, using healthy subjects as controls. Here, we provide evidence for the discovery of a novel bunyavirus associated with FTLS through high-throughput sequencing. Subsequent culture of the virus and PCR detection of the specific virus in patient specimens confirmed these findings.
This research was approved by the Review Board of the Center for Disease Control and Prevention of Henan Province, the Review Board of the Center for Disease Control and Prevention of Xinyang city, the Review Board of Beijing Institute of Genomics, the Review Board of Beijing Genomics Institute in Shenzhen, and the Review Board of the Institute of Microbiology and Epidemiology. All participants gave written informed consent for use of their samples in research.
Given the serious nature of FTLS, it was decided to handle all clinical specimens and perform all experiments involving live virus in a biosafety level-3 (BSL-3) facility.
We studied 285 Henan province patients with FTLS whose samples were submitted to the Henan Province CDC between May 2007 and July 2010. Acute-phase serum samples from all patients were collected. Paired convalescent sera were available from 95 patients.
Sera were tested by reverse transcription (RT)-PCR, PCR, and/or indirect IFA serological assays for a number of microbial agents, including A. phagocytophilum, Ehrlichia chaffeensis, Dengue fever virus, Japanese encephalitis virus, Chikungunya virus, Eastern equine encephalitis virus, Western equine encephalitis virus, Crimean-Congo hemorrhagic fever virus, Rift Valley fever virus, Sandfly fever Naples Sabin virus, and Hantavirus [3]–[6], [9]–[16], [23]. Antigen slides for diagnosis of HGA (A. phagocytophilum) were purchased from Focus Diagnostics (IF1450G, CA, USA). Antigen slides for diagnosis of other pathogens were prepared by our laboratories. Fluorescein isothiocyanate (FITC)-conjugated goat anti-human IgG (Fc) was purchased from Sihuan Sci-Technics Company (Beijing, China).
Equal quantities (100 µL) of acute-phase sera from 10 FTLS patients who had a history of tick bite were pooled and centrifuged at 1000 x g for 10 minutes. The supernatant was collected for DNA and RNA extraction. The same was done for 10 sera from healthy subjects (control).
DNA was extracted from 140 µL of each sample using the QIAamp DNA mini Kit (Qiagen, 51304) according to the manufacturer's instructions. DNA was eluted from the columns with 50 µL water containing 20 µg/mL RNaseA. After incubation at 37°C for 15 minutes to eliminate RNA, DNA was used immediately or stored at −80°C.
Total RNA was extracted from 140 µL of each sample using the QIAamp viral RNA mini Kit (Qiagen, 52904) according to the manufacturer's instructions. RNA was eluted from the columns with 50 µL of diethyl pyrocarbonate (DEPC)-treated water containing 1 U DNaseI. Samples were incubated at 37°C for 15 minutes to eliminate human DNA, followed by DNase inactivation at 95°C for 10 minutes. RNA was used immediately or stored at −80°C.
cDNA was synthesized from 6 µL of RNA by reverse transcription at 45°C for 50 minutes in a 20-µL solution containing 50 mM Tris–HCl (pH 8.3), 75 mM KCl, 3 mM MgCl2, 10 mM DTT, 100 ng of random hexamer primers, 200 U of Superscript II (Invitrogen, 18064–014), 25 U of RNasin (Promega, N2511), and 0.5 mM dNTPs.
Random hexamer PCR was carried out in a 25-µL mixture containing 4 µL of cDNA or DNA, 10 mM Tris-HCl (pH 8.4), 50 mM KCl, 2.5 mM MgCl2, 100 µM dNTPs, 1 U Taq DNA Polymerase (Promega, M1661) and 100 ng of random hexamer primers containing a linker (5'-GCCGGAGCTCTGCAGAATTCNNNNNN-3'). After denaturing at 95°C for 5 minutes, targets were amplified by 45 cycles of 95°C for 30 seconds, 40°C for 30 seconds, 50°C for 30 seconds, and 72°C for 90 seconds. The amplified products were detected by agarose gel electrophoresis. Pure water was used as a negative control.
FTLS patient and control genomic DNA/cDNA libraries were constructed according to the manufacturer's instructions (Illumina). In brief, RT-PCR and PCR products were roughly quantified by UV absorption and equal amounts of each sample were mixed. Nucleic acids within the mixtures were sheared by sonication, and fragments in the 150–180-bp range were collected by cutting bands from an agarose gel after electrophoresis. The sheared DNA and cDNA ends were repaired using Klenow DNA polymerase, after which 5' termini were phosphorylated and 3' termini were polyadenylated. The adaptors were added, PCR enrichment was performed, and 150–180-bp fragments were collected for sequencing by the Illumina method.
The sequencing procedure was performed according to the manufacturer's instructions (Illumina). In this process, template library DNA was hybridized to the surface of the flow cells and multiple copies of DNA were made to form clusters using the Illumina cluster station. Workflow steps included template hybridization, isothermal amplification, linearization, and final denaturation and hybridization of sequencing primers. Paired-end sequencing (100 cycles) was performed using a four-color DNA Sequencing-By-Synthesis (SBS) technology following the manufacturer's instructions.
Short Oligonucleotide Analysis Package (SOAP) was used to handle the large amounts of short reads generated by parallel sequencing [24]. Briefly, after filtering out highly repetitive sequences and adaptor sequences, the overlapping datasets between FTLS and healthy subjects were analyzed by subtracting fragments that mapped to both host genomic-plus-transcript and bacteria databases. The non-redundant reads were mapped onto a virus database downloaded from NCBI (ftp://ftp.ncbi.nih.gov/genbank/). The resulting alignments were filtered to identify unique sequences by examining alignment (identity ≥80%) and E-value scores (e≤10−2). Filtered unique alignments were examined in the taxonomy database (NCBI) using a custom software application written in Perl (BioPerl version 5.8.5). Unmapped reads were examined in GenBank nucleic acid and protein databases using BLASTn and BLASTx, respectively [25]–[27]. Unique alignments were examined in the taxonomy database (NCBI). Sequences without hits were placed in the ‘‘unassigned’’ category. Sequences were phylotyped as human, bacterial, phage, viral, or other based on the identity of the best BLAST hit. Considering misannotation and low-complexity for Illumina short reads, sequences assigned to the same virus family were further assembled into contigs with Velvet 1.1.04 (K-mer length = 21; coverage cutoff: default 0; Insert length: PE only; minor contig length: 42) for direct comparison with GenBank nucleic acid databases using BLASTn [28]. Contigs were also blasted with GenBank protein databases using BLASTx [28]. An E-value cutoff of 1×10−5 was applied to both BLASTn and BLASTx analyses. Sequences phylotyped as viral were placed in the “viral” category.
Our mass sequencing data revealed that some sequences showed possible infection with human bocavirus, which belongs to the Parvoviridae family. PCR performed to detect human bocavirus tentatively identified in sera from FTLS patient samples amplified a 291-bp fragment of the NS1 gene, as described previously [29]. Amplified products were detected by agarose gel electrophoresis and sequenced using an ABI 3730 DNA Sequencer.
Our mass sequencing data revealed a 168-bp sequence (C361, Accession: HQ412604) indicating the possible presence of a novel virus with closest identity (i.e., lowest E-value) to Tehran virus which belongs to the Phlebovirus genus of the Bunyaviridae family. We developed a PCR strategy based on the 168-bp sequence identified in sera from FTLS patient samples to detect the novel bunyavirus using forward (PF: 5'-GAC ACG CTC CTC AAG GCT CT-3') and reverse (PR: 5'-GCC CAG TAG CCC TGA GTT TC-3') primers designed with Primer3 (Supplementary Figure S1). PCR was carried out in a 25-µL mixture containing 4 µL of cDNA, 10 mM Tris–HCl (pH 8.4), 50 mM KCl, 2.5 mM MgCl2, 100 µM dNTPs, 1 U Taq Pol (Promega, M1661), 0.25 µM forward primer, and 0.25 µM reverse primer. Thermocycling conditions were as follows: 95°C for 4 minutes (denaturation), followed by 35 cycles of 94°C for 30 seconds, 54°C for 30 seconds, and 72°C for 30 seconds. Amplified products were detected by agarose gel electrophoresis and sequenced using an ABI 3730 DNA Sequencer.
The Vero E6 cell line (African green monkey kidney cell) was selected for isolation of the novel bunyavirus associated with FTLS because it supports the growth of many bunyaviruses [30], [31]. Vero E6 cell lines were inoculated with six serum samples that contained novel bunyavirus RNA. Each sample underwent at least three cell culture passages in Vero E6 cell line before being considered negative. Medium was replenished on day 7, and cultures were terminated 14 days after inoculation. All cultures were observed daily for cytopathic effect (CPE). Virus-infected cells and uninfected cells were also examined for the novel bunyavirus by RT-PCR at each passage. Vero cell cultures with obvious CPE and containing novel bunyavirus RNA were further analyzed by morphology, genome sequencing, and serology.
Cells showing CPE and containing novel bunyavirus RNA were collected for thin-section electron microscopy. After discarding the culture supernatant, virus-infected cells (50 mL) were mixed 1∶1 with 4% glutaraldehyde (paraformaldehyde), placed onto Formvar-carbon-coated grids, and stained with 1% methylamine tungstate. Specimens for thin-section electron microscopy were prepared by dehydrating washed cell pellets with serial dilutions of acetone and embedding in epoxy resin. Ultrathin sections were cut on an Ultracut LKBV ultramicrotome, stained with uranyl acetate and lead citrate, and examined under a transmission electron microscope (JEM-1400).
The medium from 20 mL of novel bunyavirus-infected Vero E6 cells was centrifuged at 1,000 x g for 10 minutes and then at 4,000 x g for 10 minutes, after which the supernatant was collected. PEG8000 was added to the supernatant at a final concentration of 10% (w/v) followed by centrifugation at 20,000 x g for 2 hours. The pellet was resuspended in 2 mL 1× phosphate-buffered saline (PBS) for RNA extraction. Random RT-PCR was performed, and the products (500–1500 bps) were collected and ligated into the pGEM-T vector (Promega, A3600) by incubating overnight at 16°C. Escherichia coli JM109 were transformed with the ligation mixture and cultured on LB agar containing X-gal. White clones were sequenced using an ABI 3730 DNA Sequencer. Contaminating human and extraneous sequences were eliminated using CrossMatch, and the complete sequence was assembled using Phred-Phrap-Consed [32]. Bridge RT-PCR was employed for gap-closure.
Phylogenetic analyses were performed using the neighbor-joining method in the MEGA software package, version 4.0.2 [33]. Available nucleotide or protein sequences from known viruses were obtained from GenBank for inclusion in the phylogenetic trees. Selected sequences from GenBank included those with the greatest similarity to the sequence read in question based on BLAST alignments as well as representative sequences from all major taxa within the relevant Bunyaviridae family. To further establish the relationships between the new virus and the members of the Phleboviruses genus, we included all sequences for phleboviruses available in GenBank. Branching orders of the phylograms were verified statistically by resampling the data 1,000 times in a bootstrap analysis with the branch-and-bound algorithm, as applied in MEGA.
After successful isolation of the novel bunyavirus, we developed an indirect IFA to detect specific antibodies in patient serum specimens, as previously described [4]. In brief, monolayers of virus-infected Vero E6 cells showing CPE and containing novel bunyavirus RNA were harvested, and one volume of infected cells was mixed with 0.5 volumes of non-infected cells. The mixture was centrifuged at 1,000 x g for 10 minutes, after which cells were resuspended in 1× PBS, spotted onto 12-well glass slides, and fixed with acetone for 10 minutes. Sera from patients with FTLS (including 285 acute-phase samples and 95 paired sera), patients with respiratory diseases (80 serum samples), and healthy subjects (50 serum samples) were applied to the cells. Samples (diluted 1∶20 in PBS) were screened by first spotting 50 µL of each serum sample per well and incubating for 30 minutes at 37°C. After washing for 10 minutes in PBS, 20 µL of FITC-conjugated goat anti-human IgG (Sihuan Sci-Technics Company, Beijing, China) diluted 1∶40 in buffer containing Evans blue was added to each well and incubated for 30 minutes. After washing, slides were mounted in glycerin and examined by immunofluorescence microscopy. A titer of 1∶20 was considered positive.
All 285 patients with FTLS were from the Henan Province of China and were provisionally diagnosed as suspected HGA on the basis of similar clinical manifestations [5], [7]. They represented four different epidemiologically linked sporadic cases and a few clusters of cases including 79 patients in 2007, seven patients in 2008, 47 patients in 2009, and 152 patients in 2010. The patients presented mainly between April and October, peaking in April-May during the tea-picking season in Henan. All patients resided in mountainous and hilly rural areas. In our study, 238 of 285 patients tested positive for novel bunyavirus infection by RT-PCR and/or IFA.
The median age of patients was 57.2 years (range, 23–88) and the male-to-female ratio was 1 to 2.27; 219 patients (92.02%) were farmers and 19 (7.98%) were workers or students. Among patients, 52 (21.85%) reported a tick bite within 2 weeks (5–14 days) before the onset of clinical manifestations; the remaining patients did not recall receiving a tick bite.
The main clinical features in confirmed patients included sudden onset of fever (>37.5°C −40°C) lasting up to 10 days, fatigue, anorexia, headache, myalgia, arthralgia, dizziness, enlarged lymph nodes, muscle aches, vomiting and diarrhea, upper abdominal pain, and relative bradycardia (Table 1). A small number of cases suffered more severe complications, including hypotension, mental status alterations, ecchymosis, gastrointestinal hemorrhage, pulmonary hemorrhage, respiratory failure, disseminated intravascular coagulation, multiple organ failure, and/or death. Most patients had a good outcome, but elderly patients and those with underlying diseases, neurological manifestations, coagulopathy, or hyponatremia tended to have a poorer outcome.
Laboratory tests showed that confirmed patients characteristically developed thrombocytopenia, leukopenia, proteinuria, and elevated serum aspartate aminotransferase (AST) and alanine aminotransferase (ALT) levels (Table 2). Biochemical tests revealed generally higher levels of lactate dehydrogenase, creatine kinase, AST and ALT enzymes, especially AST.
One lane for each of the two sample pools (FTLS and healthy controls), each consisting of 10 samples, was sequenced. The proportions of high-quality sequences in FTLS and control pools were 98.08% (10,198,407/10,397,161) and 97.88% (10,250,809/10,472,315), respectively. Unique, high-quality sequence reads were then classified into broad taxonomic groups based on the taxonomy of the most frequent top-scoring BLAST matches for each sequence. Virus sequences constituted 0.0065% (67,969) and 0.0048% (50,677) of the reads in FTLS and control libraries, respectively (Table 3). After contig assembly, the number of sequences phylotyped as “viral” decreased to 3,163 and 2,412, respectively, in the two pools (Table 4). To screen for possible viruses, we focused exclusively on viruses that were present in patient sera.
For detection of known viruses, non-redundant reads were directly aligned with the GenBank database of nucleic acids using BLASTn software. Some sequences from Adenoviridae (human adenovirus), Herpesviridae (herpesvirus), Papillomaviridae (papillomavirus) and Retroviridae (human endogenous retrovirus) were detected in both FTLS patients and healthy subject samples. Some hepatitis B virus (HBV) and human bocavirus sequences were detected only in patient sera, indicating the presence of HBV and human bocavirus infections among these patients. To detect novel viruses, we examined sequence data against the GenBank protein database using BLASTx. An analysis of the deduced protein sequences revealed four different virus families in sera from FTLS patients (Table 4). Among these were viruses from Hepadnaviridae, which are not known to cause FTLS; Torque teno virus (TTV) from Anelloviridae, which has been reported to be associated with certain inflammatory states [34], but is not known to be transmitted by arthropods; and viruses from the Parvoviridae family, including human bocavirus, which could cause febrile illness and signs of FTLS. Although human bocaviruses are not known to be transmitted by arthropods, feline panleukopenia virus, a parvovirus, is strongly suspected to be transmitted by arthropods [35]. All family Parvoviridae sequences detected in FTLS samples were also assembled and their protein sequences deduced. Included among these samples were four fragments, all of which were found to be highly homologous to human bocavirus; one fragment showed the greatest similarity to human bocavirus 2 isolate 53044 (identity = 86%) with the lowest E-value (8×10−17). Human bocavirus was further detected by PCR in the pool of 10 serum samples, but only one individual sample within the pool tested positive for bocavirus. The final virus family detected in sera from FTLS patients was the Bunyaviridae family, which contains viruses known to cause FTLS after tick bites [8], [30]. All family Bunyaviridae sequences detected in FTLS samples, including 11 fragments (1 S-segment, 2 M-segments, and 8 L-segment fragments), were assembled and their protein sequences were deduced. Among these 11 novel virus fragments was a 168-bp fragment (C361, Accession: HQ412604) of the polymerase gene that showed the greatest similarity to Tehran virus (identity = 36%) with the lowest E-value (3×10−7). This suggested the presence of a novel virus or a known virus whose genome had not yet been sequenced.
To more accurately assess the genetic relationships to known viruses, we constructed a phylogenetic tree using a neighbor-joining method [33]. The result showed that the potentially novel virus clustered with Toscana virus, Uukuniemi virus, and Rift Valley fever virus of the Phlebovirus genus (Figure 1A). The pools of 10 serum samples from patients and 10 serum samples from healthy subjects were also screened by PCR for the presence of the novel bunyavirus. All 10 samples from patients tested positive for the novel bunyavirus; however, all 10 samples from healthy subjects tested negative. Thus, we further focused on the virus from the family Bunyaviridae.
Using specifically designed RT-PCR primers, we detected viral RNA in 223 of the 285 acute serum samples tested (Table 5). The specificity of the RT-PCR was confirmed by sequencing selected PCR products. None of the 80 sera from patients with respiratory diseases or the 50 sera from healthy subjects was positive using the novel virus-specific RT-PCR.
Six acute serum samples that tested positive for the novel bunyavirus by specific RT-PCR were inoculated onto Vero E6 cells, and four virus strains were isolated. The initial CPE analysis showed rounded refractile cells 2–4 days after inoculation. CPE did not progress in the initial cultures, but appeared slightly at 24 hours in subsequent passages (Figure 2A). RT-PCR revealed the presence of RNA for the novel bunyavirus in all four virus strains, and all isolates reacted with the serum of a convalescent patient in IFA (Figure 2C). In addition, electron microscopy showed the presence of virus particles approximately 80–90 nm in diameter—a size compatible with a bunyavirus (Figure 3). Virus particles were presumably localized to the Golgi apparatus (Figure 3).
Genome sequencing of one isolate (HN01) revealed three segments of negative polarity, single-stranded RNA, including a large segment (L; GenBank HQ642766), a medium-sized segment (M; GenBank HQ642767), and a small segment (S; GenBank HQ642768). The deduced amino acid sequence of the L segment had the highest homology (34%, E value = 3×10−163) to RNA polymerase of the Uukuniemi virus of the Phlebovirus genus, whereas the M segment had the highest homology (26%, E value = 8×10−55) to glycoprotein genes of the Punta Toro virus in the Phlebovirus genus. Of the two proteins encoded by the ambisense S segment, one had the highest homology to nucleocapsid protein (39%, E value = 3×10−40) of the Rift Valley Fever virus, and the other had the highest homology to nonstructural protein genes (24%, E value = 0.049) of the Punique Virus, both of which belong to the Phlebovirus genus. Collectively, these findings confirm that this virus belongs to the Phlebovirus genus of Bunyaviridae.
During the revision of this manuscript, some new sequences of SFTSV (severe fever with thrombocytopenia syndrome) were released. A comparison of these new SFTSV sequences with the sequence of this novel virus showed that they were highly homologous (>99% identity). Using sequences of Phlebovirus available in GenBank, a phylogenetic analysis showed that, although most closely related to the Uukuniemi virus of the Phlebovirus genus (34%, 24%, and 29% of maximum identity, respectively, for segment L, M, S at maximum query coverage), the three genomic segments of the novel virus, along with the SFTSV sequences, were highly divergent (Figure 1B–1E).
IgG antibodies to the novel bunyavirus were detected in 80 of 285 acute-phase serum samples from patients with FTLS (Table 5). Of 95 patients from whom paired acute- and convalescent-phase sera were available, 52 had seroconversions and 21 had greater than 4-fold increases in antibody titer to the virus. Six had less than a 4-fold increase in antibody titer to the virus, but all paired sera tested positive. Sixteen patients tested negative to the virus, suggesting that some non-FTLS patients with similar symptoms were included in this study, a situation that is not surprising given that FTLS is a newly emerging disease. The acute-phase sera of four patients from whom the virus was isolated tested negative for IgG antibody to the virus. All convalescent sera obtained 2 months later from the same four patients contained IgG antibody to the virus. None of the 130 sera from patients with respiratory diseases or healthy subjects had detectable antibody.
Since 2007, there has been an increase in reported cases of FTLS in Xinyang City, Henan Province. These patients were tentatively diagnosed as having A. phagocytophilum infection. However, only a few (8.4%, 24/285) such patients had evidence for A. phagocytophilum infection, and none of the 285 patients tested positive for the many other pathogens capable of causing similar clinical and laboratory manifestations that were also investigated. These findings suggested novel infectious agents, including viruses.
Traditionally, virus culture is very important for identifying an unknown viral infection. Before performing the Illumina sequencing strategy, we attempted viral and rickettsial culture with DH82 and BHK cell lines, but the lack of an obvious CPE led us to initially abandon this approach. Here, mass sequence data obtained by Illumina sequencing revealed four virus families that appeared only in FTLS patient sera. Among these four virus families, viruses from the Parvoviridae and Bunyaviridae families reportedly can cause signs of FTLS and be transmitted by arthropods. However, only one sample from a pool of ten samples tested positive for bocavirus by PCR, suggesting that bocavirus from the Parvoviridae is not likely involved in FTLS. For viruses in the Bunyaviridae family, the incidence of infection is closely linked to vector activity. For example, tick-borne viruses are more common in the late spring and late summer when tick activity peaks. Human infections with certain Bunyaviridae, such as Crimean-Congo hemorrhagic fever virus, are associated with high levels of morbidity and mortality [30]. Considering the tick-bite history of many FTLS patients, we focused on Bunyaviridae family viruses.
The entire Bunyaviridae family contains more than 300 members arranged in four genera of arthropod-borne viruses (Orthobunyavirus, Nairovirus, Phlebovirus and Tospovirus) and one genus (Hantavirus) of rodent-borne viruses [30], [36]. The Phlebovirus genus currently comprises 68 antigenically distinct serotypes, only a few of which have been studied. The 68 known serotypes are divided into two groups: the Phlebotomus fever group (the sandfly group, transmitted by Phlebotominae sandflies) comprises 55 members, and the Uukuniemi group (transmitted by ticks) comprises the remaining 13 members. Of these 68 serotypes, eight are linked to disease in humans, including the Alenquer, Candiru, Chagres, Naples, Punta Toro, Rift Valley fever, Sicilian, and Toscana viruses [30]. Phleboviruses have tripartite genomes consisting of a large (L), medium (M), and small (S) RNA segment.
In screening for unknown viruses, species hits alone likely carry little weight. Thus, we used all sequences in the family Bunyaviridae for our analysis. A 168-bp fragment of the polymerase gene with the lowest E-value and high sequence identity was used as the sequence of the unknown virus. This virus sequence was detected in all 10 pooled samples, indicating that the virus is involved in FTLS. After detecting a possible novel bunyavirus through high-throughput Illumina sequencing, we inoculated Vero cell lines, which are known to be sensitive to phleboviruses, with sera from six positive patients and were subsequently able to detect the virus by RT-PCR [30], [31]. Although the CPE was modest, RT-PCR confirmed the infection. Genome sequencing was performed and a phylogenetic analysis of the genome sequence showed that this virus clustered into the Phlebovirus branch, but was divergent from other known phleboviruses. These results confirm the novelty of this virus within the Phlebovirus genus of the family Bunyaviridae [36]. Furthermore, virus size and propagation in cells were similar to that of the bunyaviruses.
PCR and serological tests were performed to further test the causal link between the new virus and FTLS. Although we have not completely fulfilled Koch's postulates, evidence implicating this new bunyavirus in the outbreak of the disease among patients with FTLS is compelling.
In view of the fact that the disease is caused by a novel bunyavirus, and taking into account that the disease was first discovered in Henan (HN), we propose the name "Henan Fever" for the FTLS disease cause by the novel virus (proposed name “Henan Fever Virus” [HNF virus]). Since the submission of this manuscript, a bunyavirus was identified as the cause of FTLS in Chinese patients from other regions of China, and the authors have named this virus “SFTSV” to indicate that it is the cause of severe fever with thrombocytopenia syndrome [37]. After release of the GenBank sequences referred to in the Yu paper, we compared the sequences of SFTSV with those of FTLSV and found that they were nearly identical (>99% identity). As we first identified the syndrome in 2007 and described the presence of the virus in patients between 2007 and 2010, we suggest that the name “HNF virus” should take precedence. The most distinctive feature of the current work includes the use of an unbiased metagenomic approach for viral pathogen discovery that facilitated the rapid creation and implementation of standard culture, serological, and molecular diagnostic approaches. However, there are other differences between the results described here and those reported by Yu et al; notably, we observed slight, but distinctive, CPE in Vero cells. The reason for the failure to observe CPE in Vero cells infected with the “SFTSV” bunyavirus [37], whose genome is nearly identical to that of bunyavirus isolated from our FTLS patients, is unclear. Perhaps this reflects the fact that the ensuing CPE is not dramatic. Alternately, this could indicate the existence of distinct viral strains that vary in pathogenicity, virulence, and possibly even disease manifestations. This is an area of active study in our laboratories.
The discovery of this new virus will assist in the rapid diagnosis of this disease and help to distinguish it from other diseases caused by pathogens such as A. phagocytophilum, E. chaffeensis, Crimean-Congo hemorrhagic fever virus, Hantavirus, dengue virus, Japanese encephalitis virus, and Chikungunya virus. Furthermore, the availability of the new virus will facilitate the future development of new therapeutic interventions, such as vaccines and drugs.
|
10.1371/journal.pcbi.1002869 | Oroxylin A Inhibits Hemolysis via Hindering the Self-Assembly of α-Hemolysin Heptameric Transmembrane Pore | Alpha-hemolysin (α-HL) is a self-assembling, channel-forming toxin produced by most Staphylococcus aureus strains as a 33.2-kDa soluble monomer. Upon binding to a susceptible cell membrane, the monomer self-assembles to form a 232.4-kDa heptamer that ultimately causes host cell lysis and death. Consequently, α-HL plays a significant role in the pathogenesis of S. aureus infections, such as pneumonia, mastitis, keratitis and arthritis. In this paper, experimental studies show that oroxylin A (ORO), a natural compound without anti-S. aureus activity, can inhibit the hemolytic activity of α-HL. Molecular dynamics simulations, free energy calculations, and mutagenesis assays were performed to understand the formation of the α-HL-ORO complex. This combined approach revealed that the catalytic mechanism of inhibition involves the direct binding of ORO to α-HL, which blocks the conformational transition of the critical “Loop” region of the α-HL protein thereby inhibiting its hemolytic activity. This mechanism was confirmed by experimental data obtained from a deoxycholate-induced oligomerization assay. It was also found that, in a co-culture system with S. aureus and human alveolar epithelial (A549) cells, ORO could protect against α-HL-mediated injury. These findings indicate that ORO hinders the lytic activity of α-HL through a novel mechanism, which should facilitate the design of new and more effective antibacterial agents against S. aureus.
| The mechanism controlling protein-ligand interactions is one of the most important processes in rational drug design. X-ray crystallography is a traditional tool used to investigate the interaction of ligands and proteins in a complex. However, protein crystallography is inefficient, and the development of crystal technology and research remains unequally distributed. Thus, it seems impractical to explore the structure of the α-hemolysin-ORO monomer complex by crystallography. Therefore, we used molecular dynamics simulations to investigate the receptor-ligand interaction in the α-HL-ORO monomer complex. In this study, we found that oroxylin A (ORO), a natural compound with little anti-S. aureus activity, can inhibit the hemolytic activity of α-HL at low concentrations. Through molecular docking and molecular dynamics simulations, we determined the potential binding mode of the protein-ligand interaction. The data revealed that ORO directly binds to α-HL, an interaction that blacks the conformational transition of the critical “Loop” region in α-HL and thus prevents the formation of the α-HL heptameric transmembrane pore, which ultimately inhibits the hemolytic activity of α-HL. This mechanism was confirmed by experimental data. Furthermore, we demonstrated that ORO could protect against α-HL-mediated injury in human alveolar epithelial (A549) cells.
| Staphylococcus aureus is an opportunistic pathogen in humans and other mammals that causes many different types of infections, including superficial abscesses, septic arthritis, osteomyelitis, pneumonia, endocarditis, and sepsis [1], [2]. The number of virulence factors secreted by S. aureus, including extracellular and cell wall-related proteins, determines its pathogenicity [3]. The virulence factor α-hemolysin (α-HL) is one of the most important factors produced by the majority of S. aureus strains and recent studies have demonstrated that it plays a major role in S. aureus pneumonia [4]. Previous studies using a mouse model of S. aureus pneumonia have shown that S. aureus strains that lack the hla gene (and thus do not secrete α-HL) cause less lung injury and inflammation than the hla positive strains [5].
The α-HL protein, isolated from the gram-positive pathogenic bacterium S. aureus, is a well-studied model that has been used to elucidate mechanisms of membrane insertion by soluble proteins. Studies have shown that α-HL can self-assemble on the lipid bilayers of the membranes of susceptible host cells to form a wide heptameric pore [6]. The protein is toxic for a wide range of mammalian cells, particularly erythrocytes and epithelial cells and serves primarily as a tool that converts host tissue into nutrients for any bacteria that expresses it [3].
In an effort to increase our understanding of the function of α-HL, the structure of the heptameric pore was resolved by X-ray crystallography to a resolution of 0.19 nm [6]. Contained within the mushroom-shaped homo-oligomeric heptamer is a 10 nm long solvent-filled channel that runs along the seven-fold axis and ranges from 1.4 nm to 4.6 nm in diameter. The lytic transmembrane domain forms the lower half of a 14-strand antiparallel β barrel, to which each protomer contributes two 6.5 nm long β strands. Considering the essential nature of the heptameric crystal structure, Ragle et al. used a modified β-cyclodextrin compound, IB201, to prevent the α-HL-induced lysis of human alveolar epithelial cells (A549) [7]. This protective effect does not result from the ability of β-cyclodextrin to impair formation of the oligomeric α-HL on the cell surface, supporting a role for this molecule in the blockade of the lytic pore. Previous investigations had demonstrated the use of unsubstituted β-cyclodextrin as an adapter molecule that is capable of lodging within the central pore of α-HL and can thus facilitate the use of the toxin as a biosensor [8], [9]. The investigation of β-cyclodextrin using IB201 revealed that it blocks ion conductance through the assembled hemolysin pore, which supports the finding that β-cyclodextrin inserts into the pore itself. Although the inhibitory effect of β-cyclodextrin on ion conductance and red blood cell hemolysis were both observed in the low micromolar concentration range, this treatment strategy is passive. It is clear that prior to inhibition by β-cyclodextrin, the oligomeric α-HL on the cell surface has been formed and the cell has been damaged. Therefore, further research to identify new potent inhibitors is essential.
In our previous study, we reported that baicalin (BAI), a natural compound could bind with α-HL directly and inhibit the hemolytic activity of by restraining the conformation change of the binding cavity, “triangle region” (residues 147–153) [10]. In this study, we found that another natural compound, oroxylin A (ORO) could inhibit the hemolytic activity of α-HL stronger. Surprisingly, based on molecular dynamics simulations and free energy calculations, a new mechanism of inhibition was obtained compared with baicalin (BAI), which is that ORO bind to new active sites (residues Thr12 and Ile14) of α-HL and inhibit the hemolytic activity of α-HL due to the binding of ORO to the critical “Loop” region of α-HL. All these results indicate that the “triangle region” (residues 147–153) is not the only active site of inhibitor bound with α-HL and “Loop” region of α-HL also plays an important role in the inhibition of hemolytic activity of α-HL. With these approaches, we identified that ORO, which binds to the active site (Thr11, Thr12, Ile14, Gly15 and Lys46) of α-HL, is a potent inhibitor of the α-HL self-assembly process. These results could provide useful in the design of novel drugs for α-HL.
The studies that were performed to determine the minimal inhibitory concentration (MIC) showed that the maximum concentration of ORO tested was not able to inhibit the growth of S. aureus, which indicates that ORO has no antimicrobial activity against S. aureus. Our previous study showed that many natural compounds could inhibit the hemolytic activity of the culture supernatant of S. aureus by decreasing the expression of α-HL [11]. In the present study, we found that ORO cannot affect the production of α-HL in S. aureus (Figure 1). However, ORO attenuated the hemolytic activity of purified α-HL in a concentration-dependent fashion (Figure 2B and 2C). Consequently, it is reasonable to deduce that ORO has a direct effect on α-HL.
Based on the previous result, which showed that ORO inhibited the hemolytic activity of α-HL, we studied the binding of ORO to α-HL via molecular docking and molecular dynamics simulations using the AutoDock 4.0 and Gromacs 4.5.1 software packages, respectively. The initial structure of the monomeric α-HL was obtained from homology modeling, as previously reported [12].
The complex structure based on the docking results was used as the initial structure of the 200-ns molecular dynamics simulations and the preferential binding mode of ORO to α-HL was determined. The simulations show that ORO is a ligand that can bind to α-HL via hydrogen bonding and van der Waal interaction. Over the time course of the simulation, ORO localizes to the “Loop” region of α-HL, which is reported to participate in crucial protomer-protomer interactions during α-HL self-assembly and is therefore important in heptamer formation and cell lysis [6], [13]. The predicted binding mode of ORO with α-HL is illustrated in Figure 3A and the electrostatic potentials of the residues around the binding site are mapped, as shown in Figure 3B, using APBS software [14]. In detail, the binding model of ORO to the Loop of α-HL (Figure 3B) revealed that the methyl group of the 4H-chromen-4-one moiety of ORO formed a hydrogen bond with the side chain of the Lys46 amino acid in α-HL. As shown in Figure 4, the number of hydrogen bonds fluctuates mostly between 1 and 2 throughout the simulation time, which indicates that ORO and α-HL are always interacting via a hydrogen bond.
Moreover, the neutral side chains of the Thr11, Thr12, and Ile14 residues of α-HL form Van der Waals interactions with ORO, as shown in Figure 3B. Thr11 and Thr12 anchor the benzene ring of ORO, and Ile14 and Gly15 play an important role in stabilizing the 4H-chromen-4-one moiety of ORO. In addition, the methoxy of the 4H-chromen-4-one moiety forms strong interactions with Ser16 and Lys46, which will be confirmed by energy decomposition analysis.
The root mean square fluctuation (RMSF) of the residues surrounding the ORO binding site of α-HL (residues 1–50) in the α-HL-ORO complex and in free α-HL were calculated to illustrate the flexibility of these residues. The RMSF of these residues are shown in Figure 5 and clearly depict the difference in the flexibility of the binding site of α-HL due to the presence or absence of ORO. All of the residues in the α-HL binding site that is bound with ORO show a smaller degree of flexibility, with RMSF values less than 0.3 nm, when compared with the RMSF values calculated for the free α-HL, which indicates that these residues become more rigid after binding to ORO. These results indicate that the stabilization of the α-HL binding cavity in this complex is mostly due to residues Thr11, Thr12, Ile14, Gly15 and Lys46, as shown in Figure 3B.
The MD results provide an approximate binding mode of the protein-ligand interaction of the α-HL-ORO complex. However, the contribution of the residues surrounding the binding site of α-HL is not clear. Therefore, the electrostatic, Van der Waals, solvation and total contribution of the residues to the binding free energy were calculated using the Molecular Mechanics Generalized Born Surface Area (MM-GBSA) method [15], [16]. The calculation was performed over the 200 MD snapshots obtained from the last 50-ns simulation. The summations of the interaction free energies for each residue were separated into Van der Waals (ΔEvdw), electrostatic (ΔEele), solvation (ΔEsol), and total contribution (ΔEtotal). The energy contributions from the all residues of α-HL are summarized in Figure 6A.
As shown, Lys46 has an appreciable electrostatic (ΔEele) contribution, with a value less than −2.3 kcal/mol (Figure 6A and 6B). Because Lys46 is close to the methoxy of ORO and an electrostatic interaction exists, strong H-bonds are formed between α-HL and ORO. In addition, residues Thr12 (with a ΔEvdw of −1.2 kcal/mol) and Ile14 (with a ΔEvdw of ≤−2.1) exhibit strong Van der Waals interactions with the ligand because of the close proximity between these two residues and the 4H-chromen-4-one moiety of ORO. The majority of the decomposed energy interaction, with the exception of the energy associated with Lys46, originates from Van der Waals interactions. The electrostatic contribution from these key residues does not appear to have a significant influence on the formation of the α-HL-ORO complex.
The total binding free energy of the α-HL-ORO complex, ΔGbind, and its detailed energy contributions, which were calculated using the MM-GBSA approach, are summarized in Table 1. The ΔGbind can be divided into polar (ΔGele,sol+ΔEele) and nonpolar (ΔGnonpolar+ΔEvdw) energies. As shown, the primary contributor to the free energy of the binding of ORO to α-HL is ΔGnonpolar+ΔEvdw, with a value of −11.2 kcal/mol, whereas ΔGele,sol+ΔEele have a minor contribution of −6.4 kcal/mol. This difference arises from the intermolecular Van der Waals energy, which is mainly achieved from the ORO-binding α-HL residues. After summation of the solute entropy term (5.1 kcal/mol), an estimated ΔGbind of −12.5 kcal/mol was found, which suggests that ORO can strongly bind to and interact with the binding site of α-HL.
The same procedure was followed with two α-HL mutants, T12A-HL and I14A-HL, to verify the accuracy of the binding site in the α-HL-ORO complex. A complex of each mutant with ORO was used as the preliminary structure for the MD simulations and the MD trajectories were analyzed using the MM-GBSA method. In addition, the T12A-HL and I14A-HL mutants were expressed and purified; the binding free energy and the number of binding sites between ORO and the two mutants were then determined using the fluorescence spectroscopy quenching method [17], [18].
As illustrated in Figure 6D and 6F, ORO differentially binds to the two mutants and the WT-HL, an observation that was confirmed by pair interaction decomposition analysis of the free energy, as shown in Figure 6C and 6E. The major contributions to the free energy of the complexes of ORO with the α-HL mutants were residues Thr115, Tyr118, Pro103, Phe120 and Ile142. Furthermore, as shown in Table 1, the MM-GBSA calculation predicted that T12A-HL and I14A-HL bound more weakly to ORO than WT-HL, with estimated ΔGbind values of −3.8 kcal/mol and −5.2 kcal/mol, respectively. Consequently, the calculations for T12A-HL and I14A-HL show that these mutants exhibited a decrease in the binding energy of approximately 7 to 8 kcal/mol compared to WT-HL. The experimental measurement of the binding free energy, shows that the interaction between ORO and WT-HL is highest, which means that WT-HL has the strongest ability to bind to ORO; the mutants exhibits the weaker ability, as shown in Table 2. Importantly, because the calculated binding free energies are in good agreement with the experimental data shown in Table 2, we believe that the MD simulations generated a reliable model of the α-HL-ORO complex.
ORO inhibits the hemolytic activity of α-HL and binds to the “Loop” cavity of α-HL, which has been shown to be critical to its hemolytic activity [6], [19]. Consequently, the conformations of the “Loop” region in the α-HL-ORO complex and in free α-HL were investigated using a MD simulated trajectory. As shown in Figure 7A, the distance between the Cα of Thr12 and the Cα of Thr19 in the complex ranged from 0.9 to 1.15 nm over the time course of the simulation, with an average distance of 1.05 nm (Figure 7A and 7B). In the absence of ORO, the distance between these points ranges from 1.15 to 1.3 nm, with an average distance of 1.24 nm. However, the distance between the Cα of Thr12 and the Cα of Thr19 is 1.81 nm in the crystal structure of the α-HL heptamer, which is available in the Protein Data Bank under the accession number 7AHL. Dynamic fluctuations in the distance between the Cα of Thr12 and the Cα of Thr19 likely indicate that the conformation of the “Loop” region is restrained when ORO binds to these two residues. Through comparing the structure of the loop in α-HL-ORO complex with the crystal structure of α-HL monomeric observed in the X-ray structure of the oligomer (PDB code: 7AHL), it is indicated that ORO blocks the required conformational transition of the loop by binding to the loop, which is the mechanism of decreasing the lytic activity of α-HL. The RMSD values also reflect the variation in the conformation change of the “Loop” region during the simulation time. As shown in Figure 8, the RMSD values of the “Loop” region in free α-HL are ∼0.4 nm; these values are clearly higher than that of the α-HL-ORO complex, which is in the range of 0.323 nm. In addition, as shown by the RMSD trends displayed in Figure 8, the RMSD data agreed with other types of measurements, reinforcing that the WT-HL and ORO complex displays very little variation in the conformation of the “Loop” during the MD simulation. Furthermore, significant differences were observed in the dynamics of the critical “Loop” region during the MD simulations of the wild type and free α-HL, which confirms the effect of inhibition of α-HL on cell lysis. As illustrated in Figure 9, our models predict a marked conformational transition in the “Loop” region for the wild type complex to the free α-HL.
A comparative analysis of these MD simulations suggests that the inhibitory activity of ORO is highest for the wild type α-HL, followed by the T12A mutant and then I14A. This conclusion, which is also supported by the mechanism of hemolytic inhibition, is in good agreement with experimental results. Data from a deoxycholate-induced oligomerization assay shows that the site-directed mutagenesis of T12A and I14A has no influence on the assembly of the SDS-stable oligomer, α-HL7. However, the formation of α-HL7 was inhibited when treated with 8 µg/ml of ORO. This inhibitory effect was decreased with either of the two mutants, as shown in Figure 10, with I14A showing a higher inhibition than T12A.
These findings support one possible inhibition mechanism: the binding of ORO to the “Loop” region blocks the conformational change, which inhibits the self-assembly of the heptameric transmembrane pore, thereby decreasing the lytic activity of α-HL.
Human alveolar epithelial (A549) cells have previously been employed to investigate the influence of S. aureus on lung cell injury and α-HL has been found to be the major factor associated with their injury and death [7]. Because our previous results show that ORO blocks the self-assembly of α-HL, we speculated that ORO may protect A549 cells from S. aureus-mediated death. Consequently, A549 cells were stained with a live/dead (green/red) reagent following co-culture with S. aureus 8325-4. The uninfected cells displayed the green fluorophore (Figure 11A), indicating their live status. A549 cells were significantly affected by their co-culture with S. aureus, as reflected by the increase in the amount of red fluorophore observed (Figure 11B). However, the addition of 8 µg/ml of ORO resulted in a significantly lower number of dead cells (Figure 11C). Consistent with a previous study, treatment of A549 cells with the S. aureus strain DU1090, which cannot produce α-HL, does not result in their death (Figure 11D) [11]. Furthermore, a lactate dehydrogenase (LDH) release assay was employed to quantitatively assess the influence of ORO on the protection of A549 cell injury and, as shown in Figure 11E, the addition of 1 to 8 µg/ml of ORO affords a dose-dependent protection. The 50% inhibitory concentration (IC50), which was calculated using OriginPro 8.0 (OriginLab, USA), was 3.09 µg/ml. These results highlight the potential therapeutic effect of ORO, which merits further investigation.
Historically, vancomycin and linezolid have been the recommended empirical and definitive therapies for the treatment of methicillin-resistant S. aureus pneumonia. However, the emergence of multi-drug-resistant S. aureus, such as vancomycin-resistant S. aureus, makes S. aureus infection difficult to treat and increases its mortality rate [20]. Due to our increasing understanding of bacterial pathogenesis and intercellular cell signaling, several potential strategies have been developed for drug discovery, of which the anti-virulence strategy has the interest of most researchers [21]. S. aureus can secrete numerous surface proteins and exotoxins, which are involved in the process of pathopoiesis [22]. One of the most important is α-HL, which often leads to tissue damage. Our previous research has also shown that several natural compounds can protect mice against S. aureus pneumonia by decreasing the production of α-HL [11]. Based on the results of all these studies, we theorized that α-HL can be used as a target for the development of new drugs against S. aureus.
The increasing interest in drug design based on the identification of novel virulence targets has created a demand for the structural characterization of protein-ligand complexes. X-ray crystallography is a traditional tool used to investigate the interaction of ligands and proteins in a complex, and many studies using this technique have been reported. In 1996, Song et al. discovered the crystal structure of the α-HL heptamer; however, to date, the monomeric structure of α-HL remains unknown, suggesting that crystals of the α-HL monomer may be very difficult to obtain. Thus, it seems impractical to explore the structure of the ORO-α-HL monomer complex by crystallography. In the literature, computational chemistry combined with experimental confirmation has proven an effective and reliable method for exploring the interactions between ligands and proteins. Accordingly, in this study, we used a complementary approach that includes molecular dynamics simulations (MD simulations), site-specific mutagenesis, and a fluorescence-quenching method to further explore ligand-protein binding sites. Specifically, we attempted to identify the mechanism by which ORO inhibits the biological activity of α-HL.
Moreover, to explore the formation of the interaction between a protein and a ligand at the atomic level, we used the MM-GBSA method to determine the associated free energy profiles. It has been known for years, and was restated recently [23], that free energy calculations from MD simulations can prove to be a powerful tool for exploring the process of ligand-protein binding when used in combination with mutagenesis experiments [24]–[27]. In addition, MM-GBSA calculations were performed on a series of derivatives of TIBO (a substituted tetrahydroimidazole benzodiazepine thione) to explore their potential as inhibitors of HIV-1 reverse transcriptase [28]. In the same study, the binding mode of a known drug was predicted with excellent agreement to the X-ray structure, which was discovered afterward. Other examples of studies with MD simulations include work by Biswa Ranjan Meher et al. [29], who examined the binding of the inhibitor darunavir to wild-type and mutant proteins using all-atom MD simulations and MM-GBSA calculations, and work by Lstyastono et al. [30], whose study focused on the elucidation of molecular determinants of G protein-coupled receptor-ligand binding modes by combining MD simulations and site-directed mutagenesis studies. There are a number of other such examples of studies that employ MD simulations [31]–[36]. In this study, we discovered that Van der Waals interactions play an important role in the stabilization of the binding site of α-HL-ORO. The key residues, Thr11, Thr12, Ile14, Gly15 and Lys46, in the complex were also identified, using residue decomposition analysis and mutagenesis assays. The conformational transition of the critical “Loop” region from the monomeric α-HL to the oligomer was blocked by the binding of ORO, which resulted in the inhibition of the hemolytic activity of α-HL. These findings indicate that ORO hinders the lytic activity of α-HL through a novel mechanism. This was confirmed by inducing the formation of the α-HL heptamer by Deoxycholate, the results of which indicate that addition of ORO inhibits the formation of the α-HL heptamer.
In summary, we found that oroxylin A (ORO), a natural compound without anti-S. aureus activity, can inhibit the hemolytic activity of α-HL. Based on the results of MD simulation, we confirmed that ORO could inhibit the hemolytic activity of α-HL by a new mechanism which is completely different compared with baicalin (BAI). Through the analysis of the binding free energy of the complex formation using MM-GBSA method, the results show that the residues Thr11, Thr12, Ile14, Gly15 and Lys46, which surround the binding site of the α-HL-ORO complex, are key ORO-binding residues. Due to the binding of ORO, the conformation transition of the critical “Loop” region from the monomeric α-HL to the oligomer was blocked, which resulted in inhibition of the hemolytic activity of the protein. This novel mechanism was confirmed by experimental data using a deoxycholate-induced oligomerization assay. The whole results mentioned above indicate that the “triangle region” of α-HL is not the only active site of inhibitor and “Loop” region of α-HL also plays an important role in the inhibition of hemolytic activity of α-HL, which could facilitate the design of new and more effective antibacterial agents.
S. aureus 8325-4, a high-level α-HL-producing strain, and its cognate α-HL-deficient mutant, DU 1090, were used in this study. The S. aureus strains 8325-4 and DU 1090 were cultured in TSB to an optical density of 0.5 at 600 nm. Then, either the cultures were centrifuged and resuspended in DMEM medium for the live/dead and cytotoxicity assays. ORO (purity>98.5%) (Figure 2A) was obtained from Sigma-Aldrich (St. Louis, MO, USA), and the stock solutions were prepared in dimethyl sulfoxide (DMSO) (Sigma-Aldrich). For the in vivo studies, ORO was dissolved in sterile PBS. The minimum inhibitory concentrations (MICs) of ORO for S. aureus were evaluated using the broth microdilution method according to the Clinical and Laboratory Standards Institute (CLSI) guidelines.
Hemolytic activity was measured as described elsewhere using rabbit red blood cells [7]. In brief, 100 µl of purified α-HL was pre-incubated in 96-well microtiter plates in the presence of either gradient concentrations of ORO or PBS control at 37°C for 10 min. Defibrinated rabbit red blood cells (100 µl; 5×106 cells per milliliter) in PBS were then added to the wells and the mixtures were incubated at 37°C for 20 min using 1% Triton X-100 as a positive control. After centrifugation, the supernatants were removed and their absorption at 543 nm was measured. The percent hemolysis was calculated using the supernatant reading from an equivalent number of cells that had been lysed in 1% Triton X-100.
Western blot analysis was performed as previously described [37]. Briefly, S. aureus 8325-4 and DU 1090 were cultured at 37°C in TSB and different concentrations of ORO to an optical density at 600 nm of 2.5. The cultures were collected by centrifugation and the supernatants were used in sodium dodecyl sulfate (SDS)-polyacrylamide (12%) gel electrophoresis. The proteins were then transferred onto polyvinylidene fluoride membranes (Roche, Basel, Swiss) using a semi-dry transfer cell (Bio-Rad, Munich, Germany). After blocking the membrane for 2 h with 5% Bovine Serum Albumin (BSA) (Amresco, USA) at room temperature, an anti-hemolysin primary polyclonal antibody (Sigma-Aldrich) was added at a 1 ∶ 5000 dilution. The membrane was then incubated overnight at 4°C and then for 2 h with a HRP-conjugated secondary goat anti-rabbit antiserum (Sigma-Aldrich) that was diluted to 1 ∶ 4000. The blots were developed using Amersham ECL western blotting detection reagents (GE Healthcare, UK).
For rational drug discovery, modeling and informatics play an indispensable role in the identification of lead compounds and their most plausible mechanisms of action against particular biological targets [38]. Therefore, we have performed a homology-modeling study on α-HL. To date, the structure of monomeric α-HL is unavailable and only the crystal structure of the α-HL heptamer has been reported [6].
The model of the monomeric α-HL in solution was proposed based on homology modeling, as previously reported [12]. MODELLER [3], version 9.9, was used to generate structural models of α-HL based on the template structures of LukF (PDB codes 1LKF_A), LukF-PV (PDB code 1PVL_A), Gamma-hemolysin component A (PDB code 2QK7_A) and LukS-PV (PDB code 1T5R_A). As a whole, the sequence identities between the templates Lukf-PV, LukF, Gamma-hemolysin component A, LukS-PV and the query monomeric α-HL are 30%, 31%, 26%, and 22%. The most important difference between template and query is located in “loop” region (residues 1–50), which is the critical region for the assembled hemolysin pore. The program optimizes the structure of the homology models by minimizing a global probability density function that integrates the stereochemical parameters and homology-derived restraints [39]. The best model was selected based on its DOPE score, and it was subjected to further 200 ns molecular dynamics using Gromacs 4.5.1 software package [40].
The geometry of ORO was optimized at the B3LYP/6-31G* level using the Gaussian 03 program [41].
The initial structure of α-HL was obtained from the homology modeling. To obtain the starting structure of the drug/α-HL complex for molecular dynamics (MD) simulation, a standard docking procedure for a rigid protein and a flexible ligand was performed with AutoDock 4 [42], [43]. The Lamarckian genetic algorithm (LGA) was applied in the docking calculations. All of the torsional bonds of the drug were free to rotate while α-HL was held rigid. Then, the polar hydrogen atoms were added for α-HL using the AutoDock tools, and Kollman united atom partial charges [44] were assigned. A total of 150 independent runs were carried out with a maximum of energy evaluations to 25,000,000 and a population size to 300. A grid box (50×40×49) with spacing of 0.1 nm was created and centered on the mass center of the ligand. Energy grid maps for all possible ligand atom types were generated using Autogrid 4 before performing the docking.
The clusters were ranked according to the lowest energy representative in each cluster. Then, the ligand docking poses suggesting preferential binding to the loop region are three: Pose 1, Pose 2, and Pose 3. Pose 1 has the lowest energy conformation (−6.5 kcal/mol) and the most populated cluster (28) compared with Pose 2 (−5.6 kcal/mol, 15) and Pose 3(−4.9 kcal/mol, 7), and then Pose 1 was chosen for further study. The lowest energy conformation in the most populated cluster was chosen for further study [45].
All of the simulations and the analysis of the trajectories were performed with Gromacs 4.5.1 software package using the Amber ff99sb force field and the TIP3P water model [40], [46]. The α-HL-ORO system was first energy relaxed with 2000 steps of steepest-descent energy minimization followed by another 2000 steps of conjugate-gradient energy minimization. The system was then equilibrated by a 500 ps of MD run with position restraints on the protein and ligand to allow for relaxation of the solvent molecules. The first equilibration run was followed by a 200 ns MD run without position restraints on the solute. The first 20 ns of the trajectory were not used in the subsequent analysis to minimize convergence artifacts. The equilibration of the trajectory was checked by monitoring the equilibration of quantities, such as the root-mean-square deviation (RMSD) with respect to the initial structure, the internal protein energy, and fluctuations calculated for different time intervals. The electrostatic term was described with the particle mesh Ewald algorithm. The LINCS [47] algorithm was used to constrain all bond lengths. For the water molecules, the SETTLE algorithm [48] was used. A dielectric permittivity, ε = 1, and a time step of 2 fs were used. All atoms were given an initial velocity obtained from a Maxwellian distribution at the desired initial temperature of 300 K. The density of the system was adjusted during the first equilibration runs at NPT condition by weak coupling to a bath of constant pressure (P0 = 1 bar, coupling time τP = 0.5 ps) [49]. In all simulations, the temperature was maintained close to the intended values by weak coupling to an external temperature bath with a coupling constant of 0.1 ps. The proteins and the rest of the system were coupled separately to the temperature bath. The structural cluster analysis was carried out using the method described by Daura and co-workers with a cutoff of 0.25 nm [48].
The ORO parameters were estimated with the antechamber programs [49] and AM1-BCC partial atomic charges from the Amber suite of programs [50].
Analysis of the trajectories was performed using PyMOL analysis tools and Gromacs analysis tools.
In this work, the binding free energies are calculated using MM-GBSA approach supplied with Amber 10 package. We choose a total number of 200 snapshots evenly from the last 50 ns on the MD trajectory with an interval of 10 ps. The MM-GBSA method can be conceptually summarized as:(1)(2)where ΔH of the system is composed of the enthalpy changes in the gas phase upon complex formation (ΔEMM) and the solvated free energy contribution (ΔGsol), while −TΔS refers to the entropy contribution to the binding. Eq. (2) can be then approximated as shown in Eq. (3):(3)where ΔEMM is the summation of the van der Waals (ΔEvdw) and the electrostatic (ΔEele) interaction energies.(4)In addition, ΔGsol, which denotes the solvation free energy, can be computed as the summation of an electrostatic component (ΔGele,sol) and a nonpolar component (ΔGnonpolar,sol), as shown in Eq. (5):(5)
The interactions between ORO and the all residues of α-HL are analyzed using the MM-GBSA decomposition process applied in the MM-GBSA module in Amber 10. The binding interaction of each ligand-residue pair includes three terms: the Van der Waals contribution (ΔEvdw), the electrostatic contribution (ΔEele), and the solvation contribution (ΔEsol). All energy components are calculated using the same snapshots as the free energy calculation.
The binding constants (KA of ORO to the binding site on WT-HL, T12A-HL and I14A-HL were measured using the fluorescence-quenching method, and the binding constants were converted to the binding energy by Eq. ΔGbind = RTlnKA. Fluorescence spectrofluorimetry measurements were carried out using a Horiba Jobin-Yvon Fluorolog 3–221 spectrofluorometer (Horiba Jobin-Yvon, Edison, NJ). The measurements were acquired using a 280-nm excitation wavelength with a 5-nm band-pass and a 345-nm emission wavelength with a 10-nm band-pass. Details of the measurements were described previously [51]–[53].
Oligomerization assay was performed as described previously [54], 200 ng WT-HL, T12A-HL or I14A-HL monomers was mixed with 5 mM deoxycholate separately, following the addition of ORO, the mixtures were incubated at 22°C for 20 min. Then 5× loading buffer without β-mercaptoethanol was added to the mixtures and incubated at 50°C for 10 min. 25 µl of each reaction mixture was loaded onto 12% sodium dodecyl sulfate (SDS)-polyacrylamide gel electrophoresis (PAGE) gels for electrophoresis. Gels were stained using the silver PlusOne staining kit (GE Healthcare) according to the manufacturer's instruction.
Human lung epithelial cells (A549) were obtained from the American Tissue Culture Collection (ATCC CCL 185) and cultured in Dulbecco's modified Eagle's medium (DMEM) (Invitrogen, CA, USA) supplemented 10% fetal bovine serum (Invitrogen). Cells were seeded in 96-well dishes at a density of approximately 2×105 cells each well. As described previously [55], A549 cells were incubated with 100 µl of staphylococcal suspension with the addition of different concentrations of ORO or positive control PBS for 6 h at 37°C, DU1090 suspension was used as negative control. Cell viability was determined either using live/dead (green/red) reagent (Invitrogen) or by measuring lactate dehydrogenase (LDH) release using a Cytotoxicity Detection kit (LDH) (Roche) according to the manufacturer's directions. Microscopic images of stained cells were obtained using a confocal laser scanning microscope (Nikon, Japan). LDH activity was measured on a microplate reader (TECAN, Austria).
The significance of hemolysis, LDH release assay results were determined using the two-tailed Student's t test. Differences were considered statistically significant when P<0.05.
S. aureus hla gene: NC_007795.1
|
10.1371/journal.ppat.1004545 | SUMOylation by the E3 Ligase TbSIZ1/PIAS1 Positively Regulates VSG Expression in Trypanosoma brucei | Bloodstream form trypanosomes avoid the host immune response by switching the expression of their surface proteins between Variant Surface Glycoproteins (VSG), only one of which is expressed at any given time. Monoallelic transcription of the telomeric VSG Expression Site (ES) by RNA polymerase I (RNA pol I) localizes to a unique nuclear body named the ESB. Most work has focused on silencing mechanisms of inactive VSG-ESs, but the mechanisms involved in transcriptional activation of a single VSG-ES remain largely unknown. Here, we identify a highly SUMOylated focus (HSF) in the nucleus of the bloodstream form that partially colocalizes with the ESB and the active VSG-ES locus. SUMOylation of chromatin-associated proteins was enriched along the active VSG-ES transcriptional unit, in contrast to silent VSG-ES or rDNA, suggesting that it is a distinct feature of VSG-ES monoallelic expression. In addition, sequences upstream of the active VSG-ES promoter were highly enriched in SUMOylated proteins. We identified TbSIZ1/PIAS1 as the SUMO E3 ligase responsible for SUMOylation in the active VSG-ES chromatin. Reduction of SUMO-conjugated proteins by TbSIZ1 knockdown decreased the recruitment of RNA pol I to the VSG-ES and the VSG-ES-derived transcripts. Furthermore, cells depleted of SUMO conjugated proteins by TbUBC9 and TbSUMO knockdown confirmed the positive function of SUMO for VSG-ES expression. In addition, the largest subunit of RNA pol I TbRPA1 was SUMOylated in a TbSIZ-dependent manner. Our results show a positive mechanism associated with active VSG-ES expression via post-translational modification, and indicate that chromatin SUMOylation plays an important role in the regulation of VSG-ES. Thus, protein SUMOylation is linked to active gene expression in this protozoan parasite that diverged early in evolution.
| African trypanosomes have evolved one of the most complex strategies of immune evasion by routinely switching the expression of surface proteins called Variant Surface Glycoproteins (VSG), only one of which is expressed at any given time. Previous work has suggested that the recruitment of a single VSG telomeric locus to a discrete nuclear body (ESB) underlies the mechanism responsible for VSG monoallelic expression. Our findings establish unexpected roles for SUMOylation as a specific post-translational modification that marks the ESB and the VSG-ES chromatin. We describe a highly SUMOylated focus (HSF) as a novel nuclear structure that partially colocalizes with the VSG-ES locus and the nuclear body ESB. Furthermore, chromatin SUMOylation is a distinct feature of the active VSG-ES locus, in contrast to other loci investigated. SUMOylation of chromatin-associated proteins is required for efficient recruitment of the polymerase to the VSG-ES promoter and for VSG-ES expression. Altogether, these data suggest the presence of a large number of SUMOylated proteins associated with monoallelic expression as Protein Group SUMOylation. In contrast to the wealth of literature focused on VSG regulation by silencing, our results indicate a positive mechanism via SUMOylation to regulate VSG expression in the infectious form of this protozoan parasite.
| Trypanosoma brucei displays a sophisticated mechanism of antigenic variation of the Variant Surface Glycoprotein (VSG) that allows the parasite to elude the host immune antibody response, ensuring a persistent infection [1], [2]. Antigenic variation is achieved by mutually exclusive expression of only one out of approximately 1000 VSG genes. The monoallelic expressed VSG gene is located at the end of a telomeric Expression Site (ES) locus. There are about 15 different VSG expression sites (VSG-ESs), which share highly homologous sequences at the promoter region [3]. The identification of a single extra-nucleolar RNA polymerase I-containing nuclear body, named the expression site body (ESB), which is associated with the GFP-tagged active VSG-ES promoter suggests a model whereby ESB-dependent VSG-ES recruitment leads to the expression of a single VSG on the surface of the parasite [4]–[6].
Transcription of the VSG-ES and maintaining monoallelic expression seem be controlled at multiple levels. Several proteins have been involved in silencing of inactive VSG-ESs, such as telomeric protein RAP1, DOT1 histone methyltransferase, the factor ISWI and chromatin remodeler complex FACT [7]–[10]. Recently, it has been reported that the active VSG-ES promoter is depleted of histones [11], [12]. Whilst most studies have focused on regulation of VSG-ES silencing, there must be specific factors required to guarantee high levels of transcription of the active VSG-ES. The architectural protein TDP1, a high mobility group (HMG) containing protein, facilitates RNA pol I activity, however is required for both VSG-ES and rDNA transcription [13].
In T. brucei, the VSG-ES is transcribed by RNA polymerase I (RNA pol I), an exceptional feature among eukaryotes since RNA pol I does not usually transcribe protein-coding genes. However, TbRPB7, a dissociable subunit of the RNA pol II complex, is also required for in vivo RNA pol I transcription of the VSG gene [14]. This is a controversial issue in the field since TbRPB7 does not seem to be required for in vitro transcription [15]. These discrepancies maybe explained by a possible function of TbRPB7 in vivo, as we discussed previously [16]. Based in our previous results we sought for TbRPB7-interacting proteins in search for possible factors involved in VSG-ES regulation. To do so, we directed a yeast two-hybrid screen (Hybrigenics) and identified several proteins, including a protein with a conserved SUMO E3 ligase domain (MIZ/SP-RING), that we named TbSIZ1.
SUMO (Small Ubiquitin-like MOdifier) is a reversible post-translational protein modification involved in many cellular processes, including the regulation of nuclear bodies. The first SUMO gene was identified in S. cerevisiae (SMT3); the peptide was found covalently attached to the Ran GTPase-activating protein, modifying the localization of this protein in the cell [17], [18]. SUMO are ∼12 kDa proteins with a 3D structure similar to ubiquitin, whilst sharing just 20% sequence identity. Invertebrates such as yeast, C. elegans, and D. melanogaster contain a single SUMO gene, whereas plants and vertebrates have several SUMO genes [19].
SUMOylation, like ubiquitylation, involves a pathway that requires three enzymatic steps. First, the SUMO protein is activated at its C terminus by the E1 activating enzyme [20]. The activated SUMO is then transferred to the E2 conjugating enzyme UBC9 and to the substrate forming an isopeptide bond. This last step is mediated by SUMO E3 ligases, which determine substrate specificity and catalyse the transfer of SUMO from UBC9 [21], [22].
Three protein families have been identified to date as SUMO E3 ligases. The main group is characterized by a conserved SP-RING motif, which is essential for their function. This group includes the PIAS family (Protein inhibitor of activated STAT) PIAS1-3 in mammals [23], and Siz1, Siz2 and Mms21 in budding yeast [21], [24]. One of their mechanisms consists in re-localization of transcriptional regulators to different subnuclear compartments [25]. The second type of SUMO E3 ligases is represented by the nuclear import factor RanBP2, which mediates nucleo-cytoplasmic transport [26]. The third group was discovered with the polycomb protein Pc2, which forms PcG nuclear bodies involved in gene silencing [27].
SUMO modification regulates protein activity in diverse ways. SUMO can modulate the ability of proteins to interact with their partners, alter their patterns of sub-cellular localization and control their stability. The most common group of SUMO substrates are transcription factors, whose transcriptional activity can be modulated positively or negatively as a result of SUMOylation [28].
In T. brucei, there is a single SUMO protein which has been shown to be essential in procyclic [29] and bloodstream forms [30] of the parasite. Recently, proteomic analysis of SUMO substrates in T. cruzi showed at least 236 proteins involved in several cellular processes [31]. Together these data suggest that SUMO is essential and SUMOylation is a conserved process in trypanosomatids. The lack of an anti-SUMO antibody specific for TbSUMO hampered a proper analysis of the SUMO conjugated proteins [32]. Thus, a possible SUMO function in gene expression and subcellular localization of SUMO-conjugated proteins in the infective form of this protozoan parasite are totally unknown.
We here show the presence of a single site in the nucleus highly enriched in SUMOylated proteins, which associates with the VSG-ES chromatin and the nuclear body ESB. Importantly, we identify the SUMO E3 ligase, named TbSIZ1, responsible for the VSG-ES chromatin SUMOylation. Our data indicate that SUMOylation of chromatin-associated proteins at the active VSG-ES promoter is highly enriched in a TbSIZ1-dependent manner. SUMOylation of chromatin-associated proteins contributes to efficient recruitment of RNA polymerase I to the VSG-ES promoter and is important for VSG-ES expression. In addition, RNA pol I largest subunit TbRPA1 is SUMOylated in a TbSIZ1-depending manner. However, additional chromatin-associated proteins are SUMOylated in the active VSG-ES since SUMO was detected upstream of the promoter. This epigenetic mark in chromatin was not detected in silent VSG-ESs nor in rDNA or EP transcribed also by RNA pol I, suggesting that SUMOylation is involved in VSG-ES monoallelic active expression rather than in silencing.
To investigate SUMO-conjugated protein expression we first developed a monoclonal antibody (mAb 1C9H8) against Trypanosoma brucei SUMO expressed as recombinant protein. Western blot analysis showed that the most abundant SUMO-conjugated proteins are larger than 70 kDa in bloodstream form trypanosome total extracts (Figure 1A), similar to the pattern described in other eukaryotes [21], [33]. The mAb 1C9H8 recognized free SUMO and SUMO-conjugated proteins since SUMO depletion by RNAi of the coding region showed a significant decreased signal after 48 h of depletion by Western blot analysis (Figure 1A). RNAi-induced lines were compared to the parental cell line since uninduced cell lines generally showed some depletion of the target protein due to leaky RNAi expression. Additional RNAi experiments using the TbSUMO 5′ UTR showed a similar depletion of SUMO by Western blot analysis (Figure S1A). Importantly, the use of N-Ethylmaleimide (NEM), a well-known inhibitor of de-sumoylases, reduced the signal of free SUMO in protein extracts and stabilized SUMO-conjugated proteins (Figure S1B), suggesting NEM inhibits trypanosome de-sumoylation. The previous use of the anti-Trypanosoma cruzi SUMO antiserum against T. brucei SUMO conjugated proteins [30] is controversial [32]. We compared the anti-TcSUMO rabbit antiserum on TbSUMO-depleted extracts by RNAi with the signal obtained using the anti-TbSUMO mAb on the same Western blot (Figure S1C). While the signal generated by the anti-TbSUMO mAb was abolished upon depletion, anti-TcSUMO signal was not significantly reduced. Altogether, these data suggest that the anti-TbSUMO mAb 1C9H8 showed specificity to recognize SUMO-conjugated proteins in T. brucei extracts. Comparative analysis of T. brucei total extracts in bloodstream and procyclic (insect form) developmental stages of the parasite showed differential expression pattern of several SUMO-conjugated proteins (Figure 1B).
Next, we analyzed the subcellular localization of SUMOylated proteins by three-dimensional immunofluorescence (3D-IF) microscopy using the mAb anti-TbSUMO 1C9H8. SUMO modified proteins localized mainly to the nucleus, excluding the nucleolus, in a diffuse pattern with one Highly SUMOylated Focus (HSF) (Figure 1C). Statistical IF analysis for the detection of this single HSF revealed a significant visualization in 74.9% of the nuclei, irrespective of cell cycle stage (Figures S2A and S2B). Conversely, in the procyclic insect form, where no VSG is expressed, SUMO-conjugated nuclear proteins are located in numerous small foci dispersed in the nucleus (Figure 1D).
We carried out a series of double 3D-IF experiments to investigate a possible association of SUMO with trypanosome sub-nuclear compartments in bloodstream form nuclei. Anti-TbRPA1 (RNA pol I largest subunit) affinity-purified antiserum is known to recognize not only the nucleolus but also the extra-nucleolar body named ESB [4]. Double IF analysis by 3D-deconvolution microscopy using the mAb anti-TbSUMO and the anti-TbRPA1 antiserum showed that the HSF partially colocalized with the nuclear body ESB (Figure 2A). To further investigate the association between RNA pol I and SUMOylated proteins in the nucleus, 3D-IF analysis was performed in a cell line expressing a YFP-tagged TbRPB5z [34], a subunit specific of RNA pol I complex in trypanosomes. Consistent with the nuclear localization of TbRPA1, TbRPB5z was associated with the HSF in the nucleus (Figure S2C), suggesting that the RNA pol I complex located in the extra-nucleolar ESB is associated with the HSF.
Next, we wished to investigate whether the HSF was associated with the VSG-ES chromosome position in the nucleus. To do so, we performed indirect 3D-IF analysis utilizing a cell line tagged with the GFP-Lac upstream of the active VSG-ES promoter [4]. Double 3D-IF analysis using anti-TbSUMO and anti-GFP antibodies showed that the GFP-tagged active VSG-ES partially colocalized with the HSF in a large percentage of cells (76.1%) (Figures 2B and S2D). As control, we investigated a possible association of SUMOylated proteins with the ribosomal DNA (rDNA) in the nucleus, another locus transcribed by RNA pol I. Nuclear position analysis of the rDNA locus, marked with the GFP-Lac [35], showed a lack of significant colocalization with SUMOylated proteins (Figures 2C and S2D).
The monoallelic VSG-ES transcriptional state is maintained over many generations and during S-phase, G2-phase and early mitosis the active VSG-ES locus remains associated with the single ESB [34]. Thus, we decided to investigate the dynamic of the HSF throughout the cell cycle. Trypanosome cell cycle phases are clearly distinguishable because kinetoplast mitochondrial DNA (K) segregation occurs prior to the onset of mitosis and nuclear (N) division. Thus, DAPI staining identifies a population with 1K1N cells (G1 and G1-S) and 2K1N cells (G2). To analyze HSF dynamics throughout the cell cycle we performed double indirect 3D-IF in bloodstream form trypanosomes using antiserum against TbRPA1 and the mAb anti-TbSUMO. This analysis revealed that the HSF and the ESB partially colocalized in every stage of the cell cycle (Figure S3). Anti-TbSUMO labeling allowed us now to distinguish in the nucleus the ESB when is located closed of the nucelolus (Figure S3, G1 cell).
In pre-mitotic cells a single ESB remained associated to the VSG-ES sister chromatids during segregation [34]. Thus, we investigated a possible association of the HSF with the active VSG-ES chromatids during the cell cycle using the GFP-tagged VSG-ES cell line. Double 3D-IF analysis using anti-SUMO and anti-GFP antibodies showed a single HSF in the nucleus, which was associated with the active VSG-ES locus throughout the cell cycle. Interestingly, the two sister chromatids of the active VSG-ES in pre-mitotic cells were associated with a single HSF. Once cells enter into mitosis and sister chromatids are clearly separated, two HSFs associated with each chromatid were detected (Figure S4).
Nuclear localization analysis by 3D-IF analysis suggested that SUMO-conjugated proteins associate with the active VSG-ES telomeric locus in the bloodstream form (Figure 2B). Next, we decided to investigate in detail the occupancy of SUMOylated proteins along the VSG-ES locus by chromatin immunoprecipitation (ChIP) analysis and quantitative PCR (qPCR).
To overcome the problem of highly homologous sequences among different VSG-ES promoter regions [3], ChIP experiments were performed using two cell lines containing the Firefly-luciferase (FLuc) reporter gene inserted 400 bp downstream of the VSG-ES promoter in an active (SALR) [14], or inactive (SILR) transcriptional state. The SILR cell line contains the same Fluc cassette than the SARL cell line but downstream of a silent VSG-ES promoter (BES5, VSG800), as revealed by reporter activities and sequence analysis (see Supporting Information Text S1). In addition, to monitor a RNA pol II transcribed locus, the Renilla-luciferase (RLuc) reporter gene was inserted within the tubulin locus in both cell lines (Figure 3A). ChIP analysis using anti-TbRPA1 showed that the VSG-ES chromatin is highly enriched in RNA pol I in the active transcriptional state, in contrast to the inactive VSG-ES with immunoprecipitation levels close to the background (Figures 3B and 3C). The high enrichment of TbRPA1 at the active VSG-ES compared to inactive was demonstrated using the unique sequences (FLuc) inserted downstream of the promoter. TbRPA1 immunoprecipitated 42-fold higher at the FLuc the active VSG-ES (2.52% input) compared to FLuc in the inactive VSG-ES (0.06% input). TbRPA1 levels at additional unique sequences such as the pseudo-VSG and the telomeric VSG221 showed that the active VSG-ES chromatin is highly occupied by the TbRPA1 (Figures 3B and 3C). The differences of the TbRPA1 occupancy between the active and inactive VSG-ES sequences support a transcription initiation control as one of the mechanisms involved in VSG-ES monoallelic expression.
Next, we investigated the presence of chromatin-associated SUMOylated proteins within the VSG-ES locus by ChIP using the anti-TbSUMO mAb. Interestingly, we detected SUMOylated proteins enriched at the entire active VSG-ES transcription unit, from sequences downstream of the promoter to the telomeric VSG gene (Figures 3D and 3E). SUMOylated proteins were immunoprecipitated more efficiently at the reporter inserted downstream of the active VSG-ES promoter (FLuc SALR: 0.43% input-background) than at the inactive (FLuc SILR: 0.05% input-background) and the difference was statistically significant (p value<0.01). Similarly, the active VSG221 was significantly immunoprecitated while other telomeric VSG genes such as VSG121, VSGVO2 and VSGJS1, which include also basic copies, were very low, near to background levels (Figures 3D and 3E). Furthermore, significant SUMOylation level was not detected at other RNA pol I-transcribed loci (rDNA or EP procyclin), nor at other RNA pol II or RNA pol III loci analyzed (Figures 3D and 3E).
In other eukaryotes, SUMOylated proteins were detected at RNA pol II promoters and play important roles in their activity [36], [37]. Thus, we investigated the presence of SUMOylated proteins in the chromatin upstream of the VSG-ES promoters (Figure 4A). ChIP-qPCR analysis revealed a high enrichment of SUMOylated chromatin-associated proteins upstream of the promoter region, which was notably higher in the fragments 6 and 5 (1.5% and 1.6% input-background) (Figure 4B). As a negative control, we compared with fragment 7 upstream of the 50 bp repeats, which showed no significant enrichment (0,01% input) (Figure 4B). The trypanosome genome contains at least 15 different VSG-ESs with highly conserved sequences at the promoter region, suggesting that the primers used for ChIP qPCR may anneal on many different VSG-ESs. Relative quantification revealed that sequences 4, 3 and 1 were highly conserved in many VSG-ESs (Figure S5A), suggesting that SUMO ChIP values upstream of the promoter (fragments 2, 3 and 4 in Figure 4B), represented as percentage of input, were in fact underestimated. To confirm this hypothesis, we cloned and sequenced PCR fragments from the region 4 using ChIPed and genomic DNA as templates. Sequences obtained from genomic DNA yielded 14 different sequences including one from the VSG221-ES, indicating these PCR primers amplify most of the VSG-ESs in the genome. However, analysis of the anti-TbSUMO ChIPed fragments identified 11 sequences identical to the active VSG221-ES promoter region, and 7 sequences 99% homologous (Figure S5 B). These results together indicate that the chromatin upstream of the active VSG-ES promoter is highly enriched in SUMOylated proteins.
We also analyzed in detail chromatin SUMOylation along the rDNA promoter region, however no significant levels were detected in any of the positions analyzed, including the non-transcribed upstream spacer and the coding region for the 18S rRNA (Figures 4C and 4D).
We previously proposed that TbRPB7 functions in trypanosome RNA pol I transcription by recruiting transcription or RNA processing factors to the VSG-ES chromatin [14]. Thus, we searched for TbRPB7-interacting proteins by a yeast two-hybrid screen (Hybrigenics). This approach detected several putative interacting proteins, including a topoisomerase, a ubiquitin ligase and a protein with a SP-RING conserved domain characteristic of SUMO E3 ligases [38], which we named TbSIZ1 (Tb927.9.11070), an ortholog of yeast SIZ and mammalian PIAS. Sequence alignment of SP-RING domains of previously characterized SUMO E3 ligases revealed a significant conservation with TbSIZ1 (Figure 5A). We developed a mouse monoclonal antibody (7G9D4) anti-TbSIZ1 that allowed us to identify TbSIZ1 as a 72 kDa protein highly expressed in the infective bloodstream form of the parasite (Figure 5B).
Subcellular localization analysis detected TbSIZ1 mainly in numerous nuclear foci (Figure S6A), similar to the pattern described for other SUMO E3 ligases in other eukaryotes [39], [40]. Although TbSIZ1 was not enriched in a single nuclear area as the HSF, we investigate a possible colocalization with the active VSG-ES. Statistical analysis using the cell line with the GFP-tagged active VSG-ES showed that TbSIZ1 was associated with this locus in 87% of G1 cells (1K1N cells), while in G2 and pre-mitotic cells this percentage was reduced to 40% (2K1N cells) (Figure S6B). The lack of significant colocalization of TbSIZ1 with the rDNA locus suggests the interaction of TbSIZ1 and the active VSG-ES is not accidental (Figure S6B).
SUMO E3 ligases are important for the efficient transfer of a SUMO group from the conjugating enzyme E2 to specific substrates [22]. To characterize the function of TbSIZ1 we generated bloodstream form cell lines where depletion of TbSIZ1 was performed by RNA interference (RNAi). Western blot analysis confirmed TbSIZ1 depletion after 48 h of RNAi induction (Figure 5C), while only a minor effect in cell growth or cell cycle progression was detected (Figure S7A). However, at 72 h after induction of the TbSIZ1 RNAi the protein level increased suggesting TbSIZ1 depletion is partial and transitory (Figure 5C). Importantly, TbSIZ1 partial depletion reduced the signal of SUMO-conjugated proteins by Western blot analysis (Figure 5D), indicating that TbSIZ1 is a functional SUMO E3 ligase.
Consistently with Western analysis (Figure 5D), depletion of TbSIZ1 also reduced the nuclear signal of SUMO conjugates analyzed by IF using the anti-TbSUMO mAb (Figure S7B). Unfortunately, while the nuclear signal of TbSUMO was significantly reduced by TbSIZ1 depletion, we failed to completely eliminate the HSF signal in the nucleus. TbSIZ1 depletion functioned with variable penetrance in each cell, since the SUMO conjugated protein signal was reduced with different efficiency. SUMOylation is essential in T. brucei since SUMO depletion by RNAi induced deregulation of the cell cycle [29]. Notwithstanding, we decided to analyze the stability of the HSF signal in the nucleus upon TbSUMO RNAi. Similar to TbSIZ1 depletion, TbSUMO RNAi clearly reduced SUMOylation in the nucleus, however the HSF was weaker but still detected (Figure S7B).
TbRPB7 is required for VSG-ES transcription in vivo [14], and TbSIZ1 was identified as a TbRPB7-interacting protein, thus we decided to investigate a possible function of TbSIZ1 on VSG-ES chromatin SUMOylation, To do so, we performed a series of anti-TbSUMO ChIP experiments after 48 h of TbSIZ1 depletion. Upon TbSIZ1 knockdown, reduced levels of SUMO were detected at all positions along the active VSG-ES compared with the parental cell line (Figure 6A). The reduction of SUMOylated chromatin after TbSIZ1 partial depletion was particularly significant at the region upstream of the VSG-ES promoter, where the highest enrichment of SUMO conjugated proteins was detected (Figure 4B). As expected, TbSIZ1 knockdown induced no changes in chromatin SUMOylation in loci where SUMO was undetectable, such as the silent telomeric VSGs (Figure 6A). Altogether, these data suggest that the SUMO E3 ligase TbSIZ1 is responsible for the SUMOylation of chromatin-associated proteins detected in the active VSG-ES.
The detection of SUMOylated proteins associated specifically to the active VSG-ES chromatin, contrary to inactive VSG-ES, suggests a positive role of chromatin SUMOylation in transcription driven by RNA pol I in trypanosomes. To test this hypothesis, we analyzed the effect of reduced chromatin SUMOylation induced by TbSIZ1 depletion on TbRPA1 occupancy at the VSG-ES chromatin. ChIP values obtained using anti-TbRPA1 and chromatin isolated from TbSIZ1 depleted cells were compared with the values from the original cell line (Figure 6B). These experiments detected lower levels of TbRPA1 recruited to the active VSG-ES after TbSIZ1 depletion. The reduction of TbRPA1 occupancy extended from the promoter region to the telomeric VSG221 gene. We did not detect TbRPA1 changes in the silent telomeric VSGs. The RNA pol I recruited to the active VSG-ES after TbSIZ1 depletion was about 50% less in single copy genes (FLuc, Pseudo VSG & VSG221), suggesting SUMOylation of chromatin associated proteins is important to achieve full transcription of this locus.
To determine whether reduced levels of RNA pol I occupancy affect expression levels we performed RT-qPCR analysis in cells after 48 h of TbSIZ1 depletion. We detected reduced levels of the FLuc and VSG221 mRNAs, without a significant effect in RNA pol II-transcribed RLuc or myosin genes or in U2 transcribed by RNA pol III (Figure 6C). Importantly, we also analyzed rDNA transcription driven by RNA pol I, but no significant changes were detected in either the mature 18S or the pre-spliced rRNA780 RNAs (Figure 6C). These results suggest that TbSIZ1-mediated SUMOylation of chromatin-associated proteins positively regulates VSG-ES transcription.
To further investigate SUMOylation in VSG-ES expression and to rule out the possibility of a SUMO-ligase independent function for TbSIZ1, we decided to analyze the effect of inducing a global reduction in SUMO levels. Thus, we generated cell lines where depletion of either TbUBC9 (E2 conjugase) or TbSUMO was performed by RNAi. TbUBC9 depletion reduced the SUMO-conjugated proteins very efficiently as detected by Western blot analysis (Figure S8). Depletion of TbUBC9 protein levels was confirmed using a mouse antiserum we developed against recombinant TbUBC9 (Figure S8).
Cells depleted of SUMO-conjugated proteins by either TbSUMO or TbUBC9 knockdown analyzed by ChIP using anti-TbRPA1 showed a significant reduction in RNA pol I occupancy in the VSG-ES (Figure 7A). TbRPA1 recruitment was decreased along the entire VSG-ES locus without increasing the occupancy in the silent telomeric VSGs. The reduction of the RNA pol I occupancy upon TbSUMO or TbUBC9 depletion correlated well with a decrease of mRNA derived from the active VSG-ES, while ribosomal RNA levels were not reduced (Figure 7B). TbSUMO or TbUBC9 are essential genes, however their depletion has a greater effect on the VSG-ES mRNA levels as compared to RNA pol II-derived mRNAs (Figure 7B). These results altogether suggest that SUMOylation of chromatin-associated proteins is important for active VSG-ES expression.
Next, we performed quantitative Western blot analysis of VSG expression after SUMO depletion to investigate whether VSG protein levels were affected. The VSG221 expression level was analyzed using anti-BiP antibody as loading control from three independent TbSUMO RNAi clones (Figures 7C and 7D). Quantification of VSG221 expression relative to the parental cell line extracts suggested that VSG protein level was significantly downregulated upon SUMO depletion. The extent of reduction of VSG expression after SUMO RNAi was variable but consistent, suggesting SUMO functions positively in VSG expression.
The detection of the HSF in the nucleus (Figure 2) and the high occupancy of SUMOylated proteins at the VSG-ES chromatin (Figures 3 and 4) suggest that a large number of SUMOylated proteins occur at this site, similar to Protein Group SUMOylation described previously [41]. Identification of the SUMO-conjugated proteins in the HSF is beyond the scope of this work, however the largest subunit of the RNA polymerase I is an obvious candidate since it is SUMOylated in other eukaryotes [42]. To investigate a possible TbRPA1-SUMO conjugation, we performed IP assays utilizing anti-TbSUMO mAb and affinity-purified TbRPA1 antiserum under denaturing conditions, which preserve SUMO conjugation (see Supplementary Information Text S1). IP experiments revealed that TbRPA1 is SUMOylated as shown by Western analysis using anti-TbRPA1 on a SUMO IPed extract (Figure 8A). The reciprocal experiment using anti-TbSUMO antibody on anti-TbRPA1 IPed extract reproducibly detected TbRPA1-TbSUMO conjugates (Figure 8B). The low detection of SUMO-conjugated TbRPA1 is probably due to the large number of SUMOylated proteins and the small percentage of TbRPA1 that is SUMOylated, as occurs with other SUMO targets in eukaryotes [33]. This result suggests that under normal growth conditions a fraction of TbRPA1 is SUMOylated. While this may contribute to the SUMOylation we detected by ChIP on the VSG-ES chromatin, SUMOylation is also detected upstream of the promoter (Figure 4B), suggesting additional SUMOylated chromatin proteins occur at this region.
We addressed whether the fraction of SUMO-conjugated TbRPA1 is the one that resides in the extra-nucleolar body ESB. To do so, we used the Proximity Ligation Assay (PLA) (O-link Bioscience). This technique exploits the distance requirements of a PCR reaction by linking two primers to the two secondary antibodies. The PLA assays showed that the fraction of TbRPA1 that is SUMOylated resides at an extra-nucleolar site (Figure S9). This result, together with the IF colocalization analysis of the HSF with both the active VSG-ES locus and the TbRPA1 (Figures 2A and 2B) suggest that SUMOylated TbRPA1 occurs at the nuclear body ESB.
Finally, we wished to investigate whether SUMOylation of TbRPA1 is mediated by TbSIZ1. To characterize a possible function of TbSIZ1 in TbRPA1 SUMOylation we performed a series of co-IP experiments using protein extracts isolated from TbSIZ1 depleted cells and compared to the parental cell line. Figure 8C shows that TbSIZ1 depletion significantly reduced the amount of TbRPA1 IPed using anti-TbSUMO antibody. This result suggests that TbSIZ1 mediates SUMO targeting of TbRPA1.
The importance of nuclear bodies and the three-dimensional organization of chromosomes in the regulation of gene expression is becoming evident in eukaryotes [43]. In trypanosomes, it was suggested that the recruitment of a single Variant Surface Glycoprotein Expression Site (VSG-ES) telomeric locus to a discrete, RNA pol I-containing nuclear body (ESB) underlies the mechanism responsible for VSG monoallelic expression [4], [5], [44]. Here, by nuclear localization analysis using 3D microscopy, we describe a highly SUMOylated focus (HSF) (Figure 1C). The nuclear position of the HSF partially colocalizes with the active VSG-ES locus and the nuclear body ESB (Figure 2). Unfortunately, our attempts to completely eliminate the HSF in the nucleus by TbSUMO or TbSIZ1 RNAi were unsuccessful (Figure S7B). Previous evidence whereby SUMO modifies the interaction properties of conjugated proteins and affects their subnuclear localization [45] suggests that SUMOylation of nuclear proteins at the HSF might be involved in the nuclear body ESB regulation in trypanosomes.
SUMO-conjugated proteins are localized to the active VSG-ES chromatin, in contrast to any other loci examined (Figure 3). We have investigated the possibility that SUMOylated proteins associate with other loci transcribed by RNA pol I. SUMOylated chromatin was not detected at the rDNA or EP procyclin loci. These results suggest that the association of SUMO-conjugated proteins to chromatin is a distinct feature of the VSG-ES regulation.
SUMOylation of chromatin-associated proteins at the active VSG-ES extends from ∼1 Kb upstream of the promoter down to the telomeric VSG, while SUMO was not detected at silent VSG-ESs or VSG basic copies chromatin (Figures 3 and 4). Whilst SUMOylation has been classically associated with transcriptional repression [46]–[48], there is some evidence that SUMOylation can also function as a transcriptional activator, particularly to modify gene-specific transcription factors or co-regulators [28], [49]. In HeLa cells, SUMO-1 was found at the chromatin just upstream of the transcription start site on many of the most active genes [36]. Depletion of SUMO-1 resulted in down regulation of transcription supporting the idea that marking of promoters by SUMO-1 is associated with transcriptional activation [36]. PIAS E3 ligases function as enhancers of c-Myb activity in active nuclear RNA pol II foci [39]. In trypanosomes, the transcriptionally active VSG-ES promoter and the nuclear body ESB are identified here as being highly SUMOylated (Figures 1C and 2).
TbSIZ1 is the first SUMO E3 ligase functionally analyzed in T. brucei. It contains a conserved SP-RING domain essential for the ligase activity described previously in other eukaryotes [38]. TbSIZ1 depletion has a mild effect on cell growth and cell cycle progression (Figure S7A). This result is similar to other SUMO ligases, such as S. pombe Pli1 and S. cerevisiae Siz1 and Siz2, for which deletion does not affect cell growth [20], [40]. Interestingly, TbSIZ1 depletion reduced some SUMO-conjugated protein bands more efficiently than others analyzed by Western blot (Figure 5D). This supports the idea of specificity of TbSIZ1 substrates, similarly to the role of previously described SIZ/PIAS E3 ligases [21].
In the present work, we show that TbSIZ1 functions in vivo as a SUMO E3 ligase of chromatin-associated proteins detected at the active VSG-ES chromatin by ChIP. Depletion of TbSIZ1 causes reduction in SUMOylation of the active VSG-ES with a concomitant reduction in RNA pol I occupancy and transcriptional activity (Figure 6B and C). We ruled out the possibility of a SUMO ligase independent function of TbSIZ1 by TbSUMO or TbUCB9 RNAi experiments, which also reduced both RNA pol I recruitment and VSG-ES expression (Figure 7). This finding is similar to observations in yeast, where SUMOylation of chromatin-associated proteins in actively transcribed genes is dependent on the E2 conjugating enzyme Ubc9 [37]. Interestingly, we did not detect significant levels of SUMOylated chromatin in the other RNA pol I-driven control loci as rDNA, EP or silent VSG-ESs loci, suggesting that SUMO plays a distinct function in VSG-ES positive regulation.
We showed that the constitutive rDNA promoters have no detectable levels of SUMOylated chromatin (Figure 4D), contrary to the switchable VSG-ES promoter, which is highly SUMOylated only in the active transcriptional state. Activation of inducible promoters has been shown to result in chromatin SUMOylation, suggesting that gene activation involves SUMOylation of promoter-bound factors [37]. Our results suggest a function of SUMO in VSG-ES active transcription, since upon SUMO depletion by TbSIZ1, TbUBC9 or TbSUMO RNAi, both recruitment of the RNA pol I at the VSG-ES promoter and VSG-ES derived transcripts are reduced.
The finding that SUMOylation is important for VSG-ES expression suggests that factors previously implicated in VSG regulation maybe modified by SUMOylation. An obvious candidate as SUMO substrate is the RNA pol I complex, responsible for VSG transcription. In other eukaryotes, several subunits of the RNA pol I, including RPA1, were described to be SUMOylated in large scale proteomics analyses [42], [50]. Indeed, we find by IP experiments that TbRPA1 is SUMOylated (Figure 8). However, the high SUMO enrichment detected 1 Kb upstream of the VSG-ES promoter cannot be accounted for TbRPA1. Thus, SUMOylated proteins detected upstream of the VSG-ES promoter may include transcription factors or structural components of chromatin, similar to what has been described in other eukaryotes [47].
Simultaneous SUMOylation of Protein Groups by modification of multiple targets providing synergy in a specific process has been recently described for DNA repair [41], [51]. Proteomic studies have shown that several proteins in the same complexes or biochemical pathways are SUMOylated [42], [50]. Protein group SUMOylation may also be associated with a specific subnuclear localization of SUMOylated proteins [52]. The HSF is frequently larger than the ESB detected using anti-TbRPA1, suggesting that additional factors involved in processes previously associated with VSG expression, such as transcription elongation and mRNA maturation, may be present in the HSF [2]. Our data showed that RPA1 immunoprecipitated 42-fold higher at the active VSG-ES as compared to a single inactive VSG-ES, however we still detect some TbRPA1 at the inactive VSG-ES site (FLuc SILR: 0.061% input, before removing background). Possibly, this polymerase in the inactive VSG-ESs promoter region is enough to produce detectable mRNA described recently [53]. It seems likely that the HSF described in this work represents a group of post-transcriptionally modified proteins, as Protein Group SUMOylation, functionally associated with VSG-ES transcription initiation, elongation and mRNA maturation.
Some of the proteins involved in the regulation of antigenic variation and VSG-ES expression in trypanosomes have been previously described as SUMO targets in other eukaryotes. Among the possible chromatin-associated factors that could be SUMOylated at the active VSG-ES is the architectural chromatin protein TDP1, which was reported to be enriched at the active VSG-ES and rDNA, facilitating RNA pol I transcription [13]. The yeast ortholog Hmo1 has been recently identified by proteomic analysis as SUMO-conjugated protein [50]. Thus, it seems possible that the fraction of TDP1 at the ESB is SUMOylated in the HSF.
Recent data showed that Cohesin subunits Smc1/3 and Scc1/3 are SUMOylated in yeast [54]. In Trypanosoma cruzi SMC3 was also identified as a SUMOylated protein by proteomic approaches [31]. We have recently identified Cohesin complex as a factor involved in VSG-ES switching [34]. Preliminary results suggest that SMC3 is SUMOylated in T. brucei, however and contrary to TbRPA1, SUMO is not targeted to TbSMC3 by TbSIZ1 (manuscript in preparation).
ChIP experiments showed highly-enriched SUMOylated chromatin upstream of the VSG-ES promoter (Figure 4), suggesting structural components of chromatin might be also targets for SUMO at this particular location. In trypanosomes, post-translational histone modifications are being associated with repression of silent VSG-ESs (see for review [55]). Histone SUMOylation is associated with transcriptional repression in S. cerevisiae, where all four core histones are SUMOylated [47]. However we describe the lack of SUMOylated chromatin at silent VSG-ESs, while the active VSG-ES chromatin is highly enriched in SUMO. Our results show SUMO as a post-translational modification of proteins associated with the active transcriptional state of the VSG-ES.
Ever since the finding that the ESB is associated with VSG-ES monoallelic expression [4], we and others have searched for a particular factor located exclusively at this unique nuclear body. However, specific post-translational modifications of common factors may also account for this body. Here we report that TbRPA1, the largest subunit of RNA pol I, is SUMOylated by TbSIZ1 (Figure 8C), in addition IF colocalization analysis and PLA (Figures 2B and S8), strongly suggest that the fraction of SUMOylated TbRPA1 resides at the ESB rather than in the nucleolus. SUMO modification has been involved in the re-localization of transcriptional regulators to different subnuclear compartments [56] and stabilizes interactions between the functionally related proteins [41]. Taken together, our results suggest a model whereby SUMOylation of chromatin-associated proteins mediated by TbSIZ1 at the active VSG-ES locus may function to nucleate factors to the ESB.
The complex regulation of antigenic variation involves monoallelic transcription of a single VSG-ES out of a multiallelic gene family at any given time. Our results show a positive mechanism via SUMOylation that marks the active VSG-ES chromatin. In other eukaryotes, SUMOylation of transcription factors and chromatin proteins is a negative mark that represses gene expression in most cases. The surprising observation about the specificity of chromatin SUMOylation for the active transcription state in an early-branched eukaryote suggests that the post-translational modification of proteins by SUMO play a basic role in the positive regulation of transcription in eukaryotes. Chromatin SUMOylation as an epigenetic mark for the monoallelically expressed VSG-ES could apply more widely to the regulation of antigenic variation in other protozoan parasites [57], [58].
T. brucei bloodstream form (Lister 427, antigenic type MiTat 1.2, clone 221a) and 427 procyclic form were used in this study. The dual-reporter SALR cell line in the bloodstream single-marker cell line was previously described [14], [59]. SALR dual reporter cell line contains a Firefly Luciferase (FLuc)-reporter integrated 405 bp downstream of the active ES promoter, and Renilla-Luciferase reporter integrated in the tubulin locus. The generation of dual-reporter SILR cell line was similar to the SALR cell line, but the same construct containing the Fluc gene was integrated downstream of an inactive ES promoter (see Supporting Information Text S1). The insertion site was identified by PCR and sequencing of the flanking luciferase region from SALR and SILR genomic DNA confirming FLuc is inserted in the active VSG221-ES in SALR, and in SILR downstream of the inactive VSG-ES promoter BES5/TAR98 VSG800/427-18 [3]. The VSG221-ES and rDNA GFP-LacI tagged cell lines have been previously described [4], [35]. The cell line expressing a YFP-TbRPB5z fusion was described before [35].
N-terminal fragment of TbSIZ1 (Tb927.9.11070), full-length of TbSUMO (Tb927.5.3210) and TbUBC9 (Tb927.2.2460) were amplified by PCR (See primers in Table S1). PCR products were cloned into BamHI and HindIII sites of pET28a vector (Novagen), and expressed as an N-terminal His tag. Purification of recombinant proteins was performed using NI Sepharose Fast Flow 6 (GE Healthcare). Purified recombinant proteins were inoculated in mice and used to generate anti-TbSIZ1 (7G9B4) and anti-TbSUMO (1C9H8) monoclonal antibodies (mAb), using standard procedures. Hybridomas were first screened against the recombinant proteins by ELISA and later confirmed by western blot analysis using trypanosome protein extracts since recognized a single protein of the expected size. Hybridomas 7G9B4 and 1C9H8 cell lines were grown as ascites. Anti-TbUBC9 mouse antiserum was generated using purified recombinant his-tagged TbUBC9 as antigen using standard procedures.
Mouse anti-TbSUMO (1C9H8) mAb (1∶2000), rabbit anti-TbRPA1 affinity-purified antiserum (1∶600) [4] and rabbit anti-GFP polyclonal (1∶5000; Invitrogen) were used as primary antibodies. Goat anti-mouse and anti-rabbit Alexa 488 or 594 conjugated antibodies (Invitrogen) were used as secondary antibodies. Detailed subcellular localization and colocalization analysis was performed by deconvolution 3D microscopy as described previously [35] (see Supporting Information Text S1).
ChIP was performed as described previously [60] with some modifications. In brief, T. brucei bloodstream cultures were fixed in 1% formaldehyde at 37° for 15 min. Pellets were resuspended in 1 ml of lysis buffer per 108 cells and sonicated to shear the chromatin to ∼300pb in length. Sheared chromatin was diluted 1∶5 in ChIP dilution buffer and pre-cleared with Sepharose 4B beads (Sigma). An aliquot of the input DNA (10%) was saved. 2.5 ml of pre-cleared chromatin (5×107 cells per ChIP) was incubated overnight at 4°C with each antibody (6 µg of anti-TbRPA1, 60 µg of anti-TbSUMO 1C9H8, 60 µg of unspecific antiserum). Next, protein G Sepharose (Sigma) was added and incubated for 1 hr at 4°C; Immunoprecipitates were washed and eluted from the beads. Crosslinks were reversed at 65°C for 15 h. After RNase and Proteinase K treatment, DNA was extracted with phenol∶chloroform and ethanol precipitated. DNA was resuspended in 50 µl of miliQ water and analyzed by quantitative PCR (qPCR). To compare the amount of DNA immunoprecipitated to the total input DNA, 10% of the pre-cleared chromatin saved as input was processed with the eluted immunoprecipitates beginning at the crosslink reversal step. Quantitative PCR (qPCR) was performed using the SYBR green supermix (Quanta Biosciences) in a CFX96 cycler (BioRad), as described below for RT-qPCR. qPCR mixtures contained 2 µl of a 1∶5 dilution of the ChIPed DNA or a 1∶50, 1∶100, 1∶200 dilution of the input sample and 500 nM of each primer in a final reaction volume of 10 µl. All reactions were performed in duplicate and each product was verified by melting curve analysis. The PCR primers used to analyze target fragments were designed by using the Primer3 software and synthetized by Sigma, targets and sequences are listed in Table S1. Standard curves with serial dilutions of input DNA were made to determine PCR efficiency and to determine IP percentages. The relative amount of each specific PCR fragment in the ChIPed DNA and in the input DNA was calculated against the standard curve equation, next the percentage of input immunoprecipitaded was calculated. Finally the background values from unspecific antiserum (pre-bleed rabbit antiserum ChIP) were subtracted from the values obtained with the specific antibodies. Fold values were determined using the percentage of input immunoprecipitated before the background correction, since the background values were very similar between the loci to compare. Independent ChIP experiments were performed at least three times and statistical analysis (Student's t-test) was applied to compare data sets. See supplementary information (Text S1) for more details and primer sequences are provided in the Table S1.
RNAi constructs were made using the p2T7Bla vector [14], which allows Tet-inducible expression of dsRNA from opposite T7 promoters [61]. Since most of the RNAi constructs using this vector were leaky, comparative analyses always included in addition of the dox induced (+) and uninduced (−) RNAi, the parental cell line (SALR) [14]. Fragments corresponding to 642-pb of TbSIZ1 gene, full length of TbSUMO ORF gene (345-pb) and 5′UTR TbSUMO fragment (121 bp) and full length TbUBC9 gene (594-pb) were amplified by PCR using primers described in Table S1 and cloned into BamHI and HindIII sites of p2T7Bla. The constructs were linearized and stably transfected into the dual-reporter cell line SALR [14]. dsRNA synthesis was induced by the addition of µg ml−1 of doxycycline. At least three independent clones from each construction were analyzed and depletion of the proteins was confirmed by Western blot using specific antibodies.
Total RNA samples were extracted from 40 ml parasite cultures (4×107 cells) using the High Pure RNA isolation Kit (Roche) and treated with integrated DNA digestion and DNase removal following manufacturer's instructions. RNA quality was verified by gel analysis, nanodrop quantification and A260/A280 ratio. cDNA was synthesized from 2 µg of RNA with the SuperScript IIII Reverse Transcriptase (Invitrogene) and random primers (Invitrogene) following manufacturer's instructions. RNA samples not treated with reverse transcriptase were used as a negative RT control and analyzed by quantitative PCR for DNA contamination assessment. Quantitative PCR was performed using the SYBR green supermix (Quanta Biosciences) in a CFX96 cycler (BioRad), using 96-well clear low profile plates, sealed with clear optical adhesive covers. PCR mixtures contained 5 µl of 2× SYBR green supermix, 500 nM of each primer and 1 µl of cDNA for single copy genes or 1 µl of a 1∶100 dilution for multicopy genes in a final reaction volume of 10 µl. All reactions were performed in duplicate and each product was verified by melting curve analysis. The PCR protocol used was 95°C for 3 min followed by 32 cycles of 95°C for 30 sec, 60°C for 30 sec, 72°C for 30 sec, then 72°C for 1 min and final melting curve from 55 to 90°C, increment 0.5°C/5 sec. Fluorescence readings were taken during the extension step. PCR primers were designed by using the Primer3 software and synthetized by Sigma. Primer targets and sequences are listed in Table S1. Standard curves for each primer pair were generated with serial dilutions of cDNA to determine PCR efficiency. The relative levels of gene expression between a given sample and the control sample (Parental cell line) were calculated using the ΔΔCT method with the Bio-Rad CFX Manager software. The U2 gene transcribed by RNA pol III was used as reference gene to normalize RNA starting quantity since it was stably expressed, invariant expression was confirmed using Myosin B or Renilla-luciferase (RLuc) as reference genes. Three RNAi independent clones were analyzed and statistical analysis (Student's t-Test) using SigmaPlot software was performed.
See the Supplemental Material and Methods (Text S1) in Supporting Information for additional protocols.
|
10.1371/journal.pgen.1007818 | Interactions between the mRNA and Rps3/uS3 at the entry tunnel of the ribosomal small subunit are important for no-go decay | No-go Decay (NGD) is a process that has evolved to deal with stalled ribosomes resulting from structural blocks or aberrant mRNAs. The process is distinguished by an endonucleolytic cleavage prior to degradation of the transcript. While many of the details of the pathway have been described, the identity of the endonuclease remains unknown. Here we identify residues of the small subunit ribosomal protein Rps3 that are important for NGD by affecting the cleavage reaction. Mutation of residues within the ribosomal entry tunnel that contact the incoming mRNA leads to significantly reduced accumulation of cleavage products, independent of the type of stall sequence, and renders cells sensitive to damaging agents thought to trigger NGD. These phenotypes are distinct from those seen in combination with other NGD factors, suggesting a separate role for Rps3 in NGD. Conversely, ribosomal proteins ubiquitination is not affected by rps3 mutations, indicating that upstream ribosome quality control (RQC) events are not dependent on these residues. Together, these results suggest that Rps3 is important for quality control on the ribosome and strongly supports the notion that the ribosome itself plays a central role in the endonucleolytic cleavage reaction during NGD.
| In all organisms, optimum cellular fitness depends on the ability of cells to recognize and degrade aberrant molecules. Messenger RNA is subject to alterations and, as a result, often presents roadblocks for the translating ribosomes. It is not surprising, then, that organisms evolved pathways to resolve these valuable stuck ribosomes. In eukaryotes, this process is called no-go decay (NGD) because it is coupled with decay of mRNAs that are associated with ribosomes that do not ‘go’. This decay process initiates with cleavage of the mRNA near the stall site, but some important details about this reaction are lacking. Here, we show that the ribosome itself is very central to the cleavage reaction. In particular, we identified a pair of residues of a ribosomal protein to be important for cleavage efficiency. These observations are consistent with prior structural studies showing that the residues make intimate contacts with the incoming mRNA in the entry tunnel. Altogether our data provide important clues about this quality-control pathway and suggest that the endonuclease not only recognizes stalled ribosomes but may have coevolved with the translation machinery to take advantage of certain residues of the ribosome to fulfill its function.
| The elongation phase of translation is an imperfect process, during which the ribosome moves with irregular speed along the mRNA template [1]. By and large the elongation speed is determined by sequence and structural features of the coding sequence. For instance, the identity of the A-site codon is known to have a drastic effect on the rate of protein synthesis depending on the availability of its partner tRNA and the nature of the codon-anticodon base-pairing interaction [2, 3]. Furthermore, the chemical characteristics of the locally-encoded amino acids have been shown to regulate the rate of protein synthesis based on the manner they interact with the exit tunnel of the ribosome [4]. mRNAs are also known to harbor local secondary structures that can slow down the ribosome as it unwinds them [5, 6]. Regardless of the underlying mechanism, the fluctuating rate of protein synthesis along an mRNA molecule appears to serve important biological functions such as promoting appropriate co-translational protein folding and ensuring that the encoded protein is targeted to the correct destination in the cell [7–12].
In contrast to this “programmed” regulation of ribosome traffic, the ribosome often encounters unwanted obstacles that severely hinder its progression and in some cases stall protein synthesis all together [13, 14]. Most of these impediments are typically associated with defects in the mRNA, including stable secondary structures, stretches of rare and inhibitory codons, as well as truncations and chemical damage [3, 15–17], [18]. Because multiple ribosomes are typically translating a single mRNA at any given point, one stalled ribosome is likely to impede the progression of multiple upstream ribosomes. As a result, if left unresolved, these stalling events have the potential to severely reduce cellular fitness. Notably, the stalling of the ribosome itself is not such a detriment to the cell as is the loss of valuable ribosomes from the translating net pool [13, 14]. In eukaryotes, the evolutionary solution to this predicament was No-Go Decay (NGD) [15] as a means to dissociate stalled ribosomes [19–21]. It is thought that over time, this mechanism was expanded on to include mRNA surveillance to dispose of the aberrant mRNA. In particular, the mRNA undergoes an endonucleolytic cleavage upstream of the stall site. The resulting deadenylated 5’-end and uncapped 3’-end pieces are then rapidly degraded by the exosome and Xrn1, respectively [3, 15–17].
Initial studies on NGD in yeast focused on the two factors Dom34 (Pelota in mammals) and Hbs1 [15, 22, 23]. These factors are homologs of the termination factors eRF1 and eRF3, respectively. Early reports of NGD hinted at a role for the factors in mediating the endonucleolytic cleavage of the mRNA near the stalled ribosome [15, 24]. However, later studies by the same group and others showed the cleavage to take place in the absence of the factors [22] leaving the question of the role of the factors in the process unanswered. Interestingly prior to the discovery of NGD, genetics studies suggested that Dom34 and Hbs1 are important in maintaining ribosome homeostasis of the cell [25]. To this end, both factors become essential or near-essential when ribosomes are depleted either by knocking down certain ribosomal proteins or under conditions when ribosomes are sequestered [25–27]. These observations are consistent with biochemical studies using a yeast translation reconstituted system, which showed the factors to be responsible for dissociating ribosomes into their respective subunits [18]. This splitting activity of Dom34-Hbs1 was also found to be much more efficient in the presence of Rli1 (ABCE1 in mammals) [20, 21]. In vivo data also supported this model for the role of the three factors in dissociating ribosomes [16]. Hence, this rescuing/recycling activity of these factors rationalizes the effect of their deletion on ribosome availability, especially under stress conditions.
In addition to ribosome rescue and degradation of the aberrant RNA, NGD is closely linked to a newly discovered protein-quality-control process termed ribosome quality control (RQC). This process is responsible for degrading the incomplete nascent protein resulting from stalled translation [28–34]. RQC proceeds after the splitting action of Dom34-Hbs1-Rli1, which results in a peptidyl-tRNA-associated large-ribosome subunit. This atypical form of the 60S subunit is recognized by the E3 ligase Ltn1 (Listerin in mammals) alongside Rqc2 (formerly Tae2) [30, 33, 35]. Ltn1 ligates ubiquitin chains to the nascent peptide as it is attached to the tRNA on the large subunit. The ubiquitinated nascent peptide is then extracted and delivered to the proteasome for degradation through the action of Rqc2 and Cdc48 (and its adaptor proteins Ufd1 and Npl4). Two additional factors, the ribosome-associated Asc1 and the E3 ligase Hel2 (Rack1 and Znf598 in mammals, respectively), also appear to be important for proper RQC function. Both factors are important for ribosomal protein ubiquitination and appear to play a role during stalling [36, 37]. In particular, deletion of either factor results in increased readthrough of stall sequences [38, 39]. How regulatory ribosomal protein ubiquitination interconnects with RQC and NGD is currently poorly understood.
Even though the consequences of ribosome stalling in eukaryotes was initially described in the context of its impact on mRNA steady state levels [15], as detailed above we know far more about its entanglement with ribosome rescue and quality control of the associated nascent peptide. More specifically, degradation of the mRNA is initiated by endonucleolytic cleavage, but the identity of the endonuclease remains elusive. This in turn has precluded further critical mechanistic dissections of NGD. Some of these outstanding important questions are: 1) How does the endonuclease recognize stalled ribosomes? 2) Is it associated with the ribosome? 3) Does it have a specificity for certain mRNAs 4) How is its function activated? 5) Can NGD be used to regulate gene expression? Work from our group recently provided some clues about the cleavage reaction. Using reporters and genetic manipulation of yeast we showed that the physical act of ribosome collision is important for initiating the process of RNA degradation and ribosome rescue during no-go decay (NGD) [40]. High-resolution mapping of the cleavage products also provided some important clues about the potential role of the ribosome in the reaction. Namely, cleavage appears to take place well upstream of the lead stalled ribosome with the closest most prominent one being ~45 nt upstream of the stall site. As ribosomes are likely to be stacked on the mRNA, this suggested the possibility that the cleavage is taking place inside the ribosome [18, 41]. Multiple regions of the ribosome make intimate contact with the mRNA. Most noteworthy among these is the mRNA entry tunnel, which encompasses residues of the ribosomal proteins Rps3/uS3 and Rps2/uS5 [42]. In eukaryotes additional contacts are made by helices 18 and 14 of the 18S rRNA, whereas in bacteria these contacts are carried out by Rps4/uS4 (orthologous to Rps9 in yeast and humans) [42–44]. In the entry tunnel, Rps3’s contacts with the mRNA stand out because they appear to be almost universally conserved and form an integral part of the helicase domain of the ribosome [42]. Furthermore, the protein has been implicated in translation initiation during the rearrangement of the small subunit that allows for the opening of the ribosomal mRNA binding channel and subsequent scanning of the mRNA [45] as well as start-codon selection [46].
Here we show the entry tunnel of the ribosome to play an important role during NGD. Mutation of the residues of RPS3 that form part of the entry tunnel, which have also been implicated in the helicase activity of the ribosome, were found to significantly reduce the accumulation of cleavage products. This effect on cleavage efficiency to a large extent was independent of the identity of the stall site. Combining these mutations with factors involved in other aspects of NGD revealed that the entry tunnel is also likely to be important in ribosome rescue. Our findings provide new insights into how quality control mechanisms evolved to integrate into fundamental biological machines.
To address a potential role for Rps3 in the cleavage reaction, we introduced a number of mutations to the protein and assessed their effect on cleavage of stalling reporters. Our choice of residues for the mutations was motivated by three criteria: they had to be conserved, made intimate contacts with the mRNA and have basic or acidic side chains (Fig 1A and 1B). This led us to Arg116 (R116) and Arg117 (R117). In addition to these, we also analyzed two residues that have been suggested to be important for Rps3’s extra-ribosomal activity in DNA repair [47–51], Asp154 (D154) and Lys200 (K200). Mutation of these residues abolishes the 8-oxoguanosine glycoslase and AP/endonuclease activities of the protein [51]. The variant-yeast strains were generated by introducing mutations to the chromosomal copy of RPS3 (see Methods) in different backgrounds of deletions and mutations. All in all, we generated the following mutants: Arg116 and Arg117 were substituted by Ala residues (R116A/R117A), Asp154 was substituted by an Ala residue (D154A), Lys200 was substituted by an Asn residue (K200N) and finally we generated a double mutant D154A/K200N. Of these the R116A/R117A mutation was notable as the side chain of these residues are projected into the entry tunnel of the ribosome and make electrostatic interactions with the mRNA (Fig 1B).
Next, we assessed the effect of these mutations on the cleavage of NGD substrates. We initially used an NGD reporter, which harbors a stable stem loop in the PGK1 coding sequence and was originally designed by Parker and colleagues. The stem loop presents a robust obstacle for the ribosome and is subject to an endonucleolytic cleavage as evidenced by the accumulation of 5’ and 3’ fragments when the exosome and Xrn1 are inactivated, respectively [15]. Indeed, similar to what was observed by us and others [15, 16, 40], in the ski2Δ strain- which is defective for 3’-5’ mRNA degradation- northern analysis of cells expressing PGK1-SL revealed substantial accumulation of 5’-fragments (Fig 1C). The D154A and K200N mutations in RPS3, which have been suggested to be important for an AP endonuclease activity [51], had no observable effect on the cleavage efficiency and appear to play no role in NGD. In contrast, the R116A/R117A mutations appear to reduce the accumulation of cleavage fragments and also increased heterogeneity among these products (Fig 1C). Interestingly, the mutations also appear to affect the steady-state levels of endogenous PGK1 transcript (Fig 1C). Regardless, these observations suggest that residues of Rps3 that interact with the mRNA in the entry tunnel are important during NGD.
The effects of the R116A/R117A mutations on the cleavage reaction were further studied in the context of other deletions that alter different aspects of NGD. Namely, we introduced these mutations into dom34Δ and xrn1Δ strains in addition to the wild-type parent strain. As expected, expression of the PGK1-SL in these strains does not result in the accumulation of 5’-fragments and the R116A/R117A mutations have no effect. As a control, these fragments were seen in the ski2Δ background and the rps3 mutations significantly reduced their levels (Fig 1D). Production of the 3’-fragments, as expected, was seen in the absence of XRN1 and their levels diminished in the presence of the RPS3 mutations, albeit to a lower extent than that seen for the 5’ fragments (Fig 1E). These latter observations suggested that the R116A/R117A mutations do not completely inhibit cleavage and that they may affect other aspects of NGD.
To provide further support for a role for the entry tunnel residues of Rps3 during NGD, we next examined the effect of the mutations on the stability of the PGK1-SL mRNA. Our reporters are expressed under the control of the GAL1 promoter, and as a result transcriptional-shutoff by shifting cells to glucose-containing media was used to measure the decay rate of the reporter mRNAs. As a control, we initially measured the decay rate of a non-NGD reporter (PGK1), which does not harbor any stalling sequence. The mutations were found to have little effect on the decay rate of the PGK1 mRNA reporter (Fig 2A); we measured half-lives of 28 ± 1.9 and 26 ± 4.4 minutes in the WT and the RPS3-mutant stains, respectively. As expected, the PGK1-SL mRNA decays with a faster rate relative to its PGK1 parent (Fig 2B). Its half-life of 4.7 ± 0.2 minutes is similar to previously published reports [15]. Here the RPS3 mutations result in a moderate but reproducible increase in reporter half-life to 6.0 ± 1.3 minutes, suggesting greater stabilization of the PGK1-SL mRNA (Fig 2B). Hence, these findings add additional support for the entry tunnel of the ribosome playing a role in mRNA-surveillance during NGD, whereby loss of interactions with the mRNA leads to stabilization of mRNAs harboring stalls.
So far, our analysis has focused on one type of stall—a stable RNA secondary structure in the form of a stem loop. Since the mutations under investigation here are important for the helicase function of the ribosome, any effect we saw on the cleavage reaction could be explained by defects in the unwinding activity of the ribosome and not in NGD. To rule out this potential explanation, we used two other reporters that had 12 stretches of the inhibitory arginine CGA or lysine AAA codons. Both are known to efficiently block translation and are not predicted to form secondary structures [3, 15, 23]. These new reporters were introduced to wild-type or mutant RPS3 yeast strains in the ski2Δ background. As expected, the CGA and AAA reporters accumulated 5’-fragments in the wild-type RPS3 strain, whereas the control UUU reporter did not (Fig 3). Similar to what we observed for the SL reporter, the R116A/R117A mutations significantly reduced the 5’-fragments levels for the CGA and AAA reporters, suggesting that the entry tunnel residues affect the accumulation of cleavage fragments independent of the type of stall (Fig 3). Interestingly, however, unlike the SL reporter, for which we observe an almost complete loss of cleavage products when RPS3 was mutated, cleavage fragments resulting from the CGA and AAA reporters were still visible but instead were heterogeneous in nature (Fig 3). This also made it difficult to perform any meaningful quantification. This is likely due to cleavage fragments produced by inefficient initial cleavage reactions, which lead to ribosome queuing upstream of the lead stalled ribosome.
Ski7, a component of the exosome in yeast, has been implicated in non-stop decay (NSD) [52–54]; given the similarities between NSD and NGD, the mutations in RPS3 could potentially affect the function of Ski7. To address this possibility, we deleted SKI7 from the wild-type, dom34Δ and ski2Δ stains in the absence and presence of the RPS3 mutations and assessed its effect on NGD efficiency from the SL reporter. We observed no significant changes to the accumulation of the 5’-fragments due to the SKI7 deletion suggesting that the entry tunnel residues do not affect the function of the factor (S1 Fig).
As mentioned earlier, in addition to Rps3, the mRNA entry tunnel of the small subunit also encompasses conserved residues of the ribosomal protein Rps2 [42, 55]. Namely the side-chain of Glu120 of the yeast protein protrudes into the entry tunnel and is likely to interact with the mRNA downstream of the A site (S1 Fig). Consequently, we determined whether this residue contributes to NGD or not. We mutated Glu120 to Ala in the ski2Δ strain and evaluated its effect on NGD cleavage efficiency. In contrast to the RPS3 mutations, the RPS2 mutation had no noticeable effect on the cleavage reaction; we observed comparable levels of 5’-fragments accumulation from the SL reporters in the RPS2 wild-type and mutant strains (S1 Fig). It thus appears that the changes to NGD we observe in the presence of the RPS3 mutations are the result of Rps3-dependent effects, and likely not from general alterations to the mRNA-entry tunnel.
Initial reports of NGD suggested that Dom34 plays a role in the cleavage reaction due to the loss of the cleavage products accumulation when the factor is deleted [15, 24]. Later studies, however, showed that the protein together with Hbs1 and ABCE1 dissociates stalled ribosomes [19]. In its absence ribosomes pile up on the mRNA leading to multiple cleavage events upstream of the lead stalled ribosome, which run as a long smear on a gel that appears to result in loss of cleavage efficiency [16]. Furthermore, overexpression of certain ribosomal proteins restored cleavage in the absence of DOM34, suggesting that the protein is involved in maintaining ribosome homeostasis [22]. To gain further insights into the role of the entry-tunnel residues in ribosome rescue, we deleted DOM34 from our RPS3-mutant strains and assessed its effect on the accumulation of 5’-fragments from the PGK1-SL reporter. As had been seen by others, deletion of DOM34 appeared to result in a loss of cleavage [16]. Interestingly the same deletion in the presence of the R116A/R117A mutations appears to restore cleavage with one caveat; the fragments are much more heterogeneous relative to those observed under normal conditions (Fig 4A). In particular, the products were observed to form a long smear on agarose gels. It seems that, under conditions where ribosome rescue is inhibited, mutation of the entry tunnel residues leads to a spreading of cleavage events well upstream of the stall site.
To provide further support for this notion, we examined the effect of mutations in ASC1 on cleavage in conjunction with the RPS3 mutations. Asc1 is a ribosome-associated protein that has been implicated in multiple aspects of ribosome quality control processes including NGD [38, 56–58]. For instance, cryoEM structures of a Dom34-Hbs1-bound ribosome revealed the factor to interact with Dom34 suggesting that it is critical for NGD [59, 60]. In addition, recent data from the Inada group showed that the factor is important for sequential endonucleolytic cleavage during non-stop decay (NSD) in the absence of DOM34 [58]. Instead of deleting ASC1- which harbors a snoRNA gene in its intron- from our rps3 strains, we opted to introduce the R38D/K40E mutations into the chromosomal copy of the gene. These mutations are known to affect the association of the factor with the ribosome and phenocopy its deletion in NGD [61]. Similar to the effect we saw in the dom34Δ background, the ASC1 mutations resulted in the accumulation of heterogeneous 5’-fragments from the PGK1-SL NGD substrate in the presence of the R116A/R117A mutations (Fig 4B). To verify that the effect on NGD we observe with the RPS3 mutants are not due to decreased association of Asc1 with the ribosome, we carried out polysome analysis and used western analysis to look at the binding of Asc1 to ribosomes. As can be seen in Fig 4C, ribosomal occupancy by wild-type Asc1 is not significantly altered by the mutations in RPS3; similar to the wild-type, the protein was found to primarily associate with the polysomes in the presence of the RPS3 mutation (top panels). As a control, the R38D/K40E mutant was observed in the light fractions of the sucrose gradient, that is not ribosome-associated, regardless of RPS3 status (bottom panels). We should note, though, Asc1 participates in a multitude of processes on the ribosome including translation of short ORFs, stall clearance and ribosomal protein ubiquitination [37, 38, 56–58, 62]. As a result, any interpretation of its consequence on NGD is likely to be complicated by the larger context of its effect on ribosome function.
How inhibition of ribosome rescue either by deletion of DOM34 or mutation of ASC1 restores cleavage efficiency to entry-tunnel mutants, albeit with a distinct signature of heterogeneous product accumulation, is difficult to interpret. One plausible explanation is that the R116A/R117A mutations inhibit the accumulation of cleavage fragments and under normal conditions ribosome rescue is fast enough to dissociate stalled ribosomes, which results in the observed disappearance of cleavage products. When rescue slows down due to reduced cleavage kinetics, ribosomes accumulate on the mRNA, initiating cleavage further upstream of the stall sequence.
Our Northern analysis of the NGD-cleavage products suggested that the R116A/R117A mutations affect cleavage fragments accumulation and result in ribosome queueing upstream of the stall site. This pile-up of ribosomes, in turn, results in cleavage reactions even farther upstream leading to diffusion of the NGD intermediates. We provided further support for these ideas by conducting high-throughput sequencing to map the 3’-end of the 5’-NGD fragments. Briefly, total RNA was isolated from strains harboring either the RPS3 mutants, dom34Δ, or ASC1 mutants in the ski2Δ background, each expressing one of the three NGD reporters- SL, (CGA)12 and (AAA)12. An adenylated DNA oligonucleotide was ligated to the 3’-end of the RNA samples, which was used to prime reverse transcription. The resulting cDNA was then amplified using a PGK1-specific 5’-primer and subjected to high-throughput sequencing using the Illumina Hiseq 2500 platform (GEO accession: GSE117652). Similar to what we have reported earlier [40], for otherwise wild-type cells, the 5’-fragments resulting from the PGK1-SL reporter mapped well upstream of the stall in all strains regardless of the mutational background (Fig 5). However, mapping of the fragments from the R116A/R117A mutant cells revealed extensive spreading of the cleavage events (Fig 5B). More specifically, whereas in the wild-type RPS3 background we observe one predominant peak near the ~150-nt upstream mark, in the rps3 mutant background, no predominant peak was observed (Fig 5B). Instead, fragments mapped throughout a 500-nt region upstream of the stall site and multiple peaks were observed with a near 30-nt periodicity. Interestingly, in the dom34Δ and the asc1 cells, the fragments displayed distinct mapping patterns relative to the wild-type and rps3 cells as well as to each other. Similar to what was observed for the rps3 mutant cells, in the dom34Δ cells the predominant peak at ~150-nt is lost, but here the distance between the peaks increased to 40–60 nt (Fig 5C). This is consistent with the role of Dom34 in rescuing ribosomes that run to the end of the transcript following endonucleolytic cleavage on NGD reporters. Since multiple ribosomes appear to be required for efficient cleavage, the reaction would be expected to occur every ~45-nt- with the lead ribosome protecting 15-nt, while the one behind protects 30-nt. In clear distinction to both the rps3 and the dom34Δ cells, mapping of the 5’-fragments from the SL reporter was not as diffuse in the asc1 mutant cells. Instead, only one additional predominant peak (relative to the wild-type cells) was observed at ~250 nt upstream (Fig 5D). Differences in cleavage patterns from the WT, rps3 and dom34Δ cells were also evident for 5’-fragments obtained from the (CGA)12 reporter, and to a lesser extent for (AAA)12 reporter (S2 Fig). We note that for both the (CGA)12 and (AAA)12 reporters, fewer reads were mapped in the rps3 cells, presumably due to decreased cleavage efficiency. These differences between the R116A/R117A mutant, and the DOM34 and ASC1 mutants suggest that the entry tunnel of Rps3 affects different aspects of NGD relative to these factors. It is also consistent with our model that these residues are important for the endonuclease function.
Recently we showed that ribosome collision appears to play an important role in initiating NGD during stalling [17]. In particular, decreasing ribosome concentration, and hence ribosome density per mRNA, by deleting certain ribosomal protein paralogues was found to reduce cleavage of NGD targets [40]. As a result, we wondered whether the mutations of the entry tunnel residues had similar effects on ribosome density. To address this potential explanation, we compared the polysome profile of the rps3 cells to the wild-type ones. Our analysis revealed that the mutations in RPS3 had little effect on ribosome density (Fig 6). The ratio of polysomes to monosomes in the mutant is largely similar to that observed in the wild-type background. In contrast, similar analysis of the dom34Δ cells- as has been seen before [27]- revealed elevated levels of 80S monosomes relative to polysomes (Fig 6). The finding that the RPS3 mutations do not seem to affect ribosome density has two immediate ramifications: 1) the observed inhibition of NGD in the presence of these mutations does not result from changes to ribosome collisions; 2) consistent with our mapping analysis, the mutations are not likely affecting the function of Dom34.
As discussed earlier, ubiquitination of ribosomal proteins by Hel2 (Znf598 in humans) has recently been recognized as an important feature of ribosome stalling. This modification promotes stalling on inhibitory codons as deletion of HEL2 results in significant bypassing of stalls by the ribosome [36–38, 63, 64]. Relevant to our studies is the observation that Rps3 is one of the targets for Hel2-mediated ubiquitination on K212, but it is currently unclear if its modification is important for stalling [64]. Nevertheless, if the entry tunnel mutations somehow affect Hel2 function, this could in principle explain their effect on NGD. As a result, we set out to assess stalling-induced ribosomal protein ubiquitination in the presence of R116A/R117A mutations. We took advantage of our previous observation that the addition of cycloheximide to an intermediary concentration, whereby ribosome collisions presumably occur at a global level, results in robust ribosomal protein ubiquitination [62]. We added cycloheximide to a final concentration of 2 μg/mL to wild-type, rps3 R116A/R117A, dom34Δ and double mutant cells; and isolated ribosomes. Ubiquitination patterns of ribosomal proteins resulting from cycloheximide addition, as assessed by western-blotting, was nearly identical among all strains (Fig 7). However, we noted that deletion of DOM34 had a discernible effect on the ubiquitination levels suggesting that Dom34 might affect Hel2 function (Fig 7). The rps3 mutations on their own, however, had no observable effect on the efficiency of ribosomal proteins ubiquitination. Hence, it is very unlikely that the effect of the entry-tunnel mutations on NGD are due to differences in ribosomal protein ubiquitination during stalling.
We reasoned that if the entry-tunnel residues of Rps3 are affecting NGD, then mutating them should result in increased sensitivity to cycloheximide especially at intermediate concentrations, at which ribosome collisions will occur and hence NGD is triggered. Growth of the rps3 strain was compared to the wild-type one in the presence of varying concentrations of cycloheximide (Fig 8A and 8B). To distinguish between effects on the growth rate versus lag time, we determined the first derivative of the growth curve to measure the instantaneous growth rate. The maxima of the resulting curves report on the maximal growth rate, whereas the distance between the maxima reports on the lag. As expected, the mutations had no effect on the growth rate or lag period in the absence of the drug and at very low and high concentrations (S3 Fig). In contrast and in agreement with our model, the addition of cycloheximide at intermediate concentrations (0.02–0.32 μg/mL) significantly increased the lag period for the R116A/R117A mutant. This effect was most noticeable at the 0.16 μg/mL concentration, for which we observed a lag-time difference between the wild-type and the mutant cells of more than 4 hours (S3 Fig). Our data suggests that the entry-tunnel residues are important for dealing with intermittent collision events, and likely the ensuing process of ribosome rescue.
Previous work from our lab revealed that RNA oxidation strongly stalls translation in vitro [17]. In particular, the introduction of a single 8-oxoguanosine adduct to the mRNA reduced the rate of peptide-bond formation by almost three orders of magnitude in a bacterial reconstituted system and prevented the formation of full-length protein products in wheat-germ and rabbit-reticulocyte extracts. We also provided evidence that showed oxidized mRNA is subject to NGD. Because our rps3 mutations appear to affect NGD, they should also in principle result in increased sensitivity to agents that react with RNA to produce adducts such as 8-oxoguanosine. We used the chemical 4-Nitroquinoline 1-oxide (4NQO), a UV mimetic and known to produce reactive oxygen species, to introduce 8-oxoguanosine into RNA [65] in living yeast. Wild-type and rps3-mutant cells were grown to mid-logarithmic before being challenged with 5 μg/mL of 4NQO for 30 minutes. Cells were washed with fresh media, diluted and their growth monitored. In the absence of any drugs, the rps3 mutant displayed a growth rate nearly identical to that of the wild-type (6.6 ±0.23 versus 6.3 ±0.07 hours). After incubation with 4NQO, the mutant displayed a notable lag in its growth of 1.4 hours (10.4 ±0.18 versus 9.0 ±0.79) (Fig 9A and 9B). We note that although the effects we saw are modest, they are reproducible and suggest that mutations of the entry-tunnel residues render cells sensitive to damaging agents. These effects are also reminiscent of the effects that we and others have documented for dom34Δ and xrn1Δ strains [17]. These findings together with the observation that mutations in RPS3 result in increased sensitivity towards cycloheximide provide further support for a role for the factor in NGD.
NGD is a conserved eukaryotic process that responds to stalled ribosomes [14]. The process is characterized by an endonucleolytic cleavage of the aberrant mRNA upstream of the lead ribosome [15] and as yet the identity of the culprit endonuclease remains unknown. As a result, there is a critical gap in our understanding of some of the mechanistic details of the process. Nonetheless, multiple studies have provided important hints about the enzyme. For instance, mapping experiments suggested that the endonuclease is ribosome-associated [40, 41]. In particular, cleavage takes place in frame with the ribosome and is phased by ~30 nt, the mRNA-length protected by the ribosome. Furthermore, the reaction appears to likely take place between stacked ribosomes [40]. These studies hinted at a role for the ribosome itself in activating or recruiting the endonuclease. Here we provided further evidence for this notion. More specifically, we find the entry tunnel of the ribosomal protein Rps3 to be important for the cleavage reaction. Mutation of the key-entry-tunnel residues Arg116 and Arg117 were found to drastically affect the outcome of the cleavage event; we observe a significant reduction in the accumulation of 5’-fragments from a number of NGD reporters when these residues are mutated to Ala. Consistent with these findings, although subtle, the half-life of the SL reporter increases in the presence of the mutations suggesting that these mutations may stabilize NGD reporters. Mapping of the cleavage products also revealed spreading of the cleavage reaction in the presence of the mutations. We note that Rps3 is known to interact with two key NGD factors: Dom34 and ribosome-associated Asc1 [60, 66]. Although deletion or mutation of these factors affects the cleavage pattern in the rps3 background, as evidenced by northern analysis, the effect of the mutations on NGD do not appear to phenocopy those observed in the dom34Δ and asc1 strains, which is apparent in the high-throughput mapping data. Furthermore, the mutations do not alter Asc1 occupancy on the ribosome. Collectively our data suggests that the entry-tunnel region of Rps3, and hence the ribosome, has a function in NGD upon stalling. In agreement with this proposal, mutations of this region render cells sensitive to intermediate concentrations of cycloheximide and the nucleic-acid damaging agent 4NQO; both stall the ribosome and likely trigger NGD.
Apart from the decoding center nucleotides, the Arg116 and Arg117 residues of the entry tunnel of the ribosome come closest to the mRNA. Indeed, some of the first studies on this region showed it to be important for unwinding the mRNA and make up part of the helicase domain of the ribosome [42]. While our data do not show the residues to be required for cleavage to take place–we still observe accumulation of NGD fragments in the presence of the mutations–they clearly affect the pattern of the cleavage reaction. It is feasible that the electrostatic interaction between the side chains and the phosphodiester backbone of the mRNA is important for locking the mRNA in place for the endonuclease to carry out its cleavage reaction. When these residues are mutated to Ala residues, the mRNA is more dynamic and its accessibility to the enzyme’s active site is severely affected. Alternatively, these residues might be important for recruiting or activating the endonuclease and as a result, changing their identities inhibits the cleavage reaction, although it is not clear how residues buried deep in the ribosome could be used efficiently to recruit exogenous protein factors. Instead, we favor a model whereby the endonuclease is intimately associated with the ribosome and it is activated upon stalling. In agreement with this, previous work has indicated that during non-stop decay, when the ribosome runs to the end of an mRNA, the endonucleolytic cleavage takes place near the exit tunnel of the ribosome [16, 41, 67] as evidenced partly by the accumulation of 15–18 nt fragments. Similarly, during a novel form of mRNA degradation termed ribothrypsis, it was suggested that an endonucleolytic cleavage event takes place near the exit tunnel [68]. Interestingly, recent structural data from human cells has revealed the position of multiple ribosomal proteins and associated factors at collided di-ribosomes–events that trigger NGD [69]. It appears that this higher order structure brings an entry- and exit-tunnel face of adjacent ribosomes in close proximity, which could potentially allow for interactions between otherwise distally positioned components. These include RACK1/Asc1 on the stalled ribosome with uS3, eS10, and uS10 on the collided ribosome, as well as eS26 and eS28 facing uS4 and rRNA helix 16 on the stalled and collided ribosomes, respectively. It will be exciting to see how modifications to these factors may affect endonuclease activity.
In an endonuclease-independent consequence, the residues and their interaction with the mRNA could play a role in recruiting Dom34 and Hbs1 to the ribosome. Biochemical and structural studies have suggested that Hbs1 is recruited to a ribosome with little to no mRNA downstream of the A site [20, 21, 66, 70]. The N-terminal of Hbs1 binds in the RNA entry tunnel, interacting with Rps3 [66]. It was hypothesized that Hbs1 cannot bind in the presence of mRNA in the entry tunnel [19, 20, 60, 66]. Additional recent structural studies also revealed a potential role for Dom34 in sensing the mRNA channel, whereby it uses a unique β-loop to protrude into the mRNA channel to sense its absence [60]. Together these two mechanisms ensure that ribosome dissociation only occurs when the ribosome reaches the end of the mRNA, such as during NSD or on the behind ribosomes following cleavage during NGD. It is possible that the mutations in the entry tunnel of Rps3 make the mRNA more dynamic, preventing a clash with Dom34 and Hbs1. In turn, this allows the factors to bind and dissociate the ribosomes before cleavage could take place. In agreement with this model, deletion of DOM34 in the presence of the rps3 mutations restores cleavage efficiency, and with increased heterogeneity, as expected, due to widespread ribosome queueing. This model, however, does not explain why the cleavage patterns in the double mutant do not look similar to those observed in the dom34Δ mutant. Therefore, the effects of the rps3 mutations appear to be more complex and they are likely to alter different aspects of NGD including the cleavage and the dissociation reactions. In contrast, the mutations do not appear to affect the RQC pathway, as we observe comparable ribosomal protein ubiquitination patterning and efficiency upon inducing ribosome collisions regardless of the status of Rps3.
Perhaps not surprising given its proximity to the mRNA, Rps3 plays a number of roles on the ribosome during translation. It has been shown to be important for providing the helicase activity to the ribosome; in bacteria Rps3/uS3, together with Rps4/uS4 and Rps5/uS5, encircle the incoming mRNA within the entry tunnel. When Arg131 and Arg132 in bacteria (corresponding to Arg116 and Arg117 in yeast) were mutated to Alanine, the efficiency of unwinding an RNA duplex by the ribosome was reduced [42]. Residues of Rps4 were also shown to contribute to helicase activity, but the process overall is coupled to and dependent on movements during translocation [71]. Rps3 is known to interact with other ribosomal proteins, including ribosome-bound Asc1/RACK1 [60, 66]. In addition to its aforementioned role in NGD, Asc1 is known to be involved in preventing readthrough of inhibitory codons and reading-frame maintenance [72]. In eukaryotes, the C-terminal tail of Rps3 lies further inside the mRNA channel, proximal to Asc1 [43]. It is tempting to speculate that conformational changes that involve Rps3 could be communicated to Asc1, which then may initiate additional steps in NGD. However, the convergence of phenotypes among Rps3, Asc1 and Dom34 highlight the potential for redundancy or simply subtle differences of function between these and related factors. This is also evident during non-functional 18S rRNA decay (NRD), where both Asc1 and Rps3 have recently been identified as players in the pathway [73]. The post-translationally modified C-terminal tail of Rps3 is required for 18S NRD and, as Asc1 can collaborate with either Dom34 or Hbs1, it was suggested that multiple overlapping pathways function to deal with damaged rRNA. At another step in the translation cycle, Rps3 also contributes to stabilizing the incoming mRNA during initiation. Again, yeast residues Arg116 and Arg117 were shown to promote binding of the mRNA to eIF3 dependent pre-initiation complexes (PICs) and in particular, when the exit channel is empty, they were absolutely required [46]. This demonstrates the diverse functionality of Rps3 that is likely due in part to its position at the entry tunnel where it interacts with and can survey incoming transcripts.
Collectively our findings provide further evidence for the central role of the ribosome in mRNA-surveillance pathways beyond just recognizing the aberrant mRNA and initiating the downstream events. The observation that mutations deep into the ribosome lead to dramatic changes to NGD bolsters arguments by us and others that the endonuclease is likely to be an integral part of the machine. This in turn could explain why it has been difficult to identify the endonuclease. It would be interesting to examine how quality control mechanisms evolved to integrate into fundamental biological machines. Further delineation of the details of this mechanism will also contribute to the understanding of how cells identify and degrade defective biological molecules. Finally, similar to NMD, NGD is likely to have been coopted to regulate gene expression. Indeed, recent reports have shown conditional deletion of Pelota (the human orthologue of Dom34) results in abnormal cellular differentiation [74]. The identification of the endonuclease is more than likely to provide further and important appreciation of the pervasiveness of this mode of gene regulation through NGD.
Cells were grown at 30°C in YPD or in defined media when expressing reporter plasmids. Yeast strains were made using standard PCR-based disruption techniques in the background BY4741 (MATa (his3Δ1 leu2Δ0 met15Δ0 ura3Δ0). SKI7 knockout strains were generated with a LEU2 cassette, amplified using oligos complementary to the insertion site.
RPS3 mutant strains were constructed by first cloning a fragment encoding RPS3-HIS3-rpS3 3’UTR, generated by fusion PCR, into the BamHI/XhoI sites in pPROEX-HTb. Point mutations in RPS3 were introduced by site directed mutagenesis and a cassette encoding the entire region was PCR amplified and used to transform the target yeast strains. RPS2 (E120A) strains were made using the same method and ASC1 (R38D, K40E) strains were made similarly, except using BamHI/XbaI sites in pET28a. HIS3 and LEU2 coding regions were amplified from plasmids pFAGa-6xGLY-FLAG-HIS3 and pAG415 [75] respectively.
Plasmids encoding the PGK1 gene or PGK1-SL under control of the GAL1 promoter were obtained from R. Parker [15]. PGK1-(CGA)12, PGK1-(AAA)12 and PGK1-(UUU)12 were made by annealing complementary oligos and ligating them to XbaI digested PGK1 plasmid [40].
Culture was grown overnight in a defined media (-Ura) with glucose. Cells were washed twice in media containing 2% Raffinose and 2% galactose, diluted to OD 0.1 in the same media and grown to an OD of 0.5–0.8 to permit expression of the gal-driven reporters. RNA was isolated using hot phenol extraction followed by two sets of chloroform extraction and ethanol precipitation. 2μg of total RNA was resolved on 1.2% formaldehyde agarose gel, followed by transfer to positively-charged nylon membrane (GE Lifesciences) using a vacuum blotter (Biorad). Next, nucleic acids were UV cross-linked to the membrane and baked at 80°C for 15 minutes. Membranes were then pre-hybridized in Rapid-Hyb buffer (GE Lifesciences) for 30 minutes in a hybridization oven. Radiolabeled DNA probe, which was labeled using polynucleotide kinase and [γ-32P]ATP, was added to the buffer and incubated overnight. Membranes were washed with nonstringent buffer (2 × SSC, 0.1% SDS) three times, in some cases followed by three washes in stringent buffer (0.2 × SSC, 0.1% SDS), all at hybridization temperature. Membranes were exposed to a phosphorimager screen and analyzed using a Biorad Personal Molecular Imager.
All Northern analyses were performed using at least three biological replicates. Representative images are shown.
Cells expressing PGK1-SL were grown overnight in defined media (-Ura) plus glucose. Cultures were then washed in -Ura media, resuspended at OD 0.1 in 50 mL -Ura plus galactose, and grown for 18–20 hours to allow expression of the reporter plasmid. Cells were collected at OD 0.5–0.6, washed once and resuspended in 11 mL pre-warmed -Ura media. A 1 mL aliquot was saved for the t0 timepoint and 1 mL 40% glucose added to the remainder. Cells were incubated at 30°C while shaking and aliquots taken at the indicated timepoints. For each sample, cells were pelleted, media was removed, and tubes were frozen on dry ice. RNA was isolated using a hot phenol method followed by two rounds of chloroform extraction and ethanol precipitation. 2 μg of total RNA for each sample was analyzed by Northern blot.
Yeast cultures were grown to mid-log phase before addition of cycloheximide to a final concentration of 100 μg/mL. The culture was chilled by adding an equal volume of ice and centrifuged at 4°C. Cells were then resuspended in polysome lysis buffer (20 mM Tris pH 7.5, 140 mM KCl, 5 mM MgCl2, 0.5 mM DTT, 1% Triton-100, 100 μg/mL cycloheximide, 200 μg/mL heparin), washed once and lysed with glass beads using a FastPrep (MP Biomedical). Supernatant from cleared lysate corresponding to 1 mg of total RNA was layered over a 10–50% sucrose gradient and centrifuged at 37,000 rpm for 160 min in an SW41Ti (Beckman) swinging bucket rotor. Gradients were fractionated using a Brandel tube-piercing system combined with continuous absorbance reading at A254 nm. Proteins were precipitated by the addition of TCA to 10% after a twofold dilution with water, and resuspended in HU buffer (8 M Urea, 5% SDS, 200 mM Tris pH 6.8, 100 mM DTT).
Proteins were resolved on 15% SDS PAGE gels and transferred to PVDF membranes using a semi-dry transfer apparatus (BioRad). The membranes were blocked with milk in PBST for ~ 30 minutes at room temperature followed by incubation with primary antibody overnight at 4°C. After washing with PBST, the membrane was incubated with the appropriate HRP-conjugated secondary antibody for ~ 1hr at room temperature before washing 3–4 × with PBST. Detection was carried out on a GE ImageQuant LAS 4000 using the Pierce SuperSignal West Pico Chemiluminescent Substrate. The following antibodies were used: mouse anti-PGK1[22C5D8] (ab113687) and rabbit anti-rpS9 (ab117861) from Abcam; rabbit anti-ASC1 was a gift from Wendy Gilbert (Yale University) [61]; mouse anti-rpL4 was a gift from Heather True (Washington University in St. Louis); goat anti-mouse IgG HRP (31430) and goat anti-rabbit IgG HRP (31460) from Thermo Scientific.
Total RNA from the indicated strains was ligated to a short adenylated DNA oligonucleotide, 5'rAppCTGTAGGCACCATCAAT/3ddC/ 3', at its 3’ end using truncated T4 RNA ligase 2 (NEB). For each sample, total RNA from at least two biological replicates was included. Reverse transcription using a primer complementary to the adaptor was performed, and then cDNA was amplified with a 5’-primer that annealed at position 585 of PGK1. Primers were designed for the Illumina HiSeq platform and samples were column purified to remove primers before sequencing.
Single-read HiSeq 2500 sequencing was performed by the Genome TechnologyAccess Center (GTAC) at Washington University. Raw data was analyzed for quality using the Fastx toolkit (http://hannonlab.cshl.edu/fastx_toolkit/index.html), trimmed using cutadapt [76] and aligned to our reference reporter sequence using NovoAlign (http://www.novocraft.com/). Sequencing results are available at GEO (accession #GSE117652).
Sensitivity assays were conducted essentially as described [74]. Yeast cells were grown to mid-log-phase (OD600 of 0.5–0.7), collected, washed and resuspended in YPD to a final density of OD 0.8. 5 μl of the cell suspension was added to 195 μl of YPD with CHX at various concentrations, from 0–10 μg/mL. All samples were prepared in biological triplicates as well as technical duplicates in 96-well polystyrene microplates. The plate was incubated at 30°C with shaking on a microplate scanning spectrophotometer (Biotek). Cell density was monitored every 10 min over 24–48 h at 600nm. To assay sensitivity to 4NQO, after growing cells to mid-log (OD 0.5–0.7) cultures were treated with and without 5 μg/mL 4QNO for 30 minutes. Cells were collected, washed and adjusted to OD 0.8. Samples were plated and growth monitored as above.
|
10.1371/journal.pntd.0006666 | Comparing the effectiveness of different strains of Wolbachia for controlling chikungunya, dengue fever, and zika | Once Aedes aegypti and Aedes albopictus mosquitoes that spread Chikungunya virus, dengue virus, and Zika virus are infected with Wolbachia, they have reduced egg laying rates, reduced transmission abilities, and shorter lifespans. Since most infected mosquitoes are only infectious in the last few days of their lives, shortening a mosquito’s lifespan by a day or two can greatly reduce their abilities to spread mosquito-borne viral diseases, such as Chikungunya, dengue fever, and Zika. We developed a mathematical model to compare the effectiveness of the wMel and wAlbB strains of Wolbachia for controlling the spread of these viruses. The differences among the diseases, mosquitoes, and Wolbachia strains are captured by the model parameters for the mosquito-human transmission cycle. Moreover, the model accounts for the behavior changes of infectious population created by differences in the malaise caused by these viruses. We derived the effective and basic reproduction numbers for the model that are used to estimate the number of secondary infections from the infectious populations. In the same density of Wolbachia-free Aedes aegypti or Aedes albopictus mosquitoes, we observed that wMel and wAlbB strains of Wolbachia can reduce the transmission rates of these diseases effectively.
| Mosquitoes infected with Wolbachia bacteria are less capable of transmitting dengue virus, Chikungunya virus, and Zika virus. We use a mathematical model to quantify the impact of infecting wild Aedes aegypti and Aedes albopictus mosquitoes with wAlbB and wMel strains of Wolbachia in reducing the transmission of these viruses. The model is a system of ordinary differential equations that accounts for reduced fitness of the Wolbachia-infected mosquitoes, reduced transmissibility of an infection, and the behavior changes of infected individuals caused by the infection. We derived an explicit formula for the effective reproduction number for when the host population are partially immune to new infections, as occurs in seasonal outbreaks. We compared the effectiveness of different species of mosquitoes, different strains of Wolbachia, and different diseases. Our model is a general model that can produce outputs for a specific location, once the data for the location are available to parameterize the model. The spatial heterogeneity of the populations must be considered before using this model to help guide policy decisions.
| The current pandemics of Chikungunya caused by Chikungunya virus (CHIKV), dengue fever caused by dengue virus (DENV), and Zika resulted from Zika virus (ZIKV) infect over one hundred million people each year [1]. In the past decade, CHIKV has spread around the world [2, 3] and recently over a million cases occurred in the Caribbean and Latin America. Symptoms of infection with CHIKV include high fever and headache, with arthritis affecting joints, and may sustain for weeks or months [4]. Dengue fever has spread around the world and is endemic in South America and Asia [1]. People infected with DENV have symptoms ranging from mild headaches, severe headaches and joint pains to hemorrhagic or shock syndrome fever. Recently, ZIKV has spread through the Americas, starting with a 2015 explosive outbreak in Brazil. Although most people infected with ZIKV have mild symptoms, there is a correlation between infections in pregnant women and their children born with microcephaly (an abnormally small brain). Currently no vaccines are commercially available for Zika and Chikungunya. The first dengue vaccine, Dengvaxia (CYD-TDV), registered in Mexico in December, 2015 is not effective for the younger ones and for the seronegative population due to ethical concern of non-maleficence [5].
Although both Aedes aegypti (Ae. aegypti) and Aedes albopictus (Ae. albopictus) mosquitoes can transmit these viruses, Ae. aegypti mosquitoes are more abundant in urban areas and are the primary vectors. Current prophylactic measures include individual protection from mosquito bites, such as applying mosquito repellent and avoiding exposure to mosquitoes. Limited control options focus on reducing the mosquito populations, including spraying insecticides, treatments, and removal of mosquito breeding sites. These control strategies are hard to sustain because of the vigilance needed to eradicate the breeding sites, the expense of repeated spraying, and the mosquitoes developing resistance to the insecticides.
The cost and difficulty of eliminating the mosquitoes motivate the need of developing more efficient strategies to mitigate and control the transmission of these viruses. Wolbachia is a genus of bacteria that can infect 25-75% of all insects [6] and recent studies have shown that some strains of Wolbachia can increase the resistance of mosquitoes being infected with these viruses [7]. Recent experiments have shown that wMel strain of Wolbachia infection in Ae. albopictus inhibits the growth of CHIKV [8]. Moreover, Wolbachia infection in mosquitoes reduces egg laying rates, reduces their ability of transmitting viral infections, and shortens their lifespans by a few days. Since many mosquitoes infected with DENV, ZIKV, or CHIKV are infectious for only a few days at the end of their lifespans, the shortened lifespans of Wolbachia-infected mosquitoes result in more mosquitoes dying before they can transmit the infection. This implies that Wolbachia-infected mosquitoes sustaining in a wild mosquito population will be less likely to transmit these viral diseases.
Infected females can pass the bacteria to their offsprings and spread Wolbachia vertically from one generation to the next. Wolbachia disrupts the reproductive cycle of hosts through a cytoplasmic incompatibility between the sperms and eggs. Cytoplasmic incompatibility occurs when Wolbachia-infected male mosquitoes mate with Wolbachia-free female mosquitoes, and causes the Wolbachia-free females to produce fewer progeny [9, 10]. These effects provide the Wolbachia a vertical transmission advantage. However, this is offset by a reduced lifespan and reduced number of eggs hatched by a Wolbachia-infected mosquito.
When Wolbachia-infected mosquitoes are introduced in a wild population of uninfected mosquitoes, the infection is quickly wiped out unless the fraction of infected mosquitoes exceeds a threshold θ of the total population. Recent mathematical models have established these threshold conditions as θ = 0.15, 0.24, and 0.6 for wAlbB-, wMel-, and wMelPop-infected mosquitoes to establish a stable population, respectively [11]. Note that these threshold estimates are for an ideally controlled situation where mosquitoes do not mix with surrounding uninfected mosquitoes. The thresholds could be much higher for a release in the wild. Recent studies have found that maintaining a sustained wMelPop-infection requires continually introducing new wMelPop-infected mosquitoes into the wild population [12]. A recent research has reported that large-scale releases of Wolbachia-infected Ae. aegypti in the city of Cairns, Australia, invaded and spread through the populations, while Wolbachia infection at a smaller release site collapsed due to the immigration of Wolbachia-free mosquitoes from surrounding areas [13]. Population cage experiments indicated that the wAlbB strain can be successfully introduced into populations, and subsequently persist and spread [14]. Ndii et al. analyzed a first-order differential equation and found that a significant reduction in human dengue cases can be obtained by releasing wMel-infected mosquitoes, instead of wMelPop-infected mosquitoes due to the greatly reduced lifespans [15]. Ferguson et al. developed a mathematical model of DENV transmission incorporating the dynamics of viral infection in humans and mosquitoes, and predicted that wMel-infected Ae. aegypti mosquitoes have a substantial effect on transmission [16]. Ross et al. found that, in most situations, it was easier to establish wMel than wAlbB in mosquito populations, except when the conditions were particularly hot [17]. They also observed that the wMel infected larvae survived better than wAlbB infected larvae under starvation conditions [18].
Many mathematical models have been developed to explore conditions under which Wolbachia can be used to fight against the spread of viruses effectively. The analysis of a compartmental mathematical model showed that a significant reduction in human dengue cases can be obtained provided that Wolbachia-infected mosquitoes persist when competing with Wolbachia-free mosquitoes [15]. Zhang et al. developed an ordinary differential equation (ODE) model to assess how best to replace DENV vectors with Wolbachia-infected mosquito populations and the results showed that successful population replacement will rely on the selection of suitable strains of Wolbachia and careful design of augmentation methods [19]. The analysis for an impulsive model for Wolbachia infection control of mosquito-borne diseases with general birth and death rates showed that strategies may be different due to different birth and death rate functions, the type of Wolbachia strains, and the initial number of Wolbachia-infected mosquitoes [20]. Xue et al. [21] created a two-sex model that included an egg/aquatic stage for the mosquito lifecycle and observed that an endemic Wolbachia infection can be established only if a sufficient number of infected mosquitoes are released. Recently, this model was extended by Qu et al. [22] to better account for the cytoplasmic incompatibility by considering the fact that most female mosquitoes only mate once. They used the model to investigate the effectiveness of multiple releases of infected mosquitoes in sustaining an endemic Wolbachia infection.
Manore et al. [23] used a mathematical model to compare the spread of DENV and CHIKV in Ae. aegypti and Ae. albopictus mosquitoes that are not infected with Wolbachia. Our study is based on extending these results to evaluate the effectiveness of infecting these mosquitoes with different strains of Wolbachia to show their different roles in controlling different vector-borne diseases. In our model, we assume that lifespans of the infected adult mosquitoes are slightly shorter than those of uninfected mosquitoes (reducing transmission), and the larval survival rates of wAlbB-infected mosquitoes are less than those of wMel-infected mosquitoes (making invasion somewhat potentially harder for wAlbB). We evaluated the effectiveness of infecting these mosquitoes with wMel and wAlbB strains of Wolbachia to show their different roles in controlling the transmission of DENV, ZIKV, and CHIKV. Since transmission of ZIKV is estimated to be similar to transmission of DENV but exact values of parameters are not available [24], we assume that the parameter values for ZIKV are the same as those for DENV except the fraction of infectious humans exposed to mosquito bites. Our simulation results show that the differences between the spread of DENV and ZIKV lie in different behaviors of infectious humans, and wMel is more effective than wAlbB strain of Wolbachia in simulations with the available baseline parameters.
Our compartmental model (Fig 1) divides the human population into four classes: susceptible, Sh, exposed (infected but not infectious), Eh, infectious, Ih, and recovered (immune), Rh, and splits the mosquitoes into three classes: susceptible, Sv, exposed, Ev, and infectious, Iv. We assume that humans advance from the infectious state to the recovered state, while mosquitoes do not. Model parameters are defined in Table 1, and the ODEs are:
d S h d t = μ h H 0 - α v ( t ) I v - μ h S h , d E h d t = α v ( t ) I v - ν h E h - μ h E h , d I h d t = ν h E h - γ h I h - μ h I h , d R h d t = γ h I h - μ h R h , d S v d t = B v ( N v ) N v - α h ( t ) I h - μ v S v , d E v d t = α h ( t ) I h - ν v E v - μ v E v , d I v d t = ν v E v - μ v I v . (1)
The equations are homogenous with respect to the populations of humans and mosquitoes. The solution is invariant as long as both populations are scaled by the same factors and the ratio ρvh = V0/H0 is kept fixed. That is, all the results in this study hold for other populations with the same vector-to-human ratio, ρvh.
Humans and mosquitoes leave the population through combined per capita recruitment / death / emigration rates, μh and μv, respectively. Although the model equations can accommodate variations in the human and mosquito populations, we assume a constant size of human population in simulations. Susceptible humans enter the population at a fixed rate of μhH0 per day to maintain a steady at-risk human population of H0. In our simulations, we assume that 1/μh = 20 years, where about 5% of the population turns over each year to account for individuals moving in and out of the population. Note that μh depends on the particular region being modeled and will be larger in regions where there are migrant workers, or smaller in isolated villages.
Mosquitoes are born into the population at the rate of Bv(Nv)Nv per day, where the mosquito birth and population saturation function Bv is given by [23]
B v ( N v ) = ψ v - ( ψ v - μ v ) N v V 0 .
Here ψv is the mosquito birth rate when there are abundant resources for the eggs and larvae, and V0 is the carrying capacity and steady state for the mosquitoes.
The viruses can be transmitted from infectious mosquitoes to susceptible humans, and from infectious humans to susceptible mosquitoes through blood meals. We formulate the viral transmission in terms of the rate at which infectious mosquitoes or infectious humans infect others, rather than the traditional formulation in terms of the rate at which the susceptible individuals are being infected as in [23]. The notation in this infectious viewpoint emphasizes that the infectious population drive the epidemic and clarifies the derivation for effective reproduction number.
The force from infection for humans, αv(t), and force from infection for mosquitoes, αh(t), are the rates at which infectious individuals infect others and are equal to the product
α * ( t ) = ( probability of transmission per bite, β * ) × ( number of bites per infectious individual per day , b i * ) × ( probability that an infectious individual bites a susceptible individual, P s * ) ,
where * = v or h. That is,
α v ( t ) = β h v b i v ( t ) P s h ( t ) , α h ( t ) = β v h b i h ( t ) P s v ( t ) . (2)
Here βhv is the probability that an infectious mosquito will infect a susceptible human in one bite. Similarly, βvh is the probability that an infectious human will infect a susceptible mosquito in one bite.
The parameter biv is the average number of times an infectious mosquito bites a susceptible human per day. Similarly, bih is the average number of times that an infectious human is bitten by a susceptible mosquito per day. We assume that the biting rates for the mosquitoes do not change after they become infected.
We assume that infectious people, especially those with symptoms may avoid exposure to mosquito bites. Around 20–93% individuals infected with DENV are asymptomatic [25, 26], and asymptomatic prevalence of Chikungunya was estimated in the range 16.7–27.7% during some recorded outbreaks [27, 28], while only around 20% of people infectious with ZIKV have significant symptoms [29–31]. The biting rate accounts for a fraction, π, of infectious humans do not change their behaviors due to the illnesses and continue being bitten by mosquitoes at the same rate as the susceptible population. In our model, we assume π = 0.75 for DENV, π = 0.3 for CHIKV, and π = 0.8 for ZIKV. The remaining fraction (1 − π) avoid being bitten by mosquitoes. This behavior change has a significant impact on the force of infection from humans to mosquitoes and is an important aspect of any vector-borne transmission model.
We define σv as the maximum rate at which a typical mosquito will bite humans per day, and σh is the maximum number of bites that a susceptible human will tolerate being bitten per day. We define Nv = Sv + Ev + Iv as the number of mosquitoes that bite humans. Similarly, we define Nh = Sh + Eh + πIh + Rh as the number of humans being bitten by mosquitoes. Recall that the at-risk population, Nh, is only a fraction of the total human population since some infectious people are not being bitten by mosquitoes. One of the most difficult parameters to estimate when applying any vector-borne epidemic models to a particular situation is the fraction of the population at risk of an infection. Using these definitions, σvNv is the maximum number of bites a mosquito seeks per day, while the maximum number of available human bites per day is σhNh.
The total number of times that all the mosquitoes bite humans must equal to the total number of times that humans are bitten by mosquitoes. To enforce this balance condition, we extend the harmonic average described in [23, 32] and define the total number of bites per day (total biting rate) as
b ( t ) = σ v N v σ h N h σ v N v + σ h N h . (3)
This biting rate allows a wide range of vector-to-host ratios, as opposed to the more standard frequency-dependent contact rates that are applicable only over a limited range of vector-to-host ratios [33].
The total number of bites from mosquitoes is b(t) = bvNv, where bv is the average number of bites per mosquito per day (the biting rate). Because we assume that the infection does not affect the biting rate, the average number of bites per day for an infectious mosquito is also biv = bv = b(t)/Nv. To satisfy the balance condition, the total number of bites on humans is also b(t) = bh Nh, where bh is the average number of times an infectious human being bitten per day. Because (1 − π) of infectious humans have changed their behaviors and are not being bitten, the average number of times an infectious human being bitten per day is bih = πbh = πb(t)/Nh.
We define Psh(t) as the probability that an infectious mosquito bites a susceptible human. If we assume that the bites on humans are randomly distributed, then Psh(t) = Sh(t)/Nh(t). Similarly, Psv(t) is the probability that when an infectious human is bitten, the bite is from a susceptible mosquito. Hence, Psv(t) = Sv(t)/Nv(t).
The ODES in Model (1) are formulated from the viewpoint of infectious population where the transmission parameter, α, is the force from infection. This is equivalent to the usual way of formulating the equations for force of infection from the susceptible viewpoint [23]. From the susceptible viewpoint, the ODEs for the susceptible humans and vectors have the form:
d S h d t = μ h H 0 - λ h ( t ) S h - μ h S h , d S v d t = B v ( N v ) N v - λ v ( t ) S v - μ v S v .
Here λ is force of infection and is related to α by
λh(t)Sh(t)=αv(t)Iv(t),λh(t)=βhvbsh(t)Piv(t),λv(t)Sv(t)=αh(t)Ih(t),λv(t)=βvhbsv(t)Pih(t), (4)
where the factors for λ* are all from the viewpoint of the susceptible population, instead of the viewpoint of the infectious population in Eq (2). Pih(t) = πIh(t)/Nh(t) is the probability that when a susceptible mosquito bites a human, the human is infectious; bsh = b(t)/Nh is the rate at which susceptible humans are bitten; Piv(t) = Iv(t)/Nv(t) is the probability that when a susceptible human is bitten, the mosquito is infectious; and bsv = b(t)/Nv is the rate at which susceptible mosquitoes bite humans.
The basic reproduction number, R 0, is defined as the number of new infections produced by one infected individual in a completely susceptible population. When the population is not fully susceptible, or more than one person is infected, then the effective reproduction number, R e f f ( t ), estimates the number of secondary cases produced by a typical infected individual at any time during the epidemic. We derived the effective reproduction number from infectious point of view for DENV, ZIKV, and CHIKV to estimate the reproduction rate of an epidemic at any stage.
Because this is a bipartite transmission cycle, mosquitoes only infect humans and humans only infect mosquitoes, we have different effective reproduction numbers for each part of the cycle. We define R h v ( t ) as the effective reproduction number for transmission from mosquitoes to humans, and is the average number of humans infected by one infectious mosquito. Similarly, R v h ( t ) defined as the effective reproduction number for transmission from humans to mosquitoes, is the average number of mosquitoes infected by one infectious human. These dimensionless numbers are defined by
R h v ( t ) = P v α v ( t ) τ i v , (5) R v h ( t ) = P h α h ( t ) τ i h . (6)
Here, in Eq (5), Pv = νv/(νv + μv) is the probability that an infected mosquito survives through the incubation period and becomes infectious, αv(t) is the average number of susceptible people infected by an infectious mosquito per day, and τiv = 1/μv is the average life span of a mosquito. Similarly for Eq (6), Ph = νh/(νh + μh) is the probability that an infected human becomes infectious, αh(t) is the average number of susceptible mosquitoes infected by an infectious person per day, and τih = 1/(γh + μh) is the average time that a human remains infectious.
The explicit expressions of R h v ( t ) and R v h ( t ) are:
R h v ( t ) = ν v ( ν v + μ v ) β h v b i v S h ( t ) N v ( t ) 1 μ v , R v h ( t ) = ν h ( ν h + μ h ) β v h b i h S v ( t ) N h ( t ) 1 γ h + μ h .
Because a full transmission cycle is consisted of two stages, and R e f f ( t ) measures the average effective reproduction number over one cycle. We take the geometric average of these two reproductive numbers to define:
R e f f ( t ) = R h v ( t ) R v h ( t ) .
We denote S h * as the population of susceptible people at the endemic equilibrium (EE) for Model (1). We define the fraction of humans susceptible at the EE, S h * / H 0 of the population has never been infected, as susceptibility of humans at EE,
S h * H 0 = 1 - ( 1 - π R 0 2 ) ν v σ v σ h β h v V 0 ν v σ v σ h β h v V 0 + μ h ( μ v + ν v ) ( σ h H 0 + σ v V 0 ) .
To quantify the differences in impact of different strains of Wolbachia on an epidemic, we define the coefficient for effectiveness [15] as the relative decrease in the number of people predicted to be infected if the mosquitoes are infected with Wolbachia, HW, compared with the predicted number of people who will be infected if the mosquitoes are Wolbachia-free, HF;
κ = H F - H W H F = 1 - H W H F . (7)
If κ = 1, then Wolbachia is predicted to be effective in stopping all the infections, while if κ = 0, then it is predicted to have no effects on the epidemic.
The basic reproduction number is the effective reproduction number at the disease free equilibrium where Sv(0) = V0 and Sh(0) = H0, and all other states are zero:
R 0 = R h v ( 0 ) R v h ( 0 ) . R h v ( 0 ) and R v h ( 0 ) are the effective reproduction numbers for the vectors and humans at the disease free equilibrium:
R h v ( 0 ) = ν v ( ν v + μ v ) β h v b i v ( 0 ) 1 μ v , R v h ( 0 ) = ν h ( ν h + μ h ) β v h b i h ( 0 ) 1 γ h + μ h .
The biting rates at the disease free equilibrium are biv(0) = b(0)/V0 and bih(0) = πb(0)/H0, where
b ( 0 ) = σ v V 0 σ h H 0 σ v V 0 + σ h H 0 .
The basic reproduction number derived in this way is consistent with the R 0 computed using the next generation matrix approach in [23].
After an epidemic has run its course and the infection has died out, then the previously infectious people are immune to new infections. R e f f is the average number of new infectious individuals produced in one cycle when an infectious human or mosquito is introduced into the population where some of the population is immune to infection. We define ph = Rh(0)/H0 as the fraction of people who are immune to the infection, such as those who have already had the disease or have been immunized. If we reinitialize our model at t = 0 with an infection-free equilibrium, such as the beginning of a seasonal outbreak, where ph of the humans are immune to the infection, Rh(0) ≠ 0, and Sh(0) + Rh(0) = H0. For this case, the effective reproduction number for human-to-mosquito transmission, R v h ( 0 ) e f f = R v h ( 0 ), is unchanged, while the effective reproduction number for mosquito-to-human transmission is reduced to R h v ( 0 ) e f f = R h v ( 0 ) ( 1 - p h ). Therefore, the effective reproduction number for Model (1) with ph people immune to infection becomes
R e f f ( 0 ) = R h v ( 0 ) e f f R v h ( 0 ) e f f = R h v ( 0 ) R v h ( 0 ) ( 1 - p h ) = R 0 1 - p h .
Note that, unlike human-to-human transmitted disease where the R 0 is reduced by (1 − ph) when ph of the population is immune, R 0 is only reduced by 1 - p h in this bipartite epidemic.
In the simulations, the parameters are set to the baseline values in Table 2, unless specifically stated otherwise. Most of the parameter values used in this study were derived, or extensively referenced, by Manore et al. [23] for disease transmission in Wolbachia-free mosquitoes. Manore et al. provided a comprehensive sensitivity analysis on how the model predictions change with respect to variations in the key parameters [23]. Although these baseline values are our best estimates for the parameters, they are scalar estimates from a distribution of possible values. To help quantify the uncertainty in the parameters, we will investigate the behavior of the model over a wide range of feasible parameters. The model predictions for a specific value of the basic reproduction number or the fraction of the population infected at the endemic equilibrium depend on the specific values used in the simulations. Although these specific values for these predictions are sensitive to the parameter values, we find that the qualitative differences between different diseases and strains of Wolbachia are fairly insensitive over the feasible ranges of parameters.
We assume that the probability of transmission per bite from a mosquito to a human is related to the viral load in the mosquito. Recent experimental comparisons of the growth of DENV, ZIKV, and CHIKV in mosquitoes indicate that the viral loads and the extrinsic incubation period (EIP) for an infected mosquito to become infectious are comparable [34]. Because there are no experimental estimates for the infectivity of ZIKV-infected mosquitoes, we assume that the parameter values for infectivity of ZIKV are the same as those of DENV in our simulations. Note that, although we assume that the probability of transmission per mosquito bite is assumed to be the same for ZIKV- and DENV- infected mosquitoes. The behavior changes of humans infectious with ZIKV and DENV are different, leading to very different epidemics. That is, one must be very careful in extrapolating findings between ZIKV and DENV epidemics [24].
Duong et al. [26] showed that asymptomatic individuals infected with DENV may be infectious before the onset of symptoms and continue infecting mosquitoes as they visit multiple locations during the day. They also noted that sick people who are hospitalized or stay at home are only exposed to their residential mosquitoes. Grange et al. [25] summarized data from a large number of studies, showing that often 20–93% of DENV infected individuals are asymptomatic. In our simulations, we assume that 75% of DENV-infectious people continue exposing to mosquitoes (π = 0.75).
Bloch et al. [27] concluded that about 62.5% CHIKV infections are symptomatic through extensive statistical analysis. They observed that about one-third of CHIKV-infected participants are asymptomatic, which is consistent with estimates of 3–39% asymptomatic cases in past outbreaks. Robinson et al. [28] also noted that 16.7–27.7% of the infections in Chikungunya outbreaks are asymptomatic. In our simulations, we assume that 30% of infectious people with CHIKV continue exposing to mosquitoes (π = 0.30).
ZIKV infection is a self-limiting illness that is mostly asymptomatic. Lazear et al. [29] noted that approximately 20% of the individuals infected with ZIKV progress to a clinically apparent febrile illness, although rarely hospitalized. Rajah et al. [30] also observed that 20% of the people infected with ZIKV present mild symptoms. In our simulations, we assume that 80% of ZIKV-infectious people continue exposing to mosquitoes (π = 0.80).
The Wolbachia infection changes the mosquito’s birth, death, biting rates, and the transmissibility of an infection. We account for the change in these parameters by including a scaling factor, ϕ*. We identify the rates of Wolbachia-free mosquitoes by a tilde, · ˜, and define the factors for Wolbachia-infected mosquitoes as:
The values for the factors ϕ* in Table 3 are used in Table 2 for the baseline parameter values used for this study. These factors coincide with the factors applied by [35] for comparing the effects of different strains of Wolbachia. The ranges of the lifespans for wMel-infected, wAlbB-infected, and Wolbachia-free mosquitoes in Table 2 coincide with the plot for longevity of wAlbB- and wMel- infected mosquitoes plotted by Joubert et al. [36].
We compare the differences in the spread of DENV, ZIKV, and CHIKV in wMel- and wAlbB-infected Ae. aegypti and Ae. albopictus mosquitoes with the spread in Wolbachia-free mosquitoes. In the simulations, the parameters are set to the baseline values in Table 2, unless specifically stated otherwise. The baseline values are the best estimates available for these parameter values. Since parameter uncertainty exists, it is important to investigate the behavior of the model over the wide range of feasible parameters. By investigating the model over the full range of parameters, we have focused on the qualitative differences between different infections and strains of Wolbachia.
Fig 2 shows the cumulative number of infectious humans up to time t = 700 when 0.1% of humans are infected at t = 0. The figure illustrates that the wMel-infected mosquitoes are more effective in slowing disease transmission than wAlbB-infected mosquitoes. For DENV and ZIKV, the number of people infected by Ae. aegypti mosquitoes is greater than the number of humans infected by Ae. albopictus mosquitoes carrying the same strain of Wolbachia. The opposite is true for CHIKV where the number of people infected by Ae. albopictus mosquitoes is greater than the number of people infected by Ae. aegypti mosquitoes carrying the same strain of Wolbachia. Note that in these simulations we assume the same mosquito populations for both genera. Typically the density of Ae. aegypti mosquitoes is much greater than the density for Ae. albopictus in urban areas, while the opposite is true in wooded rural areas. In this study, we only considered one mosquito genus at a time. When both mosquitoes are present, then the predictions will depend on the total mosquito population. As a first order approximation, the results are interpolated based on the fraction of each mosquito genus.
The reproduction numbers depend on the ratio of mosquitoes to humans. The model predictions are scaled for all populations with the same ρvh = V0/H0, V0 is the initial total number of mosquitoes, and V0 = Kv. In Fig 3, the reproduction numbers are plotted as ρvh varies from a ratio of ρvh = 1, with an equal number of mosquitoes as humans, to ρvh = 100, with 100 times more mosquitoes than humans. Infecting mosquitoes with Wolbachia can reduce R 0 over the full range of mosquito to human ratios.
When only few mosquitoes are present, then R 0 < 1 for all the diseases. As expected, R 0 increases as the number of mosquitoes increases, as more and more mosquitoes transmit the infection. When there are about 20 to 30 mosquitoes per human, then R 0 slowly decreases as the biting rate for the mosquitoes decreases. The rate of decrease depends upon the specific biting Eq (3) being used in the model.
Table 4 lists the basic reproduction number computed with the baseline parameters. For this case, the R 0 of DENV (ZIKV, or CHIKV) transmitted by Wolbachia-free mosquitoes is the highest, followed by R 0 of DENV (ZIKV, or CHIKV) transmitted by mosquitoes infected with wAlbB strain of Wolbachia, and R 0 for mosquitoes carrying wMel strain is the smallest. The basic reproduction number for ZIKV is the largest in all cases. The basic reproduction number of DENV is greater than the basic reproduction number of CHIKV for mosquitoes carrying the same strain of Wolbachia or Wolbachia-free mosquitoes.
The basic reproduction number is a function of the baseline parameters. Others may come up with different baseline values for different outbreaks. The readers can estimate the model response to different baseline values of a parameter using the sensitivity indices in Table 5. The relative sensitivity index of the quality of interest, q, with respect to the parameter of interest, p is
S p q≔p * q * × ∂ q ∂ p | p = p * = θ q θ p ,
as described in [23], where the notation p* indicates that a variable is evaluated at the model baseline values. For example, S π R 0 = 0 . 0136 for DENV transmitted by wAlbB-infected Ae. aegypti mosquitoes, if we reduce π, by 10%, then R 0 will be reduced by 0.00136, since −0.1 × 0.0136 = −0.00136. Sensitivity indices of R 0 varying with the fraction of people exposed to mosquitoes are shown in Fig 4.
The effective reproduction number depends on the fraction of humans who are immune to the infection. Fig 5 shows the effective reproduction number varying with the immunity of humans, assuming that all mosquitoes are susceptible and the initial total mosquito population is V0. The effective reproduction number decreases with the increase of the immunity of humans. When all humans are immune to the disease, then the effective reproduction number is the same as the basic reproduction number.
Previous examples kept most of the parameters at their baseline values. If we allow all the parameters to vary over the entire feasible sampling space, we will obtain a distribution for R 0. The distribution for R 0 is a function of the distributions for the model parameters as they vary within their allowed ranges. We assumed a triangular distribution that vanishes at the endpoints of the feasible region and has the mode at the baseline values in Table 2. If we had assumed that the distribution was uniform over the range, where the parameter was just as likely to be at the upper or lower bound as our best guess (the baseline case), then the ranges for the reproduction numbers in Fig 6 will not change, but the distributions will have fatter tails.
The histograms of the distribution for R 0 for DENV, ZIKV, and CHIKV transmitted by Ae. aegypti are shown in Fig 6. The means and medians of R 0 for wMel-infected mosquitoes are the smallest. R 0 for ZIKV is the largest, followed by DENV, and then CHIKV. In a similar analysis for Ae. albopictus, R 0 is the largest for ZIKV, followed by CHIKV, and the least is DENV. This is in agreement with the analysis where we varied the parameters one at a time over their feasible ranges.
If Wolbachia is successful in reducing the spread of the viruses, then there will be more people uninfected at the EE. In Table 6, the fraction of humans still susceptible at the EE for wMel-infected mosquitoes is the largest, while the susceptibility of Wolbachia-free mosquitoes is the smallest when a certain disease is transmitted by a certain genus of mosquitoes. For Ae. aegypti infected with the same strain of Wolbachia, the percentage of humans susceptible to CHIKV is higher than the percentage of humans susceptible to DENV or Zika. For Ae. albopictus infected with the same strain of Wolbachia, the percentages of humans susceptible to DENV and ZIKV are higher than the percentage of humans susceptible to CHIKV. Although the wMelPop-infected mosquitoes are the most effective in stopping an epidemic, it is unrealistic to consider a fully infected wild population of wMelPop-infected mosquitoes. Hence, we did not include the analysis for wMelPop strain of Wolbachia. This coefficient for effectiveness computed using Eq (7) listed in Table 7 shows that wMel is significantly more effective than wAlbB in reducing the number of infections in simulations with the baseline parameters in Table 2.
A mosquito infected with the Wolbachia bacteria is less capable of transmitting DENV, ZIKV, and CHIKV, and one of the leading new mitigation strategies is to fight the spread of these viral infections by releasing Wolbachia-infected mosquitoes. We quantified the impact of wAlbB, wMel, and wMelPop strains of Wolbachia in reducing the transmission of CHIKV, DENV, and ZIKV. The model accounts for reduced fitness of the Wolbachia-infected mosquitoes, reduced ability of transmitting viruses, and the behavior changes of infected individuals caused by the infection. Because people infectious with DENV and CHIKV are more likely to have serious symptoms, we assumed that the people infectious with these viruses were less likely to be bitten by mosquitoes than the people infectious with ZIKV. The behavior changes of humans have significant impacts on the model predictions and, unfortunately, is often let out of most vector-borne disease models.
The baseline model parameters are estimates for the general population and our best guess from the literature. The relative sensitivity indices for these parameters (Table 5) can be used to predict how slightly different assumptions about these input parameters will change the basic reproduction number. The relative sensitivity analysis quantified the relative importance of the model parameters on the model predictions and can be used to quantify the importance of obtaining more accurate data to reduce the parameter uncertainty and improve the model’s predictability.
Over the entire range of parameter values, our simulation results show that the R 0 for wMel-infected mosquitoes is the smallest. For Ae. aegypti, R 0 for ZIKV is the largest, followed by DENV, then CHIKV. In a similar analysis for Ae. albopictus, R 0 is the largest for ZIKV, followed by DENV, then CHIKV. This is in agreement with the analysis where we varied the parameters one at a time over their feasible range. Our simulation results show that the wMel strain is more effective in controlling these viruses than wAlbB strain in all of the situations we tested. We find that for the same mosquito densities, Ae. aegypti is more effective than Ae. albopictus in transmitting DENV and ZIKV, while the opposite is true for CHIKV.
The results are based on the simulations with the parameter values available in current literature, which may vary for different locations at different times. Our model is a general model that can produce outputs for a specific location, once the data for the location are available to parameterize the model.
Comparisons of the model predictions for Ae. aegypti versus Ae. albopictus must take into account the ratio of mosquitoes to humans. R 0 is sensitive to this ratio and the density of Ae. aegypti mosquitoes is typically higher in urban areas than in rural areas, while the opposite is true for Ae. albopictus mosquitoes [37]. When there are few mosquitoes per human, then R 0 < 1. As the number of mosquitoes increases, then R 0 quickly increases to be greater than one. As the number of mosquitoes per human becomes very large, R 0 eventually decreases in our model where the number of times that humans allow themselves to be bitten is limited to a maximum number of times per day. The rate of decrease depends upon the specific biting rate in Eq (3) being used in the model.
Although wMelPop-infected mosquitoes do not transmit these viruses, the increased death rate of wMelPop-infection has a high fitness cost. It is difficult for wMelPop-infected mosquitoes to survive in the wild mosquito populations because a much larger number of wMelPop-infected mosquitoes needs to be released in order to sustain in the wild mosquito population.
The analysis of the basic reproduction number assumes that when the infections first enter a population, then everyone is fully susceptible to the infection. We derived the effective reproduction number for when the host population is partially immune to new infections, perhaps due to a previous epidemic. The effective reproduction number increases with the susceptibility of humans. When more people are immune to DENV and CHIKV than ZIKV, as happened in the 2015 Zika epidemic, then the numbers of dengue and Chikungunya cases tend to be stable, while the number of Zika cases exploded. Hence, the susceptibility of the human population must be taken into account in future seasonal outbreaks. Our analysis quantified how R e f f ( t ) depends upon a fraction of the population being immune to the infection in a vector-host transmission model.
There are ongoing efforts for releasing Wolbachia-infected mosquitoes in the wild to fight against the spread of these viral infections. Wolbachia-infected mosquitoes could contribute to the reduction of transmission instead of elimination. Besides, the number of Wolbachia-infected mosquitoes released has to exceed the threshold and continual introductions are required. The real-world thresholds for sustaining an epidemic will be greater than the threshold estimates derived for ideal conditions where there is a homogenous population of infected and uninfected mosquitoes.
These field tests suggest that the spatial heterogeneity of the populations must be considered before this model will be appropriate to help guide policy decisions. Also, our simulations are based on an environment of Wolbachia-infected mosquitoes where most of the wild mosquitoes are infected with Wolbachia. When introducing the wMel or wAlbB strains of Wolbachia into a wild mosquito population, it may take several weeks or months for it to reach the equilibrium level and may require several introductions [22]. Furthermore, the model parameter values are based on average estimates from the literature, and not the parameters for a specific location. Before this model can be applied to a specific location, then model parameters, such as the average number of mosquitoes per person, must be estimated for this location.
In future studies, we will couple the model for the spread of Wolbachia [22] with the disease transmission model [23] to evaluate effectiveness of this approach for the situations where the mosquito population is only partially infected with Wolbachia and consider new human arrivals including people who are immune and infectious.
|
10.1371/journal.pbio.3000157 | Multiple functional neurosteroid binding sites on GABAA receptors | Neurosteroids are endogenous modulators of neuronal excitability and nervous system development and are being developed as anesthetic agents and treatments for psychiatric diseases. While gamma amino-butyric acid Type A (GABAA) receptors are the primary molecular targets of neurosteroid action, the structural details of neurosteroid binding to these proteins remain ill defined. We synthesized neurosteroid analogue photolabeling reagents in which the photolabeling groups were placed at three positions around the neurosteroid ring structure, enabling identification of binding sites and mapping of neurosteroid orientation within these sites. Using middle-down mass spectrometry (MS), we identified three clusters of photolabeled residues representing three distinct neurosteroid binding sites in the human α1β3 GABAA receptor. Novel intrasubunit binding sites were identified within the transmembrane helical bundles of both the α1 (labeled residues α1-N408, Y415) and β3 (labeled residue β3-Y442) subunits, adjacent to the extracellular domains (ECDs). An intersubunit site (labeled residues β3-L294 and G308) in the interface between the β3(+) and α1(−) subunits of the GABAA receptor pentamer was also identified. Computational docking studies of neurosteroid to the three sites predicted critical residues contributing to neurosteroid interaction with the GABAA receptors. Electrophysiological studies of receptors with mutations based on these predictions (α1-V227W, N408A/Y411F, and Q242L) indicate that both the α1 intrasubunit and β3-α1 intersubunit sites are critical for neurosteroid action.
| Neurosteroids are cholesterol metabolites produced by neurons and glial cells that participate in central nervous system (CNS) development, regulate neuronal excitability, and modulate complex behaviors such as mood. Exogenously administered neurosteroid analogues are effective sedative hypnotics and are being developed as antidepressants and anticonvulsants. Gamma amino-butyric acid Type A (GABAA) receptors, the principal ionotropic inhibitory neurotransmitter receptors in the brain, are the primary functional target of neurosteroids. Understanding the molecular details of neurosteroid interactions with GABAA receptors is critical to understanding their mechanism of action and developing specific and effective therapeutic agents. In the current study, we developed a suite of neurosteroid analogue affinity labeling reagents, which we used to identify three distinct binding sites on GABAA receptors and to determine the orientation of neurosteroid binding in each site. Electrophysiological studies performed on receptors with mutations designed to disrupt the identified binding sites showed that two of the three sites contribute to neurosteroid modulation of GABAA currents. The distinct patterns of neurosteroid affinity, binding orientation, and effect provide the potential for the development of isoform-specific agonists, partial agonists, and antagonists with targeted therapeutic effects.
| Neurosteroids are cholesterol metabolites produced by neurons [1] and glia [2] in the central nervous system (CNS) that are thought to play important roles in both nervous system development and behavioral modulation [3]. Neurosteroid analogues are also being developed as sedative hypnotics [4], antidepressants [5], and anticonvulsants [6]. Gamma amino-butyric acid Type A (GABAA) receptors, the principal ionotropic inhibitory neurotransmitter receptors in the CNS, have been identified as the primary functional target of neurosteroids. The major endogenous neurosteroids—allopregnanolone and tetrahydroxy-desoxycorticosterone (THDOC)—are positive allosteric modulators (PAMs) of GABAA receptors, potentiating the effects of GABA at nanomolar concentrations and directly activating currents at micromolar concentrations. GABAA receptors are members of the pentameric ligand-gated ion channel (pLGIC) superfamily and are typically composed of two α subunits, two β subunits, and one γ or δ subunit [7]. There are 19 homologous GABAA receptor subunits (including six α, three β and two γ isoforms), with each subunit composed of a large extracellular domain (ECD), a transmembrane domain (TMD) formed by four membrane-spanning helices (TMD1–4), a long intracellular loop between TMD3 and TMD4, and a short extracellular C-terminus. These distinctive structural domains form binding sites for a number of ligands: GABA and benzodiazepines bind to the ECD, picrotoxin to the channel pore [8], and general anesthetics—such as propofol [9, 10], etomidate [11], barbiturates [12], and neurosteroids—to the TMDs [13–18].
Substantial evidence indicates that neurosteroids produce their effects on GABAA receptors by binding to sites within the TMDs [13–15, 19, 20]. Whereas the TMDs of β-subunits are critically important to the actions of propofol and etomidate [11, 21–26], the α-subunit TMDs appear to be essential for neurosteroid action. Mutagenesis studies in α1β2γ2 GABAA receptors identified several residues in the α1 subunit, notably Q241 in TMD1, as critical to neurosteroid potentiation of GABA-elicited currents [14, 27]. More recent crystallographic studies have shown that, in homo-pentameric chimeric receptors in which the TMDs are derived from either α1 [16] or α5 subunits [17], the neurosteroids THDOC and pregnanolone bind in a cleft between the α-subunits, with the C3-hydroxyl substituent of the steroids interacting directly with α1Q241. Neurosteroids are PAMs of these chimeric receptors, and α1Q241L and α1Q241W mutations eliminate this modulation. These studies posit a single critical binding site for neurosteroids that is conserved across the six α-subunit isoforms [14, 27].
A significant body of evidence also suggests that neurosteroid modulation of GABAA receptors may be mediated by multiple sites. Site-directed mutagenesis identified multiple residues that affect neurosteroid action on GABAA receptors, suggestive of two neurosteroid binding sites, with one site mediating potentiation of GABA responses and the other mediating direct activation [14, 27]. Single channel electrophysiological studies as well as studies examining neurosteroid modulation of [35S]t-butylbicyclophosphorothionate (TBPS) binding, have also identified multiple distinct effects of neurosteroids with various structural analogues producing some or all of these effects, consistent with multiple neurosteroid binding sites [28–30]. Finally, neurosteroid photolabeling studies in the bacterial pLGIC, Gloeobacter ligand-gated ion channel (GLIC), demonstrate two neurosteroid binding sites per monomer [31], one analogous to the canonical intersubunit site and one located in an intrasubunit pocket previously shown to bind propofol [32, 33] and the inhalational anesthetics [33, 34]. Both of these sites contribute to neurosteroid modulation of GLIC currents, suggesting the possibility of analogous sites in GABAA receptors.
We have developed a suite of neurosteroid analogue photolabeling reagents with photolabeling groups positioned around the neurosteroid ring structure to identify all of the neurosteroid binding sites on GABAA receptors and determine the orientation of neurosteroid binding within each site. Photolabeling was performed in membranes from a mammalian cell line that stably expresses α1His-FLAGβ3 receptors (rather than in detergent-solubilized receptors) to optimize the likelihood that the receptors were in native conformations and environment. Finally, we deployed a middle-down mass spectrometry (MS) approach, coupled with a stable-heavy isotope encoded click chemistry tag for neurosteroid-peptide adduct identification to circumvent challenges associated with MS identification (predominantly neutral loss) and quantification of neurosteroid-peptide adducts [35]. Using these approaches, we have identified three clusters of neurosteroid-photolabeled residues on the human α1β3 GABAA receptor. Computational docking studies, guided by the photolabeling data, were used to describe three binding sites and the orientation of the neurosteroids within each site. The docking studies were also used to predict critical residues to test the contribution of each of these sites to neurosteroid modulation of GABAA currents. Site-directed mutagenesis of these sites and electrophysiological studies indicate that at least two of three structurally distinct sites contribute to allosteric modulation of GABA currents.
Allopregnanolone (3α-hydroxy-5α-pregnan-20-one) is a potent, endogenous PAM of GABAA receptors (Fig 1A). We synthesized three photolabeling analogues of allopregnanolone in which photolabeling moieties were placed at various positions around the steroid backbone. KK123 has a 6-diazirine photolabeling group on the C5-C6-C7 edge of the sterol, which is a likely binding interface with α-helices [36] and minimally perturbs neurosteroid structure [37]. KK123 is, however, an aliphatic diazirine and, as such, may preferentially label nucleophilic amino acids [38]. The two other reagents, KK202 and KK200, incorporate a trifluoromethylphenyl-diazirine (TPD) group at either the 3- or 17-carbon. These were designed to sample the space in the plane of the steroid off either the A-ring (KK202) or the D-ring (KK200). Following UV irradiation, TPD groups generate a carbene which can insert into any bond [39, 40]. Thus, while the TPD groups are bulky and removed several angstroms from the neurosteroid pharmacophore, they should form an adduct precisely at their binding site in the GABAA receptor. Where feasible (KK123, KK202), an alkyne was incorporated in the photolabeling reagents to allow attachment of a fluorophore, purification tag, or an MS reporter tag (FLI-tag) via click chemistry [35].
A useful photoaffinity labeling reagent must bind to the same site on a protein as the ligand it mimics and should produce the same effects on protein functions. To determine whether our photoaffinity labeling reagents mimic allopregnanolone as modulators of GABAA receptor function, we assessed modulation of α1β3 GABAA receptors currents in Xenopus laevis oocytes, and enhancement of [3H]muscimol binding in human embryonic kidney (HEK) cell membranes expressing α1β3 GABAA receptors. KK123 enhanced GABA-elicited (0.3 μM) currents 4.2 ± 3.3-fold at 1 μM (n = 5 cells) and 8.2 ± 6.7-fold at 10 μM (n = 7). KK123 (10 μM) also directly activated α1β3 GABAA receptors, eliciting 6.3% ± 3.8% (n = 5) of the maximum current elicited by a saturating concentration of GABA. KK123 potentiation of GABA-elicited currents and direct activation were absent in α1Q242Lβ3 GABAA receptors, indicating that KK123 closely mimics the actions of allopregnanolone (the human α1Q242L mutation is equivalent to rat α1Q241L and is known to selectively prevent neurosteroid action (Fig 1B and Table 1) [14, 27]). KK200 and KK202 also potentiated GABA-elicited currents at 1 and 10 μM and directly activated the channels at 10 μM (Table 1). Positive allosteric modulation by KK202 was somewhat surprising, given that an ether-linked TPD group replaces the 3α-OH group thought to be critical for neurosteroid action [41, 42]. While the effects of KK200 were abolished in α1Q242Lβ3 receptors, the potentiation by KK202 was reduced by 50% in α1Q242Lβ3 receptors, suggesting that KK202 may have actions at both the canonical neurosteroid site and other binding sites. Because photolabeling experiments were performed in membranes prepared from cells expressing α1β3 GABAA receptors, we also examined the ability of the photolabeling reagents to enhance [3H]muscimol binding in these membranes. A stable HEK-293 cell line was established with tetracycline-inducible expression of human α1His-FLAG β3 GABAA receptors (See Materials and methods); receptor density in these membranes was 20–30 pmol [3H]muscimol binding/mg membrane protein. Consistent with previous determinations [43], the average stoichiometry of the receptors was estimated at two α1 subunits and three β3 subunits using MS label-free quantitation [44] (spectral count). Allopregnanolone enhanced [3H]muscimol binding to these recombinant receptors 4-fold with a half maximal effective concentration (EC50) of 3.9 ± 5.6 μM (S1 Fig). KK123, KK200, and KK202 all enhanced [3H]muscimol binding with EC50 values similar to or lower than allopregnanolone (S1 Fig). Collectively, the electrophysiology and radioligand binding data indicate that KK123, KK200, and KK202 are functional mimetics of allopregnanolone.
To determine whether KK123—which contains an aliphatic diazirine—photolabels GABAA receptors, we utilized the butynyloxy (alkyne) moiety on KK123 to attach a biotin purification tag for selective enrichment of photolabeled GABAA receptor subunits. HEK-293 cell membranes containing α1β3 GABAA receptors were photolabeled with 15 μM KK123, solubilized in SDS, and coupled via Cu2+-catalyzed cycloaddition to MQ112 (S2A Fig), a trifunctional linker containing an azide group for cycloaddition, biotin for biotin-streptavidin affinity purification, and a cleavable azobenzene group for elution of photolabeled proteins. The photolabeled-MQ112-tagged receptors were bound to streptavidin beads and eluted by cleavage of the linker with sodium dithionite. The purified, photolabeled GABAA receptor subunits were assayed by western blot using anti-α1 and anti-β3. A band at 52 kDa was observed with both α1 and β3 subunit antibodies in the KK123 photolabeling group (S2B Fig), indicating that both α1 and β3 subunits are photolabeled by KK123. In control samples photolabeled with ZCM42—an allopregnanolone photolabeling analogue containing a diazirine at the 6-carbon but no alkyne (S2C Fig)—neither α1 nor β3 subunits were purified. These data indicated that KK123 can photolabel both α1 and β3 subunits and is thus an appropriate reagent to use for site identification. A 35 kDa band was intermittently observed in replicate anti-α1 western blots (S2B Fig); this is likely to be a proteolytic fragment of the α1-subunit that retains the antibody-recognition epitope but was not further analyzed.
Identification of sterol adducts in hydrophobic peptides has been impeded by multiple challenges, including peptide insolubility during sample digestion, ineffective chromatographic separation of hydrophobic TMD peptides, and neutral loss of sterol adducts from small hydrophobic peptides during ionization and fragmentation. To circumvent these problems, we employed middle-down MS to analyze GABAA receptor TMD peptides and their sterol adducts. This approach identifies each TMD as a single, large peptide and attenuates neutral loss of adduct, facilitating identification of the sites of neurosteroid incorporation. In our studies, α1His-FLAGβ3 GABAA receptors were photolabeled in native HEK cell membranes. The photolabeled proteins were then solubilized in n-dodecyl-β-D-maltoside (DDM)-containing lysis buffer. The pentameric GABAA receptors were purified using anti-FLAG agarose beads, and eluted receptors were digested with trypsin in the presence of the MS-compatible detergent DDM. These conditions generated peptides containing each of the GABAA receptor TMDs in their entirety. The peptides were separated using PLRP-S nano-liquid chromatography and analyzed on a Thermo ELITE orbitrap mass spectrometer. This workflow (S3 Fig) minimized protein/peptide aggregation, simplified MS1-level identification of TMD-sterol adducts, and optimized fragmentation of TMD peptides and their adducts. All eight of the TMD peptides were reliably sequenced with 100% residue-level coverage. In addition, the covalent addition of neurosteroid to the TMD peptides increased the hydrophobicity of TMD peptides and shifted their chromatographic elution to later retention times (S3 Fig). The delayed retention time was used as a critical criterion for identification of photolabeled peptides.
Two photolabeled peptides were found in the mass spectra of tryptic digests of α1β3 GABAA receptors photolabeled with KK123 (Fig 2A and S4 Fig). A KK123 adduct of the α1-TM4 peptide, 398IAFPLLFGIFNLVYWATYKK123LNREPQLK423 (m/z = 875.503, z = 4), was identified (add weight of KK123 = 316.27). Site-defining ions in the fragmentation spectra identified the site of KK123 insertion as Y415, at the C-terminus of α1-TM4 (underlined in the sequence; see Fig 2A). In a separate series of experiments, α1β3 receptors were photolabeled with KK123, which was then coupled to FLI-tag using click chemistry. FLI-tag, an azide-containing tag, adds both charge and a heavy/light stable isotope pair to a photolabeled peptide, enhancing identification by creating doublets in the MS1 spectra [35]. MS1 level search for pairs of ions differing by 10.07 mass units found two peptide ion features (m/z = 1,073.246 and m/z = 1,076.580, z = 3) that had identical chromatographic retention times (Fig 2B). Fragmentation spectra revealed both of these peptides as β3-TM4 peptide (426IVFPFTFSLFNLVYWLYKK123YVN445) with a KK123-FLI-tag adduct (adduct mass = 672.432 and mass = 682.441) on Y442 (Fig 2C). In the fragmentation spectrum, ions containing KK123 plus light FLI-tag (Fig 2C, black) were different by 10.07 mass units from the corresponding fragment ions from KK123 plus heavy FLI-tag (Fig 2C, red), confirming that KK123 photolabels Y442 of the β3 subunit. β3-Y442 is located on the C-terminus of β3-TM4 in a homologous position to α1-Y415, the KK123 photolabeling site in α1-TM4 (Fig 2D, upper right panel). Thus, KK123 labeling data identified two discrete sites, one in α1 and the other in β3. We employed additional photolabeling reagents containing TPD groups arrayed around the sterol backbone to confirm whether the KK123-labeled residues represent neurosteroid binding sites and to determine the orientation of the neurosteroids in these sites.
KK200, which has a TPD photolabeling group attached at C17 on the steroid backbone, has been previously used to map neurosteroid binding sites on GLIC [31]. Analysis of α1β3 receptors photolabeled with 15 μM KK200 detected two photolabeled TMD peptides: an α1-TM4 peptide, 398IAFPLLFGIFNKK200LVYWATYLNREPQLK423, was photolabeled with KK200 (m/z = 898.002; z = 4); site-defining ions in the fragmentation spectra identified N408 as the modified residue (Fig 3A). The N408 residue (N407 in rat) has previously been shown to be critical to neurosteroid potentiation of GABA-elicited currents [14, 15]. A β3-TM3 peptide, 280AIDMYLMGCNEM+DTTFVFVFLALLEYAFVNYIFFGRKK200GPQR313 (m/z = 1,188.352; z = 4; N-ethylmaleimide [NEM]; 1, 4-dithiothreitol [DTT]; alkylation adduct), was also photolabeled with KK200. Fragmentation spectra narrowed the possible sites of adduction to G308 or R309, both at the junction of TM3 with the M3–M4 intracellular loop (Fig 3B).
Analysis of GABAA receptors photolabeled with KK202 (Fig 3C and 3D), identified two photolabeled peptides eluting two minutes apart. Both peptides were identified as the β3-TM3 peptide, 278VKAIDMYLMGCNEMFVFVFLALLEYAFVNYIFFGRGPQR313 (m/z = 811.453, z = 6). Fragmentation spectra of the earlier eluting peptide localized labeling to a three-residue sequence, 278VKA280, at the N-terminus of β3-TM3 (Fig 3C). The fragmentation spectrum of the later eluting peptide, identified L294 as the site of adduction (Fig 3D). (The different retention time of the two photolabeled peptides is likely due to differences in peptide conformation and surface hydrophobicity resulting from incorporation of the photolabeling reagent into different residues.)
An important test of whether the photolabeled sites constitute specific allopregnanolone binding sites is the ability of excess allopregnanolone to competitively prevent photolabeling. Photolabeling studies for site identification were performed using 15 μM photolabeling reagent and achieved levels of labeling efficiency varying from 0.06% to 3.0% (S1 Table). Because allopregnanolone has limited aqueous solubility (about 30 μM) and a large competitor excess is needed to demonstrate competition (particularly with an irreversibly bound ligand), we were limited to studying competition at the photolabeled residues that could be detected following photolabeling at a concentration of 3 μM. Accordingly, we measured the photolabeling efficiency obtained following photolabeling of α1β3 GABAA receptors with 3 μM KK123, KK200, or KK202 in the presence or absence of 30 μM allopregnanolone. KK123 photolabeled both α1-Y415 (0.77% efficiency) and β3 -Y442 (0.37% efficiency). For both of these residues, photolabeling was reduced by >90% in the presence of excess allopregnanolone (Fig 4A). KK200 photolabeled β3 -G308/R309 (0.19% efficiency), and labeling was reduced by 98% in the presence of allopregnanolone. KK202 labeled both β3-L294 (0.29% efficiency) and β3-278VKA280 (0.21% efficiency) in TM3; labeling of both of these sites was undetectable in the presence of 30 μM allopregnanolone. Studies were also performed to determine whether the orthosteric agonist GABA (1 mM) enhanced photolabeling by 3 uM KK123 or KK200. Labeling efficiency was not significantly enhanced in the presence of GABA. This suggests that there is a small difference in neurosteroid affinity for closed versus open/desensitized states, which is consistent with the fact that neurosteroids have very low efficacy as direct activators of GABAA receptors [45].
Modification of ligand analogues with labeling groups at different locations has been used to determine the orientation of the ligands within their binding pockets [46]. Here, the six residues photolabeled by KK123, KK200, and KK202 were examined in a model of the α1β3 receptor created by threading the aligned sequence of the α1 subunit on the structure of the β3 subunit (PDB 4COF) [47]. The photolabeling sites grouped into the following three clusters: cluster 1 (brown circle), β3-L294 (KK202) and β3-G308/R309 (KK200); cluster 2 (red circle), α1-Y415 (KK123) and α1-N408 (KK200); and cluster 3 (blue circle), β3-Y442 (KK123) and β3-278VKA280 (KK202) (Fig 5A).
In cluster 1 (brown circle, Fig 5A and 5D), β3-L294 faces into the β(+)/α(−) intersubunit cleft, and G308/R309 is at the junction between the bottom of TM3 and the TM3–4 intracellular loop. G308/R309 is two α-helical turns below β3-F301 (i.e., toward the intracellular terminus of TM3), a residue previously photolabeled by 6-azi pregnanolone in β3 homomeric receptors [13]. These data support neurosteroid binding in the β(+)/α(−) interface, consistent with the canonical THDOC and pregnanolone binding sites identified in crystal structures of α1(+)/α1(−) interfaces in chimeric proteins [16] and in substituted cysteine modification protection studies of α1β2γ2 receptors [18]. The pattern of labeling also indicates that the A-ring of the steroid is oriented upwards in the intersubunit cleft toward the center of the membrane, the D-ring is pointing toward the intracellular termini of the TMDs, and the C5-C6-C7 edge of the steroid is pointing toward the β3(+) side of the cleft. Cluster 1 corresponds to a β3(+)/α1(−) intersubunit site.
In cluster 2 (red circle, Fig 5A and 5C), N408 and Y415 are both on the C-terminal end of α1-TM4, facing toward TM1 within the same α1 subunit, consistent with an α1 intrasubunit neurosteroid binding site. N408, the residue labeled by the C17-TPD of KK200, is two α-helical turns closer to the center of TM4 than is Y415, the residue labeled by the C6-diazirine of KK123. This labeling pattern suggests that neurosteroids orient in this site with the A-ring pointing toward the ECD and the D-ring facing to the center of the TMD. Cluster 2 corresponds to an α1 intrasubunit site.
In cluster 3 (blue circle, Fig 5A and 5B), Y442 is located at the C-terminal end of β3-TM4, and 278VKA280 is located on the TM2–TM3 loop near the extracellular end of β3-TM3. The adjacency of these two photolabeling sites suggests an intrasubunit neurosteroid binding site at the extracellular end of β3, analogous to the α1 intrasubunit site. The labeling of 278VKA280 in the extracellular loop by the C3-TPD group of KK202 suggests that neurosteroids orient in this site with the A-ring facing the ECD. Cluster 3 corresponds to a β3 intrasubunit site.
A homology model of the α1β3 GABAA receptor based on the structure of a β3 homomeric GABAA receptor (PDB 4COF) [47] was used to examine the preferred energetic poses of neurosteroid binding to the three binding sites. The homology model was embedded in a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) bilayer and the structure refined by molecular dynamics. We then docked each of the three photoaffinity labeling reagents as well as allopregnanolone to each of the proposed binding sites, using a time course series of snapshots from the simulation trajectory to account for receptor flexibility. All of the neurosteroid photolabeling reagents docked in the three sites; the identified sites are relatively shallow with respect to the protein–lipid interface. Moreover, the neurosteroid analogues were all found to adopt multiple poses in each of the sites with minimal energy differences between the poses (see Materials and methods). Photolabeling data combined with the docking scores (binding energy) and population of a given pose were used to guide selection of the preferred steroid orientation in each site.
In the α1 intrasubunit site, the poses clustered between TM1 and TM4. The preferred pose (Fig 5C) for allopregnanolone (lowest energy cluster of poses) shows the A-ring oriented toward the ECD with the walls of the predicted binding site lined on one side by N408 and Y415 and on the other by V227. Docking of KK200 in this site has a similar orientation with the A-ring oriented toward the ECD and the TPD group on the D-ring proximal to N408 (Fig 6A). Docking of KK-123 shows a preferred pose in which the A-ring is oriented toward the ECD and the C6-diazirine proximal to Y415 (Fig 6B). These data elucidate a prior finding that mutations to N408 and Y411 eliminate potentiation by steroid analogues that lack a hydrogen bonding group on the D-ring [48].
In the β3 intrasubunit site, the poses are clustered between TM3 and TM4. Allopregnanolone preferred a pose with the A-ring oriented toward the ECD near Y442 and the D-ring proximal to V290 (Fig 5B). KK123 was found to dock at the top of the TM helices with the A-ring oriented toward the ECD, placing the 6-diazirine in proximity to Y442 (Fig 6C). KK202 was found to dock in a similar orientation but lower in the TM region, with the TPD group in proximity to A280 and Y442 (Fig 6D).
In the intersubunit site, the preferred pose for allopregnanolone was one of the lowest energy clusters of poses with the A-ring proximal to α1-Q242 (equivalent to rat α1-Q241), the D-ring pointing toward the cytoplasmic termini of the TMDs, and the D-ring facing β3-F301 (Fig 5D). Docking of KK200 showed a similar orientation although shifted slightly upwards toward the ECD, placing the A-ring near α1-Q242,the benzene ring of the TPD near β3-F301, and the diazirine in proximity to G308 (Fig 6E). The preferred pose of KK202 was closer to TM3 of the β3 subunit with the A-ring near α1-Q242 and the D-ring near the β3-F301 placing the diazirine in proximity to β3-L294 (Fig 6F).
The orientation of allopregnanolone docked in our α1β3 model is nearly identical to the orientation of THDOC in the crystal structure of α1-GLIC [16] (PDB 5OSB). As a confirmation of our docking, we also docked allopregnanolone to the apo-neurosteroid crystal structures of the α1-GLIC [16] and α5-β3 chimeric [17] proteins (PDB 5OSA and PDB 5OJM, respectively) (S5 Fig). The preferred poses for allopregnanolone in the β3-α1 intersubunit site are nearly identical between the three models. The preferred poses are also very similar in the α1 intrasubunit site between our α1β3 homology model and the structures of the α-homomeric TMDs. The A-ring/D-ring orientation of allopregnanolone in the three neurosteroid sites was consistent with the orientations identified by the photolabeling data in all of the GABAA receptor structures. The calculated binding energies from the docking studies (S4 Table) indicate that the rank order of allopregnanolone affinity for the three sites is β3/α1 intersubunit site > α1 intrasubunit site > β3 intrasubunit site.
The β3-α1 intersubunit binding site identified in our photolabeling studies has been extensively validated by site-directed mutagenesis as a functionally important site. Mutations on the α(−) side of the interface, including Q241(rat)L/W and W246(rat)L, have been shown to eliminate neurosteroid potentiation and gating of α1β2γ2 GABAA receptors [14, 27, 49]; mutations on the β(+) side of the interface, including F301A and L297A, have also been shown to partially reduce neurosteroid effect [17]. In the current study, we showed that α1Q242Lβ3 prevented the action of allopregnanolone, KK123, and KK200 while reducing the effect of KK202 in α1β3 receptors, confirming the β–α interface as a functionally significant neurosteroid binding site and validating the relevance of our photolabeling reagents (Fig 1B).
Based on computational simulation and docking results, we also identified residues in the proposed α1- and β3-intrasubunit binding sites that we predicted could be involved in allopregnanolone binding or action (S2 Table and S3 Table for all mutated subunits tested). N408 and Y411 in α1-TM4 line one side of the putative α1 intrasubunit site, and V227 in α1-TM1 lines the other (Fig 5C). α1N407(rat)A and α1Y410(rat)W mutations have previously been shown to prevent neurosteroid potentiation of GABA-elicited currents in α1β2γ2 GABAA receptors [15]. Our data confirm that the double mutant α1N408A/Y411Fβ3 substantially reduces allopregnanolone potentiation of GABA-elicited currents (Fig 4B and 4C1, ***p < 0.001 versus α1β3 wild-type). Allopregnanolone (1 μM) potentiation of GABA-elicited currents and direct activation (10 μM) of α1V227Wβ3 receptors was also significantly reduced in comparison to α1β3 wild-type (*p < 0.05 and **p < 0.01; Fig 4B, 4C1 and 4C2). To test whether these mutations selectively affected neurosteroid actions, we also compared the effect of propofol in α1V227Wβ3 and α1N408A/Y411Fβ3 to its effect on wild-type α1β3 receptors. Propofol action was not different between the mutant and wild-type receptors, indicating a selective effect on neurosteroid action (Fig 4B, 4C3 and 4C4). The finding that multiple mutations lining the α1-intrasubunit binding pocket selectively reduce allopregnanolone action buttresses the evidence that the photolabeled residues identify a specific, functionally important neurosteroid binding site.
Multiple mutations within the putative β3-intrasubunit binding site were also tested. However, none of the mutations significantly altered potentiation or activation by allopregnanolone (S2 Table and S3 Table for all of the mutations that were tested). These data suggest that allopregnanolone occupancy of the β3 intrasubunit site does not contribute to channel gating. Direct activation of α1β2Y284Fγ2 receptors by THDOC has previously been shown to be markedly reduced in comparison to wild-type receptors [15], although we found no significant effect of the β3-Y284 mutation in α1β3 receptors (S2 Table and S3 Table). The difference in results between experiments in α1β3 and α1β2γ2 GABAA receptors suggests possible receptor subtype specificity in the functional effects of neurosteroid binding at a β-intrasubunit site.
Collectively, the photolabeling, modeling, and functional data indicate that heteropentameric α1β3 GABAA receptors contain at least seven binding sites for neurosteroids, of three different types. The use of multiple photolabeling reagents also enabled determination of the orientation of neurosteroids in each proposed class of sites. At least two of these classes are involved in producing the allosteric effect of steroids, the β3-α1 intersubunit site (two copies per receptor) and the α1 intrasubunit site (two copies). Mutations of residues in the proposed β3 intrasubunit site (three copies) had no effect on modulation by allopregnanolone although residues were labeled by two photolabeling reagents and labeling was prevented by excess allopregnanolone. Accordingly, the functional significance of this proposed site is not known.
Previous, site-directed mutagenesis studies using electrophysiology readout identified multiple residues, including α1-Q241, N407, Y410, T236, and β3-Y284, that selectively contribute to the positive allosteric effects of neurosteroids [14, 27]. Based on homology to the structure of the muscle nicotinic acetylcholine receptor [50], it was hypothesized that there are two neurosteroid binding sites on GABAA receptors: an α1-intrasubunit site spanning Q241 and N407 and an intersubunit site between β3-Y284 and α1-T236. Subsequent data [16–18] have clearly established the existence of a β–α intersubunit site. Our photolabeling experiments and homology modeling now show that the previously identified residues contribute to multiple distinct neurosteroid binding sites, albeit differently than originally proposed. It is noteworthy that the α1-intrasubunit site was not identified in the X-ray crystallographic structures of α1-GLIC chimeras bound with THDOC or the α5-β3 chimera bound with pregnanolone. This is likely because the proteins with steroid bound in the intrasubunit site did not form stable crystals.
Mutations in either the β3-α1 intersubunit site or the α1-intrasubunit site can ablate both potentiation and direct activation by allopregnanolone, indicating that these are not distinct sites mediating potentiation and direct activation. The data also do not conform to simple energetic additivity for the two sites. The observation that mutations in either binding site can largely eliminate neurosteroid effect indicates that these two sites do not function completely independently and suggests allosteric interaction between the two sites. Development of site-selective neurosteroid analogues (PAMs and antagonists) should facilitate clarification of the mechanisms of allosteric interaction between these two sites.
In light of the demonstration of multiple neurosteroid binding sites in α1β3 GABAA receptors, the possibility of additional isoform-specific sites must be considered. The strong sequence homology between the TMDs of the six α-subunits and three β-subunits suggests that there will not be large isoform differences in the intersubunit site [27]. In contrast, the contribution of ECD residues to the α- and β-intrasubunit sites suggests possible isoform-specific differences. The sequence homology between the γ and δ subunits and α and β subunits suggests that there may also be intrasubunit neurosteroid binding sites in these isoforms. Identification of a neurosteroid binding site on a δ-subunit would be of particular relevance because GABAA receptors containing δ-subunits are particularly sensitive to neurosteroids [51–53]. High-resolution, cryo-electron microscopy structures of α1β3γ2 GABAA receptors [54–56] have been published since initial submission of this work. The structural homology between γ2 subunits and α and β subunits suggests that there may also be intrasubunit neurosteroid binding sites in the γ2 subunit. The existence of multiple sites in which neurosteroids bind with different orientation may also offer some explanation for the difficulty in identifying neurosteroid antagonists [57] and for the differences in single-channel electrophysiological effects of various neurosteroid analogues [28, 30]. The possibility of multiple isoform-specific sites with distinct patterns of neurosteroid affinity, binding orientation, and effect offers the exciting potential for the development of isoform-specific agonists, partial agonists, and antagonists with targeted therapeutic effects.
The human α1 and β3 subunits were subcloned into pcDNA3 for molecular manipulations and cRNA synthesis. Using QuikChange mutagenesis (Agilent), a FLAG tag was first added to the α1 subunit then an 8xHis tag was added to generate the following His-FLAG tag tandem (QPSLHHHHHHHHDYKDDDDKDEL), inserted between the fourth and fifth residues of the mature peptide. The α1 and β3 subunits were then transferred into the pcDNA4/TO and pcDNA5/TO vectors (ThermoFisher Scientific, Waltham, MA), respectively, for tetracycline-inducible expression. For X. laevis oocytes, point mutations were generated using the QuikChange site-directed mutagenesis kit (Agilent Technologies, Santa Clara, CA) and the coding region fully sequenced prior to use. The cDNAs were linearized with Xba I (NEB Labs, Ipswich, MA), and the cRNAs were generated using T7 mMessage mMachine (Ambion, Austin, TX).
The tetracycline-inducible cell line HEK T-RexTM-293 (ThermoFisher) was cultured under the following conditions: cells were maintained in DMEM/F-12 50/50 medium containing 10% fetal bovine serum (tetracycline-free, Takara, Mountain View, CA), penicillin (100 units/ml), streptomycin (100 g/ml), and blastcidine (2 μg/ml) in a humidified atmosphere containing 5% CO2. Cells were passaged twice each week, maintaining subconfluent cultures. Stably transfected cells were cultured as above with the addition of hygromycin (50 μg/ml) and Zeocin (20 μg/ml).
A stable cell line was generated by transfecting HEK T-RexTM-293 cells with human α1-8x His-FLAG pcDNA4/TO and human β3 pcDNA5/TO in a 150 mm culture dish, using the Effectene transfection reagent (Qiagen). Two days after transfection, selection of stably transfected cells was performed with hygromycin and zeocin until distinct colonies appeared (usually after two weeks). Medium was exchanged several times each week to maintain antibiotic selection.
Individual clones (about 65) were selected from the dish and transferred to 24-well plates for expansion of each clone selected. When the cells grew to a sufficient number, about 50% confluency, they were split into two other plates, one for a surface ELISA against the FLAG epitope and a second for protein assay, to normalize surface expression to cell number [58]. The best eight clones were selected for expansion into 150 mm dishes, followed by [3H]muscimol binding. Once the best expressing clone was determined, the highest-expressing cells of that clone were selected through fluorescence-activated cell sorting (FACS).
FACS was done against the FLAG epitope, using a phycoerythrin (PE)-conjugated anti-FLAG antibody. Fluorescent-activated cells (1 ml containing about 10 million cells) were sorted on the AriaII cell sorter (Washington University Pathology Core), collecting 0.5% of the highest-fluorescing cells in a culture tube containing complete medium. The cells were plated in a 35 mm dish and expanded until a near confluent 150 mm dish was obtained. Cells were enriched for expression by FACS three times. A final FACS was performed to select individual cells into a 96-well plate, which resulted in only 10 colonies of cells. These colonies were expanded and assayed for [3H]muscimol binding; the highest-expressing clone was used for experiments.
Stably transfected cells were plated into fifty 150 mm dishes. After reaching 50% confluency, GABA receptors were expressed by inducing cells with 1 μg/ml of doxycycline with the addition of 5 mM sodium butyrate. Cells were harvested after 48 to 72 hours after induction.
HEK cells, after tetracycline induction, grown to 70%–80% confluency, were washed with 10 mM sodium phosphate/proteinase inhibitors (Sigma-Aldrich, St. Louis, MO) two times and harvested with cell scrapers. The cells were washed with 10 mM sodium phosphate/proteinase inhibitors and collected by centrifugation at 1,000 g at 4°C for 5 minutes. The cells were homogenized with a glass mortar Teflon pestle for 10 strokes on ice. The pellet containing the membrane proteins was collected after centrifugation at 34,000 g at 4°C for 30 minutes and resuspended in a buffer containing 10 mM potassium phosphate and 100 mM KCl. The protein concentration was determined with micro-BCA protein assay and stored at −80°C.
[3H]muscimol binding assays were performed using a previously described method with minor modification [59]. Briefly, HEK cell membranes proteins (50 μg/ml final concentration) were incubated with 1–2 nM [3H]muscimol (30 Ci/mmol; PerkinElmer Life Sciences), neurosteroid in different concentrations (1 nM-10 μM), binding buffer (10 mM potassium phosphate, 100 mM KCl [pH 7.5]), in a total volume of 1 ml. Assay tubes were incubated for 1 hour at 4°C in the dark. Nonspecific binding was determined by binding in the presence of 1 mM GABA. Membranes were collected on Whatman/GF-C glass filter paper using a Brandel cell harvester (Gaithersburg, MD). To determine the Bmax of [3H]muscimol binding, 100 μg/ml of proteins were incubated with 250 nM [3H]muscimol, with specific activity reduced to 2 Ci/mmol, for 1 hour at 4°C in the dark. The membranes were collected on Whatman/GF-B glass filter papers using manifold. Radioactivity bound to the filters was measured by liquid scintillation spectrometry using Bio-Safe II (Research Products International Corporation). Each data point was determined in triplicate.
For all the photolabeling experiments, 10–20 mg of HEK cell membrane proteins (about 300 pmol [3H]muscimol binding) were thawed and resuspended in buffer containing 10 mM potassium phosphate, 100 mM KCl (pH 7.5) at a final concentration of 1.25 mg/ml. For photolabeling site identification experiments, 15 μM neurosteroid photolabeling reagent was added to the membrane proteins and incubated on ice for 1 hour. For the photolabeling competition experiments, 3 μM neurosteroid photolabeling reagent in the presence of 30 μM allopregnanolone or the same volume of ethanol was added for incubation. The samples were then irradiated in a quartz cuvette for 5 minutes, by using a photoreactor emitting light at >320 nm [59]. The membrane proteins were then collected by centrifugation at 20,000 g for 45 minutes. All of the photolabeling experiments to identify sites of neurosteroid photolabeling were performed at least three times. The photolabeled peptides and residues described in the text were all observed in replicate experiments.
The amount of 10 mg of KK123 or ZCM42 photolabeled HEK membrane proteins were solubilized in 1 ml 2% SDS/PBS and incubated at room temperature for 2 hours. The protein lysate was collected by centrifugation at 21,000 g for 30 minutes. FLI-tag was clicked to the KK123- or ZCM-photolabeled proteins at room temperature overnight in PBS buffer containing 2% SDS, 100 μM FLI-tag [35], 2.5 mM sodium ascorbate, 250 μM Tris [(1-benzyl-1H-1,2,3triazol-4-yl)methyl]amine, and 2.5 mM CuSO4. The amount of 1% Triton/PBS was added to the protein lysate to an SDS final concentration of 0.05%. The protein lysate was loaded onto a streptavidin agarose column. The flow through was reloaded to the column two times or till the flow through was colorless and the streptavidin column was dark orange yellow. The column was washed with 10 ml 0.05% Triton/PBS and eluted by 10 ml 100 mM sodium dithionite/0.05%Triton/PBS. The column was turned into colorless after elutions. The eluted proteins were concentrated into 100 μl with 30 kDa cutoff Centricon apparatus. The supernatant of the Centricon tube was added into SDS-sample loading buffer, loaded to a 10% SDS-PAGE, and transferred to a PVDF membrane, followed by western blot with polyclonal rabbit anti-α1 raised against a peptide mapping within a cytoplasmic domain of human GABAR α1 subunit [60] (Santa Cruz Biotechnology) or monoclonal anti-β3 antibody against 370–433 of mouse GABAR β3 subunit [61] (NeuroMab).
The photolabeled membrane proteins were resuspended in lysis buffer containing 1% DDM, 0.25% cholesteryl hemisuccinate (CHS), 50 mM Tris (pH 7.5), 150 mM NaCl, 2 mM CaCl2, 5 mM KCl, 5 mM MgCl2, 1 mM EDTA, and 10% glycerol at a final concentration of 1 mg/ml. The membrane protein suspension was homogenized using a Teflon pestle in a motor-driven homogenizer and incubated at 4°C overnight. The protein lysate was centrifuged at 20,000 g for 45 minutes, and supernatant was incubated with 0.5 ml anti-FLAG agarose (Sigma) at 4°C for 2 hours. The anti-FLAG agarose was then transferred to an empty column, followed by washing with 20 ml washing buffer (50 mM triethylammonium bicarbonate and 0.05% DDM). The GABAA receptors were eluted with ten 1-ml 200 μg/ml FLAG peptide and 100 μg/ml 3X FLAG (ApexBio) in the washing buffer. The 10 ml effective elutions containing GABAA receptors (tested by western blot with anti-α1 or anti-β3 antibody) were concentrated by 100 kDa cutoff Centricon filters into 0.1 ml.
The purified GABAA receptors (100 ul) were reduced by 5 mM tris (2-carboxyethyl) phosphine (TCEP) at for 30 minutes followed by alkylation with 7.5 mM NEM for 1 hour in the dark. The NEM was quenched by 7.5 mM DTT for 15 minutes. These three steps were done at room temperature. Eight μg of trypsin was added to the protein samples and incubated at 4°C for 7–10 days. The digest was terminated by adding formic acid (FA) in a final concentration of 1%. The samples were then analyzed by an OrbiTrap ELITE mass spectrometer (ThermoFisher) as in previous work [13, 31] with some modifications. Briefly, a 20 μl aliquot was injected by an autosampler (Eksigent) at a flow rate of 800 nl/min onto a home-packed polymeric reverse phase PLRP-S column (Agilent, 12 cm × 75 μm, 300 Å). An acetonitrile (ACN) 10%–90% concentration gradient was applied in the flow rate of 800 nl/min for 145 minutes to separate peptides. Solvent A was 0.1% FA/water, and solvent B was 0.1%FA/ACN. The ACN gradient was as follows: isocratic elution at 10% solvent B, 1–60 minutes; 10%–90% solvent B, 60–125 minutes; 90% solvent B, 125–135 minutes; 90%–10% solvent B, 135–140 minutes; isocratic solvent B, 140–145 minutes. For the first 60 minutes, a built-in divert valve on the mass spectrometer was used to remove the hydrophilic contaminants from the mass spectrometer. The survey MS1 scans were acquired at acquired at high resolution (60,000 resolution) in the range of m/z = 100–2,000, and the fragmentation spectra were acquired at 15,000 resolution. Data-dependent acquisition of the top 20 MS1 precursors with exclusion of singly charged precursors was set for MS2 scans. Fragmentation was performed using collision-induced dissociation or high-energy dissociation with normalized energy of 35%. The data were acquired and reviewed with Xcalibur 2.2 (ThermoFisher). The MS experiments of identification of the photolabeling sites and competition of photolabeling were replicated at least three times.
The LC-MS data were searched against a customized database containing the sequence of the GABAA receptor 8X His-FLAG-α1 and β3 subunit and filtered with 1% false discovery rate using PEAKS 8.5 (Bioinformatics Solutions Inc.). Search parameters were set for a precursor mass accuracy of 30 ppm, fragmentation ion accuracy of 0.1 Da, up to three missed cleavage on either side of peptide with trypsin digestion. Methionine oxidation, cysteine alkylation with NEM and DTT, any amino acids with adduct of KK123 (mass = 372.16), KK200 (mass = 462.27), KK202 (mass = 500.31), KK123 with light FLI-tag (mass = 672.4322), and KK123 with heavy FLI-tag (mass = 682.44) were included as variable modification.
The GABAA receptors were expressed in oocytes from the African clawed frog (X. laevis). Frogs were purchased from Xenopus 1 (Dexter, MI) and housed and cared for in a Washington University Animal Care Facility under the supervision of the Washington University Division of Comparative Medicine. Harvesting of oocytes was conducted under the Guide for the Care and Use of Laboratory Animals as adopted and promulgated by the National Institutes of Health. The animal protocol was approved by the Animal Studies Committee of Washington University in St. Louis (approval No. 20170071).
The oocytes were injected with a total of 12 ng cRNA in 5:1 ratio (α1:β3) to minimize the expression of β3 homomeric receptors. Following injection, the oocytes were incubated in ND96 with supplements (96 mM NaCl, 2 mM KCl, 1.8 mM CaCl2, 1 mM MgCl2, 2.5 mM Na pyruvate, 5 mM HEPES, and 100 U/ml + 100 μg/ml penicillin + streptomycin and 50 μg/ml gentamycin [pH 7.4]) at 16°C for 1–2 days prior to conducting electrophysiological recordings.
The electrophysiological recordings were conducted using standard two-electrode voltage clamp. Borosilicate capillary glass tubing (G120F-4, OD = 1.20 mm, ID = 0.69 mm; Warner Instruments, Hamden, CT) were used for voltage and current electrodes. The oocytes were clamped at −60 mV. The chamber (RC-1Z; Warner Instruments, Hamden, CT) was perfused with ND96 at 5–8 ml min−1. Solutions were gravity-applied from 30-ml glass syringes with glass luer slips via Teflon tubing.
The current responses were amplified with an OC-725C amplifier (Warner Instruments), digitized with a Digidata 1200 series digitizer (Molecular Devices) and were stored using pClamp (Molecular Devices). The peak amplitude was determined using Clampfit (Molecular Devices).
The stock solution of GABA was made in ND96 bath solution at 500 mM, stored in aliquots at −20°C, and diluted as needed on the day of experiment. Stock solution of propofol (200 mM in DMSO) was stored at room temperature. The steroids were dissolved in DMSO at 10 mM and stored at room temperature.
The α1β3 wild-type and mutant receptors were tested (see Table 1 and S2 and S3 Tables) for potentiation by steroids (3α5α-allopregnanolone, 3α5β-pregnanolone, KK123, KK200, and KK-202) and direct activation by steroids (allopregnanolone KK123, KK200, KK-202, and pregnanolone). As control, several receptor isoforms were tested for potentiation by propofol. For each receptor type, we also determined constitutive open probability (Po,const).
To estimate Po,const, the effect of 100 μM picrotoxin (estimated Po = 0) on the holding current was compared to the peak response to saturating GABA + 100 μM propofol (estimated Po = 1). Po,const was then calculated as Ipicrotoxin ÷ (Ipicrotoxin − IGABA+propofol) [62].
Potentiation is expressed as the potentiation response ratio, calculated as the ratio of the peak response to GABA + modulator (steroid or propofol) to the peak response to GABA alone. The concentration of GABA was selected to produce a response of 5%–15% of the response to saturating GABA + 100 μM propofol.
Direct activation by steroids was evaluated by comparing the peak response to 10 μM neurosteroid to the peak response to saturating GABA + 100 μM propofol. Direct activation by steroids is expressed in units of open probability that includes constitutive open probability. All data are given as mean ± SD and analyzed by one-way ANOVA followed by Dunnet’s multiple comparison to the control wild-type group.
A homology model of the α1β3 GABAA receptor was developed using the crystal structure of the human β3 homopentamer published in 2014 (PDB ID: 4COF) [47]. In this structure, the large cytoplasmic loops were replaced with the sequence SQPARAA used by Jansen and colleagues [63] The pentamer subunits were organized as A α1, B β3, C α1, D β3, E β3. The α1 sequence was aligned to the β3 sequence using the program MUSCLE [64]. The pentameric alignment was then used as input for the program Modeller [65], using 4COF as the template; a total of 25 models were generated. The best model as evaluated by the DOPE score [66] was then oriented into a POPC membrane, and the system was fully solvated with 40715 TIP3 water molecules and ionic strength set to 0.15 M KCl. A 100 ns molecular dynamics trajectory was then obtained using the CHARMM36 force field and NAMD. The resulting trajectory was then processed using the utility mdtraj [67], to extract a snapshot of the receptor at each nanosecond of time frame. These structures were then mutually aligned by fitting the alpha carbons, providing a set of 100 mutually aligned structures used for docking studies.
The docking was performed using AutoDock Vina [68] on each of the 100 snapshots in order to capture the receptor flexibility. Docking boxes were built for the β3 intrasubunit site (cluster 3), the α1 intrasubunit site (cluster 2), and the β3-α1 intersubunit site (cluster 1). The boxes were centered around the residues photolabeled by KK123, KK200, and KK202 and had dimensions of 25 × 25 × 25 Ångströms, large enough to easily fit the linear dimensions of all of the steroids. For docking studies of allopregnanolone, the docking boxes were placed in the same locations but had smaller dimensions of 20 × 20 × 20 Ångströms. Docking was limited to an energy range of 3 kcal from the best docking pose and was limited to a total of 20 unique poses. The docking results for a given site could result in a maximum of 2,000 unique poses (20 poses × 100 receptor structures); these were then clustered geometrically using the program DIVCF [69]. The resulting clusters were then ranked by Vina score and cluster size and visually analyzed for compatibility with the photolabeling results, which is the photolabeling group oriented in the correct direction to produce the observed photo adducts.
The inorganic salts used in the buffers, GABA, picrotoxin, and the steroids 3α, 5α-allopregnanolone, and 3α,5β-pregnanolone were purchased from Sigma-Aldrich. Propofol was purchased from MP Biomedicals (Solon).
|
10.1371/journal.pcbi.1000987 | Lobe Specific Ca2+-Calmodulin Nano-Domain in Neuronal Spines: A Single Molecule Level Analysis | Calmodulin (CaM) is a ubiquitous Ca2+ buffer and second messenger that affects cellular function as diverse as cardiac excitability, synaptic plasticity, and gene transcription. In CA1 pyramidal neurons, CaM regulates two opposing Ca2+-dependent processes that underlie memory formation: long-term potentiation (LTP) and long-term depression (LTD). Induction of LTP and LTD require activation of Ca2+-CaM-dependent enzymes: Ca2+/CaM-dependent kinase II (CaMKII) and calcineurin, respectively. Yet, it remains unclear as to how Ca2+ and CaM produce these two opposing effects, LTP and LTD. CaM binds 4 Ca2+ ions: two in its N-terminal lobe and two in its C-terminal lobe. Experimental studies have shown that the N- and C-terminal lobes of CaM have different binding kinetics toward Ca2+ and its downstream targets. This may suggest that each lobe of CaM differentially responds to Ca2+ signal patterns. Here, we use a novel event-driven particle-based Monte Carlo simulation and statistical point pattern analysis to explore the spatial and temporal dynamics of lobe-specific Ca2+-CaM interaction at the single molecule level. We show that the N-lobe of CaM, but not the C-lobe, exhibits a nano-scale domain of activation that is highly sensitive to the location of Ca2+ channels, and to the microscopic injection rate of Ca2+ ions. We also demonstrate that Ca2+ saturation takes place via two different pathways depending on the Ca2+ injection rate, one dominated by the N-terminal lobe, and the other one by the C-terminal lobe. Taken together, these results suggest that the two lobes of CaM function as distinct Ca2+ sensors that can differentially transduce Ca2+ influx to downstream targets. We discuss a possible role of the N-terminal lobe-specific Ca2+-CaM nano-domain in CaMKII activation required for the induction of synaptic plasticity.
| Calmodulin is a versatile Ca2+ signal mediator and a buffer in a wide variety of body organs including the heart and brain. In the brain, calmodulin regulates intracellular molecular processes that change the strength of connectivity between neurons, thus contributing to various brain functions including memory formation. The exact molecular mechanism as to how calmodulin regulates these processes is not yet known. Interestingly, in other excitable tissues, including the heart, each of two lobes of calmodulin responds differentially toward Ca2+ influx and toward its target molecules (e.g., ion channels). This way, calmodulin precisely controls the Ca2+ dynamics of the cell. We wish to test if a similar mechanism may be operational in neurons so that two lobes of calmodulin interact differentially with Ca2+ ions to activate different downstream molecules that control the strength of connections between neurons. We constructed a detailed simulation of calmodulin that allows us to keep track of its interactions with Ca2+ ions and target proteins at the single molecule level. The simulation predicts that two lobes of calmodulin respond differentially to Ca2+ influx both in space and in time. This work opens a door to future experimental testing of the lobe-specific control of neural function by calmodulin.
| Calmodulin (CaM) is a ubiquitous Ca2+ buffer and signaling molecule in cells. In the excitatory synapse of hippocampal CA1 pyramidal neurons, the activation of CaM dependent enzymes results in the induction of synaptic plasticity (e.g., long-term potentiation (LTP) and long-term depression (LTD)) [1]. The induction of NMDA receptor dependent LTP and LTD require increased Ca2+ and subsequent activation of CaM-dependent downstream enzymes: CaM-dependent protein kinase II (CaMKII) and calcineurin. Injection of CA1 pyramidal cells with peptides that block CaMKII activity inhibited the induction [2], [3], but not maintenance [4] of LTP, while injection of the activated form of the enzyme also produced LTP-like plasticity [5], [6]. LTD is also critically dependent on Ca2+ and it appears that the CaM-dependent phosphatase, protein phosphatase 2B (calcineurin) is involved in LTD induction [7]. The simplest correlative explanation for these results is that LTD is induced by intermediate levels of Ca2+ that activate CaM and subsequently calcineurin but not CaMKII. Conversely, higher levels of Ca2+ initiate CaM-dependent CaMKII activation and autophosphorylation, leading to LTP induction. However, it is still unknown how Ca2+ and CaM regulate two opposing processes as distinct as LTP or LTD in such a precise and controlled manner.
Besides being a major signaling molecule, CaM also functions as a primary Ca2+ buffer in CA1 pyramidal neurons [8]. In fact, most CA1 pyramidal neurons contain CaM but not other EF-hand Ca2+ binding proteins (e.g., parvalbumin and calretinin) (reviewed in [9]). An exception is calbindin-D28K, which is expressed in a subpopulation of CA1 pyramidal neurons but only in rat ([10], [11]). CaM binds four Ca2+ ions, two in its N-terminal lobe and two in its C-terminal lobe [12]. The binding sites in the N-terminal lobe are lower affinity [13] but exhibit faster kinetics as opposed to the higher affinity, slower kinetics of the C-terminal lobe sites [14], [15]. Surprisingly little is known as to how such a protein with multiple Ca2+ binding sites influences the diffusion of Ca2+ in the cell. Most pre-existing theories of Ca2+ binding and diffusion assume a fast binding of Ca2+ and single Ca2+ binding site for the buffer (see reviews by [16]). In addition, recent experimental data suggest that each lobe of CaM has different affinity toward its downstream target (CaMKII and calcineurin) [17], [18], [19]. As each lobe differentially responds to Ca2+ signals and downstream targets, it is possible that these lobe specific properties play distinct biological roles in synaptic spines (see Discussion for more details). This motivated us to dissect the spatial-temporal dynamics of lobe specific Ca2+-CaM interaction in detail at the single molecule level.
Many elegant experimental measurements have been made of dendritic spine Ca2+ [20], [21], [22], [23], [24]. These measurements largely rely on a spatially averaged Ca2+ signal generated from fluorescence imaging of dyes whose quantum efficiency changes upon Ca2+ binding. As such, they contain no direct information relative to the issue of possible micro- or nano-domains of intracellular Ca2+. The problem is exacerbated by the high diffusion coefficients of free and dye bound Ca2+ which additionally smears the spatial signal in time frames relevant for Ca2+-imaging experiments. These and other caveats related to dye-based Ca2+-imaging experiments were recently reviewed [25]. In addition, we do not have an effective fluorescence reporter to detect and monitor Ca2+ binding to each lobe of CaM at the single molecule level. As such, mathematical models and computer simulations are presently the only tractable means of investigating this critical aspect of synaptic physiology. Furthermore, in a medium size dendritic spine (i.e., sphere-shaped spine head of 500 nm diameter), the concentration of 1 µM of any chemical species corresponds to ∼40 molecules. The basal (resting) level of spine Ca2+ is 50∼100 nM which corresponds to 2∼4 molecules of Ca2+ ions. Under such a circumstance, the behavior of single molecules within synaptic spines is not well described by the concentration-based mathematical approach such as reaction diffusion equation.
Here we report the single molecule level analysis of Ca2+-CaM interaction within a dendritic spine using a novel particle-based event-driven Monte Carlo algorithm, which we call Cellular Dynamics Simulator (CDS, [26]). Unlike other commonly used Monte Carlo simulation (e.g., MCell, [27]), it explicitly takes account of volume exclusion and collision between diffusing molecules in order to accurately simulate chemical reactions in the cellular interior. Using this simulator and first passage time theory, we dissect the mechanisms that influence the dynamics of Ca2+-CaM interaction at the single molecule level. We use a model of CaM built upon detailed kinetic data and ask if the lobe specific spatial-temporal micro-domain of Ca2+-CaM activation can exist and if so how it is biophysically regulated in a small sub-cellular compartment like dendritic spines. We employ a statistical spatial point pattern analysis [28] to understand the spatial profile of Ca2+-CaM interactions. The combination of spatial point pattern analysis and particle based Monte Carlo simulation is a unique computational strategy used in this study. Our analysis shows a higher sensitivity of the N-terminal lobe to the location and influx rate of Ca2+ from typical receptor/channel sources. Each lobe of CaM functions as distinct Ca2+ sensors and responds differentially to Ca2+ influx both in space and in time. Coupled with the experimental knowledge that different enzymes bind preferentially to either the N- or C-lobes of Ca2+ saturated CaM, we propose a possible explanation for how two opposing Ca2+/CaM-dependent enzymes can be differentially activated.
Fig. 1A illustrates the Ca2+ binding and unbinding pathway for each lobe of CaM. As shown, Ca2+ binding to the N-terminal lobe and the first Ca2+ binding event to the C-terminal lobe are diffusion limited while the second Ca2+ binding to the C-terminal lobe is the rate-limiting step in achieving the fully Ca2+-saturated state. If this Ca2+ binding step at the C-terminal lobe is much slower than the diffusion of Ca2+, the majority of Ca2+ ions that entered the spine head will have moved away from the channel without saturating local CaM molecules. The spatial profile of the C-terminal lobe or full Ca2+ saturation of CaM may then be less sensitive to the location of Ca2+ channels. On the other hand, if the N-terminal lobe Ca2+ saturation is fast as compared to the Ca2+ diffusion, its Ca2+ saturation may be more closely localized to the Ca2+ channels. Thus, three biophysical factors become important in understanding the spatial domain of Ca2+-CaM interactions. The first is how fast each lobe of CaM becomes Ca2+ saturated with a given concentration of Ca2+. The second is how fast Ca2+ ions escape from the spine. The third is how steep or flat the gradient of Ca2+ ion distribution will be in the spine head with a given Ca2+ injection rate through Ca2+ channels.
In this section, we analyze the first biophysical factor, which we call the (mean) first passage time: the (average) length of transition time required for each lobe of CaM molecule to reach the Ca2+ saturated state from a basal (apo-) state. In fact, a mathematical formula is already available to calculate this mean first passage time (Equations 5, 29 in [29]). In their single molecule biophysical analysis, Shaevitz et al. [29] used an algebraic recursive method to derive the Laplace transform of the first passage time distribution. Fig. 1B and Eq. 1∼2 explain their formalism applied to Ca2+-CaM interactions. Here we define State “0” as a Ca2+ free (apo) form, State “1” as one Ca2+ ion bound form, and State “2” as a two Ca2+ ion bound form of a given lobe. The symbols kXij in Fig. 1B denotes the rate constant between State i and State j (i, j = 0, 1, 2) of lobe X ( = N or C). Thus, each lobe has three states and the whole CaM molecule has nine states (Fig. 1C).
The resultant Laplace transform of the distribution of first passage time is:(1)where [Ca] is the given concentration of Ca2+ (Note, in order to apply Eq. 29 in [29], we needed to multiply the association rate constant by the concentration of Ca2+). Here we assume the system is well-stirred and the concentration of Ca2+ is constant (time-invariant). Then, the mean first passage time (<t>) can easily be found through differentiation (see Eq. 5 in [29]):(2)Note that the dissociation rate () of the second Ca2+ is not included in the formula. The latter rate determines the lifetime of fully Ca2+ saturated state of each lobe but it does not influence the first passage time. Therefore, three kinetic rates (, , ) and Ca2+ concentration determine the lobe specific first passage time. Note that both lobes have similar association rates for the first Ca2+ ions () (Fig. 1A). The difference in the second Ca2+ binding rates () is large as compared to the dissociation of the first Ca2+ ion () (Fig. 1A). Thus, in Eq. 2, the second Ca2+ binding rates (, ) determine the difference of the first passage time between the N- and C-lobes.
Fig. 2A is a numerical display of this formula showing that the first passage time sharply increases as we decrease the Ca2+ concentration (the unit of time, y-axis, is in seconds). As predicted, the mean first passage time for the C-terminal lobe (magenta) is much longer than the N-lobe (blue). For comparison, we show the first passage time for full Ca2+ saturation of CaM; the mean first passage time to reach the state N2C2 in Fig. 1C. As one can see from the diagram in Fig. 1C, this first passage time depends on all Ca2+ association and dissociation pathways for both lobes and is influenced by the lifetime of the Ca2+ saturated states of each lobe. The corresponding mathematical formula will be much more complicated than Eq. 1 and 2 and therefore, we calculated this quantity numerically using an extended version of the Gillespie type stochastic algorithm (see [8], [30] for more details). The results presented in Fig. 2A suggest that the N-terminal lobe may respond to a short Ca2+ transient but the C-terminal lobe may not if the transient is shorter than the first passage time of C-lobe Ca2+ saturation. For example, NMDA receptor type Ca2+ transients (∼1 µM peak with duration of ∼80–200 ms) may not result in significant CaM saturation in the spine. In fact, at a ∼1 µM Ca2+ concentration, the mean first passage time for the C-terminal lobe (or full Ca2+ saturation of CaM) is much longer than the duration of the Ca2+ transient (Fig. 2A upper right inset).
Such straightforward interpretation of the first passage time analysis, however, could be misleading. Note that we have only discussed the mean but not the entire distribution (or standard deviation) of the first passage time. In addition, we ignored the fact that the number of Ca2+ ions may be limited in the dendritic spines and that their concentration is not constant as postulated in Eq. 1∼2: the N-terminal lobe and the C-terminal lobes on the same or different CaM molecules will compete for the limited number of Ca2+ ions. As for the stochastic fluctuation, we can derive the standard deviation of the first passage time using the same analytic method described above:(3)The resultant standard deviation is very close to the mean first passage time for all Ca2+ concentrations (i.e., the coefficient of variation is >0.9 for all [Ca2+]<10 µM). The second term in the right-hand side of Eq. 3 () is small because the first Ca2+ binding rate () for both lobes are high and therefore the ratio of the right-hand sides of Eq. 3 and Eq. 2 approaches 1.
Fig. 2B and C show the histograms of the first passage time distribution for the N-terminal lobe (blue) and the C-terminal lobe (magenta) Ca2+ saturation, respectively, taken from a single stochastic simulation (the same bin size, 5ms, for both lobes and the total number of CaM molecules is 400). Fig. 2B clearly shows that the Ca2+ saturation of the C-terminal lobe is possible even if the mean first passage time is shorter than that of Ca2+ transient. However, the inset of Fig. 2B and 2C, i.e., the histogram up to 80 ms, predict that the N-terminal lobe Ca2+ saturation predominates and precedes that of the C-terminal lobe during the short Ca2+ transient. Knowing that two lobes of CaM compete for the limited amount of available Ca2+ ions in the dendritic spines, we predict that the N-terminal dominance for the short Ca2+ transient is more prominent in neurons.
This type of analysis, however, is further complicated when taking into account the non-homogeneous spatial distribution of molecules. When Ca2+ ions enter the spine head through a Ca2+ channel, a steep spatial gradient of Ca2+may be formed around the channel mouth (depending on the Ca2+ injection rate). At a single molecule level, it is the transient local (microscopic) “concentration” of Ca2+ (i.e., the number of Ca2+ collision events) felt by a CaM molecule that determines the probability of Ca2+ saturation of a given lobe of each CaM molecule. A CaM molecule can experience much higher (local) Ca2+ “concentration” than indicated by the bulk Ca2+ transient depending on its location with respect to the Ca2+ source.
The present work aims to describe a detailed analysis of this spatial stochastic phenomenon. However, before going into the detailed simulations, it is necessary to dissect each of the biophysical factors that we discussed at the beginning of this section. The last two of these factors determine the space- and time- dependent Ca2+ profile in the spines. Without such a systematic dissection, the interpretation of simulation results when trying to determine the spatial/temporal profile of CaM activation would not be possible. We next explored how fast Ca2+ ions escape from the spine.
The second factor that will determine the spatial profile of CaM activation is the escape rate of Ca2+ from the spine. Ca2+ ions that enter the spine through ion channels will eventually diffuse into the dendrites or be extruded by the Ca2+ pumps [23]. Here we focus on the impact of spine geometry and Ca2+ pumps on the escape rate of Ca2+ from the spines. We carry out this analysis in a stepwise manner. We first analyze the escape of Ca2+ via pure diffusion without Ca2+ pumps (or buffers) and establish the impact of spine morphology on the Ca2+ escape rate (Fig. 3A and B). Then we add Ca2+ pumps to examine their impact (Fig. 3C). This way we can isolate and understand the contribution of each of these factors in the regulation of the Ca2+ escape rate. In neurons, Ca2+ buffers such as CaM also influence this escape rate but in a highly complicated manner. We will study the effect of Ca2+ binding proteins (CaM) in the later sections when we combine all known biophysical factors in the detailed simulations.
Fig. 3A shows the time courses of Ca2+ decay for three different spine neck geometries. Here, we randomly placed a fixed number of Ca2+ ions ( = 400 that corresponds to ∼10 µM) in the head of a spherical spine and let them diffuse out of the spine to the dendrite. The diffusion coefficient () of Ca2+ was set to 200∼225 µm2/s (nm2/µs) [31]. Each curve in Fig. 3A represents the average of 100 simulation runs. Clearly, the longer and the narrower the neck, the slower the Ca2+ decay process. This is a so-called narrow escape problem and has been extensively investigated [32], [33]. As predicted by these theoretical studies, the simulated Ca2+ decay transient is well approximated by a single exponential decay term. These decay time constants fit well (the relative error <5%) with one of the pre-existing mathematical formula (the left-hand side of Eq. 4 below):(4)where , , and are the volume of the spine head, the length and the radius of spine neck, and the volume of neck, respectively [32]. Fig. 3B summaries our simulation results for different spine geometries. We plot the narrow escape time () against the ratio of spine head and neck volume () (x-axis). As shown all data points are aligned on straight lines, indicating that the narrow escape time is a linear function of the volume ratio () (see the right-hand side of Eq. 4). Note that Eq. 4 was previously tested against experimental data of molecular diffusion (using photo-bleaching recovery of fluorescein-dextran and enhanced green fluorescent protein) across spine-dendrite junctions in CA1 neurons [21], [24]. In other words, Eq. 4 is consistent with escape of diffusing molecules from real spines on CA1 neurons. Additional simulations confirm that Eq. 4 fits well with real spines when morphologies from 3D EM reconstructions are used (http://synapses.clm.utexas.edu/) (data not shown).
Another biophysical factor that regulates the Ca2+ decay from spines is Ca2+ pumps [20], [23]. The main Ca2+ extrusion mechanisms in CA1 spines are Na+/Ca2+ exchangers (NCX, NCKX) and plasma membrane Ca2+ ATPase (PMCA) [20], [34]. We have modeled both of them using the kinetic scheme used in [35] (see Methods for more details). Fig. 3C shows a Ca2+ clearance process with standard spine morphology (500 nm spine head diameter, 500 nm spine neck length and 150 nm spine neck diameter) with (dashed black line) and without pumps (solid black line). The fast decay time constant of Ca2+ in the presence of pump is ∼45% of the narrow escape time without pumps (∼5–6 ms). In this analysis, we have included NCX/NCKX and PMCA at the concentration close to the highest level known in the literature to examine the maximal impact that Ca2+ pumps would have on Ca2+ clearance. The Ca2+ transients with reduced number of pumps lie between the dashed and solid lines (data not shown).
Overall, the analyses in Fig. 3 show that the narrow escape time of Ca2+ without buffers in a standard spine in the presence of pumps is ∼5 ms or shorter. In the subsequent section, we will show that a major Ca2+ buffer in CA1 pyramidal neurons (i.e., CaM) slows down the Ca2+ decay to ∼10∼20 ms (the latter is close to that observed in the Ca2+ imaging analyses [20]. It is this brief time window that each lobe of CaM becomes Ca2+ saturated or not during each Ca2+ spike. The first passage time becomes a critical factor to understand the spatial profile of Ca2+-CaM interactions.
Having established the impact of spine geometry on the Ca2+ extrusion process, we now analyze the third biophysical factor that influences the spatial gradient of spine Ca2+: the Ca2+ injection rate of channels. Since the kinetics of the voltage-gated Ca2+ channels and NMDA receptors are highly complicated, we used a “model stochastic Ca2+ channel” in this section. A single stochastic Ca2+ channel was placed on the top of the head of a standard spine (black circle in Fig. 4A; see Fig. 3C for the standard morphology of CA1 dendritic spine). This channel injects Ca2+ at a given (average) rate and we examine the relation between the Ca2+ injection rate and the spatio-temporal profile of Ca2+ transients in the spine. To realize the impact of Ca2+ injection rate in isolation on the spatiotemporal Ca2+ profile, there are no pumps or Ca2+ binding buffers in this model spine. Once injected, Ca2+ ions travel via simple diffusion until they are absorbed from the compartment at the spine-dendrite boundary (see the vertical arrow in Fig. 4A). We varied the rate, but the total number of injected Ca2+ ions was set to 700 so that the peak Ca2+ concentration would be in a physiological range (∼6–16 µM, i.e., ∼250–650 Ca2+ ions; see panel C of Fig. 4). Note these numbers are taken from the lowest estimated Ca2+ injection rate of NMDA receptors and the higher Ca2+ injection rates of voltage gated Ca2+ channels ([36], [37], [38]).
Fig. 4 Panel A and B show the location of Ca2+ ions (not to scale) at designated time points after the start of Ca2+ injection. The mean Ca2+ injection rates in Panel A and B are 1.4 and 0.07 Ca2+ ions per microsecond, respectively. At the higher injection rate (1.4 ions/µs), there is a build-up of Ca2+ ions near the channel (Panel A) while such a build up is not evident in Panel B. Note that the time points chosen for Panel A and B are 20-fold different so that the total number of Ca2+ injected by the indicated time points in Panel A (10, 20, 100, 200 µs) and B (200, 400, 2000, 4000 µs) are identical. Ca2+ ions can travel ∼140 nm from the channel via diffusion before the next Ca2+ ion exits the channel at injection rate of 0.07 ions/µs. At a higher Ca2+ injection rate, Ca2+ ions will accumulate near the channel pore before they diffuse away (red in Fig. 4C). As anticipated, the lower Ca2+ injection rate (black) leads to a much lower peak Ca2+ number (concentration) than the higher Ca2+ injection rates. Ca2+ ion can travel more than 1 µm away from the channel during 1 ms. During a 10 ms Ca2+ injection period, a significant fraction of Ca2+ ions has already left the spine. Thus, we have lower Ca2+ peak than at the higher Ca2+ injection rate. After the peak, the Ca2+ level decreases with a time constant of ∼7–8 ms for all Ca2+ injection rates. This decay process is controlled by the diffusion and is consistent with the narrow escape rate we calculated in Fig. 3.
Fig. 4 clearly shows the impact of Ca2+ injection rates on the spatial and temporal dynamics of Ca2+ transients in dendritic spines. The relative lack of a Ca2+ gradient in Fig. 4B and the long first passage time of the C-terminal lobe of CaM in Fig. 2 suggest that a spatial gradient of the Ca2+-saturated C-terminal lobe may not form. However, as mentioned at the beginning of this section, we need to include CaM and examine the combined effect of all of these biophysical factors on the spatial profile of Ca2+-CaM interactions. The second half of Results provides this analysis.
In the previous sections, we studied the impact of three biophysical factors: the first passage time (Fig. 2), the narrow escape time (Fig. 3), and the impact of Ca2+ injection rate on the Ca2+ micro-domain (Fig. 4). In this section, we wish to study the combined effects of these factors on the spatial-temporal pattern of Ca2+-CaM interaction. As a first step, we placed a single “model Ca2+ channel” as in Fig. 4 but add CaM to assess the impact of Ca2+ injection rates on the Ca2+-CaM interaction. Besides the “artificial” model channel, we included CaM and Ca2+ pumps. We distributed 1600 molecules of CaM (i.e., 40 µM) uniformly within the spine volume (the estimated concentration of CaM in CA1 dendritic spines is 10∼100 µM, [8]). Before injecting Ca2+ ions the entire system is equilibrated at basal Ca2+ conditions, i.e., ∼40–46 Ca2+ bound CaM molecules with ∼2 free Ca2+ ions (the latter correspond to 50 nM of basal free Ca2+ concentration). At this basal condition, majority of CaM molecules are Ca2+ free or in a single Ca2+ bound form and none of their lobes are Ca2+ saturated. The diffusion coefficient of CaM varies between 2∼20 µm2/s (nm2/µs) [30], [39]. In this section, we set it to 20 µm2/s (nm2/µs) (but see our comments below).
The results in Fig. 5 show the dynamics of Ca2+/CaM with a channel of high Ca2+ injection rate (1.4 Ca2+ ions/µs and a total of 700 Ca2+ ions are injected as in Fig. 4). Fig. 5A shows the number of Ca2+ saturated N- and C-lobes (blue and magenta, respectively) and fully Ca2+-saturated CaM. The number of free Ca2+ ions in the spine is shown in Fig. 5B. Both Fig. 5A and 5B are taken from the same single simulation run. The result of stochastic simulation varies from one simulation run to the other; however, the overall qualitative dynamics in Fig. 5A and 5B are similar among different simulation runs. The N-terminal lobe of CaM binds Ca2+ much faster than the C-terminal lobe (Fig. 2). As a consequence, the number of Ca2+ saturated N-terminal lobes increases rapidly as Ca2+ is injected (blue line in Fig. 5A). After the termination of Ca2+ injection (at 500 µs), the N-terminal lobes quickly release Ca2+ and the C-terminal lobes slowly bind the available Ca2+ (Fig. 5A). Once bound, Ca2+ remains associated with the C-lobe for a relatively long time (the decay time constant is ∼120 ms) and the C-lobes therefore trap Ca2+ in the spine (Fig. 5A). The free Ca2+ level eventually returns to the basal level after a few hundred ms (data not shown). Another important point to note is that even at this high Ca2+ injection rate, the total number of fully Ca2+-saturated CaM molecule is less than ∼7. This number varies from simulation to simulation, but with a single Ca2+ channel, the number remains below 10 (over 100 simulation runs), a remarkably low number.
Panels C, E, and G of Fig. 5 show the spatial dynamics of each lobe of CaM taken from 15 simulation runs. During the early rising phase of their Ca2+ saturation, each lobe of CaM exhibits a nano-domain near the channel pore. For example, in Fig. 5E, we record the location (red circle) of each CaM molecule when its N-lobe becomes first Ca2+ saturated. We plot these accumulated locations of “first Ca2+ saturation event” up to the different designated time point in the figure (note each lobe may undergo multiple cycles of Ca2+ saturation, but only the first one is recorded in Panel C, E, and G in Fig. 5 and in subsequent figures). The formation of a Ca2+/CaM nano-domain is clear. A similar but less obvious nano-domain is observed for the C-terminal lobe (Panel C) and for the fully Ca2+-saturated CaM.
To further confirm these observations, we performed spatial point pattern analysis (see Methods and [28], [40], [41]). In this statistical analysis, we counted the number of the Ca2+ saturation events (e.g., as shown in Fig. 5E for the N-terminal lobe) and then randomly distributed the same number of points within the spine volume. We calculated a so-called (Besag's) L-function (see Methods for details) for this random point pattern. We repeated this process 1000 times and calculated the mean and the maximum and minimum envelope of the L-function (the black dotted lines in Fig. 5F) for the set of 1000 randomly generated spatial patterns. We then calculated the L-function for the original data point pattern of Ca2+ saturation and compared this (the red line in Fig. 5F) with that of complete spatial randomness (the black lines in Fig. 5F). The L-function of data (red) is outside of the maximum and minimum envelopes (black dotted lines) indicating that the given point pattern is not random. In this case, L-function is larger than the maximum envelope and it is typical of spatial clustering. We performed a similar analysis for the C-terminal lobe (Fig. 5D) and fully Ca2+ saturated CaM (Fig. 5H) and obtained the same conclusion (non-randomness). For all cases, we also performed (two-sample) Kolmogorov-Smirnov (goodness-of-fit hypothesis) test (significance level = 0.05) [28] to verify the conclusion of envelope test.
In summary, the high Ca2+ injection rate results in a transient Ca2+-CaM nano-domain (for both lobes of CaM). The N-terminal lobe responds to and senses the Ca2+ gradient much faster than the C-lobe (blue Fig. 5A). The C-lobe's response is resistant to the Ca2+ gradient because of its longer first passage time (i.e., slow binding kinetics of Ca2+). Note we recorded and analyzed only the first Ca2+ saturation events for each lobe of each CaM molecules. The relatively widespread C-terminal lobe Ca2+ saturation in Panel C, therefore, is not because the high affinity C-terminal lobe carries Ca2+ ions while diffusing away from the channel.
What if we reduce the Ca2+ injection rate? Fig. 4 indicates that the spatial gradient of Ca2+ is less prominent with a reduced Ca2+ injection rate. One possible scenario is that, under such a condition, only N-terminal lobe with higher Ca2+ binding kinetics (Fig. 2) can detect and sense the spatial gradient. The Ca2+ saturation of C-terminal lobe and/or full Ca2+ saturation of CaM may show relatively homogeneous spatial patterns under this condition. Fig. 6 shows results to test this prediction. The simulation conditions are the same as in Fig. 5 except the Ca2+ injection rate is reduced to 0.07 Ca2+ per microsecond. This is close to the lowest Ca2+ injection rate observed for a single NMDA receptor Ca2+ current [36], [37], [38]. Panel A and B in Fig. 6 show the population dynamics of Ca2+ saturated N- and C-terminal lobe, fully Ca2+ saturated CaM (A), and free Ca2+ ions (B). The difference in the rising phase of Ca2+ saturated N- and C- terminal lobes observed in Fig. 5A becomes less obvious at these lower rates of Ca2+ influx. The Ca2+ saturated N- and C-terminal lobes increase at a similar rate but the N-terminal lobe exhibits a larger fluctuation due to its fast Ca2+ dissociation rate. Again, the number of fully Ca2+ saturated molecules is small (less than 5∼10) over the course of a 25 ms simulation experiment.
In addition, the location of Ca2+ saturation for each lobe becomes less localized around the channel (Fig. 6C and 6E). It still looks like the N-lobe exhibits a nano-domain but it is unclear by a simple inspection of the data as to whether a nano-domain exists for the C-terminal lobe. Up to the time points 2 ms and 4 ms, the Ca2+ saturation of the C-terminal lobe takes place throughout the entire spine head. The distribution of these points appears to be random. To confirm whether this pattern is random or not, we carried out the same statistical analysis as that used in Fig. 5 (panel D, F, H). Clearly, the data point patterns in Panel D and F (red line) are closer to the maximum envelope (black dotted line) of complete spatial randomness but the N-terminal lobe data pattern shows a deviation from the complete spatial randomness. This result was again confirmed by Kolmogorov-Smirnov test. The spatial pattern of the C-terminal lobe and full Ca2+-CaM saturation lie within the maximum/minimum envelope and did not suggest significant deviations from the spatial randomness.
In conclusion, the N-terminal lobe exhibits a transient Ca2+-activated nano-domain at both lower and higher Ca2+ injection rates. This indicates that the kinetic property of the N-terminal lobe (Fig. 1 and 2) is the major determinant of the spatial pattern formation by the N-terminal lobe. In fact, we repeated simulations used to produce Figs. 5 and 6 with different spine morphologies (with shorter and longer spine neck as shown in Fig. 3A) and obtained similar results as to the N-terminal lobe specific nano-domain (Fig. S1 and Fig. S2). We also set the diffusion coefficient of CaM to 2 µm2/s (nm2/µs) and repeated simulations in Fig. 5 and 6 (Fig. S1 and Fig. S2). As long as CaM molecules are randomly distributed within the spine volume (at time 0), neither the diffusion coefficient nor the concentration of CaM (even when reduced to 10 µM) affected the high sensitivity of the N-terminal lobe to the Ca2+ influx. It appears that the Ca2+ binding kinetics of CaM (first passage time) is the major determinant of the lobe specific spatial pattern formation during Ca2+ influx.
In addition, the spatial pattern of fully Ca2+ saturated CaM was also influenced by the Ca2+ injection rate (Fig. 5A, 6A, 5H, and 6H). Recall that Ca2+ dissociation from the C-terminal lobe is slower than from the N-terminal lobe (Fig. 1A). The C-terminal lobe remains fully Ca2+ saturated for extended time (>100 ms) during which CaM (or any Brownian particle of the same diffusion coefficient) can travel a distance equal to or larger than the entire spine head volume. CaM can reach its fully Ca2+ saturated state when additional Ca2+ binds to the N-terminal lobe (note again, the first Ca2+ saturation event of the C-terminal lobe is less sensitive to the location of the Ca2+ source as compared to the N-terminal lobe). Alternatively, if Ca2+ injection rate is high and the transient Ca2+ concentration is adequate, CaM can reach the fully Ca2+ saturated state via N-terminal lobe Ca2+ saturation before Ca2+ saturates the C-terminal lobe because the first passage time for the N-terminal lobe is shorter than the C-terminal lobe (Fig. 2). The latter pathway may be responsible for the nano-domain of fully Ca2+ saturated CaM observed in Fig. 5G and Fig. 5H. If these two modes of Ca2+ saturation exist, they would have different physiological impacts of CaM signaling system as the two lobes of CaM have distinctive binding affinities for different targets. A detailed inspection of Fig. 5 and Fig. 6 simulation results in the next section reveals and confirms these two Ca2+ saturation pathways of CaM and their dependence on the Ca2+ injection rates.
Fig. 7 presents results from studies on the Ca2+ saturation pathway of CaM at the single molecule level. In Fig. 7A and 7B, we randomly selected a CaM molecule from the simulation presented in Fig. 5, and analyzed its spatial location and Ca2+ binding state. We plot the trajectory of this molecule in the spine with different colors representing the different Ca2+ occupied states. The red is for the fully Ca2+ saturated state (State N2C2 in Fig. 7A or Fig. 7E), magenta for State N1C2 and N2C1 (three Ca2+ bound state), yellow for State N1C1, N0C2 and N2C0 (two Ca2+ bound state), green for State N0C1 and N1C0 (one Ca2+ bound state), and blue for State N0C0 (apo CaM) (see Fig. 7A and Fig. 7E for the notation). Note the direct state change between the states of the same color will never occur (see Fig. 7E). The choice of color for different states seems complicated but by using this strategy, we can explicitly show the state changes of a CaM molecule with a minimum number of colors.
The CaM molecule we selected for Fig. 7A and B was located relatively close to the channel at time 0 (in blue, but not clearly visible behind other colors in Fig. 7B). It went through N0C1 (green) and N1C1 (yellow) states, reached the N-terminal Ca2+ saturated state (N2C1, magenta), and then fully Ca2+ saturated (N2C2, red) near the channel (use Fig. 7A and 7E to follow these state changes). In other words, this CaM molecule follows the sequence of N-terminal lobe Ca2+ saturation before becoming fully Ca2+ saturated (indicated by the arrow in Fig. 7A). There is no C-terminal lobe Ca2+ saturation before the N-terminal lobe. After becoming fully Ca2+ saturated, the molecule started to move away from the channel but its C-terminal lobe remained Ca2+ saturated and stays in the N2C2 (red), N1C2 (magenta), and N0C2 (yellow) states as it explores the space close to the channel (Fig. 7B).
Fig. 7C and 7D show the single molecule analysis for the low injection rate (0.07 Ca2+ ions/µs). We randomly selected a CaM molecule from the simulation presented in Fig. 6 and kept track of its state change (Fig. 7C) and spatial location (Fig. 7D). This CaM molecule was located in the middle of the spine head at the beginning of the simulation and explored a large area in the spine head in N0C0 (blue) state before reaching the N0C1 (green) state. It briefly went into the N1C1 (yellow) state and returned to the N0C1 (green) state and then it reached the N0C2 (yellow) state, the Ca2+ saturated state of the C-terminal lobe (indicated by the arrow in Fig. 7C; also follow these state changes in Fig. 7E). After the C-terminal lobe saturation, it undergoes a rapid Ca2+ binding to the N-terminal lobe (at time ∼6.5 ms) via states N1C2 (magenta) to reach the fully Ca2+ saturated state (N2C2, red) (Fig. 7C). After Ca2+ is released from the fully Ca2+ saturated C-terminal lobe, this CaM molecule undergoes multiple state changes between N0C0 (blue), N0C1 (green), and N1C1 (yellow) states (see Fig. 7C and E).
These analyses (Fig. 7A and 7C) revealed two distinctive Ca2+ saturation pathways: N-terminal first pathway and C-terminal first pathway (see Fig. 7E). Fig. 7F and 7G present results that address the generality of the single examples shown in 7A and 7C. In these figures, we use the data from Fig. 5/6 and plot the number of CaM molecules that have reached the Ca2+ saturated state (for the first time) up to each time point (cumulative sum). We plot the number of CaM molecules who have reached saturation via N-terminal lobe saturation first (blue) and via C-terminal lobe first (magenta). At the lower Ca2+ injection rate, the C-terminal lobe first is the dominant pathway (Fig. 7F). At the higher Ca2+ injection rate, the probability of CaM reaching the fully Ca2+ saturated state via the N-terminal lobe first pathway is significantly increased, especially during the first 5 ms (Fig. 7G). Note it is this first ∼5 ms time period that the number of Ca2+ saturated N-terminal lobes exceed that of the Ca2+ saturated C-terminal lobe (Fig. 5A). Overall, the C-terminal lobe first pathway exists for both low and high Ca2+ injection rates. The Ca2+ saturation of CaM via the N-terminal lobe dominant pathway only becomes prominent at higher Ca2+ injection rates.
So far we have analyzed the lobe-specific Ca2+-CaM spatial domains using a “model” channel. The purpose of this arrangement was to systematically analyze the impact of Ca2+ injection rates that may underlie possible lobe-specific Ca2+-CaM nano-domains. We now explore the same issue under a more realistic situation. Instead of a single “model” Ca2+ channel, we place multiple NMDA receptors on the spine head and analyze the impact of their spatial distribution on the lobe-specific Ca2+-CaM nano-domain.
As stated earlier, NMDA receptors are the major Ca2+ source in CA1 spines [25]. The estimated number of NMDA receptors lie between 5∼20 [42], [43]. The number and distribution of NMDA receptor may vary from one spine to the other. To gauge the impact of the spatial localization of NMDA receptors, we decided to create two extreme cases. In Fig. 8, we placed 20 NMDA receptors in a 200 nm diameter area of the spine membrane to mimic NMDA receptors embedded in the post-synaptic density. In Fig. 9, we uniformly distributed the same number of NMDA receptors over the entire spine head. In both cases, we populated the spine volume with the same number of CaM molecules and Ca2+ pumps as in Fig. 5 and 6 (see Methods for more details of simulation).
In panel A and B of Fig. 8 and 9, we show the Ca2+ binding kinetics and free Ca2+ transients of single simulation runs of each case. The stochastic fluctuation (opening and closing) of NMDA receptors dictates the Ca2+ transient as predicted by previous work [43]. Interestingly, we could not find any significant differences between the two different distribution patterns of NMDA receptors (Fig. 8 and Fig. 9) in terms of overall Ca2+ (or Ca2+ binding to CaM) transients. To show the spatial patterns of Ca2+ saturation, we compiled the results of 20 simulation runs (of 20∼25 ms, for each NMDA receptor distribution pattern) and plot the locations of the Ca2+ saturated N- and C-lobe and fully Ca2+ saturated CaM as before (Fig. 8 C∼H and Fig. 9 C∼H).
For both distribution patterns of NMDA receptors, the N-terminal lobe Ca2+ saturation exhibits deviations from spatial randomness (Fig. 8F and Fig. 9F). In the case of NMDA receptor clusters (Fig. 8), a transient nano-domain of Ca2+ saturated N-terminal lobe is formed close to the receptor cluster and visible in the 2D projection of the data. In contrast, there is no detectable focus of clustering of Ca2+ saturated N-terminal lobe for homogenous NMDA receptor distributions (compare Fig. 8E and 9E at 4 ms). However, our methodology (Ripley's K-function/Besag's L-function) still detected a slight deviation from complete spatial randomness (Fig. 9F). This may suggest that the N-terminal lobe is still sensitive to the location of NMDA receptors but their spatial pattern of Ca2+ saturation was not clearly visible in the 2D projection of the data. The C-terminal lobe exhibits a minor and weak deviation from the spatial randomness for both cases. Overall, the N-terminal lobe shows a nano-domain regardless of the spatial distribution pattern of NMDA receptors.
We have analyzed the lobe specific spatial and temporal pattern of Ca2+-CaM interactions at the single molecule level in synaptic spine compartments. Ca2+ metabolism in neuronal spines is a dauntingly complicated process that involves nonlinear interactions between channels, pumps, CaM, and other potential Ca2+ binding proteins. We focused on three primary biophysical factors, Ca2+ binding kinetics of CaM, Ca2+ clearance from the spine compartment, and Ca2+ injection rate, and dissected the spatial pattern of Ca2+-CaM interactions in a stepwise manner. Our results indicate that the N-terminal lobe and the C-terminal lobe of CaM have different functions in decoding Ca2+ signals in space and time. The N-terminal lobe is more sensitive to the Ca2+ transients while the C-terminal lobe is relatively resistant to the spatial gradient of Ca2+. Our systematic dissection (Fig. 2∼9) strongly indicated that the Ca2+ binding kinetics to each lobe of CaM is the key regulatory mechanism of the spatial pattern of the Ca2+-CaM system. Our simulation study also identified two Ca2+ saturation pathways and their Ca2+ injection-rate dependencies: the C-terminal lobe first vs. the N-terminal lobe first pathways. The simulation results showed that the former is especially prominent with the low Ca2+ injection rate.
What are the implications of the lobe specific functionalities of CaM, especially for the CaM-and NMDA receptor-dependent synaptic plasticity that involves CaMKII and calcineurin? In order to understand this issue, one must pay close attention to the details of Ca2+-CaM-target interactions. Each lobe of CaM (as well as the entire CaM molecule) undergoes a series of conformational changes upon Ca2+ and/or target binding. In fact, the Ca2+ binding and target association are thermodynamically coupled (see [8]). Target binding increases or decreases the affinity of Ca2+ of CaM while Ca2+ binding in turn changes the binding kinetics of CaM towards its targets (see below for more discussion). The changes in the Ca2+ binding kinetics upon target binding (i.e., due to the different conformational states of CaM) is a critical factor that may affect the spatial profile of Ca2+-CaM-target activation.
Another important issue to consider is that a fraction of CaM molecules may already exist in a complex with its target even at basal Ca2+ concentrations. Interestingly, recent experimental and modeling work suggested that the N-terminal lobe of CaM preferentially interacts with CaMKII before the C-terminal lobe [19], [44]. In fact, these kinetic studies suggest that CaM remained bound to CaMKII for extended periods at basal Ca2+ concentrations via the N-terminal lobe. This mode of CaM-CaMKII interaction is different from the so-called CaM-trapping by auto-phosphorylated CaMKII (see [19] for full discussion of this issue). Once bound to CaMKII via the N-terminal lobe, the C-terminal lobe of the same CaM molecule interacts with CaMKII. When bound to CaMKII, the Ca2+ binding kinetics of the C-terminal lobe are accelerated by the law of detailed balance [19]. As shown in Fig. S3, CaMKII bound C- and N-terminal lobes both have faster Ca2+ binding kinetics (Panel A) and shorter first passage time for Ca2+ saturation (Panel B). The present work (Fig. 2, 5∼9) predicts that CaMKII-bound CaM may exhibit a nano-domain as observed in the target-free N-terminal lobe as long as the distribution of CaMKII is homogeneous within the spines. The latter assumption (homogenous distribution of CaMKII) may not be the case. However, recent experimental results indicated the presence of a nano-domain of CaMKII activation in CA1 spines [45]. Since CaMKII plays a key role in LTP (long-term potentiation) induction, further investigation of this CaMKII nano-domain is critical.
What if the C-terminal lobe preferentially interacts with calcineurin which underlies LTD (long-term depression) induction? Then, each of the two lobes of CaM differentially regulates these two opposing processes of synaptic plasticity. This may seem like an attractive hypothesis and in fact, our preliminary modeling study indicated that the C-terminal lobe of CaM has a higher affinity toward calcineurin than the N-terminal lobe. However, the affinity of calcineurin for CaM is extremely high [17] and as a consequence, most of the calcineurin molecules may already be bound to Ca2+-CaM even at the basal free Ca2+ concentrations in CA1 spines. On the other hand, for full activation, additional Ca2+ must bind the regulatory subunit (subunit B, CnB) of calcineurin [17]. If the Ca2+ binding kinetics of CnB is similar to that of the C-terminal lobe of CaM, one would expect a spatial and temporal pattern of calcineurin activation to be similar to the C-lobe specific Ca2+-CaM activation domain. Detailed experimental characterization of the Ca2+ binding kinetics of the “CaM-like” subunit of calcineurin (CnB) is necessary.
In CA1 pyramidal neurons, another critical factor, RC3 (neurogranin, Ng), regulates the induction of NMDA-receptor and CaM-dependent synaptic plasticity. RC3 is highly enriched in CA1 spines and is known to regulate the transition between the induction of LTP vs. LTD [46], [47]. The biochemical analysis of RC3-CaM interactions suggested that it may have an additional impact on the spatial nano-domain of Ca2+-CaM. RC3 binds CaM (even in the absence of Ca2+) and accelerates the Ca2+ dissociation from the C-terminal lobe thereby decreasing its affinity toward Ca2+ [30], [48]. The thermodynamic reciprocal interaction between Ca2+ binding and target interaction that we mentioned earlier may play an important role in determining the spatial dynamics of Ca2+-CaM-RC3 interactions. The released Ca2+ ion can bind the N-terminal lobe of the same or another CaM molecule. We predict that RC3 has a positive impact on the N-terminal specific Ca2+-CaM nano-domain and on the nano-domain of CaMKII bound CaM. In addition, RC3 is known to interact with membrane phosphatidic acid [49]. The spatial distribution of RC3 and the mobility of CaM-RC3 may have an additional significant impact of the spatial dynamics of Ca2+-CaM activation. Overall, genetic studies clearly suggest a critical role of RC3 in the regulation of Ca2+ dynamics in spines [46], [47]. Together with CaMKII, RC3 is another molecular target for future study using the particle-based Monte Carlo simulation.
The persistent existence of N-terminal lobe specific Ca2+-CaM nano-domain (Fig. 5∼Fig. 9) may at first seem reminiscent of the traditional view on Ca2+ micro-domains. However, we must point out that “Ca2+ domains” and “Ca2+-CaM domains” are, strictly speaking, different concepts. A “Ca2+ nano-domain” is defined by the mean distance traveled by Ca2+ ions before being captured by buffer (Ca2+ binding protein) or being extruded. Only under certain conditions, for example, when the Ca2+ binding rate is faster than the diffusion of Ca2+, are “Ca2+ domain” and “Ca2+-buffer” domain closely related in space. Clearly, the C- and N- terminal lobe specific Ca2+-CaM domains respond differently for the same Ca2+ input (Fig. 5 and 6) and the spatial profile (and size) of the C-terminal lobe domain is different from the “(free) Ca2+-domain”. Fig. S4 illustrates this point and shows the distributions of Ca2+ ions, Ca2+ saturated N-terminal and fully Ca2+ saturated CaM from a single simulation run in Fig. 5 and 6. Clearly, the size and spatial profile of these domains are not identical.
The spatial profile of the “Ca2+” signal ([Ca]i below), in the presence of excess unsaturable mobile buffers, is given by the following equation [50]:(5)where, is the single channel Ca2+ current, is the diffusion coefficient of Ca2+ (defined earlier), the distance from the channel, [Ca]0 is the bulk Ca2+ concentration, and denotes the mean path length of a Ca2+ ion travels before being captured by buffer, B is the buffer concentration is the Ca2+ binding rate, and F is the Faraday constant. This and many other mathematical formulas have been developed (see reviews in [16]) but they are not very useful to study the spatial profile of Ca2+-CaM or for any other protein or buffer with multiple Ca2+ binding sites of different binding kinetics.
Furthermore, in a small sub-cellular compartment, like CA1 spines, the number but not the concentration of molecules is important. As an illustration, when the equation for the steady-state Ca2+ concentration profile is applied to an L-type Ca2+ channel, it predicts a sharp Ca2+ gradient which results in 100 µM Ca2+ concentration at a distance of ∼4 nm from the channel (see Fig. 1C in [51]). 100 µM of Ca2+ within 4 nm distance of a channel is more than sufficient to saturate the C-terminal lobe of CaM, but it corresponds to less than 1 molecule of Ca2+ ion, leading to a contradiction. In order to understand the spatial information flow of the Ca2+ signaling system in dendritic spines, one must explicitly calculate the first passage time distribution of Ca2+ saturation of CaM and their spatial profile using an accurate particle-based Monte Carlo algorithm and appropriate data analysis method (e.g., statistical point pattern analysis) as we did in this study.
In addition, it is important to note that the two lobes of CaM, with almost opposite impacts on Ca2+-CaM nano-domains, reside in the same molecule and are competing for a limited amount of Ca2+ as we discussed in the Results (Fig. 2). This again implies that the N- and C- terminal lobes decode Ca2+ signals in a different manner, and potentially serve distinct cellular functions. The current work is the first step to understand this unique functionality of CaM at the single molecule level.
The Ca2+ transient in dendritic spines is regulated by highly nonlinear interactions between voltage-gated Ca2+ channels, K+ channels, and glutamate receptors. This important issue was recently reviewed in [25]. Clearly, Ca2+-activated K+ channels (SK channels) in hippocampal neurons shape the Ca2+ transients in spines and a direct coupling between voltage-gated Ca2+ channels and SK channels via “Ca2+ nano-domains” is a critical regulatory factor of spine Ca2+ metabolism. In addition, CaM itself regulates the activities of Ca2+ channels and Ca2+ pumps (PMCA) [52]. Without the detailed knowledge of these issues, we are not able to quantitatively address their impacts on spine Ca2+ dynamics. It is also difficult to make correct interpretations of pre-existing Ca2+ imaging experimental data (e.g., roles of pump in the diffusional coupling between dendrites and spines). For these reasons, in this study we focused on the initial rising phase of Ca2+ transients and therefore only studied the impacts of Ca2+ injection rate that are relevant for any Ca2+ channels.
With these limitations in mind, we repeated all simulations in Fig. 5∼9 without Ca2+ pumps and discovered that the resultant spatial profile of lobe specific Ca2+-CaM domains were similar to the results with Ca2+ pumps (data not shown). As long as Ca2+ pumps are uniformly distributed, the Ca2+ binding kinetics of CaM dictates the spatial and temporal pattern of the Ca2+-CaM interaction. We have not, however, tested spatially non-uniform distribution of Ca2+ pumps such as clusters of PMCA/NCX/NCXK tightly coupled to Ca2+ channels. This is an open area of future research.
Finally, the smooth endoplasmic reticulum (SER) is another source of Ca2+ that potentially influences Ca2+ transients in the spine. Although our simulator is fully capable of implementing SER structures and Ca2+ release from this source, only a small subset of dendritic spines on CA1 pyramidal neurons contain SER [53]. Furthermore, a recent study suggested a strong link between the SER containing spines and metabotropic glutamate receptor dependent synaptic depression [54] which is an interesting but different topic than the focus of the present work.
CaM is a bi-lobed molecule that has two Ca2+-binding sites within each lobe. Fig. 1A shows how this kinetic mechanism is modeled. Each lobe of CaM has three different states dependent on the number of bound Ca2+ ions: (apo)-CaM, (Ca2+)-CaM and (Ca2+)2-CaM (the horizontal arrows in Fig 1A). The resultant CaM model has nine Ca2+ binding states (Fig. 1C). We assume that Ca2+ binding to the C-lobe and N-lobe are independent and that inter-lobular cooperativity is not considered. The rate constants of Ca2+ binding to each lobe are taken from our previous work [8], [30]. This model is a simplification of our more elaborate model of CaM [19]. In the latter modeling scheme, Ca2+ association and dissociation at each Ca2+ binding site of CaM were explicitly modeled. Further refinement of the latter detailed model is also possible by taking into account of open (relaxed) and inactive closed (tense) states of each EF-hand of CaM as proposed by Stefan et al. [55]. We repeated the first passage time analysis in Fig. 2 using the former detailed model and confirmed that there is no qualitative difference between the detailed and simplified models. Future efforts will be made to incorporate the idea of relaxed and tense states in our simulations to specifically examine their consequences on Ca2+/CaM/target interactions.
The Ca2+ transient in the spine (head) is regulated by a highly complicated set of nested feedback loops [25]. This includes ionotropic glutamate receptors (AMPA receptors and NMDA receptors), CaV2.3 voltage-sensitive Ca2+ channels, small conductance Ca2+-activated K+ channel (SK channels), and voltage-gated Na+ channels. The role of voltage-gated CaV2.3 channels and Na channels have been largely unknown until recently [25], [56]. On the other hand, the nature of ionotropic glutamate receptors such as NMDA receptors, the major source of Ca2+ influx into the spine compartment, has been extensively studied in the past and we used a recently published model for our simulation (Fig. 8 and 9) [43].
The functional roles [34], [57], [58], [59] and molecular expression [60], [61], [62] of Ca2+ pumps have been studied; however, very limited quantitative information is available regarding the number, (intra-spine) distribution, and detailed kinetics properties of these Ca2+ pumps. The membrane densities of the plasma membrane Ca2+-ATPase (PMCA) and the Na+-Ca2+ exchanger (NCX) are 150∼300/µm2 of membrane and 32∼60/µm2 membrane, respectively [35], [63]. Since we do not have reliable data for the intra-spine distribution of these pumps, we decided to use the maximum estimated membrane densities for each pump to evaluate their impacts on Ca2+ dynamics (Fig. 3C). The PMCA kinetic constants are 0.2 µM Km for Ca2+ and a turnover rate of ∼100 s−1 and NCX has a Km of 3 µM and a turnover rate of ∼1000 s−1 [35]. For initial investigations we fixed the resting extrusion at 25 ions per second and 48 ions per second for PMCA and NCX, respectively [35].
The reaction scheme for the Ca2+ pump is similar to the one in [35]:(6)where , , and are Ca2+ inside the spine, extruded Ca2+, pump, and Ca2+-pump complex. PMCA hydrolyzes one ATP molecule per Ca2+ ion transported, i.e., exchanges one Ca2+ for one H+ (see recent reviews by Di Leva et al. [52]). NCX exchanges three Na+ for one Ca2+ and NCKX imports four Na+ while exporting one Ca2+ and one K+ (reviewed in [64]). Provided that we do not model the diffusions of Na+ or K+ or ATP hydrolysis, Eq. 6 captures the essential characteristics of these Ca2+ pumps (see Discussion for Ca2+-CaM dependent regulation of PMCA). Finally, we randomly incorporated Ca2+ leak channels so that the net flux of Ca2+ is 0 at rest (50 nM Ca2+).
The NMDA receptor kinetics was taken from previous modeling work [43]. Although our CDS simulator is fully capable of simulating glutamate release and diffusion in the synaptic cleft, this issue was not a focus of the present study. Instead, we assumed that each NMDA receptor was exposed to a constant level of glutamate as in previous modeling work [43], i.e., we stimulated the NMDA receptors for 0.1 ms with 1 mM of glutamate application and observed the subsequent Ca2+/CaM activation in the spine. The stochastic fluctuation of Ca2+ influx is then due to the stochastic kinetics of NMDA receptors.
All other numerical analyses including spatial point pattern analysis and first passage time calculation were carried out under the Matlab environment (The MathWorks, Inc., Natick, MA, USA).
The algorithmic principle of the event-driven particle-based Monte Carlo simulator (CDS) is described in [65] and the software is downloadable from our website (http://nba.uth.tmc.edu/cds). The CDS algorithm uses the discretized Brownian motion and relies on the first passage theory and event-driven simulation scheme. The overview of pre-existing particle-based Monte Carlo simulations (Smoldyn [66], GFRD [67], the coarse-grained molecular simulator described by Ridgway et al. [68], and MCell [27]) and differences between these simulator and CDS are also discussed in [65].
Under the CDS algorithm scheme, we calculate the first passage time (and probability) of molecular collisions and chemical reactions for each molecule in the simulation and create a list of all possible future events and their timing. We execute all of these molecular collisions and chemical reactions exactly as they happen one-by-one while moving all molecules simultaneously in the space. Every time we execute an event, we update the event list based on the new location or chemical status of the molecules. The time interval between two consecutive events varies from one simulation step to the other. Therefore, unlike time-driven Monte Carlo algorithms (e.g., MCell and Smoldyn), there is no fixed time step in CDS. This event-driven scheme is the only accurate way to handle molecular collisions in a crowded cellular environment. In some cases, the interval between two successive events (collision or chemical reaction) becomes long and may result in the non-Brownian motion of molecules. To avoid this situation, we add “change of direction of move” to the event list so that the direction of molecular motion is constantly randomized at least once every10 ns (the jump length of Ca2+ ion during this time period is smaller than the size of CaM molecule).
In the CDS simulations, the radius of gyration of CaM (2.2 nm) was used to set the size of CaM molecules. The radius of Ca2+ ion was set to 0.2∼0.25 nm (larger than its atomic radius) taking into account its hydration shell [69], i.e., we modeled Ca2+ as a solvated ion while simulating its diffusion and interactions with proteins. The diffusion coefficient () of Ca2+ in non-buffered cytoplasm is 200∼225 µm2/s (nm2/µs) [31].
The idea behind the Ripley's K-function, or its derivative Besag's L-function, is that if the distribution of the points is random, the number of points within a distance is proportional to if there is no spatial boundary in the system. Suppose we have a 3D spatial distribution of points (∼) and denotes the number of all points within a distance of the particular point . The Ripley's K-function is defined by(7)where is the density of particles, the average number of particles in a unit ball [28], [41]. The expected value for a random Poisson distribution in infinite space is . The Besag's L-function is a derivative of K-function and is defined by(8)so that its expected value for a random Poisson process in infinite space is (linear). A deviation of L-function from the spatial randomness indicates a clustering or repulsion of the point distribution. We can calculate K-functions with respect to a specific point in space such as a Ca2+ channel (instead of 's), but in this work, we calculated (Besag's) L-function for all points in space. The latter type of L-function is important and very useful as the clustering of points (the location of Ca2+ saturation) can happen in the middle of the spine head when multiple channels exist or when multiple cycles of Ca2+ binding and unbinding to the same CaM molecule take place (Fig. 8 and 9). Our data represent an analysis of inter-point (inter-Ca2+-saturation point) distance distribution at all distance scales and over the entire spine compartment. The important point to note is that in a confined and complicated geometry such as dendritic spines, a simple mathematical formula of Besag's L-function for complete spatial randomness is unavailable. To overcome this constraint, we created 1000 sets of randomly distributed points in the spine of the same number of data points and then calculated the L-function for the data and for the simulated random point patterns. If the resultant L-function of the data deviates from the simulated point pattern, we can conclude that the data points are not randomly distributed.
|
10.1371/journal.pgen.1004290 | Copy Number Variation Is a Fundamental Aspect of the Placental Genome | Discovery of lineage-specific somatic copy number variation (CNV) in mammals has led to debate over whether CNVs are mutations that propagate disease or whether they are a normal, and even essential, aspect of cell biology. We show that 1,000N polyploid trophoblast giant cells (TGCs) of the mouse placenta contain 47 regions, totaling 138 Megabases, where genomic copies are underrepresented (UR). UR domains originate from a subset of late-replicating heterochromatic regions containing gene deserts and genes involved in cell adhesion and neurogenesis. While lineage-specific CNVs have been identified in mammalian cells, classically in the immune system where V(D)J recombination occurs, we demonstrate that CNVs form during gestation in the placenta by an underreplication mechanism, not by recombination nor deletion. Our results reveal that large scale CNVs are a normal feature of the mammalian placental genome, which are regulated systematically during embryogenesis and are propagated by a mechanism of underreplication.
| Generally, every mammalian cell has the same complement of each part of its genome. However, copy number variation (CNV) can occur, where, compared to the rest of its genome, a cell has either more or less of a specific genomic region. It is unknown whether CNVs cause disease, or whether they are a normal aspect of cell biology. We investigated CNVs in polyploid trophoblast giant cells (TGCs) of the mouse placenta, which have up to 1,000 copies of the genome in each cell. We found that there are 47 regions with decreased copy number in TGCs, which we call underrepresented (UR) domains. These domains are marked in the TGC progenitor cells and we suggest that they gradually form during gestation due to slow replication versus fast replication of the rest of the genome. While UR domains contain cell adhesion and neuronal genes, they also contain significantly fewer genes than other genomic regions. Our results demonstrate that CNVs are a normal feature of the mammalian placental genome, which are regulated systematically during pregnancy.
| While the accumulation of somatic copy number variations (CNVs) has been proposed to be a result of the aging process, predisposing cell types to cancer progression and neurological diseases, an alternate hypothesis is that they are a normal—or even essential—part of cell biology [1], [2]. In support of the latter, lymphocyte-specific CNVs in immunologically important genes generate the genetic diversity of receptor molecules critical to their function [3]. Although V(D)J recombination is found only in the immune system, recent reports hint that lineage-specific somatic CNVs may be essential for healthy cellular differentiation and function in a number of organs such as the liver, pancreas and skin [4], [5]. It is unknown how these lineage-specific mammalian CNVs are formed—whether by a process similar to V(D)J recombination or by an alternative mechanism.
Although the role of many cell-type specific CNVs in mammals is unclear, lineage-specific CNVs are a normal aspect of cellular development in the fruit fly Drosophila melanogaster [6]. Lineage-specific CNVs form during Drosophila egg and larval development in polyploid cells via cycles involving DNA replication in the absence of cell division (endoreplication) [6]. In egg formation, somatic CNVs form by selective amplification of genomic regions containing chorion (eggshell) genes, which facilitates secretion of chorion proteins by the ovarian follicle cells [7], [8]. Drosophila somatic CNVs can also arise due to underreplication of certain genomic regions in the salivary glands, fat body and midgut of the larva [9]–[13]. While CNVs in Drosophila polyploid cells have been observed for more than 70 years [14], it is not known whether a similar mechanism is present in mammalian cells. However, the recent observation of human tissue-specific CNVs [1]–[5] suggests that somatic CNVs are as essential in mammalian cells as they are in Drosophila.
Mammals absolutely require polyploid placental cells, corollaries to Drosophila follicle cells, for pregnancy maintenance [15]. In the placenta, polyploidy is restricted to specialized trophoblast cells that invade and remodel the uterus to promote vascularization and other maternal adaptations to pregnancy [15]. In rodents, these cells—termed trophoblast giant cells (TGCs), have 50–1,000 copies of the genome per cell. While proper TGC function depends on their polyploidy content [16], [17], it is not known what aspect of polyploidy is necessary for fetal survival. As TGCs are a class of critical polyploid support cells analogous to Drosophila follicle cells, they may similarly use differential replication of the genome to achieve highly specialized function.
Previous studies have addressed possible CNVs in rodent TGCs. Ohgane et al. [18] used restriction landmark genomic scanning (RLGS) to analyze CpG islands in rat junctional zone TGCs during late gestation (days 18 and 20). They reported that ≥97% of the spots detected by RLGS were similar to diploid controls and therefore concluded that there are no TGC CNVs. Sher et al. [19] also argued against the existence of CNVs based on array Comparative Genomics Hybridization (aCGH) and quantitative real-time PCR experiments on mouse e9.5 implantation site TGCs. However, as there are several subtypes of TGCs which all have varying ploidy and functional significance during gestation [15], [20], CNVs could be present in a subset of cell types or only at certain developmental time points. Of particular interest are parietal TGCs, which have the highest degree of polyploidy [15], and are therefore an excellent candidate for differential replication of the polyploid genome. Genetic mouse mutants affecting the parietal TGCs predominantly die before e12.5 [15]–, suggesting that this is when developmentally important CNV would be required.
Here we report that somatic CNVs are a normal part of placental cell biology. We utilized whole genome sequencing (WGS) and aCGH to identify 47 reproducibly underrepresented (UR) domains in mouse e9.5 parietal TGCs, totaling 6% of the genome. Employing a variety of genomic techniques, we demonstrate that UR domains are marked in chromatin prior to endoreplication in TGC progenitor cells and gradually form during the first half of gestation. UR domains are highly enriched for genes involved in cell adhesion and neurogenesis, as well as for gene deserts. Furthermore, we specifically show that UR domains are due to underreplication rather than somatic deletions. Together, these data reveal that lineage-specific CNVs are inherent features of the TGC genome, which are established and regulated throughout placental development.
To investigate whether the 50–1,000 genomic copies in polyploid TGCs are uniformly replicated or contain CNVs, we used aCGH to compare genomic regions of mouse parietal TGCs (TGCs) and 2N embryos at e9.5 (Figure 1A, Figure S1A). We dissected four embryos and associated TGCs from one litter, representing pairs of genetically identical tissues, performed aCGH using the Agilent SurePrint G3 Mouse CGH Microarray Kit (two embryos/TGCs pooled per biological replicate), and analyzed the data using the R/Bioconductor package cghFLasso [21]. We identified 45 regions, reproducible between biological replicates, that were underrepresented within the TGC genome compared to the embryonic genome at a false discovery rate (FDR) of 0.0001, which we termed underrepresented (UR) domains (Figure 1B, Table S1). UR domains range in size from 1,037 kb to 9,429 kb (Table S1). In addition to the 45 UR domains common to both replicates, we found 30 domains specific to only one replicate (Figure 1B). However, when we reduced the FDR (to 0.01), 19/30 of these domains are found in both replicates, suggesting that while the degree of underrepresentation varies, UR domains form in specific regions of the genome. Importantly, we did not observe any overrepresented regions in TGCs (FDR = 0.0001).
We next asked whether UR domains were specific to TGCs, or whether they existed in diploid trophoblast cells or other endocycling polyploid cells. We used aCGH to compare the DNA of megakaryocytes (up to 64N) to embryos, placental disk cells (mostly 2N) to embryos, and cultured trophoblast stem cells (TS cells; 2N) to embryonic stem cells (ES cells; Figure 1C, Figure S1B, Figure S2). Megakaryocytes have no detectable underrepresented regions and display one region of overrepresentation common to both replicates, indicating that TGC UR domains are not simply explained by endocycling (FDR = 0.0001; Table S2). Placental disk cells lack any over or underrepresentation (FDR = 0.0001; Table S3), although greatly reducing the FDR (to ≥0.05) revealed a weak trend towards UR domains in the same locations as in TGCs, likely explained by the normal presence of a small number of TGCs within this population (Figure 1C, Figure S2). Finally, we identified several TS and ES specific CNVs, but these were different from the TGC UR domains and presumably represent adaptations to cell culture (Tables S2 & S3) [22]. These data suggest that UR domains are important genomic features unique to TGCs.
As Sher et al. [19] have argued against the existence of CNVs in e9.5 TGCs, we compared our aCGH data to theirs. Consistent with Sher et al., we did not find any CNVs in their data using the R/Bioconductor package cghFLasso and an FDR of 0.0001 [21]. However, greatly reducing the FDR (to >0.05) revealed a trend towards UR domains in the same locations as in our TGC data (Figure S3), similar to the report by Sher et al. of finding reduced copy number using a smaller threshold. Moreover, the Sher et al. data bears a striking resemblance to our placental disk data (Figure S3), suggesting that their study, on implantation site TGCs, is on a population of trophoblast cells more akin to the placental disk than to the parietal TGCs of the mural trophectoderm described in our study. In support of this, while parietal TGCs surround the entire conceptus, TGCs over the central region of the placental disk are smaller and less polyploid than those at the periphery [20]. Together, these data suggest that the parietal TGCs of the mural trophectoderm not only have a higher degree of ploidy, but also have specific CNVs compared to the rest of the placenta.
To quantitatively examine the extent of underrepresentation in TGCs, we performed paired-end WGS [23]. We sequenced (at 10× coverage) six individual e9.5 TGCs and their genetically matched embryos from three separate litters (2 individuals per litter; Table S4). To identify CNVs, we used a custom R/Bioconductor program based on CNVnator [24], which identifies CNVs at a p-value of 0.01. We found 47 reproducible UR domains on the autosomes in e9.5 TGCs in all samples (Table S5). UR domains range from 75 kb to 8,965 kb and cover 6% of the genome (138 Mbs of 2,717 total Mbs; Table 1). We next calculated the fold depletion of each UR domain from the normalized log 2 ratio of sequence coverage of TGC/embryo [25] and found an average reduction between 27% and 51%, with a median between 28% and 54% (Table 1). Further, the size and degree of depletion of UR domains correlate such that the larger the size of the domain, the greater the degree of underrepresentation (Figure 2A).
Next, we examined how much variation existed between individuals. First, we compared aCGH and WGS data, and found 43 UR domains common to both platforms (Figure 2B, Table 1, Figure S4, Table S1). Of the domains that differ, five additional domains in the WGS data are likely due to the greater sensitivity of WGS, as these domains can also be found in the aCGH data if the FDR is lowered (to 0.01). Three additional domains in the aCGH data are found in a majority of the WGS samples (present in four to five out of the six samples), suggesting a small amount of variability in UR domain formation (Tables S1 & S5). To examine this variability in more depth, we examined the six individual WGS samples. Besides the 47 UR domains common to all six samples, we also found underrepresented regions present in only a subset (Figure 2C, Figure S5, Table S5). In general, samples with the least number of UR domains have a subset of the domains found in the samples with the most (Figure 2C, Figure S5, Table S5). In addition, the size of a particular UR domain is generally smaller in samples with fewer UR domains (Figure 2D, Table S5). As the samples vary slightly in age, this suggests that UR domains amass over time, such that slightly younger placentas have fewer and smaller UR domains.
To test our hypothesis that UR domains develop over time, we performed WGS on e8.0 TGCs/embryos (one litter per replicate) and compared these results to e9.5. We found 24 domains common to both biological replicates at e8.0, versus 47 domains common to all samples at e9.5 (Figure 3A & 3B, Figure S6). All e9.5 individuals have 23 of these domains with 5/6 individuals containing the remaining domain (Figure 3B). We also found 10 domains unique to one of the two biological replicates at e8.0; 10/10 of these domains are contained in all e9.5 individuals (Figure 3B). Finally, we found that both size and degree of depletion of UR domains significantly increase between e8.0 and e9.5 (Figure 3C). Overall, as all UR domains at e9.5 are also present at e8.0, and UR domains at e9.5 are also more numerous, larger and more depleted, we propose that they are gradually established during early gestation.
We next asked whether the number and degree of depletion of UR domains continues to increases throughout development. We performed aCGH on TGCs/embryos collected from the second half of gestation—e11.5, e13.5, e16.5—and compared them to e9.5. Out of 45 UR domains present in both biological replicates at e9.5 (FDR = 0.0001), 22 of these are present in all biological replicates at e11.5, e13.5 and e16.5, and an additional 10 (32/45) are present in all samples except for one of the e16.5 replicates (Figure 3D & 3E, Figure S7). We next examined size, and found that the 32 common domains are significantly larger than UR domains that arise later in development (the 147 not present at e9.5; Figure 3D & 3E, Figure S7). However, unlike between e8.0 and e9.5, where the degree of depletion expanded, we found no significant change from e9.5 to e16.5 (Figure 3F). Although, UR domains slightly trend towards becoming less depleted over time (Figure 3D & 3F, Figure S7). There is also more intrinsic variability later in gestation, as the median degree of depletion between biological replicates at both e13.5 and e16.5 is significantly different (Figure 3F). The differences between UR domains in early (e8.0–e9.0) and later (e11.5–e16.5) gestation correlate with previous data showing that TGC polyploidy drastically increases until e10.5, and endocycling ends by e13.5 [20]. These data suggest that the increase in UR domain size and degree of underrepresentation from e8.0 to e9.5 is linked to the robust endocycles of early gestation. Furthermore, the termination of endocycles in later development may free cellular machinery to increase representation levels in UR domains.
We also found 33 overrepresented regions at e11.5–e16.5 that are not present at e9.5 (Figure 3D & 3E, Figure S7). We examined gene content of overrepresented regions common to at least two staged biological replicates (10/33), but did not find any annotated genes. Thus, while new CNV regions form during late gestation, they are more stochastic, less reproducible, and significantly smaller than those conserved between all stages.
We next examined whether UR domains are also generated in vitro when differentiating TS cells into TGCs. To this end, we performed aCGH on purified TGCs harvested at 3, 5 and 7 days after differentiation [26]–[28] (Figure S8). Similar to in vivo, in vitro cells generate the same UR domains and also develop these over time (FDR = 0.0001, Figure 4A & 4B, Figure S8). At day 3, only one biological replicate has any of the UR domains found in vivo at e9.5 (3/45). At day 5, both replicates contain 1/45 domains, and one replicate contains 21/45 domains. At day 7, both replicates contain 34/45 UR domains, and one replicate contains 43/45 domains. Remarkably, in vitro cells generate the same UR domains as their in vivo counterparts (Figure 4A & 4B, Figure S8), strongly suggesting that the formation of these UR domains is a fundamental feature of TGC development.
Next, we asked whether genes contained within e9.5 TGC UR domains were enriched for certain biological functions. We found that UR domains are significantly depleted of both protein-coding and non-coding genes as expected by chance (386 observed vs. 617 expected, 0.63× enrichment, p<0.001) and when compared to the rest of the genome (Figure 4C). Further, these domains are significantly enriched for 1 Mb gene deserts (regions without any Ensembl annotations; 47 observed vs. 9 expected, 4.96× enrichment, p<0.001). In total, 386 genes are present within UR domains, 106 of which are functionally annotated. When we examined these 106 genes for function using GOTERMFINDER [29], the top enrichment categories are biological adhesion (p = 2.31×10−9) and related categories, followed by neuron projection development (p = 4.23×10−8), and related neurogenesis categories. These categories were not enriched when we performed the same analyses on a list of genes found in a random set of regions that have the same length and chromosome distribution. Finally, using 3′ RNA-Seq (3SEQ) [30] from both in vivo and in vitro TGCs, we compared expression of the genes to the degree of representation and found that genes in UR domains are either not expressed or have much lower levels of transcription than genes in regularly represented regions (Figure 4D & 4E). Overall, our data show that there are specific classes of genes enriched within the UR domains and these genes are generally not expressed, raising the possibility that UR domains function to limit the expression of a particular subset of genes in TGCs.
To test whether UR domains are characterized by a specific chromatin state, we performed ChIP-Seq using anti-H3K27ac, anti-H3K4me1, anti-H3K4me3, anti-H3K9me3, and anti-H3K27me3 in both in vitro TS cells and derived TGCs [31]. We used MACS2 to determine the normalized fold change for histone occupancy [32] and then used the Pearson correlation (R) to determine how the degree of representation (normalized log 2 of e9.5 WGS) correlates with signals from histone marks. In both TGCs and TS cells, we find that UR domains tend to co-localize with the repressive marks H3K9me3 and H3K27me3 (Figure 5). Conversely, UR domains have underrepresentation of the active chromatin marks H3K4me3, H3K4me1 and H3K27ac (Figure 5). These results demonstrate that UR domains do not occur in active regions of the genome and that they are marked in the 2N progenitor cells (TS cells). Interestingly, UR domains are only a fraction of genomic heterochromatin (Figure 5B & 5C). All UR domains have increased signals for repressive histone marks and only weak signals for active histone marks. However, not all regions of the genome having repressive marks but not active marks are associated with a UR domain. Overall, this demonstrates that UR domains have a heterochromatic signature, but represent only a subset of heterochromatin.
We further examined the relationship between UR domains and heterochromatin using an alternative statistical method. We asked whether the histone marks are significantly enriched or depleted in our defined list of UR domains compared to what would be expected by chance [31]. Similar to our correlation analysis, marks associated with transcriptional activation (H3K4me3, H3K4me1 and H3K27ac) are significantly depleted in UR domains (p<0.001; Table 2). Conversely, the repressive mark H3K9me3 is enriched within UR domains (p<0.001; Table 2). Interestingly, while the repressive mark H3K27me3 is also enriched within UR domains in TS cells, it is depleted within UR domains in TGCs (p<0.001; Table 2). This observation agrees with previous data where extraembryonic cells have lower levels of H3K27me3 methylation than embryonic cells [33], and suggests that H3k27me3 is not critical for UR domain maintenance. Together, our data show that UR domains have a heterochromatic signature, both in TGCs and in their 2N progenitors.
To examine whether UR domains are caused by genomic deletions, we carried out somatic structural variant analysis using paired-end sequencing data from the six TGC and matched embryo samples with the program SMASH [34]. If UR domains are caused by acquired genomic deletions, we would expect to find multiple library inserts that fully span the deleted regions (“discordant” paired-end reads; Figure S9). While we did detect sample-specific CNVs, we did not detect somatic deletions common to all of the six TGCs, but not the embryos. Moreover, the probability of not detecting a given deletion in each of the six samples is extremely low (p = 2•×10−5). These data show that UR domains are not a result of somatic chromosomal deletions.
Since our WGS data does not support genomic deletions as the source of UR domains, we investigated whether they may be due to underreplication (Figure S9B). In 2N cells, replication timing is precisely regulated such that specific regions of the genome are replicated early in S phase while others are replicated late in S phase [35]. To test whether UR domain formation is caused by incomplete replication of regions that are normally replicated late in 2N TS cells, we first generated a replication timing profile of TS cells. To this end, we captured early- and late-replicating regions in TS cells by pulsing an asynchronous cell culture with BrdU to label replicating DNA followed by FACS, and then used aCGH to compare early and late BrdU-containing DNA [36]. Next, we compared late-replicating regions in TS cells to UR domains. Using the Pearson correlation (R), we found that UR domains correlate with late replication (Figure 6A). Also, 47/47 TGC UR domains reside within late-replicating regions in TS cells (Figure 6B, Table S6). UR domains are significantly smaller than the late-replicating regions that they are nested in (Figure 6C; Table S6), suggesting that they are a subset of these larger regions.
Finally, as only 45 of the 211 late-replicating regions contain a UR domain (Figure 6D, Table S6), we asked what distinguishes the late-replicating regions that form UR domains from those that do not. While there is no significant difference in the degree of late replication between these classes, late-replicating regions that contain UR domains are significantly larger (Figure 6E). However, size is not the sole characteristic determining where UR domains form, as not all regions greater than a certain size contain a UR domain. We next investigated gene content and found that late-replicating regions that contain UR domains also contain significantly fewer genes than those that do not (Figure 6F). These regions are also preferentially enriched for 1 Mb gene deserts (58 observed vs. 18 expected, 3.16× enrichment, p<0.001). Together, our data show that UR domains form from a specific class of late-replicating, heterochromatic regions with low gene content, suggesting that UR domains are not simply a byproduct of late-replicating heterochromatin, but are a precisely regulated subset.
We report here the first mammalian example, outside of the immune system, of lineage-specific CNVs being an integral part of normal cell biology and development. Notably, we show that CNVs in placental cells form via a novel mechanism unrelated to V(D)J recombination. Using both aCGH and high-throughput WGS, we identified 47 reproducible underrepresented domains in mouse parietal TGCs totaling 138 Mbs, or 6% of the genome. We found that UR domains are highly enriched for genes involved in cell adhesion and neurogenesis, as well as for gene deserts. Furthermore, we specifically show that UR domains are due to underreplication of a specialized type of heterochromatin, rather than acquired genomic deletions. Our data reveal that lineage-specific CNVs are a normal aspect of the TGC genome that are established and regulated during gestation.
Only a subset of heterochromatic, late-replicating regions form UR domains, suggesting that UR domains are not simply a byproduct of late-replicating heterochromatin, but are precisely regulated. We propose that either this is dictated by genomic structure or that there are specific DNA binding proteins that define UR domains. We favor the latter model based on parallels found in Drosophila, whereby mutants for Suppressor of Underreplication (SuUR) have underreplicated domains that become replicated to normal levels [12], [13], [37]. However, SUUR protein does not appear to be present in species outside the Drosophilids, and we have not found any SuUR homologs in mice via BLAST, raising the possibility that presently unknown proteins in mammals may be regulating this process.
Lineage-specific CNVs are an overlooked aspect of the mammalian genome. Although recent data suggests that they are widespread [1]–[5], their identification and functional study has not been carried out systematically. Identification of CNVs may be particularly difficult to define in primary tissues, due to high background of cells lacking CNVs. In support of this, Abyzov et al. [4] found a low frequency of somatic CNV in human fibroblasts. Further, even in more homogenous populations, relatively small degrees of CNV may mask their presence. Van Heesch et al. [38] found tissue-specific CNVs in rat blood, brain, liver and testis, where the degree of underrepresentation does not exceed 50%. While Van Heesch et al. conclude that their findings were the result of systematic bias in DNA isolation procedures, they could never get rid of these CNVs using any analytical or experimental approach. Moreover, Manukjan et al. [39] suggest that Van Heesch et al. are identifying the signature of replication timing in their CNV analyses due to the use of proliferating cells. Intriguingly, this suggests that, analogous to polyploid TGCs in the placenta, underreplication may be crucial in organs containing a highly proliferative population of 2N cells.
While CNVs in Drosophila polyploid cells have been characterized for more than 70 years [14], our work demonstrates for the first time that CNVs are a normal aspect of mammalian development. The rarity of endoreplicating polyploid cells in animals suggests that CNVs in mouse and Drosophila arose independently [6], and therefore may have species-specific differences. While Drosophila CNVs are typically 90% underrepresented, mouse CNVs are never more than 50%. We strongly suggest that there are UR domains in both mouse and Drosophila polyploid cells, and that the presence of these domains in both taxa is an example of convergent evolution due to similar selective pressures, indicative of functional importance. As both mice and flies have a fast rate of early development compared to related species, formation of UR domains could be an integral part of accelerating the cell cycle, and therefore be a key mechanism behind their rapid life cycles.
UR domains are a unique feature of the TGC genome, suggesting that they play a central role in placental function and pregnancy. Consistent with this, UR domains are enriched for specific classes of genes involved in cell adhesion and neurogenesis. Intriguingly, there is evidence that downregulation of both classes of proteins is crucial for placental function. Downregulation of cell adhesion genes is necessary for trophoblast invasion in both mice and humans [40], [41]. Further—and quite remarkably—Liao et al. [42] found that upregulation of genes in the SLIT/ROBO neuronal guidance system in the human placenta is associated with the pregnancy disease pre-eclampsia. UR domain formation could also enable TGCs to simply save materials and time, a hypothesis that has been proposed for polyploidy in general [43]. TGCs are essential during the first half of gestation, when it is absolutely critical for the rapidly growing embryo to establish a connection with the mother [15], [44]. Formation of UR domains could allow for more rapid maturation of TGCs by allowing replication initiation to proceed without waiting for replication of nonessential regions of the genome. In support of this, UR domains represent a significant part of the genome, 6% (138 Mbs of 2,717 total Mbs), and therefore the cell would require considerable resources to fully replicate these regions. Together, functional evidence and convergent evolution suggest that UR domains are a critical element during pregnancy. Regardless, placental UR domains are the first mammalian example, outside of the immune system, of lineage-specific CNVs being an integral part of normal cell biology and development.
All animal work has been conducted according to relevant U.S. and international guidelines. Specifically, all experimental procedures were carried out in accordance with the Administrative Panel on Laboratory Animal Care (APLAC) protocol and the institutional guidelines set by the Veterinary Service Center at Stanford University (Animal Welfare Assurance A3213-01 and USDA License 93-R-0004). Stanford APLAC and institutional guidelines are in compliance with the U.S. Public Health Service Policy on Humane Care and Use of Laboratory Animals. The Stanford APLAC approved the animal protocol associated with the work described in this publication.
129-Elite, C57BL/6 and pregnant C57BL/6 mice were obtained from Charles River. Copulation was determined by the presence of a vaginal plug the morning after mating, and embryonic day 0.5 (e0.5) was defined as noon of that day. TGCs and embryos were dissected in 1× PBS (1∶10 10× PBS, pH = 7.4; Gibco) and stored on ice until further processing. After removal of the decidua, parietal TGCs of the mural trophectoderm [15] were dissected away from the placental disk, and, when possible, Reichert's membrane (Figure S1A). TGCs were identified by their extremely large cell size (Figure 1A). Using single-nucleotide polymorphism data from F1 crosses, TGCs were predicted to have, at the most, approximately 5% contamination by maternal cells (Hannibal & Baker, unpublished data). Placental disk tissue was gathered from e13.5 placental disks after the removal of the decidua and obvious parietal TGCs. For gathering 2N genomic DNA, at e8.0, the entire embryo was collected; at e9.5, the embryo body, after removal of obvious organs and head (removed at otic vesicle), was collected; and at later stages, limbs, or a mixture of limbs and the tail, were collected (Figure S1A).
For confocal imaging, TGCs/embryos were fixed in 4% paraformaldehyde at 4°C overnight. Samples were stained with 0.5 µg/mL DAPI (Life Technologies) in 1× PBS overnight, washed in 50% glycerol/1× PBS and stored in 70% glycerol/1× PBS. Confocal images were taken on a Leica DM IRE2 inverted microscope using the Leica SP2 software package, located in the Stanford Cell Sciences Imaging Facility.
Trophoblast stem cells were cultured as described in Chuong et al. [31] following [27]. TS cells were differentiated into parietal TGCs by replacing the FGF, Activin and Heparin in the media with retinoic acid [27], [28]. Mature TGCs are seen after 4–6 days of differentiation [26] and were collected on days 3, 5 and 7. TGCs/TS cells were further isolated for aCGH by placing cultured cells over a two-step density gradient (1.5% BSA over 3% BSA in a 15 mL tube; Figure S1B). TGCs sank to the bottom of the tube while the smaller TS cells stayed in the upper fraction.
The embryonic stem cell line CGR8 is a germ-line competent cell line established from the inner cell mass of a 129 e3.5 male pre-implantation embryo [45]. ES cells were cultured feeder-free on 0.1% gelatin coated plates. The ES cell medium was prepared by supplementing knockout DMEM (Invitrogen) with 15% FBS, 1 mM glutamax, 0.1 mM nonessential amino acids, 1 mM sodium pyruvate, 0.1 mM 2-mercaptoethanol, penicillin/streptomycin, and 1000 units of leukemia inhibitory factor (LIF; Millipore). Cell culture was maintained at 37°C with 5% CO2.
Megakaryocytes were derived and cultured as described in [46]. Briefly, fetal livers were dissected from e13.5 C57BL/6 embryos in Hanks' Balanced Salt Solution and placed in DMEM with 10% FBS supplemented with 100 ug/mL penicillin-streptomycin (Invitrogen). Livers were pooled based on sex of the embryo (males pooled and females pooled separately). To make a single cell solution, livers were aspirated through a progression of 18G, 21G and 23G needles. To promote differentiation into megakaryocytes, cells were cultured for five days in media containing thrombopoietin (TPO; R&D Systems) at 37°C with 5% CO2. Successful differentiation was identified by 1) the presence of large cells (megakaryocytes) and by 2) FACS to confirm up to 32N ploidy. For FACS, propidium iodide stained samples were run on a Cytek DxP10 modified Facscan (Cytek Technologies, BD Biosciences) using the blue laser. Approximately 10,000 events per sample were collected. Data was analyzed using FlowJo (Treestar, Inc.). Megakaryocytes were isolated for aCGH by placing cultured cells over a two-step density gradient (1.5% BSA over 3% BSA in a 15 mL tube; Figure S1B). Megakaryocytes sank to the bottom of the tube while smaller, undifferentiated, cells stayed in the upper fraction.
Genomic DNA was extracted from fresh tissue and cultured cells using the DNeasy Blood & Tissue Kit (Qiagen). Before column purification, in vivo and in vitro samples were digested with proteinase-K (600 mAU/ml solution or 40 mAU/mg protein) overnight and for 10 minutes, respectively, at 56°C, followed by a 4 minute incubation with RNase A (100 mg/mL; Qiagen DNeasy Blood & Tissue Kit). If necessary, DNA was further concentrated via ethanol/sodium acetate precipitation following standard protocols.
For arrays performed on DNA from TGCs, placental disks and embryonic controls, genomic DNA from two individuals in the same litter were pooled for each condition. For megakaryocyte arrays, cells derived from 5–6 livers from a single litter were pooled for each condition. For controls for the megakaryocyte array, three embryos (subset of the litter from which livers were collected from) were pooled for each condition. For arrays performed on DNA from cultured cells, two replicates from different passages were used (5 million cells each). For each condition, approximately 4 µg DNA was sent to the Biomedical Genomics Core at the Research Institute at Nationwide Children's Hospital (Columbus, OH) for processing with the SurePrint G3 Mouse CGH Microarray Kit, 4×180 k (Agilent). For all arrays performed on DNA from in vivo tissue, to ensure that the arrays detect copy number variation, duplicates consist of 1) female test versus male control and 2) male test versus female control.
aCGH data was analyzed using the R/Bioconductor package cghFLasso, which utilizes reference arrays in conjunction with a FDR [21]. An FDR of 0.0001 was used in order to examine all of the autosomes simultaneously. To determine which array to use as the reference, several analyses were performed. The TS versus ES array exhibited specific CNVs, presumably due to genomic adaptations to culturing [22]. The megakaryocytes displayed only a small region of overrepresentation and the placental disk array did not display any CNVs (FDR = 0.0001). However, as the placental disk has a small amount of underrepresentation in reproducible areas of the genome (FDR≥0.05), the megakaryocyte array was used as the reference for the remainder of the analyses. aCGH data was plotted using cghFLasso [21]. For comparison with data from Sher et al. [19], data was retrieved from Gene Expression Omnibus series: GSE45787. To compare aCGH data from Sher et al. to data presented here, results were plotted using a custom R/Bioconductor program.
For WGS, for stages e9.5 and older, genomic DNA from one individual was used for each replicate, and for stage e8.0, 5–7 individuals from one litter were used for each replicate. Libraries for WGS were prepared from 40–50 ng genomic DNA using the Nextera TruSeq Dual Index Paired End Kit (Illumina) following manufacturer's instructions with the following modification: the Qiagen MinElute Reaction Cleanup Kit (Qiagen) was used to cleanup Tagmented DNA. Library quality was assessed using Qubit and Bioanalyzer, and sequenced on the Illumina HiSeq 2000 at approximately 10× coverage (Table S4) at the Stanford Center for Genomics and Personalized Medicine. 101 bps from each of the paired-ends were sequenced and sequencing reads were aligned using either the DNAnexus mapper [47] or the Novocraft Novoalign program against the mouse reference genome (mm9). Data was analyzed using custom R/Bioconductor programs and SMASH [34]. To compare aCGH versus WGS data, results were plotted using a custom R/Bioconductor program.
The final UR domain list was generated using e9.5 WGS data and a custom R/Bioconductor program with the following criteria: neighboring data points with normalized log 2 ratio of TGCs/embryo ≤−0.3. These criteria were decided upon based on the program CNVnator [24], which, while identifying UR domains with both large and small degrees of underrepresentation at a p-value of 0.01, systematically missed UR domains that are closely spaced together, which our program rectifies.
To calculate the significance of overlap between datasets, a binomial test was used to determine whether the observed overlap for the datasets was significantly greater than an expected overlap based on the average of 1,000 randomized datasets [31]. To randomize each dataset, regions were shuffled within bins according to their chromosomal distribution and distance from gene transcriptional start sites (including 1 kb, 10 kb, 100 kb, 1,000 kb, and >1,000 kb bins).
Total RNA was extracted from fresh in vivo tissue by homogenizing in TRIzol Reagent (Life Technologies/Ambion) and total RNA was prepared following manufacturer's instructions. Total RNA from three individuals from the same litter were combined to make each library. mRNA was isolated from 10–20 µg of total RNA using Dynabeads Oligo(dT)25 (Life Technologies/Ambion). 3SEQ Libraries were prepared from mRNA following [30]. Briefly, mRNA was heat sheared for 7.5 minutes to produce an average fragment size range of 100–400 bp, then used to generate cDNA libraries using a custom oligo dT primer containing Illumina-compatible adapter sequence. cDNA fragments were end-repaired and ligated to standard Illumina adapters. Size-selection was performed using E-gel SizeSelect agarose gels (Invitrogen), products were PCR amplified for 15 cycles and purified using Ampure XP beads (Beckman Coulter). Library quality was assessed using Qubit and Bioanalyzer, and sequenced on the Genome Analyzer IIx at the Stanford Center for Genomics and Personalized Medicine.
Total RNA was extracted and 3SEQ libraries were constructed for cultured TGCs as described in Chuong et al. [31]. Two replicates from different passages (10 million cells each) were used. 3SEQ data for TS cells was retrieved from Gene Expression Omnibus series: GSE42207 [31].
Sequences were aligned to the mouse (mm9) genome using the DNAnexus mapper [47] and raw counts for sense reads were analyzed using Unipeak 1.0 [48]. Regions of transcription were associated with the nearest ENSEMBL gene 3′ UTR within 5 kb. Data were normalized and expression levels were analyzed using the R/Bioconductor package DESeq [49].
ChIP-seq and ChIP-seq analysis were performed as described in Chuong et al. [31] using the ChIP Assay kit (Millipore) following manufacturer's instructions. Briefly, 20 million cultured TGCs were cross-linked in 2% formaldehyde for 15 minutes, and sonicated for 12 cycles (30 seconds on/off) at 60% amplitude to produce a fragment range of 300–600 bp. Immunoprecipitation was performed with 2–5 µg of antibody (H3K4me3: ActiveMotif, 39159; H3K27me3: ActiveMotif, 39535; H3K27ac: Abcam, ab4729; H3K9me3: Abcam, ab8898; H3K4me1: Abcam, ab8895) conjugated to 50 µl of protein G Dynabeads (Invitrogen) overnight. Following washing and elution of DNA per manufacturer's instructions, libraries were prepared using the Illumina genomic DNA preparation kit using barcoded linker adapters, and sequenced on the Illumina HiSeq 2000 at the Stanford Center for Genomics and Personalized Medicine. ChIP-Seq data for TS cells was retrieved from Gene Expression Omnibus series: GSE42207 [31].
High-quality reads were aligned to the mm9 genome assembly using BWA 0.5.9 [50], retaining only unique alignments. Peaks were called using MACS2 2.0.10 [32]. The “bigwig_correlation” script from the Cistrome package [51] was used to generate genome-wide correlation plots between ChIP profiles and underrepresented profiles.
Cultured TS cells were incubated for two hours at 37°C in the dark with a final concentration of 100 µM BrdU (Sigma Aldrich B5002). Genome-wide replication timing was analyzed as previously described [36]. Briefly, cells were dissociated into a single-cell suspension and nuclei were isolated. DNA was subsequently stained with propidium iodide and cells were FACS sorted into early and late S-phase fractions based on their DNA content. DNA from early and late S-phase fractions was purified by immunoprecipitation of the BrdU-substituted nascent DNA (BrdU-IP). Three replicates from different passages (two million cells each) were used. Data was normalized following [36]. The R/bioconductor package DNAcopy was used to define replication timing domains based on the similarity in values (a constant value across a segment defines a domain) [36]. Regions called by DNAcopy were confirmed on the genome browser. The “bigwig_correlation” script from the Cistrome package [51] was used to generate genome-wide correlation plots between replication timing profiles and underrepresented profiles.
SuperSeries Gene Expression Omnibus (GEO) accession number for aCGH, 3SEQ, ChIP-Seq, and replication timing data: GSE50585.
Smoothed replication timing data can also be found at: http://www.replicationdomain.com/
BioProject accession number for WGS: PRJNA213010
|
10.1371/journal.pmed.1002237 | Serotonin transporter gene (SLC6A4) polymorphism and susceptibility to a home-visiting maternal-infant attachment intervention delivered by community health workers in South Africa: Reanalysis of a randomized controlled trial | Clear recognition of the damaging effects of poverty on early childhood development has fueled an interest in interventions aimed at mitigating these harmful consequences. Psychosocial interventions aimed at alleviating the negative impacts of poverty on children are frequently shown to be of benefit, but effect sizes are typically small to moderate. However, averaging outcomes over an entire sample, as is typically done, could underestimate efficacy because weaker effects on less susceptible individuals would dilute estimation of effects on those more disposed to respond. This study investigates whether a genetic polymorphism of the serotonin transporter gene moderates susceptibility to a psychosocial intervention.
We reanalyzed data from a randomized controlled trial of a home-visiting program delivered by community health workers in a black, isiXhosa-speaking population in Khayelitsha, South Africa. The intervention, designed to enhance maternal-infant attachment, began in the third trimester and continued until 6 mo postpartum. Implemented between April 1999 and February 2003, the intervention comprised 16 home visits delivered to 220 mother–infant dyads by specially trained community health workers. A control group of 229 mother–infant dyads did not receive the intervention. Security of maternal-infant attachment was the main outcome measured at infant age 18 mo. Compared to controls, infants in the intervention group were significantly more likely to be securely attached to their primary caregiver (odds ratio [OR] = 1.7, p = 0.029, 95% CI [1.06, 2.76], d = 0.29). After the trial, 162 intervention and 172 control group children were reenrolled in a follow-up study at 13 y of age (December 2012–June 2014). At this time, DNA collected from 279 children (134 intervention and 145 control) was genotyped for a common serotonin transporter polymorphism. There were both genetic data and attachment security data for 220 children (110 intervention and 110 control), of whom 40% (44 intervention and 45 control) carried at least one short allele of the serotonin transporter gene. For these 220 individuals, carrying at least one short allele of the serotonin transporter gene was associated with a 26% higher rate of attachment security relative to controls (OR = 3.86, p = 0.008, 95% CI [1.42, 10.51], d = 0.75), whereas there was a negligible (1%) difference in security between intervention and control group individuals carrying only the long allele (OR = 0.95, p = 0.89, 95% CI [0.45, 2.01], d = 0.03). Expressed in terms of absolute risk, for those with the short allele, the probability of secure attachment being observed in the intervention group was 84% (95% CI [73%, 95%]), compared to 58% (95% CI [43%, 72%]) in the control group. For those with two copies of the long allele, 70% (95% CI [59%, 81%]) were secure in the intervention group, compared to 71% (95% CI [60%, 82%]) of infants in the control group. Controlling for sex, maternal genotype, and indices of socioeconomic adversity (housing, employment, education, electricity, water) did not change these results. A limitation of this study is that we were only able to reenroll 49% of the original sample randomized to the intervention and control conditions. Attribution of the primary outcome to causal effects of intervention in the present subsample should therefore be treated with caution.
When infant genotype for serotonin transporter polymorphism was taken into account, the effect size of a maternal-infant attachment intervention targeting impoverished pregnant women increased more than 2.5-fold when only short allele carriers were considered (from d = 0.29 for all individuals irrespective of genotype to d = 0.75) and decreased 10-fold when only those carrying two copies of the long allele were considered (from d = 0.29 for all individuals to d = 0.03). Genetic differential susceptibility means that averaging across all participants is a misleading index of efficacy. The study raises questions about how policy-makers deal with the challenge of balancing equity (equal treatment for all) and efficacy (treating only those whose genes render them likely to benefit) when implementing psychosocial interventions.
Current Controlled Trials ISRCTN25664149
| It has been shown that individuals with the short form of a gene involved in nerve signaling in the brain are generally sensitive, or “susceptible,” to psychosocial interventions (i.e., they benefit), whereas individuals with the long form of the gene are insensitive, or “nonsusceptible,” deriving little benefit from the same intervention.
This study was conducted to investigate whether such a gene × intervention interaction was present in a previously conducted randomized controlled home-visiting intervention trial designed to increase levels of secure maternal-infant attachment at 18 mo of age in a population in South Africa.
Genotyping revealed that 40% of the study sample (n = 89) carried at least one short allele of the serotonin transporter gene, while 60% of the sample (n = 131) carried two long alleles of this gene.
We reanalyzed the original maternal-infant attachment results and found that the increase in secure attachment found for the intervention in the original trial (odds ratio = 1.7, effect size d = 0.29) was almost entirely attributable to children with the short form of the gene (odds ratio = 3.86, effect size d = 0.75). For children not carrying the short allele, there was no significant increase in secure attachment with the intervention (odds ratio = 0.95, effect size d = 0.03).
Children carrying the short form of the gene whose mothers received the intervention were 3.86 times more likely to be securely attached at 18 mo (84% were secure) than children carrying the short form whose mothers did not receive the intervention (58% were secure).
Among children without the short form, the number who were securely attached did not differ based on whether their mothers received the intervention (71% were secure) or not (70% were secure).
These results indicate the importance of genetic differences when considering the efficacy of psychosocial interventions aimed at improving developmental outcomes in children living in the context of socioeconomic adversity.
The findings also raise complex questions regarding spending limited resources implementing costly interventions on individuals who may not benefit because of their genetic makeup and who may comprise the majority of the population (60% in this case), but where these same interventions may nevertheless confer large benefit for a subgroup of genetically susceptible individuals (40% in this case).
| Early childhood development is increasingly recognized as a public health priority that requires attention and investment, and specific targets and indicators addressing this area are included in the recent Sustainable Development Goal (SDG) framework and the United Nations Secretary General’s Global Strategy for Women’s, Children’s and Adolescents’ Health [1,2]. With the knowledge that more than 250 million children younger than 5 y globally will fail to reach their full developmental potential [3], clear recognition of the damaging effects of poverty on early childhood development has fueled an interest in psychosocial interventions aimed at mitigating these harmful consequences in order to promote lifelong health and prosperity.
Gains in child development have been generated by psychosocial interventions to improve child nutrition and development and to address the mental health of caregivers [4]. Interventions in the early years are cost-effective [5], can reduce inequity [6], and have been shown to have an impact on adult health outcomes [7]. Parenting programs focused on early child development are a good example of a delivery mechanism for the prevention and reduction of childhood disadvantage [8]. However, the efficacy of early child development interventions has, for the most part, been quantified by averaging individual outcomes across an entire sample and then, to varying extents, controlling for other factors, mostly of an extrinsic nature (e.g., maternal, environmental, and demographic factors). However, if individual differences in susceptibility to intervention are not considered (e.g., temperament, biological stress reactivity, and genetic differences), averaging of outcomes could lead to a misleading assessment of efficacy, because weaker effects on less susceptible individuals would dilute the estimation of effects on those more liable to respond [9,10]. In order to optimize the benefit of interventions, accurate evaluation of efficacy—including for whom the intervention does and does not work—is essential.
In 2003, Caspi et al. [11] first suggested that there were genetic differences in susceptibility to environmental influence in relation to depression. They reported observational findings indicative of a gene × environment (G×E) interaction. When individuals had experienced childhood maltreatment or stressful life events in early adulthood, carriers of the short allele (short/short or short/long genotype) of a length polymorphism in the regulatory region of the serotonin transporter gene (termed the 5HTT-linked polymorphic region [5HTTLPR]) were more liable than carriers of two long alleles (long/long genotype) to suffer from depression [11]. Subsequent studies have continued to yield evidence of G×E interaction at the 5HTTLPR locus [12].
While this evidence is important, only recently have investigators begun to make use of interventions where participants are randomly allocated to different environmental conditions to test the causal status of G×E interactions. Randomized controlled trials (RCTs) have been estimated to have 13 times the statistical power of G×E studies in which individuals (and therefore genotypes) are not randomly assigned to categorically different environments (e.g., intervention and no intervention) [10]. RCTs have, inter alia, two other big advantages: precluding gene–environment correlations (where genes and environments “choose” one another) and overcoming many of the problems with accurately measuring the environment of interest (e.g., maltreatment of children) [10]. A recent meta-analysis of 11 field RCTs investigated gene × intervention (G×I) interactions involving several genes thought to confer greater susceptibility to interventions, including 5HTTLPR [9]. The meta-analysis showed that intervention benefits were significantly stronger in those with susceptible genotypes, with an overall odds ratio (OR) of 3.17 (p < 0.01) for individuals with susceptible genotypes, compared to only 1.16 (p = 0.6) among individuals with nonsusceptible genotypes. Although striking, this meta-analytic evidence of genetic differential susceptibility to intervention rests upon an empirical base beset by several limitations. These include reliance on small sample sizes and special populations (e.g., maltreated children [13–15] or stroke patients [16]), mixed results for some genes [9], lack of ethnic and socioeconomic diversity (mostly middle-class white individuals) [13], and inconsistencies across different ethnic groups [17]. Crucially, given the interest in tackling poverty through early child development interventions in low- and middle-income countries (LMICs), it is striking that, to the best of our knowledge, no study has investigated differential susceptibility to intervention outside the US and Europe.
Security of attachment, which can be objectively and reliably assessed in infancy [18], is an important indicator of positive early socio-emotional development [19–21]. A range of studies have shown that, compared to insecure attachment, secure attachment is associated with better subsequent outcomes, including reduced externalizing behavior problems and better social competence [22,23]. There is also emerging evidence that secure infant attachment and the sensitive maternal care that promotes it are linked to better growth, physical health, and cognitive development [19–21,24]. Promoting secure mother–infant attachment is therefore an important focus for prevention studies, and, indeed, a wide range of interventions have been developed that appear to be successful in promoting secure attachment [25]. The majority of these interventions target the responsiveness of the mother’s caregiving behavior in relation to the infant’s attachment cues and communications, and have been delivered as primary or secondary prevention in a wide range of contexts within high-income countries.
Here, we report the results of a reanalysis of data from a RCT in which we test for a G×I interaction in early child development. Our study focuses on 5HTTLPR as a potential genetic moderator of the efficacy of a home-visiting intervention that was designed to improve attachment in mother–infant dyads in an impoverished isiXhosa-speaking community in South Africa. The intervention, known as Thula Sana (“hush baby” in isiXhosa), was a manualized home-visiting parenting program that aimed to promote security of infant attachment (the primary outcome of the original trial) by enhancing maternal sensitivity to infant characteristics and communication and by supporting management of infant distress [26,27].
This intervention was evaluated in an individually randomized controlled trial over a 4-y period, beginning in 1999, with a sample of 449 pregnant women. At infant age 18 mo, infant attachment status was assessed using standardized laboratory procedures [26]. Compared to controls who received no intervention, infants in the intervention group were significantly more likely to be securely attached to their primary caregiver (OR = 1.7, p = 0.029, 95% CI [1.06, 2.76]) [26]. This result equates to an effect size of Cohen’s d = 0.29, consistent with previous reports of modest effect size estimates for such interventions [4]. Further results on maternal sensitivity and maternal depression are reported in Cooper et al. [26]. However, this first report of the trial did not take into account the issue of differential susceptibility, and it is therefore possible that differences in efficacy for susceptible and nonsusceptible individuals may have been overlooked.
A follow-up study of the original Thula Sana cohort at 13 y of age provided the opportunity to address the possibility of genetic differential susceptibility by collecting DNA from both children and mothers. We focused on 5HTTLPR, the polymorphism most frequently investigated with respect to attachment outcomes and related processes in G×E studies [13,28–31]. To date, all studies implicating 5HTTLPR in genetic differential susceptibility to environment for attachment outcomes have been observational and have yielded mixed results (for reviews, see [17,32]). The current study circumvented the inherent limitations of observational research designs [9–11] by testing for a G×I interaction between 5HTTLPR and a home-visiting intervention on attachment security. On the basis of previous studies implicating the short 5HTTLPR allele as the “susceptibility allele” [11,17], genetic differential susceptibility to the intervention was predicted to be high for children carrying at least one short allele and low for children carrying two long alleles.
The Health Research Ethics Committee of Stellenbosch University approved this study (Ethics Reference #S12/04/113). Adult caregivers provided written consent for their and their child’s participation, and adolescents signed assent forms prior to participating in the study.
This is a reanalysis of results from the original Thula Sana RCT. In the original trial, mothers were randomized during pregnancy to receive the Thula Sana intervention or usual care during pregnancy and the first 6 mo after birth. The primary outcomes of the original trial were maternal sensitivity and infant attachment security. The aim of this investigation was to test whether 5HTTLPR genotype moderated the intervention effect on infant attachment security measured at 18 mo. This report presents a reanalysis of the original trial’s primary outcome using genetic information gathered at 13 y.
Between April 1999 and February 2003, pregnant mothers from a racially and ethnically homogeneous black, isiXhosa-speaking population inhabiting two geographical areas within Khayelitsha were enrolled in the original Thula Sana study [26]. We made efforts to identify and recruit women who were in their last trimester of their pregnancy (on the basis of the accounts of their gestation that women had received from antenatal clinics). Throughout the recruitment period, over 22 mo, a research assistant regularly visited all the homes door-to-door in both areas to inquire whether anyone had become pregnant or a pregnant woman had moved into the area, and to invite identified women to participate in the study. We identified a consecutive series of 452 women as pregnant within the study area and invited them to take part in the study. Of these, three refused to participate. We then assigned the remaining 449 women to the intervention or control group using minimization, balancing for antenatal depression, whether or not the pregnancy was planned, and which area within Khayelitsha they lived in. In the original trial, they were assessed antenatally and at 2 mo, 6 mo, 12 mo, and 18 mo after birth. The maternal socioeconomic profile for this sample at the time of antenatal interview was as follows: 85% lived in informal housing (shacks), 89% had no formal employment, 44% had no electricity, and 39% had no running water in their home [26]. Later, from December 2012 to June 2014, we enrolled the sample for a long-term 13-y follow-up.
Lay community health workers, themselves all mothers, were selected from the local community, underwent an 8-wk training on delivering the intervention, and were given weekly support and supervision throughout the intervention period. The intervention began in the last trimester of pregnancy, and continued until 6 mo postpartum, during which a total of 16 visits of 1 h each were delivered [26]. The intervention was designed to be suitable for routine delivery within low-resource settings. The content was based closely on The Social Baby [33], but it also incorporated the key principles of the World Health Organization’s report Improving the Psychosocial Development of Children [34] and the use of items from the Neonatal Behavioral Assessment Scale [35], to sensitize the mother to her infant’s individual capacities and needs. Women in the control group received standard services provided by the local infant clinic as well as fortnightly home visits by a community health worker who assessed the physical and medical progress of mothers and infants (women in the intervention group received these services as well).
Full data collection procedures for the early trial were reported in the first outcome paper [26]. In the follow-up study, children and their caregiver were assessed at the Prevention Research for Community, Family and Child Health study center located in Khayelitsha for approximately 4 h. Only limited and out-of-date address information was available from the original study, and many of the names of areas and roads in the informal parts of Khayelitsha had changed in the period between the original study and the reenrollment period. In addition to going door-to-door to find participants at their old addresses, reenrollment strategies also included engaging local community structures. Most participants were still resident in the area, but one-quarter had migrated to other parts of the country since the infant age 18 mo assessment, with participants located in five different provinces of the country. Wherever possible, the team arranged for these child and mother participants to travel to Cape Town. However, there was a small subgroup of participants who were not able to travel to Cape Town. In these cases, a data collection team travelled to their homes for assessment purposes. At the time of assessment, saliva for DNA extraction was collected from children and whenever possible from their biological mothers as well.
The current report tested a single a priori hypothesis that 5HTTLPR genotype, operationalized as the presence versus absence of the short form of 5HTTLPR, would moderate the intervention effect on the primary outcome (secure versus insecure attachment). Security of attachment cannot be measured in children under 11 mo, and therefore it was measured only at the post-intervention follow-up when the infants were 18 mo of age. The analysis was therefore a single-level (i.e., not repeated measure) logistic regression, with the hypothesized moderating effect specified as an intervention group × 5HTTLPR genotype interaction. The primary analysis was conducted without adjustment for covariates, but sensitivity analyses were also conducted, adjusting for covariates; in these analyses we also assessed the impact of missing data (for individuals without both attachment and genetic data, including all individuals lost to follow-up) using multiple imputation, as recommended by a reviewer. We used the fully conditional specification approach to multiple imputation, which is a highly flexible approach capable of accounting for nonlinearity in the relationship between covariates and outcome and which fits an imputation model that is consistent with the substantive model (i.e., explicitly includes the gene and intervention main effects and interaction). Multiple imputations included all model variables, maternal 5HTTLPR genotype, and the only baseline measures that were associated with missingness (time to entry into the trial from start of recruitment and whether the house had electricity and water). Imputation was conducted using the package smcfcs [40] and Stata’s MI procedure based on 100 imputed samples. Details of the imputation are provided in S1 Text and S2 Data.
From December 2012 to June 2014, we reenrolled 334 (74.1%) of the children (162 intervention, 172 control; 166 males, 168 females) from the original sample of 449 mother–child pairs. At 13 y of age, 115 of the original 449 children were lost to follow-up. Of these children, 24 had died since the original randomization process. The remaining 91 could not be contacted. Derivation of the sample used in this study is depicted in Fig 1.
Of the 334 adolescent participants at 13-y follow-up, 279 (131 males, 148 females; 134 intervention, 145 control) provided DNA samples, all of which yielded 5HTTLPR genotype results. There were 220 (104 males, 116 females; 110 intervention, 110 control) adolescents for whom there were both 5HTTLPR genotype and attachment security data. 5HTTLPR genotype results for the 220 adolescents with attachment security data are shown in Table 1. Of these 220 adolescents, 185 (89 males, 96 females; 97 intervention, 88 control) had mothers for whom 5HTTLPR genotype data were available. All analyses were performed on these 220 adolescents and 185 adolescent–mother pairs. No individuals changed from the intervention to the control arm or vice versa at any point in the trial, and no individuals with both genotype and attachment security data were excluded from analysis. The 220 participants with both genotype and attachment data were compared to the rest of the original sample of 449 on a range of demographic and socioeconomic variables. As shown in Table 2, with two exceptions—water and electricity in the home—there were no significant differences between the two groups (See S3 Table for actual values). There were no significant differences on any of the variables between the 110 adolescent participants in the intervention and control groups (S4 Table).
Because the presence of at least one short 5HTTLPR allele frequently confers susceptibility to environmental influence [11,17], individuals with short/long and short/short genotypes were treated as one genotype category. Individuals carrying at least one short allele comprised 40% of the 220 participants (Table 1). Logistic regression revealed a significant G×I interaction: for infant security of attachment, the efficacy of the intervention varied as a function of serotonin transporter genotype (OR = 4.07, p = 0.028, 95% CI [1.16, 14.20]). As shown in Table 3 and in Fig 2, for those with the susceptible genotype (short/long and short/short), the intervention increased the odds of secure infant attachment nearly 4-fold relative to controls (OR = 3.86, p = 0.008, 95% CI [1.42, 10.51], d = 0.75). By contrast, for those with the nonsusceptible genotype (long/long), the intervention had no impact on the odds of secure attachment relative to controls (OR = 0.95, p = 0.89, 95% CI [0.45, 2.01], d = 0.03). Expressed in terms of absolute risk, for those with the short allele, the probability of secure attachment being observed in the intervention group was 84% (95% CI [73%, 95%]), compared to 58% (95% CI [43%, 72%]) in the control group. For those with two copies of the long allele, the probability of being secure was 70% (95% CI [59%, 81%]) in the intervention group, compared to 71% (95% CI [60%, 82%]) in the control group (Table 4; Fig 3). The results show that, on average, individuals carrying at least one short allele were susceptible to the intervention and those carrying two long alleles were nonsusceptible.
The efficacy of the home-visiting intervention on the attachment outcome in terms of percentage secure and insecure individuals according to group and genotype is shown in Fig 3. From left to right in Fig 3, for individuals carrying at least one short allele, the percentage showing secure attachment was 84% and 58% in the intervention and control groups, respectively. For individuals carrying two long alleles, the percentage showing secure attachment was 71% and 70% in the intervention and control groups, respectively. In the absence of genetic information, when results are averaged over all individuals and all genotypes (“unknown”), the apparent percentage of individuals showing secure attachment was 75% and 65% in the intervention and control groups, respectively. The numbers of individuals in each group and in each genotype category are given in Table 1.
The above logistic regression analysis was rerun controlling for covariates observed to be different between the two groups. The result was not affected by sex or any other covariates (Table 2).
Further, to address the possibility that the observed interaction effect was attributable to maternal rather than child genotype, we reran the logistic regression including terms for maternal 5HTTLPR genotype and the interaction between maternal 5HTTLPR and group (intervention versus control) in the model. The child 5HTTLPR × group interaction remained significant (OR = 4.8, p = 0.041), while neither the main effect of maternal 5HTTLPR (OR = 0.96, p = 0.93) nor its interaction with group (OR = 1.97, p = 0.37) was significant.
Finally, multiple imputation analyses based on 100 imputed samples of n = 499 (the total number of mother–infant dyads originally randomized) were also run to check the robustness of the result. These analyses confirmed the 5HTTLPR × group interaction in the logistic regression analysis, with the interaction OR equal to −1.41 (standard error = 0.64, p = 0.029, 95% CI [−2.68, −0.015]).
The current study aimed to test the hypothesis that 5HTTLPR genotype would moderate the impact of an early child development intervention aimed at promoting the security of mother–infant attachment in a middle-income country. Our reanalysis of the original trial in light of recently acquired genetic information provided support for this hypothesis. Specifically, for children with one or two copies of the short allele of 5HTTLPR, the intervention appeared to be highly effective in improving the rate of attachment security (from 58% in the control group to 84% in the intervention group), but for those with only the long allele, the intervention led to no measurable benefits (secure attachment rate 70% in the control group versus 71% in the intervention group).
There are few studies, and none outside the US and Europe, that have used the framework of experimental trials to test for G×E interaction in early child development [9]. In the specific area of attachment, which is a key domain of psychosocial functioning among young children, we are aware of only two other studies of G×I interaction. In a study of maltreated children who were randomized either to a parenting intervention or to a control condition, Cicchetti and colleagues [13] found no G×I interaction for 5HTTLPR in relation to attachment. There are, however, important differences between this study and our own that may account for the difference in findings. First, the Cicchetti et al. study involved a smaller sample size (in total, ignoring genotype: 49 intervention, 47 control) than our study (110 intervention, 110 control). Second, their sample was racially and ethnically diverse, with the frequency of short allele carriers differing markedly between the black (45.9%), white (78.6%), and other/multiracial (67.5%) categories. By contrast, our study sample was drawn from an ethnically homogeneous population. Third, maltreated children represent a special population that is not comparable to the community sample included in the Thula Sana study. In that regard, it is notable that the disorganized class of insecure attachments, which carries the highest clinical risk, was present in 88% of the maltreated sample and in only 8% of the Thula Sana sample, a figure typical of community samples. Given these important sample differences, little can be concluded from the difference in G×I findings between the two studies. The only other study in this area also relied on a special population sample, in this case, children raised in Romanian institutions. The Bucharest Early Intervention Project randomly allocated institutionalized children to either high-quality foster care or continuing institutional care before 30 mo of age. At 54 mo of age, among children carrying the short/short 5HTTLPR genotype, relative to the outcome of those in the continuing institutional care group, those provided with high-quality foster care had lower symptom levels of attachment disorder (specifically “indiscriminate social behavior”). For the children with at least one long allele (short/long and long/long), there was no difference in terms of attachment disorder symptoms between foster and institutional care conditions [14].
As Belsky recently observed, attachment research, like much research on early child development, has proceeded for the most part with the assumption that all children are equally susceptible to the effects of sensitive and insensitive care [41]. The current findings suggest otherwise and highlight the significance of genetic differential susceptibility in shaping developmental trajectories during early infancy.
An important limitation of this study is that we were not able to follow up all of the individuals from the original trial, and there were missing data for attachment and genotype. In total, our primary analysis included 49% (220/449) of the original sample of children whose mothers were randomized to intervention and control conditions. Although the intervention and control groups were highly similar in our follow-up sample, and the follow-up sample was generally very similar to the original sample, there was some evidence of selective loss to follow-up on two variables (Table 2). This means that randomization within our follow-up sample may have been imperfect. Attribution of the primary outcome to causal effects of the intervention in the present sample should therefore be treated with caution.
Another limitation of this study is its focus on only one gene. Despite extensive evidence from research using both observational and experimental approaches showing that the 5HTT promoter polymorphism influences organisms’ sensitivity to environmental influences [42], and despite evidence that 5HTTLPR influences the development of functional and structural brain networks involved in emotion regulation, stress processing, and threat sensitivity [43], it is unlikely that one single gene will explain all individual differences in intervention efficacy. Rather, it can be assumed that differential susceptibility to environmental influences is a complex, polygenic trait influenced by the combination of hundreds of common genetic variants of small effect. The first studies in the field of therapygenetics have started using genome-wide approaches [44] and polygenic scoring [45], which allow researchers to aggregate the effects of multiple variants. The use of these approaches in sufficiently large samples, much larger than the present study, could open new avenues in G×I interaction research.
A final limitation is that our attachment finding could be culturally-specific and therefore not generalizable, and certainly the study needs replicating in other cultural contexts. For example, in contexts such as the one described here, where there are multiple caregivers, infants and children are able to develop attachments to more than one person, and when the attachment status is discordant between different caregivers, it remains unclear what the longer-term outcomes are [46]. Nevertheless, we believe it is unlikely that our results would be confined to this cultural context. First, both this sample and one we previously studied in Khayelitsha [24] showed a distribution of attachment categories similar to that in high-income countries. Second, our previous research showed the same association between attachment security and the main parenting antecedents that we would expect (i.e., sensitivity and lack of intrusiveness) from a substantive body of research globally [24].
Beyond illuminating the role of genetic differential susceptibility in early childhood development, the current finding also speaks to a fundamental issue in the quest to understand and mitigate the developmental effects of poverty through psychosocial intervention. The near-large effect size reported here for the intervention in children with susceptible genotypes (d = 0.75) is at variance with the general conclusion that psychosocial interventions in the context of poverty produce only small to medium effect sizes [4]. Without taking account of genetic susceptibility, it is possible that other intervention studies have, at least in some subpopulations, underestimated the impact of their interventions, as we originally did. By the same token, as was originally reported for Thula Sana [26], other studies might also have underestimated the negative impact on susceptible subpopulations of not receiving an intervention (Figs 2 and 3). In short, averaging outcomes across all participants may well lead to an invalid conclusion about the efficacy of an intervention [47].
The launch of the SDGs and the Global Strategy for Women’s, Children’s and Adolescents’ Health in late 2015 has focused attention on a life-course perspective towards the understanding of child and adolescent development—the “thrive agenda” [48,49]. To stand a chance of meeting the ambitious SDGs and Global Strategy targets by 2030, an enhanced understanding will be required of the biological and psychological mechanisms underlying interventions aimed at improving the lives of young children. In the context of the resource constraints that characterize LMICs, ensuring that psychosocial interventions are implemented in the most efficacious manner will take on an added urgency. In this regard, it is instructive to note a parallel between genetic differential susceptibility to psychosocial interventions and genetic differential susceptibility in the emergence of personalized medicine, specifically pharmacogenomics. Just as genetic information is being used to guide the choice of medication for different individuals diagnosed with the same condition (e.g., [50]), it has been suggested that in a world of limited resources, psychosocial interventions could, once more is known, also be selectively targeted at genetically susceptible individuals [41,51]. This possibility would precipitate the daunting moral challenge of balancing equity (equal treatment for all) and efficacy (treating only those likely to benefit) [41,47].
However, while such targeting in LMICs is technically feasible, provision of intervention services on the basis of genotyping is not currently a realistic prospect. First, as already noted above, genetic prediction of intervention efficacy based on variation at one gene locus is far from sufficiently sensitive or specific to provide a reliable basis for intervention recommendations. Second, quite apart from the science, the prospect of discriminating individuals on the basis of their genetic makeup is controversial and likely to encounter strenuous social resistance.
Nevertheless, other avenues of investigation do suggest themselves. A promising approach might be to incorporate intermediate phenotypes, such as easily accessible physiological or temperamental characteristics, with genetic and epigenetic markers [52] to improve prediction by use of multiple data types. Physiological measures might include hormonal and/or sympathetic and parasympathetic nervous system stress markers [53]; relevant temperamental characteristics could include emotion regulation abilities [54] or approach avoidance tendencies [55]. Indeed, the short/long 5HTTLPR polymorphism has been associated with individual differences in epigenetic methylation [56], stress physiology [57–59], and temperament [60,61]. A combination of biological and behavioral markers could be used to identify meaningful subgroups and thus target interventions to those likely to respond. Moreover, such measures could be used not only to better target interventions to those likely to respond, but also to clarify where new or additional interventions are required.
In summary, despite a considerable body of evidence on how cumulative risk is implicated in poor child development, our understanding of pathways and mechanisms, and how dose, timing, and adversity impact on outcome, is to date quite limited [62]. Measuring genetic susceptibility together with epigenetic, physiological, temperamental, and behavioral markers in RCTs will allow better examination and greater insight into these mechanisms and pathways in LMICs. This could enhance our understanding of why certain individuals do not respond to a particular treatment and facilitate the development of new interventions for them.
|
10.1371/journal.pbio.1001728 | A Versatile Class of Cell Surface Directional Motors Gives Rise to Gliding Motility and Sporulation in Myxococcus xanthus | Eukaryotic cells utilize an arsenal of processive transport systems to deliver macromolecules to specific subcellular sites. In prokaryotes, such transport mechanisms have only been shown to mediate gliding motility, a form of microbial surface translocation. Here, we show that the motility function of the Myxococcus xanthus Agl-Glt machinery results from the recent specialization of a versatile class of bacterial transporters. Specifically, we demonstrate that the Agl motility motor is modular and dissociates from the rest of the gliding machinery (the Glt complex) to bind the newly expressed Nfs complex, a close Glt paralogue, during sporulation. Following this association, the Agl system transports Nfs proteins directionally around the spore surface. Since the main spore coat polymer is secreted at discrete sites around the spore surface, its transport by Agl-Nfs ensures its distribution around the spore. Thus, the Agl-Glt/Nfs machineries may constitute a novel class of directional bacterial surface transporters that can be diversified to specific tasks depending on the cognate cargo and machinery-specific accessories.
| Many living cells use processive cytoskeletal motors to transport proteins and subcellular organelles to specific subcellular sites. In bacteria, this type of transport has yet to be identified and it is generally thought that random protein collisions underlie most biochemical processes. In recent years, our view of the bacterial cell was changed by the discovery of subcellular compartmentalization and a cytoskeleton, suggesting that processive motors might also operate in prokaryotes. We previously characterized a mechanism of intracellular transport that drives cell motility across solid surfaces in the gram-negative bacterium Myxococcus xanthus. Since the transport apparatus was also found in bacterial species that do not move on surfaces, we postulated that intracellular transport underlies other cellular processes in bacteria. Indeed, we show here that the Myxococcus motility motor can be adapted to transport sporulation-specific proteins around the nascent spore surface. Because the transported proteins are linked to the main spore coat, this motion assists the assembly of a protective spore coat. In conclusion, the Myxococcus motility/sporulation transport machinery defines an emerging class of versatile transport systems, suggesting that processive transport has been overlooked and may well orchestrate many processes in bacteria.
| In eukaryotic cells, motor-assisted intracellular transport regulates fundamental cellular processes including cell division, macromolecule secretion, and cell migration. Because of its small size, the bacterial cell has long been considered to be a disordered compartment where biochemical reactions and cellular processes are governed by diffusion-driven random collisions. However, in recent years, it has become clear that bacteria are highly organized and contain a complex cytoskeleton [1],[2]. Despite the identification of bacterial counterparts of actin, tubulin, and intermediate filaments, processive cytoskeletal motors akin to myosin, kinesin, or dynein have yet to be found in bacteria [3]. Previously, while studying a mechanism of surface motility in Myxococcus xanthus, we have shown that the motility machinery consists of a new type of processive transport system (Agl-Glt) [4],[5]. Phylogenomic studies suggested that the motility function of the Agl-Glt machinery emerged from the recent specialization of an older system, predicting that bacterial Agl-Glt–like transporters may be adapted to other functions [5]. Here, we show that in the same bacterium, Agl, the motor component of Agl-Glt machinery, forms a second transport system that propels spore coat assembly proteins during sporulation. This finding suggests that a previously overseen type of bacterial surface transport can be adapted to mediate very different cellular tasks in prokaryotes.
M. xanthus cells move across solid surfaces by a process termed gliding (A)-motility where surface translocation occurs in the absence of extracellular organelles [6]. In recent years, remarkable progress has been made to elucidate the motility mechanism with the first identification of the motility machinery and the tracking of its localization in live gliding cells [7]. The motility machinery consists of a molecular motor, Agl (AglR, Q, and S), a three subunit flagellar-type proton channel that assembles in the bacterial inner membrane [4]. During motility, the Agl motor harvests the proton motive force (pmf) to move directionally along a looped continuous path spanning the entire cell length [8]. Helical trafficking of the motor may occur through a connection with the actin-like MreB cytoskeleton on the cytosolic side [8],[9]. In the cell envelope, mechanical work from the motor is transduced to the cell surface by the Glt (Gliding transducer, GltA–K) complex through a direct interaction involving AglR and GltG [5],[10]. Transported Glt proteins produce thrust when the machinery comes in contact with the underlying substrate at areas termed focal adhesions (FAs) [4],[11]. How exactly the Glt proteins contact the substrate is unknown [7],[12], but adhesion is facilitated by slime, a yet undefined sugar polysaccharide specifically bound by the outermost components of the motility machinery [13]. Thus, the Agl-Glt machinery may be viewed as a modular transport system where an active transport unit (Agl) combines with a specific cargo (Glt) to propel the cells on surfaces.
Remarkably, the glt genes are paralogous to the nfs (necessary for sporulation, nfsA–H, Figure 1) genes involved in assembly of the Myxococcus spore coat [5],[14]. Bacterial sporulation is a stress-induced differentiation leading to the formation of highly resistant quiescent cell types. In the model bacteria Bacillus and Streptomyces sp., a spore is formed after elaboration of a complex proteinaceous outer shell, called the spore coat. However, the coat proteins and assembly mechanisms are unrelated (for reviews see [15],[16]). Myxococcus spores are formed following starvation in multicellular fruiting bodies. During multicellular development, the initiating signals have not been identified, but sporulation can be conveniently induced in vitro by the addition of glycerol [14]. Glycerol-induced spores display certain morphological differences compared to fruiting body spores (e.g., the absence of the outermost protein cuticula) [17], but they also share many similarities: resistance to heat and sonic disruption as well as germination. Contrary to Bacillus and Streptomyces spores, the Myxococcus spore coat is mostly composed of two carbohydrates, N-acetyl-galactosamine and glucose, in a molar ratio of 3∶1 [18],[19]. The exact structure of the spore coat polymer(s) is unknown, and glucose molecules may form an alpha 1,3-glycan chain independently from the N-acetyl-galactosamine [18],[19]. Recently, a locus of nine genes, named exoA–I, has been shown to be essential for spore coat synthesis [20]. Annotation of exoA–I indicates that the main spore coat polymer is likely a capsular-type polysaccharide, exported by an outer membrane Wza-like translocon (the ExoA protein) [20]. For simplicity, the exo-dependent spore coat polymer will be named Exo throughout the rest of this article. Myxococcus sporulation is a stepwise process. During the first step of sporulation, the peptidoglycan (PG) layer is seemingly degraded, an MreB-dependent process, which leads to cell rounding [20]. Subsequently, the Exo polymer is exported by ExoA and deposited around the collapsed inner and outer membranes [20]. Tight wrapping of Exo around the spore membranes requires the Nfs complex [20]. How exactly the Nfs system promotes spore coat assembly is unknown and is addressed in this article.
The Glt and Nfs proteins are highly similar and seem to associate with extracellular sugar polymers (slime and Exo, respectively). Thus, starting from the premise that both systems share similar operating principles, we investigated the function of Nfs in spore coat assembly. Doing so, we discovered that following the onset of sporulation, the Agl motility motor dissociates from the Glt complex and becomes recruited to the Nfs complex to transport it around the spore membranes. We further show that following its secretion at discrete sites around the spore surface, the Exo polymer is recruited by mobile Nfs units, suggesting that the Agl-Nfs machinery constructs the coat by wrapping Exo strands around the cell surface. We conclude that the Agl-Glt/Nfs machineries constitute a versatile class of active surface transport machineries that may carry out multiple functions in bacterial cell surface organization.
A recent phylogenetic study showed that the glt and nfs genes likely arose from the recent duplication of a single gene system in one of the terminal branches of the deltaproteobacteria (Figure 1A) [5]. Thus, the Nfs and Glt proteins are paralogues, and when predictable, protein domains are systematically conserved between Nfs and Glt proteins (Table S1). The nfs complex consists of eight genes, nfsA–H (Figure 1B), respectively, homologous to gltA–H (Table S1). Nfs is therefore predicted to be a somewhat simpler assemblage, missing paralogues to GltI, J, and K. Gene synteny is maintained between the nfs and glt clusters (Figure 1B). Of note, however, gltA–C and gltD–H form two distinct genomic clusters, but the nfsA–H genes are clustered in a single genomic region, possibly forming an operon (Figure 1B) [20]. Consistent with the notion that the nfs genes form a complex in the spore membrane, all nfs genes are essential for sporulation and their products localize to the cell envelope (tested for NfsA, B, C, D, E, and G and predicted for NfsF and H) [20], like their glt counterpart (tested for GltD, E, F, G, and H and predicted for GltA, B, C, J, and K) [5],[10]. Therefore, we predict that the Nfs proteins assemble a Glt-like membrane sporulation complex (Figure 1B).
Previous works showed that Glt promotes motility in association with the Agl motor [4],[5],[8]. Therefore, although Nfs might be functional on its own, its function may also require an Agl-like motor. Reasoning that each complex may have its own dedicated motor, we tested whether the MXAN_3003–5 genes, which encode the only additional complete set of Agl homologues in M. xanthus and are dispensable for motility, are required for sporulation like the nfs genes [5]. Because nfs mutants are defective for sporulation whether they are extracted from fruiting bodies or after glycerol-induction [14], we measured sporulation after glycerol induction throughout this study. Under these conditions, a mutant carrying an in-frame deletion in the MXAN_3004 gene (the aglQ homologue) formed perfectly viable spores, indicating that the putative MXAN_3003–5 motor does not play a significant role in sporulation (Figure 2A).
We next tested whether AglRQS itself may be required for sporulation. Strikingly, the aglR, aglQ, and aglS mutants all showed a severe sporulation defect, comparable to the sporulation defects of the nfsD mutant and the exoA mutant lacking the Wza homologue (Figure 2A). An aglQD28N point mutant, carrying a substitution in the AglRQS channel previously shown to abolish proton conductance and to paralyze the Agl-Glt motility machinery [4], also failed to sporulate (Figure 2A). Therefore, the AglRQS complex and, importantly, its proton-conducting activity are required for both motility and sporulation.
In the nfs mutants, the sporulation program is correctly initiated following glycerol induction: the cells round up and the Exo polymer is produced and exported to the cell surface [20]. However, in absence of Nfs, Exo is loosely attached to the spore surface, which results in loss of cell integrity and abortive sporulation [20]. By Transmission Electron Microscopy (TEM), we also observed a thin and complex electron dense layer around the membranes of WT spores [20], suggesting the presence of a spore coat (Figure 2B). In both the nfsD and aglQ mutants, this layer was absent and replaced by hair-like filaments that seemed loosely attached to the spore surface at one end (Figure 2B). Since the filaments were completely absent in an exoA mutant, they may constitute Exo polymer filaments (Figure 2B) [20]. To confirm that the filaments are not artifacts of TEM sections, we tested a fluorescein-coupled lectin, Griffonia (Bandeiraea) Simplicifolia lectin I (GSL-I), with a selectivity for N-acetylgalactosamine, the major sugar component of the spore coat [18],[21]. As expected, GSL-I stained the surface of WT spores but not that of spore-coat deficient exoA mutants (Figure 2C). On WT spores, GSL-I staining decorated the entire surface with occasional brighter dots (Figure 2C). Moreover, as expected if the spore coat was loosely attached, GSL-I staining of aglQ and nfsD mutants was not compact around the cell surface (Figures 2C and S1). Taken together, the TEM and lectin-staining experiments suggest that a function of the Agl-Nfs machinery is to promote the formation of a compact spore coat layer around the spore membrane. Additionally, we identified GSL-I staining as a useful tool to detect the spore coat by live fluorescence microscopy (see below).
In the gliding machinery, AglRQS contacts the Glt complex through the specific association of AglR and GltG (Figure 2D) [5]. We used a bacterial two-hybrid assay to test whether AglR also contacts the Nfs system through an interaction with NfsG, the GltG paralogue. Consistent with interaction, a significant β-galactosidase activity (>1,000 Miller units) was obtained when AglR and NfsG were expressed together (Figure 2D). No significant β-galactosidase activity was measured when each protein was expressed alone or when MXAN_3003 (the AglR homologue) was co-expressed with NfsG or GltG (Figure 2D). Thus, the AglRQS motor can associate with both the Glt and Nfs complexes.
Myxococcus cells do not complete the sporulation process when spotted directly on an agar pad atop a microscope slide. Thus, to monitor sporulation by live microscopy and elucidate how the Agl-Nfs machinery drives spore coat assembly, we developed a microfluidic chamber assay where cells are immobilized and sporulate in liquid directly on the microscope stage (Figure S2A,B and Movie S1). Under these conditions, viable spores were obtained approximately 250–300 min after glycerol addition (Figure S2A,B). To test the dynamics of the Agl-Nfs machinery during sporulation, we constructed functional NfsD-mCherry and AglQ-sfGFP (super-folder GFP) [22] fusions to use as proxies to monitor both Nfs and Agl dynamics (Figure S3). Time-course sporulation experiments showed that NfsD-mCherry is expressed in lieu of GltD-mCherry (the motility NfsD paralogue) during sporulation, suggesting that sporulation depends on a genetic switch that results in the substitution of the Glt complex by the Nfs complex (Figures 3A–B and S4). Consistent with its role in spore coat assembly, expression of AglQ was even increased up to two-fold at the onset of sporulation and maintained at high level throughout the sporulation process (Figure 3A–B).
In sharp contrast to the focal localization of AglQ during motility [4], AglQ-sfGFP accumulated circumferentially all around the spore membrane throughout the sporulation process (Figure 3B). However, when we analyzed the localization of NfsD-mCherry, we found that this protein covered the entire cell surface during the initial stage of sporulation but later formed fluorescent-bright clusters at the end of the cell-rounding phase, a sporulation stage where the Nfs complex would be expected to become active (Figure 3B). In time-lapse experiments, the NfsD-mCherry foci were observed to move in orbital trajectories around the spore surface (Figure 4A and Movie S2). Orbital trajectories could be captured both when the microscope focal plane was set to the middle or to the top of a sporulating cell, showing that NfsD rotates all around the spore circumference (Figure 4Ai and 4Aii). Rotating NfsD-mCherry clusters could be tracked for distances between 3/4 and up to a full spore circumference, suggesting a high level of directionality (Figures 4Ai and S5). To measure the movement parameters of NfsD-mCherry precisely, we designed a computational method to measure the distance traveled by NfsD-mCherry foci at subpixel resolution. Moreover, since sporulating cells are spherical, the distance that separates two clusters at distinct times is not the euclidian distance but the orthodromic distance (Figure S5). Orthodromic distances traveled by NfsD-mCherry clusters were thus calculated for each time point by projection (see Methods and Figure S5). This analysis revealed that NfsD-mCherry foci moved at instantaneous speeds ranging from 0.1–0.3 µm.min−1 (Figure 4B). Movement was clearly directional because (i) the distance to the origin increased over the time (Figure 4C) and (ii) NfsD-mCherry movements could be described by a Mean Square Displacement (MSD) over time that increased by 4Dt+v2t2 relation (with v the mean velocity, v = 0.12±0.05 µm.min−1, and D the apparent diffusion coefficient, D = 0.012±0.007 µm2.min−1, Figure 4D). This directional movement must be linked to the activity of the Agl motor because movement was immediately stopped by addition of Carbonyl Cyanide-m-ChloroPhenylhydrazone (CCCP), a proton motive force uncoupler (Figures 4E and S7C and Movie S3) and paralyzed NfsD-mCherry foci formed both in the aglQ mutant and in the aglQD28N mutant (Figure 4B,C,F and unpublished data). MSD analysis of NfsD-mCherry clusters in the aglQ mutant showed a typical subdiffusive behavior, consistent with the absence of significant active movements (Figure 4D).
During motility, Agl functions to transport Glt proteins and slime along the cell surface [4],[11]. Thus, the Agl motor may transport the Nfs complex and associated Exo strands to construct a densely packed spore coat around the spore membrane. We used a micron-sized polystyrene beads assay [4] to test directly whether Agl has a surface transport activity and found that beads were transported along the spore surface with instantaneous speeds matching the dynamics of NfsD-mCherry (0.1–0.3 µm.min−1, Figure 5A–B). Consistent with bead transport by the Agl motor, this transport was undetectable in the aglQ mutant and abolished by the addition of CCCP (unpublished data and Figure 5B).
To test whether Exo is the terminal cargo of the Agl-Nfs machinery, we stained NsfD-mCherry–expressing spores with GSL-I and imaged each fluorophore simultaneously 4 h after the induction of sporulation. As already mentioned, GSL-I staining covers the entire spore surface but also forms prominent fluorescent clusters (Figure 2C). When these clusters were imaged by time-lapse, they rotated together with NfsD-mCherry clusters (Figure 5C and Movie S4). Co-tracking of NfsD-mCherry and GSL-I foci were occasionally observed to dissociate, which was followed by rapid dispersal of the GSL-I cluster (Figure 5C). Thus, Exo-linked rotating Agl-Nfs complexes are likely involved in the construction of a densely packed protective mesh at the surface of developing spores.
Mature spores apparently lack PG and the actin-like MreB cytoskeleton [20],[23]. Consistent with published data [20], we found that MreB is only required for the initial cell rounding phase of sporulation using A22, a specific MreB-inhibitor (Figure S6A–B) [9]. Since the activity of the Agl-Glt motility machinery requires MreB [8],[9],[11], we tested the effect of A22 on NfsD-mCherry dynamics and found that MreB is dispensable for NfsD-mCherry rotation (Figure S6C). Contrarily to Bacillus endospores, myxospores do not contain PG or only trace amounts [23]. The Exo spore coat may provide spore integrity in absence of other rigid cellular scaffolds because exo, nfs, and agl mutants all show aberrant cell morphologies after 24 h (Figure S2A) [20]. Thus, during sporulation, both MreB and the PG seem dispensable for the activity of Agl-Nfs.
What is the mechanism of Nfs transport? In absence of a rigid scaffold (i.e., MreB filaments and the PG), Agl motor units may distribute circumferentially (Figure 3B) to transport Nfs proteins from one motor unit to the next, similar to actin filaments being moved by immobilized Myosin motors [24]. Alternatively and because the Nfs proteins are terminally associated with the Exo polymer, Exo secretion itself could push Nfs proteins around the spore surface in a mechanism reminiscent of PG glycan strand insertion rotating PG synthetic complexes [25]–[27]. To discriminate between these two possibilities, we tested NfsD-mCherry dynamics in the exoA mutant. In the absence of ExoA, several critical features of Nfs transport emerged: (i) NfsD-mCherry clusters formed foci that appeared smaller in exoA mutant than in WT cells, suggesting that Exo polymers organize NfsD-mCherry clustering at the spore surface (compare Figure 4G and A); and (ii) NfsD-mCherry movement was erratic and characterized by frequent reversals and saltatory motions, suggesting that directionality is lost in the mutant (Figure 4G). To test this possibility, we computed the MSD of NfsD-mCherry clusters as a function of time in exoA mutant cells and found it to be mostly linear, a characteristic of undirected random motion (Figure 4D). In contrast, the MSD of WT cells is characteristic of directed motion (Figure 4D).
In the exoA mutant, although NfsD-mCherry clusters appear to move randomly, this movement is unlikely to be driven by diffusion alone because many clusters showed short and fast movement phases with burst speeds similar to NfsD-mCherry clusters in WT cells (Figure 4B,C,G). These fast movements depend on the activity of the Agl-motor because (i) they were completely abolished by CCCP (Figure S7A) or in an exoA aglQ double mutant (Figures 4B,C,D,G and S7B) and (ii) the MSD of NfsD-mCherry was constant over time in the exoA aglQ double mutant (Figure 4D). Finally to prove that AglQ-dependent transport can still occur in absence of the Exo polymer, we further tested bead transport in absence in the exoA mutant. Similar to the movements of NfsD-mCherry clusters in the exoA mutant, bursts of fast bead movements were observed in this strain (Figure 5B). Since movements were completely abolished by the addition of CCCP or in an aglQ mutant background and thus depend on the activity of the Agl motor (Figure 5B), all together, these results show that surface movements at the cell surface (Nfs and beads) result from active Agl-dependent transport and not spore coat polymer secretion, which may instead serve to guide motion (see Discussion).
Gene homologies suggest that the exoA–I genes responsible for spore coat synthesis encode a capsular polysaccharide synthesis and Wza-type export apparatus (Figure 6A) [20]. To further understand the link between Agl-Nfs activity and the export of Exo to the spore surface, we localized structural components of the export apparatus. Fusions to ExoA and ExoC, respectively, encoding Wza and transmembrane Wzc-domain homologues were nonfunctional (unpublished data). Although in general proteobacterial Wzc proteins carry both a transmembrane and a cytosolic BY-kinase domains, in Myxococcus the Wzc transmembrane domain and the kinase Wzc domain are carried by two distinct polypeptides (respectively named ExoC and ExoD/BtkA) [28], a conformation often found in firmicutes [29]. Nevertheless, BtkA-dependent tyrosine phosphorylation of the Wzc-like transmembrane polypeptide is essential for Myxococcus sporulation [28], suggesting that BtkA can be used to localize Exo export sites. We successfully obtained a functional BtkA-sfGFP fusion (Figure S3). Consistent with previous expression studies [28], BtkA-sfGFP was only expressed following sporulation induction. At 4 h, BtkA-sfGFP formed prominent foci in ≈50% of the cells (Figure 6B–C). Each cell contained on average 1.7±1 clusters (counted for 305 cells, with a maximum number of clusters of six per cell), all located near the spore membrane suggesting that BtkA-sfGFP is indeed recruited to the export apparatus. Z-sections of 4-h-old spores and 3D reconstructions further revealed that BtkA-sfGFP foci form at discrete sites around the spore periphery (Figure 6D and Movie S5). Importantly, the BtkA-sfGFP foci did not colocalize with NfsD-mCherry and often formed in distinct z-planes around the spore (Figures 6E and S8). In total the BtkA localization data suggest that sporulating cells only assemble a few discrete fixed export sites in the cell envelope, suggesting that rotating Agl-Nfs machineries act downstream from Exo secretion to construct the spore coat (see Discussion).
In this report, we show that the Agl motility motor is modular and interacts with Glt or Nfs proteins, depending on the growth phase. The output of these interactions is remarkably different because Agl-Glt drives gliding motility while Agl-Nfs drives spore coat assembly (Figure 7). Both Agl-Glt and Agl-Nfs interact with extracellular polysaccharides (respectively, slime and the Exo polymer) and the specific function of each system may be linked, at least partially, to the chemical properties of the cognate polymer. For example, motility could be facilitated by the chemical adhesiveness of slime, while the structure of Exo might make it particularly suited to form a coat around the spore membrane (Figure 7).
Since both slime and the Exo polymer are still detected at the cell surface in agl mutants, Agl-Glt/Nfs systems are not involved in the synthesis/export of associated sugars. This is particularly clear during sporulation where the synthesis and transport of the Exo polymer is controlled by the exo locus. Rather, several lines of evidence suggest that Agl-Glt/Nfs system transport their cognate polymers after export along the cell surface: (i) AglRQS form a flagellar motor-like complex, a predicted pmf-driven motor that interacts both with Glt and Nfs proteins (through a specific interaction between AglR and GltG/NfsG). Accordingly, both the Agl system and the pmf are required for Glt and Nfs movements (this work and [8],[11],[30]). (ii) Agl exerts transport activity at the surface of motile cells and spores, monitored by addition of polystyrene beads. Transport of slime and Exo was observed directly with fluorescent lectins, and in both cases mobile lectin patches were translocated together with mobile AglQ (slime) and NfsD (Exo) complexes (this work and [13]). (iii) During sporulation, secretion of the Exo polymer is not a significant driving force for Nfs movement. Thus, rotation does not result from a pushing action of the Exo polymer, for example, like when MreB-associated PG synthetic complexes are moved circumferentially by the incorporation of new glycan strands in the PG meshwork [25]–[27]. Importantly, however, transport directionality was lost in absence of Exo polymer secretion. When Agl-Nfs–linked Exo strands become deposited at the spore surface, they could act like a molecular ratchet, preventing motor back steps and restricting Agl-Nfs movements in one dimension. The rigidity of the growing Exo meshwork may also support Nfs movements, especially since MreB and the PG may not be involved or present. In the motility system, slime could perform analogous functions and explain the mysterious directionality of Agl-Glt complexes. Unfortunately this could not be tested because the main slime polymer has not been characterized.
Although it is still unclear how the Agl-Glt machinery accommodates the rigid PG layer during gliding, the situation may be simpler in spores: circumferential AglQ motors may transport Nfs subunits from one motor to the next, fueled by the pmf and guided by linked Exo strands. Testing this mechanism further will require defining the structure of the Exo polymer and its connection with the Nfs proteins.
The connection between Agl-Nfs and Exo and the severe spore coat assembly defect observed in nfs and agl mutants suggests that Agl-Nfs mediates spore coat assembly directly. As discussed above Nfs function can be separated from secretion. Using a BtkA-sfGFP fusion, we localized the Exo secretion apparatus and found it to form a limited number of sites around the spore periphery, suggesting that secretion is spatially constricted. This localization is not surprising because immunolocalization of Wza only suggested that the export apparatus forms discrete sites in Escherichia coli [31]. Thus, a rotary transport complex may be needed to build a homogenous glycan layer all around the spore. Our experiments and others [20] suggest that the activity of Agl-Nfs is necessary to anchor the spore coat polymer at the surface. In this process, scanning Agl-Nfs complexes may capture newly secreted glycan Exo strands and incorporate/deposit them where necessary in the growing spore coat meshwork. This could occur, for example, if a glycosyl transferase activity is linked to the Nfs complex. However, other mechanisms are also possible and more work is needed to understand how Agl-Nfs contributes to spore coat assembly at the molecular level.
The Agl-Glt/Nfs machineries likely evolved by modular expansion of a conserved system of seven proteins (Figure S9) [5]. This “core” system is found as a standalone machinery in many gammaproteobacteria and some deltaproteobacteria, suggesting that it carries function [5]. Because the core complex consists of an Agl-like motor and a simplified Nfs/Glt-like apparatus (Figure S9), this machinery may constitute a basal transport machinery. A survey of Agl-Glt/Nfs-like machineries in the deltaproteobacteria shows that these machineries adopt many potential conformations and therefore may cover a potentially broad functional repertoire (Figure S9) [5]. Functional specialization of Agl-Glt/Nfs machineries may be linked to the type of transported cargo, and thus expansion of the core system in the deltaproteobacteria may have evolved diverse tasks, sugar polymer transport (Nfs and Glt), but also potentially protein and lipid transport. Even though glt and nfs paralogues have a rather narrow distribution [5], the Agl motor itself is a member of a ubiquitous family of bacterial motors (Mot/Tol/Exb) [4]. Therefore, it is conceivable that other motor associations also promote processive transport in bacteria.
Many Agl-Glt/Nfs systems are present in bacteria that are not currently genetically tractable. Computational genetic reconstruction of ancestral assemblages is an emerging powerful means to explore both the function of a macromolecular complex and how it evolved [32]. In future works, the reductive genetic study of the Agl-Glt/Nfs machineries in Myxococcus should be instrumental to characterize functional intermediates and test the functional repertoire of these systems. Such study would also allow deciphering the evolution of this complex biological machine with high likelihood, an emerging challenge in evolution biology [33]. Finally, critical traits of the Myxococcus lifestyle evolved from the diversification of a single molecular system, showing that modifications of a single genetic system can give rise to profound ecological adaptations.
Protein sequences were analyzed by BlastP (NCBI) searches of the nonredundant (nr) and Pfam (release 24.0) (see comment on Table S1) databases [34]. Signal peptides signal and transmembrane helices were the predicted using the signalP 3.0 [35] and TMHMM v.2.0 [36] servers, respectively.
Strains, primers, and plasmids are listed in Tables S2, S3, and S4. See Tables S2, S3, and S4 for strains and their mode of construction. M. xanthus strains were grown at 32°C in CYE rich media as previously described [37]. Plasmids were introduced in M. xanthus by electroporation. Mutants and transformants were obtained by homologous recombination based on a previously reported method [37]. E. coli cells were grown under standard laboratory conditions in Luria-Bertani broth supplemented with antibiotics, if necessary.
Bacterial two-hybrid experiments, plate, and β-Galactosidase assays were performed as previously described [5] and as recommended by the manufacturer's instructions (Euromedex).
Sporulation was induced by adding glycerol to a final concentration of 0.5 M directly to a flask of CYE-grown Myxococcus cells (OD600 of 0.5) as previously described [14]. Viable spore titers were determined after 24 h based on resistance to heat and sonication. For this, 10 mL cells were harvested, pelleted at 5,000×g during 5 min at room temperature, resuspended in 10 ml sterile water, incubated at 50°C for 2 h, and sonicated three times (30 pulses, output 3, 50% duty) in ice water. The surviving spores were counted directly on the microscope after 10 min incubation in a DAPI staining solution (formaldehyde 4%—DAPI 1 µg/mL) with a detection limit of 102 spores/ml.
Protein lysates for Western blot analysis were generated by harvesting sporulating cells. Spores were pelleted and resuspended in Tris-HCl 50 mM pH 7.8 supplemented with benzonase (Sigma) and phenylmethanesulfonylfluoride (PMSF, Sigma). Spores were lysed in a Fast-Prep-24 at 6.5 m.s−1 for 45 s, 8 times (matrix lysing B, MP Biomedicals, France). Protein concentrations were determined by Bradford assay (Bio-Rad) according to the manufacturer's instructions. Lysates were solubilized in Laemmli sample buffer and resolved by sodium-dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) using 8% polyacrylamide concentration for GltD. Proteins were transferred to nitrocellulose membrane (0.45 µm, Bio-Rad) using a tank transfer system (Bio-Rad) and probed with anti-GltD rabbit polyclonal antibody [5] at a 1∶1,000 dilution and HRP-conjugated goat anti-rabbit secondary antibodies (Bio-Rad) at 1∶5,000. The signals were revealed with a SuperSignal West Pico Chemiluminescent Substrate (Thermo Scientific) and imaged in a LAS4000 Luminescent Image Analyzer (GE Healthcare).
Commercial microscopy chambers (Ibitreat uncoated, Biovalley) were filled with 100 µL of carboxymethylcellulose sodium salt (medium viscosity, Sigma-Aldrich) solution (1.5 mg/mL) diluted in de-ionized water and left 15 min at room temperature. The excess of coating solution was removed by flushing with de-ionized water first, followed by flushing with TPM buffer (10 mM Tris pH 7.6; 8 mM MgSO4; 1 mM KH2PO4). The channels were then filled with 100 µL of a CYE-grown cell suspension (OD600 = 0.5) previously washed and resuspended in TPM solution and left at room temperature during 20 min. Nonadhering cells were flushed with TPM, and sporulation was induced by adding glycerol to a final concentration of 0.5 M directly on the microscope stage. Time-lapse experiments and image processing were performed as previously described using an inverted Nikon TE2000-E-PFS inverted epifluorescence microscope [38].
All image analysis was performed under Image J (NIH). Fluorescence quantifications were performed by integrating fluorescence intensities and normalizing over the level of background fluorescence measured for cells that do not express fluorescent proteins. Kymographs were obtained as follows: a typical NfsD-mCherry cluster path was obtained by summing a stack of time-lapse images. Kymographs were then computed by reslicing along that path, defined with the “Segmented Line” selection tool.
To measure the fluorescence areas and discriminate GSL-I staining in various mutants, we first defined the specific signal by thresholding and removing background fluorescence, by subtraction. For each cell, the total area of GSLI-specific fluorescence was measured and normalized by the total area of the cell. Box plots were then computed under R.
In drug experiments, Carbonyl Cyanine-M-Chlorophenylhydrazone (CCCP, 1 mM, Sigma Aldrich) and A22 (50 µg/ml, Calbiochem) were injected manually into the flow chamber. For lectin staining, a Griffonia (Bandeiraea) simplicifolia (GSL)-FITC conjugated 2 mg/mL stock solution (CliniSciences) was diluted 1∶100 in TPM containing 1 mM of CaCl2 and 100 µg/mL of bovine serum albumin (BSA, Sigma-Aldrich) immediately prior to injection. The mixture was then injected. The unbound lectin was washed (TPM, CaCl2 1 mM, BSA 100 µg/mL) out of the flow chamber after 40 min of incubation.
For each experiment, stacks of images were first normalized to correct for background fluctuations over time. If required, the background intensity of phase contrast images was subtracted to optimize auto-thresholding operations. Cell boundary and major and minor axis were detected using a specifically developed plug-in for ImageJ. Briefly, cells were detected using an autothresholding function, and subpixel resolution refined cell contours were obtained using a cubic spline-fitting algorithm. Major axis was deduced from the skeleton and expended in both directions to the most probable point, maximizing the cell-boundary curvature and minimizing the angle between the centerline and the cell boundary. Fluorescent foci were detected using a local and subpixel resolution maxima detection algorithm. The position of each focus is first determined using a polar coordinate system (r, θ), where the radial distance (r) represents the distance from the focus position and the cell center and the angular coordinate (θ) represents the angle formed between the major cell axis and the focus (Figure S5). Since sporulating cells are spherical and NsfD-mCherry foci form close to the cell surface, the polar coordinate system was extended to a 3D spherical coordinate system (p, θ, φ), where the radial distance (p) represents the distance from the cell center to the cell boundary, the angular coordinate (θ) represents the angle formed between the major cell axis and the orthogonal projection of the focus position on the focus plane, and the azimuthal angle (φ) is deduced using the relation φ = arcos(r/p) (Figure S5). The distance between two points at the surface of a spherical spore of respective spherical coordinates (p1, θ1, φ1) and (p2, θ2, φ2) was deduced from the relation d = p·arcos(cos(θ1) ·cos(φ1) · cos(θ2) · cos(φ2)+sin(θ1) · cos(φ1) · sin(θ2) · cos(φ2)+sin(φ1) · sin(φ2)). By convention, a positive angular coordinate means that the angle θ is measured counterclockwise from the polar axis formed by the major axis.
Fluorescent foci were tracked over time with a specifically developed plug-in for ImageJ. Briefly, cells are tracked with an optimized nearest-neighbor linking algorithm using the polar (r, θ) or the spherical coordinates (p, θ, φ). From foci trajectories, the distance between two temporal points was used to calculate instantaneous speeds shown in Figure 4B and combined with the cumulated distance and the distance from the origin to compute the Mean Square Displacement (MSD). For WT cells, the mean velocity and the diffusion coefficient were extracted from the second order fit of the MSD. For each condition tested, the MSD of at least 15 individual foci trajectory was calculated.
M. xanthus cells were fixed for 1 h with glutaraldehyde 2.5% in CYE medium and postfixed 1 h in 2% OsO4. Then cells were dehydrated in ethanol and embedded in epon. Ultrathin sections were stained with aqueous uranyl acetate for 10 min and lead citrate for 5 min. Sample observations were performed on a Tecnai-G2 LaB6 microscope (FEI Company) operating at 200 kV.
Custom microscopy chambers were made of a 1 mm-thick coverslide and a thin coverslip (#1, thickness 100 µm) separated by a double layer of double-sided sticky tape (Scotch). Chambers were immersed using 1.5%-agarose in DMSO (6 M, Sigma-Aldrich) for 15 min. Then, chambers were extensively washed with TPM containing 10 mM glucose. The chambers were filled with CYE-grown cell suspension (OD600 = 0.8) exposed to glycerol (final concentration of 0.5 M) 2 h or 4 h prior to experiments, and washed and resuspended in TPM solution directly before infusion. After 30 min, access was washed out, leaving sporulated and nonsporulated M. xanthus stuck to the surfaces of the microscopy chamber. Submicron-sized polystyrene beads (diameter 520 nm) were gently placed atop the sporulating cells using an optical trap system as previously described [4]. A low-powered tracking laser (<1 mW power at the sample, wavelength 855 nm) was focused on a spore-attached bead. The forward scattered laser light was collected on a position-sensitive photodiode (Model 2931, New Focus). The bead position was recorded and used to update the stage position using a PID feedback at a frequency of 50 Hz. The accuracy of this technique was measured to be better than 4 nm. Simultaneously, we used an EMCCD camera (iXon, Andor) to record time-lapsed videos at 1 Hz [4]. Video and high-resolution tracking data were recorded for several hours for all experimental conditions.
|
10.1371/journal.pgen.1007802 | A natural antisense lncRNA controls breast cancer progression by promoting tumor suppressor gene mRNA stability | The human genome encodes thousands of long noncoding RNA (lncRNA) genes; the function of majority of them is poorly understood. Aberrant expression of a significant number of lncRNAs is observed in various diseases, including cancer. To gain insights into the role of lncRNAs in breast cancer progression, we performed genome-wide transcriptome analyses in an isogenic, triple negative breast cancer (TNBC/basal-like) progression cell lines using a 3D cell culture model. We identified significantly altered expression of 1853 lncRNAs, including ~500 natural antisense transcript (NATs) lncRNAs. A significant number of breast cancer-deregulated NATs displayed co-regulated expression with oncogenic and tumor suppressor protein-coding genes in cis. Further studies on one such NAT, PDCD4-AS1 lncRNA reveal that it positively regulates the expression and activity of the tumor suppressor PDCD4 in mammary epithelial cells. Both PDCD4-AS1 and PDCD4 show reduced expression in TNBC cell lines and in patients, and depletion of PDCD4-AS1 compromised the cellular levels and activity of PDCD4. Further, tumorigenic properties of PDCD4-AS1-depleted TNBC cells were rescued by exogenous expression of PDCD4, implying that PDCD4-AS1 acts upstream of PDCD4. Mechanistically, PDCD4-AS1 stabilizes PDCD4 RNA by forming RNA duplex and controls the interaction between PDCD4 RNA and RNA decay promoting factors such as HuR. Our studies demonstrate crucial roles played by NAT lncRNAs in regulating post-transcriptional gene expression of key oncogenic or tumor suppressor genes, thereby contributing to TNBC progression.
| Breast cancer is the most common cancer in women worldwide. The molecular mechanisms underlying the disease have been extensively studied, leading to dramatic improvements in diagnostic and prognostic approaches. Despite the overall improvements in survival rate, numerous cases of death by breast cancer are still reported per year, alerting us about the potential gap of knowledge in cancer molecular biology era. The emerging advances in new generation sequencing techniques have revealed that the majority of genome is transcribed into non-protein coding RNAs or ncRNAs, including thousands of long ncRNAs (lncRNAs) of unknown function. Natural antisense RNAs (NATs) constitute a group of lncRNAs that are transcribed in the opposite direction to a sense protein-coding or non-coding gene with partial or complete complementarity. In this manuscript, we investigate the role of NATs in breast cancer progression, focusing on the role of PDCD4-AS1, a NAT expressed from the established tumor suppressor PDCD4 gene locus. We observe that both PDCD4-AS1 and PDCD4 display concordant expression in breast cancer cell lines and patients. In mammary epithelial cells, PDCD4-AS1 promotes the stability of PDCD4 mRNA. PDCD4-AS1 by forming RNA duplex with PDCD4 RNA prevents the interaction between PDCD4 RNA and RNA decay factors in the nucleus.
| While more than 80% of the genome is transcribed to RNA, high throughput gene expression analyses have revealed that only 2% of transcribed RNAs are translated into proteins. Current studies estimate that the human genome harbors several thousands of noncoding RNA (ncRNA) genes [1,2,3,4]. NcRNAs are grouped into different subclasses; from short non-coding transcripts like miRNAs and piRNAs (~20–30 nucleotides [nts] long), to middle range ncRNAs like snRNAs and snoRNAs (~30–200 nts long), and finally the long non-coding RNAs (lncRNAs) (>200 bp in length). So far, the most studied class is microRNAs (miRNAs), which promote gene silencing by inhibiting translation of target genes and/or by destabilizing the mRNAs [5,6]. LncRNAs comprise the least studied, but most complex group of ncRNAs. Unlike miRNAs, lncRNAs are very diverse with respect to their function, localization, abundance and interacting partners [7]. For instance, lncRNAs can form complex 3D secondary structures with the capacity to bind to proteins as well as to nucleic acids (DNA as well as RNA). This dual capacity renders lncRNAs as an ideal regulator in protein-nucleic acid network. The human genome is estimated to contain ~16000 lncRNA genes [https://www.gencodegenes.org]. Based on the genome positioning, lncRNA genes could further be grouped into subclasses, including NATs or natural antisense transcripts (~5501), lincRNAs or long intergenic non-coding RNAs (~7499), sense intronic RNAs (~905), sense overlapping RNAs (~189), and processed transcripts (~544) [https://www.gencodegenes.org].
Breast cancer (BC) is the most common cancer in women, underscoring a need for research and development of more efficient treatment strategies [8]. BC is a heterogeneous disease and comprises several subtypes based on the presence or absence of three hormone receptors; estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor 2 (HER2). Based on the expressions of receptors, BC is categorized as Luminal A (ER positive and/or PR positive and HER2 negative), Luminal B (ER positive and/or PR positive and HER2 negative or positive), HER2+ (ER and PR negative, HER2 positive) and triple-negative breast cancer (ER/PR/HER2 negative). The clinical outcome is worst for triple-negative breast cancer (TNBC) patients mainly due to lack of any of the three hormone receptors and, consequently, poor response to hormone-targeted therapies [9,10,11,12]. Therefore, there is an emergent need to investigate the molecular biology of the TNBC subtype to identify efficient prognostic and diagnostic markers.
Current research on BC primarily focuses on the role of protein-coding genes in the disease progression. However, recent studies indicate that a significant number of lncRNAs show aberrant expression in BC patients (For review please see [13]). Abnormal expression of several lncRNAs is associated with chemoresistance in BC cells [14]. However, the underlying molecular mechanism remains to be determined for most cases. Mechanistic studies have indicated that several of the BC-deregulated lncRNAs play crucial roles in disease pathology. For example, HOTAIR is known to negatively regulate the expression of many protein-coding genes by recruiting repressive PRC2 and LSD1 complexes to chromatin. HOTAIR is overexpressed in a significant number of BC patients, and is shown to act as a powerful predictor of metastasis [15]. We and others have demonstrated the involvement of MALAT1 in breast cancer progression and metastasis [16,17]. MALAT1 is overexpressed in a significant number of BC patients, and its depletion compromises both tumorigenic and metastatic properties of BC cells. In a mouse mammary carcinoma model, genetic loss or systematic depletion of MALAT1 in MMTV-PyMT resulted in slower tumor growth and reduction in metastasis [16]. In addition to HOTAIR and MALAT1, both of which promote oncogenesis, lncRNAs such as GAS5 are shown to act as tumor suppressors [18]. As of now, we understand the molecular action of only a handful of the several thousands of lncRNAs that show aberrant expression in BC patients.
In order to understand the role of lncRNAs during TNBC progression, we performed RNA-seq in an isogenic tumor progressive TNBC cell line series and compared the expression of all of the annotated lncRNAs to a normal-like mammary epithelial cell line. We found that 1853 lncRNAs showed aberrant expression in the metastatic BC cells. Among these lncRNAs, >1/4 (504/1853) of them are found to be natural antisense transcripts (NATs). Interestingly, we observed that several of these NATs are transcribed in opposite orientation to key oncogenic and tumor suppressor protein-coding genes, and the expression of both sense and antisense transcripts is co-regulated in both TNBC cells and BC patient samples. Mechanistic studies of one such NAT, PDCD4-antisense RNA1 (PDCD4-AS1) in BC progression demonstrated that it regulates the expression of its sense protein-coding partner, PDCD4 (Programmed Cell Death 4) in cis. PDCD4, initially identified in a screen aimed to determine apoptosis-induced targets [19], is a well-established tumor suppressor gene [20]. We observed that the reduced levels of PDCD4-AS1 lncRNA in TNBC cells were correlated with reduced expression of PDCD4 in these cells. Further, we demonstrated that PDCD4-AS1 acted upstream of PDCD4 and induced PDCD4 expression by enhancing the stability of PDCD4 RNA. Our studies have unearthed novel NAT-mediated post-transcriptional mechanisms controlling the expression of protein coding genes in cis.
Human breast carcinomas are suggested to evolve via sequential genetic modifications from benign hyperplasia of mammary epithelial cells, into atypical ductal hyperplasia, to ducal carcinoma in situ (DCIS) and eventually to fully malignant tumors that possess the potential to metastasize into distant organs [21,22,23]. In order to understand the role of lncRNAs during various stages of breast cancer (BC) progression, we utilized a well-established isogenic mammary epithelial cell line-derived BC progression model system [21,23]. This system consists of multiple cancer cell lines of basal-like or TNBC subtype, all of which were initially derived from the spontaneously immortalized, non-tumorigenic mammary epithelial cell line, MCF10A [24]. The model system comprises of 4 isogenic cell lines, categorized as M1-M4 [21,23]. M1 represents the normal, non-tumorigenic, immortalized MCF10A cells. Transfection of MCF10A with activated T24-HRAS and selection by xenografting generated the M2 (MCF10AT1k.cl2) cell line, which is highly proliferative and gives rise to premalignant lesions with the potential for neoplastic progression. M3 (MCF10Ca1h) and M4 (MCF10CA1a.cl1) were derived from occasional carcinomas arising from xenografts of M2 cells. M3 gives predominantly well-differentiated low-grade carcinomas on xenografting, while M4 gives rise to relatively undifferentiated carcinomas and colonizes to the lung upon injection of these cells into the tail vein [22,25,26,27,28,29]. These lines represent progression through various stages of breast tumorigenesis and recapitulate key steps that mimic the progression of breast cancer in vivo [25]. In addition, the common genetic background of these cells enables us to rule out the genetic variation behind the deregulated gene expression. We hypothesized that functional characterization of lncRNAs, especially those displaying differential expression among these cell lines, would help us to determine their roles in TNBC development.
We cultured M1-M4 cells as three-dimensional (3D) acinar or organoid-like structures in Matrigel for 7–10 days, as 3D acini structurally and morphologically resemble in vivo acini of breast glands and lobules [28,30]. We performed poly A+ selected paired-end deep RNA-seq (~160–250 million reads/sample) in two biological replicates and analyzed the expression of 28905 genes in M1, M2, M3 and M4 cells (17396 protein coding and 11509 lncRNAs) (GENCODE Release v19 [GRCh37]) (Fig 1A). We identified transcripts, which were more than 2-fold deregulated in both biological repeats. Since we were primarily interested in lncRNAs that show abnormal expression during BC progression and metastasis, we initially compared gene expression between M1 and M4 cells (S1 and S2 Tables). Expression of 4668 genes (2815 protein coding and 1853 lncRNAs) were altered >2-fold change in their expression between M1 and M4 cells in both biological repeats (Fig 1B). 1159 out of the 1853 deregulated lncRNA genes showed >2-fold upregulation in M4 cells (Fig 1C, S2 Table). On the other hand, 694 lncRNA genes displayed reduced expression in M4 compared to M1 cells. Further, we noticed that natural antisense transcripts (NATs) comprised one of the largest types of lncRNAs (504 out of 1853), along with lincRNAs and pseudogenes, which showed deregulation in M4 cells (Fig 1D). Our data supports observations from a recent study, reporting deregulated expression a significant number of NATs in breast cancer samples [31].
NAT lncRNAs are typically enriched in the nucleus [1,32,33], and recent studies indicate that several of the NATs function in cis by regulating the expression of their sense partner protein-coding genes (for review please see [31,33,34]). To gain insights into the potential NAT-mediated cis-gene regulation in BC cells, we examined the status of co-regulated expression of 504 NATs and their protein-coding partner in M1 and M4 cells. We observed that 108 out of 504 deregulated NATs and their sense protein coding genes showed >2-fold change in expression (S3 and S4 Tables). Among them, 94 (~87%) NAT: mRNA pairs showed concordant pattern of deregulation (i.e., both sense/antisense pairs are up- or are down-regulated concordantly) and 14 (~13%) pairs exhibited discordant pattern of expression (Fig 1E and 1F). To assess if these NATs potentially regulate the expression of protein-coding genes that play crucial roles in BC progression, we determined the percentage of the sense protein coding genes in the sense: NAT pair that play well-established roles in cancer progression. We compiled data sets from multiple sources to identify potential cancer-associated genes, that are involved in vital cellular processes such as cell cycle and Epithelial-to-Mesenchymal transition (EMT) (https://www.qiagen.com), (http://www.bushmanlab.org/links/genelists), [17,35] (S5 Table). By such analysis, we identified 29 deregulated NAT: mRNA pairs in which the protein coding genes have established roles in cancer progression (Fig 1F and 1G, S6 Table). Furthermore, comparison of expression data of these NATs with ‘clinical survival in invasive breast carcinoma patient dataset’ (TCGA dataset, containing 105 normal samples and 814 breast tumors) revealed that the expression of 3 of these NATs was well correlated with survival outcomes in BC patients (S7 Table) [36]. Thus, BC deregulated NAT: sense protein-coding genes could potentially play vital roles in BC progression and survival.
To gain insights into the role of NATs in BC progression, we focused our attention on one NAT lncRNA, PDCD4-AS1 for the following reasons. PDCD4-AS1 is a NAT lncRNA, transcribed from the complementary strand of Programmed Cell Death 4 (PDCD4) gene (Fig 2A). PDCD4 is a known tumor suppressor gene that negatively regulates cell proliferation, neoplastic transformation and tumor invasion [37]. RNA-seq, RT-qPCR and immunoblot analyses demonstrated reduced levels of PDCD4-AS1, PDCD4 mRNA and protein in M2, M3 & M4 cells compared to M1 (Fig 2B–2D & S1A and S1B Fig). Furthermore, PDCD4 and PDCD4-AS1 RNAs showed significant positive correlation with each other in breast cancer patient RNA data set (Fig 2E). Further, gene expression data from breast invasive carcinoma patients (TCGA data set) [36] revealed that PDCD4-AS1 showed lowest levels in basal-like or TNBC patients compared to Luminal A, Luminal B and HER2 subtypes (Fig 2F). Highest levels of PDCD4-AS1 were observed in stage Tis (stage 0, pre-cancer) breast samples compared to samples from the more aggressive stages of BC (Fig 2G). Finally, the elevated levels of PDCD4-AS1 were correlated with better survival rate in a cohort of BC patients (Fig 2H). Similar to PDCD4-AS1, TNBC patient samples showed lowest levels of PDCD4 mRNA, and higher PDCD4 mRNA levels correlated with better survival in BC patients, further supporting its role as a potential tumor suppressor (S1C and S1E Fig). Our results indicate that the levels of PDCD4-AS1 and PDCD4 mRNA are co-regulated in BC cell lines and in BC patients. Low expression of PDCD4-AS1 in BC patient samples as well as better survival of patients with higher levels of PDCD4-AS1 implies that PDCD4-AS1, similar to its sense partner PDCD4, might function as a tumor suppressor.
RNA-seq and RT-qPCR analyses in M1 cells determined PDCD4-AS1 as a multi-exonic (two exons), ~778 nts long polyadenylated transcript (S1F Fig & S2I Fig). CPAT algorithm (Coding Potential Assessing Tool) identified PDCD4-AS1 as a noncoding RNA, as its coding potential score was relatively low and comparable to other well-established lncRNAs such as MALAT1 (S1G Fig). Further, cellular fractionation followed by RT-qPCR assays indicated that PDCD4-AS1 lncRNA was enriched in the nuclear fraction in mammary epithelial cells (Fig 2J). Finally, we determined the turnover rate of PDCD4-AS1 in M1 cells. RNA stability assay indicated that PDCD4-AS1 is a relatively stable transcript, and it displayed similar stability to its protein-coding partner PDCD4 mRNA (t1/2 of ~4hrs; Fig 2K). Our results identify PDCD4-AS1 as a stable, poly A+ lncRNA that is enriched in the nucleus.
PDCD4 was initially identified as a tumor suppressor gene that was upregulated during serum starvation or cellular quiescence [19]. To test whether PDCD4-AS1 is also induced under conditions that activate PDCD4, we determined the expression of PDCD4 and PDCD4-AS1 in asynchronous and quiescent (serum-starved) M1 cells (S2A Fig and S2B Fig). RT-qPCR and immunoblot data revealed elevated levels of both PDCD4 (mRNA and protein) and PDCD4-AS1 RNA in quiescent cells (S2C Fig & S2D Fig). Our results indicate that PDCD4-AS1 shows co-regulated expression with its protein-coding partner PDCD4.
Since a lower level of PDCD4-AS1 RNA was associated with poor survival in breast cancer patients, and since it showed positive correlated expression with the tumor suppressor gene PDCD4 both in breast cancer cells and in patients, we evaluated whether PDCD4-AS1 contributes to cancer-associated phenotypes. We stably depleted PDCD4-AS1 transcripts by using three independent shRNAs targeting the sequences of PDCD4-AS1 (exon 2) that were not overlapping with PDCD4 mRNA (S3A Fig & S3B Fig) in non-tumorigenic mammary epithelial (M1) cells. RT-qPCR revealed that PDCD4-AS1 shRNA successfully depleted both nuclear and cytoplasmic pool of PDCD4-AS1 (S3C Fig). Next, we analyzed the migration potential of control and PDCD4-AS1-depleted cells. M1 cells depleted of PDCD4-AS1 showed enhanced migration as observed by both transwell migration and wound healing assays (Fig 3A–3D). Next, we overexpressed the full length PDCD4-AS1 in highly tumorigenic and metastatic M4 cells (M4 cells contain lower levels of endogenous PDCD4-AS1) and determined the effect on cell migration and long-term cell proliferation. We observed that PDCD4-AS1-overexpressing M4 cells showed significant reduction in their ability to migrate (Fig 3Ea-b) and displayed reduced proliferation (Fig 3Ec-d). It is known that tumor suppressor PDCD4 inhibits cell proliferation [38]. Flow cytometric analyses revealed increased population of S and G2/M in PDCD4-depleted M1 cells (Fig 3F and 3G). Similarly, PDCD4-AS1-depleted M1 cells also showed increased population of S and G2/M cells (Fig 3H & 3I). Collectively, these results indicate that both PDCD4 and PDCD4-AS1 negatively regulate cell proliferation in human mammary cells.
We observed that depletion of PDCD4-AS1 increased cell cycle progression, and migratory properties of M1 cells. Depletion of PDCD4 is also known to promote tumorigenic properties of human cells (For review please see [37]). Similar to what we observed upon depletion of PDCD4-AS1, PDCD4-depleted M1 cells also showed enhanced cell cycle progression and cell migration (Fig 3F and 3G & Fig 3J and 3K). Based on this, we hypothesize that PDCD4-AS1 negatively regulates tumorigenic properties of cells via modulating the expression of PDCD4. To determine whether PDCD4-AS1 acts upstream of PDCD4, we exogenously expressed of PDCD4 in PDCD4-AS1-depleted M1 cells and tested the effect on cell migration phenotype (S4A Fig). Trans-well migration assays revealed that M1 cells transiently overexpressing PDCD4 alone did not show any significant change in their ability to migrate in vitro (S4A Fig), while PDCD4-AS1-depleted control cells displayed increased migration (Fig 3A and 3B & L [left and middle panels]). In contrast, overexpression of PDCD4 in cells that were stably depleted of PDCD4-AS1 rescued the enhanced migration, as these cells showed comparable levels of migration to control cells (Fig 3L and 3M; compare left and right panels in 3L). Based on these results, we hypothesize that PDCD4-AS1 negatively regulates cellular migration via modulating PDCD4 expression/activity.
To determine whether PDCD4-AS1 negatively regulates cell proliferation and cell migration by regulating the expression of PDCD4 in cis, we examined the level of PDCD4 mRNA and protein in M1 cells stably depleted of PDCD4-AS1 using shRNAs. We observed that PDCD4-AS1-depleted cells showed consistent reduction in the levels of PDCD4 mRNA and protein (Fig 4A & Fig 4B). In addition, cells depleted of PDCD4-AS1 using modified antisense DNA oligonucleotides (GAPMER ASOs) against PDCD4-AS1 also showed reduction in the levels of PDCD4 mRNA (S3D Fig). Also, cells treated with PDCD4-AS1 specific ASOs displayed cell cycle defects that were similar to PDCD4-AS1 shRNA-treated cells (S3E Fig). In addition, cell fractionation followed by RT-qPCR in control and PDCD4-AS1-depleted cells showed significant reduction in the levels of PDCD4 in the nuclear pool, supporting the argument that PDCD4-AS1 primarily functions in the nucleus (S3F Fig). Cells depleted of PDCD4 using two independent PDCD4 specific siRNAs did not show similar decrease in the levels of PDCD4-AS1 transcript (Fig 4C & 4D). In case of PDCD4-AS1-mediated regulation of PDCD4, we tested whether depletion of PDCD4-AS1 also alters the expression of other genes located in close genomic proximity. RT-qPCR analyses revealed that the expression of several other genes (BBIP1, SHOC2 and RBM20 [Fig 2A] that are located in genomic regions close to PDCD4-AS1/PDCD4 locus remained unaltered upon PDCD4-AS1 or PDCD4 depletion (S4B Fig & S4C Fig). These results imply that PDCD4-AS1 positively and specifically regulates the expression of its sense transcript.
NATs could regulate the expression of their sense partner genes either by influencing transcription or by modulating post-transcriptional processing of sense transcripts (for review please see [33]). To determine whether PDCD4-AS1 regulates the transcription of PDCD4 gene, we quantified the levels of nascent PDCD4 pre-mRNA in control and PDCD4-AS1-depleted cells by nascent RNA capture followed by RT-qPCR analysis. PDCD4-AS1-depleted M1 cells did not show any significant change in the total levels of nascent PDCD4 pre-mRNA, indicating that PDCD4 transcription remained unaffected in cells lacking PDCD4-AS1 (Fig 4E). Next, to test whether PDCD4-AS1 influenced post-transcriptional processing of PDCD4 mRNA, we performed RNA stability assay. We treated control and PDCD4-AS1-depleted cells with an RNA polymerase II transcription inhibitor Falvopiridol (1μM), collected samples at several time points post drug treatment, and performed RT-qPCR analyses to determine the relative levels of PDCD4 mRNA. Control cells displayed a half-life of ~5 hrs for PDCD4 mRNA (Fig 4F). However, cells depleted of PDCD4-AS1 showed ~50% reduction in the stability of PDCD4 mRNA (half-life ~2.5 hrs) (Fig 4F). These results indicate that PDCD4-AS1 positively regulates the stability of PDCD4 mRNA.
NATs regulate the stability of their sense RNAs by forming RNA duplex [39,40]. Among the several NATs that are involved in conferring mRNA stability, only a few have been shown to form RNA:RNA duplex with their sense RNAs [41,42]. In the case of PDCD4-AS1/PDCD4 pair, the 5’end of both the transcripts, including exon 1 and part of intron 1, showed complete complementarity (Fig 5A; relative position within PDCD4-AS1 is highlighted in red lines). In addition, two other repetitive sequence elements located within exon 2 of PDCD4-AS1 show significant complementarity with sequences within the 3’UTR of PDCD4 mRNA. A 258 nt long sequence (position 523–778 in exon 2) in PDCD4-AS1 shows 75% complementarity to a sequence within the 3’UTR PDCD4 mRNA (position 3164–3417). Besides this one, another shorter repeat of 103 nts long (position 204–306 of exon 2) in PDCD4-AS1 also shows 82% complementarity with the PDCD4 mRNA 3’UTR (position 3134–3236) (S4D Fig), indicating that multiple elements within PDCD4-AS1 and PDCD4 mRNA could complement to form RNA duplexes. To determine whether PDCD4-AS1 and PDCD4 RNA form RNA-duplex under in vivo conditions, we initially performed double-strand RNase protection assays as reported earlier [43,44]. RNaseA specifically cleaves the single-stranded RNAs but have no activity on double-stranded/duplex RNAs. RNase protection assays revealed regions within PDCD4-AS1 and PDCD4 mRNA that were protected from RNaseA treatment, implying the presence of RNA duplex under in vivo conditions (Fig 5B). We used BACE1/BACE1-AS pairs as a positive control [43] (Fig 5B). Next, we performed RNA pulldowns followed by RT-qPCR to test physical association between PDCD4-AS1 and PDCD4 RNAs [42]. Towards this, we incubated biotin-labeled PDCD4-AS1 with cell extracts and performed RNA pulldowns using streptavidin-coated beads, followed by RT-qPCR assays. We observed significant interaction between PDCD4-AS1 and endogenous PDCD4 RNA in the pulldown experiment (Fig 5C).
Next, we determined to identify sequence elements within PDCD4-AS1 that play crucial roles in promoting PDCD4 mRNA stability. To this end, we generated full length and three mutant PDCD4-AS1 constructs (PDCD4-AS1-FL, PDCD4-AS1Δ208–778, PDCD4-AS1Δ477–778, PDCD4-AS1Δ1–207), each of the mutants lacks specific sequence elements that contain PDCD4 complementary sequences (Fig 5A). We expressed these constructs in control and endogenous PDCD4-AS1-depleted M1 cells and determined the effect on endogenous PDCD4 mRNA levels. RT-qPCR assays in nuclear and cytoplasmic fractionated cell extracts revealed that the transiently expressed full-length and mutant RNAs were localized in both the nucleus and cytoplasm (S4E Fig & S4F Fig). Interestingly, endogenous PDCD4-AS1-depleted M1 cells expressing PDCD4-AS1-FL, PDCD4-AS1Δ1–207 and PDCD4-AS1Δ477–778 RNA rescued PDCD4 mRNA levels (Fig 5D). However, PDCD4-AS1Δ208–778 construct, which lacks the second exon of PDCD4-AS1, expressing cells failed to rescue the level of PDCD4 mRNA. Furthermore, RNA stability assays revealed that both PDCD4-AS1Δ1–207 and PDCD4-AS1Δ477–778 and not PDCD4-AS1Δ208–778 rescued the overall stability of PDCD4 mRNA (S4G Fig). Based on this, we conclude that sequence elements within the exon 2 of PDCD4-AS1, which display complementarity to the 3’UTR of PDCD4 mRNA play crucial roles in stabilizing PDCD4 mRNA.
Association of RNA-binding proteins (RBPs) to 3’UTRs is known to influence the cellular levels of PDCD4 mRNA. It was reported recently that RBPs such as HuR (human antigen R) and TIA1 (T-Cell intracellular antigen-1) recognize overlapping sequence within PDCD4 mRNA 3’UTR, and positively regulate PDCD4 mRNA levels [45]. Hence, we sought to determine if PDCD4-AS1 regulates the stability of PDCD4 mRNA by influencing the binding of these RBPs to PDCD4 mRNA 3’UTR. ENCODE eCLIP data set identified several potential binding sites of HuR and TIA1 on PDCD4 RNA [46]. We performed RNA-immunoprecipitation (RIP) under crosslinking conditions using HuR or TIA1 antibody followed by RT-qPCR to determine the interaction between endogenous HuR or TIA1 and PDCD4 mRNA in control and PDCD4-AS1-depleted cells. RIP assays in control cells revealed that both HuR and TIA1 interacted with PDCD4 mRNA (Fig 5E & S4H Fig). PDCD4-AS1-depleted cells showed reduced interaction between TIA1 and PDCD4 mRNA (S4H Fig). On the contrary, PDCD4-AS1-depleted cells showed significantly enhanced interaction between HuR and PDCD4 mRNA (Fig 5E). Altered interaction of TIA1 or HuR with PDCD4 mRNA in PDCD4-AS1-depleted cells was not due to overall changes in the total cellular levels of RBPs (S4I Fig). Next, we examined if the depletion of HuR and TIA1 would affect the PDCD4 mRNA levels in mammary epithelial cells. Contrary to the earlier report, [45], TIA1-depleted mammary cells did not reduce the levels of PDCD4 mRNA (S4J Fig & S4K Fig). On the other hand, HuR depletion significantly increased PDCD4 mRNA and protein levels in control cells, indicating that in mammary epithelial cells HuR negatively regulates the levels of PDCD4 mRNA (Fig 5F and 5G & S4L Fig). Finally, depletion of HuR in PDCD4-AS1-depleted M1 cells rescued the levels of PDCD4 mRNA and protein (Fig 5F–5H). On the other hand, HuR depletion did not significantly alter the levels of PDCD4-AS1 RNA, indicating that HuR functions downstream of PDCD4-AS1 in the PDCD4-AS1: PDCD4: HuR axis (Fig 5H). Thus, we conclude that PDCD4-AS1 promotes PDCD4 mRNA stability by negatively regulating HuR binding to PDCD4 mRNA. It is likely that the reduced binding of TIA1 to PDCD4 mRNA in PDCD4-AS1-depleted cells is a consequence of enhanced interaction of HuR to the same sequence elements, which also interact with TIA1.
In the present study, we have attempted to understand the involvement of lncRNAs that are differentially expressed in TNBC cell lines, in cancer cell properties. We focused our efforts on NATs and in particular, the roles played by PDCD4-AS1 in regulating the expression of its sense partner, PDCD4. We selected PDCD4-AS1/PDCD4 pair for mechanistic studies due to the following reasons. First, PDCD4 is a tumor suppressor gene, and shows reduced expression in several types of cancer, including BC [20,47,48,49,50,51,52,53,54,55,56,57]. Second, both PDCD4-AS1 and PDCD4 show concordant expression in BC cell lines and in TNBC patient samples. Finally, clinical survival data in BC patients revealed that similar to PDCD4 gene, lower expression of PDCD4-AS1 reduced overall patient survival, implying a tumor suppressor role for PDCD4-AS1. PDCD4 is a homolog of eukaryotic translation initiation factor 4G (EIF4G), and by forming a complex with EIF4A1, PDCD4 reduces the interaction between EIF4A1 and EIF4G, thereby inhibiting EIF4A1’s helicase activity. PDCD4 negatively regulates the translation of several oncogenes such as Cyclins, B-Myb and c-Myb [58,59]. Because of its critical role in several vital biological processes, its cellular level under normal physiological conditions is tightly regulated via several transcriptional and post-transcriptional regulatory mechanisms [60,61,62,63,64,65,66]. Our studies, demonstrating the role of PDCD4-AS1 in enhancing the cellular levels of PDCD4 adds another layer of complexity in PDCD4 regulation during BC progression.
NATs are widely present in the human genome, and on an average ~38% of genomic loci in cancer cells express sense: anti-sense pairs [35]. However, NATs are expressed in much lower levels compared to sense transcripts, are mostly enriched in the nucleus, and several of them are shown to influence the expression of their sense partners via cis-mediated gene regulation [35]. Similar to earlier observations, we observed aberrant expression of a significant percentage of NATs during BC progression [31,34,35]. Moreover, we observed that several of the NATs expressed from cancer-associated gene loci showed concordant expression with the oncogenic or tumor suppressor sense partner genes and also displayed survival significance in patients, implying their potential involvement in contributing to the molecular pathology of BC progression and or metastasis.
We observed that PDCD4-AS1 promotes the stability of PDCD4 mRNA in TNBC cells. PDCD4-AS1 depletion did not alter PDCD4 transcription significantly while it compromised the stability of PDCD4 mRNA. Further, we observed that PDCD4-AS1 forms RNA duplex with PDCD4 mRNA, and exon 2 of PDCD4-AS1 contains sequence elements that promote PDCD4 mRNA stability. PDCD4-AS1 could utilize multiple mechanisms to enhance RNA stability. It is possible that by forming RNA duplex, PDCD4-AS1 could prevent RNase-mediated degradation of PDCD4 mRNA, as observed in the case of FGFR3-AS1 [67]. Additionally, such RNA duplexes could prevent the binding of miRNAs to the 3’UTR of PDCD4 mRNA, thereby stabilizing the transcript, as observed in the case of BACE-AS1/BACE1 pair [33,43]. However, it is quite unlikely that PDCD4-AS1 promotes PDCD4 mRNA stability via regulating miRNA binding. Unlike BACE-AS1, PDCD4-AS1 is predominantly localized in the nucleus, and stabilizes nuclear pool of PDCD4 RNA. A recent study also reported the role of NAT in regulating the expression of its sense partner by modulating chromatin organization [68]. VIM-AS1 transcribed from Vimentin (VIM) gene locus positively regulates VIM expression by forming RNA:DNA R-loop structure [68]. Disruption of VIM-AS1-mediated R-loop structure compromised VIM expression by inducing local chromatin compaction as well as reduced association of transcription factors to VIM promoter. In the case of PDCD4-AS1, its depletion did not significantly change PDCD4 transcription, indicating that PDCD4-AS1 might not act via such a mechanism.
Alternatively, PDCD4-AS1 by forming RNA duplex with PDCD4 RNA could influence the binding of RNA-binding proteins (RBPs) to PDCD4 mRNA. We observed that PDCD4-AS1 negatively regulates the association of HuR with PDCD4 mRNA. HuR-depletion studies in M1 cells further identified HuR as a destabilizer of PDCD4 mRNA. HuR/ELAVL1 is a U-/AU-rich element interacting RBP that is known to regulate mRNA stability. (For review on HuR in breast cancer cells please see [69]. Several recent studies have described the role of HuR in destabilizing RNAs [70,71,72,73,74]. For example, HuR utilizes AUF1, Ago2 or let-7 miRNA as co-factors to enhance the decay of p16(INK4) and MYC mRNAs [73,74]. HuR also promotes the early steps of myogenesis by destabilizing nucleophosmin/NPM mRNA [72]. We recently reported that in mouse cells, double stranded RNA binding protein ADAR1 & 2 negatively regulates HuR-mediated degradation of a significant number of RNAs [70,71]. Earlier studies have observed that NATs by forming RNA duplex with regions of mRNA containing AU-rich sequences, influences that association of AU-rich interacting RNA decay factors, thereby controlling mRNA stability [75,76]. For example, an antisense RNA from HIF1α locus destabilizes one of the isoforms of HIF1α by binding to it and exposing the AU-rich sequence element within the HIF1α 3’UTR [76]. On the other hand, a NAT transcribed from the Bcl2/IgH hybrid gene stabilizes the mRNA by masking the AU-rich sequence element [75]. In the present study, we observed that HuR destabilizes PDCD4 mRNA. The molecular mechanism underlying PDCD4-AS1-mediated inhibition of HuR/PDCD4 RNA interactions remained to be determined. It is known that a significant proportion of HuR is localized in the nucleus, and we have previously shown that nuclear pool of HuR destabilizes RNA [70,71]. Based on this, we hypothesize that the formation of RNA duplex between PDCD4-AS1 and PDCD4 RNA in the nucleus occludes the binding of HuR to the PDCD4 RNA, thereby stabilizing PDCD4 RNA (Fig 5I).
At present, it is not clear how NATs, which in general are present in lower copy numbers (~10–100 fold) than their sense protein coding transcripts modulate post-transcriptional RNA processing in cis [35]. For example, Wrap53, a NAT that is expressed at 100-fold lower levels than its sense partner, the tumor suppressor p53 gene, positively regulates the stability of p53 mRNA [42]. Similarly, low copy NAT, iNOS-AS (expressed in 7 fold lower) transcribed from iNOS locus interacts with the 3’UTR of iNOS RNA and positively regulates its stability [77,78]. As a matter of fact, the question of how low copy NATs regulate post-transcriptional processing of their sense transcripts remains an “unresolved conundrum” in the antisense-RNA field [79]. At present, there is no convincing molecular explanation of how NATs regulate the stability of high copy sense RNAs. Several studies have posed models to explain potential mode of action [42,80]. It is suggested that transient association of NAT with its sense RNA allows one NAT molecule to interact with multiple sense transcripts in a ‘hit and run’ fashion [42]. Such interactions could initiate local changes in sense RNA structure that favor or inhibit the binding of RBPs [42]. In a “recycling hypothesis” model, short complementary regions within the sense RNA:NAT pair promote intermolecular RNA:RNA interactions [80]. These interactions are transient and unstable due to the low melting temperature of the small duplex, and trigger conformational changes in the sense RNA, allowing either enhanced accessibility of a stabilizing RNA-binding protein or decreased affinity of an RNA decay factor to RNA, thereby modulating RNA stability. Once an RNP complex is formed, and the sense RNA is stabilized, the NAT is released from the complex and is recycled to stabilize another RNA molecule [80]. We observed that PDCD4-AS1 is expressed ~18 fold lower than PDCD4 mRNA in total cell extracts. However PDCD4-AS1/PDCD4 ratio in the nucleus, especially at their site of transcription will be much higher due to the fact that a major fraction of PDCD4-AS1 is enriched in the nucleus, where as most of the PDCD4 mRNA is exported to the cytoplasm. Based on these data, we hypothesize that transient interaction between PDCD4-AS1 and PDCD4 RNA in the nucleus, preferentially at the site of transcription, trigger conformational changes in PDCD4 RNA, resulting in differential binding of RBPs, such as HuR (decay factor) and AUF1 (stabilizing factor) to PDCD4 RNA. In this scenario, a single PDCD4-AS1 RNA could interact with several PDCD4 RNAs during its lifetime. In general, our studies have underscored the importance of a NAT in BC progression via its role in regulating the expression of a tumor suppressor sense partner. Future studies will unravel mechanistic roles of hundreds of other BC-deregulated lncRNAs in breast cancer biology.
All of the patient RNA-seq data was obtained from the publicly available database, TCGA (https://cancergenome.nih.gov/), and no additional ethics approval was needed.
Acinar culture of M1-M4 cells was performed similar to three-dimensional culture of MCF10A cells described elsewhere [30]. Briefly, growth-factor reduced Matrigel was used to coat multi-well plates. A single-cell suspension of each of the cell lines M1-M4 was prepared. M2-M2 cells were suspended in an assay medium containing growth medium (DMEM/F12 containing 2% Horse serum, 1 mg/ml hydrocortisone, 1 mg/ml cholera toxin, 10 mg/ml insulin, 10 ng/ml EGF, and 1% penicillin/streptomycin as well as 2.5% Matrigel dissolved in the medium. M3-M4 cells are prepared in the same way but omitting the EGF in the medium. The cells were seeded at a concentration of 8000 cells/mL. Media was changed every fourth day. Cells were cultured for 8 days prior to harvesting.
M1 and M2 cells were cultured in DMEM/F12 medium containing 5% horse serum supplemented with 100 U/mL penicillin, 100μg/mL streptomycin, 20ng/mL EGF (epidermal growth factor), 0.5 μg/mL Hydrocortisone, 100ng/mL Cholera toxin, 10 μg/mL insulin and 5% horse serum. M3 and M4 cells were cultured DMEM/F12 medium containing 5% horse serum supplemented with 100 U/mL penicillin, 100μg/mL streptomycin.
8 (biological replicates of M1-M4) poly A+ RNA samples were pooled and sequenced in two lanes on HiSeq using Illumina TruSeq mRNA Prep Kit RS-122-2101 and paired-end sequencing. The samples have 163 to 256 million pass filter reads with a base call quality of above 94% of bases with Q30 and above. Reads of the samples were trimmed for adapters and low-quality bases using Trimmomatic software before alignment with the reference genome (Human—hg19) and the annotated transcripts using STAR. The average mapping rate of all samples is 96%. Unique alignment is above 87%. There are 3.74 to 4.07% unmapped reads. The mapping statistics are calculated using Picard software. The samples have 0.59% ribosomal bases. Percent coding bases are between 67–72%. Percent UTR bases are 23–26%, and mRNA bases are between 94–96% for all the samples. Library complexity is measured in terms of unique fragments in the mapped reads using Picard’s MarkDuplicate utility. The samples have 31–52% non-duplicate reads. In addition, the gene expression quantification in raw count format was performed for all samples using STAR/RSEM tools by the annotation of Gencode v19 and normalized by size factor implemented in DESeq2 package. We calculated the fold change gene expression based on FPKM data. We identified deregulated genes with >2 fold cut off and then made the overlap list between two biological repeats. RNA seq data is deposited to GEO (GEO accession number GSE120796).
Trizol reagent (Invitrogen) was used to extract total RNA according to manufacturer’s protocol. The concentration was measures using Nanodrop instrument (ThermoFisher SCIENTIFIC). RNA was treated with RNase-free DNase I (Sigma, USA) and cDNA was synthesized from RNA using High capacity reverse transcription kit (Applied Biosystem). Quantitative PCR was carried out by StepOnePlus system (Applied Biosystem). For gene specific primers please see S8 Table.
PDCD4 depletion was achieved by transfection with siRNA against GL3 (control) or siRNAs against PDCD4 (40–50 nM con, IDT) for one round using Lipofectamine RNAiMax reagent (Invitrogen). TIA1 depletion was performed using siRNA purchased from IDT. HuR depletion was carried out using siRNA as used in [81]. PDCD4-AS1 knockdown was performed by shRNA lentivirus-mediated transduction. PDCD4-AS1 depletion was achieved using gapmer ASOs at 200 nM final concentration (Ionis Pharmaceuticals Inc.). For overexpression, full-length PDCD4 was purchased as pGEX6p1-hPdcd4 from Addgene [82] and cloned into pCGT vector. Full length PDCD4-AS1 and mutants were purchased as gblocks from IDT technology, cloned and expressed in pCGT vector, and empty vector was used as control.
We used transwell migration chambers (Corning, Cat# 354578) and to perform migration assays as previously explained [17]. Briefly, cells were starved in a serum-free medium, which was then trypsinized, counted and seeded in serum-free medium in transwell chamber (8μM). We placed the cell containing chambers into a well containing serum (24-well plate). Cells were kept in incubator 37 C, 5% CO2 overnight. Migrating cells were stained by Crystal Violet 0.05% and counted the day after.
The wound was created by 200 μl filter tips. After washing with PBS, serum-free medium was added to cells in order to discourage the cell proliferation. Images were taken at Day0, 1, 2 and 3 after wound creation to monitor the wound healing.
Click-iT Nascent RNA capture kit (Invitrogen, Cat # C10365) was used to isolate nascent RNA following the product’s protocol. Then quantitative RT-qPCR was performed using gene-specific primers.
Cells were treated with Flavopiridol (1M) and were collected at different time points post treatment. RNA extraction and RT-qPCR was carried out as explained above.
RIP was conducted as described before [71,83]. Briefly, RNA-Protein interactions were reversibly crosslinked by formaldehyde in cells. Cells were lysed and lysate was immunoprecipitated using Anti-HUR (HuR (3A2): sc-5261, Santa Cruz Biotechnology) and Anti-TIA1 antibody (TIA-1 (G-3): sc-166247, Santa Cruz Biotechnology). After RIP, we reversed cross-link and RNA extraction using Trizol LS (Invitrogen). DNase I treatment, reverse transcription and qPCR was performed as described above.
As explained in [71], we washed cells with PBS and lysed in RSB buffer (10 mM Tris-HCl pH7.4, 100 mM NaCl, 2.5 mM MgCl2, RNase Inhibitor) supplemented with Digitonin (8 g/ml) (D141-100MG, Sigma-Aldrich, USA) for 10 min on ice. Lysate was centrifuged at 2000 rpm, 4°C, 10 min. Supernatant was collected as cytoplasmic fraction and RNA was extracted from with Trizol LS (Invitrogen). The pellet included the nuclear fraction. We washed the nuclear pellet with RSB-Digitonin solution one more time and then RNA was extracted using Trizol reagent (Invitrogen).
Poly(A) fractionation was performed as previously described [44]. In brief, NucleoTrap mRNA kit (Clontech) was used to fractionate Poly(A) plus and Poly(A) minus fractions following by extraction, RT and qPCR.
The experiment was performed as described previously [44]. Cells were washed with PBS and lysed in lysis buffer (10 mM Tris pH 7.4, 3 mM CaCl2, 2 mM MgCl2, and 0.7% NP-40). Cell lysate was passed through needle (27.5 gauge) five times and then incubated on ice for 10 minutes. The final solution was adjusted to DNase I (Sigma) 12.5 units/ml and 125 mM NaCl. The lysate was divided to two fractions. To one fraction RNase A (QIAgen) and the other fraction RNAse Inhibitor was added to final concentrations of 200 ng/ml and 250 units/ml; respectively. Then, solutions were incubated at 37°C for 40 minutes. RNA was extracted using Trizol LS (Invitrogen).
PDCD4-AS1 and YFP (-ve control) full-length cDNA cloned in pGEM-Teasy plasmids were in vitro transcribed to generate biotinylated RNA (Biotin Labeling Mix; Roche). M1 whole cell extract was incubated with biotin-labeled transcripts followed by streptavidin-mediated RNA pull down. Then RNA extraction, RT-PCR and qPCR were performed to analyze potential RNA: RNA interactions.
|
10.1371/journal.pgen.1007967 | A synonymous germline variant in a gene encoding a cell adhesion molecule is associated with cutaneous mast cell tumour development in Labrador and Golden Retrievers | Mast cell tumours are the most common type of skin cancer in dogs, representing a significant concern in canine health. The molecular pathogenesis is largely unknown, but breed-predisposition for mast cell tumour development suggests the involvement of inherited genetic risk factors in some breeds. In this study, we aimed to identify germline risk factors associated with the development of mast cell tumours in Labrador Retrievers, a breed with an elevated risk of mast cell tumour development. Using a methodological approach that combined a genome-wide association study, targeted next generation sequencing, and TaqMan genotyping, we identified a synonymous variant in the DSCAM gene on canine chromosome 31 that is associated with mast cell tumours in Labrador Retrievers. DSCAM encodes a cell-adhesion molecule. We showed that the variant has no effect on the DSCAM mRNA level but is associated with a significant reduction in the level of the DSCAM protein, suggesting that the variant affects the dynamics of DSCAM mRNA translation. Furthermore, we showed that the variant is also associated with mast cell tumours in Golden Retrievers, a breed that is closely related to Labrador Retrievers and that also has a predilection for mast cell tumour development. The variant is common in both Labradors and Golden Retrievers and consequently is likely to be a significant genetic contributor to the increased susceptibility of both breeds to develop mast cell tumours. The results presented here not only represent an important contribution to the understanding of mast cell tumour development in dogs, as they highlight the role of cell adhesion in mast cell tumour tumourigenesis, but they also emphasise the potential importance of the effects of synonymous variants in complex diseases such as cancer.
| The combination of various genetic and environmental risk factors makes the understanding of the molecular circuitry behind complex diseases, like cancer, a major challenge. The homogeneous nature of pedigree dog breed genomes makes these dogs ideal for the identification of both simple disease-causing genetic variants and genetic risk factors for complex diseases. Mast cell tumours are the most common type of canine skin cancer, and one of the most common cancers affecting dogs of most breeds. Several breeds, including Labrador Retrievers (which represent one of the most popular dog breeds), have an elevated risk of mast cell tumour development. Here, by using a methodological approach that combined different techniques, we identified a common inherited synonymous variant, that predisposes Labrador Retrievers to mast cell tumour development. Interestingly, we showed that this variant, despite its synonymous nature, appears to have an effect on translation dynamics as it is associated with reduced levels of DSCAM, a cell adhesion molecule. The results presented here reveal dysregulation of cell adhesion to be an important factor in mast cell tumour pathogenesis, and also highlight the important role that synonymous variants can play in complex diseases.
| Mast cell tumours (MCTs) are the most common type of skin cancer in dogs [1], and the second most frequent form of canine malignancy in the United Kingdom [2]. Recent estimates of the mean age of dogs diagnosed with a MCT range from 7.5 to 9 years [3–5]. The majority of affected dogs are successfully treated by surgery and/or local radiotherapy, but around 30% of patients require a systemic treatment, due to tumour metastasis, and have an extremely poor prognosis [6]. Canine MCTs share many biological features with human mastocytosis [7], a heterogeneous group of neoplastic conditions characterised by the uncontrolled proliferation and activation of mast cells.
Mutations in the proto-oncogene, c-kit, which encodes KIT, a member of the tyrosine kinase family of receptors, are found in 20–30% of canine MCTs and in more than 90% of adult human mastocytoses [8–10]. In the case of human mastocytosis, most of the mutations are single nucleotide polymorphisms (SNPs) in exon 17, which result in alterations in the kinase domain of the receptor, with the most reported one being the V816D substitution [11]. In canine MCTs, most c-kit alterations are tandem repeats/small indels in either exons 11 and 12 (that result in alterations in the receptor’s juxtamembrane domain), or in exons 8 and 9 that encode part of the extracellular ligand-binding domain. C-kit alterations have recently been shown to be associated with DNA copy number alterations and with increased canine MCT malignancy [12]. They have also been explored therapeutically, and tyrosine kinase inhibitors are now used for the treatment of canine MCTs that cannot be surgically removed, or that are recurrent [13]. In the case of human mastocytosis, tyrosine kinase inhibitor resistance is associated with the most frequent c-kit gene mutation [14]. Although the identification of somatic c-kit mutations has contributed to the development of therapeutics, c-kit mutations are not found in the majority of canine MCTs [15].
Human mastocytosis has been associated with underlying germline risk factors [16, 17]. Pedigree dog-breeds display significant differences in the incidence of MCTs; German Shepherd Dogs, Border Collies and Cavalier King Charles Spaniels are underrepresented amongst affected dogs, while Boxers (Odds ratio: 15.11; [18]), Golden Retrievers (Odds ratio: 6.93;[18]) and Labrador Retrievers (Odds ratio: 4.63;[18]) have an increased risk of MCT development [2, 4, 18, 19]. This suggests the involvement of inherited genetic risk factors in the development of MCTs in breeds which display increased susceptibility, although there is no evidence for the occurrence of germline c-kit risk variants.
Certain characteristics of the domestic dog’s genome make it amenable to the genetic mapping of inherited disease-associated variants. The successive bottlenecks in the recent history of modern dog breeds, which were derived from extensive selection for phenotypic traits, have resulted in long regions of linkage disequilibrium (LD) within dog breeds [20]. The consequent reduced level of genetic complexity facilitates within-breed positional mapping of disease-associated variants, reducing the required study population size from the thousands needed for mapping human disease genes to hundreds [21].
Through a genome-wide association study (GWAS) and subsequent sequence capture and fine mapping of a region containing an associated SNP marker, Arendt and co-workers identified a germline SNP that is associated with MCTs in European Golden Retrievers [22]. The SNP is located in an exon of the Nucleotide Binding Protein (G Protein) Alpha Inhibiting Activity Polypeptide 2 (GNAI2) gene on canine chromosome (CFA) 20, and causes alternative exon splicing and a truncated protein [22]. In the same study, a haplotype encompassing the HYAL4 and SPAM11 genes on CFA14 associated with MCTs in United States (US) Golden Retrievers was also identified [22]. More recently, a GWAS identified an association between MCTs in US Labrador Retrievers and a SNP marker on CFA36 [23], although a susceptibility variant has yet to be identified,
In this work, we aimed to identify germline variants that predispose Labrador Retrievers to the development of MCTs. The identification of MCT susceptibility variants in Labrador Retrievers could not only contribute to understanding of the molecular mechanisms involved in canine MCT development, but could also help to shed light onto human mastocytosis pathogenesis. With an analysis approach that combined GWAS, targeted next generation sequencing (NGS) and TaqMan genotyping, we have identified a synonymous MCT-associated variant that is associated with significantly reduced levels of a cell adhesion molecule.
We conducted an initial meta-analysis of three GWAS datasets comprising a total of 105 MCT cases and 85 controls (Sets 1, 2, and 3 in S1 Table). This analysis revealed a SNP on CFA31 that showed a strong statistical association with MCT just below the threshold of genome-wide statistical association (P-value = 7.6 x 10−7; Bonferroni correction for multiple testing of 115,432 SNPs: P = 4.3 x 10−7). The strongest associated SNP BICF2P951927 was at 34.7Mb (CanFam 3.1) (Fig 1A; S1 Fig). The common T allele at this locus was associated with an increased risk of MCT.
As MCT is likely to be a complex trait, we could not identify any clear shared haplotypes amongst cases, and examination of linkage disequilibrium (LD) amongst 2,033 GWAS SNPs on CFA31 using the pooled set of 190 dogs did not identify any other SNPs tagged by SNP BICF2P951927 at an r2 of 0.8 or above. We therefore delineated a critical region of association for further interrogation of the underlying sequence using a conservative empirical statistical threshold of P≤0.01 for SNP association results spanning SNP BICF2P951927 (Fig 1B). This resulted in an approximate 2.9Mb region (CanFam 3.1 co-ordinates CFA31:34433688–37366557).
Subsequent to selection of this region for resequencing, we received three additional datasets comprising in total a further 68 cases and 28 controls (Sets 4, 5 and 6 in S1 Table). We therefore repeated the above meta-analysis [one individual was dropped from dataset 3 (S1 Table) as it was reported to be suffering from cancer (not a MCT)], which comprised a total of 173 cases and 112 controls. The CFA31 association increased in strength to exceed genome-wide statistical association in this analysis (SNP BICF2P951927; P-value = 1.9 x 10−8; S2 Fig). We also conducted a secondary meta-analysis following individual-dataset adjustment for population stratification and the association for this SNP further increased in magnitude to P-value = 1.9 x 10−9; (S3 Fig). This analysis revealed additional genome-wide associated loci on other chromosomes. However, we have focused on the CFA31 region here as it showed the strongest association; analysis of the additional regions will be undertaken in future studies.
The associated 2.9Mb region of CFA31 was captured from libraries prepared from germline DNA samples from six Labrador Retrievers affected by a MCT and six unaffected dogs over the age of 7 years, and sequenced. All the affected dogs carried two copies of the GWAS MCT-associated BICF2P951927 allele ‘T’, and all unaffected dogs were homozygous for the alternative allele ‘C’. A total of 19,930 variants (including 4,028 that were not found in any of the unaffected dogs) were identified amongst the 12 dogs. Of the variants, 126 displayed the same segregation pattern as the GWAS MCT-associated SNP (i.e. the six cases were homozygous for the reference allele, and the six controls were homozygous for an alternative allele). However, all 126 variants were located within introns (that were part of a single gene, DSCAM), and these were not considered to be strong candidate MCT susceptibility variants. Alternatively, variants were selected for further analysis on the basis of a combination of both: (a) The potential functional consequence assessed according to the position of a variant (regardless of whether the variant was predicted, by Variant Effect Predictor and/or SIFT, to be deleterious), and (b) The extent to which a variant segregated between the six cases and six controls. Specifically, 23 variants (22 SNPs and one deletion; Tables 1 and 2) that fulfilled both of the following criteria were selected for genotyping in a large case-control set:
TaqMan Genotyping Assays were designed for the 22 SNPs. The indel variant at CFA31:34667505 was genotyped by fluorescent end point PCR fragment analysis. The 23 candidate MCT susceptibility loci were genotyped in 407 UK Labrador Retrievers comprising 191 MCT cases and 216 controls (including 71 cases and 42 controls from the GWAS study) (S2 Table). The SNP rs850787912 was excluded from the association analysis because it strongly deviated from Hardy-Weinberg distribution (P-value = 2.2 x 10−83), indicating assay failure.
One of the 22 analysed loci (SNP rs850678541, at CFA31:34760750) demonstrated statistical association with MCT (P-value = 5.2 x 10−4; Table 3). This association was stronger than that of the strongest associated GWAS SNP BICF2P951927 (Table 4). The SNP is associated with MCT with an odds ratio of 1.67 (95% confidence interval 1.24–2.24), and explains 2% (pseudo r2) of the MCT trait in this breed. The alternative ‘A’ allele is common—72% of the genotyped dogs (including 67% of controls) carried at least one copy, and 25% of the dogs (including 20% of controls) carried two copies (Table 4). This allele increases the risk of MCT development by 1.66 x (ratio of heterozygote odds: reference allele homozygote odds; 95% confidence interval 0.99–2.77) when present as one copy, and by 2.79 x (ratio of alternative allele homozygote odds: reference allele homozygote odds; 95% confidence interval 1.55–5.03) when present as two copies.
The alternative (variant) allele of SNP rs850678541 (CFA31:34760750) represents a G>A transition (plus DNA strand) located in exon 16 of the canine DSCAM gene, which encodes a cell adhesion molecule. It occurs in the third base of a codon (representing arginine), and, as such, is a synonymous mutation (changing the codon from CGC to CGT). A growing body of evidence indicates that, although synonymous mutations do not cause amino acid sequence changes, they can have an effect on factors such as mRNA stability and translation kinetics, and thus have significant biological consequences [26–30]. Consequently, we investigated if the alternative allele of SNP rs850678541 had any effect on DSCAM mRNA and protein levels. DNA, RNA and protein were simultaneously extracted from 17 RNAlater-preserved MCT biopsies borne by Labrador Retrievers (representative of the three locus CFA31:34760750 genotypes; Biopsies #1–17 in Fig 2), and from normal skin biopsies from three Labrador Retrievers (Biopsies #18–20 in Fig 2). The levels of DSCAM mRNA and protein expression were compared between the three genotypes.
As SNP rs850678541 is a synonymous variant, we investigated the possibility that it was not a causal variant, but that it tagged another DSCAM gene variant that actually caused the observed protein level effect. The variants identified by targeted resequencing of the associated 2.9Mb CFA31 region in 12 Labrador Retrievers included 2,045 at loci in the DSCAM gene. In addition to SNP rs850678541, of the remaining 2,044 DSCAM gene variants, 13 were located in exons (five synonymous variants and eight in the 3'-UTR), 1975 were located in introns (including one within a ‘splice region’), and 56 were upstream of the DSCAM gene. Consequently, we screened for LD between SNP rs850678541 and each of the remaining 2,044 loci (1,950 biallelic and 94 multiallelic). Twenty-two intronic DSCAM loci (comprising 13 SNPs and nine indels) were found to be in LD with SNP rs850678541 at an r2 of 0.8 (S6 Table). Intronic variants can disrupt splicing enhancer sites or branch points, and can also activate cryptic splicing sites [32] that compete with the canonical sites, leading to the generation of alternative splicing products [33]. The antibody employed in Western blot analysis recognises an epitope that is translated from a sequence located in exon 23 of the DSCAM gene. Consequently, an intronic mutation that generates an alternative mRNA transcript lacking exon 23 would not necessarily be detectable by RT-qPCR assay of DSCAM exon 16 expression, but could lead to a reduction in the level of the 196kDa protein encoded by the 30 exon 1,7725b DSCAM mRNA transcript (ENSCAFT00000016117), such as that observed in the MCT and normal skin biopsies homozygous for the SNP rs850678541 alternative ‘A’ allele (Figs 4 and 5). The 22 intronic variants were screened for those that could potentially affect mRNA splicing using the Human Splicing Finder web tool [34]. This analysis identified three variants that could potentially lead to the generation of new splicing products: (1) a variant (at CFA31:34767321; biallelic locus ‘16’ in S6 Table) in the intron between exons 14 and 15 that could disrupt the splicing branch point, and generate a splicing product that would include 73 additional nucleotides from the intron; (2) a variant (at CFA31:34761118; biallelic locus ‘8’ in S6 Table) in the intron between exons 15 and 16 that could activate a cryptic intronic donor splice site that (if used instead of the canonical site) would generate a splicing product including 5,980 nucleotides from the intron; and (3) a variant (at CFA31:34760052; biallelic locus ‘18’ in S6 Table) in the intron between exons 16 and 17 that could also activate a cryptic intronic donor splice site that (if used) would generate a splicing product with an additional 644 nucleotides from the intron. End point PCR assays were performed to investigate if any of the three predicted alternative splice variants were present in MCT biopsies borne by dogs homozygous for the alternative allele ‘A’ of SNP rs850678541, on the presumption that dogs homozygous for this allele would also be homozygous for the variants at the three intronic loci shown to be in LD with SNP rs850678541. The possible effect of the variant located in the intron between exons 14 and 15 was investigated using an assay (E14-15 Assay) that targets an amplicon spanning the end of exon 14 and the beginning of exon 15, whilst an assay (E15-17 Assay) targeting an amplicon spanning the end of exon 15, exon 16, and the beginning of exon 17 was employed to assess the possible effects of the variants located in the introns between exons 15 and 16, and between exons 16 and 17, respectively. End point PCR assay of MCT cDNAs prepared from two SNP rs850678541 reference ‘G’ allele homozygotes and two SNP rs850678541 alternative allele ‘A’ homozygotes showed no differences between the exonic fragments amplified (Fig 6). For both the E14-15 and E15-17 Assays only the expected exonic mRNA fragment was amplified irrespective of SNP rs850678541 genotype (Fig 6). These results indicate that the variants at the three intronic DSCAM loci in LD with SNP rs850678541 are not likely to cause the ten-fold reduction in DSCAM protein expression observed in MCTs and normal skin tissues that are homozygous for SNP rs850678541 alternative allele ‘A’.
We investigated if the SNP rs850678541 genotype was associated with a difference in the mean age at which a Labrador Retriever developed a MCT. Labrador Retrievers which were homozygous for the reference ‘G’ allele had a later mean age of onset (8.59 ± 2.75 years; n = 54) than heterozygotes (7.81 ± 2.74 years; n = 69) and dogs homozygous for the alternative ‘A’ allele (7.82 ± 2.92 years; n = 25). However, the differences between the three genotypes (Kruskal-Wallis test P-value = 0.52), and between pairs of genotypes (e.g. reference allele homozygotes v alternative allele homozygotes: Mann-Whitney U test P-value = 0.37) were not statistically significant. As the SNP rs850678541 alternative allele is associated with a significant reduction in the protein level expression of a cell adhesion molecule, we also undertook a preliminary investigation of whether it is also associated with MCT metastasis in Labrador Retrievers. The SNP was genotyped in five Labrador Retrievers that died due to MCT metastatic disease (as confirmed by abdominal/thoracic imaging and lymph node histopathological examination) and eight Labrador Retrievers for which MCT metastases could not be detected and whom were still alive 1,000 days post-diagnosis. The dogs genotyped were either heterozygotes (ten dogs: five with metastatic MCT, and five with non-metastatic MCT), or homozygous for the reference ‘G’ allele (three dogs with non-metastatic MCT). No association was found between MCT metastasis and the SNP rs850678541 genotype (Fisher exact test P-value = 0.43) in this small preliminary dataset.
SNP rs850678541 was genotyped in a MCT case-control set of UK Golden Retrievers, a breed that is both closely related to Labrador Retrievers [35] and has an elevated risk of developing MCTs [2, 4, 19]. Germline DNAs from 37 Golden Retrievers that either currently or previously had a MCT and 53 dogs aged at least 7 years of age that had never been affected by any form of cancer were genotyped. SNP rs850678541 demonstrated statistical association with MCT (P-value = 0.01) that was directionally consistent and of a similar magnitude of effect to that observed in Labrador Retrievers, and accounted for 5% (pseudo r2) of the MCT trait in Golden Retrievers (Table 5). The alternative ‘A’ allele was common in this Golden Retriever set (70% of the dogs, including 62% of controls, carried at least one copy, and 26% of the dogs, including 17% of controls, carried two copies) (Table 5). This allele increases the risk of MCT development by 1.90 x (ratio of heterozygote odds: reference allele homozygote odds; 95% confidence interval 0.65–5.54) when present as one copy, and by 4.44 x (ratio of alternative allele homozygote odds: reference allele homozygote odds; 95% confidence interval 1.34–14.77) when present as two copies.
SNP rs850678541 was also genotyped in the Border Collie (110 dogs) and Cavalier King Charles Spaniel (105 dogs), two breeds which are under-represented amongst dogs that develop MCTs [4, 19]. The alternative ‘A’ allele was present in both breeds at a frequency (Border Collie: 0.058; Cavalier King Charles Spaniel: 0.38) lower than that in the Labrador Retriever (0.49) and Golden Retriever (0.48).
We investigated if the MCT susceptibility SNP rs851590509 at CFA20: 39080161, which was previously identified in European Golden Retrievers by Arendt and co-workers [22], is also associated with MCTs in Labrador Retrievers. The variant is located in an exon of the GNAI2 gene and causes alternative exon splicing and a truncated protein. We performed TaqMan genotyping of rs851590509 in 167 cases and 193 controls from our extended MCT case-control set of UK Labradors. The alternative ‘A’ ‘risk allele’ of the SNP is rare in Labrador Retrievers (frequency in the whole set: 0.007), and no association was found with MCTs (Fisher Exact P-value = 0.09). Arendt et al. also identified a putative MCT susceptibility locus at CFA14:14.7Mb in Golden Retrievers from the United States (although the most significantly associated SNPs were not found to be associated with the MCT trait in European Golden Retrievers). A causal variant for the CFA14:14.7Mb association has yet to be identified, and for this reason we did not screen for associations between CFA14:14.7Mb SNPs and the MCT trait in our UK Labrador Retriever cohort.
Our next step was to evaluate the extent of the risk conferred by rs850678541 and rs851590509 in our UK Golden Retriever set of 37 MCT cases and 53 controls. TaqMan genotyping of SNP rs851590509 in this set showed that the alternative ‘A’ allele is extremely common (83% of the dogs, including 74% of controls, carried at least one copy, and 42% of the dogs, including 21% of controls, carried two copies), and has a statistically significant association with MCTs (P-value = 1.5 x 10−7) (Table 5). Furthermore, a combined analysis of rs850678541 and rs851590509 in this set of Golden Retrievers demonstrated a statistically significant association with MCTs (P-value = 2.6 x 10−8) and revealed that collectively these variants explain 29% of the MCT trait in this breed (Table 5). Due to the rarity of the rs851590509 SNP in the Labrador Retriever set we could not perform a combined analysis of rs851590509 and rs850678541 in this breed.
In this study we have identified a synonymous germline variant (‘A’ allele of SNP rs850678541) in the DSCAM gene that is associated with the elevated risk of MCT development in Labrador Retrievers. We revealed that, although the variant has no effect on DSCAM mRNA expression, it is associated with a significantly reduced DSCAM protein level in MCTs and in normal skin. The demonstration that intronic variants at loci in the DSCAM gene that are in LD with SNP rs850678541 do not cause alternative exon splicing (that may be reflected in a decrease in the level of the full length 196kDa DSCAM protein—UniProtKB F1PA86_CANLF) affords a strong indication that the SNP rs850678541 alternative allele may be responsible for the significant reduction in DSCAM protein expression observed in MCTs and normal skin specimens from Labrador Retrievers homozygous for the alternative allele. The variant allele is common in Labrador Retrievers, is associated with a per allele increase in MCT risk of 1.66 x, and is estimated to account for 2% of the MCT trait in the breed.
SNP rs850678541 was also shown to be a risk factor for MCT development in Golden Retrievers (accounting for 5% of the MCT trait in the breed), suggesting that the variant arose in a common ancestor at some point prior to divergence of the Labrador and Golden Retriever breeds. The strength of the association (odds ratio = 2.11) between the SNP and MCTs in our set of Golden Retrievers suggests that a lack of statistical power may be the reason why an association to SNPs in the vicinity of CFA31 34.7Mb was not detected in the European Golden Retriever MCT GWAS performed by Arendt and colleagues [22]. An alternative explanation for this is that the CFA31 SNPs on the canineHD array were not able to ‘capture’ SNP rs850678541 in Golden Retrievers due to a different haplotype structure in this breed. The association between the CFA20 SNP rs851590509 and MCTs in European Golden Retrievers reported by Arendt and co-workers [22] was reproduced in our set of Golden Retrievers. Significantly, our combined analysis showed that collectively SNPs rs850678541 and rs851590509 explain 29% of the MCT trait in Golden Retrievers. In our set of Labrador Retrievers, SNP rs851590509 was very rare, which did not allow for a combined analysis to be undertaken. To the best of our knowledge, the SNP rs850678541 described here is currently the only MCT-associated variant in Labrador Retrievers to be identified, although it is likely that other MCT-associated variants will be described because our secondary GWAS meta-analysis has suggested associations with other genomic regions. Furthermore, the demonstration that MCT susceptibility loci are shared by Labrador and Golden Retrievers, suggests that meta-analysis of genotype data from both breeds may uncover additional MCT susceptibility loci.
A recent study demonstrated the presence of Mendelian disease variants in pedigree dog breeds for which the disease/an elevated risk of developing the disease had not previously been reported [36], leading the investigators to speculate that the ‘genetic background’ may affect how a mutation is manifest. The most notable example is arguably the SOD1:c.118A allele, homozygotes and heterozygotes of which in 5 breeds are associated with degenerative myelopathy. The SOD1:c.118A allele is also present (at up to a high frequency) in many breeds [37] that are not known to develop degenerative myelopathy, suggesting that the penetrance of the allele is affected by other genetic or environmental factors. In this study we found that the SNP rs850678541 alternative ‘A’ allele, which is associated with MCTs in Labrador and Golden Retrievers, is present at a lower frequency in Border Collies (frequency 12.3 x lower) and Cavalier King Charles Spaniels (frequency 1.3 x lower), two breeds that are under-represented amongst MCT-affected dogs [4, 19]. As MCT susceptibility appears to be complex, the risk conferred by the SNP rs850678541 alternative ‘A’ allele in Labrador and Golden Retrievers has to be considered in the context of potential modifying alleles at other MCT susceptibility loci that may be present in Labrador and Golden Retrievers and absent from other breeds. Indeed, susceptibility variants have been found to modify the risk of breast cancer development associated with the BRCA1 and BRCA2 mutations, thereby accounting for the variation in breast cancer penetrance observed for these mutations in different human families [38]. Extensive GWAS of human diseases has demonstrated that genetic risk factors underlying complex diseases, such as cancer, comprise both common ancestral risk variants of intermediate effect and rarer risk variants of higher effect/penetrance [39]. However, it is likely that in the dog, as is the case for diverse human populations, the impact of these risk variants will depend on both environmental influences and other genetic risk factors that an individual possesses. In this study we have identified a common risk variant of intermediate effect that we have shown to be reproducibly associated with MCTs in two breeds. This suggests that the common disease common variant hypothesis for human complex disease also holds true in the dog, although this may vary between breeds. It will ultimately be informative to genotype all subsequently identified Labrador and Golden Retriever MCT susceptibility variants in low risk breeds to assist understanding of the contribution of the ‘interaction’ between susceptibility loci to the elevated risk of MCT development.
For some time, synonymous variants, such as SNP rs850678541, were known as silent, as it was thought that they had no effect on gene expression and cellular fitness. Genome sequencing led to the realisation that synonymous codons do not appear with the same frequency in a genome (a phenomenon known as codon usage bias) and challenged this concept [30]. Consequently, it is now acknowledged that synonymous variants can influence cellular functions through effects on mRNA stability and processing, translation kinetics and protein folding [40]. Interestingly, Vedula and co-workers have shown that the diverse functions of β and ϒ actin homologues are defined by synonymous variants in their nucleotide sequences, and consequent differences in their translation and post-translational modifications dynamics, demonstrating that synonymous variants are important factors in the regulation of the functional diversity of protein isoforms in a variety of physiological conditions [41]. With regards to medical conditions, synonymous mutations have been associated with complex diseases such as neurological disorders, diabetes and cancer [42]. In a study in which 3,000 tumour exomes and 300 tumour genomes were analysed it was estimated that 1 in 5–1 in 2 silent mutations were positively selected, and acted as driver mutations in human cancers [43]. With regard to canine MCTs, the Golden Retriever MCT-associated variant SNP rs851590509 identified by Arendt and co-workers is also of a synonymous nature [22]. In this case, the synonymous variant is located in a splicing site, and was shown to have an effect on splicing [22]. By contrast, in the present study the synonymous SNP that we have shown to be associated with MCTs in Labrador and Golden Retrievers (SNP rs850678541) appears to have an effect on the translation dynamics of the DSCAM gene.
Translation dynamics are affected by the decoding times of each of the codons present in a transcript [44]. The decoding time of each codon is a function of parameters such as the overall codon landscape in the transcript, and is also positively correlated with abundance of the cognate tRNA [45]. Transfer RNA abundance varies between different tissues [46], and is positively correlated with the frequency with which the codon that is cognate to a tRNA is used in genes that are ‘highly expressed’ in a given tissue [47]. Therefore, a synonymous mutation can conceivably lead to an increased decoding time and impaired translation of a transcript in a given tissue if it results in a rarer codon than the ‘wild type’. Indeed, Kirchner and co-workers identified a synonymous SNP in the cystic fibrosis transmembrane conductance regulator gene, which resulted in a rare codon, which had a low-frequency cognate tRNA, and decreased protein expression in bronchial tissue. Remarkably, they showed that increasing the abundance of the tRNA cognate to the mutated codon rescued the protein expression phenotype associated with the synonymous SNP [48]. We were unable to measure, in our MCT biopsies and skin specimens, the abundance of the tRNAs cognate to the ‘reference’ (CGC) codon and alternative (CGU) arginine codon generated by the synonymous variant. This is because tRNA microarrays are unable to differentiate between these two arginine isoacceptors, and the partial hydrolysis which is used to overcome the challenges imposed by tRNA secondary and tertiary structures to build a next generation sequencing library, makes it impossible to differentiate (and quantify the relative abundances of) the tRNAs by sequencing [49]. Furthermore, a sequence (canine or human) for the arginine ‘reference codon’ cognate tRNA is not available in the tRNA database [50], which made the design of primers for a RT-qPCR assay impossible. Therefore, unfortunately, we were unable to mechanistically correlate the reduced levels of the DSCAM protein that we observed with the synonymous SNP identified, although we are hopeful that future advances in tRNA analysis techniques will enable us to so do. Nevertheless, the fact that the ‘reference’ CGC codon is nearly three times as frequent as the rs850678541 alternative allele-containing CGU codon in a sample of 1,194 canine mRNA transcripts (Kazusa database [51]; S7 Table) is an indication that the synonymous variant that we identified might be capable of having a negative effect on the DSCAM gene translation dynamics. Interestingly, 10-fold differences between the translation efficiencies of arginine codons have been demonstrated in plant chloroplasts where there was parity in codon usage [52].
The DSCAM gene was first characterised as encoding a cell adhesion molecule; a member of the immunoglobulin superfamily of cell surface proteins, in a study which identified it as a Down syndrome-related gene [53]. It has an important function in nervous system development, and its conservation in arthropods and mammals reflects its role in neural circuitry formation and an innate-immunity function, specific to arthropods [54, 55]. DSCAM has also been identified as a predisposing locus for Hirschsprung’s disease that is often observed in association with Down syndrome [56]. SNPs in the DSCAM gene have also been associated with idiopathic scoliosis in adolescents [57] and with anxiety and depression disorder [58]. Although a germline SNP in the DSCAM gene has been found to be associated with shortened overall survival in response to chemotherapy in patients with non-small cell lung cancer [59], and somatic mutations in this gene have been found in approximately 40 different types of tumour ([60]; S8 Table), to the best of our knowledge, this is the first report of an association between a germline variant in the DSCAM gene and susceptibility to cancer.
It is likely to be significant that the development of MCTs in two susceptible canine breeds has now been associated with germline variants in genes involved in cell-to-cell or cell-to-extracellular matrix (ECM) interactions. MCT development has been associated with a variant (SNP rs851590509) in the gene of a G-protein subunit (GNAI2), which acts as regulator of different transmembrane signalling pathways in a study of European Golden Retrievers; and with variants located in a region containing genes that encode hyaluronidase, an enzyme which cleaves a component of the MCT ECM, in a study of US Golden Retrievers [22]. Strikingly, in a study of US Labrador Retrievers, MCT development was found to be associated with a variant suspected to be located in a gene encoding a subunit of integrin, a cell adhesion and signalling molecule [23]. Our finding that MCT development is associated with a variant in the DSCAM gene in European pet Labrador and Golden Retrievers is additional evidence that alterations in the interaction of mast cells with the microenvironment is an important step in MCT tumorigenesis. More specifically, in the case of Labrador Retrievers, it provides compelling evidence that alterations in cell adhesion molecules represent an important risk factor for MCT development. Indeed, it has been shown that cell adhesion molecules, such as E-Cadherin, and the Ig superfamily member CADM1, can act as tumour suppressors mainly through contact inhibition of cell proliferation [61–64]. For dogs affected by MCTs, there was a trend for those whose MCTs displayed a reduced expression level of SynCAM, a cell adhesion molecule of the immunoglobulin superfamily, to be more likely to suffer MCT-related death [65]. In this study we found no association between SNP rs850678541 alternative allele and MCT metastasis in Labrador Retrievers. However, the sample set was of limited size and the thirteen dogs for whom definitive confirmation of ‘MCT metastatic disease status’ was achievable did not include dogs homozygous for the SNP rs850678541 alternative (variant) allele, a significant exclusion given that the ten-fold reduction in the level of DSCAM protein expression was only observed in MCTs and skin biopsies from dogs which are homozygous for the SNP rs850678541 alternative allele. Consequently, a much larger investigation featuring dogs with all three SNP rs850678541 genotypes, and affected by both metastasising and non-metastasising MCTs, is merited. Illustration of the likely importance of dysregulation of cell adhesion in human mastocytosis is the fact that the pathway activated by the c-kit receptor, which is frequently found somatically mutated in human mastocytosis, also regulates mast cell adhesion, in addition to survival and other cellular processes [66, 67].
In conclusion, the results presented here demonstrate the importance of retaining synonymous variants as possible functional candidates when screening for germline susceptibility loci for complex diseases, such as cancer. In addition, through identifying a common genetic risk factor for MCT development in Labrador and Golden Retrievers, the contribution of dysregulation of cell adhesion to MCT pathogenesis has been demonstrated.
The blood samples and buccal swabs used in the study were collected, retained and used for research with the written consent of the dogs’ owners. Buccal swabs were collected by dogs’ owners, and blood samples were collected by clinicians with the consent of dogs’ owners. Blood samples from UK dogs were surplus to that collected for a clinical reason, or as part of a health check. MCT biopsies were dissected (with the consent of dogs’ owners) from MCTs which were surgically removed in the course of standard treatment protocols. Biopsies of normal skin were excised post-mortem from dogs whose bodies had been donated for research by their owners. The research study, and the protocol by which samples were collected for the study, were approved by the ethics committees of the participating institutions: AHT Clinical Ethics Committee, project number AHT_07–11; Committee for Animal Care at the Massachusetts Institute of Technology, approval number MIT CAC 0910-074-13; Uppsala Animal Ethical Board, approval number C2-12; Animal Experiments Committee of the Academic Biomedical Centre, Utrecht, The Netherlands, experimental protocol ID 2007.111.08.110.
Buccal swabs and blood samples were collected from Labrador and Golden Retrievers confirmed by histopathology to have/have had a MCT, and Labrador and Golden Retrievers aged at least 7 years old whom had never been affected by any form of cancer. For GWAS, Labrador Retriever samples were collected by the Animal Health Trust in the UK (153 samples), the Broad Institute in the United States (108 samples), and the University of Utrecht in the Netherlands (77 samples) (S9 Table). For genotyping of candidate germline MCT susceptibility variants, 407 Labrador Retriever samples were collected by the Animal Health Trust in the UK (S9 Table). All Golden Retriever, Border Collie and Cavalier King Charles Spaniel samples were collected by the Animal Health Trust in the UK. Genomic DNA was isolated from buccal swabs by phenol-chloroform extraction [68], and from whole blood using the Nucleon Genomic DNA Extraction Kit (Tepnel Life Sciences), or the QIAamp DNA Blood Midi Kit (Qiagen).
This protocol is available on the protocols.io database (dx.doi.org/10.17504/protocols.io.sq2edye). Seventeen RNAlater (ThermoFisher Scientific)-preserved MCT (Biopsies#1–17 in Fig 2) and three post-mortem normal skin biopsies (Biopsies#18–20 in Fig 2), in the form of 3mm cubes, were homogenised in 700μl of Qiazol (Qiagen) by shaking with 2 x 7mm stainless steel beads at 30Hz in a TissueLyser LT (Qiagen) for 10 min at room temperature. Chloroform (140μl) was added to each homogenate and the aqueous phase recovered following centrifugation (12,000 x g for 15 min at 4°C) was used for RNA extraction with the miRNeasy Mini Kit (Qiagen), following the manufacturer’s instructions.
The interphase and organic phase were used for DNA and protein extraction. Briefly, DNA was precipitated with 100% (v/v) ethanol, and washed successively in 0.1M sodium citrate and 75% (v/v) ethanol before being resuspended in 8mM sodium hydroxide. Following DNA precipitation, protein was precipitated from the interphase and organic phase with 100% (v/v) isopropanol, washed successively in 0.3M guanidine-hydrochloride in 95% ethanol, and 75% (v/v) ethanol, and resuspended in 10M urea, 1% (v/v) 2-mercaptoethanol.
Genotyping was performed at the Centre National De Genotypage, Paris, France. Genomic DNA (200ng at 100ng/μl) was genotyped using the Infinium HD Ultra Assay (Illumina) and the canineHD array (Illumina), which comprises 173,662 SNPs spanning the canine genome at a density of around 70 SNPs per Mb [69]. GWAS datasets were analysed individually by country and genotyping run before meta-analysis to preserve data quality and reduce possible biases caused by different sample preparation procedures in different laboratories, and possible population effects between countries (case-control sets did not all approximate to a 1:1 ratio). The number of cases and controls in each individual dataset following sample quality control (QC) filtering (dropping individuals with a SNP call rate of < 90%) are shown in S1 Table. SNP QC filtering was conducted in each of the individual datasets independently. SNPs that had a minor allele frequency (MAF) of <5% and/or call rate of <97% in each dataset were excluded.
Within each dataset we visually assessed the extent of population substructure using multidimensional scaling plots in two dimensions, and by calculating genomic inflation factors, which were estimated for each dataset independently from the median of the Χ2 tests of all SNPs tested following QC (S1 Table). From examination of the multidimensional scaling plot for the “Set 1” dataset it was apparent that there were two distinct clusters of dogs within this dataset there were 28 MCT cases and 20 controls that were Guiding Eye for the Blind Dogs (S6 Fig). As these dogs originate from a line of Labradors distinct from the general pet population we postulated that they could be potential confounders in the GWAS analyses. We therefore excluded Guiding Eye for the Blind Dogs from this dataset and from future analyses. Unadjusted GWAS analyses were conducted using PLINK [70] and analyses correcting for population stratification were performed using GEMMA [71].
Genome-wide meta-analyses were conducted using SNPs that had passed QC within two or more individual datasets (S2 Fig), and for the population-adjusted meta-analysis (S3 Fig) using only SNPs in common in all six datasets.
Genomic regions implicated by GWAS as containing MCT susceptibility loci were captured from DNA samples from affected and unaffected Labrador Retrievers using SureSelect Target Enrichment System RNA oligonucleotide baits (Agilent) from libraries prepared using the TruSeq DNA Sample Preparation Kit (Illumina index set A; Illumina). Enriched libraries were sequenced using a HiSeq 2000 (100bp paired-end sequencing, approximately 30-fold coverage) (Illumina). A Genome Analysis Toolkit (GATK)-based pipeline [72] was employed to align Fastq file format sequence reads to the CanFam3.1 reference Boxer genome and detect SNVs and indels. The potential functional impact of each variant was predicted using Variant Effect Predictor [73] and SIFT [25]. A locus harbouring one or more allelic variants was considered to be a candidate MCT susceptibility locus, and selected for further analysis, if it fulfilled both of the following criteria:
All CFA31 candidate MCT susceptibility variants were typed in Labrador Retrievers, and SNP rs850678541 was typed in Golden Retrievers, Border Collies and Cavalier King Charles Spaniels. For SNPs, TaqMan Genotyping Assays (ThermoFisher Scientific) were designed (S8 Table) from variant-containing genomic DNA sequences in which known SNPs, repeat sequences, and stretches of sequence displaying significant similarity to other regions of the genome were masked. TaqMan Genotyping Assays were 10μl reactions performed using 1μl of genomic DNA, according to the manufacturer’s instructions. The TaqMan Genotyping Master Mix (ThermoFisher Scientific) was used routinely, but the TaqPath ProAmp Master Mix (ThermoFisher Scientific) was employed when there was an indication of PCR inhibition. Thermocycling was performed in a StepOne Plus Machine (ThermoFisher Scientific), and the results analysed using TaqMan Genotyper Software (ThermoFisher Scientific). Every genotyping run featured DNA samples of known genotype as positive controls, and two non-template negative controls.
The indel variant at CFA31:34667505 was genotyped through DNA fragment analysis. Amplification and fluorescent end-labelling of target fragments was achieved using 10μl PCR reactions containing 2μl of 1:100 diluted genomic DNA sample, 0.2μM of each of a forward FAM-labelled forward primer and an unlabelled reverse primer (S9 Table), 4 x 0.2mM dNTPs and 0.25 units of HotStarTaq DNA Polymerase (Qiagen). Thermocycling was performed in a T100 Thermal Cycler (BioRad) using the following parameters: 95°C, 15 min; (94°C, 30s; 60°C, 60s; 72°C, 30s) x 35; 72°C, 10 min. A microlitre of each labelled PCR product was mixed with 10μl of HiDi formamide (ThermoFisher Scientific) and 0.4μl of the ABI GeneScan 400HD ROX size standard (ThermoFisher Scientific) and loaded into an ABI 3130xl Genetic Analyser machine, using POP_7 polymer (ThermoFisher Scientific) as the separation matrix. The resulting data were analysed using the ABI GeneMapper software (ThermoFisher Scientific). Every genotyping run featured positive and negative controls.
The Golden Retriever MCT-associated SNP rs851590509, identified by Arendt and colleagues [22], was genotyped in a set of Labrador and Golden Retrievers by TaqMan assay (Thermofisher Scientific) (S8 Table). Genotyping was performed using 10μl reactions, incorporating 1μl of genomic DNA and the TaqPath ProAmp Master Mix (ThermoFisher Scientific), according to the manufacturer’s instructions. Thermocycling was performed in a StepOne Plus Machine (ThermoFisher Scientific), and the results analysed using TaqMan Genotyper Software (ThermoFisher Scientific). Every genotyping run featured two non-template negative controls, and a positive control (a sample of known genotype).
Variant-harbouring loci in LD with the SNP rs850678541 were identified using genotypes derived from the resequencing data obtained for 12 Labrador Retrievers (six cases and six controls). Haplotype analysis of biallelic loci was performed using Haploview, version 4.2 [74]. The software’s “Tagger” function [75], with a r2 threshold of 0.8, was used to identify biallelic variants in LD with the rs850678541 variant. The identification of multiallelic loci in LD with SNP rs850678541 was performed in two steps. The first step involved the identification of loci for which one allele had a frequency = the frequency of the SNP rs850678541 alternative (variant) allele ± 20%. In the second step, the genotypes at the loci selected in step one were compared to the genotype of the SNP rs850678541 locus, in order to identify those that displayed ≤1 segregation event from this locus.
MCT RNAs with RIN values ≥8.0 (Agilent Bioanalyser RNA 6000 Nano Kit; Agilent) were treated with 1.5U/μg RNA of heparinase I (Sigma-Aldrich) in 5mM Tris-HCl (pH 7.5), 1mM CaCl2 at 25°C for 3h in order to eliminate heparin, a reverse transcription and PCR inhibitor commonly found in mast cells [31]. cDNA was prepared from 2.44μg of heparinase- treated RNA, using the High Capacity RNA to cDNA kit (ThermoFisher Scientific), following the manufacturer’s instructions. Each MCT cDNA sample was assessed for the presence of PCR (and potentially reverse transcription) inhibitors by adding an equal amount of a synthetic Solanum tuberosum-derived amplicon to each sample, and screening for differences between the synthetic amplicon quantification cycle (Cq) value obtained for each cDNA sample upon PCR amplification [76]. PCR reactions (10μl), comprising 1μl of cDNA, 1 x SsoAdvanced SYBR Green Master Mix (BioRad), 1.33fM of SPUD amplicon (S9 Table), and 0.3μM of forward and reverse SPUD primers (S9 Table), were run in an ABI StepOne Plus machine (ThermoFisher Scientific) using the following program: 98°C, 2min; (98°C, 5s; 60°C, 30s) x 40; Melt Curve program. Triplicate PCR assays were performed for each MCT cDNA sample and a mean Cq value calculated.
DSCAM mRNA expression was assayed using 10μl PCR reactions, comprising 1μl of cDNA, 1 x PowerUp SYBR Green Master Mix (ThermoFisher Scientific) and 0.3μM of forward and reverse DSCAM primers (S9 Table), run in an ABI StepOne Plus machine (ThermoFisher Scientific) with the following parameters: 50°C, 2 min; 95°C, 2 min; (95°C, 3s; 60°C, 30s) x 40; Melt Curve program. Triplicate PCR assays were performed for each MCT cDNA sample. To enable normalisation of DSCAM expression values, the expression level of a 70bp fragment of a SINE [77] that occurs in the 3’-untranslated region of hundreds of canine mRNAs, in each MCT RNA sample was also assayed (performing triplicate reactions for each cDNA sample). A repeat sequence that is present in hundreds of copies in any canine tissue sample transcriptome will effectively display invariant expression across all samples of a given tissue type ensuring reliable normalisation of RT-qPCR-derived gene expression data [78]. The SINE PCR reaction master mix was subject to UV irradiation (302nm) for 5 min prior to the addition of the SINE PCR primers (S9 Table), but the PCR reaction components and thermocycling parameters were as used for the DSCAM mRNA assays. For each MCT cDNA sample, a mean Cq value was determined for each PCR amplicon from the Cq values obtained for the triplicate PCR reactions by the StepOne Plus software (Thermofisher Scientific). The mean DSCAM Cq value for each MCT cDNA sample was imported into qbase+ (Biogazelle), which generated a relative measure of DSCAM expression (a calibrated normalised relative quantity) for each MCT cDNA sample, using the mean SINE Cq value obtained for the same cDNA sample [79].
Nested end point PCR assays were performed to screen for possible alternative splicing of DSCAM mRNAs. For the E14-15 Assay, first round PCR reactions (20μl) featured 1μl of cDNA, 1 x HotStar HiFidelity PCR buffer (Qiagen), 1μM of both the ‘external’ (A) forward and reverse ‘external’ (A) primers (S9 Table), and 1 unit of HotStar Taq HiFidelity DNA Polymerase (Qiagen). Thermocycling was performed as follows: 95°C, 5 min; (94°C, 15s; 52°C, 60s; 72°C, 1 min) x 40; 72°C, 10 min,. A 1μl aliquot of a 1 : 100 dilution of each first round PCR product was used in a 20μl second round PCR reaction comprising 1 x HotStar HiFidelity PCR buffer (Qiagen), 1μM of both the ‘internal’ (B) forward and reverse primers (S9 Table), and 1 unit of HotStar Taq HiFidelity DNA Polymerase (Qiagen). The thermocycling parameters were 95°C, 5 min; (94°C, 15s; 51°C, 60s; 72°C, 60s) x 30; 72°C, 10 min. For the E15-17 Assay, first round PCR reactions (20μl) featured 1μl of cDNA, 1 x HotStar Taq PCR buffer (Qiagen), 1 x Q-Solution (Qiagen), 0.5μM of both the ‘external’ (A) forward and reverse primers (S9 Table), 0.3mM of each of 4 x dNTPs (Qiagen), 2 units of HotStar Taq DNA Polymerase (Qiagen), and 0.08 units of HotStar HiFidelity Taq DNA Polymerase (Qiagen). Thermocycling was performed as follows: 95°C, 2 min; (94°C, 10s; 56°C, 60s; 68°C, 6 min 30s) x 40. A 1μl aliquot of a 1 : 100 dilution of each first round PCR product was used in a 20μl second round PCR reaction comprising 1 x HotStar Taq PCR buffer (Qiagen), 1 x Q-Solution (Qiagen), 0.5μM of both the ‘internal’ (B) forward and reverse primers (S9 Table), 0.3mM of each of 4 x dNTPs (Qiagen), 2 units of HotStar Taq DNA Polymerase (Qiagen), and 0.08 units of HotStar HiFidelity Taq DNA Polymerase (Qiagen). Thermocycling was performed as follows: 95°C, 2 min; (94°C, 10s; 59°C, 60s; 68°C, 6 min 30s) x 40. Second round PCR reaction products were analysed by 2% agarose gel electrophoresis (E14-15 Assay) and by 0.8% agarose gel electrophoresis (E15-17 Assay), respectively. Images were captured using the Alpha Imager (Alpha Innotech).
Protein samples were quantified using the Bradford Assay. Prior to polyacrylamide gel electrophoresis, protein samples were mixed with 4 x NuPage loading buffer (ThermoFisher Scientific), incubated at 70°C for 10 min, and on ice for 5 min. Twenty-five micrograms of each protein sample and 10μl of the Precision Plus Western C protein standard (BioRad) were loaded onto a TGX Stain-Free 4–20% gradient gel (BioRad) and electrophoresed at 200kV for 40 min in 1 x Tris-Glycine SDS PAGE Buffer (National Diagnostics). A single protein sample was included on every gel for use as an inter-western blot calibrator. Prior to transfer of proteins to a membrane, a gel was exposed to 365nm UV for 2.5–5 min in order to activate the Stain-Free technology. Proteins were transferred from the gel to a 0.45μm nitrocellulose membrane (BioRad) in a Mini Trans-Blot Cell system, at 100kV for 1 hour in Tris-Glycine transfer buffer, containing 20% (v/v) methanol.
The Stain-Free total protein image of a protein blot was detected under UV light using the Alpha Imager (Alpha Innotech). The membrane was subsequently agitated in Ponceau S solution (Sigma Aldrich) for 1 min, washed 3 x with MilliQ water and visualised under white reflective light in the Alpha Imager (Alpha Innotech). Ponceau S stain was removed by 3 x washes in MilliQ water, and a membrane gently agitated in Blocking Solution (WesternBreeze Chromogenic Western Blot Immunodetection Kit; ThermoFisher Scientific) for 30 min, and incubated in a 1: 1000 dilution of anti-DSCAM antibody (abcamab85362, which has a highly conserved human DSCAM protein sequence as epitope) in Blocking solution at 4°C overnight. The membrane was washed 3 x (5 min each) with the Antibody Wash Solution (WesternBreeze Chromogenic Western Blot Immunodetection Kit; ThermoFisher Scientific), incubated with a 1 : 1000 dilution of alkaline phosphatase-conjugated Goat anti-Rabbit IgG (H+L) secondary antibody (ThermoFisher Scientific) in Blocking Solution for 30 min, washed 3 x (5 min each) with the Antibody Wash Solution, and finally incubated with BCIP/NBT Chromogenic substrate (WesternBreeze Chromogenic Western Blot Immunodetection Kit; ThermoFisher Scientific) for 1–5 min. Images were captured, under reflective white light, on the Alpha Imager (Alpha Innotech). The total protein and DSCAM staining membrane images were imported into the ImageLab software (BioRad) for analysis and quantification. Normalisation of the DSCAM level in each sample involved reference to the total quantity of protein detected (by Ponceau S or the Stain-Free technology) in the sample, and inter-membrane calibration using the ratio of DSCAM protein quantity/total protein quantity measured for the 25μg protein sample loaded onto every gel.
|
10.1371/journal.ppat.1005041 | Decline of FoxP3+ Regulatory CD4 T Cells in Peripheral Blood of Children Heavily Exposed to Malaria | FoxP3+ regulatory CD4 T cells (Tregs) help to maintain the delicate balance between pathogen-specific immunity and immune-mediated pathology. Prior studies suggest that Tregs are induced by P. falciparum both in vivo and in vitro; however, the factors influencing Treg homeostasis during acute and chronic infections, and their role in malaria immunopathogenesis, remain unclear. We assessed the frequency and phenotype of Tregs in well-characterized cohorts of children residing in a region of high malaria endemicity in Uganda. We found that both the frequency and absolute numbers of FoxP3+ Tregs in peripheral blood declined markedly with increasing prior malaria incidence. Longitudinal measurements confirmed that this decline occurred only among highly malaria-exposed children. The decline of Tregs from peripheral blood was accompanied by reduced in vitro induction of Tregs by parasite antigen and decreased expression of TNFR2 on Tregs among children who had intense prior exposure to malaria. While Treg frequencies were not associated with protection from malaria, there was a trend toward reduced risk of symptomatic malaria once infected with P. falciparum among children with lower Treg frequencies. These data demonstrate that chronic malaria exposure results in altered Treg homeostasis, which may impact the development of antimalarial immunity in naturally exposed populations.
| In malaria endemic regions, immunity is slow to develop and does not provide substantial protection against reinfection. Rather, following repeated exposure, older children and adults eventually develop protection from most symptomatic manifestations of the infection. This may be due in part to the induction of immunoregulatory mechanisms by the P. falciparum parasite, such as FoxP3+ regulatory T cells (Tregs). Prior human studies have shown that Tregs are induced by malaria parasites both in vivo and in vitro, but the role of these cells in immunity in children who are chronically exposed to malaria remains unclear. In this study, we assessed the frequency and features of Tregs among children from areas of high malaria transmission in Uganda. We found that this regulatory T cell population declined markedly with increasing malaria episodes. This loss was associated with decreased expression of TNFR2, which is a protein implicated in stability of Tregs. Additionally, T cells from highly malaria exposed children demonstrated a reduced propensity to differentiate into Tregs following parasite stimulation. Together our data suggest that repeated episodes of malaria alter Treg homeostasis, which may influence the development of immunity to malaria in children.
| FoxP3+ regulatory CD4 T cells (Tregs) play a central role in preventing autoimmunity and maintaining self-tolerance. In the setting of infection, Tregs help to maintain the delicate balance between pathogen-specific immunity and immune-mediated pathology. Preserving this equilibrium requires a complicated balance between regulatory and effector T cell activity. For instance, in the murine leishmania model, Treg-mediated suppression of effector immune responses interferes with complete parasite clearance—but paradoxically, the resulting pathogen persistence fosters the long-term maintenance of effector immune responses that are required for protection from reinfection [1,2]. Given their central role in immunoregulation, the timing, magnitude, and duration of Treg activity must be fine-tuned for promote resolution of the effector immune response only after control of the pathogen has been achieved. Malaria, like many other parasite infections, has been reported to induce an expansion of the Treg population [3]. However, the factors governing Treg homeostasis in the setting of P. falciparum infection, which in high transmission regions is characterized by both recurrent symptomatic episodes in young children and persistent asymptomatic infection in older individuals, remain unclear, as does the role of Tregs in the immunopathogenesis of malaria.
P. falciparum infection in humans induces multiple immunoregulatory pathways that likely evolved to protect the host from severe malaria by down-modulating the acute inflammatory response, perhaps at the cost of interfering with clearance of parasitemia and development of immunologic memory. Several lines of evidence suggest that Tregs are induced during human P. falciparum infection and play a role in modulating the host response. Following experimental P. falciparum sporozoite infection of naïve human subjects, FOXP3 mRNA is upregulated and peripheral blood CD25+CD4+ T cells expand [4]. In rural Gambia, the percentage and absolute count of FoxP3+CD127low CD4 T cells were shown to increase following the malaria transmission season, and are significantly higher among malaria-exposed rural Gambians than among ethnically matched urban Gambians with no malaria exposure [5]. Moreover, a number of studies have shown that peripheral Treg frequencies correlate with parasite burden in infected individuals [6–8]. Together these data suggest that Tregs are induced by P. falciparum infection in vivo. This conclusion is further supported by in vitro studies demonstrating that FoxP3+ Tregs are induced by co-culture of PBMC with P. falciparum-infected red blood cells or parasite schizont extracts [9–13].
Induction of Tregs by parasite antigens may have implications for the development of a host-protective immune response. FOXP3 mRNA levels in children with acute malaria have been shown to correlate inversely with the magnitude of the subsequent Th1 memory response to P. falciparum measured 28 days after infection [6]. Similarly, FOXP3 expression among malaria-naive adults following experimental sporozoite vaccination correlates inversely with the subsequent Th1 memory response [14]. It is possible that P. falciparum induction of Tregs may contribute to the failure of the adaptive immune response to mediate parasite clearance, as has been demonstrated in other parasitic infections such as leishmania and filariasis [1,2,15]. However, the role of Tregs in protection or risk from symptomatic malaria remains unclear. High frequencies of CD25high T cells (putatively regulatory T cells) were associated with increased risk of malaria in one prospective cohort study [16]. Consistent with this, among previously naïve adults experimentally infected with malaria, Treg induction was associated with increased parasite replication rates [4]. Further, a recent study in children and adults in Indonesian Papua found a trend towards lower proportions of activated Tregs in individuals who had asymptomatic infection compared to symptomatic malaria or healthy controls, suggesting dampened activation of Tregs may be associated with decreased risk of disease [17]. However, it has also been suggested that Tregs may serve a protective role in preventing immunopathology during infection [18,19]. Murine studies have failed to provide clear resolution of this issue, as different models have yielded conflicting data. Early reports described enhanced control of parasitemia and improved survival in mice experimentally depleted of Tregs [20], but subsequent studies that used more precise definitions of Tregs, different depletion regimens, or different parasite strains have failed to demonstrate a consistent host-protective role (summarized in [19]).
To better understand the role of Tregs in the immunopathogenesis of malaria in the setting of chronic exposure, we assessed the frequencies and phenotypic features of Tregs in Ugandan children of varying ages and malaria exposure histories. Our results indicate that while Treg frequencies are expanded in a high compared to low transmission settings, in high transmission settings children with repeated malaria infection experience a marked and progressive decline in peripheral blood Tregs, accompanied by reduced in vitro induction of Tregs by parasite antigen and decreased expression of TNFR2. This loss of circulating Tregs may have implications for the development of protective immunity to malaria, and suggests that chronic antigen stimulation, such as that observed in areas of chronic Plasmodium infection, may result in pathogen-driven alteration of Treg homeostasis.
To investigate the relationship between Treg frequencies and prior malaria exposure, we measured peripheral blood Treg frequencies in 2 separate cohorts of children in the high malaria transmission region of Tororo District, Uganda (annual entomological inoculation rate (aEIR) 310 bites ppy [21]). In both cohorts, participants were followed prospectively from enrollment at approximately 6 months of age, with comprehensive documentation of all malaria episodes at a dedicated study clinic, and at the time of analysis were either 2 years old (PROMOTE cohort, no chemoprevention control arm, n = 82) or 4 years old (TCC cohort, n = 75) (S1 Table). Treg frequencies were measured as the percentage of CD4 T cells that were FoxP3+CD25+ (for gating strategy see S1 Fig). Within both the 2-year-old and 4-year-old cohorts, there was a strong inverse relationship between Treg frequencies and prior malaria incidence (Spearman’s r = -0.27, p = 0.01, and r = -0.28, p = 0.01 respectively; Fig 1A and 1B). This inverse relationship was strengthened by further gating on the CD127dim subset, which more stringently defines suppressive Tregs (Spearman’s r = -0.36, p = 0.001; assessed in 4-year-old cohort only, for gating strategy see S1B Fig). The frequency of Tregs among children who had asymptomatic P. falciparum infection at the time of assessment (determined by blood smear) did not differ from that of uninfected children (Wilcoxon ranksum p = 0.951). Furthermore, there was no relationship between the frequency of Tregs and the duration of time since the last malaria episode, which might be expected if Tregs transiently increase in response to acute malaria (Spearman’s r = 0.094, p = 0.422), similar to what has been shown in malaria-naïve adults [4]. We measured CD4 T cell responses to P. falciparum-infected red blood cells from blood samples obtained concurrently (TCC cohort, n = 56), but we observed no statistical relationship between the frequency of Tregs and other effector or regulatory T cell populations, including cells producing IFNγ (p = 0.65), TNFα (p = 0.17), or the recently described IL10-producing “self-regulatory” CD4 T cells (p = 0.99) [22–25].
The cross-sectional data above are consistent with either a malaria-driven decline in peripheral Treg frequencies or an increased susceptibility to symptomatic malaria among children whose Treg frequencies are inherently low. To distinguish between these possibilities, we measured Treg frequencies in a third cohort of children residing in the same high transmission Nagongera, Tororo District (PRISM cohort, age 1 to 11 years, n = 91 [21]), in whom mosquito exposure was directly measured using CDC light traps within the homes of individual cohort participants [26]. In this cohort, we observed an inverse relationship between Treg frequencies and mean daily household mosquito exposure, consistent with a parasite-driven decline in Tregs (Spearman’s rho = -0.265, p = 0.014, Fig 1C). In contrast to the younger cohorts of children described above, we did not observe an inverse correlation between Tregs and the incidence of prior clinical malaria in this cohort (Spearman’s rho = 0.043, p = 0.685), likely because older children do not develop symptomatic clinical malaria with each P. falciparum infection, and thus malaria incidence is not a good measure of total P. falciparum exposure beyond early childhood. There was, however, a strong inverse relationship between Tregs and age (Spearman’s rho = -0.385, p = 0.0002; Fig 1D), suggesting that Tregs progressively decline with age in this high endemnicity setting. This decline was not attributable to age-related changes in total lymphocyte counts, as a similar relationship was observed when absolute numbers of Tregs (per μl of blood) were calculated by normalization to absolute CD4 cell counts in a subset of children (r = -0.424, p = 0.025; n = 28, S2 Fig). To investigate whether the age-related decline in Treg frequencies was unique to this high malaria transmission setting, we compared Treg frequencies among children age 1.5 to 11 years who were enrolled in the observational malaria cohort (PRISM), but at the low transmission Jinja District (aEIR 2.8 bites ppy [21]). Among children at the lower transmission site, Treg frequencies did not decline with age (r = -0.05, p = 0.83; n = 34; Fig 1E). Together these data suggest that exposure to malaria parasites may contribute to a loss of peripheral blood Tregs in this high transmission setting.
To investigate whether changes in Treg frequencies within individual subjects over time correlates with their malaria incidence, we measured Treg frequencies longitudinally in 41 subjects at 16, 24 and 36 months of age (PROMOTE cohort, SP chemoprevention arm). Subjects were stratified into tertiles based on their number of malaria infections between 16 and 36 months. Among children in the highest tertile of malaria incidence (n = 14), there was a consistent decline in Treg frequencies between 16 and 36 months of age (Wilcoxon signed rank test, p = 0.009, Fig 2A). In contrast, among children in the lowest tertile of incidence (n = 13), there was no change in Treg frequencies between 16 and 36 months (Wilcoxon signed rank test p = 0.54). Repeated-measures analysis using generalized estimating equations confirmed that changes in Treg frequencies over time differed between the exposure groups (p = 0.0011, Fig 2B). Together, these data suggest that very high malaria exposure during childhood results in the loss of peripherally circulating Tregs within individuals over time.
Prior studies have shown that experimental and natural P. falciparum infection induces the expansion of regulatory T cells in vivo [4,5]. To investigate whether and how Treg dynamics differ between settings of high and low exposure, we compared Treg frequencies between children over a range of ages in the high transmission district of Tororo to children from the low transmission district of Jinja (PRISM cohort age 1 to 11 years). Children in the high transmission Tororo District experienced a much higher malaria incidence (median 3.6 vs. 0 ppy) and a much shorter duration since last infection (median 62 vs. 230 days) than children in the low transmission Jinja District (full details in S1 Table). Overall, Treg frequencies (FoxP3+CD25+CD127dim) were higher in children from the high transmission district compared to the lower transmission district across all age groups (Wilcoxon ranksum p<0.0001, Fig 3A), and this difference was most marked in the youngest age group (Fig 3B), possibly reflecting an earlier expansion of Tregs in response to initial infections during early childhood or even in utero [27–30]. The difference in Tregs frequencies between the low and high transmission study sites decreased with increasing age, and this trend extended to adulthood (Fig 3B). Thus, our data suggest that in areas of high malaria transmission malaria infections early in life induce Tregs, as has been previously described among naïve or comparatively low-exposure individuals [4,5,7,8]. However, in areas of intense and continual malaria exposure, parasite-driven induction of Tregs is diminished, and instead there is a progressive decline of Tregs with repeated malaria episodes. This decline does not appear to be transient, as Treg frequencies do not correlate with the duration of time since last malaria episode or asymptomatic parasite infection. Instead, there appears to be sustained and progressive loss of Tregs with age among children heavily exposed to malaria.
We next assessed expression of TNFR2 on Treg cells, as this receptor has been shown to be critical for both proliferative expansion of Tregs and maintenance of FOXP3 expression in inflammatory environments [31,32]. Furthermore, Tregs expressing TNFR2 have been shown to have enhanced suppressive capacity [8,33,34], and are increased during malaria infection [8]. We found that the percentage of Tregs expressing TNFR2 was significantly lower among PRISM cohort children from the high transmission Tororo District than among children of similar age from the low transmission Jinja district (p<0.0001, Fig 4A, see S3 Fig for gating strategy). Among Tororo children, expression of TNFR2 was inversely correlated with number of recent malaria episodes (Coef = -0.31, p = 0.032, Fig 4B), although expression was slightly higher on Tregs from children currently PCR-positive for P. falciparum infection (p = 0.043). These data suggest that TNFR2 expression is transiently up-regulated during parasitemia but declines over time following repeated malaria episodes. This decrease in TNFR2 expression could contribute to the loss of FoxP3+ Tregs from peripheral blood by decreasing the stability of FOXP3 expression [31,32].
The progressive decline in circulating Tregs in children heavily exposed to malaria could be explained by changes in Treg homeostasis such as decreased induction, increased loss (due to apoptosis or downregulation of FOXP3), or both. Therefore, we next examined whether heavy prior malaria infection altered the propensity for Treg induction or apoptosis. It has previously been shown that in vitro stimulation of adult PBMCs with P. falciparum antigen induces regulatory T cells [9,10,12,35]. To investigate whether the propensity of CD4 T cells to differentiate into Tregs in response to parasite antigen is influenced by age and/or prior malaria exposure, we measured induction of Tregs following in vitro stimulation with P. falciparum schizont extracts (PfSE) in malaria-naïve adults, malaria-exposed children (28 months of age), and malaria-exposed adults from the high incidence district of Tororo (gating strategy and ex vivo Treg frequencies shown in S4 Fig). As previously reported, co-culture of PBMCs from naïve adults with PfSE resulted in consistent induction of Tregs (Fig 5A and S4C Fig). However, using PBMC from malaria-exposed children and adults, induction of Tregs was reduced compared to naïve adults (Fig 5A). Further, whereas all children with low prior malaria exposure (<2 episodes ppy) exhibited Treg induction (fold change >1), only 55% of children with high prior malaria exposure (>6 episodes ppy) induced Tregs following PfSE stimulation (p = 0.03; Fig 5B). This suggests that heavy prior exposure to malaria may limit the propensity of CD4 cells to differentiate into Tregs upon re-encounter with parasite antigens.
We next investigated whether heavy prior malaria exposure increased the susceptibility of Tregs to apoptosis, as has been shown in chronic HIV-1 infection [36]. The percentage of Tregs expressing the pro-apoptotic marker CD95 increased with age among Tororo children (Rho = 0.175, p = 0.079), as did the level of CD95 expression (as calculated by MFI of CD95 on CD95+ Tregs, Spearman’s Rho = 0.44, p = 0.0001) (Fig 5C). Conversely, expression of the anti-apoptotic marker Bcl2 on Tregs declined with age (Rho = -0.266 p = 0.035) (Fig 5D). However there was no independent relationship between expression of these markers and prior malaria incidence, current parasite infection, nor time since last malaria episode, suggesting that age may independently affect the sensitivity of Tregs to apoptosis. To further investigate this, three distinct measures of apoptosis (YoPro, Annexin V and activated Caspase 3) were measured on Tregs both ex vivo and following stimulation with camptothecin (an activator of apoptosis) or PfSE in 28-month infants with low or high prior malaria incidence (PROMOTE no-chemoprevention control arm). There was no difference in sensitivity to apoptosis as measured by any of the markers either ex vivo or following stimulation with camptothecin or parasite antigen; the frequencies of positively stained cells ex vivo, and the fold change of apoptosis staining, were the same regardless of prior malaria exposure (Fig 5E and S5 Fig). Together these data suggest that Treg homeostasis may be altered in the setting of heavy malaria exposure, in part due to reduced induction of peripheral Treg cells, with little evidence for increased susceptibility to antigen-driven apoptosis.
We finally asked whether the frequency of circulating Tregs influences susceptibility to malaria. We assessed the influence of Treg frequencies on protection from malaria in both the 2-year-old PROMOTE no chemoprevention control arm and 4-year-old TCC cohorts using two methods; a time-to-event analysis (time to next malaria episode), and negative binomial regression of the relationship of Treg frequencies to malaria incidence in the year following assessment. Among 2-year-olds, we found that higher Treg frequencies were associated with an increased time to next malaria episode and a lower future malaria incidence in univariate analysis (Table 1). However, after adjusting for prior malaria incidence in a multivariate model to account for heterogeneity in environmental exposure to infected mosquitoes [22,37,38], this relationship was no longer significant, suggesting that differences in environmental exposure intensity may underlie this association [37]. Among 4-year-olds, Treg frequencies were not associated with time to next malaria infection or malaria incidence during follow-up. Similarly, no relationships between Treg frequencies and malaria incidence in follow-up or time to next malaria episode were observed in the PRISM 1–11 year old cohorts, in either the low or the high transmission study sites, even after adjustment for household mosquito exposure. Thus we did not find clear evidence that Tregs are associated with the risk of clinical malaria.
Given their immunoregulatory role, is also possible that Tregs play a role in protecting the host from the symptomatic manifestations of malaria once P. falciparum infection is established [1,2,15,19]. To assess whether Tregs may influence the risk of clinical disease once infected, we analyzed the relationship between Treg frequencies and the probability of symptoms once parasitemic using generalized estimate equations with robust standard errors, accounting for repeated measures [16,39]. Comparing children with the lowest compared to highest tertiles of Tregs, lower Treg frequencies were associated with an increased monthly probability of infection, consistent with the exposure induced decline in Tregs described. However, lower Treg frequencies were also associated with an overall decreased probability of becoming symptomatic once infected (2 year old PROMOTE cohort; OR = 0.4, p = 0.039, 4 year old TCC cohort; OR = OR = 0.37, p = 0.06), suggesting that the decline in circulating Tregs may be associated with the acquisition of clinical immunity.
Here, we have shown through both cross-sectional and longitudinal studies that the percentage and absolute number of FoxP3+ Tregs in peripheral blood are influenced by repeated exposure to malaria. While children in settings of intense exposure have higher Treg frequencies during early childhood, frequencies decline throughout childhood in settings of high (but not low) exposure, and the extent of Treg loss correlates with the intensity of P. falciparum exposure. We provide both in vivo and in vitro evidence that among children in high exposure settings, there is a reduction of parasite induced Treg expansion during infection. Further, we show a down-regulation of TNFR2, which is required for stabilization of the FoxP3+ regulatory phenotype in inflammatory environments [31]. These data demonstrate that chronic exposure to malaria results in altered Treg homeostasis in vivo, which may have a downstream impact on the acquisition of immunity and control of infection.
Our data indicate that the dynamics of Treg induction and homeostasis differ markedly between high and low malaria transmission settings. Although children residing in high transmission areas had higher Treg frequencies overall, perhaps in response to a parasite-driven Treg expansion in early childhood or in utero [27–30], we observed a marked decline in Treg frequencies beginning at 1–2 years of age, which appeared to be driven by persistent parasite exposure. Further, among highly exposed children, we saw no association between Treg frequencies and current or recent infection, suggesting that Tregs have reduced in vivo induction during infection of these chronically exposed children. Consistent with this, we demonstrated that induction of Tregs following parasite stimulation of PBMC was diminished in heavily exposed adults and children, providing in vitro evidence that chronic antigen exposure may blunt the proliferative expansion of Tregs in response to malaria. This is in contrast to published studies suggesting that Tregs expand in response to malaria in vivo and in vitro [4,7–13,18,19]. The most likely explanation for the difference in our findings is that these earlier studies were performed largely on malaria-naïve volunteers or relatively low-exposed populations. Overall our data suggest that while Tregs may be induced in initial encounters with parasites, induction capacity is diminished after repeated parasite exposure and instead Tregs undergo a steady decline in the periphery. The induction of Tregs by Plasmodium is believe to occur through activation of latent membrane-bound TGFβ [11,40], which can be blocked by antibodies to the P. falciparum thrombospondin-related adhesive protein (PfTRAP) [11]. This Treg induction mechanism is shared by related protozoal pathogens Toxoplasma and Leishmania [35] and may represent an immune evasion strategy. The reduced capacity of parasite antigen to induce Tregs in heavily malaria exposed children suggests that the host may be able to circumvent parasite induction of Tregs, potentially enabling enhanced control of infection.
Another potential mechanism for the observed decline in Tregs among children chronically exposed to malaria is via loss of FOXP3 expression by “unstable” Tregs, which has been reported to occur in highly inflammatory immune environments [41–45]. Sustained expression of the canonical transcription factor FOXP3 by Tregs is critical for maintenance of regulatory function [46]. Several recent studies suggest that Tregs can become “unstable” and lose FOXP3 expression in response to cues in the microenvironment, although the significance, extent, and triggers of this phenomenon remain subject to considerable debate [44,45,47–52]. Lineage tracking experiments have elegantly shown that antigen-driven activation and inflammation can drive a subset of FoxP3hi Tregs to lose both FOXP3 expression and suppressor function [44], and even acquire an effector phenotype [53]. Further, repeated TCR stimulation leads to the loss of FOXP3 expression and the conversion to pro-inflammatory cytokine producing cells in natural Tregs in vitro [54]. Our data suggest a potential mechanism for such Treg destabilization in malaria infection, as recurrent exposure resulted in down-regulated Treg expression of TNFR2, which has been shown to be critical for both the proliferative expansion of Tregs and stabilization of their FOXP3 expression in inflammatory environments [31,32,55]. In the setting of malaria, TNFR2+ Tregs have previously been shown to have higher FOXP3 expression and enhanced suppressive function [8]. Thus our data are consistent with mounting evidence suggesting that peripherally induced Tregs have significant plasticity in response to inflammatory environments such as that observed in malaria infection, which may culminate in loss of FOXP3 expression and suppressive function.
Several additional processes might contribute to the observed loss of Tregs in peripheral blood. Because Tregs track to sites of inflammation, it is possible that Tregs induced by P. falciparum traffic to the liver, spleen, or secondary lymphoid organs during infection. Invasive sampling was not possible in our study cohorts; therefore we were unable to exclude a preferential sequestration of Tregs in tissues or lymphoid organs. However, we did not observe any statistical relationship of Treg frequencies with the presence of parasitemia, nor with the amount of time elapsed since the last P. falciparum infection, as might be expected if Tregs migrate to sites of local inflammation during active infection. Alternatively, loss of Tregs through apoptosis might contribute to their decline following repeated malaria infections. However, we observed no relationship between prior malaria incidence and the expression of the pro-apoptotic molecule CD95 or the anti-apoptotic marker Bcl2 on Tregs (after controlling for age), nor did we observe a differential susceptibility towards apoptosis ex vivo, or following in vitro re-stimulation with parasite antigens.
The observed decline in Treg frequencies with increasing prior malaria incidence contrasts with that of another regulatory T lymphocyte population, IL10-producing Th1 cells, which we have recently shown to dominate the P. falciparum-specific CD4 T cell response among heavily malaria-exposed children, including among children from both the TCC and the PRISM Nagongera, Tororo cohorts tested here [22,56]. This “autoregulatory” population consists predominantly of IL10/IFNγ co-producing cells that express the canonical Th1 transcription factor Tbet and appear to be short-lived in the periphery, exhibiting a strong association with recent infection [22]. We observed no statistical relationship between frequencies of IL10-producing Th1 cells and conventional FoxP3+ Tregs, in contrast to an earlier small cohort study that reported a positive correlation between these two regulatory cell populations [57]. This suggests that in highly exposed children, the loss of peripherally circulating Tregs is not directly compensated by increased frequencies of IL10 producing CD4 responses. In addition, we did not observe any statistical relationships between Treg frequencies and P. falciparum-specific CD4 effector responses. In prior studies, FOXP3 mRNA levels measured during acute malaria were shown to correlate inversely with the magnitude of the subsequent Th1 memory response to P. falciparum measured 28 days after infection [6]. Similarly, FOXP3 expression among malaria-naive adults following experimental sporozoite vaccination was shown to correlate inversely with the Th1 memory response measured >100 days later [14]. Thus, while Tregs are likely to influence the development of parasite-specific T cell memory responses, no relationship between these populations could be demonstrated through our concurrent measurements in peripheral blood, which maybe due in part to the chronicity of malaria exposure in these children and/or the substantial heterogeneity within the cohort with regard to time elapsed since the last infection. Furthermore, additional parameters of Treg function that cannot readily be measured in peripheral blood in large cohorts, such as suppression of T cell proliferation and modulation of APC function, are likely to influence the cellular immune response to malaria, but could not be assessed in the present study.
While our results clearly suggest that repeated malaria impacts peripherally circulating Tregs in children, the role of these cells in protection from malaria and the development of immunity remains unclear. We observed no association between Treg frequencies and future malaria incidence or time to next malaria episodes in any of our cohorts. However, our data suggest that although children with the lowest Treg frequencies had a higher monthly probability of infection, they were less likely to become symptomatic once infected compared to children with the highest Treg frequencies over the entire study period. While these data suggest that clinical immunity is acquired as Tregs decline, the role of Tregs in mediating clinical immunity remains unclear, and may not be causal—rather, declining Treg frequencies may coincide with other immune changes that mediate protection. Because all children in our study cohorts have easy access to dedicated study clinics and prompt antimalarial drug treatment, the incidence of severe malaria was extremely low, preventing assessment of the potential role of Tregs in protection from severe disease. We were similarly unable to assess the impact of Treg activity on pathogen persistence following infection, because all cases of symptomatic malaria were promptly treated with potent artemisinin-based drugs, thus altering the natural course of infection. In other protozoal infections, such as leishmaniasis and toxoplasmosis, pathogen-induced Tregs have been reported to curb the inflammatory response, allowing long-term pathogen persistence [1,2,15]. Indeed, in murine models of leishmania, pathogen persistence resulting from Treg-mediated immune suppression has been shown to be a requirement for immunity to re-infection [1]. The long-term asymptomatic maintenance of low-burden P. falciparum infection that is commonly observed among adults in high-transmission areas [58] may represent a similar phenomenon, but the role of Tregs in mediating this process is not known. Although we did not observe higher frequencies of peripheral blood Tregs among children with asymptomatic P. falciparum infection, which is not routinely treated in Uganda, this does not exclude a role for Tregs in maintaining this state of host-parasite equilibrium.
In conclusion, we observed a progressive loss of Tregs from the peripheral blood of children following chronic repeated malaria infections, accompanied by downregulation of TNFR2 and diminished in vitro induction of Tregs by parasite antigen. Together these data demonstrate that the impact of chronic malaria antigen exposure on the FoxP3+ regulatory T cell population is quite different from that of acute infection of malaria-naïve individuals. Our findings also add to mounting data suggesting that the stability and homeostasis of FoxP3+ Tregs are perturbed under highly inflammatory conditions. The implications of this pathogen-driven Treg loss for pathogen clearance, host-parasite equilibrium, and the development of clinical immunity in regions of intense malaria transmission require further investigation.
Written informed consent was obtained from the adult individual or parent/guardian of all study participants. Study protocols were approved by the Uganda National Council of Science and Technology and the institutional review boards of the University of California, San Francisco, Makerere University and the Centers of Disease Control and Prevention.
Samples for this study were obtained from children enrolled in 3 longitudinal childhood malaria cohort studies conducted in Tororo District and Jinja District of eastern Uganda. Cohort characteristics are described in S1 Table. For all cohorts, samples were selected on the bases of availability of PBMCs.
For the PROMOTE-Chemoprevention, TCC and PRISM Nagongera high transmission area, the estimated entomological inoculation rate (aEIR) is approximately 310 bites ppy. In contrast, at the PRISM Walukuba low transmissions site the aEIR is estimated at 2.8 [21].
On enrollment all study participants were given an insecticide treated bed net and followed for all medical care at dedicated study clinics. Children who presented with a fever (tympanic temperature ≥38.0°C) or history of fever in the previous 24 hours had blood obtained by finger prick for a thick smear. If the thick smear was positive for malaria parasites, the patient was diagnosed with malaria regardless of parasite density and treatment with artemether-lumefantrine or dihydroartemisinin-piperaquine for all episodes of malaria. Incident episodes of malaria were defined as all febrile episodes accompanied by any parasitemia requiring treatment, but not preceded by another treatment in the prior 14 days. The incidence of malaria was calculated as the number of episodes per person years (ppy) from the time of enrolment into the cohort. In a subset of PRISM cohort children used to assess TNFR2 expression on Tregs parasite infection was assessed via PCR from dried blood spots as previously described [61].
Treg frequencies were enumerated from whole blood and fresh and cryopreserved PBMCs as indicated below. For enumeration of Tregs from whole blood (PROMOTE-Chemoprevention, control arm 2-year-old samples), 100 μl of fresh whole blood was stained with BD Pharmingen anti-CD3-FITC (UCHT1), anti-CD4-PE-CY7 (SK3), and CD25-APC (M-A251) for 20 minutes and then lysed and permeabilized with eBioscience RBC lysis buffer. Cells were washed and then incubated with eBioscience anti-FoxP3-PE (PCH101). Samples were acquired on Accuri C6 Cytometer.
For analysis of Tregs from fresh PBMCs (PRISM Nagongera cohort), PBMCs were isolated by Ficoll density gradient centrifugation and rested over night in 10% fetal bovine serum. PBMCs were stained with BD Pharmingen anti-CD3 PerCP (SK7), anti-TNFR2-Alexa646 (hTNFR-M1), anti-CD95-PECy7 (DX2) and Biolegend anti-CD4-APC-Cy7 (OKT4), anti-CD25-BrillantViolet510 (M-A251), anti-CD127-PacificBlue (A019D5). Following surface staining, cells were fixed and permeabilized with eBioscience FoxP3 staining set and intracellular stained with FoxP3-PE (PCH101) and BD Pharmingen anti-Bcl2-FITC (Bcl-2/100) as per manufacturers protocol. Samples were acquired on three laser BD FACsCantoII with FACSDiva software.
For analysis of Tregs from frozen PBMCs (Tororo Child Cohort 4-year-olds, PROMOTE-Chemoprevention SP arm longitudinal samples at 16, 24, and 28 months of age, PRISM Walukuba cohort), cryopreserved PBMCs were thawed using standard methods, and immediately stained with the following panels of antibodies; BD Pharmingen anti-CD3-FITC (UCHT1), anti-CD4-PE-CY7 (SK3), CD25-APC (M-A251) and Biolegend anti-CD127 Pacific Blue (A019D5); or Biolegend anti-CD3-BrilliantViolet650 (OKT3), anti-CD4-PerCP (OKT4), anti-CD127-FITC (A019D5); or Biolegend anti-CD3-PerCP (OKT3), anti-CD4-APC-Cy7 (OKT4), anti-CD25-BrillantViolet510 (M-A251), anti-CD127-PacificBlue (A019D5), anti-TNFR2-APC (3G7A02). Live/dead aqua amine (Invitrogen) was included in all panels. Following surface staining, cells were fixed and permeabilised with eBioscience FoxP3 staining set and intracellular stained with FoxP3-PE (PCH101) as per manufacturers protocol. Samples were acquired on LSR2 three laser flow cytometer (Becton Dickinson) with FACSDiva software.
For calculation of absolute Tregs counts (i.e. cells per μl, PRISM Nagongera cohort), peripheral blood CD4 T cell concentrations were measured from whole blood using counting beads, and Treg frequencies were calculated by normalization to total CD4 T cell numbers.
Analysis of CD4+ T cell responses to P. falciparum infected RBCs via intracellular cytokine staining was performed as previously described [22,56]. PBMCs were stimulated with P. falciparum infected RBCs or uninfected RBCs and CD4 T cell production of IFNγ, IL10, and TNFα were measured via intracellular staining.
PBMCs from PROMOTE subjects (28 months of age; no chemoprevention control arm) and adults from the high malaria transmission region of Tororo were thawed and washed in 10% Human serum (AB) media (Gemini), and 3–6X106 PBMC were labeled with 1 ml of 1.25 mM 5,6-carboxyfluorescein diacetate succinimidyl ester (CFSE; Molecular Probes) for seven minutes. CFSE-labeled PBMC were incubated in 96-well, deep-well culture plates (Nunc, Roskilde, Denmark) at 106 PBMC/ well in 1 ml for 7 days with P. falciparum schizont extract (PfSE) (W2 strain) or protein extract from uninfected RBCs (uE) at a effector to target ratio equivalent to 1:1 PBMC:infected RBC. PHA (1μg/ml) was used as a positive control. PfSE extracts were made from the W2 stain grown in standard culture conditions and confirmed to be free of mycoplasma contamination using MycoAlert (Lonza). Mature stage parasites were magnet purified from culture using MACs purification columns. Purified parasites or uninfected RBCs were freeze thawed 3+ times (via snap freezing on liquid nitrogen and then transfer to 37°C water bath) to produce PfSE and uE and stored at -20°C. Following culture of PBMCs with protein extracts, cells were treated with 100 units of DNase I (Invitrogen) in culture media for 5 minutes and then surface stained with Biolegend anti-CD3-BrilliantViolet650 (OKT3), anti-CD4-PerCP (OKT4), anti-CD25-PE-Cy7 (BC96), anti-CD127 Pacific Blue (A019D5), BD Pharmingen anti-CD8-ABC-H7 (SK1) and Live/dead aqua amine (Invitrogen cells). Following surface staining, cells were fixed and permeabilized with eBioscience FoxP3 staining set and stained with FoxP3-PE (PCH101). Proliferation with PHA was used to ensure cell viability, and cells incubated with uE were used as a background control. Of the infants tested, the prior median malaria incidence was 1.2 episodes ppy in the low exposed group and 8.5 episodes ppy in the high exposed group.
PBMCs from PROMOTE subjects (28 months of age; no chemoprevention control arm) were thawed and rested overnight either in standard media (untreated), 5uM camptothecin (Sigma) or P. falciparum schizont extract (PfSE) or protein extract from uninfected RBCs (uE) at an effector:target ratio of 3:1. To test for induction of apoptosis, stains for AnnexinV (Biolegend) or YoPro (Invitrogen), or activated Caspase 3 FITC (BD) were used according to the manufacturer’s instructions in combination with the following antibodies: AnnexinV and YoPro staining—CD3 (OKT3) Brilliant Violet 650, CD4 (RPA-T4) APC-Cy7, CD127 (A019D5) APC, CD25 (BC96) PE-Cy7 from Biolegend; for Caspase3—CD3 (OKT3) Brilliant Violet 650, CD4 (RPA-T4) PerCP, CD127 (A019D5) Pacific Blue, CD25 (BC96) PE-Cy7. Tregs were gated as CD3+CD4+CD25+CD127dim. Sensitivity to apoptosis was measured ex vivo (untreated control), after induction with camptothecin (fold change compared to untreated), and after stimulation with P. falciparum schizont extract (fold change comparing PfSE to uE).
Unless otherwise indicated, samples were acquired on an LSR2 flow cytometer (Becton Dickinson) with FACSDiva software. Flow cytometry data were analysed using FlowJo software (Tree Star, San Carlos, CA). Color compensation was performed using single color cell controls or beads stained for each fluorochrome. Gating strategies are outlined in Supplementary Figures. Fluorescence minus one controls were used for gating of CD25, CD95, HLA-DR and Bcl2. For FoxP3 staining, an anti-Rabbit-Isotype control was used.
Data analysis was performed using Stata version 12 (Stata Corp, College Station, Tx) and PRISM version 6 (Graph Pad). Associations between Treg frequencies and other continuous variables (prior malaria incidence, age, time since last malaria episode) were assessed using Spearman’s correlation. Changes in Treg frequencies within an individual over time were assessed using the Wilcoxon signed rank test. All other two-group comparisons of continuous variables were performed using the Wilcoxon rank sum test. Repeated measures analysis of longitudinal changes in Treg frequencies was performed using generalized estimating equations, with adjustment for concurrent parasitemia, age and duration since last malaria episode. Categorical variables were compared using Chi sq test. Associations between Treg frequencies and time to next malaria episode were evaluated using the Kaplan-Meier product limit formula, and a multivariate cox proportional hazards model was used to adjust for surrogates of malaria exposure (cumulative episodes since enrollment in study for TCC and PROMOTE cohorts, or age for PRISM cohorts). Negative binomial regression was used to estimate associations between Treg frequencies and the prospective incidence of malaria in the following year (incidence rate ratios, IRR) and prevalence of asymptomatic parasitemia in the following year (prevalence rate ratios, PRR), adjusting for malaria exposure as above. Two-sided p-values were calculated for all test statistics and p<0.05 was considered significant. In the PRMOTE and TCC cohorts, associations between the highest and lowest tertiles of Treg frequencies and the monthly risk of parasitemia, probability of symptoms if parasitemic, and incidence of malaria, stratified by year of age, were evaluated using generalized estimating equations with robust SEs accounting for repeated measures in the same patient, for the period of the inter study (6 months to 3 or 5 years of age) [62].
|
10.1371/journal.pntd.0003206 | Brugia malayi Microfilariae Induce a Regulatory Monocyte/Macrophage Phenotype That Suppresses Innate and Adaptive Immune Responses | Monocytes and macrophages contribute to the dysfunction of immune responses in human filariasis. During patent infection monocytes encounter microfilariae in the blood, an event that occurs in asymptomatically infected filariasis patients that are immunologically hyporeactive.
To determine whether blood microfilariae directly act on blood monocytes and in vitro generated macrophages to induce a regulatory phenotype that interferes with innate and adaptive responses.
Monocytes and in vitro generated macrophages from filaria non-endemic normal donors were stimulated in vitro with Brugia malayi microfilarial (Mf) lysate. We could show that monocytes stimulated with Mf lysate develop a defined regulatory phenotype, characterised by expression of the immunoregulatory markers IL-10 and PD-L1. Significantly, this regulatory phenotype was recapitulated in monocytes from Wuchereria bancrofti asymptomatically infected patients but not patients with pathology or endemic normals. Monocytes from non-endemic donors stimulated with Mf lysate directly inhibited CD4+ T cell proliferation and cytokine production (IFN-γ, IL-13 and IL-10). IFN-γ responses were restored by neutralising IL-10 or PD-1. Furthermore, macrophages stimulated with Mf lysate expressed high levels of IL-10 and had suppressed phagocytic abilities. Finally Mf lysate applied during the differentiation of macrophages in vitro interfered with macrophage abilities to respond to subsequent LPS stimulation in a selective manner.
Conclusively, our study demonstrates that Mf lysate stimulation of monocytes from healthy donors in vitro induces a regulatory phenotype, characterized by expression of PD-L1 and IL-10. This phenotype is directly reflected in monocytes from filarial patients with asymptomatic infection but not patients with pathology or endemic normals. We suggest that suppression of T cell functions typically seen in lymphatic filariasis is caused by microfilaria-modulated monocytes in an IL-10-dependent manner. Together with suppression of macrophage innate responses, this may contribute to the overall down-regulation of immune responses observed in asymptomatically infected patients.
| Lymphatic filariasis is a parasitic disease that affects over one million people worldwide, causing chronic morbidity in the majority of infected individuals. A certain proportion of individuals develop asymptomatic infection that allows persistence of the parasite and therefore transmission of disease through the blood-circulating microfilariae. We show that monocytes and macrophages, innate effector cells that circulate in the blood, migrate or reside in tissues and come into contact with microfilariae, can be stimulated by microfilarial (Mf) lysate in vitro to develop a regulatory phenotype via expression of PD-L1 and IL-10. Significantly, this regulatory monocyte phenotype was directly reflected in monocytes isolated from filaria asymptomatically infected patients in the absence of external stimuli. Mf lysate-modulated monocytes inhibited adaptive immune functions, some of which could be restored by neutralisation of IL-10. Mf lysate-modulated macrophages had reduced phagocytic capacity, while macrophages differentiated in the presence of Mf lysate displayed significantly inhibited innate responses to LPS stimulation. This suggests that microfilariae modulate the innate response by acting on macrophage differentiation and macrophages themselves, and modulate the adaptive CD4+ T cell response by acting on monocytes. The phenotypic pattern observed by modulated monocytes is recapitulated in vivo in active infection. This highlights a previously unclear mechanism of immune modulation by the parasite.
| Lymphatic filariasis is an immune-mediated spectral disease that manifests in two main clinical outcomes: chronic pathology or asymptomatic infection. These outcomes depend on a multitude of factors, including parasite-induced immunoregulation and host genetic background (reviewed elsewhere [1]).
An overt manifestation is the hyperresponsive phenotype that develops in patients with chronic lymphatic pathology (abbreviated as CP). These individuals have increased antigen-specific immunoglobulin (Ig)E and low IgG4 [2], [3], strong T helper (Th)1 and Th17 proinflammatory responses and a greatly diminished T regulatory (Treg) compartment [2], resulting in immunopathological changes in the host. CP patients carry adult worms in the lymphatics but are generally amicrofilaremic, as a strong immune response kills the microfilarial stage. Parasite death leads to the release of antigenic material that triggers inflammation and causes destruction of lymphatic vessels and inflammation [4]. In Wuchereria bancrofti, Brugia malayi and B. timori infections, this can result in the development of elephantiasis or hydrocoele, whereby the lymphatic tissue becomes dilated and hypertrophic.
The second clinical manifestation is a hyporesponsive phenotype characterised by asymptomatic infection (abbreviated as AS), which tolerates the presence of fecund adult worms due to strong parasite-induced immunosuppression and immunomodulation [5]. Importantly, adult worms are tolerated and circulating blood microfilariae are carried by these patients, ensuring transmission. This group has increased numbers of regulatory cells, high interleukin (IL)-10 and elevated levels of antigen-specific IgG4 leading to a modified Th2 response that protects the host and permits parasite survival [5]. Thus, parasite-induced immunomodulation allows persistent infection and continuous transmission while simultaneously enabling the host to tolerate infection by diminishing clinical symptoms. The proportion of individuals in a filaria-endemic area who do not develop one of these two clinical manifestations remain infection- and disease-free and are putatively immune; these individuals are known as endemic normals (abbreviated as EN) [1].
Filaria have been shown to act on host dendritic cells, monocytes, macrophages, T cells and B cells to cause immunomodulation, typically inducing Th2 type and regulatory responses [6]. Immunomodulation occurs through production of specific parasite-derived products that target mammalian host immune cells and signalling pathways. This is strictly dependent on live parasites as shown by the recovery of cellular responsiveness in patients treated with microfilaricidal chemotherapy, specifically diethylcarbamazine (DEC) [7]. While some adults are killed by DEC treatment, the main target is the microfilarial stage, suggesting a prominent role for microfilariae in modulating immune responses [8].
Monocytes and macrophages have long been described to develop a particular phenotype in filarial infection that may contribute to the dysfunction of adaptive immune responses. The spectrum of phenotypes and functions that monocytes and macrophages can develop is vast, and ranges from classical activation, through to alternative activation, wound healing, parasite resistance and immune regulation (reviewed elsewhere [9]). Thus classical activation is characterised by production of proinflammatory molecules and typically develops in response to stimulation with lipopolysaccharide (LPS) plus a secondary stimulus such as interferon (IFN)-γ or tumour necrosis factor (TNF)-α. In contrast alternative activation is induced by stimulation with IL-4 and/or IL-13 and induces macrophages that are involved in wound healing and immune regulation [9]. Murine macrophages that develop in response to these cytokines express specific markers, including arginase (arg)-1, resistin-like molecule (RELM)-α, Ym-1, Ym-2, acidic mammalian chitinase (AMCase), and mannose receptor C type (MRC)-1 [9]. This is a phenotype that is reflected in macrophages in filarial infections [10]. In contrast to the murine system, human monocytes and macrophages typically show a diverse phenotype after stimulation with IL-4, defined by expression of the scavenger receptor CD163 and the chemokine ligand CCL18 [11], [12]. MRC-1 (CD206), similar to mice, is also expressed by human macrophages in this setting [12]. In human filarial infections, expression of arg-1, programmed death-ligand (PD-L)1 and PD-L2 on monocytes has also been reported [13], [14].
Macrophages recruited in filarial infection induce specific hyporesponsiveness in T cells [15], [16]. Macrophages recruited during B. malayi infection drive CD4+ Th2 responses, deviating the immune system from inducing a proinflammatory Th1 response that could be detrimental to parasite survival [17]. During patent filarial infection monocytes encounter the microfilarial lifecycle stage in the blood, before migrating out to the tissues. This initial contact with the parasite is particularly interesting as it occurs only in AS patients where the adult worms are tolerated in the lymphatics and produce viable microfilariae [18].
We hypothesised that microfilariae in circulation act on human monocytes during patent infection. This primary contact may affect monocyte differentiation, macrophage development and the ensuing innate and adaptive responses and thus may contribute to the development and maintenance of asymptomatic infection. Therefore we aimed to characterise the phenotype of monocytes and macrophages stimulated in vitro with B. malayi microfilarial (Mf) lysate, and to determine the effect of B. malayi Mf lysate-stimulated monocytes and macrophages on defined innate or adaptive functions. We could show that monocytes stimulated with B. malayi Mf lysate in vitro develop a defined regulatory phenotype, characterised by expression of the immunoregulatory markers IL-10 and PD-L1. Significantly, this regulatory phenotype could be recapitulated in monocytes from Wuchereria bancrofti AS patients in contrast to CP patients and EN individuals. Monocytes from non-endemic normal donors stimulated with Mf lysate directly inhibited CD4+ T cell proliferation and cytokine production (IFN-γ, IL-13 and IL-10). Importantly, IFN-γ responses could be restored by neutralisation of IL-10 or PD-1. Furthermore, macrophages stimulated with Mf lysate expressed high levels of IL-10 and had suppressed phagocytic abilities. Finally we could show that Mf lysate applied during the differentiation process of macrophages in vitro interfered with macrophage abilities to respond to subsequent LPS stimulation in a selective manner.
All experiments with material from filaria non-endemic normal donors were approved by the ethical committee of the Charité, Berlin (permit number EA1/104/14). All experiments with material from W. bancrofti-exposed donors were approved by the ethical committee of the Blue Peter Public Health and Research Center-LEPRA Society, Hyderabad (permit number 5/2009). Informed written consent was obtained from all participants. All W. bancrofti infected donors were treated for lymphatic filariasis by administration of DEC and symptomatic relief after completion of the study. The study was performed according to the Declaration of Helsinki.
Live B. malayi microfilariae and adult female worms were a kind donation from the NIAID/NIH Filariasis Research Reagent Resource Center (FR3) in Athens, Georgia. Microfilariae and adult female worms were washed twice in RPMI medium containing 200 U/ml penicillin and 200 µg/ml streptomycin. To collect excretory/secretory (ES) products, microfilariae were cultured for 3–5 days in RPMI containing 1% glucose, 200 U/ml penicillin and 200 µg/ml streptomycin in a 5% CO2-incubator at 37°C in 6 well plates whereby media was replaced every 24 h. The resulting ES-containing media was concentrated using Vivacell 70 concentrators (Sartorius Stedim Biotech GmbH, Göttingen, Germany) with a membrane cut-off at 5,000 molecular weight. To prepare microfilarial (Mf) lysate, live microfilariae in suspension were subsequently harvested, washed twice in phosphate buffered saline (PBS) by centrifuging for 10 min at 1500 rpm. Pelleted microfilariae or adult female worms were homogenised directly in a glass homogeniser and ultrasonicated on ice at an intensity of 10% for 3 min. The homogenate was centrifuged at 12,000 rpm and 4°C for 10 min and sterile filtered through a 0.22 µm filter. Protein concentration was determined using the Pierce BCA protein assay kit (Thermo Scientific, Waltham, USA) as per the manufacturer's guidelines. LPS concentration was determined by Limulus amoebocyte lysate endotoxin detection kit QCL-1000 (Lonza, Walkersville, USA); the LPS content in B. malayi Mf lysate, ES or adult female (Fem) lysate used in all assays was <1 EU/ml in the final concentration.
In vitro experiments using samples from filaria non-endemic normal donors were performed in Germany, using buffy coats purchased from the German Red Cross. Experiments using samples from filaria-exposed donors examined a cohort of 56 individuals from Andhra Pradesh in South India, where lymphatic filariasis caused by W. bancrofti is endemic. Night blood smears were performed with 20 µl blood to detect circulating microfilariae, and the TropBio Og4C3 ELISA (TropBio Pty. Ltd, Townsville, Queensland, Australia) was performed using serum to detect circulating filarial antigen (CFA), as per the manufacturer's instructions. Patients with lymphatic filarial pathology (lymphadenitis, lymphoedema, hydrocoele) were examined as part of a clinical protocol approved by the institutional ethical committee of the Blue Peter Public Health and Research Center-LEPRA Society. Based on these results, 28 individuals were classed as endemic normals (EN), 21 had chronic pathology (CP) and 7 had asymptomatic infection (AS) (Table 1). Any AS patient found to be positive for CFA was classed as asymptomatic regardless of the night blood smear result; at 20 µl blood per smear, the test has low sensitivity, giving a cut off value of 50 microfilariae per ml.
For in vitro experiments using samples from non-endemic normal donors, peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats. For experiments using samples from filaria-exposed donors, PBMCs were isolated from 40 ml blood. In both cases PBMCs were isolated by density centrifugation using Lymphocyte Separation Medium (LSM). Blood from buffy coats was initially diluted in phosphate buffered saline (PBS), two parts blood to one part PBS. Briefly, blood was layered onto LSM and centrifuged at 2500 rpm, at room temperature for 25 min with zero brake and zero accelerator. The interphase was collected and washed in PBS plus 0.2% bovine serum albumin (BSA). Cells were centrifuged at 1500 rpm, 4°C for 10 min, washed in PBS plus 0.2% BSA and centrifuged at 1000 rpm, 4°C for 10 min to remove platelets. Lysis of erythrocytes was performed to remove remaining erythrocytes if necessary. For this, 5 ml of ammonium-chloride-potassium lysis buffer (0.01 M KHCO3, 0.155 M NH4Cl, 0.1 mM EDTA, pH 7.5) was added to cells for 5 min at room temperature, after which cells were washed in PBS plus 0.2% BSA and centrifuged at 1500 rpm for 10 min.
To positively select for CD14+ monocytes, anti-CD14 beads (Miltenyi, Biotec, Bergisch-Gladbach, Germany) were added for 20 min to PBMCs. A volume of 200 µl beads was used on buffy coats, while 50 µl was used on 40 ml whole blood. In both cases, cells were washed in PBS plus 0.2% BSA, 2 mM EDTA and filtered before separation using a 70 µm filter (Partek, St. Louis, USA). Cells from non-endemic normal donors were separated by an autoMACS classic, using the program ‘possel’ (Miltenyi Biotec, Bergisch-Gladbach, Germany). Cells from filaria-exposed donors were separated using MACS MS columns as per the manufacturer's instructions (Miltenyi Biotec, Bergisch-Gladbach, Germany). After isolation, cells were washed once in PBS plus 0.2% BSA, 2 mM EDTA, centrifuged at 1500 rpm and 4°C for 5 min and transferred into complete RPMI medium (RPMI 1640, 5% AB human serum, 100 U/ml penicillin, 100 mg/ml streptomycin, 1 mM L-glutamine, 1 mM MEM non-essential amino acids, 1 mM sodium pyruvate) for use in subsequent experiments. Macrophages were generated in vitro by culturing CD14+ monocytes in complete RPMI plus 10 ng/ml M-CSF in 6-well cell culture plates at a cell concentration of 0.33×106/ml and a density of 0.1×106/cm2 for 6 days, at 37°C and 5% CO2. For macrophages differentiated in the presence of Mf lysate, 20 µg/ml B. malayi Mf lysate was added to the supernatant at the beginning of culture.
Monocytes from filaria-exposed donors were used directly for ex vivo RT-PCR analysis or seeded at a concentration of 0.2×106 per well into 96-well flat bottom cell culture plates and stimulated for 24 h in vitro with 200 µl stimulus. Monocytes from non-endemic normal donors were seeded at a concentration of 2×106 per well into 24-well cell culture plates and stimulated for 4 h or 24 h in vitro with 1 ml stimulus. In vitro generated macrophages were washed on day 6 with PBS, and the culture supernatant was replaced with 1 ml stimulus for 24 h. Cells were stimulated with 20 ng/ml IL-4 (Peprotech, Rocky Hill, USA), a combination of 20 ng/ml IFN-γ (Peprotech, Rocky Hill, USA) plus 100 ng/ml LPS (Invivogen, California, USA) or 20 µg/ml B. malayi Mf lysate at 37°C and 5% CO2. Unstimulated controls were included in all experiments. The supernatant was collected and stored at −20°C for further analysis. Cells were lysed in RNA lysis buffer (Analytik Jena, Jena, Germany) for RT-PCR analysis.
IL-6, IL-8, IL-10, IL-13, IFN-γ and TNF-α protein were measured using a commercial enzyme-linked immunosorbent assay (ELISA) kit from eBioscience (San Diego, USA). IL-12p40 was measured using an ELISA kit from BioLegend (San Diego, USA). All samples were measured in duplicates. Absorbance was read at 450 nm with background wavelength subtracted at 570 nm using the Synergy HT plate reader from BioTek (Winooski, USA).
Active transforming growth factor (TGF)-β was measured using a TGF-β reporter cell line. Briefly, MFB-F11 cells were incubated for 24 h with cell culture supernatant. The resulting supernatant was assessed using the SEAP Reporter Gene Assay (Roche, Branford, USA) according to the manufacturer's instructions.
RNA was isolated from cells using a commercial kit, following the manufacturer's instructions (innuPREP RNA mini-kit, Analytik Jena, Jena, Germany). The concentration of extracted RNA was determined using the NanoDrop 1000. RNA was reverse-transcribed to cDNA using a high capacity RNA-to-cDNA kit (Life Technologies, Darmstadt, Germany). cDNA was set to a concentration of 3–10 ng/µl, depending on the experiment. Real-time PCR was performed with FastStart Universal SYBR Green Master Mix (Roche Applied Science, Indianapolis, USA) using the ABI 7300 Real-Time PCR (Life Technologies, Darmstadt, Germany). Relative changes in gene expression were calculated with ABI 7300 SDS Software (Life Technologies, Darmstadt, Germany). For monocytes and macrophages from non-endemic normal donors, expression levels of transcripts were normalized to the Ct values of the endogenous housekeeping gene by using the 2−ΔΔCt method [19]. Relative expression of genes in stimulated samples was compared to unstimulated controls (which are set at 1). For monocytes from filarial-exposed donors we chose not to normalise data to the reference group (EN), as there was a large variation in Ct values within the heterogeneous EN group. Thus expression levels of transcripts were normalized to the Ct values of the endogenous housekeeping gene by using the method 1/2ΔCt where ΔCt represents the difference between the target gene and the housekeeping gene. Baseline expression of samples from AS or CP was compared to that of EN. In all experiments ß2-microglobulin was used as a housekeeping gene. The gene-specific primer sequences and accession numbers are shown in Table S1.
Monocytes and macrophages were analysed for surface expression of HLA-DR, CD80, CD86, PD-L1, PD-L2, CD163, CD206, indoleamine 2,3-dioxygenase (IDO) or CD11b. Cells were treated with FcR Blocking Reagent (Miltenyi Biotec, Bergisch-Gladbach, Germany) and stained with Fixable Viability Dye eFluor 780 (eBioscience, San Diego, USA), and with one or combinations of the following: anti-CD80-PE (clone 2D10.4), anti-CD86-PE (clone IT2.2), anti-CD274-PE (clone MIH1), anti-CD273-PE (clone MIH18), anti-CD163-PE (clone GHI/61), anti-IDO-PE (clone eyedio), anti-CD11b-FITC (clone M1/70) (all from eBioscience, San Diego, USA), anti-CD206-APC (clone 15.2, BioLegend, San Diego, USA) or HLA-DR-APC (clone G46-6, BD, Franklin Lakes, USA). Cells were acquired using the FACSCanto II (BD, Franklin Lakes, USA) and analysed using FlowJo, version 8.8.7 (Tree Star, Ashland USA).
PBMCs were labelled using the CD4+ T cell Isolation Kit II (Miltenyi Biotec, Bergisch-Gladbach, USA) according to the manufacturer's instructions and sorted on an autoMACS classic (Miltenyi Biotec, Bergisch-Gladbach, USA) using the program ‘deplete’. Untouched CD4+ T cells were stained with CFSE and rested in complete RPMI for 24 h at 37°C and 5% CO2.
Monocytes were left unstimulated or stimulated with 20 µg/ml B. malayi Mf lysate or 20 µg/ml B. malayi microfilarial ES for 24 h in a 5% CO2-incubator at 37°C in 6-well plates. Monocytes were washed and 1×105 monocytes were cocultured with 5×105 CFSE-labelled CD4+ T cells in 96-well flat bottom plates in the presence of 2 µg/ml soluble anti-CD3 (OKT3, eBioscience, San Diego, USA). After 3–5 days, the supernatant was removed for cytokine analysis and cells were stained with Fixable Viability Dye eFluor 780 (eBioscience, San Diego, USA) and anti-CD4-PE-Cy5 (clone RPA-T4, BioLegend, San Diego, USA). Fixed cells were acquired using the FACSCanto II (BD, Franklin Lakes, USA) and analysed using FlowJo, version 8.8.7 (Tree Star, Ashland USA).
To determine a role for IL-10 signalling or PD-1-PD-L1 interactions in the suppression of CD4+ T cell effector functions induced by Mf lysate, monocytes were left unstimulated or stimulated with 20 µg/ml B. malayi Mf lysate for 24 h in a 5% CO2-incubator at 37°C in 6-well plates. Monocytes were washed and 1×105 monocytes were cocultured with 5×105 CFSE-labelled CD4+ T cells in 96-well plates in the presence of 2 µg/ml soluble anti-CD3 (OKT3, eBioscience, San Diego, USA) plus neutralising anti-IL-10 antibodies or anti-PD-1 antibodies (both at 10 µg/ml, from eBioscience San Diego, USA). After 3–5 days, the supernatant was removed for cytokine analysis and cells were stained with Fixable Viability Dye eFluor 780 (eBioscience, San Diego, USA) and anti-CD4-PE-Cy5 (clone RPA-T4, BioLegend, San Diego, USA). Fixed cells were acquired using the FACSCanto II (BD, Franklin Lakes, USA) and analysed using FlowJo, version 8.8.7 (Tree Star, Ashland USA).
To determine the ability of Mf lysate-differentiated macrophages to respond to LPS stimulation, Mf lysate-differentiated macrophages were washed on day 6 of culture and 2×105 cells were stimulated for 24 h with 100 ng/ml LPS (Invivogen, California, USA). The following day supernatants were collected for cytokine analysis by ELISA.
Macrophages were generated in vitro as above, and used after 6 days incubation. To harvest macrophages, the cells were washed three times in PBS, then 1 ml PBS containing 5 mM EDTA was added. The cells were kept at 4°C for 10 min then scraped off the well using a cell scraper. Macrophages were seeded into a 96-well flat bottom plate (0.2×106 cells per well) in complete RPMI, and stimulated using 20 µg/ml B. malayi Mf lysate for 24 h. After 24 h macrophages were washed in PBS three times, after which the prepared fluorescent pHrodo BioParticles suspension was added. pHrodo BioParticles (Life Technologies, Darmstadt, Germany) were prepared beforehand by suspending 2 mg BioParticles in 2 ml of uptake buffer (140 mM NaCl, 2.5 mM KCl, 1.8 mM CaCl2, 1.0 mM MgCl2, 20 mM HEPES, pH 7.4). The solution was briefly vortexed and then sonicated for 5 min to ensure homogenous dispersal of the particles. 100 µl of the prepared suspension was then added to the cells. Unstimulated macrophages were regarded as a positive control. Cytochalasin D (20 µM) was added as a negative control to unstimulated cells. Cells were incubated with the fluorescent particles for 3 h at 37°C (no CO2). Finally the cells were washed three times in PBS and 200 µl of 0.5% formalin was added. Fluorescence was read at 550 nm excitation and 600 nm emission using the Synergy HT plate reader from BioTek (Winooski, USA). The net phagocytosis was calculated as per the manufacturer's instructions, by subtracting the average fluorescence intensity of the negative control from the positive control and all experimental wells. The phagocytosis response to the experimental effector (% phagocytosis) could then be calculated as a percentage of the net positive control phagocytosis (% phagocytosis = net phagocytosis×100/net phagocytosis of positive control.
All statistical analyses were performed using GraphPad Prism version 6.0d (GraphPad Software, Inc., California, USA). In experiments where two paired groups were analysed, Wilcoxon signed-rank test was used to compare a condition to its unstimulated control. In experiments where more than two paired groups were analysed, Friedman's ANOVA was used to determine whether a statistically significant difference existed between any of the conditions and the unstimulated control (significance level p<0.05). In case of significance, the main analysis was followed up with a Wilcoxon signed-rank test between a condition and the unstimulated control, whereby a Bonferroni correction was applied. The Kruskal-Wallis test with Dunn's multiple comparisons post-test was used to determine statistical significance between multiple unpaired groups.
To understand if B. malayi microfilariae act on monocytes to induce immune modulation, monocytes isolated from buffy coats from filaria non-endemic normal donors were stimulated for 24 h in vitro with B. malayi Mf lysate, LPS plus IFN-γ or IL-4 (Fig. 1). Monocytes stimulated with B. malayi Mf lysate produced significant and high levels of IL-10, IL-6, TNF-α and IL-8 while IL-12p40 was not induced (Fig. 1A). IL-27 was not detected (data not shown). Control stimulation with LPS plus IFN-γ resulted in significantly elevated levels of IL-10, IL-6, TNF-α, IL-8 as well as IL-12p40, which was in stark contrast to Mf lysate-stimulated monocytes. On the other hand, control stimulation with IL-4 significantly inhibited protein production of IL-6 and IL-8 compared to unstimulated controls, while levels of IL-10, TNF-α and IL-12p40 were not significantly altered from unstimulated controls (Fig. 1A). Using a TGF-β reporter cell line, the levels of active TGF-β were measured (Fig. 1B). Stimulation of monocytes with B. malayi Mf lysate, LPS plus IFN-γ or IL-4 did not alter TGF-β production compared to unstimulated controls.
To assess the expression in monocytes of markers associated with an alternative/regulatory phenotype, mRNA expression of MRC-1 (NM_002438), CCL18 (NM_002988), PD-L1 (NM_014143) and PD-L2 (NM_025239) was analysed after 24 h stimulation (Fig. 1C). B. malayi Mf lysate induced significantly higher levels of CCL18 and PD-L1 compared to unstimulated controls, while the other markers were not altered. Monocytes stimulated with LPS plus IFN-γ or IL-4 upregulated expression of CCL18, PD-L1 and PD-L2 compared to unstimulated controls. MRC-1 was significantly inhibited by LPS plus IFN-γ compared to unstimulated controls.
To determine whether the high mRNA level of PD-L1 was reflected on a protein level, surface expression of PD-L1 was measured by flow cytometry. Expression of the activation markers HLA-DR, CD80 and CD86, the human monocyte alternatively activated markers PD-L2, CD163 and CD206 and the classical activation marker IDO was measured in parallel (Fig. 1D). In agreement with the PCR data, PD-L1 was significantly upregulated in monocytes in response to B. malayi Mf lysate stimulation (Fig. 1D). Significant upregulation of PD-L2 was also observed, however HLA-DR, CD80, CD86, CD163, CD206 and IDO were not significantly different from unstimulated controls. LPS plus IFN-γ induced expression of HLA-DR, PD-L1 and PD-L2, while stimulation with IL-4 upregulated CD86, PD-L1 and PD-L2 (Fig. 1D). The other markers analysed were not altered after stimulation with LPS plus IFN-γ or IL-4.
It has previously been shown that microfilariae affect the survival of dendritic cells by apoptosis [20], [21]. To determine whether B. malayi Mf lysate affected cell viability, monocytes, in vitro generated macrophages or macrophages differentiated in the presence of Mf lysate were stained with a dead cell exclusion dye (Fig. S1) and viability was assessed by flow cytometry. There was no difference in the percentage of viable cells in B. malayi Mf lysate-stimulated versus unstimulated cells.
Stimulation in vitro with B. malayi Mf lysate induced a defined regulatory phenotype of monocytes that significantly upregulated the proinflammatory markers IL-6, TNF-α and IL-8 as well as the alternative/regulatory markers IL-10 and PD-L1 in monocytes. Therefore we hypothesised that in filaria-infected individuals, monocytes from asymptomatically infected patients, who exhibit circulating microfilariae in their blood, develop a phenotype similar to that observed in our in vitro experiments. Thus, we performed RT-PCR on monocytes isolated from PBMCs from EN, CP and AS donors (Fig. 2). There was a trend for monocytes from AS patients to express elevated levels of the alternative/regulatory markers IL-10 (NM_000572), MRC-1, CCL18, PD-L1 and PD-L2 as well as the proinflammatory markers IL-6 (NM_000600), TNF-α (NM_000594) and IL-12p40 (NM_002187). IL-8 (NM_000584) was significantly downregulated in AS patients compared to EN individuals. MRC-1 was significantly elevated in AS patients compared to CP donors. In contrast, there appeared to be little difference between CP and EN although IL-10 and IL-6 were significantly downregulated in CP compared to EN donors. Thus, monocytes isolated from PBMCs from AS patients recapitulate to a great extent the expression profile observed in B. malayi Mf lysate-stimulated monocytes from filaria non-endemic donors.
Monocytes from AS patients have been shown to be functionally defective in terms of Toll-like receptor (TLR) expression and function [22], [23]. Thus we wanted to establish the capacity of monocytes from AS patients to respond to B. malayi Mf lysate (Fig. S2). After 24 h without stimulation in culture, monocytes from EN, CP and AS donors responded equally in terms of protein production of IL-10, IL-6, TNF-α and IL-12p40. After 24 h stimulation with B. malayi Mf lysate, monocytes from all three groups responded by producing equal amounts of cytokines. Hence, there was no inherent defect in the ability of monocytes from EN, CP and AS donors to produce cytokines in response to B. malayi Mf lysate.
To determine whether the activation phenotype seen in monocytes stimulated with B. malayi Mf lysate may account for the hyporesponsiveness of CD4+ T cells observed in ex vivo studies with PBMCs from AS patients [18], [24]–[27] a coculture assay with autologous CD4+ T cells was employed. To this end, autologous CFSE-labelled CD4+ T cells were polyclonally stimulated and incubated with B. malayi Mf lysate-stimulated monocytes (Fig. 3). T cell proliferation was significantly inhibited after coculture with B. malayi Mf lysate-stimulated monocytes (Fig. 3A). The production of IFN-γ, IL-13 and IL-10 was significantly inhibited in the supernatant of cocultures with B. malayi Mf lysate-stimulated monocytes compared to cocultures with unstimulated monocytes (Fig. 3B). Thus, Mf lysate-modulated monocytes showed a significantly impaired ability to stimulate CD4+ T cell functions.
In order to understand if monocytes treated with microfilaria-derived ES products could alter T cell responses we repeated the coculture but stimulated monocytes with 20 µg/ml B. malayi microfilarial ES. Microfilarial ES-stimulated monocytes did not suppress CD4+ T cell effector functions (Fig. S3).
As IL-10 and PD-L1 were significantly upregulated in B. malayi Mf lysate-stimulated monocytes, we hypothesised that one of these molecules could be involved in inhibiting CD4+ T cell functions as observed in Fig. 3. Thus we repeated the coculture experiment and included neutralising anti-IL-10 antibodies to block IL-10 signalling or anti-PD-1 antibodies to block PD-1-PD-L1 interactions (Fig. 4). Experiments using unstimulated monocytes in the coculture were performed as a control (Fig. S4). While proliferation of T cells was elevated after neutralisation of IL-10 to a statistically significant level (p = 0.016, Fig. 4A), the biological difference in restoration was minimal (44.78% of T cells proliferated in response to Mf lysate-stimulated monocytes compared with 48.59% of T cells proliferating after IL-10 was neutralised). Proliferation of T cells was not changed after neutralisation of PD-1 (Fig. 4B). IFN-γ production was restored in response to neutralisation of IL-10 (Fig. 4A) and PD-1 (Fig. 4B), while IL-13 responses were not restored.
To determine whether monocytes were actively expressing IL-10 upon the time of coculture, B. malayi Mf lysate-stimulated monocytes were assessed for IL-10 mRNA expression by RT-PCR (Fig. S5). After 24 h stimulation with B. malayi Mf lysate, there was a trend for monocytes to express IL-10 mRNA, although this did not reach statistical significance.
To understand if B. malayi microfilariae act on macrophages to induce immune modulation, monocytes isolated from buffy coats from filaria non-endemic normal donors were differentiated to macrophages in vitro and stimulated for 24 h in vitro with B. malayi Mf lysate, LPS plus IFN-γ or IL-4 (Fig. 5). Similar to monocytes (see Fig. 1A), stimulation with B. malayi Mf lysate led macrophages to produce significantly higher levels of IL-10 and IL-8 compared to unstimulated controls, while IL-6, TNF-α and IL-12p40 were not induced (Fig. 5A). IL-27 was not detected (data not shown). This is in clear contrast to stimulation with LPS plus IFN-γ, which led to significant and high expression of IL-10, IL-6, TNF-α, IL-12p40 and IL-8. IL-4 stimulation did not induce significant production of any cytokines measured (Fig. 5A). Using a TGF-β reporter cell line, the levels of active TGF-β were measured (Fig. 5B). Stimulation of macrophages with B. malayi Mf lysate, LPS plus IFN-γ or IL-4 did not alter production of TGF-β compared to unstimulated controls. On mRNA level, macrophages stimulated with B. malayi Mf lysate did not alter expression of MRC-1, CCL18, PD-L1 or PD-L2 compared to unstimulated controls (Fig. 5C). There were no significant changes in expression of these markers after stimulation of macrophages with LPS plus IFN-γ or IL-4 compared to unstimulated controls (Fig. 5C).
In terms of cell surface markers, stimulation with B. malayi Mf lysate did not alter expression of HLA-DR, CD80, CD86, PD-L1 or PD-L2 in macrophages compared to unstimulated controls (Fig. 5D). A similar result was observed in macrophages stimulated with LPS plus IFN-γ or IL-4 (Fig. 5D).
It is feasible that microfilariae interfere with the differentiation of monocytes to macrophages in vivo during patent infection, due to their shared anatomical locations. Thus we determined the phenotype and functions of macrophages generated in vitro from CD14+ monocytes in the presence of 20 µg/ml B. malayi Mf lysate (Fig. 6). Mf lysate-differentiated macrophages did not alter expression of the maturation markers HLA-DR, CD80 and CD86 or the macrophage markers CD11b and CD163 compared to macrophages generated in the absence of Mf lysate (Fig 6A). However when macrophages generated in the presence of Mf lysate were washed and stimulated with 100 ng/ml LPS, there was a significant and selective inhibition of IL-6, TNF-α and IL-12p40 but not IL-10, when compared to macrophages generated in the absence of Mf lysate (Fig. 6B).
To determine whether phagocytosis of macrophages was modulated by stimulation with B. malayi Mf lysate, in vitro generated macrophages were stimulated for 24 h with B. malayi Mf lysate. Subsequently the phagocytic capacity was determined by measuring the phagocytosis of fluorescently labelled bioparticles, whereby unstimulated cells were used as a positive control (Fig. 6C). B. malayi Mf lysate significantly inhibited phagocytosis in macrophages, reducing this function by approximately 20%.
Our data show that B. malayi Mf lysate induces a regulatory monocyte phenotype that curtails CD4+ T cell effector functions in vitro. This monocyte population is characterized by expression of IL-10 and PD-L1 as well as certain proinflammatory markers. This regulatory phenotype is reflected in monocytes from AS patients with active filarial infection, but not in CP patients or EN donors. Importantly, CD4+ T cell IFN-γ responses could be recovered after neutralisation of IL-10 or PD-1. Furthermore we could show that macrophages stimulated with Mf lysate expressed high levels of IL-10 and had suppressed phagocytic abilities. Finally, Mf lysate applied during the differentiation process of macrophages in vitro interfered with macrophage abilities to respond to subsequent LPS stimulation in a selective manner. A limitation to our results from filaria-exposed donors was the access to only very low numbers of AS donors used for analysis of ex vivo monocyte phenotype and for stimulation of filaria-endemic monocytes in vitro. Such low numbers of AS patients are in part due to the extensive mass drug treatment effort in South India during the last decade [28], [29].
The observed monocyte phenotype is in accordance with a previous study that described monocytes stimulated with live microfilariae to upregulate PD-L1 and to have an alternatively activated phenotype [14]. Monocytes isolated from filarial lysate-treated PBMCs from humans with asymptomatic filarial infection are characterised by increased expression of arg-1 and IL-10 and decreased levels of nitric oxide synthase (NOS, typically used as a classical activation marker in murine studies) compared to endemic normal controls [13]. Interestingly in our studies inducible NOS (iNOS) could not be detected in B. malayi Mf lysate-treated monocytes or macrophages in vitro or in monocytes from endemic patients ex vivo (data not shown). iNOS historically can be difficult to detect, thus the differences observed here may lie in the techniques used [30]. Furthermore murine markers cannot simply be translated into the human system [31], [32]. As an example, reports indicate that arg-1 may not be a reliable marker for alternatively activated monocytes or macrophages in humans as it is found in other cell types [33]. Human monocytes do not express arg-1 after stimulation with IL-4 and IL-13 (the prototypical inducers of alternative activation), unlike mouse macrophages [31]. Thus we have carefully selected a number of markers that have previously been reported for classical activation, alternative activation, or regulation in human monocytes and macrophages [9], [11], [12], [31], [34]. We have used these as the basis for characterising the cells in the current study.
Our results indicate a phenotype that consists of a spectrum of proinflammatory, alternatively activated and regulatory markers, which is in line with numerous reports stating the mixed Th1/Th2 response that is often seen after exposure to microfilariae [18], [20], [35]–[37]. The mixture of markers observed may be partly influenced by Wolbachia, the obligate endosymbiont found in all lifecycle stages of the filarial parasites W. bancrofti and B. malayi. Wolbachia are known to induce inflammatory responses in monocytes in a TLR4-independent manner [38]–[40]. Contrary to this, monocytes treated with lysate from adult female worms do not upregulate any of the cytokines tested. This highlights a microfilaria-derived factor other than Wolbachia to be responsible for the effects seen here with Mf lysate-treated monocytes (Fig. S6). Importantly in our studies, Mf lysate did not induce production of IL-12p40, in stark contrast to stimulation of monocytes or macrophages with LPS plus IFN-γ. Similarly there was no detection of TNF-α in Mf lysate-stimulated macrophages, which was highly induced by LPS plus IFN-γ. Monocytes showed some production of TNF-α however this was approximately five times less than that observed after stimulation with LPS plus IFN-γ. On the other hand, the proinflammatory markers as well as IL-10 were not detected in either monocytes or macrophages after stimulation with IL-4. Thus Mf lysate produces a clear and distinct response to that induced by the hallmark proinflammatory stimulus, LPS plus IFN-γ or the prototypical alternative activation stimulus, IL-4.
Nevertheless it is unclear to what extent the expression of proinflammatory cytokines may function in the face of high levels of immunoregulatory cytokines such as IL-10. We have previously shown that IL-10 is induced in murine macrophages by a nematode-derived cysteine-proteinase inhibitor isolated from a related filarial species, Acanthocheilomena viteae (AvCystatin) [41]. Thus it would be interesting to investigate the role of filarial cystatin in B. malayi Mf lysate-induced IL-10 expression, as B. malayi contains three different forms of cystatin, with Bm-CPI-2 being more closely related to AvCystatin [42]. Furthermore, AvCystatin was shown to mediate IL-10- and macrophage-dependent immunomodulation in a mouse model of airway hyperreactivity [43]. Thus, future experiments should determine the contribution of B. malayi cystatins to human T cell hyporesponsiveness.
Establishing the phenotype and function of B. malayi Mf lysate-stimulated monocytes and macrophages from non-endemic normal donors highlights these cells as instruments of microfilarial immune modulation. To elucidate the exact phenotype of monocytes during infection, it was necessary to analyse the cytokine and marker profile of monocytes from endemic individuals ex vivo, without prior stimulation in the presence of other immune cells as done in other studies [13]. Therefore we examined monocytes from individuals with W. bancrofti asymptomatic infection that had presumably interacted with microfilariae in circulation in the 12 hours prior to isolation. As expected, in the absence of any external stimulation, only monocytes from this group produced the specific phenotype that was previously observed in vitro. Nevertheless in response to specific stimulation, we found that monocytes from all three groups responded with a cytokine profile that closely reflected that seen under the same conditions in monocytes from filaria non-endemic normal controls. This observation revealed that monocytes from all filaria-exposed donors principally had the capacity to react to Mf lysate without an inherent defect in one of the patient groups.
Monocytes and macrophages that develop in helminth infections are believed to contribute to wound healing, regulation of Th1 and Th2 inflammation and expulsion of the parasite from the host (reviewed elsewhere [44]). Asymptomatically infected patients are the only filarial-exposed group in which monocytes in the blood come into contact with live microfilariae; thus monocytes may be influenced at this early time point in their differentiation to contribute to immune regulation and therefore the development of asymptomatic infection. Indeed it has been shown that B. malayi microfilariae act on monocytes from filaria non-endemic normal donors to reduce transendothelial migration [45]. To this end, we established that monocytes and macrophages from non-endemic normal donors stimulated with B. malayi Mf lysate in vitro develop a specific phenotype upon activation, and that these cells may influence either the adaptive or innate immune response, respectively.
B. malayi Mf lysate-stimulated monocytes could suppress CD4+ T cell proliferation as well as IFN-γ and IL-13 cytokine production in an autologous coculture assay. T cells that received a polyclonal stimulus in the presence of B. malayi Mf lysate-treated monocytes displayed significantly reduced proliferation compared to T cells stimulated in the presence of control monocytes. Furthermore, their ability to produce effector cytokines was significantly inhibited. This is in line with a previous report that demonstrated that PBMCs from microfilaremic patients produce fewer Th1 and Th2 cytokines when stimulated with live microfilariae compared to PBMCs from endemic normals [24].
PD-L1 and IL-10 were significantly induced by Mf lysate stimulation of monocytes and macrophages and thus represented prime candidates responsible for the suppression of T cell responses. Interestingly, neutralisation of IL-10 or PD-1 led to a recovery of CD4+ T cell IFN-γ but not IL-13 production. IL-10 has previously been described to be upregulated in adherent cells from patients with lymphatic filariasis [46] and in monocytes from patients harbouring tissue-dwelling filaria [38]. IL-10 has a well-defined role in filarial infections as an immunoregulatory cytokine that regulates both Th1- and Th2-derived inflammatory, potentially harmful responses [47]. PBMCs from asymptomatically infected patients spontaneously secrete significantly higher levels of IL-10 than PBMCs from patients with chronic pathology [46]. Induction of IL-10 is also associated with high levels of immune regulatory IgG4 in human asymptomatic filarial infection (reviewed elsewhere [4]). Thus the high levels of IL-10 observed in our study that resulted from B. malayi Mf lysate stimulation may contribute to T cell suppression in asymptomatic infection.
PD-L1 has been described on monocytes stimulated with live microfilariae in vitro [14]. PD-L1 together with its receptor PD-1 has an important role as a negative costimulator in numerous infection settings [48]. The high mRNA and surface expression of PD-L1 on B. malayi Mf lysate-stimulated monocytes support the idea that microfilaria-modulated monocytes may contribute to asymptomatic infection through this mechanism. Indeed IFN-γ responses were also restored after neutralization of PD-1, supporting a role for this molecule.
Neutralisation of IL-10 or PD-1 had only a minimal effect on the recovery of proliferation and had no effect on the restoration of IL-13, implying that other inhibitory mechanisms may be involved. Van der Werf et al. have previously shown that the PD-1-PD-L2 pathway is responsible for Th2 cell hyporesponsiveness in L. sigmodontis infection [49]. Nevertheless this pathway would have been similarly neutralised in our assays through application of the anti-PD-1 antibody, suggesting that other mechanisms play a role in the human T cell impairment observed in our experiments. Other candidates that have been described in murine literature to suppress T cell responses in helminth or other Th2-related diseases through a monocyte/macrophage interaction include arginase [50], [51] and RELM-α [52], [53], however as mentioned above, these may not represent reliable options in the human system. Further studies describe the roles of retinoic acid or TGF-β in directing the development of Treg cells [54]–[56]; whether the CD4+ T cells in our system have a regulatory phenotype should be further investigated.
In response to stimulation with B. malayi Mf lysate, macrophages upregulated high levels of IL-10, but in contrast to monocytes no significant upregulation of PD-L1 expression was detected. However when B. malayi Mf lysate was added during the differentiation process of macrophages there was a significant and selective impairment in the ability of macrophages to produce cytokines in response to LPS stimulation. Interestingly this was not caused by a reduction in cell survival or a change in cell activation markers. Similar results have been published by Semnani et al. for monocyte-derived dendritic cells stimulated with microfilarial antigen during the differentiation process [57]. Importantly, our results demonstrated that subsequent stimulation with LPS of macrophages differentiated in the presence of Mf lysate resulted in diminished IL-6, TNF-α and IL-12p40 production but not IL-10. This could be a result of distinct signalling pathways being activated in Mf lysate-differentiated macrophages and should be a subject of further studies. The selective impairment of expression of proinflammatory cytokines hints towards a possible involvement of NF-κB1 p50 homodimers, as shown previously [58].
It has been previously demonstrated that B. malayi live microfilariae or live microfilariae in a transwell do not induce phagocytosis in monocytes compared to monocytes exposed to M-CSF for 48 h [14]. The authors of this study argued that microfilariae failed to promote phagocytosis. In contrast we believe that B. malayi microfilariae (or, in our case, B. malayi Mf lysate) directly inhibit phagocytosis of macrophages as a form of immunomodulation. The location of macrophages in the tissues compared to monocytes in the blood may place macrophages in a more advantageous position to phagocytose. In another study, W. bancrofti endemic normal monocytes were incubated with serum from the different groups (endemic normal, patients with lymphatic pathology, microfilaremic patients) [23]. Only monocytes incubated with serum from microfilaremic patients had reduced levels of spreading but not phagocytosis. Thus future studies should determine in detail whether inhibition of phagocytosis actually translates to an increase in microfilarial survival, and which other cells or serum components contribute to microfilarial killing and phagocytosis in vivo.
In conclusion this study has elucidated the monocyte phenotype in patients with active filarial infection and the regulatory capacity of this cell. By directly acting on monocytes in the blood, microfilariae may regulate the antigen-specific T cell response. Furthermore microfilariae may interfere with the differentiation process of macrophages, thus altering their ability to respond to unrelated stimuli in the tissues. The extent to which these findings promote parasite survival and transmission is unclear and should be further investigated.
|
10.1371/journal.ppat.1005104 | Structural and Functional Analysis of Murine Polyomavirus Capsid Proteins Establish the Determinants of Ligand Recognition and Pathogenicity | Murine polyomavirus (MuPyV) causes tumors of various origins in newborn mice and hamsters. Infection is initiated by attachment of the virus to ganglioside receptors at the cell surface. Single amino acid exchanges in the receptor-binding pocket of the major capsid protein VP1 are known to drastically alter tumorigenicity and spread in closely related MuPyV strains. The virus represents a rare example of differential receptor recognition directly influencing viral pathogenicity, although the factors underlying these differences remain unclear. We performed structural and functional analyses of three MuPyV strains with strikingly different pathogenicities: the low-tumorigenicity strain RA, the high-pathogenicity strain PTA, and the rapidly growing, lethal laboratory isolate strain LID. Using ganglioside deficient mouse embryo fibroblasts, we show that addition of specific gangliosides restores infectability for all strains, and we uncover a complex relationship between virus attachment and infection. We identify a new infectious ganglioside receptor that carries an additional linear [α-2,8]-linked sialic acid. Crystal structures of all three strains complexed with representative oligosaccharides from the three main pathways of ganglioside biosynthesis provide the molecular basis of receptor recognition. All strains bind to a range of sialylated glycans featuring the central [α-2,3]-linked sialic acid present in the established receptors GD1a and GT1b, but the presence of additional sialic acids modulates binding. An extra [α-2,8]-linked sialic acid engages a protein pocket that is conserved among the three strains, while another, [α-2,6]-linked branching sialic acid lies near the strain-defining amino acids but can be accommodated by all strains. By comparing electron density of the oligosaccharides within the binding pockets at various concentrations, we show that the [α-2,8]-linked sialic acid increases the strength of binding. Moreover, the amino acid exchanges have subtle effects on their affinity for the validated receptor GD1a. Our results indicate that both receptor specificity and affinity influence MuPyV pathogenesis.
| Viruses are obligate intracellular pathogens, and all of them share one crucial step in their life cycle—the attachment to their host cell via cellular receptors, which are usually proteins or carbohydrates. This step is decisive for the selection of target cells and virus entry. In this study, we investigated murine polyomavirus (MuPyV), which attaches to host gangliosides with its major capsid protein, VP1. We have solved the crystal structures of VP1 in complex with previously known interaction partners as well as with the ganglioside GT1a, which we have identified as a novel functional receptor for MuPyV. Earlier studies have shown that different strains with singular amino acid exchanges in the receptor binding pocket of VP1 display altered pathogenicity and viral spread. Our investigations show that, while these exchanges do not abolish binding or significantly alter interaction modes to our investigated carbohydrates, they have subtle effects on glycan affinity. The combination of receptor specificity, abundance, and affinity reveals a much more intricate regulation of pathogenicity than previously believed. Our results exemplify how delicate changes to the receptor binding pocket of MuPyV VP1 are able to drastically alter virus behavior. This system provides a unique example to study how the first step in the life cycle of a virus can dictate its biological properties.
| The engagement of one or several host cell receptors is the first step in the infectious cycle of a virus. A large number of viruses, including many human pathogens, depend on carbohydrate recognition for initial attachment to the cell surface. Viral tropism and the internalization pathway are usually determined by the specificity and affinity of the receptor interaction as well as the glycan distribution on different cell surfaces (reviewed in [1]). Many viruses use glycoproteins, glycolipids, or both as receptors for cell entry [2]. Gangliosides are ubiquitous glycolipids on the outer leaflet of mammalian cell membranes that serve as receptors for a number of viruses. They are composed of a membrane-embedded ceramide moiety linked to a complex carbohydrate structure that projects away from the cell. Gangliosides almost always contain α-5-N-acetyl-neuraminic acid (sialic acid, Neu5Ac) that can be attached to the core of the molecule with [α-2,3], [α-2,6], or [α-2,8] linkages (Fig 1). Gangliosides exist on cell surfaces in complex and poorly understood patterns that are cell type-, age-, and tissue-dependent ([3,4], reviewed in [5]).
Murine Polyomavirus (MuPyV) is a double-stranded DNA virus that can induce tumors in newborn animals. It was long known to engage glycan receptors that contain a minimal motif of sialic acid [α-2,3]-linked to galactose [6,7], and more recently gangliosides GD1a and GT1b were identified as MuPyV receptors [8]. Viral attachment is mediated by the major capsid protein, VP1, which forms pentameric capsomers that assemble into the T = 7d icosahedral capsid of the virus [9,10]. Sialylated oligosaccharide receptors are engaged in a shallow groove on top of VP1 formed by loop structures on the protein surface [11–13], similar to other polyomaviruses [1].
MuPyV displays striking differences in pathogenicity and spread among three closely related prototype strains upon infection of newborn virus-free mice. The laboratory-derived RA strain [14] shows limited spread and induces few tumors of strictly mesenchymal origin after a long latency period, while the naturally occurring PTA strain [15,16] has disseminated infection and causes multiple tumors of epithelial and mesenchymal origin within a short time. LID [17,18], another laboratory isolate MuPyV strain, spreads most rapidly, causing early death by damaging host tissues, leading to brain hemorrhages and kidney failure [19]. The differences among the three strains have been mapped to amino acid variations at two positions, 91 and 296, within the receptor-binding region of VP1 [20–24]. While RA bears a glycine residue at position 91, this residue is replaced with a glutamate in both PTA and LID. An additional valine-to-alanine exchange at position 296 is present in LID (Table 1). The pathogenicity profile of one strain can be introduced into the other strains by mutating these two residues, confirming that these substitutions are necessary and sufficient to generate a specific phenotype [25]. The same substitutions have also been observed for other strains of MuPyV [21,22]. MuPyV found in feral mice has the VP1 sequence of PTA [26], but the virus is controlled by an intact immune system. As studies of viral spread can be conducted in vivo and virus infectivity can be tested in vitro using ganglioside deficient mouse cells, MuPyV represents an attractive and rare model system to define the relationships between receptor binding and viral spread and tropism.
Crystal structures of the low pathogenicity strain RA have shown how this virus engages 3’-sialyllactose, a short, linear trisaccharide terminating in [α-2,3]-linked sialic acid, as well as an oligosaccharide that additionally contains a second, branching [α-2,6]-linked sialic acid [11,12]. These structures also identified the location of residues 91 and 296 in the carbohydrate-binding region, suggesting that they might modulate interactions of VP1 with its receptors in the higher pathogenicity strains PTA and LID. Modelling suggested that a glutamate side chain at position 91 would lead to electrostatic repulsion of the [α-2,6]-branched sialic acid, thereby preventing binding of such branched structures by either LID or PTA. Branched sugars carrying an [α-2,6]-linked sialic acid could thus act as pseudoreceptors that will not facilitate productive infection but hamper the spread of RA within the host, in contrast to PTA and LID [8,12]. In line with this hypothesis, gangliosides GD1a and GT1b, which do not contain an [α-2,6]-branched sialic acid, have been identified as entry receptors for the PTA [8,16] and RA strain [27] of MuPyV. However, the molecular determinants of GD1a or GT1b receptor interactions with PTA and LID are not understood, because all structural information is limited to date to RA MuPyV.
To define the interactions of the three MuPyV strains with receptors on the cell surface, we have solved high-resolution structures of RA, PTA, and LID VP1 pentamers in complex with three ganglioside glycans that represent common motifs found in members of the four most prominent ganglioside biosynthesis series and that feature [α-2,3]-, [α-2,6]-, and [α-2,8]-linked sialic acids (for carbohydrate structures, nomenclature, and annotations see Fig 1). We have also conducted crystallographic soaking experiments at different ligand concentrations to compare the relative affinities of each of the three strains for their interaction partners. We find that expanding the well-characterized Neu5Ac-[α-2,3]-Gal epitope with a linear [α-2,8]-linked sialic acid (as found for example in GT1a vs. GD1a) leads to additional interactions between carbohydrate and VP1 in all three strains. Consequently, we identify ganglioside GT1a as an infectious receptor for all three strains. Moreover, the branching [α-2,6]-linked sialic acid is close to the strain-defining amino acids, but can be accommodated by all strains, in contrast to the earlier model. However, the amino acid exchanges defining each strain have subtle effects on their affinity for the validated receptor GD1a. Our results exemplify the effect of minimal changes in a binding pocket on the receptor binding properties of a virus.
Previous efforts to identify receptors for MuPyV used immortalized cell lines, such as Vero or C6 glioma cells that were supplemented with candidate gangliosides before infection [8,28]. We utilized a mouse embryo knock-out fibroblast cell line (Gang-/- MEFs) specifically deficient in ganglioside synthesis and completely resistant to MuPyV infection (S1A Fig and [29]) to test the ability of ganglioside receptors to rescue infection by different strains of MuPyV. Gang-/- MEFs were supplemented with individual gangliosides followed by infection with RA, PTA, and LID MuPyV (Fig 2). Importantly, it should be noted that the three MuPyV strains we used do not have the same particle to PFU ratio. The viruses have been normalized to similar MOIs, but they cannot be quantitatively compared to one another. However, each strain has been normalized to its own infection rate of WT MEFs; therefore, infection rates upon supplementation of gangliosides can be compared within a strain. The previously identified ganglioside receptors GD1a and GT1b [8] rescued RA, PTA, and LID infection of Gang-/- MEFs in a dose-responsive manner. We also analyzed the GT1a ganglioside that had not been previously investigated as a candidate infectious receptor for MuPyV. We found that GT1a, a member of the ganglio-series synthesized from GD1a (Fig 1), also rescued RA, PTA, and LID infection in a dose responsive manner (Fig 2). Moreover, GT1a supplementation of Gang-/- MEFs conferred higher levels of RA, PTA, and LID MuPyV infection than the previously identified receptors GD1a and GT1b. Finally, we tested the ability of the gangliosides GD1b and GM1 to rescue MuPyV infection of Gang-/- MEFs. GD1b and GT1b supplementation has previously been shown to restore BK polyomavirus infection of ganglioside deficient cells [30]; however, GD1b restored little to no MuPyV infection of Gang-/- MEFs. GM1 supplementation has previously been shown to restore infection by SV40 [8]; however, GM1 did not rescue MuPyV infection of Gang-/- MEFs. These data confirm that GT1a is an infectious receptor for all strains of MuPyV.
We also investigated whether MuPyV cell surface binding to infectious or non-infectious ganglioside receptors correlated with infection. To this end, we measured the levels of free (unbound) virus in each ganglioside supplemented sample at 4 hours post infection. We did not detect significant differences in MuPyV cell surface binding to different ganglioside receptors or WT MEFs, indicating that cell surface binding alone does not determine infection (S1B Fig). Instead, a considerable amount of virus binds to Gang-/- MEFs even in the absence of ganglioside supplementation (S1A Fig). MuPyV is also endocytosed in Gang-/- MEFs, which however does not lead to infection [29]. Taken together, these data confirm that gangliosides are not required for cell surface binding. They are, however, required for infection, and GT1a appears to be more efficient than GD1a and GT1b.
In order to define the mode of recognition of GT1a, particularly to the naturally occurring PTA strain of MuPyV, we have soaked VP1 crystals with the glycan portion of GT1a and solved the structure of the complex (Table 2). While the receptor interaction pocket of RA VP1 has been described [11–13], no structural information for the pathogenicity-defining amino acids at positions 91 and 296 in the pockets of PTA and LID has been available. PTA and LID both carry a glutamate at position 91, and this side chain is being held in a characteristic position with the carboxyl group facing away from the glycan receptor due to a salt bridge formed with K186 (Fig 3), as previously predicted [12]. The GT1a glycan is a branched structure with a long disialylated arm, which has the sequence Neu5Acb-[α-2,8]-Neu5Aca-[α-2,3]-Gala-[β-1,3]-GalNAc, and a second short arm, which consists of a single Neu5Acd [α-2,3]-linked to Galb (for carbohydrate structures, nomenclature, and moiety indexing see Fig 1). The disialylated arm of GT1a is clearly visible in the crystal structure of PTA VP1; it is well defined by electron density and makes extensive contacts with the protein (Fig 4B–4D). Overall, the GT1a glycan adopts a twisted horseshoe-like shape, with Neu5Aca and Neu5Acb wrapping around the side chains of Y72 and R77 of VP1. Its longer, disialylated arm contains a Neu5Aca-[α-2,3]-Gala sequence that is also present in GD1a and simpler compounds such as 3’-sialyllactose (3SL), and the interactions of this motif with VP1 are essentially identical to those seen in previous structures [11–13]. However, our structure visualizes an additional network of contacts made by the terminal [α-2,8]-linked Neu5Acb (Fig 4C and 4D). Its carboxyl group engages Y72 and forms water-mediated hydrogen bonds with Q71, Y72, as well as D85 of the neighboring monomer (D85*). In addition, the N-acetyl nitrogen of Neu5Acb forms a hydrogen bond with the backbone carbonyl of T67, and O8 and O9 in the glycerol chain of the sugar are hydrogen-bonded with the R77 side chain. The carboxyl groups of Neu5Aca and Neu5Acb are about 4 Å apart, and the positively charged side chain of R77 counteracts their negative charges (Fig 4C and 4D). Neu5Aca and Neu5Acb contribute binding interfaces of approximately 160 Å2 and 190 Å2, respectively (calculated using the PISA server [31]). The remaining Gala-GalNAc-Galb stem of GT1a forms fewer contacts with the protein, which include a hydrogen bond between G78 and the Gala O4 hydroxyl group (Fig 4) as well as several van der Waals interactions. Notably, the Cβ and Cγ atoms of E91 are within van-der-Waals range of O6 and C6 of Gala, and the E91 carboxylate group is close to C6 of GalNAc. The total contact surface for this portion of the glycan is 142 Å2.
Because the differences in tumorigenicity and host spread among strains have been mapped to the glycan binding pocket of VP1, and because GT1a appears to be particularly efficient in facilitating productive infection, we set out to determine how the three strains engage GT1a. By solving the crystal structures of RA and LID VP1 complexed with GT1a using the identical strategy used for the PTA-GT1a complex, we found that the overall binding mode of GT1a is very similar across the three strains (Fig 5A), with a conserved binding mode of the [α-2,8]-linked Neu5Acb. Although the replacement of glutamate with glycine at position 91 leads to a contact area decrease of 33 Å2 in RA, the orientation of GT1a in this strain is not altered (compare Fig 5B and 5C). Likewise, the substitution of valine with alanine at position 296 in LID removes a hydrophobic contact but does not affect the conformation of GT1a (Fig 5E; S2 Fig).
The Neu5Aca-Gala-GalNAc linkages in the long arm of GT1a adopt conformations that have been reported in numerous structures (for example [32–34]). While the [α-2,3] linkage between Neu5Aca and Gala adopts the conformation that has been reported for DSLNT and 3SL, the branching Neu5Acd-[α-2,3]-Galb linkage adopts a different conformation, which has been reported for structures containing O-4-substituted galactoses (as in [35,36]). While a higher variability is observed for Neu5Ac-[α-2,8]-Neu5Ac linkages (S2E Fig), this linkage adopts torsion angles that are in agreement with other, related structures such as in the structure of human liver fructose-1,6-bisphosphatase in complex with an allosteric inhibitor [37] or in the complex of tetanus toxin with a GT1b analog [38]. The overall structure is in good agreement with a molecular dynamics simulation using an AMBER force field in an aqueous environment [39]. A well-defined set of water molecules mediates bridged hydrogen bonds between the pyranose moieties, especially between Neu5Acb and Neu5Acd (S3 Fig). Due to these steric constraints, the GT1a complexes feature well-defined electron density not only for the binding epitope, but also for the non-binding, branching NeuNAcd in its preferred solution conformation [40], which brings this moiety to about 5 Å near the end of the long arm and gives the glycan the characteristic, horseshoe-like topology that is observable in all complex structures.
As RA, PTA, and LID VP1 all bind GT1a in a highly similar conformation, we hypothesized that the differences in pathogenicity and spread among the three strains might be due to the recognition of additional carbohydrates by only a subset of MuPyV strains. As shown in Fig 1, the many different gangliosides share a relatively small set of common sialoglycotopes. We therefore investigated the ability of all three VP1 proteins to bind other glycan structures that are representative for these epitopes. We solved structures of VP1 bound to the glycan portions of two of these gangliosides: The GD1a glycan is an established infectious receptor and essentially a truncated version of GT1a lacking the [α-2,8]-linked Neu5Acb in the long arm. The human milk hexasaccharide DSLNT is the glycan portion of the lacto-series ganglioside 3’-6’-isoLD1 (Fig 1) [41], which is overexpressed in the central nervous system. In contrast to GT1a and GD1a, DSLNT does not contain an [α-2,3]-linked Neu5Acd as a short arm but instead a branching [α-2,6]-linked Neu5Acc. This structure is similar to a very common epitope on O-linked glycoproteins [42–44]. DSLNT was used in previous studies of MuPyV as a model “pseudoreceptor” [12] and was investigated here to help rationalize these earlier data, to facilitate a comparison among strains, and to establish a binding profile for glycans containing an [α-2,6]-linked sialic acid.
Since all three MuPyV strains are able to engage the three different glycan structures in a largely identical manner, we reasoned that the differences in pathogenicity and spread might be attributable to subtle differences in affinity, rather than specificity, among the strains. The affinities of RA VP1 for 3’-sialyllactose and DSLNT were estimated to be in the low mM range [11]. Coupled with the high costs of glycans and the high amount required due to their low binding affinity, weak binding poses technical challenges for classical affinity measurements. We therefore utilized a crystallographic approach to quantitatively compare ligand binding. We crystallized all three VP1 pentamers in the same condition, and soaked each with the oligosaccharide portions of GT1a, GD1a, and DSLNT at different concentrations in parallel. X-ray data of all crystals were collected in the same manner, and the data sets were processed using the same protocol and integrated as described previously [45]. All data sets were processed in the same unit cell, scaled, and the bias-reduced difference electron density around the central Neu5Aca-[α-2,3]-Gala motif was quantified for each data set (see the Methods section for details). Our crystallization condition contains a high amount of ammonium sulfate, which competes with the carboxyl group of Neu5Aca and has to be displaced by the carbohydrates. Therefore, our observed binding is weaker than in a physiological setting. However, while not yielding dissociation constants in the traditional sense, this method enables us to compare relative levels of binding across our three different strains and three different glycans.
The GT1a glycan exhibits the strongest binding in all three VP1 variants compared with DSLNT or GD1a (Fig 7A–7C), with no detectable difference between the strains (Fig 7D). This finding is in accord with our ganglioside add-back experiments in cell culture (Fig 2), which consistently showed higher levels of infection mediated by GT1a compared to GD1a. The stronger overall binding of GT1a can be attributed to the additional [α-2,8]-linked sialic acid present in GT1a (Neu5Acb), which contributes several interactions and an increased buried surface area. These contacts seem to outweigh the differences in van der Waals contacts with the side chains of E91 or V296, at least to the extent discernable in our assay.
GD1a binds less well to all strains compared to GT1a. In addition, there are differences in binding strength among the three strains. PTA and LID VP1 appear to bind GD1a at the same level and better compared with RA (Fig 7A–7C and 7E) because these two strains gain additional interaction surface and van-der-Waals contacts from their E91 side chain. This effect is more pronounced than in GT1a, because in GD1a it cannot be masked by the additional contacts of the [α-2,8]-linked Neu5Acb.
DSLNT displays the lowest overall affinity to all strains, with levels comparable to GD1a in RA for all three strains (Fig 7A–7C) despite the DSLNT conformation being slightly different in each VP1 complex (Fig 7F). Neither the blocking of Neu5Acc binding by E91, nor the increased conformational freedom in LID appears to alter binding affinity. It is possible that Neu5Acc in RA adopts a conformation that might not be favorable and therefore not heavily contribute to affinity, in spite of the added contact surface. Combined with the fact that electron density for Neu5Acc could only be observed in one binding pocket of RA VP1 [12], we believe that this conformation is possible but not probable in solution. Instead, an increased number of conformational options might make up for a loss of binding contacts.
Many viruses engage cell-surface glycans to mount an infection, and subtle differences in the recognition of such receptors can be linked with altered tropism and pathogenicity. Examples include the canine parvovirus and feline panleukopenia virus [46,47], the human BK polyomavirus [48], B-lymphotropic polyomavirus [49,50] as well as avian and human influenzaviruses [51,52]. However, MuPyV is a rare example of a virus in which drastic differences in pathogenicity directly correlate with single amino acid substitutions in the viral capsid.
In order to provide a structural basis for understanding the profoundly different pathogenicities of the three MuPyV strains RA, PTA and LID, we have solved structures of their VP1 proteins and characterized their receptor-binding properties. We show that the ganglioside GT1a serves as a MuPyV receptor and promotes infection with higher potency than the previously identified receptors GD1a and GT1b. Structurally, the increased potency of GT1a can be directly explained by a set of additional contacts involving the [α-2,8]-linked Neu5Acb that is only present in this glycan and that gives it a characteristic horseshoe-like shape. It had previously been suggested that the G91E mutation present in PTA and LID abolishes binding to branched glycans containing [α-2,6]-linked Neu5Ac and thus allows the virus to spread more efficiently in the host [8,11]. However, our analyses show that the presence of a glutamate at position 91 still allows binding of the branched oligosaccharides GT1a, GD1a, or DSLNT to all three strains, albeit with subtle differences in binding affinity. While all three strains bind GT1a with comparable affinity, PTA and LID bind GD1a better than RA. The DSLNT glycan binds similarly to all three strains, with the lowest overall affinity. This is again in line with the structures, which show that the branched Neu5Acc of DSLNT does not engage in any specific contacts. The limited contacts between Neu5Acc and RA observed in an earlier structure [12] have to be considered a crystallization artifact as they were only observed in one out of five binding sites, and this visible Neu5Acc moiety was located near a crystal contact.
The ligand binding promiscuity of MuPyV is surprisingly high. Binding mostly requires a ubiquitous minimal Neu5Ac-[α-2,3]-Gal motif, in agreement with earlier findings [6,7]. It therefore seems plausible that the virus also recognizes other glycans bearing this motif, resulting in differences in pathogenicity and spread. Preliminary studies show that glycans with an N-acetyllactose core (Neu5Ac-[α-2,3]-Gal-[β-1,4]-GlcNAc), as found in neolacto gangliosides such as the predominant ganglioside of peripheral nerve cells, LM1 [53,54], can also be bound in a manner similar to DSLNT and with higher flexibility than GT1a or GD1a (S6 Fig).
Based on our structures, certain requirements that contribute to receptor specificity can be established. For example, branches at Gal-O4 within the minimal motif produce clashes and cannot be tolerated. Therefore, although the GD1a glycan possesses two Neu5Ac-[α-2,3]-Gal motifs, it prefers the one on its longer arm for complex formation. For the same reason, glycans such as GM1 or GM2 that only possess such a branched Neu5Ac-[α-2,3]-Gal epitope cannot engage MuPyV productively. In support of this, the GM1 ganglioside is not able to rescue MuPyV infection of Gang-/- MEFs (Fig 2, [29]), although low-level and probably non-specific interactions with cells can be detected (S1B Fig). GT1b possesses a disialylated arm at Galb and is monosialylated at Gala. We predict that GT1b engages VP1 with its monosialylated arm. The second, disialylated arm is likely to be accommodated in such a binding mode, and the [α-2,8]-linked sialic acid might contribute additional contacts. Binding via the monosialylated arm is in line with our findings that supplementation of Gang-/- cells with GT1b rescues infection at a level between GD1a and GT1a. Some gangliosides whose glycan epitopes are capable of engaging VP1 in vitro might not be infectious receptors in vivo, mainly because of steric complications in the context of the cell membrane. For example, while the crystal structure of PTA with the glycan portion of GD3 shows an identical binding mechanism to GT1a (S7 Fig), supplementation of Gang-/- MEFs with GD3 does not restore infectivity [29]. We reason that the glycan stem of GD3 (and of gangliosides with a similar length such as GM3) is too short to allow efficient attachment of the MuPyV capsid to the cell membrane.
The discrepancy in pathogenicity in MuPyV strains that differ only at one single position is stark. In sharp contrast, the differences among receptor binding between the three strains investigated here are subtle, and a correlation of the structural data with the observed pathogenicity profiles remains challenging. One reason for this is that avidity effects in the virus capsid, which can engage many ligands simultaneously, multiply subtle changes in the affinity of capsomers for single glycans. It was shown for influenza viruses that small changes between millimolar binding affinities of single binding sites can result in dramatically altered viral binding properties [52]. As discussed above, we found the main difference between RA and PTA/LID to be a differing affinity for GD1a, which appears to bind better to the latter strains due to the larger E91 side chain. This might facilitate attachment and productive infection by these strains to cells that display GD1a, and may thus give them an advantage over RA. While we could not show differences between the PTA and LID strain in terms of glycan affinity to isolated VP1 pentamers, it is unclear how this translates to avidity effects. As such, it is possible that capsid avidities differ enough to explain the more limited spread of PTA. Although direct correlations cannot be made, it becomes increasingly clear that the virus needs to uphold a delicate equilibrium between efficient infection and release from infected and lysed cells as well as selective affinity for productive receptors. The absence of the RA and LID strains outside the laboratory [26] emphasizes that this equilibrium is affected by minute changes in the receptor binding properties.
The MuPyV receptor pocket can clearly accommodate several related but distinct glycan structures (Figs 1 and 4–6). These structures also decorate glycoproteins on many cell surfaces. It therefore seems likely that MuPyV can also engage glycans that are not attached to gangliosides. For instance, the glycan stem of GD1α, which is very similar to DSLNT and prominent on glycoproteins [42–44], is a likely receptor candidate. The different cell-surface distribution patterns of glycoproteins and gangliosides may likewise influence MuPyV spread [8]. Glycoprotein receptors with unknown identity have in fact been shown to promote non-productive internalization of MuPyV, which in turn elicits innate immune responses by the host [29]. Along these lines, our results suggest that virus particles adhere to and enter ganglioside deficient MEFs to levels that are not significantly lower than for wild-type and ganglioside supplemented Gang-/- cells, although without detectable infection. Although not representative for other cell types, these results suggest that the amount of non-productive “pseudoreceptors” on the MEF cell surface is much higher than anticipated.
Our data demonstrate that varying affinities for different gangliosides are the key determinants of a successful MuPyV infection, in line with earlier reports [6–8]. Perhaps unexpectedly, we also find that (even non-specific) attachment of the virus to a host cell can lead to successful internalization, but that this does not necessarily lead to an infection. Thus, we propose that the ratio between productive (ganglioside bound) and non-productive (ganglioside and glycoprotein bound) glycotopes on the host cell itself or in its microenvironment helps to determine the productivity of infection through diverging entry routes, and that differential affinities to these receptors dictate this equilibrium. The nature of these diverging routes, their underlying driving forces, and potential biological consequences other than immune stimulation [29] remain unknown–as does the point at which they diverge. We cannot exclude the possibility that the distribution and binding properties of (pseudo-)receptors are of importance mostly for the post-entry stage rather than for events taking place at the cell surface. A better understanding of the distribution patterns and densities of glycans on specific cells is clearly needed to fully appreciate the many aspects of pathogenesis and tropism of MuPyV as well as many other glycan-binding viruses.
WT and Gang-/- MEFs were seeded onto 96-well Costar 3906 imaging plates in Dulbecco's Modified Eagle's Medium supplemented with 10% fetal bovine serum (FBS). WT (B4+/+St8+/+) and Gang-/- MEFs (B4-/-St8-/-) were provided by Thomas Benjamin at Harvard Medical School. Gangliosides were purchased from Matreya LLC and resuspended in DMSO upon arrival, aliquoted, and stored at -20°C until use. Cells were incubated overnight in serum free media prior to infection. For ganglioside supplemented Gang-/- MEFs, cells were starved in serum free media containing the indicated concentration of ganglioside. Gangliosides were then removed, and cells were washed with serum free media to remove any free ganglioside. Cells were then infected with NG59RA, PTA, and LID MuPyV (MOI ~10–30). At 24 hours post infection cells were washed in phosphate buffered saline and fixed with 4% paraformaldehyde at room temperature for 10 minutes. Cells were then permeabilized with 0.1% Triton X-100, blocked in 10% FBS in PBS, and then stained for the viral protein, T-antigen (E1). Samples were then incubated with Alexa Fluor labeled secondary antibodies (546). Plates were imaged with the Molecular Devices ImageXpress Micro XL High-Content Screener. The percent infected was calculated for each well (5 images were taken per well). Three wells were quantified per sample and the average percent infected, standard error, and standard deviation were calculated for each sample. To quantify infection, T-antigen staining was measured per each DAPI labeled nucleus. For image analysis, the DAPI channel on each image was thresholded, and nuclei were counted using ImageJ (Analyze Particles). These particles were marked as “Regions of Interest” (ROI), and then the average pixel intensity of T-antigen staining was measured for each nucleus (ROI). These were then binned into T-antigen positive or T-antigen negative nuclei to create % infected.
WT and Gang-/- MEFs were seeded onto glass coverslips in Dulbecco's Modified Eagle's Medium supplemented with 10% (FBS). Cells were incubated overnight in serum free media prior to infection. For ganglioside supplementation, Gang-/- MEFs were starved in serum free media containing the indicated concentration of ganglioside. Gangliosides were then removed and cells were washed with serum free media to remove any free ganglioside. Cells were then infected with NG59RA. At indicated times post infection the cells were fixed with 4% paraformaldehyde at room temperature. Cells were blocked in 10% FBS in PBS and then stained for GD1a using mAb MAB5606 (Millipore). Cells were then permeabilized with 0.1% Triton X-100 and stained for the viral proteins, VP1 (I58 antibody) and T-antigen (E1 antibody). Samples were washed and then incubated with Alexa Fluor labeled secondary antibodies (488, 546, 647). Slides were then mounted using DAPI prolong gold mounting media. Slides were imaged with a Nikon A1R confocal microscope. All images were taken as a 9 to 13 step (.25μm) z-stacks on a laser scanning confocal microscope. Each z-stack was aligned and compressed into a max intensity Z projection image.
WT and Gang-/- MEFs were seeded onto a 24 well dish in Dulbecco's Modified Eagle's Medium supplemented with 10% (FBS). Cells were incubated overnight in serum free media prior to infection. For ganglioside supplemented Gang-/- MEFs, cells were starved in serum free media containing the indicated concentration of ganglioside. Gangliosides were then removed and cells were washed with serum free media to remove any free ganglioside. Cells were then infected with either NG59RA, PTA, or LID at an MOI ~10–30 (250 μL/well). At 4 hours post infection 150 μL of virus supernatant was removed and placed into a microcentrifuge tube. This virus supernatant was then used to infect WT MEFs seeded onto a 96-well plate (50 μL/well). The amount of free virus was then quantified as percent of infection of the 96-well reinfection plate. At 24 hours post virus addition the plate was washed in PBS and fixed with 4% PFA at RT for 10 minutes. Cells were then permeabilized with 0.1% Triton X-100, blocked in 10% FBS in PBS, and then stained for the viral protein, T-antigen (E1). Samples were then incubated with Alexa Fluor labeled secondary antibodies (546). Plates were imaged with the Molecular Devices ImageXpress Micro XL High-Content Screener. The percent infected was calculated for each well (5 images were taken per well) as indicated by T-antigen positive nuclei. Three wells were quantified per sample and the average percent infected, standard error, and standard deviation were calculated for each sample. For image analysis, the DAPI channel on each image was thresholded and nuclei were counted using ImageJ (Analyze Particles). These particles were marked as “Regions of Interest” (ROI) and then the average pixel intensity of T-antigen staining was measured for each nucleus (ROI). These were then binned into T-antigen positive or T-antigen negative nuclei to create % infected.
DNA encoding residues 33–316 of RA (GenBank # M34958.1) or PTA VP1 (GenBank # PSU27812) was cloned into the expression vector pET15b (Novagen) in frame with an N-terminal hexahistidine tag (His-tag) and a thrombin cleavage site. DNA for LID VP1 (GenBank # PSU27813) was generated by site-directed mutagenesis of PTA VP1 residue 296. VP1 pentamers were overexpressed in E. coli (BL21) after IPTG induction, and purified by nickel affinity chromatography. The His-tag was removed by thrombin cleavage on column for 72 hours (leaving the non-native residues GSHM at the N-terminus), followed by size exclusion chromatography on a Superdex-200 column.
Pure VP1 pentamers were supplemented with 20 mM DTT, concentrated to 7.5–8 mg/mL (RA VP1) or 8.5–9 mg/mL (PTA and LID VP1), and crystallized by sitting-drop vapor diffusion. RA VP1 was crystallized at 20°C against reservoir solutions containing a range of 1.25–1.8 M ammonium sulfate and 1–10% (v/v) isopropanol. PTA and LID were crystallized at 4°C against reservoir solutions containing 0.1 M HEPES pH 7–8.5 and 1.6–1.8 M K-Na phosphate. For complex formation, the crystals were soaked in the reservoir solution supplemented with the glycan. The detailed crystallization and soaking procedures are listed in S1 Table. The GT1a and GD1a glycans were purchased from Elicityl SA (France), and the DSLNT glycan was purchased from Carbosynth (United Kingdom).
For concentration-dependent soaking VP1 pentamers of all three strains were crystallized at 20°C against a mother liquor containing 1.5 M ammonium sulfate and 6% (v/v) isopropanol. These crystals were soaked in drops of mother liquor containing the appropriate concentration of glycan for 30 minutes.
All crystals were cryoprotected by incubation in mother liquor supplemented with the appropriate concentration of glycan and 25% (v/v) glycerol. They were then flash-frozen in liquid nitrogen.
Data reduction was carried out in XDS [55], and the structure of native RA VP1 was solved in Molrep [56] using a model generated from the previously solved structure of P16 VP1 (PDB code 1VPN [12]). Other structures were solved by molecular replacement using the RA VP1 structure in Phenix [57]. All structures were completed by alternating rounds of manual model building in Coot [58], followed by restrained coordinate and isomorphous B-factor refinement including TLS refinement and five-fold non-crystallographic symmetry restraints in Refmac5 [59]. TLS parameters were obtained from the TLSMD server [60]. All models agree well with the experimental data and have good geometry (Table 2). The PDB accession codes for the structures are listed in Table 2. Structural figures were prepared in PyMOL [61].
Data collected for concentration-dependent soaking experiments was processed as described above. The unit cell parameters for all datasets were treated as equal for all datasets and isomorphous to the dataset “RA Nat” (S2 Table). They were scaled against “RA Nat” in Scaleit [62] and then subjected to B-factor refinement and simulated annealing in Phenix against models of RA, PTA, or LID VP1, which lacked atoms of all solvent molecules in the receptor binding pocket as well as those of tryptophan residues 98 and 227 as controls. The resulting bias-reduced Fobs-Fcalc electron density for Neu5Aca-[α-2,3]-Gala and the two marker tryptophans was calculated as a summation of values of the grid points in a mask generated 1 Å around these groups using the program Mapman [63]. The overall binding of a sugar at different concentrations influences the electron density of the Neu5Aca-[α-2,3]-Gala portion, which is included in GT1a, GD1a, and DSLNT. In contrast, it has no effect on the electron density of the marker tryptophan residues, which do not differ significantly for all data points. For each data point, the average density of the five chains was plotted against ligand concentration and submitted to a non-linear least squares fit using the equation
Y=X(KD+X)⋅(Bmax−B0)+B0
(1)
where Bmax was the highest observed electron density value overall (constrained to 95.03 AU) and B0 the electron density in the binding pocket at 0 mM ligand concentration. Plotting and fitting was done using the program Prism 6 (GraphPad Software, Inc., La Jolla, California, USA).
|
10.1371/journal.ppat.1002758 | Identification of a General O-linked Protein Glycosylation System in Acinetobacter baumannii and Its Role in Virulence and Biofilm Formation | Acinetobacter baumannii is an emerging cause of nosocomial infections. The isolation of strains resistant to multiple antibiotics is increasing at alarming rates. Although A. baumannii is considered as one of the more threatening “superbugs” for our healthcare system, little is known about the factors contributing to its pathogenesis. In this work we show that A. baumannii ATCC 17978 possesses an O-glycosylation system responsible for the glycosylation of multiple proteins. 2D-DIGE and mass spectrometry methods identified seven A. baumannii glycoproteins, of yet unknown function. The glycan structure was determined using a combination of MS and NMR techniques and consists of a branched pentasaccharide containing N-acetylgalactosamine, glucose, galactose, N-acetylglucosamine, and a derivative of glucuronic acid. A glycosylation deficient strain was generated by homologous recombination. This strain did not show any growth defects, but exhibited a severely diminished capacity to generate biofilms. Disruption of the glycosylation machinery also resulted in reduced virulence in two infection models, the amoebae Dictyostelium discoideum and the larvae of the insect Galleria mellonella, and reduced in vivo fitness in a mouse model of peritoneal sepsis. Despite A. baumannii genome plasticity, the O-glycosylation machinery appears to be present in all clinical isolates tested as well as in all of the genomes sequenced. This suggests the existence of a strong evolutionary pressure to retain this system. These results together indicate that O-glycosylation in A. baumannii is required for full virulence and therefore represents a novel target for the development of new antibiotics.
| Multidrug resistant (MDR) Acinetobacter baumannii strains are an increasing cause of nosocomial infections worldwide. Due to the remarkable ability of A. baumannii to gain resistance to antibiotics, this bacterium is now considered to be a “superbug”. A. baumannii strains resistant to all clinically relevant antibiotics known have also been isolated. Although MDR A. baumannii continues to disseminate globally, very little is known about its pathogenesis mechanisms. Our experiments revealed that A. baumannii ATCC 17978 has a functional O-linked protein glycosylation system, which seems to be present in all strains of A. baumannii sequenced to date and several clinical isolates. We identified seven glycoproteins and elucidated the structure of the glycan moiety. A glycosylation-deficient strain was generated. This strain produced severely reduced biofilms, and exhibited attenuated virulence in amoeba, insect, and murine models. These experiments suggest that glycosylation may play an important role in virulence and may lay the foundation for new drug discovery strategies that could stop the dissemination of this emerging human pathogen, which has become a major threat for healthcare systems.
| Acinetobacter baumannii is a strictly aerobic Gram negative, non-fermentative, opportunistic pathogen. Since the 1970's, this organism has frequently been isolated from healthcare facilities, but was easily controlled with antibiotics [1], [2]. However, many clinical isolates of A. baumannii have recently emerged with extreme resistance to antibiotics, disinfectants, and desiccation, which has permitted A. baumannii to disseminate throughout healthcare facilities worldwide [3]–[7]. One recent study showed that from 2001 to 2008 the percentage of A. baumannii isolates resistant to at least three classes of antibiotics increased from 4% to 55%, and 17% of all isolates were resistant to at least four drug classes [8]. Panresistant strains of A. baumannii have also been isolated [9]. Because of its importance as an emerging pathogen, attention towards A. baumannii has increased considerably. Most of the efforts have focused on antibiotic resistance mechanisms, but little is known about its virulence factors. A significant amount of work has been done to characterize biofilm formation, which seems to play a role in pathogenesis [10], [11]. Other suggested virulence factors for A. baumannii include the capsule, exopolysaccharide, pili and lipopolysaccharide (LPS) [11]–[14]. Undoubtedly, more research is needed in order to understand A. baumannii pathogenesis.
Genomic analysis of all sequenced A. baumannii strains revealed the presence of homologous genes to those encoding enzymes involved in the Neisseria meningitidis protein O-glycosylation system. Many different mucosal pathogenic bacteria require protein glycosylation for virulence, and glycoproteins seem to play a role in adhesion, motility, DNA uptake, protein stability, immune evasion, and animal colonization [15]. Whereas N-glycosylation seems to be restricted to epsilon and a few delta proteobacteria, O-glycosylation appears to be more widespread among bacteria. Gram negative bacteria including Neisseria spp. and Bacteroides fragilis employ en bloc O-glycosylation as a general system to modify multiple proteins [16], [17]. En bloc O-glycosylation is initiated by a specialized glycosyltransferase that attaches a nucleotide-activated monosaccharide-1P to an undecaprenolphosphate (Und-P) lipid carrier on the inner face of the inner membrane. A series of glycosyltransferases subsequently attach additional monosaccharides to the first sugar residue on Und-PP. When the carbohydrate structure is completed, the Und-PP linked glycan is flipped to the periplasmic face, where an O-oligosaccharyltransferase (O-OTase) transfers the carbohydrate to selected Ser or Thr residues in acceptor proteins [18], [19]. Campylobacter jejuni employs a similar N-glycosylation pathway to modify about 65 proteins [20].
This work demonstrates the existence of a general O-glycosylation system in A. baumannii ATCC 17978, which is required for efficient biofilm formation and pathogenesis in the Dictyostelium discoideum, Galleria mellonella, and murine septicemia virulence models. We identified seven glycoproteins carrying a branched pentasaccharide, the structure of which has been characterized by MS and NMR techniques. O-glycosylation appears to be ubiquitous in A. baumannii, which suggests that this system might be a possible target for novel antimicrobial treatments.
We initially searched the A. baumannii ATCC 17978 genome for homologues of known O-OTases. Via a BLAST analysis, we identified a homolog to the N. meningitidis O-OTase PglL (A1S_3176; E-value 1e-9) that contained a Wzy_C motif. This motif is conserved in all O-OTases, but is also found in WaaL ligases, which catalyze the transfer of the O antigen to the Lipid A core [21], [22]. To date, only experimental determination allows the assignment of an ORF containing the Wzy_C motif as either an O-OTase or a ligase [23]. No other ORFs contained a Wzy_C motif in the A. baumannii ATCC 17978 genome. A1S_3176 is not predicted to be part of an operon [24]. We carried out mutagenesis of the A1S_3176 gene by homologous recombination to evaluate if its encoded protein is an O-Otase or a WaaL ligase. PCR and DNA sequencing confirmed the creation of an A1S_3176 knockout strain, in which the targeted gene was replaced with a gentamicin resistance cassette. There was no significant difference between the growth curves of the wild-type and the A1S_3176 mutant strains at 37°C, indicating that growth in these conditions was not affected by the mutation (Data not shown).
Most of the Neisseria O-glycoproteins identified to date are associated to membranes [16]. Membrane extracts from wild type and ΔA1S_3176 A. baumannii strains were analyzed by SDS-PAGE followed by PAS staining, a technique that is specific for detecting glycans, but presents low sensitivity (Fig. 1). A broad band migrating from 25 to 35 kDa was visualized in the extract of A. baumannii WT. Although the membrane protein profile between the WT and the ΔA1S_3176 strains appeared similar, the band detected via PAS stain was not visible in the mutant strain, suggesting that A1S_3176 is required for glycosylation of at least one protein (Fig. 1B). The PAS-reactive band disappeared upon treatment with proteinase K, associating the glycan signal with proteinaceous material. Complementation of A1S_3176 was achieved in trans, and analysis of A. baumannii ΔA1S_3176-pWH1266-pglL membrane extract showed the reappearance of the PAS stained band. Due to the aforementioned similarity between O-OTases and ligases, we carried out a conventional LPS extraction and analyzed the extract of the different strains via SDS-PAGE. Silver stain showed no obvious differences in the carbohydrate pattern were observed, suggesting that A1S_3176 is not involved in LPS synthesis (Fig S1). To further determine if A1S_3716 effected LPS biosynthesis, whole cells were digested with proteinase K and analyzed by Silver stain and no differences were observed (data not shown). However, it has been reported that the O-antigen chains of certain A. baumannii strains are not detectable by Silver stain and therefore we cannot conclusively exclude a role of A1S_3176 in LPS synthesis [25]. Together these results suggest that A1S_3176 is an O-OTase responsible for O-glycosylation in A. baumannii and will be referred from here on as PglLAb, as per its N. meningitidis ortholog.
To identify the glycoprotein(s) in A. baumannii, we performed two dimensional in-gel electrophoresis (2D-DIGE) experiments [26]. Membrane samples of both WT and ΔpglL were isolated by ultracentrifugation and the lipidic components were removed as previously described [27]. Most of the signals corresponding to the wild type (Fig. 2A, green) and ΔpglL (Fig. 2B, red) proteins co-localized in the gel (Fig. 2C, yellow), indicating that these proteins were likely not glycosylated. However, a few proteins exhibited differential electrophoretic behavior (Fig. 2). These proteins spots were excised, in-gel digested, and analyzed by MALDI-TOF/TOF MS and MS/MS. We identified two separate pairs of proteins, which according to their electrophoretic migration, appeared to be larger and more acidic in the WT strain (WT1 and WT2) than in the ΔpglL strain (MT1 and MT2). Mass spectrometric analysis determined WT1 and MT1 samples to be A1S_3626 protein, whereas WT2 and MT2 were identified as A1S_3744 protein. Both, A1S_3626 and A1S_3744 are annotated as hypothetical proteins, and BLAST searches yielded homologues exclusively within the Acinetobacter genus.
Analysis of the MALDI-TOF MS spectra of a tryptic digest of WT1 (A1S_3626) revealed a peptide fragment of 2895.24 Da that was absent in MT1 (Fig. 3A). MALDI-TOF-TOF MS/MS of this ion determined that in the wild-type strain the peptide SAGDQAASDIATATDNASAK was linked to the glycan HexNAc-Hex-Hex-(HexNAc)-300, where 300 corresponded to an unknown residue of m/z 300, whereas the same peptide was unmodified in ΔpglL sample (Fig S3A). Similarly, MALDI-TOF MS analysis of a tryptic digest of WT2 (A1S_3744) revealed a peptide fragment of 3852.69 Da that was absent in MT2 (Fig. 3B). MALDI-TOF-TOF MS/MS of the 3852.69 Da peak revealed the same pentasaccharide identified on A1S_3626 on the peptide ETPKEEEQDKVETAVSEPQPQKPAK (2822.33 Da), whereas the same peptide was unmodified in ΔpglL sample (Fig S3B).. We next purified membranes from A. baumannii, digested the sample with Pronase E, and enriched glycosylated peptides using activated charcoal microspin columns. We identified a peak in the MALDI-TOF MS of 1358.4 m/z that was subsequently analyzed by MALDI-TOF/TOF MS/MS (Fig. 3C). Manual peak annotation identified the previously characterized pentasaccharide attached to a sodiated tripeptide containing the amino acids A, T and D. Overall, these results demonstrate that PglLAb glycosylates at least two different proteins with a pentasaccharide with a preliminary structure of HexNAc-Hex-Hex-(HexNAc)-300.
We observed other spots possibly corresponding to proteins migrating differently in A. baumannii WT and ΔpglL strains. The most prominent was marked as WT3, and was observed only in the WT extract (Fig. 2C). Mass spectroscopy analysis determined this spot corresponded to OmpA (A1S_2840). However, manual analysis using MS/MS of WT3 indicated that OmpA was not glycosylated. Western blot analysis of whole cell extracts of the WT and ΔpglL strains revealed no difference in OmpA expression levels, which implies that manipulation of membrane samples could account for apparent differences observed in expression levels of proteins detected by 2D-DIGE (Fig S2).
To determine if additional glycoproteins were present in A. baumannii ATCC 17978, we employed ZIC-HILIC glycopeptide enrichment. Utilizing membrane extracts previously shown to contain A1S_3626 and A1S_3744 putative glycopeptides were enriched and analyzed using an LTQ-Orbitrap Velos. HCD scans containing oxonium ion were manually inspected and searched using MASCOT resulting in the identification of at least 9 different glycosylation sites on 7 different glycoproteins in A. baumannii ATCC 17978 (Table 1; Fig. 4). This peptide-centric approach enabled multiple novel glycoproteins to be identified of which six of the seven proteins are annotated as uncharacterized hypothetical proteins, with the remaining being annotated as MotB (A1S_1193). (Table 1). This demonstrates that PglLAb is able to glycosylate multiple proteins in A. baumannii ATCC 17978.
Identification of the O-glycan of A. baumannii ATCC 17978 was achieved by 2D NMR analysis. The Pronase E digested membrane protein extracts characterized in Fig. 3C were analyzed by 1H:13C HSQC 2D NMR and revealed the structure of the pentasaccharide to be β-GlcNAc3NAcA4OAc-4-(β-GlcNAc-6-)-α-Gal-6-β-Glc-3-β-GalNAc-, with the amino acids S, E, and A attached in any combination (Fig S4, Table 2). β-GlcNAc3NAcA4OAc (corresponding to m/z 300; Fig. 3) is an O-acetylated derivative of glucuronic acid, and can account for the more acidic migration of the WT glycoproteins compared to the ΔpglL in the 2D-DIGE analysis.
It has been suggested that biofilm formation is important for A. baumannii virulence [28]. We tested if O-glycosylation has an impact on biofilm formation in this organism. Biofilm formation was detected using crystal violet staining and quantitatively analyzed by comparing the ratio between cell growth (OD600) and biofilm formation (OD580) at 30°C after 48 hours incubation (Fig. 5A). High absorbance values corresponding to a strong ability to create biofilms (1.23±0.48 and 1.12±0.40) were obtained for the WT strain and the ΔpglL strain complemented in trans respectively. On the contrary, the ΔpglL strain and the ΔpglL strain transformed with pWH1266 exhibited severely reduced levels of absorbance (0.18±0.07 and 0.20±0.04). Similar results were also observed at 37°C (data not shown). We further characterized the role of O-glycosylation in biofilm formation by employing a flow cell system. A. baumannii strains were stained with the green fluorescent stain SYTO 9, visualized by confocal laser scanning microscopy, and quantitative analysis of the biofilms was performed with COMSTAT. Assessment of the initial attachment after 2 hours shows that ΔpglL strain and vector control had significantly less surface coverage (4.12% and 2.32% respectively) than the WT and in trans complemented strain (6.41% and 6.45% respectively; Fig. 5B). Confocal microscopy and subsequent analysis of biofilms biomass, as well as average and maximal thickness after 24 hours showed significantly higher levels for the WT compared to the ΔpglL strain, and the phenotype was restored to WT levels when pglLAb was complemented in trans (Fig. 5 C, D, E, F; *P<0.05). These data indicate that the A. baumannii strain defective in O-glycosylation has a severely diminished capacity to form biofilms.
Two well-established virulence models for A. baumannii are the D. discoideum predation and the G. mellonella infection models [5], [29]–[33]. D. discoideum is an unicellular amoeba that feeds on bacteria and previous work has demonstrated similarity between phagocytosis of the amoebae and mammalian phagocytes [34]. We examined if protein glycosylation was required for virulence towards D. discoideum by co-incubation of A. baumannii strains with the amoebae on SM/5 nutrient agar. A. baumannii was previously shown to inhibit amoebae growth in the presence of 1% ethanol [5]. The WT strain was virulent and inhibited all D. discoideum growth in the presence of 1% ethanol, which resulted in no plaque being formed. However the ΔpglL strain was avirulent towards the amoeba, which resulted in plaque formation in the bacterial lawn within 48 hours and clearing of the plate within 4–5 days (Fig S5). G. mellonella have been used to study many host-pathogen interactions, and have several advantages over other virulence models including the presence of both humoral (ie. antimicrobial peptides) and cellular immune response systems (phagocytic cells) [32]. Most importantly, a correlation has been established between the virulence of several bacteria in G. mellonella and mammalian models [35], [36]. For the G. mellonella, while a similar bacterial load (2.31±1.13×105 CFU) was injected for each of the strains, only the WT and complemented strains were able to kill the wax moth larvae after 36 hours, (20% and 0% survival), whereas larvae injected with ΔpglL and the ΔpglL vector control strains had significantly higher survival rates (100% and 80%; Fig. 6). The LD50 of the WT and complemented strains were determined to be approximately 2.6×104 and 1.4×104 respectively after 36 hours. No additional killing was observed in the ΔpglL or vector control strains up to 96 hours. A PBS injected control maintained 100% survival throughout the length of the virulence assay. These results demonstrate a critical role for O-glycosylation in the virulence of A. baumannii in these two model systems.
We then tested A. baumannii ΔpglL virulence in vivo using a previously described murine septicemia competition model [37]–[39]. We first determined the LD50 of A. baumannii ATCC 17978 strain by injecting groups of 5 BALB/c mice with serially diluted bacteria cultures (Fig. 7A). A very small dose range between full survival and full killing was observed, and the LD50 was determined to be 6.49×104 CFU/mouse. The competition index (CI) was defined as the number of ΔpglL CFUs recovered/number of WT CFUs recovered, divided by the number of ΔpglL CFUs inoculated/number of WT CFUs inoculated. Cultures of each strain were mixed at a ratio of 1∶1, serial diluted, and plated to determine the initial CI. 1×105 CFU of the mixed strains were injected intraperitoneally into the BALB/c mice, which were subsequently sacrificed 18 hrs post injection. The spleens were aseptically harvested, serial diluted, and plated. All of the mice had a high spleen CFU load of 3.75±2.37×108 CFU/gram and were moribund at the time of sacrifice. While the initial prescreen showed a CI of 1.18±0.21 favoring the ΔpglL mutant, the spleen counts after 18 hrs showed a CI of 0.10±0.03 (Fig. 7B). This data suggests that Ab ΔpglL has a competitive disadvantage as compared to the WT strain. Together, these results indicate that A. baumannii strains lacking O-glycosylation are attenuated in mice.
To determine the degree of conservation of the O-glycosylation system in Acinetobacter sp., we searched for the presence of PglLAb homologues in different species within the genus. This genomic search showed that PglLAb was present in all the genomes analyzed with high sequence homology (Fig S6A). We obtained eight clinical isolates from the University of Alberta Hospital. The isolates were identified by 16S rDNA and recA sequencing to be different species within the Acinetobacter genus (A. baumannii, A. nosocomialis, A. pittii, and A. calcoaceticus). Membranes of these strains were purified and analyzed by PAS staining for the presence of glycoproteins (Fig S6B). While there appears to be variation in the size and intensity of the PAS stained band, all the isolates were positive for glycoproteins, demonstrating that PglLAb was active in all these strains. This indicates that despite the plasticity of Acinetobacter sp. genomes [40], there is a strong evolutionary pressure to retain a functional O-glycosylation system.
Isolation of MDR strains of A. baumannii is increasing at impressive rates. Despite its growing incidence as nosocomial pathogen, only a few A. baumannii virulence factors have been characterized. In this article we describe a general O-glycosylation system in A. baumannii ATCC 17978. Although once considered rare in prokaryotes, both N- and O-glycoproteins are present in all domains of life. In most bacterial species known to synthesize glycoproteins, glycosylation is restricted to a few proteins including adhesins, flagellins or pilins [15]. Only a few “general” glycosylation systems in which more than a single protein is glycosylated have been characterized. C. jejuni N-glycosylates more than 65 proteins with the same heptasaccharide. Inactivation of the glycosylation pathway does not have an effect on growth in vitro, but does reduce adhesion and invasion to cells in culture, and affects chicken and mice colonization [41]. Neisseria gonorrhoeae is able to O-glycosylate at least 12 proteins with a highly variable glycan structure [42]. The glycan has recently been shown to be important for infection of cervical epithelial cells [43]. Bacteroides fragilis also has a general O-glycosylation system, where hundreds of proteins are predicted to be glycosylated [44]. Inactivation of the glycosylation system results in severe growth defects in vitro [17]. It was then not surprising to see that the glycosylation mutant strain was outcompeted by the wild-type strain in gnotobiotic mice colonization experiments. Seven proteins are shown to be O-glycosylated by the PglL OTase encoded by the A1S_3176 gene. Cells unable to perform protein glycosylation do not show any differential growth phenotype in vitro, while exhibiting a diminished capacity to form biofilms and reduced virulence in D. discoideum, G. mellonella, and murine septicemia pathogenesis models systems.
Two glycoproteins were identified using 2D-DIGE. To our knowledge, this is the first time this technique is applied to study bacterial glycoproteomics. The structure of the glycan used to decorate these proteins in A. baumannii was determined by a combination of MS and NMR techniques. The sugar was determined to be a pentasaccharide of the formula β-GlcNAc3NAcA4OAc-4-(β-GlcNAc-6-)-α-Gal-6-β-Glc-3-β-GalNAc-S/T (Fig S4). The glycan contains a terminal O-acetylated glucuronic acid derivative that is negatively charged and has not previously been described. A similar monosaccharide was found in Pseudomonas aeruginosa and Bordetella pertussis [45]. Of the glycoproteins identified, only one (A1S_1193; MotB) has any significant homology outside of the genus Acinetobacter, with the remaining being annotated as hypothetical proteins. MotB has homology with proteins such as Pal from Haemophilus influenzae that have been shown to bind to peptidoglycan and stabilize the outer membrane [46]. Functional characterization of A. baumannii glycoproteins will be crucial to explain the phenotypes associated with lack of glycosylation.
Biofilms are proposed to be a virulence factor that is associated with increased antibiotic resistance, pathogenicity, and persistence of a bacterial population [47]–[49]. We have found that O-glycosylation enhances biofilm formation by A. baumannii ATCC 17978. Biofilm formation is a multistep process that involves an initial weak association leading to an irreversible attachment, which leads eventually to a complex maturation into sophisticated superstructures [50]. We observed by flow cell and confocal imaging that glycosylation enhances the initial attachment as well as mature biofilm mass and density. It is tempting to speculate that glycans of the glycoproteins may have a function in cell-to-cell adhesion [51]. Further work will elucidate in which aspect protein glycosylation is required for efficient biofilm formation.
The basic mechanisms of phagocytic cells are used in both amoebae and mammalian macrophages. As an infection model, the amoebae D. discoideum is considered a primitive macrophage. D. discoideum cells were unable to predate on A. baumannii WT lawns, but were able to efficiently predate on lawns of the glycosylation-deficient bacteria. It is uncertain how protein O-glycosylation protects A. baumannii from D. discoideum but we can hypothesize that glycosylation may help in the inhibition of phagocytosis by the amoebae, and/or prevent bacterial lysis by reactive oxygen species produced by the amoebae [52]. Another possibility is that glycosylation of certain proteins is required to interfere with bacterial degradation and intracellular vesicle transport and/or fusion, as shown for Legionella [53]. We also analyzed if protein O-glycosylation plays a role in pathogenesis in G. mellonella caterpillars. This model system has been recently shown to recreate the mammalian humoral immune system, with similar antimicrobial peptides, toll-like receptors, and the complement-like mechanism of melanization [54]. Similar to the D. discoideum model, A. baumannii ΔpglL strain was unable to kill G. mellonella. O-glycosylation could mediate killing of the larvae by stabilizing the bacterial outer membrane of A. baumannii, which could prevent killing by antimicrobial peptides. The negative charges of the glycan chains could play a role in this process. Alternatively, glycosylation could mask signals detected by the larvae or prevent phagocytosis by G. mellonella haemocytes, among other possibilities. The involvement of glycoproteins in virulence is further supported by the demonstration that the ΔpglL strain is outcompeted by wild type bacteria in a murine septicemia model. Thus, our experiments showed that glycosylation is critical for virulence in three different model systems. Further work using strains carrying mutations in individual glycoproteins will help to elucidate the exact role of protein glycosylation in pathogenesis.
Glycoproteins are usually immunodominant in bacteria, and therefore, the glycoproteins identified in this study may be the base of future vaccine formulations and diagnostic methods. The prevalence of the O-glycosylation machinery in Acinetobacter sp., together with its role in virulence in the three different pathogenesis models, suggest that protein O-glycosylation represents a novel target for the development of antibiotics that could be key to prevent further dissemination of this emerging human pathogen, which has become a major threat to our healthcare systems.
The bacterial strains and plasmids used in this study are listed in Table 3. A. baumannii strains were grown in Luria Bertani broth/agar at 37°C. The antibiotics ampicillin (Ap) 100 µg/mL, gentamicin (Gm) 50 µg/mL, and tetracycline (Tc) 5 µg/mL were added for selection as needed.
In order to create a ΔpglL via homologous recombination, we cloned a ∼3500 bp fragment consisting of ∼1000 bp upstream and downstream of A1S_3176 into pEXT20 using primers K/O pglLfwd and K/O pglLrev from A. baumannii ATCC 17978 genomic DNA (Table 3). The construct was subsequently subcloned from pEXT20 into pFLP2. We then digested pFLP2-pglL with PsiI and replaced A1S_3176 with a SmaI excised Gentamicin resistance cassette (aacC1) from pSPG1 [55]. The plasmid pFLP2 does not replicate in A. baumannii ATCC 17978. This final construct was transformed into electro-competent A. baumannii WT cells and selection for a single recombination event was analyzed using media supplemented with gentamicin. Positive colonies were grown in 5 mL LB at 37°C for 72 hours, with 1/1000 re-inoculations into fresh LB media every 24 hour period. After 72 hours, the liquid culture was plated on LB agar supplemented with gentamicin and 10% sucrose to select for a double recombination event. Colony PCR using both internal and external primers showed the allelic exchange of A1S_3176 with aacC1, generating a knockout mutant of A. baumannii pglL.
Bacterial cultures were pelleted by centrifugation for 15 mins at 10,000×g, washed with PBS, resuspended in PBS, and subsequently lysed by French Press. Unbroken cells were pelleted by centrifugation for 15 mins @ 5,000×g. The supernatant was ultracentrifugated for 1 hr @ 100,000×g (4°C) to pellet cell membrane. Samples were quantified by Bradford protein quantification (Biorad) and analyzed on a 12% SDS-PAGE. The PAS stain protocol used was previously described [56].
LPS was extracted according to Marolda et al [57]. Samples were resuspended in 50 µL of dH20 and analyzed by Silverstain on a 15% SDS-PAGE.
Lipid-free membranes were obtained for 2D-DIGE analysis according to [27]. The material was resuspended in: 6.5 M Urea, 2.2 M thiourea, 1% w/v ASB-14, 5 mM Tris-HCl pH 8.8, 20 mM DDT, 0.5% IPG buffer. The samples were labeled using CyDye minimal labeling protocol (Amersham Biosciences). A. baumannii WT membranes were labeled with Cy5 and ΔpglL were labeled with Cy3. Samples were quantified by 2D-Quant kit (GE Healthcare) and 600 µg of each WT and ΔpglL membranes were mixed in Destreak solution (GE Healthcare) to a final volume of 450 µL. 24 cm pH 3–11 NL IPG strips were simultaneously rehydrated and sample loaded for 24 hrs at room temperature in the dark. Isoelectric focusing was done using the Ettan IPGphor system for a total of 56,000 Vhr in the dark. The strip was then incubated in 10 mL of equilibration solution (2% SDS, 50 mM Tris-HCl, 6 M Urea, 30% (v/v) glycerol, 0.002% bromophenol blue) for 15 mins with 100 mg DTT and then 10 mL equilibration solution with 250 mg iodoacetamide. The strip was then sealed into a DALT 12.5 precast gel with 0.5% agarose. The system was run at 2.5 W/gel for 30 mins, the 17 W/gel until the dye front exited the bottom. The gel was visualized using FLA-5000 (FujiFilm) and the images analyzed by ImageQuant 5.0. The gel was subsequently stained with Coomassie brilliant blue, and individual spots excised and prepared for mass spectrometry.
Samples were in gel tryptically-digested and the peptides were desalted using C18 Zip-Tips and eluted with 60% CH3CN/40% H2O. Samples were spotted on a Bruker Daltonics MTP ground steel or Bruker Daltonics MTP AC600 Anchorchip target plate and air dried. 1 µL for ground steel and 0.4 µL for the AC600 target of 2,5-dihydroxybenzoic acid (DHB, 10 mg/mL in 30% H2O, 70% CH3CN) was spotted on top and allowed to dry. Mass spectra were obtained in the positive mode of ionization using a Bruker Daltonics (Bremen, GmbH) UltrafleXtreme MALDI TOF/TOF mass spectrometer. The FlexAnalysis software provided by the manufacturer was used for analysis of the mass spectra.The MS/MS spectra were obtained manually. The exact m/z used as the precursor m/z for MS/MS was determined first on a Bruker Daltonics (Billerica, MA) Apex Qe MALDI FTICR MS instrument and the MS/MS spectrum was automatically re-calibrated based upon this m/z.
Lipid free membrane extracts were digested for 72 hrs at 37°C with 2 µL Pronase E (20 mg/mL) being freshly added every 24 hrs. Glycosylated peptides were enriched using Active Charcoal Micro SpinColumn (HARVARD Apparatus) Briefly, the column was prewashed 3× with 400 µL of 0.1% TFA in of 80% ACN and 20% ddH2O and centrifuged at 500 RCF for 2 minutes. The column was equilibrated 3× using 400 µL of H2O. The sample was loaded 3× at 500 RCF for 2 minutes. The column was washed 2× with 200 µL of ddH2O at 500 RCF for 2 minutes. The glycan was eluted 3× with 100 µL 0.1% TFA in 50% ACN and 50% H2O at 1000 RCF for 2 minutes. The sample was dried by vacuum centrifugation and analyzed by MALDI-TOF/TOF MS and MS/MS.
For NMR analysis glycoproteins were digested with a large excess of proteinase K at pH 8 (adjusted by addition of ammonia) at 37°C for 48 hours. Products of digestion or free oligosaccharides were separated on Sephadex G-15 column (1.5×60 cm) and each fraction eluted before salt peak was dried and analyzed by 1H NMR. Fractions containing desired products were separated by anion exchange chromatography on Hitrap Q column (5 mL size, Amersham) and glycan eluted with a linear gradient of NaCl (0–1 M, 1 h). Desalting was performed on Sephadex G15 prior to analysis by NMR.
NMR experiments were carried out on a Varian INOVA 600 MHz (1H) spectrometer with 3 mm gradient probe at 25°C with acetone internal reference (2.225 ppm for 1H and 31.45 ppm for 13C) using standard pulse sequences DQCOSY, TOCSY (mixing time 120 ms), ROESY (mixing time 500 ms), HSQC and HMBC (100 ms long range transfer delay). AQ time was kept at 0.8–1 sec for H-H correlations and 0.25 sec for HSQC, 256 increments was acquired for t1. Assignment of spectra was performed using Topspin 2 (Bruker Biospin) program for spectra visualization and overlap. Monosaccharides were identified by COSY, TOCSY and NOESY cross peak patterns and 13C NMR chemical shifts. Aminogroup location was concluded from high field signal position of aminated carbons (CH at 45–60 ppm). Connections between monosaccharides were determined from transglycosidic NOE and HMBC correlations.
Dried membrane protein-enriched fractions were resuspended in 6 M urea, 2 M thiourea, 40 mM NH4HCO3. Samples were reduced, alkylated, digested with Lys-C (1/200 w/w) and then trypsin (1/50 w/w) as described previously [20]. Digested samples were then dialyzed against ultra-pure water overnight using a Mini Dialysis Kit with a molecular mass cut off of 1000 Da (Amersham Biosciences, Buckinghamshire, UK) and on completion were collected and lyophilized. ZIC-HILIC enrichment was performed according to [20] with minor modifications. Micro-columns composed of 10 µm ZIC-HILIC resin (Sequant, Umeå, Sweden) were packed into P10 tips on a stage of Empire C8 material (Sigma) to a bed length of 0.5 cm and washed with ultra-pure water prior to use. Dried digested samples were resuspended in 80% acetonitrile (ACN), 5% formic acid (FA) and insoluble material removed by centrifugation at 20,000×g for 5 min at 4°C. Samples were adjusted to a concentration of 2 µg/µL and 100 µg of peptide material loaded onto a column and washed with 10 load volumes of 80% ACN, 5% FA. Peptides were eluted with 3 load volumes of ultra-pure water into low-bind tubes and concentrated using vacuum centrifugation.
ZIC-HILIC fractions were resuspended in 0.1% FA and loaded onto a Acclaim PepMap 100 µm C18 Nano-Trap Column (Dionex Corporation, Sunnyvale, CA) for 10 min using a UltiMate 3000 intelligent LC system (Dionex Corporation). Peptides were eluted and separated on 20 cm, 100 µm inner diameter, 360 µm outer diameter, ReproSil – Pur C18 AQ 3 µm (Dr. Maisch, Ammerbuch-Entringen, Germany) in house packed column. Enriched peptides derived from tryptic digests were analysed using an LTQ-Orbitrap Velos (Thermo Scientific, San Jose CA). Samples were eluted using a gradient from 100% buffer A (0.5% acetic acid) to 40% buffer B (0.5% acetic acid, 80% MeCN) over 120 mins at a constant flow of 200 nL/min enabling the infusion of sample in the instrument using ESI. The LTQ-Orbitrap Velos was operated using Xcalibur v2.2 (Thermo Scientific) with a capillary temperature of 200°C in a data-dependent mode automatically switching between MS ion trap CID and HCD MS-MS. For each MS scan, the three most abundant precursor ions were selected for fragmentation with CID, activation time 30 ms and normalized collision energy 35, followed by HCD, activation time 30 ms and normalized collision energy 45. MS resolution was set to 60,000 with an ACG of 1e6, maximum fill time of 500 ms and a mass window of m/z 600 to 2000. MS-MS fragmentation was carried out with an ACG of 3e4/2e5 for CID/HCD and maximum fill time of 100 ms/500 ms CID/HCD. For HCD events an MS resolution of 7500 was set. A total of six HILIC enrichments were performed and analysis by the above protocol.
Raw files were processed within Proteome Discover version 1.0 Build 43 (Thermo Scientific) to generate .mgf files. To identify possible glycopeptides within exported scans, the MS-MS module of GPMAW 8.2 called ‘mgf graph’ was utilized. This module allowed the identification of all scan events within the generated .mgf files containing the diagnostic oxonium m/z 301.10 ion. These scan events were manually inspected and identified as possible glycopeptides based on the presence of the deglycosylated peptide ion with a tolerance of 20 ppm. To facilitate glycopeptide assignments from HCD scan events, ions below the mass of the predicted deglycosylated peptides were extracted with Xcalibur v2.2 using the Spectrum list function. Ions with a deconvoluted mass above the deglycosylated peptide mass and ions corresponding to known carbohydrate oxonium ions such as 204.08 and 366.14 were removed in a similar approach to post-spectral processing of ETD data [58], [59]. MASCOT v2.2 searches were conducted via the Australasian Proteomics Computational Facility (www.apcf.edu.au) with the Proteobacteria taxonomy selected. Searches were carried out with a parent ion mass accuracy of 20 ppm and a product ion accuracy of 0.02 Da with no protease specificity, instrument selected as MALDI-QIT-TOF (use of this instrumentation setting was due to the observation of multiple internal cleavage products, extensive NH3 and H2O loss from a, b, y ions, which are all included within this scoring setting) as well as the fixed modification carbamidomethyl (C) and variable modifications, oxidation (M) and deamidation (N). An ion score cut-off of 20 was accepted and all data were searched with the decoy setting activated generating a zero false positive rate generated against the decoy database.
Cultures were grown overnight and re-inoculated at an OD600 0.05 in 100 µL into replicates in a 96 well polystyrene plate (Costar). The cultures were subsequently grown without shaking for 48 hours at 30°C. Bacterial growth was determined by measuring the absorbance at OD600 nm. The cultures were removed and the wells washed with ddH20, followed by the addition of 100 µL of 1% crystal violet in ethanol to stain the cells. The plate was incubated for 30 mins with gentle agitation, then thoroughly washed with ddH20, and the stained biofilms solubilized with 100 µL of 2% SDS for 30 minutes with gentle agitation. The amount of biofilm formed was quantified by measuring the absorbance at OD580 nm. The data was normalized using the ratio between OD580/OD600.
Flow cell experiments and fluorescent staining were performed as described previously by Seper et al. [60]. Briefly, the respective overnight cultures were adjusted to OD600 = 0.1 using 50-fold diluted LB (2%). Per channel, approximately 250 µl of the dilutions were inoculated. After static incubation for 2 h, flow of pre-warmed 2% LB (37°C) was initiated (3 ml/h). Biofilms were allowed to form for a time period of 24 h and were stained with SYTO 9 (Invitrogen) for visualization. Images of attached bacteria or biofilms were acquired using a Leica SP5 confocal microscope (Leica Microsystems, Mannheim, Germany) with spectral detection and a Leica HCX PL APO CS 40× oil immersion objective (NA 1.25). For the SYTO 9 signal, the excitation wavelength was set at 488 nm and fluorescence emission was detected between 500–530 nm. Optical sections were recorded in 0.2 µm steps. For two-dimensional image visualization the Leica LAF and for three-dimensional image processing the AMIRA software (direct volume rendering with VOLREN module) was used. Quantification of image stacks was performed using COMSTAT (http://www.comstat.dk) [61] (M. Vorregaard et al., pers. comm.). For COMSTAT analysis at least six image stacks from three independent experiments were used.
This assay was performed essentially as described by [62]. Briefly, midlogarithmic cultures of D. discoideum were mixed with overnight cultures of bacteria to a final concentration of 1×103 amoebae ml−1. 0.2 ml of the suspension was then plated on SM/5 agar containing 1% ethanol. Plates were incubated at room temperature and monitored for D. discoideum plaques for 3–5 days. Wild type bacteria are toxic to the amoebae. Appearance of plaques indicates attenuation.
This assay was performed as previously described [32]. Galleria mellonella larvae were bred in sterile conditions at 37°C by Dr. Andrew Kedde (University of Alberta). After injection of bacteria, caterpillars were incubated at 37°C, and the number of dead caterpillars was scored every 5 hours. Caterpillars were considered dead when they were nonresponsive to touch. This experiment is a representative of 3 biological replicates.
A murine model of disseminated sepsis using BALB/c mice (16–20 grams) was used for bacterial challenge [63], [64]. A. baumannii strains were grown for 18 h at 37°C in Luria broth with appropriate antibiotics and adjusted to the appropriate concentration in physiologic saline. Inoculums were prepared by mixing the bacterial suspensions 1∶1 (v∶v) with a 10% solution (w/v) of porcine mucin (Sigma, St. Louis, MO) which increases the infectivity of A. baumannii, allowing for a lower concentration of bacteria to be used [65]–[67]. Mice were injected intraperitoneally with 0.2 ml of the bacterial/mucin inoculums. Bacterial concentrations were determined by plating dilutions on Luria agar. The wild type strain lethal dose for 50% of animals was determined by the limit test where groups of 5 mice were infected with dilutions of bacteria, at a range of concentrations within 2 logs of a concentration of bacteria that had previously been shown to be lethal with this species of bacteria using a disseminated sepsis model.
An in vivo competition assay was used to compare fitness between the wt and ΔpglL strains [37]–[39]. Liquid cultures containing individual strains were diluted and plated on LB agar. Mixed inoculums were established by mixing equal proportions of strains based on the OD600. Once mixed the inoculums were serially diluted and plated on LB agar and LB agar with gentamycin to select for the ΔpglL. The expected ratio of CFU on LB compared to CFU on LB with gentamycin was 2∶1.
For bacterial competition experiments in vivo an animal model of sepsis was used. Groups of 3 BALB/c female 16–20-g mice were inoculated intraperitoneally with 1×105 CFUs of mixed inoculums (50% of each strain). Groups of 3 mice were sacrificed at 18 h after inoculation. Mice at 18 hours of infection were showing clinical signs of illness and were often moribund. Spleens were aseptically removed, weighed, and homogenized via passage through a cell strainer (BD falcon 70 um cell strainer) in physiological saline before plating serial log dilutions on Luria agar plates for bacterial quantification. If the two strains had equal fitness in vivo the ratio established prior to infection should be maintained.
All procedures and experiments involving animals (mice) were approved by the Institutional Animal Care Committee of Defence Research and Development Canada Suffield (protocol # CWS-08-1-1-1), and were in accordance with guidelines from the Canadian Council of Animal Care.
|
10.1371/journal.pcbi.1005182 | WORMHOLE: Novel Least Diverged Ortholog Prediction through Machine Learning | The rapid advancement of technology in genomics and targeted genetic manipulation has made comparative biology an increasingly prominent strategy to model human disease processes. Predicting orthology relationships between species is a vital component of comparative biology. Dozens of strategies for predicting orthologs have been developed using combinations of gene and protein sequence, phylogenetic history, and functional interaction with progressively increasing accuracy. A relatively new class of orthology prediction strategies combines aspects of multiple methods into meta-tools, resulting in improved prediction performance. Here we present WORMHOLE, a novel ortholog prediction meta-tool that applies machine learning to integrate 17 distinct ortholog prediction algorithms to identify novel least diverged orthologs (LDOs) between 6 eukaryotic species—humans, mice, zebrafish, fruit flies, nematodes, and budding yeast. Machine learning allows WORMHOLE to intelligently incorporate predictions from a wide-spectrum of strategies in order to form aggregate predictions of LDOs with high confidence. In this study we demonstrate the performance of WORMHOLE across each combination of query and target species. We show that WORMHOLE is particularly adept at improving LDO prediction performance between distantly related species, expanding the pool of LDOs while maintaining low evolutionary distance and a high level of functional relatedness between genes in LDO pairs. We present extensive validation, including cross-validated prediction of PANTHER LDOs and evaluation of evolutionary divergence and functional similarity, and discuss future applications of machine learning in ortholog prediction. A WORMHOLE web tool has been developed and is available at http://wormhole.jax.org/.
| Identifying functionally equivalent proteins between species is a fundamental problem in comparative genetics. While orthology does not guarantee functional equivalence, the identification of orthologs—genes in different organisms that diverged by speciation—is often the first step in approaching this problem. Many methods are available for predicting orthologs. Recent approaches combine methods and filter candidate predictions by “voting”—assigning confidence to ortholog pairs based on the number of predictions by independent methods. Although voting is a heuristic, it maintains precision while increasing recall. Here we employ machine learning to optimize voting by learning which methods make better predictions and, in essence, giving those methods more votes. We present a new tool called WORMHOLE that predicts a strict subclass of orthologs called least diverged orthologs (LDOs) with a high level of functional specificity by learning features of orthology that are encoded in the patterns of predictions made by 17 constituent methods. We validate WORMHOLE using multiple measures of evolutionary divergence and functional relatedness, including community standards provided by the Quest for Orthologs consortium. WORMHOLE’s particular strength lies in predicting LDOs between distantly related species, where orthology is difficult to identify and is of critical importance for comparative biology.
| Comparative biology has become a central strategy in the study of human biology and disease. The availability of powerful genetic tools and our ability to control experimental conditions in model organisms often allows a much more detailed examination than directly studying a process of interest in humans. In diverse areas of biology—aging, development, stem cell differentiation, behavior—highly conserved molecular features have been described in model systems, even highly evolutionarily divergent organisms, and translated into useful interventions in humans. For example, the ability to delay aging by inhibition of the Target of Rapamycin (TOR) kinase was first discovered in the single-celled budding yeast Saccharomyces cerevisiae, and much of the work to characterize TOR signalling has been carried out in this system (reviewed by Loewith and Hall [1]). Reduced TOR signalling has since been demonstrated to increase lifespan in a range of model systems from worms to mice (reviewed by Cornu et al. [2]). Rapamycin and other drugs targeting this system are now in clinical trials for cancer [3,4] and show promise for other age-associated diseases, including Alzheimer’s disease [5]. Aging is a particularly salient example demonstrating the power of comparative biology. Lifespan studies are much shorter, much less expensive, and therefore much more tractable in invertebrate species than in vertebrates, allowing aging studies to be carried out more rapidly, on a larger scale, and in greater molecular detail for the same resource investment. To reap the practical benefits of invertebrate models in studying the genetics of human health, it is crucial to translate molecular results from invertebrates into vertebrates.
A vital step in this translation is the identification of the gene or protein that fills the functionally equivalent role in the target vertebrate species. Since functionally equivalent proteins (FEPs) are difficult to predict directly, the most commonly used surrogate is orthology. Orthologs are genes that derive from the most recent common ancestral gene by speciation (in contrast to paralogs; genes that derive from the most recent common ancestral gene by duplication) [6]. Because orthology is defined by speciation, the evolutionary history separating orthologous genes may include other categories of evolutionary event, such as duplication, deletion, and de novo mutation in one or both lineages after the defining speciation event. In addition to simple one-to-one mappings, these evolutionary processes allow for one-to-many and many-to-many mappings between genes that define an orthologous group in different species. The boundaries between orthologs and non-orthologs can be difficult to discriminate based on readily measured features of genes, such as sequence composition, leading to a difficult bioinformatics problem. A subset of all orthologs are the least diverged orthologs (LDO), defined as the pair of genes within an ortholog group for two species that have accumulated the fewest mutations after speciation and duplication-post-speciation events (i.e. have ‘diverged the least’) [7]. The identification of LDOs is a sub-problem of the ortholog identification, but its solution has many desirable properties. In particular, the gene pair in an ortholog group with the least sequence divergence is the most likely to have been functionally conserved by evolution [8,9]. More divergent gene pairs are more likely to have developed novel function, particularly in gene families that have undergone numerous duplication events. In this study we focused specifically on the identification of LDOs. The idea that orthologous genes tend to be more functionally similar than non-orthologous genes is called the “ortholog conjecture”, which states specifically that orthologs are more functionally similar than paralogs. There has been recent debate surrounding this conjecture. Contrary to the ortholog conjecture, Nehrt et al. [10] found that paralogs within either humans or mice were more predictive of gene function than orthologs between humans and mice based on comparison of microarray and gene ontology (GO) data, suggesting that cellular context, rather than shared sequence, may be the primary driver of functional evolution. However, bias in GO annotations tends to favor functional similarity between paralogs [11], and subsequent studies using RNA-seq data [8] or bias-corrected GO annotations [9] support the ortholog conjecture. Specifically, Chen and Zhang [8] found that gene expression similarity between orthologs is significantly higher than between paralogs across multiple tissue types, while Altenhoff et al. [9] found that functional GO annotation similarity was higher between orthologs than paralogs, and increased weakly, but significantly, with decreased sequence divergence, even across large evolutionary distance, when the GO annotations were controlled for common biases. Thus, while orthologs and FEPs are conceptually distinct, the preponderance of evidence suggests that they are related, and in particular that identifying an ortholog as a first step toward identifying an FEP is warranted. Because protein sequence ultimately determines function, the LDO—the ortholog with the least divergence in sequence—is therefore a strong estimate of an FEP. Likewise, observing high functional similarity between genes in different species provides evidence for, but does not guarantee, shared evolutionary history.
The past decade has seen an explosion of new methodologies and tools designed to predict orthologous genes between two or more species. The majority use one of two approaches: graph-based or tree-based ortholog prediction. Graph-based algorithms begin with pairwise alignments between all protein sequences from two species to estimate evolutionary distance between each protein pair, followed by orthology prediction made using a range of clustering criteria: reciprocal best hit (e.g. OMA [12], OrthoInspector [13], and InParanoid [14]), reciprocal smallest distance (e.g. Roundup [15]), best triangular hit (e.g. COG [16] and EggNOG [17]), or Markov clustering (e.g. OrthoMCL [18]). Tree-based systems take advantage of our understanding of evolutionary relationships between species, using simultaneous alignment of sequences from many species to build phylogenetic trees and infer orthology relationships based on tree structure. Variations on this approach are employed by many popular ortholog prediction tools: Ensembl Compara [19], metaPhOrs [20], OrthoDB [21], PANTHER [22], and TreeFam [23]. Other strategies (e.g. HomoloGene [24] and Hieranoid [25]) combine aspects of both graph- and tree-based systems, progressively applying graph-based methods at the nodes of a species tree to generate more accurate ortholog predictions while maintaining the computational efficiency inherent to tree-based methods. A further alternative strategy is to directly identify genes in a target system that fills a functionally equivalent role. For example, the Isobase algorithm infers FEPs using both sequence information and functional information encoded in protein-protein interaction (PPI) networks.
Each prediction algorithm uses a different methodology, producing overlapping but distinct sets of predicted orthologs or FEPs and displaying different strengths and weaknesses in terms of performance for the particular objective of that algorithm. Several groups have combined predictions from multiple sources in “meta-tools” to improve prediction performance. Shaye and Greenwald [26] created OrthoList, a set of human-worm orthology relationships, by simply combining the predictions from four commonly used ortholog prediction tools (InParanoid, OrthoMCL, Homologene, and Ensembl Compara) to produce a system with high recall (i.e. low false negative rate) while maintaining precision (i.e. low false positive rate) when tested on a manually curated set of human-worm ortholog pairs. MetaPhOrs was constructed by collecting phylogenetic trees from seven independent sources (PhylomeDB, Ensembl, TreeFam, Fungal Orthogroups, EggNOG, OrthoMCL, and COG) and applying a common algorithm to select orthologs between species, allowing improved ortholog prediction accuracy based on cross-tree comparison [20]. The Drosophila RNAi Screening Center Integrative Ortholog Prediction Tool (DIOPT) reports predictions from eight ortholog databases (Ensembl Compara, Homologene, InParanoid, OMA, OrthoMCL, PhylomeDB, RoundUp, and TreeFam) and one functional database (Isobase) between six species (human, mouse, zebrafish, fruit fly, nematode, and budding yeast) and includes a confidence score based on the number of algorithms predicting each pair, and a weighted score that takes into account functional similarity based on GO term comparison [27]. The recently published Multiple Orthologous Sequence Analysis and Integration by Cluster optimization (MOSAIC) combines ortholog predictions generated by four methods (Multiparanoid, Threshold Block Aligner (TBA), six-frame untranslated BLAST-like alignment tool (BLAT), and OMA) and applies a filtration process to optimize pairwise alignment between members of each ortholog cluster [28]. Pereira et al. developed Meta-Approach Requiring Intersections for Ortholog predictions (MARIO) to aggregate four ortholog prediction methods (reciprocal best hit, InParanoid, OrthoMCL, and Phylogeny [29]) to identify high-specificity ortholog groups that were then analyzed by multiple sequence alignment and hidden Markov models to predict novel orthologs [30]. In each case, the meta-tool is shown to improve prediction performance when compared to the individual input algorithms. To date, all of these methods use the number of algorithms that predict an ortholog as a heuristic to determine the confidence of a given prediction. However, while some use sophisticated post-processing to improve performance, none take into account the individual performance of each input algorithm when assigning confidence levels to aggregate predictions.
Here we present a novel strategy in this final category of meta-tools. The WORM-Human OrthoLogy Explorer (WORMHOLE) predicts LDOs between species by employing machine learning to differentially weight the output of 17 ortholog prediction strategies. WORMHOLE falls into a subcategory of meta-tools that do not predict orthology de novo (others in this category include OrthoList and DIOPT), but rather integrate information from multiple sources to refine and extend predictions. Originally developed to identify orthologous genes between humans and nematodes, we have expanded the method to include six species: Homo sapiens (humans), Mus musculus (mice), Danio rerio (zebrafish), Drosophila melanogaster (fruit flies), Caenorhabditis elegans (nematodes), and Saccharomyces cerevisiae (budding yeast). WORMHOLE considers the patterns of ortholog calls of the 17 constituent algorithms and identifies signature patterns that correspond to likely LDOs. Specifically, WORMHOLE uses the genome-wide predictions of LDOs from PANTHER (PANTHER LDOs) as a set of high-confidence examples to train machine learning classifiers.
PANTHER makes de novo predictions of LDOs based on evolutionary relationships. We expect that rigorous statistical criteria used by any de novo method will necessarily miss some true LDOs, particularly in edge cases with difficult-to-parse evolutionary history or patterns of sequence divergence (e.g. duplication-post-speciation events in both lineages). Machine learning provides a principled method to extend de novo predictions with new data. We used the PANTHER LDOs to define positive and negative examples, but reserved judgment on genes for which PANTHER does not identify an LDO. The machine learning classifier then identified a “signature” of LDO vs. non-LDO status from the PANTHER LDO examples that can be used to infer LDO status for previously unclassified genes. WORMHOLE provides rigorous confidence scores based on how strongly the pattern corresponds to the known PANTHER LDOs. We present six findings: 1) The patterns of ortholog calls by the 17 constituent algorithms contain sufficient information to strongly predict LDOs in the reference set. This is non-trivial because, as discussed below, none of the input algorithms are designed to explicitly predict LDOs. Nevertheless they encode LDO status in the patterns of their respective ortholog predictions. 2) The use of support vector machine classifiers (SVMs) strongly improves LDO prediction over simple voting, a baseline method used in other meta-tools. 3) This enhanced prediction depends on the evolutionary distance between organisms with greater improvement for distant comparisons, e.g. between vertebrates and invertebrates. 4) The WORMHOLE SVMs expands the number of LDOs relative to the PANTHER LDO training set. The novel LDOs maintain a similar evolutionary distance distribution and Basic Local Alignment Search Tool protein (BLASTp) alignment score to the PANTHER LDO training set, indicating that the novel predictions are indeed LDOs. 5) The WORMHOLE models trained on one pair of species generalize well to other species pairs, suggesting that the WORMHOLE models are identifying information about orthology in general, and not just between particular species pairs. 6) The WORMHOLE predictions have high functional specificity by several criteria, while making significantly more LDO calls than the PANTHER LDOs used to train the models. This indicates that WORMHOLE has extracted functionally relevant information from the constituent algorithms that is complementary to the PANTHER LDOs.
Most novel ortholog prediction strategies seek to increase performance by expanding the scope or improving the quality of the underlying sequence data, or through application of a new algorithm. The wealth of ortholog prediction strategies now available opens the possibility of a two-layer prediction model. To conceptualize this model, consider the individual pieces of underlying biological and genetic information—gene and protein sequences, gene and protein interactions, phylogenetic relationships between species—as first-order features (Fig 1A). Each of the established ortholog prediction algorithms (Ensembl Compara, EggNOG, etc.) uses different combinations of these first-order features to generate predicted ortholog relationships, forming the first layer of prediction (Fig 1B). These algorithms generate a pool of candidate ortholog predictions, and hence candidate LDOs, that can be considered novel second-order features (Fig 1C). In WORMHOLE, we apply a second layer of prediction to refine these candidate ortholog predictions to directly predict LDOs (Fig 1D). This refinement is accomplished by generating a confidence score for each gene pair based on the pool of predictions and considering only those pairs that meet a minimum confidence threshold.
This multilayer approach requires three ingredients: (1) genome-wide candidate ortholog predictions (i.e. second-order features) between the species of interest generated by a selected set of first-layer algorithms, (2) a second-layer algorithm to classify each gene pair as either an LDO or not based on the second-order features, and (3) a training dataset (reference set) composed of well-defined examples of both LDO and non-LDO gene pairs, which is used to train and test the second layer algorithm.
To generate a genome-wide candidate pool (ingredient 1), we collected all ortholog predictions from 17 constituent algorithms between the selected species, representing a wide array of different prediction strategies. There are more than 30 databases that predict orthologous or functional relationships between species using different methodologies. In selecting algorithms to include in WORMHOLE, we sought to sample as wide a variety of prediction strategies as possible. We examined each database that we were able to locate and access online and included the 13 data sources that met the following criteria: (1) the availability for download of complete genome-wide ortholog predictions, (2) current ortholog prediction data (updated since 2010), and (3) demonstrated performance in published literature. This set includes 5 graph-based strategies, 5 phylogeny-based strategies, 2 hybrid graph- and tree-based strategies, and 1 PPI network-based strategy (Table 1). Because some projects identify multiple categories of orthologs (e.g. EggNOG-COGs and EggNOG-KOGs), these 13 sources resulted in 17 predicted ortholog datasets (constituent algorithms). We assembled these predictions into a common database (the WORMHOLE database) and call these predicted orthologous gene pairs candidate LDOs (cLDOs).
For a second layer algorithm (ingredient 2), we trained SVMs using the predictions of the constituent algorithms. SVMs are machine learning classifiers that take as input a set of labelled examples and a set of ‘features’ describing the examples and builds a mathematical model of each class based on the relevant information within the features. In our case, we trained SVMs on known LDO and non-LDO pairs using the orthology predictions of the 17 constituent algorithms as features. To the SVM classifier, each cLDO is represented as a signature vector of binary calls by the constituent algorithms (e.g. ‘00011011101010110’) with each digit representing the prediction made by a specific algorithm (1 = predicts orthology; 0 = does not predict orthology).
The SVMs require a reference set of known LDOs and non-LDOs to use as training data (ingredient 3). A well-defined reference set should: (1) be representative of the entire set of “true” LDOs between the species considered, (2) include only high-confidence examples, and (3) include examples of both LDO and non-LDO gene pairs. We selected the PANTHER LDO dataset as the reference set for training the SVMs. PANTHER identifies orthologous gene pairs based on species structure within algorithmically constructed phylogenetic trees. PANTHER LDOs include all one-to-one orthologs and the single least divergent gene pair in one-to-many and many-to-many ortholog groups within the broader PANTHER ortholog dataset. PANTHER LDOs consistently perform well, generating conservative predictions (i.e. fewer, more closely related ortholog pairs) when compared to other ortholog datasets using the orthology benchmarking service provided by Quest for Orthologs (QfO), a consortium that provides community standards for developing and testing orthology prediction methodology (http://questfororthologs.org/) [31]. Because the PANTHER LDO set is conservative, we anticipate that it contains strong positive examples of LDOs and that we can identify gene pairs that appear “LDO-like” with additional information not available to PANTHER. We grouped each cLDO in the WORMHOLE database into one of three classes: 1) Known LDOs are cLDOs that are contained in the PANTHER LDO set. 2) Known non-LDOs are cLDOs for which one or both genes in the pair has a predicted ortholog in the PANTHER LDO set that is not the other gene in the cLDO pair (i.e. is a multiple mapping for which the cLDO is not the least diverged pair). 3) Unclassified cLDOs are cLDOs for which neither gene in the pair has a known LDO. We trained the SVMs using only the known LDOs and known non-LDOs and reserved the unclassified cLDOs for possible novel LDO identifications. These unclassified cLDOs are exactly the edge cases where PANTHER can potentially be extended.
We trained an independent SVM for each pair of query and target species using the predictions made by the 17 constituent algorithms as features and the PANTHER LDOs as a reference set for classification. As a baseline aggregation strategy to benchmark the SVM performance we used simple voting—a straightforward tally of the number of constituent algorithms that predicted a cLDO—and ranked cLDOs by their vote counts. We employed nested cross-validation to ensure that the SVM models were not overfitting the training data (see Materials and Methods). A summary of the number of genes, number of ortholog pairs, and genes with multiple ortholog mappings across species is provided in Table 2, and for each species combination in S1 Table.
As expected, the SVM models always outperformed the constituent algorithms and simple voting at predicting PANTHER LDOs in terms of precision (P, the fraction of predicted LDOs that are known LDOs) and recall (R, the fraction of known LDOs that are contained in the predicted LDOs) (Fig 2A and S1 Fig). This is because none of the constituent algorithms were designed to directly predict LDOs. The constituent algorithms display a wide range of performance at predicting PANTHER LDOs and none achieve as high performance as WORMHOLE at predicting PANTHER LDOs. While each algorithm performs well at the prediction task for which it was designed (e.g. prediction of orthologs from direct comparison of sequence, prediction of functional equivalence, identification of ortholog group with respect to a specific most recent common ancestor), the performance at predicting PANTHER LDOs depends on the similarity between PANTHER LDOs and the algorithm-specific design goal. PANTHER LDOs are a particularly conservative subset of ortholog predictions, and we observe that more conservative algorithms (e.g. Roundup) tend to achieve high precision and recall (Fig 2A and S1 Fig), while more permissive algorithms (e.g. eggNOG-KOGs; clusters of orthologs defined with respect to the most recent common ancestor, MRCA, for all eukaryotic species) tend to display high recall at the cost of low precision at PANTHER LDO prediction. PANTHER, by definition, has perfect recall of PANTHER LDOs (Fig 2A). The range of performance represented among algorithms is important, providing the SVM classifiers with a diverse set of features from which to discern “LDO-like” gene pairs and optimize LDO-prediction performance. The improved performance of WORMHOLE at predicting PANTHER LDOs demonstrates that WORMHOLE is able to consistently learn such structure, despite none of the constituent algorithms being designed to predict LDOs per se.
Identifying LDOs is of particular importance in distantly related species where evolutionary time has resulted in greater sequence divergence between orthologs, obscuring the lineal relationship between genes. In Fig 2 we examine the behavior of the SVMs as a function of the evolutionary distance between organisms. The set of species compared in WORMHOLE includes three vertebrate species (humans, mice, and zebrafish) and three invertebrate species (fruit flies, nematodes, and yeast). The three vertebrate species are substantially more closely related to each other than any vertebrate species to any invertebrate species, or any of the invertebrate species to one another. This allows LDO predictions to be grouped into those between closely related species (vertebrate-vertebrate) and more distantly related species (invertebrate-invertebrate and vertebrate-invertebrate). Fig 2A presents the performance of the SVM at predicting known LDOs as compared to each constitutive algorithm and simple voting. For each comparison the SVM has higher precision at every value of recall than simple voting or any of the constituent algorithms. Vertebrates are closely related evolutionarily; as a consequence the constituent algorithms already perform well and simple voting or the SVM yield only marginal improvement. This is ultimately due to the clarity of orthology relationships in closely related species; most orthologs are one-to-one mappings with relatively little sequence divergence. In contrast, the invertebrate species are each distantly related from each other and from the vertebrate species and the PR-curves show dramatic improvement in classification by the SVMs over voting and the constituent algorithms.
In order to normalize the outputs to make comparisons between groups, we scaled the output scores of the SVMs to the interval [0, 1] so that 0 and 1 represent the extremes of low and high prediction confidence, respectively (see Materials and Methods). We term the scaled confidence score the WORMHOLE Score. To allow direct comparison to our selected baseline, we similarly scaled the number of votes received by each algorithm to the Vote Score. A WORMHOLE or Vote Score of 0.5 is the point where the harmonic mean of precision and recall (F) is maximized. This point occurs at the “shoulder” of the PR-curve (Fig 2A) and denotes a convenient threshold of simultaneously high precision and recall. Fig 2B presents the range of F-values achieved by each constituent algorithm, simple voting, and the SVMs across species combinations. While simple voting generally outperforms the constituent algorithms, specific algorithms achieve greater performance in some cases, particularly when predicting LDOs between yeast and other species (S1 Fig). Indeed, the median F achieved by OrthoMCL between invertebrate species is 1.7% higher than simple voting (Fig 2B and S2 Table). In the vertebrate-vertebrate and vertebrate-invertebrate comparisons, simple voting achieves a median F 5.6% and 4.5% higher than the nearest constituent algorithm, respectively. In contrast to simple voting, the SVMs consistently outperform all constituent algorithms and simple voting, displaying median F 22.3%, 11.3%, and 1.4% higher than the nearest constituent algorithm at predicting LDOs between invertebrate-invertebrate, vertebrate-invertebrate, and vertebrate-vertebrate species, respectively (Fig 2B and S2 Table).
The ability of the SVM models to improve performance relative to voting appears dependent on the range of precision and recall represented in the underlying first-layer algorithms for a given species combination. Species combinations with little variation in recall in particular (e.g. human-to-zebrafish predictions, S1E Fig), result in little or no improvement in SVM performance over voting, while combinations with wide variation in both performance metrics see a much larger improvement from the SVM classifiers (e.g. human-to-worm predictions, S1E Fig).
As a measure of the generalizability of the WORMHOLE SVMs, we examined the ability of a model trained on one pair of species (e.g. human-worm) to predict orthologs between each other pair of species. While optimum performance was achieved when a model was trained and tested on the same species pair, performance was surprisingly consistent across species combination (Fig 3 and S3 Table). Two species combinations were an exception to this pattern. Models trained on human-mouse and, to a lesser extent, mouse-zebrafish reference LDOs displayed reduced performance relative to the other models when applied to predict LDOs in other species combinations. Humans and mice are the most closely related species examined and have the best annotated and least divergent ortholog datasets. We speculate that the relatively poor performance of human-mouse trained SVM models at predicting LDOs in other species is a result of the limited diversity in human-mouse ortholog prediction among constituent algorithms (S1E and S1F Fig), limiting the information available to the SVM classifiers about general orthology.
To further examine the relationship between models trained on different pairs of species, we next examined the variation in model parameters across species combinations. Each SVM is parameterized by a set of weights assigned to predictions made by each constituent algorithm that define the classifier (see Materials and Methods). While the weights differ across species pairs, the weight vectors are correlated (mean Pearson coefficient = 0.54, standard deviation = 0.21, Fig 4A), indicating that there are global trends for particular constituent algorithms to have high or low weight across species combinations. This trend is shown in Fig 4B. As expected, PANTHER receives the highest median weight. While the constituent algorithms were developed independently, all work from similar source data and many employ related strategies to predict orthologs. As a consequence, predictions between specific tools can be highly correlated. Providing prediction data from correlated algorithms introduces redundant information that results in over-representation in the case of simple voting. The SVMs respond to correlation by proportionally reducing the weight given to the predictions from correlated algorithms. For example, predictions made by Homologene and OMA are correlated (Jaccard index = 0.46, S4 Table). We speculate that this results in OMA receiving relatively low, sometimes even negative, weight, particularly in species combinations where Homologene/OMA predictions are not well suited to predicting PANTHER LDOs. Along the same lines, WORMHOLE considers predictions from metaPhOrs, which itself is a meta-predictor incorporating sequence data from several of the other WORMHOLE constituent algorithms. As expected, metaPhOrs predictions correlate well with most of these tools, including Ensembl Compara (Jaccard index = 0.37), TreeFam (Jaccard index = 0.35), and EggNOG-NOGs (Jaccard index = 0.29), while less strongly with others (OrthoMCL; Jaccard index = 0.13) (S4 Table). Higher weight is given to metaPhOrs than any of the three highly-correlated algorithms that represent metaPhOrs source data (Fig 4B), indicating that the WORMHOLE SVMs are accounting for the correlation in assigning weights.
WORMHOLE builds an image of what an LDO “looks like” by examining the PANTHER LDOs from the perspective of the amalgamated calls of the constituent algorithms. It then scans the collection of all cLDOs to identify novel gene pairs that fit that learned image. When applied across the genomes in question, we expect WOMRHOLE to capture an expanded set of LDOs that includes the majority of the PANTHER LDOs, as well as novel gene pairs. This is indeed what we observe (Fig 5A, Table 3). Importantly, WORMHOLE excludes a large portion of the broader PANTHER database that is not included in the PANTHER LDO set, removing the majority of the one-to-many and many-to-many gene-combinations. Importantly, the WORMHOLE classifier considers only the predictions made by the 17 constituent algorithms and is blind to the number of cLDOs corresponding to a specific query gene. As a consequence, WORMHOLE can generate multiple LDO predictions for a single query gene if there is sufficient evidence from the constituent algorithms. The number of query genes that generate multiple LDO predictions within a target species decreases as the WORMHOLE score threshold is increased (Fig 5B). Using a threshold of 0.5, WORMHOLE produces multiple LDO predictions for 12.4% of genes (Fig 5B and Table 2). Within the subset of genes with multiple LDO predictions, PANTHER LDOs receive higher WORMHOLE scores than gene pairs not in the PANTHER LDOs (Fig 5C), indicating that WORMHOLE predicts known LDOs with higher confidence than non-LDOs or novel LDOs.
To generate a high-confidence subset of the WORMHOLE LDOs that more closely matches the strict definition of an LDO, we identified WORMHOLE reciprocal best hits (RBHs). A WORMHOLE RBH is a predicted LDO with a WORMHOLE Score of at least 0.5 for which each gene in the pair receives the highest WORMHOLE Score when the other gene is queried (analogous to BLASTp RBHs). WORMHOLE RBHs are similar to PANTHER LDOs in that each gene in one organism will map to a single gene in the other organism. Comparing WORMHOLE RBHs to PANTHER LDOs, the WORMHOLE RBHs reproduce 81.7% of original PANTHER LDOs, but expand the total number of predicted LDOs by 17.7% (Table 3). This trend is reproduced for each comparison between vertebrates and invertebrates (Table 3). Note that in a small number of cases, multiple LDOs are predicted for a single query gene with identical WORMHOLE scores, preventing WORMHOLE from distinguishing a single RBH (Table 2). In these few cases, both predicted genes are included in the RBH category. When applied to predict PANTHER LDOs, the WORMHOLE RBHs produce similar performance to the unmodified WORMHOLE SVMs with a WORMHOLE Score of 0.75 or greater (Fig 2A).
By definition, the evolutionary divergence between genes in an LDO pair should be less than that between each gene in the pair and all other genes in the target genome. To evaluate the divergence of WORMHOLE LDOs and RBHs relative to PANTHER LDOs, we calculated evolutionary distance between all gene pairs for each species combination. We further examined alignment quality for each gene pair by calculating BLASTp bit scores. The set of all WORMHOLE LDOs and the set of novel LDOs predicted by WORMHOLE but not present in the PANTHER LDO training set both produce a similar distribution of evolutionary distance and bit score to the PANTHER LDOs (Fig 6). While the WORMHOLE SVMs are trained to predict LDOs based on the PANTHER LDOs, a subset of the PANTHER LDOs are excluded by the WORMHOLE SVMs. Gene pairs in this set of excluded PANTHER LDOs had markedly higher evolutionary distance and lower BLASTp bit scores than the WORMHOLE or PANTHER LDOs (Fig 6), indicating that the WORMHOLE SVMs specifically trimmed distantly related, low-confidence gene pairs from the PANTHER LDO dataset. A similar pattern was observed for WORMHOLE RBHs (S2 Fig). The percentage of WORMHOLE RBHs and PANTHER LDOs that identify the least evolutionarily distant gene is similar (Table 2, S1 Table). As expected, this percentage is lower for the broader category of all WORMHOLE LDOs that receive a minimum WORMHOLE Score of 0.5, which includes multiple LDO mappings for some genes (Table 2, S1 Table).
Orthology is an evolutionary concept and does not necessarily imply that a pair of genes will be functionally related. However, orthologous genes, and in particular LDOs, are often functionally similar or equivalent, and ortholog prediction is commonly used as a starting point for identifying the gene or genes in a new species that fill an equivalent functional role as a gene in another species where the role is known. To assess the ability of WORMHOLE to identify functionally-related ortholog pairs, we measured the performance of the WORMHOLE SVMs at predicting Functional Orthologs from Swissprot Text Analysis (FOSTA) FEP pairs. The FOSTA database contains high confidence FEPs based on text analysis of Swiss-Prot annotations and thus represents an assessment of functional equivalence at a high level of manual curation by experts [32]. Voting improves prediction of FOSTA FEPs relative to the constituent algorithms, with SVMs giving an additional improvement in precision, recall, and harmonic mean of precision and recall (Fig 7). As observed in the ortholog reference dataset, WORMHOLE adds almost no benefit to FEP predictions between closely related species, while performance is greatly improved in more distantly related species (Fig 7B and S3 Fig). In FEP prediction between vertebrate species (Fig 7A), and predictions between humans and mice in particular (S5E and S5F Fig), many of the first-layer algorithms produce nearly perfect performance, leaving no room for improvement. In contrast, prediction of FEPs between invertebrate species, or between vertebrates and invertebrates, receives a substantial benefit from the SVM models relative to simple voting, improving both precision and recall by more than 5% in most cases and more than 10% for certain species combinations (S3 Fig). Performance statistics for WORMHOLE, voting, and each constituent algorithm at predicting FOSTA FEPs is provided in S5 Table.
The QfO consortium provides a set of tools for benchmarking ortholog prediction datasets. One of these tools calculates gene ontology (GO) term conservation between gene pairs, an established metric of functional relatedness [33]. We used this service to assess the average functional relatedness between WORMHOLE-predicted LDOs as compared to predictions made by each of the constitutive algorithms and to PANTHER LDOs across the six examined genomes. WORMHOLE consistently maintained a similar level of functional relatedness between predicted gene pairs, but identified more gene pairs, as compared with the PANTHER LDOs (Table 4 and Fig 8). In invertebrate-invertebrate comparisons, WORMHOLE achieves nearly identical GO term conservation scores to PANTHER LDOs. In the vertebrate-vertebrate and vertebrate-invertebrate comparisons, WORMHOLE functional conservation is slightly decreased relative to PANTHER LDOs, but is higher than all methods that call a similar number of pairs. A similar result holds when comparing enzyme classification numbers (EC), which depend strictly on the catalyzed chemical reaction, between enzyme LDO pairs (Table 4, S4 Fig). The WORMHOLE RBHs receive similar functional relatedness and enzyme conservation scores to the PANTHER LDOs–and higher mean scores in invertebrate comparisons–while generating substantially more LDO predictions (Table 4, Fig 8, S4 Fig). A third measure evaluates the discordance between species and gene phylogenetic trees based on uploaded ortholog pairs [33]. Similar to GO term conservation, WORMHOLE expands the number of represented gene trees while maintaining low species-gene tree discordance and limiting the number of gene trees that do not match the phylogenetic structure of the species tree (S5 Fig).
The combined ability of WORMHOLE to improve FEP prediction and expand the pool of LDOs while maintaining functional relatedness shows that, despite non-one-to-one mapping of genes, WORMHOLE predictions are well tuned to gene function. This is demonstrated by the more restricted WORMHOLE RBHs, which maintain identical, or slightly better, functional relatedness to PANTHER LDOs while generating a larger pool of predicted LDOs. This implies that the WORMHOLE SVMs are sensitive to gene function.
To illustrate the type of LDO predicted by WORMHOLE in difficult “edge cases”, we manually inspected a set of human-to-worm LDO predictions. Specifically, we examined genes that the WORMHOLE SVMs strongly selected (WORMHOLE RBHs with WORMHOLE Score > 0.75) but were missed by simple voting (Votes < 7, Vote Score < 0.25); 17 genes fit these criteria (Table 5). As a metric of sequence conservation, we conducted protein BLASTp for each query gene against the target genome, and each target gene against the query genome (Table 5). Of the 17 human genes queried, 5 had PANTHER LDOs in worm. In all five cases, WORMHOLE predicted the same worm gene as PANTHER. Four of these genes also were the BLASTp RBH between human and worm. In the remaining case (human gene CPLX2), both WORMHOLE and PANTHER identify the worm gene cpx-1, while a BLASTp of cpx-1 against the human genome points to CPLX1.
In addition to the five gene pairs that the PANTHER LDOs called, WORMHOLE identified 12 novel LDOs that were not PANTHER LDOs (Table 5). Of these novel LDOs, 9 represent the BLASTp RBH for the query gene examined. In one of the three remaining cases, the human gene queried, RP11-343C2.11, overlaps nearly completely with another human gene, VPS4A. VPS4A is a paralog to the BLASTp RBH, VPS4B. This overlap suggests that RP11-343C2.11 may be an artifact in the human genome used by the constituent algorithms predicting the gene pair. In another remaining case (human gene TNNI1), multiple duplication-post-speciation events have occurred between human and worm, and WORMHOLE identified one member of a closely related group of genes (tni-4) instead of another that is the BLASTp RBH (unc-27/tni-2).
We next examined evolutionary distance for each gene pair. In the case of human CPLX1/2 and worm cpx-1/2, CPLX2 is less evolutionarily distance from cpx-1 than CPLX1, despite the failure of BLASTp to identify this pair as an RBH, suggesting that WORMHOLE is opting for the least divergent gene pair in this case. In contrast, the worm gene heh-1 is identified as the WORMHOLE RBH, the PANTHER LDO, and the BLASTp RBH, but not the least evolutionarily distant gene (Table 5). Similarly, only 3 of the 12 novel WORMHOLE LDOs represent the reciprocal least evolutionarily distant gene between humans and worms. Which metric is “correct” in these cases is unclear, and phylogenetic reconstruction often does not provide additional insight. Many of these edge cases represent phylogenetic trees where gene duplication has occurred in both species more recently than the orthology-defining speciation event (e.g. CPLX2/cpx-1 and TNNI1/tni-4). When this occurs, a single gene in one lineage will always be evolutionarily closer to all genes in the other lineage from the perspective of sequence divergence. For example, the CPLX2 sequence has diverged less from both cpx-1 and cpx-2 than CPLX1. Other gene pairs belong to families with an even more complex and difficult to interpret evolutionary history with multiple speciation and duplication events (e.g. HACD3/R10E4.9). While the two genes in these complex families with the least sequence divergence are technically the LDO, the relationship between other family members, particularly when attempting to infer functional relationships from orthology, is ambiguous. In these cases, direct experimental examination is necessary to confirm functional relationships between orthologs. By considering consensus predictions from multiple prediction strategies, WORMHOLE provides a disciplined strategy for selecting genes prior to these analyses.
Taken together, these examples suggest that, with the PANTHER LDOs as reference and the additional information provided by the constituent algorithms, the WORMHOLE SVMs add clarity to difficult-to-distinguish edge cases where orthology is ambiguous based solely on an examination of available ortholog prediction strategies or voting-based meta-tools. They also help define the limits of the current SVM models around gene families with complex evolutionary history involving multiple speciation and duplication events that are not clearly resolved by current phylogenetic models.
To provide convenient access to WORMHOLE LDO predictions, we developed a web tool that can be accessed publicly at http://wormhole.jax.org/. The web tool allows users to rapidly query the WORMHOLE database for LDO predictions between the six species, including options to manually define the WORMHOLE score threshold, exclude all but the highest scoring predicted LDOs for genes with multiple mappings, and select only WORMHOLE RBHs. Genome-wide ortholog predictions between each pair of species are also available for download.
The past two decades have seen the accumulation of a vast wealth of genetic information across thousands of species. On the heels of this accumulation, our ability to identify common genetic features between genomes has steadily improved, engendering dozens of methods for predicting orthologs based on sequence similarity, phylogenetic tree structure, and functional interactions. Here we introduce WORMHOLE, a novel application of machine learning to the problem of LDO prediction. In this tool we have taken advantage of the variety of available ortholog prediction strategies to develop a meta-tool that integrates predictions from many sources to specifically generate LDO predictions between six commonly used model organisms. The use of machine learning allows WORMHOLE to leverage the unique strengths of each method and the synergistic qualities between prediction methods to optimize performance and provide LDO predictions with higher confidence than other currently available methods, particularly when applied to predict LDOs between distantly related species.
In developing WORMHOLE we have taken a supervised machine learning approach to LDO prediction that combines and augments current methods by adding a second layer that intelligently aggregates the predictions of many ortholog predictors into a compound LDO prediction. Multilayer methods are standard in machine learning and were originally biologically inspired. For example, the visual cortex of primates is organized into a hierarchy of neuron layers that successively capture higher order features of the visual field as the stimulus travels deeper into the brain. The earliest layers of the visual cortex capture relatively simple features of a scene like spots of relative brightness or darkness, intermediate layers aggregate these low-level features into object boundaries, while the highest layers relate these boundaries to semantic object categories stored elsewhere in the brain allowing for object recognition.
The multilayer structure of WOMRHOLE is analogous. In the case of WORMHOLE, the primitive features (e.g. bright and darks spots in the visual field) are represented by prior biological knowledge, such as sequence similarity, physical interactions between the protein products of genes, evolutionary distance between sequences, and known mutation rates as a function of taxonomy. The first layer of WORMHOLE—the 17 constituent algorithms—transforms these primitive features into intermediate features consisting of preliminary predictions of orthology between pairs of genes (analogous to the object boundaries in the visual cortex). These intermediate features individually are not always sufficient to distinguish LDOs, indeed the constituent algorithms do not intend to make such a prediction (see below), but each is a unique assessment of the many biological features that are relevant for such predictions. The second-layer aggregation operation integrates these preliminary predictions of the individual algorithms as input features for SVM classifiers, using the patterns in these features to recognize true LDOs (as the visual cortex recognizes objects from object boundaries) (Fig 1). This second layer is separated from the raw input data (genetic sequence) by the orthology predictions made by the constituent algorithms, combining them in an intelligent way to make LDO predictions.
We stress that the constituent algorithms do not intend to explicitly predict LDOs. Rather, they predict orthology by applying various statistical criteria to input data including phylogeny, sequence alignment, and/or functional annotation that are algorithm-specific. WORMHOLE uses the orthology calls of each algorithm as features that may be relevant to predicting LDOs. Indeed, LDOs are a specific and rather small subset of all orthologs. The extent to which any constituent algorithm’s ortholog or FEP predictions align with the PANTHER LDO reference set is a function of the methodology and the orthology definition used by that algorithm. Nevertheless, we can treat the orthology calls of the constituent algorithms as predictions of LDOs. If this assumption is not valid for a specific algorithm, the SVM will simply assign a low weight to that algorithm based on the observed poor performance of that algorithm at predicting PANTHER LDOs (e.g. Isobase, Fig 5A). From this point of view the constituent algorithms display wide variation in their precision and recall on the reference set; some are very conservative and precise, while others have high recall at the cost of calling many non-LDOs. On this basis we suspected that a simple voting strategy would be a useful heuristic for capturing likely LDOs by aggregating over a range of predictions and filtering out pairs that result from algorithm-specific errors or an overly broad orthology definition. Indeed, this voting strategy is enriched for LDO prediction compared to the constituent algorithms as it improves precision and recall over the constituent algorithms when predicting the PANTHER LDO set (Fig 2). More directly, the vote counts of PANTHER LDOs are significantly higher than non-LDOs (S6 Fig), demonstrating that voting is a discriminative criterion for LDO identification. While the performance improvement is species-dependent, voting achieves higher precision at a fixed value of recall (and vice versa) in nearly all cases.
The variation in precision and recall of the constituent algorithms demonstrates that giving each algorithm equal weight in the vote count is not optimal. Conservative algorithms that predict fewer orthologs but more often identify LDOs should be given higher weight. This raises the question of how to apportion weights to algorithms. One strategy would be to try to identify commonalities directly and construct weights “by hand”, but this runs the risk of incorporating our personal biases. Instead, we learned the weights from a training set of examples of true and false LDOs using the SVM algorithm (see Materials and Methods). The SVMs clearly outperform the simple voting by learning which algorithms are more trustworthy and giving them higher weight.
In any machine learning application, the scope is defined exclusively by the training dataset. We trained our models on the PANTHER LDOs, a set of high quality LDO predictions. Because PANTHER LDOs represent a conservative set of closely related genes pairs, and because there exist edge cases for which evolutionary information becomes difficult to parse, we anticipated that the PANTHER LDOs were not comprehensive in identifying all true LDOs. Indeed, these edge cases increase in frequency for distantly related genomes that contain many duplication-post-speciation events in both lineages. PANTHER LDOs are very likely true positive LDOs, have high functional conservation, and they are more or less representative of true LDOs. However, because PANTHER LDOs are conservative, they are not comprehensive, making them a suitable reference set for predicting a larger set of LDOs. The central assumption of WORMHOLE is that we can learn a signature identifying true LDOs by inspecting the PANTHER LDOs. Our predictions are thus “PANTHER-LDO-like” as far as the input features to the SVM are concerned. We have employed four strategies to ensure that the WORMHOLE predictions are sensible: 1) nested cross-validation, which prevents overfitting on the training data, 2) estimation of evolutionary and sequence divergence between predicted LDOs, 3) prediction of known functionally equivalent proteins using a distinct set of high confidence FEPs (the FOSTA database), and 4) assessment of functional relatedness by measuring GO term conservation between predicted LDO gene pairs (using the community standard benchmarking service provided by QfO). Our results on the evolutionary and sequence divergence between WORMHOLE LDOs and RBHs are a direct test of “least divergence” between the predicted ortholog pairs. WORMHOLE LDOs and RBHs improve the PANTHER LDOs on these measures by: 1) expanding to a larger set of predicted LDOs without compromising the small divergence between predicted LDOs, and 2) excluding a subset of PANTHER LDOs that have significantly higher divergence than is typical of the PANTHER LDOs.
The tests of functional conservation and equivalence provide a completely independent assessment of the WORMHOLE predictions, but their results have to be interpreted with caution. First, as noted above, orthology is related, but not identical, to functional equivalence. Second, functional annotation of proteins is much less complete than predicted orthology. This is because sequence data are much more readily available than functional data and orthology can often be inferred with high confidence independent of any functional information. The SVMs perform better than voting and the constituent algorithms in predicting the FOSTA FEPs (S3 Fig). This relative comparison is what is important. The PR-curves for the SVMs tested on the FOSTA FEPs must be understood in light of the fact that many FEPs are likely to be missing from FOSTA. Likewise, when considering the conservation of functional annotations provided by QfO, there are many “missing” functional annotations, so performance has to be considered in a relative sense. The WORMHOLE RBHs have comparable functional similarity scores to the PANTHER LDO reference set, but WORMHOLE makes a substantial number of novel calls (Table 3, Fig 8 and S4 Fig). These novel calls are particularly important in distant species comparisons, where the methodology used to identify PANTHER LDOs is conservative. WORMHOLE employs complementary information not available to the PANTHER algorithm to improve confidence and expand the number of LDOs predicted. The functional cross-validation results suggest that WORMHOLE-predicted LDOs are sensible candidates.
Many of the WORMHOLE predictions are not one-to-one mappings, as required by the strict definition of an LDO. This can be interpreted simply as the expected “dead weight loss” of the machine learning strategy; the final model cannot reasonably be expected to perfectly predict the known LDOs and non-LDOs. An alternative interpretation is available when we observe that some LDOs will be less divergent from their non-LDO orthologs than others. Indeed, some genes will have multiple orthologs that are highly similar in both sequence and function, and selecting the LDO will amount to making an extremely fine distinction. These LDOs will be more difficult to separate using our strategy, but also much more functionally similar. The functional cross-validation shows that this is exactly what happens. Among the non-PANTHER LDOs (genes pairs in the WORMHOLE database, but not part of the PANTHER LDO dataset), a significant fraction lies within the larger PANTHER database (Fig 5A). These are the false positives that could not be reliably distinguished from true LDOs by the SVM during training. The functional cross-validation directly compares the WORMHOLE predictions to both the PANTHER LDOs and the full PANTHER set. The WORMHOLE predictions retain comparable scores to PANTHER LDO while calling many more pairs and producing better scores than other methods that call similar numbers of pairs. Simultaneously, WORMHOLE has higher performance than the full PANTHER set. We stress that this occurs purely as a side benefit of training an SVM to recognize LDOs from non-LDOs and not because WORMHOLE has explicitly included additional functional information beyond that contained in the first-layer algorithms.
Depending on the purposes of user, these functionally similar multiple mappings may be useful per se. A limitation inherent to the strict definition of an LDO as the single least diverged gene pair in an ortholog group is that it will necessarily fail to identify cases where a lineage-specific duplication results in redundant genes that are both functionally equivalent to the gene in the other species. Our functional data suggests that this is not a rare occurrence, as WORMHOLE predicts many multiple mappings that are enriched for functional conservation near the same level as the LDOs. However, there are two filters that a WORMHOLE user can use to sift through multiple hits to potentially identify the true LDO. First, within a family of hits the pair with the highest WORMHOLE score is likely to be the LDO (Fig 5C). An even stricter criteria is to select the gene pair with the reciprocal highest score (i.e. the WORMHOLE RBH), should it exist. However, some instances of multiple hits arise because the candidates have the exact same vote patterns, and hence the same WORMHOLE score. A second filter when considering multiple mappings is to use auxiliary criteria, e.g. highest-quality sequence alignment, independent of WORMHOLE to identify the LDO, which is beyond the scope of the WORMHOLE web tool.
A priori, a highly tuned model to predict LDOs in one species pair might not have any predictive power for unrelated species. However, we find that a model trained on one species pair does perform well when applied to predict LDOs between other species pairs (Fig 3 and S3 Table). This strongly suggests that the WORMHOLE SVMs are identifying patterns in the constituent algorithm predictions that are indicative of LDO status in general and not just in the species pair used to train the model. This property points to broader applicability of the supervised machine learning framework and suggests that LDOs can be inferred in a species-independent manner. This is an intriguing prospect for future work.
An examination of novel LDO predictions made by WORMHOLE in gene pairs with ambiguous orthology status (Table 5) suggests that the WORMHOLE SVMs are able to parse non-intuitive information provided by the voting patterns in the constituent algorithms to provide clarity in distinguishing orthology. WORMHOLE identifies a number of novel LDOs in this realm, picking the BLASTp RBH in most cases. A few cases of disagreement between WORMHOLE and PANTHER or BLASTp indicate that there remains room for improvement by adding additional information or updating reference LDO sets in future iterations of WORMHOLE.
WORMHOLE is the first machine learning meta-tool developed for the problem of predicting LDOs. We demonstrate the ability to improve LDO prediction using SVM classifiers. A key advantage to our approach is that it is a “meta-heuristic”, meaning that, in principle, any set of input algorithms can be used in Layer 1 and any user-preferred reference set and classification algorithm can be used in Layer 2. As more data become available and ever more sophisticated ortholog prediction tools are developed, the multilayer machine learning approach can grow to accommodate such innovations in the field. This work represents a starting point for several potential lines of future work. While WORMHOLE considers only the predictions of other orthology prediction methods, machine learning classifiers can accept any form of relational data for a given pair of potential orthologs that can be appropriately represented as input, allowing for consideration of information not implicitly captured in the constituent algorithms. In principle, future adaptations of WORMHOLE may include direct information about sequence similarity (e.g. alignment statistics), functional comparison (e.g. GO term conservation scores), or even more obscure biological information (e.g. relative expression levels in specific tissues). Beyond model systems, our results show that training a model on examples from one species pair generalizes well to other species pairs (Fig 3). It should be possible to use this property to make predictions in species not included in the design of WORMHOLE. Many current ortholog prediction projects make predictions for very large numbers of species. In principle, the machine learning framework can augment these predictions by, for example, training SVM models on a set of well-characterized and relevant models systems and using the predictions of the SVM models for less-characterized species. Some meta-tools (e.g. MOSAIC and MARIO) already use voting as a pre-processing step prior to sophisticated sequence-based analyses. Replacing simple voting with trained SVMs could supply candidates for sequence analysis at both a high level of sensitivity and specificity. The scope is only limited by availability of data and computational resources.
Ortholog and FEP datasets were acquired from the online repositories of each source database, in OrthoXML format when available. Web addresses, access dates, and version numbers for the 17 ortholog prediction datasets used to train WORMHOLE SVMs are provided in Table 1, and for all other source data in S6 Table. In building models, we were able to simply include all predictions generated by each tool under default settings in most cases. For EggNOG and Isobase, tool-specific considerations motivated additional effort.
As a first-order assessment of confidence in a given predicted ortholog pair we employ simple voting, a straightforward tally of the number of constituent algorithms that predict that pair. To improve upon simple voting, we applied machine learning to differentially weight the influence given to each algorithm based on its performance in predicting the reference LDOs. Specifically, let c denote a cLDO and let xc ϵ {0, 1}17 denote the 17-dimensional binary vector of ortholog calls from each of the 17 constituent algorithms for c (i.e. xic denotes the 0/1 prediction of the ith constituent algorithm). An SVM assigns a weight, wi, to predictions made by each constituent algorithm based on the individual performance of that algorithm at reproducing the PANTHER LDOs and defines a score for each cLDO, c:
rawSVMscore(c)=∑i=117wixic−b
where the sum is taken over the 17 constituent algorithms, wi is the weight assigned to the ith algorithm, and b is an offset that defines the boundary between positive and negative predictions. The parameters {wi,b} are learned from a set of labeled training examples. Note that if the offset is zero and all weights equal to one, then the SVM formula corresponds exactly to simple voting. Thus, the SVM is a weighted voting scheme where the weights are tuned to the training data. We fit the SVM classifiers using the R package “e1071”, which is available on the Comprehensive R Archive Network (http://cran.r-project.org).
In machine learning a key issue in model fitting is “overfitting”, i.e. setting the model parameters in such a way that the model performs well on the training data but fails to generalize to new data. The SVM has a single hyperparameter (i.e. a parameter that defines the fitting of the model, but not the model itself), called C that can be tuned to prevent overfitting. The parameter C defines the penalty for misclassifications and balances the fit to the data against generalizability [39]. For each combination of query and target species, we employed nested 10-fold cross-validation (nested CV) [40] to choose C. Nested CV splits the model selection process into an inner CV to select model parameters and an outer CV to estimate performance of the selection procedure. The outer CV, first randomly separates the data into 10 equal parts, trains the model on 9 parts, and tests the performance of the resulting model on the withheld part. The process is then iterated, withholding each 10% of the data once for testing. During each iteration of the outer CV, the model-training step is carried out using an inner CV. The 90% of the data used for training is further subdivided into 10 parts for standard CV. Within each inner CV iteration, C was chosen from a logarithmic vector, C ϵ {4−2, 4−1, 40, 41, 42}, to maximize the average testing accuracy (fraction of correct classifications) on the 10th inner CV parts. The inner-CV-tuned model is then tested on the 10% of the data that were held out for the outer CV. All assessments of generalization performance (precision, recall, harmonic mean) were estimated by their mean and standard error of mean over the 10 outer CV iterations. In this way, the inner CV ensures that the model parameters are not overfitting the idiosyncrasies of the training data, while the outer CV provides an estimate of the robustness of the model selection procedure when applied to novel data that were completely unseen by the model selection procedure during training.
Segregation of training data into training and test parts for cross validation requires care to ensure that each part is truly independent. Because there were many more negative examples than positive, we partitioned the two classes separately so that each part had the same proportion of positive and negative examples. We further ensured that all candidate ortholog pairs for a given query gene were assigned to a single part. We examined the effect of stratifying ortholog pairs into folds by gene family; however, while family-wise stratification resulted in a small increase in the variance of precision and recall across folds in the outer cross-validation, it did not affect either the overall quantitative performance of WORMHOLE or the qualitative conclusions reached in this work. Because the two strategies were qualitatively indistinguishable, we proceeded with the simpler, gene-wise stratification.
To compare SVM models between species-pairs (Fig 5B), we computed the Pearson correlation between the weight vectors, w, for each model. This correlation encodes whether or not the weight vectors assign high weight to the same set of constituent algorithms. For highly similar models this correlation is close to one, whereas for highly dissimilar models this correlation is close to zero.
As primary metrics of performance we evaluated recall (R) and precision (P). Recall is the fraction of the total number of correct ortholog pairs that are predicted by an algorithm, formally defined as:
R=TPTP+FN
while precision is the fraction of the total number of predictions made by an algorithm that are correct:
P=TPTP+FP
where TP is the number of true positives, or the number of correct ortholog pairs predicted by an algorithm, FN is the number of false negatives, or the number of correct ortholog pairs not predicted by an algorithm, and FP is the number of false positives, or the number of incorrect ortholog pairs predicted by an algorithm. These values are calculated by comparing the pairs of orthologs predicted by each algorithm to the reference ortholog dataset for a given pair of query and target species. A single performance metric is often useful for comparing a large number of predictions. In these cases we used a related metric, the harmonic mean of precision and recall (F), defined as:
F=2PRP+R
F provides a single measure that balances precision and recall.
The harmonic mean weighs P and R simultaneously and equally to summarize classification performance. A more flexible family of measures are the β-harmonic means defined by:
Fβ=(1+β2)PRβ2P+R
The β-harmonic means are a family of measures that depend on a parameter, β, which balances the importance of recall relative to precision. The measure F1 is simply the harmonic mean (defined above). The measure F0.5 gives recall half the priority of precision, while F2 gives recall twice the priority of precision.
The raw confidence values returned by each SVM model (or voting) cannot be compared across species pairs because the assigned SVM weights are specific to each species pair. To allow such comparisons, we normalized the raw SVM scores to a scale that is directly linked to the performance of the model. Specifically, we identified the thresholds T within the raw scores for which the precision and recall at T maximizes Fβ for β ϵ {0.125, 0.25, 0.5, 1, 2, 4, 8}. These thresholds are mapped onto the confidence scores 0.9375, 0.875, 0.75, 0.5, 0.25, 0.125, and 0.0625, respectively. We then interpolated that the map from raw SVM scores to confidence scores using monotonic Hermite cubic spline interpolation [41], which is implemented in the R function ‘splinefun’. Thus, the vote and SVM scores are scaled in such a way that applying a threshold of 0.5 (i.e. selecting all ortholog pairs with scores greater than or equal to 0.5) maximizes F for balanced precision and recall. Doubling β halves the distance toward 0.0/1.0 in the confidence scores. For example, a confidence threshold of 0.75 gives precision twice the weight of recall and a threshold of 0.875 gives precision four times the weight of recall. Conversely, a confidence threshold of 0.25 gives recall twice the weight of precision. We term the confidence score that scales simple voting the Vote Score and the confidence score that scales the raw SVM scores the WORMHOLE Score.
In a few cases, the score given to an ortholog pair differs depending on which species is used as query and which is used as target (e.g. a pair consisting of a human and worm gene may receive a different score if a human-to-worm query is made than when a worm-to-human query is made). This is a result of the way WORMHOLE is constructed, with a different SVM model used for each combination of query and target species. In order to harmonize scores with respect to direction of inquiry, the score given to each pair of orthologs by each Layer 2 WORMHOLE method was averaged between directions.
Evolutionary distance between genes was estimated for the longest protein encoded by each gene in a pair. Genome-wide protein sequences were obtained from Ensembl BioMart for each species [19]. All protein pairs between species were aligned using the pairwiseAlignment() function in the R package “Biostrings” [42], which implements quality-based alignment as described by Malde [43]. Evolutionary distance was calculated for each alignment using the dist.ml() function in the R package “phangorn” [44] using the BLOSUM62 substitution matrix. Both R packages are available on the Comprehensive R Archive Network (http://cran.r-project.org).
BLASTp bits scores and RBHs were determined by aligning each protein sequence against each target genome with NCBI BLAST+ (acquired from http://www.ncbi.nlm.nih.gov/blast; Table 1) using the following command options:
blastp-queryxxyy.fa-subjectxx.fa-outxx-yy-2.txt-outfmt6-max_hsps1-evalue1e-4
where “-query” and “-subject” specify input files in FASTA format, “-out” specifies the output file in text format, “-outfmt 6” requests BLASTp hits to be reported in a pairwise table with BLAST statistics, “-max_hsps 1” limits the output to a single report per matched protein pair, and “-evalue 1e-4” sets a maximum threshold on E-value for reported matches. The placeholders “xx” and “yy” indicated two letter abbreviations for species names (e.g. “ce” abbreviates Caenorhabditis elegans).
The QfO Benchmarking Service tool accepts lists of predicted ortholog pairs and returns several measures of performance. We used this service to compare predictions made by WORMHOLE to those made by the PANTHER LDOs and the constituent algorithms for three performance criteria described by Altenhoff and Dessimoz [33], which are described briefly below. QfO provides several publicly available datasets for comparison, many of which are included as constituent algorithms in WORMHOLE. To minimize potential bias introduced by differences in the version of each dataset used in QfO vs. WORMHOLE, we independently ran each QfO performance metric on the set of ortholog pairs predicted in each constituent algorithm, as included in the WORMHOLE database. Each ortholog dataset (WORMHOLE, PANTHER LDOs, and constituent algorithms) was mapped to the QfO reference proteome and uploaded to QfO for analysis.
|
10.1371/journal.pcbi.1005305 | A Topological Criterion for Filtering Information in Complex Brain Networks | In many biological systems, the network of interactions between the elements can only be inferred from experimental measurements. In neuroscience, non-invasive imaging tools are extensively used to derive either structural or functional brain networks in-vivo. As a result of the inference process, we obtain a matrix of values corresponding to a fully connected and weighted network. To turn this into a useful sparse network, thresholding is typically adopted to cancel a percentage of the weakest connections. The structural properties of the resulting network depend on how much of the inferred connectivity is eventually retained. However, how to objectively fix this threshold is still an open issue. We introduce a criterion, the efficiency cost optimization (ECO), to select a threshold based on the optimization of the trade-off between the efficiency of a network and its wiring cost. We prove analytically and we confirm through numerical simulations that the connection density maximizing this trade-off emphasizes the intrinsic properties of a given network, while preserving its sparsity. Moreover, this density threshold can be determined a-priori, since the number of connections to filter only depends on the network size according to a power-law. We validate this result on several brain networks, from micro- to macro-scales, obtained with different imaging modalities. Finally, we test the potential of ECO in discriminating brain states with respect to alternative filtering methods. ECO advances our ability to analyze and compare biological networks, inferred from experimental data, in a fast and principled way.
| Complex brain networks are mainly estimated from empirical measurements. As a result, we obtain networks where everything is connected to everything else through different strengths of interaction. Filtering procedures are typically adopted to prune weakest connections. However, network properties strongly depend on the number of remaining links and how to objectively fix such threshold is still an open issue. Here, we propose a criterion (ECO) to filter connectivity based on the optimization of fundamental properties of complex systems, i.e., efficiency and economy. Using ECO, investigators can analyze and compare connectomes in a fast and principled way, capturing network properties of different brain states to eventually quantify (re)organizational mechanisms underlying cognition and disease. Given its generality, we anticipate that ECO can also facilitate the study of networks in other fields, such as system biology.
| Network science has provided a breakthrough in the analysis and modeling of biological systems with the aim to unlock molecular mechanisms behind human disease [1–3] and quantify brain (re)organization underlying behavior, cognition and mental disorders [4–6].
In part, this has been made possible by the increasing availability of tools that indirectly infer the structure of those networks from empirical measurements, thus bypassing the current lack of accurate and complete interaction maps [3, 7]. In system biology, functional links are estimated from transcriptional or phenotypic profiling, and genetic interactions by using measures such as Pearson correlation [8] or Granger causality [9].
In neuroscience, imaging tools such as magnetic resonance imaging (MRI) and electro/magnetoencephalography (E/MEG), are extensively used to map connections and/or interactions between different brain sites, i.e., the connectome [7, 10]. Brain connectivity methods are typically used to estimate the links between the nodes. While structural connectivity (SC) measures the probability to find axonal pathways between brain areas, typically from diffusion MRI, functional connectivity (FC) rather calculates the temporal dependence between remote neural processes as recorded, for instance, by functional MRI, EEG or MEG [4, 7].
At this stage, the resulting networks correspond to maximally dense graphs whose weighted links code for the strength of the connections between different nodes. Common courses in brain network analysis use thresholding procedures to filter information in these raw networks by retaining and binarizing a certain percentage of the strongest links (S1 Fig). Despite the consequent information loss, these procedures are often adopted to mitigate the incertainty of the weakest links, reduce the false positives, and facilitate the interpretation of the inferred network topology [3, 11].
At present, there’s no objective way to fix the value of such threshold. Because network properties significantly depend on the number of remaining links, scientists are obliged to explore brain network properties across a wide range of different candidate thresholds and eventually select one representative a-posteriori [12]. Concurrently, alternative approaches can be adopted to cancel spurious links emerging from third-party interactions [13–15], or statistically validate the estimated connections [7, 16, 17]. However, these procedures lack of precise rationale, are subject to arbitrariness (e.g., the choice of the statistical significance) and make difficult the comparison of network properties between many individuals or samples [11, 18]. Furthermore, these become extremely time-consuming when considering several large connectomes due to the computational complexity of graph quantities based on paths between nodes or on communities detection [19].
To circumvent these issues, we propose a topological criterion for selecting a threshold which captures the essential structure of a network while preserving its sparsity. Based on the optimal trade-off between two desirable but incompatible features—namely high global and local integration between nodes, and low connection density—this method is inherently motivated by the principle of efficiency and economy observed in many complex systems [20], including the brain [21].
Global- and local-efficiency have revealed to be important graph quantities to characterize the structure of complex systems in terms of integration and segregation of information [22, 23].
Both structural and functional brain networks tend to exhibit relatively high values of global- and local-efficiency. At the same time they also tend to minimize, for economical reasons, the number of their links leading to sparse networks [21].
Thus, we propose to determine a density threshold that filters out the weakest links and maximizes the ratio between the overall efficiency of a network and its wiring cost. Notice that the definition of cost can have different connotations, e.g., the spatial distance between connected nodes [21]. Here, the cost in terms of number of links is a more general definition which also applies to non-spatially embedded networks (e.g., molecular interaction networks). We formally introduce a criterion to filter information in a given network by finding the connection density ρ that maximizes the quality function:
J = E g + E l ρ (1)
where Eg and El represent respectively the global- and local-efficiency of a network. By definition, the three quantities Eg, El and ρ are normalized in the range [0, 1], and both Eg and El are non-decreasing functions of ρ. More details about the formulation of J can be found in the Material and Methods.
For both regular lattices and random networks, we proved analytically that the optimal density that maximizes J follows a power-law ρ = c/(n − 1), where c is a constant and n is the network size, i.e., the number of nodes in the network. More specifically, c = 3.414 for lattices and c = e = 2.718 for random networks, so that we have approximately ρ ≃ 3/(n − 1). Hence, to maximize J, these networks have to be sparse with an average node degree k ≃ 3 or, equivalently, with a total number of links m that scales as m ≃ 3 2 n (S1 Appendix).
We confirmed this result (S2a and S2b Fig) through extensive numerical simulations (Materials and Methods), showing that it held true also in more realistic network models, such as in small-world networks [24] (Fig 1a) and in scale-free networks [25] (Fig 1b). For these simulated networks the fitted values varied progressively from c = 3.265, in lattices, to c = 2.966, in random networks, thus falling within the theoretical range found analytically (S1 Table).
Notably, the optimal density values maximizing J emphasized the intrinsic properties (random or regular) of all the implemented synthetic networks in terms of global- and local-efficiency (Fig 1d and 1e and S2d and S2e Fig).
We computed the quality function J in both micro- and macro-scale brain networks and we evaluated how the density maximizing J scaled as a function of the network size. We considered connectomes used in previously published studies that were obtained with different imaging modalities, from calcium imaging to EEG, and constructed with disparate brain connectivity methods (Table 1).
For each connectome we applied a standard density-based thresholding. We started with the empty network by removing all the links (ρ = 0). Then, we reinserted and binarized one link at time, from the strongest to the weakest, until we obtained the maximally dense network (ρ = 1). At each step we computed J and we recorded its profile as a function of ρ. The pooled density values, as returned by the maximization of the healthy group-averaged J profile in each modality (see Fig 1f for one representative), followed a power law comparable to the one that we reported for synthetic networks (Fig 1c). In particular, the fit ρ = c/(n − 1) to the data gave c = 3.06 with an adjusted r-square R2 = 0.994. Notably, we obtained a similar scaling (c = 2.87 adjusted R2 = 0.946, S2c Fig) when considering individual J profiles (S2f Fig). These results confirm that also for brain networks we can assume that the optimal density threshold maximizing J only depends on the network size according to the same rule ρ ≃ 3/(n − 1).
In conclusion, we introduced a criterion, named efficiency cost optimization (ECO), to select a threshold leading to sparse, yet informative brain networks. Such a threshold is relatively independent of the connectome’s construction and invariant to the underlying network topology so that it can be selected a-priori once the number of nodes is known.
To illustrate the methodology, we considered connectomes from four different imaging modalities, namely EEG, MEG, fMRI, and DTI (Table 1). Because we do not know the true structure for these connectomes, we evaluated the ability of ECO to discriminate network properties of different brain states, i.e., healthy versus diseased, at individual level.
We characterized brain networks by calculating graph quantities at different topological scales, i.e., large (global- and local-efficiency, Eg and El), intermediate (community partition, P; and modularity, Q), and small (node degree, ki; and betwenness, bi) (Materials and Methods). To assess network differences between brain states, we measured distances between the values of the graph quantities obtained in the healthy group and those in the diseased group. We adopted the Mirkin index (MI) to measure distances between community partitions, and the divergent coefficient (D) for other graph quantities (Materials and Methods).
We explored a wide range of density thresholds and, as expected, the value of the threshold affected the ability to separate network properties of different brain states (Kruskalwallis tests P < 0.01, S2 Table). Notably, the choice ρ = 3/(n − 1) resulted among the best candidates in producing larger distances regardless of the graph quantity (Tukey-Kramer post hoc tests P < 0.05, Fig 2 and S3 Fig). This outcome was not associated to the possible presence of disconnected components. In all the filtered brain networks the size of the largest component (> 50% of the nodes) did not differ between groups for any threshold value (Wilcoxon rank-sum tests P ≥ 0.01, Fig 3). Furthermore, ECO overall outperformed alternative methods, such as the minimum spanning tree (MST) and the planar maximally filtered graph (PMFG) [26], in giving larger distances (Tukey-Kramer post hoc tests P < 0.05, Fig 4, S4b Fig, S2 and S3 Tables). Notably, we reported good performance with respect to a hybrid method, named MST+ECO, where we added the remaining strongest links to the backbone obtained with MST, in order to reach the same average node degree as ECO, i.e. k = 3 (Tukey-Kramer post hoc tests P < 0.05, S3 Table).
Finally, brain networks filtered with ECO were more efficient (Fig 5a) and exhibited J values that better separated different brain states (Fig 5b) as compared to the other filtering methods (Tukey-Kramer post hoc tests P < 0.05, S3 Table).
We introduced ECO to filter information in networks whose links are predictions, and not direct measures, of connectivity between biological components, such as brain regions. Conventional approaches evaluate brain network properties across a large and arbitrary number of thresholds [27]. Eventually, they select a representative threshold a-posteriori that maximizes the separation between different brain states [11]. ECO allows to select an objective threshold a-priori, thus reducing the computational burden associated with typical iterative approaches.
Other methods, similar in purpose to ECO, impose unnatural constraints on the filtered network. The minimum spanning tree (MST), for instance, leads to brain networks with a null clustering coefficient [28]. The planar maximally filtered graph (PMFG) tries to alleviate this bias by allowing closed loops, but still forces planarity [26]. Conversely, ECO does not impose structural constraints, apart from favoring sparsity, and lets the intrinsic structure to emerge as illustrated in synthetic networks with known topological organization (Fig 1d and 1e, S2d and S2e Fig). This appears an important feature as different brain states (e.g., diseased versus healthy) are often characterized by networks with different topological orders (more random or more regular) depending on the underlying physiopathological neural mechanism [6].
Overall, results obtained with ECO improved the separation of all the considered network properties between different brain states as compared to other thresholds or filtering methods. In general, this does not necessarily imply a significant group difference for each graph quantity. Instead, it means that if there are underlying network differences, then ECO would be able to point them out.
Maximizing global- and local-efficiency with respect to connection density can be seen as a way to emphasize the integration and segregation properties of a connectome [29] while keeping a biologically plausible wiring cost. This rationale dovetails with current evidence showing that advantageous topological properties, such as economic small-world architectures [21], tend to be maximized in brain networks, and that, in general, sparsity increases robustness of complex systems [30].
Using ECO, networks will have a total number of links m that scales with the number of nodes as m = cn, with c ≃ 3/2. Put differently, the resulting connection density follows a fractal scaling regardless of the network size according to the power-law ρ ≃ 3n−1. Fractal scaling of size and density in self-organized systems has been recently reported and advocated as an important organizational principle to ensure optimal network functioning [31]. Although beyond the scope of this methodological study, we speculate that such characteristic scaling could result, at least for neuronal systems, from a natural optimization of the network efficiency and cost [21].
ECO makes use of density thresholds. Hence, networks having same number of nodes, will have, after pruning, the same number of links. On the one hand, this ensures that differences between network properties are not merely due to differences in the connection density [18]. On the other hand, ECO does not allow a direct evaluation of neural processes altering the number of links; however it does inform on the possible (re)organizational mechanisms.
Finally, it is important to notice that while ECO exhibits several advantageous features, it also has some limitations as described in the following section.
ECO is based on a graph theoretic approach and cannot filter out possible false positives (i.e., spurious links) resulting from biased brain connectivity estimates [7, 11]. Our criterion admits that the weighted links of the raw networks had been previously validated, either maintained or canceled. Some inference methods [32, 33] and group-based approaches [34] naturally produce sparse brain networks. In these cases, ECO would still apply as long as there is enough information to filter, i.e., a number of links m ≥ 3 2 n.
By construction, brain networks filtered with ECO (k ≃ 3) are less sparse than networks filtered with MST (k ≤ 2). However, differently from MST and PMFG, ECO does not guarantee the connectedness of the pruned networks, which can be indeed fragmented (S5 Fig). Whether this condition leads to a more realistic representation of connectomes, especially for large n, we cannot say. Current literature tends to focus on thresholded brain networks which are slightly denser than ECO, with 0.05 ≤ ρ ≤ 0.3 [35]. However, little is known on how this range depends on the number of brain nodes and future studies will have to ascertain if and how the choice of a specific threshold can give more accurate results. Here, we showed that the size of the largest components contained in average more than the 50% of the nodes (Fig 3). Therefore, caution should be used in the evaluation of the resulting network properties and, whenever possible, using graph quantities that can handle networks with disconnected nodes (e.g., the harmonic mean of the shortest path lengths [36]) appears more appropriate.
Finally, other combinations could have been considered when conceiving the quality function J. For example, in [37] authors introduced the cost-efficiency Eg − ρ, which, however, did not include the clustering counterpart. This quality function, as well as other ones that we investigated, did not exhibit meaningful analytic solutions and was therefore excluded as a possible alternative (S2 Appendix). A more general formulation would include a scaling factor in the numerator, like for example 2[αEg + (1 − α)El] where α is a control parameter ranging from 0 to 1. We proved analytically that, for both regular lattices and random graphs, the optimal density that maximizes the corresponding quality function remained ρ ≃ 3 n - 1 regardless of the α value (S3 Appendix). We confirmed this result through numerical simulations also in small-world and scale-free networks (S4 Fig) where the optimal density maximizing J corresponded to an average node degree k ≃ 3, except when α → 1 in lattices and α → 0 in random networks.
Taken together, these findings indicate that the density threshold given by ECO is relatively invariant to the specific value we assigned to the parameter α. The advantage of considering our quality function is that i) it did not depend on external parameters, ii) we could derive analytically the optimal ρ values for lattices and random networks, and iii) the density values obtained by maximizing J in real brain networks fitted the power-law that we found analytically and were able to separate different brain states. Despite these advantages, we notice that ECO could not be the definitive solution to the problem of thresholding in imaging connectomics. Other methods, possibly inspired by biology, are likely to be developed in the future and validation benchmarks will be crucial to evaluate their potential.
ECO is founded on asymptotic results in unweighted network models. Its natural application implies binarization after thresholding, a procedure widely adopted to mitigate the uncertainty carried by the weights estimated from neuroimaging data [4, 11]. Further work is needed to clarify how ECO can be extended to weighted networks, where the asymptotic expression of topological properties is less straightforward.
Interactions between biological components are not constant and need to dynamically vary to accomplish internal regulation and external function [38–40]. In neuroscience, functional brain connectivity exhibits rich temporal dynamics that are fundamental for human cognition and complex behavior [41–44]. Further studies should aim to elucidate if and how brain network differences highlighted by ECO change over time.
We introduced ECO as a possible method for filtering information in imaging connectomes. Concrete applications range from cognitive to clinical and computational neuroscience. Given its generality, we anticipate that ECO can also serve to facilitate the analysis of interconnected systems where the need of sparsity is plausible and the links are weighted estimates of connectivity. This is, for example, the case of functional networks in system biology, where links are typically derived from transcriptional or phenotypic profiling, and genetic interactions [3].
The expression of J can be seen as a particular case of a general family of functions of the form f(Eg, El, ρ). Here, we defined J as a ratio to measure the incidence of the density on the network efficiency both at global and local scale. Indeed, we were interested in a relative measure that could tell the network efficiency changes per unit of density. In addition, we did not weight the global- and local-efficiency in the numerator. While, in general, a scaling factor might be necessary to normalize changes between different graph quantities [45], here both Eg and El range between 0 and 1 and are formulated in terms of the same concept, namely the efficiency (at global and local scale) between nodes [22]. We remind to S3 Appendix and S4 Fig for more details on the introduction of a scaling parameter.
By looking at Eq (1), we have that when ρ = 0, then both global- Eg and local-efficiency El are null leading to an indefinite form. As density slightly increases (0 < ρ < ϵ, with ϵ sufficiently small) it can be demonstrated that J tends to 1. Indeed, in this range, the probability to find at least three nodes connected together (a triangle) is extremely low. By definition, El = 0 in absence of at least one triangle [22] and therefore J ≃ Eg/ρ. By considering the definitions of Eg and ρ, this quantity can be rewritten as E g / ρ = 1 / m ∑ i ≠ j n 1 / d i , j, where m is the number of existing links and di,j is the distance between the nodes i and j. In a generic network with m links there are at least m pairs of nodes directly connected (i.e., di,j = 1). This means that the sum in the latter equation is bounded from below by m in the case of isolated pairs of connected nodes (m = n/2) or in the trivial case of m = 1. It follows that J → 1 when there are relatively few links in a network.
When ρ tends to 1, it is trivial to see from Eq (1) that J → 2, as both Eg and El tend to one. For intermediate density ranges (ϵ < ρ ≪ 1 − ϵ) the analytic estimate of J is not trivial since Eg and El depend on the network topology which is, in general, unknown.
Small-world networks were generated according to the Watts-Strogatz (WS) model [24] with a rewiring probability pws = 0.1. Scale-free networks were generated according to the Barabasi-Albert (BA) model [25].
In the first simulation, we considered undirected networks. We varied both the network size and the average node degree, i.e., n = 16, 128, 1024, 16384 and k = 1, 2, 3, 4, 5. In the WS models, k is even accounting for the number of both left and right neighbors of the nodes in the initial lattice. To obtain small-world networks with k odd, we first generated lattices with k even and then, for each odd node (e.g., 1, 3, …), we removed the link with its left farthest neighbor. This procedure removes in total n/2 links leading to a new average node degree k′ = k − 1, while keeping a regular structure. As for BA models, we set the number of links in the preferential attachment mba = 3 and the initial seed was a fully connected network of n0 = mba nodes. This setting generated scale-free networks with k = 6 − 12/n, that is k ≥ 5 regardless of the selected network size. We then removed at random the exceeding number of links until we reached the desired k value. This procedure had the advantage to preserve the original scale-free structure.
In the second simulation, we considered directed networks to confirm and extend the results we obtained for undirected WS and BA networks. We selected eight representative network sizes, i.e., n = 8, 16, 32, 64, 128, 256, 512, 1024 covering the typical size of most current imaging connectomes, and we varied the connection density. Specifically, we performed a two-step procedure:
For WS models, initial lattices had k equal to the nearest even integer equal or higher than ρ(n − 1), with ρ ∈ (0, 1). For BA models, the number of attaching links was mba = log2 n to ensure an initial relatively high density; the seed was a fully connected network of n0 = mba nodes. By construction ρ ∈ ( 0 , 2 m b a n + m 0 n ( n - 1 ) ), where m0 = n0(n0 − 1)/2 is total number of links in the initial seed. For both models, we then removed at random the exceeding links until we reached the desired density value. For both simulation we generated one-hundred sample networks.
Complex networks can be analyzed by a plethora of graph quantities characterizing different topological properties [46]. Here, we considered a subset of representative ones which have been shown to be relevant for brain network analysis [47]. To characterize the entire brain network (i.e., large-scale topology), we used global- and local-efficiency, which respectively read:
E g = 2 n ( n - 1 ) ∑ i ≠ j n 1 d i j E l = 1 n ∑ i = 1 n E g ( i ) (2)
where dij is the length of the shortest path between nodes i and j, and Eg(i) is the global-efficiency of the ith subgraph of the network [22].
To characterize modules, or clusters, of brain regions with dense connections internally and sparser connections between groups (i.e., mid-scale topology), we evaluated the community structure of the brain network [4]. We extracted the partition P of the network into modules by means of the Newman’s spectral algorithm maximizing the modularity:
Q = 1 2 m T r ( G T M G ) (3)
where G is the (non-square) matrix having elements Gig = 1 if node i belongs to cluster g and zero otherwise, and M is the so-called modularity matrix [48].
To characterize individual brain areas (i.e., small-scale topology), we measured the centrality of the nodes in the brain network by means of the node degree and of the node betwenness, which respectively read:
k i = ∑ j ≠ i n A i j b i = ∑ j ≠ i ≠ h σ j h ( i ) σ j h (4)
where the element of the adjacency matrix Aij = 1 if there is a link between node i and j, zero otherwise; and where σjh is the total number of shortest paths between nodes j and h, while σjh(i) is the number of those paths that pass through i.
These quantities represent a small subset of all the possible metrics available in the market. Nevertheless, these are among the most adopted in network neuroscience thanks to their interpretability in terms of connectivity at different topological levels (e.g., network, modules, nodes) [4, 11, 27, 49–51].
To assess brain network differences between individuals (or samples) in the two groups, we measured the distance between the respective values obtained for each graph quantity. We used the Mirkin index to compute distances between two network partitions Pu and Pv:
M I ( P u , P v ) = 2 ( n 01 + n 10 ) (5)
where n01 is the number of pairs of nodes in the same cluster under Pv but not under Pu; and n10 is the number of pairs in the same cluster under Pu but not under Pv [52]. The Mirkin index is an adjusted form of the well-known Rand index and it assumes null value for identical clusterings and 1 for totally different clusterings [52]. It corresponds to the Hamming distance between the binary vector representation of each partition. Although this measure can be sensitive to the cluster sizes, it has the advantage of being a metric on the space of the clustering partitions [53].
For all other graph quantities, we used the divergent coefficient [54]:
D ( X u , X v ) = 1 M ∑ m = 1 M x u , m - x v , m x u , m + x v , m 2 (6)
where Xu = [xu,1, xu,2, …, xu,M] and Xv = [xv,1, xv,2, …, xv,M], contain the value(s) of the graph quantity for the uth and vth sample. Notably, M = 1 for global-, local-efficiency and modularity (i.e., Eg, El, Q). M = n for the node degree vector K = [k1, k2, …, kn] and the node betweenness vector B = [b1, b2, …, bn]. The divergent coefficient is a L2-norm distance similar to Euclidean distance but with a normalizing factor which is used for multidimensional scaling [55]. It ranges between 0 (equal multidimensional distribution of the features) and 1 (totally heterogeneous multidimensional distribution). This coefficient is a metric in the Euclidean space when all the values of the features are positive, as for our graph quantities [56]. Both Mirkin index and divergent coefficient are therefore metrics normalized between 0 and 1, allowing for a coherent analysis across different imaging modalities and threshold values.
We used Kruskal–Wallis one-way analysis of variance, with a 0.01 statistical threshold, to evaluate the overall effect of different thresholds, or filtering methods (i.e., MST, PMFG) on distances between individuals. A Tukey-Kramer multiple comparison post hoc test was then used to determine specific differences between pairs of thresholds or methods [57]. Here the statistical threshold was fixed to 0.05.
|
10.1371/journal.ppat.1003123 | The Tomato Prf Complex Is a Molecular Trap for Bacterial Effectors Based on Pto Transphosphorylation | The major virulence strategy of phytopathogenic bacteria is to secrete effector proteins into the host cell to target the immune machinery. AvrPto and AvrPtoB are two such effectors from Pseudomonas syringae, which disable an overlapping range of kinases in Arabidopsis and Tomato. Both effectors target surface-localized receptor-kinases to avoid bacterial recognition. In turn, tomato has evolved an intracellular effector-recognition complex composed of the NB-LRR protein Prf and the Pto kinase. Structural analyses have shown that the most important interaction surface for AvrPto and AvrPtoB is the Pto P+1 loop. AvrPto is an inhibitor of Pto kinase activity, but paradoxically, this kinase activity is a prerequisite for defense activation by AvrPto. Here using biochemical approaches we show that disruption of Pto P+1 loop stimulates phosphorylation in trans, which is possible because the Pto/Prf complex is oligomeric. Both P+1 loop disruption and transphosphorylation are necessary for signalling. Thus, effector perturbation of one kinase molecule in the complex activates another. Hence, the Pto/Prf complex is a sophisticated molecular trap for effectors that target protein kinases, an essential aspect of the pathogen's virulence strategy. The data presented here give a clear view of why bacterial virulence and host recognition mechanisms are so often related and how the slowly evolving host is able to keep pace with the faster-evolving pathogen.
| The bacteria Pseudomonas syringae is a pathogen of many crop species and one of the model pathogens for studying plant and bacterial arms race coevolution. In the current model, plants perceive bacteria pathogens via plasma membrane receptors, and recognition leads to the activation of general defenses. In turn, bacteria inject proteins called effectors into the plant cell to prevent the activation of immune responses. AvrPto and AvrPtoB are two such proteins that inhibit multiple plant kinases. The tomato plant has reacted to these effectors by the evolution of a cytoplasmic resistance complex. This complex is compromised of two proteins, Prf and Pto kinase, and is capable of recognizing the effector proteins. How the Pto kinase is able to avoid inhibition by the effector proteins is currently unknown. Our data shows how the tomato plant utilizes dimerization of resistance proteins to gain advantage over the faster evolving bacterial pathogen. Here we illustrate that oligomerisation of Prf brings into proximity two Pto kinases allowing them to avoid inhibition by the effectors by transphosphorylation and to activate immune responses.
| Plant immunity is innate and relies on two levels of pathogen perception, underpinned by different recognition strategies [1]. The first level of perception occurs at the cell surface where plasma membrane receptors called pattern recognition receptors (PRRs) recognise and respond to conserve pathogen molecules called pathogen-associated molecular patterns (PAMPs). Classically, PAMPs are invariant molecules associated with particular taxonomic classes, and are very difficult for the pathogen to modify or discard [2]. Despite the overall conservation of PAMPs, recent studies have shown that in adapted pathogens their immunogenic epitopes are under positive selection to evade host immune detection [3], [4]. Nevertheless, so-called PAMP-triggered immunity (PTI) is highly effective and is usually overcome only by adapted pathogens that have evolved specific evasive strategies [5]. Chief amongst these strategies is secretion of protein virulence molecules called effectors, which target PRRs and other nodes of the immune system to abrogate transduction of the PAMP signal within the host, or to defeat host defences [6]. Examplars of this strategy are AvrPto and AvrPtoB, two unrelated effectors of the bacterial pathogen Pseudomonas syringae, which are secreted directly into the host cell where they target a PRR complex composed of the receptor kinases FLAGELLIN SENSING 2 (FLS2) and BRI1-ASSOCIATED KINASE 1 (BAK1) that forms after perception of bacterial flagellin by FLS2 [7]–[9]. Upon direct interaction between the receptor kinases and the bacterial effectors downstream signalling events are abolished, albeit through different mechanisms. AvrPtoB also targets the receptor kinase CHITIN ELICITOR RECEPTOR KINASE 1 (CERK1) [10].
Plants have evolved a second layer of perception based on the presence of pathogen effectors within the host cell. Host resistance (R) proteins containing a central nucleotide-binding (NB) motif, of the STAND class, and C-terminal leucine rich repeats (LRRs) recognise specific effectors directly or indirectly, and induce strong defences leading to hypersensitive cell death response (HR). The responses induced by PRRs and NB-LRRs respectively have not been separated experimentally, and it is thought that these classes of proteins simply represent different points of entry to the same defence network [11]. While direct activation of NB-LRRs is self-explanatory, the paradigm of indirect effector recognition is of particular interest and illustrates the ingenuity of evolution. Indirect recognition follows a principle in which an accessory protein forms a complex with the NB-LRR protein. In theory, the accessory protein is either a molecular target of the virulence effector, or a mimic of one [1]. Examples of such accessory proteins include Pto kinase, RIN4 and PBS1 kinase, for the NB-LRR proteins Prf, RPM1/RPS2 and RPS5, respectively [5].
The mechanisms by which accessory proteins communicate effector binding to the NB-LRR protein are unknown. For the described examples, each complex exists prior to effector interaction. Accessory proteins make contact with regions of the R proteins N-terminal to the NB domain [12]. Recently, NB-LRR protein oligomerisation has been described for the Prf [13], RPS5 [14], MLA10 [15], and L6 [16] proteins, and this also occurs through N-terminal sequences. However, the significance of oligomerisation for plant immunity is not yet clear [17].
The Pto/Prf complex recognises both the AvrPto and AvrPtoB effectors. The recognition event occurs through the accessory protein kinase Pto [18]. Prf oligomerises through a novel N-terminal (N-term) domain, which also coordinates binding of Pto-like kinases, thus bringing them into proximity [13]. Although AvrPto and AvrPtoB are not related structurally, both interact with Pto predominately via the kinase P+1 loop [19]–[22]. The P +1 loop normally positions the peptide substrate within the catalytic cleft for phosphorylation. Interestingly, mutations within this loop lead to a constitutive gain-of-function (CGF) phenotype of effector-independent HR [19]. In addition, P+1 loop mutations abrogate Pto kinase activity, and activate Prf-dependent signalling [20], [23]. While Pto requires kinase activity for effector-dependent activation, it is dispensable for CGF forms [20]. Overall, the role of kinase activity in control of Prf signalling is not understood but is of critical importance.
Here we show that Pto molecules transphosphorylate each other, but only when in complex with Prf, and only under conditions of complex activation. Pto was doubly phosphorylated within the kinase activation segment, and this was necessary but not sufficient for signalling. Full activation of the complex required additional disturbance of the P+1 loop, by mutation or by interaction with the effector. We derive a model for activation of the complex by effectors, and show how the oligomeric arrangement of the complex provides a trap for the effector that is very difficult for the pathogen to avoid.
To elucidate the role of Pto kinase activity within the Prf complex, we reconstituted this complex in the model species Nicotiana benthamiana by heterologous expression of its constituent components. In this system, co-expression of the tomato Pto and Prf proteins confers recognition of the effectors AvrPto and AvrPtoB leading to HR [24]. Although Pto kinase activity is required for its effector-dependent activation [19], previous experiments to detect activatory phosphorylation have not separated uncomplexed Pto from the small fraction that is bound to Prf [25]. To overcome this, we used Agrobacterium tumefaciens to express Prf transiently as a genetic fusion with three C-terminal haemagglutinin epitopes (Prf-3HA) in stable transgenic 35S:Pto N. benthamiana plants [24], which allowed us to purify Pto within the Prf complex by co-immunoprecipitation using anti-HA antibodies. We found that co-expression of AvrPto or AvrPtoB with the Pto/Prf complex correlated with the appearance of a slow-migrating form of Pto on SDS-PAGE (Figure 1A). A similar Pto band shift was observed previously [25] and its slight appearance in the empty vector (EV) control lacking effectors (three days post infiltration) was correlated with the ligand-independent signalling phenomenon in which overexpression of Pto and Prf induces HR (Figure S1A). This band shift of Pto was previously attributed to phosphorylation as it could be removed by treatment with phosphatase, but the phosphorylation sites were not identified [25]. Prf contains a central nucleotide-binding region conserved with plant and animal proteins of the NOD family [26]. Interestingly, mutation of a conserved residue within this region required for ATP binding, Lys-1128 (prfK1128A) [27], abolished the appearance of the slow migrating Pto band after co-expression with AvrPto or AvrPtoB (Figure 1A). This mutation also strongly diminished both the ligand-independent and effector-triggered HRs (Figures S1A and S1B). Taken together, these results demonstrate that AvrPto and AvrPtoB recognition by the Pto/Prf complex correlates with the appearance of a slow-migrating form of Pto and requires a functional Prf protein.
To investigate the observed band shift of Pto, we initially attempted to purify it from within the Prf complex by co-immunoprecipitation from N. benthamiana after heterologous expression of Prf -3HA, FLAG-tagged Pto, and effectors. After immunoprecipitation of Prf using anti-HA antibodies, we were unable to identify the putative Pto phosphorylation sites in these experiments for technical reasons. Subsequently, the total Pto protein comprising both the Prf-complexed and free forms were purified from N. benthamiana. Immunoprecipitated Pto was subjected to SDS-PAGE fractionation, in-gel tryptic digestion and mass spectrometric analysis. In three independent experiments, we identified peptides spanning over 80% of the Pto protein and found that residues Ser-198 and Thr-199 within the kinase activation segment peptide K/GTELDQTHLSTVVK were modified by phosphorylation (Table 1, Tables S1 and S2), a typical regulatory event in eukaryotic kinases [28]. Before effector recognition, Pto contained a single phosphorylation event attributed predominantly to Ser-198, as found previously [13], but some MS spectra also supported a single phosphorylation event on Thr-190 or Thr-199 (Figures S2 and S3).
Doubly phosphorylated peptides were identified under conditions of effector-activated signalling. The most frequently observed and strongly supported positions were Ser-198 and Thr-199 (Figures S2 and S3), although combinations of other sites were occasionally observed (Table 1, Tables S1 and S2). Despite the significantly fewer incidences of doubly phosphorylated peptides after AvrPtoB recognition in comparison to AvrPto recognition, the complete absence of the doubly phosphorylated peptides in the EV control (2 days post infiltration) is clear. Doubly phosphorylated peptides were also identified when signalling was activated in a ligand-independent manner (Table 1, EV 3 days post infiltration). Additional experiments were performed to elucidate the role of Prf activation in the appearance of doubly phosphorylated peptides.
Data presented above with the prfK1128A loss of function mutation (Figure 1A) suggested that Prf activation influences Pto phosphorylation status. To explore this further, we created a mutation within the conserved MHD motif [29] of the NB domain that conferred an effector-independent CGF phenotype to Prf (Figure 2A) as previously described for Prf [30] and other NB-LRR proteins [27]. Using the same experimental system described above, we found that expression of the prfD1416V CGF mutant greatly increased the proportion of the slow Pto form relative to co-expression of Pto with wild-type Prf, or with the isolated Prf N-term domain that constitutes the Pto binding moiety [25] (Figure 1B). In two independent experiments, the doubly phosphorylated K/GTELDQTHLSTVVK peptide was again identified under these conditions (Tables 1 and Table S2). Thus, active forms of the Prf complex are associated with double phosphorylation of Pto on this peptide. Interestingly, single phosphorylation of Ser-198 or Thr-199 was observed previously, among other in vitro phosphosites [31] that were not identified in these experiments with the exception of Ser-11 that was phosphorylated in an activation-independent manner.
To determine the function of the identified Pto phosphorylation sites, we mutated them individually and in combination to non-phosphorylable alanine. We tested the ability of the substitution mutants ptoS198A, ptoT199A and ptoS198A/T199A to cause cell death upon effector recognition by trypan blue staining. We used this qualitative assay of cell death and an image based estimation of cell death (Relative HR index) in all subsequent HR assays. Both single mutants induced cell death after AvrPto or AvrPtoB co-expression, comparable to wild-type Pto (Figure 2A and Figure S4) possibly by phosphorylation of the secondary sites Thr-190 and Thr-195 as previously observed (Table 1, Tables S1 and S2). In contrast, the double mutant ptoS198A/T199A and the kinase-dead mutant ptoD164N [19] were severely impaired in their ability to support signalling. The CGF activity of prfD1416V was also strongly diminished by co-expression with ptoS198A/T199A or kinase-dead ptoD164N. Furthermore, co-expression of the prfK1128A mutant, that prevents the appearance of the slow migrating Pto band (Figure 1A), also diminished the CGF phenotype of prfD1416V (Figure S5A) by direct interaction (Figure S5B). Thus, phosphorylation of at least one of these residues (Ser-198 or Thr-199) is required for full activation of the Prf complex. The ability of each mutant to support cell death was again tightly correlated with the presence of a slower migrating Pto band, and the absence of the slower band in the ptoS198A/T199A mutant indicates that phosphorylation of at least one of these sites is required for the band shift (Figure 2B and C).
To assess if Ser-198 and Thr-199 are required for Pto kinase activity, we tested the mutants described above for autophosphorylation activity or the ability to transphosphorylate the substrates Pti1 [32] and AvrPtoB [33]. Wild-type Pto and the mutants ptoS198A, ptoT199A and ptoS198A/T199A were active kinases. The estimated relative autophosphorylation activities of ptoS198A, ptoT199A and ptoS198A/T199A were comparable to wild-type Pto (Table 2 and Figure S6). Most importantly, their relative transphosphorylation activities were not correlated with ability to signal. ptoT199A and ptoS198A/T199A were able to transphosphorylate AvrPtoB and Pti1K96N at comparable levels in vitro (Figure S6) but ptoT199A induced a much stronger cell death in vivo after AvrPtoB recognition (Figure 2A). Furthermore, in contrast to the kinase inactive mutant ptoD164N, the kinase active ptoS198A/T199A did not support cell death upon recognition of the E3 ligase mutant avrPtoBF479A [33] (Figure S7A and B), further substantiating the notion that the kinase activity is not correlated with the ability to signal. Therefore, double phosphorylation of Pto activation segment including phosphorylation of at least one of Ser-198 and Thr-199 is the signalling determinant, not kinase activity per se.
Prf forms oligomers through its novel N-term domain, bringing Pto monomers into proximity. This suggests the potential for Pto transphosphorylation [13]. To test this, we devised an assay for transphosphorylation of Pto within the Prf complex in the presence of AvrPto or avrPtoBF479A. The E3 ligase active AvrPtoB was not used, as it results in degradation of the kinase-dead ptoD164N [33]. Wild-type Pto was expressed as a fusion with five Myc epitopes (5Myc), whereas the Pto mutants described above were co-expressed as fusions with the FLAG tag, allowing differential detection of Pto or its mutant forms using appropriate antibodies. Both forms were recovered from the complex by immunoprecipitation of Prf-3HA or prfK1128-HA, and analysed by immunoblotting. Use of anti-Myc (to detect wild-type Pto) detected both fast and slow migrating forms in the presence of Pto-FLAG and Prf-HA, but the slow form was again suppressed by the presence of prfK1128-HA (Figure 3A and B). Importantly, the slow form was unaffected by the presence of the kinase-active ptoS198A, ptoT199A, or ptoS198A/T199A mutants, but was severely curtailed by kinase-dead ptoD164N (Figure 3 and Figure S8). These data suggest that phosphorylation of Pto within the Prf complex leading to the slow migrating form is a transphosphorylation event. Transphosphorylation between Pto molecules was previously observed in E. coli [22] in the absence of Prf, but in this study in planta, transphosphorylation required a functional Prf to induce proximity.
We next tested whether double phosphorylation on Ser-198 and Thr-199 is sufficient for activation of Pto, as are P+1 loop CGF mutants. To do this we substituted both residues for Asp (ptoS198D/T199D), which mimics the negative charge of phosphorylation. Expression of ptoS198D/T199D did not induce effector-independent HR, but the mutant was able to respond to AvrPto and AvrPtoB (Figure 4A and Figure S9A) in contrast to the ptoS198A/T199A variant described earlier (Figures 2A). Further introduction of the kinase-dead mutation (ptoD164N/S198D/T199D) weakened but did not prevent recognition of AvrPto (Figure 4A and Figure S9A), in contrast to ptoD164N (Figure 2A). This demonstrates that Pto kinase activity is required for phosphorylation of Ser-198 and Thr-199 during AvrPto recognition, but is dispensable thereafter. Conversely, AvrPtoB was not recognised by ptoD164N/S198D/T199D, consistent with its ability to degrade kinase inactive forms of Pto [33].
To test the requirement for transphosphorylation in activation, we co-expressed kinase-inactive ptoD164N with Pto or ptoS198D/T199D in the presence of AvrPto. The inactive kinase suppressed signalling by wild-type Pto (Figure 4B, Figure S9B and Figure S10A), but not by the phosphomimic form ptoS198D/T199D (Figure 4B, Figure S9B). Thus, the kinase mutant suppressed transphosphorylation of Pto, but this effect was negated in the kinase active phosphomimic mutant. These data further show that Ser-198 and Thr-199 are the major residues in Pto that require transphosphorylation for Prf activation and downstream signalling. Lastly, substitution of Ser-198 and Thr-199 for non-phosphorylable Ala in the kinase-inactive CGF mutant ptoL205D (ptoS198A/T199A/L205D) abrogated its HR-inducing ability and resulted in a marked band shift of ptoS198A/T199A/L205D to a fast-migrating, not phosphorylated form, in comparison to ptoL205D (Figure 4C and Figure S9C), suggesting that this kinase-inactive mutant must be phosphorylated in trans, perhaps in this system by a N. benthamiana Pto homolog. To test this, ptoL205D was co-expressed with ptoD164N, Pto, or the substitution mutants ptoS198D/T199D and ptoS198A/T199A. ptoD164N suppressed the ptoL205D CGF HR (Figure 4D, Figure S9D and Figure S10B), but the other molecules, which possess kinase activity, did not (Figure 4D and Figure S9D). Taken together, the data show that both P+1 loop disruption (through effector interaction or CGF mutation) and auto and trans phosphorylation are necessary for Pto activation.
We show here that activation of the Prf complex is associated with double phosphorylation of Pto kinase within its activation segment. The dual phosphorylation was seen in each signalling-active event, after effector activation or when Pto was complexed to a CGF form of Prf, and required an intact P-loop within the Prf NB subdomain. This shows that Prf is an active participant in the activation process, consistent with previous findings [23], although the role of Prf in downstream signalling is not well understood. Phosphorylation resulted in a marked band shift of Pto on SDS-PAGE to a slower migrating form. This form was never seen in the absence of tomato Prf and was detectable only when Pto was copurified with the Prf complex and further isolated with long SDS-PAGE. Therefore, in transient expression experiments where Pto is overexpressed relative to Prf, the majority of Pto within the cell is not trans-phosphorylated. Dual phosphorylation was essential for activation of Pto, could be mimicked by replacement of the phosphoresidues with Aspartate, and was destroyed by Alanine replacement. Together, our data demonstrate a strict requirement for Pto kinase activity after effector interaction. Despite this, phosphorylation was not sufficient for activation and required additional disruption of the P+1 loop.
The role of Pto kinase activity in the function of the Prf complex has previously been obscure. We showed previously that Pto kinase activity was dispensable for binding both AvrPto and AvrPtoB [19], [20]. This interpretation was challenged by Xing et al (2007) who found that Pto in complex with AvrPto was phosphorylated on Thr-199. In their model, Thr-199 is required for effector interaction, but the kinase mutant ptoD164N binds both AvrPto and AvrPtoB, and the substitution mutant ptoT199A still recognized both effectors in vivo (Figure 2A). Although Pto kinase activity is dispensable for effector binding, it is clearly required for effector-mediated complex activation [19]. Interestingly, the kinase-inactive ptoD164N did not appear to be transphosphorylated or able to signal in most of our assays. In contrast, we showed here (Figure S7) and previously [33] that ptoD164N is able to initiate weak signalling after avrPtoBF479A recognition suggesting that the requirement for autophosphorylation can be eventually bypassed by transphosphorylation. Nevertheless, is important to emphasize that the signalling mediated by the inactive kinase is weak and delayed (Figure S7). A model where initial autophosphorylation is not essential for effector interaction, but is a prerequisite for fast and efficient disruption of the P+1 loop could explain these discrepancies. Consistent with this model, the constitutively active P+1 loop mutant ptoL205D did not require kinase activity for autophosphorylation but needs to be transphosphorylated by a second kinase-active Pto for induction of cell death. Thus, co-expression of the kinase-inactive ptoD164N with ptoL205D inhibited constitutive signalling, whereas ptoD164N did not impair signalling by the phosphomimic ptoS198D/T199D mutant. Importantly, ptoS198D/T199D did not have a CGF phenotype suggesting that phosphorylation alone is not sufficient to activate the complex. Despite the need of validation of our model in tomato plants with bacterial-derived effectors, our data support a model in which Pto transphosphorylation after effector interaction is an essential step in activation of the recognition complex, but is not sufficient for complex activation which requires further disruption of the kinase P+1 loop (Figure 5 and Table S3).
Our data provides linked explanations for two phenomena: Why Prf exists in a multimeric complex, and secondly, how the kinase inhibitor AvrPto activates the Prf complex in a manner dependent on Pto kinase activity. In crystal structures, AvrPto binds to the Pto catalytic cleft occluding the active site, and inhibits Pto kinase activity in vitro. Likewise, AvrPto inhibits the kinase activities of many PRRs. How then can a kinase inhibitor act as an activator of the Pto/Prf complex? We proposed the following model derived in large part from the current data set. Binding of AvrPto to Prf-associated, previously autophosphorylated sensor Pto disrupts the P+1 loop and hence the negative regulation imposed by Prf [23]. Derepression of the P+1 loop activates a second helper Pto molecule in the Prf complex, either directly or indirectly, mediated through the Prf NB moiety. The second kinase molecule transphosphorylates the first, leading to full activation of the complex. It is tempting to speculate that similar transphosphorylation events within the activated complex lead to phosphorylation of downstream targets of Pto.
This mechanism could not work unless the Prf complex was multimeric. In a monomeric complex, the outcome of effector-Pto interaction would be kinase inhibition, as has been shown for the PRRs. In contrast, our results separate alternate Pto moieties as either sensor or helper kinases depending on which perceives the effector molecule. Such a mechanism is most likely to be successful at early stages of infection when the molecules of Pto/Prf will always outnumber the effector molecules. In this way, Pto acts as the bait in a molecular trap for effectors, which target protein kinase domains. This idea is particularly compelling because of the high similarity between Pto and the kinase domains of most plant receptor kinases, notably CERK1 [10], many of which are likely to be PRRs. We speculate that the Pto complex evolved subsequent to evolution of pathogen effectors that target PRR domains. Indeed, it is tempting to speculate further that Pto itself was derived from duplication of a genetic fragment encoding a targeted PRR kinase domain, and that the novel Prf N-term domain evolved to exploit the kinase-effector interaction which was evolved previously by the pathogen. All pathogens need to suppress PRR kinases, so in this context, the Pto/Prf complex is an ingenious molecular trap for kinase-tropic effectors.
Another example of a protein kinase targeted by an effector protein and interacting with a NB-LRR protein is the Arabidopsis PBS1. The Pseudomonas syringae effector protein AvrPphB is a cysteine protease that targets PBS1 for cleavage. The NB-LRR protein RPS5 monitors the integrity of PBS1 and is activated upon PBS1 cleavage by AvrPphB [14], [34], [35]. Similarly to Prf, RPS5 forms dimers or oligomers [14]; but in contrast to the Pto/Prf complex no need for transphosphorylation has been demonstrated within the PBS1/RPS5 complex. Initially it was shown that the PBS1 kinase activity was required for RPS5 activation [34], [36] but more recent results indicated that the kinase activity of PBS1 is dispensable for signalling [37]. The authors proposed a model where the NB-LRR protein is activated by conformational changes of the guard (sensor) protein caused by the effector protein, without the need of kinase activity [37]. Our data do not quite fit this model because Pto is clearly functional with a requirement for kinase activity documented here, and Prf plays an intimate co-regulatory role with Pto. In our model the NB-LRR protein is activated in the presence of the effector by simultaneously triggered conformational changes and transphosphorylation of the guarded (sensor) kinase. Thus Pto acts a sophisticated bait for the effector, based on its kinase activity and its highly relatedness to the kinase domains of PRRs.
Growth and transient expression conditions for N. benthamiana were as described [20] using the A. tumefaciens strain C58C1. Transgenic N. benthamiana lines used were wt11c containing ProPrf:Prf-5Myc [24] and 38-12 containing 35S:Pto [38]. Qualitative cell death assays were performed by boiling leaves in lactophenol trypan blue solution including 60% ethanol and clearing with chloral hydrate. Cell death stains dark blue in this assay. Cell death was estimated as Relative HR Index using ImageJ, as the darkly stained area in a total leaf area of 0.5 mm2 (vascular tissue was omitted from the calculations). The cell death was estimated using images of cell death from three independent experiments using two leaves in each experiment. All pictures were taken two days post infiltration.
For the analysis of protein accumulation from N. benthamiana, leaves frozen in liquid nitrogen were added to extraction buffer, 150 mM Tris-HCl (pH 7.5), 150 mM NaCl, 5 mM EDTA, 5% glycerol (v/v), 10 mM dithiothreitol (DTT), 2% polyvinylpolypyrrolidone (PVPP), 1% plant protease inhibitor cocktail (Sigma), and 0.5 mM PMSF and homogenized with a Polytron. Protein extracts were centrifuged at 20,000 g for 20 min at 4°C. Supernatants were subjected to filtration through a 0.45 µm filter. Sepharose affinity matrices used were anti-FLAG M2 and anti-HA HA-7 (both Sigma). Extracts were mixed with affinity matrices as indicated for two hours at 4°C, with gentle rotation, in batch format. Affinity matrices were washed three times with an excess of extraction buffer. Proteins were stripped from the bead fraction by boiling in SDS loading buffer. Elution from anti-FLAG beads was performed by incubation with extraction buffer containing 200 µg/mL FLAG peptide (Sigma) for 10 min at 25°C with gentle mixing. To concentrate elutions, Strataclean (Stratagene) beads were added to bind proteins, and then pelleted by centrifugation subsequently the proteins were stripped from the beads by boiling in 1× SDS-PAGE loading buffer.
In vitro kinase assays were performed as described [39] with slight modifications. Briefly, the kinase reaction mixture contained 50 mM Tris-HCl (pH 7.5), 10 mM MgCl2, 1 mM MnCl2, 1 mM DTT, 20 µM ATP, 183 kBq of γ[32P]-ATP (PerkinElmer Life Sciences) in a total volume of 30 µl. In Pto transphosphorylation assays, Pto-FLAG, ptoD164N-FLAG, ptoS198A-FLAG, ptoS199A-FLAG and ptoS198A/T199A-FLAG were transiently expressed in N. benthamiana and immunoprecipitated with anti-FLAG beads as described above. pti1K96N-His and GST-AvrPtoB were expressed and purified from Escherichia coli as previously [33]. 5 µg of pti1K96N-His and 2 µg of GST-AvrPtoB were included in the kinase reaction mixture. All reactions were initiated by addition of the kinase mixture, incubated at 30°C for 20 min, and terminated by addition of SDS-polyacrylamide gel electrophoresis (SDS-PAGE) loading buffer and boiling for 10 min. Under these assay conditions, incorporation of radiolabel was found to be linear with time, the substrate and the enzyme concentration used. At the end of each assay, samples were loaded onto SDS-PAGE. Post electrophoresis, proteins were transferred onto polyvinylidene difluoride membranes and stained with Coomassie Brilliant Blue R-250. Subsequently, the membranes were subjected to autoradiography using a FUJI Film FLA5000 PhosphorImager (Fuji, Tokyo, Japan). Relative autophosphorylation and transphosphorylation kinase activity was calculated as the ratio between incorporated radioactivity (PhosphorImager signal) and the amount of immunoprecipitated protein estimated using ImageJ based on Coomassie staining of the membranes and expressed as a percentage of Pto-FLAG relative autophosphorylation or transphosphorylation kinase activity.
Co-immunoprecipitated protein complexes were separated by SDS-PAGE and gel slices were excised. Proteins were reduced with DTT and alkylated with iodoacetamide before in-gel trypsin (Promega) digestion overnight at 37°C. After digestion, the supernatant was moved to a clean tube and the gel pieces washed sequentially with 50% and 100% acetonitrile and the washes pooled with the supernatant. Volume and the organic content of the peptide solution were reduced by lypholisation and the peptides stored at −20°C until use. Peptides were dissolved in 0.5% formic acid immediately before analysis by LC MS/MS.
LC-MS/MS analysis was performed using a LTQ-Orbitrap XL mass spectrometer (Thermo Scientific) and a nanoflow-HPLC system (Surveyor, Thermo Scientific). Peptides were applied to a precolumn (C18 pepmap100, LC Packings) connected to a self-packed C18 10-cm analytical column (BioBasic resin, Thermo Scientific. Picotip 75 µm id, 15 µm tip, New Objective). Peptides were eluted by a gradient of 2 to 50% acetonitrile in 0.1% formic acid over 50 min at a flow rate of approximately 250 nL min-1. Data-dependent acquisition of MS/MS consisted of selection of the five most abundant ions in each cycle: MS mass-to-charge ratio (m/z) 300 to 2000, minimum signal 1000, collision energy 35, 2 repeat hits, 60 sec exclusion. MS3 were triggered if the neutral loss of phosphoric acid (49 m/z for 2+ parent ions) was detected in the three most abundant ions on the preceding MS2. Collision energy for MS3 was 35. In all cases the mass spectrometer was operated in positive ion mode with a nano-spray source and a capillary temperature of 200°C, no sheath gas was employed and the source and focusing voltages were optimised for the transmission of angiotensin.
Peak lists (as .dat files) were prepared from raw data using extract_msn in BioWorks 3.3 (Thermo Electron Corp.) and collated using merge.pl (Matrix Science). The data generation parameters were: MW range 300.00–3500.00, threshold absolute 1000, group scan 10, minimum group 0, minimum ion cont 10, charge state auto (ZSA processing; default values) MS level auto. Peak lists were searched against SPtrEMBL, (containing 8385695 sequences) Taxonomy was restricted to Solaneae (7247 sequences) with the following variable modifications were allowed; oxidized methionine, phosphorylation on serine and threonine. Carbamidomethyl was specified as a fixed modification on cysteine residues. Precursor mass tolerance was 5 ppm, fragment tolerance 0.5 Da mass values were monoisotopic and two missed tryptic cleavages were allowed. Subsequently for proteins identified with >95% probability an ‘error tolerant’ search was preformed in Mascot with relaxed criteria for modifications and enzyme cleavage. Scaffold (version Scaffold_2_03_01, Proteome Software Inc.) was used to validate MS/MS based peptide and protein identifications. Peptide identifications were accepted if they could be established at greater than 95.0% probability as specified by the Peptide Prophet algorithm [40]. Protein identifications were accepted if they could be established at greater than 95.0% probability and contained at least 2 identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm [41].
In addition to the minimum PeptideProphet score provided by Scaffold, we manually evaluated the fragmentation spectra of all phosphorylated peptides of Pto to ensure that good b and y ion coverage was observed and that neutral loss of the phosphate (common in ion trap CID) supported the assigned phosphorylated residue. We used our PhosCalc algorithum to assist this interpretation (MacLean et al 2008). The peptides covering the activation loop (KGTELDQTHLSTVVK) were observed with and without a tryptic miscleavage of the N-terminal K, and single and double phosphorylation events were observed on both tryptic fragments. Furthermore, the single miscleaved peptide was observed in both 2+ and 3+ ionisation states, while most other peptides were predominantly 2+, thus providing abundant evidence of mass shifts and fragmentation patterns. This peptide contains four possible phosphorylation sites. We observed spectra, which supported single phosphorylation at T190, S198 or T199. Site S198 was most strongly supported and frequently observed. We also observed a doubly phosphorylated form of KGTELDQTHLSTVVK, again with and without the N-terminal miscleavage. Manual examination of the spectra supported phosphorylation at S198 and T199 in most cases, occasional spectra supported phosphorylation at T190 and either S198 or T199. It is important to note that even in cases where the exact site of modification remains ambiguous this does not detract from the double phosphorylation of the peptide as a whole, due to the clear difference in parent ion masses and overall fragmentation. Representative fragmentation spectra for the peptide (KGTELDQTHLSTVVK) are shown in Supplementary Figures S2 and S3. Spectrum counts were summed over both cleavage forms and scored as unmodified, single or double phosphorylation, all counted spectra were at the 95% peptide prophet threshold provided by Scaffold.
Sequence data from this article can be found in the GenBank data library under the following accession numbers: Prf (U65391), Pto (DQ019170), avrPto (L20425), avrPtoB (Q8RSY1).
|
10.1371/journal.pcbi.1004011 | Disrupted Calcium Release as a Mechanism for Atrial Alternans Associated with Human Atrial Fibrillation | Atrial fibrillation (AF) is the most common cardiac arrhythmia, but our knowledge of the arrhythmogenic substrate is incomplete. Alternans, the beat-to-beat alternation in the shape of cardiac electrical signals, typically occurs at fast heart rates and leads to arrhythmia. However, atrial alternans have been observed at slower pacing rates in AF patients than in controls, suggesting that increased vulnerability to arrhythmia in AF patients may be due to the proarrythmic influence of alternans at these slower rates. As such, alternans may present a useful therapeutic target for the treatment and prevention of AF, but the mechanism underlying alternans occurrence in AF patients at heart rates near rest is unknown. The goal of this study was to determine how cellular changes that occur in human AF affect the appearance of alternans at heart rates near rest. To achieve this, we developed a computational model of human atrial tissue incorporating electrophysiological remodeling associated with chronic AF (cAF) and performed parameter sensitivity analysis of ionic model parameters to determine which cellular changes led to alternans. Of the 20 parameters tested, only decreasing the ryanodine receptor (RyR) inactivation rate constant (kiCa) produced action potential duration (APD) alternans seen clinically at slower pacing rates. Using single-cell clamps of voltage, fluxes, and state variables, we determined that alternans onset was Ca2+-driven rather than voltage-driven and occurred as a result of decreased RyR inactivation which led to increased steepness of the sarcoplasmic reticulum (SR) Ca2+ release slope. Iterated map analysis revealed that because SR Ca2+ uptake efficiency was much higher in control atrial cells than in cAF cells, drastic reductions in kiCa were required to produce alternans at comparable pacing rates in control atrial cells. These findings suggest that RyR kinetics may play a critical role in altered Ca2+ homeostasis which drives proarrhythmic APD alternans in patients with AF.
| Atrial fibrillation is an irregular heart rhythm affecting millions of people worldwide. Effective treatment of this cardiac disorder relies upon our detailed knowledge and understanding of the mechanisms that lead to arrhythmia. Recent clinical observations have suggested that alternans, a phenomenon where the shape of the electrical signal in the heart alternates from beat to beat, may play an important role in this process, but the underlying mechanisms remain unknown. In this study, we use computational models to conduct a detailed examination of the causes and contributors to alternans associated with human atrial fibrillation. We find that in atria remodeled by atrial fibrillation, alternans appears near resting heart rates because several aspects of calcium cycling are disrupted in the atrial cells. In particular, the release and uptake of calcium from the cellular storage compartment, the sarcoplasmic reticulum, becomes imbalanced, leading to alternation in calcium signals from beat to beat. These findings provide important insights into the mechanisms of proarrhythmic alternans in human atrial fibrillation which may be used to develop novel therapeutic targets and treatment strategies in the future.
| Atrial fibrillation (AF) is currently the most common cardiac rhythm disorder, posing a significant medical and economic challenge for the US health care system [1], [2]. This burden is likely to increase as the population ages and AF prevalence rises [3]. Effective prevention and treatment of AF depends upon advances in our understanding of underlying disease mechanisms. Although several features of AF electrophysiological remodeling have been identified over the past decades [4], [5], our knowledge about the arrhythmogenic substrate remains incomplete.
Beat-to-beat alternation in the shape of cardiac electrical signals, a phenomenon called alternans, has been observed in the atria of AF patients, but the mechanism underlying these alternans is not known [6]–[11]. Narayan et al. reported differences in the rate dependence of action potential duration (APD) alternans in patients, with APD alternans occurring at pacing rates near rest in AF patients but only at fast pacing rates in controls [8]. Narayan et al. also found that APD alternans always preceded AF initiation, indicating that alternans may play an important role in establishing the arrhythmogenic substrate and creating vulnerability to AF. Thus, a better understanding of AF arrhythmogenesis will likely depend upon identification of the mechanism driving atrial alternans at heart rates near rest.
Interestingly, in AF patients the slope of the APD restitution curve was <1 during APD alternans onset at slow pacing rates. This suggests that a cellular mechanism other than voltage-driven instability underlies APD alternans at heart rates near rest [9]. Altered Ca2+ handling in atrial myocytes is known to play a crucial role in the generation of AF triggers and in AF maintenance [12], [13]. Ca2+ cycling instabilities have been shown to underlie ventricular alternans in heart failure [14], [15], as well as atrial alternans in several non-AF animal models [16]–[18]. However, it is unknown whether these represent a plausible mechanism for atrial alternans in AF patients, particularly at heart rates near rest. We therefore sought to determine, using a computer model of human atrial tissue, whether Ca2+ handling abnormalities, or other electrophysiological changes that occur in AF, lead to APD alternans. We identified a critical change in the kinetics of the ryanodine receptor (RyR) that was responsible for APD alternans onset at slower pacing rates, and subsequently aimed to elucidate the mechanistic relationship between this disruption in RyR kinetics and alternans onset. To this end, we employed single-cell clamping of ionic model parameters and iterated map analysis in order to dissect the mechanisms which drive alternans in atrial tissue, as well as to provide important insights into the pathophysiological changes that contribute to the development of alternans in AF patients.
In order to investigate ionic mechanisms in human AF that contribute to the generation of atrial APD alternans at the tissue level, we created a computer model of human atrial tissue incorporating ionic remodeling associated with chronic AF (cAF), as described in Methods. The sensitivity of APD alternans to ionic model parameters was evaluated by varying parameters one at a time and applying the clinical pacing protocol used by Narayan et al. to induce APD alternans in AF patients [8] (see Table 1 and Methods). For control, a model of normal human atrial tissue was also simulated. We then assessed the magnitude and onset pacing cycle length (CL) of APD alternans by analyzing voltage traces from the recording electrode (Fig. 1A), as outlined in Methods.
In the control model, significant APD alternans did not occur before loss of capture at 260 ms CL (Fig. 1B). However, in the cAF-remodeled tissue preparation, significant APD alternans appeared at a CL of 240 ms (Fig. 1B). Varying the RyR inactivation rate constant (kiCa) had the greatest effect on alternans onset CL in the human cAF-remodeled tissue (Fig. 2A). In fact, only reduction of kiCa resulted in alternans onset at CLs of 300–500 ms (Fig. 2B), matching alternans onset CLs observed in AF patients [8]. When other ionic model parameters were varied from their original cAF values, APD alternans either did not appear in the tissue model at CL≥300 ms (Fig. 2A, blue areas), appeared only at CL≤350 ms (Fig. 2A, red areas), or did not appear before loss of capture or conduction block occurred in the tissue (Fig. 2A, white spaces). These results suggest that altered RyR kinetics is the critical cellular component underlying the occurrence of APD alternans in AF patients at pacing rates near rest, and that kiCa plays a key role in this process.
We also tested whether differences between left and right atrial electrophysiology affect alternans susceptibility using a right atrium (RA) version of the cAF model [19] in tissue simulations. Results for RA tissue were very similar to those for the left atrium (LA), demonstrating that modulation of kiCa could reproduce alternans observed at pacing rates near rest in both the LA and RA of AF patients [8] (S2 Figure).
When kiCa was decreased by 50% in the cAF model (we refer to this as the cAFalt ionic model), APD alternans onset data from the human AF tissue model agreed well with data from persistent AF patients. Significant APD alternans began at 400-ms CL (Fig. 1B, dotted red line), mean APD at onset was 229 ms, and APD alternans magnitude at onset was 27 ms (Fig. 1C, dotted red line). These metrics were each within one standard deviation (SD) of clinical observations [8] (Fig. 3). The cAFalt model also displayed noticeable alternans in intracellular Ca2+ ([Ca2+]i) at the onset CL (Fig. 1D). For both the cAF and cAFalt models, mean APDs were shorter than in the control model (Fig. 1B–C), and diastolic and systolic [Ca2+]i were lower than in control (Fig. 1D). At 400-ms CL in the cAFalt model, on the odd (long) vs. the even (short) beat (Fig. 4, blue vs. red), there was higher sarcoplasmic reticulum (SR) Ca2+ load before release (0.288 vs. 0.273 mM), higher peak RyR open probability (RyRo) (9.0e-4 vs. 4.7e-4), a larger intracellular Ca2+ transient (CaT) amplitude (Δ[Ca2+]i = 0.13 vs. 0.067 µM), similar L-type Ca2+ (LCC) current (integrated over one beat: 144 vs. 140 mC/F), and increased Na+/Ca2+ exchanger (NCX) current (INCX, integrated over one beat: 98.4 vs. 74.5 mC/F). The positive coupling between transmembrane potential (Vm) and Ca2+, with INCX as the primary electrogenic current, is consistent with experimental findings [20]. Since the magnitude and onset of APD alternans in the cAFalt model provided the best agreement with clinical APD alternans data (Fig. 3), we chose to use this model for subsequent investigations into the underlying causes of alternans occurrence.
Since APD alternans throughout the homogenous cAFalt tissue preparation were concordant and of similar magnitude (S3 Figure), electrotonic effects and CV restitution were excluded as factors influencing these alternans. Indeed, APD and CaT alternans in the cAFalt tissue model were very similar to alternans in the isolated single-cell cAFalt model (Fig. 5, left column vs. Fig. 4, top row). We therefore concluded that cellular mechanisms gave rise to alternans in the cAFalt tissue model and decided to utilize single-cell simulations in order to investigate these mechanisms. We first used the ionic model variable clamping protocol described in detail in Methods. The percent change in APD and CaT alternans magnitudes, when each ionic model variable was clamped to its trace from either the even (short) or odd (long) steady-state beat at the alternans onset CL (400 ms), are summarized in Fig. 6 (right column: state variables, left column: currents and fluxes). Variables which resulted in >99% reduction in APD and CaT alternans magnitudes for both even and odd beat clamps were considered essential for alternans.
Clamping Vm resulted in −61.8% change in CaT alternans magnitude for even beat clamps and +6.6% for odd beat clamps, demonstrating that the alternans were not voltage-driven (see even and odd beat clamps depicted in column 2 of Fig. 5 and S4 Figure, respectively). Clamping [Ca2+]i enhanced APD alternans (+55.2% and +75.8% for even and odd beat clamps, respectively, column 3 of Fig. 5 and S4 Figure). However, when SR Ca2+ ([Ca2+]SR) was clamped to either the even or odd beat waveforms, alternans in both APD and CaT were eliminated (<−99%), demonstrating that the alternans were driven by SR Ca2+ instability (column 4 of Fig. 5 and S4 Figure). In addition, four other variables could be clamped to the even or odd beat waveforms to eliminate APD and CaT alternans: RyR inactivated probability (RyRi), RyR open probability (RyRo), junctional Ca2+ ([Ca2+]j), and SR Ca2+ release flux (JSRCarel) (Fig. 6, and S5 and S6 Figures). All five of these variables were therefore critical for enabling alternans to occur at the onset CL. Furthermore, these variables directly impact SR Ca2+ release, implicating SR Ca2+ release as the underlying source of alternans in the cAFalt model.
There were two ionic model components which greatly reduced but did not eliminate alternans when clamped: sub-sarcolemmal Ca2+ ([Ca2+]sl) and sub-sarcolemmal Na+/Ca2+ exchanger current (INCXsl). Clamping [Ca2+]sl to the even beat eliminated all alternans; clamping to the odd beat greatly reduced APD and CaT alternans (−95.8% and −96.2%, respectively), although large alternation in SR load persisted (Fig. 6 and columns 1–2 of S7 Figure). Similarly, clamping INCXsl to the even beat waveform resulted in elimination of APD but not CaT alternans (+72.9%), while clamping to the odd beat waveform resulted in elimination of all alternans (Fig. 6 and columns 3–4 of S7 Figure). Hence, the SR Ca2+-driven instabilities produced alternans in Ca2+ cycling which were positively coupled to voltage through INCXsl and [Ca2+]sl.
Increased steepness of the SR release-load relationship is a well-known mechanism for CaT alternans [21], [22]. The importance of SR Ca2+ release variables for APD and CaT alternans, as demonstrated by the results in Fig. 5, 6, and S4, S5, S6 Figures, led us to hypothesize that such a mechanism might give rise to Ca2+-driven alternans in the cAFalt model at pacing rates near rest. To test this, we compared the cAF and cAFalt ionic models under action potential (AP) voltage clamp conditions so that changes in CaT alternans would be due solely to changes in Ca2+ homeostasis rather than bidirectional coupling between Vm and Ca2+. After clamping each ionic model at a CL of 400 ms until steady state was reached, we perturbed [Ca2+]SR and tracked SR load and SR Ca2+ release on the subsequent clamped beats (see Methods for details). The SR release-load relationships for the cAF (black) and cAFalt (red) ionic models are depicted in Fig. 7 (left column, row 1). The slope of the release-load relationship in the cAFalt model ( = 3.1) was much greater than the slope in the cAF model ( = 1.7), confirming our hypothesis that differences between the cAF and cAFalt ionic models led to a steepening of the SR Ca2+ release slope.
To better explain the differences between the cAF and cAFalt ionic models that gave rise to different SR Ca2+ release slopes, we first compared [Ca2+]SR, RyRo, [Ca2+]j, and cumulative Ca2+ release for the two models at steady state (Fig. 7, left column, rows 2–5, solid lines). In the cAFalt model, [Ca2+]SR at steady state was 19.7% lower than in the cAF model as a result of increased RyR opening (Fig. 7, left column, rows 2 and 3, red vs. black solid lines). Although this led to a 15.2% decrease in peak [Ca2+]j in the cAFalt model, the duration of the release event was prolonged (Fig. 7, left column, row 4, red vs. black solid lines). Consequently, though cumulative Ca2+ release in the cAFalt model initially lagged behind, at t≈90 ms it actually surpassed the cumulative release in the cAF model, ultimately resulting in a 3.4% increase in total release by the end of the beat (Fig. 7, left column, row 5, red vs. black solid lines).
To illustrate how these differences between the cAF and cAFalt ionic models impacted SR release slope, we applied a large perturbation to [Ca2+]SR (+20 µM) at the beginning of a clamped beat and compared the unperturbed (steady state, solid line) and perturbed (dotted line) traces for each model (Fig. 7, left column, rows 2–6). Higher SR load at the beginning of the beat led to increased SR release flux due to luminal Ca2+ regulation of the RyR (causing more opening), as well as to the increased concentration gradient between the SR and junctional compartments. In both the cAF and cAFalt models, these changes led to increased peak [Ca2+]j (+54.4% and +100%, respectively) and RyR opening (+64.6% and +129%, respectively) as a result of more Ca2+-induced Ca2+ release (Fig. 7, left column, rows 2–4). The positive feedback relationship between [Ca2+]j and RyR opening was strong enough such that when SR load was increased (Fig. 7, left column, row 2, dotted vs. solid lines), this actually resulted in a lower minimum [Ca2+]SR during release (−3.6% and −13.3% for cAF and cAFalt models, respectively). However, the amount of positive feedback differed between the cAF and cAFalt ionic models. Positive feedback amplifies changes in release inputs, such as SR load; therefore, in the cAF model, where [Ca2+]j is higher and positive feedback is stronger, the increase in [Ca2+]SR produced a slightly greater change in release (compared to the unperturbed, steady state simulation) during the rising phase of [Ca2+]j (t<48 ms) than in the cAFalt model (Fig. 7, left column, row 6, black vs. red).
By contrast, termination of release occurs through a negative feedback process, with RyRs inactivating upon the binding of junctional Ca2+. Negative feedback attenuates changes in release so that robust, fast termination of release is achieved even when a disturbance (such as a transient increase in SR load) occurs. In the cAFalt model, negative feedback is decreased both directly, via reduction of kiCa, and indirectly, via reduction in [Ca2+]j that occurs as a result of decreased SR load. This causes prolongation of the Ca2+ release event and a larger peak [Ca2+]j (Fig. 7, left column, row 4, red vs. black dotted lines). Consequently, when SR load was increased by the same amount in the cAF and cAFalt models, although the cAFalt model had a lesser initial change in release because of weaker positive feedback, it also had a greater final change in release, i.e. a steeper SR release-load relationship, because of weaker negative feedback (Fig. 7, left column, row 6, red vs. black).
The results in column 1 of Fig. 7 demonstrate how the steeper SR release slope in the cAFalt ionic model (as compared to the cAF ionic model) depends upon RyR inactivation by junctional Ca2+. However, recent work suggests that termination of release does not rely on direct Ca2+-dependent inactivation of the RyR but rather on local SR Ca2+ depletion [23]–[26]. In order to test whether steepening of the SR release slope could occur in the cAF model by an alternative release termination mechanism, we implemented a version of the cAF model in which the RyR Markov model was replaced with that of Sato and Bers and the SR was divided into junctional (JSR) and network (NSR) compartments [27] (see Table 2 and S1 Text). Termination of release in this alternative RyR model relies on calsequestrin (CSQN) binding to the RyR, which occurs as luminal [Ca2+] decreases causing changes in RyR opening and closing rates.
The effects of decreased RyR termination in the Sato-Bers RyR model are shown in the right column of Fig. 7. When the CSQN-bound RyR closing rate k34 (analagous to the inactivation rate kiCa in the original model) is decreased from 100% to 50% (cAFalt), steady-state Ca2+ concentrations change modestly as compared to the original RyR formulation (Fig. 7, black vs. red solid lines), but nevertheless display similar trends: [Ca2+]JSR decreases by 1.5% (vs. 19.7%, row 2), peak [Ca2+]j is reduced by 10.5% (vs. 15.2%, row 4) and delayed, and total release increases by 3.6% (vs. 3.4%, row 5). When [Ca2+]NSR is perturbed in the Sato-Bers models by +20 µM, Ca2+ release increases more in the cAFalt model than in the cAF model (Fig. 7, right column, row 6, red vs. black dotted lines). Consequently, the SR Ca2+ release slope is steeper in the cAFalt model ( = 3.7 vs 1.9, Fig. 7, right column, row 1). Thus, although changes in SR Ca2+ release slope in the original cAF model are caused by altered junctional Ca2+-dependent inactivation, altered SR Ca2+-dependent mechanisms of release termination can produce such changes in SR Ca2+ release slope as well.
Although SR Ca2+ release slope is an important component of Ca2+ homeostasis, other aspects of Ca2+ cycling, such as SR Ca2+ uptake, could also have a significant impact. In order to understand how both SR release and uptake contribute to CaT alternans onset at slow pacing rates in human cAF cells, we used an iterated map analysis for investigating Ca2+ cycling stability under AP voltage clamp conditions. Three factors affecting Ca2+ cycling stability were included in the analysis: SR release, SR uptake, and cellular Ca2+ flux across the sarcolemma. The latter factor was included because Ca2+ content in the human atrial cell model varied significantly enough to affect alternans threshold predictions.
For each version of the human atrial cell model (cAF and control), we calculated the SR Ca2+ release slope (), the SR Ca2+ uptake factor (), and the cellular Ca2+ efflux factor () [28], [29] for a range of kiCa values and pacing rates and compared the value of to the threshold for alternans. For a typical range of parameter values (, see S1 Text), the threshold value of required for alternans is given by the following equation:(1)Theoretical analysis predicts that the system is stable when . Eq. 1 is graphed for a range of values in Figs. 8A–C (dotted lines). Each curve represents the boundary between stable (no alternans) and unstable (alternans) Ca2+ cycling in the - plane for a particular value of . As increases (Fig. 8A–C, dark blue to dark red), the threshold curve steepens, indicating that increased Ca2+ extrusion from the cell has a protective effect, helping to restore Ca2+ content back to steady state following a perturbation. Thus, a higher value of is required to reach alternans threshold for higher values of . Note that in this theoretical approach, increased Ca2+ efflux (κ) has the opposite effect as in Qu et al. [29], suppressing rather than promoting Ca2+ alternans.
The effects of changing CL and changing kiCa are explored for the cAF model in Fig. 8A. At the default kiCa value (100%), as CL is decreased from 700 ms to 200 ms (−10 ms increments), decreases, increases, and the system approaches the alternans threshold given by Eq. 1. The change in values is non-monotonic, initially decreasing (orange to green) and then increasing (green to orange) as CL is decreased. However, the change in has a minimal effect at small values, since the threshold curves for different values converge at . At CL<220 ms, the cell begins to display alternans in Ca2+ cycling, coinciding with the iterated map parameter values residing very close to the theoretically predicted boundary given by Eq. 1 (Fig. 8A, orange X's). When kiCa is set at 50% of the default cAF value (cAFalt model), a similar trend is observed. However, the 50% kiCa cAF model reaches threshold at a lower pacing rate (CL = 390 ms for the 50% kiCa cAF model vs. 210 ms for the 100% kiCa cAF model, Fig. 8A, X's). This is primarily due to increasing as kiCa is decreased, illustrated by the trajectory of the system in the - plane as CL is held constant at 390 ms but kiCa is decreased from 100% to 50% (Fig. 8A).
We next performed the same iterated map analysis for the control atrial cell model with varying CL and kiCa values (Fig. 8B). When kiCa is at 100%, decreases as CL is decreased. However, unlike in the cAF model, in the control case the value of undergoes a net decrease as CL shortens from 700 to 200 ms. Ultimately, since both and decrease as CL is shortened, the control atrial cell (with kiCa at 100%) fails to reach threshold and remains in the stable, no alternans region. This suggests that alternans in control patients, which occur at CL<250 ms [8], are driven by voltage rather than Ca2+. As in the cAF model, the alternans threshold CL in the control model can be adjusted by modulating the value of kiCa (Fig. 8B, CL = 390 ms). However, in the control model, kiCa must be decreased much more than in the cAF model in order to reach at a CL of 390 ms (kiCa reduced to 16% vs. 50%). The need for dramatic and possibly unrealistic reductions in kiCa to produce alternans at slow rates in control is consistent with the absence of alternans observed in control patients at CL≥250 ms [8].
To explain the difference in Ca2+ cycling properties of the cAF and control models, we examined the effects of cAF cellular remodeling on iterated map parameters. Stochastic ionic model parameter variation and regression analysis [30] (see S1 Text) predicted that of the ten model parameters altered in the control model to construct the cAF model, seven would have significant effects on alternans threshold CL (these are gCaL, gKur, koCa, IbarNCX, gto, gK1, and gNa, see S8 Figure). Of these seven parameters, three are involved in Ca2+ handling (gCaL, koCa, and IbarNCX). The effects of changing these three parameters from control to cAF values is depicted sequentially in Fig. 8C: starting with the default values for the control cell at a CL of 390 ms, first gCaL is decreased and then IbarNCX and koCa are increased to cAF values, resulting in an overall decrease in and . Finally, when kiCa is decreased to the cAFalt value (50%), the large increase in causes the system to reach and alternate (Fig. 8C, red X). This illustrates why the control cell is less susceptible to CaT alternans than the cAF cell: at a given kiCa value and pacing rate, SR uptake efficiency () is higher in the control model, thus requiring a large increase in the pacing rate (which decreases ) and/or a large decrease in kiCa (which increases ) in order to reach . Of the three cAF parameters which decrease , however, gCaL is the most important for alternans onset, since remodeling of IbarNCX and koCa decreases , while remodeling of gCaL increases . When gCaL is remodeled and IbarNCX and koCa remain at control values, only a 28% decrease in kiCa is required to reach (Fig. 8C, green X).
The first goal of this study was to identify the electrophysiological changes in human atrial cells that are responsible for the occurrence of APD alternans at heart rates near rest, as observed in AF patients. Using parameter sensitivity analysis, we found that of the 20 electrophysiological model variables tested, only changes in the RyR inactivation rate constant (kiCa) could produce APD alternans at relatively slow pacing rates in a tissue model of persistent/chronic AF. In particular, decreasing kiCa by 50% (the cAFalt model) produced a good match to clinical data. We next aimed to provide mechanistic insight into why disruption of RyR kinetics, together with other electrophysiological changes occurring in AF, leads to alternans onset at pacing rates near rest. We established that alternans in the cAFalt model at the onset CL were Ca2+-driven rather than voltage-driven, and that they depended upon SR Ca2+ release. Furthermore, CaT alternans occurred in the cAFalt model at relatively long CLs because of steep SR Ca2+ release slope and decreased SR Ca2+ uptake efficiency. Lastly, we demonstrated that the ability to generate alternans at slower pacing rates by modulating kiCa depended upon the negative feedback properties of SR Ca2+ release.
This study is the first to identify a possible mechanism for alternans occurring at slow heart rates in AF patients. Our novel findings show that alternans at slow rates is Ca2+-driven, brought about by AF-associated remodeling of the Ca2+ handling system in atrial cells. Clinical and experimental research has shown that atrial alternans is associated with disease progression in AF patients [8] and with increased AF susceptibility after myocardial infarction [31], [32] and atrial tachycardia [33], [34] in animal models. Additionally, CaT alternans have been studied in animal atrial myocytes [17], [18], [35] and in the intact atria of AF-prone mice [36]. However, the precise cellular mechanism underlying alternans at heart rates near rest in the remodeled human atria has not been previously identified, and a direct relationship between human AF and CaT alternans in the atria has not been established until now. Elucidating the mechanism driving alternans at slow rates is particularly important because APD oscillations appear to be closely linked to AF initiation [8]. If APD alternans play a direct role in AF initiation, the onset of alternans at slower pacing rates would indicate an increased susceptibility to arrhythmia in AF patients, consistent with clinical observations [8]. Identification of this mechanism would thus provide a significant scientific and clinical benefit, improving our understanding of arrhythmogenesis and aiding in the development of new targeted therapies for AF.
In this study, we demonstrate how different aspects of AF remodeling contribute to Ca2+-driven alternans onset at slower heart rates using a theoretical analysis of Ca2+ cycling. This analysis allowed us to quantitatively assess CaT alternans threshold under AP voltage clamp conditions in a detailed electrophysiological model, providing valuable insights into the effects of AF electrophysiological remodeling on Ca2+ handling and alternans. Furthermore, we identify a critical aspect of SR Ca2+ release—inactivation of the RyR—which is necessary for CaT alternans to occur at slow heart rates. These findings extend mechanistic insight about proarrhythmic ventricular Ca2+ remodeling [15], [37], [38] to the atria and may inform new therapeutic strategies to target the RyR and suppress Ca2+-driven alternans in the atria for the purposes of preventing or treating AF [36], [39].
The RyR has been the focus of several studies concerning trigger-mediated AF. In particular, disruption of RyR regulation has been shown to promote AF through increased RyR open probability, diastolic SR Ca2+ leak, and delayed afterdepolarizations [12], [39], [40]. Here we identify an additional pathological consequence of the disruption of RyR regulation in AF: Ca2+-driven alternans. Similar to what has been demonstrated with regards to Ca2+ sparks and triggered activity [39], we found that CaT alternans is coupled to voltage primarily through upregulated INCX, thus driving the generation of APD alternans. The RyR's central role in both alternans and triggers has important clinical implications, given the proarrhythmic consequences of interaction between ectopic activity and the arrhythmogenic substrate created by voltage alternans [41]. New drug treatments to restore the normal function of the RyR and NCX, and thereby prevent arrhythmogenic triggers and alternans, have the potential to provide more effective alternatives to current AF drug therapies which target voltage-gated ion channels and often have proarrhythmic side effects [39].
The signaling pathways involved in RyR dysfunction in AF have been the focus of much active research over the past several years [39], [40]. Possible molecular mechanisms which could account for reduced RyR inactivation include RyR hyperphosphorylation by CAMKII and PKA and dissociation of the RyR subunit FKBP12.6, which have been shown to increase RyR open probability and promote arrhythmia [42], although the exact role of these mechanisms in RyR dysregulation are still debated [43]. Calmodulin has also been shown to interact directly with the RyR to decrease its open probability [44]. Metabolic factors may play a role, since modulation of the RyR as a result of glycolytic inhibition has been linked to atrial alternans in non-AF animal models [16], [17], [35]. Such metabolic impairment is thought to contribute to profibrillatory remodeling in the atria [45]–[47]. The cAFalt model, with its reduction in kiCa, can be considered a phenomenological representation of the various signaling pathway disruptions leading to alternans, which were not represented in the original cAF model. As more information becomes available, incorporation of these signaling mechanisms into computational models may provide additional insights into how reduction in RyR inactivation leads to Ca2+-driven alternans at slow heart rates in AF patients.
There is debate over whether CaT alternans depend primarily on SR Ca2+ load alternation or on RyR refractoriness [21], [41], [48]. Recent experiments [18], [49] and simulation studies [50]–[53] have shown that RyR refractoriness can drive CaT alternans under conditions where near-identical SR loads produce different amounts of SR release. In some simulation studies, this phenomenon was restricted to limited parameter values, clamping conditions, and cycle lengths [51], [52], while in a more recent modeling study focusing on atrial cells, SR load-independent alternans occurred over a broad range of pacing rates when the number of t-tubules was reduced [53]. Of note is the fact that many of these studies [51]–[53] utilized the same RyR gating scheme as this current study, yet they identified various mechanisms for CaT alternans. This demonstrates that the relative importance of the various mechanisms, whether SR load-driven, RyR refractoriness-driven, or otherwise, is highly context-dependent.
Although exploring the issue of SR load vs. RyR refractoriness was beyond the goals of the current study, our results suggest that in human cAF, both SR load alternation and RyR refractoriness are involved in alternans genesis at slower pacing rates. In our cAFalt model, alternation in all SR Ca2+ release variables, including [Ca2+]SR, RyR open probability, and RyR inactivated probability, was necessary for alternans at the onset CL of 400 ms (Fig. 6). In addition, SR uptake flux (Jserca) enhanced alternans when clamped (Fig. 6) and therefore suppressed alternans under normal pacing conditions, suggesting that SR load is indeed an important driver of CaT alternans in cAF and that upregulation of the SERCA pump may be an important therapeutic strategy for diminishing alternans. We also showed that CaT alternans occurred in the cAFalt model at slow pacing rates because decreased RyR inactivation resulted in steepening of the SR release-load relationship. Together, these results indicate that the interplay between SR load and RyR kinetics is responsible for alternans onset in human AF.
The mechanisms for human atrial alternans susceptibility are likely to encompass a range of complex interactions at multiple scales of biology, which extend beyond the cellular-level mechanisms found here. In this study we examined the behavior of an atrial cell with well-developed t-tubules [19]. Research has shown that rat atrial cells have variable levels of t-tubule organization [54]. Such variation, if present in human atrial cells, would result in subcellular Ca2+ gradients which could make cells more susceptible to alternans [17], [55], [56]. Models of atrial myocytes incorporating detailed spatial descriptions [57] and local control of Ca2+ [58] will aid in future investigations of the subcellular mechanisms of cAF-related alternans.
In addition, the complex structure of the atria, including its normal conduction pathways [59] and fibrotic remodeling in AF [60], [61], may promote heterogeneity and discordant alternans, which significantly affect alternans dynamics and reentry initiation [9], [62]. Consideration of these factors in the future will further enrich the mechanistic insight gained from this current study and will advance our understanding of the role that alternans play in AF arrhythmogenesis.
In many cell models, the effective refractory period (ERP) is not consistent with ERP at the tissue level [63]. Electrotonic effects in tissue and the whole heart can shorten or lengthen APD depending on which structures and cell types are coupled to each other. Furthermore, alternans in single cell models may not be predictive of alternans in tissue, where conduction alternans can occur. This was the case for the control atrial tissue model, in which loss of capture occurred at a CL of 260 ms before reaching the very fast pacing rates at which APD alternans were observed in human control patients (CL = 218±30 ms) [8]. However, alternans onset at clinically observed rates occurred in the single-cell control model (200 ms CL, S9 Figure, black curve) and when kiCa was reduced by 5% (230 ms CL, S9 Figure, red curve). This suggests that the ionic model may not be well-constrained for tissue simulations at very fast rates. However, this issue did not affect the study of alternans onset at slower pacing rates, as was observed in AF patients.
Our ionic model variable clamping protocol, which involved separately clamping the even or odd beat waveforms, was used to test for model variables which could robustly suppress alternans when clamped to either of two very different waveforms. An alternative approach would be to clamp model variables to the single unstable, non-alternating waveform obtained using a control algorithm [64]. This approach would allow more precise assessment of fixed point stability, since clamping is done at the point of instability rather than during the bistable (alternans) endpoint. However, for the purposes of quantifying the most important variables influencing instability, the clamping protocol used in this study was sufficient to identify the central role of SR Ca2+ release, which was later confirmed through iterated map analysis.
Recent experimental evidence points towards local SR Ca2+ depletion, rather than Ca2+-dependent RyR inactivation, as the main mechanism of SR release termination [23]–[26]. Although alternans in the cAFalt model relied on Ca2+-dependent RyR inactivation, other termination mechanisms which rely on SR Ca2+ (used in the Sato-Bers RyR model) may have similar effects on SR release slope and alternans susceptibility (Fig. 7, column 2). However, with the Sato-Bers RyR model, alternans and other complex oscillations began at the baseline pacing rate (750 ms CL, S10 Figure) and did not display the same rate dependence observed in patients [8]. In addition, large oscillations in CaT amplitude did not couple as strongly to voltage as with the original RyR, and oscillations were also attenuated in tissue (S10 Figure). Further work is needed to develop atrial cell models which incorporate current mechanistic understanding of SR Ca2+ release and which can also reproduce AF-related alternans rate dependence in tissue.
AF is associated with progressive changes in alternans onset in the human atria, with alternans occurring at slower heart rates as AF severity worsens. We found that the differences in alternans onset between AF and control patients could be accounted for by changes in the inactivation rate of the RyR in a model of human atrial cAF-remodeled tissue. Single-cell simulations revealed that alternans at these slow heart rates were driven by abnormal Ca2+ handling and the development of CaT alternans, and that changes in CaT alternans threshold resulted from steepening of the SR Ca2+ release slope, decreased SR Ca2+ uptake efficiency, and decreased inactivation of the RyR. These findings provide important insight into the mechanisms underlying proarrhythmic APD alternans occurring at slow heart rates in cAF patients. Such insight may aid in the development of targeted therapies and new treatment strategies for AF in the future.
In order to investigate ionic mechanisms in human AF that contribute to the generation of atrial alternans at the tissue level, we created a computer model of human atrial tissue incorporating ionic remodeling associated with cAF. The atrial tissue preparation had dimensions of 0.33×0.33×9.9 mm3 (Fig. 1A), similar to the one used by Krummen et al. [65] Human atrial cell membrane kinetics were represented by a modified version of the Grandi-Pandit-Voigt (GPV) human atrial action potential model [19], which we refer to as the GPVm model. Detailed explanation and justification of the GPVm model modifications are provided in the supplement (S1, S2 Texts). Different types of human atrial tissue were modeled individually as homogenous tissue preparations, with each incorporating ionic changes appropriate for each tissue type. Both control and cAF-remodeled tissue, as well as left and right atrial tissue, were modeled using the parameter changes specified by Grandi et al. [19] (see S1 Text). The isotropic bulk conductivity value for the tissue was tuned to produce a conduction velocity of 0.62 m/s in control tissue [59], [66]. When cAF ionic remodeling was incorporated, the same bulk conductivity value produced a conduction velocity of 0.59 m/s. These values are within the reported ranges for control and AF conduction velocities [67].
We assessed alternans in the human AF tissue model by applying the clinical pacing protocol used by Narayan et al. to induce alternans in AF patients [8]. The tissue model was first initialized at all nodes with steady-state values from a single cell paced at 750-ms CL. The tissue was then paced from the stimulus electrode (Fig. 1A) for 20 beats at 750-ms CL and then for 74 beats at each subsequent CL, starting from 500 ms and shortened in 50-ms steps to 300 ms, and then shortened in 10-ms steps, until loss of capture or conduction block occurred.
Voltage traces from the recording electrode (Fig. 1A) were analyzed for APD alternans. APD was calculated as the time from maximal upstroke velocity to 90% repolarization of Vm from phase II amplitude. Alternans magnitude was quantified as the mean magnitude of change in APD over the last 10 pairs of beats (11 beats total). APD alternans normalized magnitude (ANM), obtained by dividing the alternans magnitude by the mean APD over the last 10 beats, was used to compare alternans between cells of varying APD. Alternans onset CL was defined as the longest CL for which ANM was greater than 5% [8].
To identify cellular changes which could account for the onset of alternans in AF patients at CLs of 300–500 ms [8], we explored how ANM varied in human AF tissue models of both the left and right atrium as a result of changes in ionic model parameters. Of the 20 ionic model parameters tested, 10 were parameters altered in the GPVm model to represent cAF [19]; others were associated with L-type Ca2+ current (ICaL), rapidly activating potassium current (IKr), SR uptake, or SR release (Table 1). We scaled parameter values one at a time to 25–200% of the default left or right atrium values specified by Grandi et al. [19]; for each parameter value within this range, simulations were conducted to determine the presence of alternans (282 simulations total). In AF patients, average alternans onset CL was>300 ms [8], so pacing and alternans analysis was restricted to CLs≥300 ms.
After identifying conditions under which APD alternans magnitude and onset CL matched clinical observations, we utilized two different clamping approaches in order to investigate the key cellular properties that gave rise to these alternans, as described below. Further explanation of the rationale behind these methods can be found in Results.
The monodomain and ionic model equations were solved using the Cardiac Arrhythmia Research Package (CARP; Cardiosolv, LLC) [69]. Details on the numerical techniques used by CARP have been described previously [70], [71]. A time step of 20 µs was used for all simulations.
|
10.1371/journal.pgen.1003207 | Magel2 Is Required for Leptin-Mediated Depolarization of POMC Neurons in the Hypothalamic Arcuate Nucleus in Mice | Prader-Willi Syndrome is the most common syndromic form of human obesity and is caused by the loss of function of several genes, including MAGEL2. Mice lacking Magel2 display increased weight gain with excess adiposity and other defects suggestive of hypothalamic deficiency. We demonstrate Magel2-null mice are insensitive to the anorexic effect of peripherally administered leptin. Although their excessive adiposity and hyperleptinemia likely contribute to this physiological leptin resistance, we hypothesized that Magel2 may also have an essential role in intracellular leptin responses in hypothalamic neurons. We therefore measured neuronal activation by immunohistochemistry on brain sections from leptin-injected mice and found a reduced number of arcuate nucleus neurons activated after leptin injection in the Magel2-null animals, suggesting that most but not all leptin receptor–expressing neurons retain leptin sensitivity despite hyperleptinemia. Electrophysiological measurements of arcuate nucleus neurons expressing the leptin receptor demonstrated that although neurons exhibiting hyperpolarizing responses to leptin are present in normal numbers, there were no neurons exhibiting depolarizing responses to leptin in the mutant mice. Additional studies demonstrate that arcuate nucleus pro-opiomelanocortin (POMC) expressing neurons are unresponsive to leptin. Interestingly, Magel2-null mice are hypersensitive to the anorexigenic effects of the melanocortin receptor agonist MT-II. In Prader-Willi Syndrome, loss of MAGEL2 may likewise abolish leptin responses in POMC hypothalamic neurons. This neural defect, together with increased fat mass, blunted circadian rhythm, and growth hormone response pathway defects that are also linked to loss of MAGEL2, could contribute to the hyperphagia and obesity that are hallmarks of this disorder.
| Prader-Willi Syndrome (PWS) is a genetic condition that causes insatiable appetite and severe obesity in affected children. Several genes are inactivated in children with PWS, but no one knows which gene is important for normal body weight. One of the inactivated genes is called MAGEL2. We previously found that mice missing the equivalent mouse gene, named Magel2, have more fat and are overweight compared to mice with an intact Magel2 gene. In other forms of genetic childhood obesity, there are deficiencies in the way that the brain senses a hormone called leptin, which is made by fat cells. In this study, we show that mice lacking Magel2 are defective in their ability to sense leptin. We identified the specific type of brain cell that should become activated when treated with leptin, but that is not activated in mice lacking Magel2. We then found that we could bypass this leptin insensitivity by administering a drug that compensates for the lack of activity of these neurons. We propose that loss of the MAGEL2 gene in people with Prader-Willi Syndrome may cause deficient leptin sensing, leading to the increased appetite and obesity that are hallmarks of this genetic condition.
| Energy balance is regulated in part by the coordinated action of specialized neurons within the hypothalamus of the brain, which sense circulating signals of energy stores such as the adipocyte derived hormone, leptin [1]. The arcuate nucleus (ARC) is a key hypothalamic region involved in energy balance regulation, and is a major site for leptin action. Two distinct populations of ARC neurons, expressing either Neuropeptide Y (NPY) and Agouti-related peptide (AgRP) or pro-opiomelanocortin (POMC), have opposing effects on energy balance. NPY and AgRP, via different mechanisms, stimulate food intake and reduce energy expenditure, with the overexpression of either leading to obesity [2]–[4]. In contrast, POMC is processed into several shorter peptides including α-MSH, which reduces food intake and stimulates energy expenditure through melanocortin-responsive neurons in the paraventricular nucleus and elsewhere [5]. Mutations that affect processing or lead to loss of expression of the POMC gene also cause obesity in mice and humans [6]–[8].
Impaired hypothalamic regulation of energy balance is found in numerous genetic forms of human obesity, including congenital deficiency of leptin (MIM 164160) [9], leptin receptor mutations (MIM 601007) [10], MC4R melanocortin receptor mutations (MIM 601665) [11], and Bardet-Biedl Syndrome (MIM 209900) [12]. Impaired energy homeostasis may also contribute to the severe hyperphagia and obesity seen in people with Prader-Willi Syndrome (PWS, MIM 176270), the most common genetic form of syndromic obesity in humans [13]. People with PWS typically have a loss of function of several contiguous genes, including MAGEL2, a member of the melanoma antigen (MAGE) family of proteins [14]. MAGE proteins act in intracellular signaling pathways that modulate protein modification, protein degradation, cytoskeletal rearrangement, and transcription [15]. In mice, Magel2 is predominantly expressed in the central nervous system, with highest expression levels in the hypothalamus [16], [17]. We previously showed that gene-targeted mice lacking Magel2 become overweight with increased adiposity as adults [18], and exhibit delayed puberty, irregular estrous cycles, and early onset infertility [19]. As obesity and infertility are common in animal models with impaired leptin responses [20], we hypothesized that Magel2-null mice may also respond abnormally to leptin. We now report that Magel2-null mice display physiological leptin resistance, that leptin resistance precedes the development of increased adiposity, and that leptin-mediated electrophysiological responses in POMC neurons are conspicuously absent in these animals.
Leptin maintains homeostatic control of weight, regulating ingestive behavior and energy expenditure in response to changes in nutritional energy availability. The fall in circulating leptin that occurs with food deprivation normally causes increased feeding when food is reinstated, restoring normal weight and fat mass [1]. However, refeeding-associated weight gain and hyperphagia are dysregulated in mice with diet-induced obesity [21] or mice carrying mutations that selectively ablate POMC neurons [22], [23] or that decrease levels of hypothalamic neuropeptides [24], [25]. To determine if Magel2 is important for compensatory responses after fasting, we subjected mice to a prolonged (48 h) fast. While control mice lost 16% of their body weight after fasting, Magel2-null mice lost significantly less body weight (12% of initial weight), consistent with their previously noted reduced locomotor activity (Figure 1A). We then refed the fasted mice, and measured food intake and body weight over the next 3 days. Body weight returned to baseline within 2 days of refeeding in control mice, but Magel2-null mice remained underweight even after 3 days (Figure 1A). Food intake was similar before fasting (Figure 1B), but Magel2-null mice ate less food during the initial 24 h recovery period (Figure 1C), resulting in a significantly reduced food intake ratio - the ratio of food consumed after fasting to food consumed before fasting - compared to control mice (Figure 1D). These results suggest that the hypothalamic pathways required for compensatory refeeding are defective in Magel2-null mice.
Magel2-null mice have excess adipose tissue, and high levels of circulating leptin suggesting reduced leptin sensitivity [18]. At 20 weeks of age, Magel2-null mice are 14% heavier than control mice (Figure 2A). To examine whether Magel2-null mice are sensitive to exogenous leptin, we measured food intake in singly housed male mice using a crossover study design in which the same animals received either intraperitoneal (ip) leptin (2.5 mg/kg) or phosphate buffered saline (PBS) approximately 1 week apart. In control leptin-treated mice, food intake was reduced by about 30% in the 24 h following leptin injection, as expected. However, leptin-treated Magel2-null mice showed no reduction in food intake following ip leptin (Figure 2B). Decreased sensitivity to peripherally administered leptin can occur in mice with diet-induced obesity that have very high (e.g. ten-fold elevated) leptin levels even in the absence of a genetic mutation [26], [27]. In contrast, Magel2 mice typically have only two-fold elevated leptin even as older adults. Nonetheless, we tested leptin sensitivity in younger (6-week old) mice, where there is no difference in body weight between Magel2-null and control animals (Figure 2C). Leptin treatment in young control mice again caused a reduction of approximately 35% in 24 h food intake compared to PBS treatment. In contrast, there was no reduction in 24 h food intake in leptin-treated young Magel2-null mice (Figure 2D). These results suggest that Magel2-null mice that are similar in weight to controls are nevertheless insensitive to the anorexigenic effect of peripherally administered leptin.
We next examined the activation of the leptin receptor by measuring levels of phosphorylated Signal Transducer and Activator of Transcription 3 (pSTAT3) [28], [29] in the ARC following a single ip leptin (2.5 mg/kg) injection. While few pSTAT3-positive neurons were seen in the ARC following PBS injection in both Magel2-null and control animals (Figure 3A, 3C, 3E), numerous pSTAT3-positive cells were seen in the ARC of both genotypes after leptin injection (Figure 3B, 3D). Nonetheless, detailed cell counts throughout the ARC revealed a 30–35% reduction in pSTAT3-positive cells in leptin-injected Magel2-null mice compared to leptin-injected control (Figure 3E). Next, we measured the induction of c-fos, an immediate early gene marker of neuronal activation that is detected in POMC but not NPY neurons in the ARC after leptin injection [30], [31]. Baseline c-fos immunoreactivity was observed in PBS-injected control animals (Figure 3F, 3H, 3J), and leptin treatment caused a significant increase in c-fos expression in both control and Magel2-null mice (Figure 3G, 3I, 3J), particularly in more posterior regions of the ARC where the majority of leptin-sensitive POMC neurons are located [32]. Interestingly, fewer c-fos positive cells were observed in Magel2-null mice after either PBS or leptin injection compared to similarly treated control mice (Figure 3J).
POMC neurons form an important part of the hypothalamic energy balance neural circuitry, and are activated in response to leptin [33]. Fewer leptin-induced pSTAT3 and c-fos immunoreactive cells were observed in the ARC of Magel2-null mice, particularly in areas previously shown to contain higher levels of leptin-sensitive POMC neurons. We therefore counted POMC/EGFP-positive neurons in the ARC of Magel2×POMCEGFP and control mice, and found on average 39% fewer POMC+ neurons in the Magel2-null mice than in controls (Figure 4). This reduction was most evident in the more posterior region of the ARC, where 52% fewer POMC+ cells were found (Figure 4C). The number of LepRb positive neurons (measured as EGFP positive cells in the ARC of offspring from a Magel2×LepRbEGFP cross) did not differ between mutants and controls. Thus, loss of POMC neurons can partially explain the reduction seen in leptin-induced pSTAT3 and c-fos expression in the ARC of Magel2-null mice. Alternatively, it is possible that these neurons are still present, but that the expression of POMC/EGFP has fallen below the detection limit of this experiment.
To directly examine leptin responses in ARC neurons, we performed whole-cell visualized-patch recordings of fluorescent neurons in mice expressing enhanced GFP in leptin receptor-positive (LepRb+) neurons (Figure 5A, 5B). First, the resting membrane potential (RMP) of LepRb+ neurons (Figure 5C) and the input resistance (data not shown) were comparable between Magel2×LepRbEGFP and control mice. NPY hyperpolarizes the majority of leptin-responsive ARC cells (Figure 5D) [34]. Application of either 100 nM (data not shown) or 300 nM NPY produced a robust hyperpolarization of virtually all ARC LepRb+ neurons tested in both Magel2-null and control slices, indicating that Magel2 is not required for normal NPY signaling (Figure 5E). We then examined the leptin (100 nM) responses in LepRb+ neurons in the ARC. Leptin normally activates (depolarizes) POMC neurons, and inhibits (hyperpolarizes) NPY neurons [33], [35], so we expected to observe both responses in the mixed neuronal populations represented by LepRb+ cells in the ARC. Leptin induced both hyperpolarizing and depolarizing responses in LepRb+ cells in slices from control mice, with a few unresponsive cells (Figure 5F–5H). All cells tested, including leptin-unresponsive cells, exhibited a normal electrophysiological response to 300 nM NPY. In striking contrast, LepRb+ cells in slices from Magel2-null mice never exhibited depolarizing responses to leptin. In these slices, leptin-mediated hyperpolarizing responses were seen at a frequency comparable to controls, while more leptin-unresponsive cells (which nevertheless showed normal NPY responses) were found (Figure 5H). These results suggest that the inhibitory action of leptin is unaffected at ARC LepRb+ neurons of Magel2-null mice, but that the excitatory effect of leptin, typically observed at POMC neurons, is selectively absent.
To more directly examine the population of neurons specifically activated by leptin in the ARC, we identified POMC neurons using crosses with mice expressing GFP in POMC cells (Magel2×POMCEGFP and littermate controls). As with LepRb+ neurons in the ARC, POMC+ neurons from control and Magel2-null animals did not differ in their RMP (Figure 5I). We then tested leptin (100 nM) responses in POMC+ cells located in the posterior and medial regions of the ARC, where a large number of POMC neurons are leptin-sensitive [32]. Leptin induced a depolarization in the majority of POMC neurons from control mice, but no depolarizing effects were seen in response to leptin in POMC neurons of Magel2-null mice. This confirms that POMC neurons in these animals are insensitive to the acute administration of leptin (Figure 5J).
In addition to the ARC POMC neurons, many other neurons in the hypothalamus are depolarized by leptin [36], [37]. To determine the specificity of the effect of Magel2 loss on depolarizing actions mediated by the leptin receptor, we studied leptin responses in the ventromedial hypothalamic nucleus, which comprise both depolarizing and hyperpolarizing responses [38]. The serial microscope sections stained for pSTAT3 used in the experiments on ARC above were re-imaged for the VMN using confocal microscopy, and pSTAT3-positive neurons were counted. Leptin treatment caused an increase in numbers of neurons immunopositive for pSTAT3, but in contrast to results for ARC, no significant differences in numbers of pSTAT3 neurons were seen in the Magel2-null animals compared to controls (Figure 6A). Electrophysiological recordings were made from neurons in the dorsomedial and central VMN as described [39] in slices prepared from Magel2-null×LepRbEGFP and LepRbEGFP control animals. In VMN from control LepRbEGFP mice, we observed a mixture of depolarizing and hyperpolarizing responses to 100 nM leptin, along with unresponsive neurons. In contrast to the results in ARC, there were no significant differences in the numbers of neurons depolarized or hyperpolarized by leptin (Figure 6B). Thus, leptin-mediated depolarization of VMN neurons is unaffected by the loss of Magel2.
A failure of POMC neurons to depolarize in response to leptin application is predicted to cause loss of α-MSH release. In other animal models of leptin insensitivity, an enhanced response to the direct application of either α-MSH or the synthetic melanocortin agonist MT-II is observed [40]–[42]. We therefore examined the effect of MT-II on food intake in Magel2-null mice. Mice were fasted for 24 h, and then injected with MT-II (2.5 mg/kg ip). Compared to PBS-injected control fasted mice, MT-II-injected control fasted mice consumed 50% less food over the first 2 h of refeeding. After this time, there was no significant difference in food intake between control mice injected with MT-II or PBS. In contrast, Magel2-null mice injected with MT-II had a greater reduction in food intake compared to PBS injection, and this decrease was still evident after 24 h (Figure 7). This result suggests that the melanocortin system is chronically upregulated in Magel2-null mice, likely as a result of the loss of melanocortinergic tone from ARC POMC neurons.
Mice lacking Magel2 have increased adiposity with proportionately increased leptin, suggesting leptin insensitivity [18], [43]. Here, we show that Magel2-null mice are physiologically resistant to the effects of exogenously applied leptin, both before and after the onset of increased adiposity. Further, this leptin resistance is accompanied by a 39% reduction in the number of POMC neurons in the ARC, and by a complete absence of leptin-induced depolarization responses in the remaining POMC neurons. Magel2 is therefore essential for normal leptin signaling in POMC neurons, and for the differentiation, proliferation, or survival of this population of neurons. Interestingly, the effect of Magel2 loss on leptin-mediated depolarization is not universal, even within the hypothalamus, as equivalent numbers of energy balanced-related VMN neurons not only exhibit pSTAT3 immunoreactivity, but also equal numbers are depolarized in the Magel2-null animals. Loss of POMC neuronal activation is accompanied by an exaggerated anorexigenic response to exogenous melanocortins, suggesting a compensatory upregulation of downstream melanocortin response pathways in Magel2-null mice. The role of MAGEL2 in melanocortin-associated neuronal pathways may provide important insights into dysfunctional ingestive behavior and obesity in Prader-Willi syndrome.
Insensitivity to peripheral leptin has been demonstrated in diet-induced and genetic models of obesity [44]–[47]. In principle, a failure to respond to acutely or chronically elevated leptin could be caused by reduced transport across the blood brain barrier, or by an intrinsic defect in leptin-responsive neurons. In the latter case, leptin insensitivity could be caused by failure of leptin either to inhibit the orexigenic drive (at NPY neurons), or to activate the anorexigenic drive (through POMC neurons), or both mechanisms, as in congenital leptin insensitivity in mice carrying an inactive form of the leptin receptor (LepRdb mice). Although the anorexic response to peripherally administered leptin is absent in the Magel2-null mice, the electrophysiology results demonstrate that many arcuate hypothalamic neurons that express the leptin receptor remain leptin-sensitive. Specifically, Magel2-null ARC slices have a similar proportion of neurons displaying inhibitory responses to leptin as do slices from control animals, and these responses are of similar amplitude. Moreover, the remaining POMC neurons retain sensitivity to NPY, so the loss of the leptin-mediated excitatory response is not indicative of a global cellular defect within the ARC. This retention of leptin-mediated inhibitory responses is consistent with the modest level of obesity in Magel2-null mice compared with leptin-deficient Lepob or leptin receptor null Leprdb mice. We did not test the response of VMN neurons to NPY in Magel2-null animals here.
Several mouse strains have been constructed in which leptin signaling pathways are selectively impaired in POMC neurons. Mice engineered without leptin receptor expression only in POMC neurons are mildly obese, with a significant increase in fat mass [48], [49], similar in magnitude to that previously reported in Magel2-null mice [18]. A similar degree of obesity and adiposity is seen in mice with inactivation of STAT3 in POMC neurons [23]. Unlike the Magel2-null mice, the POMC-STAT3 mutants remain sensitive to peripheral leptin, but they display defects in compensatory refeeding following food deprivation leading to reduced weight regain, similar to what we have observed in Magel2-null mice. Though the largely glutamatergic neurons of the VMN [50] remain leptin-responsive in the mutants, the loss of leptin signaling in other hypothalamic neurons in Magel2-null mice could underlie their more severe insensitivity to peripherally administered leptin. Rapid effects of leptin action on ARC leptin receptors have been linked to increased phosphatidyl inositol-3-kinase (PI3K) signaling [51], [52]. Accordingly, pharmacological blockade of PI3K signaling inhibits leptin-induced activation of POMC neurons [53]. Targeted deletion of PI3K signaling in POMC neurons also eliminates leptin-induced activation of POMC neurons, and significantly blunts the reduction in food intake provoked by intracerebroventricular leptin administration [53]. Interestingly, these mice do not appear to have any defects in weight gain or body composition, though a different strategy aimed at the downregulation of PI3K in POMC neurons does lead to a modest obesity phenotype and increased sensitivity to diet-induced obesity [54]. Investigations of a possible role of Magel2 in PI3K signaling are thus warranted.
The complete absence of a physiological response to leptin in Magel2-null mice could have several causes. First, the Magel2-null mice catch up in weight compared to control and start accumulating excessive fat mass after weaning onto a standard chow diet, albeit at a modest rate. The resulting hyperleptinemia could contribute to systemic leptin resistance through a mechanism unrelated to or secondary to defective POMC neuron activation, but in any event caused ultimately by loss of Magel2 function. Secondly, only half the normal number of ARC neurons expressed pSTAT3 in the ARC of Magel2-null mice after peripheral leptin treatment, and fewer Magel2-null neurons were activated by leptin as measured by c-fos expression. Third, Magel2-null mice had fewer POMC ARC neurons, and the remaining POMC neurons were not activated by leptin. The loss of excitatory leptin signaling at POMC neurons and their increased adiposity are consistent with a loss of key actions of leptin at ARC POMC and potentially other neurons in the Magel2-null mice [48], [49]. While our findings demonstrate a crucial role for POMC in the Magel2-null phenotype, the ARC contains a heterogeneous population of leptin-activated neurons it remains possible that the leptin-mediated activation of these neurons is also affected by loss of Magel2 [20].
Intracellular responses to leptin receptor activation are mediated by a complex signaling cascade in POMC neurons [55], and this process is similar but not identical in other leptin-responsive neurons. For example, in leptin-activated neurons in the ventromedial nucleus (VMN), some neurons depolarize in response to leptin, some cells hyperpolarize, and the majority of cells do not respond to leptin administration [39], [56], [57]. The identical rates of leptin responsiveness in VMN of Magel2-null and control mice indicates that Magel2 is required for depolarizing responses in some neuronal subtypes but not in others. Likewise, the relative increase in the number of neurons expressing pSTAT3 in the VMN of leptin-injected compared to saline injected mice did not differ between genotypes.
Fasting in rodents induces a state of negative energy balance that is reflected by dramatic decreases in circulating leptin levels [1], [58], [59] and compensatory hyperphagia on re-feeding. Deficiencies in fasting-induced hyperphagia and compensatory weight gain are found in models of POMC neuron degeneration or in POMC-specific STAT3 mutant mice [22], [23]. Thus, appropriate regulation of POMC neurons in the ARC is critical to normal responses to food deprivation, which are clearly impaired in Magel2-null mice. Other hypothalamic pathways could also contribute to dysfunctional feeding behavior in Magel2-null mice. For example, orexin neurons normally activate NPY and inhibit POMC neurons to stimulate increases in food intake [60], and ablation of orexin neurons in the lateral hypothalamus causes a loss of fasting-induced arousal and defense of body weight during fasting [61]. In fact, Magel2-null mice have fewer orexin neurons and a significant reduction in hypothalamic levels of orexin-A [43], [62], which could contribute to the impaired compensatory hyperphagic responses in the Magel2 null mice. In addition, there may be developmental defects in axonal outgrowth and synaptic contacts with other neurons in the remaining POMC, orexin, and other neuronal subtypes that require Magel2 developmentally, further impairing their leptin-mediated excitability.
Notably, the anorexic response to melanocortins is intact and hyperactivated in Magel2-null mice, suggesting that melanocortin receptors in the paraventricular nucleus and elsewhere in the central nervous system are not impaired by loss of Magel2. Further examination of melanocortin responsiveness in Magel2-null mice could provide compelling evidence for potential therapeutic intervention in PWS. The exact biochemical roles of Magel2 and how it participates in neuronal differentiation and/or survival as well as cellular activation in response to leptin remain to be determined. In summary, our results demonstrate that Magel2 is critical for leptin responses in POMC neurons in the ARC and for energy homeostasis in mice. Further experiments are required to determine whether this defect is degenerative in nature or whether mice lacking Magel2 are congenitally leptin insensitive. It will also be important to address whether loss of MAGEL2 in people with PWS likewise contributes to disrupted ingestive behavior and energy homeostasis in this disorder.
All animal procedures were approved by the University of Alberta Animal Care and Use Committee in accordance with the guidelines of the Canadian Council on Animal Care. Mice were weaned between 3–4 weeks of age and then group housed 3–5 per cage with food (PicoLab Rodent Diet 5001) and water ad lib., and housed under a 12∶12 light∶dark cycle.
Magel2-null mice were generated [43] and housed [19] as described, and are available from the Jackson Laboratories (C57BL/6-Magel2tm1Stw/J, stock 009062). To identify specific neuronal populations, Magel2−m/+p carrier males were crossed with homozygous LepRbEGFP reporter mice, which express enhanced green fluorescent protein (EGFP) in LepRb+ cells [63], or homozygous POMCEGFP reporter mice, expressing EGFP in POMC+ cells (The Jackson Laboratory stock #009593, Bar Harbor, Maine) [33]. This cross produces Magel2×LepRbEGFP or Magel2×POMCEGFP mice, lacking Magel2 but expressing LepRbEGFP or POMCEGFP, and control littermates expressing wildtype Magel2 and the reporter gene.
Male (12–16 weeks) mice were singly housed for at least one week, then weighed and fasted for 48 h beginning at 1600 h. Body weight was recorded 24 h and 48 h later, and food intake and body weight change were measured during 3 days of refeeding.
Mice were singly housed for at least one week before the start of the experiment. One week before drug injections, mice were injected daily for 3 days with phosphate buffered saline, pH 7.3 (PBS). Body weight and food intake were measured during this time. On experimental day 1, food was removed at 1500 h and mice injected intraperitoneally (ip) with either 2.5 mg/kg mouse recombinant leptin (Dr. A.F. Parlow, National Hormone and Peptide Program, NHPP-NIDDK, Torrance, California), 2.5 mg/kg synthetic melanocortin agonist melanotan-II (MT-II Phoenix Pharmaceuticals, Burlingame, California) [64], or PBS at 1600 h. Food intake was measured 2–24 h later. After a 3 day recovery, the experiment was repeated using a cross-over design.
Adult (12–16 weeks) mice were handled daily for 2 weeks, including one week of PBS injections, to minimize injection-related c-fos responses in the brain. Mice were then injected ip with either 2.5 mg/kg leptin or PBS 45 min before terminal pentobarbital anesthesia, paraformaldehyde perfusion, and preparation of coronal 30 µm hypothalamic sections for immunohistochemistry (IHC). For pSTAT3 IHC, sections were pretreated with 1% NaOH and 1% H202 in H20 for 20 min, 0.3% glycine in PBS for 10 min, and 0.03% sodium dodecyl sulfate in PBS for 10 min, blocked for 1 h with 3% normal goat serum in PBS/0.3% Triton X-100, then incubated overnight with anti-pSTAT3 antibody (1∶1000, 9131, Cell Signaling, Danvers, Massachusetts). For c-fos and POMC/EGFP IHC, sections were washed in PBS, blocked for 30 min in 3% normal serum in PBS/0.3% Triton X-100, then incubated for 48 h with primary antibody (c-fos (Ab-5), 1∶2000, PC-38, Millipore, Billerica, Massachusetts; POMC, anti-GFP, 1∶4000, ab13970, AbCam, Cambridge, Massachusetts). After primary incubation, sections were washed with PBS/0.3% Triton X-100 and incubated for 2 h with goat anti-rabbit secondary antibody (Alexa Fluor 594) or goat anti-chicken secondary antibodies (Alexa Fluor 488) (1∶500, Invitrogen, Carlsbad, California), slide-mounted then imaged using a Zeiss LSM510 confocal microscope. For cell counting in the arcuate nucleus, sections were organized in a rostral to caudal manner through the hypothalamus according to the mouse brain atlas (www.mbl.org). Cells were counted using MetaMorph Imaging Suite (Molecular Devices, Sunnyvale, California) for pSTAT3, and Image J (National Institutes of Health, Bethesda, Maryland) for c-fos and GFP.
Brains from 6–12 week old male and non-estrous female reporter mice were prepared for patch clamp electrophysiology [39]. Slices were incubated for at least 1 h at room temperature in carbogenated artificial cerebrospinal fluid (aCSF) containing (in mM): 124 NaCl, 3 KCl, 1.3 MgSO4, 1.4 NaH2PO4, 2.5 glucose, 7.5 sucrose, 26 NaHCO3 and 2.5 CaCl2 (300–305 mOsm/L). For electrophysiology, slices were continuously perfused (2–4 ml/min) with warm (32–34°C), carbogenated aCSF. Cells expressing GFP were identified by epifluorescence illumination, then the light source was switched to infrared-differential interference contrast imaging to obtain whole-cell recordings. Visualized-patch whole-cell recordings were obtained using thin-walled glass patch pipettes with resistances of 5–7 MΩ when backfilled with an internal solution containing (in mM): 126 K-gluconate, 4 KCl, 10 HEPES, 5 MgATP, 0.3 NaGTP, 1 EGTA, 0.3 CaCl2 and 0.02% neurobiotin (pH adjusted to 7.25 with KOH, 280 mOsm/L). Stock solutions were prepared in PBS, pH 7.8 (leptin) or HPLC grade water (human NPY, Peptidec Technologies Ltd., Pierrefonds, Quebec, Canada), then diluted into aCSF immediately before use, and gravity-perfused into the recording chamber for at least 3 min. Slices were washed with aCSF for at least 10 min between drugs. A stable and reversible change in membrane potential of at least 2 mV from baseline appearing within minutes after drug application was considered a valid pharmacological response.
Statistical analyses were performed using a Student's unpaired t-test or a Fisher's Exact Test (GraphPad, La Jolla, California), with differences with P<0.05 after correction for multiple t-testing considered significant.
|
10.1371/journal.pcbi.1007327 | Mechanical properties of tubulin intra- and inter-dimer interfaces and their implications for microtubule dynamic instability | Thirteen tubulin protofilaments, made of αβ-tubulin heterodimers, interact laterally to produce cytoskeletal microtubules. Microtubules exhibit the striking property of dynamic instability, manifested in their intermittent growth and shrinkage at both ends. This behavior is key to many cellular processes, such as cell division, migration, maintenance of cell shape, etc. Although assembly and disassembly of microtubules is known to be linked to hydrolysis of a guanosine triphosphate molecule in the pocket of β-tubulin, detailed mechanistic understanding of corresponding conformational changes is still lacking. Here we take advantage of the recent generation of in-microtubule structures of tubulin to examine the properties of protofilaments, which serve as important microtubule assembly and disassembly intermediates. We find that initially straight tubulin protofilaments, relax to similar non-radially curved and slightly twisted conformations. Our analysis further suggests that guanosine triphosphate hydrolysis primarily affects the flexibility and conformation of the inter-dimer interface, without a strong impact on the shape or flexibility of αβ-heterodimer. Inter-dimer interfaces are significantly more flexible compared to intra-dimer interfaces. We argue that such a difference in flexibility could be key for distinct stability of the plus and minus microtubule ends. The higher flexibility of the inter-dimer interface may have implications for development of pulling force by curving tubulin protofilaments during microtubule disassembly, a process of major importance for chromosome motions in mitosis.
| The ability to self-assemble from tubulin dimers in presence of guanosine triphosphate (GTP) and spontaneously disassemble, when GTP molecules in tubulin pockets are hydrolyzed, is a dramatic and essential feature of microtubules. This behavior has many important roles, including chromosome segregation in mitosis, rapid remodeling of the microtubule networks, establishing cell polarity and other. Nevertheless, the mechanism, linking the associated nucleotide with the conformational changes in tubulins, remains elusive. Most studies suggested that the nucleotide should affect either the equilibrium shape of tubulin dimers or the strengths of the lateral bonds between them. But existing experimental methods have lacked spatio-temporal resolution to test that. Theoretical studies, until recently, have suffered from the absence of high-resolution microtubule structures with different nucleotides to build on and the lack of computational efficiency to examine large tubulin assemblies. Here we use recent cryo electron microscopy structures of GDP and GTP-like microtubules, and employ all-atom molecular dynamics simulations to examine tubulin protofilaments. We find that the nucleotide primarily affects the interface between two tubulin dimers, making it more flexible in the GTP state. This makes the GTP-bound tubulin protofilament easier to incorporate into microtubule lattice, providing a simple mechanism for microtubule dynamic instability.
| αβ-tubulin heterodimers polymerize into microtubules, hollow cylindrical structures, usually composed of 13 laterally attached protofilaments [1]. Microtubules are about 25 nm wide and range in lengths from tens up to millions of nanometers. They form cilia and flagella and serve as tracks for long-distance transport of intracellular cargos, such as vesicles and organelles. In contrast to other polymers, microtubules are highly non-equilibrium systems [2], which can remain in growth and shrinkage phases with relatively rare spontaneous transitions between them [3]. Because of this behavior, known as dynamic instability, individual microtubules display significant length changes, even at steady state. They elongate at their tips by addition of guanosine triphosphate (GTP)-bound tubulins. Soon after incorporation into microtubule lattice, GTP molecules are hydrolyzed to guanosine diphosphate (GDP). This leads to a conformational change in tubulins, so the lattice made of GDP-tubulins becomes less stable and more prone to depolymerization. However, because of a lag between the association of GTP-tubulins with microtubules and GTP hydrolysis, there is a certain number of GTP-tubulins at the growing microtubule tip, known as GTP cap, which prevents disassembly until the stabilizing cap is lost [4]. Both ends of microtubules are dynamically unstable. The end, exposing β-tubulin subunits, is called the plus-end. It grows faster than the other end, known as the minus-end. The origin of the difference of behavior between the plus and minus-ends of microtubules is currently poorly understood. In cells, the minus-ends are usually capped, so they remain stable. The plus ends are usually dynamic and serve multiple roles. During mitosis they generate forces responsible for chromosome motions, leading to segregation of duplicated DNA between daughter cells [5,6]. This fact has been extensively exploited for therapeutics, as the inhibition of microtubules dynamics by small molecule drugs leads to arrest of cell division followed by apoptosis, leading to a powerful method to fight proliferation of actively dividing tumor cells [7].
Despite extensive studies of dynamic instability for over three decades, the molecular features of GTP- and GDP-tubulins, determining their distinct propensity to polymerize, remain unclear. Early cryo electron microscopy (EM) studies reported very distinct shapes at the ends of growing and shortening microtubules [8]. That observation informed a so called ‘allosteric’ model of the GTP cap, postulating that GTP hydrolysis induced an allosteric conformational change in straight GTP-tubulin dimers, so GDP-tubulin became curved. Further cryo EM work modified the allosteric model, proposing that GTP-tubulin was also slightly curved, but still straighter than GDP-tubulin. The latter modification was based on observations of gently curving extensions on growing microtubule tips [8,9], and the shapes of tubulin structures formed in presence of slowly hydrolysable GTP analogue, GMPCPP [10,11].
Subsequent studies have accumulated substantial evidence indirectly supporting an alternative, ‘lattice’ model of the GTP-cap, postulating that the phosphorylation state of tubulin-bound nucleotide affects the strengths of inter-tubulin bonds, while the shapes of free GTP- and GDP-bound tubulins remain similar. The following lines of evidence against a significant difference in curvature between GTP- and GDP tubulins have been reported: (1) free tubulin dimers and tetramers are similarly curved in all available crystal structures of tubulin (reviewed in [12]); (2) there is no significant difference in the shape of GTP- and GDP-tubulins according to small-angle X-ray scattering measurements [13]; (3) affinity to allocolchicine, which is thought to be able to bind only a curved intra-dimer interface, is the same for GTP- and GDP-tubulins [13]; (4) recent cryo electron tomography reveals essentially no difference in the curvatures of protofilaments at the tips of growing and shortening microtubules [14]. Finally, computational studies so far have consistently found that the relaxed tubulin conformation is curved, irrespective of the nucleotide bound, while the strengths of the inter-tubulin bonds are likely to be nucleotide-dependent [15–21]. We note, however, that at the time when many of those important pioneering simulations were carried out, no high resolution structures of both GTP- and GDP-tubulins were yet available. The first relatively high resolution cryo-EM-based structures of tubulins in microtubule walls in presence of GDP or GTP analog, GMPCPP, were presented in the seminal paper of Alushin et al., 2014 [22]. But even those structural data and their subsequent improvements have not yet produced a fully consistent picture that would clearly support only the allosteric or the lattice model. For example, reports from the Nogales group did not detect any considerable changes at the lateral tubulin-tubulin interfaces and emphasized the “compaction” and skew of GDP- tubulins, in contrast to “extended” GTP-tubulin state in the lattice [22–24]. The authors proposed that compaction could induce internal mechanical strain in GDP-tubulin. A study from Moores’ group, however, suggested that lateral bonds were not unchanged, but weakened after GTP hydrolysis [25].
These controversies, together with the inability of modern structural methods to directly visualize conformational changes of tubulins following GTP hydrolysis and breakage of lateral bonds have encouraged us to undertake a new computational study, in which we re-investigate the effects of nucleotides on the shape and mechanics of tubulins, taking advantage of the wealth of newly available structural data and new computational resources. A recent cryo electron tomography work indicated importance of tubulin protofilaments, rather than just free dimers, as structural intermediates of microtubule assembly process [14]. Therefore, both intra-dimer and inter-dimer interfaces could have an important role in dynamic instability. To take this into account, we carried out molecular dynamics simulations of tubulin protofilaments, which contained both types of inter-tubulin interfaces. The tubulin protofilaments were extracted from the microtubule wall in compacted, GDP-bound, or extended, GTP-bound, states. In our simulations, the tubulin protofilaments in each of these nucleotide states relaxed to similar non-radially curved and twisted conformations, in contrast to the expectations of the allosteric model of microtubule instability. Our further analysis suggested that GTP hydrolysis primarily affected the flexibility and conformation of the inter-dimer interface, without a strong impact on the shape or flexibility of αβ-tubulin heterodimer. The inter-dimer interfaces of GTP-tubulins were significantly more flexible than those of intra-dimer interfaces. We argue that such a difference in flexibility could be key for distinct dynamic behavior of plus and minus microtubule ends.
To characterize the nucleotide dependence of tubulin’s shape in relaxed tubulin dimers and short tubulin protofilaments, we prepared all-atom molecular dynamics models of tubulins, extracted from cryo-EM-based structures of GDP- and GTP-like (GMPCPP) microtubule lattice [22]. Flexible tubulin tails were included in the simulation to make sure that their potential effects on tubulin conformations would be taken into account [26]. For each nucleotide, we carried out two one-microsecond-long simulations of tubulin dimers, and three one-microsecond-long simulations of short protofilaments, representing two longitudinally bonded dimers (S1 and S2 Movies). In order to put the resulting conformational changes of tubulins into the context of a microtubule, we aligned the α-tubulin subunits of each simulated structure with a microtubule wall fragment, so all types of rotations could be assessed relative to the microtubule-bound coordinate system xyz (Fig 1).
Consistent with previous reports [15,21,27], in our simulations both GDP- and GTP-dimers relaxed over time to similar bent shapes (Fig 2A). Although the bending occurred predominantly in the ‘outward’ direction, it did not happen in the plane that contained the microtubule axis, as is clearly seen in Fig 2B, which shows projections of the unit orientation vector of the β-tubulin subunit relative to α-tubulin subunit. Additional molecular dynamics simulations, based on a GDP-tubulin dimer structure extracted from zinc-induced sheets (PDB code: 1JFF), indicated that such non-radial bending and a slight twist were common conformational changes shared by other structures of tubulin (S1A and S1B Fig).
To independently validate the tendency of dimers to curve in non-radial plane, we predicted major low-frequency modes of motion for three tubulin dimer structures, using normal modes analysis (NMA). The first NM in all types of tubulin dimer structures examined corresponded to predominantly twisting motions of β-tubulin relative to the α-tubulin subunit, while the second and the third NMs represented predominantly non-radial bending, similar to a previous report [28].
Major low-frequency modes of GTP- and GDP-tubulin dimers were very similar. Principal component analysis (PCA) of molecular dynamics simulation trajectories identified significant correlation of the major NMs with principal components (PCs) of motions in molecular dynamics. The first three PCs explained, on average, 65% of observed tubulin dimer motions (S1 Table). Overall, the major bending motions were represented by the second NM (Fig 2C). They occurred in a direction very similar to that of the major conformational changes observed in our molecular dynamics simulations (Fig 2B).
Conformational analysis of tubulin tetramers revealed that additional inter-dimer interface did not qualitatively change the overall fashion of bending of the whole tetramer, compared to that of a tubulin dimer. Specifically, both GDP- and GTP-tubulin tetramers also assumed outwardly curved shapes in the end of one-microsecond-long simulations (Fig 2D and Fig 2E). The overall bending was also non-radial. Likewise, the dominant NM of the whole tetramer corresponded to bending in a similar non-radial direction (Fig 2F, S3 Movie) and it overlapped significantly with PCs of motions in molecular dynamics (S2 Table). The first three PCs explained, on average, 79% of the total variance present in the molecular dynamics trajectories of tetramers (S2 Fig).
To gain more detailed insight into the conformational changes at intra- and inter-dimer tubulin interfaces during their relaxation from straight to curved shapes and their dependence on the associated nucleotide, we described the relative motions of adjacent tubulin monomers at each interface with three rotation angles, using metrics similar to those introduced previously [16]. Specifically, bending at the tubulin-tubulin interface was described with two angles: θ-angle characterized the magnitude of the conformational change, while auxiliary angle, φ, showed the direction in which the bending occurred; δ-angle characterized twist of one tubulin monomer relative to the other (see Methods for more details). To determine the direction of rotations relative to the microtubule structure, we aligned the minus-end-proximal tubulin monomer onto a corresponding subunit in a straight microtubule fragment, which was oriented relative the coordinate system as depicted in Fig 1. In this arrangement, outward strictly radial bending is described by positive θ-angles and φ = 0 degrees. Despite the advantage of being physically clear and easy to relate with microtubule geometry and with the degrees of freedom, which are usually present in higher-scale models of microtubule dynamics, the rotation angles may not optimally represent the multi-dimensional molecular dynamics data. For this reason, we also carried out an additional analysis, projecting molecular dynamics trajectories on two main PCA modes and comparing the movements at tubulin interfaces, expressed in those observables (S3 Fig, S4 Fig, S4 Movie). This yielded essentially similar conclusions about the relative properties of the inter- and intra-dimer interfaces, so we decided to stick to rotation angles throughout this report for the sake of intuitiveness and physical clarity of description.
First, we calculated rotation angles in tubulin tetramers at intra-dimer tubulin interfaces. After 500 ns of simulation, intra-dimer interfaces of both GDP- and GTP-tubulin tetramers relaxed to a conformation, in which their β-subunits were tilted relative to α-tubulins with almost identical magnitudes of intra-dimer bending, 9.4 ± 0.9 and 8.2 ± 0.7 degrees, respectively (Fig 3, Table 1). Simultaneously, β-tubulins twisted relative to α-tubulins by about 5.2 ± 0.9 and 7.0 ± 1.5 degrees, in GDP- and GTP- states. Conformational changes of free dimers at the intra-dimer interface were quantitatively similar (Table 1, S5 Fig).
In contrast to intra-dimer interfaces, the inter-dimer interfaces were less reproducible in their bending directions from run to run, suggesting the presence of multiple local minima in the energy landscape (Table 1, Fig 4). Within each run, the bending angle had a satisfactory convergence to a stable mean value, as assessed by splitting the last 500 ns of the simulations in four 125-ns-long segments and analyzing them separately (S7 Fig).
We compared equilibrium conformational angles of tubulins at the end of individual simulation runs with the respective intra- and inter-dimer angles calculated for published crystal structures of tubulins bound to microtubule associated proteins (MAPs): stathmin, darpin, TOG-domain or MCAK proteins [29–34]. Intra-dimer curvature, its direction and magnitude of twist in simulated structures were similar to characteristics of the experimental structures, confirming the ability of molecular dynamics simulations to predict equilibrium shapes of tubulin dimers (Table 2, S1C Fig). Inter-dimer angles, though, were much more variable in simulations, without clear correlation with the corresponding angles in crystallized tubulin-MAP complexes. We speculate that in complex with a MAP, the tubulin tetramer is likely to be fixed by its interaction partner, resulting in a different direction of inter-dimer bending and in lower flexibility of the inter-dimer interface. Hence, the conformational variability is reduced.
Despite the fact that in our simulations and in experimental structural data tubulin oligomers both display non-radially curved and twisted shapes, recent cryo electron tomography studies reported nearly planar protofilaments at the tips of growing and shortening microtubules [14,35]. Puzzled by this discrepancy, we performed additional molecular dynamics simulations of GTP- and GDP-tubulin hexamers, applying position restrain on Cα atoms of the minus-end proximal tubulin subunit (Fig 5A, S4 Movie). Fixation of the terminal α-tubulin was mimicking oligomer attachment to the microtubule end. Analysis of the relaxed shapes of GTP- and GDP-tubulins after 500 ns simulation revealed that the interfaces proximal to the fixed subunit (#1–3) tended to be overall straighter, less flexible, and bent in a more radial direction (Fig 5B). On the other hand, distal interfaces (#4 and #5) behaved essentially like those of free tubulin oligomers (compare with Fig 4C, Fig 4D, Fig 3C and Fig 3D). Given this ‘straightening’ effect due to the longitudinal attachment to the plus tip, overall bending direction of whole hexamers, characterized by projections of the center of mass of β-tubulin subunit onto XY-plane, was somewhat more radial on the scale of hexamers, compared with free tetramers (Fig 5A, top view). Although, we note that at the scale of longer protofilaments, the tangential component may still be significant.
We also hypothesized that the presence of adjacent protofilament neighbors could affect the bending direction of a given protofilament, attached to the microtubule tip. To test that, we constructed a molecular model of three laterally bound GDP-tubulin hexamers, whose minus-end proximal α-tubulins were fixed (Fig 6A, S5 Movie). In two independent simulations of this system, the whole assembly of three protofilaments consistently bent asymmetrically: the splaying amplitude of the right protofilament was the highest, while the left protofilament remained almost straight (‘left’ and ‘right’ are defined as in Fig 6A and Fig 6B). This result can be explained by the tendency of individual protofilaments to bend and twist in the directions depicted in Fig 6B, Fig 2D and Fig 2E. Such motions tend to stretch the left lateral bond significantly less, compared to the right bond. Therefore, it is the right lateral bond, which restricts the motion more significantly. Hence, in the absence of the right lateral bond the right protofilament undergoes relaxation to the bent and twisted state relatively easily, while the left protofilament is almost fully restricted by its right lateral bond (Fig 6B and Fig 6C). As a result, the splaying amplitude of the right protofilament is a lot more dramatic. This splaying leads to breakage of the lateral bond between the terminal β-tubulin subunits of the right and the middle protofilaments during the simulation, while the bond between the terminal β-tubulin subunits of left and the middle protofilaments remained intact, as we verified by counting the number of contacts between amino acids of those subunits (Fig 6D). Comparison of protofilament bending in the simulations in the presence or absence of adjacent protofilaments did not reveal a marked effect of lateral neighbors on the bending direction of the middle protofilament (S8 Fig).
We noticed that the variance of rotation angles at the inter-dimer GTP-interface was higher than at all other types of interfaces, which can be appreciated by the highest scatter of projections of the unit OZ-vector, characterizing direction and amplitude of tubulin bending (Fig 4D vs. Fig 4C, Fig 3C, Fig 3D). Similar increased variance is seen at the inter-dimer GTP-interface, when the data are examined by projecting on the two main PCA modes (S3 Fig, S4 Fig, S4 Table, S5 Table). Guided by this observation, we hypothesized that the nucleotide could have a distinct effect on the mechanical properties of the inter-dimer interface. We therefore used two methods to quantify the flexural stiffness of tubulin interfaces. The first method was based on the equipartition theorem. Assuming that tubulin structures were already equilibrated by 500 ns of simulation, we pooled tubulin angles after that time and calculated their variance. According to equipartition theorem, at thermodynamic equilibrium variance, σ2, of a given conformational angle (θ or δ) should be inversely proportional to the respective harmonic flexural stiffness κ:
κ=kBTσ2
(1)
where kB is the Boltzmann constant, T is the temperature. Resulting harmonic stiffness values are summarized in Table 3.
These results suggest that the inter-dimer interface of GTP-tubulin is considerably more flexible than its intra-dimer interface in all kinds of rotation. Intra-dimer stiffnesses are not sensitive to nucleotide, speaking against the presence of significant allosteric effects of GTP hydrolysis on mechanical properties of the intra-dimer interfaces at equilibrium. Inter-dimer stiffnesses, however, are significantly lower in the GTP-tubulin model compared to the GDP-tubulin. The latter finding is suggestive of at least a partial contribution of flexural stiffness modulation from the nucleotide hydrolysis state into the mechanism of dynamic instability.
The equipartition theorem-based method has allowed comparing stiffnesses, corresponding to motions in a relatively narrow high-frequency range. To identify stiffnesses, characterizing conformational changes at longer timescales, we used NMA as a complementary approach. The squared mode frequency, related to each normal mode, can be reckoned into mechanical properties, such as bending stiffness/torsional rigidity, corresponding to the motion along the given mode [36–38]. As illustrated by Table 4, NMA confirmed that inter-dimer interface was much more flexible than the intra-dimer interfaces in the GTP-state, but that was not true for the inter-dimer interface of GDP-tubulin tetramer.
It is tempting to hypothesize that significant stiffening of the inter-dimer interface may be caused by inter-dimer compaction of the GDP-tubulin. But such compaction could well be released upon relaxation of GDP-tubulin tetramer, extracted from microtubule lattice. So we questioned, whether or not free GTP- and GDP-tubulin tetramers converged in the simulations to conformations with similar extent of the inter-dimer compaction. For straight microtubule lattice, intra-dimer and inter-dimer distances were previously used to characterize the extent of tubulin compaction. They were defined as the lengths of the vectors, connecting ribose rings of the nucleotides in the corresponding pairs of longitudinally bonded tubulins [23]. Applying the same metric for our simulated tubulin tetramers and averaging over the second half of one-microsecond-long simulations, we found that both GTP- and GDP-tubulins displayed similar extended intra-dimer and similar shorter inter-dimers distances (S3 Table). Interestingly, these numbers closely matched the corresponding distances in crystal structures of curved tubulins, e.g. in the structure of tubulin tetramer in complex with stathmin and vinblastine (S3 Table, [30]).
However, we note that this metric should be used with caution for describing compaction of curved tubulin structures, because in this case, the inter-tubulin interfaces may not shrink or extend predominantly along the vector, connecting the nucleotides in adjacent tubulins. Moreover, other substantial conformational changes may be present. Therefore, we decided to additionally characterize the inter- and intra-dimer interfaces with the number of contacts between α- and β-tubulin amino acids at the interface. In fact, the number of contacts should correlate with compaction because amino acids at the more compact interface come closer together. We find that the inter-dimer interface of GDP-tubulin retains the largest number of contacts throughout the simulation (S6 Table). Thus, the high number of contacts at the GDP-tubulin inter-dimer interface might explain the enhanced flexural stiffness of this interface.
Taking advantage of the new generation of cryo-EM-based in-microtubule tubulin structures in GDP and GTP-like states, we carried out several one-microsecond-long molecular dynamics simulations of free tubulin dimers and tetramers. Our simulations reveal that initially “compacted” GDP-bound tubulins, and initially “extended” GTP-bound tubulins both adopt similar non-radially curved and slightly twisted shapes. The presence of substantial tangential bending and twist components in the resulting relaxed tubulin conformations, is fully consistent with published crystal structures of tubulins in complex with MAPs [29–34] (Tables 1 and 2). However, in contrast to those data, no pronounced out-of-plane bending was observed in a recent cryo electron tomography study, which reported essentially flat protofilaments, lying mainly in the radial planes, containing the microtubule axis [14,35]. The origin of this discrepancy is not completely clear. But it could be partially explained by the conformational effects, induced by attachment of tubulin oligomers to the plus-end of the microtubule, as suggested by our simulations of single and three tubulin hexamers, longitudinally fixed at the minus-end. It might also be possible that several longitudinally attached tubulins experience some kind of cooperative behavior, similar to described with FtsZ [39].
Strikingly, we find that the nucleotide type does not affect the curvature or mechanics of the intra-dimer interface, but it does considerably modify the stiffness of the inter-dimer interface. Prior to GTP hydrolysis the inter-dimer interface is significantly softer than the intra-dimer one. Based on our simulations, we propose that enlarged number of contacts at the inter-dimer interface in the GDP-state makes it even slightly stiffer than the intra-dimer interface.
In our opinion, these findings have at least three important implications for our understanding of the mechanisms of microtubule instability.
First, non-radial bending and twisting of tubulins during their relaxation from straight to curved configuration likely means that the lateral bonds on two sides of the splaying protofilaments at the microtubule tip experience unequal mechanical stress. This conclusion is qualitatively similar to the results of a previously study, which considered a hypothetical microtubule tip, constructed based on X-ray structures of curved tubulins in complex with stathmin [40]. We speculate that uneven distribution of mechanical stress on lateral bonds may lead to sequential, rather than simultaneous, rupture of the lateral bonds, which has not been previously considered by the majority of existing models of microtubule dynamics [41–46], with one notable exception [45]. This proposal is illustrated by our simulations of three laterally attached protofilaments, which splay apart by breaking lateral bonds between the right and the middle protofilaments sooner than the bonds between the middle and the left protofilaments (Fig 6). Given the fact that sequential breakage of the lateral bonds is energetically more feasible than their simultaneous breakage, we believe that taking this new feature of tubulin mechanics into account may lead to significant revision of current estimates for lateral bonds between tubulins. More work is needed to investigate full lateral bond rupture and its dependence on the nucleotide.
Second, stiffening of the compacted interface between GDP-tubulin dimers helps to resolve the apparent paradox of a lack of clearly visible effects of the nucleotide on lateral bonds and on the curvature of free tubulin protofilaments, despite a well-established link between microtubule stability and the type of nucleotides associated with the microtubule lattice. Indeed, most experimental work and theoretical thinking in the literature have been focused on an effort to explain dynamic instability by nucleotide-dependent curvature or by nucleotide-dependent strengths of lateral bonds. The possibility of modulating tubulin flexural stiffness by the nucleotide has been raised by only a small number of studies, e.g. [10,21,27], and underappreciated. Recently Igaev and Grubmüller suggested an allosteric mechanism for dynamic instability, based on the nucleotide-driven tubulin dimer stiffness change [21]. Our current study essentially points to a very similar idea: the softer interface between GTP-tubulin dimers requires less additional energy to straighten curved GTP-tubulin protofilaments in order to incorporate them into the microtubule wall, compared to a stiffer GDP-tubulin protofilament. However, importantly, in this study we did not observe any significant difference in flexural stiffnesses between intra-dimer interfaces of GTP and GDP-tubulins. Instead, the nucleotide dramatically affected the inter-dimer interface around the exchangeable nucleotide binding site. This inter-dimer interface was not examined in the former study, which was focused on free tubulin dimers. We do not think that the disagreement in our conclusions about tubulin dimers could be related to the tubulin structures, or molecular dynamics simulation parameters we used, as they were similar. But there were differences in the approach, with which the flexibilities were assessed. Here we measured the flexibility of tubulin interfaces around their relaxed curved conformations, while Igaev and Grubmüller used the umbrella sampling method to probe energetic landscape on a larger scale along a reaction coordinate, which did not exactly correspond to the bend and twist angles, which we used in this study. An equipartition-based analysis, similar to ours, was first carried out in the pioneering paper by Grafmüller and Voth [15]. In contrast to the present study, the authors did not find any statistically significant nucleotide-dependence of either intra- or inter-tubulin interfaces. The stiffness of inter-dimer interfaces clearly depended on the structures they used. At the time of the study, only a low resolution structure of straight GDP-bound tubulin from Zn-induced sheets with antiparallel protofilaments was available (PDB code: 1JFF). That tubulin structure was compacted around the exchangeable nucleotide-binding site, so it is unclear if simple insertion of Mg2+ ion and substituting GDP molecule for GTP would drive a correct GTP-tubulin conformation, given the complexity of tubulin’s energy landscape. The authors also examined the relaxation of tubulins from the curved structure (PDB code: 1SA0), which represented a complex of colchicine, stathmin and GDP-tubulin tetramer. Although the structure had a higher resolution, the interface between dimers in presence of stathmin could be affected by this MAP. Moreover, computational resources now allow substantially longer simulation times, which increase the chances of more complete relaxation of the tubulin structure.
Finally, we find unequal flexural stiffness of the inter-dimer and intra-dimer interfaces of GTP-tubulins, and less different but still distinct flexural stiffness of inter- and intra-dimer interfaces of GDP-tubulins. We argue that these features could be essential for explaining the difference in the rates of assembly and disassembly of plus- and minus-ends of microtubules. It has recently been proposed, that microtubules assemble and disassemble by dynamic peeling and unpeeling of curved GTP-protofilaments at the microtubule tip, so that the balance between the lateral bonds and outward bending/twisting energy controls the rate of both microtubule assembly and disassembly [14]. The minus-ends of microtubules terminate with α-tubulins, while the plus-ends terminate with β-tubulins. Obviously, if the lateral bonds between α-α and β-β tubulins were equal, microtubule rates would be identical at both ends. Distinct lateral bonds between α-α and β-β tubulins alone cannot render the rates different either, if the flexural stiffnesses between any pair of tubulin subunits is the same. However, if the inter-dimer interface is soft, compared to the intra-dimer interface, the terminal layer of subunits is under higher bending stress compared to the layer second to terminal (Fig 7). Indeed, the lateral bonds of the terminal layer are opposed by the large bending force, coming from stiff intra-dimer interface, trying to curve the subunit out and break the lateral bonds, so these layers should break lateral bonds relatively fast. The layer second to terminal, on the other hand, is under lower bending stress, because the inter-dimer stiffness is significantly softer. Therefore, lateral bonds holding the second to terminal layers would exist for longer time and become rate limiting. This means that the second to terminal layer, composed of β-tubulins at the minus-end and α-tubulins at the plus-end can control the rates of assembly from GTP-tubulin, explaining why those rates could differ.
An alternative but not mutually exclusive explanation of the difference in plus- and minus-end assembly/disassembly rates, obviously, is a possible difference in the on-rates of tubulins associating with microtubules at either end. However, it is unclear how such difference in the on-rates could arise, as association at either microtubule end seems to be dependent on the same kinds of interactions.
Overall, our study highlights a set of key mechanical properties of tubulin interfaces, shedding new light onto the mechanism of microtubule dynamics. We hope that our results will facilitate construction of more complete and realistic models of microtubule dynamics in the future.
Molecular models of the straight GDP-bound tubulin and Mg-GTP-bound tubulin structures (dimers, tetramers and hexamers) were based on 3J6F and 3J6E PDB structures [22]. The latter contained the slowly hydrolysable GTP analog (GMPCPP) in the exchangeable nucleotide-binding site of β-tubulin. GMPCPP was converted into GTP by replacing the carbon atom between α- and β-phosphate with an oxygen atom, and the new bond lengths and angle relaxed to their equilibrium values during minimization. Molecular model of three laterally bonded tubulins was based on 3J6F PDB structure. We added unresolved mobile amino acid chains in all the models, using the Modeller program [47]. Propka software [48] was used to calculate the unknown degree of protonation of ionizable amino acid residues and Dowser program [49] was used to identify and solvate cavities inside the protein. A virtual three-dimensional cubic reaction volume filled with TIP3P water with periodic boundary conditions was used for the simulation. The size of the reaction volume was set in such a way that the distance from the protein surface to the nearest box boundary was not initially less than two nanometers. The ionic strength of the solution was set at 100 mM by adding K+ and Cl− ions and the total charge of the system was zero. Simulations were performed using the GROMACS 5 software package, which allows parallel computing on hybrid architecture [50] with the CHARMM27 force field [51,52]. The parameters of the GTP and GDP molecules were also taken from the CHARMM27 force field, and parameters of their phosphate groups were set in accordance with [53], similar to the phosphate groups of ATP and ADP.
After preparing each of the tubulin systems as described above, we minimized their energies using the steepest descent algorithm. Energy minimization was followed by a two-step equilibration. First, we conducted one-nanosecond-long simulations with constrained positions of all heavy protein atoms at constant pressure and temperature. Second, we carried out five-nanoseconds-long simulations with constrained positions of protein backbone atoms, using the Berendsen barostat (time constant 4.0 fs, compressibility 4.5 × 10−5 bar−1) and the Berendsen thermostat. The production simulation runs were carried out in the NPT ensemble at 300K, using the Parrinello-Rahman algorithm [54] and the V-rescale thermostat for a duration of 1 μs each. In the simulations with single and triple tubulin hexamer systems position restrain was applied on all Cα atoms of the minus-end proximal α-tubulin subunits. The particle mesh Ewald method was used to treat the long-range electrostatics. All-bond P-LINCS constraints and mass rescaling (partial transfer of mass from heavy atoms to bound hydrogens [55]) allowed molecular dynamics simulations with 4 fs time step. VMD [56] and Pymol (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC) were used for visualization. Detailed analysis of computational performance of simulations is presented in [57].
Analyses of bend and twist angles at tubulin interfaces were carried out with Pymol (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC.) software in combination with home-made python scripts, available upon request. The scripts were realizing the following procedure. First, a coordinate system was associated with a fragment of microtubule wall structure (PDB codes: 3J6E and 3J6F), so that the orientations of the coordinate vectors were such as illustrated in Fig 1A. Then we aligned the pair of tubulin monomers under examination onto the microtubule wall fragment. For the alignment we used only the terminal globular domains of one of the tubulin subunits in the examined pair. The terminal domains were defined as described previously [15]. This way the reference subunit was aligned along the microtubule-bound coordinate system xyz (Fig 1B, blue). To determine the orientation of the second tubulin subunit in the examined pair relative to the reference subunit, another microtubule wall fragment was aligned onto the second tubulin subunit, producing three more orientation vectors: X, Y, Z (Fig 1B, red). The magnitude θ and direction φ of the bending at the tubulin-tubulin interface were calculated as:
θ=arccos(zZ)
(2)
φ={arccos(−zy1−zz2),ifZx>0−arccos(−zy1−zz2),ifZx≤0
(3)
Further, auxiliary coordinate vectors, x', y', z', shown in Fig 1B in cyan, were obtained by rotating the coordinate vectors x, y, z by angle θ around axis P. That axis was defined as cross-product of vectors z and Z. Amplitude of the twist angle was calculated as:
δ=arccos(x′X)
(4)
The sign of the δ-angle was defined positive when the triple product of vectors {x', X, Z} was positive. In other words, the clockwise direction (viewed from the microtubule plus-end), was defined as positive twist, the anti-clockwise rotation was considered to be a negative twist. Thus, the conformational angles could adopt values in the following ranges: θ adopted values from 0 to 90 degrees; φ and δ adopted values from -180 degrees to 180 degrees, unless stated otherwise.
Compaction at tubulin interfaces was characterized with inter-tubulin domain distances[23], which were measured between centers of mass of ribose rings in the nucleotides of adjacent tubulin subunits for every nanosecond frame of the last 500 ns of each simulation and then averaged over all the runs. Only the minus-end proximal intra-dimer interfaces of tetramers were considered, as the nucleotides at the plus-ends of tetramers were more mobile and in one case even escaped tubulin pocket by the end of one-microsecond simulation.
The number of contacts between tubulins at the longitudinal interfaces was determined for every nanosecond frame of the last 500 ns of each simulation run and then averaged over all the runs. A contact was defined as proximity of Cα atoms of two amino acids closer than the 8Å threshold. Lateral contacts between pairs of terminal β-tubulin subunits of the laterally bonded hexamers were determined analogously for every nanosecond of the simulation and plotted against simulation time.
We analyzed global modes of macromolecular mobility of tubulin oligomers using principal component analysis (PCA) applied to obtained molecular dynamics trajectories. PCA is a statistical technique commonly used for dimensionality reduction and determination of a subset of linearly independent variables (called principal components, PCs) explaining most of the variation observed in the original data. PCA relies on construction and diagonalization of the symmetric covariance matrix C between all the pairs of coordinates (e.g., Cartesian coordinates of atoms):
C=[1⋯〈(xi−〈xi〉)(xj−〈xj〉)〉⋮⋱⋮〈(xj−〈xj〉)(xi−〈xi〉)〉⋯1]
(5)
where, xi is a Cartesian coordinate of i-th atom, and < …> means averaging over all of the sampled conformations (e.g., frames of molecular dynamics trajectory). Diagonalization of C yields eigenvectors with corresponding eigenvalues. While the former represent the collective motions, the latter designate the respective variance. In case of molecular simulations, a limited number of eigenvectors (usually 1–10) with the largest eigenvalues describe the vast majority of variance [58].
PCA was performed with the Prody toolkit [59] using protein Cα atoms only (excluding flexible loops and C-terminal tails) and a coarse-grained representation for GTP/GDP (one bead per guanine, ribose, each phosphate group and magnesium) upon the roto-translational alignment of all the frames of molecular dynamics trajectories to the initial configurations.
Normal mode analysis (NMA) is a powerful and widely used method for prediction of functional modes of protein mobility [60] and assessment of mechanical properties of macromolecular complexes [37]. This approach is computationally similar to PCA, but instead of covariance matrix, the Hessian matrix (composed of the second-order derivatives of the potential energy by coordinates) is diagonalized, giving eigenvectors that correspond to the collective, low-frequency molecular modes, and eigenvalues that characterize the stiffness of the corresponding normal modes. Due to the large size of the systems investigated, we used NMA in its simplified form, based on the elastic network model (ENM). The built ENM was based on the straight conformations of tubulin tetramers and consisted of protein Cα atoms and a coarse-grained representation for GTP/GDP (akin to a subset of atoms used in PCA) connected with harmonic springs when closer than the cutoff distance, Rcutoff = 0.8 nm. The beads were assigned the masses of all-atom fragments replaced by them. Instead of uniform spring constants normally used for ENM, we parametrize heterogeneous ENM based on the performed all-atom simulations, which results in more accurate models [61]. We used the parametrization procedure based on the Boltzmann inversion and the iterative scheme, which we describe in details in [62]. In brief, the approach is based on fitting the fluctuations of pair distances in ENM to corresponding fluctuations computed from all-atom simulations via iterative adjustment of spring constants in ENM.
The heterogeneous ENM analysis was performed by in-house scripts exploiting the Prody toolkit [59]. The collective modes acquired by NMA were compared with principal components obtained for the same systems by computing their pairwise overlaps, which are given by the correlation cosines of corresponding eigenvectors [63].
The eigenvalue of n-th normal mode, λn, is related to its angular frequency, ωn=λn. The related vibrational frequency and mechanical stiffness for each normal mode of tubulin oligomers can be estimated applying the linear elastic beam theory [37,38]. In our analysis we assumed that tubulin oligomers behave as freely vibrating elastic filaments. Then, bending stiffness Kbend (for normal modes corresponding to bending of oligomers) and torsional rigidity Ktwist (for normal modes resembling twist-like motions of oligomers) can be found from equations:
Kbend=ρlω2k4
(6)
and
Ktwist=ρvIω2k2
(7)
where ω—angular frequency, k—corresponding wavenumber (depending on the specific boundary conditions of the wave equations and taken from [36]), ρl—mass per unit length (1.95⋅10−14 kg/m), ρv—mass per unit volume (1.3⋅103 kg/m3), and I –the moment of inertia of the cross-sectional area with respect to the long axis of the filament (3.42⋅10−35 m4).
|
10.1371/journal.ppat.1003363 | Global Organization of a Positive-strand RNA Virus Genome | The genomes of plus-strand RNA viruses contain many regulatory sequences and structures that direct different viral processes. The traditional view of these RNA elements are as local structures present in non-coding regions. However, this view is changing due to the discovery of regulatory elements in coding regions and functional long-range intra-genomic base pairing interactions. The ∼4.8 kb long RNA genome of the tombusvirus tomato bushy stunt virus (TBSV) contains these types of structural features, including six different functional long-distance interactions. We hypothesized that to achieve these multiple interactions this viral genome must utilize a large-scale organizational strategy and, accordingly, we sought to assess the global conformation of the entire TBSV genome. Atomic force micrographs of the genome indicated a mostly condensed structure composed of interconnected protrusions extending from a central hub. This configuration was consistent with the genomic secondary structure model generated using high-throughput selective 2′-hydroxyl acylation analysed by primer extension (i.e. SHAPE), which predicted different sized RNA domains originating from a central region. Known RNA elements were identified in both domain and inter-domain regions, and novel structural features were predicted and functionally confirmed. Interestingly, only two of the six long-range interactions known to form were present in the structural model. However, for those interactions that did not form, complementary partner sequences were positioned relatively close to each other in the structure, suggesting that the secondary structure level of viral genome structure could provide a basic scaffold for the formation of different long-range interactions. The higher-order structural model for the TBSV RNA genome provides a snapshot of the complex framework that allows multiple functional components to operate in concert within a confined context.
| The genomes of many important pathogenic viruses are made of RNA. These genomes encode viral proteins and contain regulatory sequences and structures. In some viruses, distant regions of the RNA genome can interact with each other via base pairing, which suggests that certain genomes may take on well-defined conformations. This concept was investigated using a tombusvirus RNA genome that contains several long-range RNA interactions. The results of microscopic and biochemical analyses indicated a compact genome conformation with structured regions radiating from a central core. The structural model was compatible with some, but not all, long-range interactions, suggesting that the genome is a dynamic molecule that assumes different conformations. The analysis also revealed new structural features of the genome, some of which were shown to be functionally relevant. This study advances our understanding of the role played by global structure in virus genome function and provides a model to further investigate its in role virus reproduction. We anticipate that organizational principles revealed by this investigation will be applicable to other viruses.
| Many viruses possess RNA genomes, and those with single-stranded plus-sense RNA genomes represent an important subgroup that includes many pathogenic plant, animal, and human viruses. These RNA genomes serve the traditional role as the blueprint for viral proteins; however, they also contain cis-acting RNA elements (RE) that direct different viral processes, such as protein translation, genome replication and transcription of subgenomic (sg) mRNAs [1]–[3]. Accordingly, REs need to be structurally and functionally integrated with coding sequences. One solution utilized is to position REs in non-coding regions, thereby physically separating coding and RE functions. Indeed, many viruses use this strategy by situating their REs terminally in 5′- and 3′-untranslated regions (UTRs) or/and internally within inter-cistronic regions [1], [3]. One drawback to this approach is that the size and/or location of the non-coding regions can be limiting. Thus, to extend this range, many plus-strand RNA viruses also position REs within coding regions [1].
One potential drawback of having an RE in a coding region is that it physically couples distinct activities to the same RNA sequence; therefore, some compromise in one or both functions may be required. Additionally, in some cases the relative location of REs within a genome may not be optimal for their activity, thus compensatory measures may be required. One strategy used by many RNA viruses to deal with suboptimally positioned REs is to reorganize their relative location within the genome via intramolecular long-range RNA-RNA interactions [4]. This tactic is employed by a number of different plant viruses to mediate translation of viral proteins from their uncapped and nonpolyadenylated genomes. Specifically, luteoviruses [5], tombusviruses [6], [7], carmoviruses [8] and umbraviruses [9] recruit translation initiation factors to a 3′-proximal RE, termed a 3′ cap-independent translational enhancer (3′CITE), that interacts with the 5′UTR via a long-range RNA-RNA interaction to mediate translation initiation. Interactions between genomic termini have been identified in other types of plus-strand RNA viruses, such as hepatitis C virus (HCV) [10] and picornaviruses [11], but their roles remain to be fully elucidated. Long-range interactions are also used for viral RNA genome replication in bacterial (i.e. phage Q-beta) [12], plant (i.e. tombusviruses) [13] and animal viruses (i.e. flaviviruses) [14], where viral polymerase binding sites are located 5′-terminally or internally in the RNA genome and the bound polymerase is repositioned near the 3′ end of the genome by long-range interactions, thereby allowing for initiation of minus-strand synthesis. In other cases, long-distance RNA-based interactions can themselves form functional RNA structures, as seen in insect nodaviruses [15], animal coronaviruses [16], [17] and plant tombusviruses [18], [19]. For these viruses, the interactions form RNA structures that act as terminators for the viral polymerase during minus-strand synthesis, which then lead to formation of truncated minus-strand templates used to transcribe viral sg mRNAs. Very recently, a second long-range interaction spanning ∼26 kb was identified in transmissible gastroenteritis coronavirus that was required for efficient transcription of its sg mRNA encoding the N gene [20]. As coronaviruses transcribe numerous distinct sg mRNAs, this finding suggests that their genomes may harbour a large number of long-distance interactions. The prevalence of long-range RNA-RNA interactions in a variety of viruses in both coding and non-coding regions suggests that their RNA genomes must maintain a considerable level of global organization. Recently, the secondary structure of the entire human immunodeficiency virus-1 (HIV-1) RNA genome was reported, which revealed many novel structural features within the global genomic context [21]. However, for standard plus-strand RNA viruses (i.e. non-retroviral), no detailed genome-scale models of higher-order structure exist, thus our understanding of genome architecture and dynamics for this important class of virus is very limited.
Tombusviruses, such as tomato bushy stunt virus (TBSV), are plus-strand RNA viruses that encode five proteins in a ∼4.8 kb long genome [22]. p33 and its readthrough product p92, the RNA-dependent RNA polymerase, are both required for genome replication and sg mRNA transcription, and both of these proteins are translated directly from the genome [23]. The more 3′-proximal p41 is the capsid protein that is translated from sg mRNA1, the larger of two subgenomic messages transcribed during infections [23]. The p19 suppressor of gene silencing and p22 movement proteins are both translated from the smaller sg mRNA2 [23]. Analyses of TBSV and other tombusviruses have led to significant understanding of many aspects of the tombusvirus reproductive cycle [23]. Numerous functional cis-acting REs have been mapped within their genomes, and viral and host proteins involved in mediating infections have been characterized [2], [24]. A particularly intriguing aspect of tombusviruses is that they use a vast network of intra-genomic long-range RNA-RNA interactions to mediate a number of different processes [4]. In total, six different functional long-distance base pairing interactions, each spanning sequences ≥1 kb, have been identified (Figure 1A). Translation of p33 requires the longest interaction, which occurs between the 3′CITE in the 3′UTR and the genomic 5′UTR [25]. Translational readthrough of the p33 stop codon to produce p92 involves an interaction between the proximal readthrough element (PRTE) located 3′ to the stop codon and the distal readthrough element (DRTE) in the 3′UTR [26]. Genome replication requires an interaction between the upstream linker (UL) sequence just 3′ to the essential replication element RII(+)SL and the downstream linker (DL) in the 3′UTR [13]. This latter interaction is also needed for efficient translational readthrough that generates p92 [26]. Transcription of the sg mRNAs utilizes three different interactions; the activator sequence 1-receptor sequence 1 (AS1-RS1) interaction mediates sg mRNA1 transcription [18] and the AS2-RS2 and distal element-core element (DE-CE) interactions facilitate sg mRNA2 transcription [19], [27].
Interestingly, the latter five interactions described are all part of a set of nested interactions, suggesting possible structural order within this grouping (Figure 1A). Additionally, some of the interacting sequences are located relatively close to each other (i.e. UL, AS2 and AS1), hinting at possible shared organizational strategies and/or functional or regulatory crosstalk (Figure 1A). Accordingly, we were interested in investigating the genome architecture of a tombusvirus, namely TBSV, in order to gain a better understanding of how local REs and multiple long-range interactions are accommodated and functionally integrated within the genome. The results from our structural analyses indicate that the TBSV genome forms a relatively compact structure that is organized into a number of differently sized domains. The largest domains mediate functional long-range interactions, whereas some of the smaller domains correspond to local REs. The analyses also revealed novel long-range interactions and local structures, some of which were shown to be functional. Collectively, this study provides considerable new insight into the structure and dynamics of a plus-strand RNA genome and supports the intriguing concept that global organization represents an integral component of genome function.
As an initial approach to gain insights into the global organization of the TBSV genome we used atomic force microscopy (AFM). This type of analysis provides information on the overall topography of an RNA molecule adsorbed to a mica surface and has been used successfully to assess global conformations of different viral RNA genomes [28], [29]. Our results revealed that TBSV genomes exist primarily as compact irregular structures, with average maximum diameters of 98 ±12 nm (n = 40) (Figure 2). A theoretical maximum ladder distance (MLD) [30] of 248 bp was calculated for the TBSV genome, which represents the number of base pairs passed when spanning the two furthest points in the optimal structure, predicted by the RNAstructure software [31]. With the average rise per base pair in dsRNA corresponding to ∼0.27 nm [32], this MLD converts to an average measurement of ∼67 nm. The MLD average over 1000 suboptimal secondary structures was calculated as 252±6 bp or ∼68 nm. These somewhat lower values are generally consistent with the measured value from AFM, because they do not include increases in length contributed by the unstructured regions connecting the base paired segments. Overall, the estimated dimensions from the RNAstructure-predicted configuration and the AFM analysis suggest a compact secondary structure for the TBSV genome.
Due to conformational flexibility, variable surface adsorption, and limits of resolution, a single well-defined structure for all genome molecules was not anticipated. However, what was evident from the AFM analysis was the general conservation of a global organization consisting of irregular extensions emanating from a central core (Figure 2). Some variation from this overall arrangement was evident, but the level of compactness and the presence of multiple tethered substructures were typically maintained. These results suggest that the TBSV genome assumes a mostly condensed structure that is composed of interconnected domain-like protrusions extending from a central hub. In this respect, its genome structure is most similar to the more compact structures observed via AFM for HCV and Hepatitis G virus RNA genomes, and contrasts the more extended arrangements seen for poliovirus and rubella virus RNA genomes [28], [29]. Extended genome configurations suggest smaller consecutive locally-folded domains, whereas more condensed structures are consistent with genome configurations that include multiple long-distance RNA-RNA interactions [28]. The compact structures observed for the TBSV genome are in agreement with the latter concept and are consistent with the numerous long-range interactions known to occur in this genome [4].
We next sought to determine the RNA secondary structure of the TBSV genome using high-throughput selective 2′-hydroxyl acylation analysed by primer extension (SHAPE) [33]. Full-length transcripts of the TBSV genome were denatured, snap-cooled, and then refolded in an effort to (i) minimize intermolecular interactions, (ii) release misfolded molecules from kinetic traps, and (iii) promote formation of the most stable secondary structure(s). The latter event would presumably reduce structural diversity in the population and allow for more readily interpretable results. As transcripts of plus-strand RNA virus genomes are infectious when transfected into cells, the refolded viral RNA was tested for replication in plant protoplasts. Progeny genomes from refolded or untreated transcripts accumulated to equivalent levels (Figure S1), indicating that the folding step did not interfere with virus infectivity. Refolded RNA transcripts were treated with 1-methyl-7-nitroisatoic anhydride, which preferentially acylates the 2′-OH position of riboses associated with conformationally flexible nucleotides [33]. Accordingly, heavily modified positions are predicted to correspond to nucleotides that are primarily single-stranded, while unmodified or weakly modified residues are considered to be mainly paired. Inflexible regions can also correspond to non-Watson/Crick base pairs or tertiary interactions. Modified positions in the viral genome were detected and quantified using primer extension and capillary electrophoresis [33].
SHAPE reactivity data were collected for ∼91% of the 4778 nt long TBSV genome, with the final values derived from the averages from two separate SHAPE analyses (Table S1). Figure 1B shows median SHAPE reactivities along the TBSV RNA genome based on a 75 nt long moving window. The average SHAPE reactivity based on this analysis was 0.488 (Figure 1B). Two regions corresponding to positions 1008–1163 and 2464–2632 could not be assessed due to strong stops in the reverse transcription reactions (Figure 1B). Also, no reactivity data were collected for the 5′-terminal 9 nt and the 3′-terminal 83 nt of the genome; the latter was due to the lack of a priming site available downstream of this segment. The absence of the terminal information was not problematic because prior analyses had thoroughly defined the functionally relevant secondary and tertiary structures in the 5′UTR and 3′UTR [34]–[36]. For the internal segments where no SHAPE data were available, structure prediction was guided by thermodynamics-based parameters only. The SHAPE reactivity data collected for the rest of the genome were normalized and incorporated into the thermodynamics-based RNAstructure software using a pseudo free-energy change term [33]. Subsequently, this program, guided by both thermodynamics and SHAPE data, was used to predict the RNA secondary structure for the TBSV genome. No limit was placed on the maximum base pairing distance for the folding analysis, because many experimentally verified long-range interactions are known to form in the TBSV genome [4]. To prevent improbable long-range interactions between internal regions and experimentally-verified local structures situated at either terminus of the genome, formation of the confirmed local RNA structures in the 5′UTR (nts 1–166) and 3′UTR (nts 4697–4778) was added as a constraint in the folding analysis. This restriction was introduced because prior comprehensive mutational analyses of these terminal structures indicated that it is highly improbable that base paired regions within these local structures participate in alternative functionally relevant long-range base pairing interactions [34]–[38]. Based on the parameters described above, an RNA secondary structure model for the entire TBSV genome was predicted by SHAPE-guided RNAstructure-based analysis (Figure 3). It should be noted that the RNA structure generated included only secondary structures formed by canonical base pairs, as this algorithm does not predict non-canonical interactions or tertiary structures such as pseudoknots.
Initially, we were interested in determining whether the SHAPE data contributed significantly to the refinement of the genomic structural model. The model that included the SHAPE data (i.e. SHAPE-plus: SHAPE and thermodynamics) predicted 60.4% of the genome to be base paired, while exclusion of the SHAPE data (i.e. SHAPE-minus: thermodynamics-only) yielded a structure with 65.5% base pairing. Accordingly, addition of the SHAPE data reduced overall predicted pairing by ∼5%. Approximately 70% of the base pairs in the SHAPE-plus structure were identical to those in the SHAPE-minus structure; therefore inclusion of the SHAPE data did result in altered pairing schemes and model refinement. Interestingly, this latter value is higher than the corresponding 47% determined during structural analysis of the HIV-1 genome [21]. Thus, compared to HIV-1, the SHAPE data for the TBSV genome were more consistent with the thermodynamics-only prediction.
At a global level, the structural model revealed distinct sub-regions of RNA secondary structure that we herein define as RNA domains. The boundaries (i.e. closing stems) of several of these domains were closely associated, leading to their emergence from a common area and an overall floret-like organization (Figure 3). Notably, this arrangement is consistent with the structures observed by AFM indicating domain-like protrusions emanating from a central core (Figure 2). The MLD calculated for the SHAPE-directed optimal and 1000 suboptimal structures was 216±19 bp (∼58 nm), consistent with compact structures. Such compressed arrangements differ from the more extended beads-on-a-string-like genomic organization proposed for HIV based on high-throughput SHAPE analysis, implying that TBSV and HIV have rather different genomic organizations [21]. Also, a very recent report of the in vitro structure of the ∼1 kb long plus-strand RNA genome of satellite tobacco mosaic virus (unrelated to tombusviruses) predicted a highly extended secondary structure [39] that contrasts the highly condensed structure proposed for TBSV.
In TBSV, the RNA domains could be divided into groups based on their relative sizes (i.e. small, 80–200 nt; medium, 200–500 nt; large, 500–2000 nt). Accordingly, three small, three medium, and four large domains were defined (Figure 4). The three small domains (sD) corresponded to the 5′UTR (termed sD1, 166 nt long, genome coordinates 1–166), the terminal 81 nts of the 3′-UTR (sD3, 81 nt, 4697–4777), and an internal region in the p33 ORF (sD2, 174 nt, 747–921) (Figure 4). The terminal sD1 and sD3 were functionally characterized previously and are involved in genome replication (sD1 and sD3) [34]–[38], translation (sD1) [25], [38] and readthrough (sD3) [26], whereas the role of sD2 is currently unknown. The three medium domains (mD) were derived from a region in the p92 ORF that overlapped with the p33 ORF stop codon (mD1, 229 nt, 1013–1241), a region downstream of the p33 stop codon in the p92 ORF (mD2, 216 nt, 1328–1543), and the 5′-portion of the 3′UTR (mD3, 275 nt, 4421–4695) (Figure 4). mD1 contains a prominent stem-loop structure containing the PRTE that interacts with the DRTE in SD3 to mediate readthrough [26] (Figure 4). mD2 contains the large stem-loop structure RII(+)-SL and the UL sequence, both of which are involved in genome replication [13], [40], [41]. mD3 corresponds to the 3′CITE that is important for translational enhancement [25], [38]. Lastly, the four large domains (lD) correspond to the 5′-proximal two-thirds of the p33 ORF (lD1, 547 nt, 172–718), the 3′-proximal two-thirds of the readthrough portion of the p92 ORF (lD2, 1047 nt, 1589–2635), the CP ORF (lD3, 1192 nt, 2639–3830), and the overlapping p22/19 ORFs (lD4, 540 nt, 3832–4371) (Figure 3 and 4). Base paired AS1 and RS1 elements, which mediate sg mRNA1 transcription [18], were present at the ends of lD2. Similarly, base paired DE and CE segments, which facilitate sg mRNA2 transcription [19], [27], closed the lD3 domain. At the 5′ end of the 1D4 domain, the RS2 sequence involved in sg mRNA2 transcription was paired to the 3′-end of 1D3, instead of to its far-upstream partner sequence AS2, suggesting that cognate pairs may have alternative partner sequences (Figure 3 and 4). The internal regions of the three consecutive lDs do not contain any known functional RNA elements; thus, their primary function may be to assist in the global positioning of transcriptional elements, as is discussed in more detail in the subsequent section.
The linear 5′-to-3′ order of the ten defined domains is sD1-lD1-sD2-mD1-mD2-lD2-lD3-lD4-mD3-sD3, with inter-domain lengths ranging from as little as 1 nt (between lD3 and lD4, and between mD3 and sD3) up to 91 nts (between sD2 and mD1). Analysis of the predicted structures of the optimal and 1000 suboptimal SHAPE-plus structures revealed that the assigned domains, with the exception of mD1, were well maintained in the population (Figure S2). Some of the inter-domain regions contained RNA elements known to be functional (Figure 4). For example, the inter-domain region between mD2 and lD2 contains the transcriptional AS2 element that is present in the loop of a local stem-loop RNA structure that can pair with its downstream partner sequence RS2 [19] (Figure 4). However, in the structure, the RS2 sequence is paired with a non-cognate partner sequence at the base of lD4 (Figure 3). The inter-domain region between lD4 and mD3 also contains a functional element, the DL sequence, which is complementary to the UL sequence in mD2 (Figure 4). This interaction is required for genome replication [13] as well as translational readthrough [26]. However, in the structure, some of DL sequence is part of a small local stem-loop structure (Figure 4). Collectively, examination of the structural model for the TBSV genome revealed several interesting features: (i) three of the four lDs are grouped consecutively, are closely linked (i.e. have small connecting inter-domain segments), and are positioned more centrally in the genome; (ii) ∼66% of the genome is present in the four lDs, internal regions of these domains do not contain known functional RNA elements, and each domain corresponds to a different viral ORF; (iii) the sDs and mDs tend to be located closer to genomic termini; (iv) known functional RNA elements are located in both domain and inter-domain regions; (v) long-range RNA-RNA interactions are evident throughout the structure, but only a third of the known functional long-distance RNA-RNA interactions (i.e. AS2-RS2 and DE-CE) are present in the model.
Previous studies uncovered six different functional long-range RNA-RNA interactions that occur during the TBSV reproductive cycle (Figure 4A). Interestingly, four of these six interactions were not predicted in our structural model, namely, 5′UTR-3′CITE, UL-DL, PRTE-DRTE, and AS2-RS2 interactions (Figure 4B). It is possible that these interactions did not form under the conditions tested or, alternatively, they did form, but limitations in the algorithm prevented their incorporation into the model. Indeed, all of the missing interactions could involve tertiary structures that are not predicted by the program. To try to distinguish between these two possibilities, the shape reactivities of the partner sequences in these interactions were assessed under the premise that high SHAPE reactivities for one or both partner sequences would be consistent with the absence of the cognate interaction, regardless of predictive limitations in the algorithm. Relative reactivities were not available for the PRTE and DRTE due to “blind spots” in the analysis, as explained earlier. Of the remaining partner elements, high reactivities (i.e. ≥0.7) were abundant in both complementary sequences for the 5′UTR-3′CITE interaction, the UL sequence, and the AS2 element (Table S2), consistent with their predicted unpaired status in the structure (Figure 3 and 4B). In contrast, DL and RS2 (the partner sequences for UL and AS2, respectively) exhibited relatively lower reactivity scores in general (Table S2) and, consequently, were predicted to form a local hairpin structure or to pair with non-cognate partner sequences, respectively (Figure 3 and 4B). Overall, the relative reactivities of one or both of the partner sequences in the 5′UTR-3′CITE, UL-DL, and AS2-RS2 interactions are consistent with both the lack of their formation in the structure model and the presence of non-cognate pairing. Furthermore, analysis of the 1000 suboptimal SHAPE-plus structures did not predict formation of any of these interactions (Figure S3). Indeed, it may not be possible for all six distinct interactions to occur simultaneously, due to local and/or global conformational constraints. Moreover, mutual exclusion between certain interactions could be advantageous for coordinating sequential or opposing processes. Interestingly, the UL-DL interaction was predicted to occur in the SHAPE-minus structure (Figure S4), indicating that thermodynamics-only structural prediction can be helpful for identifying alternative conformations that are be biologically relevant. For the SHAPE-plus structure, it is interesting to note that although several of the long-range interactions were not predicted, the cognate partner sequences tended to be in proximity of each other in the predicted secondary structure (Figure 3 and 4). Specifically, the ladder distances separating the unpaired partner sequences of 5′UTR-3′CITE, PRTE-DRTE, UL-DL and AS2-RS2 were, respectively, 78 bp (∼21 nm), 29 bp (∼8 nm), 20 bp (∼5.5 nm) and 9 bp (∼2.5 nm). Such proximity could facilitate the structural transitions required to allow these other interactions to form and suggests that the secondary structure level of viral genome structure could provide a basic scaffold for the formation of different long-range interactions. In this respect, the three consecutive large domains would play major roles in bringing together the majority of the interacting partner sequences (Figure 3 and 4).
The two long-range interactions that were predicted to occur in the structural model correspond to AS1-RS1 and DE-CE (Figure 3 and 4). Interestingly, both of these interactions act to close large domains, lD2 and lD3, respectively, and both are involved in sg mRNA transcription, albeit for different sg mRNAs (i.e. sg mRNA1 and 2, respectively). Accordingly, the simultaneous presence of these two interactions may provide a glimpse at what a transcriptionally-active genome looks like. Although the AS2-RS2 interaction is not engaged in this model, the two sequences are juxtaposed by the DE-CE and AS1-RS1 interactions, which could facilitate transient AS2-RS2 pairing. Indeed, the DE-CE interaction is critical for AS2-RS2-dependent sg mRNAs transcription [27] and the current structural model supports the previous hypothesis that the DE-CE interaction functions to position AS2 and RS2 close together within the genome structure [19]. It should be noted that long-range RNA-RNA interactions involved in regulating sg mRNA transcription are not restricted to tombusviruses. Other plus-strand viruses also require such distal intra-genomic interactions for transcription of their sg mRNAs. Interestingly, the corresponding predicted RNA secondary structures in the genomes of the insect nodavirus Flock House virus [15] and the animal coronavirus transmissible gastroenteritis virus [17] revealed a similar general arrangement for the distant interacting sequences as seen for AS1-RS1 and DE-CE in the TBSV genome model. Specifically, the interacting transcriptional segments were located near the closing ends of RNA domains, suggesting that domain formation which sequesters intervening sequences could represents a common conformational strategy used by different RNA viruses for uniting distal regulatory RNA sequences.
The AS1-RS1 interaction involving the terminal regions of the large domain lD2 is particularly intriguing. When the AS1 element is considered in its local context (i.e. within its directly flanking sequence), it is predicted to be positioned in the loop of a local stem-loop structure (Figure 5A), the formation of which was shown to facilitate sg mRNA1 transcription [42]. Consequently, it was proposed that this stem-loop structure helps to present AS1 in a single-stranded form that promotes its pairing with RS1 [42]. However, in the genomic model, the local AS1-containing stem-loop structure is not predicted and, instead, the sequences flanking AS1 are paired with segments upstream and downstream from RS1 (Figure 4). Accordingly, a sequential pairing pathway can be proposed based on (i) the relative locations of AS1 and RS1 within lD2, (ii) the requirement for the local AS1-containing stem-loop structure for efficient transcription and (iii) the local context of the AS1-RS1 interaction in the structural model. To begin with, the formation of lD2 would act to bring together distally positioned AS1 and RS1 elements (Figure 5A and 5B). The AS1 in the loop of the local stem-loop structure would initiate the interaction by pairing with RS1 (Figure 5B) and, subsequently, the stem would melt and allow those residues to pair with sequences flanking RS1, thereby stabilizing the interaction and forming the closing stems of lD2 (Figure 5C). This pairing scheme is reminiscent of the classical regulatory interaction that takes place in E. coli between Hok mRNA and Sok-antisense-RNA, where the initial contact between the loop in a stem-loop within the mRNA and an unpaired region in the antisense RNA is followed by melting of the stem and formation of additional pairing between the two RNAs [43]. However, in this latter case the interaction is intermolecular, and thus dependent on the concentrations of the two RNAs. In contrast, the AS1-RS1 interaction is intramolecular [18]; therefore, its formation would be heavily influenced by genomic context. In this respect, the scenario in Figure 5 provides an example of how both local and large-scale structures could work collectively to mediate structural rearrangements that lead to a functional long-range RNA-RNA interaction.
The SHAPE data were incorporated into our analysis with the goal of enhancing the accuracy of the structural model. Assessing the correctness of the entire model from a functional standpoint will be challenging, due to its large size and the likelihood that some of its structural components may not confer strong phenotypes. Indeed, certain regions will undoubtedly play more passive roles and thus be tolerant to structural perturbation, while other sections may be very sensitive to alteration. As an initial test of the model, two RNA structures that were predicted to form by SHAPE-plus analysis, but not by SHAPE-minus analysis, were selected for functional investigation. Accordingly, the results from these analyses would provide some feedback as to the effectiveness of the incorporated SHAPE data. The first structure examined was a local extended stem-loop structure, SL27, within mD1 in the p92 coding region (Figures 3 and 4). This structure is conserved within the genus and base pair covariations were present in two species, maize necrotic streak virus (MNeSV) and cucumber Bulgarian latent virus (CBLV), which maintained the lower stem (Figure 6A). To address the relevance of this structure, compensatory mutational analysis was performed, where the lower stem was destabilized and then restored with nucleotide substitutions. Substitutions were designed so as to maintain the amino acid identity of the encoded p92 ORF and, as mutant S27A contained two less GU base pairs than S27B, the former was predicted to be less stable than the latter (Figure 6A). In S27C, the mutations in S27A and S27B were combined to re-establish canonical base pairing at all substituted positions. When the mutant TBSV genomes containing these modifications were individually transfected into plant protoplasts, only minor differences in accumulation levels were observed (Figure 6B). However, when these genomes were co-transfected with a smaller competitor viral replicon, a TBSV genome with a single internal deletion that removed the CP and p19/22 ORFs, termed RTD-23 [42], clear differences in accumulation were evident (Figure 6C). In particular, the most destabilized mutant, S27A, showed a decrease in accumulation to ∼47%, while the relative level of the RTD-23 replicon increased to ∼190%, when compared to cotransfection with the wt TBSV genome (Figure 6C). In contrast, the S27B and S27C mutants with higher predicted stability showed genome and replicon accumulation profiles more similar to those of the wt TBSV and RTD-23 coinfection (Figure 6C). The observation that S27A with the most destabilized structure was least able to compete effectively against the replicon suggests that this structure confers a fitness advantage at the single cell level and is functionally relevant to the virus.
Next we examined a longer-range 13 base pair long interaction spanning an intervening sequence of 190 nts, which also was predicted only by SHAPE-plus analysis. This interaction formed a helix, S31, that closed mD2 (Figures 3 and 4) and was of particular interest because the intervening sequence included two important RNA replication elements, the large stem-loop structure RII(+)-SL and the UL sequence (Figure 7A). Two species in the genus, MNeSV and CBLV, contained covariations that maintained S31 but, surprisingly, the substitutions in the 5′ half of this stem were not silent and resulted in two amino acid substitutions (Gly390Ala and Arg391Ser). This finding suggests that, at these positions, maintaining this RNA structure was a higher priority than preserving the consensus amino acid identity in p92. To test the importance of this potential interaction, substitutions were introduced into the complementary sequences of S31 in the TBSV genome. Substitutions were made at degenerate codon positions so as not to change the corresponding amino acids in the p92 ORF. As no directly opposing degenerate codon positions were available, the introduction of silent compensatory mutations was not possible in the genomic context. Instead, two genomic mutants were made, TCL1 and TCL2, which were predicted to either disrupt or stabilize the interaction, respectively (Figure 7B). TCL1 did not accumulate in transfected protoplasts, whereas TCL2 accumulated to greater than wt levels (Figure 7B), consistent with the interaction being functionally relevant.
The location of the interaction close to two RNA replication-related elements (i.e. RII(+)-SL and UL) suggested a possible role in mediating genome replication. To address this question in a replication context free from translational constraints, a non-coding defective interfering (DI) RNA replicon containing the interaction was used. The DI RNA, termed DICL, was able to replicate efficiently in co-inoculations with helper T100 (which provided p33 and p92 RNA replication proteins in trans) and, as this replicon was non-coding, it was possible to generate a full set of compensatory mutations within it to test the role of S31 (Figure 7C). DICL-A and DICL-B had disruptions in either half of S31 and accumulated to 16% and 31% the levels of wt DICL (Figure 7C). If these two defects were independent, then combining them would be additive and would lead to a predicted ∼5% level of accumulation. However, when the two mutant halves were united in DICL-C, so as to restore base pairing potential, an accumulation level 55% that of wt was observed (Figure 7C), indicating an important role for S31 formation in the accumulation of the DI RNA. Its requirement in a non-coding replicon suggests that the function of S31 is related to regulating viral genome replication, possibly through modulating the activity of the replication elements in the intervening sequence. The apparent phenotype observed for disruption of S31 contrasts the more subtle differences seen for S27 mutants, however both results help to functionally validate components of the SHAPE-plus model and provide a sampling of the range of phenotypes that can be expected in future structure-function analyses of the model.
As mentioned previously, the TBSV genome structure is intimately involved in the fundamental viral processes of protein translation, genome replication, and sg mRNA transcription. Naturally, the structural knowledge gained in this study will be useful to design additional experiments to further explore these basic processes. However, genome structure is also relevant to several other aspects of viral infections. One such area is virus-induced gene silencing (VIGS), an antiviral host defence system that is based on the detection and destruction of viral RNA [44], [45]. Detection is mediated by the enzyme Dicer, which binds to and cleaves double-stranded viral RNA into small interfering RNAs (siRNAs) that are used as guides by the RNA-induced silencing complex (RISC) to target and cleave viral genomes [44], [45]. Possible targets for Dicer include viral double-stranded RNA replication intermediates [46] and highly structured regions of viral RNA genomes [47]–[49]. With respect to the latter possibility, our current structural model predicts several regions with extended uninterrupted helices, which represent possible targets for Dicer. Although in vivo structures may differ somewhat from our model, it represents a good starting point for mapping and correlating Dicer hotspots with structured regions. Our model would also be useful for in vitro studies using plant-derived Dicer extracts [50], [51] along with in vitro transcripts of the viral genome, which would allow for direct assessment of substrate preferences. Additionally, since the efficiency of RISC-based cleavage of a target RNA is known to be inhibited by secondary structure in the target sequence [52], [53], knowing which regions are less structured and, thus, better potential targets, would be beneficial for designing RNAi-based antiviral measures.
One of the driving forces in RNA virus evolution is recombination, in which discontinuous regions of a viral genome(s) are joined together [54]. The process of recombination has been studied in many RNA viruses [54], and in tombusviruses the deletion events leading to the formation of small DI RNAs have been used as a model to understand RNA recombination [55]. These deletions presumably occur when the viral polymerase dissociates from its template along with its nascent strand and then rebinds some distance upstream, where it continues nascent strand extension. In vitro and in vivo studies in tombusviruses have provided evidence that discontinuous copying can occur during genomic minus-strand synthesis [56], [57]. In such cases, the viral genome would act as the template and in our structural model the four regions maintained in a prototypical DI RNA are all positioned relatively close to each other (Figure 4). Local sequences and structures are known to mediate polymerase dissociation and/or rebinding [58], [59], however the proximity of these take-off and landing sites, as determined by the global structure, could also aid in this process. Accordingly, global structure is a likely determinant of intramolecular recombination events leading to the generation of deletions in genomic RNA.
For RNA virus characterization, a common strategy used to define the role of a viral protein is to mutate its corresponding ORF in the infectious clone. However, it is becoming increasingly evident that functional RNA elements also reside within coding regions of different RNA viruses [1], [2]. Thus, the modification of a coding region could also alter an overlapping functional RNA structure, which would greatly complicate the interpretation of results. In such cases, a reliable genomic structural model that includes local and long-range interactions would be useful, as this information could be integrated into the interpretation of data to help avoid erroneous conclusions. Additionally, the ability to manipulate infectious transcripts of RNA virus genomes has also allowed for their development as foreign protein expression vectors [60]. Challenges to this application include low levels of vector replication and/or deletion of foreign inserts, both of which lead to low protein yield. Based on our model for TBSV, we can now start to appreciate how insertion of foreign sequences or replacement of viral sequences could lead to disruption of one or more of the components of the global structure. As a result, structural models such as that for the TBSV genome should prove useful in designing viral vectors that are compatible with the natural organization and function of viral genomes.
With respect to the reliability of our SHAPE-guided structure, we feel that (i) the general agreement with the AFM structure, (ii) the presence of known local structures, (iii) the occurrence of some of the known long-range interactions, and (iv) identification of new structures that were functionally validated, all add confidence to the prediction. Nonetheless, the accuracy of many aspects of the structure still remains to be investigated. The current structural model provides a context to begin to understand the organization of local structures and long-range RNA-RNA interactions within the TBSV genome. At a global level, the genome assumes a floret-like structure that includes multiple long-range RNA-RNA interactions. The simultaneous presence of the AS1-RS1 and DE-CE interactions in our model suggests that transcriptional-related interactions are structurally compatible and points to functional coupling of this process. Conversely, the absence of other known long-distance interactions in the structure implies the requirement for dynamic structural transitions, some of which may be mutually exclusive. However, the basic framework of secondary structure observed may mediate the formation of alternate structures without the need for large-scale rearrangements. Importantly, our structure provides a foundation to further test and refine the model and to identify alternative global structures that allow other long-range interactions to occur. Such dynamic transitions may occur readily in a cellular environment that includes both viral and host proteins that could mediate structural rearrangements. A possible candidate for this role is the abundant viral protein p33, which has been shown to possess RNA chaperone activity [61]. Future structural studies will need to investigate different experimental conditions in order to develop a comprehensive understanding of all relevant genomic conformations. From a comparative perspective, as the structures of additional RNA viral genomes are determined, it will be interesting to see if related viruses (e.g. different genera in the same family) display any likeness at the global genomic level and if unrelated viruses share similar organizational features that can help to define general rules of arrangement.
The infectious clone of the wild-type TBSV genome, T100 [22], the replicon RTD-23 [42] and TBSV DI RNAs [55] have been described previously. Mutant TBSV genomes were constructed based on T100, where modifications were introduced using PCR-based mutagenesis and standard cloning techniques. The PCR-derived regions in all constructs were sequenced completely to ensure that only the intended modifications were present. The modifications introduced into the mutants are shown in the accompanying figures.
For AFM imaging, TBSV RNA genome derived from in vitro transcription of clone T100 was diluted to 1.5 ng/µl in deposition buffer (20 mM HEPES, pH 7.08, 10 mM MgCl2). A 20 µl volume of the diluted RNA sample was deposited on freshly cleaved mica and incubated for 1 min at room temperature. Following the incubation period, the sample was rinsed with double distilled water, dried under a stream of nitrogen gas, and imaged using AFM tapping mode in air. Tapping mode AFM was performed with a Dimension 3100 microscope and Nanoscope IIIa controller (Digital Instrument Inc., Vecco). Tapping mode AFM images in air were obtained using a silicon cantilever-tip assembly with a resonance frequency of 200–300 kHz. The images were captured at scan rate of 1–1.5 Hz and 512×512 pixels resolution. Image height scale was typically of the order of ca. 3 nm; exact scales are provided for each image. Images were viewed and analyzed using V614r1 Nanoscope program (Digital Instrument Inc.). Images were subjected to third order flattening to remove image bowing artifacts.
Viral RNAs were generated by in vitro transcription using T7 RNA polymerase as described previously [55]. Protoplasts (∼300,000) derived from cucumber cotyledons were transfected with viral RNA transcripts (3 µg for genomic RNA; 1 µg for replicon or DI RNA) using polyethylene glycol [18] and incubated at 22°C or 28°C for 22 hr. Total nucleic acids were extracted from transfected protoplasts using phenol/chloroform. Following ethanol precipitation they were separated in 1.4% agarose gels and subjected to Northern blot analysis as described previously [18]. Equal loading for all samples was confirmed prior to transfer via staining the gels with ethidium bromide. Viral RNAs were detected using strand-specific 32P-labeled probes and relative isotopic levels were determined using PharosFx Plus Molecular Imager.
High-throughput SHAPE analysis was carried out using the methodology described by Low and Weeks [33]. TBSV genomic RNA was prepared by in vitro transcription using AmpliScribe T7-Flash Transcription Kits, and subsequently purified by two rounds of ammonium acetate precipitation. Approximately 70% of the genomic TBSV RNA was determined to be intact, as estimated by ethidium bromide staining after gel electrophoresis. Eighty pmol of TBSV RNA in 480 µL of 0.5× TE buffer (5 mM Tris, pH 8, 0.5 mM EDTA) was heated for 5 min at 95°C and then transferred to ice for 2 min. Subsequently, 240 µL of 3.3× folding buffer (333 mM HEPES pH 8.0, 16.5 mM MgCl2, 333 mM NaCl) was added, followed by incubation at 37°C for 30 min.
Three hundred and sixty µL of the folded RNA was treated with 40 µL of 50 mM 1-methyl-7-nitroisatoic anhydride (1M7) in DMSO and another 360 µL aliquot of the folded RNA was treated with 40 µL neat DMSO. Both were subsequently incubated at 37°C for 4 min. 1M7-modified RNA (+), and DMSO control RNA (−) were recovered by ethanol precipitation (with 200 mM NaCl, 2 mM EDTA, and 800 µg glycogen) and resuspended in 400 µL of 0.5× TE.
TBSV genomic RNA secondary structure was predicted by Mfold RNA folding software [62] and complementary DNA primers were designed for regions predicted to lack strong RNA secondary structure. In total, 19 primers were designed to cover the TBSV genome (Table S3) with an average coverage distance of ∼360 nt per primer (Table S4). Each primer was separately synthesized with 4 different florescent dyes (WellRED D2, WellRED D3, WellRED D4, and LICOR IR 800) at the 5′end and purified by reverse-phase cartridge purification method (Sigma-Genosys).
Fluorescently-labeled primer (6 µL at 4 µM) was added to 20 µL RNA (+) (4 µM WellRED D4) and 20 µL RNA (−) (4 µM WellRED D3) reactions and incubated at 65°C for 5 min and 37°C for 5 min. Twelve µL SHAPE enzyme mix (167 mM Tris-HCl, pH 8.3, 250 mM KCl, 10 mM MgCl2 16.7 mM DTT, 1.67 mM dATP, 1.67 mM dTTP, 1.67 mM dCTP, 1.67 mM dITP) was added to each tube and incubated at 52°C for 2 min, to which 1.5 µL of SuperScript III (Invitrogen) was added and incubated at 52°C for 30 min. The extension reaction was stopped by incubating on ice for 2 min and then adding 4 µL of 3 M NaOAc (pH 5.2). For sequencing reactions, 7.2 pmol RNA in 40 µL of 0.5× TE was incubated at 95°C for 4 min and transferred to ice for 2 min to denature the RNA. Six µL primer (4 µM, WellRED D2) was added to 20 µL denatured RNA (Seq-G) and 6 µL primer (4 µM, LICOR IR 800) was added to another aliquot of 20 µL denatured RNA (Seq-T). After primer and RNA annealing at 65°C for 5 min and then 37°C for 5 min, 2 µL of ddGTP (0.5 mM) and 2 µL ddTTP (10 mM) were added to Seq-G and Seq-T, respectively, and the rest of the procedure was as described above for the 1M7-treated and untreated samples. The four reactions were combined and precipitated with ethanol and 2 µL of glycogen (20 mg/ml). The cDNA pellet was washed twice with 70% ethanol, dried, and resuspended in 40 µL of deionized formamide.
cDNA samples were separated in a 33 cm long (75 µm inner diameter) capillary using a Beckman CEQ800 DNA sequencer. Samples were denatured at 90°C for 120 seconds, injected at 2.0 kV for 6 to 15 sec, and separated at 4 kV for 80 min.
SHAPE electropherogram intensities were quantified using SHAPEfinder [63]. Briefly, 1) The hSHAPE data, as esd files, were loaded from the capillary electrophoresis instrument. 2) Using the base line adjustment tool, drifting baselines in each channel were corrected, with a window width of 10 times the inter-peak distance. All the channels had a common baseline. 3) The matrixing tool was used for color separation to make sure that each channel represents one dye amount, and not fluorescent intensity from the other three dyes. 4) The differences in electrophoretic migration rate of the cDNA products caused by different fluorophores were corrected using cubic mobility shift tool. 5) Because of the imperfect processivity of the reverse transcriptase, signals decay from left to right in hSHAPE electropherograms were corrected using the signal decay correction tool. Normally, the entire area of interpretable data was selected for the correction. 6) The scale of each channel in a trace was adjusted using the scale factor tool. For accurate subtraction of background intensities, corresponding to the small peak positions in the (+) channel, the peak sizes in the (−) channel were adjusted to the same intensities as the peaks in the (+) channel. 7) The align and integrate tool was used to calculate and align nucleotide activity to the RNA sequence. There are three phases in this tool. First, in the “setup” phase, every SHAPE reaction was assigned to a channel; the sequencing peak sensitivity was adjusted; the data range to be analyzed was specified; the RNA sequence file corresponding to the hSHAPE experiment was loaded. Second, in “modify” phase, the sequence offset created in the “setup” phase by missing and adding a peak was corrected by adding and deleting a peak. Finally, in the “fit” phase, the reactivity of each nucleotide was calculated. Reactivities were assigned to 4347 nts (∼91% of the genome). Due to blocks in reverse transcription, SHAPE information for the contiguous regions 1008–1163 nts and 2464–2629 nts could not be collected. Quality of SHAPE data of 9 nts at the 5′ end and 83 nts at 3′ end was not collected due to the low quality data at 5′ end and the absence of data at the 3′end. Fifteen nts throughout the entire genome had low quality data (e.g. peaks in 4 channels at one nucleotide position were high). Accordingly, SHAPE data for these regions or nts (colored in grey, Figure 3; Table S1) were not included in the computational folding (i.e. structural prediction in these regions was based only on thermodynamics.) For the rest of the genome, the raw SHAPE reactivities were normalized using the box-plot normalization method [63] in these steps: 1) All data lying outside 1.5 times the interquartile range were identified as outliers; 2) the normalization factor was calculated as the average of the top 10% in the remaining data; and 3) all data (including outliers) were then multiplied by the reverse of the normalization factor. 4) These data were smoothed by using the sigmoid functioni.e. the normalized reactivity was set to = 1 if t was equal to the average of the top 10% non-outlier data. These preprocessing steps were carried out in MATLAB.
|
10.1371/journal.pbio.1000141 | Time-Warp–Invariant Neuronal Processing | Fluctuations in the temporal durations of sensory signals constitute a major source of variability within natural stimulus ensembles. The neuronal mechanisms through which sensory systems can stabilize perception against such fluctuations are largely unknown. An intriguing instantiation of such robustness occurs in human speech perception, which relies critically on temporal acoustic cues that are embedded in signals with highly variable duration. Across different instances of natural speech, auditory cues can undergo temporal warping that ranges from 2-fold compression to 2-fold dilation without significant perceptual impairment. Here, we report that time-warp–invariant neuronal processing can be subserved by the shunting action of synaptic conductances that automatically rescales the effective integration time of postsynaptic neurons. We propose a novel spike-based learning rule for synaptic conductances that adjusts the degree of synaptic shunting to the temporal processing requirements of a given task. Applying this general biophysical mechanism to the example of speech processing, we propose a neuronal network model for time-warp–invariant word discrimination and demonstrate its excellent performance on a standard benchmark speech-recognition task. Our results demonstrate the important functional role of synaptic conductances in spike-based neuronal information processing and learning. The biophysics of temporal integration at neuronal membranes can endow sensory pathways with powerful time-warp–invariant computational capabilities.
| The brain has a robust ability to process sensory stimuli, even when those stimuli are warped in time. The most prominent example of such perceptual robustness occurs in speech communication. Rates of speech can be highly variable both within and across speakers, yet our perceptions of words remain stable. The neuronal mechanisms that subserve invariance to time warping without compromising our ability to discriminate between fine temporal cues have puzzled neuroscientists for several decades. Here, we describe a cellular process whereby auditory neurons recalibrate, on the fly, their perceptual clocks and allows them effectively to correct for temporal fluctuations in the rate of incoming sensory events. We demonstrate that this basic biophysical mechanism allows simple neural architectures to solve a standard benchmark speech-recognition task with near perfect performance. This proposed mechanism for time-warp–invariant neural processing leads to novel hypotheses about the origin of speech perception pathologies.
| Robustness of neuronal information processing to temporal warping of natural stimuli poses a difficult computational challenge to the brain [1]–[9]. This is particularly true for auditory stimuli, which often carry perceptually relevant information in fine differences between temporal cues [10],[11]. For instance in speech, perceptual discriminations between consonants often rely on differences in voice onset times, burst durations, or durations of spectral transitions [12],[13]. A striking feature of human performance on such tasks is that it is resilient to a large temporal variability in the absolute timing of these cues. Specifically, changes in speaking rate in ongoing natural speech introduce temporal warping of the acoustic signal on a scale of hundreds of milliseconds, encompassing temporal distortions of acoustic cues that range from 2-fold compression to 2-fold dilation [14],[15]. Figure 1 shows examples of time warp in natural speech. The utterance of the word “one” in (A) is compressed by nearly a factor of one-half relative to the utterance shown in (B), causing a concomitant compression in the duration of prominent spectral features, such as the transitions of the peaks in the frequency spectra. Notably, the pattern of temporal warping in speech can vary within a single utterance on a scale of hundreds of milliseconds. For example, the local time warp of the word “eight” in (C) relative to (D), reverses from compression in the initial and final segments to strong dilation of the gap between them. Although it has long been demonstrated that speech perception in humans normalizes durations of temporal cues to the rate of speech [2],[16]–[18], the neural mechanisms underlying this perceptual constancy have remained mysterious.
A general solution of the time-warp problem is to undo stimulus rate variations by comodulating the internal “perceptual” clock of a sensory processing system. This clock should run slowly when the rate of the incoming signal is low and embedded temporal cues are dilated, but accelerate when the rate is fast and the temporal cues are compressed. Here, we propose a neural implementation of this solution, exploiting a basic biophysical property of synaptic inputs, namely, that in addition to charging the postsynaptic neuronal membrane, synaptic conductances modulate its effective time constant. To utilize this mechanism for time-warp robust information processing in the context of a particular perceptual task, synaptic peak conductances at the site of temporal cue integration need to be adjusted to match the range of incoming spike rates. We show that such adjustments can be achieved by a novel conductance-based supervised learning rule. We first demonstrate the computational power of the proposed mechanism by testing our neuron model on a synthetic instantiation of a generic time-warp–invariant neuronal computation, namely, time-warp–invariant classification of random spike latency patterns. We then present a novel neuronal network model for word recognition and show that it yields excellent performance on a benchmark speech-recognition task, comparable to that achieved by highly elaborate, biologically implausible state-of-the-art speech-recognition algorithms.
Whereas the net current flow into a neuron is determined by the balance between excitatory and inhibitory synaptic inputs, both types of inputs increase the total synaptic conductance, which in turn modulates the effective integration time of the postsynaptic cell [19]–[21] (an effect known as synaptic shunting). Specifically, when the total synaptic conductance of a neuron is large relative to the resting conductance (leak) and is generated by linear summation of incoming synaptic events, the neuron's effective integration time scales inversely to the rate of inputs spikes. Hence, the shunting action of synaptic conductances can counter variations in afferent spike rates by automatically rescaling the effective integration time of the postsynaptic neuron.
We implement this mechanism in a leaky integrate-and-fire model neuron driven by N exponentially decaying synaptic conductances . Here, denotes the peak conductance of the ith synapse in units of sec−1, and τs is the synaptic time constant. The total synaptic current, measured at rest, is given bywhere denotes the reversal potential of the ith synapse relative to resting potential and ti denote the arrival times of the spikes of the ith afferent. The factor β denotes a global scaling of all incoming spike times; β = 1 is the unwarped inputs. The total synaptic conductance, Gsyn(t,β), is
For fast synapses, the total synaptic current is essentially a train of pulses, each of which occurs at the time of an incoming spike and delivers a total charge of . Changing the rate of the incoming spikes will induce a corresponding change in the timing of these pulses but not their charge. Therefore, ignoring the effect of time warp on the time scale of τs, which is short relative to the time scale of voltage modulations, the total synaptic current obeys the following time-warp scaling relation, Isyn(βt,β) = β−1Isyn(t,1). A similar scaling relation holds for the total synaptic conductance. The evolution in time of the subthreshold voltage is given by(1)
Thus, V integrates the synaptic current with an effective time constant whose inverse is 1/τeff = gleak+Gsyn(t,β). If the contribution of Gsyn is significantly larger than the leak conductance, then 1/τeff is rescaled by time-warp similar to Gsyn and Isyn, and, hence, the solution of Equation 1 is approximately time-warp invariant, namely, V(βt,β) = V(t,1). This result is illustrated in Figure 2, which compares the voltage traces induced by a random spike pattern for β = 1 and β = 0.5.
To perform time-warp–invariant tasks, peak synaptic conductances must be in the range of values appropriate for the statistics of the stimulus ensemble of the given task. To achieve this, we have devised a novel spike-based learning rule for synaptic conductances, the conductance-based tempotron. This model neuron learns to discriminate between two classes of spatiotemporal input spike patterns. The tempotron's classification rule requires it to fire at least one spike in response to each of its target stimuli but to remain silent when driven by a stimulus from the null class. Spike patterns from both classes are iteratively presented to the neuron, and peak synaptic conductances are modified after each error trial by an amount proportional to their contribution to the maximum value of the postsynaptic potential over time (see Materials and Methods). This contribution is sensitive to the time courses of the total conductance and voltage of the postsynaptic neuron. Therefore, the conductance-based tempotron learns to adjust, not only the magnitude of the synaptic inputs, but also its effective integration time to the statistics of the task at hand.
We first quantified the time-warp robustness of the conductance-based tempotron on a synthetic discrimination task. We randomly assigned 1,250 spike pattern templates to target and null classes. The templates consisted of 500 afferents, each firing once at a fixed time chosen randomly from a uniform distribution between 0 and 500 ms. Upon each presentation during training and testing, the templates underwent global temporal warping by a random factor β ranging from compression by 1/βmax to dilation by βmax (see Materials and Methods). Consistent with the psychophysical range, βmax was varied between 1 and 2.5. Remarkably, with physiologically plausible parameters, the error frequency remained almost zero up to βmax≈2 (Figure 3A, blue curve). Importantly, the performance of the conductance-based tempotron showed little change when the temporal warping applied to the spike templates was dynamic (see Materials and Methods) (Figure 3A). The time-warp robustness of the neural classification depends on the resting membrane time constant τm and the synaptic time constant τs. Increases in τm or decreases in τs both enhance the dominance of shunting in governing the cell's effective time constant. As a result, the performance for βmax = 2.5 improved with increasing τm (Figure 3B, left) and decreasing τs (Figure 3B, right). The time-warp robustness of the conductance-based tempotron was also reflected in the shape of its subthreshold voltage traces (Figure 3C, top row) and generalized to novel spike templates with the same input statistics that were not used during training (Figure 3C, second row).
Synaptic conductances were crucial in generating the neuron's robustness to temporal warping. Athough an analogous neuron model with a fixed integration time, the current-based tempotron [22] (see Materials and Methods) also performed the task perfectly in the absence of time-warp (βmax = 1); its error frequency was sensitive even to modest temporal warping and deteriorated further when the applied time warp was dynamic (Figure 3A, red curve). Similarly, the voltage traces of this current-based neuron showed strong dependence on the degree of temporal warping applied to an input spike train (Figure 3C, bottom trace pair). Finally, the error frequency of the current-based neuron at βmax = 2.5 showed only negligible improvement upon varying the values of the membrane and synaptic time constants (Figure 3B), highlighting the limited capabilities of fixed neural kinetics to subserve time-warp–invariant spike-pattern classification.
Note that in the present classification task, the degree of time-warp robustness depends also on the learning load, i.e., number of patterns that have to classified by a neuron (unpublished data). A given degree of time warp translates into a finite range of distortions of the intracellular voltage traces. If these distortions remain smaller than the margins separating the neuronal firing threshold and the intracellular peak voltages, a neuron's classification will be time-warp invariant. Since the maximal possible margins increase with decreasing learning load, time-warp invariance can be traded for storage capacity. This tradeoff is governed by the susceptibility of the voltage traces to time warp. If the susceptibility is high, as in the current-based tempotron, robustness to time warp comes at the expense of a substantial reduction in storage capacity. If it is low, as in the conductance-based tempotron, time-warp invariance can be achieved even when operating close to the neuron's maximal storage capacity for unwarped patterns.
In the conductance-based tempotron, synaptic conductances controlled, not only the effective integration time of the neuron, but also the temporal selectivity of the synaptic update during learning. The tempotron learning rule modifies only the efficacies of the synapses that were activated in a temporal window prior to the peak in the postsynaptic voltage trace. However, the width of this temporal plasticity window is not fixed but depends on the effective integration time of the postsynaptic neuron at the time of each synaptic update trial, which in turn varies with the input firing rate at each trial and the strength of the peak synaptic conductances at this stage of learning (Figure 4). During epochs of high conductance (warm colors), only synapses that fired shortly before the voltage maximum were appreciably modified. In contrast, when the membrane conductance was low (cool colors), the plasticity window was broad. The ability of the plasticity window to adjust to the effective time constant of the postsynaptic voltage is crucial for the success of the learning. As is evident from Figure 4, the membrane's effective time constant varies considerably during the learning epochs; hence, a plasticity rule that does not take this into account fails to credit appropriately the different synapses.
The evolution of synaptic peak conductances during learning was driven by task requirements. When we replaced the temporal warping of the spike templates by random Gaussian jitter [22] (see Materials and Methods), conductance-based tempotrons that had acquired high synaptic peak conductances during initial training on the time-warp task readjusted their synaptic peak conductances to low values (Figure 5, inset). The concomitant increase in their effective integration time constants from roughly 10 ms to 50 ms improved the neurons' ability to average out the temporal spike jitter and substantially enhanced their task performance (Figure 5).
To address time-warp–invariant speech processing, we studied a neuronal module that learns to perform word-recognition tasks. Our model consists of two auditory processing stages. The first stage (Figure 6) consists of an afferent population of neurons that convert incoming acoustic signals into spike patterns by encoding the occurrences of elementary spectrotemporal events. This layer forms a 2-dimensional tonotopy-intensity auditory map. Each of its afferents generates spikes by performing an onset or offset threshold operation on the power of the acoustic signal in a given frequency band. Whereas an onset afferent elicits a spike whenever the log signal power crosses its threshold level from below, offset afferents encode the occurrences of downward crossings (see Materials and Methods) (cf. also [6],[23]). Different on and off neurons coding for the same frequency band differ in their threshold value, reflecting a systematic variation in their intensity tuning. The second, downstream, layer consists of neurons with plastic synaptic peak conductances that are governed by the conductance-based tempotron plasticity rule. These neurons are trained to perform word discrimination tasks. We tested this model on a digit-recognition benchmark task with the TI46 database [24]. We trained each of the 20 conductance-based tempotrons of the second layer to perform a distinct gender-specific binary classification, requiring it to fire in response to utterances of one digit and speaker gender, and to remain quiescent for all other stimuli. After training, the majority of these digit detector neurons (70%) achieved perfect classification of the test set, and the remaining ones performed their task with a low error (Table 1). Based on the spiking activity of this small population of digit detector neurons, a full digit classifier (see Materials and Methods) that weighted spikes according to each detector's individual performance, achieved an overall word error rate of 0.0017. This performance matches the error rates of state-of-the-art artificial speech-recognition systems such as the Hidden Markov model–based Sphinx-4 and HTK, which yield error rates of 0.0017 [25] and 0.0012 [26], respectively, on the same benchmark.
To reveal qualitatively some of the mechanisms used by our digit detector neurons to selectively detect their target word, we compared the learned synaptic distributions (Figure 7A) of two digit detector neurons (“one” and “four”) to the average spectrograms of each neuron's target stimuli aligned to the times of its output spikes (Figure 7B; see Materials and Methods). The spectrotemporal features that preceeded the output spikes (time zero, grey vertical lines) corresponded to the frequency-specific onset and offset selectivity of the excitatory afferents (Figure 7A, warm colors). These examples demonstrate how the patterned excitatory and inhibitory inputs from both onset and offset neurons are tuned to features of the speech signal. For instance, a prominent feature in the averaged spectrogram of the word “one” (male speakers) was the increase in onset time of the power in the low-frequency channels with the frequency of the channel (Figure 7B, left, channels 1–16). This gradual onset was encoded by a diagonal band of excitatory onset afferents whose thresholds decreased with increasing frequency (Figure 7A, left). By compensating for the temporal lag between the different lower-frequency channels, this arrangement ensured a strong excitatory drive when a target stimulus was presented to the neuron. The spectrotemporal feature learned by the word “four” (male speakers) detector neuron combined decreasing power in the low-frequency range with rising power in the mid-frequency range (Figure 7B, right). This feature was encoded by synaptic efficacies through a combination of excitatory offset afferents in the low-frequency range (Figure 7A, right, channels 1–11) and excitatory onset afferents in the mid-frequency range (channels 12–19). Excitatory synaptic populations were complemented by inhibitory inputs (Figure 7A, blue patches) that prevented spiking in response to null stimuli and also increased the total synaptic conductance. The substantial differences between the mean spike-triggered voltage traces for target stimuli (Figure 7C, blue) and the mean maximum-triggered voltage traces for null stimuli (red) underline the high target word selectivity of the learned synaptic distributions as well as the relatively short temporal extend of the learned target features.
In the examples shown, the average position of the neural decision relative to the target stimuli varied from early to late (Figure 7B, left vs. right). This important degree of freedom stems from the fact that the tempotron decision rule does not constrain the time of the neural decision. As a result, the learning process in each neuron can select the spectrotemporal target features from any time window within the word. The selection of the target feature by the learning takes into account both the requirement of triggering output spikes in response to target stimuli as well as the demand to remain silent during null stimuli. Thus, for each target neuron, the selected features reflect the statistics of both the target and the null stimuli.
We have performed several tests designed to assess the ability of the model word detector neurons to perform well on new input sets, different in statistics from the trained database. First, we assessed the ability of the neurons to generalize to unfamiliar speakers and dialects. After training the model with the TI46 database, as described above, we measured its digit-recognition performance on utterances from another database, the TIDIGITS database [27], which includes speech samples from a variety of English dialects (see Materials and Methods). This test has been done without any retraining of the network synapses. The resulting word error rate of 0.0949 compares favorably to the performance of the HTK system, which resulted in an error rate of 0.2156 when subjected to the same generalization test (see Materials and Methods). Across all dialects, our model performed perfectly for roughly one-quarter of all speakers and with at most one error for half of them. Within the best dialect group, an error of at most one word was achieved for as many as 80% of the speakers (Table S1). These results underline the ability of our neuronal word-recognition model to generalize to unfamiliar speakers across a wide range of different unfamiliar dialects.
An interesting question is whether our model neurons are able to generalize their performance to novel time-warped versions of the trained inputs. To address this question, we have tested their performance on randomly generated time-warped versions of the input spikes corresponding to the trained word utterances, without retraining. As shown in Figure 8, the neurons exhibited considerable time-warp–robust performance on the digit-recognition task. For instance, the errors for the “one” (Figure 8A, black line) and “four” (blue line) detector neurons (cf. Figure 7) were insensitive to a 2-fold time warp of the input spike trains. The “seven” detector neuron (male, red line) showed higher sensitivity to such warping; nevertheless, its error rate remained low. Consistent with the proposed role of synaptic conductances, the degree of time-warp robustness was correlated with the total synaptic conductance, here quantified through the mean effective integration time τeff (Figure 8B). Additionally, the mean voltage traces induced by the target stimuli (Figure 8C, lower traces) showed a substantially smaller sensitivity to temporal warping than their current-based analogs (see Materials and Methods) (Figure 8C, upper traces).
We also found that our model word detector neurons are robust to the introduction of spike failures in their input patterns. For each neuron, we have measured its performance on inputs which were corrupted by randomly deleting a fraction of the incoming spikes, again without retraining. For the majority of neurons, the error percentage increased by less than 0.01% for each percent increase in spike failures (Figure 9). This high robustness reflects the fact that each classification is based on integrating information from many presynaptic sources.
The proposed conductance-based time-rescaling mechanism is based on the biophysical property of neurons that their effective integration time is shaped by synaptic conductances and therefore can be modulated by the firing rate of its afferents. To utilize these modulations for time-warp–invariant processing, a central requirement is a large evoked total synaptic conductance that dominates the effective integration time constant of the postsynaptic cell through shunting. In our speech-processing model, large synaptic conductances with a median value of a 3-fold leak conductance across all digit detector neurons (cf. Figure 8B) result from a combination of excitatory and inhibitory inputs. This is consistent with high total synaptic conductances, comprising excitation and inhibition, that have been observed in several regions of cortex [28] including auditory [29],[30], visual [31],[32], and also prefrontal [33],[34] (but see ref. [35]). Our model predicts that in cortical sensory areas, the time-rescaled intracellular voltage traces (cf. Figure 3C), and consequently, also the rescaled spiking responses of neurons that operate in the proposed fashion, remain invariant under temporal warping of the neurons' input spike patterns. These predictions can be tested by intra- and extracellular recordings of neuronal responses to temporally warped sensory stimuli.
A large total synaptic conductance is associated with a substantial reduction in a neuron's effective integration time relative to its resting value. Therefore, the resting membrane time constant of a neuron that implements the automatic time-rescaling mechanism must substantially exceed the temporal resolution that is required by a given processing task. Because the word-recognition benchmark task used here comprises whole-word stimuli that favored effective time constants on the order of several tens of milliseconds, we used a resting membrane time constant of τm = 100 ms. Whereas values of this order have been reported in hippocampus [36] and cerebellum [21],[37], it exceeds current estimates for neocortical neurons, which range between 10 and 30 ms [35],[38],[39]. Note, however, that the correspondence of our passive membrane model and the experimental values that typically include contributions from various voltage-dependent conductances is not straightforward. Our model predicts that neurons specialized for time-warp–invariant processing at the whole-word level have relatively long resting membrane time constants. It is likely that the auditory system solves the problem of time-warp–invariant processing of the sound signal primarily at the level of shorter speech segments such as phonemes. This is supported by evidence that primary auditory cortex has a special role in speech processing at a resolution of milliseconds to tens of milliseconds [11]–[13]. Our mechanism would enable time-warp–invariant processing of phonetic segments with resting membrane time constants in the range of tens of milliseconds, and much shorter effective integration times.
The proposed neuronal time-rescaling mechanism assumes linear summation of synaptic conductances. This assumption is challenged by the presence of voltage-dependent conductances in neuronal membranes. Since the potential implications for our model depend on the specific nonlinearity induced by a cell-type–specific composition of different ionic channels, it is hard to evaluate the overall effect on our model in general terms. Nevertheless, because of its immanence, we expect the conductance-based time-rescaling mechanism to cope gracefully with moderate levels of nonlinearity. As an example, we tested its behavior in the presence of an h-like conductance (see Materials and Methods) that opposes conductance changes induced by depolarizing excitatory synaptic inputs and is active at the resting potential. As expected, we found that physiological levels of h-conductances resulted in only moderate impairment of the automatic time-rescaling mechanism (Figure S1).
For the sake of simplicity as well as numerical efficiency, we have assumed symmetric roles of excitation and inhibition in our model architecture. We have checked that this assumption is not crucial for the operation of the automatic time-rescaling mechanism and the learning of time-warped random latency patterns. Specifically, we have implemented the random latency classification task for a control architecture in which all synapses were confined to be excitatory except a single global inhibitory input that, mimicking a global inhibitory network, received a separate copy of all incoming spikes. In this architecture, all spike patterns have to be encoded by the excitatory synaptic population, and the role of inhibition is reduced to a global signal that has equal strength for all input patterns. Due to the limitations of this architecture, this model showed some reduction of storage capacity relative to the symmetric case, but the automatic time-rescaling mechanism remained intact. For a time-warp scale of βmax = 2.5 (cf. Figure 3), the global inhibition model roughly matched the performance of the symmetric model when the learning load was lowered to 1.5 spike patterns per synapse, with an error fraction of 0.18%.
To utilize synaptic conductances as efficient controls of the neuron's clock, the peak synaptic conductances must be plastic so that they adjust to the range of integration times relevant for a given perceptual task. This was achieved in our model by our novel supervised spike-based learning rule. This plasticity posits that the temporal window during which pre- and postsynaptic activity interact continuously adapts to the effective integration time of the postsynaptic cell (Figure 4). The polarity of synaptic changes is determined by a supervisory signal that we hypothesize to be realized through neuromodulatory control [22]. Because present experimental measurements of spike-timing–dependent synaptic plasticity rules have assumed an unsupervised setting, i.e., have not controlled for neuromodulatory signals (but see [40]), existing results do not directly apply to our model. Nevertheless, recent data have revealed complex interactions between the statistics of pre- and postsynaptic spiking activity and the expression of synaptic changes [41]–[44]. Our model offers a novel computational rationale for such interactions, predicting that for fixed supervisory signaling, the temporal window of plasticity shrinks with growing levels of postsynaptic shunting. One challenge for the biological implementation of the tempotron learning rule is the need to compute the time of the maximum of the postsynaptic voltage. We have previously shown for a current-based neuron model that this temporally global operation can be approximated by temporally local computations that are based on the postsynaptic voltage traces following input spikes [22]. We have extended this approach to plastic synaptic conductances and checked that the resulting biologically plausible implementation of conductance-based tempotron learning can readily subserve time-warp–invariant classification of spike patterns. Specifically, in this implementation, the induction of synaptic plasticity is controled by the correlation of the postsynaptic voltage and a synaptic learning kernel (see Materials and Methods) whose temporal extend is controlled by the average conductance throughout a given error trial. A synaptic peak conductance is changed by a uniform amount whenever this correlation exceeds a fixed plasticity induction threshold. When tested on the time-warped latency patterns with βmax = 2.5 (cf. Figure 3), the correlation-based tempotron roughly matched the voltage maximum–based version at a reduced learning load of 1.5 patterns per synapse with an error fractions of 0.35%.
In our model, dynamic time-warp–invariant capabilities become avaliable through a conductance-based learning rule that tunes the shunting action of synaptic conductances. This learning rule enables neurons to adjust the degree of synaptic shunting to the requirements of a given processing task. As a result, our model can naturally encompass a continuum of functional specializations ranging from neurons that are sensitive to absolute stimulus durations by employing low total synaptic conductances, to time-warp–invariant feature detectors that operate in a high-conductance regime. In the context of auditory processing, such a functional segregation into neurons with slower and faster effective integration times is reminiscent of reports suggesting that rapid temporal processing in time frames of tens of milliseconds is localized in left lateralized language areas, whereas processing of slower temporal features is attributed to right hemispheric areas [45]–[47]. Although anatomical and morphological asymmetries between left and right human auditory cortices are well documented [48], it remains to be seen whether these differences form the physiological substrate for a left lateralized implementation of the proposed time-rescaling mechanism. Consistent with this picture, the general tradeoff between high temporal resolution and robustness to temporal jitter that is predicted by our model (Figure 5) parallels reports of the vulnerability of the lateralizion of language processing with respect to background acoustic noise [49] as well as to abnormal timing of auditory brainstem responses [50].
The architecture of our speech-processing model encompasses two auditory processing stages. The first stage transforms acoustic signals into spatiotemporal patterns of spikes. To engage the proposed automatic time-rescaling mechanism, the population rate of spikes elicited in this afferent layer must track variations in the rate of incoming speech. Such behavior emerges naturally in a sparse coding scheme in which each neuron responds transiently to the occurrences of a specific acoustic event within the auditory input. As a result, variations in the rate of acoustic events are directly translated into concomitant variations in the population rate of elicited spikes. In our model, the elementary acoustic events correspond to onset and offset threshold crossings of signal power within specific frequency channels. Such frequency-tuned onset and offset responses featuring a wide range of dynamic thresholds have been observed in the inferior colliculus (IC) of the auditory midbrain [51]. This nucleus is the site of convergence of projections from the majority of lower auditory nuclei and is often referred to as the interface between the lower brain stem auditory pathways and the auditory cortex. Correspondingly, we hypothesize that the layer of time-warp–invariant feature detector neurons in our model implements neurons located downstream of the IC, most probably in primary auditory cortex. Current studies on the functional role of the auditory periphery in speech perception and its pathologies have been limited by the lack of biologically plausible neuronal readout architectures; a limitation overcome by our model, which allows evaluation of specific components of the auditory pathway in a functional context.
Psychoacoustic studies have indicated that the neural mechanism underlying the perceptual normalization of temporal speech cues is involuntary, i.e., it is cognitively impenetrable [16], controlled by physical rather than perceived speaking rate [17], confined to a temporally local context [2],[18], not specific to speech sounds [52], and already operational in prearticulate infants [53]. The proposed conductance-based time-rescaling mechanism is consistent with these constraints. Moreover, our model posits a direct functional relation between high synaptic conductances and the time-warp robustness of human speech perception. This relation gives rise to a novel mechanistic hypothesis explaining the impaired capabilities of elderly listeners to process time-compressed speech [54],[55]. We hypothesize that the down-regulation of inhibitory neurotransmitter systems in aging mammalian auditory pathways [56],[57] limits the total synaptic conductance and therefore prevents the time-rescaling mechanism from generating short, effective time constants through synaptic shunting. Furthermore, our model implies that comprehension deficits in older adults should be linked specifically to the processing of phonetic segments that contain fast time-compressed temporal cues. Our hypothesis is consistent with two interrelated lines of evidence. First, comprehension difficulties of time-compressed speech in older adults are more likely a consequence of an age-related decline in central auditory processing than attributes of a general cognitive slowing [56],[58]. Second, recent reports have indicated that recognition differences between young and elderly listeners originate mainly from the temporal compression of consonants, which often feature rapid spectral transitions, but not from steady-state segments [54],[55],[58] of speech. Finally, our hypothesis posits that speaking rate–induced shifts in perceptual category boundaries [2],[16],[17] should be age-dependent, i.e., their magnitude should decrease with increasing listener age. This prediction is straightforwardly testable within established psychoacoustic paradigms.
In a previous neuronal model of time-warp–invariant speech processing [5],[6], sequences of acoustic events are converted into patterns of transiently matching firing rates in subsets of neurons within a population, which trigger synchronous firing in a downstream readout circuit. The identity of neurons whose firing rates converge to an identical value during an input pattern, and hence also the pattern of synchrony emerging in the readout layer, depends only on the relative timing of the events, not on the absolute duration of the auditory signal. However, for this model to recognize multiple input patterns, the convergence of firing rates is only approximate. Therefore, the resulting time-warp robustness is limited and, as in our model, dependent on the learning load. Testing this model on our synthetic classification task (cf. Figure 3) indicated a substantially smaller storage capacity than is realizable by the conductance-based tempotron (Text S1). An additional disadvantage of this approach is that it copes only with global (uniform) temporal warping. Invariant processing of dynamic time warp as is exhibited by natural speech (cf. Figure 1C and 1D) is more challenging since it requires robustness to local temporal distortions of a certain statistical character. Established algorithms that can cope with dynamically time-warped signals are typically based on minimizing the deviation between an observed signal and a stored reference template [59]–[61]. These algorithms are computationally expensive and lack biologically plausible neuronal implementations. By contrast, our conductance-based time-rescaling mechanism results naturally from the biophysical properties of input integration at the neuronal membrane and does not require dedicated computational resources. Importantly, our model does not rely on a comparison between the incoming signal and a stored reference template. Rather, after synaptic conductances have adjusted to the statistics of a given stimulus ensemble, the mechanism generalizes and automatically stabilizes neuronal voltage responses against dynamic time warp even when processing novel stimuli (cf. Figure 3C). The architecture of our neuronal model also fundamentally departs from the decades-old layout of Hidden Markov Model–based artificial speech-recognition systems, which employ probabilistic models of state sequences. These systems are hard to reconcile with the biological reality of neuronal system architecture, dynamics, and plasticity. The similarity in performance between our model and such state-of-the-art systems on a small vocabulary task highlights the powerful processing capabilities of spike-based neural representations and computation.
Although the present work focuses on the concrete and well-documented example of time-warp robustness in the context of neural speech processing, the proposed mechanism of automatic rescaling of integration time is general and applies also to other problems of neuronal temporal processing such as birdsong recognition [3], insect communication [9], and other ethologically important natural auditory signals. Moreover, robustness of neuronal processing to temporal distortions of spike patterns is not only important for the processing of stimulus time dependencies, but also in the context of spike-timing–based neuronal codes in which the precise temporal structure of spiking activity encodes information about nontemporal physical stimulus dimensions [62]. Evidence for such temporal neural codes have been reported in the visual [63]–[65], auditory [66], and somatosensory [67], as well as the olfactory [68] pathways. As a result, we expect mechanisms of time-warp–invariant processing to also play a role in generating perceptual constancies along nontemporal stimulus dimensions such as contrast invariance in vision or concentration invariance in olfaction [4]. Finally, time warp has also been described in intrinsically generated brain signals. Specifically, the replay of hippocampal and cortical spiking activity at variable temporal warping [69],[70] suggests that our model has applicability beyond sensory processing, possibly also encompassing memory storage and retrieval.
Numerical simulations of the conductance-based tempotron were based on exact integration [71] of the conductance-based voltage dynamics defined in Equation 1. With the membrane capacitance set to 1, the resting membrane time constant in this model is τm = 1/gleak. Implementing an integrate-and-fire neuron model, an output spike was elicited when V(t) crossed the firing threshold Vthr. After a spike at tspike, the voltage is smoothly reset to the resting value by shunting all synaptic inputs that arrive after tspike (cf. [22]). We used Vthr = 1, Vrest = 0, and reversal potentials and for excitatory and inhibitory conductances, respectively. Unless stated otherwise, the resting membrane time constant was set to τm = 100 ms throughout our work [20]. For the synaptic time constant, we used τs = 1 ms for the random latency task, which minimized the error of the current-based neuron, and to τs = 5 ms in the speech-recognition tasks.
To check the effect of nonsynaptic voltage-dependent conductances on the automatic time-rescaling mechanism, we implemented an h-like current Ih after [72] as a noninactivating current with HH-type dynamics of the form
Here, is the maximal h-conductance, with reversal potential and m is its voltage-dependent activation variable with kineticswhereand
The voltage dependence of the rate constants α and β were described by the formwith parameters aα = −39.015 s−1, bα = −259.925 s−1, kα = 1.77926 and aβ = 365.85 s−1, bβ = −2853.25 s−1, kβ = −1.28889.
In Figure S1, we quantified the effect of the h-conductance on the fidelity of the time-rescaling mechanism by measuring the time-warp–induced distortions of neuronal voltage traces for different values of the maximal h-conductance . Specifically, for a given value of and a time warp β, we measure the voltage traces and and their standard deviations across time σ1 and σβ, respectively. We define the time-warp distortion index as the mean magnitude of the time-warp–induced voltage difference across time normalized by the mean standard deviation, ,
In Figure S1, values of are normalized by Λ(0,β). The voltage traces were generated by random latency patterns and uniform synaptic peak conductances as used in Figure 2. As increasing values of depolarized the neuron's resting potential, excitatory and inhibitory synaptic conductances were balanced separately for each value of .
In the current-based tempotron that was implemented as described in [22], each input spike evoked an exponentially decaying synaptic current that gave rise to a postsynaptic potential with a fixed temporal profile. In Figure 8C (upper row), voltage traces of a current-based analog of a conductance-based tempotron with learned synaptic conductances , reversal potentials , and effective membrane integration time τeff (cf. Figure 8B) were computed by setting the synaptic efficacies ωi of the current-based neuron to and its membrane time constant to τm = τeff. The resulting current-based voltage traces were scaled such that for each pair of models, the mean voltage maxima for unwarped stimuli (β = 1) were equal.
Following [22], changes in the synaptic peak conductance of the ith synapse after an error trial were given by the gradient of the postsynaptic potential, , at the time of its maximal value tmax. To compute the synaptic update for a given error trial, the exact solution of Equation 1 was differentiated with respect to and evaluated at tmax, which was determined numerically for each error trial. Whenever a synaptic peak conductance attempted to cross to a negative value, its reversal potential was switched.
A voltage correlation-based approximation of tempotron learning was implemented by extending the approach in [22] such that the change in the synaptic peak conductance of the ith synapse due to a spike at time ti was governed by the correlation of the postsynaptic potential V(t) with a synaptic learning kernel Klearn(t) = (exp(−t/τlearn)−exp(−t/τs))/(τlearn−τs). The two time constants of the synaptic learning kernel were the synaptic time constant τs and the learning time constant , which was determined by the time-averaged synaptic conductance of the current error trial and approximated the effective membrane time constant during that trial. The voltage maximum operation was approximated by thresholding νi, yieldingfor changes of excitatory conductances on target and null patterns, respectively, and changes with the reversed polarity, ±1, for inhibitory conductances. The plasticity induction threshold was set to κ = 0.45.
As in [22], we employed a momentum heuristic to accelerate learning in all learning rules. In this scheme, synaptic updates consisted, not only of the correction , which was given by the learning rule and the learning rate λ, but also incorporated a fraction μ of the previous synaptic change . Hence, . We used an adaptive learning rate that decreased from its initial value λini as the number of learning cycles l grew, λ = λini/(1+10−4(l−1)). A learning cycle corresponded to one iteration through the batch of templates in the random latency task or the training set in the speech task.
Global time warp was implemented by multiplying all firing times of a spike template by a constant scaling factor β. In Figure 3A, random global time warp between compression by 1/βmax and dilation by βmax was generated by setting β = exp(qln(βmax)) with q drawn from a uniform distribution between −1 and 1 for each presentation.
Dynamic time warp was implemented by scaling successive interspike intervals tj−tj−1 of a given template with a time-dependent warping factor , such that warped spike times with and . The time-dependent factor resulted from an equilibrated Ornstein-Uhlenbeck process ξ(t) with a relaxation time of τ = 200 ms that was rescaled by the complementary error function erfc to transform the normal distribution of ξ(t) into a uniform distribution over [−1 1] at each t.
To ensure that the symmetry of excitation and inhibition in our model architecture was not crucial for the time-warp–invariant processing of spike patterns, we implemented a control architecture in which all afferents were confined to be excitatory, except one additional inhibitory synapse, which mimicked the effect of a global inhibitory network. In the time-warped random latency task, spike patterns were fed into the excitatory population as before; however, in addition, the inhibitory synapse received a copy of each incoming spike. All synaptic peak conductances were plastic and controlled by the conductance-based tempotron rule. In this model, synaptic sign changes were prohibited.
Spike time jitter [22] was implemented by adding independent Gaussian noise with zero mean and a standard deviation of 5 ms to each spike of a template before each presentation.
Sound signals were normalized to unit peak amplitude and converted into spectrograms over NFTT = 129 linearly spaced frequencies fj = fmin+j(fmax+fmin)/(NFTT+1) (j = 1… NFTT) between fmin = 130 Hz and fmax = 5,400 Hz by a sliding fast Fourier transform with a window size of 256 samples and a temporal step size of 1 ms. The resulting spectrograms were filtered into Nf = 32 logarithmically spaced Mel frequency channels by overlapping triangular frequency kernels. Specifically, Nf+2 linearly spaced frequencies given by hj = hmin+j(hmax−hmin)/(Nf+1) with j = 0…Nf+1 and hmax,min = 2,595log(1+fmax,min/700) were transformed to a Mel frequency scale between fmin and fmax. Based on these, signals in Nf channels resulted from triangular frequency filters over intervals with center peaks at . After normalization of the resulting Mel spectrogram SMel to unit peak amplitude, the logarithm was taken through log(SMel = ε)−log(ε) with ε = 10−5 and the signal in each frequency channel smoothed in time by a Gaussian kernel with a time constant of 10 ms. Spikes were generated by thresholding of the resulting signals by a total of 31 onset and offset threshold-crossing detector units. Whereas each onset afferent emitted a spike whenever the signal crossed its threshold in the upward direction, offset afferents fired when the signal dropped below the threshold from above. For each frequency channel and each utterance, threshold levels for onset and offset afferents were set relative to the maximum signal over time to and . For , onset and offset afferents were reduced to a single afferent whose spikes encoded the time of the maximum signal for a given frequency channel.
We used the digit subset of the TI46 Word speech database [24]. This clear speech dataset comprises 26 isolated utterances of each English digit from zero to nine spoken by 16 adult speakers (eight male and eight female). The data is partitioned into a fixed training set, comprising 10 utterances per digit and speaker, and a fixed test set holding the remaining 16 utterances per digit and speaker. We also tested our neuronal word-recognition model on the adult speaker, isolated-digit subset of the TIDIGITS database [27]. This subset comprises two utterances per digit and speaker, i.e., a total of 20 utterances from 225 adult speakers (111 male and 114 female), that are dialectically balanced across 21 dialectical regions (tiling the continental United States). Because the TI46 database only provides utterances of the word “zero” for the digit 0, we excluded the utterances of “oh” from our TIDIGITS sample.
Based on the spiking activity of all binary digit detector neurons, a full digit classifier was implemented by ranking the digit detectors according to their individual task performances. As a result, a given stimulus was classified as the target digit of the most reliable of all responding digit detector neurons. If all neurons remained silent, a stimulus was classified as the target digit of the least reliable neuron.
To preserve the timing relations between the learned spectrotemporal features and the target words, we refrained from correcting the spike-triggered stimuli for stimulus autocorrelations [73].
Test errors in the speech tasks were substantially reduced by training with a Gaussian spike jitter with a standard deviation of σ added to the input spikes as well a symmetric threshold margin v that required the maximum postsynaptic voltage on target stimuli to exceed Vthr+v and to remain below Vthr−v during null stimuli. Values of λini, μ, σ, and v were optimized on a 4-dimensional grid. Because for each grid point, only short runs over maximally 200 cycles were performed, we also varied the mean values of initial Gaussian distributions of the excitatory and inhibitory synaptic peak conductances, keeping their standard deviations fixed at 0.001. The reported performances are based on the solutions that had the smallest errors fractions over the test set. If not unique, we selected the solution with the highest robustness to time warp (cf. Figure 8B). Note that this naive optimization of the training parameters did not maintain a separate holdout test set for cross-validation and might therefore overestimate the true generalization performance. We adopted this optimization scheme from [25],[26] to ensure comparability of the resulting performance measures.
HTK generalization performance was tested with the HTK package version 3.4.1 [74] with front-end and HMM model parameters following [26]. Specifically, speech data from the TI46 and TIDIGITS databases were converted to 13 Mel-cepstral coefficients (including the 0th order coefficient) along with their first and second derivatives at a frame rate of 5 ms. Mel-coefficients were computed over 30 channels in 25-ms windows with zero mean normalization enabled (TARGETKIND = MFCC_D_A_Z_0). In addition, the following parameters were set: USEHAMMING = T; PREEMPCOEF = 0.97; and CEPLIFTER = 22. Ten HMM models, one for each digit plus one HMM model for silence, were used. Each model consisted of five states (including the the two terminal states) with eight Gaussian mixtures per state and left-to-right (no skip) transition topology.
|
10.1371/journal.pcbi.1003746 | Different Inward and Outward Conduction Mechanisms in NaVMs Suggested by Molecular Dynamics Simulations | Rapid and selective ion transport is essential for the generation and regulation of electrical signaling pathways in living organisms. Here, we use molecular dynamics (MD) simulations with an applied membrane potential to investigate the ion flux of bacterial sodium channel NaVMs. 5.9 µs simulations with 500 mM NaCl suggest different mechanisms for inward and outward flux. The predicted inward conductance rate of ∼27±3 pS, agrees with experiment. The estimated outward conductance rate is 15±3 pS, which is considerably lower. Comparing inward and outward flux, the mean ion dwell time in the selectivity filter (SF) is prolonged from 13.5±0.6 ns to 20.1±1.1 ns. Analysis of the Na+ distribution revealed distinct patterns for influx and efflux events. In 32.0±5.9% of the simulation time, the E53 side chains adopted a flipped conformation during outward conduction, whereas this conformational change was rarely observed (2.7±0.5%) during influx. Further, simulations with dihedral restraints revealed that influx is less affected by the E53 conformational flexibility. In contrast, during outward conduction, our simulations indicate that the flipped E53 conformation provides direct coordination for Na+. The free energy profile (potential of mean force calculations) indicates that this conformational change lowers the putative barriers between sites SCEN and SHFS during outward conduction. We hypothesize that during an action potential, the increased Na+ outward transition propensities at depolarizing potentials might increase the probability of E53 conformational changes in the SF. Subsequently, this might be a first step towards initiating slow inactivation.
| Voltage gated sodium channels are essential components of living cell membranes. They regulate the cell potential by facilitating permeation of ions across the membrane. In the past decades, studies revealed that the bacterial selectivity filter (SF) exhibits a constricted architecture lined with electronegative carboxyl oxygens of four glutamic acid side chains (EEEE motif), which repulse anions but attract Na+ ions. Crystal structures enable the investigation of structural dynamics with computational methods. Ion selectivity and conduction mechanisms between Na+, K+ and Ca2+ are progressively elucidated by molecular dynamics simulations and free energy calculations. The structural dynamics of the protein, especially the flexibility of SF and its fundamental role in kinetics underpinning ion selectivity, conduction and channel gating are less well understood. To shed light on this question, we use computational simulations to simulate ion conduction with membrane potentials. Our results suggest different dynamical behaviors of the EEEE locus and distinct ion distribution patterns in the SF with respect to permeating directionalities. These findings indicate a novel mechanism in differentiating reciprocal transitions of ion flow, preventing large sodium efflux during action potential initiation and may further suggest that increased flipping propensities at depolarizing potentials, might initially trigger channel slow inactivation.
| Na+ flux through voltage gated sodium channels (NaV) is crucial for initiating action potentials in the membranes of electrically excitable cells. They mediate a variety of biological functions such as muscle contraction, propagation of nerve impulses, release of hormones and many more [1]. As a consequence, mutations in NaV channels lead to a variety of channelopathies, such as congenital epilepsy, cardiac arrhythmias or chronic pain [2], [3].
Recently, homotetrameric crystal structures of several bacterial NaV channels were successfully resolved [4]–[10], providing a tremendous opportunity to investigate the structure and function of these channels on the atomistic level. They are composed of four membrane spanning subunits and contain six transmembrane (TM) helices per subunit. Helices S1 to S4 form the voltage sensing module. Helices S5, P1 segments, the selectivity filter (SF) region, P2 segments and S6 helices, lining the inner pore cavity, form the pore module. The SF of most bacterial channels contains the amino acid sequence TLESW. The four glutamic acid side chains [11] form a high field strength binding site (HFS) [12] which is essential for ion selectivity. In eukaryotic sodium channels, this site consists of the amino acids motif DEKA.
The molecular mechanisms underlying ion conduction and selectivity in NaV are beginning to emerge. Computational methods, particularly molecular dynamics (MD) simulations are extensively adopted to address these questions [13]–[23].
As reviewed recently [24], sodium ions were illustrated to spontaneously traverse the SF into the cavity with energy barriers between ∼2–5 kcal/mol. Compared to potassium coordination in KV channels, sodium ions partially or fully preserve their first hydration shells [13], [15]–[17], [20], [22]. A loosely coupled knock-on mechanism with an average ion occupancy around two in the SF was predicted during ion conduction [14], [15], [21], [22]. The incoming ion repulses the present ion out of the SF. The wide radius (≥9 Å) of the SF enables double occupancy of ions at the same level [20], [21]. Further, Na+ vs. K+ [15], [16], [20], [20]–[22] and Na+ vs. Ca2+ [17], [20] discrimination studies were carried out. These studies revealed higher energy barriers in the SF for K+ and Ca2+ compared to Na+. Subsequently, non-equilibrium simulations were performed to investigate conduction under applied membrane potentials and to study kinetics [21], [22]. The estimated inward conductance rate successfully reproduced electrophysiology data [22].
A recent study by Chakrabarti et al. [23], suggested that conformational changes at the EEEE motif (corresponding to E177 in NaVAb) might play an important role in ion conduction. However, this observation was not reported in other simulations, except for simulations using Ca2+ as a charge carrier [17]. In K+ channels, subtle structural changes in the SF, involving rotations around a highly conserved glycine residue result in different non-conductive conformations [25], [26]. This regulation of ion flow by conformational changes of the selectivity filter is termed C-type inactivation.
It is not clear to which extend structural changes at the EEEE locus in Nav channels are crucial for conductance and inactivation in Nav channels.
To investigate these issues, we conducted MD simulations using the open conformation of the bacterial sodium channel homologue NaVMs (Magnetococcus sp. (strain MC-1)) [7] pore domain focusing on the structural changes of the SF during inward and outward conduction.
The four-fold symmetrical structure of NaVMs (pdb identifier: 4F4L) was generated using chain A (splayed outward by 25° rotation about its Ψ-bond at position T84), which creates an open pore with a diameter of ∼14 Å [7]. As described by Ulmschneider et al. [22], a harmonic restraint was exerted on the S5 and S6 TM helices to keep the gate in the open conformation throughout simulations. This structure was then embedded into a POPC lipid patch and duplicated in the Z direction (pore axis). A constant charge imbalance of four elementary charges (4 e) across each lipid bilayer between the central electrolyte bath and the two outer ones was maintained during simulation (supplementary movie S1) [27].
Four times 500 ns double-patch MD simulations with 500 mM NaCl were performed, with the first 100 ns treated as equilibration. Figure 1 shows the cumulative ion conducting events from MD simulations with depolarized and hyperpolarized membrane potentials of ΔV: 565±126 mV (Figure 2A). The estimated sodium current in the inward direction is γ = 27±3 pS. This value agrees with previously observed single channel conductance measurements (γ∼33 pS) [22]. Our double bilayer simulations enabled us to further estimate outward conduction, which amounts to γ = 15±3 pS. Interestingly, this process is distinguishably slower than inward ion flux (P<0.01, N = 4).
To explore the underlying differences between inward and outward ion permeation, we plotted the ion probability density map across the pore region from all four simulations. Several favorable ion-interacting sites (SEX, SHFS, SCEN and SIN) from periplasm to cytoplasm were assigned as proposed previously by Payandeh et al. [4]. The ion-interacting sites across the SF were determined by measuring the axial distance (Z axis) along certain atoms from −5.00 to 10.25 Å as shown in Figure 3. Side chain oxygens of S54 from all four chains were taken as the origin (Z = 0.0 Å) of the SF. Additionally, in two previous studies, a site with an energy barrier (∼2 kcal/mol) distinguishing between site SHFS and SCEN was identified [17], [22]. In this study, we refer to this site as “SBAR”, indicating this barrier (2.75≤Z<4.75 Å).
During influx (Figure 4A and B), short-lived Na+ binding at site SEX was observed (2.6±0.5 ns) in an asymmetrical manner. SHFS is the dominant site with the highest ion density. At this site, ions tended to be directly coordinated with side chain oxygens of E53 and S54 in an off-axis manner. Additionally, a less densely populated configuration was observed in the center of this site consistent with previous studies [21], [24]. Moving inward from site SHFS, ions further translocated transiently via site SBAR (1.5±0.3 ns) to site SCEN (3.4±0.3 ns). Subsequently, ions reciprocally traversed between sites SCEN and SIN. These results are in good agreement with previous simulation studies [14], [15], [17], [21], [22].
Our simulations revealed a distinct ion distribution pattern for efflux compared to influx as shown in Figure 4C and D. After entering into site SIN from the cytosol, ions mainly populated sites SCEN and SBAR with extended dwell times compared to inward conduction, (SCEN: 11.9±1.1 ns vs. 3.4±0.3 ns; SBAR: 8.2±0.9 ns vs. 1.5±0.3 ns) suggesting a putative barrier for efflux between sites SBAR and SHFS. Additionally, during efflux, Na+ ions tended to traverse in an on-axis manner through the filter.
Conformational isomerization of the E53 side chains has been reported previously [17], [23]. Generally, the glutamic acid side chain might adopt two main conformations (Figure 5A and B): inward-facing (χ2 angle ∼60°, flipped) and outward-facing (χ2 angle ∼290°, non-flipped). In our simulations, during influx, flipping events were observed only in 2.7±0.5% of the simulation time, thus the E53 side chain mainly adopted a non-flipped conformation. In contrast, during efflux 32.0±5.9% flipping events were observed (P value = 0.015, see Figure 5C). A more detailed investigation of this flipping events revealed that 80% of these changes occurred in only one of the four glutamic acid side chains (“one-flip”) (Figure 5D).
To investigate the influence of the presence of Na+ ions on E53 side chain dynamics, three repeated simulations without ions in the SF (“no salt”) were performed. Irrespective of the directionality of the applied potentials, the flip probability is less than 0.6% in all simulations (Figure 5C). This indicates, that a depolarizing potential per se does not significantly influence the number of E53 flipping events. This suggests that the combination of local positive charge carried by outward Na+ flux in the SF especially at sites SHFS and SBAR and the outward attracting membrane potential might collectively induce the rotation of the χ2 angle from ∼290° to ∼60°.
A detailed investigation of the ionic binding modes and their relations to free energy profiles enabled us to describe mechanisms regarding different conducting directionalities (Figure 6 and 7). During inward conduction, the largest barrier in the SF occurs between sites SHFS and SBAR which amounts to 2.1 kcal/mol (Figure 6B). At site SHFS, the probe ions (yellow) mainly distributed in an off-axis manner, the first coupling Na+ ions (blue) may occupy site SCEN (IN), and there existed a second binding site for coupling ions at site SEX (Figure 6A, II). Subsequently, the probe ions distributed in the middle of channel axis when traversing the short lived site SBAR, with the other two coupling ions populating sites SEX and SIN (CAV) respectively (Figure 6A, III). The probe ions then occupied site SCEN in both on-axis and off-axis manners, other coupling ions in the SF were distributed mainly at sites SIN and SEX. Only a few coupled ions occupied sites SHFS and SBAR (Figure 6A, IV).
In these ionic binding modes, under hyperpolarized potential, the coupling ions in the SF generally demonstrated a loosely coupled knock-on mechanism with only a few of them present in the adjacent binding sites to the probe ions (Figure 6A, II–IV). This is in agreement with a study by [21], [22], where it was shown that during inward conduction the ions displayed a combination of mono-ionic and multi-ionic mechanism with an overall occupancy of 1.8 ions in the pore region.
The flipping probability analysis indicates that the conformational changes of the E53 side chains play a minor role for ion inward conduction as shown in Figure 6A, II′–IV′.
Compared to inward permeation, the maximum energy barrier during outward conduction amounted to 2.3 kcal/mol between sites SBAR and SHFS. It is interesting that the free energy difference between sites SCEN and SHFS is 2.2 kcal/mol, which is close to the largest energy barrier (Figure 7B). In addition, the ionic binding modes demonstrate a distinct conduction mechanism compared to Na+ influx. Traversing outward from the cavity, ions at site SIN were tightly coupled with ions at site SBAR (Figure 7A, III and V) corresponding to two energy wells in Figure 7B, III and V). When probe ions located at site SCEN, the coupling ions distributed in the upper part of the cavity (Figure 7A, IV) which corresponds to the energy well at site SCEN (Figure 7B, IV). Generally, the translocation of probe ions from the cytoplasm to site SBAR is readily stepwise by a tight knock-off mechanism without significant energy barriers. At all three energy wells (Figure 7B III–V) the E53 side chains maintained non-flipped conformations.
When probe ions faced the energy barrier at site SBAR, a delicate tightly-coupled “knock-off” conducting mechanism occurred. Initially, E53 started to flip and one of the carboxyl oxygens started to coordinate the probe ions (Figure 7A, III′ and S5, B). Compared to ions located in the close energy wells (Figure 7A, III), the probe ions were meanwhile expulsed by the outward movements of approaching coupling ions at site SIN (Figure 7A, III′). If this knock-off mechanism was successful, the probe ions would then migrate to site SHFS, as a result, the coupling ions would move outward to site SCEN simultaneous (Figure 7A, II′ and IV′). At this time, two carboxyl oxygens of the flipped E53 side chain tended to coordinate with the probe ions and coupling ions respectively (Figure 7A, II′ and IV′ and S5, B). Afterwards, ions left site SHFS promptly (t = 2.6 ns) into the periplasm via site SEX.
If the attempt to overcome the barrier failed, the aforementioned mechanism was easily reversed, the probe ions and coupling ions occupied the two stable energy wells at sites SBAR with the coupling ions at site SIN (Figure 7A, III and V) and site SCEN with the coupling ions in the cavity (Figure 7A, IV) again. That is the reason why ions stayed longer in sites SBAR and SCEN.
One the one hand, larger Pi values (flip inducing probability of number of probe ions, see methods for details) values of sites SBAR, SCEN and SHFS indicated the flipped conformations of E53 were crucial (Pi>90%) in overcoming the dual energy barriers between SCEN, SBAR and SHFS. On the other hand, smaller Pt values (flipping time probability for all probe ions, see methods for details) values indicated that the flipping events were easily reversible. Because of these flipping events, the major ion distribution for outward conduction is limited to the center of the channel axis during translocation within the SF.
To further explore the correlation between flux directionality and E53 conformation, we performed two sets of inward and outward conduction simulations (four times 300 ns) with dihedral restraints to maintain “non-flip” and “one-flip” configurations during sampling. The influx rate was independent of the E53 conformations as shown in Figure 8A. This observation disagrees with recent data from Chakrabarti et al. [23] on the NaVAb channel. The outward conduction with “one-flip” simulation displayed an increased efflux rate compared to simulations without dihedral restraints on E53, where the flipping events would be reversible when conducting ions (Figures S1, S2, S3, S4). Interestingly, if E53 was restrained to a “non-flip” configuration, sodium ions translocation slowed down (Figure 8B). These results suggest a clear influence of filter dynamics on the efflux rate.
Comparison of the free energy profiles from outward simulations of these three types of configurations revealed that the largest energy barrier of the “non-flip” simulations is increased from 2.3 kcal/mol (non-restraint) to 3.4 kcal/mol (Figure 9A and C). The lowest energy well at site SBAR was also replaced by site SCEN. The energy profile of “one-flip” simulations indicated that the energy barrier between sites SCEN and SHFS was diminished, although the original energy barrier increased slightly by 0.3 kcal/mol (Figure 9A and B). Therefore, the outward conduction would be more straightforward without reversible backward translocation compared to the simulations without the dihedral restraints from a kinetic point of view.
The mechanism of ion conduction and selectivity of bacterial voltage gated sodium channels is gradually emerging. The inverted tepee shape architecture of the SF lined with TLESW sequence enables sodium influx at the diffusion rate. The glutamic acid side chains are responsible for recruiting Na+ ions from the outer vestibule. Ions will then translocate via sites SHFS, SBAR, SCEN and SIN spontaneously to complete a conduction event [21]–[23]. Details of conduction are only partially understood. Large conformational changes of the glutamic acid side chains were described recently [23]. However, its role for conduction is still under discussion [24].
To gain further insights into these questions, we performed MD simulations to compare the different binding patterns and characterize the structural dynamics of glutamic acids during ion permeation. Double bilayer simulations with the open NaVMs structure enabled us to investigate influx and efflux separately. The calculated inward conductance rate is in good agreement with a previously reported experiment and computational data [22]. The estimated outward conductance rate obtained from MD simulations is predicted to be markedly lower than inward permeation (15±3 pS vs 27±3 pS, Figure 1). Ion translocation between sites SBAR and SHFS is substantially prolonged (8.2±0.9 ns vs 1.5±0.3 ns, Figure 4) during Na+ efflux. From the energetic point of view, this would imply a potential barrier. This agrees with previous two-ion free energy calculation studies, revealing a higher energy barrier in this region for outward current compared to inward conduction (ΔG: 4.6 kcal/mol vs 0.4 kcal/mol, Stock et al. [21]; ΔG: 3.5±0.5 kcal/mol and 2.4±0.3 kcal/mol, Furini and Domene [15]). In our studies, this barrier is also higher for outward conduction (ΔG: 2.3 kcal/mol vs. 2.1 kcal/mol).
In agreement with previous studies [21], [22] during inward conduction, our simulations revealed that ion translocations in the SF generally involve a loosely coupled knock on mechanism with an average ion occupancy of 1.8 (Figure 6).
A possible outward conduction mechanism was described by Stock et al., [21] using a “fully activated-open” NaVAb channel structure [28]. They have found a third ion denoted k, directly coupling with the probe ions triggering outward conduction by a “nudging” collision effect. Similar results were obtained in our study, which shows that the coupling ions directly couple with the probe ions by a tight “knock-off” mechanism. Moreover, our simulations further elucidated that this “knock-off” mechanism is highly dependent on the conformational isomerization of the glutamic acid side chains in the SF. In other words, to overcome the energy barriers of outward conduction, at least one of the glutamic acid side chains has to be flipped to an inward facing conformation (Figure 7).
A recent simulation study under ∼0 mV membrane potential with a closed gate NaVAb structure suggested that Na+ in- and outward movement involves variable configurations of multiple glutamic acid side chains giving rise to non-simple degenerated ion binding modes [23]. Remarkably, detailed investigations of the structural dynamics of E53 in our study revealed distinct isomerization patterns between forward and backward translocations respectively. When the ion moved into the SF from the outer vestibule under hyperpolarized membrane potential, the E53 remained mostly in the non-flipped conformation. In contrast, during outward conduction, the flipping occurrence increased significantly with a typical “one-flip” configuration (Figure 5C and D) when coordinating ions occupied the SF (Figure 7). In our simulations, the depolarized and hyperpolarized membrane potentials of approximately ΔV: 565 mV enabled the detailed investigation of ion permeation directionalities. This was not possible in previous simulations at ∼0 mV [23]. The conductive, open gate structure used in this study may also reduce the repulsive effect which could have been induced by ions present in the cavity in previous simulations with a closed gate [23]. In addition, different forcefields used in these two studies may also play a substantial role for these discrepancies. As reported by Cordomi et al [29]), compared to the combination of OPLS-AA protein with Berger lipids parameters, combining Amber99sb protein and Berger lipids gives more accurate free energies of solvation in water and water to cyclohexane transfer with respect to experimental data for glutamic acid side chains. This may explain the reduced flexibility of the glutamic acid side chain dynamics observed in our study. Further, the force field discrepancies might explain the contrasting results for the “no salt” simulations in these two studies. In the study by Chacrabarti et al [23], the E side chains are more favorable to form flipped conformations even in the “no salt” conformation. This is in contrast to our simulations, where flipping events occurred rarely (Figure 5C) in the “no salt” simulations.
While the inward flow exhibited indistinguishable flux rates irrespective of the E53 conformation (Figure 8A), efflux displayed different rates depending on the configurations of the E53 side chain. The highest efflux rate was observed in our “one-flip” simulations and the lowest rate with all four glutamic acid side chains restrained to an outward-facing conformation (Figure 8B). PMF calculations further confirmed that the energy barrier for outward conduction increased from 2.3 kcal/mol to 3.4 kcal/mol when the flipping conformation is prohibited (Figure 9C). That indicates that this flipping conformation provides direct coordination for Na+ ions, which lowers the energy barrier and aids outward conduction.
A simulation study published [30] after the submission of this manuscript, indicates that the SF dynamics, especially the side chain conformational changes of the EEEE locus in the SF, may lead to the conformational changes of the cavity lining helix on the µs timescale, subsequently initiating slow inactivation in NaV channels. We hypothesize that the E53 dynamics under depolarizing potentials uncovered in this study provide further insights into slow inactivation, especially the fast slow inactivation for prokaryotic species during action potentials. When the membrane potential depolarizes, the probability of Na+ outward transitions increases. As a result, the inactivation probability of the channel is increased probably due to a series of conformational changes starting from the EEEE locus in the SF.
A general limitation of current force fields is that the simulated linear current–voltage regime can only be achieved at higher membrane potentials compared to experimental conditions, resulting from the large electrostatic barriers in the transmembrane region [31], [32]. It should be noted that the computational electrophysiology simulations in this study were not done at constant membrane potentials (565±126 mV). This may result from the movement of the ions inside the channels and the fluctuation of the ions in the aqueous compartments (Figure 2A and B). However, a single ion permeation event under physiological conditions will also exist as a non-equilibrium process. Thus, to which extent, current simulation methods resemble ion channels' electrophysiology needs to be further validated. In addition, inaccuracies in the interaction parameters (from the forcefield) between ions and surrounding atoms could also influence the conduction rates [33]. Thus, further structural and computational studies (including optimized strategies for ion interaction with surrounding atoms and polarizable force fields with different lipid species) will be required to further investigate the conformational changes of the SF under different electrochemical drives and the influence of different protonation states of the EEEE locus. In addition, experimental validation is essential to further uncover the structural determinants and the importance of the protonation states of the EEEE locus on ion conductance and selectivity.
Summarizing, our simulations, using applied membrane potentials, reveal different conduction mechanisms for ion inward and outward transitions respectively. An inward facing conformation (flip) of one glutamic acid side chain in the SF would reduce the energy barrier for ion outward transition by providing direct coordination with interacting Na+ ions. This local change can provide insights into the slow inactivation of NaV channels as suggested by Boiteux et al [30] during an action potential, when the membrane potential is depolarized.
The coordinates of NaVMs (PDB Entry: 4F4L; Resolution: 3.49 Å) [7] with a conductive pore gate were used. The symmetric tetrameric structure consists of residues 8 to 94. All charged residues were treated keeping their charge states at physiological pH 7.4. In order to investigate ion conductance under two opposite membrane potentials, we used the computational electrophysiology method developed by Kutzner et al. [27] with a double-bilayer scheme. Each bilayer leaflet consists of 242 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) lipids encompassing the protein structure, solvated with 500 mM NaCl solution. The system was then duplicated in the Z direction (pore axis). The virtual-site model was adopted for hydrogen atoms [34].
MD simulations were performed with Gromacs version 4.5.5-dev [35], [36]. Simulations were carried out with the AMBER99sb [37] all atom force field, POPC lipids parameter were taken from Berger et al. [29], [38] with the TIP3P water model [39]. All covalent bonds were constrained using the LINCS algorithm [40], allowing for an integration time step of 4 fs with virtual sites. A 10 Å cutoff was adopted for calculating short-range electrostatic interactions and the Particle Mesh Ewald [41] summation was used for calculating long-range electrostatic interactions. The corrected monovalent ion Lennard-Jones parameters for the amber forcefield [42] were implemented in this study and the vdW interactions were calculated with a cutoff of 10 Å. The Nose-Hoover thermostat [43], [44] and the semi-isotropic Parrinello-Rahman barostat algorithm [45] was used to maintain simulation temperature and pressure constantly at 300 K and 1 bar, respectively.
Prior to MD simulations, 3000 conjugate gradient energy-minimization steps were performed, followed by 5 ns equilibration in order to fully solvate mobile water and lipids around the restrained protein with a force constant of 1000 kJ/mol/nm2 on all heavy atoms. Hereafter, an equal number of Na+ ions and a net difference of 4 e of Cl− across each lipid bilayer between the central electrolyte bath and the two outer ones were sustained during the simulation by a swapping mechanism [27]. In this scheme, a new form of Poisson equation [46] was adopted to derive the potential profile as a function of system length (z). The well-defined transmembrane voltage across each lipid patch was directly assessed by twice integration of this sustained charge density differences between the central electrolyte bath and the two outer ones [27]. In our simulations, depolarized and hyperpolarized membrane potentials were calculated as ΔV = 565±126 mV (Figure 2). Harmonic restraints (1 kcal/mol/Å2) were exerted on the α-carbon atoms of the TM helices (S5 and S6) throughout the simulations to maintain the open configuration in the absence of the voltage-sensing domain as suggested by Ulmschneider et al. [22]. Four times 500 ns MD simulations were performed; the first 100 ns were treated as equilibration. Simulation trajectories were saved every 100 ps; as a result, 4000 snapshots (analyzing windows) were recorded for analyzing data. Three repeated 500 ns simulations (400 ns were adopted for analysis) with only ions neutralizing the system and no ions in the SF (“no salt”) were performed as control to investigate the influence of the membrane potential on the E53 side chain dynamics. In this setup, only ions used to generate the charge imbalance and neutralize the net charge were kept. Further, four repeated 300 ns simulation (200 ns were adopted for analysis) for “non-flip” and “one-flip” configurations respectively were carried out, where the dihedral restraints were applied on the χ2 dihedral of E53 for all four subunits with a force constant of 500 kcal/mol/rad2. This allows dynamic ranges of 56±10° for “one-flip” configuration and 288±10° for “non-flip” configuration of the χ2 angle.
The total number of ions (i) which completed their conduction in the SF were analyzed (i = 158, from four inward simulations; and i = 79 from four outward simulations). All snapshots of the probe ions (yellow) and coupling ions in the SF (blue) were rendered for five different interaction sites (SEX, SHFS, SBAR, SCEN and SIN). For sites SHFS, SBAR and SCEN, the side chain isomerization states were separated into flipped and non-flipped categories and analyzed. For flipped ones, all four protein chains and the relative ion positions were aligned to chain A to achieve a better representation of the ionic binding patterns. Two probability parameters Pi and Pt were calculated to characterize the influence of E53 dynamics on ion conduction for these three sites. Pi = (Fi/i)*100%, where Fi denotes the number of ions which generated at least one E53 flipping event during their permeation through each site. Pt = Ft/Tt*100%, where Ft denotes the number of snapshots where E53 flipped when the probe ions traversed each site and Tt denotes the total number of snapshots when the probe ions traversed each site.
The 1-D potential of mean force profile of the ions under membrane potentials were calculated by taking the logarithm of the Na+ probability distribution along the channel axis (z) in the SF region, according to G(z) = −kBT ln [p(Ri)], where kB is the Boltzmann constant, T is the temperature, and p(Ri) is the probability distribution of the probe ions. 100 bins were used to achieve a bin width of 0.15 Å depicting the details of the profile. Error bars are S.E.M. from four different simulations.
|
10.1371/journal.pcbi.1004408 | Detecting Horizontal Gene Transfer between Closely Related Taxa | Horizontal gene transfer (HGT), the transfer of genetic material between organisms, is crucial for genetic innovation and the evolution of genome architecture. Existing HGT detection algorithms rely on a strong phylogenetic signal distinguishing the transferred sequence from ancestral (vertically derived) genes in its recipient genome. Detecting HGT between closely related species or strains is challenging, as the phylogenetic signal is usually weak and the nucleotide composition is normally nearly identical. Nevertheless, there is a great importance in detecting HGT between congeneric species or strains, especially in clinical microbiology, where understanding the emergence of new virulent and drug-resistant strains is crucial, and often time-sensitive.
We developed a novel, self-contained technique named Near HGT, based on the synteny index, to measure the divergence of a gene from its native genomic environment and used it to identify candidate HGT events between closely related strains. The method confirms candidate transferred genes based on the constant relative mutability (CRM). Using CRM, the algorithm assigns a confidence score based on “unusual” sequence divergence. A gene exhibiting exceptional deviations according to both synteny and mutability criteria, is considered a validated HGT product. We first employed the technique to a set of three E. coli strains and detected several highly probable horizontally acquired genes. We then compared the method to existing HGT detection tools using a larger strain data set.
When combined with additional approaches our new algorithm provides richer picture and brings us closer to the goal of detecting all newly acquired genes in a particular strain.
| The transfer of genetic material between organisms, usually denoted as horizontal (or lateral) gene transfer (HGT or LGT), is a prime mechanism in microbial evolution and responsible for genetic innovation and the evolution of genome architecture. Detecting HGT between closely related species or strains is imperative as drug-resistant pathogenic strains most often acquire their virulence from closely related bacteria. The proposed method combines two evolutionary signals that were not employed in the past for this task. One is the synteny index (SI), measuring the loss of synteny in an organism, and the other is a novel concept—constant relative mutability (CRM), maintaining that genes preserve their relative evolution rate along linages (although the latter ones may each change).
We show both in simulation and real biological data that the method is sound and, in the cases examined, provides stronger sensitivity than existing methods. We therefore believe this novel approach represents a significant advance, for the first time enabling the detection of previously ignored HGT events that will bring us closer to the goal of detecting all newly acquired genes in a particular strain.
Availability: The method is publicly available at http://research.haifa.ac.il/~ssagi/software/nearHGT.zip
| Most microbial genomes have experienced extensive gene mobility between lineages during their evolution, a phenomenon known as horizontal gene transfer (HGT). This process has been critical in shaping microbial genome evolution both in terms of functional repertoires and of genome architecture [1, 2, 3, 4, 5, 6]. Many HGT events result in a gene being copied from the donor genome to the recipient genome (see Fig 1), and this process can be mediated by integration of viruses (bacteriophages), transposable elements, or integrative plasmids, often by non-homolgous recombination.
The study of the HGT is of paramount importance for several reasons. First, from a clinical perspective, HGT plays a major role in the emergence of new human diseases, as well as promoting the spread of antibiotic resistance in bacterial species [7, 8]. From the fundamental, evolutionary standpoint, HGT links distant branches in the tree of life, turning it into an evolutionary network [9, 3, 10]. Genetically, HGT is an important, if not the primary, source of genetic novelty by bacteria and archaea and often results in adaptations to new environments and conditions [11]. Recent advances of comparative genomics and especially metagenomics indicate that the complexity of the genetic material that is horizontally transferred, is vast and often exceeds by orders of magnitude the complexity of the set of conserved genes that are mostly vertically inherited [12]. Therefore, correct identification of HGT can shed light on many significant evolutionary processes some of which are adaptive.
Currently, there are two prevailing methods for detecting HGT. The phylogeny based approach takes a relatively large set of copies of the investigated gene (may contain several copies at a species due to duplication), constructs their corresponding phylogeny and contrasts it to the phylogeny of their originating species. When conflicts are found between the two trees, they are reconciled by introducing HGTs or other events (see e.g. [13, 14, 15, 16, 17]). While this approach has the advantage of identifying relatively ancient events, it is based on a very stringent assumption of where to seek the events—which is the transferred gene. Additionally, it also requires a multiple sequence alignment (MSA) of the sequences, and inferring a reliable species tree (two major problems by themselves [18, 19], in particular where phylogenetic signal is weak). In contrast, the composition based approach contrasts genomic sequences of different compositional features such as G+C content, dinucleotide frequencies or codon usage biases, striving to detect genes with different origins than the rest of the genome (e.g. [20, 5, 6, 21, 22]). The latter approach suffers from the fact that the species involved might share similar compositional patterns. Moreover, the length of a transferred segment may be too short to reliably reveal these differences. As concluded in [23, 24], “atypical G+C content and pattern of codon usage are not reliable indicators of horizontal gene transfer events”.
Both the phylogenetic and the sequence composition based approaches must rely on strong enough signals for detecting HGT: The phylogenetic approach requires the transferred gene to be relatively distinct from its close relatives’ counterparts and at the same time resemble a relatively distant species in the taxa set [14, 25]. The sequence composition based approach, requires the transferred segment to be of relatively distant origin, so that enough divergence has accumulated to result in different compositional features. Thus, to maximize sensitivity and accuracy HGT detection should use an array of approaches to either detect new events or confirm events detected by one method, using rival methods [26].
The discussion above raises the problem of detecting HGT between closely related species or even strains of the same species, where a strong enough signal for existing HGT detection methods may not exist. This calls for a new distinction of intra-clade HGT in which both donor and recipient organisms are from the same broadly defined lineage, and inter-clade HGT where the donor and recipient are from different, and distant lineages. Such a solution is required when exploring the sudden emergence of drug-resistant pathogenic strains, which most often acquire their virulence from closely related bacteria. In this work we make a first step in this direction and present a novel technique for detecting HGT between closely related species or strains, that we refer to as intra-clade HGT. The technique builds on the concept of—the synteny index (SI) between two genomes (species) that we previously developed [27].
Gene synteny [28, 29] is the conservation of gene order across species along the evolutionary course. Synteny (or lack of) was already employed for defining a distance measure between genomes (species). Under this formulation, two genomes over the same set of genes are viewed as a permutation of one another and the task is to find the minimal number of legal operations to transform one genome to another [30, 31, 32]. Nevertheless the rearrangement distance is irrelevant in the context of a particular gene and therefore cannot be used to detect HGT. In contrast, SI measures how much a gene, orthologous to the two species, is in its “natural place”, or in other words, shares the same neighborhood in both genomes. The two underlying assumptions are that a newly acquired gene is inserted at a random location and therefore with high probability in a new neighborhood and that, closely related species have undergone low level of HGT activity (since they are closely related). We also define the average SI between two genomes that is a weighted average of SI’s and extends the SI from the gene-level to the genome-level. Average SI provides a measure of divergence in a population exposed to frequent HGT activity. Since low average SI is indicative of high divergence (and vice verse for high average SI) [33, 27], we can exploit a gene-specific low SI between closely related species (that exhibit high average SI), to detect potential HGTs for that gene. Hence, the core set of genes shared by two organisms, can be a basis to generate the SI distribution between them where genes of exceptionally low SI are marked as SI HGT candidates. As low SI can be a result of other global genomic rearrangements [34], we need to account for these events (see in Fig 1, genome G2 can equally be resulted by a translocation of gene d from between c and e to between j and k). Here we rely on the constant relative mutability (CRM) property that is a direct product of the Universal Pacemaker (UPM) of genome evolution [35, 36] phenomenon. This property asserts that, in general, and across all lineages of the tree of life, any two genes preserve the same ratio between their respective evolutionary rates. In particular, this measure was tested and validated in bacteria [35, 37], the organisms we analyze here. Using this property, we can calculate the expected distance between the two copies of a gene that SI has indicated to be a HGT candidate in the studied organisms. Using a statistical confidence check, a reinforcement for the HGT hypothesis is obtained. We applied our method to real biological data, the three strains of E. coli that were studied in [38] and were found to exhibit a very high rate of HGT. Understanding and detecting HGT within the strains, could be of great importance, for instance in understanding the origin of pathogenicity of certain pathogenic strains, particularly those whose ancestors were not pathogenic. While [38] focused on inter-HGT among these species by means of codon usage, they could not detect intra-HGTs between the strains themselves. Our method detected several genes with high probability of being horizontally transferred. For a sample of them, we checked for HGT by other complementary methods, such as RIATA-HGT [39] and PhylTr [40], and obtained supporting evidence for our inferences. These results suggest a combined approach in which the lightweight approach Near HGT is first used to detect putative HGTs where the signal is weak (e.g. among strains). Next heavier approaches such as the phylogenetic approaches, are used where the signal is more pronounced or to confirm putative specific events first found by Near HGT.
The method with an accompanying documentation and examples, along with the procedures used for this study is available at http://research.haifa.ac.il/~ssagi/software/nearHGT.zip. Supplementary material used in this study is available at http://research.haifa.ac.il/~ssagi/SI-HGT/suppl.zip
In this section we describe our novel algorithm, Near HGT for detecting putative HGTs between closely related species, and subsequently, results from applying it on a set of E. coli strains.
Since SI is defined for a single specific gene shared by two genomes, we can exploit that property for gene specific studies. As demonstrated in [27], closely related species exhibit high average SI reflecting the fact that their respective genes normally share the same neighborhood. Our underlying assumption is that an acquired gene is inserted in a random location. Hence, between closely related species (and in particular strains of a species), if a gene has exceptionally low SI, we might suspect it has undergone HGT. Indeed looking at the histogram of SI between three strains of E. coli: CFT073, EDL933 and MG1655 in section [Analysis of Real Biological Data] below, reveals very high gene counts at the high SI values (bars at the right end corresponding to SI ∈ [17, 20]) and very low gene counts for the low SI, SI ∈ [1, 5]. The absolute values for these SI distributions can be found at table in S2 Table in the supplementary material. A notable rise is found for SI = 0. We suspect this reflects genes acquired by HGT. Therefore, given some threshold SI value 0 < δSI < 1, we define an SI cutoff C(δSI), such that the fraction of genes g0 for genomes Gi, Gj, SI(g0, Gi, Gj) ≤ C(δSI), is less then δSI. We denote these genes as SI HGT suspected. We note though, that by low SI we cannot distinct between donor and recipient. Moreover low SI is exhibited between the recipient and generally every other genome. Therefore, as we indicate in our real data analysis, when multiple genomes are analyzed, a clearer view is provided.
Next it is important to verify that these genes are indeed the result of a HGT event. This is important as low SI can also be a product of other large scale genomic events: a translocation, an event where a gene moves to a different location in a genome, or a Duplication, a similar event where a copy of the gene remains in the original location.
The following observation follows intuitively from Fig 1.
Observation 0.1. Let G1 and G2 be two genomes sharing a common gene g. Assume g was either translocated or duplicated in G2 (we assume g corresponds to the copied instance rather than the original). Assuming no other large scale genomic events occurred, then with high probability SI(g, G1, G2) = 0.
Indeed, based on SI only, it cannot be distinguished whether a gene has been horizontally transferred or simply translocated within the genome. Therefore we cannot rely on low SI as the sole evidence for HGT. To establish that a gene has undergone HGT we rely on the fact that a translocated (duplicated) gene has resided in its host genome a sufficiently long time since its split from another genome (one belonging to another strain or species), in contrast to a gene recently acquired through HGT. This implies that the translocated gene was subjected to small scale substitutions (such as point mutations) for the time period since its split from the other genome. Hence the inferred distance between orthologous genes in two genomes, is proportional to the time since their divergence.
Therefore, to distinguish an HGT from translocations or duplications, we rely on the fact that a translocated (duplicated) gene has been in its hosting genome since its split from another genome, in contrast to a gene recently acquired through HGT.
We now rely on a very basic evolutionary effect recently demonstrated, dubbed as Universal Pacemaker (UPM) of genome evolution [35, 36]. The UPM principle states that along every lineage in the evolution of cellular life, most genes change their mutation rate in unison, as if adhering to a universal (but lineage specific) pacemaker.
We now observe the basic property, denoted as constant relative mutability (CRM), which we exploit in this part and is a direct outcome of the UPM: For every two genes g and g′ residing in a genome G mutating at (not necessarily constant) rates α and α′, the ratio ρg, g′ = α/α′ is (approximately) constant at all times.
The CRM property can be utilized for our task in the following way. If a gene gh has undergone a HGT between two species s1 and s2, then the evolutionary distance between these very species according to this gene gh has shortened, proportionally to the time of the HGT event. However, since the HGT is unknown, this short distance between s1 and s2 according to gh cannot be attributed with certainty to a HGT event, but rather to conservation of gh, or to the case that gh has slowed its rate along these specific lineages (recall that the evolutionary tree is not known and in particular, this tree according to gh is substantially jumbled). Now, the CRM property comes to play. It manifests that regardless of the characteristic rate of gh, and even if it slowed down, it maintains (relatively) the same ratio to all other gene rates along that lineage. Therefore, the following is done: An additional witness gene gw, and two additional reference organisms r1 and r2 are taken arbitrarily and assume the time separating between r1 and r2 is t(r1, r2). Now, the rate ratio between gh and gw, ρgh, gw is calculated,
ρ g h , g w = d g h ( r 1 , r 2 ) / t ( r 1 , r 2 ) d g w ( r 1 , r 2 ) / t ( r 1 , r 2 ) = d g h ( r 1 , r 2 ) d g w ( r 1 , r 2 ) . (1)
This is the expected ratio that is expected to prevail along all lineages and between any two organisms. Hence the same ratio but between s1 and s2 is now computed and this is the observed rate ratio ρ g h , g w ′:
ρ g h , g w ′ = d g h ( s 1 , s 2 ) / t ( s 1 , s 2 ) d g w ( s 1 , s 2 ) / t ( s 1 , s 2 ) = d g h ( s 1 , s 2 ) d g w ( s 1 , s 2 ) . (2)
Now, by the CRM hypothesis, ρ g h , g w ′ = ρ g h , g w and this is indeed our null hypothesis. As we suspect the “rate” of gh has changed as a result of HGT (we use quotation marks as the rate of gh has not really changed, but rather the time of divergence is different), and hence also the respective observed distance dgh(s1, s2), or for short just dgh. We now set
d g h ′ = ρ g h , g w d g w ( s 1 , s 2 ) , (3)
and denote it as the expected distance between s1 to s2 according to gh.
To decide whether gh has undergone HGT, we use Chi-square significance test between observed and expected values [41]. In our case dgh and d g h′ serve as observed and expected “coin probabilities” respectively, gene length is the coin flips, and we use degree of freedom (DoF) 1 as follows:
χ 2 = ∑ i ( O i - E i ) 2 E i = ( ℓ d g h - ℓ d g h ′ ) 2 ℓ d g h ′ + ( ℓ ( 1 - d g h ) - ℓ ( 1 - d g h ′ ) ) 2 ℓ ( 1 - d g h ′ ) (4)
We refute the null hypothesis, i.e. decree if gh undergone HGT, if the χ2 probability with one degree of freedom is below another threshold value δρ.
Fig 2 describes the situation. At the top, the tree for the reference organisms and the two strains is illustrated with proportional branch lengths. The SI-suspected gene between the two strains S1 and S2 should be compared with respect to the reference organisms. At the bottom left, HGT at the suspicious gene “shortens” the distance between the two strains, violating the constant ratio between rates (or distances).
Example 1. To illustrate the use of our inference rule we show an example from our real data below. The evolutionary model with which we use is the Jukes-Cantor [42] (JC) evolutionary model(While we are aware it is not a realistic model, it serves here only for illustration.).
Let the two strains s1 and s2 be the E. coli strains CFT073 and MG1655 and the reference organisms, r1 and r2, be Bacteroides fragilis and Wolbachia. The HGT suspected gene gh is engA and the witness gene is gmk. We abbreviate for dh(r) for dgh(r1, r2) and analogously for the other cases. The distances obtained are:
dh(s) = 0.0080
dw(s) = 0.0237
dh(r) = 0.583
dw(r) = 0.541
n = 1472.
we get: ρ = d h ( r ) d w ( r ) = 0 . 583 / 0 . 541 = 1 . 077. Now, by Eq (3) we set d g h′ = ρ d w ( s ) = 1 . 077 * 0 . 0237 = 0 . 0255.
However, we have dgh(s) = 0.0080.
We convert the two distances to hamming distance:
h
d
g
h
(
s
)
=
(
3
/
4
)
(
1
−
e
−
(
4
/
3
)
d
g
h
)
=
(
3
/
4
)
(
1
−
e
−
(
4
/
3
)
*
0
.
0080
)
=
0
.
00795
h
d
g
h
′
=
(
3
/
4
)
(
1
−
e
−
(
4
/
3
)
d
g
h
′
)
=
(
3
/
4
)
(
1
−
e
−
(
4
/
3
)
*
0
.
0255
)
=
0
.
02507
Therefore, by Eq (4), our χ 2 = ( 1472 ( 0 . 00795 − 0 . 02507 ) ) 2 1472 ( 0 . 02507 ) + ( 1472 ( 1 − 0 . 00795 ) − 1472 ( 1 − 0 . 02507 ) ) 2 1472 ( 1 − 0 . 02507 ) = 17 . 65
Now, if we set δρ = 0.01 we see that χ2 = 17.65 with one DoF is obtained with probability below δρ and we can infer that the gene has undergone HGT.
There are few cases that we can miss a gene having undergone HGT. As depicted in Fig 2 at the bottom right(marked with yellow dashed line), the SI-suspected gene might have undergone a HGT also between the reference organisms. In that case we will not detect the HGT since the rate ratio is biased in both the strains and the reference genome. It might also be that the witness gene has undergone HGT in the strains (but not in the reference organisms). Here as well the rate ratio is maintained and the HGT will not be detected. Finally, as the strains are evolutionarily close, for many genes, the phylogenetic signal is very weak and does not provide the distinction between HGT and vertical descent. For these reasons the complete algorithm iterates over all possible witness genes and reference organisms. Here is the complete algorithm, Near HGT, for detecting all putative intra HGT genes within a group of species (strains) 𝓢 and a reference set of organisms 𝓡:
Procedure Near HGT(𝓢, 𝓡, δSI, δρ)
for all S1, S2 ∈ 𝓢
for every HGT suspected gene gh ∈ S1 ∩ S2 s.t. SI(gh, S1, S2) < C(δSI)
let ℓ = |gh|
for R1, R2 ∈ 𝓡 s.t. gh ∈ R1 ∩ R2
– for all witness genes gw ∈ S1 ∩ S2 ∩ R1 ∩ R2
* set ρ g h , g w ← d g h ( r 1 , r 2 ) d g w ( r 1 , r 2 )
* set d ′ g h ← ρ g h , g w d g w ( s 1 , s 2 )
* set χ 2 ← ℓ ( d g h − d ′ g h ) 2 d ′ g h ( 1 − d ′ g h )
* if the probability for χ2 with 1 DoF is at most δρ, then mark gh as putative HGT
It is important to note here that since we perform many tests for many witness genes and reference organisms, a correction for multiple hypothesis testing should be performed. We chose the standard Bonferroni correction, considered to be highly conservative, multiplying the bound obtained by the number of tests for a given gene.
We conducted a simulation study to assess the power of the new proposed method. Obviously, the longer the gene the greater the confidence that is obtained (more samples). Similarly, the more recent the event is (closer to the extant species) the stronger the signal. We wanted to show these effects in a simulation study.
In the study we created a random Yule [43] tree over 20 taxa that was used as the species tree. Edge lengths represent the time that passed between speciation events and distribute exponentially (see more details in supplementary text in S5 Text). We chose two pairs of organisms from the tree: r1 and r2 that were used as the reference pair, and s1 and the s2 pair between which the HGT event occurred. We evolved the witness gene gw on the original tree. Then we simulated a HGT event along the path from s1 to the least common ancestor of s1 and s2, LCA(s1, s2). This HGT resulted in a lower ancestor to s1 and s2. Then, the HGT gene gh was evolved on this tree. Both genes evolved on their respective tree, according to the Jukes-Cantor model. The four distances were taken between the resulting sequences at leaves s1, s2, r1, and r2, for both gw and gh. We used the χ2 test (with 1 DoF) to reject the null hypothesis (i.e., no HGT occurred). Every point in the plotted graphs is an average of 20 runs.
Our first study focused on the effect of how recent the HGT event and is depicted in Fig 3 The event’s height signifies how close the event was to the leaves (i.e. recent) as a fraction of the length of the path from the leaves (s1 or s2) to the LCA, LCA(s1, s2), where zero implies HGT at the very leaves, and one—at the LCA. In the figure, gene length is held constant at 70bp while the HGT height varies. The top graph shows HGT identification success rate and the bottom graph shows the four distances (only one distance should change and it is the dgh(s1, s2) when the event height changes). As can be seen the distance between the s1 and s2 according to the gh grows the higher the HGT (closer to the LCA(s1, s2)), while all other distances are not affected, yielding fewer HGT event identifications. HGT identification is perfect until HGT height reaches 0.4 and then starts to drop. However, we still see some significant identification rate of 0.4 even at very high position of the HGT—0.9 where the sequences are almost identical, implying that under “laboratory conditions” such as these, our method is quite effective, even for short gene fragments.
In the bottom graph, we see that the distances between r1, and r2 according to gh and gw are the same, and hence the rates are also equal, while the distance between s1, and s2 according to gh reaches its reciprocal dgw(s1, s2) only when HGT height is one—at the LCA LCA(s1, s2).
Our second study focused on the effect of the length of the transferred fragment and is depicted in Fig 4. Here we set the event height constant at 0.7 and varied only the length of the transferred gene. The simulation parameters remained the same as before. We see from the figure that identification starts even at quite low lengths of transferred fragments, for instance 0.4 identification rate for gene length of 20bp and achieves perfect identification (rate 1) at length 80. We note that event height 0.7 is quite challenging and a better rate is achieved for events closer to the leaves.
Also here the bottom graph in Fig 4 depicts how the four respective distances change as a result of the HGT. Unsurprisingly, distances do not change as a result of the HGT in this experiment. We see that, similarly to Fig 3, the distances dgw(r1, r2) and dgh(r1, r2) are the same since the two rates are the same (and of course the separating time is the same as no HGT occurred). The other two lines, representing dgw(s1, s2) and dgh(s1, s2), do not coincide although mutation rates are the same as HGT did occur between s1 and s2, causing the distance dgh(s1, s2) to shrink. However, as the HGT height is constant, same is that line. It is noteworthy that the misidentification at short gene length is partly due to “incorrect” distances as a result of the stochastic process of gene evolution that we simulate.
Our third study addressed the question of false positive (FP) rate. As HGT is believed to be a stochastic process, our method is subjected to FP errors in the sense of alerting HGT even in the case no real HGT event took place. The first part of the study investigated the effect of sequence length on FP errors. The distance between the organisms was held fixed at 0.2 (i.e. expected number of mutations at a site 0.2). Sequence length grew exponentially from 20bp to 10k. The results are depicted in Fig 5. The second part of the study focused on the effect of the distance between the donor and recipient organisms on FP rate while the gene length is held fixed. The results appear in Fig 6. The figure shows four curves for gene length 40, 640, 2.5k, and 10k bp respectively.
As can be seen in Fig 5, as opposed to the sensitivity (or false negative) case, FP is almost entirely unaffected by sequence length. This is due to the Chi-square property that while the true parameters (distances and hence ρ’s) are estimated more precisely, Chi-square tends to refute the null hypothesis quicker given more data (gene length). In the contrary Fig 6 readily shows that the distance between organisms does affect FP rate. For a very short distance (closely related organisms) the signal is weak and the method is more prone to false alerts (and this holds for any sequence length, in accordance with 4.a). However, as the distance between organisms grows, the signal increases and FP rate declines.
Escherichia coli is the best-studied bacterial species, with much variation between strains, some of which are pathogenic. From an evolutionary perspective, different strains of E. coli exhibit highly diverse gene repertoires, reflecting much gene gain and gene loss. As such, it was of interest to look into three E. coli strain genomes for genes that underwent HGT and by so doing to test our method for detecting HGT between strains of the same species. Here, we used the three well-known and sequenced strains of E. coli studied extensively by [38]: the uropathogenic CFT073, the enterohemorrhagic strain EDL933, and the non-pathogenic laboratory K-12 strain MG1655. In general, all strains of E. coli underwent changes in the ancestral backbones genes at a slow rate resulting in the conserved synteny apparent across strains today. However, the remainder of these genomes is highly variable, probably reflecting numerous independent HGT events along the evolution of the different strains, and tracing back these events is challenging. Studying these three strains, one of which is an extra-intestinal pathogen, the other an intestinal pathogen and the third a non-pathogenic commensal, can shed light on the contribution of HGT to the genome evolution of pathogens.
As a first step we reconstructed the three pairwise S I ¯ 10 ( G i , G j ) values for these three strains. The results are shown in Fig 7 and also in the table at the supplementary material (see table in S2 Table). To get some intuition on these species’ relatedness, their rate of evolution, and ancestry, we reconstructed their phylogeny based on their 16S rRNA genes obtained from the Ribosomal Database Project (RDP) [44, 45]. To root the tree, a related species Escherichia fergusonii was used as an outgroup. The tree (without the Escherichia Fergusonii outgroup) appears in Fig 8. While we are aware that several other works [46, 47, 48] found different topologies over this set (i.e. different rooting), these works used different inputs and methods and also reported on conflicts between themselves. Our tree was built by the accurate maximum likelihood (ML) approach, supported by synteny data as we detail below, and also agrees with the tree obtained using seven housekeeping genes by [49]. We therefore found it sufficient for this part.
From the tree it appears that the strains CFT073 and MG1655 are sister taxa while EDL933 is an outgroup. This is in absolute agreement with our synteny-based findings, reflected in Fig 7 that we explain next. As argued before, high synteny between organisms indicates evolutionary relatedness. Therefore, between closer pairs of species we expect to find more genes with high synteny than between more distant pairs. Indeed, in Fig 7, we see greater numbers of genes with SI ∈ [14–19] for the CFT073- MG1655 pair (the tall green bars in the figure) than for the two other pairs (red and violet bars).
Next we set δSI = 0.05. From the table in S2 Table at the supplementary material, it can be seen that all genes with SI ≤ 5 are SI-based HGT candidates. Hence we applied the algorithm Near HGT for each SI-based candidate gene. The genes found significant for having undergone HGT between each of the three pairs of strains appear in Fig 9. The height of the bars represents the (log) number of witness genes found to testify for HGT of the studied gene. The value −1 indicates that the gene was not found to be an SI-based HGT candidate in the pair of genomes.
Conspicuously, the three most prominent HGT events, detected in a pairwise genome comparisons are supported by almost exactly the same number of witness genes. This may enforce the latter finding as every gene witnessing in one pair of reference taxa, also witnessing in the other pair. Because a gene’s SI values are computed pairwise, when a gene is transferred into a recipient organism, it incurs a low SI not only between the recipient and the donor, but also between the recipient and all other organisms that contain this gene in its original (usually ancestral) location. Hence, in cases when a gene has low SI values in both pairwise comparisons, the organism in the intersection of the two pairs, is probably the recipient. That gene will have high SI values between the other two remaining genomes. Accordingly, the recipient genome is that of strain MG1655 for the genes engA and ribF, and strain EDL933 for gene speG. By our rate check in Eq (4) we can hypothesize regarding the donor organism. In the case of the speG gene, where the strain EDL933 appears in both pairs (that is, in the red and green bars corresponding to pairs EDL933-MG1655 and EDL933-CFT073 in Fig 9. respectively), the event could have occurred before the MG1655- CFT073 split (See the 16S rRNA tree in Fig 8), or after the split. Both scenarios yield low SI and also unexpected rate (distance) decrease at both sister strains MG1655 and CFT073.
The case of the engA gene is more complicated. Here the recipient is the strain MG1655, which causes low SI with both EDL933 and CFT073. However, the rate check found this gene significant for both pairs MG1655- CFT073 and MG1655- EDL933. It cannot be that the distance to both species became shorter. Indeed a BLASTN search for the engA gene at the strain MG1655 in the nr database at NCBI revealed that the closest homolog is present in Shigella flexneri (See BLAST output file in S1 Fig in the supplementary material). We can infer that the engA gene was transferred to the strain MG1655 from an organism that was not included in the 3 strain set we investigated (in this case from a close relative of Shigella flexneri), causing an unexpected increase (as opposed to decrease) in distance as evidenced in the rate check algorithm.
In terms of nucleotide composition. these three genes have a composition that is far from striking—with G+C% of 46.34%, 53.6% and 52% for speG, ribF and engA respectively, similar the the E. coli genomic average, and confirming the hypothesis they were transferred from a recently diverged taxa. Conceivably, such similar composition is unlikely to be picked up by composition-based HGT-detection methods.
Finally, genes with only a single bar in Fig 9, may indicate existence in only that pair of organism (specifically the case of genes ydaO and cspB)
Since our approach relies on new ideas that were not explored before in the realm of HGT detection, we set to compare our approach with representative existing HGT methods.
To substantiate the set of detected genes and allow reliable application of the phylogenetic method, we added to the strains analyzed above five more strains of E. coli: Enteroaggregative E. coli 042 (denoted 042 below), uropathogenic E. coli 536 (denoted 536), enterotoxigenic E. coli W (denoted w), enterohemorrhagic E. coli O157:H7 str. TW14359 (denoted TW14359) and enteropathogenic E. coli O55:H7 str. CB9615 (denoted CB9615). The HGT events detected when applied to the entire data (including the previously described strains), containing the eight strains, are shown in Fig 10. A list of these genes, sorted by incongruent pairs and number of witnesses is given in table S2 Table in the supplementary material.
We start with the phylogenetic approach. This approach concentrates on a specific gene and contrasts its history (phylogeny) with the species history. As was shown in the three strains analysis in Section [Analysis of Real Biological Data], a single HGT event may yield synteny incongruence between several pairs of taxa. Therefore, when working with multiple species, our approach highlights “incongruent pairs” of species that may result from one single HGT event. A closer inspection of the kind done in Section [Analysis of Real Biological Data] can reveal the source and target of the event. In this part we chose two genes that were detected as putative HGT-derived with significant support by our method but are also present in all selected strains, and additionally, perform important functions within the bacterial cell: valS and speG.
valS ([50, 51]) is a Valyl-tRNA synthetase, an amino-acyl tRNA synthetase which catalyzes the attachment of valine to tRNA(Val). tRNA amino-acyl synthetases have been shown to frequently being horizontally transferred in evolution[52].
speG([53]) is spermidine N1-acetyltransferase (SAT) which regulates polyamine concentration by its degradation, and is involved in the prevention of spermidine toxicity at low temperatures in E. coli[54]. Detoxification functions are often horizontally transferred across bacterial species [55].
We tested the speG and the valS genes for HGT within the eight E. coli strains using two phylogenetic methods: RIATA-HGT [39] and PhylTr [40].
RIATA-HGT [39] is a relaxed version of a problem of minimum-cardinality [56] which looks for the minimum number of HGT events (SPR moves, see [57]) occurring on a given species tree S which give rise to a given gene tree. As the problem is NP-hard, RIATA-HGT is a heuristic for that problem that runs in polynomial time but was found to provide fairly accurate results [39].
In order to use RIATA-HGT, a species tree based on 16S rRNA gene and two gene trees based on valS and speG genes, were constructed. Next we applied RIATA-HGT over the three described trees. Examination of the RIATA-HGT results for valS gene (Fig 11) reveals two HGT events, while our method detected twelve incongruent pairs. While a single HGT event may yield several incongruent pairs, careful inspection of the pairs in Fig 11 gives rise to at least three events. For speG gene, RIATA-HGT detected three HGT events (Fig 12), largely in agreement with our incongruent pairs findings.
The other phylogenetic method is PhylTR [40] that reconciles the incongruence between given species and gene trees. The chosen reconciliation is the one with a minimum number of gene duplications, losses, and lateral transfers. This method defined the DTL-scenario (Duplication-Transfer-Loss scenario), which is a formal equivalent of a reconciliation. A scenario explains how a gene tree has evolved within a species tree using duplications, HGTs, and losses. The output of this method is the trees with the most parsimonious (MP) DTL-scenarios. Applying this method (with its built-in parameter values) to our data (the 3 trees described earlier—a species tree and two gene trees) yielded the following results: valS—one MP tree was found with two HGT events; speG—nine MP trees were found with between one to three HGT events. In contrast, the Near HGT method was applied to all (82) pairs and found eleven incongruent pairs for valS, and ten incongruent pairs for speG. This result indicates that HGT event took place. However, further analysis as was done for the three strains (Section [Analysis of Biological Data]) for determining donors and recipients and number of events was not performed here.
Sequence composition based methods [6, 58, 59, 60, 5] rely on the fact that certain genomic characteristics have wide variation across different bacterial species. Therefore, genes from alien origins (i.e. that were transferred horizontally) exhibit different characteristics than the typical genomic one. The characteristics that are normally investigated are the frequency of certain “words” in the genome. In order to detect such alien, atypical segments, methods work by applying a sliding window approach, in which the characteristics inside the window are constantly compared to those of the whole genome. When a significant difference between the window’s characteristics and those typical to the entire genome is found, it is reported as HGT suspected. However, this distinction between “alien” segments and the prevailing genome characteristics, normally “fades” throughout the time due to the phenomenon of amelioration [58] in which the acquired segment is adapted to the host’s genomic composition.
HGT-DB [60] is a genomic database that combines statistical parameters such as codon and amino-acid usage as well as G+C content and information about which genes deviate in these parameters from the complete prokaryotic genome. A gene is declared as HGT if it deviates by more than 1.5 standard deviations from the mean (i.e. genomic) values [22]. Additionally, there are also minimal length requirements for a transferred segment.
The HGT-DB contains only three out of the eight strains: CFT073, 536 and EDL933. In addition, out of all genes detected by SI, only cspB was reported as HGT in CFT073 by HGT-DB. Since segments transferred between closely related strains cannot differ too much from their host, there is no wonder that only one gene was found.
In order to apply general sequence based criteria for HGT to the genomes under study, we pursued the following general procedure [61]. For a given word length ℓw and a segment S, the Sℓw-spectrum is a 4 ℓ w dimensional vector holding the relative frequency of every ℓw long word in S. For a window I (a segment of a pre-determined length along the genome), we compute the Euclidean distance between Iℓw-spectrum and its host genome’s spectrum. This defines a distribution over the distances pertaining to the various windows along a genome. For a 0 < δ < 1, we say that a window I is δ-atypical if its distance to the genome is greater than 1 − δ fraction of all the other distances (i.e. a p − value of δ). We note that for a genome with a uniform (or any other constant) distribution over the words, if window sizes are large enough, then no window will be atypical. According to the law of large numbers, every window will have very similar spectrum to the genome’s spectrum, and no window will be more distant than 1 − δ fraction of all the other distances.
We implemented this approach for dinucleotide [62], trinucleotide [63] and tetranucleotid content [64] (i.e. ℓw = 2, 3, 4). G+C content was implemented using a 2-dimensional vector holding the frequency of G+C versus A+T. Window size was set to 2000 bp and the window was moved along the genomes in steps of 10bp. We constructed the respected di-, tri-, tetra-spectra of each of the eight strains, and checked each of our suspected genes if it is 0.05-atypical. In all our tests, only one gene was found (by the tri-nucleotide experiment).
Concluding this part, comparing the Near HGT method with a variety of HGT detection methods, we found out that Near HGT extends, sometimes significantly, the other methods. The difference originates from the fact that between closely related species it is much harder to detect HGT events. On the other hand, composition-based methods facilitate detection of singleton/orfan horizontally acquired genes, as the rate check of Near HGT (but also phylogenetic methods) needs a genome related to the donor. For the phylogenetic methods, when reconstructing phylogenetic trees of closely related species any difference between the trees is hardly seen, even if they are not based on a conserved tree. Another source for lack of sensitivity in the phylogenetic approach, is that most of these methods are NP-hard [56] and therefore use heuristics [39] with no real guarantee on the results returned. As was shown here, Riata-HGT and PhylTR detected only a fraction of the HGT events found by Near HGT.
On sequence composition-based grounds, when a gene is transferred within closely related taxa, their genomeic signature is naturally highly similar, making atypical composition impossible to detect. Therefore, we observed poor sensitivity by the sequence-based methods of HGT detection, unlike the efficiency of Near HGT.
In this work we have exploited the notion of synteny index (SI) [27] that is useful in settings of inter-species recombination to devise a novel approach, Near HGT, to detect HGT between closely related taxa. We first applied it to three strains of E. coli and subsequently to five more (a data set of eight strains in total) and found several genes highly suspected of having undergone HGT. Our method also provides indications regarding the donor and recipient lineages by phylogenetic analysis as we demonstrated in the case of the three strains.
HGT between closely related organisms is a domain that is not covered by existing HGT methods as the signal available to these methods is very weak in this particular case. The method applies two stages of HGT detection. The first stage relies on synteny conservation between the species and discovers genes with unusual location. The second stage, exploits the key property of relative rate conservation that is maintained across species [35]. If a gene is found to exhibit both low synteny conservation with respect to another species, and also a significant deviation from the rate conservation, it is considered a validated HGT candidate.
Near HGT may shed light on recent gene acquisition events between related organisms, possibly only recently diverged. Identifying such events is important for the study of evolution as well as for molecular epidemiology. The latter field will benefit greatly from a more sensitive reconstruction of the emergence of virulent, often drug-resistant, strains. In the future this method will be applied to additional organisms and strains, for which genome sequences are available and integrate it with existing approaches for HGT detection so that cross validation and accurate tracing of the donors and recipients are facilitated.
We now define our working model that will serve to locate HGT between genes. A genome is a sequence of genes (g1, g2, …, gn) and each gene is a sequence of DNA letters. That is, our view of a genome is at a resolution of genes, and of a gene at a resolution of nucleotides (See Fig 13.).
The k-neighborhood of a gene g0 in genome G, Nk(G, g0) is the set of genes at distance at most k from g0 in G (i.e. at most k genes upstream or downstream). The conservation of gene order between two genomes is called synteny. Let g0 be a gene common to two genomes Gi, Gj. Then the k synteny index (k-SI), or just SI when it is clear from the context, of g0 in Gi, Gj is the number common of genes in the k neighborhoods of g0 in both Gi and Gj: SI(g0, Gi, Gj) = |Nk(Gi, g0) ∩ Nk(Gj, g0)|. For the sake of completeness, for g0 ∉ Gi ∩ Gj, SI(g0, Gi, Gj) = 0. See Fig 14 for illustration.
Given two genomes G1, G2, and let 𝓖 be the set of genes in at least one genome, 𝓖 = G1 ∪ G2. Then the average k-SI between G1 and G2 is defined by
S I ¯ k ( G 1 , G 2 ) = 1 | 𝒢 | ∑ g ∈ 𝒢 S I k ( g ) 2 k . (5)
We observe that for two identical genomes, S I ¯ k ( G 1 , G 1 ) = 1 and for two genomes with disjoint sets of genes S I ¯ k ( G 1 , G 2 ) = 0. The average SI gives us a measure of similarity between pairs of species.
A genome undergoes events of gene gain and loss in which genes are added or removed respectively. As we are focused in the core set of genes that are common to two organisms, we are not interested in the latter processes. Every gene undergoes a process of sequence evolution according to some stochastic evolutionary model [65]. The evolutionary model we consider is such that the nucleotides along a gene are identically and independently distributed (IID). The value of the nucleotide is the state (we sometimes use just “nucleotide” to denote its state). A single mutation (or point mutation or just a mutation for short) is the event of a nucleotide changing its value to a different one. An evolutionary model 𝓜 models the (stochastic) process of mutations occurring at a site as a function of mutation rates αi, j modeling the rate of transitions from state i to j, and a specified time period t. We use the transition notation in the context of Markov chains and note that it has nothing to do with the type of mutation bearing the same notation (see [65] for more details). Given 𝓜, mutation rates [αi, j], and a time period t, the transition probability pi, j from nucleotide i to j during t is uniquely defined by an appropriate function (determined by 𝓜). An evolutionary model 𝓜 is said to be time reversible if it is not possible to determine the direction of time given two states of a nucleotide, separated by a time period t. The evolutionary distance (or mutation distance or simply distance), d(s1, s2), is the number of mutations separating between two homologous sequences s1 and s2. The Hamming distance h(s1, s2) between two homologous sequences counts the number of sites with different states. Using the model 𝓜 we can convert between the two distances. These distances are usually normalized by the length of the sequences and are normally denoted by d and h respectively. As every gene exhibits a different distance between the respective sequences, we use the gene as a subscript in the distance notation, e.g. dg(s1, s2). In the Results section, we used the simple Jukes-Cantor [42] (JC) evolutionary model for illustration.
A horizontal gene transfer (HGT) is the event in which a gene of a genome, the donor genome, being copied and inserted at some (random) position at another genome, the recipient genome. Since we view the genome as a sequence of genes (see Fig 1), the new gene is always between two genes (or at the ends of the genome). colorblack By the assumption of randomness we expect the gene to have a new neighborhood.
All genomes analyzed were downloaded from the NCBI microbial genomes resources [66] (http://www.ncbi.nlm.nih.gov/genomes/lproks.cgi). Appropriate 16S-rRNA genes were downloaded from the Ribosomal Database Project (RDP) [44, 45]. RDP provided two sources for trees, namely a distance based, ready made tree for selected organisms and pre-aligned sequences, based on rRNA secondary structure alignment, that are available from RDP for further independent comparative analysis (including phylogenetics). As maximum likelihood (ML) reconstruction is considered more reliable than distance based analysis, we chose to use the aligned sequences.
The names and order of genes were extracted using RefSeq annotation [67] as it provides an easy to use source of such data, especially for the well-annotated E. coli genomes.
The gene trees for genes speG and valS (see Figs 11 and 12) were obtained as follows. Gene sequences for the eight orthologs were extracted from the GenBank sequences and aligned using ClastalW [68].
All phylogenetic reconstruction (including the 16S rRNA was done using ML reconstruction under the GTR + Gamma evolutionary model (designed for sequences with significant between-site rate heterogeneity). We used the PhyML software [69] to build tree from the aligned sequences (with the parameters indicated above).
|
10.1371/journal.pcbi.1000665 | Predicted Functions of MdmX in Fine-Tuning the Response of p53 to DNA Damage | Tumor suppressor protein p53 is regulated by two structurally homologous proteins, Mdm2 and MdmX. In contrast to Mdm2, MdmX lacks ubiquitin ligase activity. Although the essential interactions of MdmX are known, it is not clear how they function to regulate p53. The regulation of tumor suppressor p53 by Mdm2 and MdmX in response to DNA damage was investigated by mathematical modeling of a simplified network. The simplified network model was derived from a detailed molecular interaction map (MIM) that exhibited four coherent DNA damage response pathways. The results suggest that MdmX may amplify or stabilize DNA damage-induced p53 responses via non-enzymatic interactions. Transient effects of MdmX are mediated by reservoirs of p53∶MdmX and Mdm2∶MdmX heterodimers, with MdmX buffering the concentrations of p53 and/or Mdm2. A survey of kinetic parameter space disclosed regions of switch-like behavior stemming from such reservoir-based transients. During an early response to DNA damage, MdmX positively or negatively regulated p53 activity, depending on the level of Mdm2; this led to amplification of p53 activity and switch-like response. During a late response to DNA damage, MdmX could dampen oscillations of p53 activity. A possible role of MdmX may be to dampen such oscillations that otherwise could produce erratic cell behavior. Our study suggests how MdmX may participate in the response of p53 to DNA damage either by increasing dependency of p53 on Mdm2 or by dampening oscillations of p53 activity and presents a model for experimental investigation.
| A Molecular Interaction Map (MIM) akin to a circuit diagram of an electric device can give a comprehensive view of cellular processes and help understand complex protein functions in cells. To this end, we generated a MIM focused on the p53-Mdm2-MdmX network proteins and performed computer simulations to help understand how Mdm2 and MdmX may regulate p53. Proper regulation of p53 is important for cell survival: elevated levels of p53 can lead to cell death, and decreased levels of p53 can lead to cancer. Mdm2 and MdmX are structurally homologous proteins that regulate p53. Mdm2 negatively regulates p53 by degradation, but MdmX regulation of p53 is not well understood. Recently, Mdm2 and MdmX have been recognized as potential cancer therapeutic targets. In an effort to better understand how MdmX can alter the p53 activity under various conditions, we used mathematical models based on the MIM network to generate hypotheses that can be tested by experiments. Our simulations suggest that MdmX may increase the dependency of p53 on Mdm2 or dampen p53 oscillations during DNA damage response.
| The transcription factor p53 is a tumor suppressor that causes cell cycle arrest or apoptosis in response to stress signals [1]. Loss of p53 function by mutation or by disregulation often leads to cancer [1]. Excessive p53 protein results in premature aging [2] and cell death [1]. Thus, maintaining appropriate levels of p53 is essential for cell survival. Mdm2 and MdmX are structurally-related p53-binding proteins that play a key role in regulating the level of p53 [3]. Although several models of the p53-Mdm2 network have been proposed [4]–[14] and reproduced experimentally observed oscillatory or pulsatile behaviors, those models did not include MdmX and did not address potential impacts of MdmX on dynamics of the p53-Mdm2 network. MdmX can interfere with p53-Mdm2 negative feedback loop by interacting with those two molecules. As indicated by diverse effects of MdmX in biological experiments [15]–[18], it is not intuitively obvious what the consequence of that interference would be. In this work, we examined mathematical network models to address the role of MdmX in regulating the dynamics of p53 activity in relation to Mdm2.
Mdm2 is a ubiquitin ligase that negatively regulates p53 by promoting ubiqutin-dependent p53 degradation [3]. In response to DNA damage, Mdm2 is transcriptionally induced by p53 generating a delayed negative feedback loop between p53 and Mdm2 [19]. Mdm2-null mice die during the embryonic stages due to apoptosis induced by elevated p53 activity [19]. MdmX is structurally similar to Mdm2 and interacts with p53, but MdmX lacks ubiquitin ligase activity [20]. However, MdmX is critical for negative regulation of p53 as indicated by p53-dependent embryonic death in MdmX-null mice [21]–[23]. MdmX can bind Mdm2, but the effects of the ubiquitin ligase activity of Mdm2∶MdmX heterodimer on p53, if any, is not well understood. The role of MdmX in p53 regulation has been suggested to be either (1) a negative regulator, functioning as Mdm2 cofactor to enhance Mdm2-dependent p53 degradation [15], (2) a stabilizer that increases the level of p53 [16],[18], or (3) a positive regulator of p53 activity under stress condition [17].
In our model, MdmX is a key component. MdmX interacts with p53 or Mdm2 to form transcriptionally inactive p53∶MdmX or enzymatically inactive Mdm2∶MdmX. In order to model the interactions of MdmX with Mdm2 and p53, we selected a system of elementary processes that focused on the non-enzymatic interactions between MdmX and other molecules. Since the values of many kinetic constants for our model are unknown, we selected initial kinetic parameter sets to explore potential interesting behaviors. Then, we searched for regions of parameter space that showed biologically interesting behaviors and reproduced previously observed dynamic behaviors, including DNA damage induced oscillations [9] as well as the effects of Nutlin-induced inhibition of p53-Mdm2 binding [24]. Simulations of the model showed that simple binding interactions with p53 or Mdm2 by MdmX generated remarkably complex effects on the dynamics of the p53-Mdm2-MdmX network, such as amplification of p53 activity and damping of p53 oscillations.
To lay the foundation of a model, we first assembled the known molecular interactions among p53, Mdm2, and MdmX in the form of a heuristic MIM using the previously described notation [25]–[27] (Figure 1). The heuristic MIM organizes information from which an explicit model (Figure 2) for simulations was extracted and portrayed as an explicit MIM (Kohn 2001). To facilitate understanding, the model was also represented as an informal diagram (Figure S1) [28]. Additionally, SBML model was provided in the Supporting Information (Text S8).
Four pathways of p53 regulation, functioning simultaneously, are highlighted in Figure 1 and summarized as follows. (Details and references for each interaction are provided in an annotation Table S1. The heuristic map with annotations can be found at http://discover.nci.nih.gov/mim)
After specific behaviors of interest were identified from the initial simulations, a wider range of kinetic parameter space was surveyed with the full model (Figure 2-B) using a previously described algorithm [29] to map the variability observed in initial simulations and to test validity of the predictions made by the simple model (Figure 2-A) (see Methods). Fitness scores were designed to represent how well the model can regenerate the oscillatory patterns [9], or total protein ratio and Nutlin response data [24] (see Methods).
In this search, we found regions of parameter space that exhibit oscillations and Nutlin responses that are consistent with experimental observations, and used the resulting parameter sets to simulate the effect of MdmX. These surveys were performed in two steps (see Methods). In the first step, we searched for values of 15 kinetic constants that simulate previously published average oscillatory patterns [9]. After the first step, the kinetic parameter space was visualized in two dimensions using the first and second principal components [30] (Figure 8). Two oscillatory regions (OSC1 and OSC2 in Figure 8) that gave good fits for the average oscillatory patterns were identified. In the next step, two data points were selected in each oscillatory region (OSC1P1, OSC1P2, OSC2P1, OSC2P2 in Figure 8), and these points were used as center points for the second search to fit the Nutlin response [24]. In this second search, 12 additional kinetic parameters were optimized using additional constraints.
Nutlin is a drug that blocks p53 and Mdm2 interactions, and it was shown that Nutlin induces basal transcription activity of p53 without phosphorylation [24]. When this second search was initiated from OSC2, kinetic parameter sets that fit Nutlin response were found easily. 95% of searches initiated from OSC2 gave good fits (scored≥0.25) (Table 2). However, when the search was initiated from OSC1, less than one percent of the searches gave good fits (scored≥0.25) (Table 2). The kinetic parameters that gave good fits were selected for further simulations to evaluate the role of MdmX in regulating p53 activity during early and late DNA damage responses to be described below. Simulated Nutlin responses using two example sets among the selected kinetic parameters are shown in Figure S5. Simulated oscillations and average oscillation patterns are compared for one example kinetic parameter set, and the simulated oscillations show approximately previously observed a 6 hour interval (Figure S6).
Sensitivity analyses were performed for two selected sets of kinetic parameters that were used for further simulations. The normalized local sensitivity of mRNA (species 15 in Figure 2-B) with respect to various parameters was calibrated during an early DNA damage response (Figure S7), and the time integral of the local sensitivity values were used to rank order the parameters based on their mRNA sensitivity (species 15 in Figure 2-B) (Table S3-A,B). In general, the two kinetic parameter sets showed similar levels of sensitivities. In both sets, the most sensitive kinetic parameters were parameters involved in p53 oligomerization and were involved in the production and degradation of p53 and Mdm2. In the kinetic parameter set derived from OSC1, parameters relevant to the production and degradation of p53 and Mdm2 were most sensitive; in the kinetic parameter set derived from OSC2, parameters relevant to p53 oligomerization were most sensitive.
In order to investigate the role of MdmX in the p53-dependent response to DNA damage, we prepared a heuristic MIM of the pathways leading from DNA damage to p53 activity (Figure 1), from which we selected interactions for an explicit MIM (Figure 2) that defines the network to be simulated.
The known pathways of p53 regulation are highlighted in four panels in Figure 1. The pathway steps highlighted in Figure 1 were all included in the simulation model displayed in Figure 2, except for the following:
A recent mouse study suggested that p53 phosphorylation may be less important for stabilization and activation of p53 than previously thought [20]. This however remains consistent with our model, in which stress-dependent modification (phosphorylation and/or ubiquitylation) of p53, Mdm2, and MdmX operate coherently to Sstabilize and activate p53.
For simulations, two versions of the network model were used: a simple network model (Figure 2-A) and a full network model (Figure 2-B). These two versions differ with respect to the basal transcription activity of p53 (k34, k35, k36), p53 induced Mdm2 production (k33, k37, k38) and mRNA (species 15): the simple network model does not include the p53 basal production rate and the mRNA species (in that case, species 14 directly generates Mdm2, species 3). Initial simulations were performed with the simple network model (Figure 2-A), because it allowed us to interpret simulation results more easily. The biologically interesting network behaviors, found from the initial simulations, were further surveyed in a wider range of kinetic parameter space using a full network model including basal p53 activity and mRNA (Figure 2-B).
Reliable values for rate constants that would apply to the peculiar and diverse conditions existing in cells are elusive, because local molecular crowding in different regions of the cell can substantially affect the thermodynamic parameters [32], and conditions are likely to vary with time and place in the cell. That difficulty exists for the cell as well as for the computer modeler, and may be ameliorated for both if the networks operated in a somewhat digital mode, that is to say with switch-like behavior. Then connections in the network would tend to be either on or off, and quantitative degree of function would be less of an issue. In fact, switch-like behavior has been observed in biological systems [33]–[35]. The problem could also be mitigated if the range of parameter values existing in the cell were in a region of parameter space where function is not much changed over local regions in the parameter space. Network function would then be robust in being relatively invariant over a region of parameter space. Moreover, functional invariance over a region of parameter space would be a condition relatively easy for evolution to access, and therefore would be more likely to exist in modern cells. Based on these considerations, we focused our attention on switch-like behavior and exploration of parameter space. We have also centered some of our studies at plausible sets of parameter values, derived from published experimental data.
The switch-like behavior induced by MdmX in our simulation was dependent upon the heterodimers p53∶MdmX and Mdm2∶MdmX. When DNA damage-induced phosphorylation perturbs the system by depleting free p53, Mdm2, and/or MdmX, the heterodimers dissociate to compensate those depletions. The dissociation of the two MdmX heterodimers, p53∶MdmX and Mdm2∶MdmX, have opposite effects on p53 activity; therefore, the net effect on p53 activity depends on the relative level and contribution of those heterodimers. In initial simulations, switch-like behavior occurred when p53∶MdmX (species 10) became a major reservoir during the pre-equilibration with low Mdm2 or when Mdm2∶MdmX (species 9) became a major reservoir during the pre-equilibration with high Mdm2 (Figure 11-C, D). Figure 11 illustrates the idea using an initial model. Most simulations (over 80%) using the fitted parameter sets with a full model also generated Mdm2 dependent MdmX effects suggestive of reservoir-based amplification during an early response.
Unlike some previously reported switch-like behaviors arising from steady states [34],[36],[37], memory-less switch-like behavior was observed during transient response in our models. This kind of switch may usefully generate rapid transient responses to perturbations when the system has not had time to reach steady state. The switch-like dependence of p53 activity on Mdm2 may produce different degrees of susceptibility to DNA damage in various cell states and cell types. The question of whether such reservoir-based switch-like behavior is employed during response to DNA damage could be addressed by measuring p53 activity at different levels of Mdm2 and MdmX early during the response.
Previous studies showed sustained oscillations of p53 during the long-term response to DNA damage [9], [38]–[40]. In the mathematical models described here, either time delay or TR3-mediated positive feedback was required to display sustained oscillations, consistent with a previous study [10]. TR3 promotes Mdm2 self-ubiquitination in a p53-dependent manner [41], and the relation may form a positive feedback loop. The feedback loop via TR3 was implied by k14 in Figure 2. Our initial study with a set of kinetic parameters that generated oscillations showed that increased level of MdmX can dampen oscillation via non-enzymatic interactions. The reason may be that the concentrations of the unbound p53 and Mdm2 were buffered by the p53∶MdmX and Mdm2∶MdmX heterodimer reservoirs. When some kinetic parameters were varied, sometimes p53 activity converged to a stable steady state without oscillations; in this system with a stable steady state, MdmX transiently affected p53 activity in either a negative or positive manner, but eventually p53 activity converged to the same stable steady state. The simulation with full model also showed damped oscillations with the increased MdmX with some sets of parameters. Interestingly, with some kinetic parameter sets, transient increases of MdmX levels dampened oscillations over an extended time period, with sustained quiescent intervals following the reduction of MdmX to prior levels. The bifurcation diagram of p53 activity as a function of MdmX basal production rate suggests that the transient change in the MdmX basal production rate takes the system from conditions near a stable oscillatory state to a stable steady state. The suppression of oscillations following the return to the original production rate is likely a ‘memory effect’ in which the system tracks back along the steady state solution branch, ending in the vicinity of an unstable equilibrium point where the oscillatory instability is slow to build up.
Kinetic parameter space was explored in two stages. (1) During the first stage, two oscillatory regions, OSC1 and OSC2, were identified. The characteristic difference between the OSC1 and OSC2 clusters was the phosphorylation rate constant (k3 in Figure 2). Narrow distributions of k2, k10, and k37 were common in both clusters, which may mean that oscillatory behavior is sensitive to the degradation rate of p53, Mdm2, and mRNA (see Figure S10). (2) In the second step, two parameter sets from each region were selected as center points to search neighbor parameter space to fit experimental data with a new fitness function. The searches showed marked difference depending on the initial center point in their success rates (a ratio to converge to the second criteria, Nutlin response and total protein ratios in MCF7 cells), and OSC2P1 and OSC2P2 were better center points than OSC1P1 and OSC1P2. Notable, only when the model was simulated with the parameter sets derived from OSC1 were oscillations dampened by MdmX. Thus there may be two separate clusters of kinetic parameters corresponding to p53 oscillations during the late response to DNA damage: kinetic parameter sets derived from OSC1 are MdmX-sensitive, while kinetic parameter sets derived from OSC2 are MdmX-insensitive. Since negative regulation of p53 oscillations by MdmX correlates with parameter sets derived from OSC1, it is tempting to conclude that models defined by these kinetic constants best describe the regulatory behavior of the p53-Mdm2–MdmX network. However, it is not obvious how nature could have evolved such a finely tuned set of parameters considering the small success rate (<0.5%) in fitting Nutlin and total protein ratio data.
The two groups of parameter sets derived from OSC1 or OSC2 underlie MdmX-sensitive and MdmX-insensitive models. It is likely that physiological condition will be explained better by one of those kinetic parameter sets. Experiments to test whether the over-expressed MdmX can in fact dampen DNA damage induced p53 oscillations will allow us to discriminate between the two models.
Our theoretical study showed, with little prior knowledge, potentially interesting roles of MdmX in p53 regulation. Many of the kinetic constants used in the study were not biologically constrained, and some of the assumptions made in the models need to be biologically validated. The lack of prior knowledge demands rigorous experimental validation of the simulated results; however, it also allows the unbiased investigation of possible functions that are not intuitively obvious. The MdmX roles predicted in this study are not simply linear increases or decreases of p53 activity dependent upon the level of MdmX; the study suggests that manipulation of MdmX to alter p53 activity requires careful investigation of the dynamics of three proteins: MdmX, Mdm2, and p53. An experimental system with inducible MdmX and Mdm2 and a reporter protein to monitor p53 activity should allow us to test whether the switch-like behavior (increased dependency on Mdm2 by MdmX) or dampening of oscillations are observed. To observe the switch-like behavior, transient p53 activity has to be monitored, and it may require monitoring the p53 activity relative early (10, 30, 60, 120, 180 min) after DNA damage. To observe dampening of oscillations by MdmX, a long period of observation (>12 hr) will be necessary since the known peak-to-peak interval is approximately 6 hours. As shown in one example plot in the result section with the full model, the degree of level change in the switch-like effects by the increased MdmX was approximately 30%∼50% increase (low Mdm2) or decrease (high Mdm2) of p53 activity, and that level of change may require high sensitivity to measure p53 activity in experiments.
We used highly simplified p53 models in this study because we can interpret the simulation results more easily. However, p53 is involved in many other regulations and feedback loops, and it is not obvious whether the conclusions will still hold in a larger p53 system. Although experimental verification will partially address the question, we could also proceed theoretically by introducing additional components successively to the current model and by surveying parameter space with an updated fitness function. The extension of the current model might include Wip1, which is known to trigger recurrent initiation of ATM activity in some cells [13], detailed phosphorylation and dephosphorylation of molecules, degradation of p53 by ubiquitination steps, and additional interacting partners with Mdm2 and MdmX. Those interactions may introduce additional switch-like steps [42],[43], and it may be interesting also to investigate how dynamic behaviors change when these multiple switches operate together in the same system.
One goal of this study was to understand how non-enzymatic interaction of MdmX with p53 and Mdm2 can regulate p53 during the response to DNA damage. The mechanism of p53 regulation by MdmX is poorly understood, at least in part because the role of Mdm2∶MdmX heterodimer in p53 ubiquitination and degradation is not clearly understood. It was previously speculated that p53∶MdmX heterodimer may serve as a reservoir to maintain a pool of p53 [44]. The findings reported here suggest how such biological reservoirs, which lack any enzymatic activity, can account for previous studies that proposed negative regulation [21]–[23] and various effects [16]–[18] observed on p53 activity by MdmX. Furthermore, the results suggest that heterodimer reservoirs can differentially alter dynamic behaviors during short- and long-term responses to system perturbations. Lastly, this study uncovers a new potential mechanism for inducing a switch-like behavior of p53, which operates transiently in response to cellular events such as DNA damage and depends on heterodimer reservoirs. These results can guide future experiments to elucidate mechanisms by which Mdm2 and MdmX may regulate p53 responses to DNA damage.
First, a heuristic molecular interaction map (MIM) (Figure 1) was constructed by using a previously described notation [25]–[27]. Next, we extracted from the heuristic MIM a sub-network in the form of an explicit MIM that defines a computer simulatable model (except for parameter values and initial conditions) (Figure 2). In an explicit map of the model, all interactions and contingencies are displayed as molecular association/dissociation or as stoichiometric conversions and can be represented as reactions. A list of reactions obtained from the explicit map (Figure 2) was used to write ordinary differential equations (ODEs) (Table S4), which mathematically describe the p53-Mdm2-MdmX network model, based on mass action law. The model was simulated by numerically solving the ODEs (using ODE15S or DDE23 functions) in Matlab.
The kinetic constants used for initial simulations were generated based on collected kinetic constants from previously published models [8],[10],[39] (see Table 1 and Table S5). For kinetic constants whose values were not available from previous models, values were chosen randomly based on the assumptions in Table S6. Although reported qualitative behaviors of molecular interactions support some of the assumptions in Table S6, many of them were assumed for simplicity. Some of the kinetic constants that were available from previous models differed over a wide range and did not represent equivalent processes. We therefore chose values randomly within or near the range of collected values to generate an initial set of kinetic constants. We simulated the network by varying one or more kinetic constants from the initial set to understand the effects on network dynamics. The assumptions in Table S6 were relaxed when the kinetic constants were varied, and the used kinetic parameter set for initial simulations are presented in Table 1.
As input, rate constants of phosphorylation (k3 = k8 = k17) and basal production rate of MdmX (k15 in Figure 2) and Mdm2 (k6) were varied. As output, p53 activity was quantified as p53-dependent promoters occupied by p53 tetramers (species 14 in Figure 2). In these initial simulations with the simple model (Figure 2-A), p53 tetramers were assumed to be the only form of p53 that can bind promoters (tetramer-occupied promoters are represented by species 14 in Figure 2). In the model used here, MdmX modified transient dynamics of p53 activity without altering the p53 steady state activity, as can be deduced from the equations and verified in the numerical solutions. Output was therefore collected at a fixed time point regardless of whether or not the system had reached a steady state. Simulation results with the simple model were presented as [AU] because the kinetic parameters were not fit to any experimental data.
We considered that DNA response may display bi-phasic responses: early and late responses. During initial simulations with the simple model (Figure 2-A), early response was measured at t = 55 [AU] after DNA damage. The time t = 55 [AU] was chosen because the system did not reach steady state and various types of system responses were observed depending on the choice of kinetic parameters at this time (Figure 3, 4, 5, 11). These initial simulations were performed without p53-Mdm2 transcription feedback to interpret simulation results more easily. For late response, the system was simulated for a long period of time until multiple numbers of sustained oscillatory peaks were observed (Figure 6, 7, 8, 9). In subsequent simulations with the full model, early response was measured by the maximum level of mRNA achieved during 0∼180 min after DNA damage. For late response, the system was again simulated for a long period of time until multiple numbers of sustained oscillatory peaks were observed.
To set realistic initial condition for early DNA damage response, cells without DNA damage were simulated by pre-equilibrating the model in the absence of DNA damage induced kinase activities (k3 = k8 = k17 = 0). For late DNA damage response, simulations were performed without pre-equilibration. DNA damage was simulated by setting all the rate constant of phosphorylation as a same positive value (k3 = k8 = k17>0) unless otherwise indicated. The rate constant of phosphorylation was kept constant during the simulation of DNA damage response because quite rapid saturation of ATM activity followed by a constant level of ATM activity up to 24 hr were observed in a previous study [45].
The bifurcation diagram shown in Figure 7 was generated for the parameter set given in column 5 of Table 1, with k3 = 0.0086851. Steady state values of the species in the system can all be expressed in terms of the steady state value of p53, which satisfies a single nonlinear equation. The steady state values of all of the species that do not involve MdmX were found to be independent of the levels of MdmX. The stable steady state values shown in the diagram were readily computed by integrating the ODEs for long times until transient effects decayed. The branch of both stable and unstable steady states was then computed using a quasi-Newton method with continuation along the branch from stable values. The linear stability of the steady state solution was determined by computing the eigenvalues of the Jacobian matrix at each point. The stable oscillatory solution branches could also be computed by integration of the ODEs for long times. In addition, both stable and unstable (subcritically bifurcating solutions) oscillatory solutions were computed by a shooting method that enforces periodic boundary conditions over a single period of oscillation. Continuation was used when necessary to follow solution branches from stable to unstable regions of parameter space.
To search parameter space, we used a previously described algorithm [29]. Briefly, the algorithm first randomly selects an initial point as a center point in a k dimensional space (where k is the number of rate constants) and randomly searches for neighbor points within a radius R. Once a point with a higher fitness score is found, the center point is updated. The method is iteratively repeated with larger values of R until the searching criteria (high fitness score) is satisfied.
The searches were done in a two step procedure. First, a subnetwork including a subset of interactions and kinetic constants was selected, based on their previously predicted roles in oscillatory behaviors [38],[39]. This subnetwork included 15 unknown kinetic constants (k2, k3, k10, k11, k12, k13, k27, k28, k29, k30, k31, k32, k33, k37, k38), and we searched this 15 dimensional parameter space for oscillatory behaviors. To limit the extent of parameter space to search, we assumed k7 = k5 = k2, k14 = k13*0.2, k8 = k3, k9 = k4 = k3*0.02. During the search, fitness scores were calculated based on cross-correlation [30] between simulated time series data and a published average oscillation pattern [9]. Among the high scored sets of parameters, we selected sets of kinetic constants that generated oscillations with certain maximum values of p53 activity (species 14>0.0001 [AU]). The selected kinetic sets were used as initial center points for the next searches to fit additional experimental data. In this second search, we searched 12 dimensional space (k2, k3, k6, k7, k15, k16, k23, k25, k34, k36, k38). To limit parameter space, we assumed k10 = k17 = k3, k4 = k9 = k18 = k3*0.02. The remaining kinetic constants, determined from the first search, were retained during the second parameter search. In the second search, fitness scores were calculated based on three types of published data: average oscillatory patterns [9], Nutlin response [24], and Mdm2/p53 and MdmX/Mdm2 total protein ratios [24] (Figure 12).
The normalized local sensitivity of mRNA (species 15 in Figure 2-B) with respect to various kinetic parameters was defined by the following equation
The local sensitivity calculation was performed using the Simbiology toolbox in Matlab. The time integral (LSj) of the normalized local sensitivity was obtained by trapezoidal numerical integration [46] and used to rank the parameters based on their sensitivity value,where T = 180 was chosen as the length of the interval of integration.
DNA damage was simulated similarly as in initial simulations. As output, species 15 (mRNA) was measured because the full model include basal p53 activity (transcription activity by p53 monomer and promoter complex). Early DNA damage response was quantified by the maximum p53 activity (species 15) achieved during 0∼180 min after DNA damage; this period corresponds to the first rise in p53 activity after DNA damage [47]. Late DNA damage response was quantified by the maximum p53 activity achieved during 10020 min∼10980 min (time window arbitrarily chosen after long period of simulation) after DNA damage. Such long term simulations were performed to allow the initial trajectory to converge to a limit cycle from an initial condition. Maximum p53 activity was taken to be the maximum level of mRNA (species 15 in Figure 2) achieved during the indicated time frame. In the initial simulations for early DNA damage response, we measured p53 activity at one time point. In the subsequent simulations (Figure 3, 4, 5), however, we selected the maximum values of p53 activity during a time window 0<t<180 min to avoid any sensitivity to a single measurement time (Figure 9). To measure the effects of MdmX, we compared the maximum p53 activities achieved in simulations with and without MdmX. When MdmX was included in the model, various levels (1/100, 1/10, 1, 10, 100 fold) of the basal production rate constant (k15) of MdmX were used.
The network was pre-equilibrated without Nutlin (k11>0) and without DNA damage (k3 = k8 = k17 = 0). For the simulation of Nutlin treatment, we simulated a network with k11 = 0, because Nutlin is known to affect Mdm2-p53 interaction only [48]. The fold increase in total p53 and Mdm2 was measured as a function of time. To fit Nutlin response data, it was assumed that unphoshorylated p53 could display basal Mdm2 induction activity [24].
Some of example matlab scripts used for simulations were provided as supplemental information (Text S1, S2, S3, S4, S5, S6, S7).
|
10.1371/journal.pgen.1008073 | Genetic determinants of gut microbiota composition and bile acid profiles in mice | The microbial communities that inhabit the distal gut of humans and other mammals exhibit large inter-individual variation. While host genetics is a known factor that influences gut microbiota composition, the mechanisms underlying this variation remain largely unknown. Bile acids (BAs) are hormones that are produced by the host and chemically modified by gut bacteria. BAs serve as environmental cues and nutrients to microbes, but they can also have antibacterial effects. We hypothesized that host genetic variation in BA metabolism and homeostasis influence gut microbiota composition. To address this, we used the Diversity Outbred (DO) stock, a population of genetically distinct mice derived from eight founder strains. We characterized the fecal microbiota composition and plasma and cecal BA profiles from 400 DO mice maintained on a high-fat high-sucrose diet for ~22 weeks. Using quantitative trait locus (QTL) analysis, we identified several genomic regions associated with variations in both bacterial and BA profiles. Notably, we found overlapping QTL for Turicibacter sp. and plasma cholic acid, which mapped to a locus containing the gene for the ileal bile acid transporter, Slc10a2. Mediation analysis and subsequent follow-up validation experiments suggest that differences in Slc10a2 gene expression associated with the different strains influences levels of both traits and revealed novel interactions between Turicibacter and BAs. This work illustrates how systems genetics can be utilized to generate testable hypotheses and provide insight into host-microbe interactions.
| Inter-individual variation in the composition of the intestinal microbiota can in part be attributed to host genetics. However, the specific genes and genetic variants underlying differences in the microbiota remain largely unknown. To address this, we profiled the fecal microbiota composition of 400 genetically distinct mice, for which genotypic data is available. We identified many loci of the mouse genome associated with changes in abundance of bacterial taxa. One of these loci is also associated with changes in the abundance of plasma bile acids—metabolites generated by the host that influence both microbiota composition and host physiology. Follow up validation experiments provide mechanistic insights linking host genetic differences, with changes in ileum gene expression, bile acid-bacteria interactions and bile acid homeostasis. Together, this work demonstrates how genetic approaches can be used to generate testable hypothesis to yield novel insight into how host genetics shape gut microbiota composition.
| The intestinal microbiota has profound effects on host physiology and health [1–3]. The composition of the gut microbiota is governed by a combination of environmental factors, including diet, drugs, maternal seeding, cohabitation, and host genetics [4–7]. Together, these factors cause substantial inter-individual variation in microbiota composition and modulate disease risk [8,9]. Alterations in the composition of the microbiota are associated with a spectrum of cognitive, inflammatory and metabolic disorders [10–12], and a number of bacterial taxa have been causally linked with modulation of disease [13–15]. A major challenge in the field is deciphering how host genetics and environmental factors interact to shape the composition of the gut microbiota. This knowledge is key for designing strategies aimed at modifying gut microbiota composition to improve health outcomes.
Several mouse and human studies have examined the role of host genetics in shaping the composition of the gut microbiota [16]. Mouse studies comparing gut bacterial communities from inbred mouse strains [17,18] and strains harboring mutations in immune-related genes [19–22] support this notion. Additionally, quantitative trait locus (QTL) analyses in mice have identified genetic regions associated with the abundance of several bacterial taxa and community structure [23–26]. Twin studies and genome-wide association studies (GWAS) in humans have identified heritable bacterial taxa and SNPs associated with specific gut microbes. While comparing these studies is often difficult due to differences in environmental variables among populations, some associations are consistently detected among geographically discrete populations, such as the association between Bifidobacterium abundance and the lactase (LCT) gene locus [27–29], indicating the abundance of specific taxa is influenced by host genetic variation.
Gut microbes and the host communicate through the production and modification of metabolites, many of which impact host physiology [30–34]. Bile Acids (BAs) are host-derived and microbial-modified metabolites that regulate both the gut microbiome and host metabolism [35–37]. BAs are synthesized in the liver from cholesterol, stored in the gallbladder and are secreted in the proximal small intestine where they facilitate absorption of fat-soluble vitamins and lipids. Once in the intestine, BAs can be metabolized by gut bacteria through different reactions, including deconjugation, dehydroxylation, epimerization, and dehydrogenation, to produce secondary BAs with differential effects on the host [33,35]. In addition to their direct effects on the host, BAs shape the gut microbiota composition through antimicrobial activities [38,39]. The detergent properties of BAs cause plasma membrane damage. The bactericidal activity of a BA molecule corresponds to its hydrophobicity [40]. Additionally, the microbiota modulates primary BA synthesis through regulation of the nuclear factor FXR [41]. Thus, we hypothesized that host genetic variation associated with changes in BA homeostasis mediates alterations in gut microbiota composition.
To investigate how genetic variation affects gut microbiota and BA profiles, we used the Diversity Outbred (DO) mouse population, which is a heterogenous population derived from eight founder strains: C57BL6/J (B6), A/J (A/J), 1291/SvImJ (129), NOD/ShiLtJ (NOD), NZO/HiLtJ (NZO), CAST/EiJ (CAST), PWK/PhJ (PWK), and WSB/EiJ (WSB) [42,43]. These eight strains capture a large breadth of the genetic diversity found in inbred mouse strains. Additionally, the founder strains harbor distinct gut microbial communities and exhibit disparate metabolic responses to diet-induced metabolic disease [18,44,45]. The DO population is maintained by an outbreeding strategy aimed at maximizing the heterozygosity of the outbred stock. The genetic diversity and large number of generations of outbreeding make it an ideal resource for high-resolution genetic mapping of microbial and metabolic traits [43].
We characterized the intestinal microbiota composition and plasma and cecal BA profiles in ~400 genetically distinct DO mice fed a high-fat/high-sucrose diet for ~22 weeks and performed quantitative trait loci (QTL) analysis to identify host genetic loci associated with these traits. Specifically, we focused our analysis on potentially pleiotropic loci, which we defined as a single genetic locus that associates with both bacterial and BA traits. Our analysis revealed several instances of bacterial and metabolite traits attributed to the same DO founder haplotypes mapping to the same position of the mouse genome, including a locus associated with plasma BA levels and the disease-modulating organism Akkermansia muciniphila. Additionally, we identified the ileal BA transporter Slc10a2 as a candidate gene that regulates both the abundance of Turicibacter sp. and plasma levels of cholic acid.
We investigated the impact of genetic variation on gut microbiota composition and bile acid (BA) profiles using a cohort of ~400 DO mice maintained on a high-fat high-sucrose diet (45% kcal from fat and 34% from sucrose) for ~22 weeks (range 21–25 weeks), starting at weaning. We previously showed that this diet elicits a wide range of metabolic responses in the eight founder strains that are associated with microbiome changes [18,46]. Furthermore, we incorporated in our analyses previously published clinical weight traits collected from the same DO mice [47]. All animals were individually housed throughout the duration of the study to measure food intake and minimize microbial exchange.
We performed LC-MS/MS analyses of plasma and cecal contents to assess relative variation in the levels of 27 BAs. Both plasma and cecal bile acids were measured to provide a comprehensive picture of systemic BA homeostasis. There was substantial variation in the plasma and cecal BA profiles across the 384 mice (Fig 1A and 1B; S1 Table). Additionally, we examined gut microbiota composition (n = 399) using 16S rRNA gene amplicon sequencing of DNA extracted from fecal samples collected at the end of the experiment. Within the cohort, there were 907 unique Exact Sequence Variants (ESVs), (100% operational taxonomic units defined with dada2 [48]), which were agglomerated into 151 lower taxonomic rankings (genus, family, order, class, phyla). The microbial traits represented each of the major phyla found in the intestine and the relative abundance of these phyla was highly variable among the DO mice (Fig 1C). For instance, the abundance of taxa classified to the Bacteroidetes phylum ranged from 1.17–89.28%.
For subsequent analysis, we identified a core measurable microbiota (CMM), which we defined as taxon found in at least 20% of the mice [24]. This was done to remove the effects of excessive variation in the data due to bacterial taxa that were low abundance and/or sparsely distributed. In total, the CMM was comprised of 86 ESVs and 42 agglomerated taxa (S2 Table). The CMM traits represent a small fraction of the total microbes detected, but account for 94.5% of the rarefied sequence reads, and therefore constitute a significant portion of the identifiable microbiota.
Since mice were received in cohorts (i.e., waves) of 100, we examined whether animals in each wave were more similar to each other than mice in other waves. The fecal microbiota composition significantly clustered by wave (p < 0.001, PERMANOVA) and sex (p < 0.001, PERMANOVA) (S1 Fig). PCA analysis of plasma and cecal bile acids showed a significant effect of sex, but not wave, on both plasma (p < 0.0001, Kruskal Wallis) and cecal BA profiles (p < 0.05, Kruskal Wallis) (S2 Fig).
There is substantial evidence implicating gut microbiota and BAs in metabolic disease development [36,37]. To identify potential relationships among these traits, we performed correlation analysis which yielded many significant associations after FDR correction (FDR < 0.05) (S3 Table, discussed in S1 Data).
To identify associations between regions of the mouse genome and the clinical and molecular traits discussed above, we performed QTL analysis using the R/qtl2 package [49]. We used sex, days on the diet, and experimental wave as covariates. We identified 13 significant QTL (LOD ≥ 7.66; P ≤ 0.05) and 50 suggestive QTL (LOD ≥ 6.80; P ≤ 0.2) for bacterial [36], bile acid [13], and body weight [1] traits (Fig 2, S4 Table).
Of the microbial QTL, we found 23 QTL for 17 distinct bacterial ESVs from the Bacteroidetes and Firmicutes phyla that met the LOD ≥ 6.80 threshold. ESVs with the strongest QTL (LOD > 8) are classified to the Clostridiales order and map on chr 12 at ~33 Mbp, the Lachnospiraceae family on chr 2 at 164 Mbp, and the S24-7 family on chr 2 at ~115 Mbp. We also identified 12 QTL for microbial taxa collapsed by taxonomic assignment (i.e., genus to phylum). The genera Lactococcus and Oscillospira were also associated with host genetic variation, which is consistent with previous studies [23,24,50,51].
Similarly, BA QTL mapped to multiple loci spanning the mouse genome and most BA traits mapped to multiple positions. BA synthesis and metabolism are regulated by multiple host signaling pathways: there are >17 known host enzymes involved in the production of BAs [36], transporters, which play a critical role in maintaining the enterohepatic circulation and BA homeostasis, and receptors that respond to BA in a variety of host tissues [52–54]. Therefore, it is not surprising that our results indicate that BA levels are polygenic and shaped by multiple host factors.
To identify instances of overlapping QTL, we applied a less stringent threshold of LOD ≥ 6.1 (P < 0.5). We observed multiple instances of related BA species associating to the same genetic locus, indicating the presence of pleiotropic loci. Interestingly, several of these loci associate with levels of related BA species in different stages of microbial modification. For example, cecal taurocholic acid (TCA) and plasma CA QTL overlap on chr 7 at 122 Mbp. Likewise, QTL for plasma TDCA and cecal DCA, overlap on chr 12 between ~99–104 Mbp. For the cecal DCA, the WSB founder haplotype was associated with higher levels of this BA, while the NOD founder haplotype was associated with lower levels. The opposite pattern was observed for plasma TDCA, where the NOD and WSB haplotype were associated with higher and lower levels, respectively (S3A and S3B Fig).
We also identified overlapping QTLs on chr 11 at ~71 Mbp for cecal levels of the secondary BAs lithocholic acid (LCA) and isolithocholic acid (ILCA), the isomer of LCA produced by bacterial epimerization (S3C Fig). Higher levels of these cecal BAs are associated with the 129 founder haplotype and lower levels are associated with the A/J founder haplotype (S3D and S3E Fig). We identified the positional candidate gene Slc13a5 (S3F Fig), which is a sodium-dependent transporter that mediates cellular uptake of citrate, an important precursor in the biosynthesis of fatty acids and cholesterol [55]. Recent evidence indicates that Slc13a5 influences host metabolism and energy homeostasis [56–58]. Slc13a5 is a transcriptional target of pregnane X receptor (PXR) [59], which also regulates the expression of genes involved in the biosynthesis, transport, and metabolism of BAs [60].
We searched for regions of the chromosome that were associated with both BA and bacterial abundance, as this may provide evidence of interactions between the traits [61]. We identified 17 instances of overlapping microbial and BA QTL on 12 chromosomes (LOD ≥ 6.1; P ≤ 0.5). This QTL overlap indicates there might be QTL with pleiotropic effects on BAs and the microbiota, suggest that genetic variation influencing host BA profiles has an effect on compositional features of the gut microbiota, or genetic-driven variation in microbiota composition alters BAs. Examples of notable instances of overlapping bacterial and BA QTL, including Akkermansia muciniphila and Peptostreptococcaceae family are discussed in the Supporting Information (S1 Data).
We focused our co-mapping analysis on chr 8 at ~ 5.5 Mbp, where Turicibacter sp. QTL and plasma cholic acid (CA) QTL overlap (Fig 3A and 3B). These traits were particularly interesting because both have been shown to be influenced by host genetics by previous studies. Turicibacter has been identified as highly heritable in both mouse and human genetic studies [24,27,45,50], and multiple reports have found differences in CA levels as a function of host genotype [18,46]. Furthermore, CA levels are influenced by both host genetics and microbial metabolism since it is synthesized by host liver enzymes from cholesterol and subsequently modified by gut microbes in the intestine. Notably, these co-mapping traits also share the same allele effects pattern, where the A/J and WSB haplotypes have strong positive and negative associations, respectively (Fig 3C and 3D).
To assess whether the trait patterns observed in the DO founder strains correspond to the observed allelic effects in the QTL mapping, we performed a separate characterization of the fecal microbiota composition and plasma bile acids in age-matched A/J and WSB animals fed the HF/HS diet. The founder strain allele patterns inferred from the QTL mapping closely resembled the observed levels of Turicibacter sp. (Fig 3E) and plasma CA in the founder strains (Fig 3F), where A/J animals had significantly higher levels of Turicibacter sp. and CA than WSB animals. However, Turicibacter levels in the founder strains do not completely mirror the estimated allele effects. This may be due to other genetic factors that also influence Turicibacter levels, as this taxa may be influenced by multiple host genes and levels of Turicibacter have previously been associated on chr 7 [24], 9 and 11 [50] in mice. Furthermore, Turicibacter and plasma CA were positively correlated in the DO mice (r = 0.43, p = 3.53e-10). This finding is consistent with a previous study that found positive correlations between Turicibacter and unconjugated cecal BAs [62]. Taken together, the overlap between the Turicibacter sp. QTL and plasma CA QTL, along with the similar allele effects pattern, which reflect the values observed in the founder strains, provide strong evidence that these traits are related and they are responding to the common genetic driver.
We searched in the QTL confidence interval for candidate genes via high-resolution association mapping on chr 8 and identified SNPs associated with both microbial and BA traits. Among these we identified SNPs upstream of the gene Slc10a2, which encodes for the apical sodium-bile transporter (Fig 3G). Slc10a2 is responsible for ~95% of BA reabsorption in the distal ileum and plays a key role in BA homeostasis [63]. In humans, mutations in this gene are responsible for primary BA malabsorption, resulting in interruption of enterohepatic circulation of BAs and decreased plasma cholesterol levels [64]. Likewise, Slc10a2-/- mice have a reduced total BA pool size, increased fecal BA concentrations and reduced total plasma cholesterol in comparison to wild-type mice [63]. Additionally, a comparison between germ-free and conventionally-raised mice found that expression of Slc10a2 is downregulated in presence of the gut microbiota, suggesting microbes may influence the expression of the transporter [41].
Our analysis identified SNPs associated with levels of Turicibacter sp. and plasma CA at the QTL peak (Fig 3G). The SNPs with the strongest associations were attributed to the WSB and A/J haplotypes and fell on intergenic regions near Slc10a2. There is growing evidence that non-coding intergenic SNPs are often located in or closely linked to regulatory regions, suggesting that they may influence host regulatory elements and alter gene expression [65,66]. To assess if candidate gene expression patterns in the DO founders corresponds to the estimated allelic effects in the QTL mapping, we quantified Slc10a2 expression in distal ileum samples from A/J and WSB mice by quantitative reverse transcriptase PCR (qRT-PCR). A/J mice exhibited significantly higher expression of Slc10a2 compared to WSB mice (Fig 3H), which is consistent with estimated allele patterns for the overlapping Turicibacter and plasma CA QTLs on chr 8 (Fig 3A and 3B). Remarkably, several studies have noted concomitant changes in microbiota composition and Slc10a2 mRNA levels [67–69].
We mapped QTL for Turicibacter sp. and for plasma CA levels to a common locus on chr 8 at 5–7 Mbp. Since the LOD profiles and allelic effects are highly similar, the QTL may be due to a single shared locus (pleiotropy) or multiple closely linked loci. We examined this question using a likelihood ratio testing of the null hypothesis of pleiotropy versus the alternative of two independent genetic regulators of these traits [70]. Analysis of 1000 bootstrap samples resulted in a p-value of 0.531, which is consistent with the presence of a single pleiotropic locus that affects both traits.
We next sought to understand the causal relationships between the microbe and the BA. We asked whether the relationship between the microbe and BA was causal, reactive or independent. To establish the directionality of the relationship, we applied mediation analysis where we conditioned one trait on the other [71]. When we conditioned Turicibacter sp. on plasma CA (QTL → BA → Microbe), we observed a LOD drop of 3.2 (Fig 4A and 4B). Likewise, when we conditioned the plasma cholic acid on the microbe (QTL → Microbe → BA) there was a LOD drop of 3.32 (Fig 4C and 4D). The partial mediation seen in both models suggests that the relationship between the microbe and the BA could be bidirectional, where they exert an effect on one another.
From this analysis, we can hypothesize this relationship can be explained by a pleiotropic model, where a single locus influences a microbial and a BA trait, and the microbial trait is also reactive to changes in the BA trait. It is important to note that statistical inference only partially explains the relationship between the traits and there may be other hidden variables that may further explain the relationship. The complex relationship depicted by the causal inference testing is consistent with the interplay between gut microbes and BAs in the intestine and their known ability to influence the other.
Due to the strong correlative relationship between the QTL, we tested whether there was a direct interaction between bile acids and Turicibacter. Turicibacter inhabits the small intestine where BAs are secreted upon consumption of a meal [72,73]. We screened the human isolate Turicibacter sanguinis for deconjugation and transformation activity in vitro by HPLC/MS-MS. We found that T. sanguinis deconjugated ~96–100% of taurocholic acid and glycochenodeoxycholic acid (Fig 5A) within 24 hours. It also transformed ~6 and 8% of CA and CDCA to 7-dHCA and 7-ketolithocholic acid (7-KLCA), respectively (Fig 5B and 5C). Both of these transformations require the action of the bacterial 7α-hydroxysteroid dehydrogenase.
Based on these results, we asked if conjugated and unconjugated bile acids differentially modulate T. sanguinis growth. BA concentrations range from ~1–10 mM along the small intestine [74] to ~0.2–1 mM in the cecum [75]. Therefore, we grew T. sanguinis in the presence of either conjugated or unconjugated bile acids at physiologically relevant concentrations ranging from 0.1–5 mM. T. sanguinis growth decreased with increasing concentrations of BAs and growth was completely inhibited at 1 mM for unconjugated BAs and 5 mM for conjugated BAs (Fig 5D and 5E). Growth rate was significantly slower in the presence of 1 mM conjugated and 0.5mM unconjugated bile acids (Fig 5F). These results suggest that levels of BAs may affect abundance of Turicibacter in the gut.
To compare T. sanguinis sensitivity to conjugated bile acids relative to other small intestine colonizers, we grew four taxa (Bacteroides thetaiotaomicron, Clostridium asparagiforme, Lactobacillus reuteri and Escherichia coli MS200-1) known to colonize this region of the intestine with or without 1 mM conjugated bile acids. Members of these genera are known to have bile salt hydrolase (BSH) activity to deconjugate bile acids [35]. Unlike T. sanguinis, the addition of high levels of conjugated bile acids had little to no effect on the growth of these four gut microbes (S4 Fig). Consistent with these findings, Turicibacter abundance was negatively correlated with cecal TCA levels in the DO mice (r = -0.262, p = 0.0035).
Taken together, these data indicate that T. sanguinis is sensitive to higher concentrations of BA compared to other small intestine colonizers. These reciprocal effects between the BA and the bacterium provide biological evidence for the correlative relationship shown by the causal model testing. In summary, using a genetic approach, we identified and provide validation of a relationship between a genetic locus containing the BA transporter Slc10a2, and levels of Turicibacter and plasma cholic acid. Based on our findings, we hypothesize that the identified locus regulates expression of Slc10a2, altering active BA reabsorption in the ileum, leading to increased intestinal BA concentrations and alterations in the intestinal BA environment. Consequently, the resulting environmental change provides an unfavorable habitat for Turicibacter. In turn, lower levels of Turicibacter BA deconjugation activity leads to a decrease in circulating free plasma cholic acid levels.
In this study, we performed the first known genetic mapping integration of gut microbiome and BA profiles. Using DO mice, we identified multiple QTL for gut microbes and bile acids spanning the host genome. These included loci that associated with individual microbial and BA traits, as well as loci with potential pleiotropic effects, where a single genetic region influenced both the abundance of a gut microbe and levels of a BA. While several studies suggest that host genetic variation has a minor impact on microbiota composition, there are overlapping findings among different studies in both human and mouse populations that indicate that specific bacterial taxa are influenced by host genetics. Our results in the DO population corroborate several of these key findings (discussed in S1 Data). Turicibacter sp. is among the microbes consistently associated with host genetics. This work plus data from previous reports suggest that alterations in the BA pool driven by Slc10a2 genetic variation and concomitant changes in expression/activity elicit an impact on gut microbiota community structure and influence the ability of Turicibacter to colonize and persist in the intestine. Although this microbe deconjugates primary BAs, we found that it is also sensitive to elevated concentrations of both conjugated and unconjugated BAs. Future experiments are needed to examine how a decrease in Slc10a2 expression changes intestinal BA profiles and the consequences on Turicibacter colonization. Additionally, this work identified multiple host-microbe-metabolite interactions that need to be validated with additional molecular studies. More broadly, our work demonstrates the power of genetics to identify novel interactions between microbial and metabolite traits and provides new testable hypotheses to further dissect factors that shape gut microbiota composition.
Animal care and study protocols were approved by the University of Wisconsin-Madison Animal Care and Use Committee (A005821) and were in compliance with all NIH animal welfare guidelines.
Animal care and study protocols were approved by the University of Wisconsin-Madison Animal Care and Use Committee. DO mice were obtained from the Jackson Laboratories (Bar Harbor, ME, USA) at ~4 weeks of age and maintained in the Department of Biochemistry vivarium at the University of Wisconsin-Madison. Mice were housed on a 12-hour light:dark cycle under temperature- and humidity-controlled conditions. Five waves of 100 DO mice each from generations, 17, 18, 19, 21, and 23 were obtained at intervals of 3–6 months. Each wave was composed of equal numbers of male and female mice. All mice were fed a high-fat high-sucrose diet (TD.08811, Envigo Teklad, 44.6% kcal fat, 34% carbohydrate, and 17.3% protein) ad libitum upon arrival to the facility. Mice were kept in the same vivarium room and were individually housed to monitor food intake and prevent coprophagy between animals. DO mice were sacrificed at 22–25 weeks of age.
The eight DO founder strains (C57BL/6J, A/J, 129S1/SvImJ, NOD/ShiLtJ, NZO/HILtJ, PWK/PhJ, WSB/EiJ and CAST/EiJ) were obtained from the Jackson Laboratories. Mice were bred at the University of Wisconsin-Madison Biochemistry Department. Mice were housed by strain and sex (2–5 mice/cage), with the exception of CAST that required individual housing. Inbred founder mice were housed under the same environmental conditions as the DO animals. Like the DO mice, the eight founder strains were maintained on the HF/HS diet and were sacrificed at 22 weeks of age, except for NZO males that were sacrificed at 14 weeks, due to high mortality attributable to severe disease.
For both DO and founder mice, fecal samples for 16S rRNA sequencing were collected immediately before sacrifice after a 4 hour fast. Cecal contents, plasma, and additional tissues were harvested promptly after sacrifice and all samples were immediately flash frozen in liquid nitrogen and stored at -80°C until further processing.
DNA was isolated from feces using a bead-beating protocol [18]. Mouse feces (~1 pellet per animal) were re-suspended in a solution containing 500μl of extraction buffer [200mM Tris (pH 8.0), 200mM NaCl, 20mM EDTA], 210μl of 20% SDS, 500μl phenol:chloroform:isoamyl alcohol (pH 7.9, 25:24:1) and 500μl of 0.1-mm diameter zirconia/silica beads. Cells were mechanically disrupted using a bead beater (BioSpec Products, Barlesville, OK; maximum setting for 3 min at room temperature), followed by extraction with phenol:chloroform:isoamyl alcohol and precipitation with isopropanol. Contaminants were removed using QIAquick 96-well PCR Purification Kit (Qiagen, Germantown, MD, USA). Isolated DNA was eluted in 5 mM Tris/HCL (pH 8.5) and was stored at -80°C until further use.
PCR was performed using universal primers flanking the variable 4 (V4) region of the bacterial 16S rRNA gene [76]. Genomic DNA samples were amplified in duplicate. Each reaction contained 10–30 ng genomic DNA, 10 μM each primer, 12.5 μl 2x HiFi HotStart ReadyMix (KAPA Biosystems, Wilmington, MA, USA), and water to a final reaction volume of 25 μl. PCR was carried out under the following conditions: initial denaturation for 3 min at 95°C, followed by 25 cycles of denaturation for 30 s at 95°C, annealing for 30 s at 55°C and elongation for 30 s at 72°C, and a final elongation step for 5 min at 72°C. PCR products were purified with the QIAquick 96-well PCR Purification Kit (Qiagen, Germantown, MD, USA) and quantified using Qubit dsDNA HS Assay kit (Invitrogen, Oregon, USA). Samples were equimolar pooled and sequenced by the University of Wisconsin–Madison Biotechnology Center with the MiSeq 2x250 v2 kit (Illumina, San Diego, CA, USA) using custom sequencing primers.
Demultiplexed paired end fastq files generated by CASAVA (Illumina) and a mapping file were used as input files. Sequences were processed, quality filtered and analyzed with QIIME2 (version 2018.4) (https://qiime2.org), a plugin-based microbiome analysis platform [77]. DADA2 [48] was used to denoise sequencing reads with the q2-dada2 plugin for quality filtering and identification of de novo exact sequence variants (ESVs) (i.e. 100% exact sequence match). This resulted in 20,831,573 total sequences with an average of 52,078 sequences per sample for the DO mice, and 2,128,796 total sequences with an average of 34,335.4 sequences per sample for the eight DO founder strains. Sequence variants were aligned with mafft [78] with the q2-alignment plugin. The q2-phylogeny plugin was used for phylogenetic reconstruction via FastTree [79]. Taxonomic classification was assigned using classify-sklearn [80] against the Greengenes 13_8 99% reference sequences [81]. Alpha- and beta-diversity (weighted and unweighted UniFrac [82] analyses were performed using q2-diversity plugin at a rarefaction depth of 10000 sequences per sample. For the DO mice, one sample (DO071) was removed from subsequent analysis because it did not reach this sequencing depth. For analysis of the eight DO founder strains, one sample (NOD5) was removed because it did not reach this sequencing depth. Subsequent processing and analysis were performed in R (v.3.5.1), and data generated in QIIME2 was imported into R using Phyloseq [83]. Sequencing data was normalized by cumulative sum scaling (CSS) using MetagenomeSeq [84]. Summaries of the taxonomic distributions were generated by collapsing normalized ESV counts into higher taxonomic levels (genus to phylum) by phylogeny. We defined a core measurable microbiota (CMM) [24] to include only microbial traits present in 20% of individuals in the QTL mapping. In total, 86 ESVs and 42 collapsed microbial taxonomies comprised the CMM.
40 μL of DO plasma collected at sacrifice (30 μL used for founder strains) were aliquoted into a tube with 10 μL SPLASH Lipidomix internal standard mixture (Avanti Polar Lipids, Inc.). Protein was precipitated by addition of 215 μL MeOH. After the mixture was vortexed for 10 s, 750 μL methyl tert-butyl ether (MTBE) were added as extraction solvent and the mixture was vortexed for 10 s and mixed on an orbital shaker for 6 min. Phase separation was induced by adding 187.5 μL of water followed by 20 s of vortexing. All steps were performed at 4°C on ice. Finally, the mixture was centrifuged for 4 min at 14,000 x g at 4°C and stored at -80°C. For targeted bile acids analysis, samples were thawed on ice. 400 μL of ethanol were added to further precipitate protein, as well as 15 μL of isotope-labeled internal standard mix (12.5 μM d4-TαMCA, 10 μM d4-CDCA). The samples were vortexed for 20 s and centrifuged for 4 min at 14,000 g at 4°C after which the supernatant (ca. 1000 μL) was taken out and dried down. Dried supernatants were resuspended in 60 μL mobile phase (50%B), vortexed for 20 s, centrifuged for 4 min at 14,000 g and then 50 μL were transferred to vials with glass inserts for MS analysis.
30 ± 7.5 mg cecal contents along with 10 μL SPLASH Lipidomix internal standard mixture were aliquoted into a tube with a metal bead and 270 μL MeOH were added for protein precipitation. To each tube, 900 μL MTBE and 225 μL of water were added as extraction solvents. All steps were performed at 4°C on ice. The mixture was homogenized by bead beating for 8 min at 25 Hz. Finally, the mixture was centrifuged for 4–8 min at 11,000 x g at 4°C. Subsequent processing for the DO mice and eight DO founder strains differed due to other analyses performed on the samples that are not presented in this paper. For DO samples, 100 μL of the aqueous and 720 μL of organic layer were combined and stored at -80°C. For analysis, these were thawed on ice and 400 μL of ethanol were added to further precipitate protein, as well as 15 μL of isotope-labeled internal standard mix (12.5 μM d4-TαMCA, 10 μM d4-CDCA). The samples were vortexed for 20 s and centrifuged for 4 min at 14,000 g at 4°C after which the supernatant (ca. 1000 μL) was taken out and dried down. Dried supernatants were resuspended in 100 μL mobile phase (50%B), vortexed for 20 s, centrifuged for 8 min at 14,000 g and then 50 μL were transferred to vials with glass inserts for MS analysis. For the eight DO founder strains, the mixture was dried down including all solid parts and stored dried at -80°C. For targeted bile acid analysis, these dried down samples were then thawed on ice and reconstituted in 270 μL of methanol, 900 μL of MTBE, and 225 μL of water. 400 μL of ethanol were added to further precipitate protein, as well as 15 μL of isotope-labeled internal standard mix (12.5 μM d4-TαMCA, 10 μM d4-CDCA). The mixture was bead beat for 8 min at 25 Hz and centrifuged at 14,000 g for 8 minutes after which the supernatant (ca. 1500 μL) was taken out and dried down. Dried supernatants were resuspended in 100 μL mobile phase (50%B), vortexed for 20 s, centrifuged for 4 min at 14,000 g and then 90 μL were transferred to vials with glass inserts for MS analysis.
LC-MS analysis was performed in randomized order using an Acquity CSH C18 column held at 50°C (100 mm × 2.1 mm × 1.7 μm particle size; Waters) connected to an Ultimate 3000 Binary Pump (400 μL/min flow rate; Thermo Scientific). Mobile phase A consisted of 10 mM ammonium acetate containing 1 mL/L ammonium hydroxide. Mobile phase B consisted of MeOH with the same additives [85]. Mobile phase B was initially held at 50% for 1.5 min and then increased to 70% over 13.5 min. Mobile phase B was further increased to 99% over 0.5 min and held for 2.5 min. The column was re-equilibrated for 5.5 min before the next injection. Twenty microliters of plasma sample or ten microliters of cecum sample were injected by an Ultimate 3000 autosampler (Thermo Scientific). The LC system was coupled to a TSQ Quantiva Triple Quadrupole mass spectrometer (Thermo Scientific) by a heated ESI source kept at 325°C (Thermo Scientific). The inlet capillary was kept at 350°C, sheath gas was set to 15 units, auxiliary gas to 10 units, and the negative spray voltage was set to 2,500 V. For targeted analysis the MS was operated in negative single reaction monitoring (SRM) mode acquiring scheduled, targeted scans to quantify selected bile acid transitions, with two transitions for each species’ precursor and 3 min retention time windows. Collision energies were optimized for each species and ranging from 20–55 V. Due to insufficient fragmentation for unconjugated bile acids, the precursor was monitored as one transition with a CE of 20 V. MS acquisition parameters were 0.7 FWHM resolution for Q1 and Q3, 1 s cycle time, 1.5 mTorr CID gas and 3 s Chrom filter. In total, 27 bile acids, including 14 unconjugated, 9 tauro- and 4 glycine-conjugated species, were measured. The resulting bile acid data were processed using Skyline 3.6.0.10493 (University of Washington). For each species, one transition was picked for quantitation, while the other was used for retention time confirmation. Normalization of the quantitative data was performed to the internal standard d4-CDCA as indicated in Eq 1.
Genotyping was performed on tail biopsies as previously described [42] using the Mouse Universal Genotyping Array (GigaMUGA; 143,259 markers) [86] at Neogen (Lincoln, NE). Genotypes were converted to founder strain-haplotype reconstructions using a hidden Markov model (HMM) implemented in the R/qtl2 package [49]. We interpolated the GigaMUGA markers onto an evenly spaced grid with 0.02-cM spacing and added markers to fill in regions with sparse physical representation, which resulted in 69,005 pseudomarkers.
We performed QTL mapping using the R package R/qtl2 [49]. QTL mapping was done through a regression of the phenotype on the founder haplotype probabilities estimated with an HMM designed for multi-parental populations. Genome scans were performed for each phenotype with sex, cohort (wave), and days on diet included as additive covariates. Genetic similarity between mice was accounted for using a kinship matrix based on the leave-one-chromosome-out (LOCO) methods [87]. For microbial QTL mapping, normalized gut microbiota abundance data transformed to normal quantiles. For bile acid QTL mapping, normalized plasma and cecal bile acid levels were log2 transformed. The mapping statistic reported is the log10 likelihood ratio (LOD score). The QTL support interval was defined using the 95% Bayesian confidence interval. Significant and suggestive QTL were determined at a genome-wide threshold of P ≤ 0.05 (LOD ≥ 7.66) and P ≤ 0.2 (LOD ≥ 6.80), respectively. We used a common significance threshold for all phenotypes, by pooling the permutation results for the individual phenotypes. No adjustment was made for the search across multiple phenotypes.
To assess whether two co-mapping traits were caused by a pleiotropic locus, we used a likelihood ratio test implemented with the open source R package R/qtl2pleio [70]. Here, we compared the alternative hypothesis of two distinct loci with the null hypothesis of pleiotropy for two traits that map to the same genetic region. Parametric bootstrapping was used to determine statistical significance. Mediation analysis was applied to identify whether a microbe or bile acid were likely to be a causal mediator of the QTL as presented in Li et al. [88]. This analysis was adapted from a general approach previously described to differentiate target from mediator variables [89]. The effect of a mediator on a target was evaluated by performing an allele scan or SNP scan using the target adjusted by mediator. Only individuals with both values for both traits were considered for mediation analysis. Traits with a LOD drop >2 after controlling for the mediator were considered for further causality testing. To statistically assess causality between microbial and bile acid trait sets (causal, reactive, independent, undecided), a causal model selection test [90] was applied using the R packages R/intermediate and R/qtl2. Causal model selection tests were evaluated on both alleles and SNPs in peak region.
Total RNA was extracted from flash-frozen distal ileum tissues by TRIzol extraction and further cleaned using the RNeasy Mini Kit (Qiagen, Germantown, MD, USA). DNA was removed by on-column DNase digestion (Qiagen). Purified RNA was quantified using a Nanodrop 2000 spectrophotometer.
SuperScript II Reverse Transcriptase with oligo(dT) primer (all from Invitrogen, Carlsbad, CA, USA) was used to synthesize 20 μl cDNA templates from 1 μg purified RNA. cDNA was diluted 2X before use and qRT-PCR reactions were prepared in a 10 μl volume using SsoAdvanced Universal SYBR Green Supermix (Bio-Rad, Hercules, CA, USA) and 400 nM specific primers targeting the gene of interest (SLC10A2-F [5’- TGGGTTTCTTCCTGGCTAGACT-3’]; SLC10A2-R [5’- TGTTCTGCATTCCAGTTTCCAA-3’] [91]). All reactions were performed in triplicate. Reactions were run on a CFX96 Real-Time PCR System (Bio-Rad, Hercules, CA, USA). The 2-ΔΔCt method [92] was used to calculate relative changes in gene expression and all results were normalized to GAPDH.
Bacterial strains were obtained from DSMZ and ATCC. All strains were cultured at 37°C under anaerobic conditions using an anaerobic chamber (Coy Laboratory Products) with a gas mix of 5% hydrogen, 20% carbon dioxide and 75% nitrogen. Strains were grown in rich medium (S5 Table) that was filter sterilized and stored in the anaerobic chamber at least 24 hours prior to use. L. reuteri was grown in medium supplemented with 20 mM glucose. For all in vitro assays, cultures used for inoculation were grown overnight at 37°C in 10 mL 14b medium in anaerobic Hungate tubes. Stock solutions of conjugated bile acids (TCA, GCDCA) and unconjugated bile acids (CA, CDCA, DCA) were prepared to a final concentration of 100 mM and used for all in vitro assays. All bile acids used were soluble in methanol.
Stock solutions of conjugated and unconjugated bile acids (100 mM) were added to 3 ml 14b medium to obtain a final concentration of 100 μM total bile acid. Tubes were inoculated with a T. sanguinis cultured overnight, then incubated in the anaerobic chamber at 37°C for 48 hours. At the 24- and 48-hour timepoints, 1 mL of each culture was removed and the supernatant was collected after brief centrifugation. Each culture supernant was diluted 10x in initial running solvent (30:70 MeOH:10 mM ammonium acetate). Samples were spun at max speed for 3 minutes to remove suspended particles prior to loading on the uHPLC. Samples were analyzed using a uHPLC coupled with a high-resolution mass spectrometer.
10 μL aliquots of diluted supernatant samples were analyzed using a uHPLC-MS/MS system consisting of a Vanquish uHPLC coupled by electrospray ionization (ESI) (negative mode) to a hybrid quadrupole-high-resolution mass spectrometer (Q Exactive Orbitrap; Thermo Scientific). Liquid chromatography separation was achieved on an Acquity UPLC BEH C18 column (2.1-by 100-mm column, 1.7-μm particle size) heated to 50°C. Solvent A was 10 mM Ammonium acetate, pH 6; solvent B was 100% methanol. The total run time was 31.5 minutes with the following gradient: 0 min, 30% B; 0.5 min, 30% B; 24 min, 100% B; 29 min, 100% B; 29 min, 30% B; 31.5 min, 30% B. Bile acid peaks were identified using the Metabolomics Analysis and Visualization Engine (MAVEN) [93].
Bacterial growth rate was measured in medium 14b supplemented with either 100 μM, 300 μM, 1 mM bile acids or methanol control. Medium was dispensed inside an anerobic chamber into Hungate tubes. Tubes containing 10 mL of medium were inoculated with 30 μL of an overnight culture and incubated at 37°C for 24 hours. T. sanguinis was grown with shaking to disrupt the formation of flocculent colonies. Growth was monitored as the increase in absorbance at 600 nm in a Spectronic 20D+ spectrophotometer (Thermo Scientific, Waltham, MA, USA). Growth rate was determined as μ = ln(X/Xo)/T, where X is the OD600 value during the linear portion of growth and T is time in hours. Values given are the mean μ values from two independent cultures done in triplicate.
All statistical analyses were performed in R (v.3.5.1) [94]. Unless otherwise indicated in the figure legends, differences between groups were evaluated using unpaired two-tailed Welch’s t-test. For multiple comparisons, Krustkal-Wallis test was used if ANOVA conditions were not met, followed by Mann-Whitney/Wilcoxon rank-sum for multiple comparisons and adjusted for multiple testing using the Benjamini-Hochberg FDR procedure. The correlation between the abundance of microbial taxa was performed using Spearman’s correlation in the “Hmisc” (v.4.1–1) R package [95]. The p-values were adjusted using the Benjamini and Hochberg method, and correlation coefficients were visualized using the “pheatmap” (v.1.0.10) [96]. Multiple groups were compared by Kruskal-Wallis test and adjusted for multiple testing using the Benjamini-Hochberg FDR procedure. Significance was determined as p-value < 0.05. To assess magnitude of variability of the CMMs, summary statistics were calculated on each CMM (taxa and ESVs). Non-parametric-based PERMANOVA statistical test [97] with 999 Monte Carlo permutations was used to compare microbiota compositions among groups using the Vegan R package [98].
|
10.1371/journal.pgen.1003991 | Canine Hereditary Ataxia in Old English Sheepdogs and Gordon Setters Is Associated with a Defect in the Autophagy Gene Encoding RAB24 | Old English Sheepdogs and Gordon Setters suffer from a juvenile onset, autosomal recessive form of canine hereditary ataxia primarily affecting the Purkinje neuron of the cerebellar cortex. The clinical and histological characteristics are analogous to hereditary ataxias in humans. Linkage and genome-wide association studies on a cohort of related Old English Sheepdogs identified a region on CFA4 strongly associated with the disease phenotype. Targeted sequence capture and next generation sequencing of the region identified an A to C single nucleotide polymorphism (SNP) located at position 113 in exon 1 of an autophagy gene, RAB24, that segregated with the phenotype. Genotyping of six additional breeds of dogs affected with hereditary ataxia identified the same polymorphism in affected Gordon Setters that segregated perfectly with phenotype. The other breeds tested did not have the polymorphism. Genome-wide SNP genotyping of Gordon Setters identified a 1.9 MB region with an identical haplotype to affected Old English Sheepdogs. Histopathology, immunohistochemistry and ultrastructural evaluation of the brains of affected dogs from both breeds identified dramatic Purkinje neuron loss with axonal spheroids, accumulation of autophagosomes, ubiquitin positive inclusions and a diffuse increase in cytoplasmic neuronal ubiquitin staining. These findings recapitulate the changes reported in mice with induced neuron-specific autophagy defects. Taken together, our results suggest that a defect in RAB24, a gene associated with autophagy, is highly associated with and may contribute to canine hereditary ataxia in Old English Sheepdogs and Gordon Setters. This finding suggests that detailed investigation of autophagy pathways should be undertaken in human hereditary ataxia.
| Neurodegenerative diseases are one of the most important causes of decline in an aging population. An important subset of these diseases are known as the hereditary ataxias, familial neurodegenerative diseases that affect the cerebellum causing progressive gait disturbance in both humans and dogs. We identified a mutation in RAB24, a gene associated with autophagy, in Old English Sheepdogs and Gordon Setters with hereditary ataxia. Autophagy is a process by which cell proteins and organelles are removed and recycled and its critical role in maintenance of the continued health of cells is becoming clear. We evaluated the brains of affected dogs and identified accumulations of autophagosomes within the cerebellum, suggesting a defect in the autophagy pathway. Our results suggest that a defect in the autophagy pathway results in neuronal death in a naturally occurring disease in dogs. The autophagy pathway should be investigated in human hereditary ataxia and may represent a therapeutic target in neurodegenerative diseases.
| The hereditary ataxias are an important, heterogeneous group of movement disorders unified by the presence of degeneration of the cerebellar cortex, and in particular of the Purkinje neurons [1]–[5]. They may be inherited as autosomal dominant (also known as the spinocerebellar ataxias or SCAs), recessive, X linked and mitochondrial traits with the autosomal dominant SCAs representing the most common group of ataxias in humans [3], [4]. Associated mutations range from possible toxic gain of function mechanisms in polyglutamine diseases [6], to ion chanelopathies such as the proposed calcium channel dysfunction in SCA6 [7], to abnormalities in growth factors such as FGF14 in SCA27 [8], and to structural proteins such as β-III spectrin in SCA5 [9]. While over 50 different genetic loci have been shown to be associated in humans, in 20–40% of patients, the genetic cause remains elusive [6]. Purebred dogs suffer from comparable neurodegenerative diseases affecting the cerebellar cortex referred to as cerebellar cortical degeneration, cerebellar abiotrophy or canine cerebellar or hereditary ataxia. Currently, hereditary cerebellar degenerative disorders have been described in over 20 breeds of dog [10], [11] with many more sporadic cases reported. In most dog breeds, the disorder causes slowly progressive degeneration of the cerebellar cortex, with dramatic Purkinje neuron loss resulting in a progressive gait dysfunction [10], [11]. Forms that primarily target the granular layer or produce ataxia without cell loss are also reported [12]–[18]. The canine form of the disease targets the cerebellum primarily and is most commonly inherited as a fully penetrant autosomal recessive trait [11], [13], [19], [20]. Onset may be neonatal [13], juvenile [21] or even older [22]. The genetic cause of three of these canine disorders has been described and include a mutation in SEL1L, a protein important in endoplasmic reticulum associated protein degradation in the Finnish hound [23], SPTBN2 gene encoding β-III spectrin in the beagle [24], and GRM1, the metabotropic glutamate 1 receptor in the Coton de Tulear [25]. Autosomal recessive cerebellar degenerative disorders have been described in the Old English Sheepdog [20] and the Gordon Setter [26], [27]. The clinical phenotype is identical in both breeds with an onset of cerebellar ataxia first noted in juvenile to young adult dogs aged from six months to four years. Dogs develop pronounced hypermetria, a truncal sway and intention tremor, and signs progress to cause severe gait disturbances. Cerebellar atrophy can be identified by magnetic resonance imaging (MRI) [28] and histopathological findings include loss of Purkinje cell, granule cell and molecular layer neurons causing atrophy of the cerebellar cortex. In more detailed work on Gordon Setters, profound changes in cerebellar neurotransmitter levels and synapses have been described [29], along with the development of Purkinje neuron axonal spheroids [30].
The aim of this project was to investigate the genetic cause of hereditary ataxia in Old English Sheepdogs. To that effect, we genotyped an extended family of Old English Sheepdogs that suffer from canine hereditary ataxia, and identified a chromosomal locus associated with the trait using linkage and genome-wide association analyses. Sequence capture of the associated region was performed to facilitate fine mapping and this identified a mutation that segregated with the condition. Dogs from other breeds affected with cerebellar degeneration were also screened for the mutation and the identical mutation was found in Gordon Setters with canine hereditary ataxia.
Targeted sequence capture was performed to sequence the entire genomic region from 34 MB to 46 MB on CFA4, thereby covering the whole homozygous region. Six dogs were sequenced, three cases, and three controls, including two obligate heterozygotes and one unrelated, phenotypically normal dog. A Roche NimbleGen array (Madison, WI) was designed to provide coverage of 95.7% of the region using unique probes. The design was reviewed to ensure that all predicted and known gene exons had adequate coverage. Following next-generation sequencing on Illumina's Hi-Seq 2000 machine (Illumina; San Diego, CA), approximately 96% of the 12,000,000 targeted bases had at least 2× coverage with approximately 72% having at least 30× coverage. Quantitative PCR confirmed that the region of interest was enriched appropriately.
Data processing using the Genome Analysis Toolkit (GATK) [35] revealed 40,711 total variants (32,475 SNPs and 8,236 indels). After filtering the total variants for known SNPs and those present in all six sequenced dogs, 28,061 variants remained of which only 288 were present in coding regions. Screening of these 288 variants revealed nine SNPs that segregated in an autosomal recessive inheritance pattern, six of which resulted in an amino acid change and were considered potential causative mutations for the disease trait. The six SNPs were located in the genes RGR, RAB24, NSD1, GPRIN1, and CDHR2 (Table S1).
The six SNPs identified as mutations of interest were verified by Sanger sequencing in the six dogs initially sequenced. Additional cases and controls were genotyped on the six SNPs (Table S2). SNPs in RGR, GPRIN1, and CDHR2 did not segregate with phenotype in this larger population and were eliminated from further analysis. The two SNPs present in RAB24 and NSD1 showed segregation in the cases and controls consistent with an autosomal recessive mode of inheritance (Table S2). When these two SNPs were tested for association with the trait in the same 53 dogs as the GWAS, the p-values were both highly significant at pbonferroni = 5.7×10−9. When additional dog breeds were tested (Table S3), the NSD1 SNP was present in the heterozygous state in three neurologically normal Labrador Retrievers (from a group of 7 dogs) and two neurologically normal Standard Poodles (from a group of 7 dogs). Cerebellar degeneration has not been reported in Standard Poodles but has been reported in Labrador Retrievers [36],[37]. However, an affected Labrador retriever was genotyped for this SNP and it was not present in the affected dog (see results below). The presence of this SNP in five of 14 dogs from two breeds implies a relatively high prevalence of this polymorphism within these breeds. The lack of reports of hereditary ataxia in one of these breeds and the absence of this SNP in an affected dog from the other breed make it unlikely that it is a pathogenic mutation and focused our attention on RAB24.
The RAB24 SNP polymorphism was an A to C transversion located at position 113 in the first of its eight exons (Figure 2e) and it produced an amino acid change from glutamine (Q) to proline (P) at position 38. In order to investigate the hypothesis that the Rab24 p.Q38P change was the causative mutation for hereditary ataxia in Old English Sheepdogs, a total of 376 Old English Sheepdogs were genotyped, all of which came from case blood lines, including the 14 cases used in the GWAS and an additional six cases. Of these dogs, all 20 confirmed cases were homozygous for the alternate allele. 109 controls were heterozygotes, four controls were homozygous for the alternate allele while the remainder were homozygous for the wild type allele. This cohort of dogs had been sampled approximately 20 years previously and at the time, their owners provided the dogs' phenotype. All dogs were old enough to be expected to exhibit neurological signs if affected. Of the four dogs that were reported as normal but exhibited the case genotype, two were littermates of affected dogs, with one confirmed affected parent and one parent confirmed as a carrier. The remaining two dogs were descended from parents who were carriers. Physical examinations were not performed on these dogs by a veterinarian and so their phenotype could not be confirmed.
The effect of the Rab24 p.Q38P change on protein function was investigated using two different homology-based on-line tools, Polyphen-2 [38] and SIFT [39]. Both predicted that the change would probably damage protein function, with a score of 0.989 for Polyphen-2 (sensitivity: 0.72 and specificity: 0.97) and 0.01 for SIFT. The Rab24 protein is a GTPase from the large Rab protein family important in vesicle trafficking, endocytosis and exocytosis [40]. Several Rab proteins, including Rab24, have now been shown to play a vital role in autophagy [41], the process by which proteins and organelles are moved to lysosomes for degradation. Domains known to be important to Rab protein function include the nucleotide binding and Mg2+sites necessary for GTPase activity; two switch regions play a vital role in facilitating GTPase activity. The p.Q38P change lies in a highly conserved amino acid (Figure 3a) located within the putative switch I region, suggestive of an effect on GTP binding (Figure 3b).
An additional 254 DNA samples from Old English Sheepdogs were obtained from the Orthopedic Foundation for Animals Inc. (OFA) CHIC DNA repository. Detailed phenotypic information was not available on these dogs although none were reported to have canine hereditary ataxia on health questionnaires completed by owners at the time of DNA submission. This cohort of dogs came from approximately 70 different breeding kennels. These samples were genotyped in order to determine the prevalence of the mutation in a wider population of dogs. Twenty-eight of these dogs were heterozygous while the remainder was homozygous for the reference allele, giving an overall alternate allele prevalence in the original pedigrees studied and the random selection of dogs (a total of 630 dogs) of 14.3%.
To test for the potential of the mutation to discriminate cases and controls, receiver operating characteristic (ROC) curve analysis was performed in the total population of Old English Sheepdogs genotyped. The area under the curve was 99.5% (S.E. = 0.0023, 95% confidence interval = 0.9909–0.9999), which was highly statistically significant (p<0.001).
Hereditary ataxia is not unusual in dogs and has been reported in numerous breeds. In order to determine whether the p.Q38P mutation was found in affected dogs from other breeds, DNA samples from a Dalmatian, a beagle (both confirmed by neurological evaluation and the presence of cerebellar atrophy on MRI of the brain), 2 Rhodesian Ridgebacks (necropsy confirmed), 2 Gordon Setters (one of which was necropsy confirmed and the other confirmed by neurological evaluation and the presence of cerebellar atrophy on MRI of the brain), 2 Scottish Terriers (one MRI and one necropsy confirmed) and a Labrador retriever (diagnosed by neurological evaluation and clinical history only) were genotyped for the RAB24 and the NSD1 SNPs. In addition, all exons of the RAB24 gene were sequenced in these affected dogs. The alternate allele for RAB24 or NSD1 was not present in any of the breeds except the affected Gordon Setters. Two heterozygous synonymous SNPs were identified in exon 3 of RAB24 in the beagle which were also present in the Old English Sheepdog sequencing data. Heterozygous non-exonic SNPs were also identified in the Dalmatian. None of these were consistent with the mode of inheritance or pathologic in nature.
A total of 18 affected Gordon Setters were then genotyped, all of which were homozygous for the RAB24 mutation. DNA from ten of these cases was obtained from archived material from a research dog colony [19]. An additional 82 normal Gordon Setters were genotyped from Scandinavia and the US and 24 dogs were heterozygotes and 58 were homozygous for the wild type allele. None were homozygous for the alternate allele. Excluding the 10 cases that were archival material and therefore not part of the general breeding population, there was an alternate allele frequency of 22.2%. The phenotype of cerebellar degeneration in Gordon Setters has been described in detail [19], [26], [27] and is identical to the description of Old English Sheepdogs in terms of age of onset and progression of signs. To compare the genetic background of Gordon Setters and Old English Sheepdogs further, a cohort of seven affected and 26 control Gordon Setters were genotyped on the Illumina Infinium Canine HDBeadchip and the haplotypes on CFA4 were compared to those of the Old English Sheepdogs. This revealed an identical region of homozygosity extending from 39,245,536 bp to 41,172,873 bp in affected dogs from both breeds, including the NSD1 mutation also identified in the region (Figure S1), suggesting the mutation dates back to a common European ancestral dog population, from which these two separate breeds were founded. This region contains 29 genes inferred from human sequence data (Table S4). When targeted sequencing was performed in the Old English Sheepdogs, mean sequencing coverage of this region of shared homozygosity was 47×, making it unlikely that additional variants were missed. These findings are supportive of the hypothesis that the RAB24 mutation is the causative mutation of hereditary ataxia in these two breeds of dog.
In order to determine whether this mutation was present in other breeds of dog representing diverse ancestral lineage, DNA samples were collected from at least eight individuals from breeds in each of 10 breed clusters reported as related ancestrally [42], [43]. A total of 194 individuals from 43 different breeds were genotyped (Table S3). All additional breeds were homozygous for the wild type allele on the RAB24 SNP.
Expression levels of RAB24 in the cerebellum were compared between case (n = 4) and control (n = 2) dogs by qRT-PCR. There was no significant difference in level of RAB24 expression between cases and controls (p-value = 0.71). The A>C change in the RAB24 transcript was present in all affected dogs and absent in the control dogs. The predicted exon/intron boundaries were also confirmed.
Samples of brain from eight affected Old English Sheepdogs ranging in age from 2.5 to 13 years, and one affected 2.5-year-old Gordon Setter, and from 10 neurologically normal, age matched dogs were fixed in 10% neutral buffered formalin and embedded in paraffin for histological evaluation. Six of the eight Old English Sheepdogs were examined at the University of Pennsylvania in 1999 and the paraffin embedded blocks were retrieved from the archives. In the remaining two Old English Sheepdogs, the brains were harvested by a local veterinarian following euthanasia, placed in formalin and shipped to NCSU for paraffin embedding. The Gordon Setter was euthanized at NCSU and the brain was removed immediately and placed in formalin within 30 minutes of euthanasia. Sections were stained with hematoxylin and eosin, Periodic Acid Schiff (PAS, to evaluate glycogen storage products), Bielschowsky silver stain (to evaluate axonal processes), and luxol fast blue (to evaluate myelination). Immunohistochemical staining was performed for GFAP, ubiquitin, and Rab24. Samples of cerebellum from the affected Gordon Setter were fixed in McDowell's and Trump's 4F:1G fixative and processed for electron microscopy.
Pathological changes were largely restricted to the cerebellum, with the majority of changes affecting the cerebellar cortex. There was dramatic loss of Purkinje neurons, with atrophy of the molecular and granular layers (Figure 4a). Vacuoles were visible in the white matter throughout the cerebellum and axonal spheroids were identified in the granular layer (Figure 4a). In addition, there were vacuoles in the cerebellar peduncles, the vestibular and cochlear nuclei and the nucleus of the dorsal trapezoid body. There were minimal changes in the cerebellar nuclei. The Bielschowsky stain highlighted the processes of basket cells and the lack of Purkinje neurons (Figure 4b). There was no evidence of glycogen accumulation on PAS stained sections and the luxol fast blue staining confirmed the presence of myelin around the axonal spheroids identified in the granular layer. Immunostaining for GFAP showed increased astrocytic expression and highlighted mild to moderate astrocytosis.
Ubiquitin immunostaining of controls revealed granular accumulations of ubiquitin positive material in the white matter, the density of which increased with age (Figure 5a). In older control dogs, some positive staining was also seen in the granular layer and very fine positive granules were found at the junction of the molecular and granular layers around Purkinje neurons. In cases, the ubiquitin positive staining within the white matter was comparable to the age matched controls. However, there were also multiple, large ovoid bodies containing ubiquitin positive material staining in a punctate pattern, lying within the granular layer, at the junction of the granular and molecular layers, and in the cerebellar white matter (Figure 5b and c). In some instances, these ubiquitin positive bodies appeared to be emanating from a Purkinje neuron and to co-localize with axonal spheroids seen on the hematoxylin and eosin stained sections (Figure 5b and d). Ubiquitin positive bodies were limited to the cerebellum of cases, with no inclusions seen elsewhere in the brain. In addition, some Purkinje neurons, molecular layer neurons and Golgi neurons within the granular layer had strong diffuse cytoplasmic ubiquitin staining (Figure 5b). The axonal spheroids were examined on electron microscopy and were packed with organelles such as mitochondria and the Golgi apparatus, and vesicular structures, many of which had the classic double membrane of autophagosomes (Figure 6).
There was no significant difference in the intensity or pattern of Rab24 immunostaining in the cerebellum of cases and controls. There was positive staining within the cytoplasm of Purkinje neurons; the staining was granular and located in an eccentric perinuclear position as described in cell culture studies [44]–[46] (Figure 7a). The terminal dendrites of the basket cells on Purkinje neurons stained intensely (Figure 7a). In addition, neurons in the molecular layer, and the granular layer stained positively, as did the neurons of the deep cerebellar nuclei (Figure 7b) and oligodendrocytes (Figure 7c) throughout the white matter of the cerebellum. Axonal spheroids were negative for Rab24.
The results of our work implicate a founder mutation in the GTPase Rab24 as the cause of autosomal recessive hereditary ataxia in both the Old English Sheepdog and the Gordon Setter although dysfunction of this protein has not been established. Genome wide linkage and association studies in a cohort of related Old English Sheepdogs identified a strong association between the disease phenotype and an approximately 12 MB region of homozygosity on CFA 4 in affected dogs. High throughput sequencing identified an A>C variant that predicted a Q>P change in the Rab24 protein. Evaluation of Gordon Setters with the same clinical phenotype revealed the same mutation in affected dogs and a shared block of homozygosity extending over 1.9 MB. While LD extends over long distances within breeds, across breeds, LD rapidly drops off. This is a characteristic resulting from two different bottlenecks in the history of domesticated dogs, the first dating back to their separation from wolves over 10,000 years ago, and the second more recent bottleneck occurring as modern breeds developed from a limited number of founder dogs only a few hundred years ago [47], [48]. These breeds of dog are placed in different ancestral clusters with Old English Sheepdogs classified as herding dogs and Setters clustering with working dogs. This long shared haplotype between the two breeds is unusual, and the implications of our findings are that this mutation dates back to a time before these two breeds of dog were developed, that the two breeds have a shared ancestry, and that this shared founder mutation is the cause of hereditary ataxia in these breeds. This has been found with other mutated canine genes, for example, the mutation in prcd-PRA [49].
Rab24 is an atypical member of the large Rab family of small GTPases [45]. These enzymes play a vital role in membranous transport within the cell allowing movement of cell organelles, and endocytosis and exocytosis (reviewed in [40], [50]). They work in concert with the SNARE proteins to bridge membranes and drive fusion. The mechanism by which they achieve this has been well characterized for certain members of the family, and depends on their GTP state, and their ability to prenylate and thus cycle on and off membranes. Rab24 is unlike the other members of the family, having poor GTPase activity, and reduced prenylation and its mechanism of action remains elusive [45]. It has been shown to localize to the cis Golgi and ER and to co-localize with autophagosomal markers such as LC3, and it is proposed to play a role in the late stages of autophagy related to the fusion of the autophagosome and lysosome [46]. Proteomic evaluation of autophagy networks also show interactions between Rab24 and other vesicle trafficking autophagic proteins [51]. While expressed at low levels in many tissues, it is expressed most highly in the brain [44] and there is evidence that it is upregulated during neuronal differentiation [45] and as a response to nerve injury [52]. There has been speculation that Rab24 dysfunction might play a role in neurodegenerative disease because mutations placed in a region known as the G2 domain that reduce the affinity of Rab proteins for GTP produce nuclear inclusions that disrupt the nuclear membrane, and stain positive for ubiquitin and Hsp70, typical of protein aggregates in polyglutamine diseases [53]. However, more recently, Rab24 has been shown to play a role in cell division [54]. The mutation described here lies in the putative switch 1 region, important for GTP binding [55]. In support of this region being functionally important, in another Rab associated disease, Griscelli Syndrome, a mutation in the switch 1 region of Rab27A results in a profound phenotype due to failure of the Rab protein to interact with its target melanophilin [56]. Mutations in RAB7 have been associated with Charcot-Marie-Tooth type 2B neuropathy, a progressive neurodegenerative condition of the peripheral nervous system. One of the mutations described lies immediately adjacent to a highly conserved GTP binding domain [57]. Rab7 is involved in transport between endosomes and lysosomes, demonstrating the importance of subcellular trafficking in neuronal health.
There are two main systems that allow for turn-over of cellular organelles and proteins, the ubiquitin proteasome system (UPS) and autophagy. The ubiquitin proteasome system (UPS) is well described and is a system by which short-lived regulatory and misfolded proteins undergo non-lysosomal degradation [58], [59]. The UPS system has been shown to be vital to normal development of the central nervous system, to synaptic plasticity and to cellular homeostasis. Reflective of these important roles, many neurodegenerative diseases have been linked to abnormalities in the UPS system [58], [59]. More recently, attention has shifted to the role that autophagy might play in neurodegenerative disease. Autophagy is the process by which more long-lived proteins and organelles are incorporated into autophagosomes for delivery to vacuoles or lysosomes for degradation. There is abundant cross communication between the UPS and autophagy systems, with specific ubiquitinated proteins being moved to autophagosomes for disposal. As a result of this convergence of disposal pathways, dysfunction of autophagy can result in ubiquitin accumulation [60], [61] and accumulation of autophagosomes and ubiquitin positive inclusions, both evident in our affected dogs, are considered to be hallmarks of neurodegenerative diseases (reviewed in [62]). However, these findings are quite non-specific, being present in most neurodegenerative diseases.
Autophagy is believed to be vital for cell survival in several different ways. Constitutive autophagy plays an important role as a basal source of energy in cells with high metabolic needs, such as the Purkinje neuron, in the global turnover of cellular organelles and in the clearance of potentially toxic protein aggregates [62]–[65]. While there are a variety of hypotheses for the role of autophagy in neurodegenerative diseases [66]–[68], the most compelling evidence that a primary defect in autophagy can induce neurodegenerative disease was generated by two studies in which neuron specific dysfunction of two proteins involved in the formation of autophagasomes, Atg5 and Atg7, was induced in mice [65], [69]. In both studies, mice developed progressive motor incoordination and balance deficits accompanied by progressive neurodegeneration. There was dramatic loss of Purkinje neurons accompanied by axonal swellings and development of ubiquitin positive inclusions and diffuse intraneuronal ubiquitin accumulation. The clinical and neurodegenerative phenotypes of these mice are similar to the dogs in our study in which we saw dramatic Purkinje neuron loss, axonal spheroids containing autophagosomes and large ubiquitin positive inclusions in addition to more diffuse intracellular ubiquitin accumulation. However, we were unable to demonstrate an alteration in the level of expression of the RAB24 gene or Rab24 protein by qPCR or immunohistochemically. It can be argued that Purkinje neurons are a natural target for many different pathological processes due to their high metabolic activity, requiring higher levels of protein and organelle turn over. Indeed, in hereditary ataxia affecting the Finnish Hound, a defect in SEL1L, a gene important in the endoplasmic reticulum associated degradation (ERAD) process that targets misfolded proteins to the UPS, causes early onset rapidly progressive Purkinje neuron loss [23]. Taken together, these two naturally occurring canine models of hereditary ataxia suggest that protein and organelle turnover is of vital importance to the maintenance of neuronal, and in particular, Purkinje neuron health. Additional in vitro work needs to be completed to demonstrate Rab24 dysfunction as a result of this mutation.
We have described a mutation that results in an amino acid change in Rab24, a protein as yet poorly understood, but believed to play a role in the late stages of autophagy. Our findings of Purkinje neuron degeneration in this naturally occurring canine neurodegenerative condition support the evidence that Rab24 is necessary for autophagy, and our clinical and histopathological findings are strongly reminiscent of the mouse studies demonstrating that defects in autophagy produce neurodegeneration targeting the Purkinje neuron. This finding may provide a novel tool for the investigation of the mechanisms of autophagy and defects in Rab proteins should be considered when investigating neurodegenerative diseases.
Old English Sheepdogs with canine hereditary ataxia were identified by referral from veterinarians, breeders and owners. DNA samples and pedigrees from these U.S. and Canadian dogs and their unaffected family members were collected. Affected status was determined by compatible clinical signs of cerebellar disease including ataxia, typical age of onset, physical exam, and neurological exam. Cerebrospinal fluid analysis was undertaken in many dogs, as well as a necropsy, when possible. All protocols were performed with approval from North Carolina State University's Institutional Animal Care and Use Committee. All normal dogs reached the age of four years of age with no evidence of clinical signs of ataxia. Dogs that could not definitively be classified as “affected” or “normal” based on collected information were classified as of “undetermined” status. DNA was extracted from whole blood using QIAamp DNA Blood Midi Kit (Qiagen; Valencia, CA), frozen tissues; DNeasy Blood and Tissue Kit (Qiagen; Valencia, CA), buccal swabs; QIAamp DNA Mini Kit (Qiagen; Valencia, CA), and from saliva using Oragene Animal kit (DNA Genotek; Kanata, Ontario). DNA concentrations were measured using a ND-1000 UV-Vis NanoDrop spectrophotometer (Thermo Scientific, Wilmington, DE).
Related dogs were genotyped using a genome-wide panel of 311 canine microsatellite markers (representing an average of 9 MB resolution), organized into multiplex PCR groups [31]. Four fluorescent labels (FAM, VIC, NED and PET) were incorporated into the PCR primers to allow multiplexing. PCR fragments were analyzed on an ABI-3700 automated Genetic Analyzer (Applied Biosystems), and genotypes were assigned using GeneMapper v3.7 software (Applied Biosystems). Homozygous (uninformative) markers were excluded from further analysis.
Genotyping of SNPs was performed using Illumina's Canine SNP20 (22,362 SNPs) and CanineHD (170,362 SNPs) genotyping beadchips (Illumina; San Diego, CA). Nine cases and 13 control dogs were genotyped on the SNP20 chip and five cases and 28 controls on the CanineHD chip. One control dog was genotyped on both chips, enabling the comparison of genotype calls between them. The assays were performed at the National Institutes of Health's Laboratory of Neurogenetics (Bethesda, MD) according to the manufacturer's instructions. The amplified DNA products were imaged using a BeadArray Reader (Illumina; San Diego, CA) and analyzed using Illumina's Bead Studio and Genome Studio software (Illumina; San Diego, CA).
Data was pruned such that individuals with less than a 95% call rate and those having a significant number of Mendelian errors were removed from further analysis. SNPs having a minor allele frequency of less than 1%, missing genotype calls greater than 10% and showing inconsistent calls between the two chips were also removed from further analysis. The pruned dataset was then used to perform a case-control and family-based association analysis. The PLINK toolset v1.07 (http://pngu.mgh.harvard.edu/~purcell/plink/) was used to perform data pruning and the case-control allelic and genotypic association tests using a Bonferroni correction for multiple comparisons [33]. As most of the Old English Sheepdogs genotyped belonged to a large pedigree, a family-based association test was also performed using the Family Based Association Test (FBAT) toolkit (v2.0.3) (http://www.hsph.harvard.edu/~fbat/fbat.htm) [34]. The FBAT toolkit improves upon the traditional transmission disequilibrium test (TDT) [70] by handling factors such as missing parents, additional family members, different genetic models and phenotypes, as well as controlling for false positive associations due to population structure [34]. A Bonferroni correction for multiple comparisons was implemented using JMP Genomics (SAS; Cary, NC) with adjusted P-values less than 0.05 considered significant.
Receiver operator characteristic curve analysis was performed using Stata v10 (www.stata.com).
A 12 MB genomic region of interest was targeted using a custom array designed and manufactured by Roche NimbleGen (Madison, WI). 19,720 tiled probes approximately 60–90 bp in length covered approximately 96% of the targeted bases between 34,000,000–46,000,000 bp on CFA4. Unique probes were determined using the Sequence Search and Alignment by Hashing Algorithm (SSAHA) [71]. Targeted bases not captured were due to SSAHA's inability to determine valid probes possibly resulting from non-unique sequence, repetitive sequence, homopolymer runs or ambiguous bases.
Three micrograms of genomic DNA from three cases and three controls was used to prepare libraries for sequencing. DNA samples were fragmented by sonication (Covaris; Woburn, MA) and the fragments end-repaired, A-tailed, and ligated to indexing oligonucleotide adapters using NEBNext reagents (New England Biolabs; Ipswich, MA). Indexing adapters were provided by the Broad Institute (Cambridge, MA). The indexed DNA fragments were enriched for by PCR using AccuPrime (Life Technologies; Grand Island, NY) and Phusion (New England Biolabs, Ipswich, MA) DNA polymerases. DNA purification was done using QIAquick and MinElute PCR purification kits (Qiagen; Valencia, CA). Size selection and purification was done using Agencourt AMPure XP beads (Beckman Coulter; Beverly, MA). Analysis of DNA libraries was done using the Agilent Bioanalyzer DNA 1000 (Agilent Technologies; Santa Clara, CA) and Quant-iT dsDNA HS assay (Life Technologies, Grand Island, NY). Libraries were hybridized onto the array using a NimbleGen hybridization system (Roche NimbleGen; Madison, WI) at 42°C for 70 hours. Arrays were washed and samples eluted using NimbleGen's elution system (Roche NimbleGen; Madison, WI). Post-capture amplification was done using the primers 5′-AAT GAT ACG GCG ACC ACC GAG-3′and 5′-GAA GCA GAA GAC GGC ATA CGA-3′ with Phusion (New England Biolabs; Ipswich, MA) and AccuPrime (Life Technologies; Grand Island, NY) enzymes. Quantitative PCR (qPCR) using QuantiFast SYBR Green PCR kit (Qiagen; Valencia, CA) was done to estimate relative fold-enrichment of the target region (Table S5).
Paired-end sequencing was done on Illumina's Hi-Seq 2000 machine (Illumina; San Diego, CA) at the Broad Institute (Cambridge, MA).
Sequence reads were aligned to the canine reference genome CanFam2 using the Burrows-Wheeler Aligner (BWA) [72]. SNPs and insertion/deletions (indels) were determined using the Genome Analysis Toolkit (GATK; http://www.broadinstitute.org/gatk/) [35]. Sequence data was processed according to the current best practices for data analysis found on the online GATK Wiki (http://www.broadinstitute.org/gsa/wiki/index.php/Main_Page) and included; initial SNP and indel calling, correcting alignment errors due to indels and inaccurate base quality scores, SNP and indel calling after the realignment and recalibration, and filtering the data using standard filtering parameters. Sequence and variants were viewed with the Integrative Genomic Viewer (IGV) [73]. Variants were identified in which all three affected dogs were homozygous for the non-reference allele, the two control carriers were heterozygous and the final control dog was either heterozygous or homozygous for the reference allele. Variants resulting in amino acid changes of coding regions of the genome were considered potential causative mutations for hereditary ataxia.
Variants considered as potential causative mutations were genotyped via Sanger sequencing in Old English Sheepdogs and additional dog breeds. The RAB24 gene was also sequenced in affected dogs of breeds in which the A>C RAB24 variant was not present. Primers were designed using Primer3 (http://frodo.wi.mit.edu/) [74] or NCBI's Primer-BLAST (http://www.ncbi.nlm.nih.gov/tools/primer-blast/) [75] (Table S6). PCR reactions included: 10× Buffer B (Thermo Fisher Scientific; Waltham, MA), 25 mM MgCl2 (Thermo Fisher Scientific; Waltham, MA), 10× MasterAmp (Epicentre Biotechnologies Madison, WI), 25 mM dNTPs (Apex BioResearch Products, San Diego, CA), 10 µM forward and reverse primer (IDT; Coralville, IA), Taq DNA polymerase (Apex BioResearch Products; San Diego, CA) and molecular grade water to a volume of 27 µL. DNA amounts varied from approximately 25 ng to 100 ng per reaction. Thermocycler conditions varied by primer set (Table S7). PCR products were purified using either Agencourt AMPure XP kit (Beckman Coulter; Beverly, MA), QIAquick PCR Purification Kit (Qiagen; Valencia, CA), or MinElute PCR Purification Kit (Qiagen; Valencia, CA). The sequencing of each purified PCR product (20–40 ng/µl for 8 µl) was carried out using the same forward and reverse primers used for PCR. Sequencing was performed by Eurofins MWG Operon (Huntsville, AL) following the standard BigDye Terminator v3.1 manufacture's protocol (Applied Biosystems; Foster, CA) with capillary electrophoresis carried out on the ABI 3730×l DNA Analyzer (Applied Biosystems; Foster, CA).
Sections of cerebellum were collected from euthanized affected Old English Sheepdogs (n = 2), affected Gordon Setters (n = 2) and normal beagles (n = 2), and the tissue immediately frozen in liquid nitrogen. RNA was extracted from approximately 30 mg of the frozen tissue using Qiagen's RNeasy Mini Kit (Qiagen; Valencia, CA) and concentrations measured using a ND-1000 UV-Vis spectrophotometer (Thermo Scientific; Wilmington, DE).
Reverse transcription was performed using the Stratagene AffinityScript Multiple Temperature cDNA synthesis kit (Agilent Technologies; Santa Clara, CA) and 0.5 µg oligo(dT) and 0.2 µg random primers. Thermocycler conditions were set to 25°C for 10 minutes, 42°C for 5 minutes, 25°C for 60 minutes, and 72°C for 15 minutes.
Primers to amplify the mRNA were designed using Primer3 (http://frodo.wi.mit.edu/) [74], with forward primer 5′-CGTGTCTCCAGGCGTAGC-3′ and reverse primer 5′-ACTGGGGGTAGCTCAGAC-3′. The approximately 850 bp PCR product was excised from an ethidium bromide stained 2% agarose gel and cleaned using the QIAquick Gel Extraction Kit (Qiagen; Valencia, CA). The cDNA of an affected Old English Sheepdog and Gordon Setter as well as a normal beagle were sequenced using the same forward and reverse PCR primers. Sequencing was performed as was done for genotyping the variants.
Quantitative PCR was performed using Applied Biosystems' StepOne Plus instrument and Invitrogen's Power SYBR Green Master Mix (Life Technologies; Grand Island, NY). RAB24 primers were designed using Primer3 (http://frodo.wi.mit.edu/) [74] and PerlPrimer (http://perlprimer.sourceforge.net/) [76]. The housekeeping gene RPS19 (primers published [77]) was used for normalization. 20 µl reaction volumes containing 200 nM primers were done in triplicate. Cycling conditions consisted of 95°C for 10 minutes followed by 40 cycles of 95°C for 15 seconds and 60°C for 1 minute with a melt curve performed at completion. Reaction efficiencies were calculated using a 6 point 2∶1 dilution series of 100 ng cDNA. RAB24 and RPS19 efficiencies were estimated to be 90% and 88% respectively with both r2 values of 0.999. Relative expression differences were calculated using the ΔΔCt-method and statistical significance determined using Microsoft Excel (2010) to perform an unpaired t-test on the relative differences between cases and controls.
For histopathology, tissues were fixed in 10% neutral buffered formalin, embedded in paraffin, sectioned to five micrometers and stained with hematoxylin and eosin, PAS, Luxol Fast Blue/Cresyl Violet and Bielschowsky Silver Stain. Immunohistochemistry for GFAP, ubiquitin and Rab24 was performed on formalin fixed paraffin embedded brain tissue. Rabbit polyclonal antibodies were used for all three antigens purchased from Dako (GFAP), Santa Cruz Biotechnology (ubiquitin) and Proteintech Group Inc. (Rab24). 5 µm sections were cut, deparaffinized and rehydrated. GFAP and ubiquitin immunostaining were performed using the Dako Autostainer. Rab24 immunostaining was performed manually. The autostaining method included a five-minute retrieval for GFAP using Dako Proteinase K (Dako; Carpinteria, CA), followed by blocking of endogenous peroxidase activity using 3% hydrogen peroxide for 10 minutes. The primary antibody was applied (concentrations and incubation time and conditions are given in Table S8), and the product was visualized by incubation with Dako Envision and rabbit polymer (Dako; Carpinteria, CA) for 30 minutes followed by application of DAB for five minutes. Slides were counterstained with hematoxylin, dehydrated through sequential alcohol immersion to xylene and cover-slipped. Rab24 immunostaining involved a citrate retrieval using a Pascal Pressurized Heating Chamber at 120°C, followed by blocking of endogenous peroxidase activity using 3% hydrogen peroxide for 10 minutes. A protein block was then performed using normal goat serum (protein block/normal goat serum, Biogenex; Fremont, CA) applied for 20 minutes, followed by incubation with rabbit anti-Rab24 at a concentration of 1∶50 for 1 hour at room temperature. Slides were washed with phosphate buffered saline (PBS) then biotinylated goat anti-rabbit antibody (Link; Biogenex, Fremont, CA) was applied diluted to 1∶20 and incubated for 20 minutes at room temperature. Following washing, peroxidase conjugated streptavidin (Label; Biogenex, Fremont, CA) was applied and incubated for a further 20 minutes at room temperature. The product was developed under the microscope using DAB for about one minute to an appropriate level of staining. The slides were counterstained with hematoxylin, dehydrated through sequential alcohol immersion to xylene and cover-slipped.
Sections were prepared for transmission electron microscopy from freshly harvested cerebellar tissue as described previously [11]. Samples of cerebellar cortex were cut into approximately 1 mm2 cubes and fixed in McDowell's and Trump's 4F:1G fixative prior to incubation in 1% osmium tetroxide for one hour. Following this period of fixation, samples were placed in a 1∶1 mixture of Spurr resin (EMS Spurr resin kit 14300, Hatfield, PA) and acetone for 30 minutes and then placed in 100% resin changed three times at two hourly intervals. The final change of resin was polymerized at 70°C for eight hours. Semi-thin sections (0.25 µm) were cut, stained with 1% toluidine blue-O in 1% sodium borate and used to identify areas of interest. Ultrathin sections (70–90 nm) of these areas were stained with methanolic uranyl acetate (EMS 22400, Hatfield, PA), followed by lead citrate, and examined by TEM (FEI/Philips EM208S TEM, Oregon, USA). Reagents were obtained from Fisher Scientific (Pittsburgh, PA) unless otherwise indicated.
All animals evaluated in this study were privately owned pets. DNA samples were obtained with client consent and approval of the Institutional Animal Use and Care Committee or from DNA banks. Euthanasia and necropsies were performed at the owners' request.
|
10.1371/journal.pgen.1008319 | Distinct roles of RAD52 and POLQ in chromosomal break repair and replication stress response | Disrupting either the DNA annealing factor RAD52 or the A-family DNA polymerase POLQ can cause synthetic lethality with defects in BRCA1 and BRCA2, which are tumor suppressors important for homology-directed repair of DNA double-strand breaks (DSBs), and protection of stalled replication forks. A likely mechanism of this synthetic lethality is that RAD52 and/or POLQ are important for backup pathways for DSB repair and/or replication stress responses. The features of DSB repair events that require RAD52 vs. POLQ, and whether combined disruption of these factors causes distinct effects on genome maintenance, have been unclear. Using human U2OS cells, we generated a cell line with POLQ mutations upstream of the polymerase domain, a RAD52 knockout cell line, and a line with combined disruption of both genes. We also examined RAD52 and POLQ using RNA-interference. We find that combined disruption of RAD52 and POLQ causes at least additive hypersensitivity to cisplatin, and a synthetic reduction in replication fork restart velocity. We also examined the influence of RAD52 and POLQ on several DSB repair events. We find that RAD52 is particularly important for repair using ≥ 50 nt repeat sequences that flank the DSB, and that also involve removal of non-homologous sequences flanking the repeats. In contrast, POLQ is important for repair events using 6 nt (but not ≥ 18 nt) of flanking repeats that are at the edge of the break, as well as oligonucleotide microhomology-templated (i.e., 12–20 nt) repair events requiring nascent DNA synthesis. Finally, these factors show key distinctions with BRCA2, regarding effects on DSB repair events and response to stalled replication forks. These findings indicate that RAD52 and POLQ have distinct roles in genome maintenance, including for specific features of DSB repair events, such that combined disruption of these factors may be effective for genotoxin sensitization and/or synthetic lethal strategies.
| We have examined the role of two factors, RAD52 and POLQ, in genome maintenance pathways. While these factors are biochemically distinct, they are both synthetic lethal with loss of the BRCA1 and BRCA2 tumor suppressor genes, and hence are emerging therapeutic targets. Furthermore, RAD52 and POLQ have been implicated in chromosomal break repair events that use flanking repeats to restore the chromosome. We identified distinct features of chromosomal break repair events that are mediated by RAD52 vs. POLQ. Additionally, we have found that combined disruption of RAD52 and POLQ causes at least additive hypersensitivity to cisplatin and a synthetic reduction in replication fork restart velocity. These findings indicate that POLQ and RAD52 have distinct roles in genome maintenance, such that combined disruption of these factors could be a potential therapeutic strategy.
| Exploiting synthetic lethal relationships in cancer cells has emerged as a promising therapeutic approach [1, 2]. As a key example, cells deficient in BRCA1 or BRCA2 are hypersensitive to inhibitors of Poly-ADP-ribose Polymerase (PARP) [1, 2]. Both BRCA1 and BRCA2 are important for homology-directed repair (HDR) of chromosomal breaks, which involves RAD51-mediated invasion of a homologous sequence to template nascent DNA synthesis [3]. In addition, BRCA1 and BRCA2 are important for protection of stalled replication forks by blocking recruitment of the MRE11 nuclease to reversed forks [4–6]. PARP inhibitors appear toxic to cells deficient in BRCA1 and BRCA2, by causing DNA lesions that require HDR for repair, and/or replication defects that require protection from degradation via BRCA1 and BRCA2 [6]. However, since PARP inhibitors are effective in only a fraction of cancer patients [7], it is important to develop additional targets for this synthetic lethality approach.
In particular, deficiencies in BRCA1 or BRCA2 are synthetic lethal with disruption of either RAD52 or POLQ [8–11], which have distinct biochemical activities. RAD52 forms multimeric ring structures and has a strong affinity for ssDNA [12]. Moreover, RAD52 is capable of facilitating the displacement of the ssDNA binding protein replication protein A (RPA) to anneal complementary strands of ssDNA [13, 14]. RAD52 also interacts with dsDNA, although with a weaker affinity than with ssDNA [15]. Consistent with a role in promoting stable DNA annealing, RAD52 appears to protect dsDNA from force-induced strand separation [16]. POLQ is an A-family DNA polymerase that has also been shown to anneal complementary ssDNA [17]. The polymerase domain of POLQ has a unique structure that consists of three insertion loops, which are not conserved among other A-family DNA polymerases [18]. This distinct polymerase domain structure allows for the interaction, annealing, and extension of short ssDNA primers [17, 19]. In addition to its C-terminal polymerase domain, POLQ also has an N-terminal helicase domain [20, 21].
To develop RAD52 and POLQ as therapeutic targets for synthetic lethal approaches, it is important to understand their role in genome maintenance. As one possibility, disruption of POLQ or RAD52 may cause similar effects as PARP inhibitors, e.g., cause defects at replication forks that require BRCA1 and BRCA2 [6]. Although, an additional potential mechanism of such synthetic lethality is that these factors mediate alternative DSB repair pathways to HDR. In particular, one class of repair pathways involves annealing of homologous repeat sequences that flank the break. These pathways are referred to as Single Strand Annealing (SSA) and Alternative end-joining (Alt-EJ), which generally are distinguished by the use of long vs. short repeat sequences (the latter referred to as microhomology), and the involvement of RAD52 vs. POLQ, respectively [22–28]. However, a limitation of these terms is that the precise parameters that define the mechanism of these events remain poorly understood. Such parameters include repeat length, and influence of a non-homologous intervening sequence. Thus, we refer to these events collectively as repeat-mediated repair (RMR) to avoid a presumption of mechanism.
In this study, we have sought to define whether RAD52 and POLQ have distinct vs. redundant functions in chromosomal break repair or response to replication stress. Specifically, we have examined the influence of these factors on several distinct features of DSB repair, as well as in response to genotoxic agents and replication stress. To test whether these factors have distinct (i.e., non-epistatic) roles in these aspects of genome maintenance, we have also compared cells with combined deficiency in POLQ and RAD52 vs. cells with disruption of the individual factors. Finally, we posited that RAD52 and POLQ have distinct roles in genome maintenance vs. BRCA2, due to their synthetic lethality with BRCA2 loss. Thus, we have also compared the influence of these factors vs. BRCA2 on DSB repair events and in response to replication stress.
We have sought to examine the relative roles of RAD52 and POLQ in cellular response to genotoxic stress, including distinct DSB repair events. For this, we developed cell lines with disruptions of these genes (both single and double mutants) using the RNA-guided nuclease Cas9. For our parental cell line, we used human osteosarcoma U2OS cells [29, 30], which retain intact cell cycle checkpoints [31, 32]. Notably, these cells rely on the ALT-pathway of telomere maintenance, which could possibly influence repair mechanisms [33]. Our parental cell line was also stably transfected with pFRT/lacZeo (i.e., U2OS Flp-In T-Rex) [29, 30], which is used to integrate the reporter assays described below.
We used single guide RNAs (sgRNAs) and Cas9 to generate cell lines deficient in POLQ and RAD52. To generate a POLQ-deficient cell line, we used two sgRNAs targeting exon 16 (Fig 1A). We targeted this region of POLQ to disrupt expression of the C-terminal polymerase domain, and thereby cause loss of POLQ-mediated primer extension [20]. We screened for clones with deletion mutations by PCR, and identified a clone with three mutations in exon 16 (POLQ exon 16 mutant, POLQe16m): 1) one allele with deletion of the segment between the two DSBs, causing mutation of I862 to a termination codon (I862X), 2) a second allele with an inversion of this segment causing mutation of I862 to V, and encoding another 8 amino acids followed by a termination codon (I862V8X), and 3) a third allele with a single nucleotide insertion at the 3' DSB site causing an S1152 to K mutation, and encoding 2 amino acids followed by a termination codon (S1152K2X) (Fig 1A). These mutant alleles disrupt the coding sequence for POLQ upstream of the C-terminal polymerase domain (Fig 1A). We also used Cas9 to generate a RAD52 knockout (RAD52KO) cell line, and a RAD52KOPOLQe16m cell line from the POLQe16m cell line, both of which were identified using RAD52 immunoblotting (Fig 1B). Using these cell lines, we first examined cell cycle profiles using BrdU and propidium iodide labeling, and found that RAD52KO and RAD52KOPOLQe16m cells, but not POLQe16m cells, showed a modest, but statistically significant increase in G1 cells compared to the parental cell line (Fig 1C).
To examine the response to genotoxic stress, we exposed cells to DNA damaging agents and measured clonogenic survival based on colony formation. In addition to testing the cell lines described above, we also examined POLQ using RNA-interference (RNAi). Specifically, we treated parental and RAD52KO cells with siRNAs targeting POLQ (siPOLQ), or a non-targeting siRNA (siCTRL). We confirmed that siPOLQ treatment causes depletion of the POLQ mRNA in both the parental and RAD52KO cells (Fig 1D). Beginning with the crosslinking agent cisplatin, we examined the effect of two doses of cisplatin on clonogenic survival. At the higher dose, we found that both the RAD52KO and POLQe16m cell lines were hypersensitive, compared to the parental cell line (Fig 1E). Furthermore, the RAD52KOPOLQe16m cells were hypersensitive compared to both the parental cells and the single mutants, at both doses (Fig 1E). Notably, the fold-effect on clonogenic survival for the RAD52KOPOLQe16m cells was at least additive, compared to the effects of the single mutants (Fig 1E). Similarly, we found that siPOLQ treatment caused hypersensitivity to cisplatin at both doses, in both the parental and RAD52KO cells (Fig 1F). Finally, the RAD52KO cells treated with siPOLQ showed at least additive hypersensitivity to cisplatin, as compared to the effects of siPOLQ treatment in the parental cell line, and the RAD52KO cells vs. the parental cells (Fig 1F). Thus, disruption of RAD52 and POLQ appear to cause hypersensitivity to cisplatin, which is at least additive with combined disruption of these factors.
We also examined clonogenic survival in response to ionizing radiation (IR), and the PARP inhibitor Olaparib. Using two doses of IR, we found that the single mutant cell lines either showed no hypersensitivity, or showed a modest hypersensitivity (Fig 1E, < 2-fold). Similarly, siPOLQ treatment did not caused an obvious effect on IR response in either the parental or RAD52KO cells (Fig 1F). In contrast, the RAD52KOPOLQe16m cells showed significant hypersensitivity to both doses of IR (Fig 1E). These findings indicate that RAD52 and POLQ have modest effects on resistance to IR. Although, the results from the RAD52KOPOLQe16m cell line indicate that combined genetic disruption of these factors can cause IR hypersensitivity. Using two doses of Olaparib, the RAD52KO and RAD52KOPOLQe16m cells were both hypersensitive compared to the parental cell line at both doses (Fig 1E). The RAD52KO and RAD52KOPOLQe16m cells were not statistically different from each other (Fig 1E). The POLQe16m and siPOLQ-treated parental cells showed a modest hypersensitivity to Olaparib (≤ 2.1-fold), and siPOLQ-treatment in the RAD52KO cell line caused hypersensitivity to Olaparib at both doses (Fig 1F). Thus, RAD52 and POLQ appear important for resistance to Olaparib, although RAD52 appears to have a greater effect.
We then sought to examine the influence of RAD52 and POLQ on distinct DSB repair events. Both RAD52 and POLQ have been implicated in DSB repair that uses homologous repeat sequences that flank a DSB to bridge the break [23, 34]. These events often cause a deletion between the repeat, along with one copy of the repeat, such that we refer to all of these events as repeat-mediated repair (RMR). The parameters of RMR events that are mediated by RAD52 vs. POLQ have remained unclear. Thus, we sought to establish a reporter assay platform to examine two variable features of RMR events: repeat length and non-homologous tail removal. For this, we generated a set of reporter assays in which an expression cassette for green fluorescent protein (GFP) was disrupted by a non-homologous insert sequence (Fig 2A). We then added a homologous repeat of varying lengths (200–6 nt), by expanding the size of the 3' GFP sequence (S1A Fig). Each reporter was integrated in the U2OS cell lines, using the FRT/Flp system [35] (S1B and S1C Fig). In these reporter assays, the RMR events are induced by expression of Cas9 and various sgRNAs.
We tested these reporters in the parental U2OS cells with different combinations of DSBs. To begin with, we targeted a DSB at the 5' edge of the non-homologous insert, such that an RMR event that uses the flanking homology would restore the GFP expression cassette (Fig 2B). In the parental cells, we found that inducing this DSB in the reporters with repeat lengths of 200–72 nt caused similar frequencies of GFP+ cells (Fig 2B). However, with a repeat length of 50 nt, the frequency of GFP+ events was reduced approximately 2-fold compared to the longer repeats, and with repeat lengths of 23–6 nt, induction of GFP+ cells was nearly abolished (Fig 2B). We then considered that the inability to detect RMR events at the shorter repeats was due to the presence of the non-homologous insert. So, we examined these reporters using two DSBs to excise the insert: the first sgRNA targets the edge of the 5' GFP sequence as described above, and the second sgRNA targets the edge of the 3' GFP sequence, which is distinct for each reporter (5' & 3' edge; Fig 2C). With this approach, we were able to readily detect GFP+ events at each of the shorter repeat lengths (50–6 nt, Fig 2C). Notably, as with the 5' edge DSB alone, the 3' edge DSB alone was insufficient to significantly induce GFP+ cells for repeat lengths of 23–6 nt (Fig 2C). Thus, both DSBs are required to significantly induce these repair events. The restoration of the GFP coding sequence was confirmed for each of the reporters by sorting cells to enrich for GFP+ cells, followed by PCR amplification and sequencing analysis (S2A Fig).
We then analyzed the influence of RAD52 and POLQ on this series of RMR events. Since the RAD52KO, POLQe16m, and RAD52KOPOLQe16m cell lines were generated using the U2OS Flp-In T-REx cell line, we were able to integrate each reporter into these lines using the FRT/Flp system. At least two independent integrants of each reporter for each cell line were analyzed. A technical limitation of DSB reporter assay experiments is that different cell lines and experimental replicates can show variations in transfection efficiency, although this issue is partially mitigated by normalizing each experiment to transfection frequency using a parallel well with a GFP expression vector. To address this technical limitation via another method, we used transient complementation, which enables examination of the same cell line with parallel transfections of the complementation vector vs. empty vector (EV). The complementation vector was included in the transient transfection with the sgRNA/Cas9 plasmid(s). However, a drawback of this approach is that complementation vectors do not readily mimic endogenous levels of the respective protein. Indeed, for the POLQ complementation vector, while we confirmed expression using the Flag-immunotag (Fig 3A), we were unable to identify an antibody that is sensitive to detect endogenous POLQ. Thus, we were unable to compare endogenous POLQ levels vs. expression from the complementation vector. Furthermore, while we used a relatively low concentration of the RAD52 complementation vector, we found that these experimental conditions caused a marked increase in RAD52 protein levels (Fig 3B).
Accordingly, in addition to using complementation analysis, we also independently assessed the influence of RAD52 and POLQ on these RMR events using RNAi, by treating cells with siRNAs targeting these factors (siRAD52 and siPOLQ, respectively). As with complementation experiments, RNAi enables comparisons of the same cell line with parallel transfections (i.e., the targeting siRNA vs. siCTRL). As mentioned above, we confirmed that siPOLQ treatment causes depletion of the POLQ mRNA in these cell lines (Fig 1D). We also confirmed that siRAD52 causes a reduction in RAD52 protein (Fig 3B).
Beginning with POLQ, we found several repair events were reduced in POLQe16m cells compared to the parental line (S3A Fig). However, POLQ expression in the POLQe16m cells promoted only one RMR event: the RMR with the 6 nt repeat, which was induced using two DSBs to excise the non-homologous insert (5' & 3' edge; Fig 3A). Similarly, siPOLQ caused a significant decrease in the 6 nt repeat RMR event, but not any of the others (i.e., RMR events with ≥ 18 nt repeats) (Fig 3A).
For RAD52, beginning with the RMR events with repeat lengths of 200–50 nt and using the 5' edge DSB, the RAD52KO cell line exhibited lower frequencies vs. the parent cell line for each of these events (S3B Fig). Regarding complementation, we found that RAD52 expression in the RAD52KO cells significantly promoted these events for repeat lengths of 101, 72, and 50 nt, but not for repeats of 200 and 140 nt (Fig 3B). We found similar results with the RAD52KOPOLQe16m cell line, although in this case RAD52 expression promoted each of these repair events (i.e., 200–50 nt and using the 5' edge DSB, Fig 3C, S3C Fig). Similarly, siRAD52 treatment caused a significant reduction in each of these repair events (200–50 nt and using the 5' edge DSB, Fig 3B).
We then examined the influence of RAD52 on the events with shorter repeats (50–6 nt) by inducing two DSBs to excise the non-homologous insert (i.e., 5' & 3' edge). The RAD52KO and RAD52KOPOLQe16m cells showed reduced frequencies of several of these events, compared to the parental line (S3B and S3C Fig). However, we found that RAD52 expression in RAD52KO and RAD52KOPOLQe16m cells showed only a modest increase in events for the 50 nt repeat (1.2-fold), and did not promote events involving 23, 18, or 6 nt (Fig 3B and 3C). Similarly, again using the two DSBs to excise the insert, siRAD52 treatment caused a modest reduction in the event with the 50 nt repeat, but did not affect the frequencies of the events with the shorter repeat lengths (Fig 3B).
Altogether, considering effects of both complementation and RNAi, these findings indicate that RAD52 is important for RMR events using ≥ 50 nt repeats, whereas POLQ promotes RMR events using 6 nt repeats, but not ≥ 18 nt. Regarding the double-mutant cell line, as mentioned above, the findings were similar for the RAD52KO and RAD52KOPOLQe16m cell lines with this panel of reporters and complementation analysis (Fig 3C). We also found that POLQ complementation in the RAD52KOPOLQe16m cell line showed the same results as with the POLQe16m cell line (i.e., promoted only the RMR event using the 6 nt repeat; Fig 3C). Thus, the analysis with the RAD52KOPOLQe16m cell line indicates that combined disruption of RAD52 and POLQ does not appear to generate a synthetic defect in RMR events, but rather shows a combination of two independent defects found in the single mutant cell lines.
In the above analysis, for RMR events with a 50 nt repeat, we found that RAD52 is more important for such events when induced by one DSB at the 5' edge, compared to when the non-homologous sequence was excised with the 5' & 3' edge DSBs (Fig 3B). The distinction between these events is that the former requires the removal of the non-homologous sequence upstream of the 3' GFP segment. Accordingly, we sought to also examine RMR events requiring removal of non-homologous sequences from both sides of the DSB. To test this, we used an sgRNA to induce a DSB approximately in the middle of the non-homologous insert (mid-ins; 0.3 kb from both 5' GFP and 3' GFP). Using this mid-ins DSB, the repeat lengths are increased by 1 nt, compared to the above analysis (Fig 4A), since the 5' DSB cleaves upstream of this single nucleotide of homology between the repeats.
In the parental cell line, we found that the frequency of RMR events restoring GFP was highest for the 201 nt repeat, and decreased with the length of the repeat (Fig 4A). Indeed, such repair using the 51 nt repeat was largely undetectable. Therefore, inducing a DSB with non-homologous sequences on both sides of the repeats causes a greater requirement for a longer repeat to induce RMR events. We then analyzed the influence of RAD52 on RMR events using the mid-ins DSB, and found that RAD52 complementation promoted these events for each of the repeat lengths (i.e., 201, 141, 102, and 73 nt repeats, Fig 4B, S3D Fig). We found similar results for RAD52 complementation in the RAD52KOPOLQe16m cell line, and RNAi depletion of RAD52 (siRAD52 treatment), whereas POLQ complementation did not promote these events (Fig 4B). Notably, overexpression of RAD52 in the parental cells also promoted RMR events induced by the mid-ins DSB with the 201, 141, 102 nt repeats, but not any of the other RMR events (S4A Fig). This finding indicates that the level of RAD52 is a limiting factor for RMR events with repeats flanked by non-homologous sequences. In summary, these results with the mid-ins DSB, combined with the above finding (Fig 3B) that RAD52 has a greater effect on 50 nt RMR events that are induced by the 5' DSB vs. excision of the non-homologous sequence (i.e. the 5' & 3' edge DSBs), indicate that RAD52 is particularly important for RMR events that involve removal of a non-homologous sequence.
Since POLQ appears important only for the RMR event using the 6 nt repeat, we considered that POLQ might also be important for other repair events. In particular, we considered that POLQ might be important for DSB repair events that require nascent DNA synthesis. We based this hypothesis on previous studies showing that POLQ mediates annealing of oligonucleotides using short complementary ssDNA to template nascent DNA synthesis [19, 24]. To examine events that require nascent DNA synthesis, we modified our chromosomal RMR reporter system by deleting 7 nt from the 5' edge of the 3' GFP segment (Δ7 reporter; Fig 5A). Repair using an oligonucleotide with microhomology as a template that contains the missing 7 nt would restore GFP expression (i.e., oligonucleotide microhomology-templated repair). We used oligonucleotides that contained the missing 7 nt, which are flanked by equal lengths of homology to both the 5' and 3' GFP sequences, using several different lengths: 12, 14, 16, 18, or 20 nt (referred to as 12-7-12, 14-7-14, 16-7-16, 18-7-18, and 20-7-20, respectively, Fig 5A). The oligonucleotides also contain phosphorothioate linkages at the two terminal bases at both ends to promote stability [36]. These oligonucleotides were co-transfected with the sgRNA/Cas9 plasmids to induce DSBs at the edge of the 5' GFP and 3' GFP segments (i.e., the 5' & 3' edge DSBs, as described above). Using the parental U2OS cells, we found that each of the oligonucleotides induced GFP+ cells, which increased in frequency with the length of the flanking sequence homology (Fig 5A). To confirm the restoration of GFP in the Δ7 reporter with each of the oligonucleotides, the cells were sorted to enrich for GFP+ cells, and examined by PCR and sequencing (S2B Fig). We also confirmed that both the 5' and 3' DSBs are required to induce these oligonucleotide microhomology-templated events (S4B Fig).
We then analyzed the role of RAD52 and POLQ on the Δ7 reporter assay using each of the oligonucleotide templates. We performed both complementation and RNAi analysis, although since the 12-7-12 oligonucleotide events were near background levels, we found it difficult to examine effects of RNAi in potentially reducing these events (S4C Fig). In any case, both types of analysis were feasible for the rest of the oligonucleotides (14-7-14 and longer). From both complementation and RNAi analysis, we found that RAD52 was dispensable for such repair with each of the oligonucleotides (Fig 5B and 5C and S4B and S4D Fig). In contrast, we found that POLQ expression in the POLQe16m cells significantly promoted the induction of GFP+ cells using all of the oligonucleotides (Fig 5B and S4B and S4D Fig). POLQ expression had a similar effect on RAD52KOPOLQe16m cells, and the fold-effects were magnified (Fig 5B and S4D Fig). Importantly, and consistent with the complementation analysis, siPOLQ treatment caused a reduction in each of these events (Fig 5C, i.e., with the 14-7-14 oligonucleotide and longer).
To provide a contrast for these assays, we also examined end joining (EJ) events that do not require annealing of a homologous repeat or nascent DNA synthesis. Specifically, we used EJ7ins (S5A Fig), in which the non-homologous insert is flanked by the first two bases (GG) and the final base (C) of the GGC codon for Glycine 67 for GFP. Following DSBs to excise the non-homologous insert, EJ without indels between the distal DSBs would restore the GGC codon. Thus, restoration of GFP+ does not involve any nascent DNA synthesis nor annealing of microhomology. This assay is a variant of EJ7-GFP [30]; the only difference is the size of the non-homologous insert. We also performed experiments with an oligonucleotide that is homologous to the EJ junction that could possibly bridge the DSB ends during repair. Specifically, we used an oligonucleotide with 14 nt of homology to each side of the EJ junction, but with no bases in between (i.e., 14-0-14, S5A Fig). We included this experiment to provide a contrast to the Δ7 reporter assays, which use oligonucleotides to template nascent DNA synthesis. We found that including the 14-0-14 oligonucleotide did not promote the EJ event measured by EJ7ins, compared to a control oligonucleotide (luciferase/LUC), or to transfections without any oligonucleotide (S5A Fig).
We then examined the influence of POLQ and RAD52 on these EJ events. We found that siPOLQ and siRAD52 treatments did not cause a decrease in the frequency of such EJ events, with or without the 14-0-14 bridging oligonucleotide (Fig 6A). From analysis of the mutant cell lines, expression of POLQ from the complementation vector caused a modest increase in these EJ events, irrespective of whether an oligonucleotide was included (Fig 6A, S5B Fig). Notably, these effects of the POLQ complementation vector on EJ were less than for the oligonucleotide microhomology-templated repair events (Figs 5B and 6A, in the POLQe16m cells, EJ7ins promoted ≤1.36-fold, whereas the Δ7 reporter promoted between 1.6 to 2.6-fold, depending on the oligonucleotide). Furthermore, as mentioned above, the oligonucleotide microhomology-templated events (Fig 5C), but not the EJ events (Fig 6A), were reduced by siPOLQ treatment. Altogether these findings indicate that POLQ promotes oligonucleotide microhomology-templated repair to a greater degree than EJ without use of microhomology.
For another contrast to the above DSB repair events, we also examined HDR, using the DR-GFP reporter, which measures use of a homologous sequence as a template for gene conversion [37]. For these experiments, we used Cas9 and an sgRNA to induce the DSB in DR-GFP [38]. We found that neither RAD52 nor POLQ complementation vectors caused an increase in HDR in the respective mutant cell lines (Fig 6B, S5D Fig). Similarly, siRAD52 treatment did not cause a decrease in HDR, although siPOLQ caused a modest decrease in HDR (Fig 6B, 1.3-fold). These findings indicate that RAD52 and POLQ do not have a substantial role in HDR, as measured using the DR-GFP reporter.
To provide a contrast with RAD52 and POLQ, we also examined the influence of BRCA2 on several DSB repair events. BRCA2 is important for RAD51 recruitment to DNA damage and HDR [39]. We first sought to confirm that BRCA2 is important for HDR using the DR-GFP reporter [37], using siRNAs targeting BRCA2 (siBRCA2, depletion of BRCA2 validated by immunoblotting, Fig 6C). As expected, we found that siBRCA2 treatment caused a marked decrease in HDR (Fig 6D). We then examined the RMR reporter assays, and found that siBRCA2 treatment caused a decrease in nearly all of the RMR events (Fig 6D). Accordingly, BRCA2 appears to promote RMR events irrespective of the repeat length or DSB induced (Fig 6D), which is distinct from the results with RAD52 and POLQ. Finally, siBRCA2 treatment did not have a substantial effect on EJ (EJ7-ins reporter), nor the oligonucleotide microhomology-templated events (the Δ7 reporter, Fig 6D). In summary, BRCA2 promotes several RMR events, but to a much lesser degree than its requirement for HDR.
Disruption of BRCA1 and BRCA2 causes defects not only in HDR, but also the cellular response to replication stress [6]. Thus, we next examined whether disruption of RAD52 and POLQ may also affect replication stress responses, using DNA fiber analysis [40]. We first examined how the disruption of RAD52 and/or POLQ would affect the rate of replication fork progression in unstressed cells. Specifically, we pulse labeled cells with the thymidine analog CldU, followed by a pulse label with the thymidine analog IdU for equal amounts of time (40 min). Antibodies against each analog that are conjugated to different fluorophores allowed for the visualization of the fibers. We measured the lengths of the labels for individual fibers to calculate the IdU/CldU ratio, and thereby measure the rate of fork progression, which we refer to as replication fork velocity (Fig 7A). We found that the POLQe16m cells showed a modest but significant increase in replication fork velocity, whereas disruption of RAD52 had no effect (Fig 7A). In contrast, the RAD52KOPOLQe16m cells showed a significant reduction in replication fork velocity (Fig 7A). Similarly, siPOLQ treatment caused a reduction in replication fork velocity in the RAD52KO cells, but not parental cells (Fig 7A). As described above, depletion of POLQ mRNA via siPOLQ was confirmed in both parental and RAD52KO cells (Fig 1D). We also examined the fraction of stalled replication forks (i.e., CldU-labeled fibers only). We found that POLQe16m and RAD52KOPOLQe16m cells, but not RAD52KO cells nor siPOLQ treated cells, showed a modest decrease in the frequency of stalled replication forks (S6A Fig).
We next examined the influence of RAD52 and POLQ on the restart of replication forks after replication stress. In this analysis, cells were pulse labeled with CldU, and then treated with hydroxyurea (HU), which causes a depletion of dNTPs, thereby causing replication fork stalling [40]. Following release from HU, cells were pulse labeled with IdU, and the DNA fibers were analyzed for the IdU/CldU ratio to measure the rate of replication restart, which we refer to as replication restart velocity (Fig 7B). We also quantified the frequency of stalled replication forks (S6A Fig). We found that replication fork restart velocity was not distinct between the POLQe16m cell line and the parental cells line, but was higher in the RAD52KO vs. parental (Fig 7B). Strikingly, the RAD52KOPOLQe16m cell line showed a marked decrease in replication fork restart velocity, compared to the parental cell line (Fig 7B). Similarly, siPOLQ treatment in the RAD52KO cell line caused a marked decrease in replication fork restart velocity, whereas siPOLQ treatment only caused a modest decrease in the parental cell line (Fig 7B). Apart from fork velocity, we did not observe major effects on the percentage of stalled replication forks, apart from a modest increase with the RAD52KOPOLQe16m cell line (S6B Fig).
We also examined replication fork protection during stalling, which has been shown to require BRCA2, among other factors [5, 41]. In this analysis, cells are pulse labeled with CldU, followed by IdU, and then treated with HU for 5 hr [5, 41]. To begin with, we examined cells treated with siRNAs targeting BRCA2, and consistent with prior studies, found that depletion of BRCA2 (confirmed by immunoblotting) causes a reduction in the IdU/CldU ratio, reflecting fork degradation [5, 41] (Fig 7C). In contrast, BRCA2 depletion did not cause an obvious effect on the IdU/CldU ratio when the HU treatment was positioned between the two labels (S6B Fig). Regarding the influence of RAD52 and POLQ on fork protection during stalling, we found that the POLQe16m and RAD52KO cell lines were not distinct from the parental cell line (Fig 7C). The RAD52KOPOLQe16m cells showed a modest decrease in the IdU/CldU ratio in these experiments (P = 0.045, Fig 7C), similar to the findings without replication stress (see Fig 7A). Taken together, these finding indicate that the combined disruption of RAD52 and POLQ causes a significant decrease in the velocity of replication fork progression, particularly during restart of stalled replication forks, but does not have an obvious effect on protection of stalled replication forks from degradation.
As RAD52 and POLQ are each synthetic lethal targets for cells deficient in BRCA1 and BRCA2 [8–10], we have sought to test whether RAD52 and POLQ have distinct vs. redundant functions in chromosomal break repair, sensitivity to genotoxins, and/or response to replication stress. Beginning with genotoxin sensitivity, we found that disruption of RAD52 and POLQ each caused hypersensitivity to cisplatin, and combined disruption of these factors caused an at least additive hypersensitization. Accordingly, RAD52 and POLQ appear to have non-epistatic roles in cisplatin resistance. We also found that RAD52 and POLQ have different effects on DSB repair, using a series of novel assays for RMR and oligonucleotide microhomology-templated repair events. The DSB reporter analysis involved multiple approaches to examine RAD52 and POLQ, i.e, both complementation analysis in mutant cell lines, and RNAi. We suggest that identifying DSB repair phenotypes that are relatively consistent between these approaches, and that reveal patterns among multiple reporter contexts, has provided insight into the influence of RAD52 and POLQ on such DSB repair events.
Beginning with RAD52, we found that this factor is important for RMR events using ≥ 50 nt, and when repeat sequences also require removal of non-homologous sequence flanking at least one of the repeats. The influence of RAD52 on events with this range of repeat length is consistent with biochemical properties of RAD52. In particular, single-molecule studies have shown that multimeric rings of RAD52 interact with ssDNA by optimally binding ~30 nt around the protein ring [12, 16, 42–44]. Regarding removal of non-homologous sequences flanking a region of homology, other studies also support a role of RAD52 in such events. For example, our laboratory recently reported that an RMR event in mouse cells requiring removal of several kb of non-homologous sequence was particularly dependent on RAD52 [45], and another recent study showed that HDR events requiring removal of a non-homologous sequences were also promoted by RAD52 [46]. Thus, we suggest that RAD52 may have a specific role in synapsis of ≥ 50 nt of homology that is embedded within a non-homologous sequence, and thereby stabilize this intermediate to facilitate cleavage of the non-homologous sequence to complete repair.
For POLQ, we found that this factor was important for RMR events using 6 nt, but not ≥ 18 nt, as well as DSB repair events requiring nascent DNA synthesis from oligonucleotide templates with 12–20 nt of microhomology. These findings are consistent with studies of POLQ-dependent extension of oligonucleotide substrates that are annealed via a very short (e.g., 4 nt) sequence [17, 19]. This activity of POLQ is consistent with the structure of its C-terminal polymerase domain, which contains additional insertions loops that are not found in other A-family DNA polymerases [18]. Within these unique insertions loops, multiple residues facilitate specific interactions with the primer strand, which appear to enable extension of minimally annealed DNA substrates [17, 18, 47]. Notably, combined loss of POLQ and RAD52 did not reveal any synthetic defects in DSB repair events (e.g., repair events promoted by POLQ were the same in the POLQe16m cells as the POLQe16mRAD52KO cells), which altogether indicate that these factors have distinct roles in such repair.
We also found that RMR events involving 18–23 nt of homology were unaffected by RAD52 and POLQ. Notably, events with ≤ 23 nt of homology are nearly undetectable if the repeat is flanked by a non-homologous sequence. Accordingly, the mechanisms that mediate such RMR events with ≤ 23 nt of homology may be insufficient to facilitate cleavage of a non-homologous tail. Alternatively, ≤ 23 nt of homology may not be sufficient to compete with shorter lengths of homology that are closer to the DSB end. In any case, other factors besides RAD52 and POLQ appear to be sufficient to mediate RMR events involving 18–23 nt of homology. Indeed, beyond these particular repair events, we suggest that other factors are likely involved in RMR events of diverse repeat lengths, since each RMR event we examined remains readily detectable in cells deficient in RAD52 and/or POLQ.
The factors apart from RAD52 and POLQ that mediate RMR events remain unclear. Although, we found that BRCA2 mediates several RMR events at multiple repeat lengths (i.e., 201 nt– 6 nt), which is distinct from our findings with RAD52 and POLQ. However, the influence of BRCA2 on these RMR events was markedly lower than its influence on HDR. Furthermore, in other studies, BRCA2 has been shown to suppress RMR events, likely due to competition with HDR [26, 48]. However, BRCA2-mediated HDR may not be a substantial competitive pathway for the RMR events measured here. Namely, the DSBs in these assays are not readily repaired by HDR, which requires a repair template with homology on both sides of the DSB. Nevertheless, our findings support the notion that BRCA2 has a distinct role in DSB repair vs. POLQ and RAD52, since BRCA2 is required for HDR, whereas POLQ and RAD52 do not appear to have substantial roles in HDR.
Consistent with BRCA2 having distinct roles in genome maintenance vs. POLQ and RAD52, these factors differentially affect the response to replication stress. As in other studies [5, 41], we found that depletion of BRCA2 caused a defect in protecting stalled replication forks from degradation, but did not cause obvious effects on the restart of stalled forks. In contrast, disruption of POLQ and RAD52, either alone or in combination, caused no major effects on protection of stalled replication forks, using the same experimental conditions that reveal a role for BRCA2. We also found that disruption of RAD52 or POLQ individually did not obviously cause defects in the frequency of restart of stalled replication forks. These findings are consistent with other studies of RAD52, in which this factor appears dispensable for restart of stalled replication forks, but rather appears important for restart of collapsed forks (i.e., following long-term HU treatment) [49]. Although, a recent report found that combined treatment of a small molecule that targets RAD52, along with a CDC7 inhibitor, caused an increase in the frequency of stalled replication forks after HU treatment [50]. Furthermore, our findings with POLQ are distinct from a report that cells depleted of POLQ via RNAi show an increase in the frequency of collapsed forks following recovery from HU [9].
Nevertheless, we found that combined disruption of POLQ and RAD52 caused a marked decrease in replication fork restart velocity, as indicated by a substantial reduction in the length of the labeled DNA fiber after release from HU. The cause of this effect on fork restart velocity could be due to several mechanisms. For example, RAD52 could promote an annealing intermediate important for stabilizing the stalled fork, and/or re-establishing the replisome [49]. Indeed, a recent report found that RAD52 is important to suppress excessive ssDNA formation at stalled forks [50]. Similarly, POLQ could stabilize the stalled fork via its primer extension activity [51]. RAD52 or POLQ could also recruit other factors important for these processes [47, 52]. Alternatively, loss of one of these factors could cause accumulation of an intermediate that requires the other factor for resolution to enable rapid fork restart. Along these lines, disruptions of POLQ and/or RAD52 may affect other aspects of DNA replication that may not have been revealed in our analysis, such as suppressing fork discontinuities [50, 53], which could contribute to the reduced fork velocity that we observed. In summary, these findings indicate that RAD52 and POLQ have distinct roles in genome maintenance, including DSB repair and replication fork restart velocity. Since these factors are emerging therapeutic targets [8–11], these findings indicate that combined disruption of these factors may be an effective approach for genotoxin sensitization and/or synthetic lethality strategies.
All sgRNAs, primers, and oligonucleotide template sequences are found in S1 Table. The parental cell line in this study is the human osteosarcoma U2OS Flp-In T-REx cell line, which is stably transfected with pFRT/lacZeo [29, 30]. Cells were cultured as previously described [54], and using the Lonza MycoAlert PLUS Mycoplasma Detection Kit, cell lines tested negative for mycoplasma contamination. To generate plasmids for inducing DSBs, sgRNA sequences were cloned into the px330 vector (Addgene #42230) that expresses an sgRNA and Cas9 [55]. To generate the mutant cell lines, these sgRNA/Cas9 plasmids were co-transfected (400 ng of each sgRNA vector) with the dsRED expression plasmid (120 ng) and 3.6 μl of Lipofectamine 2000. After 3 days, the cells were sorted (using an Aria 3 or Aria SORP, Becton Dickinson) to enrich for dsRED-positive transfected cells followed by low-density plating. To generate the POLQe16m cell line, two sgRNAs were used to target exon 16 of POLQ, and clones were screened by PCR amplification and sequencing. For the RAD52KO cell line, two sgRNAs were used to target exon 3 and exon 9 of RAD52 [49]. The RAD52KOPOLQe16m cell line, was generated in the POLQe16m cell line using the RAD52 exon 3 sgRNA and an sgRNA that targets RAD52 exon 4. The RAD52KO and RAD52KOPOLQe16m cell lines were identified by screening individual clones using RAD52 immunoblot analysis.
The RMR200 reporter plasmid was generated by inserting two gBLOCK fragments (IDT) into the pcDNA5-FRT-EJ7-GFP vector [30]: 1) a non-homologous sequence derived from the puromycin-resistance gene [54] to generate the EJ7-ins reporter, and 2) the 3' GFP fragment, which contains 200 nt of homology to the 5' GFP sequence. This RMR200 reporter plasmid was used to generate the variants with the different 3' repeat sequences. These RMR reporter plasmids (100 ng) were integrated into the U2OS Flp-In T-REx cells by co-transfection with the PGK-Flp vector (400 ng) [35], using Lipofectamine 2000 (Thermofisher) as described below for the DSB reporter assays. Integrated clones were selected using hygromycin (0.2 μg/μl), and subsequently screened with PCR analysis to confirm integration (S1B, S1C Fig). To integrate the DR-GFP reporter into the parental U2OS Flp-In T-Rex cell lines, and the various mutant cell lines, 10 μg of XhoI linearized Pim-DRGFP plasmid [54] was electroporated into each cell line (0.8 ml volume), followed by selection of stably transfected cells in 0.8 μg/ml puromycin, which were pooled together for analysis. The RNAi experiments to examine HDR used the previously described U2OS DR-GFP reporter cell line [54].
Cell cycle analysis was performed as previously described [45]. Briefly, the cells were pulse labeled with BrdU (BD Pharmingen, 51-2420KC) for 30 min at 37°C. The cells were then fixed with 70% ethanol, and stained with FITC-conjugated anti-BrdU (BD Pharmingen, 51-33284X), followed by and propidium iodide (Sigma, P4170) supplemented with RNase (Sigma, R4642) for 30 min at 37°C. Each sample was analyzed by flow cytometry using a CyAn-ADP (Dako).
Cells with integrated reporter cassettes were seeded at 0.5 x 105 cells per well (24 well plate). The following day, the cells were transfected with 200 ng of each sgRNA/Cas9 vector and 1.8 μl of Lipofectamine 2000 with 0.5 ml of antibiotic-free media. To normalize the frequency of repair events between experiments, parallel transfections with GFP expression vector (200 ng, pCAGGS-NZE-GFP [54]) were included. In the RAD52 complementation experiments, the reactions were performed as describe above with the addition of 25 ng of empty vector (pCMV6-XL5) or RAD52 expression vector (Origene RC238113). For the POLQ complementation, 100 ng of empty vector (pCAGGS-BSKX) [56] or POLQ expression vector [57] was added to the reactions. Similar transfections with equivalent concentrations of expression vectors were used to generate samples for immunoblotting analysis. For RAD52 and POLQ complementation in the double mutant cell line, additional empty vector (pCAGGS-BSKX) was included to ensure an equivalent amount of total plasmid in each transfection. For the Δ7 reporter, transfections were scaled 2-fold onto a 12 well dish, and transfections were performed as described above with the addition of 10 nM (final concentration) of the indicated oligonucleotide to the reaction. Each oligonucleotide contained phosphorothioate linkages on the first two and last two terminal bases (IDT). In the experiments with siRNA, 5 pmol of either non-targeting siRNA (siCTRL; Dharmacon, D-001810-01-20) or a pool of four siRNAs targeting RAD52, POLQ, or BRCA2 (Dharmacon siGENOME siRNAs, sequences from manufacturer in S1 Table) was included in the respective Lipofectamine 2000 transfections. In addition, for POLQ siRNA experiments, the day before the above transfections with Lipofectamine 2000, cells were first treated with 5 pmol of either siCTRL or the four siRNAs targeting POLQ, using Lipofectamine RNAiMAX (Thermofisher). For immunoblotting analysis to confirm BRCA2 and RAD52 depletion, an equivalent concentration of cells and siRNA as for the reporter assays was used for a transfection with Lipofectamine RNAiMAX. For each reporter assay, three days after transfection, the percentage of GFP+ cells were determined by flow cytometry using a CyAn-ADP (Dako), as previously described [54]. The repair value for each sgRNA(s)/CAS9 transfection was first normalized to transfection efficiency using the parallel transfection with a GFP expression vector. For comparisons vs. EV or siCTRL, each repair value normalized to transfection efficiency was divided by the mean repair value for the parallel control transfections (i.e., siCTRL and/or EV). To confirm the sequence of GFP+ products for each reporter, transfected parental cells were sorted (Aria III or Aria SORP, Becton Dickinson) to enrich for cells expressing GFP, which were analyzed by PCR-amplification and sequencing (S2 Fig).
Clonogenic survival was assessed by plating 103 cells on 6 well plates in media containing cisplatin (1.0 or 3.0 μM, Pfizer) or Olaparib (0.75 or 1.5 μM, Selleckchem), or were untreated (equivalent volume of DMSO added as a control). For ionizing radiation, each cell line was exposed to 1.5 or 3 Gy (Gammacell 3000), or left untreated, prior to plating. Cells were cultured for 9 days, and stained with crystal violet (Sigma). Colonies of approximately 50 or more cells were quantified under a 10x objective, and fraction survival was calculated relative to the number of colonies on the untreated control wells that were plated in parallel. For experiments with siRNA depletion, 105 cells were plated on a 12 well plate with either control siRNA (siCTRL) or a pool of four POLQ siRNAs (40 pmol; Dharmacon, sequences in S1 Table), using Lipofectamine RNAiMAX. Two days after transfection, cells were treated with genotoxins to test clonogenic survival, as described above.
To test for depletion of POLQ mRNA, cells were transfected on a 6 well dish with 20 pmol of siCTRL or pool of four POLQ siRNAs (see above) using Lipofectamine RNAiMAX (2 ml total volume). On the following day, cells were transfected with the respective siRNA (20 pmol) and two plasmids (400 ng of pgk-PURO and 1200 ng EV) [54], using Lipofectamine 2000 (2 ml total volume), as for the reporter assays. The day after transfection, cells were treated with puromycin (2 μg/ml concentration) for one day to enrich for transfected cells, and then RNA was isolated using the RNeasy Plus Minikit (Qiagen 74134). The RNA was treated with M-MLV Reverse Transcriptase (Promega M170A) to generate cDNA, which was amplified in an Applied Biosystems 7900HT Fast Real Time PCR system using SYBR-green, with the primer sequences shown in S1 Table.
Immunoblotting analysis was performed by lysing the cells using NETN buffer (20 mM Tris pH 8, 100 mM NaCl, 1 mM EDTA, 0.5% IGEPAL, 1.25 mM DTT, and protease inhibitors, Roche) followed by several freeze-thaw cycles. The blots were probed with antibodies against: RAD52 1:500 (Santa Cruz Biotechnology, sc365341), FLAG 1:1000 (Sigma, A8592), BRCA2 1:1000 (Millipore, OP95-10006), or ACTIN 1:3000 (Sigma, A2066); and with the HRP-conjugated secondary antibodies rabbit anti-mouse 1:3000 (Abcam, ab205719) or goat anti-rabbit 1:3000 (Abcam, ab205718). ECL Western Blotting Substrate (Thermo Fisher Scientific, 32106) was used to detect HRP signal on film.
For DNA fiber analysis, cells were plated at 105 cells/well on a 6 well plate. The following day, the cells were pulse labeled with CldU (50 μM, Sigma C6891) for 40 min followed by IdU (250 μM, Sigma I7125) for 40 min. When testing replication stress recovery, the cells were pulse labeled with CldU for 30 min, hydroxyurea (2 mM) for 2 hr, then IdU for 30 min. In the fork protection assay the cells were pulse labeled with CldU for 30 min, IdU for 30 min, then hydroxyurea (4 mM) for 5 hr. In the experiments with siRNA, 105 cells were plated on a 12 well plate with either control siRNA (siCTRL), a pool of four BRCA2 or POLQ siRNAs (40 pmol; Dharmacon, sequences in S1 Table), using Lipofectamine RNAiMAX, and the following day the cells were seeded on a 6 well plate. The next day (two days after transfection) cells were treated with the nucleotides and HU as above. DNA was isolated from cells using the FiberPrep DNA extraction kit (Genomic Vision, EXT-001). These DNA preparations were combed onto vinylsilane coated coverslips (Genomic Vision, COV-002-RUO) using the FiberComb Molecular Combing System (Genomic Vision, MCS-001). After combing, the coverslips were dehydrated, and then denatured using 0.5 M NaOH and 1 M NaCl. The coverslips were blocked with 5% BSA in PBS, followed by treatment with a rat antibody to detect the CldU signal and a mouse antibody to detect the IdU signal (1:50; Abcam ab6326 and BD Biosciences 347580, respectively), and then with goat anti-rat Alexa Fluor 488 and goat anti-mouse Alexa Fluor 555 (colored green and violet, respectively, by the image capture software to clearly distinguish the signals) (1:50; Thermo Fisher Scientific, A110060 and A28180, respectively). The coverslips were mounted using ProLong Gold Antifade (Thermo Fisher Scientific), and the slides were imaged using a Zeiss Observer II with a 40x oil immersion objective, and fiber lengths were quantified using Image J [58].
|
10.1371/journal.pgen.1005747 | Arabidopsis COP1 SUPPRESSOR 2 Represses COP1 E3 Ubiquitin Ligase Activity through Their Coiled-Coil Domains Association | CONSTITUTIVE PHOTOMORPHOGENIC 1 (COP1) functions as an E3 ubiquitin ligase and mediates a variety of developmental processes in Arabidopsis by targeting a number of key regulators for ubiquitination and degradation. Here, we identify a novel COP1 interacting protein, COP1 SUPPRESSOR 2 (CSU2). Loss of function mutations in CSU2 suppress the constitutive photomorphogenic phenotype of cop1-6 in darkness. CSU2 directly interacts with COP1 via their coiled-coil domains and is recruited by COP1 into nuclear speckles in living plant cells. Furthermore, CSU2 inhibits COP1 E3 ubiquitin ligase activity in vitro, and represses COP1 mediated turnover of HY5 in cell-free extracts. We propose that in csu2 cop1-6 mutants, the lack of CSU2’s repression of COP1 allows the low level of COP1 to exhibit higher activity that is sufficient to prevent accumulation of HY5 in the dark, thus restoring the etiolated phenotype. In addition, CSU2 is required for primary root development under normal light growth condition.
| CONSTITUTIVE PHOTOMORPHOGENIC 1 (COP1) is a key regulator of light mediated developmental processes and it works as an E3 ubiquitin ligase controlling the abundance of multiple transcription factors. In the work presented here, we identified a novel repressor of COP1, the COP1 SUPPRESSOR 2 (CSU2), via a forward genetic screen. Mutations in CSU2 completely suppress cop1-6 constitutive photomorphogenic phenotype in darkness. CSU2 interacts and co-localizes with COP1 in nuclear speckles via the coiled-coil domain association. CSU2 negatively regulates COP1 E3 ubiquitin ligase activity, and repress COP1 mediated HY5 degradation in cell-free extracts.
| Sunlight provides not only the major energy source, but also a main environmental signal that regulates multiple developmental processes in plants, such as seed germination, photomorphogenesis, flowering, phototropism and root growth [1]. In Arabidopsis thaliana, phytochromes (phyA-phyE) sense red and far-red light (600–750 nm) [2, 3]; cryptochromes (CRY1 and CRY2) and phototropins (PHOT1 and PHOT2) perceive blue and UV-A light (315–500 nm) [4, 5]; and UVR8 acts as the UV-B (~280 nm) photoreceptor [6]. In response to light, photoreceptors can directly act on numerous gene promoters throughout the genome to regulate the expression of their target genes in order for plants to rapidly adapt to their changing light environment [7–9]. In the absence of light, plants develop long hypocotyls, apical hook, unopened cotyledons and etioplastids, a unique developmental program known as skotomorphogenesis or etiolation. In the light, plants undergo photomorphogenesis, which is characterized by short hypocotyls, expanded cotyledons, and developed chloroplasts [1]. The skotomorphogenesis program is vital for terrestrial plants when their lives often start in the darkness of soil. The program prepares the plants for exposure to sunlight with vigor (a process known as greening), while inability to etiolate in darkness would be lethally damaged when exposed to light irradiation.
The CONSTITUTIVELY PHOTOMORPHOGENIC 1 (COP1) gene is essential for etiolation by acting as a repressor of photomorphogenesis, and its loss of function mutant display a constitutive photomorphogenic phenotype in darkness [10]. COP1 protein contains a RING finger, a coiled-coil domain, and WD-40 repeat domain, and it functions as an E3 ubiquitin ligase that targets a subset of photomorphogenic promoting factors for ubiquitination and degradation. In plant cells, COP1 exists as homodimers, which further stably associates with two SPA proteins, forming a tetrameric protein complex [11, 12]. Both COP1 dimerization and the interaction with SPA proteins are mediated through the coiled-coil domain of respective proteins. Association with SPA proteins enhances the activity of COP1 to targets substrate ubiquitination [12–14]. The substrates of COP1 in seedlings include LONG HYPOCOTYL (HY5), HY5 HOMOLOG (HYH), LONG HYPOCOTYL IN FAR-RED 1 (HFR1), LONG AFTER FAR-RED LIGHT 1 (LAF1), SALT TOLERANCE HOMOLOG 3 (STH3/BBX22) and PHYTOCHROME INTERACTING FACTOR 3-LIKE1 (PIL1) [14–20]. Besides seedling photomorphogenesis, COP1 also mediates the degradation of CONSTANS (CO), GIGANTEA (GI), EARLY FLOWERING 3 (ELF3), HYPERSENSITIVE RESPONSE TO TCV (HRT), SCAR1, GATA TRANSCRIPTION FACTOR 2 (GATA2) and MYC2, and plays critical roles in various developmental processes including flowering time, circadian clock, viral defense, root development, hormone signaling and controlling miRNA biogenesis [21–27]. COP1 is evolutionarily conserved from plants to animals. Mammalian COP1 has been reported to act as a tumor suppressor that targets oncoproteins c-Jun and ETS via its E3 ubiquitin ligase activity [28–31].
As a key regulator, COP1 protein level, activity, and localization are tightly controlled to ensure appropriate protein accumulation of its targets in response to developmental and environmental cues. In the dark, COP1 is enriched in the nucleus where it targets substrates for ubiquitination. Light triggers photoreceptors, including phyA, phyB, CRY1 and CRY2, to associates with SPA proteins or COP1, resulting in repression of the COP1-SPA E3 ubiquitin ligase activity [32–36]. This event is then followed by repartitioning of COP1 from the nucleus to the cytoplasm [37–40]. In addition, recent studies reveal that CSU1, SPAs and PIFs contribute to the modulation of COP1 protein level and activity as well [12,14, 41, 42].
In search of novel factors that modulate COP1 function or mediate its output, we have conducted a genetic screen for suppressors of cop1-6, a hypomorphic allele of cop1 mutants [43]. This screen has previously led to successful identification of CSU1, an E3 ubiquitin ligase that targets COP1 [41]. Here we report another novel COP1 suppressor, designated as CSU2. Mutations in CSU2 nearly completely suppress the constitutive photomorphogenic phenotype of cop1-6 in darkness. CSU2 physically interacts and co-localizes with COP1 in nuclear speckles via a coiled-coil domain association. CSU2 is able to repress the COP1 E3 ubiquitin ligase activity. In addition, CSU2 has an important role in root development. Collectively, our genetic and biochemical data demonstrate that Arabidopsis CSU2 functions as a negative regulator of COP1, which serves to optimize the development of plants.
A forward genetic screen was performed to explore cop1 suppressors as described previously [41]. Six independent recessive alleles, located at a novel extragenic locus (At1g02330) named cop1 suppressor 2 (csu2), were recovered from this screen (Fig 1). Each of the mutant alleles (csu2-1 to csu2-6) nearly completely suppressed cop1-6 constitutive photomorphogenic phenotype in the dark (Fig 2). Since the mutation in cop1-6 causes a splicing defect that leads to low expression of the COP1 gene product [41, 43], we first tested whether mutations in CSU2 affected cop1-6 splicing profiles by a RNA pattern analysis. csu2 cop1-6 produced five cryptically spliced profiles at intron 4 of COP1, similar to cop1-6 (S1 Fig), suggesting that csu2 suppressed cop1-6 not by correcting its splicing defect. Thus, csu2 was further characterized.
Via a combined chromosomal mapping and re-sequencing approach (see materials and methods for detail), we found that the csu2-4 mutation changes the splicing junction “AG” at the 3' end of intron-2 to “AA”, thus disrupting the splicing principles of CSU2. Five additional mutant alleles from the same genetic complementation group were analyzed by PCR amplification followed by sequencing, which led to identification of distinct point mutation in each of the csu2 mutant allele in At1g02330 (Fig 1B). Thus, At1g02330 defines the CSU2 gene.
CSU2 is a single-copy gene encoding a predicted 279 amino acid protein in Arabidopsis (Fig 1C). Only one putative domain, a coiled-coil domain, was identified in CSU2. CSU2 is evolutionarily conserved. The amino acid sequence identity of Arabidopsis CSU2 to its orthologs from Homo sapiens, Mus musculus, Danio rerio, Drosophila melanogaster, and Oryza saliva is 34%, 34%, 35%, 40% and 61% respectively, with the coiled-coil domain being the most conserved region (S2 Fig).
cop1-6 mutant is unable to etiolate in darkness [43], and is defective in greening upon transfer to white light [44]. Mutations in CSU2 almost completely restored cop1-6 constitutive photomorphogenic phenotype to WT phenotype in the dark (Fig 2). Hypocotyl length of all six different csu2 cop1-6 mutant lines was essentially indistinguishable from that of WT seedlings (Fig 2A and 2B). Although the cotyledons of csu2 cop1-6 were slightly open, the cotyledon apertures of all six independent csu2 cop1-6 mutant lines were significantly smaller than that of cop1-6 (Fig 2C and 2D). Moreover, although most dark-grown cop1-6 seedlings were unable to green when transferred from dark to white light, the greening rate of etiolated csu2 cop1-6 mutant seedlings was restored to a level comparable to that of WT (Fig 2E).
To verify that the suppression of the cop1-6 phenotype in csu2 cop1-6 etiolated seedling was indeed caused by the mutation in CSU2 gene only, we introduced CSU2-GFP and YFP-CSU2 into the csu2-2 cop1-6 double mutant. Consistently, CSU2-GFP csu2-2 cop1-6 and YFP-CSU2 csu2-2 cop1-6 transgenic seedlings displayed constitutive photomorphogenic phenotype similar to that of cop1-6 single mutant in the dark, indicating that a functional CSU2 could complement the phenotype conferred by csu2-2 in cop1-6 background in darkness (Fig 2F).
Not only did csu2 rescue the dark phenotype of cop1-6, csu2 also partially suppressed the short hypocotyl phenotype of cop1-6 seedlings grown under various light conditions tested (white, red, far-red and blue) (S3 Fig). The dwarf phenotype of cop1-6 adult plants under the long-day condition (16 h light / 8 h dark) for 30 days was also partially suppressed by csu2 (S4 Fig). All together, these genetic data suggest that csu2 almost completely suppress cop1-6 in the dark and partially in the light.
To examine whether mutations in CSU2 have defect in light responses by themselves, single mutants of all six alleles (csu2-1 to csu2-6) were isolated from the F2 generation of csu2 cop1-6 crossed with Col and grown under various light conditions (dark, white, blue, red and far-red) for five days. At low fluence rate of white light (15.7 μmol/m2/s), the hypocotyl length of csu2 mutant seedlings was indistinguishable from that of WT (S5 Fig). At the higher fluence of white light (33.3 and most evidently 112.5 μmol/m2/s), all six independent csu2 single mutants displayed statistically significantly longer hypocotyls than did WT seedlings (Fig 3 and S5 Fig). However, csu2 mutant seedlings did not differ significantly from WT seedlings under all monochromatic light (blue, red and far-red) conditions tested (S6 Fig). The fact that csu2 mutant seedlings were specifically hyposensitive to higher fluence rate of white light suggests that CSU2 acts as a positive regulator in the high fluence white light induced inhibition of hypocotyl elongation.
HY5 transcription factor is a major downstream effector of COP1, whose mutation can also suppress cop1-6 [15, 44, 45]. The hypocotyl length of hy5-215 csu2 double mutant seedlings was similar to that of hy5-215 single mutants in all light conditions tested including high fluence of white light, in which csu2 exhibited longer hypocotyls than WT (Fig 3B–3K). This result indicates that hy5-215 is epistatic to csu2 in the control of hypocotyl growth. Although either csu2 or hy5 alone only partially suppressed cop1-6 in the light, both mutations together (hy5 csu2 cop1-6) restored cop1-6’s hypocotyl length to that of WT seedlings under all light conditions tested (white, blue, red and far-red) (Fig 3B–3K). It appeared that CSU2 and HY5 act additively in the suppression of cop1 hypocotyl phenotype in the light. We suggest from these genetic data that CSU2 and HY5 work independently and additively, with HY5 acting downstream of CSU2, to counter COP1’s action in the control of hypocotyl elongation.
To understand the mechanism of CSU2, we examined a possible protein-protein interaction between CSU2 and COP1 by a yeast-two-hybrid assay. As shown in Fig 4, CSU2-COP1 interaction was evident as indicated by increased β-galactosidase activity compared to BD-CSU2 and AD-COP1 alone. COP1 possesses three protein-protein interaction domains, Ring-finger, coiled-coil and WD40 domains, while CSU2 contains only one predictable coiled-coil domain. To identify which COP1 domain is responsible for the interaction with CSU2, a deletion analysis of the COP1 fragment was carried out. Interestingly, COP1 N282, COP1 Δring and COP1 coil containing the COP1 coiled-coil domain, showed even stronger interaction with CSU2 than full-length COP1 (Fig 4). In contrast, COP1 Ring and COP1 WD40, which lack COP1 coiled-coil domain, were unable to interact with CSU2. Thus, the coiled-coil domain of COP1 is necessary and sufficient for interaction with CSU2. Next, we examined whether the coiled-coil domain of CSU2 was sufficient for the CSU2-COP1 interaction. Similar to the full-length CSU2, the CSU2 coil domain was capable of interacting with COP1, COP1 N282, COP1 Δring and COP1 coil, but not COP1 Ring and COP1WD40. In addition, CSU2 Δcoil, which lacks the coiled-coil domain, was unable to interact with full-length COP1 or any of the COP1 deletion constructs (Fig 4). Taken together, those data indicates that CSU2 interacts with COP1 through their respective coiled-coil domains.
We next performed the Bimolecular Fluorescence Complementation Assays (BiFC). Constructs of CSU2 fused with N-terminal of YFP (YN-CSU2) and COP1 fused with C-terminal of YFP (YC-COP1) were generated. When YN-CSU2 and YC-COP1 were co-transformed into onion (Allium cepa) epidermal cells, strong YFP fluorescence signals were observed in the nucleus, indicating that CSU2 can interact with COP1 (Fig 5A). Furthermore, we examined whether Fluorescence Resonance Energy Transfer (FRET) could occur between the two fusion proteins CFP-CSU2 and YFP-COP1 using the acceptor photobleaching technique. Here, we co-expressed CFP-CSU2 with YFP-COP1 in onion epidermal cells and excited them with 405- and 514-nm wave lengths light sources. Both CFP and YFP fluorescence were detected before bleaching. CFP-CSU2 produced uniform fluorescence throughout the nucleus, while YFP-COP1 formed nuclear speckles (S7A and S7B Fig). Since FRET occurs only at nanometer scale distances [46], only YFP-COP1 speckles areas were chosen for bleaching by 514-nm laser. After bleach, emission of YFP-COP1 was reduced dramatically, whereas emission from CFP-CSU2 in the region of interest increased (S7A and S7B Fig), indicating that FRET had occurred between the two proteins prior to the bleach. As a control, we did not detect FRET between YFP and CFP-CSU2 (S7C and S7D Fig). Together, these data support a conclusion that the CSU2 interacts with COP1 in living plant cells.
COP1 forms nuclear speckles in darkness and is able to recruit several interacting proteins to those loci [14, 16, 45, 47]. Our FRET assay data indicated that COP1 and CSU2 might co-localize in the nuclear speckles (S7A Fig). To further substantiate this finding, we performed transient co-localization assays using GFP tagged CSU2 fusion protein in onion epidermal cells (Fig 5B and 5C). Unlike COP1, CSU2 localized uniformly throughout the nucleus (Fig 5C). However when we co-expressed COP1 (35S:COP1) together with CSU2-GFP, we detected consistent nuclear speckles (Fig 5C). Since CSU2-GFP by itself only produces a uniform fluorescence, the observation of nuclear speckles when co-expressed with untagged COP1 suggests that CSU2 is recruited into nuclear speckles by COP1. Moreover, untagged COP1 (35S:COP1) could confer nuclear speckle formation to a co-expressing CSU2 coil-GFP but not CSU2 Δcoil-GFP (Fig 5C). These observations provide further evidence that interaction of COP1, via the coiled-coil domain of CSU2, is required and sufficient for recruitment of CSU2 into the nuclear speckles in living plant cells.
To determine whether CSU2 is a nuclear protein in planta, we examined its localization pattern in 35S:CSU2-GFP csu2-2 transgenic Arabidopsis seedlings where CSU2-GFP has been shown to be functional (Fig 2F). As shown in S8 Fig, CSU2-GFP was found within the nucleus both in darkness and light, confirming that CSU2 is a nuclear protein in planta.
COP1 targets a group of interacting proteins for ubiquitination and degradation. Therefore, we investigated whether COP1 regulates CSU2 abundance. YFP fluorescence signal intensity was comparable in the YFP-CSU2 csu2-2 and YFP-CSU2 csu2-2 cop1-6 transgenic seedlings (S9 Fig). In addition, similar protein levels of YFP-CSU2 were detected in these two transgenic lines grown in various light conditions tested (dark, white, blue red and far-red) (S10 Fig). These findings suggest that COP1 does not regulate CSU2 abundance.
The coiled-coil domain of COP1 is necessary for its dimerization [13] and for interacting with SPA proteins [11,12,48]. These interactions enhance COP1’s E3 ubiquitin ligase activity [12, 14]. Given that CSU2-COP1 association is through COP1 coiled-coil domain, we wanted to test whether CSU2 can affect COP1 activity. Consistent with previously described in vitro ubiquitination assay [14, 17], we detected a robust COP1 dependent ubiquitination activity, and this activity was drastically inhibited when CSU2 was present in the reaction (Fig 6A). Remarkably, COP1’s ubiquitination activity was not affect by CSU2 Δcoil, which lacks coiled-coil domain (Fig 6A). Therefore CSU2 can inhibit COP1 E3 activity in vitro, and the inhibition is dependent on CSU2’s COP1-binding domain.
HY5 is a major ubiquitination substrate of COP1 in seedlings, and its level of accumulation correlates with seedling photomorphogenesis [15, 45]. To examine the effect of CSU2 on COP1’s activity toward a specific substrate, we performed a cell-free HY5 degradation assay in cell lysates, in which degradation of HY5 was dependent on the presence of COP1 (Fig 6B and 6C). Notably, with decreasing amounts of CSU2 in the mixture, the protein level of HY5 also decreased (Fig 6B). In contrast to full length CSU2, CSU2 Δcoil had no effect on COP1 mediated degradation of HY5 (Fig 6C). As a validation of the assay, degradation of HY5 protein could be blocked by proteasome inhibitor MG132 treatment. The GFP protein, as an internal control, remained relatively stable under all the tested conditions (Fig 6B and 6C). Together, these data show that CSU2 represses the COP1 ubiquitination activity in vitro, and repress COP1-dependent degradation of HY5 in a cell-free degradation assay. In both cases, the coiled-coil domain of CSU2 is required for the repression of COP1 activity.
Prompt by CSU2’s activity in repressing COP1’s E3 ubiquitin activity in vitro, and in inhibiting HY5 degradation in the cell-free assay, we determined the steady state levels of COP1 and HY5 proteins in the seedlings of csu2 cop1-6 compared to cop1-6, and wild type (Fig 6D). The levels of COP1 in csu2 cop1-6 appeared slightly higher than that of cop1-6, but still substantially lower than WT in both dark- and light-grown seedlings (Fig 6D). The reason of the slight increase of COP1 is discussed later. The important point is that, even with clearly reduced amount of COP1, the dark-grown csu2 cop1-6 seedlings nevertheless managed to keep HY5 protein level as low as in WT, which was drastically decreased compared to cop1-6 (Fig 6D). Presumably, despite of reduced level of COP1 in csu2 cop1-6, but due to lack of CSU2-mediated inhibition, the total activity of COP1 seems sufficient to prevent HY5 accumulation in the dark. The slight increase of COP1 level in csu2 cop1-6 might also have contributed to the suppression of HY5 in the dark.
We next asked whether csu2 mutant seedlings display altered protein accumulation of additional components of light signaling. Under both dark and light conditions, phyA, phyB, COP1, HY5 and SPA1-4 (dark only) accumulated at comparable levels in WT and csu2 mutant seedlings (S11 Fig). Thus we have not detected an effect of CSU2 on protein abundance of these light-signaling components under normal growth conditions.
Light-grown seedlings display longer primary roots than etiolate seedlings, and cop1 mutant seedlings display an opposite root growth pattern [49]. To investigate the role of CSU2 in the root development, the six different csu2 mutant lines were germinated on vertical plates and grown for five days under dark or constant white light conditions. In the dark, cop1-6 displayed longer roots than did WT, while csu2 displayed the same root length to that of WT. csu2 cop1-6 double mutants exhibited roots similar to those of csu2 or WT seedlings, indicating that the long root phenotype of cop1-6 was completely suppressed by csu2 (Fig 7A and 7B). In the light however, all six different csu2 single mutants displayed dramatically shorter roots than did WT or cop1-6 (Fig 7C and 7D), and csu2 cop1-6 showed similar root length as csu2 single mutants. To further confirm that the short primary root phenotype is caused by disruption of CSU2, we investigated the primary root phenotypes of 35S:myc-CSU2 csu2-2 as well as 35S:CSU2-GFP csu2-2 and 35S:YFP-CSU2 csu2-2 transgenic lines (S9 Fig). In all cases, expression of CSU2 transgene rescued the shortened primary root phenotype of csu2-2 (S9 Fig), indicating the short primary root phenotype is resulted from lack of a functional CSU2. Taken together, these findings show that csu2 completely suppresses cop1 long primary root phenotype in the dark, that CSU2 is required for light stimulated primary root development, and that csu2 is epistatic to cop1 with respect to the primary root phenotype in the light.
To further investigate the genetic relationship among csu2, hy5 and cop1 with respect to root phenotypes, we studied the hy5 csu2, hy5 cop1 and hy5 csu2 cop1 double and triple mutants. In the dark, all the double and triple mutants exhibited root phenotypes similar to those of WT (Fig 8A and 8B). Under white light condition, the root length of hy5 csu2, or hy5 csu2 cop1 double and triple mutant seedlings resembled csu2 short roots phenotype (Fig 8C and 8D), suggesting a different genetic relationship of those three loci in mediating light regulation of root development and in hypocotyl growth. With regard to primary root growth, the requirement for functional CSU2 overrides the regulatory functions of COP1 and HY5.
COP1 is a central player of light regulated developmental processes. The mechanism of COP1 in the regulation of these processes is by working as an E3 ubiquitin ligase that targets an array of important gene expression regulators for proteolysis in a manner that is dependent on developmental stages and/or environmental cues [23, 50, 51]. Using seedling photomorphogenesis as a model, we have isolated six different alleles of csu2 mutants, each of which can completely suppress cop1-6 phenotype and restore etiolation when grown in the dark. In this system, the extent of photomorphogenic development of seedlings correlates quantitatively with HY5 protein abundance in planta, and HY5 protein levels normally correlates inversely with the nuclear abundance of COP1 [15]. Here we report that CSU2 interacts and co-localizes with COP1 in the plant cells, and it negatively regulates COP1 E3 ubiquitin ligase activity, which directly affects HY5 stability. Thus, CSU2 functions as a repressor of COP1 to regulate aspects of plant development.
COP1 is regulated in a number of different ways. Not only is COP1 nucleocytoplasmic partitioning regulated by light, low temperature, heat shock and ethylene [37, 52–54], its protein abundance is regulated by CSU1, an E3 ubiquitin ligase identified by the same screen as CSU2 [41] (Fig 9). COP1 activity is rigorously regulated as well. It has been demonstrated that PIFs and SPAs interact with COP1, and enhance COP1 ubiquitylation activity [12, 14, 42], while photoreceptor activation inhibits COP1 E3 activity [32–36] (Fig 9).
In etiolated seedlings, two SPA proteins associate with COP1 homo-dimers and form stable core complexes through their respective coiled-coil domains, which in turn, serve to enhance the COP1 activity possibly by increasing substrate recruitment [11, 12, 14]. Upon exposure to light, phyA, phyB and CRY1 interact with SPA, while the CRY2 binds to COP1. These interactions result in destabilization and disruption of the COP1-SPA complex, and consequently inhibition of COP1 E3 ubiquitin ligase activity [32–36]. In a similar fashion, we speculate that CSU2 mediated repression may also be directed at dismantling COP1-SPA complex and/or blocking COP1 dimerization.
CSU2 and COP1 interact through their coiled-coil domains, and CSU2 coiled-coil domain is necessary for the repression of COP1 activity in vitro (Figs 4, 5 and 6). Moreover, CSU2 can inhibit COP1-mediated HY5 turnover in a cell-free plant extract assay, also in a coiled-coil domain dependent manner. The coiled-coil domain of COP1 is responsible for its self-dimerization, a necessary conformation for its E3 ubiquitin ligase activity [13]. Thus it is possible that CSU2-COP1 association may interfere with the COP1 self-dimerization (in vitro) as well as COP1-SPA interaction (in vivo), which may result in destabilization COP1 dimer and COP1-SPA complexes, in a similar mechanism to activated photoreceptors.
We found that csu2 cop1-6 seedlings contained slightly higher amount of COP1 protein than cop1-6 alone, although still substantially lower than in wild type (Fig 6D). This could also be explained by above mentioned hypothesis: lack of CSU2’s competitive binding to COP1 coiled-coil domain would stabilizes COP1 dimerization and COP1-SPA complex, which would protect COP1 protein to certain extent. Nonetheless the slight increase of COP1 protein alone cannot fully account for the complete suppression of HY5 level in cop1-6 csu2 double mutants in the dark (Fig 6D). We postulate that both stabilization of COP1, and more importantly an increase of COP1 activity, occur in the absence of CSU2, which most likely underlie the mechanism of suppression of cop1-6 by csu2.
csu2 specifically suppresses the cop1-6 allele, but not cop1-1 and cop1-4 (Fig 2, S13A and S13B Fig). In cop1-6, the mutation causes a splicing defect that eventually produces COP1-6 mutant protein with five additional amino acids insertion at severely decreased level [43]. COP1-6 protein is largely biologically functional [41]. The strong allele cop1-1 has a 66-bp deletion, causing a deletion from amino acid 355 to 376 (~74 kD) [43]. The COP1-1 protein is produced to wild-type levels (S13C Fig), but is severely functionally defective, as indicated from the mutant phenotype. cop1-4 mutant accumulates a truncated COP1 protein (~33 kD) containing only the N-terminal 282 amino acids, and it is expressed at same or reduced level compared to wild type [43] (S13C Fig). Interestingly, when COP1-4 (N282) protein is overexpressed, it can cause a dominant negative phenotype in wild type background [55]. Thus the loss-of-function mechanism of cop1-4 mutation is rather complicated. Among the three cop1 mutant alleles, cop1-6 is the only hypomorphic allele, as it produces a functional protein at a lower level. Since csu2 suppression works by releasing the repression on a functional COP1 protein, only cop1-6 can be effectively suppressed by lack of CSU2. The failure of suppression of cop1-1 and cop1-4 by csu2 may primarily attribute to the nature of COP1-1 and COP1-4 mutant gene products, which are functionally defective. It should be mentioned that hy5 is able to partially suppress cop1-1 and cop1-4, as well as cop1-6 [44], because HY5 is a downstream factor that mediates COP1’s output.
Arabidopsis exhibited longer roots in the light and shorter roots in darkness, while cop1-6 displayed a revered phenotype [49]. In the dark, COP1 directly targets SCAR1, a positive regulator of root growth for ubiquitination and protein turnover [25], which contributes to the longer primary root phenotype of cop1 in darkness. The drastic long primary root length of cop1-6 grown in darkness was completely suppressed by csu2 (Fig 7). In the light however, both CSU2 and COP1 function as positive regulators of root development. The csu2 mutant seedlings developed severely shortened roots in the light, suggesting CSU2 is required for primary root growth in response to light (Fig 7).
Our study revealed that CSU2 may act upstream of COP1 in the hypocotyls, whereas may genetically act downstream of COP1 in the roots, and that a functional CSU2 protein is required for primary root growth both in WT and in cop1-6 (Fig 8). Thus, it appears that different regulatory module of CSU2-COP1 pair may exist in the hypocotyl and root cells. Nevertheless, the exact functional relationship between COP1 and CSU2 in regulation of root growth needs further investigation.
The cop1-6 [43], hy5-215 [44], csu2 cop1-6 (csu2-1 cop1-6 to csu2-6 cop1-6), and csu2 (csu2-1 to csu2-6) (this study) mutants are in the Columbia-0 (Col-0) ecotype. Seeds were surface sterilized with 30% commercial Clorox bleach and 0.02% Triton X-100 for ten min and washed three times with sterile water, and sown on 1×Murashige and Skoog (MS) medium supplemented with 0.4% Bacto-agar (Difco) and 1% sucrose. The seeds were stratified in darkness for three days at 4°C, and then transferred to light chambers maintained at 22°C.
The genetics screen, identification and characterization were previously described [41]. Genetic complementation tests showed that six different csu2 (csu2-1 cop1-6 to csu2-6 cop1-6 lines) EMS mutations were allelic to each other. Homozygous mutant suppressor plants were crossed to wild-type plants (Col-0), and segregation in the F2 generations was analyzed in the dark to distinguish between intragenic and extragenic suppressors. Meanwhile, the suppressor mutants were backcrossed to cop1-6. The phenotype of F1 and the segregation ratio in the F2 generations in the dark were analyzed to identify whether the suppression phenotype is caused by a monogenic recessive mutation.
Rough mapping was performed as described [41]. We crossed csu2-4 cop1-6 (Col background) with Landsberg containing a cop1-6 mutation to generate the mapping population. F2 generation seeds were sown on plates containing 1×MS medium, and grown in darkness at 22°C for five days. The suppressor seedlings with long hypocotyl and apical hook were then picked for Genomic DNA extraction and mapping. The markers used for mapping were designed based on the Arabidopsis Mapping Platform (http://amp.genomics.org.cn) and the standards described previously [56]. CSU2 was rough mapped to a ~250 kb region between markers 1-U89959-0145 and 1-AC022521-0169 on the left arm of chromosome 1.
SOLiD sequencing and mutation identification was performed as previously described [41]. The fragment libraries were created using the SOLiD Fragment library construction procedures according to the manufacturer’s instructions (Life Technologies, Carlsberg, USA). The libraries were sequenced on a SOLiD5500 sequencer according to the manufacturer’s instructions (Life Technologies, Carlsberg, USA). Mapping of sequencing reads to the Arabidopsis thaliana reference genome (TAIR10) and single nucleotide polymorphism (SNP) calling were accomplished using LifeScope v2.5. SNPs were then sorted into four categories (EMS induced homozygous, EMS induced heterozygous, other homozygous and other heterozygous). Candidate homozygous EMS induced SNPs were identified in windows with reduced heterozygosity in the regions identified by physical mapping using in house scripts.
To measure the hypocotyl and root length of seedlings, seeds were sown on horizontal or vertical plates and stratified at 4°C in darkness for three days, and then kept in continuous white light for eight h in order to induce uniform germination. The seeds were then transferred to dark, white, blue, red, and far-red light conditions, and grown at 22°C for five days [41]. The hypocotyl and root length of seedlings was measured using ImageJ software.
The full-length CSU2 open reading frame (ORF), CSU2 coiled-coil domain fragment and CSU2 lacking coiled-coil domain fragment were cloned into the pDONR-221 vector (Invitrogen) and introduced into the plant binary vector pEarley Gateway 103, pEarley Gateway 104 or pEarley Gateway 203 [57] under the 35S promoter using Gateway LR Clonase enzyme mix (Invitrogen). pEarley Gateway-CSU2-GFP, pEarley Gateway-YFP-CSU2, pEarley Gateway-Myc-CSU2, pEarley Gateway-CSU2 coil-GFP, and pEarleyGateway-CSU2Δcoil-GFP constructs were generated.
pB42AD-COP1, pB42AD-COP1N282, pB42AD-COP1ΔRing, pB42AD-COP1 Ring, pB42AD-COP1 coil, and pB42AD-COP1 WD40 constructs were described previously [17].To generate pLexA-CSU2, pLexA-CSU2 coil and pLexA-CSU2 Δcoil constructs, full-length CSU2, CSU2 coiled-coil domain and CSU2 lacking coiled-coil domain fragment were amplified by PCR with the respective pairs of primers and then cloned into the EcoRI/XhoI sites of the pLexA vector (BD Clontech). To produce the constructs for BiFC assays, each full-length CSU2 or COP1 fragments was amplified by PCR with the respective pairs of primers and then cloned into the NcoI/NotI sites of pSY728 or pSY738 vector [58], respectively. COP1-Flag construct was prepared with modified versions of pCombia1300 plasmid. pJIM-35S-HA-HY5 [59], pCombia1300-35S-GFP [60], and pCombia1300-35S-P19 [61] constructs were described previously. To produce pCold-TF-COP1, full-length COP1 were amplified by PCR and then cloned into the KpnI/PstI sites of the pCold-TF vector (Takara). To generate pET28a-CSU2 and pET28a-CSU2 Δcoil, full-length CSU2 or CSU2 Δcoil fragment lacking CSU2 coiled-coil domain were amplified by PCR and then cloned into the NdeI/XhoI sites of the pET28a vector, respectively. The primers used for plasmids construction were listed in S1 Table.
The LexA-based yeast two-hybrid system (BD Clontech) was used for the assays. The respective combinations of LexA and AD fusion plasmids were co-transformed into the yeast strain EGY48. Yeast transformation and the β-galactosidase activity assays were performed as described in the Yeast Protocols Handbook (BD Clontech).
Each pair of recombinant constructs encoding nYFP and cYFP fusions was co-bombarded into onion epidermal cells and incubated in 1×MS solid media containing 4% sucrose for 24 h at 22°C in darkness, followed by observation and image analysis by using confocal microscopy.
FRET and co-localization assay experiments were performed according to the standards outlined in previous research [19]. For FRET assays, the pAM-PAT-35SS-YFP-COP1 [41], pAM-PAT-35SS-CFP-CSU2 (this study), overexpression constructs were introduced into onion epidermal cells by particle bombardment and incubated, and live cell images were acquired using an Axiovert 200 microscope equipped with a laser scanning confocal imaging LSM 510 META system (Carl Zeiss). Cells were visualized at 24 h after particle bombardment using the confocal microscope. The multitracking mode was used to eliminate spillover between fluorescence channels. The CFP was excited by a laser diode 405 laser and the YFP by an argon-ion laser, both at low intensities. Regions of interest were selected and bleached with 100 iterations using the argon-ion laser at 100%. For co-localization assays, respective combination of pRTL2-35S-COP1 [19], pEarly Gateway-35S-CSU2-GFP (this study), pEarly Gateway-35S-CSU2 coil-GFP (this study), and pEarly Gateway-35S-CSU2Δcoil-GFP (this study) constructs were introduced into onion epidermal cells by particle bombardment, and incubated in darkness for 24 h. The cells were then analyzed by confocal microscopy.
Total RNA was extracted from five-d-old Arabidopsis seedlings grown under white light using the RNeasy plant mini kit (QIAGEN). cDNAs were synthesized from 2 mg of total RNA using the SuperScript II first-strand cDNA synthesis system (Fermentas) according to the manufacturer’s instructions. Then, cDNA were subjected to PCR or real-time qPCR assays. Quantitative real-time qPCR was performed using the CFX96 real-time PCR detection system (Applied Biosystems) and SYBR Green PCR Master Mix (Takara). PCR was performed in triplicate for each sample, and the expression levels were normalized to that of a PP2A gene.
In vitro ubiquitination assays were performed as previously described [41], with some minor modifications. Ubiquitination reaction mixtures (60 μL) contained 30 ng of UBE1 (E1; Boston Biochem), UbcH5b (E2; Boston Biochem), and 500 ng of HA-tagged ubiquitin (HA-Ub; Boston Biochem) in a reaction buffer containing 50 mM Tris at pH 7.5, 10 mM MgCl2, 2 mM ATP, and 0.5 mM DTT. 500 ng 6×His-TF, 500 ng 6×His-TF-COP1 (previously incubated with 20 μM zinc acetate), 500 ng 6×His-CSU2, and 500 ng 6×His-CSU2 Δcoil were applied in the reactions as indicated. After 2 h incubation at 30°C, the reactions were stopped by adding 5×sample buffer. One-half of each mixture (30 μL) was then separated onto 8% SDS-PAGE gels. Ubiquitinated TF-COP1 was detected using anti-ubiquitin (Santa Cruz), and anti-HA (Sigma-Aldrich) antibodies, respectively.
In vitro protein degradation assays were performed as described [62] with minor modification. For in vitro protein degradation analysis, Agrobacterium tumefaciens strains carrying constructs of p19 (for suppressing PTGS) together with HA-HY5, COP1-Flag, myc-CSU2, myc-CSU2Δcoil, or GFP (internal control) plasmids were co-infiltrated in Nicotiana benthamiana leaves, separately. One day after infiltration, a HA-HY5 sample was harvested. COP1-Flag sample, myc-CSU2 sample and GFP sample were collected after three days infiltration, individually. These four samples were separately extracted in native extraction buffer (50 mM Tris-MES pH 8.0, 0.5 M sucrose, 1 mM MgCl2, 10 mM EDTA, 5 mM DTT, 10 mM PMSF, 1×protease inhibitor cocktail (Roche)). Then, 100 μg HA-HY5 extract was mixed with 100 μg Flag-COP1, 100 μg GFP, 100 μg or 200 μg myc-CSU2 and myc-CSU2 Δcoil extract as indicated. A final concentration of 10 μM ATP was added to the reaction samples to preserve the function of the ubiquitination and 26S proteasome. For the proteasome inhibition, a final concentration of 50 μM MG132 was added to the corresponding mixtures. The mixtures were incubated at 4°C with gentle shaking for 6 h. Reaction was stopped by the addition of 5×SDS sample buffer and boiling for 10 min before protein gel analysis. The primary antibodies used in this study were anti-Flag (Sigma-Aldrich), anti-HA (Sigma-Aldrich), anti-GFP (BD Clontech), and anti-myc (Sigma-Aldrich).
Statistical analysis was performed by using GraphPad Prism 6 (GraphPad Software). To determine statistical significance, we employed one-way ANOVA with Tukey’s posthoc test. The difference was considered significant at P < 0.05.
Sequence data from this article can be found in the Arabidopsis Genome Initiative database under the following accession numbers: CSU2 (At1g02330), COP1 (AT2G32950), HY5 (AT5G11260).
|
10.1371/journal.pgen.1003014 | Morphogenesis and Cell Fate Determination within the Adaxial Cell Equivalence Group of the Zebrafish Myotome | One of the central questions of developmental biology is how cells of equivalent potential—an equivalence group—come to adopt specific cellular fates. In this study we have used a combination of live imaging, single cell lineage analyses, and perturbation of specific signaling pathways to dissect the specification of the adaxial cells of the zebrafish embryo. We show that the adaxial cells are myogenic precursors that form a cell fate equivalence group of approximately 20 cells that consequently give rise to two distinct sub-types of muscle fibers: the superficial slow muscle fibers (SSFs) and muscle pioneer cells (MPs), distinguished by specific gene expression and cell behaviors. Using a combination of live imaging, retrospective and indicative fate mapping, and genetic studies, we show that MP and SSF precursors segregate at the beginning of segmentation and that they arise from distinct regions along the anterior-posterior (AP) and dorsal-ventral (DV) axes of the adaxial cell compartment. FGF signaling restricts MP cell fate in the anterior-most adaxial cells in each somite, while BMP signaling restricts this fate to the middle of the DV axis. Thus our results reveal that the synergistic actions of HH, FGF, and BMP signaling independently create a three-dimensional (3D) signaling milieu that coordinates cell fate within the adaxial cell equivalence group.
| How specific genes and signals act on initially identical cells to generate the different tissues of the body remains one of the central questions of developmental genetics. Zebrafish are a useful model system to tackle this question as the optically clear embryo allows direct imaging of forming tissues, tracking individual cells in a myriad of different genetic contexts. The zebrafish myotome, the compartment of the embryo that gives rise to skeletal muscle, is subdivided into a number of specific cell types—one of which, the adaxial cells, gives rise exclusively to muscle of the “slow twitch” class. The adaxial cells give rise to two types of slow muscle cell types, muscle pioneer cells and non-muscle pioneer slow cells, distinguished by gene expression and different cellular behaviours. In this study we use lineage tracing live imaging and the manipulation of distinct genetic pathways to demonstrate that the adaxial cells form a cell fate “equivalence group” that is specified using separate signaling pathways that operating in distinct dimensions.
| The mechanisms that are utilised to generate individual cell types from a set of equivalently fated set of precursors remains a central experimental focus of developmental biology. Studies from invertebrate systems have defined the concept of an equivalence group, where small clusters of lineage related cells are determined by a combination of inductive and intrinsic signals to adopt individual fates [1]–[6]. This concept faces many difficulties when applied to complex three dimensional tissues such as those that typify vertebrate development, where the direct lineage relationships of many cells remain ill defined and the complicated morphogenesis of many tissues precludes definition of models of equivalence.
Zebrafish provides perhaps one of the most tractable contexts in which to examine concepts of cell fate determination in a vertebrate embryo, as a variety of lineage tracing techniques can be deployed in different genetic contexts in real time within an optically accessible embryo. One zebrafish lineage that has been examined in some detail is the embryonic myotome of zebrafish. As in all vertebrates, the majority of skeletal muscle in zebrafish forms from precursor cells present in the somites, which arise by segmentation of paraxial mesoderm in a rostral to caudal progression on either side of neural tube and notochord along the main body axis of the embryo. This process, referred to as myogenesis, gives rise to distinct slow and fast twitch muscle populations that differ in contraction speeds, metabolic activities and motoneuron innervation. In zebrafish, the location and origin of these two different cell populations are topographically separable [7], [8]. The early differentiating slow-muscle cells arise from a particular subset of presomitic mesodermal cells, termed the adaxial cells, which at the end of gastrulation align medially against the notochord [8]. These precursors initially adopt a pseudo epithelial morphology but shortly after their incorporation within the formed somite, undergo stereotypic morphogenetic cell shape changes, moving from their columnar shape to flatten and interleave, adopting a triangular shape, that upon further differentiation results in single adaxial cells extending from one somite boundary to the other. These cells collectively flatten medio-laterally to form a set of elongated myocytes that span the somite, positioned against the notochord [9].
Ultimately, adaxial cells give rise to two distinct sub-types of slow muscle fibers: the superficial slow-twitch muscle fibers (SSFs) and the muscle pioneer cells (MPs). SSFs and MPs possess distinct morphological, molecular and functional properties. After undergoing the initial morphogenetic cell shape changes described above, SSFs migrate from their notochord-associated midline position to traverse the entire extent of the forming myotome and come to lie at its most lateral surface. There, the SSF precursors complete their differentiation to form a monolayer of approximately 20 slow twitch muscle fibers. By contrast, MPs (2 to 6 per somite) do not migrate from the midline and are the first cells of the zebrafish myotome to differentiate, forming slow twitch muscle fibers immediately adjacent to the notochord [10]. All slow fibers express slow isoforms of myosin heavy-chain (SMyHC) as well as the homeodomain protein Prox1 and are mono nucleated cells [11]. MPs, in addition, express high levels of homeodomain-containing Engrailed proteins [12], [13]. By contrast to slow precursors, differentiating fast precursors originate from the lateral somite and fuse to form multinucleated fibers, subsequently to SSF migration, and are distinguished by their expression of fast MyHC. A subset of these fibers, known as medial fast fibers (MFFs) also expresses Engrailed at lower levels than MPs [14]. The timing of the fate determination of these distinct cell types has been examined by a rigorous in vivo transplantation assays. By interchanging slow and fast muscle precursors at specific points in their development it has been demonstrated that at the time of gastrulation, although slow and fast muscle precursors are already spatially segregated, they remain uncommitted to their individual fates until they have entered into the segmental plate. Furthermore, the subdivision of adaxial compartment in to MP and non MP cell fates occurs at a similar period of development, with MP becoming irreversibly fated within the posterior part of the segmental plate during early somite formation [7].
In vertebrates, the specification and differentiation of the somite into specific cell types is under the influence of inductive signals from the somite itself or those derived from the surrounding tissues (reviewed in [15], [16]). In the case of zebrafish myogenesis, by far the most well understood inductive signals controlling myogenesis are the Hedgehog (HH) family of secreted glycoproteins, which emanate from the embryonic midline. Numerous studies in the last two decades have demonstrated that HH is necessary and sufficient for induction of the slow twitch muscle fate. Indeed, analysis of loss of function mutants in HH pathway genes and the use of the HH pathway inhibitor cyclopamine have demonstrated that the timing and the level of HH signaling are critical for the formation of different muscle identities, including the MP cells, which require the highest level of HH signaling for their formation [14], [17]–[19]. However, even though HH over-expression can induce supernumerary MP cells, this is not sufficient to convert the dorsal and ventral extremes of the myotome into MP cells [20], [21] suggesting that other signals could induce MP in the midline region or repress MP differentiation in the dorsal and ventral muscle cells [21].
A further complication of these analyses is that they fail to explain how the symmetry of the adaxial cell compartment is initially broken to generate the dichotomy of MP and SFF fates within equivalent sets of cells. As the adaxial cells flank the notochord and floorplate, the source of HH peptide secretion, all adaxial cells would initially be exposed to the same level of secreted HH peptides. Hence, it is unclear how different levels of HH could act to generate the MP cell fate within a subset of adaxial cells and suggests that additional signals must influence adaxial cell fate. Recent studies have shed some light on the nature of other secreted signals that may act to influence muscle cell formation. Several studies have shown that manipulation of BMP signaling can alter MP number [21], [22]. Furthermore, Smad5, a downstream effector of BMP signaling has been shown to be activated in the dorsal and ventral adaxial cells and is absent within the central region of the somite [22], [23]. In addition, Smad binding sites have been shown to regulate activity of the eng2a promoter, the eng gene expressed the earliest within MP precursors [12], [21]–[23]. Collectively, these studies suggest that BMP signaling can influence the number of different cell types within the embryonic zebrafish myotome, but exactly how this is achieved has yet to be determined mechanistically.
In this study, we utilize a combination of live imaging, retrospective and indicative fate mapping, molecular and genetic studies to demonstrate that MP and SSF precursors arise from distinct regions along the anterior-posterior (AP) and dorsal-ventral (DV) axes of the adaxial cell compartment. Uniquely, this regionalization is controlled by the action of different signal transduction pathways that act specifically to direct specification in distinct axial dimensions. We demonstrate that the sprouty4-mediated inhibition of FGF signaling induces MP cell fate in the anterior-most adaxial cells in each somite and that radar-mediated BMP signaling restricts this fate to the middle of the DV axis. Our results indicate that HH, FGF and BMP signaling synergize to determine cell fate within the adaxial cell equivalence group.
In order to understand the origins of SSF and MP precursors from within the adaxial cell compartment (Figure 1A–1B), we examined adaxial cell behaviors during the first phase of their differentiation via continuous 4D time-lapse analysis and retrospective fate map analysis of the entire forming myotome. The position and shape of the adaxial cells were followed using a membrane-bound GFP and a nuclear localized mCherry whose expression in all cells was achieved after mRNA injection at 1-cell-stage. This analysis identified that the first adaxial cells to initiate differentiation and elongation arise adjacent to the anterior border of each somite at its DV mid-point (Figure 1C–1M and Video S1). These cells are most likely MPs, which previously have been shown to differentiate precociously [10]. To confirm this, we analyzed the expression of the MP marker gene engrailed2a (eng2a) during early somitogenesis by in situ hybridization. At the 10-somite stage eng2a transcripts were detected within newly formed somites exclusively within a subset of adaxial cells, adjacent to the anterior somitic border, located precisely at the mid-point of the DV axis of the somite (Figure 1N). To more precisely localize eng2a expression within the somite, we undertook dual in situ hybridization with myod, which marks the adaxial cells and the posterior aspect of the lateral somite, which contains the differentiating fast muscle progenitors (Figure 1O). This analysis confirmed that the expression of eng2a initiates specifically in the anterior-most cells of the newly formed somites. The positioning of cells initiating eng2a expression to the dorsal ventral midline of the forming myotome was confirmed in transverse sections of similarly staged embryos individually stained for eng2a and slow myosin heavy chain 1 (smyhc1) gene expression (Figure 1P, 1Q).
Collectively, these results suggest that SSF and MP precursors arise from distinct positions within the adaxial equivalence group. To test this hypothesis, we fate mapped the entire adaxial compartment by systematic iontophoretic injection of tetra-methyl rhodamine dextran (TMRD) lineage tracer dye into individual adaxial cells located at various AP and DV positions. Adaxial cells were labeled within the three most newly formed somites at the 10–15-somite stage and the fates of individually labeled cells were analyzed after the muscle fibers had terminally differentiated at 30 hpf. Individual injected embryos were sequentially incubated and imaged, first with an anti-Eng antibody and secondly with an anti-SMyHC antibody to unambiguously determine the fate of marked cells. This analysis confirmed that MP cells arise from the anterior-most adaxial cells at the dorso ventral midline of the somite (n = 8/8, Figure 2A, 2C, 2B, 2H) while posterior adaxial cells at this DV level make SSFs (n = 32/32, Figure 2A, 2D, 2E). Furthermore, we found that based on the initial position of a SSF precursor within the adaxial cell pool, we could predict its final location with the post migratory slow muscle palisade such that the dorsal- and ventral-most adaxial cells generate the dorsal and ventral-most post-migratory differentiated slow fibers respectively (n = 83, Figure 2B, 2F–2G, 2I–2J). This analysis not only demonstrates that MP and SSF precursors segregate at the beginning of somitogenesis but also determines the exact position of the precursors of every slow fiber. To further validate the fate of the adaxial cells located in the anterior somite at the DV mid-point, we examined their behaviour during the migration period. We thus performed a time-lapse analysis during a 20 hour period on embryos that were injected with a DNA construct containing the GFP gene under the control of the slow-twitch muscle-specific, slow myosin heavy chain 1 (smyhc1) promoter. When located in the anterior margin of the somite, the transgenically labeled adaxial cell elongates in an anterior to posterior movement but remains adjacent to the notochord identifying the labelled cell as a MP (Video S2).
We next turned our attention to the molecular basis of the adaxial cell fate specification events that we had defined by our fate mapping strategies. A candidate approach, examining AP restricted inductive signals within the myotome, highlighted the FGF pathway as a putative regulator of AP patterning in the adaxial progenitors. Indeed, in zebrafish, at least two of the genes encoding fgf ligands, fgf8 and fgf17b have been shown to be restricted in expression to the anterior somite [24], [25], [26]. However an analysis of the expression of the downstream targets of the FGF cascade, erm and pea3 surprisingly revealed that asymmetric FGF responses occur specifically within the adaxial cells such that the anterior-most cells lose expression of FGF target genes during somite formation (Figure 3A, 3B–3B″ and data not shown). The temporal and spatial regulation of FGF signal activation during zebrafish myogenesis suggests a simple hypothesis. Distinct levels of FGF activation along the AP axis of the somite inform the adaxial cells of their position within this axis and consequently control their fate. In order to test this hypothesis we disrupted FGF signaling by the addition of the pharmacological inhibitor SU5402, a drug that blocks the phosphorylation of FGF receptors (FGFRs) and so prevents downstream signaling, as revealed by the downregulation of the target genes erm, pea3 and spry4 in SU5402 treated embryos (Figure 4A–4C, [25], [27]). SU5402 treatments at the 6-somite stage did not affect the number of slow muscle fibers (Table S1) but instead increased the number of MPs at the expense of SSFs, as revealed by a failure in slow-muscle fiber migration to the surface of the myotome and a corresponding increase in Engrailed positive MP cells evident at the midline (Figure 4D, 4E). Furthermore FGF inhibition does not alter the number of En positive medial fast fibers (Figure S1). The increase in MP number is foreshadowed by an expansion of the eng2a expression domain throughout the AP dimension of the adaxial cell compartment at the 10-somite stage (Figure 4F). Furthermore, the heat shock induced expression of a dominant negative form of FGFR1 that blocks the FGF/ERK signaling cascade also causes a similar increase of eng2a expression at the expense of SSF migration at 1 dpf (Figure 4G–4J). Collectively, these results show that FGF inhibition promotes the specification of the MP fate. Importantly, delayed addition of SU5402 until the 10-somite stage revealed that the more rostral 5–6 somites, which had already formed at the time of treatment, remained unaffected revealing a discrete temporal window of action for FGF signaling in MP specification within the newly formed somite (Figure 4K). This correlates specifically with the period of development when cuboidal cells are arrayed along the AP axis, prior to their differentiation (Figure 1A–1B). These data shows that FGF signaling inhibition specifies anterior identity and consequently MP fate within the adaxial cell equivalence group.
As described above the adaxial cells, and thus the slow twitch muscle lineage are highly dependent on Hedgehog (HH) signaling with the MP fate requiring higher levels and longer exposure to HH for proper specification than SSFs [14]. To test a possible cross talk between FGF and HH signaling, we analyzed the expression of HH target gene patched1 (ptc1). However ptc1 expression remains unaffected by SU5402 treatment (Figure S2A). FGF signaling was also recently shown to control the length of motile cilia within Kupffer's vesicle [28]. Although non-motile cilia are a distinct class of cell organelle, one possible mechanism for FGF action could be to regulate HH signal reception through the length or number of primary cilia on adaxial cells, as reception and activation of the HH pathway is controlled within the primary cilia in vertebrate cells [29]. However, our analysis suggests that SU5402 treatment doesn't affect the length or the number of primary cilia within the adaxial cells (Figure S2B). Therefore, the effect of FGF signaling on MP specification cannot be explained by modulation of HH transduction within adaxial cells.
To understand how the precise spatial activity of FGF is regulated to control the dichotomy of the cell fate decision evident with the adaxial cells, we systematically examined known inhibitors of the FGF pathway for their expression within the adaxial cells. This analysis revealed that sprouty4 (spry4), which encodes a known intracellular inhibitor of receptor tyrosine kinases (RTKs), including the Fgfrs [30]–[33], becomes specifically activated in the anterior adaxial cells. Furthermore, the loss of expression of FGF target gene erm in the anterior adaxial cells correlates spatially and temporally with the induction of expression of the spry4 gene in the identical cells (Figure 3C, 3D–3D‴). To test whether spry4 expression influences MP and SSF fate specification, we ectopically expressed it within the adaxial cell compartment. Mosaic overexpression of spry4 from the promoter of the smyhc1 gene (smyhc1:spry4-IRES-GFP), which drives expression throughout the adaxial cell compartment [34], [35] doubled the number of MP cells (47.29% of transgenic fibres, nfibres = 143) within the embryo compared to control embryos expressing GFP alone (24.9%, nfibres = 521) (Figure 4L–4N). Furthermore, over-expression of spry4 induces a third population of transgenic fibers that possess attributes of both MPs and SSFs. These rare fibers (4.19%, nfibres = 143) are able to migrate to the surface of the myotome and express Engrailed, a unique behavior never observed in control embryos (untreated or smyhc:GFP injected) (Figure 4L, 4O, 4O′ and 4O″). Reciprocally when we express a dominant negative form of spry4, using the identical smyhc1 promoter (smyhc1:dn-spry4-IRES-GFP) cell autonomous loss of spry4 leads to a loss of MP identity and adaxial cells that express dnspry4 are incapable of making MPs (0% of transgenic fibres, nfibres = 48, Figure 4L, 4P, [36]).
We next analyzed muscle development in mutants that have had the spry4 gene inactivated. spry4fl117 mutants carry a single A-to-T transversion, which introduces a stop codon early in the ORF of the gene (Figure 5A). The mutant allele encodes for a truncated protein, which lacks the putative activation domain involved in FGF signaling inhibition and is consequently predicted to engender a full loss of function in spry4 (Figure 5B). Maternal zygotic (MZ) spry4fh117 homozygous mutant embryos, but not heterozygous or zygotic (Z) mutants, exhibit a marked increase in FGF target gene expression (erm, n = 9/9; dpERK, n = 13/13; and spry4, n = 8/8, Figure 5C–5F and 5G–5H and data not shown), showing that FGF activity is increased in MZ spry4fh117 mutants. Furthermore, while we could show that both the number of slow fibers and their position was unaffected in MZ spry4fh117 mutant embryos (Figure 5I–5J, Table S1 and Figure S3) the number of MPs was less than half that of controls (n = 31, Figure 5K, 5L and 5N) a deficit that was rescued by SU5402 treatment (n = 32, Figure 5K, 5M and 5O), indicating that the deficiency of MPs associated with the loss of spry4 is directly due to FGF over-activation and not modulation of other RTKs.
Although the regional inhibition of FGF signaling can explain the localization of the MP precursors to anterior adaxial cells, it cannot explain the positioning of these progenitors to the DV midline of the somite. Several recent studies have shown that manipulation of BMP signaling can alter MP number [21], [22] and these studies also show that Smad5, a downstream effector of BMP signaling is activated in the dorsal and ventral adaxial cells and but not within cells of the central region of the compartment [22], [23]. Furthermore, Smad binding sites have been shown to regulate activity of the eng2a promoter [22], [23]. This has led to the suggestion that BMP activity could influence the fate of the myotome along the DV axis, although direct evidence for this assertion is lacking. Furthermore, Smad5 is also known to be activated by the TGF-ß signaling pathway in many biological systems and a number of tgf-ß genes are expressed during zebrafish during myogenesis complicating interpretation of these data [37]. To visualize BMP signaling more specifically we generated a transgenic line that expresses GFP under the control of a BMP Responsive Element which contains 5 tandem BRE elements derived from the Xenopus vent2 gene coupled to a minimal Xenopus id3 promoter, promoter elements known to specifically respond to BMP signal transduction. The activation of this transgene (Tg(5XBRE[vent2]:-20lid3:GFP) has been shown to occur specifically via the BMP signaling pathway, and not by other TGF-ß-related ligands [38], [39], [40] Figure 6A–6D). The expression of GFP in Tg(5XBRE[vent2]:-20lid3:GFP) embryos correlates with the distribution of phospho-Smad5 (Figure 6F). By early somitogenesis, BMP signaling is activated in the adaxial cells specifically in cells of the dorsal and ventral edges of the myotome, and reporter expression decreases in the midline (n = 14/14, Figure 6A, 6B, 6D, 6H) where MP precursor formation occurs (Figure 7C). Subsequent activity of the transgene is restricted to migrating adaxial cells but not to MPs (n = 12/12, Figure 6C and 6G). These data suggest that the different levels of BMP activation along the DV axis could control the dichotomy of the MP/SSFs cell fate choice.
As mentioned above, several BMP-like ligands are present in the tissues surrounding the myotome. gdf6a/radar exhibits polarized expression in the DV axis, with expression evident in the dorsal neural tube, hypochord, and the primitive gut endothelium [41], [42]. The specific temporal and spatial aspects of its expression suggest radar/gdf6a is the most likely BMP ligand to influence the DV patterning of the zebrafish myotome [41], [42]. To examine this question, we genetically down-regulated radar/gdf6a by the injection of antisense morpholinos specifically targeted to the zygotic radar/gdf6a mRNA (rdrMO, Figure 7A). Loss of zygotic radar/gdf6a function in Tg(5XBre[vent2]:-201id3:gfp) embryos causes a reduction of BMP activation evident within this line (n = 5, Figure 6E versus 6A), and a concomitant medio-ventral expansion of both the MP precursor domain (n = 6/6, Figure 7D versus 7C, Figure S5) and the number of differentiated MP cells at 24 hpf (nsomite = 21, Figure 7B and 7F versus 7E, Figure S5) consistent with previously reported results [43]. To confirm the specific effect of the rdrMO we generated a p53 and radar double morphant in which the number of MPs was similarly increased (Figure 7B and 7G) but non-Eng-positive SSFs now migrated properly at 24 hpf compared to the single radar morphant, (Figure 7M versus 7K, 7L). Furthermore, the phenotype of the p53/rdrMO injected embryos was identical to homozygous rdrs327 mutant embryos [44] (nsomite = 17, Figure 7B and 7H), an phenotype that could be reversed by careful titration with WT rdr mRNA injection (Figure S4). Embryos treated with Dorsomorphin (DM) (nsomite = 17), a specific pharmacological inhibitor of BMP signaling [45], exhibited a dose dependent increase in MP number (Figure 7B and 7I, 7N–7Q) and a concomitant reduction of GFP expression in Tg(5XBre[vent2]:-201id3:gfp) embryos (Figure 7R, 7S and [22]). A similar increase in MP number is also seen when adaxial cells are cell autnomously inhibited from responding to BMP like ligands through use of a dominant negative form of the BMP receptor (dnbmpr) expressed from the adaxial specific smyhc promoter (smyhc:dn-BMPr GFP) (39.01% of transgenic fibres, nfibres = 326, Figure 4L and 4Q).
To elucidate whether FGF and BMP signaling co-operate to control adaxial cell fate, we examined the formation of MPs and SSFs when both pathways were simultaneously knocked down. rdr morpholino injections into SU5402-treated embryos caused an increase in MPs and eng2a expression compared to controls (DMSO, SU5402 treatment or rdrMO alone) that was essentially additive (nsomite = 17, Figure 7B, 7J and 7T–7W), demonstrating that FGF and BMP cooperate to control the MP/SSF decision, and do so independently of one another.
While the experiments outlined above, together with those of previously published studies, clearly show that BMP and FGF signaling can influence MP formation, they do not provide direct evidence for a role in DV or AP axis specification. It is possible that these signals could influence proliferation of MP precursors or recruitment to the adaxial cell compartment. In order to examine these issues more directly, we fate mapped the adaxial cell compartment using iontophoresis of TMRD into embryos where FGF signaling (SU5402 treatment) or FGF and BMP (SU5402+DM treatment) signaling had been inhibited (Figure 8A). According to our model, the MP domain should expand in the AP axis without FGF signaling and along both the AP and DV axes in the absence of either signal. Consistent with these predictions we found that MPs in SU5402-treated embryos could be derived from posterior adaxial cells (n = 8/12), a situation never observed in untreated embryos, but remained restricted to the mid-point of the DV axis (Figure 8B–8D). MPs in SU5402+DM-treated embryos arose from a pool of progenitors expanded in both the DV and AP axes of the adaxial cell equivalence group (n = 7/11, Figure 8E). Collectively, these results demonstrate that FGF and BMP signaling synergize to control specification of adaxial cells in the AP and DV axis, respectively.
At the beginning of segmentation all adaxial cells are columnar shaped epithelial-like precursors that align medially along the notochord, and display no morphological asymmetry. By initially undertaking fate map analyses of the entire forming myotome we have defined the adaxial cell compartment as a cell fate equivalence group that gives rise to these two specific slow muscle cell fates, the MPs and the SSFs. We have further defined mechanistically how these precursors are induced to give rise to these two distinct populations. The adaxial cells differentiate asynchronously within newly formed somites, with the cells adjacent to the anterior somitic border and located at the mid-point of the DV axis of the somite being the first to initiate the morphogenetic and differentiation movements we have previously describe [9]. This morphogenetic asymmetry is mirrored at the molecular level where the same cells that undergo precocious differentiation simultaneously initiate expression of the MP specific marker gene eng2a. This analysis suggests that these cells are the progenitors of the MP cells. In order to examine this question directly we generated a fate map of the adaxial compartment and found that each slow muscle fiber type (SSF and MP) arose from a specific region of the adaxial cell array. While the anterior adaxial cells at the DV mid-point of the somite give rise to MP within the midline, the non-MP precursor adaxial cells go on to form the SSF palisade at the lateral surface of the myotome in direct topographical reflection of their position in the pre-migratory adaxial compartment. These data indicate that both dorso-ventral and the anterior-posterior identities need to be determined coordinately within the adaxial cell equivalence group for cell fate determination to occur correctly.
Previous analyses have indicated that HH signaling is required to specify the adaxial cells prior to the onset of segmentation and that levels of HH influence the fate of these cells [7], [14], [18]. However, in the absence of HH signaling, cells with a distinct morphology still form adjacent to the notochord, indicating that not all aspects of adaxial cell morphogenesis are controlled by HH signal transduction [7]. In the absence of HH signal activation, a fast twitch muscle gene expression profile is activated within these cells instead of genes indicative of the slow muscle lineage. Consequently, these cells differentiate as fast MyHC expressing, cells stochastically dispersed throughout the myotome [7]. Despite the ability of HH signaling to control the determination of the slow muscle fate, the three HH ligands expressed in the embryonic midline (ehh/ihhb, shh/shha, twhh/shhb, [12], [20], [46], [47] are not restricted in the anterior-posterior direction, nor is there any indication that HH target genes are asymmetrically activated within the nascent adaxial cell compartment in either the anterior-posterior or dorso-ventral planes. Furthermore, we could also find no variation in the length of the primary cilia in adaxial cells, in line with the lack of modulation of HH target gene expression within adaxial cells. Thus, a model involving distinct regulators of cell fate needed to be invoked in order to conceptually generate the MP fate from the anterior-most cells of the dorso-ventral midline of the adaxial cell equivalence group.
Many studies have examined the role of FGF signaling during myogenesis in vitro, where it has been shown to promote cell proliferation and represses myoblast differentiation. It has also been shown that early myoblast precursors require FGF in order to subsequently express their myogenic phenotype [48], [49]. However, despite these extensive in vitro studies the exact function of Fgf in the activation or the repression of muscle differentiation in vivo is controversial and appears to often to contradict this simple repressive role defined in vitro [50]. For example, zebrafish Fgf8-mediated signaling has been shown to drive the terminal differentiation of fast-twitch but not slow-twitch muscle fibers, and simultaneously also controls proliferation of the external cell progenitor layer, the equivalent of the amniote dermyotome [25], [51]. In amniote embryos, FGF signaling has been implicated in myogenesis in vivo, both in promoting progenitor cell proliferation [52] and in promoting their differentiation [53], [54]. In chick embryos most, if not all, replicating myoblasts present within the skeletal muscle masses of the limb express high levels of the FGF receptor FREK/FGFR4 and the inhibition of FgfR4 leads to a dramatic loss of limb muscle [54], [55]. Conversely, over expression of FGF in the chick somite leads to muscle differentiation suggesting that, as in the zebrafish lateral myotome, myogenic differentiation is positively controlled by FGF signaling [25], [54]. This is consistent with observations in mouse where ectopic expression of the cell autonomous negative regulator of FGF signaling sprouty2 in myogenic progenitors inhibits their differentiation [53].
Here we show that the FGF pathway does play a role in muscle formation but it is downstream of the HH dependent process of slow-twitch fiber specification. FGF signaling is asymmetrically activated in the adaxial cells. Specifically, within anterior adaxial cells it is strongly reduced, to the point of complete inhibition of specific FGF target genes. We have shown, using a combination of genetic and pharmacological approaches that down regulation of the FGF pathway promotes MP formation at the expense of SSFs within the adaxial cell compartment. This does not appear to be driven by the restriction of the expression of FGF ligands, since the FGF encoding genes, Fgf8a and Fgf17, are both localized to the anterior somite [25]. Rather, FGF signaling in anterior adaxial cells is inhibited by a cell autonomous negative regulator of the FGF signaling cascade, spry4 [30]. spry4 expression is induced by FGF signaling and has been shown to act in a negative feed back loop on the FGF pathway in a number of contexts (this present study and [27], [31]). The direct role of spry4 in MP formation is demonstrated by data that shows that the ectopic expression of spry4 in the adaxial cells induces MPs while its inactivation in spry4 mutant embryos inhibits this fate. Therefore, our results suggest a model where spry4 is activated within the anterior adaxial cell compartment in response to high levels of adjacent FGF ligands that ultimately suppress FGF signaling within these cells, thereby breaking equivalence in the anterior posterior dimension. This role appears to be more analogous to that played by FGF signaling during organogenesis rather than those outlined above for myogenesis, where the fate of various stem and progenitor cells are partitioned by activation or inhibition of FGF signaling in organs as diverse as the liver and pancreas [56], ear [57], [58] and teeth [59] often in conjunction with opposing cell fate determining signals, including BMP signaling.
While FGF signaling restricts the fate of the adaxial cells to the anterior most cells of the myotome, a second signal is needed to restrict the positioning of these cells in the dorso-ventral dimension. Recently, studies have demonstrated that the downstream effector of BMP signaling, p-Smad5 is specifically restricted to the dorsal and ventral adaxial cells, and is absent from cells of the dorso-ventral midline of the myotome [22], [23]. Furthermore, several previous studies have shown that manipulation of BMP signaling can influence the number of engrailed positive MPs [21], [43]. Indeed, the ectopic expression of chick Dorsalin-1, a BMP-like family member, in the zebrafish notochord inhibits MP development [21]. More recent studies have shown that inhibition of BMP via use of the small molecule inhibitor Dorsomorphin, or morpholinos against the BMP receptor bmpr1ba, results in an increase of MPs [22], [23]. However, exactly how BMP influences the formation of these muscle subtypes has remained unclear. Here we show that the fate of the adaxial cells is specified in the DV axis by a radar-mediated BMP signaling. This statement is supported by several lines of evidence. Firstly, a transgenic reporter line specific for BMP signaling reveals that at the onset of segmentation, BMP signaling is active in the dorsal and ventral most adaxial cells, but absent from in the DV mid-point of the forming myotome. This region of low BMP activity of correlates with the location of MP precursor specification, as specifically determined via our fate map analysis. Secondly, BMP signaling is mediated by gdf6a/radar in the adaxial cells and knockdown of BMP activity modifies the fate muscle precursors in the adaxial compartment and promotes MP formation in a dose dependant manner.
Previous analysis of the activity of BMP signaling during muscle formation in amniotes has provided evidence that it negative regulates the myogenic program [60], [61] a role it appears to also play in controlling the proliferation and the onset of myogenesis within the external progenitor cell layer of the zebrafish myotome [62]. However, in the context of the adaxial cells it does not appear to influence the proliferation of these progenitors, the timing of entry of these cells into myogenesis or the differentiation of the adaxial cells themselves. Our lineage analysis specifically illustrates that it alters the fate of this progenitor compartment.
In this study we show that in contrast to HH signal transduction, FGF and BMP signaling has no effect on the slow muscle fate but instead regulates the decision of adaxial cell progenitors to become either SSF or MP cells. Indeed, as discussed above, the activation of these signaling pathways promotes SSF formation while their decrease or absence promotes MP formation. Modulation of FGF or BMP signals does not affect HH signaling and the consequences of their knockdown on the adaxial fate are additive (this study and [22]). Similarly, manipulation of the level of HH signaling (mutants within the HH pathway or cyclopamine treatment) does not affect the expression pattern of phospho-Smad 5, suggesting that HH signaling does not influence cell fate indirectly through BMP signaling [22], [23]. Thus the FGF and BMP signals act independently of, and synergistically with, each other to control the SSF/MP cell fate dichotomy. Intriguingly, the application of both FGF and BMP is required for the induction of a specific muscle cell fate, the Pax7-positive satellite cell progenitors, in Xenopus animal caps [63]. This suggests that the synergistic action of BMP and FGF may operate to specify other muscle cell types.
While HH and BMP signalling have been demonstrated to coordinate cell fate determination in the chick neural tube [64] and HH, BMP and FGF signalling collectively control the specification of numerous cell types in vertebrate and invertebrate systems, the majority of these studies do not examine the fate of individual cells in real time. The developmental paradigm of the adaxial cells allows single cells to be labelled and tracked and their fate determined within a genetically defined cellular equivalence group in the living animal a set of attributes that is to our knowledge unique in vertebrate developmental systems. We therefore believe that our study suggests that the adaxial of zebrafish could emerge as a paradigmatic example of a vertebrate cell fate equivalence group, in the same manner as the, Drosophila neuroectoderm, parasegment and imaginal disc and the C. elegans vulva [1]–[6] which have provided exquisite cellular and genetic resolution to generate a detailed understanding of cell specification mechanisms within invertebrate systems.
Our results also demonstrate an integrated signaling milieu that coordinates the specification of muscle cell fates within the adaxial cell compartment. The adaxial cell pool is initially specified in the somitic region adjacent to the notochord by HH signal transduction from the embryonic midline. This, together with regional inhibition of FGF in the anterior-most adaxial cells and a lack of BMP signaling at the DV midpoint of the somite, creates a 3-Dimensional network of signals that restricts the MP fate to the most anterior cells within a specific cellular equivalence group in the developing myotome (Figure 9). These signals act independently from each other to determine fate and uniquely MP specification is controlled by the action of different signal transduction pathways that act specifically to direct specification in distinct axial dimensions. This essentially Cartesian system of cell fate determination is somewhat reminiscent of that deployed during the development of the ventral nerve chord of Drosophila where a complex series of patterning genes are deployed in gradients along the DV and AP axes to induced specific fate determining genes within individual neuroblasts within the neuroectodermal sheet [3], [4], [5]. However, in the case of the adaxial cells there is no evidence for a role of lateral inhibition, which in the Drosophila ventral neuroectoderm is required for the expression of individual proneural genes and adoption of specific fates [4], [5]. Furthermore, our results reveal that individual secreted signals act in specific dimensions within this Cartesian system, rather than in a cooperative or mutually exclusive manner to specify cell fate, the prevalent ways by which cells are determined in vertebrate systems.
Fish maintenance, staging and husbandry were as described previously [65]. Wild-type embryos of the TE strain were used in all staining and manipulation. Mutant alleles used were spry4fh117 (ZIRC, direct submission from the laboratory of Cecila Moens), radar/gdf6as327 (kind gift of Herwig Baier). Transgenic lines used were Tg (hsp70:dnFgfr1-EGFP)pd1 and Tg(5XBre[vent2]:-201id3:gfp).
In situ hybridization, antibody staining, and microtome sectioning were performed as previously described [65]. Probes were obtained by PCR amplification or from existing clones: sprouty4, erm (cb805), pea3 (IRBOp991G0430D) and eng2a [65]. In situ hybridizations on whole mount embryos were performed using digoxigenin (DIG)-labeled (Roche) antisense RNA probes and nitro blue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate (NBT/BCIP) or fast red (Sigma). Microtome sectioning was performed on ISH stained embryos. Antibodies used were: anti-sMyHC (1/10, F59, DSHB Iowa, USA), anti-GFP (1/500, Rockland), anti-Engrailed (1/10, 4D9, DSHB Iowa, USA), anti-Prox1 (1/150, Fitzgerald), anti-Phospho-Smad5 (1/100, Cell signaling technology), anti-diphospho-ERK (1/10000, Sigma) streptavidin-alexa546 (1/1000, Molecular Probe). Vibratome sectioning was performed before antibody staining when necessary. 3D reconstructions were performed using Nikon C1 and Leica SP5 Confocal microscopes and Imaris software.
Counts of the number of differentiated MPs or SSFs were performed in the yolk extension region of 6 to 15 embryos. Analysis of variance (ANOVA) determined statistical significance of differences within a 95% confidence interval. In specific figures the following statistics were applied: Figure 4L: ANOVA analysis, Figure 5K: ANOVA analysis, Figure 7B: ANOVA analysis, Figures S1, S3, S5: Student test, 2 tails, unpaired, Table S1: ANOVA analysis.
All constructs were assembled from entry clones using the Tol2kit (Kwan et al 2007). For transcription of RNA for whole-somite imaging, we assembled CMV/SP6-EGFPcaax and CMV/SP6-H2/afz-mCherry. Plasmids were linearized with NotI before transcription of capped RNA using an mMessage-mMachine kit (Ambion). Vectors used for mosaic analysis of single cells were smyhc1:spry4-IRES-EGFP, smyhc1:EGFP, smyhc1:dnBMPr GFP and smyhc1:dnspry4-IRES-EGFP. The new entry clone p5E-smyhc1 was made by subcloning the smyhc1 promoter from the plasmid p9.7kbsmyhc1:GFP-I-SceI (Elworthy et al 2008) into p5E-MCS (Kwan et al 2007). The pME-spry4 clone was made by cloning the full-length spry4 ORF into pDONR221. Similarly, the ORF of Xenopus type Ia BMPr truncated in C terminal (BMPrΔC) from BMPR22 construct ([66] or of dominant negative form of spry4 (spry4Y52A) from the pCS2-spry4Y52A were also cloned into pDONR221.
Injections were performed as described previously [65]. 40 ng/µl of DNA encoding smyhc:spry4 ires GFP or smyhc:GFP were injected in one cell stage. Adaxial cells were imaged in embryos where 25 ng/µl of both CAAX-GFP and NLS-mCherry encoding mRNAs were injected at the one cell stage. 3 ng/µl of radar morpholino alone (5′-GCAATACAAACCTTTTCCCTTGTCC-3′) or in combination with 3 ng/µl of p53 morpholinos (5′-GCGCCATTGCTTTGCAAGAATTG-3′) were injected at the once cell stage. SU5402 (calbiochem) was added to the embryo medium at gastrulation or between 6- to 10-somites at a final concentration of 80 µM and maintained until the appropriate stage. 10 to 50 µM Dorsomorphin (Sigma) was applied to similarly staged embryos. Heat shock induction of dn-fgfr1 expression was carried out at 6-somite stage. (hsp70:dnFgfr1-EGFP)pd1 transgenic embryos in there plate were placed at 38° during 2 hours. GFP expression was visualized immediately after heat shock to confirm the expression of the transgenic protein.
Iontophoresis injections as described in [67] with the following modifications: rhodamine dextran (10,000 MW, Molecular Probes, 5 mg/ml) combined with Biotin dextran (10,000 MW, Molecular Probes, 1.5 mg/ml) were injected into cells of agarose-imbedded, 10- to 15-somite stage embryos. Adaxial cell labelings were positioned on the dorso-ventral axis via references to adjacent tissue landmarks within injected embryos and were imaged as previously described. The labeled embryo was dissected free of agarose and was allowed to develop until 30 hpf; it was then remounted in a 3% solution of methylcellulose (Sigma) and imaged. Subsequently, the embryo was fixed 2H in 4% paraformaldehyde and sequentially stained for Engrailed and sMyhc as described above.
|
10.1371/journal.pcbi.1002671 | Sub-diffraction Limit Localization of Proteins in Volumetric Space Using Bayesian Restoration of Fluorescence Images from Ultrathin Specimens | Photon diffraction limits the resolution of conventional light microscopy at the lateral focal plane to 0.61λ/NA (λ = wavelength of light, NA = numerical aperture of the objective) and at the axial plane to 1.4nλ/NA2 (n = refractive index of the imaging medium, 1.51 for oil immersion), which with visible wavelengths and a 1.4NA oil immersion objective is ∼220 nm and ∼600 nm in the lateral plane and axial plane respectively. This volumetric resolution is too large for the proper localization of protein clustering in subcellular structures. Here we combine the newly developed proteomic imaging technique, Array Tomography (AT), with its native 50–100 nm axial resolution achieved by physical sectioning of resin embedded tissue, and a 2D maximum likelihood deconvolution method, based on Bayes' rule, which significantly improves the resolution of protein puncta in the lateral plane to allow accurate and fast computational segmentation and analysis of labeled proteins. The physical sectioning of AT allows tissue specimens to be imaged at the physical optimum of modern high NA plan-apochormatic objectives. This translates to images that have little out of focus light, minimal aberrations and wave-front distortions. Thus, AT is able to provide images with truly invariant point spread functions (PSF), a property critical for accurate deconvolution. We show that AT with deconvolution increases the volumetric analytical fidelity of protein localization by significantly improving the modulation of high spatial frequencies up to and potentially beyond the spatial frequency cut-off of the objective. Moreover, we are able to achieve this improvement with no noticeable introduction of noise or artifacts and arrive at object segmentation and localization accuracies on par with image volumes captured using commercial implementations of super-resolution microscopes.
| Biological function at its fundamental level involves molecular interactions on a nanometer scale, and it is this reason that biological imaging has pushed for increasingly better resolution. Light microscopy is highly prevalent in biology due to its combination of large field of view, simple sample preparation, cost effective usage and relatively high tolerance by biological samples. The problem with light microscopy is that diffraction of light limits the resolution of achievable images to hundreds of nanometers in volumetric space, which is much too low for the accurate localization of proteins in subcellular organelle or structures, such as the synapse of a neuron. Super-resolution light microscopy is now available, but its implementation usually requires technically complex and expensive imaging systems. In this paper, we demonstrate a method that combines physical thin sectioning of tissue with Bayesian based deconvolution of conventional, fluorescent microscopy to achieve volumetric resolution well below the diffraction limit, and that using this method we are able to greatly improve the computational segmentation and localization of labeled proteins in a reconstructed volume of brain tissue.
| The spatial resolution and definition of the cellular protein matrix is fundamental to the characterization and analysis of cellular function. The accurate resolution of sub-organelle protein localization, in tissue, on a proteomic scale is immensely useful. It is with this in mind that we developed Array Tomography (AT), a proteomic imaging technique. AT uses ribbon arrays of ultrathin (50–100 nm) physical sections of resin-embedded, fixed tissue for multiple rounds of immunohistological detection, which produces a rich, high-dimensional matrix of protein information in an ex-vivo context [1], [2]. AT allows the collection of 30+ channels of protein information in a cubic millimeter volume of brain tissue [1], [2]. This information is only useful if we can, with spatial accuracy, localize spatially aggregated protein units within cellular structures and in relation to all other imaged protein channels. This places a premium on the computational segmentation of objects in the image volume, and is highly dependent on resolution and contrast.
The axial resolution of AT image volumes is limited only by the physical sectioning, which is 50–100 nm and is far smaller than the diffraction limited axial resolution of most microsocopes (∼385 nm). However, the lateral resolution of AT image volumes is still limited by the Abbe diffraction limit (∼200 nm for visible wavelengths) [3], [4]. At that lateral resolution, the segmentation of densely packed proteins, such as Synapsin (a highly abundant presynaptic protein in the brain), is unreliable and difficult. Recently, AT was combined with direct stochastical optical reconstruction microscopy (dSTORM) to achieve lateral resolution of ∼40 nm [5]. However, dSTORM imaging is time consuming and requires specialized microscopes. Thus, we investigated deconvolution as a simple and efficient method to improve our resolution in AT. The reason for considering deconvolution is that the physical sectioning of AT provides full removal of out of focus light, and the ideal correction of refractive index, astigmatism, coma, spherical aberration and curvature of field [1]. Moreover, the thinness of the tissue coupled with the direct placement of the sample onto glass also means that the heterogeneity of refractive indexes in normal biological samples is not present, which further eliminates sources of aberration and wave-front distortions. These properties, which are not present in most imaging techniques, allow AT to produces image volumes where the point spread function (PSF) is truly spatially invariant throughout, which makes these images an ideal substrate for deconvolution.
Deconvolution is a method by which the diffracted light is computationally returned back into its actual source using either an idealized or empirically measured PSF [6], [3], [4]. The PSF describes the diffraction of light from a point source. Specimens in the image are blurred by the PSF at a point by point basis. This blurring can be considered a convolution operation on the image [7], [8], [3], if it is linear (each point source in the image sums their intensity linearly) and shift invariant (the PSF is the same for the entire field of view). Wide-field is such an imaging systems [8], although in actual biological tissue the heterogeneity and depth of the tissue volume does introduce aberrations, wave-front distortions and out of focus light contributions that can cause significant deviations in the PSF across the image volume, which adversely affect the quality of deconvolution. This is not the case for AT thin sections where the PSFs are truly spatially invariant. Moreover, it might be easier to appreciate the advantages of thin physical sections by thinking about the analogy to conventional optical sectioning microscopes such as confocals. Confocals achieve optical sectioning by using a pinhole to reject out of focus light. This improves image quality by increasing the collection of high spatial frequency information in the image, but this comes at a cost of reduced signal to noise, due to the rejection of in focus light by the pinhole. AT physically removes all out of focus light sources, which means that AT does not need to use a pinhole for optical sectioning thus allowing it to provide both high signal to noise (which, in normal confocal microscopy, would be maximized by a large-diameter pinhole) and measurement of high-frequency spatial information (which would be maximized by a small-diameter pinhole) [9], [3].
The content of high-frequency information in the image is reflected in the bandwidth of the Optical Transfer Function (OTF), which is the Fourier Transform (FT) of the PSF. In confocal the OTF bandwidth varies inversely with pinhole diameter [9], [3]. The OTF determines the actual spatial frequencies transferred to the recorded image. Thus, if the OTF were small at high spatial frequencies (as is the case for an expanded confocal pinhole or a conventional wide-field setup), the high-frequency components of the specimen would be greatly attenuated, causing blurring and decreased resolution. Interestingly, the OTF of a theoretical infinitely-small pinhole would have twice the bandwidth of a standard wide-field OTF [10], [9]. In AT, we approximate this ideal pinhole with physical sectioning, and combined with the spatially invariant PSF, allow us to perform deconvolution at its mathematical optimum, which should, with the correct algorithm, allow us to greatly increase the magnitude of recovery for high spatial frequency information in the OTF up to the physical bandwidth limit, which is defined by diffraction.
Richardson-Lucy deconvolution (RL) is a Bayesian based expectation maximizing deconvolution method originally developed for the restoration of images in astronomy [11]–[14]. RL has several advantages for AT images. It assumes the non-negativity of the observations and that the statistic of the associated noise follows a Poisson distribution, which is appropriate for fluorescent images [15], [13], [16]. RL is globally and locally intensity-conserving at each iteration [11], [12], thus ensuring that intensity data remain quantifiable after deconvolution [13], [15]. RL is computationally efficient, and the restored images are robust against small errors in the image and the point-spread function (PSF) [12], [11], [17], [15], which makes its real world implementation realistic. Finally, in our tests on AT images, RL significantly out performs other non-Bayesian based deconvolution methods, and has demonstrated a greater than 8 fold increase in the magnitude of spatial frequency recovery up to the diffraction limit, without any measurable introduction of artifact or noise into the images. Moreover, RL in our application demonstrated mathematically a potential for the recovery of spatial frequencies beyond the diffraction limit, which likely contributes to the analytical improvements seen in the analysis of the deconvolved tissue volumes.
Thus, the confluence, in AT, of an essentially two-dimensional sample imaged at the optical optimum of the imaging system (e.g., minimal spherical aberration, optimal refractive index correction, ideal flatness of field, high signal to noise and a spatially invariant PSF) [1], [2] allows AT in combination with RL to achieve volumetric resolution significantly better than the diffraction limit. Using this technique, we demonstrate accurate and clean computational separation of objects in densely labeled tissue volumes.
Two-dimensional RL deconvolution is used to improve the resolution of protein structures. Initial deconvolution trials using ultra-thin sections seeded with 110 nm beads using RL with a high-quality, low-noise empirical PSF (Figure 1) or blind deconvolution using a hypothetical Gaussian as an initial PSF (Figure 2A) demonstrated that RL performed significantly better, returning most of the diffracted light back into the central pixel (1 pixel = ∼100 nm, 1.4NA Oil objective). Further tests using RL on volumes of YFP labeled dendrites of Layer 5 pyramidal neurons, imaged in traditional wide-field AT (ATW), demonstrated significant improvements in contrast and the visible recovery of high spatial frequency information in the image, which lead to a dramatic qualitative improvement in image quality (Figure 2B).
This qualitative increase in image quality accompanies a quantitative increase in object separation that can be further demonstrated through a simulation of improved point source discrimination by deconvolution of two adjoining points of light (Figure 3A–D). Within a fluorescent image measured intensity from point sources of light sum linearly [3], [8]. In figure 3 and Figure S1, two point sources are progressively moved further apart, and it is clear in both the image and the cross-sectional plot that after deconvolution the two point sources start to become visibly separate with only a single pixel between them (Figure 3B, S1B), while in the original image the two points only become noticeably separate with 3 pixels between them (Figure 3D, S1D). This demonstrates a theoretical improvement in resolution that pushes the resolvability of point sources in the image to 1pixel separation or 100 nm in our setup.
Although the simulations approximate real imaged objects in a noise free environment, a real world demonstration of improved resolvability is critical. Thus, we imaged in AT a volume of microtubules, and after deconvolution (Figure 4A–C) we demonstrated that indeed the resolvability of nearby microtubules, including those that are separated by a single pixel (Figure 4C) is improved. Furthermore, the most important aspect of this work is that, because array tomography generates large and information-rich datasets, we need methods of image processing and segmentation that are simple, fast and computationally efficient. Two-dimensional Bayesian based deconvolution significantly improves the performance and accuracy of finding the weighted centers of Synapsin puncta, an abundant presynaptic protein [2], by a simple 26 neighborhood connected component analysis, in 3D volumes of cortical tissue. (Figure 4D).
The apparent improvement of object separation in ATD images requires us to verify this result with imaging of AT ribbons using previously described and commercially available forms of super resolution microscopy. We first compared ATD with Structured Illumination Microscopy (SIM). SIM images the specimen using gratings of several orientations, which creates moiré fringes along the boundaries of the gratings. These moiré fringes provide extra spatial frequency information that can be extracted in Fourier space and used to reconstruct a new image with 100 nm resolution [18], [19]. We imaged AT ribbon arrays stained and labeled for tubulin, first using a commercial SIM, then using our wide-field AT setup. The result is a direct comparison of SIM, ATW, and ATD images of the exact same tissue volume with the exact same labeling (Figure 5). Qualitatively, the ATD images and the SIM images are virtually identical, whereas the wide-field AT image appears to have significantly lower contrast and definition (Figure 5A). Furthermore, looking at the intensity profiles of two microtubules running side by side it is clear that SIM and ATD provide similar quantitative separation of the two intensity profiles as well as matching intensity peaks and valleys, which suggest similar localization accuracy (Figure 5B–C). Finally, it is informative to look at the FT of the image volumes in the three modalities, which show that in the ATD and SIM case there is a significant increase in high spatial frequency information as demonstrated by the expansion of the magnitudes in the frequency domain (Figure S2).
Next we compared ATD to Continuous Wave Stimulated Emission Depletion microscopy (CWSTED) [20], which uses an excitation beam that is perfectly aligned with an annular depletion beam that limits the fluorescent release of photons to only a small nanometer size spot in the imaged specimen [21], [22], [20]. For this experiment, we were able to achieve 90 nm resolution with CWSTED. We imaged ribbon arrays in CWSTED and AT in a setup similar to the SIM experiments with the exception that instead of tubulin we stained the brain tissue for Synapsin. Again, the CWSTED and ATD images are extremely similar by visual comparison (Figure 6A, B). More importantly, the locations of the calculated centers of mass using CWSTED and ATD are similar, even with the expected jitter caused by the alignment and scaling of images due to the differences in the two imaging setups (100× objective with 50 nm pixels for CWSTED and 63× objective with 100 nm pixels for AT) (Figure 6C–E). A histogram of point to point distances between the modalities shows that the majority of points are within 1.5 pixels of each other (Figure 6F). The most striking difference between ATW and ATD in comparison to CWSTED is the number of objects computationally segmented in the image volume using 3D connected component analysis, with ATW lagging CWSTED and ATD due to the poor 3D object separation in the image volume (Figure 6G).
Finally, it is of interest to look at the empirical OTFs of the above modalities. More specifically, we are interested in the modulus of the OTF or the Modulation Transfer Function (MTF), which describes the amount of signal power present at each spatial frequency, or more practically, the amount of contrast that can be generated for each spatial frequency and relates directly to the resolvability of that spatial frequency in the actual image. The measured MTF was generated by applying FT to PSFs generated with 100 nm beads imaged at 488 nm wavelength for AT images and single sub-diffraction primary with secondary fluorescent antibodies at 488 nm in CWSTED. The MTF of ATW falls off dramatically as we approach the theoretical cut-off frequency of a 1.4NA objective (Figure 7). The cut-off frequency is described by the equation 2NA/λ (λ = wavelength, NA = numerical aperture). This clearly demonstrates the bandwidth-limited nature of the MTF in AT imaging. Two dimensional blind deconvolution of the ATW images increases the amount of signal at the higher spatial frequencies, but it only serves to bring the MTF edge closer to the theoretical cut-off (Figure 7). CWSTED's major gain in the MTF is at the higher spatial frequencies and as expected for a super resolution technique it surpasses the cut off value (Figure 7). The most significant aspect of the ATD MTF is the dramatic increase in modulation at all frequencies within the frequency cut-off. This massive improvement in modulation is the most likely cause of the image improvement seen in ATD, however intriguingly the ATD MTF, like CWSTED was able to extend beyond the frequency cut-off of the objective. The MTF of the actual AT images are bandwidth limited by diffraction, but it appears that in ATD, our deconvolution algorithm has mathematically extended the high spatial frequency information, which does eventually hit a hard limit, that is set by the image pixel size (100 nm), whereas CWSTED does not (pixel = 50 nm) (Figure 7). Further testing of ATD with 50 nm pixels using a 1.6× optivar and the 63× objective revealed that the higher spatial frequency component can be further pushed out approaching CWSTED levels (Figure 7). While this is a curious result and has interesting implications to the interpretation of our result, this phenomenon has been demonstrated in astronomical imaging. RL, but not blind deconvolution, applied to images with high signal to noise and band-limited OTFs can recover, through analytic continuation in the Fourier domain, frequency information beyond that of the measured object, thus allowing the extension of the MTF beyond the diffraction limit [23]–[25], [15]. Analytic continuation is a method in complex analysis that allows the extension of the domain over which a function is defined [26], [23], [25], [27]. Analytic continuation requires an original function to be analytic within its domain of definition, and not every complex function is analytic. In essence analytic continuation states that knowing the value of a complex function in some finite complex domain uniquely determines the value of the function at every other point. In image restoration, if a 2D object is compact in the space domain, i.e., confined within a finite region, its FT is analytic [28], [29]. In wide-field fluorescence images with diffraction limited OTFs, the image is an analytic function restricted to the pass-band, which analytic continuation maybe be applied to extrapolate it beyond the pass-band [23], [25], [30]. In practice, analytic continuation is highly sensitive to noise [26], [23], [31], [32] (Figure 8), and applied without constraints on real images results in little resolution improvement [33]. However, if we apply the reasonable constraint that all observations in our images are non-negative, which is an intrinsic assumption in RL, significant improvements in resolution can be obtained even with a moderate signal to noise ratio [23], [16], [25].
Finally we thought it might be of interest to test whether deconvolution of confocal images from our thin sections would improve our results further, because the confocal PSF is the multiplication of the excitation PSF and the emission PSF, which sharpens the lateral PSF and improves lateral resolution. Empirically we show, as we stated earlier, confocal have better native lateral resolution and spatial frequency capture, as can be seen in its MTFs as compared to ATW (Figure 9 A, B, E, F). Moreover, as one expects, by decreasing the pinhole size the MTF does see an appreciable increase in all spatial frequencies (Figure 9 A, B, E, F). RL deconvolution of confocal images, much like ATD, allowed the extension of spatial frequencies beyond the cut-off limit of the objective, especially when 50 nm pixels were used, and in some cases (when the pinhole is 1 airy unit (au) or smaller), RL plus confocal actually out performs ATD (Figure 9 D, H). This suggests that for array tomography, confocal imaging is a viable alternative to wide-field, although the gain in spatial frequency capture and recovery might not outweigh the increased image acqusition time, equipment cost and illumination intensity (especially, for small pinhole sizes (< = 1au) where confocal deconvolution beats ATD).
It must be noted that although our comparison of ATD with commercial SIM and CWSTED appear to suggest that ATD images in certain instances can approach the resolution of those techniques, we must caution that ATD is purely a mathematical process based on reasonable, but not perfect assumptions. It does not record extra spatial frequencies as SIM and CWSTED does through the use of deterministic light patterns. Moreover, the proper implementation of RL requires that the algorithm to converge through the iterations [34], and although in practice applying RL to AT images has always converged, one must be aware that this is a mathematical process that can fail, and the results of any deconvolution must be carefully interpreted. That said, the ideal optical characteristics of ultra-thin (50–100 nm) sectioning (minimal non-linear aberrations, optimal refractive index correction, ideal flatness of field, high signal to noise and a spatially invariant PSF) creates optimal circumstances for two-dimensional Bayesian based deconvolution (RL) to dramatically improve the MTF of AT images and perhaps even mathematically extend it, thus improving the resolution and computational segmentation of imaged protein structures. Our application of deconvolution, in the AT framework, truly allows RL to shine, because of the ideal data characteristics, which in many ways mimic the astronomical images that RL was originally designed for. Interestingly this does suggest that optical methods, such as evanescent field microscopy, that have extremely fine optical sectioning, could also benefit greatly from RL deconvolution.
The combination of deconvolution and AT creates volumetric images of intact tissue with a combination of speed, resolution, coverage and cost that cannot be matched by any other imaging modality. This coupled with the highly multiplexed imaging of proteins that is native to the AT procedure opens the door for the detection of biologically relevant protein localization in intact tissue samples at a scale and detail that will be crucial for understanding the function and dysfunction of biological systems. This spatial proteomic approach, where protein localization is maintained with sub-organelle precision from the in-vivo context can provide an essential piece of information that is missing in traditional proteomic approaches. It has become increasingly clear that the analysis of total expression level of proteins lacks the nuance that will be required to understand function at a complex cellular and systems level. The localization of a protein within a cell in relation to other proteins within its interaction repertoire is as important to the function of that protein as its modification state or its intrinsic structural and catalytic capabilities. The collection and analysis of this data is the information space that is uniquely occupied by ATD. It is this convergence of proteomic breadth with sub-organelle localization accuracy that will allow a much deeper analysis of biological function that can contribute significantly to our understanding of biological processes.
Tissue preparation, array creation and immunohistochemistry are described in detail in previous publications [1], [2]. In short, a small piece of tissue (∼2 mm high by 1 mm wide by 1 mm deep), in our case cortical tissue from the somatosensory cortex of the mouse brain, is microwave fixed in 4% Paraformaldehyde. The fixed tissue is then dehydrated in graded steps of ethanol, and then embedded in LR White resin overnight at 50°C. The embedded tissue is section on an ultramicrotome at a thickness of 70 nm and placed as a ribbon array directly on gelatin or carbon coated glass coverslips.
Immunohistochemistry is then carried out on the arrays using primary antibodies targeting antigens of choice (alpha-Tubulin, Abcam ab18251 and Synapsin, Cell Signaling Technology 5297S). The primary antibodies are visualized via fluorescently labeled secondary antibodies (Alexa 594, Invitrogen A11037, Alexa 488, Invitrogen A11034, and Alexa 647, Invitrogen A21245), and mounted in SlowFade Gold antifade with DAPI (Invitrogen).
Wide-field imaging of ribbons were accomplished on a Zeiss Axio Imager.Z1 Upright Fluorescence Microscope with motorized stage and Axiocam HR Digital Camera as previously described [1], [2]. A position list was generated for each ribbon array of ultrathin sections using custom software modules written for Axiovision. Single fields of view were imaged for each position in the position list using a Zeiss 63×/1.4 NA Plan Apochromat objective.
SIM imaging of ribbons were performed on a Zeiss ELYRA PS.1 super resolution scope using an Andor iXon 885 EMCCD camera. Positions on the ribbons were manually acquired across each section of the ribbon, and each fluorescent channel was imaged with five pattern rotations with 5 translational shifts, using a Zeiss 63×/1.46 NA Plan Apochromat objective. The final SIM image was created using modules build into the Zen software suite that accompanies the imaging setup.
CWSTED imaging was performed on a Leica TCS SP5 II using Lecia HyD hybrid PMT detectors. Positions on the ribbons were manually acquired across each section of the ribbon, and CWSTED images were acquired with a calibrated 90 nm resolution using a Lecia HCX PL APO 100× 1.40NA objective.
Confocal imaging was performed on a Zeiss LSM-510 using a Zeiss 63×/1.46 NA Plan Apochromat objective. Images of 100 nm beads, seeded on AT thin sections, were acquired using manually set pin-hole sizes ranging from 0.5 airy unit to 8 airy unit using either 100 nm pixels or 50 nm pixels.
Image stacks from ATW, SIM and STED were imported into FIJI and aligned using both rigid and affine transformations with the Register Virtual Stacks plugin. The aligned image stacks were further registered across image sessions using MultiStackReg.
The aligned and registered image stacks were imported into Matlab (Mathworks) and deconvolved using the native implementation of Richardson-Lucy deconvolution with empirical or theoretical PSFs with 10 iterations [15]. Custom functions were written to automate and facility this work flow. Blind deconvolution is also natively implemented in Matlab.
Matlab native function (regionprops) was used to calculate the centers of mass of punctas in the image volumes using 26 neighborhood 3D connected component analyses with an assumed background threshold that is 0.1 of the total dynamic range, which is 6553.5 for a 16bit image, and is in line with previous background thresholds used for AT analysis [2]. Custom functions were implemented to facility the handling and processing of the data.
|
10.1371/journal.ppat.1005168 | The White-Nose Syndrome Transcriptome: Activation of Anti-fungal Host Responses in Wing Tissue of Hibernating Little Brown Myotis | White-nose syndrome (WNS) in North American bats is caused by an invasive cutaneous infection by the psychrophilic fungus Pseudogymnoascus destructans (Pd). We compared transcriptome-wide changes in gene expression using RNA-Seq on wing skin tissue from hibernating little brown myotis (Myotis lucifugus) with WNS to bats without Pd exposure. We found that WNS caused significant changes in gene expression in hibernating bats including pathways involved in inflammation, wound healing, and metabolism. Local acute inflammatory responses were initiated by fungal invasion. Gene expression was increased for inflammatory cytokines, including interleukins (IL) IL-1β, IL-6, IL-17C, IL-20, IL-23A, IL-24, and G-CSF and chemokines, such as Ccl2 and Ccl20. This pattern of gene expression changes demonstrates that WNS is accompanied by an innate anti-fungal host response similar to that caused by cutaneous Candida albicans infections. However, despite the apparent production of appropriate chemokines, immune cells such as neutrophils and T cells do not appear to be recruited. We observed upregulation of acute inflammatory genes, including prostaglandin G/H synthase 2 (cyclooxygenase-2), that generate eicosanoids and other nociception mediators. We also observed differences in Pd gene expression that suggest host-pathogen interactions that might determine WNS progression. We identified several classes of potential virulence factors that are expressed in Pd during WNS, including secreted proteases that may mediate tissue invasion. These results demonstrate that hibernation does not prevent a local inflammatory response to Pd infection but that recruitment of leukocytes to the site of infection does not occur. The putative virulence factors may provide novel targets for treatment or prevention of WNS. These observations support a dual role for inflammation during WNS; inflammatory responses provide protection but excessive inflammation may contribute to mortality, either by affecting torpor behavior or causing damage upon emergence in the spring.
| White-nose syndrome is the most devastating epizootic wildlife disease of mammals in history, having killed millions of hibernating bats in North America since 2007. We have used next-generation RNA sequencing to provide a survey of the gene expression changes that accompany this disease in the skin of bats infected with the causative fungus. We identified possible new mechanisms that may either provide protection or contribute to mortality, including inflammatory immune responses. Contrary to expectations that hibernation represents a period of dormancy, we found that gene expression pathways were responsive to the environment. We also examined which genes were expressed in the pathogen and identified several classes of genes that could contribute to the virulence of this disease. Gene expression changes in the host were associated with local inflammation despite the fact that the bats were hibernating. However, we found that hibernating bats with white-nose syndrome lack some of the responses known to defend other mammals from fungal infection. We propose that bats could be protected from white-nose syndrome if these responses could be established prior to hibernation or if treatments could block the virulence factors expressed by the pathogen.
| White-nose syndrome (WNS) is an epizootic disease that has killed millions of bats in North America [1, 2]. WNS is caused by the psychrophile Pseudogymnoascus destructans (Pd) (formerly Geomyces destructans), an ascomycete fungal pathogen [3–5] that affects bats during hibernation. Pd grows at temperatures between 2 and 18°C and can infect bats while they hibernate [4, 6]. Pd is invasive and damages the cutaneous tissues of bats, including the wing [7], forming characteristic cupping erosions that are diagnostic of Pd infection [8]. Mortality rates due to WNS vary by species. In the little brown myotis, Myotis lucifugus, the mortality rate is up to 91% in affected caves [9, 10] whereas WNS resistance has been reported in the big brown bat, Eptesicus fuscus [11]. Bats in Europe are exposed to endemic Pd, but do not exhibit WNS mortality and appear to be resistant to the disease [12], despite cutaneous invasion by Pd [13].
Cutaneous infection by Pd causes some species of bats to arouse more frequently from torpor [5, 14, 15]. Although hibernating mammals spend less than 1% of their time euthermic [16], they use up to 90% of their stored energy during these periods [17, 18]. Because each arousal in little brown myotis utilizes an estimated 108 mg of stored fat [18], the increase in arousal frequency caused by WNS explains 58% of the morbidity rate associated with Pd infection [14]. Other factors that are also associated WNS pathology include effects of Pd infection on the integrity of wing tissue [7, 19], electrolyte balance and hydration [7, 20, 21], chronic respiratory acidosis [22], oxidative stress [23], and immune function [24]. The relative importance of each of these mechanisms in causing death in WNS is not clear, and the most likely model that has emerged is a multi-stage progression of WNS with contributions of several of these factors [22]. Differences in susceptibility to WNS between species in North America may be explained in part by different responses to Pd infection such as changes in thermoregulatory behavior. Understanding host responses to Pd infection may provide insight that could be useful for improving survival of affected species.
Cutaneous fungal infections in mammals are first recognized by components of the innate immune system, including C-type lectin receptors and Toll-like receptors [25]. Conserved components of the fungal cell wall activate pattern recognition receptors on phagocytes such as neutrophils, macrophages, and dendritic cells, and on epithelial cells [26]. Activation of these cells can lead to induction of the inflammasome, the production of inflammatory cytokines, and generation of reactive oxygen species that can mediate fungal cell killing [25]. The importance of the innate immune response to the initial recognition of fungal infections is demonstrated by the observation that deficiencies in these signaling pathways can lead to chronic fungal infections in humans [27, 28]. In the absence of invasion, colonization by commensal fungi can be maintained through tolerance mechanisms mediated by interactions with dendritic cells and epithelial cells in the skin [29]. Local activation of innate immune pathways can slow the growth of invasive pathogenic fungi and promote tolerance, possibly leading to a commensal relationship with the fungus [30], but is not usually sufficient to clear infections. Clearance of infections typically requires T helper (Th) cells, as demonstrated by the susceptibility of patients with acquired immune deficiency syndrome, immunosuppressant therapy, or chemotherapy to fungal infections [31]. These T cell responses can be mediated by Th17 cells [32, 33] or, in some cases, Th1 cells [34], with Th2 responses typically associated with greater susceptibility [35]. Th17 responses can contribute to clearance of invasive fungal infections through the actions of IL-17A and IL-22 [36] and the further recruitment and activation of neutrophils [37]. These T cell subsets have not been well characterized in bats, but those T-cell mediated immune mechanisms that have been studied appear to be conserved between bats and other mammals [38–41].
Fungal infections in animals are typically life-threatening only upon suppression of adaptive immune responses in the host, such as when chytrid fungus (Batrachochytrium dendrobatidis) blocks lymphocyte-mediated inflammatory responses [42]. Hibernation produces a natural suppression of some immune responses in mammal species where it has been studied. During hibernation, when the conservation of energy is critical, certain immunological mechanisms are downregulated while others remain unaffected [43–51]. Changes during hibernation can include depressed antibody responses [44, 52], decreased ability of T and B lymphocytes to proliferate in response to challenge [53, 54], and reduced complement activity [47]. Hibernation does not affect all immune responses equally, as shown in thirteen-lined ground squirrels (Ictidomys tridecemlineatus) that have a suppressed T-independent antibody response but are capable of mounting a T cell-dependent response during hibernation [44]. Studies of transcriptome-wide changes during hibernation in squirrels [55–59] have shown expression changes in genes involved in metabolism, oxidative stress, protein folding, ischemia/hypoxia, and other processes, but these studies were not examining an active immune response. Hibernation is also known to affect the distribution of leukocytes [45, 60] and platelets [61]. However, we have an incomplete understanding of how hibernation affects the suppression, or subsequent recovery, of immune responses [43], or how immune physiology in bats during hibernation may differ from that of rodents.
The cost of immune suppression during torpor is presumably outweighed by the benefits of energy conservation because most pathogens are not capable of proliferating at the low body temperatures of hibernating animals. However, the psychrophilic nature of Pd allows it to infect bats within hibernacula [2, 4]. The brief euthermic bouts of hibernating bats are shorter than most other hibernating mammalian species [14, 62] and it may not be possible for a bat naïve to a pathogen to mount a primary immune response in the few hours that it is euthermic throughout the hibernation season. We have observed antibody responses to Pd in bats, but these responses are strongest in active bats exposed to Pd after emergence from hibernation [63]. Therefore, hibernating bats may keep pathogens in check by relying on hypothermia, innate immune responses, and/or memory immune responses. The psychrophilic nature of Pd overcomes the first of these barriers to infection and the difficulty in fighting fungal pathogens with innate mechanisms alone may allow Pd to proliferate and invade the cutaneous tissues of bats.
The WNS panzootic has created an urgent need to understand if North American bat populations can persist in the presence of the fungal pathogen [1, 10]. Understanding the complete array of host responses mounted by bats afflicted with WNS may help illuminate sources of variation in survival within and among bat species. To determine which host responses are activated by Pd infection, we measured transcriptome-wide gene expression levels in bat wing tissue from hibernating bats affected by WNS. Gene expression was compared to bats that were hibernated in captivity in the absence of Pd exposure. We hypothesized that Pd infection would cause changes in gene expression that would reveal physiological responses during WNS that might be either protective or pathological. By using next-generation RNA sequencing to examine transcriptome-wide gene expression changes we expected to discover consistent patterns of host responses that occur in Pd-infected tissues. Combined with changes in gene expression within the Pd pathogen, these results have provided a survey of the host and pathogen interactions occurring during WNS.
To determine the host response mounted by little brown myotis to Pd during hibernation, we measured changes in gene expression at the whole transcriptome level. Wing tissue samples were obtained from hibernating little brown myotis with no known exposure to Pd and bats exhibiting physical signs of WNS, as shown in Table 1. Histopathology [8] and quantitative PCR (qPCR) for Pd [64] were used to confirm the WNS status of each bat (Table 1). Cupping erosions diagnostic of WNS were found on all 6 bats captured in Kentucky, but on none of the 5 bats from states negative for WNS at the time of capture. Low levels of neutrophilic inflammation were found in all 11 wing samples (Table 1; Infl), although this inflammation was not associated with sites of Pd infection. All 6 WNS-affected bats tested positive for Pd by qPCR, although the fungal load measured on wing swabs (Table 1; qPCR) did not correlate with the number of cupping erosions found by histology (Table 1; WNS). As previously shown [5, 14, 15], WNS-affected bats had significantly lower body condition (Table 1; SMI; p = 0.017, t = 2.9255, df = 9).
Next generation RNA sequencing (RNA-Seq) was performed using poly-A selected RNA isolated from each RNAlater-preserved wing tissue sample (S1 Table). Using expression levels of Pd-derived transcripts, we confirmed that all 6 WNS-affected bats had abundant expression of Pd genes. The Pd-derived transcripts were not present at significant levels in any of the 5 samples from unaffected bats (S2 Table; p = 2.2x10-6, t = 21.5, df = 5.33), including the MN090 sample that had tested positive for Pd by qPCR in one of the two replicates (Table 1). Because high levels of differential expression of Pd transcripts would make it more difficult to detect significant changes in host gene expression, the assembly was filtered [65] to remove Pd-derived sequences. Comparison of the filtered assembly with the original revealed that removing the Pd sequences did not significantly decrease the completeness of the assembly (S3 Table) as determined by BUSCO [66]. This filtered assembly (S1 Dataset) was used to calculate differential expression in host genes between the unaffected and WNS-affected samples.
We compared host gene expression across all samples (S2 Dataset) using DESeq2 [67] to identify transcript clusters that were expressed at a minimum of 2-fold difference and significant at a false discovery rate (FDR) of 0.05 (S1 Fig). We found 1804 transcript clusters that were expressed at higher levels, and 1925 transcript clusters expressed at lower levels, in WNS-affected bat tissues (S4 Table). Hierarchical clustering (Fig 1) revealed that expression of these transcripts from all 5 bats without WNS was similar to each other. Gene expression in wing tissue from WNS-affected bats was different from unaffected bats and more similar to each other, as predicted. The normalized expression levels of the 3729 identified transcript clusters differentially expressed are listed in S4 Table.
Differential expression of individual gene isoforms was further analyzed using EBSeq [68], an empirical Bayesian approach to modeling gene expression. For each transcript cluster identified as differentially expressed by DESeq2, we used EBSeq to determine if any of the individual transcripts were differentially expressed at a posterior probability greater than 0.99 (S4 Table). Of the 3729 differentially expressed transcript clusters identified by DESeq2, EBSeq identified at least one differentially expressed transcript for 1427 (38% of total, 43% of upregulated genes and 33% of downregulated genes). These results indicate that differences in gene expression are likely due to alternative splicing or other isoform differences for many of the differentially expressed genes.
To annotate the functions of these genes and identify those likely to be involved with host responses to Pd infection, we used the Trinotate pipeline. BLAST was used to identify 1365 upregulated transcripts and 325 downregulated transcripts in WNS-affected tissues with significant homology to known genes from vertebrates in the Swissprot database. Of the 2295 remaining transcripts, 13 were mapped to genes from non-vertebrates in the Swissprot database, presumably due to environmental contamination or incomplete removal of Pd transcript sequences. Of the 2842 trinity transcript clusters without a BLASTx match in Swissprot, 2731 (96%) were found to align to sequences (e-value < 0.0001) in the little brown myotis genome. Of the aligned transcripts, 204 (7.4%) were found to correspond to previously identified non-coding RNA sequences. Of the 111 transcript clusters without a transcript that aligned to the little brown myotis genome or Swissprot, BLAST was used to align their transcripts to the UniRef90 database. We found that 7 genes aligned to vertebrate homologs, 9 aligned to fungal homologs, and 15 aligned to other metagenomic sequences. We were unable to identify homologous sequences for any transcripts from 80 (2.1%) of the transcript clusters that were differentially expressed.
Expression levels for the Swissprot-identified transcript clusters with the 100 lowest adjusted p values are shown in Fig 2 (see S4 Table for all results). Some of the differentially expressed genes with putative functions that were predicted to associate with host responses to a fungal pathogen are listed in Table 2. WNS caused dramatic changes in expression of genes involved in inflammation, immune responses, wound healing, metabolism, and oxidative stress, even though the bats were hibernating during the Pd infection. Most of these genes were upregulated in WNS-affected tissues, while a much smaller number of identified genes with putative functions in these categories were downregulated (Tables 2 and S4).
To determine if all 6 little brown myotis with WNS exhibit similar changes in gene expression, we performed clustering analysis of the differentially expressed transcripts (Fig 1). To confirm the significance of these patterns of gene expression, bootstrap analysis of clustering was performed [69]. The clustering of the unaffected samples together and the clustering of the WNS-affected samples together was verified with a confidence of 99% (Fig 3A). Principal component analysis was performed to better understand the relationships between the transcripts expressed in the 11 samples (Fig 3B). All 5 samples from unaffected bats were very similar based on the first three principal components identified, which account for 71% of the variance in these transcripts. The WNS-affected bat samples have more diverse gene expression (S5 Table) and PC1 (accounting for 44% of the variance) differentiates all 6 from the unaffected bat samples. The genes represented by PC1 include those that are more highly expressed in unaffected than WNS-affected wing tissue (Fig 2). PC2 (17% of the variance) and PC3 (10% of the variance) distinguish the KY19, KY23, and KY39 samples from the other two WNS-affected samples and from the unaffected samples. The rotation values of principal component analysis (S5 Table) reveal that inflammatory genes made the greatest contribution to PC2. Clustering analysis revealed diverse host responses among the bats infected with Pd.
We next examined the functional pathways that were most affected in little brown myotis infected with Pd. For this gene ontology analysis, DESeq2 results on transcript isoforms were used with a higher FDR threshold of 0.1, as is typical for this type of analysis. From WNS-affected bat tissue, 3104 upregulated transcripts were aligned with BLAST to the human Uniprot database. Homologs for these transcripts were identified and a list of 1937 unique Ensembl IDs associated with upregulated genes was generated (S6 Table). GOrilla [70] was used to determine significantly upregulated gene ontology categories from the Uniprot GO ID database (Table 3 and S7 Table) and REVIGO [71] was used to visualize biological processes that were significantly overrepresented in the WNS-affected transcriptome (Fig 4). The functional analysis revealed that Pd infection increases expression of genes involved in metabolism, defense responses, and other pathways (Table 3).
For the transcripts that showed lower expression in WNS-affected tissue (Fig 2), the same analysis was performed. Of the 1152 identified transcripts that were downregulated at an FDR of 0.1, 694 homologous human genes were identified by BLAST and mapped to Ensembl gene IDs (S6 Table). GOrilla did not identify any Biological Process categories that were significantly downregulated in the WNS-affected bat tissue.
To examine the gene expression of the Pd pathogen using a dual RNA-Seq approach [72], we separately generated a genome-guided Trinity assembly (S3 Dataset) with the Broad Institute G. destructans genome. The reads from each of the WNS-affected tissues were mapped onto this assembly with Bowtie and gene expression estimated using RSEM (S4 Dataset). Expression levels for the Pd genes with the greatest variance are shown in Fig 5. Hierarchical clustering (Fig 6) and principal component analysis (S2 Fig) of the differentially expressed transcripts indicated that Pd gene expression was most similar in the wing tissues from bats obtained from the same hibernaculum (Table 1). The expression patterns of Pd genes were more similar for KY06, KY07, and KY11, which corresponds to bats captured in Cave 1 in Kentucky, and for KY19 and KY23, which were captured from Cave 2.
The possible functions of the Pd genes expressed among the WNS-affected samples were analyzed by sequence homology. We first examined the expression levels of a family of secreted proteases that have been proposed to be involved in Pd virulence [73, 74] and found that these alkaline proteases were expressed by Pd in all 6 wing samples (Table 4). Destructin-2 was the most highly expressed isoform in all WNS-affected bat Pd samples.
We next examined the Pd transcript clusters for additional factors that could affect virulence. Alignment by BLAST to the Swissprot and Uniprot90 databases identified 12 056 transcripts with significant homology to known fungal genes (S8 Table). For the remaining 67 Pd transcript clusters, Trinotate was not able to identify known functional domains or signal peptides present in these previously uncharacterized Pd transcripts. The results from the BLAST alignment were examined for genes known to be involved in processes that could affect Pd virulence, such as secreted proteases [73–75], metal binding proteins [76], fungal cell wall remodeling [76, 77], and other virulence factors [75, 77, 78]. This analysis identified 46 Pd genes that could be involved in pathogenesis (Table 5), including additional secreted proteases that could be involved in tissue invasion.
Because the tissue samples were collected from bats from 6 different hibernacula for this study, it is possible that differences in host or pathogen gene expression reflect differences in the environmental conditions present in each location, including the microbiome. In addition, the housing of the unaffected bats in captivity for 13 weeks prior to analysis could also have affected the microbiome. To examine the differences in the skin microbiome between the bats, we used MG-RAST to identify the lowest common ancestor of metagenomic sequences present (S8 Table). Although there were some differences observed in the bacterial microbiomes present on the wings of the 11 bats, there were no significant changes between the WNS-affected and unaffected samples when bacteria were identified at the class level. Several strains of Pseudomonas fluorescens isolated from bat tissues have been identified with Pd growth inhibiting properties [79]. MG-RAST analysis showed that Pseudomonas species are present in all 11 samples (S8 Table). P. fluorescens transcripts represented 2.8±0.6% of transcripts identified from gammaproteobacteria and 0.40±0.05% of all bacteria on the wings of unaffected bats and 0.37±0.07% of all bacteria on WNS-affected bats. P. fluorescens was present on all little brown myotis sampled, but was rare and relative abundance was not statistically different between WNS-affected and unaffected bats (p = 0.49, t = -0.71, df = 9).
The comparison of host gene expression between WNS-affected and unaffected little brown myotis clearly demonstrates that Pd infection causes physiological responses in wing tissue, where substantial fungal invasion of the skin occurs in WNS-affected bats [8]. The changes in transcript levels that we have observed indicate that host responses to fungal infection remain intact during hibernation and are similar to those observed during the initial stages of fungal infection in euthermic mammals [32]. These host responses include acute inflammation, wound healing, and metabolic changes. Pathogen gene expression varies among bats with WNS, suggesting host-pathogen interactions that mediate pathogenesis. Together, these results lay a foundation to determine which host and pathogen responses contribute to WNS resistance and susceptibility and identify targets to increase survival.
The gene expression changes we observed in the wing tissue of WNS-affected bats are similar to those observed in other cutaneous fungal infections [80]. Cutaneous Candida albicans infections in humans and mice typically initiate an immune response by activating pattern recognition receptors of the C-type lectin family [81–83] and the toll-like receptor family, both of which we found upregulated in WNS-affected bat wing tissue (S4 Table). These included C-type lectin domain (CLEC) family members CLEC4D (MCL), CLEC4E (MINCLE), CLEC7A (Dectin-1), CLEC6A (Dectin-2), and Toll-like receptor 9. In mice and humans, protective host responses to C. albicans are usually characterized by many of the same cytokines and chemokines [29] that we have found upgregulated in WNS-affected wing tissue, including the cytokines IL-1β, IL-6, G-CSF, IL-23A, and IL-17C. Little brown myotis infected with Pd are increasing transcription of the key genes necessary for initiating a host response that provides protection from fungal infection. This clearly demonstrates that hibernation does not prevent innate immune responses in bats infected with Pd and that, although they are not closely related to rodents and primates [41], bats respond to fungal infections similarly to these other mammals.
The responses to Pd infection within bat wing tissue may be mediated by keratinocytes in the epithelial tissue. Activation of pattern recognition receptors by fungal ligands is expected to induce keratinocytes to produce many of the cytokines that we have found upregulated at the transcript level in WNS-affected bat wing tissue [84]. In addition to the cytokines typically involved in C. albicans responses described above, keratinocytes are also known to express the chemokine Ccl2 and the cytokines IL-20 and IL-24 in response to pattern recognition receptor activation [85]. Keratinocytes and fibroblasts are also known to exhibit a paracrine loop of IL-1 and IL-6 activation [86] that enhances wound healing and host defense to microbial infection and we found evidence of IL-1 and IL-6 receptor activation in the increased RNA levels for transcription factor p65, NFκB, and P-selectin glycoprotein ligand 1 (S4 Table). Another important cytokine produced by epithelial cells in response to infection is IL-17C [87]. This is an atypical IL-17 family member that is expressed by epithelial cells and causes autocrine responses in the epithelial cells that also express the IL-17RA and IL-17RE heterodimeric IL-17 receptor [87]. The wing tissue transcriptomes from WNS-affected and unaffected bats show similar expression levels of both IL-17RA and IL-17RE (S2 Dataset) and would, therefore, be expected to be responsive to IL-17C. The gene ontology analysis also found evidence for functional enrichment of genes involved in keratinocyte differentiation, presumably due to wound healing responses. Keratinocytes or other epithelial cells in bat wing tissue appear to have responded to the invasion of the epidermis by fungal hyphae.
Genes for pro-inflammatory mediators characterized the innate immune response that we observed in the wing tissue of Pd infected bats. Under euthermic conditions this would be expected to provide protection by the recruitment of monocytes and neutrophils, mediated by G-CSF, IL-23A, Ccl2, IL-17C and IL-6 [88], and the initiation of an adaptive Th17 or Th1 response. However, under the constraints of hibernation, responses that require leukocyte migration do not appear to occur in Pd-infected bats. We do not find strong evidence of increased expression for genes characteristic of either innate or adaptive leukocytes, except for L-selectin, which is expressed on T cells, and CD177, which is expressed on neutrophils. Lower than expected levels of monocyte, neutrophil, Th1, and Th17 cell recruitment may be related to the sequestration of leukocytes during hibernation [45]. However, we have observed neutrophil recruitment in hibernating little brown myotis in response to another fungal infection (Table 1). In the histological examination of the current samples, we found neutrophilic inflammation in both WNS-affected and unaffected wing tissue (Table 1). However, this inflammation did not occur at the sites of Pd infection. Curiously, we found a significant increase in WNS-affected tissue for transcripts for CD3γ and CD45 that could be expressed by gamma-delta T cells or other innate lymphocytes that reside in the skin [89]. It is possible that Pd is specifically suppressing neutrophil and/or T cell recruitment by interfering with chemotactic signals, similar to the suppression of inflammatory immune responses during chytridiomycosis in amphibians [42]. However, analysis of tissue levels of the cytokines and chemokines is necessary to confirm the secretion of these proteins. Because neutrophils and T cells do not appear to be recruited to sites of Pd infection during hibernation, only local inflammatory mediators may be available and they appear to be unable to control the infection in little brown myotis.
In addition to immune responses, hibernating bats also respond to Pd infection in other ways. We found transcripts for proteins from many pathways involved in metabolism, signaling, gene expression, transport, migration, and differentiation that were altered in WNS-affected bats (Fig 4). We cannot exclude the possibility that some of these differences were due to the different hibernation conditions of the two groups of bats. However, the differential expression of the genes in these pathways demonstrates that they are subject to regulation during hibernation and can respond to infection, tissue damage, and/or environmental changes.
Host responses to fungal infection can be influenced by changes in the pathogen, including gene expression changes in the colonizing fungus, such as C. albicans [29]. We found significant variability in the gene expression by Pd, which is particularly interesting because all Pd in North America is presumed to be a clone of the same mating type [90]. The pathogen has adopted different gene expression profiles in the 6 bat tissues (Fig 5), perhaps in response to differences in the host environments. Correspondingly, host gene expression patterns also show differences between the WNS-affected tissue samples. Of particular interest is the observation that the cytokine and chemokine genes found in principal component 2 of our PCA analysis (Fig 3B and S5 Table) are expressed at very different levels in the 6 Pd-infected samples. From this study we cannot determine whether the differences in pathogen gene expression are driven by differences in the host environment or vice versa. Although all 6 WNS-affected bats had visible signs of WNS, had similar Pd burdens, and similar histopathology, it is possible that the differences in host or pathogen gene expression that we observed may have affected progression of WNS and survival.
Because the increased frequency of arousals from torpor appears to be a primary cause of WNS mortality [5, 14, 22], we considered possible mechanisms that could affect torpor bout length. The increased gene expression of IL-1, IL-6, and other pro-inflammatory cytokines mediates a local acute inflammatory response to Pd. These cytokines also have systemic effects that modify behavior and thermoregulation [91]. In addition to cytokine and chemokine transcript increases, we also found increased transcripts for the enzyme cyclooxygenase-2 (prostaglandin G/H synthase 2) and both secreted and cytosolic phospholipase A2 that form critical inflammatory lipid mediators such as prostaglandin H2. The eicosanoids generated by these enzymes, along with the actions of the upregulated genes kallikrein-6 and cathepsin S, are expected to generate pain and itching by locally activating neuronal nociceptors [92, 93]. This, in turn, could affect torpor bout length and/or behavior during periodic arousals. Indeed, we have documented significantly more grooming in WNS-affected bats infected in the wild [94], although a different study on laboratory-infected bats did not find similar behavior changes [95]. Together, the upregulated genes will likely generate an inflammatory microenvironment within the wing that may contribute to the robust wound healing response that we observe in WNS-affected bats. However, inflammation can also play a detrimental role in some diseases [96]. Further tissue damage and subsequent wound healing occurs in surviving bats upon emergence from hibernation [19]. These local affects of inflammation (pain and itching) as well as systemic effects are likely to play a key role in WNS pathology.
In addition to the gene expression changes that may contribute to acute inflammation locally within the epithelial tissues invaded by Pd, the systemic release of febrile cytokines such as IL-6 could affect the signals that control hibernation arousal. However, an exogenous pyrogen, lipopolysaccharide, is not able to provoke arousals in hibernating golden-mantled ground squirrels [97], so it may be unlikely that inflammation or febrile cytokines can directly trigger arousal in WNS-affected bats. Intracerebroventricular injection of prostaglandin E2 in golden-mantled ground squirrels induces arousal from torpor and a febrile response during an extended periodic arousal [97]. Our observation of increased expression of the enzyme that generates prostaglandin H2 may provide a mechanism that explains the shortened torpor bouts in WNS-affected bats, if it can be shown that this enzyme is active in the tissue and produces enough prostaglandin H2 to act systemically.
In addition to the changes in expression of genes involved in immune responses and wound healing, we also found significant changes in metabolic genes. We found evidence of gene expression changes consistent with increased fat metabolism, including changes in transcripts for apolipoproteins, lipid transport proteins, protein metabolism, and carbohydrate metabolism. Of particular interest, we found increases in the expression of hydroxycarboxylic acid receptors 2 and 3 that are known to mediate adiponectin secretion [98]. This suggests that infection with Pd may directly trigger changes in lipid and carbohydrate metabolism that contribute to WNS pathology. These changes stand in contrast to the changes that have been seen in the brain transcriptome of hibernating horseshoe bats (Rhinolophus ferrumequinum) [99] and the brain proteome of hibernating Rickett’s big-footed bats (Myotis pilosus) [100], which show decreased fat metabolism during torpor. By leading to premature depletion of fat stores, these gene expression changes could contribute to WNS mortality.
The other changes in host gene expression that we observed are consistent with a multi-stage progression model of WNS [22]. We also found support for changes in genes involved in oxidative stress [23] and body fluid levels, which may contribute to WNS progression. Together, the pattern of gene expression changes that we find in little brown myotis with WNS suggests that a combination of maladaptive responses may contribute to mortality. However, the number of upregulated genes involved in the acute inflammatory response suggests that excessive inflammation may also be a factor contributing to pathology even prior to emergence from hibernation when it is suspected to contribute to wing damage [101].
The changes in host transcript levels that we have found are presumably caused by physiological responses of the host to infection. However, caution must be used when extending these transcriptional responses to functional mechanisms because the current study does not measure protein or metabolite levels directly. Future studies will be necessary to determine which of the gene expression changes observed affect which host response mechanisms.
The little brown myotis chosen for the WNS-affected samples were exhibiting WNS pathology and appeared unlikely to survive at the time of sample collection. For this reason, it is presently uncertain which of the gene expression changes that we have observed are contributing to protection and which are pathological. Another factor that likely contributes to the variation in gene expression that we observed among the samples collected from free-ranging bats is the time since the most recent arousal from torpor. Prior to collection of each wing tissue sample, bats were artificially aroused for 30 to 120 minutes. This period of arousal is similar in duration to the natural arousals during hibernation for little brown myotis [14], and presumably of sufficient duration for some innate immune responses to occur and for transcript levels to be altered. One reason for this procedure was to avoid disparities between the elapsed time from the most recent arousal bout until tissue collection. For the WNS-affected bats we could not determine when the most recent natural arousal would have occurred, but it would have likely been more recently than in unaffected animals, as affected animals arouse from torpor more frequently [14]. In the current study we cannot resolve whether the changes in gene expression that we observed occurred during the most recent arousal, during previous periodic arousals, or during torpor. Future studies will be needed to determine which of the changes in gene expression that we observed during WNS in bats in the wild also vary in controlled captive hibernation conditions when prior arousal patterns are known. Further studies are also needed to compare the physiological responses in bats exhibiting WNS morbidity to responses in less susceptible bats, such as European species, North American species that are less susceptible like the big brown bat [11], and the remnant populations of little brown myotis that appear to have developed tolerance or resistance to Pd [1]. Such studies should point to a path forward for bats in North America to persist in a landscape where Pd is endemic.
Little brown myotis mount a host response to Pd infection during hibernation. Which components of this response are protective or contribute to WNS pathology remains to be resolved. The innate immune response we have observed would be expected to promote a Th17-directed adaptive immune response that could clear the infection. However, the energetic constraints of hibernation may prevent little brown myotis from execution of the Th17- and neutrophil-mediated phases of the immune response. This may lead to excessive inflammatory responses, either during hibernation or upon emergence. The changes in host gene expression that we observed demonstrate that during Pd infection, little brown myotis also alter other defense responses, metabolic pathways, and transcription. Numerous Pd genes that may contribute to virulence were identified and these represent potential pathogen responses to host defense. Hibernation does not prevent a host response to infection and a better understanding of the differences between host and pathogen responses in bats susceptible to WNS and those resistant may lead to ways for increasing survival.
This study was carried out on bats from non-endangered species in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All methods were approved by the Institutional Animal Care and Use Committee at Bucknell University (protocol DMR-016). Animals were humanely euthanized by isoflurane anesthesia overdose followed by decapitation. In Illinois, animal collection was conducted by state wildlife officials and a numbered permit was not required. Scientific collector’s permits were obtained in Michigan (SC1448), Minnesota (201174), and Kentucky (SC1411147).
We collected hibernating little brown myotis from cave or mine walls at the locations listed in Table 1. Bats collected from all locations are expected to be from the same genetic population of eastern little brown myotis [102]. For bats unaffected by WNS, little brown myotis were first swabbed on the left forearm for quantitative PCR analysis. After measurements were taken, bats were individually placed in cloth bags and hung in constant temperature thermoelectric coolers (Koolatron PC-3) maintained at ~7°C. Water-saturated sponges were placed in the bottom of each cooler to maintain humidity during transportation to Bucknell University. Bats were housed for 13 weeks in a Percival (model I36VLC8) environmental chamber with conditions set to 4°C and 95% relative humidity. Bats were provided water throughout hibernation. Bats were aroused from hibernation for 30–120 minutes prior to euthanasia. For WNS-affected bats, little brown myotis were collected in the field, measured, swabbed for quantitative PCR, and humanely euthanized after being aroused from hibernation for 60–120 minutes. Scaled mass index (SMI) was calculated using the formula (mass(in g))*(38.01/(forearm length(in mm))^1.406 [103]. Wing tissue was placed in formalin for histology and placed in RNAlater (Sigma-Aldrich) for gene expression analysis. RNAlater samples were stored at ambient temperature for up to 24 hours before long-term storage at -80°C. RNA was purified from 50 mg of wing tissue using a QIAGEN RNeasy Mini Kit. All samples used for RNA sequencing had RNA integrity values greater than 7.0 using an Agilent Bioanalyzer.
Wing skin tissue was removed from the bones of the arm and digits and rolled onto 2 cm paraffin wax logs. The logs were then fixed in 10% neutral buffered formalin for at least 24 hours. Each log was cut into 3 pieces that were processed into paraffin blocks overnight in a Tissue-Tek VIP processor (Sakura Finetek). The pieces were embedded in paraffin blocks, sectioned at 3 microns, and stained with periodic acid Schiff with a hematoxylin counterstain [8]. WNS lesions (Table 1; WNS) were identified as cupping erosions with fungal hyphae and conida present. Inflammatory foci (Table 1; Infl) were identified as clusters of infiltrating neutrophils and were not associated with the asymmetrical curved conidia of Pd.
To determine presence or absence of Pd on bats unaffected by WNS, each swab was tested twice by quantitative PCR [64] by Jeffrey T. Foster at University of New Hampshire. A cycle-threshold less than 40 was used as a positive result. One of the 5 unaffected bats had one positive and one negative test (Table 1), but histology (Table 1) and subsequent RNA sequencing determined this to most likely be a false positive (S2 Table; p = 2.2x10-6). For bats affected by WNS, we performed quantitative PCR to measure the Pd load, in genomic equivalents normalized to swabs spiked with 10 000 Pd conidia, that were detected on each bat [15].
The Genome Sequencing and Analysis Facility at the University of Texas at Austin performed all library preparation and quality control procedures. Directional RNA libraries were prepared with poly-A mRNA enrichment, dUTP/UDG strand-specific labeling, fragmentation, and 200 base pair size selection. RNA-Seq was performed in two lanes of an Illumina HiSeq 2500 with 101 base pair length reads obtained.
The paired reads from all samples were preprocessed by removing adapters and using trimmomatic PE [104] with settings of Illumina clip:2:30:10, seed mismatches:2, palindrome threshold:30, clip threshold:10, leading:5, trailing:5, minlength:36. The remaining paired reads were then combined and Trinity (v2.0.4) was used in strand-specific mode (RF) to construct a de novo assembly [105]. K-mer in silico read normalization with maximum coverage of 50 resulted in 22 482 456 read pairs that were used for assembly out of 177 755 004 total. The assembly was then filtered to remove Pd sequences using the program Deconseq [65] with the Broad Institute Geomyces destructans genome 20631–21 used to identify pathogen sequences and with the little brown myotis genome (Myoluc2.0) used to retain host sequences. Bowtie 1.0.1 [106] was used to determine the number of reads that mapped to each transcript in the assembly.
The script align_and_estimate_abundance.pl included in the Trinity v2.0.6 distribution [105] was used to estimate expression levels for each transcript. Bowtie 1.0.1 [106] was used to map reads (including unpaired reads after quality trimming) from each sample onto the assembly. RSEM v1.2.20 [107] was used to apply an expectation maximization algorithm to predict gene expression counts for each transcript. Expression levels are presented after trimmed mean of M-values (TMM) normalization in fragments per kilobase of transcript per million mapped reads (FPKM). DESeq2 v1.8.1 [67] was used to determine the probability of differential expression for each Trinity transcript cluster that had a minimum RSEM-estimated count, before normalization, of 5 across all samples. For DESeq2 analysis, the default values for removing outliers and filtering lowly expressed transcripts were used. An alpha value of 0.05 was used instead of the default of 0.1 to decrease the number of differentially expressed genes identified. Posterior probabilities of differential expression for individual transcript isoforms were estimated using a Bayesian approach with EBSeq v1.8.0 [68]. False discovery rate [108] was used to control for multiple comparisons. NCBI BLAST v2.2.29+ [109] was used to identify the highest-ranking match for each isoform in the UniProt Swissprot database (downloaded on Sep 17, 2014) with an e-value cutoff of 1x10-5.
Hierarchical clustering of samples and genes was performed within R 3.1.2 using the hclust function with the complete linkage method. Bootstrap analysis of clustering was performed using the pvclust 1.3–2 package and 1000 replications [69]. Principal component analysis was performed using the prcomp function and visualized with the rgl 0.93.1098 package.
NCBI BLAST v2.2.29+ [109] was used with an e-value cutoff of 1x10-5 to identify homologs in the Uniprot Swissprot human protein database (downloaded on Nov 25, 2014) for transcripts significantly upregulated in WNS-affected bat wing tissue with an FDR of less than 0.1 (in order to increase the number of genes prior to subsequent analysis with higher stringency FDR). Unique Ensembl gene IDs were identified for 1144 of the 1922 upregulated transcripts and 481 of the 1356 downregulated transcripts. GOrilla [70] was used with a p value cutoff of 0.001 to identify upregulated or downregulated biological processes by comparison to the background list of 12 828 human genes identified by BLAST in the Trinity assembly. Multiple testing correction [108] was used with an FDR cutoff of 0.01. Results were visualized as a treemap with REVIGO [71].
Trinity v2.0.4 was used to generate a Pd assembly in genome-guided mode with jaccard clipping and using the Broad Institute G. destructans genome 20631–21. This assembly was used to assess pathogen gene expression in the samples from WNS-affected bats using RSEM v1.2.20 [107]. Trinotate v2 was used to annotate the Pd transcripts by using NCBI BLAST v2.2.29+ [109] and both the Swissprot and Uniref90 databases (downloaded on Sep 17, 2014).
Reads for each sample were analyzed using MG-RAST v.3.5 [110] to identify metagenomic sequences after filtering against the B. taurus genome (the taxonomically closest genome available for filtering). For assignment of organism abundance, the best hit classification was used with the M5NR database, maximum e-value cutoff of 1x10-5, minimum identity cutoff of 60%, and minimum alignment length cutoff of 15.
|
10.1371/journal.ppat.1002809 | IFNγ Inhibits the Cytosolic Replication of Shigella flexneri via the Cytoplasmic RNA Sensor RIG-I | The activation of host cells by interferon gamma (IFNγ) is essential for inhibiting the intracellular replication of most microbial pathogens. Although significant advances have been made in identifying IFNγ-dependent host factors that suppress intracellular bacteria, little is known about how IFNγ enables cells to recognize, or restrict, the growth of pathogens that replicate in the host cytoplasm. The replication of the cytosolic bacterial pathogen Shigella flexneri is significantly inhibited in IFNγ-stimulated cells, however the specific mechanisms that mediate this inhibition have remained elusive. We found that S. flexneri efficiently invades IFNγ-activated mouse embryonic fibroblasts (MEFs) and escapes from the vacuole, suggesting that IFNγ acts by blocking S. flexneri replication in the cytosol. This restriction on cytosolic growth was dependent on interferon regulatory factor 1 (IRF1), an IFNγ-inducible transcription factor capable of inducing IFNγ-mediated cell-autonomous immunity. To identify host factors that restrict S. flexneri growth, we used whole genome microarrays to identify mammalian genes whose expression in S. flexneri-infected cells is controlled by IFNγ and IRF1. Among the genes we identified was the pattern recognition receptor (PRR) retanoic acid-inducible gene I (RIG-I), a cytoplasmic sensor of foreign RNA that had not been previously known to play a role in S. flexneri infection. We found that RIG-I and its downstream signaling adaptor mitochondrial antiviral signaling protein (MAVS)—but not cytosolic Nod-like receptors (NLRs)—are critically important for IFNγ-mediated S. flexneri growth restriction. The recently described RNA polymerase III pathway, which transcribes foreign cytosolic DNA into the RIG-I ligand 5′-triphosphate RNA, appeared to be involved in this restriction. The finding that RIG-I responds to S. flexneri infection during the IFNγ response extends the range of PRRs that are capable of recognizing this bacterium. Additionally, these findings expand our understanding of how IFNγ recognizes, and ultimately restricts, bacterial pathogens within host cells.
| Shigella flexneri, the major cause of bacillary dysentery worldwide, invades and replicates within the cytoplasm of intestinal epithelial cells, where it disseminates to neighboring cells and ultimately increases the likelihood of transmission to uninfected hosts. A hallmark of the mammalian immune system is its ability to inhibit the growth of such intracellular pathogens by upregulating intracellular resistance mechanisms in response to the cytokine IFNγ. We found that in non-myeloid host cells stimulated with IFNγ S. flexneri remains able to invade the cells efficiently and gain access to the host cytoplasm. Once in the cytoplasm of IFγ-activated cells, the RIG-I/ MAVS immunosurveillance pathway is activated, enabling the stimulated host cells to inhibit S. flexneri replication. Interestingly, RIG-I only played a minor role in the cellular response to this pathogen in the absence of IFNγ, suggesting that the IFNγ response ensures the recognition of the infection through an immunosurveillance pathway that is otherwise dispensable for controlling S. flexneri growth. Together, these findings implicate the RIG-I pathway as a crucial component in the cellular response to this devastating bacterial pathogen.
| Shigella flexneri is a Gram-negative bacterial pathogen that causes bacillary dysentery, resulting in significant morbidity and mortality worldwide. Following ingestion, S. flexneri translocate through the colonic epithelial cell barrier, where they infect resident macrophages and rapidly induce caspase-1-dependent pyroptotic cell death in these cells [1], [2], [3]. After escaping from the dying macrophages, S. flexneri invade nearby colonic epithelial cells using a Type III secretion system (TTSS) and become temporarily enclosed within a membrane-bound vacuole. The bacteria rapidly escape from the vacuole using a poorly defined mechanism and enter the host cytoplasm, where they engage in both intra- and inter- cellular motility by inducing local actin polymerization at one pole of the bacterium [4]. Invasion, vacuole escape, and intercellular spreading augment the dissemination of S. flexneri throughout the epithelium. Simultaneously, however, these virulence mechanisms also inadvertently allow greater recognition of the bacterium by the host through various intracellular immunosurveillance pathways. The stimulation of these immunosurveillance pathways ultimately leads to the induction of a robust proinflammatory response and the eventual resolution of infection [5], [6], [7], [8].
A critical mediator of the proinflammatory response to S. flexneri is the cytokine IFNγ (also known as Type II IFN), which acts on a wide variety of cells types to regulate the expression of over 2,000 genes [9]. In the past decade, significant progress has been made in identifying and characterizing the downstream IFNγ-inducible intracellular resistance mechanisms that coordinate the killing or growth inhibition of microbial pathogens. Some of these mechanisms include the targeting of bactericidal reactive oxygen species (ROS) to pathogen containing vacuoles (PCVs), the direct vesiculation and destruction of PCVs [10], [11], [12], [13], and the induction of antimicrobial autophagy [14]. Although advances have been made in identifying IFNγ-inducible intracellular resistance mechanisms, the mechanisms responsible for restricting many cytosolic bacterial pathogens have largely remained elusive, presumably a result of redundancy among effector mechanisms. One study found that Francisella tularensis escapes to the cytosol of IFNγ-activated primary macrophages but is subsequently restricted for cytosolic growth by an unknown mechanism, independently of reactive nitrogen species (RNS) or ROS [15]. However, a parallel study found that inhibition of RNS was able to block F. tularensis killing but did not restore intracellular replication [16]. In contrast, the cytosolic pathogen Listeria monocytogenes fails to escape the phagosome and is subsequently killed in IFNγ-activated peritoneal macrophages due to the functional disruption of the hemolysin listeriolysin O (LLO) by RNS and ROS [17], [18].
The importance of IFNγ in host defense during S. flexneri infection was demonstrated by Way, et al., who showed that the lethal dose of S. flexneri is 5 logs greater in immunocompetent mice compared to IFNγ−/− mice [9]. Furthermore, immunocompetent mice challenged with 105 CFU of S. flexneri were able to clear the infection by 5 days post infection, while IFNγ−/− mice were unable to inhibit S. flexneri replication and eventually succumbed to the infection. The effect of IFNγ on cell autonomous resistance to S. flexneri has also been demonstrated. Primary mouse macrophages or rat L2 fibroblasts pre-treated with IFNγ prior to infection significantly inhibit S. flexneri growth compared to untreated cells [9]. Although IFNγ is a critical mediator of innate immunity against this bacterium, the IFNγ-inducible host factors mediating cell autonomous resistance against this bacterium are completely unknown. Moreover, unlike F. tularensis and L. monocytogenes, no data are available on the specific step of S. flexneri pathogenesis that is blocked by IFNγ in macrophages or in other cell types also naturally infected by this bacterium, such as non-myeloid epithelial cells.
Here we sought to identify which step of S. flexneri intracellular infection is inhibited in IFNγ-activated non-myeloid cells and to begin to define the cellular mechanism(s) and pathways that are enabled by IFNγ to recognize or restrict intracellular S. flexneri infection. We found that S. flexneri efficiently invades and escapes from the vacuole of IFNγ-activated MEFs and are inhibited at the step of cytosolic replication. Furthermore, we found that the detection of S. flexneri infection by the cytoplasmic RNA sensor RIG-I was required for the inhibition of S. flexneri cytosolic growth by IFNγ. Interestingly, S. flexneri genomic DNA and RNA were sufficient to induce RIG-I dependent immune responses. Additionally, chemical inhibition of host RNA polymerase III partially blocked the ability of IFNγ to inhibit S. flexneri growth, suggesting that S. flexneri DNA is a stimulus of the IFNγ-dependent immune response against this bacterium. Collectively, these findings implicate the RIG-I/MAVS signaling pathway as a crucial component of cell autonomous IFNγ-mediated restriction of cytosolic bacterial pathogens.
The replication of S. flexneri within the colonic epithelium is an essential determinant of this bacterium's pathogenesis. Previously it was demonstrated that IFNγ inhibits the growth of this bacterium in both mouse macrophages and rat L2 fibroblasts [9]. To determine how IFNγ might inhibit the intracellular growth of this bacterium in the epithelium, we examined S. flexneri growth in mouse primary MEFs, as a model for non-myeloid epithelial cells. While unstimulated MEFs were highly permissive for S. flexneri replication over a 15 hour infection, pre-stimulation of MEFs with IFNγ prior to infection drastically inhibited bacterial growth by 15 hours post infection (hpi) (Fig. 1A). Interestingly, S. flexneri grew similarly well in unstimulated and IFNγ-stimulated cells for at least 5 hpi, indicating that the antimicrobial mechanisms that block S. flexneri growth are not immediately felt by the bacteria. The addition of IFNγ to the cell culture media only at the time of the infection had no effect on S. flexneri growth (data not shown), suggesting that IFNγ-mediated priming of the cell prior to infection is necessary for the restriction.
Invasion of host cells and vacuolar escape by S. flexneri are essential for the evasion of extracellular and intracellular immune mechanisms, respectively. An intriguing hypothesis to explain the ability of IFNγ to inhibit intracellular S. flexneri is that this cytokine affects a specific stage of the bacterium's intracellular pathogenic cycle to prevent otherwise efficient escape from antimicrobial mechanisms. While a non-invasive virulence plasmid-cured strain of S. flexneri was over 100-fold less invasive than the wild-type (WT) strain, WT S. flexneri invaded unstimulated and IFNγ-stimulated cells with similar efficiencies (Fig. 1B). Many vacuolar– and even cytosolic– pathogens are killed in IFNγ-activated cells following invasion by mechanisms targeting nascent pathogen containing vacuoles formed during microbial invasion or uptake [18]. Therefore, one possibility was that IFNγ-induced mechanisms either destroy S. flexneri in the vacuole prior to vacuole escape or functionally disrupt the action of bacterial effectors necessary for vacuole escape. To assess the ability of S. flexneri to escape from the vacuole of IFNγ-activated MEFs, unstimulated and stimulated cells were infected for 30 minutes and subsequently treated with chloroquine, which concentrates in phagosomes of host cells at bactericidal levels. As expected, the plasmid-cured strain (which is deficient for vacuole escape) was killed in the presence of chloroquine, confirming that chloroquine effectively killed bacteria trapped in the vacuole. Interestingly, similar numbers of bacteria were recovered from unstimulated and stimulated MEFs both in the absence and presence of chloroquine, suggesting that S. flexneri efficiently escapes to the cytoplasm of IFNγ-treated cells (Fig. 1C). Collectively these findings demonstrate that S. flexneri enters into cells and accesses the cytoplasm of IFNγ-activated MEFs, demonstrating that inhibition occurs after the organisms reach the cytoplasm.
Once in the host cytoplasm, S. flexneri becomes motile and spreads to adjacent cells through the activity of IcsA, a bacterial cell surface-associated protein required for the polymerization of host actin in the cytoplasm [19], [20]. Since access to the cell cytoplasm is a prerequisite for actin tail formation, we examined the ability of S. flexneri to form actin tails at 2 hpi to confirm that S. flexneri reaches the cytoplasm of IFNγ-activated cells. S. flexneri formed actin tails in both untreated and IFNγ-treated cells with comparable frequency (30% and 26%, respectively) (Fig. 1D), confirming their presence within the cytoplasm. Although IFNγ did not inhibit the intracellular motility of S. flexneri at early time points, we hypothesized that a S. flexneri mutant that was unable to move might be more easily targeted by potential IFNγ-induced mechanisms and therefore be more susceptible to IFNγ-mediated killing. To test this hypothesis, we examined the survival of the non-motile S. flexneri icsA mutant, which is fully invasive but deficient for actin tail polymerization and motility [21]. We found that the icsA strain was more restricted for growth compared to the WT strain in unstimulated MEFs at 15 hpi, suggesting that non-motile S. flexneri are more efficiently targeted by IFNγ-independent mechanisms (Fig. 1E). Additionally, like the WT strain, ΔicsA was significantly inhibited by IFNγ stimulation. However, when these data were normalized against the observed IFNγ-independent killing (CFU recovered from IFNγ-stimulated cells/CFU recovered from unstimulated cells), icsA was not more susceptible to IFNγ-mediated killing compared to the WT strain (8% icsA versus 2% WT recovered from IFNγ-treated cells over untreated cells). This suggests that the mechanisms that inhibit S. flexneri replication during the IFNγ response are not specifically targeted to non-motile bacteria. The finding that icsA is significantly inhibited by IFNγ also indicates that inhibition of S. flexneri occurs intracellularly, by a mechanism that does not depend on the escape or spreading of S. flexneri to the extracellular space or to other cells. Collectively, these experiments demonstrate that, in non-myeloid cells, S. flexneri growth is inhibited by IFNγ following bacterial invasion and vacuolar escape, at the stage of cytosolic replication.
Although previous reports had established the ability of IFNγ to inhibit S. flexneri growth [9], the cellular mechanisms responsible for this resistance remained completely undefined. One major component downstream of IFNγ signaling that is often required for microbial inhibition by IFNγ is the transcription factor interferon regulatory factor 1 (IRF1), a member of the IRF family of transcription factors, which play broad roles in immunity, oncogenesis, and apoptosis [22], [23], [24]. IFNγ signaling induces the direct transcriptional upregulation of IRF1 and other genes containing gamma-activated-site (GAS) elements in their promoters [25]. IRF1 then translocates to the host cell nucleus, where it binds to interferon stimulated response elements (ISREs) of IFN-stimulated genes (ISGs) and induces a second wave of IFNγ-dependent gene transcription. Although IFNγ-dependent pathogen restriction can occur independently of IRF1, this second wave of transcription induced by IRF1 is often required for IFNγ-mediated growth restriction. To begin to define the cellular pathways and/or gene products that inhibit S. flexneri replication in MEFs, we first tested whether IRF1 is induced by IFNγ during infection and whether IRF1 contributes to IFNγ-mediated restriction of this bacterium. We found that IRF1 gene expression was highly induced by IFNγ in uninfected cells, as reported previously (Fig. 2A). Interestingly, S. flexneri partially inhibited the ability of IFNγ to upregulate IRF1, even though S. flexneri alone induced the upregulation of IRF1 10-fold compared to unstimulated uninfected cells. We next compared S. flexneri growth in WT and Irf1−/− MEFs to determine whether IRF1 was required for IFNγ-mediated growth restriction of S. flexneri. Although an IFNγ-independent effect of IRF1 on S. flexneri replication was not observed, IRF1 significantly contributed to the restriction of S. flexneri growth in IFNγ-activated cells by 15 hpi (Fig.2B), confirming the role of IRF1 in the innate immune response to this pathogen.
Since IRF1 is required for IFNγ-mediated restriction of S. flexneri, we hypothesized that an effector mechanism downstream of IRF1-dependent transcription was ultimately required for the inhibition of S. flexneri growth. Therefore, we used transcriptional profiling to identify genes that are regulated by IFNγ and dependent on IRF1. Although the identification of IRF1 target genes by microarray analysis has previously been conducted in mouse peritoneal macrophages [26], our goal was to identify genes induced by IFNγ in MEFs specifically during S. flexneri infection. We reasoned that IFNγ-inducible gene products that inhibit S. flexneri growth might require the cooperation of S. flexneri-specific pathogen-associated molecular pattern (PAMP)-mediated signaling in addition to IFNγ for their induction. WT and IRF1−/− MEFs were stimulated with IFNγ or left unstimulated, and all of the cells were subsequently infected with WT S. flexneri for 6 hours before total RNA was harvested. Affymetrix mouse whole genome microarrays, representing approximately 20,000 genes, were used to identify IFNγ-dependent, IRF1 target genes. Analysis of the data from unstimulated and IFNγ-stimulated WT MEFs revealed that 365 genes were induced and 100 genes were weakly repressed more than 2-fold by IFNγ during S. flexneri infection (Fig. 3A). IFNγ-altered genes included many well-described IFNγ-dependent genes, such as Stat1, chemokines including Cxcl16 and Cxcl9, anti-viral genes Oasl1, Mx1, and Rsad2, members of the GBP family Gbp2, -3, and -6, as well as several previously uncharacterized genes (). To next identify IFNγ-inducible genes that were dependent on IRF1, we analyzed expression profiles of IFNγ-regulated genes (those identified as altered in Fig. 3A) from IFNγ-stimulated WT cells and IFNγ-stimulated IRF1−/− cells (Fig. 3B). We found that 174, or almost half, of the IFNγ-dependent genes were induced and 17 were repressed by IRF1 (Table S2). This is similar to what was found in peritoneal macrophages in which 387/1,009 IFNγ-induced genes were dependent on IRF1 [26]. The reduction in absolute number of IFNγ-dependent and IFNγ- IRF1-dependent genes identified in our microarray experiments is consistent with previous findings demonstrating that the IFNγ-dependent response in macrophages is more robust than in MEFs [27]. Microarray data from these experiments are available in the ArrayExpress database (www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-713.
Ultimately, we were interested in identifying IFNγ-induced, IRF1-dependent genes capable of blocking S. flexneri growth. As we began to think about how to prioritize testing genes identified from the microarray analysis, we first considered our finding that S. flexneri growth was not significantly inhibited by IFNγ until at least 5 hpi (Fig. 1A). We hypothesized that this delay in growth restriction might correlate with the amount of time that it would take for full transcriptional or post-transcriptional activation of important antimicrobial genes that require both IFNγ and a signal transmitted to the cell following infection. In support of this hypothesis, we found that blocking host protein synthesis with cycloheximide (CHX), a specific inhibitor of eukaryotic translational elongation, 1 hour prior to the infection (but, importantly, after IFNγ stimulation) blocked the ability of IFNγ to inhibit S. flexneri growth in a dose dependent manner (Fig. 4A). Although continued IFNγ-mediated gene induction would also be blocked following CHX treatment, the robust expression of IFNγ-dependent, infection-independent genes would have been strongly upregulated prior to treatment with CHX. These data support the hypothesis that IFNγ-dependent restriction absolutely requires a transcriptional event after infection.
PRRs, which induce changes in gene transcription following the detection of conserved microbial products, would be prime candidates for linking pathogen detection to IFNγ-induced antimicrobial restriction. The Toll-like receptors (TLRs) form one class of PRRs that are expressed at the cell surface or inside of endocytic vesicles and recognize microbial components derived from the extracellular space. Although TLRs are crucial for the detection of most pathogens, the detection of microbial products in the host cytosol requires the action of NLRs and RIG-I-like receptors (RLRs). Cytoplasmic nucleotide-binding oligomerization domain-containing protein 1 (Nod1) and Nod 2 of the NLR family recognize specific structures within peptidoglycan, leading to the recruitment of the adaptor molecule receptor-interacting serine/threonine protein kinase 2 (RIP2) and subsequent RIP2-dependent MAPK and NF-κB activation [28], or RIP2-independent recruitment of autophagsomes at sites of bacterial invasion [29]. The cytoplasmic RLRs RIG-I and melanoma differentiation-associated antigen 5 (MDA5) belong to the phylogenetically conserved DExD/H box family of RNA helicases that recognize various RNA species in the host cytoplasm. Ligand recognition by RIG-I leads to conformational changes that facilitates its association with the downstream signaling adaptor MAVS (also known as VISA, Cardif, IPS-1) at mitochondria and peroxisomes [30], [31]. This interaction results in both the activation of NF-κB and the phosphorylation of IRF-3, leading to the induction of intracellular antimicrobial gene expression and the secretion of type I IFNs, such as IFNβ. To identify a potential PRR that might be required for IFNγ-mediated restriction of S. flexneri, we examined the data from our gene expression experiments for PRRs and/or associated signaling adaptor molecules that were identified as IFNγ-dependent, IRF1-dependent genes. Genes encoding several PRRs were highly induced by IFNγ during S. flexneri infection and were dependent on IRF1, including the TLRs TLR2 and TLR3 and the RLRs RIG-I (encoded by Ddx58) and MDA5 (encoded by Ifih1) (Fig. 4B). Since the IFNγ-mediated inhibition of S. flexneri occurs in the host cytoplasm, we first focused on the cytosolic RLRs to determine if these molecules play a role in inhibiting S. flexneri growth. Additionally, RLRs were interesting because they had not previously been implicated during S. flexneri infection. Consistent with our microarray data, it has previously been reported that RIG-I can be induced by IFNγ and is a target gene of IRF1, due to a single IRF1 binding site in its proximal promoter [32], [33], [34]. To determine if RIG-I plays a role in restricting S. flexneri during the IFNγ response, we analyzed bacterial growth in WT and Ddx58−/− (referred to as Rig-I−/− for clarity) MEFs. Surprisingly, we found that RIG-I was critical for the ability of IFNγ to inhibit S. flexneri growth at 15 hpi (Fig. 4C). Interestingly, RIG-I was also important for IFNγ-independent inhibition of this bacterium at 5 hpi, although it was completely dispensable by 15 hpi. To determine if we could observe a requirement for RIG-I in IFNγ-mediated restriction of a different pathogen that is both restricted by IFNγ and activates the RIG-I pathway, we examined the growth of L. pneumophila in IFNγ-stimulated MEFs but failed to observe an IFNγ-dependent effect of RIG-I (Fig. S1).
Although S. flexneri has been shown to activate members of both the TLR and NLR families [1], [5], [6], [35], a role for RLRs during S. flexneri infection had not previously been demonstrated. Additionally, the requirement for RIG-I in the cell autonomous inhibition of bacterial growth downstream of IFNγ signaling had not been previously described. To determine whether RIG-I-dependent IFNγ-mediated restriction of S. flexneri occurs via the canonical RIG-I signaling pathway, we next examined S. flexneri growth in WT and Mavs−/− MEFs. MAVS, the downstream signaling adaptor for RIG-I, was critically important for IFNγ-dependent growth restriction of S. flexneri (Fig. 4D), demonstrating that RIG-I functions as a signaling molecule acting through its canonically described pathway during S. flexneri infection, and not as a MAVS-independent effector of bacterial growth. Interestingly, the requirement for RIG-I and MAVS for S. flexneri growth inhibition in non-myeloid cells did not extend to primary macrophages, in which IFNγ-dependent killing appeared to occur independently of both RIG-I and MAVS (Fig. S2).
To determine whether other well-described PRR pathways might similarly play a role in blocking S. flexneri replication downstream of IFNγ, we tested the role of the NLR signaling adaptor RIP2, as well as Nod1, which is known to inhibit S. flexneri growth independently of RIP2 [29]. Nod1−/− and Rip2−/− MEFs were more permissive for S. flexneri growth compared to WT cells, both in the absence and presence of IFNγ (Fig. 4E). However, in contrast to RIG-I, these proteins were found to be completely dispensable for IFNγ-dependent restriction of this bacterium after normalizing for the IFNγ-independent effects (Fig. 4F). In fact, RIP2−/− MEFs were slightly more efficient than WT MEFs in IFNγ-induced restriction of S. flexneri growth (0.2% versus 0.6% CFU recovered in IFNγ-stimulated/unstimulated cells, respectively). In addition to cell-autonomous growth restriction, stimulation of some NLR family members can mediate the induction of the caspase-1 inflammasome in response to microbial infection, leading to the secretion of proinflammatory cytokines, cell death, and the restriction of bacterial replication. However, we found that IFNγ-induced restriction of S. flexneri occurred completely independently of caspase-1, suggesting that NLR-mediated inflammasome induction was also not required (data not shown). In contrast to Nod1 and RIP2, the TLR signaling adaptor MyD88 contributed to, but was not required for, the ability of IFNγ to block S. flexneri replication (Fig. 4G). Unlike the finding for RIG-I, this result was partially expected, since MyD88 has been shown to be important for TLR-independent IFNγ signaling [36] and for the translocation of IRF1 to the nucleus, at least in myeloid dendritic cells [37]. From our findings, it is unclear whether the role of MyD88 here is solely attributable to one of these previously described functions of MyD88 in IFNγ-mediated signaling or to an alternate mechanism, such as an adaptor of TLR signaling.
PRR-mediated detection of microbial pathogens, including S. flexneri, often results in the transcriptional induction and subsequent secretion of type I IFNs, such as IFNβ, by the host cell [38]. Therefore, the analysis of type I IFN production can be used to assess the cellular immune response to a microbial challenge. To further investigate the role of RIG-I in the cellular immune response to S. flexneri, we analyzed IFNβ production from uninfected and S. flexneri-infected cells under IFNγ-stimulating conditions. To quantify the bioactivity of IFNβ secreted from infected cells, supernatants from the cells were collected and added to cells that harbor an IFNβ-dependent luciferase reporter (L929-ISRE cells) as described previously [39]. While supernatants from quiescent uninfected WT and Rig-I−/− cells activated the reporter cells with a comparably low efficiency, supernatants from cells transfected with the known RIG-I ligand low molecular weight (LMW) poly (I∶C),a synthetic analog of dsRNA, significantly induced the activation of the reporter cells, and this effect was partially dependent on RIG-I (Fig. 5A). Supernatants from unstimulated cells infected with S. flexneri for 8 hours also activated the reporter cells over uninfected cells, but this activation was not dependent on RIG-I. Interestingly, S. flexneri-mediated induction of IFNβ secretion was dependent on RIG-I if the MEFs were stimulated with IFNγ prior to the infection. In contrast to WT infection, the IFNβ response to virulence plasmid-cured S. flexneri was not dependent on RIG-I, even in the presence of IFNγ (Fig. 5A), suggesting that either access to the cell cytoplasm (where RIG-I is located) or another activity of the Type III secretion system is required for recognition by RIG-I. Collectively, these findings suggest that in unstimulated MEFs, PRRs other than RIG-I dominate the innate immune response to S. flexneri, whereas in the presence of IFNγ, RIG-I emerges as an important player in the cellular immune response to this bacterium.
To further characterize the RIG-I-dependent immune response to S. flexneri, we also examined RIG-I expression in uninfected and infected cells under stimulating and non-stimulating conditions. Similar to previous reports on other cell types [32], [33], we found that RIG-I is induced 2.5 fold by IFNγ in uninfected MEFs (Fig. 5B). S. flexneri infection alone did not induce significant RIG-I expression by 5 hpi, but S. flexneri-induced RIG-I expression was apparent by 10 hpi. Despite the lack of induction by S. flexneri alone at 5 hpi, IFNγ-induced RIG-I expression was significantly enhanced by the presence of S. flexneri at both 5 and 10 hpi. These findings demonstrate that IFNγ and S. flexneri synergistically induce RIG-I expression in MEFs, potentially facilitating S. flexneri recognition during the IFNγ response.
RIG-I contains two N-terminal caspase activation and recruitment domains (CARDs), a DExD/h helicase domain, and a C-terminal repressor domain. In the absence of an activating ligand, the repressor domain maintains RIG-I in an auto-inhibited state in the cell cytoplasm [40]. Therefore, RIG-I signaling does not occur until an activating RNA ligand binds to RIG-I and induces a conformational change that exposes its CARD domain and allows for CARD-CARD interactions with its downstream signaling adaptor MAVS. Due to this mechanism of autoregulation, overexpression of full-length RIG-I is insufficient to induce downstream RIG-I signaling [41]. Therefore, although we found that RIG-I was upregulated in IFNγ-activated cells, the upregulation of RIG-I without an activating ligand is unlikely to be sufficient to induce downstream RIG-I signaling and cannot fully explain the ability of RIG-I to restrict S. flexneri growth. Rather, the finding that RIG-I can inhibit S. flexneri growth in the presence of IFNγ suggests that S. flexneri infection provides or generates a ligand capable of activating RIG-I. Since we found that S. flexneri activates RIG-I in the presence of IFNγ, we began to explore the possibility that the ability of S. flexneri to activate RIG-I depends upon the presence of S. flexneri nucleic acids in the host cytoplasm of activated cells. Although there is debate over the exact nature of the ligand(s) recognized by RIG-I (reviewed in [42]), this molecule is largely thought to recognize cytoplasmic short, double-stranded RNA containing a 5′-triphosphate group (thereby avoiding the recognition of host mRNA, which contains 5′ modifications such as capping). Although RNA species are currently thought to be the only ligands directly recognized by RIG-I, it has been reported that foreign dsDNA introduced into the host cytoplasm can activate RIG-I signaling [43], [44]. More recent reports expanded on these findings to show that RIG-I indirectly recognizes dsDNA introduced into the host cell cytoplasm through the RNA polymerase III-dependent transcription of this DNA into an RNA intermediate that can be recognized by RIG-I [45], [46]. The RNA polymerase III pathway has been shown to be an important component in the recognition of viruses, L. monocytogenes [47], and possibly L. pneumophila, despite some controversy [45], [48]. To directly test whether S. flexneri RNA or DNA is sufficient to induce a RIG-I-dependent cellular response, we extracted S. flexneri genomic DNA and total RNA from exponential-phase cultures and then transfected these preparations into WT and Rig-I−/− cells. Importantly, the DNA and RNA preparations were treated with RNase and RNase-free DNase I, respectively, prior to the transfection to eliminate contaminating nucleic acids. Eight hours post-transfection, supernatants from these cells were collected and added to the L929-ISRE IFNβ-dependent luciferase reporter cells. Supernatants from cells transfected with either S. flexneri DNA or RNA activated WT cells over basal levels (Fig. 6). Interestingly, the observed activation was dependent on RIG-I in each case, suggesting that both nucleic acid species have the potential to activate a RIG-I-dependent response.
Our results suggested that S. flexneri DNA has the potential to activate the RIG-I pathway. Currently, the only pathway known to detect bacterial DNA through RIG-I is via the transcription of cytosolic DNA into a 5′-ppp-containing RNA intermediate via the RNA polymerase III pathway [45], [46]. Therefore, we next tested whether host RNA polymerase III is important for IFNγ-mediated suppression of S. flexneri growth using an RNA polymerase III-specific inhibitor, ML-60218, described previously [49]. We found that pre-treatment of MEFs with ML-60218 partially blocked the ability of IFNγ to inhibit S. flexneri growth by 15 hpi (Fig. 7), potentially suggesting that RNA polymerase III contributes to the detection and subsequent inhibition of this pathogen. It should be noted that significant IFNγ-dependent killing was still observed in the presence of ML-60218, and complete reversal of IFNγ-dependent killing by the inhibitor was not apparent. Interestingly, we did not observe a significant effect of ML-60218 on IFNγ-independent growth restriction, again suggesting that IFNγ enables the functional recognition of intracellular S. flexneri by the RIG-I/RNA polymerase III pathway to inhibit S. flexneri growth.
We next wanted to elucidate the downstream mechanism by which RIG-I enables IFNγ-dependent restriction of S. flexneri. RIG-I signaling not only induces the expression of antimicrobial ISGs, but also induces the secretion of type I IFNs, which act in an autocrine manner by binding to the IFN alpha receptor (IFNAR) and sustaining an antiviral state in the host cell. Therefore, we sought to determine whether RIG-I-mediated type I IFN production during infection could explain the role of RIG-I in the inhibition of S. flexneri growth. First, we examined the ability of the type I IFN IFNβ to inhibit S. flexneri growth in MEFs. At 1 hpi, similar numbers of CFU were recovered from untreated cells and from cells stimulated with IFNγ or IFNβ for 24 hours prior to infection, demonstrating that there was no difference in the invasion of S. flexneri into cells under these conditions (Fig. 8A). By 15 hpi, IFNγ potently blocked S. flexneri replication by 100-fold, as expected. IFNβ also inhibited S. flexneri growth at 15 hpi, however this restriction was much less robust (5-fold growth inhibition), even at very high concentrations. To definitively determine whether type I IFN signaling is important for IFNγ-mediated S. flexneri growth restriction, we next examined S. flexneri growth in MEFs lacking the type I IFN receptor (Ifnar−/− MEFs), which are completely defective in the ability to respond to type I IFNs. IFNγ-stimulated Ifnar−/− MEFs were significantly more restrictive for S. flexneri replication compared to WT MEFs. These findings demonstrate that while type I IFNs are capable of inhibiting S. flexneri growth (albeit to a lesser extent compared to IFNγ), they are dispensable for IFNγ-mediated growth restriction of this bacterium. Together these findings suggest that the inhibition of S. flexneri by IFNγ occurs due a type I IFN secretion-independent mechanism following RIG-I signaling.
To further investigate the importance of the IFNγ/RIG-I/MAVS pathway in protection against S. flexneri in vivo, we analyzed S. flexneri growth in IFNγ−/−, Mavs−/−, and Mavs+/+ mice (Rig-I−/− mice are embryonic lethal and were therefore not included). Mice were infected intranasally with 3×105 S. flexneri, and the lungs were harvested at 4 hours and 1, 3, and 5 days post-infection. Mavs−/− mice and IFNγ−/− mice harbored bacterial burdens that were nearly 1 log greater than those observed in Mavs-sufficient mice by day 1 post-infection (Fig. 9). Although this difference was not statistically significant among groups, this trend towards greater bacterial burden in Mavs−/− mice on day 1 post-infection was consistently observed in 3 independent experiments. Surprisingly, while the burdens in IFNγ−/− mice continued to increase through day 5 (consistent with previous reports, [9]), bacterial burdens in Mavs−/− mice dramatically declined by day 3 by several logs, to levels below those seen in MAVS-sufficient mice. Although the role of the MAVS pathway during S. flexneri infection remains to be fully elucidated, these findings suggest that early during infection the MAVS pathway plays a small role in the inhibition of S. flexneri replication in vivo. Collectively, the findings presented here implicate the RIG-I/MAVS immunosurveillance pathway as an important component in IFNγ-mediated cell autonomous restriction of a cytosolic bacterial pathogen.
During an infection, host cells must simultaneously respond to multiple signals, including both host-derived factors and microbe-derived factors in order to exert the appropriate immune response. Here we describe the unexpected mutual requirement for both host-derived IFNγ and infection-dependent stimulation of an RNA-sensing pathway in order to mediate the inhibition of a cytosolic bacterial pathogen. For years, cytosolic RNA-sensing pathways were thought to respond only to viral pathogens. More recently, the microbes capable of stimulating RNA-sensing pathways has expanded to include the recognition of bacterial pathogens as well, including L. pneumophila by RIG-I and MDA5 [48], L. monocytogenes by the RLR laboratory of genetics and physiology 2 (LGP2) [47], and now also S. flexneri by RIG-I. We found that RIG-I is key to the recognition and subsequent elimination of S. flexneri during the IFNγ response, potentially through the recognition of both S. flexneri RNA and an RNA polymerase III-transcribed RNA intermediate derived from S. flexneri DNA. Together these findings implicate RNA-sensing pathways as critical players in the IFNγ-mediated cell autonomous restriction of a cytosolic bacterial pathogen.
Similar studies on other cytoplasmic pathogens have been performed, although these studies have largely been conducted in macrophages. In macrophages, IFNγ prevents the escape of L. monocytogenes and F. novicida to the host cytoplasm, while F. tularensis is not inhibited by IFNγ until the bacteria have reached the host cytoplasm [15], [16], [17]. Although macrophages are often the primary mediators of IFNγ-inducible killing, complete clearance of S. flexneri must result from intracellular resistance mechanisms induced at the site of S. flexneri replication, in the epithelial cell layer. Therefore, in these studies, we used non-myeloid MEFs as a model for epithelial cells to understand how IFNγ enables the host to control, and finally clear, an infection with this pathogen. We found that IFNγ inhibits the growth of S. flexneri in MEFs at the step of cytosolic replication, and not at the earlier stages of entry and vacuolar escape (Fig. 1). Taken together, these reports highlight the complexity of the IFNγ response in terms of its ability to recognize and inhibit different pathogens—even different cytosolic bacterial pathogens—through distinct mechanisms and at different locations. In the case of S. flexneri, we hypothesize that the >5 hour delay in the ability of IFNγ-activated cells to block S. flexneri replication corresponds to the recognition of the bacterium by RIG-I and subsequent induction of antimicrobial mechanisms. It is possible that during human infection this delay is sufficient for S. flexneri to begin its cycle of replication and cell-to-cell dissemination prior to being killed by IFNγ-induced mechanisms, further pushing the host-pathogen balance in favor of the bacteria. Indeed, IFNγ added to cells at the time of the infection had no effect on S. flexneri growth by 15 hpi (data not shown); it is possible that S. flexneri is capable of actively interfering with IFNγ-mediated signaling and can establish a productive infection if the cells are not stimulated prior to the infection. Alternatively, this delay could be a reflection of the time that it would take for IFNγ-dependent IRF1 induction and subsequent IRF1-, RIG-I- mediated immune mechanisms to be induced. These possibilities certainly warrant further investigation.
Synergy between IFNγ and PRRs has long been appreciated as a crucial component of innate immunity. In some cases, IFNγ effectively lowers the concentration of PAMP ligands required to affect downstream PRR-dependent gene regulation. In other cases, IFNγ priming is absolutely required for downstream PRR-dependent gene induction [50]. Conversely, PRR signaling can also enhance IFNγ-dependent gene induction. In the case of RIG-I, it has been reported that RIG-I can potentiate IFNγ-induced expression of the chemokines CXCL9–11, although the functional consequences of this finding on microbial infection has not been explored [51], [52]. We extended these findings by identifying a role for RIG-I in mediating IFNγ-dependent cell autonomous resistance to a cytosolic bacterial pathogen. The findings presented here raise two fundamentally important but distinct questions. How might RIG-I link IFNγ signaling to inhibition of S. flexneri growth? Secondly, how does IFNγ potentiate the recognition of S. flexneri by RIG-I? To first address how RIG-I links IFNγ signaling with inhibition of S. flexneri growth, we considered that RIG-I is important for the expression of important antimicrobial ISGs induced by IFNγ during S. flexneri infection. In support of this hypothesis, our preliminary data suggest that ISG expression is altered by RIG-I downstream of IFNγ during infection (data not shown). Although previous reports have demonstrated that PRRs can inhibit microbial growth independently of their canonical downstream signaling adaptors [29], the absolute requirement for MAVS in the inhibition of S. flexneri by IFNγ suggests that RIG-I functions as a signaling molecule and not through a MAVS-independent mechanism. Furthermore, our results demonstrate that the RIG-I-dependent effect occurs in a mechanism independent of type I IFN signaling. Therefore, overall we favor a model in which RIG-I modulates ISG expression or other cell-intrinsic MAVS-dependent antimicrobial mechanisms to inhibit S. flexneri growth during the IFNγ response. What remains unclear from our findings is whether RIG-I acts upstream or downstream of IRF1. An alternative, and equally valid, model to the one presented here is that the cell upregulates RIG-I upon sensing of S. flexneri infection, allowing for the activation of RIG-I and subsequent gene transcription through IRF1, which will have been previously upregulated by IFNγ.
Secondly, to address how IFNγ potentiates a RIG-I-dependent response against S. flexneri, we considered several findings. We found that RIG-I expression was induced by IFNγ but was not induced by S. flexneri in the absence of IFNγ at early timepoints. Because RIG-I failed to be upregulated by S. flexneri early during infection, it is possible that greater RIG-I expression in the presence of IFNγ accounts for the ability of RIG-I to ultimately restrict S. flexneri growth. While upregulation of RIG-I alone may not fully activate its downstream signaling cascade, the stimulation of a greater number of RIG-I molecules might be necessary to overcome a potential evasion of RIG-I signaling by S. flexneri. However, we also considered an alternative possibility, in which IFNγ-mediated, RIG-I-independent effector mechanisms induce some amount of lysis or damage to a small number of S. flexneri early during infection, releasing bacterial nucleic acids into the cytoplasm that lead to the activation of RIG-I. This model, or an alternate model requiring more than the simple upregulation of RIG-I, is more consistent with our findings, since IFNβ was a weak inhibitor of S. flexneri growth (compared to IFNγ), despite its demonstrated ability to upregulate RIG-I expression.
We found that S. flexneri RNA and genomic DNA can each induce a RIG-I dependent immune response in MEFs (Fig. 6), however the actual ligand for RIG-I during S. flexneri infection remains to be identified. Whether S. flexneri DNA or RNA actually reach the cytoplasm during infection is not known. The finding that the RNA polymerase III pathway partially contributed to the inhibitory effect of IFNγ on S. flexneri (Fig. 7) suggests that S. flexneri DNA accesses the cytoplasm and is involved in the IFNγ-mediated immune response during infection. It remains to be determined whether S. flexneri DNA reaches the cytoplasm by being shed during normal bacterial replication, bacterial cell lysis, direct translocation of DNA into the host cytoplasm, or other alternate mechanisms. The RNA polymerase III pathway appeared to only partially contribute to the inhibition of S. flexneri replication, however, supporting the idea that S. flexneri RNA also potentially activates RIG-I during infection. In support of this hypothesis, activation of RIG-I signaling by bacterial RNA has previously been demonstrated for L. pneumophila [48]. Finally, in the case of S. flexneri infection, we cannot discount the possibility that the infection induces the recognition of host-derived RIG-I ligands through the disruption of cellular processes or damage to host organelles. In fact, in response to viral infections, host nuclease RNase L can produce small RNAs from host RNA that can serve as RIG-I ligands [53].
The RNA polymerase III pathway has been shown to be activated by certain viral and AT-rich DNA [46], however its role in antibacterial immunity remains to be fully elucidated. Using the RNA polymerase III inhibitor ML-60218, Chiu, et. al reported that L. pneumophila activates type I IFN production via the RNA polymerase III/RIG-I pathway in RAW macrophages, resulting in inhibition of bacterial growth [45]. A counter report by Monroe, et al. showed that the RNA polymerase III pathway failed to affect L. pneumophila replication in bone marrow-derived macrophages (BMDMs) [48], calling into question the role of the RNA polymerase III pathway during L. pneumophila infection. More recently, it was demonstrated that type I IFN production induced by L. monocytogenes in BMDM is dependent on RNA polymerase III [47]. Here, we report that the ML-60218 inhibitor partially relieved IFNγ-dependent killing of S. flexneri in MEFs, suggesting that RNA polymerase III may play a role in this pathway. Despite this finding, future studies using alternate inhibitors of this pathway and/or additional techniques will be needed to firmly establish the role of the RNA polymerase III/RIG-I pathway in immunity against S. flexneri.
Interestingly, not all PRRs stimulated by S. flexneri exhibited IFNγ-dependent effects on S. flexneri replication. It is possible that NLR ligands are simply equally available to NLRs in both the absence and presence of IFNγ, whereas RIG-I ligands are significantly more accessible under IFNγ stimulation, as discussed above. Although MAVS was crucial for S. flexneri restriction by IFNγ, we did observe some IFNγ-dependent, RIG-I-independent effects on growth inhibition (Fig. 4C), suggesting that MDA5 or other MAVS-dependent host factors might also be involved in S. flexneri recognition during the IFNγ response. Finally, it will be interesting to investigate the role of MyD88 in the IFNγ response against S. flexneri. One possibility is that TLR2, which we identified as an IFNγ-inducible, IRF1-dependent gene, mediates the MyD88-dependent effect. However, the strict requirement for MAVS in the IFNγ response suggests an alternate role for MyD88, such as the MyD88-dependent translocation of IRF1 into the nucleus, which has been shown to occur in myeloid dendritic cells [37]. Collectively, we favor a model in which early in infection NOD-like receptors inhibit the initial growth of S. flexneri in the epithelial cell layer, but at later times (following the recruitment of IFNγ-producing NK cells and IFNγ secretion at the site of infection) RIG-I becomes a crucial component in the ability of the host to clear S. flexneri infection.
In vivo, the MAVS pathways appeared to play only a minor role in inhibiting S. flexneri growth, contributing to protection only at very early timepoints. One explanation is that IFNγ-dependent control of S. flexneri occurs in non-myeloid cells (such as epithelial cells) early during infection, while at later timepoints IFNγ-activated macrophages dominate the IFNγ-dependent response; consistent with this hypothesis, we found that neither RIG-I nor MAVS is important for IFNγ-dependent killing of S. flexneri in primary macrophages (Fig. S2). However, considering that even unactivated macrophages are non-permissive for S. flexneri replication, the importance of macrophages in the IFNγ response awaits further investigation. Equally valid is the possibility that RIG-I/MAVS-dependent effector mechanisms are activated prior to other intracellular resistance mechanisms, making this pathway important only until other pathways have been activated, at which time they become dispensable. This hypothesis is consistent with the finding that bacterial burdens in Mavs−/− mice decreased dramatically by Day 3, suggesting that other antimicrobial pathways are in place and that these pathways can compensate for MAVS in its absence.
Collectively these experiments deepen our understanding of the many pathways used by host cells to inhibit infections with cytosolic bacterial pathogens. While the discovery of RIG-I as a mediator of the antimicrobial host response downstream of IFNγ and IRF1 is exciting, ultimately it will be interesting to identify the downstream effector mechanisms that inhibit or kill S. flexneri in host cells. Studies on other cytosolic bacterial pathogens such as L. monocytogenes have found that IFNγ-mediated growth restriction depends upon the induction of RNS and ROS delivered to nascent pathogen-containing vacuoles [17], [18]. While these pathways are established resistance mechanisms in macrophages and other phagocytic cells, these pathways are thought to play a relatively minor role in IFNγ-dependent immunity in non-myeloid cells. Indeed, our preliminary results suggest that RIG-I-dependent killing of S. flexneri in IFNγ-activated MEFs occurs independently of ROS (unpublished data). Recent advances in the elucidation and discovery of IFNγ-dependent antimicrobial pathways in non-myeloid cells such as MEFs have led to the characterization of a multitude of protein families and pathways capable of exerting cell autonomous resistance against microbial pathogens, such as the p47 GTPase and p65 GTPase families (reviewed in [54], [55]). Due to the abundance of pathways induced by IFNγ, the microarray experiments described in this paper will provide a crucial starting point for unraveling the complexity of the IFNγ response against this pathogen. Which of these described, or previously undescribed, genes are important for blocking S. flexneri replication in host cells awaits further investigation.
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 Harvard's Animal Care and Use Committee.
Shigella flexneri serovar 2a WT strain 2457T [56] and WT strain 2457T transformed with p-GFPmut2 [57], virulence-plasmid cured strain BS103 [58], and MGB283 (icsA) on the 2457T background were described previously. Legionella pneumophila WT serogroup 1 strain was used [59].
WT and Irf1−/− mouse embryonic fibroblasts (MEFs) were isolated from day 12.5–14.5 embryos. Cell lines that were not generated in our lab were obtained as follows: Rig-I−/− and matched WT MEFs (L. Gherke, Harvard Medical School); Nod1−/−, Rip2−/− and matched WT MEFs (D. Philpott, University of Toronto); Myd88−/− MEFs (E. Kurt-Jones, University of Massachusetts), Mavs−/− MEFs, WT, Rig-I−/−, and Mavs−/− immortalized primary BMMs (J. Kagan, Harvard Medical School); Ifnar−/− MEFs (B. Burleigh, Harvard School of Public Health); L929-ISRE fibroblasts (B. Beutler, The Scripps Research Institute). Cells were grown in DMEM (Invitrogen) supplemented with 10% FBS, 1× non-essential amino acids, 1× sodium pyruvate, 100 µM streptomycin, and 100 U/ml penicillin.
Unless indicated otherwise, 100 U/ml recombinant mouse IFNγ (Chemicon International) was added to cells 24 hours prior to infection and was maintained throughout the infection. Cells were infected with S. flexneri by centrifuging exponential phase bacteria diluted in PBS onto semi-confluent monolayers of cells at an MOI of 1∶1 at 700×g for 10 minutes. The cells were subsequently incubated for 20 minutes at 37°C and 5% CO2, washed 3 times with PBS, and resuspended in media containing gentamicin (25 µg/ml) to kill extracellular bacteria. To assess intracellular bacterial number, the cells were then incubated for indicated amounts of time in media containing gentamicin, washed 3 times with PBS, and lysed in 0.1% sodium deoxycholate/PBS. Cell lysates were then plated on tryptic soy agar (TSA) plates, and CFU were counted after overnight incubation at 37°C. For L. pneumophila experiments, L. pneumophila were grown on charcoal yeast extract (CYE) agar for 2 days prior to infections. Heavy patch cultures were subsequently resuspended in PBS and centrifuged onto semi-confluent monolayers of cells at an MOI of 30∶1 at 700×g for 10 minutes. The cells were subsequently incubated for 50 minutes at 37°C and 5% CO2, washed 3 times with PBS, and resuspended in media containing gentamicin (25 µg/ml) to kill any extracellular bacteria. To assess intracellular bacterial load, cell monolayers were lysed in sterile water, plated on CYE agar plates, and incubated at 37°C for 48 hours prior to CFU enumeration.
B6;129-Mavstm1Zjc/J (Mavs+/−) and B6.129S7-Ifnγtm1Ts/J (IFNγ−/−) mice were obtained from Jackson Laboratories (Bar Harbor) and bred in specific pathogen-free breeding rooms at Harvard Medical School. Mavs+/− mice were bred with each other, and Mavs+/+ and Mavs−/− littermates were used in the experiments. For infections, S. flexneri was subcultured from an overnight culture to exponential phase (OD600 = 0.4–0.6), and diluted with PBS to the appropriate concentration prior to inoculation. Numbers of bacteria per inoculums were confirmed by plating serial dilutions of the inoculum. For inoculation, 6–8 week-old mice were lightly sedated with 5% isoflurane (Vedco, Inc) in oxygen and inoculated by pipetting 40 µL PBS containing 2.5×105 CFU of bacteria onto the external nares. For quantification of bacteria numbers, mice were sacrificed via CO2 inhalation and lungs were excised, homogenized in 2 mL PBS, serially diluted and plated onto TSA plates containing Congo red (0.01%). Colonies were counted after incubation at 37°C for 12–18 hours. The lower limit of detection was 20 CFU.
Cells were grown and infected on glass cover slips for indicated amounts of time. Cells were then washed, fixed in 4% paraformaldehyde for 10 min, washed in PBS, and permeabilized with 0.1% Triton-X for 10 min. Actin was visualized by staining with an Alexa Fluor 647-conjugated phalloidin (Invitrogen) according to the manufacturer's directions.
MEFs were stimulated with IFNγ for 18 hours or left unstimulated. All cells were subsequently infected at an MOI of 1∶1 for 6 hours, a time when the transcription of genes that require both IFNγ and molecular sensing of S. flexneri for their regulation would be altered. Total RNA was harvested at 6 hpi using the RNeasy kit (Qiagen) and subsequently treated with DNase I. Generation of cDNA, cRNA, biotinylation, fragmentation, and hybridization to Affymetrix mouse whole genome 430 2.0 arrays were performed at the Harvard Biopolymers Facility. The array was repeated two times using biological replicates. Data from each of the 4 samples in one array were first normalized using the MAS 5.0 algorithm using Gene Spring GX software. Next, probe sets in which 0/4 samples exhibited expression between 20% and 100% were dropped from further analysis. Out of the 45,101 probe sets represented on the arrays, 37,569 probe sets had at least one sample with a value within the cut-off threshold and were kept for analysis.
Total RNA from MEFs was harvested 6 hpi using the RNeasy kit (Qiagen) according to the manufacturer's directions. RNA samples were treated with DNase I prior to reverse transcription and amplification with SYBR Green One-Step Quantitative RT-PCR kit (Qiagen). Transcript levels were normalized to 18S rRNA. The following primer sequences were used: Irf1: F, 5′-TTAGCCCGGACACTTTCTCTGATGG-3′ and R, 5′-GTCCCCTCGAGGGCTGTCAATCTCT-3′; Rig-i: F, 5′-ATTGTCGGCGTCCACAAAG-3′ and R, 5′-GTGCATCGTTGTATTTCCGCA-3′; 18S rRNA: F, 5′-CATTCGAACGTCTGCCCTATC-3′ and R, 5′-CCTGCTGCCTTCCTTGGA-3′.
Total bacterial genomic DNA was isolated using the DNeasy kit (Qiagen) in conjunction with RNaseA treatment (Qiagen) at 100 mg/ml for 2 min. Total bacterial RNA was isolated using RNAprotect Bacterial Reagent (Qiagen) and the RNeasy kit (Qiagen); isolated RNA was subsequently treated with DNAase I at 10 U/ml for 10 minutes at 37°C and re-purified using the RNeasy kit. Transfections of isolated bacterial DNA, RNA, or low molecular weight poly (I∶C) (Invivogen) into MEFs were performed by mixing indicated nucleic acids with DMEM and Attractene (Qiagen) to a final ratio of 0.25 µg nucleic acid/µl Attractene and incubated for 20 minutes. Lipid-ligand complexes were added to cells at a quantity of 0.4 µg ligand/well of a 24 well plate. Eight hours post-transfection or post-infection with S. flexneri, supernatants were collected and overlaid onto pre-seeded L929-ISRE cells, which harbor an IFNβ-dependent luciferase reporter, for 4 hours. Luciferase expression from L929-ISRE cells was quantified using Bright Glo (Promega) according to the manufacturer's directions.
Where indicated, cycloheximide was added to cells 2 hours prior to infection at indicated concentrations and was maintained throughout the experiment. The RNA polymerase III inhibitor ML-60218 (Calbiochem) was added to cells 2 hours prior to infection at 20 µM and maintained throughout the experiment.
As indicated, a two-tailed Student's t test for paired samples or a one-way ANOVA was used to determine statistical significance. A p value<0.05 was considered statistically significant.
Microarray data are available in the ArrayExpress database (www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-713. Entrez gene ID numbers for genes mentioned in the text are as follows: Irf1: 16362; Rig-I: 230073; Mavs: 228607; Nod1: 107607; Rip2: 192656; Myd88: 17874; Ifnar: 15975.
|
10.1371/journal.pgen.1006596 | The genetic basis of resistance and matching-allele interactions of a host-parasite system: The Daphnia magna-Pasteuria ramosa model | Negative frequency-dependent selection (NFDS) is an evolutionary mechanism suggested to govern host-parasite coevolution and the maintenance of genetic diversity at host resistance loci, such as the vertebrate MHC and R-genes in plants. Matching-allele interactions of hosts and parasites that prevent the emergence of host and parasite genotypes that are universally resistant and infective are a genetic mechanism predicted to underpin NFDS. The underlying genetics of matching-allele interactions are unknown even in host-parasite systems with empirical support for coevolution by NFDS, as is the case for the planktonic crustacean Daphnia magna and the bacterial pathogen Pasteuria ramosa. We fine-map one locus associated with D. magna resistance to P. ramosa and genetically characterize two haplotypes of the Pasteuria resistance (PR-) locus using de novo genome and transcriptome sequencing. Sequence comparison of PR-locus haplotypes finds dramatic structural polymorphisms between PR-locus haplotypes including a large portion of each haplotype being composed of non-homologous sequences resulting in haplotypes differing in size by 66 kb. The high divergence of PR-locus haplotypes suggest a history of multiple, diverse and repeated instances of structural mutation events and restricted recombination. Annotation of the haplotypes reveals striking differences in gene content. In particular, a group of glycosyltransferase genes that is present in the susceptible but absent in the resistant haplotype. Moreover, in natural populations, we find that the PR-locus polymorphism is associated with variation in resistance to different P. ramosa genotypes, pointing to the PR-locus polymorphism as being responsible for the matching-allele interactions that have been previously described for this system. Our results conclusively identify a genetic basis for the matching-allele interaction observed in a coevolving host-parasite system and provide a first insight into its molecular basis.
| Negative frequency-dependent selection, whereby common genotypes are disfavored, resulting in cyclic change of gene frequencies and maintenance of genetic diversity in host and parasite populations, is one the mechanisms predicted to drive host-parasite coevolution. Specific matching-allele interactions between hosts and parasites are a mechanism predicted to underpin this mode of selection. In spite of in depth research, little is known about the genetic basis of such matching-allele interactions and few empirical examples have been described. Recent research has suggested that the Daphnia-Pasteuria host-parasite system follows a model of negative frequency-dependent selection. We map a Daphnia magna locus of resistance to Pasteuria ramosa. We use next-generation genome and transcriptome sequencing to characterize resistant and susceptible haplotypes of the resistance locus. We find large-scale structural polymorphism between resistance locus haplotypes and we find evidence that gene conversion, segment duplication and restricted homologous recombination contribute to produce the observed polymorphisms. We analyse natural populations and find that the resistance locus structural polymorphisms reproduce the matching-allele interactions predicted for the Daphnia-Pasteuria system. This work presents rare and conclusive evidence of the genetic basis of matching-allele interactions in host-parasite systems while opening research avenues to find the underlying molecular mechanisms.
| Host-parasite interactions are ubiquitous among all living organisms and are thought to represent one of the strongest contributing factors to shaping the evolution of biological organisms [1]. The antagonistic nature of host-parasite interactions leads to reciprocal selection of the antagonists on each other that can drive rapid coevolutionary change [1–3]. Hosts are expected to evolve mechanisms to reduce the likelihood of infection and to minimize the fitness costs associated with infections, while parasites are expected to evolve mechanisms to evade the hosts’ defense mechanisms. Host-parasite interactions are thought to contribute to diversification, speciation, maintenance of sexual reproduction, and maintenance of genetic diversity in natural populations [1, 4–6]. Multiple evolutionary mechanisms have been proposed to underlie host-parasite evolutionary dynamics. These include heterozygote advantage, selective sweeps, and negative frequency-dependent selection (NFDS) [2, 7–9]. NFDS, whereby common host genotypes have a selective disadvantage, can result in balancing selection and is therefore proposed to contribute to the maintenance of genetic diversity in natural populations. The selective disadvantage for common host genotypes comes about because parasites are expected to adapt to these common genotypes [10, 11]. Signatures of balancing selection have been found in gene families associated with disease resistance in vertebrates (the Major Histocompatibility Complex, MHC) and plants (R-gene) [12, 13]. An assumption underlying this form of coevolution is that no parasite can infect all host types and no host can resist all parasite types. The matching-allele-model is one of the genetic mechanisms suggested to prevent the rise of such super-genotypes and thus contributing to the maintenance of genetic diversity [10, 11, 14]. However, despite of in-depth knowledge of the molecular structure of immune-related loci, the genetics underlying the interactions between hosts and parasites have not yet been resolved [15–17].
The Daphnia–Pasteuria system is a model for studies in host-parasite coevolution. Pasteuria ramosa is an obligate bacterial pathogen of the crustacean Daphnia magna that causes strong disease phenotypes with major fitness consequences for the host [8]. In short, feeding hosts pick up dormant P. ramosa spores. Contact with the host results in the activation of spores, which then attach to the hosts’ foregut. If attachment is successful, the spores penetrate into the D. magna body cavity initiating infection and disease. P. ramosa eventually kills the host and its spores are then released into the environment [18]. Importantly, spore attachment is genetically determined and fully consistent with infection success, i.e. resistant host genotypes prevent spore attachment whereas attachment is successful in susceptible host genotypes [19–22]. Here we use the terms resistance and susceptibility to refer to both spore attachment and overall infection.
In this host-parasite system fluctuating selection in natural populations have been observed [23] and the D. magna—P. ramosa interactions follow a matching-allele model with no universally resistant host genotype being found [20–22, 24]. Thus, the Daphnia-Pasteuria host-parasite system fulfils the core assumptions of models for coevolution by NFDS [10, 11, 14], making it a promising model to explore the underlying genetic mechanisms of host-parasite interactions.
We aimed to investigate the molecular genetic basis of this host-pathogen system and to gain insight into the genetic basis of coevolution by NFDS. Using a Quantitative Trait Locus (QTL) approach on a D. magna F2 recombinant panel, one large effect QTL associated with resistance to infection by the P. ramosa C19 genotype was detected [25]. The F2 recombinant panel showed Mendelian segregation of approximately 75% resistant and 25% susceptible genotypes. We build upon this work to explore and characterize the Pasteuria Resistance locus (PR-locus) in D. magna. We show that the PR-locus is highly polymorphic with striking structural genetic polymorphisms and, additionally, gene content and gene expression divergence in the PR-locus between resistant and susceptible haplotypes. The most striking aspect of these differences in gene content is related to a cluster of glycosyltransferase genes located within the PR-locus. Finally, we show that genetic variation at the PR-locus explains variation in resistance to spore attachment observed in natural D. magna populations following the predictions of a matching-allele model.
Routtu and Ebert (2015) detected one major effect QTL underlying D. magna resistance to infection by the P. ramosa C19 genotype located within a scaffold of approx. 2.3 Mb of the D. magna draft genome 2.4 (Fig 1A)[25]. We reduced the interval of the D. magna resistance locus and fine-mapped the QTL interval using microsatellites and SNP markers to find recombination breakpoints within the QTL interval (S1 File). Microsatellite marker P34 and SNP g311b (S1 Table) defined the closest recombination breakpoints at positions 1369860 and 1506194 of scaffold00944 in the D. magna genome draft 2.4, leaving a mapping interval of approximately 130 kb that we call here the PR-locus (Fig 1B). Within this region no further recombination event was detectable among 360 F2 clones. Interestingly, we detected a genomic region of approximately 50 kb within the interval map where none of the designed genetic markers (g294 and g350) could be amplified in the resistant parental D. magna clone Iinb1, while the genetic markers placed outside this region (g292 and g351) did amplify in both parent clones (Fig 1C). As genetic markers were designed to match the D. magna Xinb3 based draft genome (D. magna 2.4), this result could be explained by structural polymorphism—a single indel polymorphism where the entire 50 kb region is absent in D. magna Iinb1 genotype or by a genomic region of such high sequence divergence between haplotypes that all the primer pairs based on D. magna Xinb3 clone would not produce an amplicon with D. magna Iinb1 DNA.
In order to understand the polymorphism between the parental genotypes we applied high-throughput sequencing and long-read PacBio sequencing of both parental clones with the goal to improve the existing assembly of PR-locus in the D. magna Xinb3 clone and to obtain an independent de novo assembly of the same region in the Iinb1 clone. We obtained two complete haplotypes from the D. magna clones Xinb3 and Iinb1 for the PR-locus that correspond to the interval between positions 1366653 and 1520041 of the scaffold00944 in draft genome 2.4 and call them the xPR-locus and iPR-locus, respectively. The most striking feature found was that each haplotype contains a large genomic region where little homology was found corresponding to the region where we had previously found amplicon presence/absence polymorphism (Fig 1C). We call this the Non-Homologous Region (NHR), and the haplotypes we obtained from clones Xinb3 and Iinb1 are called xNHR and iNHR, respectively (Fig 2).
xPR-locus and iPR-locus differ in their nucleotide lengths: xPR-locus is 159 kb long while iPR-locus is 215 kb long. In addition, considering the entire PR-locus haplotypes 34% of xPR-locus and 46% of iPR-locus have no homology to each other (Fig 2) (S2 Table). However, these differences in length and lack of homology are unevenly distributed across PR-locus. It is the NHR that differs substantially in length: iNHR (from the Iinb1 clone) was 121 kb in length, in contrast to xNHR (Xinb3 clone) with only 55 kb (Figs 2 and 3). The two NHR haplotypes contain only few fragments with homologous sequences: in iNHR a total of 25 kb had a significant alignment in xNHR, representing only 20% of the total sequence; in xNHR only 13.7 kb could be homologized to iNHR (Figs 2 and 3)(S2 Table). This region of non-homology at the NHR contrasts to high homology (>90%) at the flanking regions of the NHR, i.e. in the remainder of the PR-locus (Figs 2 and 3)(S2 Table).
A large proportion of both PR-locus haplotypes was composed of repeated sequences. We divide the repeated sequences in two groups according to the location of their copies: sequences that are repeated in the host genome but outside PR-locus–extra-locus repeats; and sequences that were repeated within PR-locus–intra-locus repeats. A large proportion of both PR-locus haplotypes sequences were made of extra-locus repeats. In spite of the differences observed in length between xPR-locus and iPR-locus haplotypes, both had approx. 25% of their total sequence composed of these extra-locus repeats representing 54.7 kb and 39.9 kb, respectively (Fig 2)(S3 Table). Looking into the distribution of extra-locus repeats we observed that they were unevenly distributed as the NHR contains by far the largest proportion of these extra-locus repeat elements, representing 33% of iNHR and 38% of xNHR (Fig 2)(S3 Table). In addition, the remaining extra-locus repeats found outside the NHR were concentrated in a 20 kb region immediately upstream of NHR (Fig 2)(S3 Table). Interestingly, extra-locus repeats accounted for a significant proportion of sequences non-homologous between PR-locus haplotypes. Specifically, 53% of the non-homologous iPR-locus sequences and 51% of the non-homologous xPR-locus are extra-locus repeats. Second, iPR-locus and xPR-locus diverged in number and nature of intra-locus repeats. In xPR-locus, we detected 14 intra-locus repeats, covering 17.3 kb or 11% of the sequence total (Fig 2)(S4 Table). In contrast, in iPR-locus haplotype we detected 30 intra-locus repeats, representing 68 kb and nearly 32% of the total sequence (Fig 2)(S4 Table). Most of these intra-locus repeats were located within the NHR, specifically 97% and 67% of the intra-locus repeat sequence in iNHR and xNHR, respectively (Fig 2)(S4 Table).
In summary, PR-locus is characterized by dramatic structural polymorphism that in its overwhelming majority is contained within a defined genomic region, the NHR. In particular a large proportion of PR-locus sequences here investigated are non-homologous between the resistant and susceptible haplotypes; a large proportion of both PR-locus haplotypes was composed of repeat elements; the repeat sequences could be repeated extra-locus, intra-locus or both; a large part of the sequence that was non-homologous between the PR-locus haplotypes was composed of extra-locus and/or intra-locus repeats; PR-locus haplotypes diverged in their sequence nucleotide length and in the number and nature of both extra and intra-locus repeats (Figs 2 and 3). The NHR, where most of the variation described here is found, is therefore a strong candidate to harbor variation underlying D. magna resistance to P. ramosa.
We annotated the expressed genes in each PR-locus haplotype. Orsini et al. (2016) produced an RNAseq database for D. magna Xinb3 and Iinb1 clones investigated in this article, as well as for D. magna F1 lineage resulting from a cross between D. magna Xinb3 and Iinb1 clones [26]. This D. magna (Xinb3 x Iinb1) F1 clone was in turn used to generate the F2 recombinant panel genotypes used for QTL mapping [27]. In addition to control conditions, the Orsini et al. (2016) study also investigated gene expression in the same genotypes when exposed to multiple environmental stress factors, including exposure to spores of P. ramosa [26]. Using this resource we produced a de novo transcriptome and carried out reciprocal blasts between this transcript database and the PR-locus haplotype sequences that we generated from D. magna Xinb3 and Iinb1 genotypes in order to find which expressed transcripts map to each of the PR-locus haplotypes. We annotated a total of 83 expressed genes that map to the PR-locus haplotypes. Of these, 20 mapped exclusively to the iPR-locus and 18 exclusively to the xPR-locus, whereas 45 annotated expressed transcripts mapped to both haplotypes (S5 Table). The 20 annotated genes that mapped only to the iPR-locus represented one sulfoquinovosyltransferase, and 19 uncharacterized proteins (UP) (S5 Table). The 18 annotated genes that mapped only to the xPR-locus represented five fucosyltransferases, one alpha 1,4-glycosyltransferase, one PC-Esterase and 11 UPs (S5 Table). These observations revealed that the differences in gene content between PR-locus haplotypes resulted for the most part from differences in the representation of fucosyltransferases and UPs. Importantly, all the genes that were exclusive of one or another haplotype, mapped entirely to the NHR region at the center of the PR-locus with the exception of one fucosyltransferase mapping to xPR-locus. This result is consistent with the lack of homology between haplotypes at the NHR. Finally, the 45 expressed transcripts that were shared between the PR-locus haplotypes represented four PC-Esterases, two fucosyltransferases, one methyltransferase, one alpha 1,4-glycosyltransferase, one galactosyltransferase, one sestrin, one DNA mismatch-repair protein, one zinc-finger binding domain, one glutamate synthase, one calcipressin, one spermidine synthase, one acyl-CoA Thioesterase and 29 UPs (S5 Table).
We investigated differences in expression of genes shared between clones Xinb3 (susceptible to P. ramosa C19) and Iinb1 (resistant to P. ramosa C19). Among the 45 transcripts resulting in annotated genes that mapped to both PR-locus haplotypes, 20 were differentially expressed between Xinb3 and Iinb1 clones (S6 Table). Using the Xinb3 clone (the chosen clone for the 2.4 D. magna draft genome) as the focal genotype we identified 11 upregulated and nine downregulated expressed transcripts (S6 Table). The 11 transcripts upregulated in the Xinb3 clone represented one methyltransferase, one fucosyltransferase, one DNA mismatch-repair protein, one PC-esterase and seven UPs (S6 Table). The nine transcripts downregulated in the Xinb3 clone represented one calcipressin, one DNA mismatch-repair protein, one fucosyltransferase, one sestrin and five UPs (S6 Table). In order to narrow down the number of candidate genes in the PR-locus haplotypes, we compared expression of transcripts mapping to the PR-locus haplotypes between the Xinb3 and Iinb1 clones and the hybrid F1 (Xinb3 x Iinb1) clone. The hybrid F1 clone was resistant to the P. ramosa C19 genotype just as the Iinb1 clone and in contrast to the Xinb3 clone. Thus, we searched for those transcripts that were consistently down- or upregulated in the Xinb3 clone in comparison to both of the Iinb1 and F1 clones, as those represented the best candidates to underlay the variation in resistance to P. ramosa observed in the previous QTL study [25]. Only one transcript of calcipressin was downregulated in the Xinb3 clone when compared to both of the Iinb1 and F1 clones. In contrast, seven transcripts were upregulated in the Xinb3 clone, including one methyltransferase, one DNA mismatch-repair protein, and five UPs (S6 Table).
In Orsini et al (2016), a number of transcripts were differentially expressed between P. ramosa infected and non-infected individuals of the same genotype (same D. magna clone) [26]. We investigated these transcripts to find if any of them would map to our interval. Importantly, we found no significant differences in gene expression between controls and P. ramosa treatments for transcripts mapping to PR-locus (data not shown) (but see McTaggart et al. 2015) [28]. Rather, the significant differences in expression were identified when comparing the control treatments of the Xinb3 and Iinb1 clones. This is not surprising given that we are here investigating the host’s first line of defense, while genes expected to be expressed differently are genes whose expression is induced once the parasite succeeds in infecting its host—the second line of defense [18].
One model was suggested, whereby three D. magna resistance loci govern the Daphnia-Pasteuria host-pathogen system, regarding the two P. ramosa genotypes, C1 and C19 [22]. In this model, variation in locus C determines resistance to both P. ramosa genotypes whereas variation in loci A and B determines D. magna resistance to P. ramosa genotypes C1 and C19, respectively. Epistasis between loci can be described as follows: the presence of the resistant allele in C masks the genotypes at loci A and B, and the presence of the resistant allele in A masks the genotype at locus B (Fig 4). A hierarchy of dominance between D. magna resistance phenotypes is observed: RR (C1, C19 double resistant) > RS (C1 resistant, C19 susceptible) > SR (C1 susceptible, C19 resistant) > SS (double susceptible) [20, 22]. Our analysis so far allows us to conclude that the predicted locus C (Fig 4) is located within PR-locus. However, it does not resolve if different locus C alleles result from structural variation at the NHR or from variation in the flanking region. In addition, since all F2 recombinant clones were either RR (double resistant) or RS (C1 resistant/C19 susceptible) resistance phenotypes, we cannot withdraw any conclusions on whether loci A and B are located within PR-locus even though the three loci are expected to be closely linked [22] (Fig 4). Therefore, we undertook an association study, testing for a link between structural variation at the PR-locus and variation in resistance to P. ramosa spore attachment in D. magna clones collected from a metapopulation in the Tvärminne archipelago in Finland. We tested 447 Tvärminne clones from 27 different populations (rock pools) (on average 16.5 clones per population) for resistance to P. ramosa genotypes C1 and C19 using the attachment test and observed high resistance phenotype diversity between and within the rock pool populations (S7 Table). We then tested two genetic markers (g294 and g350) designed within xNHR unique coding sequences based upon the current draft genome (ver 2.4) for the susceptible D. magna clone Xinb3 for presence/absence patterns. We had two predictions: i) that these markers (S1 Table) would produce an amplicon when the xNHR haplotype was present either in a homozygote or heterozygote form, but not when it was absent from the tested genotype; and ii) that since the RS phenotype (observed in Xinb3 clone) is dependent on the dominant allele of locus A, these amplicons would be produced irregularly in RR clones, always in RS clones and never in SR (C1 susceptible/C19 resistant) and SS (double susceptible) clones. Our analysis revealed two groups of host genotypes. There were genotypes where the xNHR diagnosis markers amplified together (as does the Xinb3 clone) and other genotypes where none of the markers could be amplified (as is the case for the Iinb1 clone) (Fig 1C). As expected, this amplification pattern was strongly associated to resistance to P. ramosa C1 genotype. Specifically, clones susceptible to C1 almost never showed xNHR diagnostic marker amplification (resistance phenotypes SR and SS). Clones that are at the same time resistant to C1 genotype and susceptible to C19 genotype (RS) always show amplification (this is also the case for the Xinb3 genotype), whereas double resistant clones (RR) could show amplification or not (Fig 5). The double resistant Iinb1 clone does not show amplification of any of these xNHR diagnostic markers. We tested whether these results would be confirmed within a single D. magna population. We chose a rock pool population (K-8) with only RS and SR resistance phenotypes being present and predicted that this polymorphism is associated with presence and absence of the xNHR. In our K-8 population sample we found that 56 out of 60 RS clones showed xNHR marker amplification, whereas only one out of 36 SR clones showed such amplification (Table 1). Thus, we find a strong association between presence of xNHR haplotype and RS resistance phenotype, and xNHR absence and C1 susceptibility both within and between populations.
The fine mapping and sequence analysis of the Daphnia magna PR-locus revealed an unusual pattern of structural polymorphism between haplotypes. Remarkably, we find lack of homology between PR-locus haplotypes in restricted regions of 55 kb and 121 kb, the xNHR and iNHR, respectively (Figs 2 and 3)(S2 Table). In the PR-locus haplotypes, and particularly within the NHR sequences we found a complex pattern of repeated sequences, which likely represent a history of evolutionary events with multiple classes of structural mutations playing a role. The existence of large-scale repetition of sequences found elsewhere in the D. magna genome, the extra-locus repeats (Fig 2)(S3 Table), argues against horizontal gene transfer in creating the NHR, while suggesting that gene conversion might be a recurrent phenomenon influencing its evolution. The difference in length between the two haplotypes is explained by a far higher prevalence of intra-locus repeats in the iNHR in comparison to the xNHR that suggests a higher number of segment duplication events in iNHR (Fig 2)(S4 Table). Finally, the lack of homology between the two NHR haplotypes together with the observation that this region seem to segregate as one unit in natural populations, suggests the absence of, or very low rates of local recombination.
Taken together, our results indicate that the NHR represents a defined and highly divergent genomic region whose structural genetic variation underlies the natural variation in D. magna resistance to P. ramosa.
The characteristics that we find in the NHR of the D. magna PR-locus largely overlap with what is known of the genetics, origin, structure and evolution of supergenes. Supergenes are clusters of multiple loci, each affecting different traits that together control complex phenotypes within a species and segregate as a block that is characterized by restricted or suppressed recombination [29]. Supergenes can emerge due to new mutations leading to beneficial interactions with closely linked loci, or to structural large-scale mutations such as gene duplication and translocation [29]. Large-scale structural polymorphisms are one of the main reasons for recombination suppression in supergenes and there are examples of supergenes being located in genomic fragments that are absent in alternative haplotypes [29, 30]. Finally, NFDS seems to be the main evolutionary mechanism to maintain supergene polymorphism [29]. Thus, it is tempting to suggest that the NHR of D. magna PR-locus may represent an immunity supergene.
We collected more than 400 clones from a well-studied D. magna metapopulation located in the Tvärminne archipelago in South-Western Finland and made an association study between their resistance phenotypes for P. ramosa genotypes C1 and C19 and the presence of diagnostic markers of the xNHR. We find that the presence of the xNHR haplotype is tightly associated to the RS phenotype (C1 resistance and C19 susceptibility), while xNHR markers are absent in D. magna clones with SR and SS phenotypes (Fig 5). On the other hand, the presence of xNHR markers shows no association with RR phenotypes (Fig 5). We verified the association between xNHR and the RS phenotype in a single population (rock pool K-8), which was polymorphic only for RS and SR phenotypes. In this population the matching-allele matrix–already described for this host-parasite system–is clearly seen [21, 22]. D. magna clones showing RS phenotype are homozygote or heterozygote for the dominant xNHR, while this haplotype is absent in SR clones (Fig 4) (Table 2). Gene conversion, rare events of homologous recombination at NHR, or errors while determining the resistance phenotypes or the marker could explain the few instances where xNHR diagnostic markers are absent in RS clones or present in SR clones (Table 1)(S7 Table).
Our results are consistent with previous work showing a dominance hierarchy between D. magna resistance phenotypes and epistasis between resistance loci [20, 22]. The NHR corresponds to the A-locus in these earlier studies. The xNHR contains the dominant allele of the A-locus whereas the iNHR contains the recessive allele. The phenotype associated to xNHR is hidden in RR clones, as its effect is suppressed by the dominant C allele at the C-locus (Fig 4). Conversely, the presence of the xNHR is strongly associated with the RS phenotype and completely absent in SR and SS clones. The presence of the xNHR masks the effect of the B-locus, which defines the SR and SS resistance phenotype polymorphism (Fig 4). In population K-8 the C-locus is apparently fixed for the recessive c-allele, while the B-locus is fixed for the dominant B-allele (Table 2). On the other hand, the results of the QTL mapping leading to PR-locus, is based on a polymorphism at the C-locus (parents are CC—Iinb1, and cc—Xinb3, while the F1 is Cc), because the parental genotypes used, Iinb1 and Xinb3 clones, have RR and RS phenotypes and no other phenotype was found in over 400 tested F2 recombinants [22]. Thus, the C-locus is also located within the PR-locus (Fig 4). Finally, a report of recombination between the three linked resistance loci concluded that the B-locus is located between loci A and C [22], suggesting loci A, B and C loci would all sit within the PR-locus (Fig 4).
Until now few empirical examples of matching-allele interactions have been described in host-parasite systems [31], which can result from this type of genetic interactions being rare. However, in the D. magna-P. ramosa system the ease of collecting, large samples are easily available for collection, genotyping and phenotyping. Furthermore, the clonal system of reproduction of D. magna permits the maintenance of stable genotypes without the need to produce inbred lines [8, 21, 22, 24]. Together, these traits increase the probability of finding existing matching-allele interactions. In addition, many studies of host-parasite systems rely on the overall infection results whereas the infection process requires a series of steps, each with its own genetic basis [18]. In the D. magna-P. ramosa system the spore attachment step is the only infection step that fulfils the requirements of a matching-allele model: binary response; lack of environmental variability and; host-parasite genotype-to-genotype interactions. It is possible that by focusing on infection steps that show the same characteristics and using large numbers of host and parasite genotypes, future studies reveal more examples of matching-allele interactions.
In parallel to large structural polymorphisms found in the NHR region of D. magna PR-locus we found differences in the gene content between the i- and the x- haplotypes at the PR-locus. Most differences in gene content are associated with genes that map to the NHR region (S4 Table and S5 Table). Gene annotation reveals that genes of the glycosyltransferase family are over-represented within xPR-locus including seven fucosyltransferases, two alpha 1,4-glycosyltransferase and one galactosyltransferase transcripts (S5 Table). In contrast, iPR-locus has only two fucosyltransferase transcripts, one alpha 1,4-glycosyltransferases and one galactosyltransferase (S5 Table). Glycosyltransferases are known to play fundamental roles in innate and acquired immunity-related traits in multiple organisms [32–34]. Thus, the differences in the presence and activity of fucosyltransferases and alpha 1,4-glycosyltransferases indicate that these are good candidates genes that may determine variation in D. magna resistance to P. ramosa.
D. magna—P. ramosa is a host-pathogen system where growing evidence suggests NFDS as the primary responsible of the coevolutionary process [20–23]. Here we describe the first steps into the molecular basis of evolution by NFDS and find evidence that suggest a role for glycosyltransferase genes in our study system. Next, it is important to identify which particular genes are responsible for the observed polymorphism. That requires to fine-map the A, B and C loci (Fig 4) and to then carry out functional tests on the remaining candidate genes (e.g. gene knock-outs) to verify their role. Furthermore, it is important to describe more D. magna PR-locus haplotypes associated with different resistance phenotypes to better understand the extent of the genetic variation associated to D. magna resistance to P. ramosa and the relative roles that gene conversion and homologous recombination have in shaping it.
The D. magna (Xinb3 x Iinb1) F2 recombinant panel is a resource available at the Ebert laboratory in Basel, Switzerland, that was generated from a single cross between the Xinb3 mother clone and the Iinb1 father clone [27]. A QTL analysis based on this resource revealed one major effect QTL for resistance against P. ramosa genotype C19 [25]. In the region of the major QTL for resistance to P. ramosa, single nucleotide polymorphism (SNP) and microsatellite markers were designed based on the D. magna 2.4-genome draft (S1 Table). We amplified each marker via standard PCR and Sanger sequenced them in all F2 clones with a recombination event in the region around the resistance QTL. We then searched for the recombination breakpoints in each F2 recombinant clone.
Since the region around the QTL was poorly assembled in version 2.4 of the D. magna draft genome (http://wfleabase.org/), we undertook a number of additional sequencing and assembly methods in order to better resolve the focal region. For Xinb3 we generated high coverage (~60X) PacBio sequencing in order to perform de novo genome assembly. For Iinb1 we took a hybrid Illumina short-read/PacBio long-read approach, generating ~80X 125bp PE Illumina coverage and ~ 15X PacBio long-read coverage (see S1 Methods). We used the D. magna Xinb3 and Iinb1 haplotype sequences obtained to search for homologies within and between haplotypes and other genomic regions (see S1 Methods). In order to understand how expression of individual genes localized to the focal genome regions and to other parts of the genome differed between the Xinb3 and Iinb1 clones, we conducted a de novo transcriptome assembly of the data set described in Orsini et al. (2016) (see S1 Methods) [26]. Finally, we constructed a de novo annotation of each of the transcripts mapping to PR-locus by performing blastx (nucleotide to protein) searches in the NCBI database (see S1 Methods).
The aim of this assessment was to link the structural polymorphism observed in the QTL panel with genetic variation for resistance in natural populations. D. magna females were collected from fresh water rock pools in the long term study area of the Tvärminne archipelago, South-Western Finland. The Tvärminne archipelago is composed of many skerry islands of varying sizes, each with multiple rock pools that freeze in winter, forcing the Daphnia to survive as sexually produced resting stages called ephippia. It is the location where the ancestor of the D. magna Xinb3 genotype (our three times selfed reference genome clone) was first collected. Each rock pool represented one population, but together these populations form a metapopulation with frequent migration. Females were freshly hatched from sexually produced resting stages (ephippia) in the wild right after the winter season and thus each of them represented a unique genotype (clone). In the laboratory, we separated females into individual jars initiating a clonal line. Clones were kept in ADaM media at 20°C, fed with Scenedesmus sp. three times a week and moved to fresh media once a week [20, 35]. Resistance phenotypes were determined using the attachment protocol described in Duneau et al. (2011) [19]. Two cloned P. ramosa genotypes, C1 and C19, were used in this study [24]. In short, three replicates of each D. magna clone were placed individually into 96-well plates and exposed for one hour to spores of P. ramosa C1 or C19 genotypes marked with fluorescein5(6)isothiocyanite [19], after which the attachment of spores to an individual was assessed under fluorescent microscope. Attachment of spores to the esophagus of the host indicated that this host genotype was susceptible to the pathogen genotype tested whereas absence of spore attachment implied host resistance [19]. Primers for genetic structural markers were designed based on the available Xinb3 D. magna genome draft (version 2.4) at the time. Each primer pair was selected so that it amplified one coding sequence predicted to be present in the annotated genome (S1 Table). Absence or presence of visible amplicons on an agarose gel (1.5% w/v) was used as indicator of PR-locus genotypes (absence indicating homozygotes for absence, while presence indicates homozygotes for presence or heterozygotes). Statistical analysis was based on contingency tables of expected vs. observed values to which a Chi-square test was applied to test statistical significance to both the full dataset and to pairwise comparisons between resistance phenotypes.
|
10.1371/journal.pntd.0007439 | Insecticide resistance levels and mechanisms in Aedes aegypti populations in and around Ouagadougou, Burkina Faso | Recent outbreaks of dengue and other Aedes aegypti-borne arboviruses highlight the importance of a rapid response for effective vector control. Data on insecticide resistance and underlying mechanisms are essential for outbreak preparedness, but are sparse in much of Africa. We investigated the levels and heterogeneity of insecticide resistance and mechanisms of Ae. aegypti from contrasting settings within and around Ouagadougou, Burkina Faso.
Bioassays were performed on larvae and adults to diagnose prevalence of resistance, and to assess levels where resistance was detected. Investigation of resistance mechanisms was performed using synergist bioassays, knockdown resistance (kdr) target site mutation genotyping and quantitative PCR expression analysis of candidate P450 genes.
Larval dose-response assays indicated susceptibility to the organophosphates tested. Adult females were also susceptible to organophosphates, but resistance to carbamates was suspected in urban and semi-urban localities. Females from all localities showed resistance to pyrethroids but resistance prevalence and level were higher in urban and especially in semi-urban areas, compared to the rural population. Environment was also associated with susceptibility: adults reared from larvae collected in tires from the semi-urban site were significantly less resistant to pyrethroids than those collected from large outdoor drinking water containers (‘drums’). Susceptibility to both pyrethroids tested was largely restored by pre-exposure to Piperonyl Butoxide (PBO), suggesting a strong metabolic basis to resistance.
The 1534C kdr mutation was nearly fixed in semi-urban and urban areas but was far less common in the rural area, where the 1016I kdr mutation frequency was also significantly lower. P450 gene analysis detected limited over-expression of single candidates but significantly elevated average expression in the semi-urban site compared to both a susceptible laboratory colony, and females from the other collection sites.
Our results reveal pyrethroid resistance and paired kdr mutations in both urban and semi-urban sites at levels that are unprecedented for mainland Africa. The combination of target site and metabolic mechanisms is common in Ae. aegypti populations from other continents but is a worrying finding for African populations. However, organophosphate insecticides are still active against both larvae and adults of Ae. aegypti, providing useful insecticidal options for control and resistance management.
| Several African countries including Burkina Faso have experienced dengue outbreaks recently. In outbreaks, dengue control relies on the control of its vector but data on Aedes aegypti resistance to insecticide, a key for efficient control, are often lacking, especially for the African continent. We conducted a study in localities within and around Ouagadougou to assess the Ae. aegypti resistance to insecticides and investigate the mechanisms involved. We collected larvae of Aedes aegypti from three different localities and different breeding sites to assess larval and adult susceptibility to insecticides. Aedes aegypti adults showed high resistance to pyrethroid insecticides with the prevalence and intensity of resistance depending on the locality and type of breeding site. Adults showed less pronounced resistance to carbamates, and both larvae and adults remain susceptible to organophosphate insecticides. The resistance to pyrethroid insecticides is partly explained by a high frequency of a pair of kdr mutations (1534C and 1016I) and the overexpression of genes of the P450 family linked to insecticide degradation in the mosquito. Datasets on both resistance and mechanisms in Ae. aegypti from the African continent are quite rare and are important for dengue control in Burkina Faso and elsewhere.
| The African continent is particularly at risk of arbovirus-disease outbreaks, a situation enhanced by the high number of vector species, the lack of organised vector control, a deficit of vector biologists and the absence of prevention policies for neglected tropical diseases [1]. Aedes aegypti is the main vector involved in the transmission of the most important arboviruses—dengue, yellow fever, Zika and chikungunya—which occur as recurrent outbreaks in parts of the African continent [2]. Approximately 50% of the world’s population lives in areas at risk of dengue virus infection, and in Africa, dengue has been recorded in 34 countries in the past 50 years [3,4]. West Africa has been identified as a potential dengue hotspot because of the co-occurrence of rapid urbanization without adequate sanitation and the widespread presence of Ae. aegypti [5].
Burkina Faso has a long history of dengue epidemics, with the first reported in 1925 [3] and another in 1982 [6]. Dengue cases were recorded regularly but at low levels from 2006 [7,8] until an outbreak in 2016 resulted in 2,600 cases and 21 deaths [9]. In 2017, a larger outbreak in the city of Ouagadougou spread to other regions, ultimately resulting in 14,455 cases and 29 deaths nationwide [10]. These outbreaks highlighted the vulnerability of Burkina Faso to dengue and other Aedes-borne arboviral diseases.
Typically, dengue prevention relies on vector control through larval source reduction and case management, but in outbreak periods, insecticidal space spraying is usually employed to target adult mosquitoes [11]. Worldwide, dengue vectors have developed resistance to most insecticides used in public health [12] but data from Africa are sparse, and the mechanisms involved in resistance in African Aedes populations are very poorly understood [1].
The best documented mechanisms of Ae. aegypti insecticide resistance involve mutations in the voltage gated sodium channel (VGSC) target site of pyrethroids and Dichloro-Diphenyl-Trichloroethane (DDT), and metabolic detoxification. Multiple VGSC knockdown resistance (kdr) mutations have been identified in Ae. aegypti but only V410L, V1016G, I1011M and F1534C have been validated as being directly causally-associated with resistance to pyrethroid insecticides [13–15]—and only one of these, F1534C, has been detected in Africa to date [16]. Other mutations (e.g. V1016I and S989P) are involved in the resistance at least when associated with other VGSC variants [12]. Metabolic resistance is also important in Ae. aegypti populations at multiple geographic locations, and many genes of the P450 family, especially from the CYP9 and CYP6 subfamilies have been associated with resistance to pyrethroids [12,17].
Aedes aegypti is rarely a target for vector control in Africa, though in Cape Verde insecticide-based vector control has been established since 2009 after the first dengue cases were diagnosed [18]. Current data suggest a more heterogeneous picture of insecticide resistance across Africa than in Latin America and South-East Asia, but in mainland West Africa there is evidence of established or emerging resistance to DDT, carbamates and pyrethroids [1]. Knowledge of the underlying mechanisms is very limited in African populations, but resistance to DDT and permethrin has been linked to a high frequency of the 1534C kdr mutation in Ghana, whilst the 1016I mutation which, when co-occurring with 1534C yields broader and stronger pyrethroid resistance [19], was very rare [16]. This is in contrast to Cameroon, where resistance to pyrethroids has evolved within a decade from susceptibility to well-established resistance, apparently in the absence of kdr mutations [20,21]. On the island of Madeira, the Ae. aegypti population exhibit strong pyrethroid resistance, underpinned by dual kdr mutations (V1016I and F1534C) and overexpression of metabolic genes. Since the Madeira population was founded only recently, it suggests, worryingly, that a suite of resistance mechanisms can establish rapidly after introduction to an area [22].
Information on insecticide resistance is a basic requirement when considering tools or approaches for dengue control. In this study, we characterise the resistance of Ae. aegypti and investigate underlying mechanisms in urban, semi-urban and rural localities of Ouagadougou, the capital city of Burkina Faso. Key findings include the vector population’s susceptibility to organophosphates and strong but variable resistance to pyrethroids between localities, linked to the 1534C and 1016I kdr mutations and to P450 gene overexpression.
Aedes aegypti larvae were sampled from two localities within, and one beyond the perimeter of the city of Ouagadougou (Fig 1), selected for differences in their ecological characteristics, human population size and housing type.
1200 Logements (12° 22' 3.569''N, 1° 29' 50.24''W): located within central Ouagadougou close to the international airport; housing has piped water supply with good sanitation and drainage, electricity and waste management systems. Potential Aedes breeding sites are predominantly discarded tires and small containers (volume <5L).
Tabtenga, (12° 21' 58.039''N, 1° 26' 59.074''W) located approximately 5 km East of 1200 Logements, Tabtenga is a semi-urban district, lacking a centralized water supply, electricity or waste management systems. Potential Aedes breeding sites include tires, drums (large ceramic water containers) and small containers.
Goundry (12°31´4.262”N, 1°20´25.771”W): a small rural farming community situated 25 km north-east of Ouagadougou; mostly small scale cultivation and livestock, with no water supply, electricity or waste management systems. Potential Aedes breeding sites are primarily drums, or water containers provided for animals.
During the rainy season from August to October 2016, larvae were collected in tires from 1200 Logements, in drums and tires from Tabtenga, and in drums from Goundry. Water containing larvae from breeding sites was filtered using sieves and the larvae transferred to the laboratory. Many breeding sites were sampled from each locality and larvae were pooled for subsequent rearing. Larvae were reared using dried cat food in the insectary until F0 adult mosquitoes of three to five days old were obtained for bioassay tests. The insectary conditions were 27.7±1.4°C temperature, 79.1±5.5% relative humidity, 12h light/dark photoperiod.
Larval bioassays were performed according to WHO dose-response assay protocols [23] on 3rd and 4th instar larvae after one day of acclimation to the laboratory conditions as described above. Three organophosphate insecticides were tested, at five concentrations for each insecticide: temephos (0.25 mg/L to 20.25 mg/L); fenitrothion (0.25 mg/L to 20.25mg/L) and malathion (6.25 mg/L to 168.75 mg/L). One hundred Ae. aegypti larvae were exposed to each concentration, along with a negative control (no insecticide); numbers of dead and alive larvae were recorded at the end of the 24 h exposure period. We were unable to obtain a laboratory susceptible strain as a standard on which to simultaneously perform bioassays locally. Therefore, to compare with values estimated from the data (see below) we obtained LC50 values (in mg/L) from published literature to allow computation of resistance ratios. Moyes et al. [12] reviewed published data on temephos larval assays using the Ae. aegypti Rockefeller strain, and calculated a mean LC50 = 0.0042 (N = 30). We found three studies that estimated malathion LC50 values for susceptible strains: Rockefeller = 0.27 [24]; Rockefeller = 0.40 [25]; GA1 strain = 0.097 [26], and took the median of these values (0.27 mg/L) as the reference LC50. For fenitrothion we identified only one study with the Rockefeller strain (0.009 mg/L; [25], which we used as our reference value.
Adult bioassays used females (three to five days post-eclosion) reared from field-collected larvae and were performed according to standard protocols [27]. Five insecticides were tested: 0.75% permethrin; 0.05% deltamethrin; 0.1% bendiocarb; 1% fenitrothion and 5% malathion; and the resistance status of each population interpreted according to WHO criteria [27]. Whilst these are not all accepted diagnostic doses for Ae. aegypti, they are the most commonly used [12]. One hundred adult mosquitoes from each locality were exposed to each insecticide, along with 50 mosquitoes exposed to control (no insecticide) papers. Immediately following 1h exposure, knockdown was recorded, and mortality recorded after 24h. Mosquitoes were stored over silica gel at -20°C for later DNA analysis, and samples of survivors and control were kept in RNAlater at -20°C for gene expression analysis.
The CDC bottle assay technique was used also to test for higher levels of resistance to those insecticides for which resistance was detected in the WHO assays. For this purpose, a prolonged exposure time of 2h was employed with knockdown recorded at the end of the exposure period and mortality recorded 24h later. Technical grade insecticides (>90% purity; Sigma-Aldrich) were used at discriminating concentrations [28]. Stock solutions were prepared at concentrations of 12.5 μg/ml for bendiocarb, 10 μg/ml for deltamethrin, 50 μg/ml for fenitrothion and 15 μg/ml for permethrin. For each insecticide, 1 ml was used to coat the 250 ml bottles, which were dried and kept in a fridge until use. In addition, the synergist piperonyl butoxide (PBO), which primarily blocks activity of P450s and some esterases [29], was used to assess a possible role for metabolic resistance mechanisms, at the recommended concentration of 400 μg/ml [28] in pre-exposure assays, followed by pyrethroid exposure.
DNA was extracted from two legs of each mosquito, which were transferred using clean forceps to PCR plate wells. The wells were sealed and the plate briefly centrifuged to ensure the legs were at the bottom of the well. Twenty μl of STE buffer (0.1M NaCl, 10 mM TrisHCl pH = 8.0, 1mM EDTA pH8.0) was added to each well containing the mosquito legs. The plates were heated at 95°C for 90 min and briefly spun again. The resultant extracts were stored at -20°C until use. The V1016I, F1534C, S989P and V1016G VGSC mutations were genotyped using the Taqman qPCR method [30]. Reactions were performed in 96 well plates by adding 5 μl of Taqman gene expression SensiMix (Applied Biosystem, Foster city, USA), 0.125 μl of primer/probe, 3.875 μl of molecular grade sterile water and 1 μl of the DNA extract. Reactions were run on an Agilent MX3000P qPCR thermal cycler using cycling conditions of an initial denaturation of 10 min at 95°C, followed by 40 cycles of 92°C for 15 min and 60°C for 1 min.
Control mosquitoes from bioassays preserved in RNAlater were used for RNA extraction and cDNA preparation. Mosquitoes were washed in distilled water and pooled in batches of five per tube for RNA extraction using the PicoPure RNA Isolation Kit (ThermoFisher) according to manufacturer’s instructions. Quantity and quality of RNA was checked using a NanoDrop spectrophotometer and was kept at -80°C until further use.
Complementary DNA (cDNA) was synthetized using reverse transcriptase with Oligo(dt)20 primer according to manufacturer’s instructions.
Candidate genes for expression analysis were chosen based on previous implication of involvement in metabolic resistance [12,17]. All primer sequences and their origins are shown in S1 and S2 Tables. Standard curve analyses were performed for each primer pair to check the specificity and efficiency of amplifications. Seven cytochrome P450 candidate genes were chosen for analysis, along with two normalising genes (S1 Table). Real-time quantitative PCR reactions were performed in a total volume of 20 μl (7.8 μl DDW+10 μl SYBRgreen, 0.6 μl of each primer and 1 μl of cDNA) under the following conditions: 95°C for 3 min, followed by 40 cycles of 95°C for 10 s and 60°C for 10 sec. The relative expression level and fold change (FC) of each candidate gene relative to the susceptible Rockefeller strain was calculated using the ΔΔcT method [31].
Larval 50% and 95% lethal concentrations and their confidence limits were calculated by fitting a logistic regression to mortality after 24h, using an R script for analysis of bioassays and probit graphs [32]. To interpret results in terms of susceptibility, we calculated resistance ratios compared to susceptible strain values, as described above, and interpreted resistance ratios as follows: <5, little resistance; 5–10, moderate resistance; >10, substantial resistance [33]. Adult bioassay data were analysed according to WHO criteria [27]: a population is considered resistant if the mortality after 24h is less than 90% and susceptible when the mortality is over 98%. Between the two values, the population is considered suspected to be resistant and confirmation is needed.
A generalized linear model (GLiM) was fitted to the 2h CDC bottles bioassay mortality for pyrethroid insecticides. The model initially included pyrethroid insecticide type, container type, pre-exposure to PBO, and all interactions, with terms removed sequentially until the minimal model was obtained. Gene expression data (ΔcT) values were compared between each locale and the Rockefeller strain using t-tests, following checks using F-tests for homogeneity of variances. Variation in fold change (calculated as 2-ΔΔcT, relative to the average of Rockefeller) among collection locations was tested using multivariate analysis of variance (MANOVA) for all genes, following checks for normality (Kolmogorov-Smirnov test) and homogeneity of variances (Levene’s test), and ANOVA for individual genes with Tukey’s test for pairwise comparisons. All tests were performed using SPSS v 23. Allele frequencies were compared among localities using χ2 tests.
The research protocol entitled (16–030) “Dengue in Burkina Faso: establishing a vector biology evidence base for risk assessment and vector control strategies for an emerging disease” (16–030) received ethical approval from the National Ethical Committee for Medical Research, Ministry of Health in Burkina Faso (Deliberation N°2016-6-073) on 6th June 2016 and the Research Ethics Committee at the Liverpool School of Tropical Medicine on 15th July 2016. When larvae were collected inside or near a residence, permission (signed consent) from the owners/residents was obtained before entering their property or land.
Effect of exposure to malathion, fenitrothion and temephos on Ae. aegypti larvae collected from urban, semi-urban and rural localities of Ouagadougou are shown in Table 1. There were significant differences in the estimates obtained for both LC50 values and LC95 among the localities, although the rank order of these was inconsistent across the three insecticides. Importantly, in all cases the resistance ratios calculated from the LC50 values were low, and in no case did they indicate any evidence of significant resistance.
Adult mosquito bioassay results are shown in Fig 2 for the WHO tube bioassays and Fig 3 for the CDC bottle bioassays. As there was no evidence of resistance to the organophosphate insecticides malathion and fenitrothion in any of the collection localities using the WHO bioassays, CDC assays were not performed for these insecticides. WHO test results for bendiocarb differed between collections (χ23 = 53.4, P<<0.001) with Goundry susceptible, but 1200 Logements (1200LG) and Tabtenga resistant. However, in the Tabtenga collections, mosquitoes sampled from drums were significantly more resistant than those from tires (χ21 = 10.9, P<0.001), while tire-collected mosquitoes showed similar bendiocarb bioassay results to those collected from 1200LG (χ21 = 2.4, P = 0.12). No evidence of higher-level bendiocarb resistance was detected in either site in the CDC bottle bioassays (Fig 3).
For both pyrethroids, WHO bioassays showed confirmed resistance (i.e. <90% mortality) in each collection but with significant variation in prevalence (permethrin: χ23 = 298.9, P<<0.001; deltamethrin: χ23 = 186.4, P<<0.001). In each case, mortality was by far the highest in Goundry and lowest in the Tabtenga drum collections (Fig 3), but there were no significant differences between WHO bioassay results for the tire-collected mosquitoes from Tabtenga and 1200LG (permethrin: χ21 = 1.3, P = 0.25; deltamethrin: χ21 = 1.4, P = 0.24). Results from the 2h CDC assays indicated some reduction in susceptibility in Goundry, with mortalities >90% but <98% for permethrin and deltamethrin, respectively. In both cases, PBO pre-exposure restored full susceptibility (Fig 3). Mortality was lower in the 1200LG females (and largely restored by PBO) and lower still in those from both container types from Tabtenga (Fig 3; non-overlapping confidence limits indicate significance). In the Tabtenga collections, insecticide type, container and PBO all exerted highly significant influences on mortality, but the lack of any significant model interaction terms (Table 2), suggests independent effects, i.e. the differences between insecticides and the effects of PBO synergism were similar across container types.
The target site mutations kdr S989P, V1016G, F1534C, and V1016I were genotyped in 48, 75, and 43 Ae. aegypti females from 1200LG, Tabtenga and Goundry, respectively (S5 Table). Only the V1016I and F1534C mutations were detected. The 1534C mutation was almost fixed in the urban (1200LG) and the semi-urban (Tabtenga) localities (which are separated by about 5 km) with allele frequencies of 0.94 and 0.97, respectively, but was far less common in rural Goundry (separated by about 25 km from the other localities), with an allele frequency of 0.34 (χ22 = 149.8, P<<0.001). The V1016I kdr mutation was less common in all collection sites, but again it was found at higher frequencies in 1200LG (0.22) and Tabtenga (0.27) than in Goundry (0.06) (χ21 = 18.6, P<0.001). Similarly, the frequencies of combined genotypes differed markedly (Fig 4) with the dual wild type genotype absent from 1200LG and Tabtenga but common in Goundry. Assuming that a single allele (across the two loci) is unlikely to exert much influence on resistance phenotype [34], we compared the frequencies of genotypes with zero or one mutant alleles with those with two or more mutations across populations (Fig 4). Frequencies differed dramatically (χ21 = 48.3, P<0.001) as a result of the Goundry genotypes, with no significant differences (χ21 = 0.19, P = 0.66) between 1200LG and Tabtenga.
Expression of Cyp6 and Cyp9 subfamily P450 candidate genes was analysed by comparing insecticide-unexposed mosquito samples from each locality relative to the Rockefeller susceptible strain. The P450 genes were generally expressed at low-moderate levels relative to Rockefeller and whilst expression was highest in Tabtenga females for every gene (Fig 5), no significant differences in expression from Rockefeller or among collections were detected for any individual gene after correction for multiple testing (minimum uncorrected P = 0.02). However, comparing fold differences (relative to Rockefeller) of the P450 genes jointly revealed significant variation among the three collection locations (MANOVA, Hotelling’s T212,14 = 7.3, P = 0.027), with higher average expression for the Tabtenga collections than 1200LG or Goundry, neither of which differed from Rockefeller (Fig 5). This suggests that whilst none of the candidate P450 genes showed strong variation, their aggregate expression level may contribute to the variation in resistance phenotypes observed.
Resistance to insecticides in Aedes vectors of arboviruses is a major challenge for disease control globally. Here we investigated the susceptibility to commonly-used larval and adult insecticides in three contrasting localities of Ouagadougou, Burkina Faso, to provide essential information to aid rational insecticide choices for preventative control and dengue outbreaks. Unfortunately, a severe outbreak of dengue started before the end of our investigation, but our preliminary data showing Ae. aegypti susceptibility to malathion supported its use for outdoor spraying in hotspots during the outbreak. We also investigated the resistance mechanisms that may be involved in resistance, with particular relevance to pyrethroids. Overall, our study recorded no resistance to organophosphates, moderate and spatially-variable resistance to carbamates, and strong but highly-variable resistance to pyrethroids, likely driven by dual kdr mutations and metabolic resistance.
Aedes aegypti control commonly employs organophosphates, particularly temephos and malathion, to control larval and adult stages, respectively. The LC50 values we obtained for temephos are very similar to values obtained for the Rockefeller susceptible strain [12] and whilst fewer published data are available for fenitrothion and malathion as larvicides, our data are compatible with a fully susceptible phenotype. These results reflect the overall picture from studies in Africa where, to date, temephos resistance appears absent [1], in contrast to the situation in Asia and especially Latin America [12]. We did not test other common larvicides such as Bacillus thuringiensis var israeliensis (Bti) or pyriproxyfen in this study, but there are currently no reports of resistance to either in Ae. aegypti [12], suggesting that multiple options for larval control exist in Burkina Faso, and likely elsewhere in mainland Africa.
In different parts of the world with longer histories of dengue control, Ae. aegypti populations are resistant to organophosphates [35–37], though inconsistencies in the diagnostic doses applied limit comparability [12]. In fact, the dose we applied for malathion is correct for Anopheles, but five-fold higher than that recommended for Ae. aegypti [27], although the recommended dose is very seldom applied [12]. Nevertheless, the dose used for fenitrothion is recommended for both Aedes and Anopheles [27], and, since full susceptibility was recorded for both insecticides, a conclusion of no resistance to organophosphates seems reasonable. Though uncommon, there are a few reports of relatively low-prevalence resistance to organophosphate adulticides from elsewhere in Africa [1]. Therefore, more detailed investigation of variation in the susceptibility profiles in adult female Ae. aegypti to malathion is recommended using the more informative dose-response methodologies. At present though, susceptibility to organophosphates in the localities we surveyed, coupled with the general rarity of organophosphate resistance in Africa, suggests that insecticides from this class are viable options for outbreak control. In contrast, we detected WHO-defined resistance to a carbamate, bendiocarb, in the urban (1200LG) and semi-urban (Tabtenga) sites, though not in the rural site (Goundry). Bendiocarb is less commonly used for Aedes control than organophosphates and pyrethroids, but a recent trial in Mexico found that bendiocarb was much more effective than deltamethrin for indoor residual spraying against a pyrethroid resistant population [38]. Full mortality in the 2h CDC bottle bioassays suggests that bendiocarb resistance is not at a high level in our survey sites, but it would still appear to be a less favourable option for adult control than organophosphates at present in Burkina Faso.
Populations from the urban and semi-urban areas showed moderate to very high resistance to both of the pyrethroid insecticides tested, and though at much lower prevalence, resistance was also detected to permethrin and deltamethrin in the rural site. From the WHO assays it was unclear whether resistance might differ between Tabtenga and 1200LG because results were almost identical when comparing adult females raised from collections from tires, but much lower mortalities were found in the collections from drums in Tabtenga. The longer duration CDC assays resolved this uncertainty, with significantly lower mortality in the Tabtenga than the 1200LG tire collections. Yet, in both the WHO and CDC assays, the Tabtenga drum collections showed significantly lower mortality than those from tires. The cause of this difference is unclear but seems most likely to be environmental, perhaps related to poorer developmental conditions in tires (e.g. lower food availability in these shaded habitats) or toxins leaching from the tires, although these have previously been linked to induction of P450s and potentially increased resistance in Ae. albopictus [39,40]. We are not aware of any previous demonstrations of such an effect of natural environmental variation on Ae. aegypti resistance, but given variation in the frequency of types of breeding sites found among areas [41], this could have an important impact on local resistance and deserves further investigation.
Pyrethroid resistance is found worldwide in Ae. aegypti [12], though the prevalence and higher-level resistance we detected in Tabtenga appears to be as strong as any yet reported from Africa [1]. The source of selection that may have driven resistance to this level is unclear. There is no history of targeted vector control for Ae. aegypti using pyrethroid insecticides in Burkina Faso, and the nature of the breeding sites means that run-off from agricultural application is a far less likely selective pressure than for Anopheles [42]. Increased insecticide pressure from malaria control interventions, is frequently linked to rising pyrethroid resistance in Anopheles [43–45]. Indeed in Goundry, Anopheles gambiae pyrethroid resistance and kdr mutation frequency increased between 2008 to 2011, which has been attributed to successive bednet distribution campaigns in Burkina Faso [46]. These may also have affected Ae. aegypti pyrethroid resistance in the three localities, although domestic use of insecticides may also constitute an important source of selection [47,48], especially in more affluent urban and semi-urban localities. Further work to identify sources of selection is clearly required if resistance management programs are to be a successful part of Aedes control programs.
We genotyped four possible kdr mutant positions in our survey. The S989P and V1016G mutants which are important for pyrethroid resistance in Asia [12], and also Saudi Arabia [30] were absent in our study site. The 1534C kdr mutation is common in Ae. aegypti and has a worldwide distribution [12]. We found this mutation to be almost fixed in the urban and semi-urban localities, though far less common in the rural area. This mutation is known to occur in neighboring Ghana [16] though the highest allele frequency reported there (60%) is much lower than in Tabtenga or 1200LG. Similarly, the V1016I mutation, which was only detected in a single individual in Ghana, [16] is much more common in the localities we surveyed (>20%), and again was significantly less common in Goundry, than in 1200LG and Tabtenga. Excluding Goundry, the frequencies of these mutations in our sites confirm results from recent (2017) collections in another urban area of Ouagadougou [34], and are very similar to those detected on the island of Madeira, where the Ae. aegypti population is thought to have been very recently introduced [22]. Apart from Ghana and the recent data from Burkina Faso, there are few other results from mainland Africa, though in central Africa, neither kdr mutation has yet been found, despite established pyrethroid resistance [21].
In South and Central America [49–51], and in the Caribbean [37] the co-occurrence of the 1534C and 1016I mutations is common and usually present as either (1) 1014V/1534C or (2) 1014I/1534C. Both of these haplotypes can confer pyrethroid resistance when present as a homozygote for haplotype 1, a heterozygote of haplotypes 1 and 2, and especially a homozygote of haplotype 2 [19]. In Goundry, these mutant genotypes comprised only 14% of the sample, but in the other sites the combined mutant genotype frequency was around 90%, although double mutant homozygotes were rare (<10%). The much higher frequencies of resistant genotypes in 1200LG and Tabtenga than Goundry are likely to explain a significant portion of the difference in permethrin and deltamethrin resistance.
Pre-exposure to PBO restored a substantial part of the susceptibility to permethrin and deltamethrin, most noticeably in Tabtenga where the resistance level (as measured in the 2h CDC assays) was highest. Although enhanced insecticide penetration may also be involved, the PBO result suggests the involvement of metabolic resistance mechanisms involving P450s, and perhaps esterases [52]. Metabolic resistance in Ae. aegypti mediated by P450s appears very common [12,17], and whilst several genes are frequently implicated in resistance, some of which are proven pyrethroid-metabolizers, the role of specific genes remains unclear [12]. We examined seven candidate P450 genes, most of which have been shown to metabolize pyrethroids, and including four from the geographically-closest population (Madeira) from which transcriptomic data have been obtained [22]. Whilst individually we did not detect strong or significant overexpression of individual genes, a trend of stronger overexpression was evident in the Tabtenga collection, which overall was significantly greater than the susceptible Rockefeller strain or the other two collection sites. Some of these genes may play a role in the metabolic resistance phenotype suggested by the strong action of PBO, which underpins variation between Tabtenga and 1200LG, but perhaps more as an aggregate overexpression than one dependent on specific genes. Alternatively, it is possible that the P450 genes probably important for resistance elsewhere, may be less so in African mainland populations and we did not assay the most important genes for metabolic resistance. Transcriptomic studies of Ae. aegypti from Africa will be required to help resolve this uncertainty.
Our results show a variable but alarmingly high level of pyrethroid resistance, underpinned by dual kdr mutations and metabolic resistance, perhaps involving some of the P450 genes we screened. Operational consequences of this resistance are unknown, but the use of pyrethroids for spraying as an outbreak control method would now appear to be unlikely to have significant impact. Moreover, with the source of selection unknown, return to susceptibility seems improbable. In contrast, susceptibility to both larvicidal and adulticidal organophosphates indicates that effective options for control still exist. Additional insecticide classes, and non-insecticidal interventions, should also be tested in order to operate a successful resistance management programme. Typically, insecticide resistance is considered a highly heritable trait, but our results also highlight how differences in breeding habitats can exert a strong influence on resistance. This phenomenon has been little-investigated and warrants further research.
|
10.1371/journal.pgen.1000300 | The Stringent Response and Cell Cycle Arrest in Escherichia coli | The bacterial stringent response, triggered by nutritional deprivation, causes an accumulation of the signaling nucleotides pppGpp and ppGpp. We characterize the replication arrest that occurs during the stringent response in Escherichia coli. Wild type cells undergo a RelA-dependent arrest after treatment with serine hydroxamate to contain an integer number of chromosomes and a replication origin-to-terminus ratio of 1. The growth rate prior to starvation determines the number of chromosomes upon arrest. Nucleoids of these cells are decondensed; in the absence of the ability to synthesize ppGpp, nucleoids become highly condensed, similar to that seen after treatment with the translational inhibitor chloramphenicol. After induction of the stringent response, while regions corresponding to the origins of replication segregate, the termini remain colocalized in wild-type cells. In contrast, cells arrested by rifampicin and cephalexin do not show colocalized termini, suggesting that the stringent response arrests chromosome segregation at a specific point. Release from starvation causes rapid nucleoid reorganization, chromosome segregation, and resumption of replication. Arrest of replication and inhibition of colony formation by ppGpp accumulation is relieved in seqA and dam mutants, although other aspects of the stringent response appear to be intact. We propose that DNA methylation and SeqA binding to non-origin loci is necessary to enforce a full stringent arrest, affecting both initiation of replication and chromosome segregation. This is the first indication that bacterial chromosome segregation, whose mechanism is not understood, is a step that may be regulated in response to environmental conditions.
| Management of cell growth and division in response to environmental conditions is important for all cells. In bacteria, nutritional downturns are signaled by accumulation of the nucleotide ppGpp. Amino acid starvation causes a programmed change in transcription, known as the “stringent response”; ppGpp also causes an arrest of cell cycle in bacteria, whose mechanism has not been thoroughly investigated. Here, we show that E. coli cells, when the stringent response is in effect, complete chromosomal replication but do not initiate new rounds and arrest with an integer number of chromosomes. The number of chromosomes corresponds to the growth rate prior to arrest. In polyploid arrested cells, the chromosomal regions at which replication initiates are segregated, whereas the termini regions remain colocalized. The E. coli chromosome remains decondensed and unsegregated during arrest and rapidly resumes replication and segregation, concomitant with chromosome condensation, upon release. The protein SeqA, a DNA binding protein and negative regulator of replication, is necessary for enforcing this arrest.
| Bacterial cells encounter varied environmental stresses and make appropriate adjustments to ensure survival. One such stress is amino acid starvation, which triggers physiological reprogramming known as the “stringent response”. Signaling of the stringent response is achieved by accumulation of effector nucleotides, guanosine tetra- and pentaphosphate ppGpp and pppGpp (reviewed in [1]), with the former the predominant and more stable of the two species. Two enzymes, RelA and SpoT, control the levels of ppGpp: RelA (PSI, or ppGpp synthetase I) synthesizes ppGpp in response to uncharged tRNA, the consequence of amino acid starvation; SpoT (PSII) possesses weak synthetase activity and is the sole hydrolase for ppGpp degradation.
The stringent response is RelA-dependent and is the most well-studied of stress responses signaled by ppGpp (reviewed in [1]). RelA associates with the ribosome and is triggered to produce ppGpp when the ribosome stalls with an uncharged tRNA in the acceptor site. This causes a dramatic alteration in gene expression, including the reduction of rRNA synthesis and increased transcription of amino acid biosynthesis genes [2]. These changes result from direct binding of ppGpp to RNA polymerase, aided by the transcription factor, DksA [3],[4]. Although it has been reported that ppGpp accumulation promotes arrest of replication, in addition to the well-studied transcriptional changes, the mechanism of the cell cycle arrest remains unclear [5],[6] and is the subject of our investigation.
The cell cycle of E. coli under low nutrient conditions is similar to that of eukaryotic cells, with a period prior to initiation of replication (“B period”, equivalent to eukaryotic G1), a period with ongoing DNA synthesis (“C period”, equivalent to S phase) and a period after completion of replication but before cell division (“D period”, equivalent to eukaryotic G2 phase). In medium rich in nutrients, E. coli cell cycle is accelerated such that it becomes faster than the time necessary to replicate the entire chromosome; under this circumstance, replication cycles overlap and cells are born with partially replicated chromosomes, whose replication was initiated in its mother or even grandmother [7].
In E. coli, replication initiates at the oriC locus and proceeds bidirectionally to completion within the terminator region, ter [8]. Replication initiation is tightly controlled by the AAA+ ATPase, DnaA [9]. Binding to oriC followed by loading of the DnaB helicase depends on levels of DnaA-ATP [9], correlated with achievement of a critical cell mass [10]–[12]. Firing of sister origins occurs in synchrony, such that cells contain 1, 2, 4, 8, 16, etc. copies of the oriC locus.
“Sequestration”, the binding of SeqA to a hemimethylated origin is a component of initiation control [13]–[15]. SeqA binds newly replicated, hemimethylated GATC sequences throughout the entire genome [16],[17]. Within oriC, SeqA binding prolongs its hemimethylated status and blocks DnaA-origin interaction until the region becomes fully methylated by Dam methyltransferase [13],[18],[19]. This contributes to the “eclipse period”, the time within a cell cycle when reinitiation is actively prevented [20], which is defective in mutants lacking SeqA and Dam [15], [21]–[23].
In addition, SeqA is implicated in chromosome segregation and nucleoid organization [24],[25]. In the absence of SeqA, DNA becomes more negatively supercoiled and nucleoids appear more decondensed [15],[26],[27]. Immunofluorescence microscopy of SeqA and GFP-tagged SeqA reveals foci colocalized with the replisome, consistent with SeqA binding to hemimethylated DNA as it emerges from the replication fork [24], [25], [28]–[30]. Under conditions of fast growth with overlapping replication cycles, SeqA promotes the colocalization of sister origins in E. coli, in a manner independent of sequestration at oriC [31]. SeqA overexpression also interferes with chromosome segregation [32].
It has been reported that stringent response in E. coli promotes replication arrest via inhibition of initiation [33]–[36]. Phase and fluorescent microscopy show that cells experience a reduction in cell size and contain a single nucleoid at mid-cell after ppGpp accumulation [34]. Initial reports showed limited replication at oriC after stringent onset [36] and later studies revealed that cells have a significant reduction, though not a complete block, in septum formation and cell division [34]. We reexamine replication arrest induced by the stringent response using flow cytometric techniques and implicate SeqA as a contributor to this arrest.
Our studies reveal that under the stringent response, cells appear to complete replication and arrest at an integer DNA content corresponding to their pre-arrest growth conditions. This arrest appears to be dependent on RelA-dependent ppGpp synthesis. Stringent cells always contain one decondensed nucleoid; in relA mutants that fail to arrest, nucleoid compaction is apparent, similar to that seen upon treatment of cells with the translational inhibitor, chloramphenicol. Although marker frequency analysis shows that the ratio of oriC to ter is 1 after stringent arrest, visualization of oriC and ter by ParB-GFP binding shows that sister loci at the termini remain colocalized, whereas the oriC loci have separated. Stringent cell cycle arrest is dependent on SeqA and Dam, in a manner that is only partly dependent on GATC sites near the origin.
To determine the extent of DNA replication in populations of Escherichia coli K12 cells growing in defined media, we used flow cytometry for DNA content using the stain PicoGreen (Invitrogen). We examined DNA content in cells at different growth rates, with and without treatment with serine hydroxamate, a derivative of serine that acts as a competitive inhibitor of seryl-tRNA synthetase, eliciting the stringent response [37]. This was compared to cells incubated with rifampicin and cephalexin, so-called “run-out” conditions that are used routinely to judge cell cycle status of E. coli. These antibiotics block initiation of replication and cell division, respectively; under this regimen, ongoing replication forks are completed and cells arrest in the D period. Under run-out conditions, the DNA content reflects the number of oriC loci in the cell at the time of treatment.
With no treatment, there is a broad distribution of cellular DNA content, indicative of DNA replication, asynchronous in the population (Figure 1). After treatment with rifampicin and cephalexin, replication is completed and cells assume an integer DNA content reflective of the growth rate. In low glucose medium, the majority of cells appear to be replicating and arrest at 2N DNA content after run-out; a subpopulation appears not to have initiated DNA replication at the time of rifampicin and cephalexin treatment and arrest at 1N. In higher glucose minimal medium, the cell cycle is accelerated such all cells appear to be replicating. Some arrest after run-out as 2N but a proportion of cells have initiated a second round of replication prior to completion of the first and arrest at 4N. In high glucose with casamino acids (CAA), all cells have initiated a second round, some with a third round, and arrest after run-out as a mixture of 4N and 8N cells.
In each growth medium, treatment with serine hydroxamate to induce the stringent response for 1.5 hr caused the population to assume an integer DNA content, similar to treatment with rifampicin and cephalexin, indicating an inhibition of cell cycle progression and DNA replication. In each case, the integer DNA content at stringent arrest was somewhat lower than that seen with rifampicin and cephalexin treatment. This indicates that some stringent arrested cells undergo cell division after cessation of replication. Although some stringent cells may arrest in D period (after completion of replication and prior to division, G2-like) as do rifampicin and cephalexin-treated cells, there is a preference for stringent arrest in B period (after division but prior to replication re-initiation, G1-like).
Under the fastest growth condition, we examined relA mutants, defective in ppGpp production elicited by serine hydroxamate. Mutants in relA were noticeably impaired for arrest after serine hydroxamate treatment and the cellular DNA content of the population was broadly distributed, indicative of ongoing replication (Figure 2A). In the presence of rifampicin and cephalexin, relA mutants assumed integer DNA content similar to wild-type cells, but with slightly more 8N relative to 4 N peaks. This finding confirms our expectation that ppGpp production by RelA is required for cell cycle arrest during the stringent response.
We also examined wild-type cells in which the stringent response was induced by high levels of a catalytically active truncated RelA fragment, RelA′. Overexpression of RelA′ causes ppGpp production, independent of idle ribosomes [38]. This method has the advantage of ppGpp accumulation independent of ribosome signaling and which is not accompanied by translational stalling. This allows us to distinguish ppGpp-specific effects from those associated indirectly with starvation or translation inhibition. If ppGpp accumulation is sufficient to induce cell cycle arrest, we would expect DNA profiles from pRelA′ induced cells to resemble those after treatment with serine hydroxamate. Indeed, cell cycle arrest and integer chromosome number was apparent by flow cytometry in IPTG-induced pRelA′ cells as compared to non-induced cells (Figure 2B). This confirms that accumulation of ppGpp, rather than inhibition of translation or starvation itself, is responsible for the arrest. We did note that the integer chromosome number after RelA′ overexpression was somewhat lower, with a more predominant 2N peak, and more asynchronous, with a strong 6N peak, than that seen for serine hydroxamate. This may be a consequence of higher ppGpp levels induced by RelA′ overexpression.
Because of the implication of the DNA binding protein SeqA in control of both chromosome initiation and segregation, we tested ΔseqA mutants for DNA content after induction of the stringent response. We found them to be virtually blind to serine hydroxamate treatment (Figure 2A) and DNA profiles resemble untreated cells. As has been noted previously [15],[39],[40], seqA mutants are also unable to assume integer DNA content after run-out with rifampicin and cephalexin, potentially because initiation is no longer sensitive to rifampicin or because ongoing forks fail to proceed to completion. The former explanation is suggested by the fact that the mean DNA content per cell in the seqA mutant cell population appears to increase substantially even after rifampicin/cephalexin treatment, as compared to untreated cells.
SeqA binds preferentially to GATC sites hemi-methylated by Dam (DNA adenine methylase) over unmethylated sites [20] and therefore we also tested dam mutants, which are expected to resemble seqA defective strains. We observed strains lacking Dam methylase showed a stringent cell cycle arrest defect similar to ΔseqA strains (Figure 2A). As seqA, dam mutants show no “run-out” after treatment with rifampicin and cephalexin (Figure 2A), although average cellular DNA content increases, suggesting rifampicin-resistant replication.
Because of SeqA's preference for hemi-methylated sites, Dam-overproducing cells also perturb SeqA binding to GATC sequences and cause defects in initiation control. Overproduction of Dam causes rapid conversion of hemimethylated DNA to fully methylated DNA, decreasing SeqA's ability to bind to these sites [13]. We would therefore expect Dam-overexpressing cells to exhibit defects in stringent cell cycle arrest, similar to seqA and dam mutants. To test this, we engineered otherwise wild-type cells to overexpress Dam from an arabinose-inducible promoter. This strain exhibited cell cycle arrest after treatment with serine hydroxamate when grown in the presence of glucose but failed to shift DNA content fully to an integer amount when Dam was induced by arabinose (Figure 2C). There was, however, evidence of a slight 2N peak in Dam-induced wild-type cells: this may represent a subpopulation that arrests replication because Dam levels are not sufficiently high or because they have lost the plasmid. After treatment with rifampicin and cephalexin, glucose-grown cells were primarily 2N and 4N whereas many arabinose-grown, Dam-overexpressing cells failed to assume integer DNA content. As seqA and dam mutants, Dam-overexpressing cells may exhibit rifampicin-resistant replication initiation that prevents “run-out” to integer DNA content.
DNA content is notably higher in seqA mutant strains due to loss of initiation restraint and subsequent over-replication [15]. To test whether the tendency to overinitiate obscured a functional stringent arrest in ΔseqA cells, we used a genetic suppressor of seqA in dnaA. The hypomorphic dnaA(Sx) allele lowers the frequency of initiation at cold-temperatures and suppresses the overinitiation property of ΔseqA mutants at its permissive temperature [40],[41]. We confirmed that dnaA(Sx) itself does not influence stringent cell cycle arrest. dnaA(Sx) cells displayed a stringent arrest after treatment with serine hydroxamate as indicated by a shift to an integer number of chromosomes, similar to run-out conditions (Figure 2A). When combined with the dnaA(Sx) allele, ΔseqA cells maintain a DNA content similar to wild-type cells in the absence of any treatment (Figure 2A). The dnaA(Sx) ΔseqA mutant did not successfully arrest after treatment with serine hydroxamate, although DNA content remained within the range that was comparable to arrested wild-type cells. This suggests that failed stringent arrest in ΔseqA cells is due to a lack of SeqA function and is not an indirect consequence of hyper-initiation or excessive DNA content.
SeqA appears to have two distinct functions: it binds to hemimethylated DNA sites at the origin and, somewhat more transiently, to newly replicated DNA throughout the chromosome. To distinguish between the origin and other sites of SeqA binding, we examined DNA content after stringent arrest in an oriCm3 strain, which is specifically defective in oriC sequestration. oriCm3 strains have eight of the GATC sequences within oriC converted to GTTC, causing a decreased affinity for SeqA, asynchronous replication, shortened eclipse periods and asymmetric cell division, equivalent to strains lacking SeqA altogether [39]. These mutations should not affect SeqA binding to hemi-methylated DNA revealed after replication of other regions of the chromosome. Flow cytometry of oriCm3 cells shows a modest increase in DNA content, presumably due to hyperinitiation caused by defective origin sequestration. After treatment with serine hydroxamate, the DNA content indicates that stringent arrest is partially intact in the oriCm3 strain, with a prominent 4N arrest peak. This suggests that there is a critical role of SeqA in enforcing the stringent response by interactions to DNA sites outside the origin. Mutants in oriCm3, like those in dam and seqA, showed poor run-out to integer chromosomal content after treatment with rifampicin and cephalexin.
To examine segregation patterns of specific locations of the chromosome, we used strains that contain a parS sequence inserted adjacent to the origin or to the terminus. The parS sequence provides a binding site for the plasmid-expressed fusion protein, GFP-ParB, to allow visualization of the locus of interest [42]. To enhance GFP fluorescence, these experiments were performed at lower temperature, 34°, and flow cytometry was performed on these cultures in parallel to discern their cell cycle status.
For growth in minimal medium plus casamino acids at 34°, examination of oriC and ter foci via ParB-GFP showed that stringently-arrested wild-type cells have two or four distinct oriC foci (Figure 3A and 3B). However, the number of visible ter foci was one or two, half that seen for oriC and similar to that seen for untreated cells. Marker frequency analysis by quantitative Southern blots for oriC and ter showed that with no treatment, the oriC to ter ratio was 2.5 (Figure 3C), indicating that replication was ongoing. After induction of the stringent response, the oriC/ter ratio in wild-type cells was 1.0 (Figure 3C), indicating that replication was complete. In contrast, relA mutant cells retained an oriC/ter ratio of greater than 2, even after serine hydroxamate treatment (Figure 3C). In similar analysis, the oriC/ter ratio of dnaA(Sx) seqA mutant cells was 2.8 before treatment and 2.7 after serine hydroxamate, confirming that replication is ongoing after induction of the stringent response in these strains.
The 2∶1 ratio of oriC to ter foci after serine hydroxamate treatment of wild-type cells, despite the fact that the chromosome appears to have completely duplicated from the marker frequency analysis, suggests that ter loci remain colocalized after replication. In contrast, colocalization of ter was not observed in wild-type cells arrested in the cell cycle by rifampicin and cephalexin: under these conditions the number of ter foci are equivalent to those seen for oriC (Figure 3A and 3B). This confirms that colocalization of ter seen during the stringent response is not likely to be a simple artifact of the ParB-GFP visualization system. Rather, there appears to be a distinct chromosome organization pattern enforced by the stringent response, and it is unlike that seen with other types of cell cycle inhibitors.
We examined the nucleoid morphology of stringent cells by DAPI-staining and fluorescence microscopy. Untreated wild-type cells growing in minimal CAA medium showed 1 or 2 nucleoids and had apparent signs of division and chromosome segregation (Figure 4A, left panel). DNA-free zones of the cytoplasm were apparent. After treatment with serine hydroxamate, signs of cell division were absent and all cells contained one nucleoid (Figure 4A, middle panel). The nucleoids appear decondensed, filling nearly the entire cell volume. This was quite different from cells treated with the translation inhibitor chloramphenicol, with a single, very condensed nucleoid structure at midcell (Figure 4A, right panel), consistent with previous reports [43]. From the observation of chloramphenicol-collapsed nucleoids, it has been argued that translation and anchoring of membrane proteins is required to extend the nucleoid throughout the cytoplasm. Despite a potential down-regulation of translation induced by serine hydroxamate, the nucleoid remained surprisingly decondensed, even more so than untreated cells. The decondensed nucleoid was not just a consequence of serine hydroxamate treatment: we saw similar appearance in cells in which the stringent response had been induced by pRelA′ overexpression (Figure 4B).
We also examined nucleoid morphology in ΔrelA strains (Figure 4A). Untreated ΔrelA nucleoids appear similar to wild-type; however, when treated with serine hydroxamate, nucleoids in ΔrelA cells have a condensed appearance, similar to cells treated with chloramphenicol. This may indicate that the absence of capacity to synthesize ppGpp causes a collapse in translational capacity upon serine hydroxamate treatment, reflected in nucleoid appearance. Or, it could mean that ppGpp induces some active nucleoid decondensation process that is absent in relA strains. Interestingly, mutants in seqA, despite their failure to arrest replication, appeared to retain the decondensed nucleoid character of wild-type cells after serine hydroxamate treatment. Although this is difficult to see in seqA strains because of their excessive DNA content, nucleoid decondensation after serine hydroxamate is also apparent in dnaA(Sx) seqA strains that have more normal DNA content per cell. This may indicate that seqA affects only a subset of the responses to ppGpp (see below). Mutants in oriCm3 also exhibited nucleoid decondensation similar to wild-type cells after serine hydroxamate treatment.
We wondered how cell cycle patterns would be reset after release from arrest. Would cells resume replication in a pattern similar to that prior to arrest? Or would cells be obligated to segregate each chromosome, divide and initiate replication from 1N progeny? To examine this, after 90 minutes of stringent arrest induced by serine hydroxamate, wild-type cells in minimal M9 Glucose CAA medium were washed and allowed to resume growth. DNA content was followed over time by flow cytometry and nucleoid appearance was monitored by DAPI staining (Figure 5A and 5B). As early as 15 minutes after release, some signs of nucleoid segregation were apparent. Segregation of two nucleoids appeared first at midcell; later segregation at the ¼ positions to form 4 nucleoids was detected. We think this reflects sequential and ordered segregation of chromosomes: sister-chromosome pairs held in cohesion segregate first, followed by separation of sisters (see schematic in Figure 5C). From 15 to 30 minutes, DNA content in the entire population appeared to increase, concomitant with nucleoid condensation and segregation. Therefore, cells appear to assume DNA content equivalent to conditions before the arrest, suggesting that replication patterns are reset quickly upon release.
Growth is inhibited in wild-type Escherichia coli cells with elevated ppGpp levels and they are unable to form colonies [44]. We wondered whether this is a consequence of cell cycle arrest induced by the stringent response and, moreover, if the stringent cell cycle arrest defects in Δdam and ΔseqA would permit colony formation during chronic ppGpp production. Wild type, ΔseqA, Δdam, and oriCm3 cells containing the IPTG-inducible pALS13 plasmid producing truncated RelA′, were grown on M9 minimal CAA ampicillin plates containing various concentrations of IPTG. A control plasmid, pALS14, contains a truncated 1.2 kb fragment of RelA that has no enzymatic activity [38]. Colony formation in wild-type cells was severely inhibited by induction of RelA′ (Table 1) but not with the nonfunctional RelA- protein. Similar inhibition of plating was obtained by induction of pALS10, expressing wild-type RelA+ (data not shown). ΔseqA and Δdam cells, however, maintained the ability to form some colonies in the presence of elevated ppGpp concentrations, with plating efficiencies dramatically elevated, over 1000-fold, in the presence of IPTG (Table 1, and data not shown). On the other hand, oriCm3 cells had only a modestly increased plating efficiency of about 10-fold. This is consistent with our flow cytometric results and suggests that an origin compromised for SeqA interaction has only a partial defect in the stringent cell cycle arrest, considerably less severe than strains lacking all SeqA function or Dam methylase. Cells that successfully formed colonies under stringent response induction in ΔseqA or Δdam genetic backgrounds were no more resistant than the original population upon rechallenge with IPTG (data not shown), indicating that no heritable change arising during growth, which is a special concern for dam mutator strains, had allowed them to escape preferentially.
The inability of cells to form colonies under constitutive ppGpp accumulation appears to be, in part, a consequence of cell cycle arrest. We do note that the rescue of plating by seqA and dam is not 100%, indicating that transcriptional reprogramming, translational inhibition or other aspects of the stringent response, which remain intact in seqA and dam mutants (see below), may also be inhibitory to growth and division.
One possible explanation for the inability of Δdam and ΔseqA strains to arrest cell cycle during the stringent response is that production of ppGpp, or its ability to interact with RNA polymerase, for some reason, is defective. To investigate this, we examined ppGpp production directly by 32P-phosphate labeling and separation of nucleotides by thin-layer chromatography (Figure 6A and 6B). The more slowly-migrating form corresponding to ppGpp was apparent in wild-type, seqA, dam and oriCm3 strains after treatment with serine hydroxamate; this was not seen for relA strains.
Another characteristic of stringent control in wild type cells is the reduction of stable RNA synthesis [44]. Uridine uptake is also reduced by induction of the stringent response [45]. By pulsing cells with [3H]-uridine, treated with or without serine hydroxamate, we examined the amount of newly labeled rRNA synthesis over time (Figure 6C). Wild-type cells treated with serine hydroxamate display a decrease in labeled rRNA compared to untreated cultures. ΔrelA cells, defective in stringent control, maintain labeled rRNA production at high levels, blind to treatment with serine hydroxamate. Like wild-type, ΔseqA and Δdam cells showed reduced labeled rRNA after serine hydroxamate addition.
Because the latter experiment reflects both down-regulation of rRNA synthesis, as well as reduction in uridine uptake, we also examined pulse-labeling of rRNA with inorganic phosphate, relative to total levels as determined by ethidium bromide staining, which should be more reflective of rRNA synthesis. Labeling of rRNA decreased over 2-fold after treatment with serine hydroxamate in wild-type cells and 15-fold in ΔseqA mutant cells; incorporation of 32P into rRNA was not strongly affected by serine hydroxamate treatment in relA mutants that fail to synthesize ppGpp. These experiments suggest that seqA and dam mutants have no defect in production of ppGpp or other aspects of the stringent response, including transcriptional down-regulation of rRNA synthesis and possibly uridine uptake, and therefore are specifically impaired in cell cycle response to ppGpp.
In this study, we have characterized the replication arrest that occurs in Escherichia coli cells under the stringent response. We confirm a previous report that initiation but not elongation of replication is blocked and that RelA-dependent ppGpp synthesis is required for full arrest [34]. The majority of cells appear to arrest in the B period, prior to initiation, and contain an integer number of chromosomes appropriate for the growth medium prior to arrest. A minority of cells may arrest in the D period, after replication is completed but prior to chromosome segregation. In either case, stringent E. coli cells have unsegregated, but apparently fully-replicated, chromosomes, indicating a new cell cycle point of stringent control.
We have identified mutants, ΔseqA and Δdam, that cannot control replication in response to ppGpp and continue to divide in the presence of chronic stringent induction that is sufficient to inhibit growth of wild-type strains. Overproduction of Dam methylase is also sufficient to uncouple the control of cell cycle from the stringent response, suggesting that stringent cell cycle arrest is regulated by SeqA interactions with hemi-methylated GATC sites. Other aspects of the stringent response, including down-regulation of rRNA synthesis, decondensation of nucleoids, and a diminution in cell size (data not shown), appeared to be intact in seqA mutants. This implies that SeqA and Dam specifically control cell cycle and do not have global effects on the stringent response, including levels of ppGpp.
In bacteria, as well as other organisms, control of the cell cycle is likely to be an important component of stress responses. Our previous work suggested that SeqA may limit replication in the face of chronic DNA damage [40]. In the case of the stringent response, it may be advantageous to limit replication in the presence of diminishing translational capacity since stalled forks may be subject to cleavage and collapse, causing the demise of both chromosomes. Arrest with two or more replicated chromosomes provides an opportunity for recombinational repair if one of the chromosomes becomes damaged and two potential templates for gene expresssion. Decondensation of the nucleoid, potentially induced by the stringent response, may facilitate the dynamic interactions between sister chromosomes that are required for recombinational repair.
Bacterial cells fail to proliferate when ppGpp levels rise, such as caused by overexpression of RelA or loss of the ppGpp hydolase SpoT (reviewed in [1])–our experiments suggest that this is due in large part to SeqA-dependent arrest of cell cycle. Mutants in dam or seqA are dramatically relieved of growth inhibition by overexpression of RelA and are able to form colonies, albeit with some loss of plating efficiency, likely the result of down-regulation of translational capacity [46].
Previous results suggested that initiation of replication is blocked when ppGpp levels are high and that elongation of replication continues to completion after ppGpp accumulation [34]–[36]. Our findings confirm this and suggest that chromosome segregation is an additional point of stringent regulation. Stringent cells arrest with unsegregated nucleoids with colocalized termini. This is consistent with reports showing that cell division, though not completely inhibited, does not likely proceed until each cell has 1 chromosome [34].
Stringent cell cycle arrest points differ between Escherichia coli and Bacillus subtilis. B. subtilis arrests replication elongation by ppGpp inhibition of primase activity [47]. It is unclear why E. coli does not perform this C-period (“intra-S phase”-like) arrest, although it remains possible that elongation is slower in stringent cells or punctuated with stalling events. We also do not know the threshold levels of ppGpp or starvation that elicit replication arrest in the two organisms. It remains possible that elongation arrest in E. coli could occur under more stringent conditions or occurs in a subpopulation of cells, obscured by arrest in B and D periods. During the stringent response in E. coli, GTP levels do not fall more than 50% [46]; in contrast, Bacillus subtilis experiences a drop in GTP concomitant with ppGpp accumulation and its stringent response may be an indirect consequence of depletion of GTP pools [48]. For this reason, E. coli may respond more sensitively to ppGpp, enabling it to complete replication before collapse of nucleotide pools that would stall replication. E. coli and its relatives are unique in their acquisition of adenine methylation and SeqA function, perhaps allowing an extra level of control of replication and chromosome segregation, lacking in other bacteria.
The origin region is particularly rich in GATC sites and SeqA binding to these sites is relatively long-lived, up to 1/3 of the cell cycle, when the sites remain hemimethylated [18],[23]. Using a mutant in many of the GATC sites near the origin, oriCm3 [39], we addressed the role of the methylation pattern of oriC during the stringent response. Our data indicates that in this mutant stringent arrest is partially intact. Some arrest, as judged by flow cytometry for DNA content, was seen in this mutant upon serine hydroxamate treatment. During chronic exposure to ppGpp, this mutation relieved inhibition of colony formation only modestly, approximately 10-fold, relative to wild-type strains. This was in contrast to the ΔseqA strain, in which no evidence of replication arrest was apparent and which was 1,000-fold more efficient in colony formation during chronic exposure to ppGpp. It is possible that the oriCm3 mutation only produces a partial loss of SeqA binding to the origin, although the first characterization of this mutant suggests that loss of sequestration at oriC is complete [39]. A recent study [31] shows that transient colocalization of sister origins in cells with overlapping replication cycles depends on SeqA in a manner that is not affected by oriCm3 but dependent on the property of SeqA for self-aggregation. It may be this mode of SeqA binding that is required for stringent arrest.
This study provides new evidence that late stages of chromosome replication or chromosome segregation can be regulated in response to environmental conditions. The nature of this arrest is intriguing and suggests possible mechanisms. We observed segregated oriC regions marked by GFP-ParB but only one half the predicted number of ter regions. In contrast, after cell cycle arrest by rifampicin and cephalexin, the number of observed ter foci is equivalent to those of oriC. Loss of ter cohesion may be a prerequisite for chromosome segregation and a point of regulation by the stringent response.
Interestingly, the nature of stringent arrest of chromosome segregation is very similar to the arrest seen in cells depleted of the GTPase ObgE, which is required for chromosome segregation and survival after treatment with replication inhibitors [49],[50]. Both the Bacillus subtilis Obg and E. coli ObgE bind ppGpp ([51]; Persky and Lovett, unpublished results); E. coli Obg (also known as CgtA) interacts with ppGpp synthetase/hydrolase SpoT [52] and CgtA may regulate SpoT hydrolase activity in Vibrio cholerae [53]. One possibility is that Obg controls ppGpp levels; alternatively, Obg may be required to license cell cycle progression in a manner that is inhibited by ppGpp binding.
Replication and chromosome segregation are concurrent processes in growing E. coli cells and loci segregate as they are duplicated. In some studies [54],[55], there appears to be a delay of segregation of ter relative to other regions of the chromosome, suggesting that there could be special factors that control its segregation. The 2∶1 ratio of oriC to ter foci may suggest a post-replication cohesion, either because regions near ter have not fully duplicated or because fully-replicated sister chromosomes are held together by DNA or protein linkages. Although the marker frequency analysis showing a 1∶1 oriC/ter ratio appears to support the latter explanation, we cannot rule out the possibility that short or heterogeneous unreplicated sequences in the ter region after ppGpp accumulation escape our detection. In any case, this “cohesion” may assist in the organization of chromosome segregation that will occur when cells are released from arrest with multiple chromosomes. We observed a rapid segregation of 4 nucleoids upon release, concomitant with nucleoid compaction, with segregation occurring first at midcell and later at the quarter position. Cohesion at the termini (Figure 5C) may restrain segregation of recent sister chromosomes until all others have segregated, a means of enforcing sequential patterns of segregation.
In vitro, SeqA wraps DNA, producing positive supercoiling and can form cooperative self-aggregates [56]. This aggregative property, specifically altered in seqA N-terminal mutants, is implicated in the formation of visible foci colocalized to the replication fork and for promoting organization of the origin [31],[57]. The effects exerted by SeqA in origin-proximal cohesion could act in the same manner at the termini to promote their cohesion. We do note that, as ter is the last region to be replicated, its hemimethylated GATC sites should be bound by SeqA prior to segregation, giving the opportunity for SeqA to control separation and segregation of this region of the chromosome. Chromosome cohesion by prolonged binding during ppGpp accumulation could, in turn, signal a block to cell division.
Alternatively, apparent cohesion of sister chromosome could be mediated by DNA topological links. SeqA interacts with Topoisomerase IV, with the potential to modulate decatenation of the replicated chromosomes [58]. In vitro, at moderate levels of SeqA, decatenation and relaxation by Topo IV is stimulated by specific recruitment of the enzyme. At high levels, SeqA promotes catenane formation by Topo IV by promoting intermolecular aggregates. In either case, an increased probability of intertwined, catenated chromosomes upon induction of the stringent response could explain failure of the terminus regions of the chromosome to segregate after replication.
Although our genetic analysis places SeqA and Dam in the cell cycle aspect of the stringent response, the direct connection between ppGpp, changes in the transcriptional program and modulation of SeqA or Dam methylase activity remains to be elucidated. Neither SeqA nor Dam bind guanine nucleotides; therefore some factor responsive to ppGpp is implicated in their control. An obvious candidate is RNA polymerase. Although transcription is required to initiate replication (and is therefore blocked by the RNA polymerase inhibitor, rifampicin), the mechanism of this control is still obscure, since promoter activity at oriC is not required for initiation under normal conditions [59]. We note that many of the strains defective in stringent control of replication, such as seqA, dam, oriCm3 and dam overexpressors, may also exhibit rifampicin-resistant replication, as evident in run-out experiments, suggesting some connection between the two phenomena. An attractive model is that transcription sensitive to ppGpp may be required to reverse inhibitory effects of SeqA on replication initiation and chromosome segregation. This could be because SeqA binding blocks some DNA site required for cell cycle progression, with the act of transcription of this locus directly reversing this. Or alternatively, ppGpp-sensitive transcription of some unknown gene may be required to dissociate SeqA.
Previous reports have shown that DnaA expression decreases during the stringent response [35]. Although this may play some role in stringent control of replication initiation, this alone appears unlikely to mediate all stringent response effects on cell cycle. We were not able to suppress seqA defects in stringent arrest by dnaA mutants with reduced initiation efficiency. Moreover, DnaA is unlikely to directly influence chromosome segregation arrest during the stringent response. Nonetheless, it is possible that multiple mechanisms of cell cycle control, such as the DnaA levels and the regulatory inactivation of DnaA (RIDA), cooperate to control cell cycle in response to ppGpp.
Escherichia coli K-12 strains (Table 2) were grown at 30°C, 34°C, 37°C as previously described on Luria-Bertani (LB) medium, consisting of 1% Bacto Tryptone, 0.5% yeast extract, 0.5% sodium chloride and, for plates, 1.5% agar or in M9 minimal medium (48 mM Na2HPO4-7H2O, 22 mM KH2PO4, 8.5 mM NaCl, 19 mM NH4Cl, 2 mM MgSO4 and 0.1 mM CalCl2) with 0.2% glucose, 0.4% glucose, or 0.4% arabinose, and 0.2% casamino acids with 1.5% agar for plates. For P1 transductions and phage lysates, cultures were grown in LCG, LB medium supplemented with 1% glucose with an additional 2 mM calcium chloride; for plates, 1% agar was added. Antibiotics were used in the following concentrations: ampicillin (Ap), 100 µg/ml; kanamycin (Km), 60 µg/ml; tetracycline (Tc) and chloramphenicol (Cm), 15 µg/ml. Isogenic strains in MG1655 were constructed by P1 vira transduction. Cultures employed M9 minimal media described above with the addition of 1 mg/ml DL-Serine hydroxamate (Sigma) or, in the case of strains harboring pALS13 (pRelA′), 1 mM isopropyl β-D-1-thiogalactopyranoside IPTG for 1.5 hrs. Addition was made in early logarithmic growth of the culture, at OD600 of ∼0.2. Experiments examining chloramphenicol-induced translational inhibition were performed with the addition of 300 µg/ml chloramphenicol for 1.5 hrs. Release from DL-serine hydroxamate was accomplished by centrifugation of the treated culture and resuspension of the cell pellet in an equal volume of fresh medium without the drug. For experiments employing radioactive inorganic phosphate labeling, MOPS-buffered minimal medium was used (50 mM MOPS pH 7.2, 43 mM NaCl, 93 mM NH4Cl 1 mM MgSO4, 3.6 µM FeSO4-7H2O, 0.12 mM CaCl2 and either 2.2 mM or 0.4 mM KH2PO4).
DNA content per cell was determined as described [49]. Briefly, 1 ml of culture was fixed in 9 mls of 70% ethanol and stored at 4°C until staining. For staining, fixed cultures were resuspended in 1 ml phosphate-buffered saline (PBS) pH 7.4. The samples were incubated with 100 µl PicoGreen dye (Invitrogen), diluted in 1∶100 in 25% DMSO for 3 hr at room temperature, and then further diluted with an additional 1 ml PBS containing PicoGreen (1∶1000). Cultures were analyzed by using a FACSCalibur flow cytometer and FloJo 6.4.1 software. As a control for chromosome number, a stationary phase wild type culture and an isogenic dnaA46 strain, temperature sensitive for replication initiation, was analyzed similarly after 3 hr of growth at 42°C.
Overnight cultures were inoculated into fresh medium at a dilution of 1∶100 and grown with aeration for 3 hours. Nucleoid staining was performed as previously described [28]. For DAPI staining alone, cultures were fixed in 3∶1 methanol acetic acid. 10–20 µl of cells was placed on poly-L-lysine hydrobromide (1 mg/ml)-coated slides and air-dried. Cells were washed 3 times with 1× Phosphate-Buffered-Saline (PBS [pH 7.4]) and allowed to air dry. Cells were then stained with 10 µl of 10 µg/ml DAPI (4′,6′ diamido-2-phenylindole) for 10 min and washed three times with PBS and mounted in 1 mg/ml p-phenylenediamine 90% glycerol in PBS, mounted with 5 µl VectaShield mounting medium and analyzed as described below. Living cells harboring pALA2705 (ParB-GFP) were grown at 34°C without supplementation of isopropyl β-D-1-thiogalactopyranoside (IPTG) to induce synthesis of the GFP fusions. A suspension of growing cells was added to a 2% agarose pad (MP Biomedicals, Inc) and covered with a cover slip over the residual media. Slides were analyzed with an Olympus BX51 microscope equipped with a RGB liquid crystal color filter. Images were acquired with a Qimaging Retiga Exi camera by using the manufacturer's software. Foci counts were obtained using Openlab Darkroom imaging software (Improvision, Coventry, United Kingdom) and edited with Openlab and Adobe Photoshop Elements 4.0.
Cells were grown to exponential growth phase in M9 0.4% glucose CAA medium. Samples were taken and chromosomal DNA was extracted using the MasterPure DNA Purification Kit (Epicentre). Restriction digest and preparation of the probe were done as described previously [60]. Briefly, the chromosomal DNA was triple digested with EcoRI, HindIII, and EcoRV and the fragments were separated on a 1.0% agarose gel. The DNA was vacuum-transferred to a nylon membrane (Amersham). The membrane was prehybridized with BSA for more than 1 h at 65°C and hybridized overnight at 65°C with 32P a-dATP. The probe consisted of two DNA fragments that anneal to the chromosomal regions gidA (84.3 min), and relB (34.8 min). The DNA fragments were labeled using Random Primer Labeling (Molecular Cloning). After hybridization, the membrane was washed with 0.21 M Na2HPO4 (pH 7.3) / 6% SDS / 0.85 mM EDTA 3× for 10 minutes at 25°C followed by 2× for 5 minutes at 65°C. The membrane was exposed on a Phophoimaging screen (Molecular Dynamics) and scanned on a Bio–Rad Molecular Imager FX (Bio-Rad). Analysis of bands was carried out using Quantity One imaging software (Bio-Rad). Normalization of the bands was accomplished using genomic DNA from dnaA46 that was grown at the non-permissive temperature for 2 h.
Overnight cultures grown in M9 CAA media were diluted 1∶100 in fresh media and grown to an OD600∼0.3. 10-fold serially diluted cultures were plated on M9 minimal media plates containing ampicillin and 100 µM IPTG. Total colony numbers were determined by plating on M9 ampicillin medium without IPTG. Colonies were counted after 48 hours of growth at 37°C. The number of independent cultures, as indicated in figure legends, was determined on at least three different days.
Overnight cultures were diluted 1/50 in M9 CAA media and grown to an O.D.600∼0.2. Where indicated, cells were treated with serine hydroxamate. Pulse labeling was initiated by the addition of 20 µCi of [3H] uridine per 2 ml of culture. Uracil was added to 0.9 mg/ml after a ten minute pulse. Total RNA was extracted from 0.5 ml by the RNAeasy kit (Qiagen). Purified RNA was resuspended in RNase free water and stored at −20°C. Approximately 1 µg of RNA sample was analyzed by electrophoresis in a 1% non-denaturing agarose gel. After electrophoresis, the gel was photographed under UV light. Amount of rRNA was determined by band intensity of the 16S, 23S, and unprocessed rRNA bands in the image using Quantity Oneâ imaging software (Bio-Rad). To determine the amount of [3H] labeled rRNA, corresponding rRNA bands were cut out from the gel, dissolved, and analyzed in a scintillation counter. Normalization was achieved by dividing the obtained cpm by the volume of rRNA band intensity.
Overnight cultures grown in MOPS pH 7.2 medium containing 2 mM phosphate (KH2PO4) and 0.2% casamino acids were diluted 1∶100 into MOPS medium containing 0.4 mM phosphate and 0.2% casamino acids. Treated samples received 1 mg/ml SHX before the addition of [32P]H3PO4 to a final concentration of 200 µCi/ml when the culture OD600 reached ∼0.05. Total RNA was isolated after 15 minutes of incubation with [32P]H3PO4. RNA was isolated using the RiboPure Bacteria Kit from Ambion according to the manufacturer's instructions. Total RNA was electrophoresed on a native agarose gel and stained with ethidium bromide. rRNA bands were visualized and quantified by UV- light induced fluorescence using Quantity One software (Bio-Rad). RNA was then transferred to a nylon membrane (Amersham) using the Vacuum Blotter 785 (Bio-Rad). The amount of radiolabeled rRNA was determined by autoradiography and quantified using MolecularImager FX PhosphorImager and Quantity One software (Bio-Rad). Amount of radiolabeled rRNA was normalized to the amount of RNA observed by UV-light induced fluorescence.
(p)ppGpp measurements were performed as previously described [61]. Overnight cultures grown in MOPS minimal medium containing 0.4% Glucose 2 mM phosphate (KH2PO4) and 0.2% casamino acids were diluted 1∶100 into the same MOPS medium except containing 0.4 mM phosphate. [32P]H3PO4 was added to a final concentration of 100 µCi/ml when the OD600 reached ∼0.05 and cultures grew for an additional 3 hours before the first sample was taken. SHX was added at time 0, and samples were isolated every ten minutes by mixing 100 µl of culture with an equal volume of 13 M formic acid and chilling on dry ice. The samples were subjected to two rounds of freezing and thawing before microcentrifugation at 14, 000 rpm for 2 minutes to remove cellular debris. 6 µl of supernatant were spotted onto 20×20 cm polyethyleneimine cellulose on polyester TLC plates (Sigma) in 1.5 KH2PO4 (pH 3.4) for 2 h. After chromatography, nucleotides were visualized by autoradiography and quantified with a MolecularImager FX PhosphorImager and Quantity One software (Bio-Rad). Unlabeled GDP and GTP were spotted on the plates as markers and visualized after chromatography by UV light-induced fluorescence. The identities of the labeled (p)ppGpp were inferred from their positions in the chromatograph relative to the origin and GTP. (p)ppGpp levels are normalized to levels of GTP observed in the same sample.
|
10.1371/journal.pgen.1000427 | Sfrp Controls Apicobasal Polarity and Oriented Cell Division in Developing Gut Epithelium | Epithelial tubular morphogenesis leading to alteration of organ shape has important physiological consequences. However, little is known regarding the mechanisms that govern epithelial tube morphogenesis. Here, we show that inactivation of Sfrp1 and Sfrp2 leads to reduction in fore-stomach length in mouse embryos, which is enhanced in the presence of the Sfrp5 mutation. In the mono-cell layer of fore-stomach epithelium, cell division is normally oriented along the cephalocaudal axis; in contrast, orientation diverges in the Sfrps-deficient fore-stomach. Cell growth and apoptosis are not affected in the Sfrps-deficient fore-stomach epithelium. Similarly, cell division orientation in fore-stomach epithelium diverges as a result of inactivation of either Stbm/Vangl2, an Fz/PCP component, or Wnt5a. These observations indicate that the oriented cell division, which is controlled by the Fz/PCP pathway, is one of essential components in fore-stomach morphogenesis. Additionally, the small intestine epithelium of Sfrps compound mutants fails to maintain proper apicobasal polarity; the defect was also observed in Wnt5a-inactivated small intestine. In relation to these findings, Sfrp1 physically interacts with Wnt5a and inhibits Wnt5a signaling. We propose that Sfrp regulation of Wnt5a signaling controls oriented cell division and apicobasal polarity in the epithelium of developing gut.
| The gastrointestinal tract is generated from the primitive gut tube during embryogenesis. The primitive gut differentiates regionally along the cephalocaudal axis. Individual regions simultaneously acquire specific morphologies through morphogenetic mechanisms. The regional specification of the gut tube is controlled by cross-talk between the mesenchyme and epithelium. However, the morphogenetic mechanisms governing gut formation remain poorly understood. Secreted Frizzled-related protein (Sfrp) is an inhibitor of the Wnt pathway, members of which are expressed in the developing gut. A deficiency of Sfrp genes (Sfrp1, Sfrp2, and Sfrp5) results in reduction of fore-stomach length in mice. During normal fore-stomach formation, cell division is oriented along the cephalocaudal axis; in contrast, reduced fore-stomach length in Sfrps-deficient mice is associated with the divergence of oriented cell division in tubular epithelial cells. Thus, oriented cell division is one of the essential components in fore-stomach morphogenesis. In addition, Sfrps-deficient small intestine epithelium fails to maintain proper apicobasal polarity. We also found that Wnt5a-inactivation leads to a phenotype similar to that induced by Sfrps-deficiency in the developing gut, and that Sfrp1 inhibits Wnt5a signaling. We propose that Sfrp regulation of Wnt5a signaling is required for oriented cell division and that it modulates apicobasal polarity in gut epithelium during organ elongation.
| Generation of the gastrointestinal (GI) tract is initiated by formation of the primitive gut tube during embryogenesis. Subsequently, this tube differentiates regionally along the cephalocaudal axis, giving rise to the esophagus, stomach, small intestine and colon, as well as acquiring specific morphologies, which are generated through morphogenetic mechanisms. Regional specification of the gut tube involves interactions between splanchnic mesoderm and endoderm epithelium [1]. However, the morphogenetic mechanisms governing gut formation remain poorly understood.
Wnt family members are secreted glycoproteins that play important roles in controlling tissue patterning, cell fate, cell proliferation and tissue morphogenesis [2] (http://www-leland.stanford.edu/̃rnusse/wntwindow.html). Wnts are classified into two groups [3]. Wnt1 class ligands (e.g., Wnt1, Wnt3a and Wnt8) activate the canonical Wnt/ß-catenin pathway, which stabilizes ß-catenin as a transcriptional regulator in the nucleus [2],[3]. Wnt5a class ligands (e.g., Wnt5a and Wnt11) stimulate non-canonical Wnt pathways, such as the Ca2+ and Fz/PCP pathways, through the Frizzled receptor [3]. Although the role of Wnt signaling in the developing gut is ill-defined, a number of Wnts, Fzs, and their inhibitors, especially Sfrps, are expressed in the tissue [4],[5].
Secreted Frizzled-related protein (Sfrp) is a secreted Wnt antagonist that interacts directly with the Wnt ligand [6]. The Sfrp gene family, which consists of five members in both the human and mouse genomes, is classified into the Sfrp1 (Type 1) and FrzB subfamilies based on amino acid sequence similarity [6]. Sfrp1, Sfrp2 and Sfrp5 comprise the Sfrp1 subfamily (referred to as Type 1 Sfrps) [6]. Type 1 Sfrps inhibit the Wnt/ß-catenin pathway in vitro. Type 1 Sfrps exhibit characteristics of Wnt inhibition that differ from those of FrzB Sfrps (Sfrp3 and Sfrp4), a phenomenon that can probably be attributed to Wnt ligand specificity [7],[8]. Genetic analysis has revealed the functional redundancy of Sfrp1, Sfrp2 and Sfrp5; moreover, Sfrp1/2/5 genetically interact with Stbm/Vangl2 (also known as Ltap), an ortholog of Drosophila Strabismus/Van Gogh Fz/PCP core component [9]. These observations suggest a redundant role for Type 1 Sfrps in the regulation of the Wnt/ß-catenin and the Fz/PCP pathways.
The body axis of Sfrp1/2/5 compound mutants is shortened [9]. This observation suggests that a concomitant shortening of the axial visceral organs, i.e., the GI tract, may occur. Therefore, we focused on the forming gut tube and epithelial morphogenesis. Our results suggest that Sfrp-regulation of Wnt5a signaling is required for the regulation of epithelial cell polarity, oriented cell divisions and apicobasal (AB) polarity, and lengthening of the developing gut.
During mouse embryonic development, the primitive gut tube is generated by embryonic day (E) 9. Subsequently, the gut tube develops the organ buds of the lung, stomach, liver and pancreas, which are apparent at E10.5. In the developing gut, Sfrp1, Sfrp2 and Sfrp5 are regionally expressed along the cephalocaudal axis. At E10.5, Sfrp1 was expressed in the splanchnic mesoderm from the caudal region of the prospective stomach to the midgut. At E12.5, expression was observed in the mesenchyme of the colon, as well as in the caudal region of the fore-stomach and the small intestine (Figure S1A, D, G, J, N). Sfrp2 was expressed in the splanchnic mesoderm of the prospective esophagus at E10.5. Later, at E12.5, Sfrp2 expression expanded to the rostral region of fore-stomach mesenchyme (Figure S1B, E, H, K, N). Sfrp2 expression was also detected in colon epithelium at this stage (Figure S1L). In addition, Sfrp5 expression was present in endoderm cells of the presumptive midgut region at E8.75 [10]. Sfrp5 expression remained in evidence in the midgut endoderm at E9.5, a stage lacking obvious expression of Sfrp1 and Sfrp2 in the gut tube (data not shown). During the later stages of E10.5–12.5, Sfrp5 expression was observed in the duodenum epithelium (Figure S1C, F, I, M, N). Despite expression in the developing gut, no obvious morphological abnormality was identified in the gut of Sfrp1, Sfrp2 and Sfrp5 single knock-out embryos as far as we examined [9],[11],[12], possibly because, as in other tissues, the long-range effect of an Sfrp as a secreted factor can compensate for the function of other Sfrps in those mutants [9],[12].
In order to establish the redundant role of Sfrp1, Sfrp2 and Sfrp5 in gut formation, the gut tube was examined in Sfrp1 subfamily compound mutant mice. Sfrp1 and Sfrp5 (Sfrp1−/− Sfrp5−/−) and Sfrp2 and Sfrp5 (Sfrp2−/− Sfrp5−/−) double homozygous mutants appeared to be normal in terms of GI tract formation. In contrast, embryos carrying a double homozygous mutation in both Sfrp1 and Sfrp2 (Sfrp1−/− Sfrp2−/−) displayed severe shortening of the gut tube, e.g., smaller stomach and shorter intestine, at E13.5 (Figure S2B, E). The reduction in the length/size of the gut in the E13.5 Sfrp1−/− Sfrp2−/− embryos was enhanced in the presence of an Sfrp5 heterozygous mutation (Sfrp1−/− Sfrp2−/− Sfrp5+/−), which is suggestive of a redundant role for Sfrp1, Sfrp2 and Sfrp5 in gut formation (Figure S2E, F).
To gain insight into the defect in stomach formation, we examined regional marker expression, Shh [13], Pitx1 [14] and Nkx6.3 [15], in the epithelium of Sfrp1−/− Sfrp2−/− and Sfrp1−/− Sfrp2−/− Sfrp5+/− stomachs at E13.5. Shh was expressed in fore-stomach and intestinal epithelium of control (Wild-type, Sfrp1−/−, Sfrp1−/− Sfrp2+/−, Sfrp1−/− Sfrp5+/− and Sfrp1−/− Sfrp2+/− Sfrp5+/−) embryos, while diminished Shh expression was evident in the hind-stomach (Figure 1A). Pitx1 expression was strong in fore-stomach epithelium, but weaker in the hind-stomach (Figure 1B). Nkx6.3 expression was observed specifically in epithelium extending from the caudal hind-stomach to the duodenum (Figure 1C). All of these epithelial markers were expressed in the stomach of Sfrp1−/− Sfrp2−/− and Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos; however, Shh and Pitx1 expression exhibited a significant reduction in terms of the size of the fore-stomach (carved arrows in Figure 1A, B). Furthermore, the size reduction of the fore-stomach was enhanced by an Sfrp5 heterozygous mutation in an Sfrp1−/− Sfrp2−/− background (Figure 1A, B). As suggested by the negative region of Shh expression (the region indicated by a broken line in Figure 1A), the region marked by Nkx6.3 expression (Figure 1C) and expression of Islet1 [16] in hind-stomach mesenchyme (Figure 1D), the hind-stomach at E13.5 appeared to be unaffected by the mutations.
Epithelial specification in the gut is tightly controlled by cross-talk between splanchnic mesoderm and endoderm epithelium [17]. The stomach of compound mutant embryos demonstrated normal expression of Islet1 (Figure 1D) and Barx1 [18], which are specific markers for stomach mesenchyme (Figure 1E). The non-glandular stomach of compound mutant embryos was significantly smaller than that of control embryos at E16.5; however, normal characteristic cell types were detected at histological levels in the glandular and non-glandular stomach. The mucosa appeared to be thicker and tightened in Sfrp1−/− Sfrp2−/− Sfrp5+/− non-glandular stomach in comparison with control non-glandular stomach (Figure S3).
To determine correlation between reductions of the anterior-posterior (a-p) body axis and abnormal gut formation in Sfrp1−/− Sfrp2−/− and Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos, we examined the shape of the gut at earlier stages. Reductions in the length of the hindgut and the caudal half of the midgut were already apparent at E10.5. The shortening of the caudal gut tube was closely related to reduction of the a-p body axis in compound mutant embryos at earlier stages [12] (Figure 1F). In contrast, marker analysis of Shh, Pdx1 [19], Barx1 and Sfrp5 in Sfrp1−/− Sfrp2−/− and Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos demonstrated that the regions corresponding to the prospective stomach and duodenum were unaffected at E10.5 (Figure 1F–I). Thus, the region corresponding to the stomach and the duodenum in Sfrps-deficient embryos is specified and generated in normal length at E10.5; moreover, organ bud formation is initiated in a manner consistent with that in control embryos. Organ bud formation occurs following the establishment of the a-p body axis; consequently, we concluded that deficiency of Type 1 Sfrps leads to a reduction in the size of the fore-stomach in a manner that is independent of the mechanism that shortens the a-p body axis.
Although the stomach region is enlarged at E11.5, one day after the initiation of organ bud formation, the greater curvature of the fore-stomach is not well expanded as observed at later stages. The expansion of the greater curvature becomes obvious from around E12. We measured the size of E12.5 Sfrps-deficient fore-stomach, in order to gain insight into the character of the size reduction defect in the fore-stomach. The length of the greater curvature was greatly reduced in the fore-stomach of Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos in comparison with that of control embryos (790±75 µm in control and 406±53 µm in Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs, n = 3, P<0.01). In contrast, the width of the Sfrp1−/− Sfrp2−/− Sfrp5+/− stomach was significantly increased at the junction of the fundus and the body (347±33 µm in control and 453±44 µm in Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs, n = 3, P<0.05; Figure S4A–C). Thus, Sfrps deficiency induces lateral expansion of the fore-stomach, which may be suggestive of a defect in morphogenesis.
Since this defect might be associated with an abnormality in epithelium, the histology of Sfrps-deficient fore-stomach epithelium was examined. The cell number per area of epithelium (2000 µm2) was slightly increased in the greater curvature of Sfrps-deficient fore-stomachs; however, this observation was statistically insignificant (Figure S4E). Due to the low frequency of multi-nuclei along the apicobasal (AB) axis, the greater curvature epithelium at E12.5 is considered a mono-cell layer in both Sfrp1−/− Sfrp2−/− Sfrp5+/− and control fore-stomachs (Figure S4D, F).
To elucidate the defect in morphogenesis of the mono-cell epithelial layer in Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs at E12.5, we examined oriented cell division in the greater curvature. The basolateral cellular membrane, microtubule spindles and chromosomes were visualized with anti-ß1-integrin antibody, anti-acetylated α-tubulin antibody [20] and DAPI (4′, 6′-diamidino-2-phenylindole hydrochloride) staining, respectively (Figure 2A, B). The staining of E12.5 fore-stomach epithelium revealed cell division in approximately 3% of cells in the greater curvature of the fore-stomach (3.64±0.32% in control and 3.27±1.12% in Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomach, n = 3). Approximately 20% of the cell division axis was oriented along the AB axis in both control and Sfrps-deficient fore-stomach epithelium (19.6±1.18% of 204 cells in three controls and 18.9±5.06% of 196 cells in three Sfrp1−/− Sfrp2−/− Sfrp5+/− mutants; Figure 2E). Approximately 80% of cell divisions occurred within the horizontal plane of the epithelium, with significant convergence within ±45° of the cephalocaudal axis along the fundus to the pylorus in controls (68.9±3.69% of 164 horizontal mitotic cells in three fore-stomachs; Figure 2C, D). In contrast, oriented cell division was not apparent in the fore-stomachs of Sfrp1−/− Sfrp2−/− Sfrp5+/− mutants (39.6±1.71% of 159 horizontal mitotic cells in three fore-stomachs; Figure 3C, D) (P<0.001).
The distinctive abnormality in oriented cell division was maintained at E13.5. Approximately 35% of the cell division axis was oriented along the AB axis in control and Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs (36.5±1.56% of 416 cells in four control and 33.3±5.78% of 465 cells in four Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs; Figure 2H). In the remaining mitotic cells, the orientation of cell division in the horizontal plane displayed convergence within ±45° of the cephalocaudal axis in control fore-stomach epithelium (81.6±5.37% of 266 horizontal mitotic cells in four fore-stomachs; Figure 2H, G). However, cell division orientation diverged markedly in Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs (34.0±3.78% of 312 horizontal mitotic cells in four fore-stomachs, P<0.0001; Figure 2F, G). Oriented cell division was not observed along the cephalocaudal axis of the greater curvature of hind-stomachs of either control or Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos (56.3±3.9% of 135 horizontal mitotic cells in three control and 56.0±0.9% of 134 horizontal mitotic cells in three Sfrp1−/− Sfrp2−/− Sfrp5+/− hind-stomachs; Figure 2I–K). Hence, these observations suggest that Type 1 Sfrps are required for oriented cell division in the fore-stomach.
Cell proliferation and apoptosis ratios were also examined in Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs. No difference was detected in cell proliferation at E13.5 (14.5±2.5 and 15.4±2.9 phospho-Histone H3-positive cells in 1×105 µm3 of control and Sfrp1−/− Sfrp2−/− Sfrp5+/− epithelium, respectively; n = 2). The TUNEL assay detected less than 1 apoptotic cell per section of fore-stomach in control and Sfrps-deficient embryos; thus, no observations were possible. Similarly, total epithelial cell number in the fore- and hind-stomachs of compound mutant embryos was identical to that in control embryos (Figure S5C, D); however, cell number per area of the greater curvature epithelium (2000 µm2) was slightly increased (approximately 27%) in Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs at E13.5 (n = 4, P = 0.0001; Figure S5A, B, E, F).
Since cell density in the epithelium of Sfrps-deficient fore-stomachs appeared to be increased in comparison with the controls, we examined AB polarity. Sub-cellular distribution of marker proteins (e.g., atypical Protein Kinase C (aPKC), ß1-integrin, E-cadherin and F-actin) [21]–[23] in Sfrp1−/− Sfrp2−/− Sfrp5+/− fore-stomachs was identical to localization in control fore-stomachs (data not shown). However, the distribution patterns did not suggest a strong establishment of AB polarity even in the controls. Thus, these observations suggest that the defect of cell division orientation is associated with fore-stomach morphogenesis phenotype.
We next examined which Wnt pathway is regulated by Sfrps in the fore-stomach, since Sfrps regulate the Wnt/ß-catenin and the Fz/PCP pathways [9]. The Wnt/ß-catenin pathway is highly activated in fore-stomach epithelium at E13.5, as evidenced by TOPGAL reporter activity [24] (Figure S6A, B). The activity levels were not altered in fore-stomach epithelium of Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos in comparison with that in control embryos. In contrast, TOPGAL activity, which was markedly diminished at the boundary of the control fore- and hind-stomachs, extended into the hind-stomach region in Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos (Figure S6C, D). However, as shown by FoxA2 [25] and Sox2 [26] protein expression, no significant patterning defect was observed at the junction of fore- and hind-stomach epithelium (Figure S6E–J). Moreover, the TOPGAL activation pattern in Sfrp1−/− Sfrp2−/− Sfrp5+/− stomach was not correlated with the defect observed in fore-stomach morphogenesis.
We also examined fore-stomach morphogenesis in Lp/Lp embryos carrying mutations in Stbm/Vangl2, a component of the Fz/PCP pathway, because Type 1 Sfrps genetically interact with Stbm/Vangl2 [9]. Stbm/Vangl2 is expressed in stomach epithelium [27]. Marker analysis revealed that fore-stomach size/length was greatly reduced in E13.5 Lp/Lp embryos (Figure 3A). In addition, a morphological defect was associated with divergence of cell division orientation in the epithelium of the greater curvature epithelium (Figure 3B, C, also see Figure 2B). These observations suggest that the Fz/PCP pathway modulates lengthening of the fore-stomach during oriented cell division.
These data present the possibility that dys-regulation of the Fz/PCP pathway perturbs oriented cell division in Sfrps-deficient fore-stomach epithelium. To investigate whether the Fz/PCP pathway is affected in Sfrps-deficient fore-stomach, we examined the sub-cellular distribution of Frizzled3 (Fz3) and Dishevelled-2 (Dvl-2) [28],[29]. Proper sub-cellular distribution of Fz/PCP pathway components is essential for the pathway activity [30]. Fz3 and Fz6 are mammalian homologues of the Drosophila Fz receptor in the Fz/PCP pathway. Fz3 localized to the apical surface of epithelial cells in control greater curvature at E13.5; in addition, local enrichment of the protein at the site of cell-cell adhesion was not observed. Dvl-2 co-localized with Fz3 in the apical region of the fore-stomach epithelium (arrowheads in Figure 3Da-c). Co-localization of Fz3 and Dvl-2 was less apparent in the apical surface of hind-stomach epithelium (data not shown). Dvl-2 expression, which was also observed in the basal side of the epithelium, overlapped with that of ß1-integrin (Figure 3Db; data not shown). Significantly, Fz3 and Dvl-2 displayed diffuse distribution in the middle of the greater curvature of the Sfrps-deficient fore-stomachs (n = 3; Figure 3Dd-f). Thus, this finding indicates that the Fz/PCP pathway is affected in Sfrps-deficient fore-stomach. Moreover, Sfrp regulation of the Fz/PCP pathway appears to be correlated with the defect observed in fore-stomach morphogenesis.
Wnt signaling inhibition by Sfrp usually involves an associating Wnt ligand [6]. The following observations suggest that Wnt5a is inhibited by Sfrps during fore-stomach morphogenesis: Wnt5a, a typical non-canonical Wnt ligand gene, is expressed in fore-stomach mesenchyme, where defects of the Sfrps-deficient stomach were found; Wnt5a and Type1 Sfrps genetically interact with Stbm/Vangl2 to regulate the Fz/PCP pathway [9],[31].
To address the possibility that Wnt5a is an inhibitory target of Sfrps in the fore-stomach, fore-stomach phenotype in Wnt5a homozygous (Wnt5a−/−) mutant embryos was surveyed. Significantly, the Wnt5a−/−gut displayed similarities to the Sfrps-deficient gut, with the exception of ectopic branching of the small intestine (Figure 4A, double arrows in the panel depicting Shh expression). First, fore-stomach formation was defective in the Wnt5a−/− embryos at E13.5 (Figure 4A). Hind-stomach formation was less affected in the E13.5 Wnt5a−/− embryos, although the hind-stomach appeared to be affected and was smaller at later stages, such as E16.5 (data not shown). Second, canonical Wnt/ß-catenin signaling was not altered in that region as evidenced by TOPGAL activity, a reporter of the Wnt pathway [24] (Figure 4A). Third, fore-stomach malformation was associated with divergence of cell division orientation in the greater curvature epithelium (Figure 4B, C, also see Figure 2B). Oriented cell division in the horizontal plane of the greater curvature of the fore-stomach, which was evident in the controls, was not obvious in the Wnt5a−/− mutants (Figure 4B).
Statistical analyses revealed a significant difference between convergence of cell division orientation in the control (83.2±4.14% of 232 cells, n = 3) and Wnt5a−/− fore-stomach epithelia (37.0±4.48% of 257 cells, n = 3, P<0.0001; Figure 4C, left). The cell proliferation ratio determined with anti-phospho-Histone H3 staining was not altered in Wnt5a−/− fore-stomach epithelium (18.3±0.2 positive cells in1×105 µm3, n = 3) in comparison with control fore-stomach epithelium (18.7±1.0 positive cells in1×105 µm3, n = 3). Oriented cell division was not observed in either control (54.0±3.59% of 137 cells, n = 3) or Wnt5a−/− hind-stomach epithelium (53.4±1.22% of 131 cells, n = 3; Figure 4D). No significant difference in the frequency of cell division along the AB axis was detected between control and Wnt5a−/− fore-stomachs (35.4±3.91% and 34.0±2.65% of 359 and 389 cells, respectively, n = 3); a similar situation was apparent with respect to corresponding hind-stomachs (34.1±1.54% and 36.7±4.76%, respectively, n = 3; Figure 4C, D, right). Thus, oriented cell division is disrupted in the epithelium of Wnt5a−/− fore-stomachs as well as in the Sfrps-deficient fore-stomach.
In addition to a defect in fore-stomach morphogenesis, the intestine was shortened substantially in the Sfrps-deficient embryos in association with shortening of the anterior-posterior (a-p) body axis (Figure S7). Moreover, the observed reduction in the length of the intestine in E13.5 Sfrp1−/− Sfrp2−/− embryos was enhanced in the presence of an Sfrp5 heterozygous mutation (Sfrp1−/− Sfrp2−/− Sfrp5+/−) (Figure S7F). Thus, cell migration associated with a-p axis elongation may be involved in gut morphogenesis. The small intestine was remarkably shortened in Sfrp1−/− Sfrp2−/− and Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos at E13.5; therefore, Sfrps-deficient gut tubes were examined to determine the effect of these molecules on regionalization of the small intestine. Cdx2 is expressed in the epithelium of the small and large intestines [32]. In the gut derived from Sfrp1−/− Sfrp2−/−, Sfrp1−/− Sfrp2−/− Sfrp5+/− and control embryos, the rostral boundary of Cdx2 expression was observed at the pyloric sphincter, a junction between the stomach and the duodenum (Figure S7A, arrow). Hoxa4 is expressed in the mesenchyme from the prospective duodenum to a portion of the ileum (rostral small intestine) in control embryos [33], whereas Sfrp5 is expressed in the epithelium (Figure S7B, C). Wnt5a and Hoxc6, which are marker genes for the caudal small intestine, are expressed in the mesenchyme (Figure S7D, data not shown) [34],[35]. The expression of these markers was indicative of the regionalization of the small intestine in Sfrp1−/− Sfrp2−/− and Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos. Furthermore, Hoxd13 expression was observed in the caudal large intestine of Sfrps-deficient and control embryos (Figure S7E). Thus, the expression of these markers indicates that (Type 1) Sfrps do not affect regional specification of the gut tube at E13.5.
Interestingly, the small intestine of Sfrp1−/− Sfrp2−/− and Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos at E13.5 displayed cell clumps, which disrupted the internal surface of the epithelial tube (Figure 5A, arrowhead). The clump of epithelial cells occurred continuously from the jejunum to the ileum. In control embryos at E13.5, the region corresponding to the prospective jejunum and ileum within the small intestine exhibited a smooth apical surface (Figure 5A). In contrast, cell clumps were not obvious in the Sfrps-deficient small intestine at E16.5. It is possible that an increase in cell proliferation may contribute to the generation of cell clumps within the epithelium. However, a BrdU incorporation assay failed to detect an increase in cell proliferation rates in the clumps as well as the entire gut epithelium (36.1±3.38% of total 1030 control epithelium cells, n = 3; 33.6±4.63% of total 1265 Sfrp1−/− Sfrp2−/− Sfrp5+/−epithelium cells, n = 3) (Figure S8A–C). In addition, cell density was not significantly increased in the epithelium of Sfrp1−/− Sfrp2−/− Sfrp5+/− small intestine (Figure S8D). Based on these observations, we hypothesize that this histological abnormality appears to be related to a defect in epithelial morphogenesis.
To address this possibility in greater detail, we analyzed the localization of AB polarity markers. Activated aPKC (phospho-aPKC) has been implicated in the establishment of AB polarity in mammalian cells [23]. We observed proper localization of aPKC to the apical region of control small intestine epithelium. In contrast, specific sub-cellular localization of aPKC disappeared in a clump of epithelial cells in Sfrp1−/− Sfrp2−/− Sfrp5+/− small intestine (Figure 5B–G). Antibody staining against ß1-integrin revealed a round cell shape in the clump of epithelium (Figure 5D, E; Video S1, S2). In addition, E-cadherin was concentrated at the apicolateral cytoplasmic membrane in control epithelium (Figure 5F). However, it was widely distributed in the cytoplasmic membrane of the epithelial cell clump (Figure 5G, asterisk; Video S3, S4). These observations indicate the involvement of Type 1 Sfrps in the regulation of AB polarity in the small intestine epithelium.
Since a relationship between AB polarity and the PCP pathway was suggested previously [30],[36], we assessed sub-cellular distributions of Fz3 and Dvl-2 in the small intestine. In control epithelium derived from small intestine corresponding to the jejunum and the ileum at E13.5, Fz3 occupied the apical region and co-localized with Dvl-2 (Figure 6F). However, Fz3 and Dvl-2 were not concentrated in the apical region of the Sfrps-deficient small intestine epithelium, especially in the clump of epithelial cells (Figure 6G). No difference in sub-cellular distributions of Fz3 and Dvl-2 was detected in other regions of the Sfrp1−/− Sfrp2−/− Sfrp5+/− gut tube relative to that of the control gut tube. Thus, Type 1 Sfrps affect AB cell polarity in conjunction with the regulation of the sub-cellular distribution of core Fz/PCP factors in the small intestine epithelium.
TOPGAL reporter activity [24] indicated that up-regulation did not occur in the canonical pathway within the Sfrp1−/− Sfrp2−/− Sfrp5+/− small intestine epithelium (Figure S9A, B). Since similar defects in terms of oriented cell division were observed in the stomachs of Sfrp1−/− Sfrp2−/− Sfrp5+/− and Wnt5a inactivated embryos, we examined AB polarity in the small intestine of Wnt5a−/− embryos at E13.5. The epithelial structure was disrupted by epithelial cell clumps in the region corresponding to the jejunum and the ileum of Wnt5a−/− embryos (Figure 6A). In addition, the apical distribution of aPKC was disturbed in Wnt5a−/− small intestine epithelium (Figure 6B–E), suggesting a defect in AB polarity. Moreover, this epithelial abnormality was associated with defective Fz3 and Dvl-2 sub-cellular distributions in the Wnt5a−/− small intestine (Figure 6H). The gut phenotypes observed in Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos appeared to be more severe than those in Wnt5a−/− embryos. However, Wnt5a−/− embryos displayed significant phenotypic similarities to Sfrps-deficient embryos.
Although the morphological abnormalities detected in the gut of Wnt5a−/− embryos resembled those in Sfrps-deficient gut, this observation did not necessarily equate to a similarity in signaling regulation. In fact, the loss or gain of Wnt5a function results in dys-regulated convergent extension (CE) movements in vertebrates [37],[38]. Additionally, previous reports imply that Sfrp2 antagonizes Wnt5a signaling [39].
To establish a molecular relationship between Sfrps function and Wnt5a signaling, the signaling activity of the Wnt5a pathway in the gut was assayed in terms of phospho c-Jun levels. It is well established that Wnt5a activates c-Jun N-terminal kinase (JNK). In turn, JNK phosphorylates c-Jun [40],[41]. Currently, phospho c-Jun is the only available marker in the pathway detectable with antibody staining (Figure 7A). Phospho c-Jun-positive cells were observed in small intestine epithelium at E13.5; however, it was scarcely detected in fore-stomach epithelium. In Sfrps-deficient small intestine, phospho c-Jun levels were elevated significantly in the epithelium in comparison with the control small intestine epithelium (Figure 7A, B). Moreover, phospho c-Jun-positive cells were frequently observed in the mesenchyme of Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos (Figure 7A). In contrast, phospho-c-Jun levels were decreased in Wnt5a−/− epithelium (Figure 7A, B). Immunofluorescence staining was repeated three times, followed by statistical analysis of the staining intensity. The results were statistically significant (i.e., Control, 100±8.16%; Sfrps-deficient, 139±9.24%; Wnt5a−/−, 52.5±11.4%; P<0.01) upon comparison between control and Sfrps-deficient or Wnt5a−/− small intestines (Figure 7 B). These observations suggest that Type 1 Sfrps may inhibit Wnt5a signaling. Therefore, we tested this possibility. In a co-culture immunoprecipitation assay, Wnt5a co-precipitated with Sfrp1 from the lysate containing Sfrp1 and Wnt5a; this observation indicated that Sfrp1 interacts with Wnt5a (Figure 7C). We also examined the effect of Sfrp on JNK activation induced by Wnt5a. In conditioned medium, Wnt5a elevated the levels of phospho-JNK, an active form, in HEK293T cells. However, Sfrp1, in the presence of Wnt5a, reduced the levels of active JNK (Figure 7D). Thus, Type 1 Sfrps are capable of inhibiting Wnt5a signaling.
In Zebrafish, oriented cell division is a driving force of a-p axis elongation during gastrulation and neural tube morphogenesis during neurulation [42]–[44]. Moreover, in Drosophila, oriented cell division in the imaginal disc affects adult organ shape [45]. In both processes, oriented cell division is disrupted by a defect in the Fz/PCP pathway [42],[43],[45]. Thus, the Fz/PCP pathway regulates cell division orientation with respect to tissue elongation during embryonic development. However, this involvement was unknown in mammals.
We observed divergence of cell division orientation in the greater curvature of the fore-stomach in Sfrps-deficient and Wnt5a−/− embryos. Both Wnt5a and Type1 Sfrps genetically interact with Stbm/Vangl2 [9],[31]. Moreover, oriented cell division was disrupted in the fore-stomach epithelium of Lp/Lp embryos. Thus, these observations are indicative of Fz/PCP pathway regulation of epithelial oriented cell division in the fore-stomach. Significantly, the components of the Fz/PCP pathway, i.e., Fz3, Dvl2 and Vangl2, are expressed in the epithelium [27]; in contrast, co-expression is not evident in the mesenchyme. Furthermore, sub-cellular distribution of Fz3 and Dvl-2 was affected in Sfrps-deficient fore-stomach epithelium. Therefore, the Fz/PCP pathway components in the epithelium appear to be involved in the regulation of oriented cell division.
Following the initiation of organ bud formation at around E10.5, the stomach is dramatically enlarged over several days. Previous data of Nyeng et al. [26] suggest that the epithelium of the fore-stomach initiates terminal differentiation at E15.5. Hence, most of the cells in fore-stomach epithelium at E12.5 and E13.5 could be immature cells continuing cell division during these stages. When we observed cell division in fixed samples, approximately 3% of epithelial cells demonstrated division in the greater curvature of the fore-stomach at E12.5. However, cell proliferation occurs rapidly in the epithelium; additionally, an increment of epithelial cell number generates the largest number of dividing cells in the entire fore-stomach epithelium during organ development. Therefore, cell division orientation could be one of essential components contributing to fore-stomach morphogenesis.
The phenotype described by fore-stomach shortening and mitotic orientation defect in Sfrp1−/− Sfrp2−/− Sfrp5+/− stomachs might be most severe in all mutants; additionally, the phenotype in Lp/Lp stomachs might be relatively milder in comparison to that found in other mutants. Although the difference between the mitotic orientation defects of the mutants was statistically insignificant, a weak tendency in the severity of mitotic orientation defect, which could be correlated with the severity of fore-stomach shortening defects, may occur. The weak tendency might be due to our observation of cell division involving an instantaneous event of an individual cell, whereas the morphological features of the fore-stomach shortening defect was a result of accumulated cellular events.
A defect in the cell rearrangement process (e.g., increased radial intercalation) [46] might be a possible element in the induction of fore-stomach shortening. However, epithelial cell division along the AB axis occurred at low frequencies and the mono-cell layer was maintained in Sfrps-deficient fore-stomach epithelium at E12.5. Therefore, oriented cell division and allocation of the divided cells along the cell division axis could contribute to organ lengthening as one of the earlier events in fore-stomach morphogenesis. The fore-stomach is the most remarkable structure in the developing gut tube, which is generated during mid-gestation. Most internal organs are generated from an epithelial tubule structure, with the tube altering its shape depending on the function of the organ. It is possible that oriented cell division is a common mechanism that is essential for tubular morphogenesis of the internal organs.
Our results implicate the Fz/PCP pathway, in association with Sfrp regulation of Wnt5a, in the regulation of oriented cell division. The reason that mutations in opposing regulatory components lead to similar defects in oriented cell division may be due to the loss of planer cell polarity in both defects. Animals with mutations in opposing Fz/PCP pathway regulatory components frequently exhibit similar phenotypes [47].
The current in vitro data and data representing the response of downstream effectors of Wnt5a signaling suggested that Sfrp inhibits Wnt5a signaling. Genetic analysis was also conducted via the generation of Wnt5a, Sfrp1 and Sfrp2 compound mutant embryos. A survey of the morphology and internal organs of these embryos was suggestive of no rescue and no enhanced phenotype in Wnt5a+/− Sfrp1−/− Sfrp2−/− embryos in comparison with Sfrp1−/− Sfrp2−/− embryos at E10.5, E12.5 and E13.5. Thus, the genetic analysis was not beneficial in terms of evaluation of an interaction between Sfrps function and Wnt5a signaling. This observation is likely attributable to the insufficient capacity of the heterozygous mutation to reduce Wnt protein expression under an effective dosage.
The Wnt5a transcript is highly expressed in fore-stomach mesenchyme; in contrast, the expression is weaker in hind-stomach mesenchyme at E13.5. Based on transcriptional expression patterns, an active Wnt5a protein gradient may be generated in conjunction with Sfrps along the cephalocaudal axis in the stomach. Further investigation is necessary in order to understand the mechanism by which the protein gradient is involved in epithelial oriented cell division through the Fz/PCP pathway.
Previous studies have identified the role of various molecules including the conserved PAR-aPKC complex in the regulation of epithelial AB polarity [23]. Additionally, the involvement of PCP pathway components in AB polarity of epithelial cells has been suggested [36],[48]. However, little is known regarding intracellular signaling regulation. The intercellular signaling regulation could coordinate AB polarity in developing organs. We observed a similar AB polarity defect in both Sfrps-deficient and Wnt5a inactivated small intestines. In contrast, we could not identify clear evidence suggesting abnormality of AB polarity in the fore-stomach; AB polarity was not well established even in wild-type fore-stomach epithelium throughout the assessment of sub-cellular distribution of protein markers.
Although Sfrps are capable of inhibiting Wnt5a signaling, it is possible that mutations in opposite regulatory components result in similar defects as both defects lead to a loss of AB polarity. Previously published data indicate that Wnt5a activates JNK [40],[41], which is an essential effecter of CE movement [49]. JNK phoshorylates paxillin, a component of the focal adhesion complex [50]. In addition, Wnt5a is able to activate focal adhesion kinase (FAK) [39]. Paxillin and FAK are known to play a role in cell migration [39],[50]; thus, the possibility exists that Wnt5a promotes cell migration. However, FAK and paxillin are also required for the maintenance of adherence junctions via N-cadherin-based cell-cell adhesion. Regarding this function, FAK and paxillin are involved in a mechanism of the down-regulation of the activity of the small GTP-binding protein Rac1 [51]. In contrast, Wnt5a signaling is also involved in the up-regulation of Rac1 [39]. Interestingly, the constitutive active and dominant negative forms of Rac1 lead to a phenotype similar to that observed upon the loss of E-cadherin at sites of cell-cell contact [52],[53]. Furthermore, E-cadherin provides a clue with respect to the development of AB polarity leading to the recruitment of the PAR-aPKC complex to immature adherence junctions [23]. In this respect, appropriate levels of Wnt5a signaling activity may be essential for the modulation of AB polarity. Hence, we propose that Sfrps and Wnt5a are putative components of intracellular signaling regulation in order to coordinate AB polarity in the developing small intestine.
Control of AB polarity via Wnt signaling has been suggested. In Xenopus and Drosophila, Dvl is necessary for basolateral membrane localization of Lgl (Lethal giant larvae), which encodes a protein with multiple WD-40 motifs that regulates AB polarity [48]. Dvl interacts with Lgl. Moreover, Fz8, but not Fz3 and Fz7, regulate Lgl sub-cellular localization [48], which suggests that the basolateral localization of Fz8, Dvl and Lgl is required for the establishment of AB polarity [48]. In contrast, we demonstrated that Dvl-2 is concentrated at the apical surface of gut epithelium and that this apical localization overlaps with Fz3 localization. Thus, these observations suggest that AB polarity regulation in gut epithelium is distinct from that previously described in Xenopus and Drosophila.
In summary, our results indicate a link between Sfrps function and Wnt5a signaling in the regulation of epithelial cell polarity including oriented cell division and AB cell polarity in developing gut. Sfrps are known as tumor suppressor genes, which are epigenetically silenced in many types of cancer, especially in colorectal cancer [7]. In contrast, up-regulation and down-regulation of Wnt5a are observed in gastric and colon cancer, as well as in cancer progression [39],[54]. Sfrps-regulation of Wnt5a signaling may provide novel insight into the progression and aggressiveness of GI tract cancer.
Sfrp1−/− Sfrp2+/− and Sfrp1−/− Sfrp2+/− Sfrp5−/− mice were maintained in a 129 and C57BL/6 mixed genetic background [9],[12]. Sfrp1−/− Sfrp2−/− Sfrp5+/− embryos were derived from crosses between Sfrp1−/− Sfrp2+/− and Sfrp1−/− Sfrp2+/− Sfrp5−/− mice. Lp (LPT/LeJ) mice and Wnt5a heterozygous mutant (Wnt5atm1Amc/J) mice were obtained from the Jackson Laboratory [31],[55]. Crosses were utilized to introduce the TOPGAL reporter [24] into Sfrps-deficient and Wnt5a+/− mice. The activity of the TOPGAL reporter was visualized via standard LacZ staining involving a short reaction time of 30 minutes to compare activation levels.
An Sfrp1 cDNA fragment obtained from cDNA subtraction screening was used as a probe for whole mount in situ hybridization [56]. Sfrp2, Sfrp5 and Barx1 cDNA clones were obtained as I.M.A.G.E. clones; Cdx2, Hoxa4, Hoxd13, Islet1, Nkx6.3 and Pdx1 probes were generated from FANTOM cDNA clones [57].
The GI tract was isolated from control, Sfrp1−/− Sfrp2−/−, Sfrp1−/− Sfrp2−/− Sfrp5+/−, Lp/Lp and Wnt5a−/− embryos in phosphate buffered saline (PBS) containing 10% fetal calf serum. Whole mount immunofluorescence staining of stomach epithelium was performed as described previously [12] employing anti-acetylated α-tubulin antibody (Sigma, mouse monoclonal clone 6-11B-1) and anti-ß1-integrin antibody (Chemicon, rat monoclonal MAB1997). Chromosomes were visualized by DAPI staining. Images were captured on a BioRad Radiance 2100 Laser Scanning Confocal Microscope System equipped with a Zeiss Axiovert and processed using Adobe Photoshop. Immunofluorescence staining of sectioned tissue was conducted utilizing the following primary antibodies: anti-ß-galactosidase antibody (from Rabbit, Cappel), anti-E-cadherin antibody (Sigma, mouse monoclonal anti-Uvomorulin clone DECMA-1), anti-phospho-aPKC antibody (from rabbit, Cell Signaling), anti-Fz3 antibody (from rabbit, MBL), anti-Dvl-2 antibody (from goat, Santa Cruz, N-19), anti-FoxA2 antibody [58] (from Rabbit) and anti-Sox2 antibody (from goat, Santa Cruz, Y-17), anti-phospho-c-Jun (from Rabbit, Cell Signaling) and anti-BrdU (mouse monoclonal, BD). F-actin was visualized with Rhodamine-conjugated phalloidin. Images, which were captured on a BioRad Radiance 2100 Laser Scanning Confocal Microscope System equipped with a Zeiss Axiovert, were processed using Adobe Photoshop. Images of anti-phospho-c-Jun staining were analyzed by imaging software MultiGauge (Fujifilm) to calculate averages of signaling intensity per area in the epithelium. With respect to this calculation process, staining background intensity in epithelium and mesenchyme was measured, followed by subtraction from the specific signal intensity. For the BrdU incorporation assay, BrdU/PBS solution was injected into a pregnant mouse intraperitoneally (100 µg/g body weight) 1 h before embryo collection.
The immunoprecipitation assay was performed by co-culture immunoprecipitation of L cells expressing Wnt5a (L Wnt-5A, ATCC) and L cells expressing Sfrp1-FLAG (c-terminal tagged with FLAG, L Sfrp1) (1∶1 ratio). Sfrp1-FLAG was precipitated with anti-FLAG M2 affinity gel (Sigma) according to the manufacturer's protocol. Goat anti-Wnt5a antibody (R&D Systems, Inc.) was used to detect Wnt5a protein. JNK activity was evaluated based on the levels of phospho-JNK determined by an anti-phospho-JNK antibody (Cell Signaling). Total JNK was detected by anti-JNK antibody (Cell Signaling). Lysates for Western blotting were derived from HEK293T cells incubated for 3 h in L cell conditioned medium. L Sfrp1 was established as a stable transformant cell line. Conditioned media from L cell, L Wnt5a and L Sfrp1 were obtained following a 4-day incubation of the culture medium.
Histological sections (5 µm) stained with Hematoxylin and Eosin (H&E) were prepared from Bouin-fixed and paraffin-embedded specimens. Cell number per 2000 µm2 in fore-stomach epithelium was calculated by counting the nuclei in a 400-µm wide area of a single section. Total cell number in an epithelial cell suspension was determined with a hemocytometer. Cell suspensions were prepared from stomach epithelium separated from mesenchyme following incubation of the sample in PBS containing 0.5 unit dispase and 1.25% pancreatin at room temperature for 30 minutes. Statistical significance, which was evaluated using Welch's t-test, was defined as P<0.05. Error bars indicate SD.
All animals were handled in strict accordance with good animal practice as defined by the relevant national and/or local animal welfare bodies, and all animal works were approved by the appropriate committee.
|
10.1371/journal.pcbi.1002166 | Hydrophobicity and Charge Shape Cellular Metabolite Concentrations | What governs the concentrations of metabolites within living cells? Beyond specific metabolic and enzymatic considerations, are there global trends that affect their values? We hypothesize that the physico-chemical properties of metabolites considerably affect their in-vivo concentrations. The recently achieved experimental capability to measure the concentrations of many metabolites simultaneously has made the testing of this hypothesis possible. Here, we analyze such recently available data sets of metabolite concentrations within E. coli, S. cerevisiae, B. subtilis and human. Overall, these data sets encompass more than twenty conditions, each containing dozens (28-108) of simultaneously measured metabolites. We test for correlations with various physico-chemical properties and find that the number of charged atoms, non-polar surface area, lipophilicity and solubility consistently correlate with concentration. In most data sets, a change in one of these properties elicits a ∼100 fold increase in metabolite concentrations. We find that the non-polar surface area and number of charged atoms account for almost half of the variation in concentrations in the most reliable and comprehensive data set. Analyzing specific groups of metabolites, such as amino-acids or phosphorylated nucleotides, reveals even a higher dependence of concentration on hydrophobicity. We suggest that these findings can be explained by evolutionary constraints imposed on metabolite concentrations and discuss possible selective pressures that can account for them. These include the reduction of solute leakage through the lipid membrane, avoidance of deleterious aggregates and reduction of non-specific hydrophobic binding. By highlighting the global constraints imposed on metabolic pathways, future research could shed light onto aspects of biochemical evolution and the chemical constraints that bound metabolic engineering efforts.
| What governs the identity and concentrations of metabolites within living cells? The first part of this question has received much attention. Organisms were found to qualitatively prefer hydrophilic and charged metabolites, a phenomenon that was explained to be a result of constraints imposed by contemporary as well as archaic metabolism. However, among the metabolites that are used, a quantitative preference has never been analyzed systematically. Here we use the most comprehensive data sets of metabolite concentrations available to explore such trends. We find that in various organisms and growth conditions, living cells minimize the concentrations of non-polar, un-charged metabolites. More specifically, metabolites' hydrophobicity alters concentrations by two orders of magnitudes on average and explains up to half of the variation of metabolite concentrations within cells. We suggest that this can be attributed to an evolutionary pressure to avoid an unspecific hydrophobic effect: the preference of hydrophobic surfaces in an aqueous environment to adhere to other hydrophobic surfaces. Our findings shed light on the evolution of the internal makeup of living cells and can assist in establishing metabolic models that support synthetic biology and metabolic engineering efforts.
| Living cells exhibit a preference towards certain types of metabolites. Many of these tendencies can be explained as consequences of chemical constraints imposed on metabolism. For example, the cellular ubiquity of charged metabolites, like those containing phosphoryl or carboxyl groups, is attributed to increasing solubility and decreasing leakage through the membrane [1].
Several studies suggest that contemporary structural preferences can be attributed to characteristics of archaic metabolism [2], [3]. For example, it has been suggested that positively charged surfaces played a central role in archaic metabolism, selecting for negatively charged molecules, mainly carboxylates and phosphates [3], [4]. Such conditions also favored water-eliminating polymerization reactions, resulting in the formation of large biomolecules like those that make up most of the biomass in contemporary cells [3]. In addition, early energy demands probably involved the use of iron and sulfur [3], [4], elements that still play a central role in living organisms. Focusing on carbon fixation, the availability of various reduced metals and volatile C1 compounds in the highly reduced early environment probably account for the structure of some of the contemporary carbon fixation pathways [5].
In this study we explore whether the qualitative preferences for specific types of metabolites represent a systematic, quantitative trend across multiple organisms. We suggest that a quantitative perspective on the chemical preferences of living cells could help elucidate the evolutionary forces shaping the structure of metabolic systems, facilitate genome-scale metabolic reconstructions and advance the design and implementation of novel metabolic pathways [6].
A previous study [7] demonstrated that the specific chemical groups composing metabolites explain a fraction of the variance in their concentrations. However, this previous work collected concentration values from separate sources, each employing different conditions and measurements techniques. In our study we use data sets of simultaneously measured concentrations of dozens of metabolites. We report a comprehensive correlation analysis between physico-chemical parameters of metabolites and their in-vivo concentrations. We find consistent trends which suggest that, beyond specific metabolic effects on concentrations, such as the kinetics of the enzymes producing and consuming a metabolite, there are global evolutionary tendencies that shape the internal makeup of living cells.
We employed two large data sets of measured metabolite concentrations in E. coli which represent the most comprehensive data sources currently available (Bennett et al. [8], containing 93 metabolites and Ishii et al. [9], 108 metabolites). To strengthen our analysis we have further used five smaller data sets: three are from S. cerevisiae (Ewald et al. [10], 29 metabolites, Fendt et al. [11], 29 metabolites, and Kummel et al. [12], 33 metabolites); one from B. Subtilis (Kleijn et al. [13], 35 metabolites) and another contains measurements of the 20 common amino acids in human muscle (Bergstorm et al. [14]). Most of these data sets contain at least three different conditions in which concentrations were measured. Overall, 21 conditions were analyzed independently. The full concentration data is given in the Dataset S1.
We analyzed various physico-chemical parameters associated with the different metabolites, including molecular mass (MW), polar surface area (PSA), non-polar surface area (NPSA), number of charged atoms (NCA), hydrogen bond inventory (HBI), number of rotatable bonds (NRB), solubility in water (LogS) and lipophilicity (LogP, the ratio of the equilibrium concentrations of a compound in octanol and water), as shown in Figure 1 (Materials and Methods). We focus our discussion on small metabolites (MW≤300 Da) as we find that these show the most prominent correlations. This group contains most (≥80%) of the metabolites in each of the original data sets. The excluded metabolites includes mostly co-factors (e.g. NADPH, ATP etc), which are expected to be subject to a different and stronger set of selective pressures, alongside phosphorylated nucleotides and CoA substituted compounds. Notably, the qualitative trends we describe below also persist in the full data set, albeit less clearly (Figure S1).
In Figure 2 we show the level of correlation between the physico-chemical parameters analyzed and the logarithm of metabolite concentrations for each of the 21 experimental conditions. Even though the data sets are known to be noisy for experimental reasons we find that some parameters are consistently correlated with metabolite concentrations whereas others show no consistent correlation. The non-polar surface area (NPSA), LogP, LogS and the number of charged atoms (NCA) correlate with concentrations across the data sets and conditions (Figure 2) and point to a systematic phenomenon: the concentrations of non-polar, un-charged metabolites are significantly lower within cells. Specifically, in the two large data sets (Figures 3A and S2), metabolite concentrations decrease on average ∼100 fold with increasing NPSA. In the S. cerevisiae data sets, concentrations increase ∼100 fold with decreasing LogP or increasing LogS (depending on the data set, Figure 2). The lower correlation observed in the data set of Kleijn et al. can be attributed to the multiple analytical platforms that the authors used for the measurement of the metabolites, which might introduce different experimental biases.
In Bennett et al. [8], the most reliable and comprehensive data set (see below), we find that a regression analysis using only NPSA and NCA accounts for almost half of the variation in metabolite concentrations within the cell (R2 = 0.43, glucose-fed E. Coli, Figure 3B). Moreover, while ∼55% of metabolites' concentrations are within one order of magnitude of the mean metabolite concentration in glucose-fed E. coli, we find that a linear model using NPSA and NCA predicts concentrations to within an order of magnitude with a significantly higher ∼80% accuracy (Materials and Methods and Figure S3). The difference between the measured concentrations and those predicted by our linear model is about 5-fold on average. This variation can be attributed to other global or local factors which affect metabolite concentrations. Also, error inherent to the measurement procedures limits the accuracy of the fit between model and data set.
Could the observed correlations stem from a systematic bias in the extraction and measurement procedures, which might prefer polar and charged metabolites over non-polar and un-charged ones? Indeed, some of the published data sets were obtained using extraction methods which risk losing lipophilic metabolites. For example, the two-phase water/chloroform extraction system used by Ishii et al. [9] may be biased towards the extraction of hydrophilic compounds. In order to control for such extraction biases and calibrate the intracellular metabolite concentrations, most studies spiked internal standards directly into the extraction fluid. Bennett et al. [8] and Fendt et al. [11] took the most stringent approach and added known concentrations of labeled standards of all compounds measured to the extraction solvent. Consequently, cellular metabolites and internal standards experienced the same opportunities for adsorptive losses or degradation [15]. This methodology enabled the authors to minimize sources of bias in the extraction and measurement procedures, indicating that the observed trends are unlikely to be the result of experimental artifacts (see Text S1 for further discussion).
When we restrict our analysis to amino acids, we find a significantly higher correlation between their hydrophobicity and measured concentrations (Figure S4). For amino acids, NPSA (or LogP) yields an R2 of more than 0.3 in all data sets, and in several cases it even surpasses 0.5. This trend is apparent when using LogP instead. We note that the concentration differences between free amino acids span two or three orders of magnitude. This large range cannot be explained by the well-known observation that hydrophobic amino acids are less abundant in proteins by about an order of magnitude [16].
The increased correlation observed for amino acids suggests that the observed trends might be more prominent when inspecting a group of metabolically similar compounds. Indeed, we find confirmation of this notion in phosphorylated nucleic acids, the concentrations of which correlate with NCA with R2>0.4, where each additional phosphate group increases concentration roughly three-fold on average. The observation that trends sharpen for groups of metabolically similar compounds suggests that the observed preference for polar, charged metabolites is present at multiple scales of inquiry and is indeed systematic.
There are, however, metabolites which display a consistent deviation from predicted concentrations. Most significant deviations from predicted concentrations occur only in specific conditions or data sets. Notably, glutamate and, to a lesser extent, glutamine are the only non-cofactor metabolites with MW<300 that display a consistent deviation from concentrations predicted using the four main physico-chemical parameters (NPSA, LogP, LogS and NCA) across most data sets. The concentration of glutamate is >30-fold higher than predicted, which has been explained by its role as a cellular nitrogen donor and counter-ion to potassium [8]. Notably, glutamate and glutamine can be regarded as co-factors, serving as nitrogen donors for the biosynthesis of essentially all other amino-acids.
Why should the concentration of hydrophobic, un-charged metabolites be lower in living cells? We hypothesize that concentrations are governed by evolutionary constraints. Here, we summarize and shortly discuss several previously suggested selective pressures acting in cells and how they might account for the observed trends.
A cellular preference for low hydrophobicity and high NCA can be attributed to a selection for decreased membrane permeability [17]. High permeability can result in metabolite leakage [17] or in metabolite accumulation within the membrane, which can lead to membrane instability [18]. Indeed, lipophilicity has become an important criterion in the pharmaceutical industry for estimating the permeability of small molecules through the intestinal membrane and their potential for use as oral drugs [17], [19]. In contrast, charged molecules are orders of magnitude less permeable as compared to their un-charged counterparts [20]. However, previous studies demonstrated that the negative effect of polar surface area (PSA) on permeability is considerably higher than the positive effect of NPSA [17]. As PSA does not exhibit consistent correlation with concentrations, permeability can only provide a partial explanation of the observed trends.
Another explanation for generally lower concentrations of hydrophobic metabolites is that non-polar and un-charged small compounds are at the risk of forming large colloid-like “aggregates” within the cell [21], [22]. These aggregates have been shown to enhance protein unfolding [23], and many synthetic aggregating compounds begin to aggregate at the low µM concentrations [22]. Furthermore, different compounds may promote aggregation synergistically when present in the same mixture [24]. Indeed, it has been shown that lipophilicity, solubility and lack of charged atoms are the most central factors determining the tendency of a compound to form aggregates [21].
Finally, a reduction in the concentration of non-polar metabolites can serve to decrease non-specific binding. Hydrophobic compounds can bind non-specifically to hydrophobic surfaces within the cells, including enzymatic active sites [25], [26], protein surfaces that participate in protein-protein interactions, or even nucleic acid strands [27]. Such hydrophobic stickiness is also associated with promiscuous activity of enzymes towards substrates other than their natural ones [28]. Indeed, in a study examining a large set of enzymes, the lipophilicity of a substrate was found to correlate with its participation in promiscuous drug binding [29]. According to this line of reasoning there is selective pressure to decrease the concentrations of metabolites that are highly hydrophobic and able to bind non-specifically to hydrophobic surfaces. Strengthening this explanation, a selection against non-specific binding of proteins and peptide ligands was demonstrated in the cellular protein interaction network of yeast [30].
We note that each of the above hypotheses actually refers to the phenomenon known as the hydrophobic effect: the preference of hydrophobic surfaces in an aqueous environment to adhere to other hydrophobic surfaces [31]. The “aggregation” hypothesis relates to self-adhesion while the “hydrophobic stickiness” and “membrane permeability” hypotheses refer to adhesion to other hydrophobic surfaces in the cell, the latter involving a specific hydrophobic organelle: the membrane. However, when hydrophobic metabolites are present in low enough concentrations, they are much less likely to diffuse out, aggregate, or bind non-specifically. That is, the “cost” of a metabolite, considering the above constraints, is a function of its concentration as well as its physico-chemical parameters.
From this perspective it is clear that the selective pressures we discuss do not necessarily predict a correlation between absolute concentrations and physico-chemical parameters relating to hydrophobicity. Rather, they predict a correlation when the absolute concentrations are high enough that the costs imposed by the various constraints discussed above are not negligible. In this light it is striking that we observe the significant level of correlation that we do, as several of the metabolites measured are present in extremely low concentrations (<10−6M), likely low enough to not be significantly affected by any of the above constraints. Conversely, a metabolite that is found in high concentration must be soluble and polar enough to meet the constraints imposed by the aqueous environment of the cell or it will certainly impose the costs we have described.
In conclusion, our study suggests that the concentrations of metabolites within the cell is not only a result of specific metabolic effects (i.e. kinetic parameters of the enzymes utilizing them), but also follows systematic global trends. Various large metabolomics data sets have accumulated in recent years and their number is predicted to increase rapidly as the technology improves and becomes more accessible. We believe that our study could raise the interest of the scientific community in the general questions addressed here and pave the way for future and more elaborate analysis. Such future studies could test and refine our findings and pinpoint the exact forces that shape the in-vivo concentrations of metabolites. Of special interest are the questions we addressed only partially: what is the relative importance of each of the discussed selective pressures? How do the differences between the internal environments of different organisms and organelles affect their distributions of metabolite concentrations? Do the constraints associated with different organisms and environments translate into preferences for different, parallel metabolic pathways, each employing different metabolites? We believe that the methodology put forward in this study enables inquiry into these questions and provides a better understanding of the forces shaping cellular life.
The physico-chemical parameters for all compounds analyzed are given in Dataset S1.
We used Pybel, the Python wrapper for OpenBabel (http://openbabel.sourceforge.net) to calculate the molecular mass, number of hydrogen bond acceptors, number of hydrogen bond donors, number of charged atoms and number of rotatable bonds [32]. Using the same software package we corrected all compounds to be in the protonation level most abundant at pH 7. The total hydrogen bond inventory of the molecule [33] was taken as hydrogen bond donors + hydrogen bond acceptors. The number of rotatable bonds refers to the internal molecule bonds that are able to freely rotate in solution but become restricted on passing from a free to a bound state, resulting in an entropic cost [34].
The molecular 3D-structure, essential for determining the surface area of the molecules, was also estimated using OpenBabel. We used asa.py (http://boscoh.com/protein/asapy) [35] to calculate the total surface area of the 3D-structure. We used the solvent-excluded surface area, representing the “cavity” the molecule creates in bulk solvent [36]. We also computed the polar surface area, i.e. the area contributed by polar atoms only (oxygen, nitrogen and the hydrogen atoms attached to them). The non-polar surface area is the difference between total surface area and polar surface area.
The logarithm of the octanol-water (LogP) partition coefficient for un-ionized compounds, was estimated using three different programs: XLOGP3 [37], ALogPS [38] and SciFinder (https://scifinder.cas.org/scifinder). In the paper, we use the ALogPS values since they were found to have the lowest RMSE for small molecules [37] and indeed they produce higher overall correlations. LogS, the logarithm of the solubility in water, was also estimated using ALogPS [38].
We calculated the correlation between the metabolite concentrations in each data set and each of the physico-chemical parameters. For each such calculation, metabolites that were not measured in a given data set or did not have a value for that parameter, were discarded. To find a p-value for each R2 we used a Monte-Carlo permutation test. We created a distribution of randomized R2 values by shuffling the parameter values, randomly assigning them to metabolites and then correlating shuffled values with concentrations. We repeated this process 105 times. The p-value was defined to be the fraction of times for which the randomized R2 values were higher than the original R2. To account for multiple hypothesis testing, we used false discovery rate (FDR) control [39], with a rate of 0.01 (n = 168, 21 data sets X 8 physico-chemical parameters).
Metabolite concentrations were predicted using least-squares multiple linear regression of log10 concentrations against the metabolite NPSA and NCA values. As before, high molecular weight compounds were removed from the analysis. In order to avoid potential over-fitting, the concentration of each metabolite was predicted using a model trained on all other metabolites and excluding the one to be predicted. As we are interested in global trends in concentration, the accuracy of the prediction was taken to be the fraction of predictions within an order of magnitude of the true concentration. In order to quantify the predictive power of our model, we compared the prediction accuracy to the accuracy of predicting the mean concentration for a given data set. For the case of glucose-fed E. Coli from Bennet et. al. we found that 78% of predictions were within one order of magnitude of the true concentrations while only 57% of measured concentrations were within one order of magnitude of the mean concentration (Figure S3).
|
10.1371/journal.pntd.0001914 | Live Brugia malayi Microfilariae Inhibit Transendothelial Migration of Neutrophils and Monocytes | Lymphatic filariasis is a major tropical disease caused by the parasite Brugia malayi. Microfilariae (Mf) circulate in the peripheral blood for 2–3 hours in synchronisation with maximal feeding of the mosquito vector. When absent from the peripheral blood, Mf sequester in the capillaries of the lungs. Mf are therefore in close contact with vascular endothelial cells (EC) and may induce EC immune function and/or wound repair mechanisms such as angiogenesis. In this study, Mf were co-cultured with human umbilical vein EC (HUVEC) or human lung microvascular EC (HLMVEC) and the transendothelial migration of leukocyte subsets was analysed. In addition, the protein and/or mRNA expression of chemokine, cytokine and angiogenic mediators in endothelial cells in the presence of live microfilariae were measured by a combination of cDNA arrays, protein arrays, ELISA and fluorescence antibody tests.
Surprisingly, our findings indicate that Mf presence partially blocked transendothelial migration of monocytes and neutrophils, but not lymphocytes. However, Mf exposure did not result in altered vascular EC expression of key mediators of the tethering stage of extravasation, such as ICAM-1, VCAM-1 and various chemokines. To further analyse the immunological function of vascular EC in the presence of Mf, we measured the mRNA and/or protein expression of a number of pro-inflammatory mediators. We found that expression levels of the mediators tested were predominantly unaltered upon B. malayi Mf exposure. In addition, a comparison of angiogenic mediators induced by intact Mf and Wolbachia-depleted Mf revealed that even intact Mf induce the expression of remarkably few angiogenic mediators in vascular EC. Our study suggests that live microfilariae are remarkably inert in their induction and/or activation of vascular cells in their immediate local environment. Overall, this work presents important insights into the immunological function of the vascular endothelium during an infection with B. malayi.
| Brugia malayi is a nematode which causes lymphatic filariasis in South and South-East Asia. Most infected people harbour many millions of the microfilarial stage of the parasite in their blood stream and yet they show few visible symptoms of disease. Vascular endothelial cells (EC) line the blood vessels and are therefore in direct contact with microfilariae. Since vascular EC are potent immune cells functioning in the production of both immune mediators and regulating the migration of immune cells from the blood into the tissue, we have established an in vitro model in which to test the effect of live Mf upon vascular EC function.
Strikingly, we observed that Mf exposure caused reduced transendothelial migration of neutrophils and monocytes, but not lymphocytes. However, microfilariae stimulated EC production of few pro-inflammatory mediators. Additionally, while filarial infection is known to stimulate mediators that increase blood vessel formation in vivo, live microfilariae promoted only a limited number of these regulators in cultured vascular EC. Our study suggests that the live microfilariae are remarkably inert in their induction and/or activation of vascular cells in their immediate local environment.
| The filarial parasite Brugia malayi is a causative agent of human lymphatic filariasis in South and South-East Asia. B. malayi is transmitted by mosquitoes, which take up the blood-borne microfilarial stage (Mf) of the parasite. For the majority of the day, Mf sequester predominantly in the lungs of the host and they only appear in the peripheral blood circulation for a few hours, which coincides with maximal mosquito feeding [1], [2]. While sequestered in the lungs, B. malayi Mf are likely to interact with vascular endothelial cells (EC) and we have observed them binding to the surface of vascular EC (manuscript in preparation). Helminths are potent modulators of the immune response and filarial nematodes, in particular, have been shown to influence the secretion of inflammatory mediators from a number of different cell types [3], [4], [5]. Vascular EC themselves can modulate the immune response by producing pro-inflammatory cytokines and chemokines, in addition to several angiogenic mediators. Vascular EC also play a critical role in extravasation of leukocytes to the site of inflammation [6], [7].
To our knowledge no studies have addressed induction of local immune or inflammatory responses by vascular EC to live microfilariae of lymphatic filarial parasites. However, Bennuru et al. (2009) have shown that lymphatic EC (LEC) proliferate in response to adult, but not microfilarial, antigen and live parasites can induce tube formation by LEC in a contact-dependent manner. B. malayi microfilarial antigen also induced a number of angiogenic mediators in LEC. These data, together with an increased expression of angiogenesis and lymphangiogenesis mediators found in sera of humans infected with Wuchereria bancrofti, suggest that lymphatic filarial parasites may directly influence inflammation and angiogenesis [8], [9], [10]. Other helminths have been shown to induce pro-inflammatory mediators in EC, for example, Schistosoma mansoni schistosomulae stimulate production of the inflammatory cytokines, IL-6 and IL-7 [11], [12].
In this study, we investigated B. malayi Mf-induced immune responses in the local environment by modelling the interaction of Mf and vascular EC in vitro. Live B. malayi Mf directly inhibited extravasation of both neutrophils and monocytes, but not lymphocytes. However, Mf induced limited immune and angiogenic mediator expression. Several previous studies have shown that the filarial endosymbiotic bacteria, Wolbachia are partially responsible for induction of inflammatory and angiogenic mediators in filarial patients [8]. However, a comparison of angiogenic mediator mRNA expression induced by Wolbachia-depleted and live intact Mf, revealed that few angiogenic mediators were specifically induced by Wolbachia in vascular EC.
Ethical approval was obtained from the East London Local Research Ethics Committee to collect human umbilical cords from mothers from the Royal London Hospital and blood from healthy donors. All study participants provided written informed consent. Parasites were obtained from infected animals in accordance with our Home Office project licence, which was approved under the Home Office (1986) Scientific Procedures Act.
Human umbilical vein endothelial cells (HUVEC) were isolated from human umbilical cords using a modified previously published method [13]. In all experiments, HUVEC were used at passage 5. Cell morphology was confirmed by phase contrast microscopy. HUVEC were cultured in HUVEC medium (M199 supplemented with 150 U/ml penicillin, 150 U/ml streptomycin, 2 mM L-glutamine, 20% heat-inactivated FBS, 1 U/ml heparin and 0.03 mg/ml endothelial cell growth supplement from bovine neural tissue). Cryopreserved human lung microvascular endothelial cells (HLMVEC) were purchased from Clonetics (UK) and were cultured according to the supplier's recommendations. HLMVEC were used for experiments at passage 7–9.
Infected gerbils (Meriones unguiculatus) were obtained from TRS Laboratories, Athens, Georgia, USA. Infection of gerbils was performed by i.p. injection of 400 B. malayi L3. B. malayi Mf were obtained by peritoneal lavage with RPMI-1640, 100–400 days post infection. Mf were isolated by centrifugation of recovered lavage fluid over lymphocyte separation medium (MP Biomedicals, USA).
To harvest Wolbachia-depleted Mf, gerbils were treated with tetracycline in their drinking water (2.5 mg/ml) for a period of 6 weeks [14]. Following treatment, Mf were isolated, genomic DNA extracted and the ratio of Brugia glutathione S-transferase (gst) to Wolbachia surface protein (wsp) copy numbers was measured by qPCR as previously described [15]. Using this measurement, the two batches of Mf isolated for use in Wolbachia-depletion experiments were shown to be 98.46% and 99.84% Wolbachia-free.
1×106 confluent HUVEC at passage 4 were cultured in HUVEC medium. After 60 hours of incubation the medium in each flask was replaced with co-culture medium (50% HUVEC medium (as above) plus 50% RPMI-1640 supplemented with 150 U/ml penicillin, 150 U/ml streptomycin, 2 mM L-glutamine, 20 mM HEPES, 20% heat-inactivated FBS and 20% of glucose solution) containing 125,000 B. malayi Mf. Co-culture medium without B. malayi Mf was added to HUVEC in control flasks. After 24 hours of co-culture, EC or the EC supernatant were collected for further investigation. In some experiments, EC were stimulated with 10 ng/ml IFN-γ (ImmunoContact, USA) for 24 or 48 hours prior to co-culture with Mf. When HLMVEC were co-cultured with Mf, 50% EGM-2 MV BulletKit medium (Clonetics) was used in place of HUVEC medium.
With approval from East London Local Research Ethics Committee whole human blood was collected in 20 U/ml heparin. Peripheral blood mononuclear cells (PBMC) were isolated using lymphocyte separation medium. The intermediate layer of PBMC was collected, washed twice and re-suspended in complete RPMI-1640 and 10% FBS.
To isolate granulocytes, the pellet remaining from the lymphocyte separation medium was re-suspended in a 50∶50 mix of RPMI-1640/10% FBS and 0.9% NaCl. The final solution was supplemented with 3% dextran. After one hour the upper layer was removed and centrifuged at 129×g for 10 minutes at 4°C. The pellet was re-suspended in ice cold 0.2% NaCl for 30 seconds. An equal volume of ice cold 1.6% NaCl was added and the mixture was centrifuged at 129×g for 6 minutes at 4°C. This process was repeated until the cell pellet was free of red blood cells. After the final wash, granulocytes were re-suspended in RPMI-1640 supplemented with 10% FBS and kept on ice until use.
1×105 HUVEC were added to human fibronectin-coated cell culture inserts in the wells of a 24-well plate (Greiner Bio-One). 6 h later HUVEC were stimulated with human TNF-α at a concentration of 20 ng/ml. After another 18 h, 50% medium was removed from each transwell and replaced with complete RPMI-1640 20% FCS 20% glucose solution supplemented with 12,500 Mf. In control conditions, no Mf were added. After another 24 h, 50% of medium was removed from each transwell and replaced with RPMI-1640 supplemented 10% FBS and either 1×106 PBMC or 1×106 granulocytes. RPMI-1640 plus 10% FBS was added into the lower wells. After 4 hours transmigrated cells were harvested from the lower wells and analysed by flow cytometry and/or cytospin.
For cytospin analyses, cells were centrifuged in a cytospin at 800×g for 5 minutes. The slides were fixed with 50% acetone: 50% methanol for 2 minutes and stained with May-Gruenwald stain for 10 minutes. Cell morphology was examined by phase contrast microscopy. Granulocytes transmigrating through the endothelial monolayer were 100% neutrophils.
For chemotaxis experiments, unstimulated HUVEC were co-cultured with Mf in the lower well and after 24 h either 1×106 PBMC or granulocytes were added in RPMI-1640 supplemented with 10% FBS into the upper well. After 4 h, cells that had migrated into the lower well were analysed by flow cytometry.
Mouse anti-human CD8 antibodies were prepared by growing the OKT8 hybridoma in RPMI-1640/10% FBS in vitro. Supernatant was harvested after 7 days and centrifuged at 2,057×g for 10 minutes. Antibodies were purified over protein G sepharose.
Mouse anti-human CCR5 (BD), mouse anti-human CD14 (26ic) (in-house), mouse anti-human CD8 (OKT8) (in-house), PE-conjugated mouse anti-CD56 (eBioscience), FITC-conjugated mouse anti-human CD3 (eBioscience), PE-Cy5-conjugated mouse anti-human CD16 (BD Pharmingen) and the isotype control antibodies FITC–conjugated mouse IgG1 (BD Pharmingen), PE-Cy5–conjugated mouse IgG1 (eBioscience) and PE–conjugated IgG2a (BD Pharmingen) were used to stain cells. Goat anti-mouse IgG FITC conjugated antibodies (Sigma) were used as a secondary antibody with unconjugated primary antibodies. Negative control samples for unconjugated primary antibodies were solely stained with this secondary antibody. Data was acquired using a FACS Canto II (BD Oxfordshire UK) and analysed with FlowJo software (Tree Star Incorporation).
HUVEC were washed twice with ice-cold PBS. Cells were lysed using RIPA Buffer (20 mM MOPS, 150 mM NaCl, 1 mM EDTA, 1% Igepal, 1% Sodium deoxycholate and 0.1% SDS supplemented with a 1∶1000 concentration of protease inhibitor mix (Sigma)). Genomic DNA was broken up by mechanical syringe action. The lysates were centrifuged at 10,400×g for 10 minutes at 4°C. The supernatant was kept at −80°C until use. The protein concentration was measured using a BCA protein assay (Pierce).
The cytokine protein levels in supernatants were analysed using protein arrays (RayBioTech) according to the manufacturer's instructions. The dot intensity on the membranes when exposed to X-ray film was measured using QuantityOne Software (BioRad Laboratories). Subsequently, the levels of cytokines were analysed using the RayBio Analysis Tool for the human cytokine antibody array I (RayBioTech).
Protein expression was detected in cell lysates by SDS-PAGE followed by Western blotting. Blots were incubated with mouse anti-human heme oxygenase-1 (HO-1) (BD transduction Laboratories) or mouse anti-human β-actin antibodies. HRP-conjugated rabbit anti-mouse (Dako) antibody was used for detection and developed with SuperSignal West Pico Chemiluminescent Substrate (Pierce).
ELISA was used to measure human IL-1β, TNF-α, IL-6, IL-8, CCL2, TGF-β1 (R&D Systems) and IL-13 (Pelikine Compact) in EC supernatant or lysate according to the manufacturer's instructions.
Total EC RNA was harvested using QIAShredder and RNeasy kit as advised by the manufacturer (Qiagen, Brighton UK). RNA quantity and quality were evaluated using the NanoDrop ND-1000 spectrophotometer. Total RNA integrity was verified using agarose gel electrophoresis and ensuring that 18 and 28S ribosomal RNA bands were intact.
Oligo microarrays for chemokines and chemokine receptors (Supp. Table 1–2), and angiogenesis mediators (Supp. Table 5) were purchased from SuperArray Bioscience (UK) and were also used according to the manufacturer's instructions. The dot intensity of the oligo microarrays when exposed to X-ray film was measured using GE Array Analysis Suite (SuperArray Bioscience). Sample values were considered to be different, if both values from two duplicate experiments were either lower or higher in gene expression units than the comparative samples, and if the means of the duplicates differed by at least a factor of 4 in gene expression units. The reference value for β-actin in these experiments was 1 (chemokine & chemokine receptors) or 11 (angiogenesis mediators) gene expression units.
cDNA was synthesised from 1 µg total RNA using the QuantiTect Reverse Transcription kit (Qiagen) following the manufacturer's instructions. Primer sequences used and Entrez accession numbers for each gene are outlined in Supp. Table 4. Primers were designed using the Primer-3 Web-Software (Whitehead Institute for Biomedical Research, MA, USA) and purchased from MWG-Biotech (Ebersberg, Germany). Real-time qRT-PCR was performed as previously published [16] and quantification analysis was carried out using the MJ Research Opticon 3.1 software from standard curves with correlation coefficient (r2) greater than 0.98. Gene expression data was normalised to total RNA and presented as copy numbers. Specificity and purity of amplificons were verified from melting curves and agarose gel electrophoresis.
The Students t-test for paired data was used in all statistical analyses and performed with Prism 4 software (GraphPad Software, Inc). P values<0.05 were taken to be statistically significant. All data are presented as mean ± standard deviation.
Presence of B. malayi microfilariae in the blood vessels may alter transmigration of leukocytes across the vascular endothelium. In order to investigate this, TNF-α - stimulated HUVEC were co-cultured with or without live Mf in transwells of a transmigration assay plate and the ability of PBMC or neutrophils to transmigrate through the confluent HUVEC monolayer was analysed. The presence of live Mf did not affect the total number of extravasated lymphocytes, CD3+ (T cells), CD8+ cells or CD3−CD56+CD16− (NK cell subset) (Figure 1a–d). However, the transendothelial migration of neutrophils and CD14+, CD3−CD56−CD16+ and CD3−CD56+CD16− monocytes was significantly inhibited in the presence of live Mf (p<0.05) (Figure 1e–h).
To investigate whether Mf altered the chemotactic ability of leukocytes and/or the ability of vascular EC to chemoattract, the chemotaxis of lymphocytes and neutrophils to HUVEC in the presence of Mf was analysed. Interestingly, neutrophils were more strongly attracted to HUVEC than lymphocytes, however Mf presence did not significantly alter the chemotactic ability of either leukocyte subset (Figure 2).
Altered expression of adhesion molecules and/or chemokines by vascular EC may inhibit the extravasation of leukocytes. To further investigate the mechanism of reduced monocyte and neutrophil extravasation in Mf presence, the key adhesion molecules, ICAM-1 and VCAM-1, expressed by HUVEC were measured. The presence of B. malayi Mf did not alter the surface expression of these adhesion molecules (Figure 3a–b). The chemokines, CCL2 (MCP-1) and IL-8 are potent inducers of monocyte and neutrophil extravasation, respectively. In endothelial cell biology the amount of chemokine secreted corresponds to the level of chemokine presented at the vascular surface. Therefore, CCL2 and IL-8 were measured by ELISA, in the supernatants of HUVEC cultured with or without Mf. However, the presence of B. malayi Mf did not have a significant effect on the up- or down-regulation of either CCL2 or IL-8 (Figure 3c–d).
To further investigate whether B. malayi Mf alter the EC expression of immune mediators in their immediate environment, a comprehensive analysis of cytokine and chemokine mRNA expression was performed by oligo microarray (Figure 4a, Supp. Table 1 and 2). Since Mf are situated in the lung capillaries for long periods of time, in addition to HUVEC, we also used HLMVEC, as the latter may more closely resemble EC in the locality of Mf in vivo. Within the stringency criteria of our experiments, neither cytokine nor chemokine mRNA expression in vascular EC was found to be altered in live Mf presence (Figure 4a). However, the mediators with the highest fold increases in mRNA expression in the presence of Mf were almost identical between the two different vascular EC, HUVEC and HLMVEC. These mediators were CCL1, CCL23, IL-1α, C5, and the chemokine receptors CCR5 and CCR10 (Figure 4a).
To investigate whether live Mf stimulate and/or down-regulate the immune function of vascular EC, secretion of pro- and anti-inflammatory cytokines and chemokines was measured in the supernatants of HUVEC following exposure to Mf (Figure 4b–d). Initially, an exhaustive exploration of cytokines and chemokines produced by HUVEC in the presence of B. malayi Mf, was conducted by protein expression array in culture supernatants (Figure 4b, Supp. Table 3). Mf presence appeared not to significantly alter the secretion of any of the immune mediators tested. Indeed the array confirmed our previous results that CCL2 and IL-8 are not altered in Mf presence (Figure 3c–d and 4b). Although high levels of GRO family members (CXCL1, CXCL2, CXCL3) were detected, expression of these chemokines was not significantly enhanced by Mf.
Key inflammatory cytokines known to be produced by EC were also measured by ELISA (Figure 4c–d, and data not shown). In confirmation of the protein array data, the secretion of IL-6, TGF-β1, TNF-α and IL-1β by HUVEC was not altered in Mf presence. Indeed, IL-1β and TNF-α were not detected in the HUVEC supernatant in the presence or absence of Mf (data not shown). IL-13 was not found in HUVEC supernatants but was detected in HUVEC lysates by ELISA (Figure 4e), however, live Mf did not alter the protein expression of this cytokine.
While the oligo microarray analysis did not show any mediators significantly up- or down-regulated in HUVEC or HLMVEC in the presence of Mf; we sought to more definitively determine whether the mediators with highest mRNA expression levels were altered by Mf. Therefore, we used qRT-PCR to analyse the mRNA levels of selected genes. In accord with the oligo microarray data and the applied analysis criteria, qRT-PCR confirmed that HUVEC mRNA expression of CCL1 (not detected), CCL23 and IL-1α were not altered by Mf presence (Figure 5a–b). In addition, Mf presence caused no alteration in CCL1 (not detected) and IL-1α mRNA expression in HLMVEC (Figure 5b). However, CCL23 mRNA was significantly (p<0.0001) downregulated upon Mf exposure. Interestingly, Mf presence also caused a down-regulation of the mRNA levels of pro-inflammatory C5 in both HUVEC and HLMVEC as analysed by qRT-PCR (Figure 5c).
Expression of the chemokine receptors, CCR5 and CCR10, were also further analysed in both HUVEC and HLMVEC exposed to live Mf (Figure 6a–c). Initially, HUVEC co-cultured with or without B. malayi Mf were analysed by flow cytometry for surface expression of CCR5 (Figure 6a). CCR5 was found to be significantly upregulated on the surface of HUVEC exposed to live Mf (p<0.05). However, mRNA expression analysis of HUVEC and HLMVEC CCR5 by qRT-PCR did not show alteration in the presence of live Mf. Furthermore, qRT-PCR for CCR10 revealed that live Mf increased mRNA in both HUVEC and HLMVEC, however this was only significant in HLMVEC (p<0.05) (Figure 6c). The oligo microarray analysis of other chemokine receptors revealed no other differences in mRNA expression in either HUVEC or HLMVEC in the presence of Mf (Figure 4a). In some instances therefore, discrepancies existed between the recognition of mRNA by the primers used in qRT-PCR and the sensitivity of the probes used on the oligo-microarray.
Live Mf could be responsible for increased levels of pro-angiogenic mediators found in the sera of filarial patients [8], [9]. Furthermore, previous studies have suggested that the filarial endosymbiotic bacteria, Wolbachia, may induce angiogenesis [8].
In order to investigate whether live Mf induce angiogenic mediators in EC, the mRNA expression of these mediators in HUVEC and HLMVEC following co-culture with B. malayi Mf was analysed (Figure 7, Supp. Table 5). In addition, to determine whether Wolbachia endosymbionts are responsible for any angiogenic mediator induction, oligo microarray of mRNA from EC cultured with either intact Mf or Mf-depleted of Wolbachia were compared in two separate experiments. These studies showed that angiogenic factors were not altered in HUVEC in the presence of either intact Mf or Wolbachia-depleted Mf (Figure 7a). A qRT-PCR analysis of several mediators with the highest fold change expression in Mf presence (angiopoietin-2 (Ang-2), brain-specific angiogenesis inhibitor-1 (BAI-1), tumor necrosis factor superfamily member 15 (TNFSF15), cyclooxygenase-2 (COX-2) and CCL11), confirmed these results (Figure 7b, Supplementary Figure 1). However, in HLMVEC, Ang-2 mRNA was downregulated and the angiostatic factor TNFS15 was upregulated by live Mf (Figure 7b,d). Further analysis of the pro-angiogenic mediator COX-2, showed that this mediator was upregulated in HLMVEC, but not HUVEC, by Mf presence (Figure 7b).
Live B. malayi Mf. enhanced, protein expression of the hypoxia-induced product, heme oxygenase-1 (HO-1), in HLMVEC after 6, 12 and 24 h of co-culture with Mf (Figure 7f). Following IFN-γ-stimulation, HO-1 was no longer detectable at any of these time points. Interestingly live Mf also induced relatively high levels of hypoxia-inducible factor (HIF-1α) mRNA in HLMVEC (Figure 4a). Presence of HIF-1α is an indicator of hypoxia which in turn is a potent promoter of angiogenesis.
In an area endemic for lymphatic filariasis, the majority of people have asymptomatic infection and harbour several million Mf in their blood stream. Vascular EC play an important role in mediating immune and angiogenic responses. Therefore, maintenance of this asymptomatic condition, as well as survival of B. malayi Mf in the blood stream, could depend upon Mf-driven modulation of EC activity. In this study we sought to investigate the vascular EC response upon exposure to live B. malayi Mf. We found that the transendothelial migration of monocytes and neutrophils, but not lymphocytes, is inhibited by live Mf presence; while either intact or Wolbachia-depleted Mf stimulate few cytokines, chemokines or angiogenic mediators.
Both macrophages and neutrophils are capable of killing B. malayi Mf in vitro [17], [18], [19]. Reduced extravasation of monocytes and neutrophils could therefore lead to retention of effector leukocytes in the vascular location of the parasite, resulting in increased clearance of Mf. Indeed, reduced eosinophil extravasation, in eotaxin-1−/− mice infected with B. malayi Mf, lead to eosinophil retention in the blood stream and enhanced Mf clearance [20]. Both monocytes and neutrophils appear to kill Mf via production of reactive intermediates, however, in turn, Mf can partially neutralise the toxic effects of these intermediates by secreting anti-oxidant enzymes such as peroxidases and superoxide dismutase [17], [21], [22], [23].
There are a number of potential mechanisms, which could result in the inhibition of monocyte and neutrophil transendothelial migration in the presence of Mf. In general, leukocyte extravasation is a multi-step cascade involving rolling mediated by selectin-selectin ligand axes, tethering mediated by integrin-adhesion molecule axes strengthened by chemokine-triggered activation and finally, diapedesis [24]. Selectin-selectin ligand axes are unlikely to have a functional role in the static transendothelial migration experiments performed in this study. Furthermore, HUVEC surface expression of the adhesion molecules ICAM-1 and VCAM-1, which have crucial roles at the tethering step, was not modulated upon Mf exposure. Another potential mechanism investigated was the possibility that alteration(s) in chemokine expression in the presence of Mf selectively interfered with leukocyte tethering.
IL-8 and CCL2 are considered to be the most important chemokines for the transendothelial migration of neutrophils and monocytes respectively [25]. However, Mf presence had no effect on IL-8 or CCL2 production by EC. In accord with this, a comprehensive examination (by oligo microarray, qRT-PCR and protein array) of chemokines in EC exposed to Mf, did not reveal any major differences in these or other chemokines that may have a role in extravasation of monocytes and neutrophils. No alteration in the transendothelial migration of whole T cells, CD8+ cells or NK cells was observed in Mf presence.
Interestingly, in agreement with our study, experiments in mice implanted with adult B. malayi or Mf in vivo showed that in the presence of Mf alone, infiltration of leucocytes into the peritoneal cavity is reduced in comparison to adult nematode implanted mice [26]. In addition, the proportion of macrophages within these leukocyte populations was significantly lower in Mf implanted rather than adult implanted mice [26]. Previous work has also shown that a serine protease derived from B. malayi Mf abolishes C5a-mediated chemotaxis of granulocytes [27]. Furthermore both adult B. malayi and Mf extracts inhibit hyper-permeability induced by TNF-α or IL-1α, of lymphatic EC monolayers to dextran. Although neither extract showed any effect on the permeability of confluent EC per se [10]. Dirofilaria immitis adult extracts, however, did reduce the transendothelial permeability of a human EC line. Enhanced expression of tight junction and/or adherence molecules by both Brugia and Dirofilaria extracts has been shown to be the likely mechanism of this reduced permeability [10], [28], [29]. Indeed, Wolbachia surface protein (WSP), but not whole D. immitis extract, induced the expression of ICAM-1 and VCAM-1 on a human EC line [28], and, WSP or D. immitis extracts up-regulated CD31 on this EC line [28], [29]. Strikingly, lymphatic EC exposed to Brugia adult or Mf extract also had higher mRNA levels of CD31, in addition to, VE-cadherin and Junctional Adhesion Molecule-C (JAM-C) [10]. If live Mf presence also causes elevated expression of these intercellular adhesion molecules, this may provide an explanation for the retention of monocytes and neutrophils while the transmigration of smaller lymphocytes is not affected.
In addition, we investigated whether live B. malayi Mf initiate immune responses in their local environment. Perhaps, not surprisingly, live B. malayi Mf (as opposed to extracts [10], [21]) appear to be relatively inert in their local vascular environment and do not induce significant levels of pro-inflammatory immune mediators from EC, such as IL-6, TNF-α or IL-1β. Interestingly, Mf also did not induce increased levels of IL-13, which promotes alternatively-activated macrophages (AAMø), or the down-regulatory cytokines IL-10 or TGF-β1. However, Mf presence did down-regulate mRNA expression of the inflammatory complement component, C5, in HUVEC and HLMVEC. In light of the recent report, that B. malayi Mf secrete a C5a-cleaving serine protease, this suggests that C5 products may be potentially damaging to filarial nematodes [27]. Similarly, mRNA expression of CCL23, a chemoattractant for monocytes, neutrophils and T cells, was downregulated in HLMVEC. Both of these latter observations indicate that Mf may modulate inflammatory responses. This is in accord with the fact that most filarial patients are asymptomatic and have down-regulated cellular immune responses to filarial antigens, however, when given therapeutic treatments, patients subsequently regain responses to filarial antigen [30]. This also suggests that while the inflammatory potential of B. malayi is dependent on the presence of Wolbachia [31], [32], Wolbachia and/or their products are not released or secreted from living Mf to induce inflammatory mediators in their local environment. However upon death of worms, Wolbachia and their inflammatory products such as lipoprotein, which has been shown to stimulate both innate and adaptive immunity are released [31], [33], [34].
Interestingly, we observed that Mf exposure induced upregulation of the hypoxia-responsive mediator HO-1 in HLMVEC, indicating that Mf may induce hypoxia, which is an angiogenesis-promoting condition. Furthermore in HLMVEC Mf upregulated mRNA for hypoxia-inducible factor (HIF-1α) which is known to induce HO-1. In addition to hypoxia and HIF-1, HO-1 can be induced by other components such as heme, IL-6, IL-1 or LPS in a number of model systems [35], [36], [37], [38], [39], [40], and is often used as a marker of inflammatory as well as oxidative stress. Neither IL-1 nor IL-6 increased in the EC supernatant following incubation with Mf. However, as Wolbachia spp. are an endosymbiotic bacteria of Brugia malayi, they produce heme [41]. Therefore, it is possible that Wolbachia spp. derived heme is a trigger of HO-1 production by HLMVEC. Additionally, HO-1 expression was repressed upon IFN-γ-stimulation of HLMVEC. Previous work has also shown that IFN-γ inhibits HO-1 in various cell types [42], [43] however, to the best of our knowledge the role of this mechanism has not been investigated in a functional context. While HO-1 has anti-inflammatory properties, Mf are also known to induce IFN-γ [44], [45], [46] thus the role HO-1 induction by Mf warrants further investigation.
Surprisingly, live intact Mf did not stimulate the expression of many angiogenic mediators including the key mediator, VEGF-A, in vascular EC, although, live Mf did stimulate pro-angiogenic COX-2 in HLMVEC. This is in line with previous work in which Simόn et al. found that Wolbachia surface protein from D. immitis and adult somatic antigen from D. immitis, both induce COX-2 in a human EC cell line [28], [29].
Mf also enhanced the surface expression of CCR5, which binds CCL3 (MIP-1α), CCL4 (MIP-1β) and CCL5 (RANTES), and mRNA expression of CCR10, which binds to CCL27 and CCL28 in vascular EC. The role of this increased expression of chemokine receptors is not clear, however, CCR5 itself is known to mediate angiogenesis [47].
Other work has investigated the potential of filarial antigen, and/or live worms, to initiate vessel dilatation and/or angiogenesis by measuring EC proliferation and tube formation in vitro [10], [28], [29], [48]. The results varied depending on differing use of HUVEC, lymphatic EC (LEC) or a human vascular EC cell line and parasite extracts or live nematode stages. For example, live female B. malayi decreased HUVEC proliferation [48] while LEC, but not HUVEC, cultured with adult Brugia or Mf extract showed increased proliferation [10] and D. immitis adult extract had no effect on the proliferation of a human EC cell line [29]. Live Mf and adults and their extracts all induced tube formation in LEC, however, in our experiments using live B. malayi Mf with vascular EC, both HUVEC and HLMVEC, we did not observe these structures [10].
In this study we report new insights into the EC response to live B. malayi Mf in their vascular environment, albeit within by the limitations of an ex vivo model which uses an EC isolate under static conditions incubated with parasites. Upon Mf exposure, extravasation of monocytes and neutrophils was partially blocked, while the transendothelial migration of lymphocytes was not altered. However, overall, Mf induced the expression of only a small number of cytokines, chemokines or pro-angiogenic mediators in human vascular EC. Furthermore, depletion of Wolbachia from live Mf did not significantly alter mRNA expression of these mediators. Taken together, our study suggests that live Mf are either relatively inert or that they are able to modulate local responses to promote their own survival and limit infection-induced pathology. Alternatively, Mf may induce a highly localised response mediated by other cells not present in this model system, rather than, a direct interaction between Mf and endothelium.
|
10.1371/journal.pmed.1002152 | The Global Burden of Latent Tuberculosis Infection: A Re-estimation Using Mathematical Modelling | The existing estimate of the global burden of latent TB infection (LTBI) as “one-third” of the world population is nearly 20 y old. Given the importance of controlling LTBI as part of the End TB Strategy for eliminating TB by 2050, changes in demography and scientific understanding, and progress in TB control, it is important to re-assess the global burden of LTBI.
We constructed trends in annual risk in infection (ARI) for countries between 1934 and 2014 using a combination of direct estimates of ARI from LTBI surveys (131 surveys from 1950 to 2011) and indirect estimates of ARI calculated from World Health Organisation (WHO) estimates of smear positive TB prevalence from 1990 to 2014. Gaussian process regression was used to generate ARIs for country-years without data and to represent uncertainty. Estimated ARI time-series were applied to the demography in each country to calculate the number and proportions of individuals infected, recently infected (infected within 2 y), and recently infected with isoniazid (INH)-resistant strains. Resulting estimates were aggregated by WHO region. We estimated the contribution of existing infections to TB incidence in 2035 and 2050.
In 2014, the global burden of LTBI was 23.0% (95% uncertainty interval [UI]: 20.4%–26.4%), amounting to approximately 1.7 billion people. WHO South-East Asia, Western-Pacific, and Africa regions had the highest prevalence and accounted for around 80% of those with LTBI. Prevalence of recent infection was 0.8% (95% UI: 0.7%–0.9%) of the global population, amounting to 55.5 (95% UI: 48.2–63.8) million individuals currently at high risk of TB disease, of which 10.9% (95% UI:10.2%–11.8%) was isoniazid-resistant. Current LTBI alone, assuming no additional infections from 2015 onwards, would be expected to generate TB incidences in the region of 16.5 per 100,000 per year in 2035 and 8.3 per 100,000 per year in 2050.
Limitations included the quantity and methodological heterogeneity of direct ARI data, and limited evidence to inform on potential clearance of LTBI.
We estimate that approximately 1.7 billion individuals were latently infected with Mycobacterium tuberculosis (M.tb) globally in 2014, just under a quarter of the global population. Investment in new tools to improve diagnosis and treatment of those with LTBI at risk of progressing to disease is urgently needed to address this latent reservoir if the 2050 target of eliminating TB is to be reached.
| Addressing the latent TB infection reservoir is critical to achieving TB elimination.
The current estimate that “one-third” of the global population is infected with tuberculosis is widely cited but has not been formally estimated for nearly 20 y.
Changes in demography, the size and distribution of TB burden, as well as new scientific insights and the availability of new data mean a re-estimation is needed.
We generated an annual risk of infection between 1934 and 2014 and applied this to a country-level demographic model, quantifying uncertainty wherever possible.
We estimated that approximately 1.7 billion individuals were infected with LTBI in 2014; just under a quarter of the global population.
If left unaddressed, the current LTBI burden alone will likely prevent achieving the global TB targets for TB elimination.
For long-term TB control to be successful, an aggressive approach to LTBI is needed.
Research and development should focus on developing better tools to identify individuals who will benefit from LTBI treatment.
Estimates would be strengthened by additional empirical data from new population-based studies of LTBI prevalence.
| Infection with Mycobacterium tuberculosis (M.tb) is the precursor to TB disease, which is responsible for 1.5 million deaths each year—more than any other infectious disease [1]. Once infected, the individual is at highest risk of developing TB disease within the first two years, but can remain at risk for their lifetime [2]. The population carrying a latent TB infection (LTBI) is commonly quoted as “one-third” of the global population, a reservoir of approximately 2.3 billion individuals [3–6].
As the global community looks to meet ambitious targets for reduction (90% reduction in TB incidence by 2035) and even elimination of TB (less than 1 incident case per 1,000,000 per year) by 2050 [7], our ability to address the LTBI reservoir will be critical in our chance to succeed.
Despite its clear importance to global TB control efforts, the most recent attempt to estimate the global burden of LTBI was in 1998 [3]. Since then, the size and distribution of the global population [8] and TB burden [1] has changed dramatically, as has our understanding of prevalent disease as a driver of infection [9,10]. Global population growth from around 6 billion in 1998 to over 7 billion in 2014 has been mainly driven by areas with the highest TB burden, such as Southeast Asia and sub-Saharan Africa [1,8]. The previous estimation method relied on a fixed relationship between TB burden to the annual risk of acquiring LTBI, the so-called “Styblo rule” [3]. Since then, two groups have shown this long-held rule of thumb substantially overestimates infection risk in modern populations [9,10].
Given these changes and the drive towards eliminating TB [7,11], an updated estimate of the global burden of LTBI that incorporates the available data and applies current scientific insights is urgently needed [4,12].
An updated estimate of the size and distribution of the LTBI reservoir should also address questions about the likely contribution of LTBI to TB disease over the coming decades. Specifically, how many active TB cases would arise from the currently infected individuals alone if all transmission was halted now? Updated LTBI burden estimates also indicate the population in need of interventions and new tools, thus catalyzing new research and potential investment from commercial partners in, for example, vaccines, and tools for the diagnosis and treatment of LTBI [4].
Critical questions include the number of those with LTBI at highest risk of developing disease, i.e., those infected within the past 2 y [12]. This population is a focus of proposed “test and treat” strategies in TB, which would use an RNA expression profile test to identify individuals most likely to develop TB [13], improving on the low predictive value of existing tests for LTBI [14]. As resistance to TB drugs is rising, an estimate of the proportion of LTBI that involves isoniazid (INH)-resistant strains is important, since INH remains the cornerstone of most treatment regimens for LTBI [4]. Finally, as TB becomes rarer, the epidemiology of LTBI will have a renewed and increasing importance for monitoring the progress of control efforts.
In this paper, we estimate the global burden of LTBI and its distribution by country, geographical region, and age group. We also estimate the number of recent infections and the number of recent infections with INH-resistant strains. Finally, we predict the TB incidence in 2035 and 2050 solely due to the existing LTBI reservoir.
To estimate the burden of LTBI, we reconstructed country trends in annual risk of M.tb infection (ARI) and combined these historical projections with demographic data to estimate burden of recent and all-time infection by age. ARI trends were modelled for 168 countries (comprising >99.9% of the world population) using a flexible non-parametric regression framework accounting for measurement uncertainty and applied to two sources of data.
The first source of data was direct estimates of ARI from tuberculin skin test (TST) surveys. We abstracted data on estimated ARI, survey sample size, and mean age for 100 country-years in 24 countries from Cauthen et al. [15] and undertook a systematic review of nationally representative ARI estimates in the years 1990–2014 (see Methods and Figure A in S1 Text), yielding further data on 31 country-years in 19 countries, to give a total of 37 unique countries with TST survey data. Historically, ARIs have been estimated from TST surveys without presentation of uncertainty. In the Methods section of S1 Text, we show how sample size and mean age can be used to approximately and conservatively quantify measurement precision. We used this method for studies not stating ARI estimate precision.
The second source of data was indirect: combining WHO estimates of TB prevalence (5,373 country-years for 218 countries) with an uncertain representation of the revised Styblo ratio that accounts for uncertainty and relates the prevalence of smear positive disease to ARI [1,9,10]. A previously published study of childhood tuberculosis [16] characterized this ratio in the modern era by fitting a log-normal distribution to data from reviews of studies in which both ARI and prevalence estimates were available [9,10]. To estimate the proportion of prevalent TB that is smear positive for each country, we averaged estimates of smear positivity for 0–4, 5–14, and ≥15 y age groups from a recent systematic review [17] against the proportion of cases in these age-groups calculated using the model of Dodd et al. [16]. To calculate the impact of HIV on the proportion of TB in a country that is smear positive, we first calculated the fraction of prevalent TB in people living with HIV (PLHIV) by adjusting the WHO estimates of HIV prevalence in incident TB for each year using estimates of the duration of prevalence in PLHIV and HIV-uninfected individuals [1]. The mean smear positivity of HIV-TB was then reduced by a fraction reported in Corbett et al. [18]. The uncertainty of each ingredient in these ARI calculations was propagated using the delta method. More details are described in the Methods section of S1 Text.
Gaussian process regression with a linear trend was applied to the data on ARI (on a log scale), using the measurement precision calculated for each data point. This implicitly combines the data feeding into WHO prevalence estimates (and Styblo ratio) and the TST data feeding into ARI estimates, with the assumption of a normal approximation to the likelihood. A sensitivity analysis re-analysed these data with a constant rather than linear trend assumption. To allow a comparison with the 1998 estimate, we assumed a constant ARI before 1934.
For each country, 200 simulated ARI trajectories from 1934 to 2014 were used to compute the cumulative hazard of infection for individuals by age. The cumulative hazard was converted into a probability of infection and combined with UN Population Division estimates of country demography in 2014 to give our estimates of all-time infection. We computed the probability of infection for the first time within 2 y, using the cumulative hazard up to 2012 to calculate the fraction at each age who had escaped infection until then, and the cumulative hazard from 2012 to 2014 to calculate the fraction of these who were then infected. To calculate the fraction of those at a given age infected or re-infected during the last 2 y, we introduced a beta distribution characterizing an uncertain partial protection against re-infection of 79% (70%–86%) from Andrews et al. [19]. See the Methods section of S1 Text for details. As estimates of protection have varied [20,21], we conducted a sensitivity analysis using a protection of 50% with the same variance. Estimates of infection prevalence were summarized by region, age, and medians mapped by country.
To calculate the number of infections within the last 2 y with resistance to isoniazid, we combined our estimates of infection within 2 y in each country with a recent analysis of the proportion of new infections in each country that are isoniazid-resistance using data from the Global Project on Anti-tuberculosis Drug Resistance Surveillance at WHO [22]. This proportion was treated as uncertain and sampled from the output of this analysis. A conceptual overview diagram for the methods is presented in Figure R in S1 Text.
Finally, we estimated the regional prevalence of latent infection in 2035 and 2050 under the assumption of no M.tb transmission after 2014, using UN Population Division demographic projections, and calculated the likely implications of existing M.tb infections for future TB incidence, assuming a 0.15% per year remote activation rate [21,23,24].
Results are reported as medians together with 95% uncertainty intervals (95% UI), calculated as the 2.5% to 97.5% percentile range. No specific funding was received for this work.
The ARI estimates from TST surveys were comparable with the ARI estimates from WHO TB prevalence estimates via the updated Styblo rule and typically did not exhibit discontinuities (Figures C–H in S1 Text). Fig 1 shows the results for the WHO Southeast Asia region. Fig 1 also illustrates uncertainty increasing at earlier times, away from data.
We estimate a global prevalence of latent M.tb infection in 2014 of 23.0% (95% UI: 20.4%–26.4%) (Table 1). This amounts to an estimate for the worldwide prevalence of M.tb infection of 1.7 billion (95% UI: 1.5 billion–1.9 billion) in 2014 (Table 2). Fig 2 highlights the substantial regional and sub-regional variation in LTBI prevalence. The WHO Southeast Asia, Western Pacific, and Africa regions were all estimated to have LTBI prevalence in the general population of above around 20% (see first column of Table 1), whereas the WHO Eastern-Mediterranean, Europe, and Americas regions all had general population LTBI prevalence of below 17%. The large populations and high proportion infected imply that around 80% of the number of people with latent infection are in the WHO Southeast Asia, Western Pacific, and Africa regions, compared to 65% of the total population. On the country level, China and India had the highest LTBI burden, approximately 350 million infections, followed by Indonesia at around 120 million infections and fewer than 60 million infections in all other countries. The USA had the 20th highest burden, at an estimated 13 million (Figure J in S1 Text).
The proportions of each age group infected by region are shown in Fig 3, and column 2 in Table 2 shows the percentage of all LTBI in children under 15 y old. While around 6% of M.tb infections are in children globally, 13% of infections in Africa are in children, compared to 2% in the Americas. In all regions, the proportion infected rises with age and, with the exception of the WHO Europe and Americas regions, exceeds 50% in the oldest age groups. The substantial increases in TB burden in Africa and Southeast Asia are reflected in the shape of Fig 3, with more rapid increases in the younger age groups compared to other regions. Uncertainty in these proportions is largest for the Western Pacific region, particularly in older age groups, due to larger uncertainty in historical ARIs there.
Around 1% of the global population, approximately 56 million individuals, was infected within the last 2 y and would therefore be at appreciable risk of progressing to active TB (see Tables 1 and 2). Of these recent infections, the vast majority are infections for the first time; however, this, too, varies by age (Figure J in S1 Text). We estimate that around 11% of these recent infections involved an isoniazid-resistant strain of M.tb, amounting to around 6 million individuals at elevated risk of TB in whom isoniazid-preventive therapy would be ineffective (see Tables 1 and 2). There is strong regional variation, with around 30% of the recent infections in the European region involving an isoniazid-resistant strain.
If we assume no ongoing transmission from 2015 onwards, our projections of current LTBI burden imply that in 2035, 961 (95% UI: 870–1,113) million individuals would still be infected (around 11% of the population). By 2050, these numbers would be 599 (95% UI: 557–668) million (around 6%). Assuming a remote LTBI activation rate of 0.15% per year, this implies a TB disease incidence from these latent pools of 16.5 (95% UI: 14.9–19.2) per 100,000 per year in 2035, which is above the 10 per 100,000 per year target in the End TB Strategy. In 2050, the rate would be 8.3 (95% UI: 8.6–10.6) per 100,000 per year; nearly two orders of magnitude higher than the 2050 elimination target of 1 per million per year.
Sensitivity analysis assuming constant trends in the Gaussian process regression (Figures L–Q in S1 Text) led to smaller estimates of LTBI burden due to lower extrapolated ARIs at earlier times. In general, estimates in prevalence were around 20% lower with this assumption, giving global estimates of 18.5% (95% UI: 17.0%–20.7%) LTBI prevalence or 1.3 (95%UI: 1.2–1.5) billion infected individuals. Considering a lower protection against reinfection of 50% made little difference to overall numbers but resulted in more recent infections at older age groups (see Tables A–D and Figure Q in S1 Text). Applying our method to estimate the population in 1997 yielded an LTBI prevalence of 26.9% (95% UI: 22.4%–32.7%).
The global burden of LTBI is just under a quarter of the population—around 1.7 billion individuals—with substantial geographical and age variation. We estimate 56 million people are at high risk of developing TB disease because of a recent (re-)infection, 11% of whom are carrying an isoniazid-resistant strain. With reasonable assumptions for reactivation risks, incident TB disease arising from the 2014 LTBI reservoir alone would prohibit reaching the 2035 and 2050 End TB Strategy goals.
Using our method to calculate LTBI prevalence in 1997 yielded 27%, suggesting the difference between our estimate for 2014 and the 1997 estimate [3] is mainly due to changes in methodology, including the revised Styblo rule [9,10]. Our results also matched a recent survey-based estimate of 13 million latent infections for the United States [25].
One limitation of this study was that we were not able to consider the effects due to different tuberculin strains and cut-off choices for TST tests or the potential impact of BCG vaccination on these test outcomes. Recent work has suggested that LTBI may be a dynamic and heterogeneous state [12] and highlighted the difference between M.tb infection and positive tests for infection. Our work does not make this distinction but is in line with literature relating TB prevalence to infection risk and risks of progression to disease. In addition, we assumed lifelong LTBI in common with previous estimates and consistent with observation [2]. While it may be biologically plausible that some individuals clear their LTBI in absence of treatment, there is no published evidence to support a separate model scenario.
Our approach to quantifying the typical infectiousness of prevalent TB cases by country entailed some necessary simplifications. Notably, we assumed the same TB case-detection rate applied independent of HIV status and anti-retroviral treatment status. However, these assumptions had relatively little influence on ARI estimates and their uncertainty (see S1 Text, page 8), with the dominant contributions to uncertainty coming from the TB prevalence estimates themselves and uncertainty in the Styblo ratio. We also assumed that the same ARI applied to all individuals in a country and neglected migration between countries.
A major strength of this study is its treatment of uncertainty. We were able to characterize and include measurement precision for data on ARI derived from TST surveys as well as indirectly from prevalence estimates. While uncertainty in ARI estimates grew substantially for the earliest years, the number of people alive today who were alive then was small, limiting the impact of this uncertainty on our overall estimate. This also limited the impact of our assumption of linear trends extrapolating backward from data; our very conservative assumption of flat trends resulted in less than a five-percentage-point difference in our overall prevalence estimate.
Our estimate of TB incidence in 2035 and 2050 from currently existing LTBI uses a single reactivation rate. This parameter has been estimated at widely varying levels, [23,24,26], and we have not attempted to include potential geographic heterogeneity or anticipate trends due to changes in population health, for example, through achievements of the Sustainable Development Goals (SDG) [27] or due to cohort aging. However, our estimates of TB disease incidence in 2035 and 2050 are proportional to this parameter, and even large reductions do not alter the conclusion that incidence will exceed the 2050 elimination target.
It is a limitation that there were not more recent direct national estimates of ARI, which would have strengthened our estimates. More empirical data on the epidemiology of LTBI, including from the use of modern tests that are less prone to biases of interpretation and cross-reaction [28], should improve our understanding of these features and generate more data upon which to base estimates. As active TB becomes rarer, surveys of infection will become an increasingly appealing option for monitoring trends, but this relies on understanding the relationship between infection and other burden metrics.
Policies to address the LTBI reservoir have to balance the potential of harm versus the benefit for the individual [29]. In the current landscape of diagnostic tools, WHO recommends LTBI testing and treatment only in high-risk groups, such as people living with HIV, and close contacts of TB cases. While these guidelines are sensible, it is clear that a more aggressive approach is needed to reduce the threat to long-term TB control targets stemming from a LTBI reservoir of approximately 1.7 billion individuals. Future work with this model could inform current policies by estimating the burden of LTBI in specific risk groups, such as people living with HIV or diabetes. A test to more precisely identify those at substantial risk of progressing to disease could enable targeted LTBI treatment beyond known risk groups [12]. Emerging tests based on RNA signatures may come to provide a more practicable method of identifying individuals for LTBI treatment [13]. Among biomedical interventions, a vaccine that prevents progression to disease from LTBI could make a major contribution, depending on global availability [30]. Beyond the biomedical perspective, improvements in social and economic conditions globally have been associated with reductions in TB burden in historic and contemporary contexts [31,32], and could also contribute to reducing the TB burden originating from the LTBI reservoir.
Treatment for LTBI still relies heavily on isoniazid, either as monotherapy or as part of a combination regimen [4,11]. We found that just under 11% of all recent M.tb infections are likely to be isoniazid resistant, with much higher rates in some regions, and this proportion is likely to increase. While less common, rifampicin resistance also has the potential to threaten the usefulness of rifampicin-containing prophylactic regimens. New treatments that bypass the rising resistance to isoniazid and rifampicin are needed to fully operationalise interventions to test and treat LTBI.
We estimate that approximately 1.7 billion individuals were latently infected with M.tb globally in 2014, just under a quarter of the global population. Investment in new tools to improve diagnosis and treatment of those with LTBI at risk of progressing to disease is urgently needed to address this latent reservoir if the TB community is to reach the 2050 target of eliminating TB.
|
10.1371/journal.pntd.0005709 | Identifying wildlife reservoirs of neglected taeniid tapeworms: Non-invasive diagnosis of endemic Taenia serialis infection in a wild primate population | Despite the global distribution and public health consequences of Taenia tapeworms, the life cycles of taeniids infecting wildlife hosts remain largely undescribed. The larval stage of Taenia serialis commonly parasitizes rodents and lagomorphs, but has been reported in a wide range of hosts that includes geladas (Theropithecus gelada), primates endemic to Ethiopia. Geladas exhibit protuberant larval cysts indicative of advanced T. serialis infection that are associated with high mortality. However, non-protuberant larvae can develop in deep tissue or the abdominal cavity, leading to underestimates of prevalence based solely on observable cysts. We adapted a non-invasive monoclonal antibody-based enzyme-linked immunosorbent assay (ELISA) to detect circulating Taenia spp. antigen in dried gelada urine. Analysis revealed that this assay was highly accurate in detecting Taenia antigen, with 98.4% specificity, 98.5% sensitivity, and an area under the curve of 0.99. We used this assay to investigate the prevalence of T. serialis infection in a wild gelada population, finding that infection is substantially more widespread than the occurrence of visible T. serialis cysts (16.4% tested positive at least once, while only 6% of the same population exhibited cysts). We examined whether age or sex predicted T. serialis infection as indicated by external cysts and antigen presence. Contrary to the female-bias observed in many Taenia-host systems, we found no significant sex bias in either cyst presence or antigen presence. Age, on the other hand, predicted cyst presence (older individuals were more likely to show cysts) but not antigen presence. We interpret this finding to indicate that T. serialis may infect individuals early in life but only result in visible disease later in life. This is the first application of an antigen ELISA to the study of larval Taenia infection in wildlife, opening the doors to the identification and description of infection dynamics in reservoir populations.
| Although tapeworm parasites of the genus Taenia are globally distributed and inflict enormous socioeconomic and health costs on their hosts, which include humans, little is known about taeniid tapeworms that infect wildlife. This gap in knowledge prevents an assessment of the potential for these parasites to infect humans and production animals and is largely due to the difficulty of conducting standard diagnostic tests on wildlife. To address this gap, we adapted a standard diagnostic assay to be used with dried urine samples. We used urine from geladas, primates endemic to Ethiopia, which are frequently infected with the larval stage of a taeniid tapeworm and exhibit protuberant cysts during advanced infection. The use of this diagnostic test in a wild gelada population allowed us to detect that individuals can be infected without exhibiting observable cysts, and that some individuals may control infection in its early stages. This tool provides information about how a neglected tapeworm functions in a wildlife system and opens the door to the non-invasive identification of tapeworm reservoir hosts that may threaten humans.
| Tapeworm parasites of the genus Taenia are globally distributed in numerous mammalian hosts, frequently exploiting predator-prey relationships and posing considerable risk to humans. Although the life cycles and zoonotic potential of some taeniids are among the most well known of all tapeworms, due to their importance in human health and evolution [1], the descriptions of other taeniids have been neglected. Particularly enigmatic is Taenia serialis, conventionally thought to infect dogs in its adult stage and rodents and lagomorphs in its intermediate stage [2]. Over the past century, extensive taxonomic and morphological confusion and disagreement [1,3] have made it difficult to identify the geographic and phylogenetic distribution of this parasite. Thus, we begin by providing what is, to our knowledge, the first thorough review of T. serialis biology and zoonotic potential by synthesizing previous case reports. We then describe the antigen enzyme-linked immunosorbent assay (ELISA) that we validated for use with gelada urine samples. Finally, we demonstrate the application of this assay in a free-living population of Ethiopian geladas (Theropithecus gelada), the only known primate host of the larval stage of T. serialis, and provide recommendations for future implementation of this assay in wildlife systems.
Singular among cyclophyllidean tapeworms, taeniid species parasitize mammals in both their adult and larval stages [1]. Taeniid adult stages infect humans and carnivorous species that include canids, felids, hyaenids, mustelids, and viverrids [1, 3] and cause few severe symptoms in healthy hosts [2, 4, 5]. By contrast, taeniid larval stages (metacestodes) generally infect herbivorous artiodactyl, rodent, and lagomorph species [1, 3] and regularly cause extensive muscular and visceral damage [2, 4, 6, 7]. Intermediate hosts become infected when they ingest eggs shed by adult tapeworms harbored in the definitive host, and definitive hosts become infected when, via predation or scavenging, they ingest larvae in infected intermediate hosts [1, 3].
The scientific study of T. serialis is marked by a tendency to make species-level designations that may not be warranted and, consequently, to underestimate the range of hosts that T. serialis infects. The T. serialis metacestode is a thin-walled, translucent structure (coenurus) containing multiple protoscolices, the precursor to the mature scolex that constitutes the attachment end of the adult tapeworm in the definitive host [6]. This metacestode morphology is indistinguishable from that of T. multiceps, a zoonotic parasite found primarily in sheep [2]. Before the relatively recent emergence of molecular tools [8–13], cases of coenurosis were ascribed to either T. serialis or T. multiceps based on now-outdated morphological cues [2, 14, 15] or on infection site predilection (e.g., central nervous system or subcutaneous tissue) [2, 16, 17]. Furthermore, some researchers employed synonyms for T. serialis (e.g., T. brauni, T. glomeratus) based on geographic location or occurrence in a non-rodent or lagomorph host [17, 18].
In addition to taxonomic confusion surrounding metacestode identification, the occurrence of coenurosis ascribed to T. serialis in non-rodent or lagomorph hosts has been largely overlooked. Although parasitological texts invariably refer to T. serialis as a parasite of rodents and lagomorphs in its larval stage, it has been reported in a wide range of phylogenetically and geographically diverse hosts. Case studies have described T. serialis coenurosis in three rodent species [14, 19–22], domestic cats [23–29], two marsupial species [30, 31], two lagomorph species [32–37], and two nonhuman primate species (the greater spot-nosed guenon (Cercopithecus nictitans) [38], and the gelada (Theropithecus gelada) [39–44]. To our knowledge, only two studies of naturally occurring T. serialis coenurosis have used molecular tools for species identification [42, 43]. Given the lack of confirmed T. serialis diagnoses in the literature, including cases in ‘standard’ rodent and lagomorph hosts, it stands to reason that T. serialis may be more widespread and flexible in its selection of intermediate hosts than previously described.
The historic difficulty of definitively diagnosing T. serialis coenurosis may have also led to an underestimation of its zoonotic potential. Coenurosis has been recorded in humans across the globe [45, 46], including in Europe [47–59], Africa [60–65], the Middle East [66, 67], and the Americas [68, 69]. Certain authors declined to assign a species [17, 65], while the others ascribed infection to T. serialis or T. multiceps based on morphological analysis. Only one study used molecular tools, identifying T. serialis coenurosis in a man in Nigeria [46]. In sum, the taxonomic uncertainty of coenurosis occurring in animals, including humans, has led to a fragmented record of the global occurrence and distribution of T. serialis and a potential underestimation of its zoonotic potential and importance to public health.
As humans come into increasing contact with wildlife, understanding the biology and zoonotic potential of T. serialis is crucial to preventing its transmission to humans and domestic animals. Little is known about the natural dynamics of Taenia spp. in wildlife hosts, largely because of the impracticality of obtaining and storing biological samples or performing medical imaging in remote settings and on wildlife. To obtain a more accurate assessment of the prevalence of larval T. serialis infection in wildlife host species, we adapted an existing monoclonal antibody-based sandwich enzyme-linked immunosorbent assay (ELISA) for the detection of Taenia antigen in dried urine samples [70–73]. The monoclonal antibodies (B158C11 and B60H8) used in this assay are specific to the Taenia genus, which permits its use in the detection of larval infections of all taeniid species. Indeed, this assay has been used as an epidemiological tool, often complementary to other diagnostic methods, in studies of porcine, bovine, and human cysticercosis [70, 71, 74–78]. Because this assay detects circulating metacestode (larval) antigens, it identifies active infections rather than past exposure identified by antibody assays [75, 77].
Despite the success of this assay in studies of cysticercosis in livestock, the difficulty of obtaining blood or serum samples from humans limited its use in human populations [77, 78]. Thus, two teams [77, 78] adapted the monoclonal antigen test to non-invasively diagnose these diseases in urine. However, the existing protocols for Taenia antigen detection in urine are still impractical for implementation in wildlife studies because they require that urine samples be stored at -20°C until processing [77, 78]. Because many wildlife studies are carried out in areas where electricity is absent or inconsistent, the need for refrigeration limits the practicality of these tests in remote areas. We therefore validated the use of dried urine with a modified protocol to investigate sylvatic cycles of Taenia transmission.
Geladas—herbivorous primates endemic to the Ethiopian highlands–are known to exhibit protuberant cysts characteristic of infection with the larval stage of T. serialis (Fig 1). Coenuri have been recorded in wild-caught captive geladas for nearly a century and were often ascribed to T. serialis based primarily on morphological cues [39–41, 79–83]. Recently, this identification was confirmed with molecular diagnosis of cystic material obtained from protuberant cysts [42, 43]. Prevalence of T. serialis-associated cysts in geladas ranges from 4–13% in an ecologically disturbed area [42, 44, 84] to 30% in an ecologically intact area [43], and cysts in both areas are associated with significant increases in mortality and decreases in reproductive success [43, 44]. However, not all infections necessarily manifest as conspicuous cysts, a point illustrated by the presence of non-protruding cysts revealed during necropsies on infected captive geladas. Thus, prevalence of T. serialis in geladas based on protuberant cysts is likely to be underestimated.
We implemented the monoclonal antibody-based sandwich ELISA in a wild population of geladas in the Simien Mountains National Park (SMNP), Ethiopia, where individuals are parasitized with T. serialis [42]. Recent work in this population demonstrated sex- and age- biased distribution of T. serialis cysts, with higher prevalence in adults and females [44]. This sex bias may reflect either patterns of data collection that bias towards observing infected females and uninfected males, or the estrogen affinity exhibited by the larvae of many taeniid species [85, 86]. The increased prevalence of T. serialis cysts in adults compared to immatures may arise either from increased susceptibility of adults due to the immunosuppressive effects of hormones related to sexual maturity, or as a function of the time required for infection to develop into observable cysts. The adaptation of the urine antigen ELISA to non-invasively diagnose T. serialis in dried gelada urine allowed us to investigate infection dynamics that cannot be detected solely by analyzing the presence of observable cysts.
We conducted our study in the Sankaber area of the SMNP, Amhara Region, Ethiopia. The SMNP was established in 1969 and has been classified as a UNESCO World Heritage Site in Danger since 1996 due to substantial anthropogenic impact [87]. The park covers 13,600 hectares, is characterized by Afro-montane and Afro-alpine habitats, and contains a number of mammals of potential importance to the T. serialis life cycle. These include the black-backed jackal (Canis mesomelas), the golden jackal (Canis aureus), the spotted hyena (Crocuta crocuta), the Ethiopian wolf (Canis simiensis), Starck’s hare (Lepus starcki), and the gelada [88]. The substantial human population in the SMNP has contributed to the loss of natural vegetation and the expansion of crops and grazing seen in many areas of the park [88, pers. obs.]. Dogs, jackals, hyenas, and Ethiopian wolves are among the carnivores living in the SMNP that potentially prey on or scavenge the corpses of geladas [88], and are thus of potential importance for the T. serialis life cycle as definitive hosts.
From August 2014 to June 2015, we collected a total of 527 urine samples from 204 geladas (117 females, 87 males; 37 infants, 60 subadults, 107 adults) in 2 habituated groups under long-term study by the Simien Mountains Gelada Research Project (SMGRP) in the SMNP. Geladas in the habituated groups are each assigned a three-letter code and are individually identifiable by the field team based on suites of morphological characteristics and corporeal idiosyncrasies [88]. Thus, all samples collected in this population were from known individuals, with most individuals sampled more than once over time (n = 97 individuals; median: 2 samples/individual, range: 1–10). Sampling included 58 samples from 10 individuals exhibiting the cysts characteristic of T. serialis infection to serve as ‘true positives’, and 57 samples from 37 unweaned infants to serve as ‘true negatives’ (unweaned infants are unlikely to ingest eggs because they do not yet eat grass; see below for further explanation). All other samples (412 from 158 individuals) were collected for evaluation in the Ag-ELISA as samples of ‘unknown status’ (median = 2, range: 1–10). These included 94 females and 64 males; 60 subadults and 98 adults.
Urine samples were collected from the ground immediately after urination using Whatman Qualitative Filter Papers (Grade 4, 11.0 cm). After urination, as much urine as possible was soaked up from the ground with a filter paper. The filter paper was folded and stored in a 2-oz Whirl-Pak bag, which was labeled with the unique code associated with the individual, date, and time. Approximately 1 g of indicating silica desiccant was added to each bag to ensure samples remained dry and to prevent mold growth.
Samples were processed and analyzed using the B158/B60 ELISA (Institute for Tropical Medicine, Antwerp) in the Immunochemistry Laboratory of the Division of Parasitic Diseases and Malaria at the Centers for Disease Control and Prevention (CDC) in Atlanta, Georgia. To aid in identifying urine stains on the filter papers, we viewed each paper under a UV light (long-wave, 365 nm; Spectroline Model ENF-240c), and used an office hole puncher to remove four circles (~6 mm diameter) from the part of each filter paper that was soaked on both sides. The hole puncher was sterilized and dried after each use to prevent cross-contamination. The four circles taken from each sample were placed into a single labeled 2 mL sample tube. Each sample was reconstituted with 1 mL blocking buffer (PBS-Tween 20 + 1% newborn calf serum (NBCS), existing CDC collection) and vortexed.
Following [73], polystyrene ELISA plates (Nunc Maxisorp flat-bottom 96 well) were coated and incubated with the capture antibody (B158C11A1 monoclonal antibody in a sensitization buffer (carbonate bicarbonate buffer, pH 9.5)). Each plate included 80 unknown samples, 4 known negative human samples, and 2 positive control samples created by spiking known negative human samples (existing CDC collection) with 0.125 μg antigen/1 mL urine T. crassiceps antigen (soluble protein extract). A standard curve (2-fold serial dilutions of known negative human urine samples spiked with T. crassiceps antigen) was included on each plate as an additional control. After a washing step (1x), plates were coated and incubated with 150 μL/well of blocking buffer, and then loaded and incubated with 100 μL from each sample. After a washing step (4x), plates were coated and incubated with 100 μL of detecting antibody dilution (B60H8A4 + blocking buffer). Plates were washed (1x) and subsequently loaded and incubated with 100 μL of Streptavidin-horseradish peroxidase (HRP) dilution (Peroxidase-conjugated Streptavidin 1:10,000 dilution, Jackson ImmunoResearch Laboratories, West Grove, PA, in blocking buffer (0.1ug/ml)). Plates were washed (1x) and then loaded with 100 μL of Tetramethylbenzidine (TMB) (1-step Ultra TMB-ELISA, ThermoFisher Scientific, USA) and shaken at room temperature for two minutes. After the addition of 100 μL of stop solution (1M sulfuric acid; H2SO4, EMD Millipore, Darmstadt, Germany) to each well, the optical densities (OD) of samples were read in the VersaMax ELISA Microplate Reader (Molecular Devices, Sunnyvale, CA, USA) at 450 nm (see S1 Text for detailed protocol). If more than one control on a plate failed, the entire plate was repeated. The index value (IV) for each sample relative to the positive and negative controls on each plate was calculated using the following formula:
IV=SampleOD−Average(NegativeControlsOD)/Average(PositiveControlsOD−Average(NegativeControlsOD)
We assessed the sensitivity and specificity of the Ag-ELISA with a receiver operating characteristic (ROC) curve [89]. The nature of working in a wild system precludes establishing a negative ‘gold standard’ because we are unable to confirm negative diagnoses with serological or imaging techniques. Thus, we used unweaned infants as ‘true negatives’ (n = 58 samples), because they do not yet consume grass and are thus minimally exposed to T. serialis eggs and can be considered likely to be negative. We used individuals presenting with T. serialis cysts as ‘true positives’ (n = 58 samples). We selected the point on the ROC curve at the shortest distance from the coordinate (0, 1) as the optimal threshold IV for classifying a sample as positive or negative. ROC analysis was performed with the package “pROC” [90] in R [91].
To investigate if sex and age predicted the occurrence of cysts among adults and subadults (n = 158 individuals), we used logistic regression implemented in the ‘glm’ function in the R package ‘stats’ [91]. We coded age as a continuous variable based on known or estimated birthdates for individuals. Model selection was performed with Akaike information criterion (AICc), which selects the optimal model based on maximum likelihood [92] with a finite sample size [93].
To investigate if sex and age predicted the occurrence of antigen-positive samples (i.e., those with an IV greater than the IV threshold from the ROC analysis) among adults and subadults without cysts (n = 412 samples, 158 individuals), we used a generalized linear mixed effects model (GLMM) implemented with the ‘glmer’ function in the ‘lme4’ package in R [94]. We used binomial errors with a logit link function, and included age and sex as fixed effects. Because individuals were sampled at varying intensities and may have had different individual risks of infection, we included individual identity as a random effect. We coded age in the following two ways: (1) as a continuous variable based on known and estimated birthdates; and (2) as an ordered categorical variable with two levels based on developmental stage (i.e., subadult or adult). Continuous age is expected to be a relevant predictor of infection if accumulated exposure to T. serialis eggs in the environment drives risk, whereas categorical age based on developmental stages may be more relevant if hormonal factors are a major driver of risk. We compared the fit of the continuous age and categorical age models using AICc and calculated averaged coefficients for each variable using model averaging.
All research was approved by the University Committee on the Use and Care of Animals at the University of Michigan (UCUCA protocol #09554), the Duke University Institutional Animal Care and Use Committee (IACUC protocol #A218-13-08), and followed all laws and guidelines in Ethiopia. This research adhered to the standards presented in the Guide for the Care and Use of Laboratory Animals (National Research Council of the National Academies, 8th Edition) and the Animal Care Policy Manual (United States Department of Agriculture, 2016).
Our measurement of infection status using the described Ag-ELISA was highly accurate. The ROC analysis revealed the optimal threshold IV to be 42.1, with 98.4% specificity (95% CI: 95.1–1), 98.5% sensitivity (95% CI: 95.6–1) and an area under the curve (AUC) of 0.99 (95% CI: 0.9937–1; Fig 2). We identified only one likely false positive (i.e., an infant with a positive sample) (98.2%, 56/57), and one false negative (i.e., an individual with a cyst and a negative sample) (98.3%, 57/58, Table 1). Because sample quality was difficult to evaluate with our collection technique, we binned samples into either ‘positive’ or ‘negative’ categories based on the IV cutoff instead of conducting analysis at the level of sample OD. This conservative approach permits for the broad designation of samples as positive or negative for antigen presence, but precludes analyses that address fluctuations or activity in sample OD.
Twenty-six of 158 individuals without visible cysts (16.4%) tested positive at least once. This included 14 females and 12 males, of which 6 were subadults and 20 were adults. All but one sample from an individual with a visible cyst fell above the optimal cutoff (Fig 3), indicating that samples from individuals with cysts had generally higher logged index values (IVs) than individuals without cysts. Importantly, 2 individuals without cysts that tested antigen-positive developed observable cysts within 7 months of sampling. One of these individuals had one negative and one positive sample in the 3 months prior to exhibiting an observable cyst, after which all of his samples were positive. The other individual had one positive sample 7 months before exhibiting an observable cyst, after which all of her samples were positive.
To search for evidence of established T. serialis infection in individuals without visible cysts, we focused on individuals that were sampled at least 5 times during the study period (21 adults, 2 subadults). We found that some individuals without cysts were consistently positive for T. serialis antigen, others were consistently negative, and still others switched between antigen-positivity and antigen-negativity throughout the study period. Twelve individuals showed no antigen-positive samples, 2 showed a clear majority of positive samples (one with 8/9 positive samples, one with 9/10 positive samples), and 7 individuals had a single positive sample within a sequence of negative samples. The remaining 2 individuals showed an interesting mixture of positive and negative samples: one individual tested positive in 3 consecutive months, and then negative 7 months later. The other displayed a sequence of negative and positive samples within 6 months.
We investigated the predictors of visible cysts, focusing on age, sex, and the interaction between these two variables. AICc model selection revealed the models with the most support to include age (in years), sex, and an interaction between age and sex as predictor variables, with the model including only age garnering the most support (Table 2). The importance of age and lack of effect of sex were reinforced with the results of full model averaging, which showed that increasing age was the strongest predictor of cysts across all models (Table 3).
We then investigated the predictors of antigen-positivity in urine samples, again including age, sex, and the interaction between these two variables as predictors. One analysis included age coded categorically, whereas the other included age coded continuously, and both included individual ID as a random intercept to account for repeated sampling from individuals. In the first analysis (categorical age), AICc model selection showed that the model with the most support included only the random intercept (individual ID) and no fixed effects (i.e., age, sex, and the interaction did not appear as predictors of Taenia antigen-positivity in samples, Table 4). A model including age and the random intercept was less supported than the model containing only the random intercept (Table 4). Full model averaging revealed age to be a weaker predictor of antigen-positivity than the random intercept (Table 3). Results were similar for the analysis that used age coded as a continuous variable. The model with the most support included only the random intercept and no fixed effects (Table 4), which was also reflected in the model averaging estimates (Table 3).
We adapted and evaluated a monoclonal antibody-based sandwich ELISA protocol for the detection of Taenia antigen in dried gelada urine, finding that our adaptation was able to detect Taenia antigen with high accuracy in geladas infected with T. serialis. We implemented this assay in a wild gelada population in order to understand basic patterns of T. serialis infection, providing the first evidence for widespread T. serialis infection in individuals that do not exhibit external cysts. Our results indicate that T. serialis infection is more widespread than are visible cysts, with 18% of the sampled population testing positive for Taenia antigen where only 4.8% exhibited visible cysts. However, our results demonstrate the occurrence of short-term antigen presence in individuals sampled multiple times, suggesting that individuals may eliminate initial infection with T. serialis and that a single positive sample may not necessarily indicate an established infection (as do cysts).
Positive antigen samples are highly likely to reflect active larval growth (i.e., true infections) and not merely the presence of eggs passing through the gastrointestinal tract, because this assay identifies active infection by detecting glycoproteins produced by taeniid metacestodes and not oncospheres (this also precludes the possibility that positive antigen samples reflect atypical growth of the adult stage of the tapeworm in geladas) [75, 95]. We postulate that individuals without cysts that presented with high log(IV) samples should be considered positive for Taenia antigen and are likely to harbor active infections that are not visible as cysts to observers, whether because (1) the infection is young and has not yet had time to develop into a visible cyst; or (2) the infection is advanced but is located deep in the abdominal cavity or somatic tissues and will never become visible. It is highly unlikely that the samples positive for antigen presence are all false positives: based on the false positive rate of 1.79% calculated using the “known negative” infant set (in which 1 out of 57 samples from unweaned infants tested positive), the expected number of false positives is 8.4, and the probability of observing 50 or more false positives in 412 samples is less than p = 10−25. These two possibilities–that positive assay results indicate young infections or fully developed internal cysts–are not mutually exclusive.
In support of the interpretation of a positive antigen result as (1) reflecting the presence of young cysts that are not yet observable externally, 2 individuals that tested positive with no external cysts at the time of sample collection developed cysts within a year of sampling. In support of the interpretation of a positive antigen result as (2) reflecting the presence of advanced infections in deep tissue that will never become visible to observers, early necropsies of wild-caught captive geladas revealed fully developed, non-protruding cysts in the abdominal cavities, deep musculature, and viscera [39–41, 79–83]. Thus, positive assay results in the absence of observable cysts may reflect either young infections or advanced infections in undetectable locations.
Interestingly, we observed switches in infection status (antigen-positive or antigen-negative) within individuals without cysts (i.e., positive to negative and vice versa). Among 23 well-sampled individuals without cysts (i.e., 5 or more samples), only 2 had a clear majority of antigen positive samples, whereas 12 had no positive samples, 7 had just 1 positive sample, and the remaining 2 flipped from positive to negative during the study period. The observed switches in infection status may reflect either (1) the inability of some larvae to persist; or (2) the ability of hosts to control or eliminate their infections through calcification (although caveats in data certainty must also be considered, such as incorrect individual identification during sample collection). Importantly, the values of samples from these individuals were strikingly different enough (i.e., not close to the cutoff on either side) to make it unlikely that variation in sample quality was behind this pattern. A similar phenomenon was described in humans with T. solium cysticercosis, with 3.5% of 867 participants exhibiting a single positive sample in between 2 negative samples [96]. The authors postulated that this short-term antigen presence could owe to incomplete parasite formation or to effective host defenses that enable clearance of the parasite.
In geladas, short-term antigen presence may indicate low T. serialis egg viability or highly effective host immune responses that result in stunted infections or incomplete parasite establishment. Indeed, experimental infection of swine with T. solium eggs demonstrated low rates of infection establishment even with high infectious doses [97]. Attempts by the host immune system to control infection may not always be successful; for example, one individual tested positive once and negative once in the 3 months before developing an external cyst, after which he consistently tested positive. This may indicate a process in which the host attempted to mount an immune response and was fleetingly able to control the infection before succumbing. Early stages of infection may also release antigens less reliably, which would make early infection difficult to detect. Future work that combines frequent longitudinal urine sampling from known individuals while monitoring for external signs of disease is needed to better understand the frequency and health consequences of transient T. serialis infections.
The higher occurrence of cysts among older individuals is consistent with previous studies of T. serialis cyst prevalence in geladas [43, 44], whereas the lack of support for a strong relationship between age and antigen-positive samples was unexpected. Together, these results suggest that susceptibility to infection does not vary strongly with age, and that cysts may take years to develop to a stage at which they protrude and are visible to observers. Contrary to our predictions based on the increased female susceptibility observed in other larval taeniid systems [85, 86] or the female-bias in data collection, we found no evidence for a sex bias in either T. serialis cysts or antigen-positivity in samples. The lack of support for increased susceptibility with age or sex suggests that susceptibility to T. serialis in geladas may not be hormonally modulated.
Further research is needed to elaborate the physiological and ecological drivers of susceptibility and exposure in this system. Ongoing research is exploring the relationship between co-occurrence of gastrointestinal parasites and T. serialis infection in geladas, and research is planned to investigate the associations between measurements of stress (fecal glucocorticoid concentrations) and susceptibility to T. serialis infection and the development of cysts. Future studies should additionally consider other potential drivers of susceptibility and exposure to T. serialis, such as seasonal changes in T. serialis egg distribution and gelada ranging patterns and differences in social behavior that affect risk. Articulating the risk factors associated with infection in geladas may inform the understanding of the danger T. serialis poses to other primates, including humans, as well as the control of infections. If exposure is the central driver of infection, then humans and nonhuman animals that overlap significantly with T. serialis definitive hosts may be at the highest risk for infection and can thus be targeted for control efforts. While research has shown that T. serialis cysts substantially increase gelada mortality [45], there is no indication that this infection threatens population-level persistence. Continuous monitoring of T. serialis and mortality in this population will determine whether future interventions are necessary.
The use of dried urine for larval Taenia infection diagnosis provides the substantial benefits of not requiring refrigeration or invasive procedures; thus, it is well suited to the identification of Taenia infections in wildlife inhabiting remote areas. However, this approach has one notable drawback: this assay is genus-specific, not species-specific, and will pick up antigens from any Taenia species. Thus, other methods must be used for species-level identification. If it is possible to obtain tissue from the cyst of an infected individual (from a dead individual, as in [42], or from leaked cystic material, as in [43]), genetic methods can be used to identify the parasite to the species-level. Non-lethal traps may be employed in studies of smaller species (e.g., lining the trap floor with filter paper for urine collection prior to release), and fecal analysis of carnivore hosts sympatric with the target intermediate host species may also be employed to identify the taeniid species active in a given system.
In future applications of this method, the potential for cross-reactions should be considered. Infection with parasites in the Trypanosoma genus may give rise to a cross-reaction on this assay [98], and thus must be taken into account in the interpretation of assay results in Trypanosoma-endemic areas. Because geladas inhabit cool, high-altitude habitats that are free of the tsetse flies that carry Trypanosoma parasites [99, pers.obs.], and because infection with mechanically transmitted Trypanosoma spp. is unlikely in African primates, this cross-reaction was not considered in the interpretation of our results.
In conclusion, the global distribution and flexibility in intermediate host selection of many taeniid species make them critically important to monitor for global human and animal health. The adaptation of a serum protocol for the detection of Taenia infections for use with dried urine samples is a useful and pioneering step towards a complete understanding of the dynamics of Taenia infection in wildlife. While this assay cannot be used as a stand-alone diagnostic technique, particularly given its genus-wide specificity, it holds great value for studies of infection dynamics in host populations where regular invasive monitoring is impractical and in areas where sample storage prohibits the collection of wet urine samples.
|
10.1371/journal.ppat.1002431 | Role of Permissive Neuraminidase Mutations in Influenza A/Brisbane/59/2007-like (H1N1) Viruses | Neuraminidase (NA) mutations conferring resistance to NA inhibitors were believed to compromise influenza virus fitness. Unexpectedly, an oseltamivir-resistant A/Brisbane/59/2007 (Bris07)-like H1N1 H275Y NA variant emerged in 2007 and completely replaced the wild-type (WT) strain in 2008–2009. The NA of such variant contained additional NA changes (R222Q, V234M and D344N) that potentially counteracted the detrimental effect of the H275Y mutation on viral fitness. Here, we rescued a recombinant Bris07-like WT virus and 4 NA mutants/revertants (H275Y, H275Y/Q222R, H275Y/M234V and H275Y/N344D) and characterized them in vitro and in ferrets. A fluorometric-based NA assay was used to determine Vmax and Km values. Replicative capacities were evaluated by yield assays in ST6Gal1-MDCK cells. Recombinant NA proteins were expressed in 293T cells and surface NA activity was determined. Infectivity and contact transmission experiments were evaluated for the WT, H275Y and H275Y/Q222R recombinants in ferrets. The H275Y mutation did not significantly alter Km and Vmax values compared to WT. The H275Y/N344D mutant had a reduced affinity (Km of 50 vs 12 µM) whereas the H275Y/M234V mutant had a reduced activity (22 vs 28 U/sec). In contrast, the H275Y/Q222R mutant showed a significant decrease of both affinity (40 µM) and activity (7 U/sec). The WT, H275Y, H275Y/M234V and H275Y/N344D recombinants had comparable replicative capacities contrasting with H275Y/Q222R mutant whose viral titers were significantly reduced. All studied mutations reduced the cell surface NA activity compared to WT with the maximum reduction being obtained for the H275Y/Q222R mutant. Comparable infectivity and transmissibility were seen between the WT and the H275Y mutant in ferrets whereas the H275Y/Q222R mutant was associated with significantly lower lung viral titers. In conclusion, the Q222R reversion mutation compromised Bris07-like H1N1 virus in vitro and in vivo. Thus, the R222Q NA mutation present in the WT virus may have facilitated the emergence of NAI-resistant Bris07 variants.
| The H275Y neuraminidase (NA) mutation conferring resistance to oseltamivir was shown to impair old influenza H1N1 strains both in vitro and in vivo. By contrast, an oseltamivir-resistant A/Brisbane/59/2007 (Bris07)-like H1N1 H275Y NA variant emerged in 2007 and completely replaced the wild-type (WT) strain in 2008–2009. This discrepancy could be attributed to permissive NA mutations (R222Q, V234M and D344N) that were identified in most Bris07-like oseltamivir-resistant variants. To verify this hypothesis, we developed a reverse genetics system for a sensitive Bris07-like isolate (275H) whose NA protein contains the 3 permissive mutations (222Q, 234M, 344N). Using mutagenesis, we first introduced the H275Y then reverted codons at positions 222, 234 and 344. The resulting 5 recombinants (WT, H275Y, H275Y/Q222R, H275Y/M234V and H275Y/N344D) were compared with regard to NA enzyme properties, replicative capacities in vitro as well as infectivity and contact-transmissibility in ferrets. Among the studied permissive mutations, Q222R was associated with a significant reduction of both affinity and activity of the NA enzyme resulting in a virus with a reduced replicative capacity in vitro and decreased replication in lungs of ferrets. Thus, the R222Q mutation may have been the major permissive NA change that facilitated the emergence and spread of NAI-resistant Bris07 variants.
| Influenza viruses are respiratory pathogens associated with significant public health consequences. Each year, influenza epidemics can be responsible for significant morbidity in the general population and excess mortality in elderly patients and individuals with chronic underlying conditions. Influenza A viruses of the H1N1 subtype have been associated with seasonal influenza epidemics for many decades and, in presence of immunological pressure, such viruses continue to evolve through genetic variability which is mainly confined to virus segments encoding surface glycoproteins i.e., the hemagglutinin (HA) and neuraminidase (NA) [1]. Consequently, viral strains to be used in annual influenza vaccines should be regularly updated to ensure optimal protection. Besides vaccines, neuraminidase inhibitors (NAI) including inhaled zanamivir, oral oseltamivir and intravenous peramivir provide an important additional measure for the control of influenza infections [2]. These antivirals target the active center of the influenza NA molecule, which is constituted by 8 functional (R-118, D-151, R-152, R-224, E-276, R-292, R-371, and Y-406; N2 numbering) and 11 framework (E-119, R-156, W-178, S-179, D-198, I-222, E-227, H-274, E-277, N-294, and E-425; N2 numbering) residues that are largely conserved among influenza A and B viruses [3]. However, the emergence of NAI-resistant viruses, as a result of drug use or due to circulation of natural variants, may compromise the clinical utility of this class of anti-influenza agents.
The H275Y (H274Y in N2 numbering) NA mutation conferring resistance to oseltamivir and peramivir has been detected with increasing frequency in seasonal A/H1N1 viruses since 2007 to the extent that almost all characterized A/Brisbane/59/2007-like (Bris07) (H1N1) influenza strains that circulated worldwide during the 2008–09 season were H275Y variants [4], [5]. Interestingly, this drug-resistant strain seemed to have emerged independently of NAI use [6], [7]. The rapid dissemination of the H275Y Bris07 variants in the absence of antiviral pressure suggests that the H275Y NA mutation may not compromise viral fitness and transmissibility in this recent H1N1 viral background. This contrasts with previous studies that analyzed the role of the H275Y mutation using older (A/Texas/36/91 [8] and A/New Caledonia/99/01 [9]) drug-selected H1N1 variants. Recent reports by our group and others have confirmed the differential impact of the H275Y mutation on viral fitness and enzymatic properties in the context of old and recent influenza H1N1 isolates [10], [11]. In an attempt to provide a molecular explanation for this observation, previous authors suggested that secondary NA mutations such as D344N that emerged in H1N1 variants isolated after the 2006–07 season were associated with higher NA activity and affinity and could have facilitated the emergence of the H275Y mutation [11], [12]. Such drug-resistant mutants may have a better HA-NA balance than the susceptible viruses and indeed completely replaced them in a short period of time. In addition, Bloom and colleagues recently described two other secondary NA mutations at codons 222 and 234 that may have counteracted the compromising impact of the H275Y mutation [13]. In that study, the V234M and R222Q mutations were shown to restore the viral fitness of an A/New Caledonia/20/99 H1N1 variant containing the H275Y mutation [13].
To further investigate which secondary NA mutations may have facilitated the introduction of the H275Y mutation in contemporarily seasonal H1N1 viruses and allowed their dissemination, we developed a reverse genetics system using a clinical Bris07 (H1N1) isolate as genetic background and evaluated the impact of the H275Y oseltamivir resistance mutation as well as several potential compensatory NA mutations on enzyme activity, viral fitness and transmissibility.
In the present study, five recombinant Bris07 influenza viruses were generated i.e., the WT virus (containing the putative permissive mutations) that briefly circulated during the 2007–08 season, the single H275Y oseltamivir-resistant variant and three double mutants containing the H275Y mutation as well as reversion of potential permissive mutations (H275Y/Q222R, H275Y/M234V and H275Y/N344D). NA enzymatic properties using equivalent titers of recombinants were first analyzed with determination of relative NA enzymatic activity (Vmax values), which reflects the total NA activity per virion, and Km values, which reflect the affinity for the substrate. As shown in Table 1, the single H275Y mutation had no significant impact on NA affinity and activity compared to the WT virus in the context of the Bris07 background. By contrast, the double H275Y/Q222R mutation was associated with a significant reduction of both NA affinity (Km of 40.31 vs 11.95 µM, P<0.001) and relative NA activity (7.01 vs 28.19 U/sec, P<0.001) compared to the WT (Table 1 and Fig. 1). The H275Y/M234V mutant had a Km value comparable to that of the WT, whereas its relative NA activity was significantly reduced (Vmax of 21.89 vs 28.19 U/sec, P<0.05). The H275Y/N344D mutant showed a significantly reduced affinity (Km of 50.77 vs 11.95 µM, P<0.001) with no change in NA activity compared to the WT. When comparing the double mutants to the single H275Y mutant, the Km values were significantly increased for the H275Y/Q222R and H275Y/N344D mutants (P<0.001) whereas only the double H275Y/Q222R mutant had a significantly lower relative NA activity (P<0.001).
Using recombinant NA proteins expressed in 293T cells, we further investigated the impact of NA mutations on the amount of NA activity at the cell surface. As shown in Fig. 2, all studied mutations were associated with a significant reduction of total surface NA activity compared to the WT with relative total surface activities of 66% (P<0.01), 9.72% (P<0.001), 32.07% (P<0.001) and 54.89% (P<0.01) for the H275Y, H275Y/Q222R, H275Y/M234V and H275Y/N344D mutant proteins, respectively. When compared to the single H275Y mutant, H275Y/Q222R (P<0.001), H275Y/M234V (P<0.001) and H275Y/N344D (P<0.05) double mutants also had significantly reduced surface NA activities. The differences observed in total surface NA activity between the different recombinant NA proteins may be due to a decreased number of NA molecules that reached the cell surface or to less activity per enzyme.
We next determined the phenotype of resistance to NAIs for the 5 recombinant viruses. As expected, the presence of the H275Y mutation was associated with resistance to oseltamivir (mean fold increase of 2627 in IC50 values) and peramivir (mean fold increase of 998) with no impact on zanamivir susceptibility (Table 2). Interestingly, comparison of the levels of resistance for the double recombinant mutants versus the single H275Y mutant revealed a significant reduction in the level of resistance to peramivir for the double H275Y/Q222R mutant (IC50 of 35.25 nM vs 59.85 nM, P<0.01). A similar trend was observed for oseltamivir (IC50 of 651.86 nM vs 1024.54 nM) although, in this case, the difference between IC50 values was not statistically significant.
Viral fitness of recombinant A/Brisbane/59/2007-like viruses was assessed in vitro using ST6Gal1-MDCK cells. The double H275Y/Q222R mutant produced viral plaques with a significantly reduced area compared to the recombinant WT (0.13 mm2 vs 0.53 mm2, P<0.001) whereas the remaining recombinants generated plaques of comparable sizes (Table 1). Of note, the reduction in plaque size for the H275Y/Q222R mutant was also significant compared to that of the single H275Y mutant (P<0.001). In replication kinetics experiments, the peak viral titers for all recombinants were obtained at 36 h post-infection (PI) with viral titers ranging from 5.6×106 PFU/ml (H275Y/Q222R) to 5.3×107 PFU/ml (WT) (Fig. 3). The WT, the single (H275Y) and the double (H275Y/N344D) mutants had comparable viral titers at all time points. By contrast, and in accordance with plaque size data, the double H275Y/Q222R mutant was associated with a significant reduction in viral titers at 36 h (P<0.001) and 48 h (P<0.05) PI compared to the WT (Fig. 3). There was also a significant reduction in the viral titer obtained at 36 h PI for the double H275Y/M234V mutant compared to the WT (P<0.001). When compared to the single (H275Y) mutant, viral titers of the double H275Y/Q222R and H275Y/M234V mutants were significantly lower at 36 h (P<0.001).
Intranasal inoculation of ferrets with the WT and two mutant (H275Y and H275Y/Q222R) Bris07 recombinant viruses resulted in a febrile response that peaked on day 2 PI (Fig. 4A). The area under the curve (AUC) of temperatures between days 0 and 6 PI was similar for the 3 groups of ferrets i.e. 6.81±1.19 for the WT virus, 5.99±1.9 for the H275Y/Q222R mutant and 7.26±0.55 for the H275Y mutant. There was no significant difference in body weight between the three groups of animals at any time points (data not shown). As shown in Fig. 5A, mean viral titers in nasal wash samples collected on day 2 PI from ferrets infected with the recombinant WT and the single H275Y mutant were comparable (4×105±2.9×104 PFU/ml for the WT and 2.6×105±8.7×104 PFU/ml for the H275Y mutant) whereas the H275Y/Q222R mutant had a reduced mean viral titer (4.6×104±4.2×103 PFU/ml; P<0.05 vs WT). Similarly, mean viral titers in nasal wash samples of ferrets infected with the H275Y/Q222R were significantly lower than those of the H275Y mutant (P<0.05) and WT virus (P<0.01) on day 4 PI (3.4×103±1.7×103, 1.1×104±6.7×103 and 1.5×104±9.6×102 PFU/ml, respectively). On the other hand, the three recombinants were associated with comparable mean viral titers on day 6 PI (2×102±4.6×101 PFU/ml for the WT, 1.1×102±5.8×101 PFU/ml for the H275Y/Q222R and 1.3×102±8.1×10PFU/ml for the H275Y).
All contact ferrets seroconverted for A/Brisbane/59/2007 when tested 14 days after contact, with geometrical mean hemagglutination inhibition (HAI) titers of 160±33, 145±119 and 95±55 for the WT, H275Y and H275Y/Q222R recombinant viruses, respectively. A febrile response could be observed on days 4 and 5 in the WT and the H275Y groups, respectively, but not in the H275Y/Q222R group (Fig. 4B). The AUC of temperatures between days 2 and 6 PI was similar between groups of ferrets infected with the recombinant WT (5.29±0.34) and its H275Y variant (4.54±0.19) whereas the AUC of the H275Y/Q222R group was significantly lower than that of the WT group (4.09±0.96; P<0.05). Viral titers in nasal wash samples collected on days 2, 4 and 6 PI are shown in Fig. 5B. Only the WT virus was detected on day 2 PI. Mean viral titers were comparable for the H275Y mutant and the WT virus on days 4 and 6 PI. In contrast, the H275Y/Q222R mutant was associated with significantly lower mean viral titers compared to WT on both day 4 (2.7×102±1.2×102 vs 1.2×104±3.5×103 PFU/ml, P<0.01) and day 6 PI (3.8×103±2.1×103 vs 1.2×104±2.5×103 PFU/ml, P<0.01).
In this study, we used recombinant viruses derived from a clinical WT Bris07 strain to demonstrate using both in vitro and ferret experiments that the R222Q NA mutation was the main but possibly not the only permissive mutation that allowed the widespread dissemination of the oseltamivir-resistant H275Y mutant during the 2007–09 influenza seasons. Although such mutant seems to have disappeared since the emergence of the pandemic H1N1 virus in April 2009, understanding the mechanisms leading to the transmission of this unique virus is of great importance and could have an impact on the future use of NAIs.
The influenza NA protein plays a major role during the viral replication cycle. Its sialidase activity promotes virion release by removing sialic residues from viral glycoproteins and infected cells [14]. The NA enzyme also mediates virus penetration in the mucin layer of the respiratory tract, facilitating virus spread [15]. Importantly, the catalytic site of the NA enzyme has been shown to be conserved in all influenza A subtypes and influenza B viruses [3]. Therefore, the influenza NA protein has been considered as a suitable target for designing anti-influenza agents for both prophylactic and therapeutic purposes. Besides its functional role, the NA protein is a major structural surface glycoprotein that is exposed to the host immune pressure [14]. The NA gene, like the HA one, is therefore subject to more genetic variations than the rest of the influenza genome. Consequently, some amino acid (a.a.) changes, part of antigenic sites of the NA protein, may significantly contribute to the emergence of drifted variants, whereas certain substitutions located in or near the catalytic site may also affect the NA enzyme properties. For instance, Hensley and colleagues have recently identified NA mutations conferring resistance to zanamivir in variants of an influenza A/Puerto Rico/8/1934 H1N1 virus that was subjected to anti-HA monoclonal antibodies pressure [16].
In this study, we focused on a.a. changes that occurred in the NA protein during the evolution of recent seasonal influenza H1N1 viruses and that may have been involved in the development and dissemination of resistance to NAIs. These changes included the well-known framework H275Y mutation, responsible for the resistance phenotype to oseltamivir and peramivir, as well as other substitutions (V234M, R222Q and D344N) that may have contributed to the emergence and dissemination of resistance by acting as permissive/compensatory mutations.
Phylogenetic analyses previously demonstrated that the V234M mutation was already present in oseltamivir-susceptible A/Solomon Islands/3/2006 (SI06) viruses [13]. In another report, NA enzyme properties of SI06 viruses were found to be similar to those of older oseltamivir-susceptible strains such as A/New Caledonia/99/2001 in terms of relative NA activity (Vmax) and affinity (Km) [11]. By contrast, the appearance of the R222Q and D344N mutations in H1N1 viruses isolated after 2007 was associated with a significant increase in NA affinity (decreased Km values) in both 275H and 275Y strains [11]. In accordance with these observations, we demonstrated a sharp impact for the Q222R and N344D reversion mutations on Km values using our Bris07 recombinants (Table 1). Besides its effect on NA affinity, the Q222R reversion mutation was also associated with a significant decrease in relative NA activity (Table 1 and Fig. 1) and total NA activity that was expressed on the cell surface (Fig. 2), in line with previously-reported results in another viral background [13]. As a result, the H275Y/Q222R mutant virus was significantly compromised in vitro based on plaque size and replication kinetics patterns. Such decreased viral replication of the H275Y/Q222R mutant was also evident in vivo, resulting in lower viral titers in nasal wash samples and an absence of febrile response in contact ferrets. However, the H275Y/Q222R mutant was transmitted to all naïve ferrets by direct contact meaning that the combination of several permissive NA mutations and/or mutations elsewhere in the viral genome may be necessary to recapitulate the epidemiological observations showing increased transmission of the oseltamivir-resistant Bris07 virus. Also, it should be noted that naïve (non-immune) ferrets may not completely capture the fitness of Bris07 in humans with pre-existing immunity. Alternatively, the Q222R mutation could affect airborne transmission which has not been evaluated in our study. Of note, possibly due to the lower affinity of Q222R for MUNANA, less NAIs were required for competitive inhibition of the H275Y/Q222R mutant compared to the H275Y mutant. Residue 222 is located in the vicinity of the catalytic site of the N1 enzyme based on 3-D structure analysis [17]. Thus, substitution of a charged (R) by an uncharged (Q) a.a. at codon 222 may be the main change that dramatically altered the NA enzyme properties of recent seasonal H1N1 viruses. Of interest, only one NA substitution (R194G) was sufficient to restore the viral fitness of an influenza A/WSN/33 (H1N1) virus containing the compromising H275Y NA mutation [13].
In addition to the R222Q mutation, a permissive role was also suggested for V234M and D344N substitutions [11], [13]. Interestingly, in a recent report on the evolution of influenza NA genes, positive epistasis (i.e. combination of mutations that are substantially more beneficial than single mutations alone) was detected in pairs of codons within the NA gene of the N1 subtype including 275−222, 275−234, and 275−344 [18]. In our study, although the M234V and N344D reversions were associated with decreased relative NA activity and affinity, respectively (Table 1 and Fig. 1), none of these mutations significantly altered the viral fitness in vitro. Nevertheless, a possible synergy between these mutations and Q222R cannot be completely excluded.
Our study revealed that the H275Y NA mutation was not deleterious to fitness in the Bris07 genetic context in contrast to older H1N1 strains. However, this mutant did not have a replicative advantage compared to the WT as suggested by epidemiological studies. Indeed, the recombinant WT virus and its H275Y variant demonstrated similar replication kinetics during in vitro experiments. In addition, these recombinants had comparable infectivity and contact transmissibility in ferrets. Thus, the presence of the permissive mutations (R222Q, V234M and D344N) in the NA protein of our WT strain was apparently not sufficient to alter the viral fitness to the level that a compensatory change, such as the H275Y mutation, would be necessary. Therefore, we believe that changes in the NA gene alone may not provide a complete explanation for the emergence and spread of the oseltamivir-resistant H275Y Bris07 variant. Other changes in the genome might have been involved in this event. For instance, Yang and colleagues recently demonstrated that the dominant H275Y variant that emerged in Taiwan in 2007–2008 was a result of intra-subtypic reassortments between HA, NA, PB2 and PA genes from one clade (clade 2B) and the remaining 4 genes from another one (clade 1) [19]. Furthermore, the H275Y NA substitution and other changes in NA, HA, PB1 and PB2 proteins occurred in that background [19]. Thus, it would be also interesting to assess the effect of HA and particularly polymerase mutations that differed between WT and H275Y mutant clinical Bris07 isolates on replicative capacities and transmissibility.
Despite the fact that the secondary mutations described here were not investigated individually but in conjunction with H275Y, our study provides a comprehensive analysis of relevant permissive NA mutations in the contemporarily seasonal H1N1 background. This included in vitro characterization, assessment of viral fitness and contact transmission in ferrets as well as NA enzyme properties of recombinant mutants. In particular, our investigation clearly demonstrated the positive impact of one specific NA substitution (i.e. R222Q) in conjunction with the oseltamivir resistance H275Y mutation on enzymatic properties and viral fitness of the Bris07 H1N1 strain. Noteworthy, our results suggest that total NA activity was more likely predictive of in vitro and in vivo viral fitness than the enzyme affinity (Km) parameter. Whether the Q222R mutation is also deleterious in the absence of H275Y was not investigated here; however, in a previous work, influenza A/Paris/497/2007 (222Q/275H) and A/Solomon Islands/3/2006 (222R/275H) seasonal H1N1 isolates grew to comparable titers in in vitro kinetics experiments [11]. Although clinical 2009 pandemic H1N1 variants containing such permissive mutations have not been reported, a computational approach had recently led to the identification of R257K and T289M as potential secondary mutations in that context [20]. Thus, monitoring for resistance in influenza viruses should take into consideration not only NA resistance-mutations themselves but also permissive/secondary ones as the latter may significantly affect the clinical and epidemiological impacts of seasonal or pandemic influenza viruses.
All procedures were approved by the Institutional Animal Care Committee at Laval University according to the guidelines of the Canadian Council on Animal Care.
Reverse transcription-PCR using universal influenza primers [21] was used to amplify the eight genomic segments of an oseltamivir-susceptible A/Quebec/15230/08 (H1N1) isolate whose HA and NA genes shared respectively 99.53% and 99.71% nucleotide identity with those of the influenza A/Brisbane/59/2007 vaccine strain [10]. All segments were cloned into the pJET plasmid (Fermentas, Burlington, ON, Canada) and sequenced. Sequence analysis confirmed the presence of histidine (H), glutamine (Q), methionine (M) and asparagine (N) residues at residues 275, 222, 234 and 344 (N1 numbering), respectively, of the NA protein. The PB1, PB2 and PA segments were sub-cloned into pLLBG whereas the HA, NA, NP, M1/M2 and NS1/NS2 segments were sub-cloned into pLLBA bidirectional expression/translation vectors as described [22]. The pLLBA plasmid containing the NA gene was used for the introduction of the H275Y mutation using appropriate primers and the QuikChangeTM Site-Directed Mutagenesis kit (Stratagene, La Jolla, CA). The resulting pLLB-NA275Y mutant plasmid was then used for reverting potential compensatory mutations (Q222R, M234V or N344D) as described above. All recombinant plasmids were sequenced to confirm the absence of undesired mutations. The eight bidirectional plasmids were cotransfected into 293T human embryonic kidney cells using the LipofectamineTM 2000 reagent (Invitrogen, Carlsbad, CA) as previously described [23]. Supernatants were collected 72 h post-transfection and used to inoculate ST6Gal1-MDCK cells kindly provided by Dr. Y. Kawaoka, University of Wisconsin, Madison, WI). The recombinant wild-type (WT) and H275Y, H275Y/Q222R, H275Y/M234V and H275Y/N344D mutant viruses were subsequently sequenced and titrated by standard plaque assays in ST6Gal1-MDCK cells.
A fluorometric based assay using MUNANA (Methylumbelliferyl-N-acetylneuraminic acid) (Sigma, St-Louis, MO) as substrate was performed to determine total NA enzymatic activity per infectious virus [24]. Briefly, recombinant viruses were standardized to an equivalent dose of 106 plaque forming-units (PFU)/ml and incubated at 37°C in 50-µl reactions with different concentrations of MUNANA. The final concentration of the substrate ranged from 0 to 3000 µM. Fluorescence was monitored every 90 s for 53 min (35 measures). The Michaelis-Menten constant (Km) and the relative NA activity (Vmax) were calculated with the Prism software (GraphPad, version 5), by fitting the data to the Michaelis-Menten equation using nonlinear regression [25].
Recombinant NA plasmids and pCAGGS-PA, -PB1, -PB2 and -NP plasmids were used to co-transfect 293T cells in order to express recombinant NA enzymes [26]. Twenty-four hours after transfection, the cells were briefly treated with trypsin-EDTA and neutralized by the addition of serum followed by centrifugation at 3000 RPM for 5 min. After washing twice with PBS, the cells were resuspended in a non-lysing buffer (15 mM MOPS, 145 mM sodium chloride, 2.7 mM potassium chloride and 4 mM calcium chloride, adjusted to pH 7.4) and used in an NA assay using the MUNANA substrate [13].
The drug resistance phenotype was determined by NA inhibition assays using the MUNANA substrate as previously described [26], with minor modifications. Briefly, recombinant viruses were standardized to a NA activity ten-fold higher than that of the background and then incubated with serial three-fold dilutions of the drugs (final concentrations ranging from 0 to 1800 nM), including oseltamivir carboxylate (Hoffmann-La Roche, Basel, Switzerland), zanamivir (GlaxoSmithKline, Stevenage, UK) and peramivir (BioCryst, Birmingham, AL). The 50% inhibitory concentration (IC50) was determined from the dose-response curve.
Replicative capacities of the recombinant viruses were evaluated by infecting ST6Gal1-MDCK cells with a multiplicity of infection (MOI) of 0.001 plaque-forming units (PFUs)/cell. Supernatants were collected every 12 h until 60 h PI and titrated by plaque assays. The mean viral plaque area of recombinant viruses was determined from a minimum of 16 plaques obtained after 60 h of incubation under agarose overlay using the ImageJ software (version 1.41), developed by Wayne Rasband of the National Institutes of Health as previously described [25].
Groups of 4 seronegative (900–1500 g) male ferrets (Triple F Farms, Sayre, PA) were lightly anesthetised by isoflurane and received an intranasal instillation of 1.25×105 PFUs of the recombinant Bris07-like WT, H275Y or H275Y/Q222R variants. Temperature of ferrets was measured by rectal thermometers every day until day 10 PI. Ferrets were weighed daily and nasal wash samples were collected from animals on days 2, 4 and 6 PI. Virus titers from nasal wash samples were determined by plaque assays using ST6Gal1-MDCK cells. Serum samples were collected from each ferret before intranasal infection and on day 14 PI to evaluate specific antibody levels against the seasonal Bris07 strain using standard HAI assays. To evaluate contact-transmissibility, inoculated-contact animal pairs were established by placing a naïve ferret into each cage 24 h after inoculation of the index ferret [27]. Contact animals were monitored for clinical signs and nasal wash and serum samples were collected as described above for determination of viral titers and serological status, respectively.
NA kinetic parameters (Km and Vmax values), NAI IC50 values and viral titers in vitro and in nasal washes of ferrets were compared by one-way ANOVA analysis of variance, with the Tukey's multiple comparison post test. The amount of NA activity on the cell surface and plaque sizes of the recombinants were compared to those of the WT virus and/or the H275Y mutant by the use of unpaired two-tailed t tests.
|
10.1371/journal.pbio.0060207 | Dual Role of Topoisomerase II in Centromere Resolution and Aurora B Activity | Chromosome segregation requires sister chromatid resolution. Condensins are essential for this process since they organize an axial structure where topoisomerase II can work. How sister chromatid separation is coordinated with chromosome condensation and decatenation activity remains unknown. We combined four-dimensional (4D) microscopy, RNA interference (RNAi), and biochemical analyses to show that topoisomerase II plays an essential role in this process. Either depletion of topoisomerase II or exposure to specific anti-topoisomerase II inhibitors causes centromere nondisjunction, associated with syntelic chromosome attachments. However, cells degrade cohesins and timely exit mitosis after satisfying the spindle assembly checkpoint. Moreover, in topoisomerase II–depleted cells, Aurora B and INCENP fail to transfer to the central spindle in late mitosis and remain tightly associated with centromeres of nondisjoined sister chromatids. Also, in topoisomerase II–depleted cells, Aurora B shows significantly reduced kinase activity both in S2 and HeLa cells. Codepletion of BubR1 in S2 cells restores Aurora B kinase activity, and consequently, most syntelic attachments are released. Taken together, our results support that topoisomerase II ensures proper sister chromatid separation through a direct role in centromere resolution and prevents incorrect microtubule–kinetochore attachments by allowing proper activation of Aurora B kinase.
| Successful cell division requires that chromosomes are properly condensed and that each sister chromatid is self-contained by the time the sister pairs are segregated into separate daughter cells. It is also essential that the kinetochores at the centromeres of each pair of sister chromatids bind microtubules from opposite spindle poles. Topoisomerase II is a highly conserved enzyme that removes interlinks from DNA and is known to be essential to proper chromosome segregation during cell division. In this work, we have used state-of-the-art four-dimensional fluorescent microscopy to follow progression through mitosis in living cells depleted of topoisomerase II. We find that when the enzyme is absent, the two sister centromeres do not separate, and chromosomes missegregate. Moreover, the inappropriate centromere structure that results prevents the correct activation of the Aurora B kinase, which forms part of a regulatory mechanism that monitors correct segregation of chromosomes; as a result, cells exit mitosis abnormally.
| Ordered segregation of the genome during cell division requires bipolar attachment to spindle microtubules [1] and maintenance of sister chromatid cohesion until anaphase onset [2]. Cohesin provides a physical link between sister chromatids, and cleavage of cohesin subunits results from separase activation after the spindle assembly checkpoint (SAC) is satisfied [3]. However, before segregation occurs, proper chromosome condensation and sister chromatid resolution must be completed. The condensin complex has been shown to play a key role in these processes by organizing an axial structure where topoisomerase II (TOPO II) localizes and decatenates entangled DNA strands that result from replication or transcription [4,5]. Indeed, the requirement of TOPO II activity in mitosis has been amply documented. In Saccharomyces cerevisiae, circular DNA molecules accumulate as catenated dimers in top2 mutants [6], and TOPO II activity prevents nondisjunction and DNA breakage during mitosis [7–9]. Injection of antibodies against TOPO II in Drosophila embryos [10], the addition of TOPO II inhibitors or RNA interference (RNAi) in mammalian culture cells and Xenopus extracts [11–14] caused severe defects in chromosome segregation during anaphase. More specifically, TOPO II activity has been suggested to affect normal centromere structure [15] where the protein normally accumulates in its catalytically active form [15–20]. These data strongly suggest that prior to segregation, TOPO II has a general role in promoting the resolution of sister chromatids. However, how this correlates with TOPO II activity at the centromeres remains a critically unanswered question.
To study the function of TOPO II during mitosis, we first analyzed the consequences of depleting the enzyme by RNAi or treating Drosophila S2 cells with specific inhibitors (Figure 1). Significant levels of TOPO II depletion were obtained by RNAi treatment as shown by western blot analysis in which the protein is barely detectable after 72 h (Figure 1A). However, we found that these cells apparently progress normally through early stages of mitosis but show severe segregation defects during anaphase and telophase, and cell proliferation is significantly inhibited without altering the mitotic index (Figure 1B–1D). Quantification of chromosome segregation abnormalities shows that after long RNAi treatment, a significant proportion of cells display either chromatin bridges or lagging chromatids during anaphase (Figure 1B, 1E, and 1F). Immunofluorescence analysis of chromosome morphology with antibodies against condensin subunits reveals that depletion of TOPO II does not significantly affect mitotic chromosome structure (Figure 1G). Cells were also treated with the TOPO II inhibitor ICRF-187, a bisdioxopiperazine-type chemical that has been shown to interfere with the catalytic activity of TOPO II [21]. However, treatment of cells with ICRF-187 results in a more pronounced alteration in chromosome structure (Figure 1H). The exact role of TOPO II in mitotic chromosome structure remains highly debatable. This is due to the fact that the use of different procedures to disrupt TOPO II function and localization in several model organisms has led to conflicting results [22]. Moreover, previous studies have shown that TOPO II inhibitors may also result in the activation of the G2 checkpoint because they block the activity of the enzyme in different conformation states [23]. Therefore, we have resorted to depleting TOPO II by RNAi for most of our study.
To directly assess whether TOPO II function is required at centromeres, we depleted the single TOPO II isoform from Drosophila S2 cells stably expressing fluorescent markers for chromatin (mRFP-H2B) and centromeres (CID-GFP) [24] by RNAi treatment (Figure 2 and Videos S1–S4). In order to visualize individual sister centromeres at high temporal and spatial resolution, TOPO II–depleted cells were imaged by four-dimensional (4D) time-lapse fluorescence microscopy in which the distance between the optical layers of the Z-stack was kept to less than 1 μm. In control cells, chromosomes congression occurs normally, and as anaphase starts, CID-GFP pairs disjoin and move poleward (Figure 2A and Video S1). However, in TOPO II–depleted cells, while chromosomes appear to exhibit normal congression, centromeres of sister chromatids remain on the same side of the metaphase plate, fail to disjoin, and move towards the same pole during anaphase (Figure 2B and 2C and Video S2). Nevertheless, in many cells after 72 h of RNAi treatment, chromatin bridges presumably linking chromosome arms are clearly observed (Figure 2C and Video S3). At later times after RNAi treatment (96 h), most cells show chromosome nondisjunction (Figure 2D and Video S4). Noteworthy, most of the analysis after TOPO II depletion was carried out in cells that did not show extensive polyploidy, as ascertained by chromosome and centromere labeling, indicating that they had not undergone multiple cell cycles. After long RNAi treatment, a small proportion of polyploid cells were observed (Figure S1). These cells are characterized by the presence of chromosomes that are attached by their nondisjoined centromeres, as would be expected if in the previous cycles, proper centromere separation failed. This effect is apparently not due to a failure to replicate centromeric DNA as shown by Southern blotting analysis (Figure S1).
In order to determine why sister centromeres fail to disjoin after TOPO II depletion, cells stably expressing GFP-α-tubulin and CID-mCherry to specifically label spindle microtubules and centromeres were treated with RNAi, and mitotic progression was followed by time-lapse fluorescence microscopy (Figure 3A and 3B and Videos S5 and S6). As expected, in control cells, chromosomes show mostly amphitelic attachments, congression occurs normally, and bundles of spindle microtubules are easily observed associated with individual kinetochores (Figure 3A and Video S5). However, 72 h after TOPO II RNAi, most chromosomes show syntelic attachment and are oriented towards the same spindle pole (Figure 3B and 3C and Video S6). To confirm these observations, we performed an assay designed to quantify the nature of microtubule–kinetochore interactions in S2 cells [25]. For this assay, cells were arrested in mitosis with the proteasome inhibitor MG132 and subjected to a high dose of Taxol, which over a short period of time causes the collapse of the bipolar spindle into a monopolar configuration. This monopolar structure now contains the chromosomes distributed at the periphery of the aster, and microtubule–kinetochore interactions can be easily scored (Figure S2). We find that in control and TOPO II–depleted cells, more than 95% of the chromosomes had both kinetochores attached to microtubules in a syntelic configuration soon after significant depletion of TOPO II occurs. If these cells are analyzed when the two asters are in the process of collapsing, it is possible to ascertain whether the attachment is amphitelic, with chromosomes localized between the asters, or syntelic/monotelic, when located at the periphery of the aster (Figure 3D–3H). In control cells (n = 24), most chromosomes (78%) exhibited a clear amphitelic configuration, whereas in TOPO II–depleted cells (n = 20), most chromosomes (65%) localized at the periphery of the two asters with a clear syntelic configuration (Figure 3F and 3G). Consistently, CID fluorescence intensity for the centromere marker CID in TOPO II–depleted cells is almost double that of control cells (unpublished data), and intercentromere distances never increase during mitotic progression (Figure S3). Analysis of microtubule–kinetochore interactions after TOPO II depletion was also performed in asynchronous cells. For this, either control or TOPO II–depleted cells were fixed and stained for CID, α-tubulin, and CENP-meta, the Drosophila CENP-E homolog, a kinetochore motor protein whose levels decrease significantly at kinetochores during anaphase [26] (Figure 3I–3O). As expected, we find that CENP-E is present at kinetochores of control and TOPO II–depleted cells in prometaphase (Figure 3I and 3L), but when chromosomes move poleward, CENP-E is undetectable (Figure 3N), indicating that all kinetochores are attached to spindle microtubules. Rendering these images for the staining of CID and tubulin confirms that in the absence of TOPO II, kinetochore bundles are associated with pairs of CID dots, unlike control cells in which they associate with single CID dots (Figure 3J, 3M, and 3O). Taken together, these results indicate that in the absence of TOPO II, sister centromeres fail to disjoin and chromosomes show mostly syntelic microtubule attachments.
Long-term inhibition of TOPO II with drugs affects S2 cells during G2 and mitotic entry, resulting in severe abnormalities in chromosome structure that prevented us from using inhibitors to carry out a thorough analysis of the role of TOPO II during mitotic progression. However, we hypothesized that short incubations with the inhibitors might allow us to study its effects in living cells as they enter mitosis. Accordingly, cells were treated for short periods of time with ICRF-187. S2 cells stably expressing GFP-α-tubulin and CID-mCherry were imaged by 4D microscopy during ICRF-187 treatment, before and after establishment of the metaphase plate (Figure 4 and Videos S7–S9). We find that inhibition of TOPO II activity after chromosomes have reached the metaphase plate and established bipolar attachment does not have an effect on the kinetochore–microtubule interaction (Figure 4A and 4B). However, if the TOPO II inhibitor is added before nuclear envelope breakdown, as inferred by the exclusion of GFP-α-tubulin from the nucleus (unpublished data), during prometaphase, we observe that 62.5% of chromosomes/cell (n = 12) exhibit syntelic chromosome attachments, with kinetochore pairs moving poleward without separating their centromeres (Figure 4C). These observations not only confirm our live-cell analysis of TOPO II–depleted cells after RNAi, but also indicate that TOPO II activity at early stages of mitosis plays an important role at centromeres to promote normal chromosome biorientation. However, once amphitelic attachments are achieved, TOPO II activity is not required for their maintenance.
Previously, it has been demonstrated that as cells progress through late prometaphase and metaphase, cohesin is removed from the chromosome arms, remaining only at the centromere until the metaphase–anaphase transition [27–29]. Therefore, it remains possible that sister centromeres fail to disjoin after depletion of TOPO II due to an inappropriate accumulation of cohesin. In order to test this hypothesis, we determined the localization of the cohesin subunit RAD21/SCC1 in control and TOPO II–depleted cells before and after incubation with colchicine. In control cells at metaphase or after colchicine incubation, cohesin localizes as a clearly defined stripe between centromeres (Figure 5A), but after depletion of TOPO II, cohesin shows a very abnormal distribution, which extends into the chromosome arms even after mitotic arrest (Figure 5B). To test whether the inappropriate localization of cohesin accounts for the observed centromere nondisjunction in TOPO II–depleted cells, we performed simultaneous depletion of TOPO II and RAD21 by RNAi and followed mitotic progression and sister chromatid segregation (Figure 5C). Although both proteins were efficiently depleted, cells progressed through mitosis, showing lagging chromatids or chromosomes (Figure S4). The 4D microscopy studies of mitotic cells stably expressing CID-GFP and RFP-H2B after simultaneous depletion of TOPO II and RAD21 show that the behavior of sister chromatids is identical to that of cells depleted of TOPO II alone and very different from RAD21 RNAi (Figure 5F and Videos S10 and S11). Cells depleted of RAD21 enter mitosis with separated sister chromatids and arrest in a prometaphase-like state (Figure S5, and Video S12) because they fail to inactivate the SAC [30,31]. However, in cells depleted of both TOPO II and RAD21, closely paired sister chromatids reach the metaphase plate and, during anaphase, fail to disjoin, segregating together to the same spindle pole. These results demonstrate that inappropriate localization of RAD21 is unlikely to be responsible for the centromere nondisjunction phenotype observed after depletion of TOPO II.
Together with the data, our results strongly suggest that depletion of TOPO II causes the formation of a physical linkage between sister centromeres, likely provided by DNA concatameres, which is not resolved during the metaphase–anaphase transition. As a consequence, TOPO II–depleted cells segregate entire chromosomes rather than sister chromatids. Apparently, this is not due to the inability of TOPO II–depleted cells to satisfy the SAC since they showed no significant mitotic delay and anaphase-promoting complex (APC/C)-dependent proteolysis of RAD21 and cyclin B occurs (Figure S6). If depletion of TOPO II causes failure to resolve sister chromatids and syntelic attachments, how then do these cells satisfy the SAC? To address this question, we first determined the localization of SAC proteins after TOPO II RNAi. We find that in prometaphase, BubR1 accumulates strongly at kinetochores, while during anaphase, the level of BubR1 was significantly reduced, suggesting that TOPO II–depleted cells are able to inactivate the SAC, just like control cells (Figure 6A and 6B). To further determine whether syntelic chromosomes undergo proper tension during mitosis, TOPO II–depleted cells were immunostained with the 3F3/2 monoclonal antibody that specifically detects kinetochore phosphoepitopes in the absence of tension [32–34]. We find that control and TOPO II–depleted cells behave very similarly so that 3F3/2 kinetochore phosphoepitopes are strongly labeled in prometaphase, become significantly reduced during prometaphase/metaphase, and are undetected in anaphase (Figure 6C and 6D). These results indicate that SAC satisfaction in TOPO II–depleted cells with syntelic attachments correlates with the dephosphorylation of 3F3/2 epitopes but does not require extensive interkinetochore stretching.
Previously, it has been shown that the correction of improper microtubule–kinetochore attachments requires Aurora B kinase activity as part of a tension-sensing mechanism on centromeres [1]. Aurora B is part of the chromosomal passenger complex (CPC) that localizes to the inner centromere during prometaphase/metaphase and transfers to the spindle midzone at anaphase onset and the mid body in telophase [35]. Therefore, we analyzed whether the localization of CPC proteins after TOPO II depletion was compromised and could be responsible for the inability of these cells to release syntelic attachments (Figure 7). In control cells, Aurora B localizes to the inner centromere region during prometaphase/metaphase and is transferred to the spindle midzone when cells initiate anaphase [36] (Figure 7A). However, in TOPO II–depleted cells Aurora B localization is abnormal (Figure 7B). In prometaphase, Aurora B remains associated with sister kinetochores, does not stretch across the metaphase plate, remains associated with inner centromeres of syntelic chromosomes during anaphase, and fails to transfer to the spindle midzone. A very similar abnormal pattern of localization was also observed for INCENP, a member of the CPC, which regulates Aurora B activity (Figure 7C and 7D). These observations indicate that TOPO II is essential for the organization of the inner centromere so that the CPC can show a normal pattern of localization during mitosis.
Previous studies in human cells have shown that chromosomes with syntelic attachments experience significant distortion of their centromeres and are not easily identified by the SAC, and that these errors are enhanced when aurora B kinase activity is inhibited [37]. Moreover, it has been shown that Aurora B is normally enriched at sites associated with erroneous microtubule attachments where it promotes microtubule depolymerization [38]. Since in TOPO II–depleted cells Aurora B remains associated with syntelic attachments throughout mitosis, it would be expected that these erroneous attachments would be corrected, unless Aurora B activity is compromised due to structural changes resulting from TOPO II depletion. To test this hypothesis, we analyzed the levels of phosphorylation of histone H3 at Ser 10 (PH3), a known Aurora B substrate [39] (Figure 8A). Immunofluorescence analysis revealed that after TOPO II depletion at 96 h, PH3 levels were reduced almost by half (41%) when compared with control cells (Figure 8B) and not much different from the reduction (62%) observed after RNAi depletion of Aurora B (Figure 8A and 8B). Similar results were obtained after treatment of cells with the TOPO II inhibitor ICRF-187 (Figure S7). Western blot of total protein extracts of TOPO II– and Aurora B–depleted cells confirmed that PH3 levels were significantly reduced (Figure 8C). However, although TOPO II–depleted extracts show a significant reduction in PH3 reactivity, the total Aurora B levels appear unaffected, suggesting that depletion of TOPO II specifically affects the kinase activity of Aurora B. To directly address this possibility, Aurora B was immunoprecipitated from total protein extracts from control or TOPO II–depleted cells and its kinase activity tested in vitro with unphosphorylated histone H3 (Figure 8D). We find that phosphorylation of histone H3 is reduced by half relative to controls when Aurora B is immunoprecipitated from TOPO II–depleted cells. This indicates that either directly or indirectly, TOPO II is required to promote Aurora B kinase activity. To further analyze how TOPO II regulates Aurora B activity at centromeres, we turned to HeLa cells. Cells were treated with the TOPO II inhibitor ICRF-187, and Aurora B kinase activity was quantified by measuring the phosphorylation of Ser7 of the centromeric protein CENP-A that has been found to be a direct substrate of Aurora B [40] (Figure 8E and 8F). In control cells, we find that P-Ser7CENP-A immunoreactivity is very high in most kinetochores during prometaphase, revealing the normal activity of Aurora B at this stage of mitosis. As cells reach metaphase and kinetochore–microtubule interactions become stabilized, P-Ser7CENP-A immunoreactivity is significantly reduced, suggesting that Aurora B kinase activity is normally down regulated at this stage. However, after inhibition of TOPO II, we find that cells in prometaphase display a significant reduction in the level of P-Ser7CENP-A immunoreactivity (Figure 8E–8G), and more than 60% of chromosomes per cell (n = 26) display syntelic attachment. These results indicate that TOPO II is also required to establish amphitelic attachment in HeLa cells, similar to what we observed for Drosophila, and further demonstrate that TOPO II activity regulates Aurora B kinase activity on chromosomes and more specifically at centromeres.
The observations described above demonstrate that TOPO II activity at the centromere is required for the normal function of Aurora B. However, these studies do not distinguish whether TOPO II controls Aurora B kinase activity directly or indirectly. Previous studies showed that inhibition of Aurora kinase activity suppresses the misalignment/attachment defects in BubR1-depleted cells [41]. This effect was shown to be due to an increase in Aurora B kinase activity after BubR1 depletion [41]. Similarly, small interfering RNAi (siRNAi) depletion of Aurora B in cells where BubR1 was also knocked down, results in more stable kinetochore attachment [42]. Therefore, given that in the absence of TOPO II, sister centromeres appear unable to resolve, bind microtubules syntelically, and segregate to the same pole, we tested whether BubR1 might be responsible for negatively regulating Aurora B activity in these cells (Figure 9). To address this issue, we measured microtubule–kinetochore attachment using the Taxol-MG132 assay, as well as mitotic PH3 reactivity in control, BubR1-, TOPO II–, and TOPO II/BubR1–depleted S2 cells. Single or double RNAi treatments were carried out and the respective protein levels quantified by western blotting (Figure 9G). As described before, the Taxol-MG132 assay shows that in control or TOPO II–depleted cells, most chromosomes are attached to spindle microtubules (Figure 3), but when BubR1 is depleted alone, many cells show either unattached or mono-oriented chromosomes [25] (Figure 9A and 9B). Interestingly, when BubR1 and TOPO II are simultaneously depleted, we observed a large increase in the number of unattached kinetochores (Figure 9A and 9B). This result indicates that removing BubR1 from TOPO II–depleted cells can reactivate the correction mechanism and allow the release of syntelic attachments. To determine whether this was the result of Aurora B kinase activity, we then analyzed PH3 levels by immunofluorescence microscopy. The results show that depletion of BubR1 in the absence of TOPO II is able to restore normal PH3 levels on chromatin, suggesting that Aurora B kinase is now active (Figure 9C and 9D). We also analyzed the localization of Aurora B during mitotic exit of cells after simultaneous depletion of TOPO II and BubR1 (Figure 9E). Interestingly, during early anaphase, Aurora B is found tightly associated with the centromeres but in late anaphase is no longer detected and accumulates in the spindle midzone (Figure 9E and 9F). These results suggest that TOPO II is unlikely to have a direct role in regulating the kinase activity of Aurora B. Instead, the abnormal configuration of the centromere resulting from TOPO II depletion appears to cause inappropriate inhibition of Aurora B through BubR1.
Our live-cell studies show that TOPO II has a central role in promoting structural changes of the centromeric DNA that are essential for their individualization and separation at metaphase–anaphase transition. This process is clearly independent of the cohesin complex since depletion of RAD21 causes a SAC-dependent prometaphase-like arrest with separated sister chromatids [4,43], which can be overcome by simultaneous depletion of condensin [4] or TOPO II (Figure 5). Therefore, whereas the role of cohesin degradation in defining the initial steps of sister chromatid separation is well established, it is clear that these events must be tightly coordinated with TOPO II activity. Although it was previously suggested that TOPO II might have a role at the centromere [15,44], previous functional studies have failed to detect any effect on centromere separation during mitotic exit [12,31,45]. Therefore, our results provide the first direct evidence that TOPO II activity is required for centromere disjunction during mitosis.
Our results further show that the structural changes of centromeric DNA resulting from the decatenation activity of TOPO II appear to be essential for the establishment of amphitelic microtubule–kinetochore attachments. In the absence of TOPO II, the SAC appears unable to detect sister kinetochores that are attached to the same pole. One possible explanation is that in TOPO II–depleted cells, Aurora B kinase activity is down-regulated, and given its role in activating the SAC in response to loss of tension, cells cannot respond properly and therefore do not activate the correction mechanism. Interestingly, it has been shown that during exit from meiosis I, when sister chromatids do not disjoin, Aurora B and INCENP remain at the inner centromere [46], similar to what we observe after depletion of TOPO II. Thus, chromosome segregation in TOPO II–depleted cells resembles the first meiotic division when both sister kinetochores are oriented towards the same pole, suggesting that TOPO II may play a role in modulating centromere structure required for proper bivalent biorientation.
The functional interaction between TOPO II and Aurora B has been explored before. In human cells, TOPO II was demonstrated to be an in vitro substrate of Aurora B [47]. Here, we show that depletion of TOPO II causes a down-regulation of Aurora B kinase activity. We observed that the levels of chromosome-associated PH3 staining during prometaphase and metaphase are significantly reduced after depletion of TOPO II but also find that after treatment of S2 cells with a TOPO II inhibitor ICRF-187, which compromises TOPO II activity without changing its chromosomal localization, PH3 reactivity is also significantly reduced. In agreement, inhibition of TOPO II catalytic activity in human cells also results in a dramatic reduction on the phosphorylation levels of Ser 7 CENP-A phosphoepitope, indicating that Aurora B activity is affected not only on chromosomes, but also specifically at centromeres.
The reduction in Aurora B kinase activity could either result from a direct effect of TOPO II or, more likely, through an alteration of the structure of the centromere that occurs as a consequence of TOPO II depletion, and therefore likely represents an indirect effect. We addressed this issue by codepleting BubR1, a SAC protein thought to be involved in inhibiting Aurora B at kinetochores, and find that indeed, codepletion of TOPO II and BubR1 restores normal Aurora B kinase activity and releases syntelic chromosome attachments. Previous work in HeLa cells has shown that either inhibition of Aurora B kinase activity [41] or depletion of Aurora B by RNAi [42] suppresses the misalignment/attachment defects observed in BubR1-depleted cells. In agreement, an increase in Aurora B kinase activity has been reported in the absence of BubR1 [41]. Taken together, our results suggest that BubR1 is able to inhibit Aurora B when in close proximity, so that in early stages of prometaphase, when microtubule attachment is being established and there is not sufficient tension, Aurora B is not activated. However, the increase in tension upon chromosome biorientation, which increases the distance between BubR1 and the centromere during prometaphase, allows Aurora B activation. Indeed, after TOPO II depletion, sister centromeres remain very close, and therefore BubR1 could be responsible for inhibiting Aurora B. Depletion of both BubR1 and TOPO II results in reactivation of Aurora B kinase activity, release of syntelic attachments, and the formation of unattached or mono-oriented chromosomes. These data support recent observations suggesting that activation of microtubule–kinetochore correction mechanisms during mitosis is dependent on centromere plasticity, but not on centromere elasticity [48]. In summary, our observations demonstrate that TOPO II is required for structural changes at the centromere during their resolution, and in turn, this allows normal function of Aurora B, maintenance of SAC activity, and eventual activation of the mechanisms that correct abnormal microtubule–kinetochore attachments.
RNAi was performed in Drosophila S2 tissue culture cells as previously described [49]. A 1,000-bp EcoRI-HindII and an 800-bp EcoRI-KpnI fragment from the 5′ end of TOPO II (RE49802) and RAD21 cDNAs [4], respectively, were cloned into both pSPT18 and pSPT19 expression vectors (Roche). The recombinant plasmids were used as templates for RNA synthesis using the T7 Megascript kit (Ambion), and 15 μg of double-stranded RNA (dsRNA) were added to 106 cells in all RNAi experiments. At each time point, cells were collected and processed for immunoblotting or immunofluorescence. For immunoblotting, cells were collected by centrifugation, washed in PBS supplemented with protease inhibitors (Roche), and resuspended in 20 μl of SDS sample buffer before loading on a 5%–20% gradient SDS-PAGE. When required, cells were incubated with 30 μM colchicine prior to fixation (Sigma).
Live analysis of mitosis was done on S2 cells stably expressing GFP-CID and RFP-H2B [24], as well as in CID-Cherry and GFP-α-tubulin. A cell line stably expressing both GFP-α-tubulin and CID-mCherry was created by transfecting S2 GFP-α-tubulin cells (a kind gift from Gohta Goshima [50]) with pMT_cid_mCherry_BLAST vector (designed from the pMT_cid_gfp, a kind gift from Karpen), pCoBLAST (Invitrogen), and pRSET-B_mCherry (Invitrogen). Control or TOPO II RNAi-treated cells were incubated for 72–96 h and plated on glass coverslips treated with 30 μg/ml concanavalin A (Sigma). Time-lapse images were collected every 20 s for CID-GFP and RFP-H2B and every 45 s for CID-mCherry and GFP-α-tubulin, by Scanning Confocal Microscope Leica SP2 AOBS SE (Leica Microsystems), using the software provided by the manufacturer, Software LCS (Leica Microsystem). Each Z-stack is composed of ten images at 0.8–1-μm intervals. Data stacks were deconvolved with the Huygens Essential version 3.0.2p1 (Scientific Volume Imaging). Image sequence analysis and video assembly was done with ImageJ Software (NIH) and Quicktime 7 (Apple Computer).
For immunostaining, 2 × 105 cells were centrifuged onto slides, simultaneously fixed and extracted in 3.7% formaldehyde (Sigma), 0.5% Triton X-100 in PBS for 10 min, and then washed three times for 5 min in PBS-T (PBS with 0.05% Tween 20). Blocking and incubating conditions were performed as described previously [4].For immunofluorescence with the monoclonal antibody (mAb) 3F3/2, cells were grown on glass coverslips, after which they were simultaneously lysed and fixed in lysis/fixation buffer for 2 min (1.5× PHEM, 2% Triton X-100, 0.15% glutaraldehyde, 2% formaldehyde, 10 μM microcystin LR) by 1:1 dilution directly in the culture dishes. Detergent-extracted S2 cells were fixed in 1% formaldehyde in 1× PHEM with 10 μM microcystin for 12 min at room temperature. Coverslips were then washed with 0.5× PHEM, and immunofluorescence was done as described previously [51]. Images Z-stacks were collected using the Scanning Confocal Microscope Leica SP2 AOBS SE (Leica Microsystems) and the software provided by the manufacturer, Software LCS (Leica Microsystems). Data stacks were deconvolved, using the Huygens Essential version 3.0.2p1 (Scientific Volume Imaging).
HeLa cells were cultured in DMEM medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS) and grown at 37 °C in a 5% CO2 humidified chamber. Cells were fixed for 12 min in freshly prepared 2% paraformaldehyde (Sigma) in 1× PHEM, permeabilized with 0.5% Triton X-100 in PBS 3 times for 5 min, washed in PBS, and blocked with 10% FBS. Incubation with primary and secondary antibodies was performed in 1× PBS with 10% FBS.
Primary antibodies were anti–α-tubulin (mouse mAb B512), used at 1:3,000 (Sigma-Aldrich); antiphosphorylated histone H3 rabbit polyclonal, used at 1:500 (Upstate Biotechnology); anti-CID chicken polyclonal, used at 1:200 [52]; anti-SMC4 antibodies (rabbit, 1:500, or sheep, 1:200), as described previously [53]; anti-Bubr1 rat polyclonal [25] used at 1:3,000; anti-RAD21 [29] rabbit polyclonal (1:500); anti-Polo mouse mAb MA294, used at 1:50 [54]; anti–Aurora B polyclonal antibody [36], used at 1:1,000; anti–Aurora B polyclonal antibody [55], used at 1:500 in western blot; antiphospho(Ser7)CENPA (Upstate Biotechnology), used at 1:500; anti-3F3/2 mAb [56], used at 1:1,000; anti-CENP-meta rabbit polyclonal (Byron Williams) used at 1: 1,000; anti–cyclin B rabbit polyclonal [57], used at 1: 3,000; and anti–TOPO II mouse mAb P2G3 [58], used at 1:20.
MG132 (Sigma) was used to inhibit the proteasome activity in S2 Drosophila cells according to the conditions previously described [25]. ICRF-187 at 50 μg/ml was used to inhibit TOPO II activity both in S2 Drosophila cells and in HeLa cells. Incubations were always performed for 2 h, excepted for in vivo–timed TOPO II inhibition. The TOPO II inhibition was done according to the description in each figure.
Fluorescent in situ hybridization to mitotic chromosomes, 2 × 105 cells were centrifuged onto slides, after which cells were simultaneously fixed and extracted as described previously. Cells were dehydrated by incubation for 5 min in 70%, 80%, and 100% ethanol at 4 °C. Cells were air dried and denatured in 2× SSC; 70% formamide for 2 min at 70 °C. Cells were dehydrated once again as described before. We labeled the pericentromeric probe dodeca-satellite DNA with biotin-14-dATP using the BionickTM DNA labeling system (Invitrogen). Detection of the biotinylated probe was done with avidin-D conjugated with fluorescein (Vector Lab).
A total of 30 ml of S2 cells were grown exponentially to 4–5 × 106 cells/ml, incubated on ice for 45 min, and centrifuged at 1,500g for 15 min at 4 °C. Cells were resuspended in 1 ml of cold 1× PBS in the presence of protease inhibitors, kept on ice for 45 min, and lysed in 1 ml of lysis buffer (15 mM Tris-HCl [pH 7.4], 0.2 mM spermine, 0.5 mM spermidine, 2 mM K-EDTA, 1 mM EGTA, 150 mM KCl, 15 mM NaCl, 1 mM DTT, 1% [V/V] Triton X-100) with 2× protease inhibitors-EDTA free (Roche) and 1× phosphatase inhibitor cocktail I (Sigma). Cells were lysed with a B-type pestle in a Dounce homogenizer after which they were incubated on ice for 1 h. Samples were precleared with Protein A-Sepharose CL-4B ( Sigma) for 30 min at 4 °C. Antibody anti–Aurora B (5 μl) and 100 μl of 10% Protein A-Sepharose beads were added to the samples and rotated for 1 h at 4 °C. The beads containing the immune complexes were washed three times in 1 ml of lysis buffer. The pellet was resuspended in 20 μl of serine/threonine kinase assay buffer (10 mM Tris-HCL [pH 7.4]; 0.1% Triton X-100; 10 mM MgCl2) with 1× phosphatase inhibitor cocktail I (Sigma), 50 μCi (185 KBq) of [γ32P]-adenosine 5′-triphosphate (ATP; >5,000 Ci/mmol; Amersham) and 10 μg of histone H3 (Upstate Biotechnology).
The preparation of the genomic DNA was done using 107 cells both for control and for TOPO II double-stranded RNAi (dsRNAi) cells. Cells were collected and spun down at 1,000g for 3 min. Pellets were resuspended in 300 μl of STE (150 mM NaCl; 30 mM Tris-HCl [pH 8.0]; 2 mM EDTA). We added 3 μl of 10% NP40, 30 μl 10%SDS, and 30 μl of 10 mg/ml proteinase K (Sigma) and incubated at 55 °C for 3 h. We performed phenol/chloroform extraction. Aqueous fraction was recovered and extracted once again with chloroform. DNA was precipitated with 20 μl of 5 M NaCl and 400 μl of 100% ethanol. Genomic DNA was digested with the restriction enzyme HindIII (Biolabs) according to the manufacturer's conditions. Electrophoresis of the genomic DNA was performed in a 0.7% agarose gel, and the gel was prepared for standard alkaline Southern blotting. DNA probes were radioactively labeled with [α-32P]dCTP using a multiprime labeling kit (Amersham).
|
10.1371/journal.ppat.1006457 | Potassium is a key signal in host-microbiome dysbiosis in periodontitis | Dysbiosis, or the imbalance in the structural and/or functional properties of the microbiome, is at the origin of important infectious inflammatory diseases such as inflammatory bowel disease (IBD) and periodontal disease. Periodontitis is a polymicrobial inflammatory disease that affects a large proportion of the world's population and has been associated with a wide variety of systemic health conditions, such as diabetes, cardiovascular and respiratory diseases. Dysbiosis has been identified as a key element in the development of the disease. However, the precise mechanisms and environmental signals that lead to the initiation of dysbiosis in the human microbiome are largely unknown. In a series of previous in vivo studies using metatranscriptomic analysis of periodontitis and its progression we identified several functional signatures that were highly associated with the disease. Among them, potassium ion transport appeared to be key in the process of pathogenesis. To confirm its importance we performed a series of in vitro experiments, in which we demonstrated that potassium levels a increased the virulence of the oral community as a whole and at the same time altering the immune response of gingival epithelium, increasing the production of TNF-α and reducing the expression of IL-6 and the antimicrobial peptide human β-defensin 3 (hBD-3). These results indicate that levels of potassium in the periodontal pocket could be an important element in of dysbiosis in the oral microbiome. They are a starting point for the identification of key environmental signals that modify the behavior of the oral microbiome from a symbiotic community to a dysbiotic one.
| Homeostasis of the human microbiome plays a key role in maintaining the healthy status of the human body. Changes in composition and function of the human microbiome (dysbiosis) are at the origin of important infectious inflammatory diseases such as inflammatory bowel disease (IBD) and periodontal disease. However, the environmental elements that trigger the development of dysbiotic diseases are largely unknown. In previous studies, using community-wide transcriptome analysis, we identified ion potassium transport as one of the most important functions in the pathogenesis of periodontitis and its progression. Here, we confirm with a series of in vitro experiments that potassium can act as an important signal in the dysbiotic process inducing pathogenesis in the oral microbiome and altering the host response in front of the microbial challenge that could lead to microbial immune subversion. Our study provides new insights into the important role that ion potassium plays a signal in oral dysbiosis during periodontitis.
| Dysbiosis, or the imbalance in the structural and/or functional properties of the microbiome, leads to the breakdown of host-microbe homeostasis, and has been associated with the pathogenesis of several important inflammatory diseases mediated by the activity of the microbial community, such as inflammatory bowel diseases (IBDs) [1,2] and periodontal diseases [3].
Periodontitis is a polymicrobial disease caused by the coordinated action of a complex microbial community, leading to inflammation and periods of active destruction of the tissues supporting the teeth. Periodontal inflammation has adverse impacts on a wide variety of systemic health conditions, such as diabetes, cardiovascular and respiratory diseases [4,5]. It is the sixth most prevalent disabling health condition in the world affecting approximately 750 million people worldwide [6]. It has been postulated that changes in the composition of subgingival biofilms could explain these periods of disease activity. In fact, a few studies have found differences in the levels of subgingival species when comparing progressing and non-progressing sites [7,8]. These studies also demonstrated considerable overlap in the composition of the microbial communities associated with progressing and non-progressing lesions, suggesting that the difference in the periodontal status of the sites could not be explained solely by differences in subgingival microbial composition. Recently, it has been proposed that certain organisms could act as 'keystone-pathogens' that modulate the behavior of the oral microbial community, which becomes dysbiotic [9]. Indeed, we have previously reported that organisms not considered pathogens express large numbers of putative virulence factors during chronic severe periodontitis and disease progression [10,11]. However, the environmental signals that trigger this change in behavior of the community remain for the most part unknown.
In two previous studies focused on the oral microbial metatranscriptome in health, disease and during periodontitis progression, gene ontology (GO) enrichment analysis showed that potassium ion transport was a key signature of microbial metabolic activities associated with disease [10,11]. In the present study we focus our interest on confirming our previous in vivo observations [11] on the potential role that potassium ion has as a signal that initiates changes in the oral microbiome, leading to dysbiosis of the microbial community. Potassium is the most abundant monovalent ion inside the cells. However, in healthy periodontal tissues potassium is present at low concentrations in the gingival crevicular fluid in contact with the oral biofilm [12–14]. A positive and statistically significant correlation has been found between the concentration of potassium in crevicular fluid and mean pocket depths [12], probably due to cell lysis of host cells. Here we performed a series of experiments to test the effects of potassium on plaque community gene expression, virulence, and inflammation and showed that levels of potassium ion act as an important environmental signal for microbial dysbiosis and epithelial response to the microbial challenge.
Potassium ion (K+) transport has previously been identified as an important signature among the metabolic activities of the oral microbiome in periodontitis [10,11]. However, the exact mechanisms by which potassium exerts its activity as an environmental signal leading to microbial dysbiosis remain unknown. To test the effects of potassium on plaque community gene expression, plaque was collected from a healthy human volunteer, exposed to saliva with or without added potassium, and subjected to metatranscriptomic analysis. We used K+ concentrations akin to those found in gingival crevicular fluid in severe periodontitis [13,14]. After only 3 hours of incubation RNA was extracted for analysis. We thus identified the initial reaction of the community to higher levels of K+ in the environment. We detected between 73.7 and 96.2% of all genes in our libraries, which represents a high sequencing depth across all samples (S1 Fig).
Phylogenetic assignment of the transcripts showed that several members of the community responded immediately to the presence of K+ in their surroundings (Fig 1). Transcripts were assigned to different taxa using Kraken and LEfSe was used to determine differentially transcriptionally active taxa. Among those that contribute a significantly higher fraction of transcripts to the metatranscriptome, we found organisms that previously have been associated with periodontal disease such as Leptotrichia spp., Campylobacter spp. and Fusobacterium spp. and Prevotella spp. [15], but also organisms that have been considered to be associated with health such as Streptococcus spp. Nonetheless Streptococcus spp. have been also found in large numbers in periodontal disease [15], and we previously identified them as producing large numbers of putative virulence factors at early states of dysbiosis during periodontitis progression [11]. Genus Lautropia, which has been associated with health [16], was significantly less active in the presence of K+. However, other groups of microorganisms that have been considered to be associated with periodontal disease such as Corynebacterium and Campylobacter [15,16] were less active in the presence of high concentration of K+.
To determine changes in metabolic activities in the whole community due to the increase in K+ concentration, we performed GO terms enrichment analysis. In order to mimic as much as possible the conditions present in the periodontal pocket during disease, we utilized levels of K+ of the same order as the ones found in severe periodontitis [12–14]. After only 3 hours of incubation, we observed at the community-wide level an over-representation of activities associated with disease, such as iron ion transport, oligopeptide transport, flagellum assembly and cobalamin biosynthesis (Fig 2a).
Among the GO molecular functions over-represented in the presence of K+ there are some linked to proteolysis, we found metallo-exopeptidase activity and aminopeptidase activity (S1 Table). Protease activity is a well established player in the pathogenesis of periodontal disease [19,20]. Consistent with low overall environmental levels of K+ in the environment, we observed an over-representation of potassium transport activities in the oral microbiome incubated without K+ added (Fig 2b, S1 Table).
We next determined whether the addition of K+ increased the synthesis of putative virulence factors. As a model we used hemolysins, which are recognized as potential virulence factors in a large number of anaerobic species [21].
In the case of whole dental plaque growing on plates with added K+, after 6 days incubation we observed hemolysis at all concentrations, with the 50mM concentration showing higher hemolytic activity, while lower concentrations and the control with no K+ added behaved similarly (Fig 3a). To show that the increase in hemolytic activity of the whole plaque at 50mM was due to changes in metabolic activities and not to changes in community composition, we characterized the microbial communities from the final plaque growing at the different concentrations of K+ from 3 different subjects (Fig 3b). The results of the phylogenetic composition based on 16s rRNA deep sequencing of those communities showed no differences between the communities growing at different concentrations of K+ and the control (Fig 3b and S2 Table). Additionally, we characterized the composition of the initial inoculum used in the experiments and the microbial communities from a periodontally healthy and from a sample with severe periodontitis. The initial inocula for all three patients analyzed was identical in total number of cells and after 6 days of incubation all plates from a single experiment had the same number of total cells growing (S3 Table). The microbial composition of the original inoculum from the different subjects was different as well as the final composition of the communities growing on plates (Fig 3b). However, those communities showed no differences between the communities growing at different concentrations of K+. Collectively, these results indicate that an increase in K+ in the environment leads to expression of genes associated with pathogenicity in the oral microbiome, which could be important in the process of dysbiosis, from commensal to pathogenic microbiome.
We observed an up-regulation of hemolysins in 45 different species, with the majority belonging to genera Prevotella and Streptococcus (S4 Table). Among the species presenting high up-regulation of hemolysins were Prevotella nigrescens and Streptococcus mitis. Increased numbers of P. nigrescens have been associated with severity of periodontitis [22] and in previous work we showed that S. mitis expresses a high number of putative virulence factors in periodontitis [10,11]. To determine whether these genera increase hemolysin expression in response to potassium, we assayed hemolytic activity in culture supernatants of P. nigrescens and S. mitis. We observed an increase in hemolytic activity in the supernatant of P. nigrescens and S. mitis growing in liquid media at 0.5mM and 5mM K+ added but an inhibitory effect at 50mM of K+ added to the media (Fig 4a and S4 Table). S. mitis had a lower hemolytic activity. S. mitis hemolysins tend to accumulate in the cytoplasm rather than in the extracellular environment [23]. These changes in hemolytic activity were not associated with different levels of growth. The final OD600 and CFU/mL of the different tubes used for analysis were not significantly different (S5 Table) indicating that activity was increased without a change in the total number of cells.
P. nigrescens expresses β-hemolytic activity when grown on blood agar with a peak of hemolytic activity on the fifth day of incubation [24]. After 6 days of incubation we observed β-hemolytic activity at all concentrations of K+ but with lower activity at 50mM K+ in P. nigrescens (Fig 4b and 4c). Interestingly, as described above, the range where the effect occurs is well defined and once a certain concentration threshold is exceeded the effect is repressed.
To test the effect of potassium on the host plaque interaction, we challenged a three-dimensional gingival multi-layered tissue model with cornified apical layers similar to in vivo gingival tissue (EpiGingival GIN-100, MatTek Corp.) with dental plaque. These three-dimensional tissue models have been used in a wide variety of studies and organs [25]. Tissue-engineered 3D culture systems of the oral mucosa provide an organizational complexity that lies between the culture of single cell types and organ cultures in vivo. We first confirmed that the histological morphology of the tissue used in the experiments was not altered by the different concentrations of K+ or the addition of bacteria. Haematoxylin-eosin stained sections showed a normal morphology of the gingival tissue models in both unchallenged and challenged tissues (S2 Fig). In the tissues that were challenged with dental plaque we observed that bacterial invasion was already occurring regardless of the addition of K+ (S3 Fig).
Cytokines are important markers of inflammation. To test the effects of potassium on the inflammatory response to plaque, we measured cytokine production in the presence and absence of plaque and increasing concentrations of potassium. We observed that K+ had a major effect on the expression profiles of the cytokines assessed. As shown in Fig 5a, the profiles of expression clustered as a function of K+ concentration, regardless of the presence or absence of dental plaque interacting with the tissue. Two cytokines, IL-6 and TNF-α, presented significant differences in their levels of expression associated with the levels of K+ (S6 and S7 Tables). IL-6 showed higher levels of expression at 0mM, 5mM and 50mM of K+ than at 100mM of K+ while TNF-α was significantly up-regulated at 50mM and 100mM of K+ added (Fig 5). Most of the other cytokines analyzed (IFN-γ, IL-17A, IL-1β and IL-10) did not change their pattern of expression significantly either with different K+ concentrations or with the addition of bacteria to the system (S4 Fig). One of the cytokines, the anti-inflammatory IL-4, was not detected under any of the conditions studied.
Interestingly, interaction analysis of plaque and K+ concentration showed that they indeed had an interacting effect on the values of TNF-α and IL-6 but not in the rest of cytokines values (S5 Fig). In case they did not interact we would expect a parallel plot for the lines as it was observed for rest of cytokines. To confirm that conclusion we fit a two-way ANOVA with an interaction term (see S1 Bioinformatic Analisys in Supplementary Information). Those results show evidence of significant interaction between plaque and potassium concentration in TNF-α and IL-6 response (S9 Table).
Human β-defensin-3 (hBD-3) is widely expressed in the oral cavity and exerts strong antibacterial and immunomodulatory activities [26]. hBD-3 plays an important role in periodontitis [27] and it is reduced in individuals with severe disease [28]. More importantly, the appropriate expression of hBD-3 peptide may contribute to the maintenance of periodontal homeostasis, possibly through its antimicrobial effect and promotion of adaptive immune responses [29].
The three-dimensional tissue model used in these experiments expresses hBD-3 in all layers except the stratum corneum, expresses hBD-1 weakly only in the apical layers, and does not express hBD-2 at all. Using immunohistochemistry we assessed the effect that K+ and bacterial plaque had on the levels of expression of hBD-3. Fig 6a shows representative examples of the results. Expression of hBD-3 was observed in all layers including the apical areas of the tissue but was more intense on the basal layers (Fig 6aii and 6aiii). K+ had a major effect on hBD-3 expression. The intensity of the signal was normalized by the signal obtained by DAPI, which represents an estimate of the number of cells. The addition of K+ by itself inhibited the production of hBD-3 regardless of the presence of plaque. Individually, plaque and potassium each reduced hBD-3 production to similar levels. Combined, plaque and potassium had an additive effect, significantly reducing hBD-3 production more than either treatment alone. (Fig 6b). We compared the statistical significance of those differences using a non-parametric analysis (Kruskall-Wallis correcting for multiple comparisons) and found that all values shown in Fig 6 were significantly different, except for the results with no plaque plus 50mM K+ and plaque without K+ added, and plaque plus 50mM of K+ and no plaque plus 5mM of K+ (S8 Table). These results indicate that K+ exerts an inhibitory effect on the production of hBD-3, which would clearly weaken the antimicrobial response of the gingival tissue in response to bacterial challenge.
ANOVA analysis revealed that there was not an interaction effect between plaque and potassium on hBD-3 production by the gingival epithelial (F = 0.4148, p = 0.661) (S10 Table, S6 Fig). The test for the effect of the presence of plaque shows a significant effect on the levels of hBD-3 (F = 31.3052, p<0.0001). Similarly, the test for the effect of potassium concentration (F = 23.1869, p<0.0001) indicates a significant effect on the levels of hBD-3.
Dysbiosis of the oral community is mediated by changes in composition and functional activities of the microbiome. In this paper, we show that K+ could be an important environmental signal in dysbiosis of the oral microbiome. We have shown that K+ induces an increase in pathogenicity of the oral microbiome, and at the same time inhibits mechanisms of defense such as production of hBD-3 that could lead to microbial immune subversion. These results represent a first step in unveiling the role that K+ could have in disease.
We are just beginning to understand the mechanisms of oral dysbiosis that lead to disease, where certain organisms can behave as 'keystone-pathogens' (e.g. P. gingivalis) orchestrating the activities of the rest of the biofilm to their advantage and leading to an aggressive bacterial attack on the gingival tissue [9]. However, the presence of these 'keystone-pathogens' does not necessarily explain why disease is initiated or progresses, since they can be detected in healthy individuals and non-progressing sites [30,31]. In addition, other environmental signals may be responsible for the initiation of the shift from commensal to dysbiotic communities. Our findings provide a first report of one of such key signals leading to dysbiosis. Despite the novelty of the current study, some limitations must be considered when interpreting the results. In our metatranscriptome and tissue-engineered 3D culture systems experiments we used plaque from only one human volunteer and other plaque samples may behave differently and we performed short term experiments that resemble an acute response rather that the chronic nature of periodontitis. In future studies it would be more appropriate to perform long-term steady-state experiments. However, to this date it is extremely difficult to maintain reproducible oral microbial communities that resemble the complexity of the oral microbiome for long periods of time [32–34]. There are models for oral microbial communities but the selection of the organisms used on those models is mainly based on how easy is to grow them than in the real importance of those organisms for the homeostasis of the microbial community. Moreover, even the addition of few new members would completely alter the expression profiles of the rest of members of the oral biofilm [35].
Using metatranscriptome analysis of the oral microbiome during periodontitis progression, we identified several metabolic signatures associated with periodontitis progression [11] and with severe periodontitis [10], being potassium ion transport one of the most significant in our analysis of GO biological processes. Although we can not discard the possibility that other ions could, and likely do have an effect on virulence, the fact that we only observed changes in expression of potassium ion transport in our in vivo studies at the time of disease progression and severe periodontitis seems to indicate that this particular ion acts as an important signal in the transition from health to dysbiosis [10,11]. Whether potassium exerts its effect directly or through an increase on osmolarity K+ has to be revealed in future studies.
Potassium, which plays an essential role in cellular homeostasis, is the most abundant ion in the cytoplasm of prokaryotic and eukaryotic cells. The internal concentration of K+ in a typical mammalian cell is 139mM [36]. Maintenance of high internal K+ concentration and consequently potassium uptake are of crucial physiological significance. Moreover, K+ acts as a cytoplasmic signaling molecule, activating and/or inducing enzymes and transport systems. The signal could be ionic strength or specifically K+ itself [37]. However, not until recently it has been shown that ion potassium channels in bacteria enable bacterial communication in biofilms and that potassium is the key signal in propagating the signal through a monospecific biofilm of Bacillus subtilis [38]. The natural environment in which the periodontal biofilm grows, the subgingival crevice, is in direct contact with the gingival crevicular fluid (GCF) rich in inflammatory mediators that are important in maintaining the homeostasis of the system. Levels of K+ in the GCF increase with severity of periodontal disease while levels of other ions such calcium remain stable [12]. In GCF from healthy patients, Kaslick et al. reported mean values of K+ of 10mM [13], while in severe periodontitis K+ concentrations of more than 20mM have been reported [13,14]. Although the source of such levels of K+ in GCF is not known it may come from host cells lysis, which could be sensed by the oral microbiome as a signal of tissue damage that in turn triggers up-regulation of genes involved in pathogenicity and induction of a local immune response, thus initiating a positive feedback loop in which lysis of host cells by microbial proteolytic activity or tissue immune responses lead to the release of more K+ into the GCF. Tissue irritation, caused for instance by the placement of a ligature in murine animal models, leads to disruption of the normal commensal microbiota [39]. Plaque accumulation on the ligature has been suggested as the primary reason for the initiation of periodontitis in those models, nonetheless release of K+ may also a role in inducing virulence in the oral microbiome.
In the present study we showed that an increase in K+ in the ex vivo model used to mimic the oral environment lead to a rapid metabolic response of the microbial community as a whole, increasing the activity of known putative periodontal pathogens. Using metatranscriptomic analysis of biofilms at high concentrations of potassium we show an up-regulation of community-wide virulence factors expression such as iron transport and motility. Virulence of pathogenic organisms is related to the availability of iron, therefore, microbial iron acquisition mechanisms are an important determinant of infection potential. In the major periodontopathogen Porphyromonas gingivalis iron limitation up-regulates the genes involved in iron uptake and its ability to invade host cells [40,41]. It is also well established that flagellated bacteria are abundant in samples from patients having periodontal disease [42]. Motility is not considered to be a classic virulence factor of bacteria. However, in other major periodontopathogen, Treponema denticola, motility is a key element for the virulence of the bacterium during disease progression [43,44]. Additionally, we have recently found that cobalamin biosynthesis and oligopeptide transport activities are associated with the progression of periodontitis [11].
The increase in virulence was experimentally confirmed by the increase in hemolytic activity of specific bacteria that had previously showed up-regulation of hemolysins in periodontitis progression [11] (P. nigrescens and S. mitis) as well as the whole oral biofilm. Interestingly, a high concentrations of potassium (50 mM K+) the isolated species P. nigrecens and S. mitis had the opposite behavior that the whole dental plaque. While hemolytic activity was inhibited in P. nigrescens and S. mitis at 50 mM K+ it was induced in the whole dental plaque biofilm sample, most likely due to the different behavior that bacteria show in a mixed biofilm and isolated from it [45].
Proteolysis was one of the community-wide activities observed as up-regulated in the presence of high concentration of K+, which has been shown to contribute to virulence, and most importantly in degradation of host extracellular matrix proteins that leads to a loss of the epithelial barrier in the oral cavity [46,47]. Our results confirm previous observations where dysbiotic activities were mainly driven by organisms that are generally considered commensals [10,11]. Thus in our previous study of periodontitis progression we observed that the vast majority of up-regulated putative virulence factors were expressed by organisms not consider periodontal pathogens [11]. Indeed, adding K+ increased the up-regulation of hemolysins and hemolytic activity in S. mitis, which under the right environmental conditions is capable of causing severe infectious diseases outside the oral cavity [48,49].
The short-term response, akin to an acute response, to the presence of K+ of a multilayered tissue model similar to in vivo gingival tissue was similarly rapid. We demonstrated that high levels of potassium modulated the response of a gingival epithelial three-dimensional (3D) model to the bacterial challenge altering the expression of cytokines. Oral keratinocytes express a variety of pro-inflammatory cytokines, including IL1-α, IL-1β, IL-6, IL-8 and TNF-α [50] after just a few hours of incubation. Among the host mediators produced after microbial recognition, innate immunity cytokines such as TNF-α, IL-1, and IL-6 were the first to have their roles in periodontal disease pathogenesis unraveled [51]. Our results showed a pro-inflammatory profile of cytokine expression associated with elevated concentrations of K+, while the anti-inflammatory cytokines IL-10 and IL-4 were not altered by the presence either K+ or bacteria. In fact IL-4 was not detected under any of the conditions assayed. IL-4 has been reported to possess a protective role in periodontal disease [52] and could mediate the remission or improvement of periodontal lesions [53]. The literature on the protective role for IL-10 is much stronger than for IL-4. Knockouts of IL-10 have rapid periodontal destruction in response to microbial challenge while knockouts of IL-4 do not produce the same effect [54] and have naturally-occurring IBD [55].
On the other hand, the profiles of expression of two assessed cytokines: TNF-α and IL-6, were altered by the concentration of K+ in the medium. TNF-α and IL-6 are major pro-inflammatory cytokines, although it has been suggested that IL-6, under certain conditions, could act also as anti-inflammatory [50,56]. Expression of TNF-α was up-regulated at high concentrations of K+ but not altered by the presence of bacteria, indicating that K+ by itself had a pro-inflammatory effect on the gingival tissue model. Data from human studies as well as from animal models clearly demonstrated that TNF-α plays a central role in inflammatory responses and the loss of connective tissue attachment [57]. TNF-α is present at high levels in both gingival crevicular fluid (GCF) and tissues in periodontitis, where positively correlates with matrix metalloproteinases and receptor activator of nuclear factor kappa-B ligand expression [57,58]. Moreover, TNF-α plays a key role on bone resorption by promoting osteoclastogenesis by stimulating RANKL expression and bone-resorbing osteoclasts exposed to permissive levels of RANKL [59–61].
IL-6 also showed significant changes under different K+ concentrations. Although IL-6 and TNF-alpha are normally thought of as pro-inflammatory cytokines, IL-6 followed a complete different pattern of expression than TNF-α, being expressed at higher levels at low concentrations of K+ This results seem to indicate that in our gingival epithelial model TNF-α and IL-6 are independently regulated. Previous studies have shown that in some instances IL-6 and TNF-alpha are coordinately regulated, while in other instances they can be regulated independently. Increased levels of IL-6 have been detected in the crevicular fluid of active sites compared with healthy sites of patients with refractory periodontitis [62] and exposure to lipopolysaccharide from the P. gingivalis induces elevated levels of IL-6 and TNF-α in primary gingival mouse cell lines [63]. Nonetheless, it has also been shown that IL-6 inhibits production of TNF-α in culture human monocytes, as a part of a TNF-α/IL-6 regulatory circuit [64].
Our findings showed that the presence of bacteria increased IL-6 expression regardless of K+ concentration. Expression of IL-6 is induced in epithelial cells following adhesion by organisms as diverse as E. coli and H. pylori [65]. More interesting is the fact that IL-6 can also be induced by bacterial invasion [66]. We observed invasion in all K+ assayed which could explain the induction of IL-6 when dental plaque was added to the tissue.
A central aspect of both homeostasis and dysbiosis in the oral cavity is the subversion of the immune response by the oral microbiome to overcome the host mechanisms of defense. Among them we find immunosuppression, complement subversion and blocking the production of human β‐defensins [67].
hBD-3 is a potent antibacterial peptide that is expressed at similar levels in samples of both healthy and inflamed gingival tissue [26]. We observed a decrease in hBD-3 expression when K+ was added as well as additional inhibition when dental plaque was present. It has been shown that the oral pathogen T. denticola suppresses the expression of hBD-3 in gingival epithelial cells but has no effect on the production of TNF-α, adding to the virulence of this bacterium and its role during the progression of periodontal inflammation [68]. In the presence of LPS, hBD-3 (but not hBD2), effectively inhibit TNF-α and IL-6 accumulation [69]. The suppression of hBD-3 by K+ and bacteria my favor biofilm growth and proliferation during periodontitis.
Despite the limitations of our study, these findings represent an important advance in our understanding of signaling during the initial stages of breakdown of host-microbiome homeostasis in the oral cavity, and may also be important in other polymicrobial inflammatory diseases such as IBDs where microbial dysbiosis is an essential factor in the evolution of pathology.
Our findings highlight the importance of ion potassium as a signal for dysbiosis in periodontitis. In the presence of high concentration of ion potassium we observed an increase in virulence of the whole microbial community, which agrees with our previous in vivo observations on severe periodontitis and progression of the disease. Moreover, we demonstrated the effect of potassium on the expression of virulence factors of isolated oral microorganisms including S. mitis, which is considered a commensal under normal conditions. Furthermore, pro-inflammatory cytokines were up-regulated as a response to the presence of K+ and bacteria and expression of the antimicrobial peptide hBD-3 was inhibited by both potassium and dental plaque. Future studies are needed to confirm our results on a periodontitis mouse model as well as to identify the mechanisms by which the oral biofilm senses the different levels of potassium in the environment
To assess the effect that K+ has on the oral microbiome we performed metatranscriptome analysis of its effect on dental plaque. Dental plaque from a periodontally healthy subject sample was resuspended in his own saliva, vortexed for 30 seconds and split in 3mL aliquots, each placed in a well of a 6 well, flat bottom Corning Costar cell culture plate. To 3 of the well containing the saliva/plaque suspension we added KCl to a final K+ concentration of 50mM and the other 3 well containing saliva/plaque suspension were used as controls. The culture plate containing the samples was incubated at 37°C for 3 hours under anaerobic conditions. Cells were collected by centrifugation at 10,000 x g for 5 minutes and RNA was extracted immediately for further analysis.
Detailed protocols for community RNA extraction, RNA amplification and Illumina Sequencing are described in Yost et al. [11].
Genomes of archaea and bacteria as well as their associated information were downloaded from the HOMD database server (http://www.homd.org/), the PATRIC ftp server (https://www.patricbrc.org/) [70] and the J. Craig Venter Institute (www.jcvi.org). A total of 524 genomes from 312 species of bacteria and 2 genomes from 1 archaea species were used in the analysis. Detailed explanation of genomes is reported in Yost et. al. [11].
Low-quality sequences were removed from the query files. Fast clipper and fastq quality filter from the Fastx-toolkit (http://hannonlab.cshl.edu/fastx_toolkit/) were used to remove short sequences with quality score >20 in >80% of the sequence. Cleaned files were then aligned against the bacterial/archaeal database using bowtie2. We generated a.gff file to map hits to different regions in the genomes of our database. Read counts from the SAM files were obtained using bedtools multicov from bedtools [71].
Counts from the mRNA libraries were used to determine their phylogenetic composition. Phylogenetic profiles of the metatranscriptomes were obtained using the latest version of Kraken [17]. Phylogenetic profiles were used to identify significant differences between active communities under the different conditions studied. We performed linear discriminant analysis (LDA) effect size (LEfSe) as proposed by Segata et al. [18] with default settings except that LDA threshold was raised to 3 to increase the stringency of the analysis.
To identify differentially expressed (DE) genes from the RNA libraries, we applied non-parametric tests to the normalized counts using NOISeqBio function of the R package 'NOISeq' with 'tmm' normalization, with batch and length correction and removing genes whose sum of hits across samples was lower than 10. We used a threshold value for significance of q = 0.95, which is equivalent to a FDR adjusted p-value of 0.05 [72].
To evaluate functional activities differentially represented we mapped the DE genes to Gene Ontology (GO) terms (http://www.geneontology.org/). GO terms for the different ORFs were obtained from the PATRIC database (https://www.patricbrc.org/). GO terms not present in the PATRIC database and whose annotation was obtained from the HOMD database or from the J. Craig Venter Institute were acquired using the program blast2GO under the default settings [73]. Enrichment analysis on these sets was performed using the R package 'GOseq', which accounts for biases due to over-detection of long and highly expressed transcripts [73]. Gene sets with ≤ 10 genes were excluded from analysis. We used the REVIGO web page [74] to summarize and remove redundant GO terms. Only GO terms with FDR adjusted p-value < 0.05 in the 'GOseq' analysis were used.
We used the EpiGingival GIN-100 (MatTek Corp.) a multilayered tissue model with the apical layers cornified, similar to in vivo gingival tissue. The tissues used in these experiments had more than 10 layers of cell with 300,000 to 500,000 cells per tissue. After arrival, the tissues were maintained overnight in a humidified incubator at 37°C and in the presence of 5% CO2 in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% (vol/vol) fetal bovine serum (GIBCO/BRL) and 1% (vol/vol) penicillin-streptomycin (GIBCO/BRL). Next day the tissues were washed 3 times with PBS to remove any traces of antibiotics and were reinoculated with DMEM without K+ (USBiological Life Sciences D9800-15) or antibiotics. We challenged the tissues with different K+ concentrations (0, 5, 50 and 100mM) and with bacteria from dental plaque and saliva. 4 different tissues were used for all concentrations. Supra and subgingival plaque from the same healthy volunteer used in all experiments was collected in saliva and diluted in DMEM -K to a McFarland standard #1 value (McFarland Standards Gibson Laboratories), which is equivalent to approximately 3 x 108CFU/mL. Based on the number of cells per tissue supplied by MatTek Corp. (see above) we inoculated with a multiplicity of infection (MOI) of 100, which had been used previously with success with other oral bacteria in invasion experiments [75].
At the end of the experiment the tissues were fixed with 4% formalin for 24 hours, paraffin embedded and cut into 5μm slices that were mounted on poly-lysine glass slides. Some slides were haematoxylin-eosin stained to check the integrity of the tissues. The rest were used for FISH analysis and immunohistochemistry as describe below.
Cytokine levels from 3 biological replicates of the medium surrounding the tissue cultures under the conditions described above were determined using MILLIPLEX MAP Human Cytokine/Chemokine Magnetic Bead Panel-Immunology Multiplex Assay (EMD Millipore, Billerica, MA, USA). Seven cytokines: IFN-γ, IL-10, IL-17A, IL-4, IL-6, IL-1β and TNF-α were measured. Samples were thawed at 4°C prior to assay and kept on ice throughout the assay procedures. Manufacturers’ protocols were followed for all panels, with a general protocol as follows. Reagents were prepared as per kit instructions. Assay plates (96-well) were loaded with assay buffer, standards, samples, and beads and then covered and incubated on plate shaker (500 rpm) overnight at 4°C. After primary incubation, plates were washed twice and then detection antibody cocktail was added to all wells; the plates were covered and left to incubate at room temperature for 1 hour on plate shaker. After one hour incubation, streptavidin-phycoerythrin fluorescent reporter was added to all wells, plates were covered and incubated for 30 minutes at room temperature on plate shaker. Plates were then washed twice and beads were resuspended in wash buffer, placed on shaker for 5 minutes, and then read on Bio-Plex200 following manufacturers’ specifications and using Bio-Plex Manager software v6.0.
Mounted slides were deparaffinized using standard protocols. Antigen retrieval was performed in 10mM Na Citrate pH 6 buffer in a microwave with an initial cycle of 2 minutes at 80% power and a final cycle of 8 minutes at 40% power. Slides were blocked on blocking solution (2% goat serum in PBS) for 1 hour, washed three times with PBS and incubated overnight with Anti hBD-3 antibody L3-18b-E1 (Abcam, Cambridge, MA, USA) at 4°C. As a negative control we used as a primary antibody a polyclonal rabbit IgG whose serum was obtained from naive (non-immunized) rabbits (RD Systems, Minneapolis, MN, USA). After incubation with the primary antibody slides were washed three times in PBS and incubated with goat anti-Rabbit IgG (H+L) secondary Antibody, Alexa Fluor 594 conjugate (LifeTechnologies). We finally counter stained the slides with DAPI (100ng/mL) for 10 minutes before analysis. Tissues were dried and mounted with Prolong Gold anti-fade (Life Technologies). Images were captured using an Inverted Widefield Fluorescence—Zeiss Cell Observer Z using a 10x and 20x objectives and analyzed using Fiji Software [76]. The number of histological sections analyzed are indicated in Fig 6 legend. Sections for each condition tested came from a single three-dimensional multilayered gingival tissue model with cornified apical layers (EpiGingival, MatTek Corporation). Because of the low magnification used for analysis we analyzed one field per section to avoid fields with wrinkles or creases on the slide that could interfere with fluorescence measurements. Files with.czi extension were open in Fiji with ‘Autoscale’, ‘Split channels’ and ‘Color mode = colorized’ options marked. Brightness and contrast of the images were auto-adjusted, merged (channels ‘red’ and ‘blue’) and converted to RGB and analyzed using ‘Color histogram’. Results for the color histogram gives values of intensity for the different channels. We used the ratio of red fluorescence (Alexa Fluor 594 from hBD-3 expression) and blue fluorescence (DAPI from the cells' nuclei) as an estimate of the levels of expression of hBD-3 in the tissues.
We performed FISH on the mounted slides with a universal probe for bacteria labeled with EUB probe (EUB388 5′-GCT GCC TCC CGT AGG AGT) [77] (Life Technologies). The hybridization oven was pre-warmed to 46°C and humidifying solution (20% formamide non-HiDi in water) was added to the hybridization chamber. Slides were placed in hybridization chamber and covered with 40μl of probe mixture (0.9 M NaCl, 0.02 M Tris pH 7.5, 0.01% SDS, 20% Hi-Di formamide and 2pmol/μl) on top of whole mount. The hybridization chamber was sealed with parafilm and samples were incubated at 46°C for 3 hours. After hybridization in fume hood excess of hybridization solution from the slides was drained and slides were washed in 50 ml of pre-warmed wash buffer (215mM NaCl, 20mM Tris pH 7.5, 5mM EDTA) to 48°C for 15 minutes in the hybridization oven. Finally, slides were dipped in ice-cold water followed by a final dip in 100% ethanol at room temperature. The slides were air-dried and mounted in Prolong Gold antifade (LifeTechnologies) before being observed under the microscope as described above.
DNA was extracted from the fresh plaque inoculum used in the plate hemolysis assays as well as from the cultures growing and showing hemolysis after 6 days of incubation as described below. DNA extraction was performed using Ultraclean Microbial DNA Isolation kit. (MoBio, Carlsbad, CA) following manufacturers’ specifications. We used HOMINGS (http://homings.forsyth.org/index2.html) for species-level identification of oral bacteria using 341F (5' ATGATACGGCGACCACCGAGATCTACACTATGGTAATTGTCCTACGGGAGGCAGCAG) and 806R (5' CAAGCAGAAGACGGCATACGAGATNNNNNNNNNNNNAGTCAGTCAGCCGGACTACHVG GTWTCTAAT) primers to amplify the V3-V4 region of the 16s rRNA gene. The underlined stretch of 12N are designated barcode sequences. HOMINGS uses species-specific, 16S rRNA-based oligonucleotide 'probes', designed to target oral species, as a database in a BLAST program ('ProbeSeq' for HOMINGS) to identify the frequency of oral bacterial targets.
Bacterial load was quantified by qPCR following the method described in Nadkarni et al. [78]. Total DNA was extracted as described above. Escherichia coli DNA was used as the standard for determining bacterial number by qPCR. Amplification and detection of DNA by real-time PCR were performed with iCycler (BioRad) using optical grade 96-well plates. Triplicate samples were used for the determination of DNA by qPCR. The PCR reaction was performed in a total volume of 20 μl using the TaqMan Universal PCR and Master Mix (Integrated DNA Technologies), containing 100 nM of each of the universal forward and reverse primers and the fluorogenic probe.
Two different hemolysis assays were performed to assess the activity of individual organisms and whole dental plaque. The first was a plate hemolysis assay that gave a visual assessment of the hemolytic activity on agar plates to which different amounts of K+ were added. The second was a quantitative assay that measures release of hemoglobin from erythrocytes due to hemolysis. Agar hemolysis assays: we used P. nigrescens ATCC 33563 as a model organism. P. nigrescens was grown O/N in Schaedler Anaerobic Broth (Oxoid, Thermo Scientific, Lenexa, KS) and 10μl of the culture were spotted on horse blood TSBY plates with 0, 0.5, 5 and 50 mM of K+ added. The plates were incubated at 37°C under anaerobic conditions and checked at 48 hours and 6 days. For the biofilm assay we used fresh collected dental plaque from the same healthy individual who was used for the rest of experiments and resuspended in 100μl Schaedler Broth (Oxoid, Thermo Scientific, Lenexa, KS). 10μl of the suspension were spotted on horse blood TSBY plates with 0, 0.5, 5 and 50 mM of K+ added. The plates were incubated at 37°C under anaerobic conditions and checked at 48 hours and 6 days.
Quantitative hemolysis assays: To quantitatively measure hemolytic activity we used supernatants from P. nigrescens ATCC 33563 and S. mitis NCTC 1226. We followed the protocol describe by Maltz and Graf [79] with minor modifications. Briefly, Schaedler Anaerobic Broth (Oxoid, Thermo Scientific, Lenexa, KS) diluted to 50% with water was pre-reduced in an anaerobic chamber for 48 h. Prior to pre-reduction the different concentrations of K+ were added (0, 0.5, 5 and 50mM K+). P. nigrescens cells were inoculated from TSBY plates to OD600 of 0.2 and grown O/N 37°C under anaerobic conditions. S. mitis was grown on Todd-Hewitt Yeast Broth (THY: Bacto Todd–Hewitt Broth supplemented with 0.2% yeast extract) diluted to 50% with a minimal streptococci medium [80]. THY/MM media was pre-reduced in anaerobic chamber for 48 hrs. As mentioned above, prior to pre-reduction the different concentrations of K+ were added (0, 0.5, 5 and 50mM K+). S. mitis cells from TSBY plate were resuspended in 1.0 ml THY/MM and 100μl were added to tubes containing 3.0 ml of the pre-reduced THY/MM to OD600 of 0.2 and grown O/N at 37°C under anaerobic conditions.
Horse red blood cells (Horse RBC; Northeast Laboratory Services) were washed 3 times in 1.0 ml PBS and resuspended in PBS to final concentration 10% v/v. The O/N cultures were spun (5 min, 7,500 rpm) and 250 μl of culture supernatant was added to 250 μl of washed erythrocytes and incubated for 3h at 37°C. Horse RBC were also incubated with 250 μl PBS (negative control) or ultra pure dH2O (positive control). Reactions aliquots were centrifuged and 100 μl of each biological replicate were analyzed for hemolytic activity at OD540 with a BioTek Synergy HT plate reader.
We used the R package 'agricolae' to perform the non-parametric multiple comparison Kruskal-Wallis analysis on our results. Shapiro test analysis of our results showed that they did not follow a normal distribution. FDR adjusted p-value were obtained by setting the 'p.adj' argument of the 'kruskal' function as “fdr”. A cut-off value of 0.05 was used to determine the significance of the results. We used a Two-way ANOVA interaction test in R to assess whether plaque and potassium had an additive or an interaction effect on levels of cytokines and hBD-3. The detailed protocol is described in the section ‘Two-way ANOVA interaction test in R’ in the S1 Bioinformatic Analysis file of the supplementary information.
|
10.1371/journal.pntd.0007747 | High specificity and sensitivity of Zika EDIII-based ELISA diagnosis highlighted by a large human reference panel | Zika virus (ZIKV) and Dengue virus (DENV) are often co-endemic. The high protein-sequence homology of flaviviruses renders IgG induced by and directed against them highly cross-reactive against their antigen(s), as observed on a large set of sera, leading to poorly reliable sero-diagnosis.
We selected Domain III of the ZIKV Envelope (ZEDIII) sequence, which is virus specific. This recombinant domain was expressed and purified for the specific detection of ZEDIII-induced IgG by ELISA from ZIKV-RT-PCR-positive, ZIKV-IgM-positive, flavivirus-positive but ZIKV-negative, or flavivirus-negative sera. We also assessed the reactivity of ZEDIII-specific human antibodies against EDIII of DENV serotype 4 (D4EDIII) as a specific control. Sera from ZEDIII-immunized mice were also tested.
Cross-reactivity of IgG from 5,600 sera against total inactivated DENV or ZIKV was high (71.0% [69.1; 72.2]), whereas the specificity and sensitivity calculated using a representative cohort (242 sera) reached 90% [84.0; 95.8] and 92% [84.5; 99.5], respectively, using a ZEDIII-based ELISA. Moreover, purified human IgG against D2EDIII or D4EDIII did not bind to ZEDIII and we observed no D4EDIII reactivity with ZIKV-induced mouse polyclonal IgGs.
We developed a ZEDIII-based ELISA that can discriminate between past or current DENV and ZIKV infections, allowing the detection of a serological scar from other flaviviruses. This could be used to confirm exposure of pregnant women or to follow the spread of an endemic disease.
| The serological detection of Zika virus (ZIKV) is a challenge, as ZIKV infection generally leads to an immune response with a high level of cross reactivity against related viruses, such as Dengue virus. Although seroneutralization assays are the gold-standard to address specificity, a rapid and cost-effective detection assay with good specificity and sensitivity could be used for first-line screening. We used a large cohort to define a set of human reference sera to validate an ELISA based on a recombinant ZIKV antigen. The assay showed 90% specificity and 92% sensitivity, providing a good basis for the development of diagnostic assays. Characterization of both DENV-EDIII-purified human and murine IgG induced by ZIKV infection or ZEDIII, respectively, confirmed the good specificity of the antigen.
| Zika virus (ZIKV) was isolated in 1947 in the Zika forest in Uganda [1] and has been responsible for sporadic cases for several decades. Epidemics in Yap State, Micronesia (2007) [2], French Polynesia (2013) [3], and more recently the Americas 2015 [4], have dramatically changed its status. The association of ZIKV infection with Guillain-Barré syndrome [5] and severe outcomes during pregnancy, including microcephaly in fetuses and neonates [6, 7], have been brought to light during the last outbreaks. ZIKV can be sexually transmitted. This is uncommon for flaviviruses, which are arthropod-borne viruses (arbovirus) [8]. Infective ZIKV particles have also been found in breast milk but whether neonatal infection or perinatal transmission can occur is unclear [9]. Due to the ZIKV threat, the World Health Organization (WHO) declared a Public Health Emergency of International Concern on the 1st February 2016 [10].
According to WHO recommendations, the diagnosis is based on detection of the ZIKV genome by real-time reverse transcription PCR, serology, and neutralization assays, such as plaque-reduction neutralization tests (PRNT) [11]. However, induced antibodies can show high cross-reactivity between antigens of the same family, as Flaviviruses are phylogenetically very close. Moreover, Dengue virus (DENV) and ZIKV can co-circulate [12]. In addition, cross-seroneutralization of DENV and ZIKV, described recently [13], further complicates the development of relevant target antigens for reliable serological diagnosis. Several immunodiagnostics based on IgM detection have been developed [14, 15]. However, most cases are asymptomatic [16] and ZIKV-IgG detection is relevant for dating or confirming infections, especially those of pregnant women, retrospective diagnosis for evaluating transmission intensity to decide on the use of a new prophylaxis, or confirming future ZIKV protection by vaccination.
Domain III of the ZIKV-envelope protein (EDIII) shares 29% amino-acid identity with DENV-EDIII and 90% of EDIII-antibodies (EDIII-Abs) elicited by ZIKV infection are virus specific [17]. ZIKV EDIII-Abs tested in ELISA do not bind to DENV-2 or West-Nile virus (WNV) -EDIII [18]. Studies in mice have shown that treatment with EDIII-Abs protect against ZIKV infection [19] by neutralizing the virus [18].
In this context, we produced a recombinant ZIKV-EDIII protein (ZEDIII) and assessed its recognition by IgG. We constructed four representative panels of reference sera, (1) ZIKV-RT-PCR-positive, (2) ZIKV-IgM-positive, (3) flavivirus-positive but ZIKV-negative, and (4) flavivirus-negative, from more than 5,000 sera serodiagnosed at the French National Reference Center for Arboviruses (NRC). The objective was to evaluate a ZEDIII-based ELISA for its specificity (ability to detect IgG raised after infection by ZIKV, but not other flaviviruses) and sensitivity (ability to detect ZIKV-specific IgG in the serum of a patient confirmed to be ZIKV-positive). We thus designed three experiments to evaluate: (1) the specificity and sensitivity of recognition by the IgG of cohort sera, (2) potential purified DENV-IgG cross-reactivity, and (3) the specificity of ZEDIII-induced IgG. The specificity of the assay represents its ability to not detect false-positive sera, that is positivity due to cross-reactive antibodies that are able to recognize a common antigen with the same affinity shared by various viruses.
As French National Reference center, all samples used in this study were send to the laboratory for arboviruses diagnosis including the specificity of the antibodies response against different flaviviruses, as Zika virus. For this study, samples were analyzed anonymously. All samples sent in the laboratory are associated with a file with clinical data, travel, date of symptoms and also the consent to used the end of tube for technic development or comparison of diagnosis technics.
All sera were submitted to the French National Reference Center for arboviruses (NRC, Marseille) for routine diagnosis and stored at −20°C in an anonymized biobank before testing. No specific sampling dedicated to the study was performed. There were no legal or ethical restrictions for sample use. Associated documents contained the following information: date of birth, gender, date, nature of sampling, place of stay and return date, symptom onset date, clinical symptoms, and clinical diagnosis results.
Ethical approval of the ZIFAG cohort was given by the “Comité de Protection des Personnes Sud-Mediterranée” corresponding to the “Etude descriptive prospective de la maladie a virus Zika au sein de la communaute de defense des Forces Armees en Guyane—ZIFAG” and was registered on 29 February 2016 as RCB: 2016-A00394-47. Written consent was obtained from all participants [20].
A total of 5,600 sera from the NRC Arbovirus serum bank were serodiagnosed by ELISA using inactivated virus as antigen. This group was selected to follow antibody cross-reactivity between ZIKV and DENV. Eighty-one sera were not tested for the ZIKV-IgM response.
A set of sera was selected (n = 242) from the 5,600 of the NRC Arbovirus serum bank, as well as a set from the ZIFAG cohort, and divided into four groups based on their genome detection, reaction against inactivated arboviruses, and epidemiological data (Table 1, SD1 Fig). The “ZIKV-RT-PCR-positive” group, from the ZIFAG one-year clinical follow-up cohort, is composed of 43 sera collected in French Guyana during the 2016 ZIKV outbreak [20, 21]. The ZIKV genome was detected by RT-PCR during the acute phase and tested sera were obtained during convalescence: 6 to 195 days post symptoms onset (DPSO). Eighty-four percent of this group (n = 36) was sampled during the first month post symptoms onset: two sera were obtained less than 10 days DPSO and 34 between 13 and 25 DPSO. The other sera (n = 7) were obtained between 35 and 195 DPSO [21]. The “ZIKV-IgM-positive” group was composed of 50 randomly selected sera, collected mainly from the Caribbean Islands or another DENV-endemic area during the 2015–2016 ZIKV epidemic (total n = 241), based on the following criteria: ZIKV-IgM-positive, DENV-IgM negative, and ZIKV-IgG positive, all tested sera neutralizing ZIKV in PRNT assays. “Flavivirus” sera (n = 99) group was selected to be flavivirus-positive but not ZIKV-positive (total n = 797). All were flavivirus-IgG positive and sampled between 2013 and 2014 in Guadeloupe, Martinique, or Saint Martin, DENV-endemic areas, where and when ZIKV was not circulating, assuming that these patients were not ZIKV infected. The “Negative” sera group was composed of 50 negative sera that were both IgM- and IgG-negative for DENV, WNV, Chikungunya virus, Encephalitis St Louis virus, ZIKV, Toscana virus, and Rift Valley fever virus. The ZIKV-IgM-positive, flavivirus and negative groups were representative of the entire parent groups in terms of their mean age and immune responses. We determined the sensitivity of the ZEDIII-based ELISA with the ZIKV-RT-PCR-positive group and compared it to that of the ZIKV-IgM-positive group, and the specificity to that of the flavivirus-positive group. The serology of the samples was tested upon their arrival to the arbovirus NRC and then stored at -20°C. The antibodies were stable over time and no difference in response was observed several months or years after freezing.
Sequences of the Asian ZIKV strain from the French Polynesia outbreak of 2013, DENV2, the most frequently detected serotype by the NRC for Arbovirus, and DENV4, the DENV strain phylogenetically closest to ZIKV (accession numbers AHZ13508, P09866, and M29095, respectively), were selected for EDIII production using the structure of the ZIKV envelope protein (PDB 5JHM). The ZEDIII, D2EDIII, and D4EDIII coding sequences, optimized for production in E.coli, were synthesized in fusion with a His-tag coding sequence at their 5’ end by GenScript. The sequences were then cloned into pET-24a or pET-19b plasmids (Novagen). The recombinant proteins were produced in E. coli T7 Iq Express (New England Biolabs) and purified under denaturing conditions prior to in vitro refolding according to a protocol described previously [22]. The purity, conformation, and homogeneity of folding of the protein were controlled by size-exclusion chromatography coupled to multi-angle light-scattering analysis (SEC-MALS) and Coomassie blue-stained SDS PAGE gels.
IgG were purified from pools of 20 sera from the ZIKV-positive group against ZEDIII, and 20 positive sera from the flavivirus-positive but ZIKV-negative groups against D2EDIII or D4EDIII (25 μL per sera for a total of 500 μL per pool). Antibody purification was carried out in two steps, first on a protein-G column and then on a column with immobilized recombinant EDIII. For the immobilization of EDIII, 1 mg of recombinant ZEDIII, D2EDIII, or D4EDIII in 2.5 mL phosphate-buffered saline (PBS) was covalently bound to an NHS-activated column following the manufacturer’s protocol (GE Healthcare). Four milligrams of protein-G-purified antibodies from EDIII positive sera in 4 mL were then loaded onto the EDIII-coupled columns. After washes with PBS, EDIII-specific antibodies were eluted from the column with 1 mL 0.1 M glycine (pH 2.7) and neutralized with 100 μL 1 M Tris (pH 9).
Vero cells were inoculated with 0.01 MOI ZIKV (African ZIKV strain, accession number ArB41644) or DENV2 (Martinique DENV2 98–703 strain of 1998, accession number AF208496) and grown in Dulbecco’s modified Eagle medium (DMEM) complemented with 2% heat-inactivated fetal calf serum (FCS) at 37°C in 5% CO2 for 3.5 (ZIKV) or 7 (DENV2) days. The time of growth depended on the virus. Culture supernatants were centrifuged, and viral particles precipitated with polyethylene glycol 6000 (PEG 6000) and NaCl. The precipitates were washed and resuspended in PBS/Hepes solution. This viral solution was inactivated with beta-propiolactone.
Six-to-eight-week old female AG129 mice, weighing approximately 20 g at the start of the study, were subcutaneously injected with 10 μg ZEDIII in 20 μL PBS on day 0 and received a boost on days 15 and 30. Blood was collected on day 46 via the retro-orbital route.
All procedures were in accordance with the guidelines set by the Noble Life Sciences (NLS) Animal Care and Use Committee. Noble is fully accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care International (AAALAC). Veterinary care for all lab animals was in accordance with the Public Health Service Policy, U.S. Dept. of Agriculture (USDA) and AAALAC International requirements. All mice were bred at B&K Universal Laboratories under specific pathogen-free conditions. All experiments were carried out at Integrated BioTherapeutics, Inc.
96-well plates (Maxisorp, NUNC-Immuno Plate) were coated with 200 ng/well of recombinant protein diluted in PBS and incubated over night at 4°C. After removing antigen, wells were blocked with blocking buffer (3% milk in PBS) for 1 h at 25°C. After two washing steps with washing buffer (1% Tween 20 in PBS), they were incubated with 100 μL diluted human serum samples (1:500), diluted purified human IgG (0.15 or 0.06 μg/mL), or sera from vaccinated mice diluted (1:25) in dilution buffer (3% milk and 0.1% Tween 20 in PBS), then incubated 1 h at 25°C. After four washing steps, the plates were incubated with horseradish peroxidase-conjugated anti–human IgG (1:10,000) (Jackson Laboratories, Immuno Research) or horseradish peroxidase-conjugated anti–mouse IgG (1:40,000) (Sigma-Aldrich) diluted in dilution buffer for 1 h at 25°C. After four washing steps, TMB (KPL, TMB microwell Peroxidase Substrate System) was added and the reaction stopped after 5 min with 50 μL 0.25M H2SO4 per well. The optical density (OD) was read at 450 nm and the final values, OD ratio (ODr), obtained by dividing the average OD of duplicate wells from that of the corresponding blank non-coated wells. The threshold of positivity was calculated following the equation: m + 3σ (m = mean, σ = standard deviation, α (error) = 1%) on the ODr of the negative-sera group. Self-recognition of human purified IgG ODr was normalized to 100% for each EDIII.
96-well plates (Maxisorp, NUNC-Immuno Plate) were coated with inactivated viruses diluted (1:200) in PBS and incubated over night at 4°C. After removing antigen, wells were blocked with blocking buffer (3% milk and 1% sodium azide in PBS) for 1 h at room temperature (RT) in the dark. After two washing steps with washing buffer (0.05% Tween 20 in PBS), they were incubated with 100 μL human serum samples diluted (1:500) in dilution buffer (3% milk, 0.05% Tween 20, and 1% sodium azide in PBS), then incubated 1 h at 41°C. After four washing steps, the wells were incubated with human secondary antibody, horseradish peroxidase-conjugated anti–human IgG (1:10 000) (Jackson Laboratories, Immuno Research), diluted in dilution buffer and incubated 1 h at 41°C. After four washing steps, TMB (KPL, TMB microwell Peroxidase Substrate System) was added and the reaction stopped after 5 min with 50 μL 0.25M H2SO4 per well. The optical density (OD) was read at 450 nm and final values were obtained by dividing the average OD of duplicate wells from that of the corresponding blank wells coated with negative antigen. The threshold of positivity was fixed to an ODr of 3.
IgG responses of human sera samples were tested against NS1 (Euroimmun kit) following the manufacturer’s protocol (Euroimmun). Ambiguous results were considered to be negative.
Statistical analyses were performed with Prism 6 (GraphPad). IgM and IgG responses to DENV and ZIKV were correlated and R-values determined by the Spearman rank correlation test with n = 5,600.
The IgG response data of the selected groups (ZIKV-RT-PCR-positive, n = 43; ZIKV-IgM-positive, n = 50; Flavivirus, n = 99; Negative, n = 50) were plotted showing the median, 25th, and 75th percentiles. The non-parametric Kruskal-Wallis test with Dunn’s correction was used to identify the differences between the medians of the IgG responses with two-tailed and alpha = 0.01 for each test. The two-tailed Z-test with alpha = 0.05 was used to compare the difference in sensitivity of the ZEDIII- and NS1-based ELISAs according to the group or cohort used. The overlap of the IC 95% was assessed to compare the difference of misdiagnoses between the ZEDIII- and NS1-based ELISAs. The non-parametric Wilcoxon rank-sum test was used to compare the difference of the mouse IgG responses to ZEDIII or D4EDIII.
The study was performed in compliance with the STAndards for Reporting of Diagnostic accuracy (STARD) statement [23].
IgM cross-reactivity [95% Confidence Interval] between related flaviviruses (Fig 1A, Table 2) represented 23.8% [20.4; 27.2] of the sera (n = 121) that reacted against at least one virus (n = 606). In contrast, IgG cross-reactivity [95% Confidence Interval] (Fig 1B) represented 71.0% [69.1; 72.2] of the sera (n = 1,485) that reacted against at least one virus (n = 2,091). This cross-reactivity was evaluated by calculating the Spearman r correlation (p < 0.0001; r(IgM) = 0.4191, r(IgG) = 0.7986). The r value for IgG indicates a strong correlation between the recognition of ZIKV and DENV (Table 1). We obtained similar results concerning the cross-reactivity of IgM (11.6% [8.3; 14.9]; n = 42) and IgG (79.7% [77.4; 82]; n = 971) against WNV and ZIKV (Table 3).
IgG from the sera of the ZIKV-positive group and the flavivirus-positive but ZIKV-negative group (colored dots, Fig 1B) cross reacted with DENV and ZIKV similarly to all 5,600 sera: 68.7% versus 71% of IgG, showing that these two panels are representative of the 5,600 sera (Table 1).
The ZEDIII amino-acid sequence shares 46.3% and 47.2% identity and 64.8% homology with the amino-acid sequences of D2EDIII and D4EDIII, respectively (Fig 2). The DENV2 and DENV4 EDIII amino-acid sequences share 61% identity and 78.1% homology. The purified EDIIIs analyzed by SEC-MALS and Coomassie blue-stained SDS Page gels showed the elution of a pure protein in a single and symmetric peak with an average molecular weight of 14.5 kDa and a polydispersity of 1.01 (Fig 3A and 3B). The presence and importance of the conformational epitopes for ZEDIII recognition was highlighted by the loss of recognition of ZIKV-positive sera against chemically and thermally denatured ZEDIII and a gain of recognition of flavivirus-positive but non ZIKV-positive or negative sera (Fig 3C).
We investigated the reactivity of IgG of the four sera groups against inactivated DENV, ZIKV, ZEDIII, and NS1 proteins (Fig 4 and [21]) by ELISA. The ZIKV-RT-PCR-positive, ZIKV-IgM-positive, and flavivirus groups recognized DENV with medians of 14.4, 16.6, and 12.3 ODr, respectively, with no statistical difference (K = 5.02, p = 0.0811), whereas no negative serum samples recognized DENV (median ODr of 1.1). The ZIKV-RT-PCR-positive, ZIKV-IgM-positive, and flavivirus groups recognized ZIKV (median ODr of 8.6, 7.3, and 2.8, respectively). The levels of recognition between the ZIKV-RT-PCR-positive or ZIKV-IgM-positive groups and flavivirus group were statistically different (Mean rank diff. = 89.61, p < 0.0001; Mean rank diff. = 64.98, p < 0.0001) but similar for the ZIKV-RT-PCR-positive and ZIKV-IgM-positive groups (Mean rank diff. = 24.63, p = 0.0993). The negative group did not recognize ZIKV (median ODr of 1.1). The difference between the ability of the ZIKV-RT-PCR-positive and flavivirus groups (median ODr of 2.2 and 1.1, respectively) or ZIK-IgM-positive and flavivirus groups (median ODr of 3.6 and 1.1, respectively) to recognize ZEDIII was highly significant (Mean rank diff. = 103.3, p < 0.0001, Mean rank diff. = 123.2, p < 0.0001, respectively), whereas there was no statistical difference between the flavivirus and negative sera (median ODr of 1.1 and 1.2; respectively, Mean rank diff. = -13.75, p > 0.9999) or the ZIKV-RT-PCR-positive and ZIKV-IgM-positive sera (median ODr of 2.2 [1.7; 4.0] and 3.6 [2.0; 8.5]; respectively, Mean rank diff. = -19.82, p > 0.9999). The median of the OD obtained for the ZIKV-RT-PCR-positive and ZIKV-IgM-positive sera against ZEDIII was 0.12 (IQR [0.09; 0.24] min = 0.06; max = 1.96) and 0.29 (IQR [0.17; 0.64] min = 0.07; max = 2.10) respectively, whereas the median OD obtained against the blank was 0.05 (IQR [0.05; 0.06] min = 0.05; max = 0.08) and 0.08 ([0.07; 0.10], min = 0.05, max = 0.23), respectively.
With a calculated positive threshold based on the negative-group values (m + 3σ) of 1.54 (m = 1.17 and σ = 0.13), and tolerating a 1% error rate, one serum (2%) of the negative sera was positive against ZEDIII and 49 (98%) were negative. The smallest OD value considered to be positive with this positive threshold was 0.085, with a blank of 0.055, and the highest OD value was 2.12. The sensitivity was calculated as the number of positive sera from the ZIKV-RT-PCR-positive or ZIKV-IgM-positive groups, of which 39 of 43 (90.7%) or 46 of 50 (92.0%), respectively, were positive against the ZEDIII-based assay (Table 4) and four (9.3% or 8.0%, respectively) did not react. The sensitivity [95% confidence interval] was thus 90.7% [82.0; 99.4] or 92.0% [84.5; 99.5], respectively. There was no difference in sensitivity between these two groups (Z-test, p = 0.2219). The specificity was calculated as the negative sera in the flavivirus group (n = 99) or in both the flavivirus group plus the negative group (n = 149). Ten sera (10.1%) of the flavivirus group or 11 sera (7.4%) of the flavivirus group plus the negative group were positive against ZEDIII and 89 or 138 were negative (89.9% and 92.6%, respectively). Thus, the specificity [95% confidence interval] was 89.9% [84.0; 95.8] or 92.6% [84.4; 100]. These groups were also tested against NS1 protein using the Euroimmun kit. We determined a specificity of 83.8% [76.8; 91.2] with the flavivirus group and up to 89.3% [84.3; 94.3] with the flavivirus group plus the negative group and a sensitivity of 88.4% [78.8; 98.0] with the ZIKV-RT-PCR-positive group and 96.0% [90.6; 100.0] with the ZIKV-IgM-positive group (Table 4). There was no statistical difference in sensitivity between these two groups (Z-test, p = 1.3535). The specificity obtained with the NRC ELISA against ZIKV was 46.5% [36.7; 56.3]; the sensitivity was 100% with the ZIKV-RT-PCR-positive group and 96.0% [90.6; 100] with ZIKV-IgM-positive group.
We compared the reliability of the ZEDIII- and NS1-based ELISA by calculating the positive predictive value (PPV), the false discovery rate (FDR), the negative predictive value (NPV), and the false omission rate (FOR) within a ‘hypothetical population’ with a given prevalence of 33.5% (50/149), composed of ZIKV-IgM-positive (sensitivity identical to that of the ZIKV-RT-PCR-positive group) and flavivirus groups (n = 149; Table 5). The likelihood of being positive when detected as positive (PPV) was 82.1% [72.1; 92.1] with the ZEDIII- and 75.0% [64.4; 85.6] with the NS1-based ELISA, whereas the likelihood of being negative when detected as negative (NPV) was 95.7% [91.6; 99.8] and 97.6% [94.3; 100] with the ZEDIII- and NS1-based assays, respectively. The likelihood of being positive when not detected (FOR) was 4.3% [0.2; 8.4] and 2.4% [0; 5.7] with the ZEDIII- and NS1-based ELISAs, respectively, and the likelihood of being negative when detected as positive (FDR) was 17.9% [7.9; 27.9] and 25.0% [14.4; 35.6] with the ZEDIII- and NS1-based ELISAs, respectively.
We further characterized the lack of cross reactivity of anti-EDIII IgG of the three sub-groups selected from the ZIKV-positive and flavivirus-positive panels (Table 6). IgG purified against D2EDIII and D4EDIII cross-reacted with D4EDIII and D2EDIII (39.7 ± 4.2% and 47.0 ± 4.2% respectively, Fig 5A). Conversely, D2EDIII and D4EDIII-purified IgG did not recognize ZEDIII. ZEDIII-purified IgG recognized ZEDIII well (ODr = 15.9).
We next addressed the immune response induced by the recombinant ZEDIII. The median ODr of the sera obtained after the immunization of eight mice by ZEDIII was 2.25 and 1.02 against ZEDIII and D4EDIII, respectively. The difference between the ability of the sera to recognize the two recombinant proteins was significant (W = 36, p = 0.0078). None of the five ZEDIII-positive sera from mice (n = 8) recognized D4EDIII (Fig 5B), with the sequence closest to that of ZEDIII (Fig 2).
To date, no broad antibody cross-reactivity study has been performed for flavivirus with a particular emphasis on ZIKV, despite the often high co-endemicity of flaviviruses. Our study included more than 5,000 samples, providing sufficient power to observe statistically significant cross-reactivity results and select sera for the design of reference groups of patients. First, we observed low IgM cross-reactivity and high IgG cross reactivity against inactivated ZIKV, DENV, and WNV. This is generally observed with flavivirus-positive human sera, resulting in low diagnostic reliability of IgG-based assays [14, 24]. The use of peptides or recombinant proteins can partially improve the specificity of IgG-based diagnostic assays. Indeed, ZEDIII is specifically recognized with high sensitivity by anti-ZIKV IgG and can distinguish between DENV and ZIKV in late phase or post-infection. IgM detection shows relatively good specificity (only 23.8% cross-reactivity) permitting the determination of the infecting flavivirus, as already shown [25]. However, the IgM response does not persist. Thus, we studied ZEDIII recognition by IgGs, the most persistent antibodies.
We divided the set of sera into four groups based on various characteristics that allowed us to confirm their membership in each group. First, we determined the sensitivity of our assay using the reference group, the ZIKV-RT-PCR-positive group, which recognized the ZEDIII domain with the same sensitivity (90.7% [82.0; 99.4]) as the selected ZIKV-IgM-positive group (92.0% [84.5; 99.5]), allowing us to validate the ZIKV-IgM-positive group as a ZIKV-positive group. Thus, IgM that recognized only the total inactivated ZIKV, but not DENV (56.3% of ZIKV-infected patients), were induced by ZIKV, allowing the diagnosis of a recent ZIKV infection. Second, IgG from the sera of “flavivirus” patients, selected from 2013 to 2014 in Guadeloupe and Martinique, recognized all flaviviruses where and when ZIKV was likely not circulating. The large size of our group and the rational to select the sera allowed us to assemble relevant panels of samples suitable to support a robust and accurate study of ZIKV IgG detection using recombinant ZEDIII antigen.
The specificity for ZEDIII reached 89.9% and sensitivity 92.0%, whereas IgG presented cross-reactivity against total inactivated viruses (71% of IgG recognized both DENV and ZIKV). This makes recombinant EDIII one of the most robust tools for diagnosis [14, 24]. Of note, most of the 10 false-positive sera from the flavivirus groups gave a value close to the positive threshold. The ODr of only one serum sample was 6.26. Based on denaturation experiments (Fig 3), correct conformational folding of the EDIII protein is required for specific recognition. Indeed, chemical and heat denaturation of the protein led to the loss of recognition in ELISA assays, likely due to the loss of conformational epitopes, thereby decreasing specificity.
Given the high DENV seroprevalence observed in the area from which the ZIKV-IgM-positive group (mean age = 40 years) originated (93.5% of the Caribbean island population (mean age = 38 years), [26]) it is likely that they were infected with DENV prior to ZIKV infection, although we have no way of knowing whether the antibodies that recognized ZEDIII were actually induced by DENV. We thus purified IgG against immobilized EDIII domains to discriminate between IgG induced by and directed against ZIKV from that induced by DENV but cross-reacting to both ZIKV and DENV. No IgG from the selected pools purified against D2EDIII or D4EDIII recognized ZEDIII. Such a lack of cross-reactivity underscores the high specificity of the immune response against this antigenic domain. It is likely that the anti-ZEDIII antibodies were induced by ZIKV infection and not DENV infection, as we were unable to observe recognition of ZEDIII by the sera of patients who had never been infected by ZIKV. Thus, the high sensitivity obtained with our assay is due to ZIKV-induced IgG recognition and not potential cross-reactive IgG. The binding experiments developed to validate the low cross-reactivity in the ZEDIII-based ELISA showed that IgG purified against D4EDIII partially recognized D2EDIII, and vice-versa. This result is not surprising, as the experimental conditions were designed to force cross-reactivity further than in a classical sero-diagnostic test with high IgG concentrations. Instead, the experiment validates ZEDIII as a relevant antigen for sero-diagnosis. We further validated the high specificity of the humoral immune response against EDIII by assessing the ability of ZEDIII-immunized mouse sera to recognize recombinant EDIII. The sera of ZEDIII-positive mice did not recognize D4EDIII. As the D4EDIII amino-acid sequence is that closest to ZEDIII, it is possible that the result would be the same with any other flavivirus EDIII. Finally, we designed experiments using DEDIII-purified IgG to show that the observed sensitivity and specificity were due to only ZIKV-IgG binding to ZEDIII. The epitopes present on the surface of DEDIII were only recognized by IgG that did not recognize ZEDIII. IgG induced by ZEDIII immunization recognized only epitopes present on the surface of ZEDIII. Thus, the epitopes recognized by ZEDIII-IgG are highly different from those carried by D4EDIIIand the EDIII epitopes are virus specific. We hypothesized that WNV induced IgG will not recognize ZEDIII.
To date, the most robust and specific ZIKV sero-diagnostic test is the PRNT, which is costly and time-consuming. ELISA is also commonly used but usually serves as a first-line screen prior to PRNT, due to its low specificity. Indeed, we showed that 71% of IgG diagnoses were not reliable with an ELISA performed with inactivated ZIKV and that 48% of flavivirus sera were ZIKV-cross-reactive. We assessed our ZEDIII-based-ELISA and the NS1-based-ELISA (Euroimmun Assay) on the selected sera of the four groups. The ZEDIII- and NS1-based-ELISAs for IgG were 89.9% to 92.6% and 83.8% to 89.3% specific (calculated with the flavivirus group (n = 99) or the flavivirus group plus the negative group (n = 149), respectively), and had a sensitivity of 92.0% and 96.0%, respectively (Table 4). The sensitivity determined for the NS1-based ELISA varies significantly, depending on the cohort used: from 70.7% with the cohort of Matheus et al. [21] to 88.4% (ZIKV-RT-PCR-positive group) or 92.0% (ZIKV-IgM-positive group) (Z-test between the ZIKV-RT-PCR-positive or ZIKV-IgM-positive groups and the cohort of Matheus et al., p = 2.5257 and p = 4.4080, respectively). With our assay, the sensitivity varied non-significantly from 90.7% (ZIKV-RT-PCR-positive group) to 92.0% (ZIKV-IgM-positive group). The sensitivity of the ZEDIII-based ELISA can thus be considered stable. The ZIKV-IgM-positive group sera were collected in a DENV-endemic area, where the ZIKV infection is a second flavivirus infection, whereas this infection was the first for 65.6% of the ZIKV-RT-PCR-positive group. Thus, the difference in sensitivity could be due to this immunological scar.
Specificity is an important criterion for flavivirus diagnosis, as flaviviruses can induce cross-reactive IgG and high DENV prevalence is observed in the ZIKV endemic area. The specificity of the ZEDIII-based-ELISA was in the same range as that of the commercial test: within a 33.5% prevalence population, the probability of misdiagnosis would be 9.4% for the ZEDIII-based ELISA versus 12.1% for the NS1-based assay, which was not significantly different when considering the IC 95% overlap. However, unlike the NS1-based ELISA, the sensitivity of our assay did not depend on the cohort tested. Relative to the NS1-based ELISA, our assay would lead to the misdiagnosis of 1.9% additional positive patients but 7.1% fewer negative patients. The recombinant refolded ZEDIII domain could thus be a good target for diagnosis. According to Matheus et al. [21] and our study (data not shown), IgG raised against ZIKV and ZEDIII were still present, with at least 50% of the maximum ODr, after 300 DPSO and up to several years after infection [27]. This minor limitation could allow us to carry out retrospective seroprevalence studies of the ZIKV outbreak, whereas detecting ZIKV-induced IgG with ZEDIII could be used to detect recent past infections in populations at risk. A specific diagnosis must also be reliable for pregnant women or those planning to become pregnant to indicate their ZIKV immunization status, and not necessarily at the first stages of putative ZIKV infection, to orient medical care, as ZIKV can cause miscarriage [28, 29].
Recombinant EDIII proteins have already been used as IgG-targets in ELISA and Microsphere ImmunoAssay to detect the immune response of horses and humans in seroprevalence studies [30–33] and, in particular, ZEDIII to link pathology to infection in humans [34, 35]. However, no study has yet clearly shown the limits, specificity, and sensitivity of a ZEDIII-based ELISA. Our study reinforces precedent EDIII findings and shows the use of recombinant flavivirus protein-based ELISAs, such as that based on ZEDIIIE, to be potentially reliable, offering opportunities to develop rapid, inexpensive, and specific first-line assays. Antibodies specific to the infecting flavivirus are able to protect against a new infection by neutralization [13, 36], whereas cross-reacting antibodies have been extensively linked to antibody-dependent enhancement, increasing viremia [37]. This phenomenon has been observed for both DENV- and ZIKV-induced antibodies [38]. Our results suggest that ZEDIII, by inducing a strong and specific immune response, could also be a safe model for the development of vaccines.
|
10.1371/journal.pntd.0002358 | Epidemiology of and Impact of Insecticide Spraying on Chagas Disease in Communities in the Bolivian Chaco | Chagas disease control campaigns relying upon residual insecticide spraying have been successful in many Southern American countries. However, in some areas, rapid reinfestation and recrudescence of transmission have occurred.
We conducted a cross-sectional survey in the Bolivian Chaco to evaluate prevalence of and risk factors for T. cruzi infection 11 years after two rounds of blanket insecticide application. We used a cubic B-spline model to estimate change in force of infection over time based on age-specific seroprevalence data. Overall T. cruzi seroprevalence was 51.7%. The prevalence was 19.8% among children 2–15, 72.7% among those 15–30 and 97.1% among participants older than 30 years. Based on the model, the estimated annual force of infection was 4.3% over the two years before the first blanket spray in 2000 and fell to 0.4% for 2001–2002. The estimated annual force of infection for 2004–2005, the 2 year period following the second blanket spray, was 4.6%. However, the 95% bootstrap confidence intervals overlap for all of these estimates. In a multivariable model, only sleeping in a structure with cracks in the walls (aOR = 2.35; 95% CI = 1.15–4.78), age and village of residence were associated with infection.
As in other areas in the Chaco, we found an extremely high prevalence of Chagas disease. Despite evidence that blanket insecticide application in 2000 may have decreased the force of infection, active transmission is ongoing. Continued spraying vigilance, infestation surveillance, and systematic household improvements are necessary to disrupt and sustain interruption of infection transmission.
| Despite significant gains in the reduction of the burden of Chagas disease in many South American countries, active transmission and significant burden remain in areas such as the Gran Chaco. High initial vector density, poor housing material, peridomestic infestation, insecticide resistance, and a lack of systematic insecticide spraying and vector surveillance have previously been incriminated for failure to interrupt and sustain interruption of transmission. We conducted a census, seroprevalence, and epidemiologic study in a rural area in the Bolivian Chaco. The prevalence of infection was almost 20% in children and over 80% in adults. We estimated the intensity of transmission over time based on infection prevalence by age. We found that after the first spray program, transmission appeared to fall transiently but then increased again quickly. Sleeping in a structure with cracks in the walls, age and village of residence were associated with increased likelihood of infection. These findings suggest that consistently repeated systematic spraying campaigns accompanied by housing improvements are necessary to interrupt and sustain interruption of vector-borne transmission.
| An estimated 8–10 million persons in Latin America are infected with Trypanosoma cruzi, the parasite responsible for Chagas disease [1], [2]. In the Western Hemisphere, Chagas disease is responsible for the highest burden of disability-adjusted life years lost among the neglected tropical diseases [3]. Without antitrypanosomal treatment, infection is life-long. Morbidity results largely from cardiomyopathy, which occurs in 20–30% of infected individuals [1]. T. cruzi is transmitted by more than 100 species of hematophagous triatomine insects. The parasite is deposited in the vector's feces during a blood meal and can enter the mammalian host through the bite site or intact mucus membranes.
In the 1960s, Chagas disease control programs were initiated in Brazil, Argentina, Chile and other South American countries. Efforts in southern South America were consolidated in 1991 in the first coordinated regional effort, the Southern Cone Initiative (SCI) [4]. The two pillars of the SCI strategy are prevention of transfusional T. cruzi transmission through serological screening of blood donations and domestic vector elimination through residual pyrethroid insecticide application and housing improvement. Insecticide campaigns begin with an attack phase in which all houses and peridomestic structures in an endemic community area are sprayed once or twice, followed by surveillance for residual vector foci and reinfestation with focal spraying of affected houses [5]. While housing improvement programs have been limited, insecticide spray campaigns were widely implemented throughout the region [4], [6]. The SCI has led to a marked decrease in the geographic range of T. infestans and interruption of transmission by this vector in Chile, Uruguay, Brazil and parts of Argentina [6]–[8]. In Brazil, the seroprevalence in children, used as a proxy for recent transmission, fell from 7.9% in 1990 to 0.1% in 2008 [9], [10]. However, the Gran Chaco, a 1.3 million km2 ecological zone shared among Bolivia, Argentina and Paraguay is an exception to this success story [8], [11]. In this region, reports of rapid reinfestation after spray campaigns, emergence of insecticide resistance and sylvatic T. infestans populations challenge the strategy of the SCI [5], [12]–[15]. We conducted an epidemiological study in an area of the Bolivian Chaco with serologic evidence of recent transmission despite control efforts. Our objectives were to describe patterns of and risk factors for T. cruzi infection, and to examine age-specific prevalence to estimate force of infection as an indicator of transmission over time through the use of catalytic models.
The protocol was approved by the Institutional Review Boards of the Centers for Disease Control and Prevention (Atlanta, GA), Asociación Benéfica Proyectos en Informatica, Salud, Medicina y Agricultura (Lima, Peru) and Hospital Universitario Japones (Santa Cruz, Bolivia). All participants 18 years of age or older provided written informed consent. A parent or guardian provided written informed parental consent for all children aged 2–17 years, and in addition, all children 7–17 years signed a simply-worded assent.
The Eiti health sector (19°43′52.4994″S, 63°23′9.4812″W, altitude 800 m) is a catchment area composed of 18 villages with a total estimated population of 8320 persons, located in Gutierrez municipality, Cordillera province, Santa Cruz department. The local ecology is sub-humid dry Chaco [8], [16]. Average temperatures range from 16.0–29.1°C, but drop to 5–10°C between the months of May to September [16]. Average annual rainfall is 860 cm, with a rainy season from November to April. The population is almost exclusively of the Native American Guaraní ethnicity, and the local economy is based on subsistence farming and animal husbandry. Houses were predominantly constructed of mud and sticks (tabique) or adobe, with packed dirt floors, and straw or corrugated metal roofs; each household comprised 1 to 4 separate structures, usually of one or two rooms. There was no wired electricity and no sewage system. The major water sources were small ponds (atajados) in which animals also bathe and drink, or shared or private outdoor taps.
The study area, the Eiti health sector, was chosen based on Ministry of Health Chagas control program data suggesting high likelihood of active T. cruzi transmission, including infection prevalence >15% in a sample of children tested in 2006 and high rates of household vector infestation (personal communication, N. Suarez and R. Vargas, Santa Cruz Chagas Disease Control Program, Ministry of Health, 2011). Within the Eiti health sector, we performed a census of seven neighboring communities chosen non-randomly based on size, relative lack of previous interventions, and proximity to our laboratory. The first systematic spray campaign targeting domestic T. infestans began in early 2000. Blanket spraying with alphacypermethrin 20% was conducted in 2000 and 2003. From 2005 to 2009, focal spraying of infested houses was conducted; no systematic spraying against triatomines was performed from 2009 to the time of the study. Prior to the 2000 spray campaign, reported household infestation rates ranged from 81.6–100% in the study villages, with an average of 94.3% (based on the reported village level infestation rates weighted by the number of houses in each village). Prior to the 2003 blanket spray, the weighted average infestation rate was 6.9%. Subsequently, the weighted average infestation rates were 3.0%, 18.4% and 57.7% in 2005, 2006 and 2008 respectively.
The field team attended community meetings in order to introduce, explain, and answer questions about the study, and to request permission to work in the area. Data were collected from July 2011 to May 2012. Households were defined as all structures that were utilized by a group of people who ate together; the head of household was determined by a consensus of the inhabitants of the household. The target sample size was 2000 based on anticipated T. cruzi infection and cardiac morbidity prevalence. All households in the study communities were invited to participate in the census. After obtaining a detailed household roster, the epidemiologic survey was performed, in which the interviewee provided data on the level of education of the head of household, ownership of animals, presence of electricity, and the most recent year in which the house had been sprayed with insecticide. A sleeping structure was defined as a structure in which a person sleeps, irrespective of whether the structure served another purpose. Sleeping structures that share a common entrance were considered a single sleeping structure, while those that did not were considered separate sleeping structures regardless of whether they shared a common wall. For each sleeping structure within a household, observations were recorded regarding the construction material, condition, such as the presence of cracks large enough for vectors to enter, presence of vector fecal traces or exuvia, and traces of lime-based whitewash on the outer walls of the house. Some of the whitewash was provided as part of an intervention program by Caritas, a non-governmental organization, and was sometimes, but not always, accompanied by other housing improvements such as wall plastering and/or substitution of tin roofs for those of thatch.
After census and epidemiologic data were collected, trained study nurses visited houses to collect venous blood specimens (5 cc; 3 cc for children 2–5 years old) from all consenting participants 2 years of age or older. Samples were maintained on ice packs and transported within 4 hours to the laboratory of the Hospital Municipal Camiri. Serum and cellular portions were separated by centrifugation, and stored at −20°C until they could be transported to the laboratory at Hospital Universitario Japones in Santa Cruz. All specimens were tested for T. cruzi antibodies by the indirect hemagglutination test (IHA; Chagas Polychaco kit, Lemos Laboratories) and Chagatest whole parasite lysate enzyme-linked immunosorbent assay (EIA; Wiener Laboratories, Rosario, Argentina) following the manufacturers' instructions. The IHA was tested in serial dilutions; specimens detectable at 1∶16 dilution were considered to have positive results. Discordant specimens were tested using the Chagatest Recombinante 3.0 ELISA (Wiener Laboratories, Rosario, Argentina). Participants with positive results by at least two assays were considered to have confirmed T. cruzi infection [17].
Our analytic approach was to use cross-sectional seroprevalence data to model the population age-specific point seroprevalence using a catalytic model function as described below. After identifying the “best-fit” catalytic model for our data, we applied a force of infection model to this, which we used to estimate the incidence within the population over time (calendar year), taking into account a number of assumptions as described below.
Serologic test agreement was calculated as the proportion of concordance between the two tests. Differences in seropositivity between the census population and those who participated in the epidemiological survey were tested by chi-square and Wilcoxon rank-sum tests. T. cruzi infection prevalence rates by age group were calculated as means, and an age-adjusted prevalence was calculated by applying age-specific rates among those tested to the total population for whom age data were available in the census.
We used catalytic modeling from the cross-sectional seroprevalence data to derive age-specific point seroprevalence within our population as a function of age [18]. We tested two parametric catalytic models, constant and Weibull, and a non-parametric cubic B-spline catalytic model to find the best fit to our data [19]. For all models, we assumed (1) no reversion to negative serology (i.e., seropositive individuals will never again be susceptible); (2) the entire seronegative population was susceptible and exposed to the same degree of risk; (3) mortality due to Chagas disease was negligible compared to all-cause mortality and could be ignored; and (4) there was no in- or out-migration other than births and deaths. The constant model also assumed time homogeneity, i.e., that the force of infection remained constant over the study interval, and there were no changes over time in infectivity (infectiousness of the vector and susceptibility to infection of the host remained the same) or transmissibility (including risk factor reduction efforts such as housing improvements) [20]. The Weibull and spline models relaxed the assumption of time homogeneity, as described below.
Model fit statistics and residual plots were used to evaluate which model best fit the data [18]. In the parametric constant and Weibull models, the generated change in prevalence by age cohort is described by a single equation [18]. The constant model has no inflection points within the curve, while the Weibull model had a single inflection point. In the spline model, the participants are divided into multiple cohorts based on consecutive age intervals and the model is fit with a piecewise function using local, smoothed polynomials. The piecewise functions are then connected end-to-end to allow for interval-specific changes in prevalence by age cohorts. The points where the polynomials join are known as knots [21].
Each model was fit to data from all participants for whom age and infection status data were available. In order to test the effect of spray campaigns on transmission, knots were introduced into the B-spline model for the age cohorts whose birth years coincided with the blanket sprays. Additional knots were placed at the minimum and maximum ages, 2 and 91, respectively. We used exploratory data analysis to determine whether and when adding additional knots improved model fit. We used a GLM with a complementary log-log link function in all catalytic models. Thus, our model had the formwhere π(age) is the disease prevalence as a function of age, S(age) is the linear predictor, and i denotes each participant (i = 1, …, 1545). With the B-spline basis functions, the linear predictor becomesThe knots are ξj with boundary knots of ξ0 = 2 and ξ4 = 91, the minimum and maximum ages. I is an indicator function which equals one if the age falls between ξj-1 and ξj and is zero otherwise. Finally, π(age) is continuous by forcing S(ξj) = S(ξj+1), S′(ξj) = S′(ξj+1), and S″(ξj) = S″(ξj+1) for the non-boundary knots.
Model fit was assessed by the Akaike information criterion (AIC) [22] using rescaled AIC cutoffs [23] and Pearson's χ2 testing, with the cutoff for significance for the latter at p<0.05. The best-fit model was used to determine force of infection. Confidence limits for age-specific seroprevalence were calculated by bootstrap resampling the data 10,000 times with replacement.
Force of infection is defined as the proportion of the susceptible population that becomes infected over a given period of time, which we use as an estimate of incidence. In order to calculate this, we used the following function to derive our force of infection, ℓ,where π(age) is the disease prevalence as a function of age. This function assumes:
The resultant curve estimates the force of infection experienced by the entire susceptible population in by calendar year. Bootstrap 95% confidence intervals were computed for the force of infection at calendar yearby resampling the data 10,000 times with replacement., Any negative incidence estimates as a result of the resampling were truncated at zero [20]. The presented 95% confidence intervals thus represent the set of the linked age-specific confidence intervals. Sample sizes were insufficient to model force of infection at the village level.
We used Pearson's χ2 tests and p-values to test for differences among the variables of interest between villages. To reduce temporal bias between disease acquisition and risk factor presence, we limited the evaluation of associations with risk of T. cruzi infection to the subset of our population aged 2–15 years. We constructed logistic regression models with generalized estimating equations to account for clustering at the level of the sleeping structure [24]. Potential risk factors were evaluated in univariable and trivariable models controlling for village and age. Variables to be tested in the full multivariable model (accounting for clustering and all factors in the model) were chosen based on biological plausibility and the change in AIC value [23] and were evaluated for confounding, interaction, and collinearity.
Catalytic and force of infection model analyses were performed in R (R Core Development Team, version 2.13.0, Vienna, Austria). SAS (version 9.3, SAS Institute Inc., Cary, NC) was used for logistic regression models. All statistical tests used the p<0.05 level of significance.
Of the 2233 persons recorded during the census, 40 individuals had neither age data nor serum specimens and were excluded from further analyses. Of the remaining 2193 individuals, 107 were younger than 2 years and 541 others declined participation or were absent during multiple house visits. The 1545 serosurvey participants were more likely to be female (54.9% vs 38.5%, p<0.0001) and younger (median age 15 [Interquartile range (IQR) 9–34] vs 20.5 [IQR 12–35], p<0.0001) than the 541 eligible non-participants (Table 1). Participants were more likely to reside in Eiti (30.2% vs 20.7%, p<0.0001) or Paja Colorada (7.6% vs 4.4%, p = 0.01), and less likely to be from Itapicoe (14.7% vs 23.3%, p<0.0001) or El Cruce (17.9% vs 22.3%, p = 0.025) than non-participants.
Of the 1545 individuals tested for T. cruzi infection, concordance between IHA and EIA was 95.9%. Concordance between the tests in the subset of 2–15 year old children was 97.0%. The T. cruzi seroprevalence among serosurvey participants was 51.7%. The prevalence increased from 19.8% among children 2–15 years old to 72.7% among those 15–30 years old and 97.1% among participants older than 30 years. The age-adjusted prevalence, based on the census population, was 54.1%. The infection prevalence among children 2–15 years old varied significantly by village, ranging from 11.2% in Eiti to 38.7% in Karavaicho (Table 2). There was no difference in prevalence by sex.
Several other variables differed significantly by village, including the educational level of the head of household, presence of a protected water source, access to a latrine, and housing materials and conditions (Table 2). In Karavaicho, Sinai and Guasuanti, the majority of houses were built of tabique (mud over a stick frame), whereas the majority of houses in other villages were of adobe brick. Traces of whitewash were observed on the walls of 36% of houses overall, but this finding was much more common in some villages (Eiti, Guasuanti, Sinai) than in others (Karavaicho, Paja Colorada, Itapicoe). Cracks in the house walls were much more common in villages where tabique houses predominated than in the other villages.
All three catalytic models demonstrated a steep increase in infection prevalence from age 2 to 30 years; nearly all participants 30 years or older had T. cruzi infection (Figure 1). The cubic B-spline catalytic model (AIC = 1212) had better fit statistics than either the constant model (AIC = 1349; χ2 = 149.2; df = 6; P = <0.0001) or the Weibull model (AIC = 1227; χ2 = 25.1; df = 5; P = 0.0001). The spline model (Figure 2) was therefore used as the basis for the force of infection analysis.
The force of infection curve was unstable for earlier calendar years due to the small number of susceptible participants in the denominator of the catalytic model, so the analysis focused on the years 1996–2011, corresponding to the birth years of the children aged 2–15 whose data were included in the risk factor analysis and including the blanket spray campaigns in 2000 and 2003. In the exploratory data analysis, we found that an additional knot corresponding to the year of birth of 25-year-olds improved our model fit. Although this knot does not to our knowledge correspond to the year of any intervention, it does suggest a separation of the seroprevalence among persons older than 11 years into one age range (12–24 years) in which prevalence increased with age and another (25 and older) in which nearly all participants were already seropositive.
The estimated mean force of infection over the time period 1996–2008 was 3.6% per year (Figure 3). Over the two years immediately before the first blanket spray in 2000, the estimated force of infection was 4.3% per year. For the 2 year interval immediately following the first blanket spray, 2001–2002, the estimated force of infection was 0.4% per year. The estimated annual force of infection for 2004–2005, the 2 year period following the second blanket spray, was 4.6%. However, the 95% bootstrap confidence intervals overlap for these estimates, indicating that the differences did not reach statistical significance.
In the model adjusted for village, we found a 16% increase in the odds of T. cruzi infection with each successive year of life (Table 3). In the age-adjusted model, children from Itapicoe had more than twice the odds, while children in El Cruce and Karavaicho had more than four times the odds of infection compared to those living in Eiti. Because village of residence and age were strongly associated with T. cruzi infection, all subsequent analyses were adjusted for these variables.
Of 444 sleeping structures in 377 households, 348 structures in 313 households were used by children aged 2–15; serological data were available for 808 children sleeping in 307 structures in 278 households. There was at least one T. cruzi-infected child in 75 structures (24.4%) in 72 households. In models adjusted for age and village, sleeping in a structure with traces of whitewash was associated with 40% decrease, whereas observed cracks in the walls and evidence of vector infestation were associated with approximately two-fold increase in the odds of infection. Exploratory analyses suggested that evidence of vector infestation and cracks in the walls were collinear variables. In addition to age and village, only the presence of cracks in the walls remained significantly associated with T. cruzi infection in the multivariable model (Table 4).
The Gran Chaco has the highest prevalence of Chagas disease in the world [8], [11]. In common with other investigators working in the Bolivian Chaco [25], we found an extraordinarily high T. cruzi prevalence, with close to universal infection among adults over 30 years old. Even more disturbing, however, is the fact that nearly 20% of children were infected. Although some study children likely had infection acquired via congenital transmission, the steep rise in prevalence between age 2 and 15 years suggests that vector-borne transmission was responsible for most infections in this age group. Based on our models, the estimated force of infection was above 2.5% per year for most of the past decade, despite major insecticide spray campaigns in 2000 and 2003. Evidence of ongoing transmission in the Chaco poses an urgent challenge to the effort by the Southern Cone Initiative to eliminate T. cruzi transmission by domestic T. infestans [8], [26].
We used catalytic models to derive the estimated force of infection from age-specific seroprevalence data. Similar models have been used to estimate changes in transmission of T. cruzi [27]–[31] and other infections [18], [32]–[34]. The spline model allowed us to include inflection points or knots at the time of the blanket spray programs. The model indicates that transmission was likely already decreasing prior to 2000, possibly as a result of secular changes or activities of local organizations promoting housing improvement. The force of infection dramatically declined in 2000 after the first blanket spray, falling below 1%, but paradoxically increased in each year after the second blanket spray. This suggests that the first blanket spray may have been quite effective, whereas the second blanket spray appears to have had little impact. However, the confidence intervals around these force of infection estimates overlap across all years in this range, suggesting that any impact of blanket spraying was ephemeral. Bolivian Ministry of Health data showed swift reinfestation after the second blanket spray campaign, increasing from 3% of households in 2003 to 58% in 2008 (personal communication, N. Suarez and R. Vargas, Santa Cruz Chagas Disease Control Program, Ministry of Health, 2011), possibly explaining the increasing force of infection estimates over this same time period.
In our study site, several of these factors may have impeded the effectiveness and durable impact of the spray campaigns, including high vector density, vulnerable housing materials, highly infested peridomestic areas, and diminishing resources for intensive surveillance and focal spraying. Although pyrethroid resistance has been reported in T. infestans in other parts of the Bolivian Chaco [15], [35], no frank resistance was found in preliminary testing of vectors from our study site (P. Marcet, unpublished data, 2011).
The insecticidal effect of pyrethroids on a house wall is estimated to last from 3 to 9 months, and depends on the type and condition of the wall surface [14]. Nearly all the houses in our study area were of tabique (mud over a stick frame) or adobe. Neither is the ideal surface for insecticide application and tabique is particularly problematic. Moreover, cracks in the walls of sleeping structures, which provide a refuge for triatomines to avoid contact with insecticide, were associated with an elevated risk of T. cruzi in children who slept there. Cracks in the walls were also associated with increased infection risk in children in Argentina and Peru [29], [36]. Studies in diverse sites have demonstrated increased risk of vector infestation and/or human T. cruzi infection with adobe or mud walls, and decreased risk with stucco or plastered walls [36]–[39].
In the Paraguayan Chaco, investigators found mortality of T. infestans nymphs exposed on mud walls was only 45%, 25% and 0% at 1, 3 and 6 months after insecticide application [14]. A coat of lime was found to increase mortality to 57.5% at 1 and 3 months, but at 6 months, vector mortality was 0% [14]. The improved effectiveness of insecticide applied on lime-painted mud walls may be relevant to our finding that children in whitewashed houses had a lower risk of T. cruzi infection. However, the brevity of the entomological effect in the bioassay [14] is consistent with the rapid reinfestation documented in Chagas disease control program data in our study area after 2005. In a study in the Argentine Chaco, a single round of spraying drove infestation rates from close to 100% to undetectable levels, but reinfestation was nearly complete within 5–6 years, attributed to lack of systematic surveillance for reinfestation and inadequate targeted spraying [8], [29], [40]. In that study, the average incidence during a 3-year period beginning four years after the initial blanket spray campaign was 4.3% [29], comparable to our average estimated force of infection of 4.6% in 2004–2006.
Domestic animals, especially chickens and dogs, were found in more than 90% of the houses in our study area. Animal enclosures in the peridomiciliary area are thought to be a frequent source of residual colonies leading to house reinfestation [41]. Peridomestic sites such as chicken coops and pig sties can sustain large vector colonies, are generally of unfinished materials on which insecticide is less effective, and surveillance is insensitive because of the multiplicity of vector refuges [41]–[44]. Mammals may also act as T. cruzi reservoir hosts. Dogs appear to be the most important T. cruzi infection reservoir in the Argentine study communities mentioned above, with risk for children increasing with the number of infected dogs in the household [29], [45], [46]. In Peru, children who slept in a bedroom with dogs or cats had a significantly increased risk of T. cruzi infection compared to those without animals in the bedroom [36]. In our study area, the lack of association between disease status and ownership of dogs or other animals may be due to the ubiquitous ownership of animals, and the fact that dogs freely migrated from household to household, and did not sleep indoors. Data on prevalence of infection in dogs would be useful to further assess their possible role as infection reservoirs.
Age and village of residence were the strongest predictors of T. cruzi infection risk in our data. In trivariable models that included these two variables, the only factors that retained statistical significance were traces of vector feces indicating recent or current infestation, cracks in the house walls and whitewash on the house walls. We found no association with socioeconomic indicators or the material of the sleeping structure, factors noted elsewhere to be associated with infection [5], [29], [39], [47], [48]. This may be due to the relative homogeneity in housing structures as many were of vulnerable materials and in poor condition. The strength of association with village of residence suggests that this variable may be acting as a proxy for multiple unmeasured factors.
Our force of infection model requires a number of assumptions. First, the model assumes that all uninfected persons in the population are equally susceptible, implying no difference in immunologic susceptibility or risk factor exposure. Chagas disease, unlike diseases such as malaria which are acquired and cleared repeatedly in persons living in endemic areas, causes lifelong infection and seropositivity without reversion, and so there is no evidence of naturally acquired immunity. The primary risk factor for disease acquisition is sleeping in a structure infested with infected vectors; in our study site housing structures and conditions were fairly similar. The next assumption is that the difference in seroprevalence between two age cohorts separated by one year is the difference in disease acquired by virtue of living that additional year. Finally, the model assumes that disease incidence is independent of age in any year, meaning that the entire population experiences the same force of transmission in a given year. If one already accepts that all uninfected persons are equally susceptible, then it would follow that unless there is a differential in vector affinity to a specific age, that force of transmission is age-independent.
The limitations of our study include the cross-sectional design, which precludes our ability to establish temporality of risk factors with disease acquisition. We cannot say with certainty that a child was sleeping in their current sleeping structure at the time of infection, although in general there was little movement of children between households. As we did not collect systematic data regarding housing improvements other than whitewash, and the organization providing the whitewash often included other interventions for the same houses, we were unable to differentiate the true effect of whitewash versus plastering, roof replacement, focal spraying or other changes. Although we recorded the presence of cracks in the wall large enough for a vector to enter, we did not further differentiate size of cracks. The question of reported insecticide spraying was likely subject to recall bias. Furthermore, comprehensive vector data are not available from the time of the study, impeding a direct assessment of household-level infestation and T. cruzi risk.
A large proportion of the population refused to take part in the serosurvey. Those refusing were more likely to be adult males. Thus, as demonstrated by the population adjusted prevalence, the true prevalence in the entire population was likely higher. Refusals in the younger population were much more infrequent. As this was the population from which the risk factor analysis was performed, we expect that refusals played a minimal role in the risk factor analysis. However, due to the age limits imposed upon these analyses we cannot say with any certainty that these same risk factors applied to older populations.
In summary, we found that the study area had an extremely high prevalence of T. cruzi infection, with evidence of continued vector-borne transmission despite earlier vector control efforts, and that catalytic modeling provided a useful tool to estimate the force of infection over time and examine the impact of past interventions. As part of a renewed vector control effort in the Bolivian Chaco, municipal teams together with the regional Chagas disease control program performed blanket spraying at the conclusion of our serosurvey. A second round of spraying is planned for 2013. The findings that cracked walls, evidence of vector infestation and whitewashing of walls affected risk of recent T. cruzi infection were not unexpected, but highlight the need for systematic housing improvement and effective vector surveillance in the future.
|
10.1371/journal.pcbi.1006260 | How adaptive plasticity evolves when selected against | Adaptive plasticity allows organisms to cope with environmental change, thereby increasing the population’s long-term fitness. However, individual selection can only compare the fitness of individuals within each generation: if the environment changes more slowly than the generation time (i.e., a coarse-grained environment) a population will not experience selection for plasticity even if it is adaptive in the long-term. How does adaptive plasticity then evolve? One explanation is that, if competing alleles conferring different degrees of plasticity persist across multiple environments, natural selection between genetic lineages could select for adaptive plasticity (lineage selection). We show that adaptive plasticity can evolve even in the absence of such lineage selection. Instead, we propose that adaptive plasticity in coarse-grained environments evolves as a by-product of inefficient short-term natural selection: populations that rapidly evolve their phenotypes in response to selective pressures follow short-term optima, with the result that they have reduced long-term fitness across environments. Conversely, populations that accumulate limited genetic change within each environment evolve long-term adaptive plasticity even when plasticity incurs short-term costs. These results remain qualitatively similar regardless of whether we decrease the efficiency of natural selection by increasing the rate of environmental change or decreasing mutation rate, demonstrating that both factors act via the same mechanism. We demonstrate how this mechanism can be understood through the concept of learning rate. Our work shows how plastic responses that are costly in the short term, yet adaptive in the long term, can evolve as a by-product of inefficient short-term selection, without selection for plasticity at either the individual or lineage level.
| Organisms respond to different environments by changing how they act, look or function. When these responses improve the chances of survival, we call them adaptive plasticity. But observing adaptive plasticity does not prove that the response evolved because it improved survival. Being plastic is only selected for if individuals experience environmental variation, so that in slow changing environments plasticity may be selected against even if it is adaptive in the long term. Can adaptive plastic responses still evolve under these conditions? Yes. We use learning theory to describe how genetic changes accumulate when individual lifespan is shorter than the time between environmental changes, and show that adaptively plastic responses can evolve even when they are selected against. This is because adaptive plastic responses can evolve as the by-product of selection for different functions in different environments, as long as organisms retain some plasticity until the next environmental change. Our work demonstrates that evolution can reach general solutions even when each individual is only presented with a simple fraction of a more complex problem. This intuition could explain why plastic responses to past environments can be adaptive even to environments the entire lineage has never seen before.
| Organisms that live in variable environments are often subject to opposing selective pressures, either temporal or spatial, such that intermediate generalist phenotypes have decreased fitness across all environments. Rather than evolving a generalist phenotype, populations can keep adapting to each environmental condition as they encounter them, a process known as adaptive tracking [1, 2]. Populations that evolve via adaptive tracking need time to adapt to each new environment. As a result of this adaptation, the population experiences reduced fitness after each environmental change. Both populations that evolve a generalist phenotype and those that evolve by adaptive tracking thus have reduced fitness in the long term. By contrast, adaptive phenotypic plasticity allows individuals to maintain an adaptive fit between phenotype and environment: plastic individuals produce only high fitness phenotypes by responding appropriately to environmental cues. Populations evolving adaptive plasticity thus avoid both the fitness loss arising from trade-offs of generalist phenotypes and the fitness loss that tracking populations suffer after environmental change. Within this framework, the question of whether plasticity evolves can be interpreted as the comparison between the average fitness across all environments for populations which evolve plastic responses, evolve generalist phenotypes or evolve via tracking [3, 4]. As such, a considerable amount of effort has been invested in characterizing the conditions that determine the fitness of plastic rather than non-plastic solutions, and to document if plasticity itself incurs a fitness cost [5–7].
While adaptive plasticity is common in nature and demonstrably superior to non-plastic solutions for a wide range of conditions, the process by which it evolves remains a matter of debate. The standard assumption that natural selection favours the best available solution is problematic, since natural selection only discriminates between phenotypes that are expressed. Natural selection is thus unable to detect that a plastic organism is adapted to more environments than a non-plastic one unless individuals encounter multiple environments within their life spans, a condition known as environmental fine-grain [8]. Even when individuals experience more than one environment per lifetime, each individual may express only a single phenotype if plastic responses are irreversible [9–11], too slow (e.g [12]) or too costly (e.g. [5]) relative to the fitness advantage of producing the right phenotype for the current conditions [2, 7].
This creates an evolutionary dilemma: adaptive plasticity maximizes fitness in the long-term, but natural selection favours non-plastic phenotypes in each short-term environment. In other words, experiencing one environment per lifetime (environmental coarse grain) does not allow individual selection for plasticity, so that if plastic responses incur any cost compared to non-plastic phenotypes they will be selected against in the short-term. Since costly plastic responses in coarse-grained environments provide fitness benefits only when individuals are selected over multiple generations, we refer to those responses as long-term adaptive plasticity. While long-term adaptive plasticity is selected against in the short-term, adaptive responses to coarse-grained environments commonly evolve, and include environmental determination of resistance and dispersal phenotypes [13, 14] and seasonal morphs of short-lived species [9, 15].
How can we explain the process by which costly adaptive plasticity evolves in such coarse-grained environments? While individual-level selection does not favour plasticity in coarse-grained environments, alleles that determine an organism’s plasticity are transmitted between generations, and their fixation or loss will depend on their fitness across the set of environments they encounter [16, 17]. Natural selection may therefore discriminate between plastic and non-plastic alleles if both are maintained long enough to be selected across multiple environments, even if each individual organism experiences only a single environment. Plastic adaptations to coarse-grained environments could therefore evolve if multiple alleles (genetic lineages) persist long enough to be subject to natural selection across multiple generations and environments, a process known as lineage selection [4, 18, 19]. More precisely, we define lineage selection as a specific type of natural selection acting on multiple alleles which persist for multiple generations (see [20]). This is in contrast with Strong Selection and Weak Mutation regimes (SSWM) in which each new allele is either lost or fixed before more genetic variation can arise. Under SSWM genetic variation is provided only by new mutations (rather than standing genetic variation), so that repeatedly comparing multiple alleles is impossible.
The availability and persistence of standing genetic variation on plastic responses is thus a key requirement for the evolution of adaptive plasticity in coarse grained environments (e.g. [4, 17, 19]). This implies that plasticity will not evolve in populations that are small or under strong selection, since these conditions remove the genetic variation lineage selection requires to operate (e.g. [21]). Because small population size and strong selection are representative for populations experiencing rapid environmental change, evolution of plasticity appears unlikely to play a role in evolutionary rescue or successful colonization [22, 23]. The evolution of costly adaptive plasticity will only be possible if genetic diversity is available, but high genetic diversity will also cause rapid removal of costly plastic variants in favour of non-plastic short-term solutions, so that costly adaptive plasticity should only evolve as an intermediate step towards non-plastic solutions.
We apply a core concept of learning theory—learning rate—to propose an alternative mechanism for the evolution and maintenance of costly adaptive plasticity without lineage selection. In machine learning, learning rate measures the amount of change a system accumulates with each example shown. Existing literature demonstrates that the process of learning by trial and error is mechanistically analogous to evolution by natural selection [24]. In the context of adaptation, genetic learning rate measures the ability of a population to change in response to new environments by accumulating adaptive mutations. More specifically, we can define genetic learning rates as the amount of genetic change fixed by a population in each new environment. Genetic learning rate (henceforth just learning rate) depends both on the ability to generate variation (mutation rate and effect size, population size) and to fix particular variants (strength of selection). Since both the processes that produce and fix variants require time to operate, increasing the time spent in each environment will allow populations to accumulate more adaptive change. Thus, the more generations a population spends in a single environment the higher its learning rate will be.
As we show in our simulations, populations initially produce phenotypes matching their current environment by accumulating both mutations that change the mean phenotypic value and mutations that change plasticity. Populations with high learning rates find optimal phenotypes for the current environment and remove costly plasticity before each new environmental shift: When populations can quickly reach current optima in each current environment, plastic adaptations to past environments cannot evolve. Populations with low learning rates cannot reach current optima before the next environmental shift, and pass on to the next environment all genetic changes which brought them closer to the previous phenotypic optimum, whether or not these genetic changes cause phenotypes to be plastic. Selection in the new environment thus starts from a population which already accumulated adaptively plastic changes, so that the overall plastic responses can be further refined over time.
In evolutionary terms, low learning rates maintain directional selection for plastic development, with the end result of directing evolution towards the production of long-term adaptive plastic responses. Unlike the lineage selection explanation, the learning theory explanation does not require the prolonged co-existence of alleles with different effects on plasticity: adaptive plastic responses will evolve even in populations which exhibit only a single reaction norm at any given time. Rather, learning theory only requires that the population accumulates limited genetic change per environment, so that the average genotype retains some of the adaptive plasticity accumulated in past environments. Learning theory thus predicts that, as long as natural selection is inefficient in bringing about genetic change, long-term adaptive plasticity should evolve even in the extreme case when only one lineage is present in the population at any given time (strong selection weak mutation) and plasticity is selected against in each current environment.
In this paper, we provide a first exploration of the evolution of adaptive plasticity from a learning theory perspective. To do so, we employ a classic linear reaction norm model [25, 26] to simulate the evolution of costly adaptive plasticity in temporally coarse-grained scenarios. This allows us to contrast the predictions made by learning theory and lineage selection regarding when and how plasticity should evolve. First, we demonstrate that plasticity can evolve in coarse-grained environments, showing that individual-level selection for plasticity is not necessary to evolve adaptive plasticity. Second, we demonstrate that adaptive plasticity evolves in coarse-grained environments even in the absence of multiple lineages, counter to the predictions of lineage selection. Third, we show that limiting mutation rates biases populations towards adaptive plasticity rather than adaptive tracking, in accordance with the predictions of learning theory. These results reveal that long-term adaptations can evolve even when each current environment selects against them, as long as natural selection is inefficient.
We simulate a population that experiences temporal environmental heterogeneity. Each individual receives information from the environment and develops into an adult phenotype, upon which selection can act. We follow standard approaches for the evolution of plasticity [18, 27, 28] and model development as a linear reaction norm, whose intercept a represents the mean genetic trait value across environments (also known as G, e.g. [17]) and slope b the degree of plasticity, or genotype by environment interaction (GxE, see Reaction norm model). The developed phenotype P is thus
P = a + b * C
where C is the univariate environmental cue.
We model a heterogeneous varying environment with 10 environmental states, so that each environmental state, Ei produces a single, unique value of the cue C E i and requires a single specific univariate phenotype P E i. We model the matching between cues and trait optima as a linear function (see Environmental variability). This implies that a linear reaction norm with appropriate slope and intercept can achieve perfect fit for all environments in our set. We assume non-overlapping generations of individuals with a constant fixed lifespan. This assumption allows us to control the granularity of environmental variability with a single parameter, K. If K ≥ 1 the environment changes every K generations, indicating coarse-grained (K = 1) or slow coarse-grained (K > 1) environmental variability. If instead K < 1 the population encounters on average 1/K environments per generation, indicating fine-grained environmental variability.
We evaluate the fitness of each individual based on the distance of its developed phenotype from the optimal target phenotype in the current environment. In case the individuals experience more than one environment, we calculate their fitness as the mean match between the developed phenotypes and the selective environments experienced. We further impose a fitness penalty proportional to the individual’s responsiveness to its environment (absolute reaction norm slope b, see above). This cost of plasticity ensures that plastic individuals will have lower fitness than non-plastic ones regardless of their phenotypes, and effectively represents a trade-off incurred by plastic organisms (see Evaluation of fitness). While few empirical studies have found evidence for costs of plasticity (see Conclusion), including a cost means that plasticity is selected against, and thus serves as a form of conservative bias against plasticity. Since we measure fitness as relative to a pre-specified optimal phenotype, we express it as phenotypic mismatch or lack-of-fit: a measure which decreases quadratically from zero as the phenotype diverges from the optimum (see section Evaluation of reaction norms).
Organisms reproduce asexually with a probability proportional to their relative fitness within the population (see Evolutionary process). Every individual inherits the same slope and intercept as their parents, which are then mutated by adding a random value selected from a normal distribution with mean 0 and standard deviation equal to the mutation size (0.01 unless otherwise specified). Thus, both intercept and slope mutate every generation (effective mutation rate = 1), but most mutations have small effects. Unless otherwise stated, we set a population of 1000 individuals and choose strength of selection ω of 0.2. In addition, we set the associated cost of plasticity, λ, to be 0.1.
While we assume that the cost of plasticity is a property of the genotype, the fitness losses caused by adaptive tracking depend on both the frequency of environmental changes and the amount of time required to reach new short-term optima after each environmental change. Thus, if environmental changes are rare or if the population can quickly reach new optima, the cost of adaptive tracking can be lower than the cost of adaptive plasticity. To verify whether or not adaptive plasticity is the optimal long-term strategy, we analytically tested all parameter combinations used in our simulations (see S1 Appendix). Our analysis confirmed that the fitness cost of adaptive tracking is greater than the cost of adaptive plasticity for all parameter combinations used in this paper. Since adaptive plasticity is the optimal strategy across all our simulations, we can rule out that the eventual evolution of adaptive tracking is because of its greater long-term fitness. In other words, lineage selection should select for adaptive plasticity across all our simulations, since adaptive plasticity incurs lower fitness costs compared to adaptive tracking.
In this section, we compare the evolution of plasticity in fine-grained environments, which allow individual-level selection for plasticity, with coarse-grained ones, which do not. We initially assess the evolution of phenotypic plasticity when individuals encounter multiple environmental states per life-time (i.e., a fine-grained environment; here 10, K = 0.1). We further assume that the phenotype can change during individuals’ lifespan (reversible plasticity), and this change is both immediate and incurs in no fitness costs.
In fine grained environments, the evolved reaction norms converge the optimal intercept and slope in less than 3000 generations (Fig 1A, inset). This means that individuals produce trait values that perfectly match the optimal trait value of all environmental states they encountered during their lifetime, as we can see from the fact that the distance between realised and optimal phenotypes decreases to zero for all environments in our set (Fig 1A). We find minimal residual genetic variation on both the slope and intercept terms of the reaction norm (Fig 1B). This is reflected in the limited differences between the reaction norms of top and mean performing individuals (Fig 1A). Note that the reaction of the average (yellow dots) and best individual (green dots) are perfectly aligned and match the optimal reaction norm (red crosses).
We contrast the previous fine-grained scenario with a slow coarse-grained environment in which conditions change every 4000 generations on average (K = 4000). As such, each individual experiences only one environment, and environmental change between generations is also slow. In this coarse-grained environment, the population fails to evolve adaptive long-term plasticity (Fig 2). After each environmental change we observe a drop in fitness to the current environment, followed by a distinctive two-step pattern in their adaptive paths. During the first phase, organisms evolve towards the new target phenotype, as indicated by the steep increase in current fitness (Fig 2A, inset, green line). Crucially, the increase in current fitness during this phase is accompanied by a corresponding increase in fitness to past environments (Fig 2A, blue line), which indicates evolution of adaptive plasticity. During this phase, mutations which increase plasticity can be selected for if they cause the production of fitter phenotypes, offsetting the cost of plasticity (see S1 Appendix). After organisms are able to produce phenotypes which match the current phenotypic optima, we observe a decrease in their fitness to past environments (Fig 2A, blue curve). This indicates that the same organisms would no longer be able to produce adaptive phenotypes when exposed to past environments, consistent with a decrease in costly adaptive plasticity. During this phase plasticity is directly selected against in order to decrease its fitness costs.
In other words, the population reaches the optimal phenotype using a combination of slope and intercept (phenotypic adaptation) and then minimizes the slope (plasticity minimization). From a fitness perspective, selection during the phenotypic adaptation phase increases fitness by producing the current target phenotype, whereas selection in the plasticity minimization phase increases fitness by maintaining the current target phenotype while removing costly plasticity. It is worth noting that these two phases match those described in the analogous model presented in [17]. After the plasticity minimization phase we still observe some genetic variation in reaction norm slope (grey lines in Fig 2B), but the average slope is 0: adaptive plastic responses are approximately as likely as maladaptive ones. Populations evolving under slow, coarse-grained environments thus fail to evolve adaptive plasticity and instead re-adapt upon each environmental change, consistently with adaptive tracking.
Next, we test whether or not direct selection for plasticity is required for its evolution. To do so, we set the environment to change every generation (K = 1), which is the fastest rate we can set under a coarse-grain scenario: every individual experiences only a single environment, but every generation experiences a different one. Since each individual only experiences one environment, we can rule out direct selection for adaptive plasticity. Furthermore, costly plasticity is selected against within each short-term environment.
In this fast coarse-grained environment, populations evolved adaptive plasticity (Fig 3). We observe that the deviation from the optimal phenotype for both current and past environments decreased to zero, indicating optimal fit to all environments within the range experienced (Fig 3A). In addition, we observe less residual genetic variation compared to the case of slow coarse-grained environmental variability (Fig 3B). This is also indicated by the narrow gap between the top and the mean performance curve in Fig 3A.
Looking at the evolutionary trajectory of the population, we can see that while fitness to the current environment (green line) fluctuates, fitness to the whole environment set (past environment; blue line) gradually increases over time. Moreover, we see no gap between performance in current and past environments. This indicates that increasing fitness to the current environments does not cause loss of fitness in past environments. Instead, the population accumulates responses that are adaptive for all previously experienced environments. These results demonstrate that populations evolving in fast-changing environments produce adaptive plastic responses even when plasticity is costly and environmental change only occurs between generations.
At this stage, we have merely confirmed well-known results (e.g., [17]). We now consider two explanations for the evolution of adaptive plasticity in coarse-grained environments. The standard interpretation is based on a lineage selection model, where faster environmental change will increase the odds that each allele is tested in more than one environment. Adaptive plasticity can evolve since plastic alleles have greater mean fitness than non-plastic alleles when compared across multiple environments, even though the latter have higher fitness within each current environment. The learning theory interpretation instead is based on the prediction that decreasing the number of generations in each environment will decrease the genetic change accumulated within each environment (i.e., the learning rate), ensuring that the changes accumulated during the phenotypic adaptation phase are not lost because of optimization to current environments. While both mechanisms cause a shift from short to long-term adaptation, each has distinct requirements: lineage selection relies on the transmission of genetic variants in order to compare the fitness of multiple alleles; learning theory requires that populations accumulate little genetic change in each environment, so that the system retains some information from the past. In contrast with lineage selection, learning theory does not require that past information is stored in separate lineages. Rather, past information can also be stored in developmental parameters, such as the slope of plasticity. As long as plasticity does not revert to zero, the system retains some information about past adaptive plasticity and can be progressively improved after each environmental change, regardless of the presence of trans-generational genetic variation. In the next two sections, we make use of this key difference to determine which of the two processes can better explain the evolution of plasticity in coarse environments.
To test the need for lineage selection, we repeat the scenarios for the evolution of plasticity in fine-grained (K = 0.1), coarse-grained (K = 1) and slow coarse-grained (K = 40000) environments enforcing strong selection and weak mutation (SSWM). Under SSWM, the speed at which mutations arise is much slower compared to the speed at which they are fixed or lost, driving standing genetic variation to zero. Comparing the fitness of alleles across different environments is therefore impossible. We model SSWM using a hill-climber algorithm: each evolutionary step produces only one mutation. If the new mutation is fitter than the previous one it is fixed, otherwise it is lost (see Hill-climbing model). SSWM leads to a constant effective population size of 1 and makes lineage selection impossible. Therefore, if the lineage selection hypothesis is correct, we expect that adaptive plasticity will fail to evolve in all coarse-grained environments. To rule out that the potential failure to evolve plasticity is due to insufficient time, we verify the results under an extended simulation time of 2*107 generations.
Contrary to the predictions of the lineage selection explanation, we find that the results from the above simulations are qualitatively and quantitatively similar to those obtained using a population size of 1000, despite the SSWM selection regime (Fig 4). That is, populations fail to evolve plasticity when environments change every 40000 generations (Fig 4A), and succeed in doing so when provided with either fine environmental grain (Fig 4B) or a rapid coarse-grained (i.e., trans-generational) change (Fig 4C).
The evolutionary trajectory of populations under SSWM also remains remarkably similar to that of populations with standing genetic variation (compare Fig 4 with Figs 1, 2 and 3). Populations evolving in fine-grained and fast coarse-grained environments both show a gradual increase in fitness to past environments, which remains comparable to fitness in the current environment. This indicates that they adapt to all previously seen environments rather than just the current one. Populations in slow coarse-grained environments instead perform consistently better in current environments compared to past ones, showing the repeated evolution of phenotypes adapted to current conditions, or adaptive tracking. Their evolutionary trajectory also displays the same two-step cycle after each environmental change: fitness increase in both current and past environments (phenotypic adaptation) followed by fitness decrease in past environments only (plasticity minimization) (Fig 4A).
Taken together, these findings demonstrate that both the final results and the evolutionary trajectories of our simulations are largely unaffected by the lack of standing genetic variation. Since standing genetic variation is required for adaptation via lineage selection, these results falsify the hypothesis that plasticity needs to evolve by averaging the fitness benefits of alternative variants across multiple environments. In the next section, we make further predictions based on the learning theory explanation and try to falsify them.
Using a learning theory framework, we can define the conditions that allow evolution in coarse-grained environments to approximate evolution in fine-grained ones. The two scenarios will produce the same outcome only as long as the average of evolutionary changes in coarse-grained environments is the same as the evolutionary changes that would happen in fine grained environments.
In our specific example, individuals selected in slow coarse-grained environments evolve non-plastic solutions after each environmental change. On average, evolutionary changes in slow coarse-grained environments decrease plasticity until it reaches zero. This is in contrast with fine-grained environments, which evolve plasticity towards the optimal adaptive slope. Since the average change in plasticity in coarse-grained environments is different from the change in plasticity under fine-grained environments, the two scenarios have different outcomes. Conversely, individuals selected in fast coarse-grained environments retain some plasticity between environments. Furthermore, on average, the change in plasticity induced by each new environment points towards optimal adaptive plasticity: inherited maladaptive plasticity will be selected against, and inherited adaptive plasticity will be conserved. Therefore, as long as plasticity does not reach zero before the environment changes, evolution in coarse-grained environments will follow the same direction as evolution in fine-grained environments. This is the reason why we expect lower learning rates to cause the evolution of adaptive plasticity in coarse-grained environments: lower learning rates ensure that the population does not find short-term, non-plastic optima before the next environmental change, which allows the averaging of plasticity across environments.
Since we define learning rates in biological systems as the amount of genetic change accumulated by the population in each new environment, they can be affected by several parameters other than rate of environmental change. Population size, mutation size and mutation frequency will all increase the amount of genetic change produced within each environment and thus increase the population’s learning rates. Stronger selective pressure will speed up the fixation of beneficial variants, and therefore also increase learning rates. If the learning rate explanation for the evolution of adaptive plasticity in coarse-grained environments is correct, these factors should be interchangeable with the rate of environmental change.
For example, small populations or populations with low mutation frequency should be able to find long-term plastic solutions even when environmental change is rare. It is important to point out that decreasing population size or mutation frequency would instead hinder the action of lineage selection, which benefits from the maintenance of a large pool of genetic variants to select from.
While a full exploration of all possible parameter space is beyond on the scope of this paper, we evaluate the learning theory explanation by testing the specific prediction that adaptively plastic responses can evolve even when environmental changes are slow, provided that mutation sizes are sufficiently small (and hence learning rate is low). This question can be answered using the same model, and in particular the case of slow coarse-grained environments (environments change every 4000 generations) with a population size of 1000 individuals. As shown above, adaptive plasticity fails to evolve under these conditions. Learning theory explains this failure with the high learning rates in this population. Rather than decreasing the learning rate by decreasing the number of generation spent in each environment, we lower the standard deviation of mutation sizes from 10−2 to 10−4.
As we can see in Fig 5B, the population eventually evolves an optimally adaptive plastic reaction norm, with negligible amounts of variation around both slope and intercept. Their evolutionary trajectories (Fig 5A) are also qualitatively similar to those of populations evolving in fast, coarse-grained environments. In both scenarios, fitness to the current environment (green) fluctuates around average fitness to past environments (blue), indicating that the populations are not evolving phenotypes that increase current fitness at the expense of past adaptation. The steady increase in average fitness to past environments instead indicates the evolution and retention of more general, plastic solutions.
While the two trajectories are similar in shape, the population experiencing slower environmental changes and smaller mutation rates takes a significantly longer to reach optimal plasticity. An increase in the number of generations required to find solutions is a known consequence of lower learning rates. Intuitively, we can explain the longer time required to adapt as a consequence of the slower rate at which variants become available.
While lineage selection is technically viable in this simulation, decreasing mutation sizes would also decrease the amount of available genetic variation, making it even less effective. A potential alternative explanation to our findings is that the reduced amount of genetic change per generation would enable multiple lineages to persist for longer, thus enabling the action of lineage selection. To test for this alternative explanation we run a simulation with K = 40000 and σμ = 10−5 using a hill-climber to model SSWM. The results are both qualitatively and quantitatively similar to those obtained in the previous simulation (see Fig 6). Since our results are unaffected by the absence of lineages, we can rule out that the observed evolution of plasticity with smaller mutation rates is due to the longer persistence of multiple lineages. Taken together, our simulations provide falsifying evidence for a number of frequent assumptions on the requirements for the evolution of costly adaptive plasticity in coarse-grained environments, which we summarize in Table 1.
The evolution of costly adaptive plasticity has often been framed as a necessity caused by environmental change outpacing the ability of natural selection to generate new adaptations [2, 3, 29, 30], but the process by which organisms achieve plasticity in these conditions have seldom been clarified.
We demonstrate that neither individual nor lineage-level selection for adaptive plasticity are necessary for the evolution of adaptive plasticity. Rather, the speed of adaptation relative to environmental change (modelled as learning rates) is by itself a causal factor in the evolution of plastic responses that are adaptive across a range of coarse-grained environments. High learning rates allow optimization of phenotypes in each current environment, at the expense of more general solutions that improve their fitness across all environments experienced. Low learning rates instead make it impossible for phenotypes to chase short-term optima, yet allow individuals to reach long-term optimal plasticity despite the presence of short-term trade-offs.
If approached from a purely adaptationist perspective, these results seem counter-intuitive: the conditions which allow natural selection to work most effectively (high population sizes, high mutation rates, strong selective pressure and rare changes in the environment) result in an evolutionary outcome (adaptive tracking) which has lower fitness than adaptive plasticity across all of our simulations (see S1 Appendix). Conversely, changes in the same parameters that decrease the ability of natural selection to effectively cause phenotypic change result in an evolutionary outcome (adaptive plasticity) which maximizes fitness of the population in the long-term. We explain these counter-intuitive findings by using learning rates, a core concept of learning theory. Specifically, we demonstrate that low learning rates prevent populations from reaching short-term optima before a new environmental change occurs. This in turn allows evolved plastic reaction norms to be transferred across environments, so that they are effectively selected across multiple environments. The end result is that, as long as learning rates are sufficiently low, selection in coarse-grain environments converges on the same outcome as selection in fine-grained ones: adaptive plasticity. In learning theory terms, the cumulative effect of testing models sequentially on each individual example (online learning) will be the same as testing them on the entire set at once (batch learning) only if learning rates are low enough to prevent overfitting to the last example seen [31].
While low learning rates are necessary to evolve general solutions in the presence of trade-offs in performance, none of the factors that affect learning rates is necessary by itself. This is because learning rate is a composite measure, so any given factor may be offset by the others. We demonstrate this by showing that low mutation rate is sufficient to evolve costly adaptive plasticity even in slow, coarse-grained environments. Increasing population size and selection strength should instead decrease the odds of evolving costly adaptive plasticity, as both factors increase learning rates. As a consequence, even populations with no measurable genetic variation in plasticity could evolve adaptive plastic responses as long as (1) new genetic variation can be produced over time and (2) short-term optima change before natural selection can reduce plasticity to zero.
This observation reverses the suggested causal link between plasticity and the rate of genetic evolution. Current theory proposes that plastic individuals experience weaker selection because they are able to cope with a wider range of environments [4]. Because of the reduced selective pressure, the amount of genetic change that accumulates in the population (learning rate) is also reduced. We instead suggest a low learning rate itself may skew populations towards evolving more general solutions, including plastic responses that are costly in current conditions but optimal across the entire set of previously experienced environments. As such, weak selection could facilitate the evolution of plasticity.
Since low learning rates promote the evolution of adaptive plastic responses by reducing the relative importance of minimizing plasticity costs, they are irrelevant to the evolution of inexpensive plastic responses. When there are no costs of plasticity, every combination of slope and intercept that generates the optimal short-term phenotype is fitness equivalent within each environment. Because plastic and non-plastic solutions have the same short-term fitness, adaptive plasticity is selected for when the population moves towards the current phenotypic optimum and randomly drift after the optimal phenotype has been reached. The population will thus inevitably find the optimum for all past environments, and learning rates will only determine the speed at which the population reaches the optimum.
Learning rates are likewise irrelevant for the evolution of costly adaptive plasticity in fine-grained environments, which are sufficient (but not necessary) for the evolution of adaptive plasticity across all our simulations (see S1 Fig). Fine-grained environments allow natural selection to directly compare the fitness of phenotypes across multiple environments at the individual-level within each generation, so that adaptive plasticity is optimal even in the short-term. Direct selection for plasticity is unsurprisingly sufficient to ensure the evolution of adaptive plasticity. Under those conditions, learning rates can only determine the speed of selective process rather than its outcome.
Our simulations consider the specific case of maintenance costs for plasticity. That is, we assume that plasticity directly decreases fitness, regardless of whether it is expressed. This assumption has a long history in modelling the evolution of plastic responses, but has been largely unsupported by empirical data which does not find costs of plasticity for the vast majority of traits analysed [32, 33]. However, several alternative scenarios can create mathematically equivalent trade-offs between selection in current and past environments. A well-studied example is that of inaccurate cues, either due to imperfect perception or noise in the cues themselves [3, 22, 34]. Alternatively, the target phenotypes may not perfectly match with the best possible reaction norm. This scenario can happen for any reaction norm which is selected on a set of environments larger than its degrees of freedom (3 in the case of linear reaction norms) [35] or if there are limits to the maximum amount of plastic changes that an organism can evolve [27, 32, 33, 36]. In all of the above mentioned cases, optimal long-term plasticity would cause loss of fitness across current environments and consequently be selected against. Learning rates will thus be relevant for the evolution of plastic responses across all of them.
In our simulations, mutations that lead to adaptive plasticity are selected since they increase phenotypic fitness within current environments, eventually causing the evolution of adaptive long-term plasticity. This is in contrast with lineage selection models, in which mutations that cause adaptive plasticity are selected because of their long-term benefits, but are (at best) selectively neutral in current environments. Since the evolution of plasticity in our model is driven by a short-term (rather than lineage) selection process, we predict it to be both faster and more robust to the presence of trade-offs. Similar dynamics apply to the evolution of modularity as a by-product of short-term phenotypic selection, and are proven to be scalable to arbitrarily complex systems [37].
From a learning theory perspective, low learning rates cause the evolution of adaptive plasticity because they constrain populations to evolve new adaptive solutions starting from previous genetic adaptations of the reaction norm rather than ‘from scratch’. As a result, evolved reaction norms do more than just ‘remember’ which specific phenotype associated with each specific environment: they capture the logic that connects all cues to all phenotypes. In learning theory terms, organisms learn the regularities of the (evolutionary) problem, a process also known as generalization [31]. Therefore, as long as regularities remain the same, each individual will be able to produce adaptive phenotypes even in environments it has never experienced in its evolutionary history (extrapolation), without the need for further adaptation. Conversely, several studies show that systems that learn a problem’s regularities are also able to quickly adapt to new problems which share a similar logic [38, 39]. This ability to more rapidly evolve new adaptive phenotypes in response to new environments can instead considered as an increase in their evolvability. Our demonstration that organisms can learn regularities between environments even when each organism only ever experiences a single environment opens up the possibility that evolved plastic responses may both prepare organisms for future, more extreme, environments (via extrapolation) and enable them to more rapidly evolve new adaptive solutions (via evolvability). This demonstrates that past evolution can shape evolutionary trajectories by biasing the phenotypic variants that are exposed to selection [24, 40].
In summary, we use a simple reaction norm model to demonstrate that costly adaptive plasticity can evolve even when natural selection is unable to compare competing alleles over multiple environments (i.e., lineage selection). A learning theory framework helps us interpret this finding: Populations evolving in coarse-grained environments can evolve adaptive plasticity if the amount of adaptive change accumulated per environment—the learning rate—is low. Populations with high learning rates evolve via repeated short-term adaptation even if this pattern is maladaptive in the long term. Low learning rates facilitate adaptation to the entire set of environments experienced over adaptation to just the current environment, favouring adaptive plasticity even in the presence of short-term functional trade-offs. Thus, long-term adaptive plasticity can evolve even when it is not selected for at either the individual nor lineage level. Whether a population evolves phenotypes that optimize fitness in the short or long term instead depends on the amount of adaptive changes it accumulates within each environment.
For plasticity to evolve, the environment needs to fulfill two roles: determining the selective conditions (selective role) and providing information about those conditions (constructive role) [41]. We simulate the selective role by assigning each environmental state (current or short-term environment) a target single trait optimum ϕ, represented by a single real number. We simulate the constructive role by assigning each target optima an environmental cue represented by a real number C sampled from a normal distribution with mean 1 and standard deviation 1. Each of our simulations cycles between 10 short-term environments, which make up the long-term environment. For simplicity, we consider a linear relationship between phenotypic targets and environmental cues, so that ϕ = g(C) = g1 * e + g0. Hence, the targets are directly proportional to the respective cue. We choose g1 = −2 and g0 = 6. This ensures that the relationship between selective environment and cues remains constant across environmental states.
We assume that the lifespan of the individuals is fixed and the same for all. As a result, environmental grain is solely determined by the parameter K. K < 1 indicates fine-grained environmental variability, where the population encounters an average of 1/K environments per generation. On the other hand, K >= 1 indicates coarse-grained (K = 1) or slow coarse-grained (K > 1) environmental variability where the population encounters a new environment every K generations on average. We choose small K values compared with the total number of generations in our simulations so that each population is able to evolve for multiple environmental cycles.
Our simulations were designed with temporal variation in mind, but the conclusions should be applicable to spatial variation as well. In fact, the environmental fluctuations described within our model match those experienced by a population in which all individuals migrate after fitness evaluation and before reproduction, or in which all propagules are dispersed to the same new environment. In this scenario environmental change rates are effectively interchangeable with migration rates, with other findings remaining unchanged.
We model plastic responses using a univariate linear reaction norm model [42]. A reaction norm can be defined as the set of phenotypes that would be expressed if the given individual would be exposed to the respective set of environments. Since we consider univariate and linear reaction norms, we can describe the development of an organism’s phenotype as P = a + b * C. Each organism’s genotype can thus be described by the factors a and b. Of those, a determines the organism’s breeding value and b the direction and magnitude of its plasticity.
We model the evolution of a population of asexual individuals as follows. First, we select a parent using a fitness proportional criterion [43, 44]. Each individual can be selected with a probability of f / f ¯, where f ¯ corresponds to mean fitness in the current population and f to the parent’s own fitness (see section Evaluation of fitness for details on how we calculate f). Then, we generate a new individual with the same genotype (reaction norm intercept a and slope b) as the parent. Finally, we independently mutate both the offspring’s intercept and slope by adding a random value sampled from a normal distribution with mean μ = 0 and standard deviation equal to mutation size (σμ = 0.01 unless otherwise specified). We repeat this process until we generate a number of offspring equal to the set population size. The parameters a and b are initialized at zero.
Following previous work [35, 37, 38], we define an organism’s overall fitness f in terms of a benefit-minus-cost function, which allows us to consider both positive (benefits) and negative (costs) contributions to its fitness. The benefit of a given genotype, b E i, for each environment, Ei, is determined based on how close the developed adult phenotype, Pa, is to the target phenotype, P E i, of the given selective environment, Ei. Since we deal with an univariate phenotype, we can calculate this amount as
b E i = w ( P a , P E i ) = - | P a - P E i | , (1)
where |*| corresponds to the absolute distance between the two phenotypes. Note that the selective advantage of respective genotypes is solely determined by its immediate fitness benefits on the currently encountered selective environment(s). We consider that individuals experience a distribution of selective environments during their lifetime with occurring probabilities, q E 1 , q E 2 , . . , q E N. Each environment contributes to the selection process in proportion to its occurrence [45]. The overall fitness benefits of an individual over all experienced environments in its lifetime, bE is determined by the arithmetic mean of the fitness benefits in each environment, b E i, weighted by the occurrence, q E i, of each environment:
b E = ∑ i q E i b E i . (2)
In cases of coarse-grained environmental variability, where each individual encounters a single environment in its lifespan, q E i = 1 for the respective environment, i = j, and q E i = 0 for i ≠ j. On the other hand, in cases of fine-grained environmental variability, we assume a uniform distribution of environments experienced during individual’s lifespan, that is, q E i = 1 / K. The cost represents how maintaining plasticity reduces the organism’s fitness. Unlike the benefit, the cost of plasticity is a property of the genotype and does not change in different environments. Thus, we can calculate the overall performance, d, of a genotype over a range of selective environments as
d = b E - λ | b | , (3)
where parameter λ indicates how steeply fitness decreases in proportion to the reaction norm slope b. The final fitness score is calculated with the following formula:
f = e x p ( d 2 ω ) , (4)
which penalizes lower performances exponentially and re-scales them to a 0-1 range. ω is a scaling factor on the relation between f and d. Lower ω values cause greater loss of fitness per loss of performance, and correspond to steeper selection gradients. We choose ω = 0.2, which corresponds to a scenario of strong selection (see [38]).
We evaluate the adaptive potential of the population due to plasticity by estimating how close the reaction norm of each individual in the population is to the (theoretical) optimal reaction norm. The optimal reaction norm here corresponds the function that given any environmental cue, C E i, produces the appropriate target phenotype, P E i, which best matches the current selective environment, Ei (Evaluation of fitness). We evaluate the performance of reaction norms based on how different they are from the optimal reaction norm. The lack of fit, LackD of a given reaction norm, D, is estimated as a function of the phenotypic trait values in each of the past selective environments (here 10), Ei, whose magnitude increases quadratically with the distance from each phenotypic optimum, P E i:
LackD=−ΣEi| D(eEi)−PEi |NE (5)
Where NE stands for the number of past selective environments. The evaluation of lack of fit is performed for each individual at the end of each environmental period. We report the average and best performance in the population.
A hill-climbing evolutionary model simulates a scenario of strong selection and weak mutation, where each new mutation is either fixed or lost before a new one can arise. Therefore, the entire population shares the same values of a and b. Each evolutionary step introduces a single mutant genotype with parameters a′ and b′ equal to a and b plus a random value sampled from a normal distribution with mean 0 and standard deviation equal to mutation size. We develop both the reference and mutant phenotypes P and P′ (section Reaction norm model) and compare their fitness values f and f′ (section Evaluation of fitness). If f′ > f, the mutation is beneficial and therefore adopted so that at+1 = a′ and bt+1 = b′. Otherwise, the mutation is deleterious and a and b remain unchanged.
The code used to generate the results shown in this paper is provided in S1 File.
|
10.1371/journal.pbio.0060085 | Darwinian Evolution on a Chip | Computer control of Darwinian evolution has been demonstrated by propagating a population of RNA enzymes in a microfluidic device. The RNA population was challenged to catalyze the ligation of an oligonucleotide substrate under conditions of progressively lower substrate concentrations. A microchip-based serial dilution circuit automated an exponential growth phase followed by a 10-fold dilution, which was repeated for 500 log-growth iterations. Evolution was observed in real time as the population adapted and achieved progressively faster growth rates over time. The final evolved enzyme contained a set of 11 mutations that conferred a 90-fold improvement in substrate utilization, coinciding with the applied selective pressure. This system reduces evolution to a microfluidic algorithm, allowing the experimenter to observe and manipulate adaptation.
| The principles of Darwinian evolution are fundamental to understanding biological organization and have been applied to the development of functional molecules in the test tube. Laboratory evolution is greatly accelerated compared with natural evolution, but it usually requires substantial manipulation by the experimenter. Here we describe a system that relies on computer control and microfluidic chip technology to automate the directed evolution of functional molecules, subject to precisely defined parameters. We used a population of billions of RNA enzymes with RNA-joining activity, which were challenged to react in the presence of progressively lower concentrations of substrate. The enzymes that did react were amplified to produce progeny, which were challenged similarly. Whenever the population size reached a predetermined threshold, chip-based operations were executed to isolate a fraction of the population and mix it with fresh reagents. These steps were repeated automatically for 500 iterations of 10-fold exponential growth followed by 10-fold dilution. We observed evolution in real time as the population adapted to the imposed selection constraints and achieved progressively faster growth rates over time. Our microfluidic system allows us to perform Darwinian evolution experiments in much the same way that one would execute a computer program.
| The scientific community will soon celebrate the 200th anniversary of the birth of Charles Darwin and the 150th anniversary of the publication of his seminal work On the Origin of Species by Means of Natural Selection [1]. The principles of Darwinian evolution are fundamental to understanding biological organization at the level of populations of organisms and for explaining the development of biological genomes and macromolecular function. Darwinian evolution also has become a chemical tool for discovering and optimizing functional macromolecules in the test tube (for recent reviews, see [2–5]). Laboratory evolution is greatly accelerated compared with natural evolution but requires substantial manipulation by the experimenter, which is imprecise, time consuming, and usually performed in an ad hoc manner.
Many laboratory procedures have been miniaturized using microfluidic technology, which reduces cost and increases precision over manual methods [6]. In the present study, a system is described that relies on computer control and microfluidic chip technology to automate the directed evolution of functional molecules, a process that is subject to precisely defined parameters. A population of billions of RNA enzymes with RNA ligase activity was made to evolve continuously, with real-time monitoring of the population size and fitness. Whenever the population size reached a predetermined threshold, chip-based operations were executed to isolate a fraction of the population and mix it with a fresh supply of reagents. These steps repeated automatically as the population adapted to the imposed selection constraints within a period of several hours.
The RNA enzyme that was chosen for this study is a descendant of the class I RNA ligase, first described by Bartel and Szostak [7]. It is one of only two RNA enzymes that have been made to undergo continuous in vitro evolution [8,9], a process in which all of the components necessary for evolution are contained within a common reaction vessel. The enzyme is challenged to ligate a promoter-containing oligonucleotide substrate to itself by catalyzing nucleophilic attack of the 3′-hydroxyl of the substrate on the 5′-triphosphate of the enzyme. The reaction mixture also contains two polymerase enzymes (reverse transcriptase and T7 RNA polymerase) that amplify any RNA molecules that have acquired the promoter sequence as a consequence of RNA-catalyzed ligation. Multiple copies of progeny RNA are generated, which then can enter another cycle of reaction and selective amplification. In a population of variant RNA enzymes, those that react most efficiently grow to dominate the population in the competition for limited chemical resources. A serial transfer or serial dilution protocol is used to refresh periodically the supply of reagents, allowing the evolution process to continue indefinitely.
The chip-based evolution system consists of a microfluidic device mounted on a temperature-controlled stage and monitored by an inverted confocal fluorescence microscope (Figure 1A and 1B). A laptop computer controls the actuation of valves on the chip and the acquisition and processing of fluorescence data indicate the concentration of RNA enzymes within the microfluidic circuit. The circuit consists of a mixing loop (1 cm diameter, 400 nl volume), with three in-line valves for mixing and two bus valves that control the input of fresh reagents and the output of spent reaction materials (Figure 1C). This device can perform serial dilutions in a rapid and precise manner [10]. Each iteration of events on the chip entails an incubation step with slow mixing, an isolation step in which one-tenth of the reaction mixture is retained in part of the circuit while fresh reaction materials are drawn into the remainder of the circuit, and a rapid mixing step in which the isolated aliquot is combined with the fresh reagents (Figure 1D). The reaction mixture contains thiazole orange, which intercalates into nucleic acids and upon laser excitation gives a characteristic fluorescent signal. When the fluorescence reaches a predetermined threshold, correlating with a 10-fold increase in the concentration of RNA, the computer initiates an automated 10-fold dilution.
Continuous evolution on the chip was initiated with randomized variants of the “B16–19” form of the class I ligase RNA enzyme [11]. Like other forms of the class I ligase, this enzyme has an impressive catalytic rate of 20 ± 2 min−1, but a somewhat poor Michaelis constant (Km) of 35 ± 8 μM (measured in the presence of 10 mM MgCl2 and 50 mM KCl at pH 7.5 and 37 °C). Random mutations were introduced throughout the molecule at a frequency of ∼0.7% per nucleotide position using a mutagenic PCR procedure [12], followed by in vitro transcription. A starting population of 2 × 109 variants was introduced to the chip and challenged to catalyze RNA ligation under conditions of progressively reduced substrate concentrations. This selection pressure was expected to favor individuals with an improved Km. At the outset, the substrate concentration was 1 μM, causing the starting B16–19 enzyme to operate with an observed rate of only 0.6 min−1. Any individuals with an improved Km would operate at a faster rate, and therefore would undergo more rapid amplification and grow to dominate the population. As the evolving population adapted to the reduced substrate concentration, the concentration was reduced further, eventually reaching just 0.05 μM.
The course of evolution was monitored continuously based on fluorescence, tracking the time needed to achieve 10-fold overall amplification of the RNA population (Figure 2). The log-linear growth rate of the starting population during the first 40 min was used to set the fluorescence threshold for the circuit, before executing the first 10-fold dilution. One hundred iterations of log-growth and dilution were executed in the presence of 1 μM substrate. The instantaneous fitness of the population was reflected by the interval between successive dilutions, which decreased monotonically over the first 100 log-growth iterations.
Materials collected from iterations 95–100 were pooled, subjected to mutagenic PCR, and reintroduced to the chip, but now the substrate concentration was reduced to 0.5 μM. This resulted in a temporary decrease in fitness (increased dilution interval), but the population quickly adapted, achieving 10-fold growth every 10 min by iteration 198. At that point, and at iterations 280, 363, and 428, materials again were collected, mutagenized, and returned to the chip. The substrate concentration was reduced to 0.3 μM at iteration 280, to 0.1 μM at iteration 363, and finally to 0.05 μM at iteration 428.
After 500 iterations of log-growth and dilution (70 h on the chip), the evolution process was deemed complete. Individuals were cloned from the population at iterations 198 and 500 and sequenced. At iteration 500, all of the sequenced clones contained 11 mutations, which could be divided into four subgroups (M1, M2, M3, and M4) based on their relationship to the known secondary structure of the class I ligase (Figure 3). The M1 mutations occur immediately on the 3′ side of the ligation junction, replacing the pppA•U pair by a pppG•C pair. These mutations restore the pppGpG transcription initiation sequence that is preferred by T7 RNA polymerase [13], while maintaining Watson-Crick pairing of the 5′-terminal guanosine. The M2 mutation is a C insertion that appears to extend the template region of the enzyme so that it binds six additional nucleotides in the upstream portion of the substrate. The M3 mutations (one transition, one transversion, and one insertion) all occur within a hairpin loop that is thought to lie in close proximity to the template region, based on modeling of the three-dimensional structure of the class I ligase [14]. The M4 mutations change a U•A pair to a G•C pair within a stem adjacent to the ligation junction.
A representative clone that contained all 11 conserved mutations, as well as three mutations near the 3′ end, was examined with regard to its catalytic properties. It exhibited a kcat of 21 ± 0.8 min−1, which is nearly identical to that of the starting enzyme, and a Km of 0.4 ± 0.05 μM, which corresponds to a 90-fold improvement (Figure 4A). The fact that only Km improved reflects the selective pressure that had been placed on the population. The starting B16–19 form of the enzyme was evolved to operate in the presence of 5 μM substrate [8,11], and this concentration was reduced by 100-fold during the course of 500 logs of on-chip evolution. Thus, the improvement in Km closely parallels the degree of selective pressure that was applied.
To assess the individual contribution of the M1–M4 mutations, each was added to the starting enzyme and each was removed from the final evolved enzyme. The observed reaction rate in the presence of 0.1 μM substrate was determined for each construct (Figure 4B). Adding the M1 or M2 mutations to the starting enzyme caused a 9-fold increase in the observed rate, whereas adding the M3 mutations caused a 24-fold increase. Surprisingly, adding the M4 mutations caused a 2-fold decrease. Reverting the M1, M2, or M4 mutations within the evolved enzyme caused a 2- to 3-fold decrease in the observed rate, while reverting the M3 mutations caused a 10-fold decrease. Thus the M1, M2, and M3 mutations appear to exhibit independent effects, whereas the M4 mutation only confers selective advantage on the background of the other mutations. Clearly, the M3 mutations are the most significant. When they are added to the starting enzyme, there is no change in kcat and there is a 10-fold improvement in Km. Conversely, when the M3 mutations are removed from the evolved enzyme, there is no change in kcat and there is a 10-fold worsening of Km (Figure S1).
Returning to iteration 198, when the population had only been challenged to adapt to 0.5 μM substrate, none of 20 cloned sequences contained the M4 mutations. However, all of the cloned sequences contained the M1 and M2 mutations. The M3 mutations were beginning to appear at that time, with all 20 clones containing the G insertion (Figure 3), but only eight containing the A→C transversion and only two containing the G→A transition. The insertion and transversion mutations are intriguing because they result in the sequence 5′-GACCCAG-3′ (mutations underlined), which is identical to the sequence 5′-GACCCAG-3′ (M2 mutation underlined) that occurs within the extended template region of the enzyme. It is possible that one or both of these regions engages in pairing interactions with the substrate.
Site-directed mutagenesis studies were carried out on the final evolved enzyme to examine potential enzyme–substrate interactions enabled by the M2 and M3 mutations. An alternative substrate was prepared, leaving unchanged the eight nucleotides that are complementary to the original template region of the enzyme, but changing the six nucleotides that are complementary to the extended template region (Figure 3). No activity was observed with this substrate. Activity was largely restored, however, by introducing compensatory mutations within the enzyme that reestablished Watson-Crick complementarity with the substrate (Figure S2). An enzyme containing these six mutations could not react with the original substrate, also consistent with the requirement for complementarity in this region. Interestingly, the region of the M3 mutations could compensate for a partial mismatch within the extended template region. If just four of the six nucleotides in the extended template region were mutated, activity was retained so long as the M3 region was left unchanged. If the M3 region also was mutated, then there was no detectable activity (Figure S2). Thus the region of the M3 mutations appears to assist the extended template region in recognizing the substrate.
The constellation of mutations that arose over the course of evolution could not have been anticipated, especially the triple mutations within the M3 region that had the most dramatic effect on Km and the paired M4 mutations that only had benefit when combined with the other mutations. It is not surprising that these more complex traits arose later in the evolution process. The snapshot of the evolving population that was obtained at iteration 198 revealed intermediate forms, with two advantageous traits (M1 and M2) already acquired and the acquisition of a third (M3) still in progress. If sequence analyses were carried out at more frequent intervals during the 500 iterations of log-growth and dilution, it would provide a more detailed picture of the ebb and flow of genetic traits. This genetic information could be correlated with measurements of the catalytic behavior and growth rate for each of the cloned individuals. Methods exist for microfluidic-based clonal isolation and amplification [15], DNA sequencing [16], and analysis of enzyme kinetics [17], raising the possibility that Darwinian evolution and analysis of the evolving population could be carried out in an integrated microfluidic format.
It is possible that further evolution on the chip, carried out in the presence of even lower concentrations of substrate, would lead to further improvement in Km. Ultimately, however, this improvement would be limited by three constraints: (1) reduced size of the evolving population when operating at very low substrate concentrations, yet maintaining conditions of substrate excess; (2) technical limitations in fluorescence monitoring of sub-nanomolar concentrations of RNA; and (3) intrinsic limitations on the ability of RNA to catalyze templated RNA ligation. With regard to the latter, the catalytic efficiency, kcat/Km, of the final evolved enzyme is 5 × 107 M−1 min−1. This is close to the rate of association of two complementary oligonucleotides, which is ∼109 M−1 min−1 under similar reaction conditions [18–20]. If the catalytic rate of the RNA enzyme remained about 20 min−1, then it would not be possible for Km to improve to better than about 20 nM (a 20-fold improvement compared with the present value), unless the enzyme evolved a means to bind the substrate faster than the inherent rate of duplex formation.
The microfluidic system could be used to obtain RNA enzymes with a variety of phenotypes, including those that have been obtained by conventional in vitro evolution methods. Microfluidic technology also might be used to evolve proteins, viruses, and even cellular organisms. Bacterial populations have been maintained in a microfluidic bioreactor [21], and that system could, in principle, be used to conduct evolution experiments. However, when evolution is carried out at the level of molecules rather than cells, one has ready access to the genotype and phenotype of individuals in the population throughout the course of their evolutionary history. Such access makes it possible to witness evolutionary adaptation and to determine the particular genetic mutations and corresponding phenotypic changes that are responsible for that adaptation.
The chief advantages of chip-based evolution are its precision and ease of operation. The runtime parameters for evolution are established at the outset and are enforced precisely throughout the course of an experiment. The continuous stream of real-time data provides a high-resolution record of an evolutionary trajectory, which can be obtained as a function of population size, population heterogeneity, growth conditions, and the availability of limiting resources. Each microchip contains multiple microfluidic circuits that can be addressed independently, and the chip as a whole can be produced at nominal cost. Thus, Darwinian evolution becomes commoditized, allowing one to perform many evolution experiments with little more difficulty than the execution of a computer program.
Plasmid DNA encoding the B16–19 form of the class I RNA ligase [11] was PCR amplified using a primer that converted the 3′-terminal nucleotides of the enzyme to 5′-ACGAGCAUGGAGGGACU-3′, so as to bind a different cDNA primer. The PCR product was purified by agarose gel electrophoresis, then subject to error-prone PCR, which introduced random mutations at a frequency of ∼0.7% per nucleotide position [12]. The resulting DNA was transcribed in vitro to generate the starting pool of RNA, which was purified by denaturing polyacrylamide gel electrophoresis (PAGE) and desalted on Sephadex G-25. The microfluidic circuit was primed with a solution of 100 nM starting pool RNA, 15 mM MgCl2, 50 mM KCl, 50 mM EPPS (pH 7.5), 4 mM DTT, 0.1% IGEPAL-CA630 (used to reduce surface tension), and 0.1 μM fluorescein dye (used as a tracer).
A microfluidic serial dilution circuit [10] with a carryover fraction of 0.1 was used for the on-chip continuous evolution experiments. The microfabrication process [22] and membrane valve technology [23] have been described previously. The circuit design was modified by splitting the input channel into separate inputs for the polymerase enzymes (E) and mono- and oligonucleotide components (S), converging at the deflection chamber for the in bus valve (Figure 1C). Fluidic and vacuum channel features were etched into separate glass wafers to a depth of 50 μm. The fluidic and vacuum channels had widths of 300 μm and 55 μm, respectively. The mixing loop had a diameter of 1 cm and a volume of 400 nl. The microfluidic device was vacuum-chucked to an aluminum stage fitted with an annular thin-film heater (5548, Minco) and K-type thermocouple probe, controlled by a PID temperature controller (CNi32, Omega Engineering). Data acquisition, pneumatic control, and temperature control were handled by a laptop computer equipped with a NI6715 data acquisition card and software written in-house (LabVIEW, National Instruments).
Computer-controlled pneumatic actuation of valves on the chip was accomplished with a solenoid valve array (HV011, Humphrey). PEEK capillary tubing (25 μm inner diameter, Upchurch Scientific) was used to deliver reagents and collect samples from the device, interfaced with fluid access reservoirs using finger-tight capillary tubing fittings (N-123s, Upchurch Scientific). The sample collection line was pierced through the septum of a 2-ml silanized glass autosampler vial. An additional pneumatic control line fitted with a 24 gauge syringe was inserted through the septum and used to control depressurization of the sample collection vial.
Diode laser excitation (490 nm, Coherent) was coupled into the detection optical train using a dichroic long-pass mirror (505DRLP, Omega Optical) and focused on the fluidic channel using a microscope objective (40×, 0.6 NA, Newport). Fluorescence emission was collected by the same objective, spectrally filtered using a bandpass filter (535DF60, Omega Optical) and spatially filtered with a 100-μm pinhole prior to illuminating a PMT detector (H7827, Hamamatsu Photonics). Fluorescence data were acquired at 0.1 Hz and processed with a five-point averaging filter.
Fluid handling on the chip was accomplished by three valve actuation programs: prime, mix, and isolate. The prime program consists of opening valves a, b, c, in, and out, then depressurizing the sample collection vial (Figure 1C). This draws reagent in through the E and S sample lines, flushing the entire circuit with reagent and depositing the waste into the collection vial. The mix program pumps fluid around the mixing loop by serially actuating valves a, b, and c. The wait time between valve actuations is 300 ms for slow mixing during incubation steps and 80 ms for rapid mixing following the introduction of fresh reagents (Figure 1D). The isolate program consists of opening valves a, b, in, and out and depressurizing the sample collection vial. This flushes reagent through the outside portion of the mixing loop containing a and b, while isolating an aliquot of material in the region containing c and bounded by in and out. Executing isolate followed by rapid mix results in 10-fold dilution of the carryover materials and constitutes one iteration of the continuous evolution process.
Prior to use, each new circuit was flushed with a rinse solution containing 50 mM EPPS (pH 7.5) and 0.1% IGEPAL-CA630, executing prime for 10 s every min over a 60-min period. During this process, the stage temperature was stabilized at 38.5 °C to maintain 37 °C within the device (previously calibrated). Continuous evolution was initiated on the device by immersing both the E and S lines in the solution of starting pool RNA. The circuit was primed until the fluorescence intensity of the fluorescein tracer had stabilized. Then the E and S lines were immersed in the rinse solution, and the circuit executed a 60-s isolate program to flush the input lines and isolate an aliquot of the starting pool of RNA within the circuit.
The collection vial was repressurized, and the E line was immersed in a solution containing 20 U/μl Superscript II reverse transcriptase (Invitrogen), 5 U/μl T7 RNA polymerase, 0.001 U/μl yeast inorganic pyrophosphatase (Sigma-Aldrich), 15 mM MgCl2, 50 mM KCl, 50 mM EPPS (pH 7.5), 4 mM DTT, 0.1% IGEPAL-CA630, and 6 μM TO-PRO-1 (Invitrogen). The S line was immersed in a separate solution containing the oligonucleotide substrate, 5 μM cDNA primer having the sequence 5′-AGTCCCTCCATGCTCGT-3′, 4 mM each NTP, 0.4 mM each dNTP, 15 mM MgCl2, 50 mM KCl, 50 mM EPPS (pH 7.5), 4 mM DTT, 0.1% IGEPAL-CA630, and 6 μM TO-PRO-1. The oligonucleotide substrate, which was present at progressively lower concentrations during the course of evolution, had the sequence 5′-CCGAAGCCTGGGATCAATAATACGACTCACUAUA-3′ (T7 RNA polymerase promoter sequence underlined; RNA residues in bold). Aqueous glycerol (50%) was added to the substrate-containing solution to match the viscosity of the polymerase-containing solution. Both solutions were thermoelectrically cooled to preserve the activity of the polymerase enzymes.
The collection vial was again depressurized to draw a 1:1 mixture of the substrate- and polymerase-containing solutions, which primed the input lines and filled the circuit in the region containing a and b and bounded by in and out. The circuit then was directed to execute rapid mix for 40 s, followed by slow mix during the incubation phase of the evolution procedure. The slow cyclic mixing prevented photobleaching of the fluorescent dye. The intercalating dye TO-PRO-1 was chosen for its superior enhanced fluorescence quantum yield upon binding double-stranded nucleic acids [24]. The background fluorescence prior to RNA amplification was typically ∼10 kCPS. The time required for 10-fold growth of the starting RNA enzyme in the presence of 1 μM substrate was determined to be 40 min. This established the incubation time for the first round and set the threshold for dilution at 30 kCPS. Subsequently, whenever the detector registered 30 kCPS, the circuit was directed to execute the steps of isolate, delivery of fresh reagents with rapid mix, and incubation with slow mix.
The population of RNA enzymes underwent continuous evolution on the chip until the time between successive log dilutions either decreased below 2 min or exhibited no further improvement. Materials from the last five iterations were collected in a fresh vial that contained 50 μl of 0.1 N NaOH. This mixture was incubated at 95 °C for 10 min to hydrolyze the RNA components, and the remaining DNA then was amplified by both standard PCR (as a control) and error-prone PCR. The primers for PCR amplification had the sequence 5′-AGTCCCTCCATGCTCGT-3′ and 5′-CCGAAGCCTGGGATCAATAA-3′. The sample collected after iteration 198 was mutagenized at a frequency of ∼10% per nucleotide position using a hypermutagenic PCR protocol [25]. Samples collected after iterations 280, 363, and 428 were mutagenized using standard error-prone PCR [12]. The PCR products were transcribed, and the resulting RNAs were purified by PAGE, desalted, and resuspended in a solution containing 15 mM MgCl2, 50 mM KCl, 50 mM EPPS (pH 7.5), 4 mM DTT, 0.1% IGEPAL-CA630, and 0.1 μM fluorescein, which was used to start the next set of iterations on the chip. The concentration of RNA in the start solution was 100 nM following iterations 198 and 280, 20 nM following iteration 363, and 10 nM following iteration 428, thereby maintaining the substrate in excess of the RNA enzyme throughout the evolution process.
The PCR products obtained following iterations 198 and 500 were cloned and sequenced. An individual corresponding to the consensus sequence of 10 clones that were sequenced after iteration 500 was PCR amplified, transcribed in the presence of [α-32P]ATP, purified by PAGE, and desalted. Its catalytic activity was measured in the presence of 10 mM MgCl2, 50 mM KCl, and 50 mM EPPS (pH 7.5) at 37 °C, determining the observed rate constant in the presence of 10 nM enzyme and varying concentrations of substrate. Reactions were initiated by adding equal volumes of enzyme and substrate solutions, each containing all of the other reaction components and pre-equilibrated at 37 °C. Aliquots were taken at various times and quenched by the addition of 15 mM EDTA. For very short reaction times (<5 s), the reaction was carried out in a quench-flow apparatus (KinTek), using separate syringes to deliver the enzyme, substrate, and quench solutions.
The reaction products were separated by PAGE and quantitated using a PharosFX molecular imager (Bio-Rad). Biphasic kinetics were observed at all substrate concentrations. The overall maximum extent of the reaction was determined empirically by measuring the fraction reacted at 2- and 3-h time points. The data were fit to the equation: F(t) = Fmax – A1e–k1t – A2e–k2t, where Fmax is the maximum extent, A1 and k1 are the amplitude and rate of the initial fast phase, and A2 and k2 are the amplitude and rate of the subsequent slow phase, respectively. The amplitude of the fast phase typically was 0.6–0.7 and the overall maximum extent typically was 0.9. A saturation plot was constructed by plotting k1 as a function of substrate concentration, and these data were fit to the Michaelis-Menten equation to determine kcat and Km.
Variants of the starting and final evolved enzymes that contained different combinations of the four critical mutations were prepared by PCR amplification, using appropriate primers to introduce the desired mutations. The PCR products were transcribed in the presence of [α-32P]ATP, purified by PAGE, and desalted. The maximum extent of reaction and observed rate constant were determined in the presence of 0.1 μM substrate for each variant. In addition, a variant of the starting enzyme that contained the M3 mutations and a variant of the evolved enzyme that lacked the M3 mutations were subject to formal kinetic analysis, as described above.
|
10.1371/journal.ppat.1003630 | HTLV-1 bZIP Factor Induces Inflammation through Labile Foxp3 Expression | Human T-cell leukemia virus type 1 (HTLV-1) causes both a neoplastic disease and inflammatory diseases, including HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP). The HTLV-1 basic leucine zipper factor (HBZ) gene is encoded in the minus strand of the proviral DNA and is constitutively expressed in infected cells and ATL cells. HBZ increases the number of regulatory T (Treg) cells by inducing the Foxp3 gene transcription. Recent studies have revealed that some CD4+Foxp3+ T cells are not terminally differentiated but have a plasticity to convert to other T-cell subsets. Induced Treg (iTreg) cells tend to lose Foxp3 expression, and may acquire an effector phenotype accompanied by the production of inflammatory cytokines, such as interferon-γ (IFN-γ). In this study, we analyzed a pathogenic mechanism of chronic inflammation related with HTLV-1 infection via focusing on HBZ and Foxp3. Infiltration of lymphocytes was observed in the skin, lung and intestine of HBZ-Tg mice. As mechanisms, adhesion and migration of HBZ-expressing CD4+ T cells were enhanced in these mice. Foxp3−CD4+ T cells produced higher amounts of IFN-γ compared to those from non-Tg mice. Expression of Helios was reduced in Treg cells from HBZ-Tg mice and HAM/TSP patients, indicating that iTreg cells are predominant. Consistent with this finding, the conserved non-coding sequence 2 region of the Foxp3 gene was hypermethylated in Treg cells of HBZ-Tg mice, which is a characteristic of iTreg cells. Furthermore, Treg cells in the spleen of HBZ-transgenic mice tended to lose Foxp3 expression and produced an excessive amount of IFN-γ, while Foxp3 expression was stable in natural Treg cells of the thymus. HBZ enhances the generation of iTreg cells, which likely convert to Foxp3−T cells producing IFN-γ. The HBZ-mediated proinflammatory phenotype of CD4+ T cells is implicated in the pathogenesis of HTLV-1-associated inflammation.
| Viral infection frequently induces tissue inflammation in the host. HTLV-1 infection is associated with chronic inflammation in the CNS, skin, and lung, but the inflammatory mechanism is not fully understood yet. Since HTLV-1 directly infects CD4+ T cells, central player of the host immune regulation, HTLV-1 should modulate the host immune response not only via viral antigen stimulation but also via CD4+ T-cell-mediated immune deregulation. It has been reported that Foxp3+CD4+ T cells are increased in HTLV-1 infection. It remains a central question in HTLV-1 pathogenesis why HTLV-1 induces inflammation despite of increase of FoxP3+ cells, which generally possess immune suppressive function. We have elucidated here that most of the increased Foxp3+ cells in HBZ-Tg mice or HAM/TSP patients is not thymus-derived naturally occurring Treg cells but induced Treg cells. Since the iTreg cells are prone to lose FoxP3 expression and then become cytokine-producing cells, the increase of iTreg cells could serve as a source of proinflammatory CD4+ T cells. Thus HTLV-1 causes abnormal CD4+ T-cell differentiation by expressing HBZ, which should play a crucial role in chronic inflammation related with HTLV-1. This study has provided new insights into the mechanism of chronic inflammation accompanied with viral infection.
| Human T-cell leukemia virus type 1 (HTLV-1) is known to be the causal agent of a neoplastic disease of CD4+ T cells, adult T-cell leukemia (ATL) [1]. In addition, this virus perturbs the host immune system, causing inflammatory diseases and immunodeficiency. Inflammatory diseases associated with HTLV-1 includeHTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) [2], [3], uveitis [4], [5], alveolitis [6], infective dermatitis [7] and myositis [8]. Increased expression of inflammatory cytokines and immune response to the Tax antigen has been proposed as mechanisms of these inflammatory diseases [9]. However, the detailed mechanisms of inflammation remain elusive.
The HTLV-1 bZIP factor (HBZ) gene is encoded in the minus strand of the provirus and consistently expressed in ATL cases and HTLV-1-infected individuals [10]. In vitro and in vivo experiments have shown that the HBZ gene promotes the proliferation of T cells and increases their number [10], [11]. Recently, we reported that HBZ transgenic (HBZ-Tg) mice develop both T-cell lymphomas and inflammatory diseases [12]. In HBZ-Tg mice, we found that the number of CD4+ T cells expressing Foxp3, a master molecule for regulatory T (Treg) cells, was remarkably increased. HBZ induces transcription of the Foxp3 gene via interaction with Smad2/3 and a co-activator, p300, resulting in an increased number of Foxp3+ T cells [13]. Concurrently, HBZ interacts with Foxp3 and decreases the immune suppressive function [12]. This interaction could be a mechanism of the inflammatory phenotype observed in HBZ-Tg mice. However, detailed mechanisms to induce inflammation by HBZ remain unsolved.
Treg cells suppress excessive immune responses, and control the homeostasis of the immune system [14]. Foxp3 is considered a marker of Treg cells, yet several lines of evidence have shown that there is heterogeneity within Foxp3+cells [15]. Natural Treg (nTreg) cells are generated in the thymus while induced Treg (iTreg) cells are induced in the peripheral lymphoid organs. It has been reported that Treg cells that have lostFoxp3 expression (exFoxp3 T cells) produce interferon-γ (IFN-γ), indicating thatFoxp3+ Treg cells are not terminally differentiated cells but susceptible to conversion into effector T cells according to their environment [16]. Recently, Miyao et al. have reported that Foxp3+ T cells induced by activation exhibit transient Foxp3 expression, and become exFoxp3 T cells [17]. Even though the plasticity of Treg cells remains controversial [18], these reports suggest that Foxp3+ T cells possess not only suppressive function but also proinflammatory attributes.
In this study, we found that iTreg cells increased in HBZ-Tg mice and that Treg cells of HBZ-Tg mice tend to lose Foxp3 expression, leading to increased IFN-γ-expressing proinflammatory cells. Cell adhesion and migration are enhanced in CD4+ T cells of HBZ-Tg mice. Thus, these HBZ-mediated abnormalities of CD4 T cells play critical roles in inflammatory diseases caused by HTLV-1.
We have reported that HBZ-Tg mice develop both T-cell lymphoma and inflammatory diseases including dermatitis and alveolitis [12]. To further study the inflammatory changes affecting HBZ-Tg mice, we analyzed various tissues and organs in detail. In HBZ-Tg mice, moderate lymphoid cell infiltration was detected in the peri-bronchial space of the lung (Figure 1A), the peri-follicular area of the skin (Figure 1B), the mucosa of the small intestine (Figure 1C), and the mucosa of the colon (Figure 1D). Meanwhile, there was no obvious evidence of inflammation in liver, kidney or spinal cord. In non-Tg littermates, infiltration of lymphoid cells was not observed in skin, lung or intestine. These findings suggest the inflammatory involvement of multiple tissues and organs in HBZ-Tg mice.
Infiltration of lymphocytes into various tissues suggests that the lymphocytes of HBZ-Tg mice have increased adhesive ability. We first studied the expression of LFA-1, which is a heterodimer of CD11a and CD18. As shown in Figure 2A, both CD11a and CD18 were upregulated on HBZ-Tg CD4+ T cells of spleen, lung and lymph nodes compared with CD4+ T cells from non-Tg mice. In addition, the expression of CD103 (alpha E integrin) on HBZ-Tg CD4+ T cells was also higher than that on non-Tg CD4+ T cells. These findings suggest an increased adhesive capability of CD4+ T cells in HBZ-Tg mice. Immunohistochemical analyses of lung and intestine of HBZ-Tg mice confirmed increased expression of these molecules, particularly CD18 (Figure 2B, C).
Expression of CD11a, CD18 and CD103 was also studied in HAM/TSP patients. In addition to healthy donors, we analyzed expression of these molecules on HTLV-1 infected cells that are identified using anti-Tax antibody. As shown in Figure 2D, CD11a and CD18 expression of CD4+Tax+ T cells was upregulated compared with CD4+ T cells from healthy donors and CD4+Tax− T cells of HAM/TSP patients while expression of CD103 was not different among these cells. These results show that enhanced expression of LFA-1 is also observed in HTLV-1 infected cells in HAM/TSP patients.
We next investigated adhesion of CD4+ T cells to ICAM-1, since ICAM-1 is critical for lymphocyte migration and adhesion to vascular epithelial cells in an inflammatory lesion. We isolated CD4+ T cells from non-Tg or HBZ-Tg splenocytes, placed them on ICAM-1-coated 96-well plates, and evaluated cell adhesion activity to ICAM-1. CD4+ T cells from HBZ-Tg mice showed increased adhesion in the absence of stimulation, while no difference was found when cells were stimulated by anti-CD3 antibody (Figure 3A). Furthermore, we evaluated the migration activity of CD4+ T cells on ICAM-1-coated plates. To induce cell migration, we stimulated CD4+ T cells with CCL22 as reported previously [19]. Cell migration of HBZ-Tg CD4+ T cells was also increased compared with migration of non-Tg CD4+ T cells (Figure 3B). These results demonstrate an infiltrative phenotype of CD4+ T cells in HBZ-Tg mice.
Infiltration of LFA-1 expressing T cells into various tissues suggests that ICAM-1 expression is enhanced. Indeed, expression of ICAM-1 was increased in intestine of HBZ-Tg mice (Figure 2B). Enhanced migration of CD4+ T cells suggests involvement of chemokine(s)-chemokine receptor for HBZ-Tg mice. We analyzed expression of chemokine receptors on CD4+ T cells of HBZ-Tg mice. As shown in Figure 3C, CXCR3 expression of CD4+ splenocytes was increased while expression of CCR5 and CCR7 were not different compared with control mice (Figure S1). CXCR3 expression of CD4+ T cells was upregulated in both lung and lymph node (Figure 3C). Although the ligands for CXCR3, CXCL9 and CXCL10, were not increased in the sera of HBZ-Tg mice (Figure S1), CXCR3 might be implicated in infiltration of CD4+ T cells.
To elucidate the mechanism of the pro-inflammatory phenotype observed in HBZ-Tg mice, we investigated cytokine production in CD4+ T cells of the spleen. After stimulation by PMA/ionomycin, production of IFN-γ was increased in CD4+ T cells while that of TNF-α was suppressed (Figure 4A). There were no significant differences between HBZ-Tg mice and non-Tg mice in IL-2, IL-4 and IL-17 production by CD4+ T cells. We have reported that the number of Foxp3+CD4+ Treg cells is increased in HBZ-Tg mice. Therefore, we simultaneously stained both intracellular cytokines and Foxp3 to distinguish the cytokine production of CD4+Foxp3− T cells from that of CD4+Foxp3+ T cells. Production of TNF-α, IL-17 and IL-2 was slightly increased in CD4+Foxp3+ T cells of HBZ-Tg mice (Figure 4B, C). Since Foxp3 suppresses production of cytokines [19], and HBZ impairs function of Foxp3 [12], HBZ-mediated impairment of Foxp3 function might be a mechanism of this increased expression of these cytokines. However, TNF-α production was suppressed in CD4+ Foxp3− T cells and total CD4+ T cells (Figure 4A, C). In particular, IFN-γ production of splenic CD4+Foxp3− T cells from HBZ-Tg mice was remarkably increased compared with those from non-Tg mice (Figure 4B). We also studied IFN-γ production in CD4+ T cells of PBMCs and lung-infiltrating lymphocytes. The production of IFN-γ was remarkably increased in PBMC and lung from HBZ-Tg mice (Figure 4D). Taken together, these results suggest that increased IFN-γ production, especially in CD4+Foxp3− T cells, is related to the chronic inflammation observed in HBZ-Tg mice. Immunohistochemical analyses also showed that IFN-γ production was increased in both lung and intestine of HBZ-Tg mice (Figure 2B, C).
We have reported that HBZ enhances the transcription of the Foxp3 gene in cooperation with TGF-ß, leading to an increased number of Treg cells in vivo [12], [13]. Two types of Treg cells have been reported: natural Treg (nTreg) cells and induced Treg (iTreg) cells in CD4+Foxp3+ cells. The expression of Helios, a member of the Ikaros family of transcription factors, is considered a marker of nTreg cells [20]. To determine which Treg cell population is increased in HBZ-Tg mice, we analyzed the expression of Helios. Expression of Helios in CD4+Foxp3+ T cells in HBZ-Tg mice was lower than that in non-Tg mice (Figure 5A, C), suggesting that the number of iTreg cells is increased in HBZ-Tg mice. A higher proportion of CD4+Foxp3+Helioslow cells were found in the lungs of HBZ-Tg mice (Figure S2). Next, we analyzed the expression of Helios in Treg cells from HAM/TSP patients. As shown in Figure 5 B and D, Helios expression of Treg cells in HAM/TSP patients was lower than that of Treg cells in healthy controls. We also analyzed Helios expression in Foxp3+ T (nTreg) cells of the thymus. The level of Helios expression in nTreg cells in HBZ-Tg mice was equivalent to that of non-Tg mice (Figure S3). These data collectively suggest that the iTreg cell population is increased not only in HBZ-Tg mice, but also in HAM/TSP patients.
Recent studies have reported that Helios expression is not always associated with nTreg cells [21]–[23]. A previous study reported that conserved non-coding DNA sequence (CNS) elements in the Foxp3 locus play an important role in the induction and maintenance of Foxp3 gene expression [24]. Among these elements, CNS2, methylated in iTreg cells, was suggested to be responsible for the lack of stable expression of Foxp3 in these cells [24]. This region is not methylated in Helios- nTreg cells, indicating that unmethylation of this region is a suitable marker of nTreg cells [21]. Therefore, we sorted the Treg fraction from HBZ-Tg or non-Tg mice splenocytes, extracted genomic DNA, and determined the DNA methylation status in the CNS2 region of the Foxp3 gene. The results revealed that in HBZ-Tg CD4+Foxp3+ T cells, the CNS2 region had a higher methylation status than in non-Tg CD4+Foxp3+ cells (Figure 6), indicating that the increase in CD4+Foxp3+ cells in HBZ-Tg mice indeed mostly consists of iTreg cells.
Recent studies have revealed that CD4+Foxp3+ T cells are not terminally differentiated but have the plasticity to convert to other T cell subsets [25]. When Treg cells lose the expression of Foxp3 (exFoxp3 T cells), such cells produce pro-inflammatory cytokines [16]. It has been reported that Foxp3 expression in nTreg cells is stable but that it is not in iTreg cells [15]. These findings suggest that in HBZ-Tg mice, which have greater numbers of iTreg cells as shown in this study, Foxp3 expression in these cells tends to diminish, letting these cells acquire an effector phenotype associated with the production of pro-inflammatory cytokines such as IFN-γ. To investigate this possibility, we sorted Treg cells from the spleens of HBZ-Tg or non-Tg mice based on their expression of CD4, CD25 and GITR; cultured them for 7 days; and analyzed Foxp3 expression by flow cytometry. After 7 days in culture, the percentage of Foxp3+ T cells diminished remarkably in HBZ-Tg mice compared with non-Tg mice (Figure 7A, B). We investigated the production of IFN-γ at this point, and found that it was increased in Foxp3− T cells from HBZ-Tg mice compared with those from non-Tg mice (Figure 7C). In sharp contrast to this finding, Foxp3 expression of nTreg cells did not change in CD4+ thymocytes of HBZ-Tg mice (Figure 7D). Collectively, these data indicate thatFoxp3 expression in nTreg cells is stable in HBZ-Tg mice, while most of the Treg cells in the periphery are iTreg cells. The enhanced generation of exFoxp3 T cells in the periphery is a possible mechanism of the increase in IFN-γ -producing Foxp3− T cells in HBZ-Tg mice. We reported that HBZ induced the Foxp3 gene transcription via interaction with activation of TGF-β/Smad pathway [13]. Reduced expression of Foxp3 in HBZ-Tg CD4+Foxp3− T cells might be caused by low HBZ expression in that cell population. To investigate this possibility, we analyzed the relationship between HBZ and Foxp3 expression in CD4+ T cells of HBZ-Tg mice. We isolated CD4+CD25+GITRhigh T cells as Foxp3+ T cells, and CD4+CD25−GITRlow T cells as Foxp3− T cells from HBZ-Tg mice. Although Foxp3+ T cells are contaminated in CD4+CD25−GITRlow T cells, level of the Foxp3 gene transcript was much higher in CD4+CD25highGITRhigh T cells (Figure S4). However, level of HBZ transcript was no different among these cells, indicating that level of HBZ expression is not associated with reduced Foxp3 expression.
HTLV-1 is a unique human retrovirus with respect to its pathogenesis, since it causes not only a neoplastic disorder, but also various inflammatory diseases. For most viruses, tissue-damaging inflammation associated with chronic viral infection is generally triggered by the immune response against infected cells, which involves both antigen specific and non-specific T cells that produce pro-inflammatory cytokines, chemokines, and other chemical mediators that promote tissue inflammation [26]. However, this study shows that HTLV-1 can induce inflammation by a different mechanism that does not involve an immune response against infected cells, but instead, involves deregulation of CD4+ T-cell differentiation mediated by HBZ. Since transgenic expression of HBZ does not induce an immune response to HBZ protein itself, the inflammation observed in this study is attributed to an intrinsic property of HBZ-expressing cells.
Studies of the pathogenesis of inflammatory diseases related to HTLV-1 are usually focused on HAM/TSP, since it is the most common inflammatory disease caused by this virus [9]. Two different mechanisms of HAM/TSP pathogenesis have been reported: one mechanism involves the immune response to viral antigens, and another mechanism implicates the proinflammatory attributes of HTLV-1-infected cells themselves. Previous studies reported a strong immune response to Tax in HTLV-1-infected individuals [9], [27]. In lesions of the spinal cord, CD4+ T cells expressing viral gene transcripts were identified by in situ hybridization [28]. The presence of CTLs targeting Tax in cerebrospinal fluid and lesions in the spinal cord suggest an important role of the immune response and the cytokines produced by CTLs in the pathogenesis of HAM/TSP by HTLV-1 [29]. Those studies showed the involvement of the immune response to Tax in the pathogenesis of HAM/TSP. In addition, cell-autonomous production of proinflammatory cytokines by HTLV-1-infected cells has been reported. HTLV-1-transformed cells produce a variety of cytokines, including IFN-γ, IL-6, TGF-ß, and IL-1α [30]. It was speculated that Tax was responsible for the enhanced production of these cytokines. In this study, we have shown a new role of HBZ in inflammatory diseases. CTLs against HBZ have been reported in HTLV-1 carriers and HAM/TSP patients; this immune response might be involved in inflammation caused by HTLV-1 [31]. However, an immune response to HBZ does not occur in HBZ-Tg mice, indicating that the proinflammatory phenotype of HBZ expressing T cells is sufficient to cause the inflammation.
Does HBZ induce IFN-γ production in CD4+ T cells? HBZ and Tax have contradictory effects on many pathways. For example, Tax activates both the canonical and non-canonical NF-κB pathways, while HBZ suppresses the canonical pathway [32], [33]. Conversely, HBZ activates TGF-ß/Smad pathway, while Tax inhibits it [13], [34], [35]. Tax activates the IFN-γ gene promoter, whereas HBZ suppresses the transcription of the IFN-γ gene through inhibition of AP-1 and NFAT, which are critical for IFN-γ gene transcription [36]. These findings collectively suggest that the enhanced production of IFN-γ is not due to a direct effect of HBZ, but may be attributed to the increased presence exFoxp3 T cells triggered by HBZ as shown in this study. Recent studies reported that exFoxp3 T cells produce higher amount of IFN-γ [17], [37]. This indicates that increased production of IFN-γ in exFoxp3 T cells surpasses the suppressive function by HBZ. In this study, HBZ inhibited the production of TNF-α as we reported [36], indicating that enhanced production is specific to IFN-γ. However, it remains unknown how the production of IFN-γ is enhanced in exFoxp3 T cells.
We have shown that the Foxp3+ T cells of HBZ-Tg mice tend to lose Foxp3 expression and change into IFN-γ-producing proinflammatory cells. This observation makes sense in the light of several other studies on Treg cells. It was reported thatFoxp3+ T cells convert to Foxp3− T cells [37]–[39]. Recently, Miyao et al. reported that Foxp3 expression of peripheral T cells induced by activation is promiscuous and unstable, leading to conversion to exFoxp3 T cells [17]. Peripheral induced Foxp3+ T cells show lower expression of CD25 and Helios, which corresponds to the phenotype we observed in the Foxp3+ T cells of HBZ-Tg mice. Thus it is likely that HBZ induces unstable Foxp3 expression and generates iTreg cells, which then convert to exFoxp3 T cells with enhanced production of IFN-γ as shown in this study. It has recently been reported that CD4+CD25+CCR4+ T cells in HAM/TSP patients were producing extraordinarily high levels of IFN-γ, when compared to cells of healthy donors. These findings are consistent with those of this study. Importantly, the frequency of these IFN-γ-producing CD4+CD25+CCR4+Foxp3− T cells was increased and found to be correlated with disease severity in HAM/TSP patients [40]. In addition, it has been reported that HBZ expression is correlated with the severity of HAM/TSP [41]. Thus, the presence of abnormal HBZ-induced IFN-γ-producing cells is a plausible mechanism that leads to inflammation in HAM/TSP patients.
FOXP3 expression is detected in two thirds of ATL cases, suggesting that ATL cells originate from Treg cells in these cases [42], [43]. Human FOXP3+ T cells have been divided into three subgroups based on their functions and surface makers: resting Treg cells (rTreg), activated Treg (aTreg) cells, and FOXP3lownon-suppressive T cells [44]. Recently, we reported that HTLV-1 infection is frequently detected in Treg cells, which include FOXP3low non-suppressive T cells and FOXP3high activated Treg cells, and concordantly, some ATL cells also belong to the population of FOXP3low non-suppressive T cells [44], [45]. This suggests that HTLV-1 increases the population of aTreg and FOXP3low non-suppressive T cells and induces leukemia/lymphoma of these cells. It is thought that most of nTreg are resting and activated Treg cells and iTreg cells contain both aTreg cells and Foxp3low non-suppressive T cells in human. The CNS2 region in the Foxp3 locus is highly methylated in FOXP3low non-suppressive T cells [44], like we report for the iTreg cells of HBZ-Tg mice. It is likely that a fraction of FOXP3low non-suppressive T cells lose FOXP3 expression and change to FOXP3− proinflammatory T cells as reported in HAM/TSP patients [40], suggesting that the finding of this study is indeed the case in HTLV-1 infection.
It has been widely believed that nTreg cells represent a highly stable lineage in which few cells lose Foxp3 expression under normal homeostatic conditions [46]. In contrast, small subsets of CD25−Foxp3+ Treg cells have recently been reported to be unstable and to rapidly lose Foxp3 expression after transfer into a lymphopenic host [16]. The CNS2 sequence is methylated in iTreg cells [24]. Consistent with this finding, CNS2 was heavily methylated in Treg cells of HBZ-Tg mice, indicating that Treg cells in HBZ-Tg mice largely belong to the iTreg cell subset. Foxp3 expression of CD4+ thymocytes in HBZ-Tg mice did not decrease after in vitro culture, a fact which shows that loss of Foxp3 expression is not a direct effect of HBZ, but is due to the increased number of iTreg cells converting to exFoxp3 cells. Recently, it was reported that Foxp3+ T cells without suppressive function convert to exFoxp3 T cells [17]. We recently reported that HBZ enhances Foxp3 gene transcription by activating the TGF-ß/Smad pathway [13]. Collectively, it is likely that HBZ increases Foxp3+ T cells in HBZ-Tg mice and most of Foxp3+ T cells are iTreg and/or non-suppressive Foxp3+ T cells. Foxp3 expression in HBZ-Tg mice is unstable as shown in this study, and such cells easily convert to exFoxp3 T cells, which produce excess amounts of IFN-γ, leading to inflammation.
Helios expression has been reported to be high in nTreg cells, and low in iTreg cells [20]. This study showed that Helios expression in CD4+Foxp3+ cells of HBZ-Tg mice was low although it was higher than control iTreg cells. Recently, it has been reported that stimulation enhances Helios expression of iTreg cells, which might account for increased Helios expression in CD4+Foxp3+ cells of HBZ-Tg mice compared with control iTreg cells [22]. In particular, inflammation caused by HBZ expression might increase Helios expression of iTreg cells of HBZ-Tg mice. In addition, it has been reported that Helios is not expressed in a part of nTreg cells and its expression is induced in iTreg cells, indicating that only Helios expression cannot discriminate nTreg cells from iTreg cells [21]–[23]. However, CNS2 is not methylated in Helios− nTreg cells, which shows that the methylation status of CNS2 is critical [21]. In this study, analysis of DNA methylation of CNS2 confirms that most of CD4+Foxp3+ cells in HBZ-Tg mice are iTreg cells. Importantly, the similar pattern of Heilos expression was observed in HAM/TSP patients.
The present study has demonstrated that HBZ-Tg mice develop inflammation in the intestines, skin and lungs. These tissues are always exposed to extrinsic antigens and commensal microbes, where Treg cells are critical for maintaining the homeostasis of the host immune system. In addition to the increased production of IFN-γ by HBZ-expressing cells, it is likely that the cell adhesion attributes of these cells also play a role in their pro-inflammatory phenotype. Treg cells express a variety of molecules that are important for cell adhesion, including LFA-1, CCR4, and CD103 [12]. We have shown that these molecules are also present on HBZ-expressing CD4+ T cells. In this study, we showed that HBZ increases the number of iTreg cells, which subsequently convert into exFoxp3 T cells. The proinflammatory phenotype of HBZ-expressing T cells indicates that HBZ plays an important role in the inflammatory diseases caused by HTLV-1.
In conclusion, HBZ-Tg mice developed chronic inflammation accompanied with hyper IFN-γ production, which is consistent with the findings in HAM/TSP patients. CD4+Foxp3+ T cells, especially iTreg cells, were increased in HBZ-Tg mice. The expression of Foxp3 was not stable and tended to be lost, which resulted in the enhanced generation of exFoxp3 cells producing IFN-γ. This could be a mechanism for the development of chronic inflammation in HBZ-Tg mice and HTLV-1-infected individuals.
Transgenic mice expressing HBZ under the murine CD4 promoter have been previously described [12]. Genotypes were determined by means of PCR on mouse ear genomic DNA. All the mice were used at 10–20 weeks of age. Animal experimentation was performed in strict accordance with the Japanese animal welfare bodies (Law No. 105 dated 19 October 1973 modified on 2 June 2006), and the Regulation on Animal Experimentation at Kyoto University. The protocol was approved by the Institutional Animal Research Committee of Kyoto University (permit number: D13-02). All efforts were made to minimize suffering. A total of 10 HAM/TSP patients and 10 healthy donors participated in this study. Written informed consents were obtained from all the subjects in accordance with the Declaration of Helsinki as part of a clinical protocol reviewed and approved by the Institutional Ethics Committee of Kyoto University (approval number: 844). Blood samples were collected from the subjects and peripheral blood mononuclear cells (PBMC) were isolated by Ficoll-Paque Plus (GE Healthcare Bio-Sciences) density gradient centrifugation.
Production of recombinant mouse ICAM-1 was performed as described previously [47]. A 96-well plate was coated with 100 µl/well of 0.25 µg/ml mouse mICAM-1-Ig (R&D Systems) at 4°C overnight, followed by blocking with 1% BSA for 30 min. Mouse CD4+ cells were labeled with 2′, 7′-bis-(2-carboxyethyl)-5-(and-6) carboxyfluorescein (Molecular Probes, Inc.), suspended in RPMI 1640 containing 10 mM HEPES (pH 7.4) and 10% FBS, transferred into the coated wells at 5×104 cells/well and then incubated at 37°C for 30 min. Non-adherent cells were removed by aspiration. Input and bound cells were quantitated in the 96-well plate using a fluorescence concentration analyzer (IDEXX Corp.).
Random cell migration was recorded at 37°C with a culture dish system for live-cell microscopy (DT culture dish system; Bioptechs). Thermoglass-based dishes (Bioptechs) were coated with 0.1 µg/ml mouse ICAM-1. CD4+ mouse splenocytes were loaded in the ICAM-1-coated dish, and the dish was mounted on an inverted confocal laser microscope (model LSM510, Carl Zeiss MicroImaging, Inc.) Phase-contrast images were taken every 15 s for 10 min. The cells were traced and velocity was calculated using ImageProR Plus software (Media Cybernetics).
Single-cell suspensions of mouse spleen, lung or PBMC or human PBMC were made in RPMI 1640 medium supplemented with 10% FBS. To detect Tax, CD8+ cells were depleted from human PBMC using the BD IMAG cell separation system with the anti-human CD8 Particles-DM (BD Pharmingen) according to the manufacturer's directions and then the cells were cultured for 6 hours. Surface antigen expression was analyzed by staining with the following antibodies: anti-mouse CD4 (RM4-5), CD11a (2D7), CD18 (C71/16) or CD103 (M290) (all purchased from BD Pharmingen) or anti-human CD4 (RPA-T4), CD11a (HI111), CXCR3 (G025H7) (all purchased from BioLegend), CD18 (6.7), CD103 (Ber-ACT8) (all purchased from BD Pharmingen). For intracellular cytokine staining, cells were pre-stimulated with 20 ng/ml phorbolmyristate acetate (PMA, NacalaiTesque), 1 µM ionomycin (NacalaiTesque) and Golgi plug (BD Pharmingen) for 4 h prior to surface antigen staining. After this stimulation period, cells were fixed and permeabilized with Fixation/Permeabilization working solution (eBioscience) for 30 min on ice and incubated with antibodies specific for the following cytokines: IFN-γ (XMG 1.2), IL-17 (TC11-18H10), IL-2 (JES6-5H4) (all BD Pharmingen), TNF-α (MP6-XT22, eBioscience) and IL-4 (11B11, eBioscience). Intracellular expression of mouse Foxp3 (FJK-16s, eBioscience), human FoxP3 (PCH101, eBioscience), Tax (MI73), human IFN-γ (4SB3, BD Pharmingen) and Helios (22F6, BioLegend) was detected following the protocol for cytokine staining. Dead cells were detected by pre-staining the cells with the Live/dead fixable dead cell staining kit (Invitrogen). Subsequently, the cells were washed twice, and analyzed by FACS CantoII with Diva software (BD Biosciences).
Mouse tissue samples were either fixed in 10% formalin in phosphate buffer and then embedded in paraffin or frozen in embedding medium Optimal Tissue-TeK (SAKURA Finetek Japan). Hematoxylin and eosin staining was performed according to standard procedures. Tissue sections prepared from the frozen samples were also stained with anti-mouse IFN-γ (RMMG-1, Abcam), CD11a (M17/4, BioLegend), CD18 (N18/2, BioLegend), CD103 (M290, BD Pharmingen) and CD54 (ICAM-1)(YN1/1.7.4, BioLegend). Images were captured using a Provis AX80 microscope (Olympus) equipped with an OLYMPUS DP70 digital camera, and detected using a DP manager system (Olympus).
The α chemokines CXCL9 and CXCL10 were analyzed using an enzyme linked immunosorbent assay (ELISA). For α chemokines, capture and detection antibody concentrations were optimized using recombinant chemokines from R&D Systems Inc. (Minneapolis, MN, U.S.A.) according to the manufacturer's guidelines.
Genomic DNA was extracted from sorted Treg cells as described below. One mg of genomic DNA (10 µl) was denatured by the addition of an equal volume of 0.6 N NaOH for 15 min, and then 208 µl of 3.6 M sodium bisulfite and 12 µl of 1 mM hydroxyquinone were added. This mixture was incubated at 55°C for 16 hours to convert cytosine to uracil. Treated genomic DNA was subsequently purified using the Wizard clean-up system (Promega), precipitated with ethanol, and resuspended in 100 µml of dH2O. Sodium bisulfite-treated genomic DNAs (50 ng) were amplified with primers targeting the specified DNA regions, and then PCR products were subcloned into the pGEM-T Easy vector (Promega) for sequencing. Sequences of 10 clones were determined for each region using Big Dye Terminator (Perkin Elmer Applied Biosystems) with an ABI 3100 autosequencer. The primers used for nested PCR were as follows:
for the mouse Foxp3 promoter:
mproF, 5′-GTGAGGGGAAGAAATTATATTTTTAGATG-3′;
mproR, 5′-ATACTAATAAACTCCTAACACCCACC-3′;
mproF2, 5′-TATATTTTTAGATGATTTGTAAAGGGTAAA-3′;
mproR2, 5′-ATCAACCTAACTTATAAAAAACTACCACAT-3′.
For mouse Foxp3 intronic CpG:
mintF, 5′-TATTTTTTTGGGTTTTGGGATATTA-3′;
mintR, 5′-AACCAACCAACTTCCTACACTATCTAT-3′;
mintF2, 5′-TTTTGGGTTTTTTTGGTATTTAAGA-3′;
mintR2, 5′-TTAACCAAATTTTTCTACCATTAAC-3′.
To sort Treg cells, we isolated mouse splenocytes and resuspended them in FACS buffer for subsequent staining with the following antibodies purchased from BD Pharmingen: anti-mouse CD4 (RM4-5), GITR (DTA-1), CD25 (PC61). CD4+CD25+GITRhigh cells and CD4+CD25−GITRlowcells were sorted as Foxp3+ or Foxp3−cells using FACS AriaII with Diva software (BD Biosciences). To confirm the purity of the sorted Treg cells, we measured the percentage of Foxp3 expression by intracellular staining, as described above. Sorted Treg cells were cultured in RPMI1640 containing 10% FBS, antibiotics, and 50 µM 2-mercaptoethanol (Invitrogen).
Total RNA of sorted cells was extracted with TRIZOL reagent (Invitrogen) according to the manufacturer's instructions. Approximately 200 ng of RNA were used to prepare cDNA using the SuperScript III enzyme (Invitrogen). Levels of HBZ and Foxp3 transcripts were determined with FastStart Universal SYBR Green Master reagent (Roche) in a StepOnePlus real time PCR system (Apllied Biosystems). Data was analyzed by the delta Ct method. The sequence of the primers used were as follows:
HBZ Forward: 5′-GGACGCAGTTCAGGAGGCAC-3′, Reverse: 5′-CCTCCAAGGATAATAGCCCG-3′; Foxp3 Forward: 5′-CCCATCCCCAGGAGTCTTG-3′, Reverse: 5′-ACCATGACTAGGGGCACTGTA-3′; 18S rRNA Forward: 5′-GTAACCCGTTGAACCCCATT-3′, Reverse: 5′- CCATCCAATCGGTAGTAGCG -3′.
|
10.1371/journal.ppat.1000385 | Effective but Costly, Evolved Mechanisms of Defense against a Virulent Opportunistic Pathogen in Drosophila melanogaster | Drosophila harbor substantial genetic variation for antibacterial defense, and investment in immunity is thought to involve a costly trade-off with life history traits, including development, life span, and reproduction. To understand the way in which insects invest in fighting bacterial infection, we selected for survival following systemic infection with the opportunistic pathogen Pseudomonas aeruginosa in wild-caught Drosophila melanogaster over 10 generations. We then examined genome-wide changes in expression in the selected flies relative to unselected controls, both of which had been infected with the pathogen. This powerful combination of techniques allowed us to specifically identify the genetic basis of the evolved immune response. In response to selection, population-level survivorship to infection increased from 15% to 70%. The evolved capacity for defense was costly, however, as evidenced by reduced longevity and larval viability and a rapid loss of the trait once selection pressure was removed. Counter to expectation, we observed more rapid developmental rates in the selected flies. Selection-associated changes in expression of genes with dual involvement in developmental and immune pathways suggest pleiotropy as a possible mechanism for the positive correlation. We also found that both the Toll and the Imd pathways work synergistically to limit infectivity and that cellular immunity plays a more critical role in overcoming P. aeruginosa infection than previously reported. This work reveals novel pathways by which Drosophila can survive infection with a virulent pathogen that may be rare in wild populations, however, due to their cost.
| The fruit fly is commonly used as a model organism to understand the mechanistic nature of the immune response to bacterial pathogens. The fly is also commonly used to understand what immunity costs hosts in terms of other traits such as life span and reproductive success. Here, we examine these two questions together in flies selected for improved defense against the bacterium Pseudomonas aeruginosa. We show that selected flies develop from egg to adult more rapidly than unselected flies. It appears that the selected flies invest more heavily in a wing of the immune system that involves engulfment and walling off of invading bacteria. This investment can also explain the shift in developmental rate, as these two biological pathways are controlled by shared sets of genes. These latter two findings are counter to the conventional wisdom and reveal a costly, but effective, means for the fly to circumvent the virulence of Pseudomonas aeruginosa. This bacterium is normally deadly, as it has specific mechanisms to evade the host immune response. Our work is significant for demonstrating a pathway for flies to survive bacterial infection with Pseudomonas aeruginosa and for offering a reason why such a defense is not normally present in wild populations.
| It costs insects to invest in immunity. Highly immune Drosophila mate less and produce fewer offspring [1],[2], more immune bee colonies are less productive [3], and crickets with heightened immunity exhibit reduced sexual displays and longevity [4]. Recently, it has been shown that resource availability can also play a role in determining the strength and direction of these trade-offs between immunity and life history traits for insects [5]. While it is clear that individual insects vary with respect to their immune performance, only in the fly are we beginning to identify the genetic basis of this phenotypic variation [6]–[8]. With an understanding of which genetic changes confer enhanced immunity we can begin to elucidate how selection drives and balances investment into immunity in general and more specifically into different aspects of the immune response.
The innate immune response of insects is generally classified into cellular and humoral components [6], [9]–[11]. Cellular aspects of defence involve both phagocytosis by hemocytes and encapsulation of pathogens with biotoxic melanin. These aspects of the immune response are constitutively expressed and broad spectrum in target [12]. The key features of the humoral reaction, in contrast, are its inducibility upon exposure to infection and its specificity of response. Selective initiation of the Toll and/or the immune deficiency (Imd) pathways that depend on the specific pathogen, ultimately lead to the production and secretion of different sets of antimicrobial peptides (AMPs) [10], [12]–[14]. A recent study in the beetle, Tenebrio molitor, has suggested a challenge to the conventional wisdom, that the humoral response is the stronger partner of the two arms of the immune response. In the beetle, it appears that the cellular response clears the majority of infecting bacteria in the first hour after infection and that the humoral response acts secondarily to remove any persisting bacteria [15].
Here, in Drosophila melanogaster recently caught from the wild, we have artificially selected for defense against a virulent, opportunistic pathogen, Pseudomonas aeruginosa [16],[17]. In three highly resistant lines we have examined the relationship between correlated changes in life history and patterns of immune gene transcription. In contrast to traditional approaches that tend to compare gene expression of infected with uninfected flies, our microarray experiments have paired selected lines with unselected lines both post infection. The approach has lead to the identification of transcriptional changes that explain the evolved defense response instead of the genetic basis of the induced immune response. The evolved lines exhibited an effective genetic mechanism for defense against a highly virulent pathogen characterized by an increased transcriptional investment in cellular immunity. This genetic change was costly to females in particular in terms of longevity and fecundity. Antibacterial defense also correlated with an increase in developmental rate in both males and females, which was counter to expectation. Expression changes in a handful of genes that participate both in cellular immunity and host development provided a possible mechanism for this positive correlation through the action of pleiotropy.
Three independent lines stemming from a single base population were selected for improved defense against P. aeruginosa infection over 10 consecutive generations. Three additional populations, unexposed to infection, but reared with the same population size bottlenecks served as pair matched controls. In selected lines, the proportion of flies surviving P. aeruginosa infection rose from ∼15% at G1 to ∼30% by G3 (see Figure 1). Survival then increased again to ∼70% at G5 where it remained for the duration of the selection regime. There was a significant effect of selection at both G6 (treatment effect: F1,2 = 426.02, P<0.0023) and G10 (treatment effect: F1,2 = 117.44, P<0.0084), with selected lines showed significantly higher survivorship compared to corresponding controls. There was no sexual dimorphism in survivorship for these two generations G6 (sex effect: F1,4 = 1.68, P = 0.265) and G10 (treatment effect: F1,4 = 0.47, P = 0.531) nor was there any indication of sex-dependent evolution of survival, G6 (sex×treatment: F1,4 = 0.32, P = 0.601) and G10 (sex×treatment: F1,4 = 0.04, P = 0.843). The mean realized heritability of the evolved survival across the three lines was 16.7±1.3% (s.e.m). Unlike survivorship, the time it took for infected flies to die following infection did not change under the selection regime (data not shown). After the selection experiment, all fly lines were passaged without infection for a further 5 generations (G15). In the absence of selection, survival in the selected lines returned to pre-selection baseline levels and was no different from G15 controls (treatment effect: F1,2 = 0.5, P = 0.848) (Figure 1).
To assess the fitness cost of evolved defense in the selected flies, six life-history traits representing major aspects of host fitness were measured at G9. Longevity was quantified by rearing virgin males and females separately and then recording their time to death in days. A general linear model demonstrated there was no sex or sex×treatment effect on longevity (data not shown). While there was no effect of selection on longevity (Figure 2B) in males (t2 = 1.70, P = 0.14) in the absence of infection, a significant reduction (t2 = 4.07, P<0.01) in average lifespan of female flies was observed in selected flies relative to control flies (Figure 2A). A general linear model demonstrated there was no sex or sex×treatment effect on body mass (data not shown). The mean body mass for selected female (1.21±0.010 g, Figure 2C) and male (0.71±0.008 g, Figure 2D) flies were not different (data not shown) from their respective controls, 1.20±0.013 g and 0.69±0.007 g. Selected flies developed from egg to eclosion (Figure 2D) on average ∼12 hours faster (t2 = 13.0, P<0.01) than controls. Mean egg viability (Figure 2F) of the selection lines (54% egg hatch) was lower (t2 = 73.1, P<0.001) than that of controls (78%). Number of offspring produced from a single mating between a pair of virgin flies was recorded as female productivity. The mean number of offspring produced (Figure 2G) in selected lines, in contrast, did not differ when compared to controls (t2 = 3.3, P = 0.08). To assess the effect of selection on male attractiveness, a selected male and a control male were allowed to compete for a female from the base population. The mating success of male flies from selected lines did not differ compared with controls (F1,1 = 0.68, P = 0.56).
Both selected and control lines were infected at G10 and their RNA was extracted for transcriptional profiling experiments. This comparison specifically revealed the changes in expression due to selection for defense. This is in contrast to the traditional approach of comparing infected lines to uninfected, where the question is instead about which genes are induced after infection. A total of 414 (337 up, 77 down) transcripts showed shared patterns of altered expression in all three lines after selection (Figure 3). Expression profiles of S1 and S2 were most similar to one another. Approximately, 69 immune related genes were significantly up-regulated in at least 2 of the 3 selected lines and 46 of these genes showed similar increases across all three lines (Table S1). Eighteen genes with known roles in either the cellular or humoral immune response showed parallel changes in expression in at least 2 of the 3 selected lines (Table 1).
Three peptidoglycan-recognition protein (PGRP) genes showed up-regulation in at least two of the three selected lines (Table 1). Both PGRP-SB1 and PGRP-SD are produced in the fat body and are only induced upon infection. PGRP-SB1 codes for a bactericidal amidase [18], while PGRP-SD, which functions as a receptor for gram-positive bacteria is involved in Toll activation [19]. PGRP-SC2 is a predicted amidase and was up-regulated in S2 and S3 [20]. Three AMP genes belonging to two families are also up-regulated in selected flies (Table 1). Drosomycin-4 and -5, which are primarily antifungal and target gram-positive bacterium, showed increased expression in all three selected lines [21]. Diptericin B, which has previously been shown to be stimulated upon P. aeruginosa infection, showed the strongest expression changes among AMP genes [22]. Both persephone and easter which encode serine endopeptidases and that regulate the Toll signalling pathway [13] were significantly up-regulated in all selected lines (Table 1).
In previous studies examining the expression profiles of infected flies in response to a range of pathogens, including Pseudomonas, the humoral response dominates in terms of numbers of responsive genes (Table 2). Here, as best seen by the ratio of the number of humoral/cellular responding genes, the nature of evolved defence has shifted toward the cellular. The cellular genes responding to selection in this study are associated with both recognition/phagocytosis and melanization/coagulation. Many of these genes (N = 8) were up-regulated in all three selected lines (Table 1).
The complement related, Thioester-containing proteins (Tep)1 and Tep2 function as opsonins that bind to pathogen surface to promote the detection and phagocytosis of the invading microbes [23]. Tep2 has previously been shown to be required for effective phagocytosis of Gram-negative bacterium E. coli [24]. Two phagocyte specific receptor molecules Scavenger receptor class C type 1 (SR-C1) and eater, which are found on hemocyte surface that bind to a broad range of pathogens [25],[26], are up-regulated in all selected lines. Nimc1 is another phagocytosis gene, which is structurally related to phagocytosis receptors such as eater and Draper, plays an important role in both phagocytosis and development as they are efficient in removing microbes as well as apoptotic cells [27]. Annotation of CG10345 and CG2736 suggest they have cell adhesion and scavenger receptor activities [28]. CG30427, CG7593 and CG3891 are genes required for phagocytosis [24],[28] and CG7593, CG8193 [24] and Black cells [29] have monophenol monooxygenase activity and are essential for the production of melanin from tyrosine (Table 1).
The rapid response to selection by G5, indicates that the initial population of D. melanogaster harbored substantial additive genetic variation for defense against P. aeruginosa infection. The proportion of surviving individuals in the selected population, however, did not increase above 80% despite continued selection pressure. This in combination with the rapid decrease in population survivorship after selection was removed also suggests the presence of antagonistic pleiotropy and/or physiological constraints at work. Corresponding reductions in fitness attributes in selected flies, namely female longevity and fecundity also provide evidence of a trade-off. Such negative correlations between immunity and other aspects of host fitness are predicted [30] and well-documented in the literature [1],[2],[4],[31].
The consistent correlated increase in antibacterial defense and developmental rate in the selected lines was, however, surprising. An elevated investment in immune defense predicts a lengthening of the development processes caused by the depletion of essential nutrients [32]. Indeed the direction of this predicted trade-off has been confirmed in a selection experiment for sexual competitiveness in Drosophila [33] and virus resistance in moths [32]. Here the increase in developmental rate occurred without a reduction in body mass that may be attributed to a lack of competition for food under laboratory conditions. An examination of the transcriptional profiles of our selected lines revealed expression changes in a number of genes that have dual roles in both development and immunity. We, therefore, propose that pleiotropy between developmental and cellular immune processes and the multi-tasking functional role of hemocytes may underlie the shift toward faster development.
The Toll signaling pathway, which is an essential component of humoral immunity, also plays a key role in dorsal-ventral pattern formation in Drosophila embryos [34],[35]. The signal for dorsal-ventral axis formation is conveyed by serine proteases and Easter, which is the last serine protease in a cascade that modifies the transmembrane Toll receptor and leads to activation of the pathway [36],[37]. The process of melanization requires the activation of prophenoloxidase (PPO) to PO. The activation of PPO and Easter are negatively regulated by a single serine protease inhibitor (serpin27) [36],[38]. Transcriptional profiling of our selected lines showed that four POs genes and Easter were up-regulated in all lines. The decrease in developmental time can thus be explained in part by the selection for PPO activation, which would consequently activate Easter and alter the timing of the dorsal-ventral axis formation in the embryo [38].
In addition to patrolling the hemolymph for invading microorganisms, the hemocytes are known to play important roles during embryonic development. Hemocytes are the prominent producer of embryonic basement membrane proteins including proteoglycan papilin and the major connective tissue collagen IV [39],[40], both of which are up-regulated in all selected lines. Hemocytes migrate along conserved pathways in the embryo and shape various tissues by removing apoptotic cells and depositing extracellular matrix. Hemocyte migration and number are both tightly controlled [40]. In Drosophila, the number of hemocytes is shown to influence the outcome of the infection specifically, greater numbers of circulating hemocytes confer greater immunity [41],[42]. We found that the selected flies evolved a greater investment in cellular immunity that could translate into increases in hemocyte number and/or activity. This in turn could also alter the rate of development in selected flies.
The hallmark of the humoral immune response is the production of AMPs as regulated by the Toll and Imd pathways. The signaling cascades that lead to AMP activation are well studied and it is now generally accepted that whether one or both pathways respond to infection depends on the specific pathogen [43]. Shared components that exist in both pathways also provide for some level of cross-regulation [21],[44],[45]. Gene knockout studies have found that flies deficient for either Toll or Imd pathways are more susceptible to P. aeruginosa infection than the wild type [46]. We compared the transcriptome of selected flies to that of controls during early infection in an attempt to identify mechanisms for limiting the initiation and the early progression of P. aeruginosa infection. Components of the Toll pathway including persephone and PGRP-SD were up-regulated in all selected lines. AMP genes from both pathways including drosomycin (Toll) and diptericin (Imd), showed similar patterns of expression increase across all lines. Our data indicate that the Toll and Imd pathways work synergistically as part of the evolved defense against Pseudomonas aeruginosa.
P. aeruginosa synthesize an extensive collection of virulence associated factors that suppress the host immune defense. Drosophila hemocytes, which are the target of several P. aeruginosa toxins, are impaired by the bacterium leading to suppression of phagocytosis [22],[47]. We found a strong involvement of cellular immunity in selected lines that appears to have overcome this immune suppressive effect, possibly acting very early in the infection process [12],[15] before toxins could be produced. All major aspects of cellular immunity including recognition, phagocytosis and melanization are involved in fighting the bacterium. The comprehensive list of cellular immune genes begins with opsonins and surface receptors that recognize and phagocytose bacteria. An array of lysosomal enzymes, proteases, lipases and DNases was up-regulated in selected flies that are involved in the break down of the bacterium in the phagosome (Table S1). Melanization and coagulation genes, including PO genes, which produce melanin that physically impede the growth of intruding microorganisms [14], are up-regulated in selected flies. The conserved pattern of cellular immunity gene expression among the selected lines emphasizes the crucial role of hemocytes in suppressing P. aeruginosa. This also suggests that the synergistic activation of phagocytosis, AMP production and melanization together in selected flies is the best strategy in limiting bacterial infection [41],[42].
The selected flies have evolved mechanisms to overcome the immune suppressive effects of P. aeruginosa that involve a substantial mobilization of cellular immunity as well as investment in the humoral response. We think we see greater evidence of a cellular component in our study as compared to previous work with Pseudomonas as well as other pathogens due to a combination of both methodology and the role of selection. First, it is important to remember that our control lines were also infected and so we are focusing only on the evolved aspects of the response. Evolution of greater investment into the cellular response may be the most effective means of pathogen control. This is in keeping with recent experimental work showing the efficacy of the cellular response over the humoral in early clearing of systemic infections [12],[15]. It may also be that given the inducible nature of the humoral response that it is already operating at the upper limits of its functionality determined by cellular constraints instead of lack of genetic variation. In either case, the investment in both aspects of immunity has come at a cost particularly for females in terms of longevity and fecundity. Both selected males and females also exhibit accelerated development that may be due to changes in expression of shared gene sets in both processes and the multifunctional role of hemocytes. These experiments have revealed highly effective mechanisms of defense available to genetically diverse flies that are nonetheless unsustainable in the absence of continuous pathogen pressure due to their cost.
Brisbane (BNE) base stock was founded from 26 females D. melanogaster caught around the University of Queensland St Lucia campus in August 2006. The flies were treated with 0.5% penicillin and streptomycin in the diet for one generation [48] and then passaged without antibiotic for more than 10 generations before the start of the selection experiment. A large inbred population was maintained as the base stock and reared on standard yellow corn meal medium.
P. aeruginosa PA01 was cultured as in LB medium supplemented with 100 mg ml−1 ampicillin at 37 °C [49]. For infection, the concentration of an overnight bacterial culture was adjusted to an OD of 0.5±0.05 measured spectrometrically at 600 nm. The culture was then diluted 100 fold using sterile LB. This OD was determined at the start of the selection experiments to achieve a population kill rate of 80–90%.
The base stock was split into 3 control and 3 selected lines. These replicate populations were used to test the reproducibility of the selection given the genetic variation present in the base population. Selected lines were infected each generation with PAO1 and the survivors allowed to produce the subsequent generation. Selection was applied for 10 generations. For each round of selection, 8 sub-replicate populations consisting of 20 flies each per gender (160 flies per gender per line per generation) were infected with P. aeruginosa PA01. Mated flies aged to 4–7 days old were anaesthetized with CO2 and infected as previously described by dipping a sterile needle in the bacterial culture and piercing the intrathoracic region of the fly [11]. Fly mortality was then monitored for each population over 48 hours. Survivors from each of the 8 sub-replicates were pooled into a single population to seed the subsequent generation. The control lines were not infected, but were exposed to the same bottleneck in population size as their paired selected lines by randomly selecting a set of individuals to found the next generation. All flies were passaged for a further 5 generations after G10 without selection.
Survival in selected lines was monitored each generation. A subset of control line flies not used to found subsequent generations were also tested for survival post infection at G6 and G10. After G10, the lines were passaged for another 5 generations without infection followed by an additional assessment of survivorship at G15. Realised heritability of infection survival was calculated for each of the selected lines with sexes pooled as the ratio of the cumulative selection response to the cumulative selection differential [50]. For this calculation, we modelled infection survival as a threshold character following transformation [51].
Longevity. Virgin female and male flies were kept in separate vials in populations of 20 (5 replicate populations per gender per line) and moved onto fresh food weekly. Fly death was recorded at each food change. Body mass. Flies were placed in vials on the first day of eclosion and aged for a further three days. Flies were then briefly anaesthetized with CO2 and weighed individually on an electronic balance. Traits were measured for both sexes (40 individuals per gender per line). Developmental time and viability. Twelve eggs laid by a female within a 6 hour window were placed onto a vial after mating with a single male (40 replicates per line). The eggs were monitored every 6 hours. The period of time (hours) from the point of oviposition to the recorded time of eclosion was recorded as the development time. Viability was calculated as a percentage of eggs hatched from a possible of twelve. Female Productivity. Pairs of virgin flies were placed together in a vial and males were removed after 24 hours. The mated females (40 replicates per line) were moved onto fresh vials every 5 days to lay eggs. The total number of viable adult offspring produced by each female was recorded as its productivity. Male Mating Success. A selected male and a control male each powdered with micronized dust of distinct colors were placed with a female from the base population for 90 minutes (Variable N, 137 to 215 replicates per line). Female choice was scored by identifying the male that the female had chosen as a mate.
Microarrays were used to screen for genes demonstrating expression changes in selected lines relative to control lines after bacterial infection in G10. Only male flies were extracted and compared in this analysis. A dual colour reference design paired each selected and control line. Each pair was represented by technical replicates (N = 3) that were then replicated with a dye swap (total N = 6). Microarrays were of the 4×44 K format (Agilent) each containing controls and 3 replicates of each 60 mers feature randomly distributed across the layout. The D. melanogaster genomic sequence (Release 5.4) was obtained from Flybase [28] and was used for construction of oligonucleotides using eArray Version 5.0 (Agilent Technologies Inc., Santa Clara, CA). After removing probes that cross hybridised, a total of 13,875 transcripts which represented 12,041 genes were spotted onto each microarray. Pools of 20 males representing each line were snap frozen in liquid nitrogen and extracted for Total RNA using Trizol (Invitrogen Corp., Carlsbad, CA). RNA was then purified using Qiagen RNeasy kits according to manufacturer's instructions. Further sample preparations and hybridizations were then carried out by the Special Research Centre Microarray Facility at the University of Queensland. Sample quality was examined using the Agilent 2100 Bioanalyzer (Agilent Technologies Inc., Santa Clara, CA). Fluorescent cDNA was synthesized using Agilent Low RNA Input Linear Amplification Kit with Cyanine 3 or Cyanine 5-CTP.
For each transcript, median signal intensity, background signal intensity, flag and saturation were extracted and analyzed using Genesping v.7.0 (Agilent Technologies Inc., Santa Clara, CA). Probes that were not detected in at least one hybridization were considered uninformative and excluded from further consideration. An intensity dependent (Lowess) normalization (Per Spot and Per Chip) was used to correct for non-linear rates of dye incorporation as well as irregularities in the relative fluorescence intensity between the dyes. Hybridizations from each line were used as replicate data to test for significance of expression changes using the cross-gene error model. The Bonferroni multiple testing correction was used to reduce the occurrence of false positives. All array data have been deposited in ArrayExpress (http://www.ebi.ac.uk/microarray-as/ae/) under the accession # E-MEXP-2054.
Quantitative real-time PCR (RT–PCR) was used to validate the expression of a subset of 6 immune genes showing increased expression across all three selected lines on the arrays (Table 3) and that represented some of the major functional categories of the immune response. RNA was extracted as above and then treated with 2 µl of DNase I (Roche) for 30 minutes at 37°C to eliminate genomic DNA. Approximately 0.5 µg of total RNA was reverse transcribed using random primers and SuperScript III reverse transcriptase (Invitrogen) according to manufacturer's protocols. Quantitative PCR (qPCR) was performed on Rotor-gene 6000 (Corbett Life Science, Sydney, NSW) using Platinum®SYBR®Green (Invitrogen Inc, Carlsbad, CA) according to manufacturer's instructions. For each sample a mastermix of 2 µl RNase-free water, 5 µl of SYBR Supermix and 0.5 µl of each primer (10 µM) was added to 2 µl of cDNA. Three replicates were run for each sample. The cycling protocol was as follows; 1 cycle UDG incubation at 50 °C for 2 minutes, 1 cycle Taq activation at 95°C for 2 minutes, 40 cycles of denaturation at 95 °C for 5 s, annealing at 60 °C for 5 s, extension at 72°C for 15 s, fluorescence acquisition 78 °C, and 1 cycle of melt curve analysis from 68–95°C in 1°C steps. The raw output data of Cycle Threshold (CT) was normalized by taking into consideration the differences in amplification efficiency of target and the reference genes using Q-gene software [52]. The linear normalized expression (NE) was analyzed using Statistica 8.0 (StatSoft, Inc.). D. melanogaster ribosomal protein rpS17 was used as the reference gene (Table 3).
|
10.1371/journal.pgen.1005762 | Base J and H3.V Regulate Transcriptional Termination in Trypanosoma brucei | Trypanosoma brucei is a protozoan parasite that lacks many transcription factors found in other eukaryotes, such as those whose binding demarcates enhancers. T. brucei retains histone variants and modifications, however, and it is hypothesized that it relies on epigenetic marks to define transcription-related boundaries. The histone H3 variant (H3.V) and an alternate nucleotide, base J (ß-D-glucosyl-hydroxymethyluracil), are two chromatin marks found at both transcription termination sites (TTSs) and telomeres. Here, we report that the absence of both base J and H3.V result in transcription readthrough and the appearance of antisense transcripts near TTSs. Additionally, we find that maintaining the transcriptional silencing of pol I-transcribed telomeric Variant Surface Glycoprotein (VSG) genes appears to be dependent on deposition of H3.V alone. Our study reveals that gene expression depends on different epigenetic cues depending on chromosomal location and on the transcribing polymerase. This work provides insight into how these signals may have evolved into the more nuanced and fine-tuned gene regulatory mechanisms observed in other model systems.
| In eukaryotes, DNA is wrapped around histones to form chromatin. Modifications on the DNA itself, or on the canonical histones around which the DNA is wrapped, can lead to changes in gene expression. Alternate histones are also used to regulate gene expression. The African trypanosome, the causative agent of African sleeping sickness, transcribes its genes in long arrays called polycistronic transcription units (PTUs). In this study, we describe how the African Trypanosome uses two chromatin marks to regulate genes that lie close to the end of the PTU or close to the ends of chromosomes, called telomeres. One mark, base J, is on the DNA itself, while the other utilized mark is an alternate histone protein called H3.V. In the absence of these marks, there is an increase in antisense RNA that corresponds to genes that lie near the ends of the PTUs, and an increase in sense RNA for genes near telomeres. Since trypanosomes are evolutionarily distant from humans, these findings shed light on how gene expression mechanisms that are modulated by chromatin marks may have evolved to produce the complex gene regulatory networks found in our own tissues.
| Trypanosoma brucei, a eukaryotic parasite that causes African sleeping sickness, belongs to the order Kinetoplastida and diverged from mammals ~500 million years ago. T. brucei is heteroxenous, requiring two obligatory hosts to complete its life cycle: a mammalian host and an insect host (Glossina spp, the tsetse). Trypanosomes proliferate in the tsetse gut as procyclic forms (PF) [1] and differentiate into non-proliferative metacyclic forms in tsetse salivary glands. After T. brucei infects a mammalian host, it differentiates into the bloodstream form (BF) and proliferates in the host’s extracellular spaces. The cycle is completed when BF T. brucei reaches a quiescent state and gets ingested by a tsetse.
T. brucei branched early in eukaryotic evolution, and differs substantially from other eukaryotes in its regulation of gene expression. It is thus of great evolutionary interest. One prominent difference concerns transcription of protein-coding genes by RNA polymerase II (pol II), which diverges in two important ways from other well-studied eukaryotes. First, transcription occurs polycistronically, and most genes are arranged in polycistronic transcription units (PTUs). Mature mRNAs are then generated post-transcriptionally by coupled trans-splicing and polyadenylation reactions that trans-splice a 39-nucleotide leader sequence onto every mRNA [2]. Second, some general transcription factors and one pol I-specific factor have been identified in T. brucei [3–5], but the parasite lacks many of the sequence-specific transcription factors found in other eukaryotes that bind to cis-regulatory sequences, and cis-regulatory sequences themselves have not been well characterized. The lack of these regulatory factors, and the fact that histone modifications and histone variants are deposited at sites of putative transcription initiation and termination, has led to the idea that chromatin marks are important for demarcating PTUs [6]. Prior experiments using chromatin immunoprecipitation and sequencing (ChIP-seq) have correlated the presence of certain histone modifications and histone variants with transcription initiation sites (H3K4me3, H4K10ac, and H2 variants H2.AZ and H2.BV at transcription start sites (TSSs)) or transcription termination sites (H3.V and H4.V at transcription termination sites (TTSs)) [6,7]. However a mechanistic involvement of these marks with transcriptional regulation has not been demonstrated.
In addition to these distinct chromatin features, pol II transcription initiation and termination regions contain a kinetoplastid-specific DNA modification, known as base J (ß-D-glucosyl-hydroxymethyluracil) [8]. It arises via two steps that modify dT: the J-Binding Protein-1 (JBP1) and JBP2 (homologs of the mammalian TET proteins) mediated hydroxylation of thymidine to generate a hydroxymethyl-dU (hmU) intermediate [9], followed by the glucosylation of hmU by the J-associated Glucosyl Transferase (JGT) [10]. Base J synthesis requires both JBP1/2 and JGT [10,11]. Base J is enriched at PTU flanks, where it coincides with, but does not depend on, the presence of H3.V [12]. It has recently been demonstrated that depletion of base J in Leishmania results in readthrough transcription at sites of transcription termination [13,14], while deletion of H3.V does not have this phenotype [15]. In contrast, depletion of base J in T. brucei does not cause readthrough at convergent strand switch regions, though it does affect termination at specific sites proximal to J deposition [13]. Deletion of H3.V has not been characterized with respect to transcriptional readthrough in T. brucei. Based on these results, we hypothesized that while base J may be the dominant mark for transcription termination in L. major, H3.V may compensate for the lack of base J in T. brucei, and its role in the epigenetic network of T. brucei may differ from that of L. major. We surmised that the effects of base J and H3.V in T. brucei might best be assessed by removing them simultaneously, and sought to ask whether perturbation of chromatin marks at sites of transcription termination would have any functional consequence on gene expression.
In addition to TTSs, J and H3.V (but not H4.V) marks coincide over telomeric regions in T. brucei. Variant Surface Glycoprotein (VSG) genes, encoding the major T. brucei surface antigen, are frequently found near telomeres. One genomic location for telomere-proximal VSGs is the Bloodstream-form Expression Site (BES), which contains a polycistronic transcription unit that is transcribed by RNA pol I. Fifteen BESs have been found in the Lister 427 strain [16], and each BES terminates at the telomere with a telomere-proximal VSG. Another class of telomere-proximal VSGs is found in metacyclic expression sites (mVSGs). These sites are also transcribed by pol I and mVSG is expressed during the metacyclic stage in the tsetse salivary gland. Minichromosomes have also been found to contain VSG genes, but these sites lack obvious pol I promoter sequences. A large number of other VSG genes are scattered throughout the genome, but not all of their genomic locations are known [17]. In BF cells, only one BES is transcriptionally active at any given time, ensuring the monoallelic expression of a single coat protein that is essential to immune evasion and, therefore, survival of the organism in the mammalian host. Given the colocalization of base J and H3.V at telomeric sites, we sought to test whether the combination of J and H3.V marks are necessary to maintain silencing of VSG expression units in BF trypanosomes, taking advantage of the telomeric location of VSG genes to interrogate telomeric silencing in general.
Using high-throughput, strand-specific sequencing of poly-A+ mRNA (RNA-seq) on genetic mutants that lack H3.V, base J, or both marks, we demonstrate that simultaneous deletion of base J and H3.V increases the amount of antisense transcripts for genes proximal to transcription termination sites, where both marks are found. We also demonstrate that of the two marks, H3.V appears to be more dominant than base J with respect to maintaining silencing of telomeric VSG genes. Thus our experiments suggest that perturbations of the chromatin landscape in different regions of the genome (transcription termination sites vs. telomeres) have functional effects on gene expression, and that the dependence on a particular chromatin mark differs depending on chromosomal location and/or specific features of the transcribing polymerase (RNA pol II vs. pol I).
To study the roles of base J and H3.V, we generated knockout cell lines in BF T. brucei using previously published methods [18]. To make BF trypanosomes deficient in base J, we first removed both alleles of JBP1, then both alleles of JBP2. In both cases, we used deletion-cassettes containing hygromycin- or puromycin-resistance genes fused to Herpes simplex virus thymidine kinase (HYG-TK or PUR-TK) and flanked by loxP sites, allowing the markers to be removed by transient expression of Cre-recombinase and reused. JBP1Δ/Δ JBP2Δ/Δ double mutants, which we will henceforth refer to as JΔ mutants, are completely null for base J [11]. To generate BF trypanosomes deficient in only H3.V or both H3.V and base J, respectively, we sequentially deleted both alleles of H3.V either from a WT or JΔ background using the deletion cassette H3.VΔPUR (pJEL76) and H3.VΔHYG (pJEL38) [19].
H3.V and base J are not synthetic lethal, as JΔ H3.VΔ trypanosomes were viable. Growth curves revealed that there was no obvious defect in growth for any of the mutants (Fig 1) and there were no obvious morphological differences.
Both base J and H3.V are highly enriched at telomeres in BF cells [19–21]. As some VSG genes are located near telomeres (Fig 2A, diagram, BES VSGs or mVSG), we asked whether base J and H3.V are important for maintaining silencing of inactive telomeric VSGs by assessing expression levels of these telomeric VSGs in cells lacking one or both chromatin marks. We also assessed expression levels for minichromosomal VSGs (MC VSGs), which are assumed to be proximal to telomeres. We isolated polyA+ RNA from each of the mutants described above, as well as from the parental cell line (referred to as WT). Libraries for high-throughput sequencing were generated using 3 independent cultures from each genotype (for a total of 12). Reads from each library were uniquely aligned allowing for two mismatches to the megabase chromosomes. Unmapped reads were then uniquely aligned allowing for two mismatches to the VSGnome [17]. A log2(RPKM) (Reads Per Kilobase Per Million mapped reads) value was generated for reads aligning to each VSG and these values were averaged across the 3 replicates. We then compared WT and mutant log2(RPKM) values for VSGs with known subtelomeric locations in Lister427 cells (BESs, minichromosomes) as well as those known to be associated with metacyclic promoters [17,22] (Fig 2B–2D). We also generated notched boxplots for log2(RPKM) values of BES, metacyclic, and minichromosomal VSGs to elucidate the overall distribution of the data (Fig 2E–2G, respectively).
In the H3.VΔ mutant, 40% (6 of 15) of BES-associated VSGs, and 4 of the 5 metacyclic VSGs were >4-fold upregulated (Fig 2C, S1 Table). The median log2(RPKM) values for BES and metacyclic VSGs were higher in the H3.VΔ mutant and the distribution of log2(RPKM) values was shifted upward (Fig 2E and 2F). This effect was recapitulated in the JΔ H3.VΔ mutant (Fig 2D), with 7 BES-associated and 5 metacyclic VSGs >4-fold upregulated in these cells (S1 Table). Median log2(RPKM) values were also higher in the JΔ H3.VΔ mutant (Fig 2E and 2F). A Mann-Whitney U test was performed between WT and mutant cells to address whether the change in median values was significant. This test revealed that the difference in VSG expression at BES VSGs in both the H3.VΔ mutant and the JΔ H3.VΔ mutant were significant. P values for these tests are provided in S2 Table. While the majority of individual metacyclic VSGs are upregulated in the H3.VΔ and JΔ H3.VΔ mutants, the difference in gene expression for this group of genes was not statistically significant according to our statistical test. We used antibodies against VSG3 (which is located at a BES but transcriptionally silenced in our strains) and detected its presence by western blot analysis (S1 Fig), thus further validating BES-associated VSG derepression. On the other hand, in the absence of base J alone, no BES or metacyclic promoter associated VSGs were >4-fold upregulated (Fig 2B, S1 Table). We conclude that the H3.V mark is important for maintaining silencing of BES VSGs, while base J appears dispensable for silencing of these VSGs.
Unlike the BES and metacyclic VSGs, transcripts of minichromosomal VSGs (which lack canonical pol I promoters) were upregulated in the absence of both base J and H3.V. Of 56 MC VSGs, 13 minichromosomal VSGs showed >4-fold upregulation in the JΔ H3.VΔ mutant (Fig 2D), while 2 and 1 minichromosomal VSGs were >4-fold upregulated in the JΔ and H3.VΔ strains respectively (Fig 2B and 2C, S1 Table). Boxplots also show an increase in the median log2(RPKM) values in the JΔ H3.VΔ mutant (Fig 2G), which were statistically significant. The increase in expression of subtelomeric VSGs by deleting H3.V alone suggests that H3.V is the dominant epigenetic mark for telomeric silencing. The fact that more minichromosomal VSGs become upregulated when base J is also absent may indicate that base J is able to partially compensate for the lack of H3.V at these genomic locations. The absence of canonical pol I promoters from minichromosomes would imply that the upregulation could be due to promiscuous transcription by pol II or unknown factors.
When we applied RNA-seq analysis to ~2,500 annotated VSG sequences [17], we found that transcription of 45 and 57 VSG genes were >4-fold upregulated in the absence of base J or H3.V, respectively (S2A and S2B Fig, S1 Table). A more pronounced effect was observed in JΔ H3.VΔ cells, where 193 VSG genes were >4-fold upregulated (S2C Fig, S1 Table), representing about 7.5% of the annotated VSGs. Median log2(RPKM) values increased slightly in the absence of base J or H3.V, with the greatest difference found in the JΔ H3.VΔ mutant (S2D Fig). Statistical testing indicated that differences in transcript levels between WT and mutant cells was significant in each of the mutant strains tested (JΔ, H3.VΔ, and JΔ H3.VΔ) (S2 Table). All VSG raw read counts are reported in S1 Data.
Combined with the analysis of VSGs at known chromosomal locations described above, these results highlight a role for both base J and H3.V in maintaining a chromatin structure that promotes silencing of a subset of VSG genes. The additive effect of removing base J and H3.V on transcription of the set of VSGs with uncharacterized locations suggests that collaborative reading of these chromatin marks may be more common in regions further from the telomeres. The effect observed upon removal of both marks implies that the factors responsible for reading these marks and maintaining silencing may operate in different pathways that have some functional redundancy [23]. Overall, it appears that the role for certain chromatin marks (and by extension chromatin structure) in maintaining silencing of VSG genes appears to differ depending on chromosomal context.
Increased VSG levels could indicate either derepression of silent VSGs or an increase in VSG switching rates, possibly due to effects on telomere length [24]. However, the VSG switching rate was not significantly elevated in H3.VΔ or JΔ mutant cells, and was only slightly elevated in JΔ H3.VΔ cells (S3A Fig). In mammals, telomere shortening led to reduction of repressive epigenetic marks such as H3K9me3 and H4K20me3 [25]. As base J and H3.V are both located at telomeres in T. brucei, we examined telomere length in the H3.VΔ and JΔ mutants but did not find differences between the genotypes either in speed of lengthening or in maintenance (S3B–S3E Fig).
Regions flanking PTUs can be binned into three distinct types. Sites where both PTUs initiate but run in opposite directions are putative sites of transcription initiation, termed divergent strand switch regions (dSSRs). Sites where both PTUs terminate coming from opposite directions are putative sites of transcription termination, termed convergent strand switch regions (cSSRs). Sites where one PTU ends and another starts are referred to as head-to-tail or HT (S4 Fig). Previous studies indicated that pharmacologic depletion of base J results in increased gene expression for genes downstream of specific sites of base J deposition in T. brucei [13], but did not find consistent evidence for readthrough transcription at cSSRs, indicating that an additional mark might help maintain termination signals in the absence of base J at cSSRs. Both base J and H3.V are found at cSSRs in T. brucei (S4 Fig), and it is hypothesized that these marks might act as signals for transcription termination. We thus wondered whether the absence of both base J and H3.V might result in an increase in the number of transcripts found within cSSRs.
Stranded RNA-seq libraries were prepared from 3 independent cultures from each genotype, and sequencing reads were aligned uniquely to the megabase chromosomes allowing for 2 mismatches. We investigated polyA+ transcript levels extending past the coding regions of the genes that flank cSSRs. We obtained polyA+ reads that mapped to regions between the coding regions of genes that flank convergent PTUs (Fig 3A), which we will refer to as convergent strand switch regions (or cSSRs for simplicity). We defined this region as starting 1 bp past the coding region of the (+) strand gene at the end of one PTU and ending 1 bp before the start of the (-) strand gene at the end of the second PTU (Fig 3A). log2(RPKM) values were generated for each of these regions and the three values for each replicate within one genotype were averaged. Polyadenylation sites have previously been shown to vary quite widely both within a gene and across different genes, and can extend up to 6,000 nucleotides past the end of the coding region of the gene, with the median length estimated at ~400 bp [26]. Overall, we found an increase in polyA+ transcripts specifically in cSSR regions in the JΔ H3.VΔ mutant (Fig 3B–3E). Median log2(RPKM) levels and inner quartile ranges for these regions were also increased in each of the mutants, and this increase was significant in JΔ H3.VΔ cells according to a Mann-Whitney U test (Fig 3E, S4 Table). We were able to confirm an increase in total transcript levels within these regions by performing q-PCR on selected sites (Fig 3F, TTS-105 and MCM-BP did not show increased log2(RPKM) values in the RNA-seq dataset and are used here as negative controls).
We analyzed the number of cSSRs with a significant increase in gene expression in each of our mutant lines by performing statistical tests on each set of replicates using Benjamimi and Hochberg correction. We found that 60% of all cSSR sites showed a significant change in log2(RPKM) values in the JΔ H3.VΔ mutant, while 42% and 7% of these sites showed significant changes in log2(RPKM) values in the JΔ and H3.VΔ mutants, respectively (S3 Table).
We also performed an analysis of head-to-tail sites. H4K10ac marks have previously been shown at both divergent strand switch regions (dSSRs) and at head-to-tail sites, and these regions are also bound by Bdf3 [6]. We defined head-to-tail sites as regions that are bound by Bdf3 that lie within the middle of a PTU (as opposed to a dSSR). When we analyzed these regions by stranded RNA-seq, we found that there was a significant increase, as measured by a Mann-Whitney U test, in both sense and antisense reads at these regions in JΔH3.VΔ mutant cells, but that the effect was more pronounced in antisense transcripts (S5 Fig and S5 Table). Taken together, these results led us to speculate that the absence of both base J and H3.V might result in transcriptional readthrough at regions of transcriptional termination.
To explore the idea that readthrough takes place at regions of transcription termination, we wanted to ask whether the absence of base J and H3.V might result in polyA+ reads that extend past the 3´ UTRs of genes immediately proximal to cSSRs. We interrogated the set of cSSR proximal genes for which 3´ UTR information was defined [26]. Because 3´ UTR length is highly variable, we used the maximum 3´ UTR length defined for each of these genes [26]. We then restricted our analysis to the set of cSSRs whose boundaries were delimited by defined 3´ UTRs and whose length was between 1,000 and 5,000 bp. To determine whether increased transcripts within the cSSRs result from readthrough of cSSR proximal genes, we computed the average difference in coverage between WT and mutant cells for sliding windows starting from up to 1 kb upstream of the 5´ end of the cSSR and ending 1 kb downstream of the 3´ end of the cSSR (Fig 4A). Sliding windows were calculated by dividing the length of each region (upstream 1kb, the cSSR itself, and downstream 1kb) into 10% intervals and sliding these intervals by increments of 2%. Because cSSR and cSSR proximal region lengths vary, intervals were computed using percentages rather than length to ensure an equal number of windows for comparison across cSSRs. In this analysis, we found that the greatest difference in transcript levels between WT and JΔ H3.VΔ mutant cells is found immediately following the end of the last 3´ UTR for the last gene at the end of the PTU (Fig 4B and 4C). That is, if a PTU is found on the (+) strand, the difference in transcript levels for WT and JΔ H3.VΔ mutant cells for the last gene in the PTU remain minimal until the end of the 3´ UTR. More transcripts are then observed in the JΔ H3.VΔ mutant cells when compared to WT cells in cSSR regions, and that difference increases crossing into the opposite PTU. While antisense transcription does persist into the cSSR past the 3´ UTR in WT cells, the abundance of antisense transcripts approaches 0 moving into the opposite PTU. Conversely, an appreciable level of antisense transcripts persists through the cSSR, extending further into the opposite PTU. Consequently, a more pronounced difference in antisense transcript levels within cSSR proximal regions is observed.
To more directly measure the level of antisense transcription in cSSR proximal genes, we performed stranded q-PCR on five candidate genes that showed a marked difference in antisense transcript levels between WT and JΔ H3.VΔ (Fig 4D–4F). q-PCR analysis confirmed that more antisense transcripts were produced in the absence of base J and H3.V with respect to WT (Fig 4E). Conversely, the difference in sense transcripts in JΔ H3.VΔ compared to WT was much lower than for the antisense transcripts (Fig 4F). Together, these data suggest that the absence of base J and H3.V may cause a perturbation at the site of transcription termination.
Readthrough transcription proceeding from one PTU into the converging PTU produces antisense transcripts. Polyadenylated antisense transcripts have previously been identified in T. brucei [26], so we next asked whether antisense transcripts could be detected in our stranded polyA+ RNA-seq libraries. Indeed, we were able to identify antisense transcripts and furthermore found that antisense transcripts are more prevalent in base J and H3.V mutant cells. We divided the regions upstream and downstream of each cSSR into sliding windows of 5,000 bp, with a step size of 100bp, and then calculated the median difference in log2(RPKM) values between the WT and mutant cells for all the genes within each window (Fig 5A, diagram). We then plotted these medians as a function of distance of the window from the cSSR. For sense transcripts, the median difference between WT and mutant log2(RPKM) value for genes within each window is very subtle, but is greater at closer distances to the cSSR (Fig 5B), indicating that the minor effect of base J and H3.V on sense transcription decreases with distance from the cSSR. When we examined antisense transcripts, we found that the deletion of base J and H3.V had a much greater effect (Fig 5C). In addition, this effect was distance dependent, with greater differences closer to the cSSR. The most pronounced effects were seen in JΔ H3.VΔ mutants (Fig 5C, blue line). We conclude that readthrough transcription in the absence of base J and H3.V results in the production of antisense transcripts in the neighboring PTU.
As an extension of our analysis above, we asked whether transcription readthrough and the production of antisense transcripts might affect gene expression for those genes that are proximal to cSSRs. We used stranded polyA+ RNA-seq libraries generated from each of the mutants to address this question. We calculated log2(RPKM) values for all reads generated from both sense and antisense transcripts, and separately calculated log2(RPKM) values specifically from reads generated from sense strand transcripts or from antisense transcripts. A list of genes that are significantly up- or down- regulated by >4-fold, as measured by a T-test followed by Benjamimi and Hochberg correction in the mutants with associated p adjusted values is provided in S6 Table. Distances from the start of the gene to the nearest cSSR and to each flanking head-to-tail site is also provided in S6 Table. We first compared log2(RPKM) values for genes that fall within 1,000 bp of a cSSR (Fig 6A, top and S6A Fig). The results for sense reads are shown in the left-hand panels of Fig 6, and antisense reads in the right-hand panels. The results for all reads (both sense and antisense) are shown in S6 Fig. For each cSSR, we defined the 5´ interval as -1,000 bp from the end of the last (+) strand gene within the 5´ PTU, and the downstream interval as the first 1,000 bp from the start of the first (-) strand gene in the 3´ PTU. A gene was only included in the analysis if it fell within either interval. Boxplots comparing log2(RPKM) values for each genotype revealed an increase in the median log2(RPKM) for genes falling within 1,000 bp of the cSSR in each of the mutant lines for both sense and antisense reads, with the most pronounced increase in median log2(RPKM) found in the JΔ H3.VΔ mutant (Fig 6A). However, only the increase in antisense reads was statistically significant in both the JΔ mutant and the JΔ H3.VΔ mutant (S7 Table). Frequency distributions for the number of genes that fall within 1,000 bp upstream or downstream of a given cSSR reveal that at the majority of sites, 1–4 genes are within this range (S6A Fig, right). We conducted a similar analysis of genes that fell within 5,000 bp of the cSSR (Fig 6B, top diagram). For reads generated from sense transcripts, we observed only a subtle increase in median log2(RPKM) in each of the mutant lines, none of which were significant (Fig 6B, left panel, S7 Table). For reads generated from antisense transcripts, we observed an increase in the median log2(RPKM) value and a shift upward in the interquartile range for each of the mutant cell lines. The largest effect was once again seen in the JΔ H3.VΔ mutant cells (Fig 6B, right panel), but differences in transcript levels were also significant for the JΔ mutant when compared to WT transcript levels (S7 Table). For most sites, 1–8 genes lie within 5,000 bp of each cSSR (S6B Fig, right panel).
When we extended our analysis to 10,000 bp on either side of each cSSR, we found that the difference in median log2(RPKM) was again unaltered between the WT and the mutant lines for reads generated from sense transcripts (Fig 6C). However, expression levels for antisense transcripts were significantly higher in the JΔ and JΔ H3.VΔ mutants when compared to the WT, with the greatest effect again observed in the JΔ H3.VΔ mutant (S7 Table). These results indicate that deletion of both base J and H3.V does not change the level of sense transcripts for genes within 10kb of the cSSR. In contrast, the effect on antisense transcription of genes proximal to cSSRs appears to extend further out (up to at least 10,000bp) in the absence of both base J and H3.V (Fig 6C). A list of the genes analyzed for each distance window along with RPKM values is provided as S2 Data.
Here, we have investigated the effects of base J and H3.V in maintaining the transcriptional landscape surrounding their sites of deposition, which include cSSRs and telomeres. In general, we find that H3.V appears to be required for the maintenance of silencing of telomeric genes (VSGs), while both marks are important for proper transcription termination at the ends of PTUs. In the absence of these marks, the number of antisense transcripts following the end of a PTU is increased in the neighboring PTU. Previous work on the role for base J in pol II transcription termination was done using small RNA-seq on cells treated with dimethyloxalylglycine (DMOG) [13], and thus, in the context of pharmacologic inhibition of J production, these authors found no evidence for antisense transcription at cSSRs or at head-to-tail (HT) sites. However, they did detect changes in transcript levels for genes immediately downstream of some sites of J localization within PTUs in JBP1Δ JPB2Δ mutants using total RNA-seq. Our experiments on readthrough transcription were performed slightly differently, using polyA+ enriched libraries for genetic mutants. In contrast to previous results, we did find evidence for transcriptional readthrough at cSSR regions in JΔ mutant cells, which was exacerbated in the absence of H3.V (Figs 4 and 5). It is possible that the antisense transcripts that we detected might be processed by the cellular machinery in a way that does not lead to the production of small RNA degradation products, or that they are so short lived that they are undetectable, thus explaining why they were not detected using small RNA-seq. Alternatively, differences in the method for eliminating J (DMOG vs. genetic mutation) may have led to differences in the results from these two analyses. The total of number of genes where sense transcript abundance changes by > 2-fold in the JΔ is relatively low (total 62), most of these changes represent an increase in expression in the mutant, and many of the genes affected are VSGs, ESAGs, and RHS hotspots. This is in agreement with previous results using total RNA-seq on JΔ genetic mutants [13]. The number of transcripts whose abundance changes by > 2-fold in the JΔ H3.VΔ mutant is ~3 times higher than that of J alone, and ~13 times higher than H3.VΔ alone. However, these changes may be the result of secondary effects, since we did not find a specific pattern for changes in sense transcription immediately proximal to cSSRs.
Our data support a mechanism that requires the collaborative reading of histone variants as well as of a DNA modification in suppressing readthrough and antisense transcription at transcription boundaries, with the difference in severity of these transcriptional defects in T. brucei and L. major possibly due to a third chromatin mark that remains in BF JΔ H3.VΔ cells (perhaps H4.V or another, uncharacterized mark) to demarcate the ends of PTUs. At subtelomeres however, H3.V appears to be the dominant mark for suppressing transcription of inactive VSG genes by an unknown mechanism. Functional effects for the loss of base J and H3.V during infection remain to be studied in T. brucei, but studies in T. cruzi indicate that loss of JBP1 and JBP2 has functional effects on host cell invasion and egress [27]. Our findings that two independently deposited marks are coordinately read to affect transcription have relevance beyond T. brucei biology. For example, JBP1/2 are members of a larger family of α-ketoglutarate and Fe2+-dependent dioxygenases whose mammalian homologs are the ten-eleven translocation (TET) proteins (now referred to as the "TET/JBP family"; reviewed in [28]). TET proteins hydroxylate 5-mC to convert it to 5-hmC, but have also retained their ability to convert dT to 5-hmU [29]. A number of recent reports revealed functional links between TET-mediated DNA marks (hmC, hmU) and chromatin modifications (reviewed in [28,30,31]). For example, TET proteins can be attracted to particular loci on the basis of certain chromatin marks (e.g. promoter regions enriched with H3K4me3 and H3K27me3 mark [32]). In light of our work with H3.V and base J, we hypothesize that the binary reading of TET-mediated DNA marks together with a subset of chromatin marks results in the regulation of transcription close to termination sites. The transcriptional effects we observe in the homologs of these proteins in T. brucei might indicate that the DNA mark and histone mark, which do not depend on one another for deposition, might be read either by the same protein complex or by independent complexes that function within the same time frame. These complexes could potentially recruit different downstream factors to control repression at transcription boundaries and subtelomeres. Epigenetic marks and the proteins that modify or read them have been recognized as important factors in regulation of the expressed genome. Our study demonstrates a functional relationship between DNA and histone marks at transcription boundaries, and sheds light on the evolution of transcription regulation.
Trypanosoma brucei bloodstream forms (strain Lister 427) of the ‘single marker’ (SM) background expressing T7 RNA polymerase and Tet repressor (TetR) [33] were cultured in HMI-9 at 37°C [34]. Stable BF clones were obtained using electroporation (Lonza) and maintained in HMI-9 media containing antibiotics, as required, at the following concentrations: 2.5 μg/ml G418 (Sigma); 5 μg/ml blasticidin (Invivogen); 5 μg/ml hygromycin (Invivogen); 0.1 μg/ml puromycin (Sigma); 1 μg/ml phleomycin (Invivogen). The mutant cell lines and plasmids used for this study are listed in S8 and S9 Tables. Detailed construction information and maps are available upon request.
Equal numbers of cells were collected and suspended in Laemmli buffer and separated on an SDS-PAGE gel. Following transfer to nitrocellulose, the proteins were analyzed using anti-VSG3 or anti-tubulin antibody.
Switching assays were performed as described previously [35,36]. Briefly, cells were maintained in the presence of blasticidin to exclude switchers from the starting population, as the blastidicin-resistance gene (BSD) was placed at the active BES promoter. Cells were then allowed to switch in the absence of selection for 3–4 days to ensure the same number of population doublings. Switchers were enriched using a MACS-column method. Flow-through enriched with switchers was collected and stained with anti-VSG2 antibody and propidium iodide (PI) for dead cell exclusion and analyzed by flow cytometry. Cells that lost VSG2 signal and are PI negative are live cells that have switched. Switching mechanism was assessed as described previously [36].
RNA was extracted from treated or control cells using RNA Stat-60 (Tel-Test) according to the manufacturer’s protocol and quantified on a NanoDrop2000c and further cleaned using RNase easy kit (Qiagen). Poly-A sequencing libraries were prepared by Rockefeller University Genomics Resources Center using the Illumina platform. Stranded sequencing libraries were prepared using the NEBNext Ultra Directional Library Kit (E7420). Sequencing was performed on an Illumina HiSeq 2000 sequencer using 50 bp reads. Reads were trimmed for quality using the TrimGalore program from Babraham Bioinformatics (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) and aligned with Bowtie [37] to the reference genome (Tb927v5) allowing only for uniquely aligning reads with a maximum of 2 mismatches. Reads that did not align to the genome were then aligned to the Lister427 VSGnome (http://129.85.245.250/index.html) using the same parameters. Reads were quantified using SeqMonk (http://www.bioinformatics.babraham.ac.uk/projects/seqmonk) from Babraham Bioinformatics or, in the case of VSG aligned reads, custom python scripts.
For the plots shown in Fig 6, log2(RPKM) values for each gene were included in the analysis if they fell within a defined length of the cSSR. This length was defined as within 1,000 bp, 5,000 bp, or 10,000 bp upstream of the 3´ end of the last gene in the (+) strand PTU or within 1,000 bp, 5,000 bp or 10,000 bp downstream of the 5´ end of the first gene in the (-) strand PTU (see the diagrams in Fig 6). Genes falling within PTUs that were shorter than 1,000bp, 5,000bp, or 10,000bp were discarded from the analysis. For the plots in Fig 5, the distance of each gene from the cSSR was computed starting at the halfway point between the 3´ end of the last gene in the (+) strand PTU (for upstream genes) and 5´ end of the first gene in the (-) strand PTU (for downstream genes)(halfway point is demarcated by a black line in the Fig 5 diagram). Median difference in log2(RPKM) value between mutant and WT cells were computed for genes that fell within 5 kb windows proceeding upstream or downstream of the halfway point within the cSSR, defined as between the 3´ end of the last gene in the (+) strand PTU and the 5´ end of the first gene in the (-) strand PTU. For Fig 3, log2(RPKM) values were computed for the region that lies between the 3´ end of the last gene in the (+) strand PTU and the 5´ end of the first gene in the (-) strand PTU for each analyzed cSSR. For Fig 4, the maximal length 3´ UTR was defined using the data in [26]. The analysis was restricted to cSSRs 1–5 kb in length and flanked by genes with defined 3´ UTRs. For each cSSR, analysis was focused on three regions: (1) up to 1kb of the gene immediately upstream of the cSSR, (2) the cSSR itself, and (3) up to 1kb of the gene immediately downstream of the cSSR. For cSSR proximal genes shorter than 1kb, analysis was restricted to the length of the gene. Sliding windows were calculated by dividing each region into intervals 10% of the total length and sliding by a step size of 2%. For instance, 100 bp windows were slid 20 bp down the length of a 1 kb-long region. Median difference in log2(RPKM) values between mutant and WT cells were computed for windows in each cSSR and cSSR-flanking region. Median log2(RPKM) differences were then averaged for each window across all cSSR and cSSR-flanking regions. Notched boxplots were computed using python’s matplotlib library with the following command pylab.boxplot(data, notch = True, outliers = 'b.').
Differences between WT and mutant cells were summarized using descriptive statistics performed in R, version 3.2.0 (The R Foundation for Statistical Computing). To compare VSG expression (Fig 2E, 2F and 2G), expression of cSSRs (Fig 3E), expression of cSSR proximal genes (Fig 6), and expression of HT sites (S5 Fig) a Kruskal-Wallis one-way analysis of variance test (command kruskal.test()) was first applied to determine whether a significant difference exists across median log2(RPKM) counts in WT and mutant cells. Once significance was established across groups, a Mann-Whitney U test (command wilcox.test()) was performed between median log2(RPKM) counts for each mutant genotype and the WT. Significance values in Figs 2E–2G, 3E and 6 correspond to the results of the Mann-Whitney U test. P < 0.05 was considered significant for all tests. For all graphs, P < 0.05 is indicated by * and P < 0.01 is indicated by **. Test statistics and P values are reported in S2, S4, S5 and S7 Tables.
Total mRNA was prepared using RNA Stat-60 (Tel-Test) as described in manufacturer’s protocol and cDNAs were generated using random hexamer and reverse-transcriptase (Stratagene). RNA was amplified using primers specific to individual selected TTS regions by quantitative PCR using the LightCycler 480 (Roche). Amplified double-stranded DNA product during 40 cycles was detected by SYBR Green I. All measurements were in triplicate and compared with a 4,000-fold range of serially diluted standard genomic DNA prepared from the wild-type strain. The sequences of primers are available upon request.
Stranded RT-PCR was performed as described above with the following differences. Gene specific primers were used to generate cDNAs from antisense or sense transcripts. These cDNAs were amplified using two sets of primers specific to the 5´ or 3´ region of each gene by quantitative PCR using the LightCycler 480 (Roche). Amplified double-stranded DNA products during 40 cycles were detected by SYBR Green I. All measurements were in quadruplicate and compared with a 16,000-fold range of serially diluted standard genomic DNA prepared from the WT strain. The sequences of primers are available upon request.
To examine telomere growth, WT, JΔ, H3.VΔ, and JΔ H3.VΔ cells were transfected with pSY37, which targets the HYG selection marker right downstream of the active VSG2 and leaves about 200 bp telomere seed at the end of the chromosome. Over time, telomere seeds should elongate and elongation rate of telomeres was measured by Southern blot. About 200 million cells were collected and genomic DNA was prepared using QIAmp kit from Qiagen. Genomic DNA was digested with AflII to measure the lengthening of BES-VSG2-telomere or with BglI to measure the length of silent BES-VSG3 telomere, and separated on an agarose gel. Specific VSG2 or VSG3 probes were used for detection.
|
10.1371/journal.pcbi.0030056 | Query-Dependent Banding (QDB) for Faster RNA Similarity Searches | When searching sequence databases for RNAs, it is desirable to score both primary sequence and RNA secondary structure similarity. Covariance models (CMs) are probabilistic models well-suited for RNA similarity search applications. However, the computational complexity of CM dynamic programming alignment algorithms has limited their practical application. Here we describe an acceleration method called query-dependent banding (QDB), which uses the probabilistic query CM to precalculate regions of the dynamic programming lattice that have negligible probability, independently of the target database. We have implemented QDB in the freely available Infernal software package. QDB reduces the average case time complexity of CM alignment from LN2.4 to LN1.3 for a query RNA of N residues and a target database of L residues, resulting in a 4-fold speedup for typical RNA queries. Combined with other improvements to Infernal, including informative mixture Dirichlet priors on model parameters, benchmarks also show increased sensitivity and specificity resulting from improved parameterization.
| Database similarity searching is the sine qua non of computational molecular biology. Well-known and powerful methods exist for primary sequence searches, such as Blast and profile hidden Markov models. However, for RNA analysis, biologists rely not only on primary sequence but also on conserved RNA secondary structure to manually align and compare RNAs, and most computational tools for identifying RNA structural homologs remain too slow for large-scale use. We describe a new algorithm for accelerating one of the most general and powerful classes of methods for RNA sequence and structure analysis, so-called profile SCFG (stochastic context-free grammar) RNA similarity search methods. We describe this approach, called query-dependent banding, in the context of this and other improvements in a practical implementation, the freely available Infernal software package, the basis of the Rfam RNA family database for genome annotation. Infernal is now a faster, more sensitive, and more specific software tool for identifying homologs of structural RNAs.
| Many functional RNAs conserve a base-paired secondary structure. Conserved RNA secondary structure induces long-distance pairwise correlations in homologous RNA sequences. When performing database searches to identify homologous structural RNAs, it is desirable for RNA similarity search programs to score a combination of secondary structure and primary sequence conservation.
A variety of approaches for RNA similarity searching have been described. There are specialized programs for identifying one particular RNA family or motif, such as programs that identify transfer RNAs [1,2], small nucleolar RNAs [3,4], microRNAs [5,6], signal recognition particle (SRP) RNAs [7], and rho-independent transcription terminators [8]. There are also pattern-matching algorithms that rely on expertly designed query patterns [9]. However, the most generally useful approaches are those that take any RNA (or any multiple RNA alignment) as a query and use an appropriate scoring system to search a sequence database and rank high-scoring similarities [10,11], just as programs like Blast (http://www.ncbi.nlm.nih.gov/BLAST/) do for linear sequence comparison [12].
In a general search program, one wants to score a combination of RNA sequence and structural conservation in a principled rather than an ad hoc manner. A satisfactory solution to this problem is known, using probabilistic models called stochastic context-free grammars (SCFGs). SCFGs readily capture both primary sequence and (non–pseudo-knotted) RNA secondary structure conservation [13,14]. Just as hidden Markov models (HMMs) are useful for many different linear sequence modeling applications, including gene finding, multiple alignment, motif finding, and similarity search [14], SCFGs are a generally useful paradigm for probabilistic RNA sequence/structure analysis, with applications including secondary structure prediction and gene finding. A particular SCFG architecture called covariance models (CMs) was developed specifically for the RNA similarity search problem [15]. CMs are profile SCFGs, analogous to the use of profile HMMs in sequence analysis [15,16]. The Rfam database of RNA families [17] is based on CM software (Infernal [inference of RNA alignment]; http://infernal.janelia.org) in much the same way that the Pfam database of protein families is based on profile HMM software (HMMER; http://hmmer.janelia.org) [18,19].
The most serious problem with using CMs has been their computational complexity. Applying standard SCFG dynamic programming (DP) alignment algorithms to the particular case of CMs results in algorithms that require O(N3) memory and O(LN3) time for a query of length N residues (or consensus alignment columns) and a target database sequence of length L. The memory complexity problem has essentially been solved, by extending divide-and-conquer DP methods (the Hirshberg or Myers/Miller algorithm) to the case of CMs [16], but the time complexity problem still stands.
Weinberg and Ruzzo [20–22] have described several filtering methods for accelerating CM searches. The original idea (“rigorous filters”) was to score a target sequence first by a linear sequence comparison method, using a profile HMM specially constructed from the query CM such that the profile score was provably an upper bound on the CM score; the subset of hits above threshold would then be passed for rescoring with the more expensive CM alignment algorithm [21]. Subsequently a “maximum likelihood heuristic” filter profile was developed that gives up the guarantee of recovering the same hits as the unfiltered search but offers greater speedups [22]. For most current Rfam models, Weinberg–Ruzzo filters give about a 100-fold speedup relative to a full CM-based search at little or no cost to sensitivity and specificity. However, because these filters depend on primary sequence conservation alone, they can be relatively ineffective for RNA families that exhibit poor sequence conservation—unfortunately, precisely the RNAs that benefit the most from SCFG-based search methods. Indeed, in this respect, we are concerned that the overall performance of rigorous filters on the current Rfam database may be somewhat misleading. Rfam currently uses a crude Blast-based filtering method to accelerate the CM searches used in curating the database. This step introduces a bias toward high primary sequence similarity in current Rfam alignments. As Rfam improves and incorporates more diverse structural homologs, the effectiveness of sequence-based filters will decrease. To address this worry, Weinberg and Ruzzo [20] have also described additional heuristics (“sub-CMs” and the “store-pair” technique) that should capture more secondary structure information in the filtering process. Bafna and coworkers [23] have described further improvements to sequence filtering methods. Currently, the Infernal codebase includes Weinberg's C++ implementation of rigorous filters but not, as yet, the ML heuristic, sub-CM, or store-pair methods. All these methods are important, but it also remains important to us to identify yet more methods for accelerating CMs.
Here, we describe a method for accelerating CM searches using a banded DP strategy. In banded DP, one uses a fast method to identify a band through the DP matrix where the optimal alignment is likely to lie and then calculates computationally expensive DP recursions only within that band. In most cases, including our approach, banded DP is a heuristic that sacrifices guaranteed alignment optimality. Banding is a standard approach in many areas of sequence analysis. Gapped Blast uses banded DP to convert ungapped high-scoring pairs (HSPs) to full gapped alignments [12]. LAGAN and Multi-LAGAN (http://lagan.stanford.edu) use banded DP (referred to as limited-area DP) to stitch together alignments between anchored sequences when aligning long genomic sequences [24]. Banding has also been applied to profile SCFGs by Michael Brown in his RNACAD program by using information from a profile HMM alignment to define bands for the expensive SCFG alignment [25]. The key to developing a banded DP strategy is in deciding how the bands are identified. Usually, including all the examples just mentioned, banded DP involves performing some sort of rapid approximate sequence alignment between the query and the target.
In contrast, the method we describe here, called query-dependent banding (QDB), takes advantage of specific properties of CMs in order to predefine bands that are independent of any target sequence. QDB depends on the consensus secondary structure of the query, so it is complementary to acceleration methods such as the Weinberg–Ruzzo filters that rely on sequence but not structure.
Briefly, the key idea is the following. Each base pair and each single-stranded residue in the query RNA is represented in a CM by a state. States are arranged in a treelike structure that mirrors the secondary structure of the RNA, along with additional states to model insertions and deletions. The standard CM DP alignment algorithm works by calculating the probability that a substructure of the query rooted at state v aligns to a subsequence i… j in the target sequence. The calculation is recursive, starting at the leaves of the CM (ends of hairpin loops) and subsequences of length 0, and working upward in larger substructures of the CM, and outward in longer and longer subsequences.
To guarantee optimality, at each v, the DP algorithm must score all possible i… j subsequences in the target sequence. However, most of these subsequences are obviously too long or short, when one considers the size of the query substructure under state v. For example, when state v models the closing base pair of a consensus four-base loop, only i… j subsequences of length six are likely to occur in any optimal alignment to state v; that is, (j − 5,j) being the base pair and (j − 4… j − 1) being the four bases of the hairpin loop. Likewise, the optimal subsequence aligned to the next consensus base pair in that stem is almost certainly of length eight.
Because insertions and deletions may occur in the target sequence, no subsequence length is known with certainty, but because the CM is a probabilistic model, a probability distribution for subsequence lengths under each state (including the probability of insertions and deletions) can be analytically derived from the query CM. These distributions can be used to determine a band of subsequence lengths that captures all but a negligible amount of the probability mass. A CM DP algorithm can then look not at all subsequences i,j for each state v but only those i within a band of minimum and maximum distance relative to each j.
To formalize this idea, we start with a description of CMs, followed by the QDB algorithms for calculating the subsequence length distributions, using these length distributions to determine bands, and using the bands in a banded CM DP alignment algorithm. Calculation of the bands is sensitive to transition parameter estimation, so we describe Infernal's new implementation of informative Dirichlet priors for CM parameter estimation. Finally, we present results from a benchmark that suggest the sensitivity and specificity of a QDB-accelerated search are negligibly different from those of a nonbanded search.
CMs are a convention for mapping an RNA secondary structure into a treelike, directed graph of SCFG states and state transitions (or, equivalently, SCFG nonterminals and production rules). The CM is organized by a binary tree of nodes representing base pairs and single-stranded residues in the query's structure. Each node contains a number of states, where one state represents the consensus alignment to the query, and the others represent insertions and deletions relative to the query. Figure 1 shows an example of converting a consensus structure to the guide tree of nodes and part of the expansion of those guide tree nodes into the CM's state graph. Here, we will only concentrate on the aspects of CMs necessary to understand QDB, and a subset of our usual notation. For full details on CM construction, see [16,26].
A guide tree consists of eight types of nodes. MATP nodes represent consensus base pairs. MATL and MATR nodes represent consensus single-stranded residues (emitted to the left or right with respect to a stem). BIF nodes represent bifurcations in the secondary structure of the family, to deal with multiple stem-loops. A ROOT node represents the start of the model. BEGL and BEGR nodes represent the beginnings of a branch on the left and right side of a bifurcation, respectively. END nodes end each branch.
The CM is composed of seven different types of states, each with a corresponding form of production rule, with notation defined as follows:
That is, for instance, if state v is a pair state, it produces (aligns to and scores) two correlated residues, a and b, and moves to some new state, Y. The probability that it produces a residue pair a,b is given by an emission probability ev(a,b). The probability that it moves to a particular state Y is given by a transition probability tv(Y). The set of possible states Y that v may transit to is limited to the states in the next (lower) node in the guide tree (and insert states in the current node); the set of possible children states Y is called Cv, for “children of v.” The indicators
and
are used to simplify notation in CM DP algorithms. They are the number of residues emitted to the left and right of state v, respectively. Bifurcation rules are special, in that they always transition to two particular start (S) states, at the root of subtrees in the guide tree, with probability 1.0.
These state types essentially define a “normal form” for SCFG models of RNA, akin to SCFGs in Chomsky normal form where all productions are in one of two forms, Y → a or Y → YY. We describe CM algorithms (including QDB) in terms of this normal form. CMs define a specific way that nodes in the guide tree are expanded into states and how those states are connected within each node and to states in the next node in the guide tree. For example, a MATP node that deals with a consensus base pair contains six states called MATP_MP (a P state for matching the base pair), MATP_ML and MATP_MR (an L and an R state for matching only the leftmost or rightmost base and deleting the right or left one, respectively), MATP_D (a D state for deleting the base pair), and MATP_IL and MATP_IR (L and R states with self-transitions, for inserting one or more residues to the left and/or right, respectively, before going to the next node).
Thus, a CM is a generative probabilistic model of homologous RNAs. A sequence is emitted starting at the root, moving downward from state to state according to state transition probabilities, emitting residues and residue pairs according to emission probabilities, and bifurcating into substructures at bifurcation states. An important property of a CM is the states can be numbered from 0 . . . M − 1 (from root to leaves) such that for any state v, the states y that it can transit to must have indices y ≥ v. There are no cycles in a CM, other than self-transitions on insert states. This is the property that enables the recursive calculations that both CM DP alignment algorithms and QDB rely on.
Without any change in the above description, CMs apply to either global or local alignment, and to either pairwise alignment to single RNA queries or profile alignment to a consensus query structure of a multiple RNA sequence alignment. CMs for single RNA queries are derived identically to profiles of a consensus structure, differing only in the parameterization method [27]. Local structural alignment to substructures and truncated structures (as opposed to requiring a global alignment to the whole RNA structural model) is achieved by adding state transitions from the ROOT that permit entering the model at any internal consensus state with some probability, and state transitions from any internal consensus state to an END with some probability [26,27].
Observe that for any state v, we could enumerate all possible paths down the model from v to the END(s). Each path has a certain probability (the product of the transition probabilities used by the path), and it will emit a certain number d of residues (two per P state, one per L or R state in the path). The sum of these path probabilities for each d defines a probability distribution γv(d), the probability that the CM subgraph rooted at v will generate a subsequence of length d. Given a finite limit Z on maximum subsequence length (defined later), we can calculate γv(d) by an efficient recursive algorithm, working from the leaves of the CM toward the root and from smallest subsequences to largest:
for v = M − 1 down to 0:
v = end state (E):
v = bifurcation (B):
else (v = S,P,L,R):
For example, if we are calculating γv(d) where v is a pair state, we know that v must emit a pair of residues and then transit to a new state y (one of its possible transitions Cv), and then a subgraph rooted at y will have to account for the rest of the subsequence of length d − 2. Therefore, γv(d) must be the sum, over all possible states y in Cv, of the transition probability tv(y) times the probability that the subtree rooted at y generates a subsequence of length d − 2, which is γv(d − 2), guaranteed to have already been calculated by the recursion. For the B state (bifurcation) calculation, indices y and z indicate the left and right S (start) state that bifurcation state v must connect to.
A band dmin(v)…dmax(v) of subsequence lengths that will be allowed for each state v is then defined as follows. A parameter β defines the threshold for the negligible probability mass that we are willing to allow outside the band. (The default value of β is set to 10−7, as described later.) We define dmin(v) and dmax(v) such that the cumulative left and right tails of γv(d) contain less than a probability
:
Larger values of β produce tighter bands and faster alignments, but at a cost of increased risk of missing the optimal alignment. β is the only free parameter that must be specified to QDB.
Because CMs have emitting self-loops (i.e., insert states), there is no finite limit on subsequence lengths. However, we must impose a finite limit Z to obtain a finite calculation. Z can be chosen to be sufficiently large that it does not affect dmax(v) for any state v. On a digital computer with floating point precision ɛ (the largest value for which 1 + ɛ = 1), it suffices to guarantee that, for all v:
Empirically, we observe that the tails of the γv(d) densities decrease approximately geometrically. We can estimate the mass remaining in the unseen tail by fitting a geometric tail to the observed density γv(d). Our implementation starts with a reasonable guess at Z and verifies that the above condition is true for each v, assuming these geometrically decreasing tails; if it is not, Z is increased and bands are recalculated until it is.
A QDB calculation needs to be performed only once per query CM to set the bands. Overall, a QDB calculation requires Θ(MZ) in time and space, or, equivalently, because both M and Z scale roughly linearly with the length L in residues of the query RNA, Θ(L2). The time and space requirement is negligible compared with the requirements of a typical CM search.
A standard algorithm for obtaining the maximum likelihood alignment (parse tree) of an SCFG to a target sequence is the Cocke–Younger–Kasami (CYK) DP algorithm [28−30]. Formally, CYK applies to SCFGs reduced to Chomsky normal form, and it aligns to the complete sequence. The CM database search algorithm is a CYK variant, specialized for the “normal form” of our seven types of RNA production rules and for scanning long genomic sequences for high-scoring subsequences (hits) [14].
The CM search algorithm recursively calculates αv(j,d), the log probability of the most likely CM parse subtree rooted at state v that generates (aligns to) the length d subsequence xj−d+1… xj that ends at position j of target sequence x [14,15]. This calculation initializes at the smallest subgraphs (E states) and shortest subsequences (d = 0) and iterates upward and outward to progressively larger subtrees and longer subsequences up to a preset window size W. The outermost loop iterates over the end position j on the target sequence, enabling an efficient scan across a long target like a chromosome sequence. Banding is achieved simply by limiting all loops over possible subsequence lengths d to the bounds dmin(v)…dmax(v) derived in the band calculation algorithm, rather than all possible lengths 0…W. The banded version of the algorithm is as follows:
For example, if we are calculating αv(j,d) and v is a pair state (P), v will generate the base pair xj−d+1,xj and transit to a new state y (one of its possible transitions Cv), which then will have to account for the smaller subsequence xj−d+2… xj−1. The log probability for a particular choice of next state y is the sum of three terms: an emission term log ev (xj−d+1,xj), a transition term log tv(y), and an already calculated solution for the smaller optimal parse tree rooted at y, αy(j – 1,d – 2). The value assigned to αv(j,d) is the maximum over all possible choices of child states y that v can transit to.
The W parameter defines the maximum size of a potential hit to a model. Previous Infernal implementations required an ad hoc guess at a reasonable W. The band calculation algorithm delivers a probabilistically derived W for database search in dmax(0), the upper bound on the length of the entire sequence (the sequence generated from the root state of the CM).
In our implementation, this algorithim is encoded in a more memory-efficient form that allocates space for only two sequence positions in j (current and previous) for most states rather than for all j = 0…L, using essentially the same techniques described for CYK search in [14]. We have omitted the necessary details here for clarity. QDB does not reduce the asymptotic computational complexity of the CM search algorithm. Both the banded algorithm and the original algorithm are O(MW + BW2) memory and O(L(MW + BW2)) time, for a model of M states containing B bifurcation states, window size W of residues, and target database length L. M, B, and W all scale with the query RNA length N, so roughly speaking, worst-case asymptotic time complexity is O(LN3).
The subsequence length distributions calculated by QDB depend on the CM's transition probabilities. Transition probability parameter estimation is therefore crucial for obtaining predicted subsequence length bands that reflect real subsequence lengths in homologous RNA targets. Transition parameters in Infernal are mean posterior estimates, combining (ad hoc weighted) observed counts from an input RNA alignment with a Dirichlet prior [26]. Previous to this work, Infernal used an uninformative uniform Dirichlet transition prior, equivalent to the use of Laplace “plus-1” pseudo-counts. However, we found that transition parameters derived under a uniform prior inaccurately predict target subsequence lengths, as shown in an example in Figure 2. The problem is exacerbated when there are few sequences in the query alignment, when the choice of prior has more impact on mean posterior estimation. To alleviate this problem, we estimated informative single component Dirichlet prior densities for CM transition parameters, as follows.
The training data for transition priors consisted of the 381 seed alignments in the Rfam database, version 6.1 [17]. For each alignment, we built CM structures by Infernal's default procedure and collected weighted counts of observed transitions in the implied parse trees of the training sequences. Considering all possible combinations of pairs of adjacent node types, there are 73 possible distinct types of transition probability distributions in CMs. To reduce this parameter space, we tied these 73 distributions into 36 groups by assuming that certain distributions were effectively equivalent. Thirty-six Dirichlet densities were then estimated from these pooled counts by maximum likelihood as described in [31], with the exception that we optimize by conjugate gradient descent [32] rather than by expectation–maximization (EM). The results, including the Dirichlet parameters, are given in Table 1. Using these priors for transition probability parameter estimation results in an improvement in the utility of QDB calculations, often yielding tighter, yet accurate subsequence length distributions, as illustrated by anecdotal example in Figure 2.
We also estimated informative mixture Dirichlet density priors for emission probabilities. Emission probabilities have no effect on QDB, but informative emission priors should improve sensitivity and specificity of CM searches, as they do for profile HMMs [31,33]. We collected filtered counts of aligned single-stranded residues and base pairs from annotated ribosomal RNA alignments from four alignments in the 2002 version of the European Ribosomal RNA Database [34,35]: large subunit rRNA (LSU), bacterial/archaeal/plastid small subunit rRNA (SSU-bap), eukaryotic SSU rRNA (SSU-euk), and mitochondrial SSU rRNA (SSU-mito). These alignments were filtered, removing sequences in which either less than 40% of the base-paired positions are present or more than 5% of the nucleotides are ambiguous, and removing selected sequences based on single-linkage clustering such that no two sequences in a filtered alignment were greater than 80% identical (in order to remove closely related sequences). Summary statistics for the filtered alignments and collected counts in the training data set are given in Table 2. These data were used to estimate a nine-component Dirichlet mixture prior for base pairs and an eight-component Dirichlet mixture prior for single-stranded residues. The base pair prior is given in Table 3, and the singlet residue prior is given in Table 4.
The reason to use two different data sets to estimate transition versus emission priors is the following. Rfam provides many different structural RNA alignments but of uneven quality and varying depth (number of sequences). The European rRNA database provides a small number of different RNA alignments but of high quality and great depth. A transition prior training set should be maximally diverse, so as not to bias any transition types toward any particular RNA structure, so we used the 381 different Rfam alignments for transitions. Emission prior estimation, in contrast, improves with alignment depth and accuracy but does not require broad structural diversity per se, so we used rRNA data for emissions.
Inspection of the Dirichlet α parameters shows sensible trends. In the transition priors, transitions between main (consensus) states are now favored (higher α values) relative to insertions and deletions. In the base pair emission mixture prior, all components favor Watson–Crick and G-U pairs, with different components preferring different proportions of pairs in a particular covarying aligned column (for instance, component 1 likes all four Watson–Crick pairs, component 2 describes covarying conservation of CG,UA,UG pairs, and component 3 specifically likes conserved CG pairs), and the mean α parameters prefer GC/CG pairs over AU/UA pairs. In the singlet emission mixture prior, some components are capturing strongly conserved residues (component 1 favors conserved U's, for example) while other components favor more variation (components 4 and 5, for example), and the marginal α parameters show a strong A bias, reflecting the known bias for adenine in single-stranded positions of structural RNAs (especially ribosomal RNAs).
There is redundancy between some components (notably 5 and 8 in the base pair mixture and 2, 3 and 8 in the singlet mixture). This is typical for statistical mixture estimation, which (unlike, say, principal components analysis) does not guarantee independence between components. The decision to use nine pair and eight singlet components was empirical, as these priors performed better than priors with fewer components on the benchmark we describe below (unpublished data).
Note that all singlet positions are modeled with one singlet mixture prior distribution, and all base pairs are modeled with one base pair mixture prior. These priors do not distinguish between singlet residues in different types of loops, for example, or between a stem-closing base pair versus other base pairs. In the future, it may prove advantageous to adopt more complex priors to capture effects of structural context on base pair and singlet residue preference.
In another step to increase sensitivity and specificity of the program, we adopted the “entropy weighting” technique described for profile HMMs [36] for estimating the total effective sequence number for an input query alignment. This is an ad hoc method for reducing the information content per position of a model, which helps a model that has been trained on closely related sequences to recognize distantly related homologs [37]. In entropy weighting, one reduces the total effective sequence number (which would normally be the actual number of sequences in the input alignment), thereby increasing the influence of the Dirichlet priors, flattening the transition and emission distributions, and reducing the overall information content. We approximate a model's entropy as the mean entropy per consensus residue, as follows. Let C be the set of all MATP_MP states emitting consensus base pairs (a,b), and let D be the set of all MATL_ML and MATR_MR states emitting consensus singlets (a); the entropy is then calculated as:
For each input multiple alignment, the effective sequence number is set (by bracketing and binary search) so as to obtain a specified target entropy. The target entropy for Infernal is a free parameter, which we optimized on the benchmark described below to identify our default value of 1.46 bits.
To assess the effect of QDB, informative priors, and entropy weighting on the speed, sensitivity, and specificity of RNA similarity searches, we designed a benchmark based on the Rfam database [17]. The benchmark was designed so that we would test many RNA query/target pairs, with each query consisting of a given RNA sequence alignment, and each target consisting of a distantly related RNA homolog buried in a context of a random genome-like background sequence.
We started with seed alignments from Rfam version 7.0. In each alignment, sequences shorter than 70% of the median length were removed. We clustered the sequences in each family by single-linkage clustering by percent identity (as calculated from the given Rfam alignment) and then split the clusters such that the training set and test sequences satisfied three conditions: (1) no training/test sequence pair is more than 60% identical; (2) no test sequence pair is greater than 70% identical; and (3) at least five sequences are in the training set. Fifty-one families satisfy these criteria (listed in Table 5), giving us 51 different query alignments (containing 5 to 1,080 sequences each) and 450 total test sequences (from 1 to 66 per query). We embedded the test sequences in a 1-Mb “pseudo-genome” consisting of twenty 50-kb “chromosomes,” generated as independent, identically distributed (iid) random sequences with uniform base frequencies. The 450 test sequences were embedded into this sequence by replacement, by randomly choosing a chromosome, orientation, and start position, and disallowing overlaps between test sequences. The total length of the 450 test sequences is 101,855 nucleotides, leaving 898,145 nucleotides of random background sequence.
The benchmark proceeds by first building a CM for each query alignment and then searching the pseudo-genome with each CM in local alignment mode. All hits above a threshold of 8.0 in raw bit score for each of the 51 queries were sorted by score into 51 ranked family-specific lists, as well as one ranked master list of all 51 sets of scores. Each hit is classified into one of three categories: “positive,” “ignore,” or “negative.” A “positive” is a hit that significantly overlaps with a true test sequence from the same family as the query. An “ignore” is a hit that significantly overlaps with a test sequence from a different family, where “significantly overlap” means that the length of overlap between two sequences (either two hits, or one hit and one test sequence embedded in the pseudo-genome) is more than 50% of the length of the shorter sequence. (Although it would be desirable to measure the false-positive rate on nonhomologous structural RNAs, we cannot be sure that any given pair of Rfam families is truly nonhomologous. Like most sequence family databases, Rfam is clustered computationally, and more sensitive methods will reveal previously unsuspected relationships that should not be benchmarked as “false positives.”) A “negative” is a hit that is not a positive or an ignore. For any two negatives that significantly overlap, only the one with the better score is counted.
The minimum error rate (MER) (“equivalence score”) [38] was used as a measure of benchmark performance. The MER score is defined as the minimum sum of the false positives (negative hits above the threshold) and false negatives (true test sequences that have no positive hit above the threshold), at all possible choices of score threshold. The MER score is a combined measure of sensitivity and specificity, where a lower MER score is better. We calculate two kinds of MER scores. For a family-specific MER score, we choose a different optimal threshold in each of the 51 ranked lists, and for a summary MER score, we choose a single optimal threshold in the master list of all hits. The summary MER score is the more relevant measure of our current performance, because it demands a single query-independent bit score threshold for significance. A family-specific MER score reflects the performance that could be achieved if Infernal provided E-values (currently, it reports only raw bit scores).
For comparison, BlastN was also benchmarked on these data using a family-pairwise search (FPS) procedure [39]. For each query alignment, each training sequence is used as a query sequence to search the pseudo-genome, all hits with an E-value of less than 1.0 were sorted by increasing E-value, and the lowest E-value positive hit to a given test sequence is counted.
Using this benchmark, we addressed several questions about QDB's performance.
What is the best setting of the single QDB free parameter, β, which specifies how much probability mass to sacrifice? Figure 3 shows the average speedup per family and summary MER score as a function of varying β. There is no clear choice. The choice of β is a tradeoff of accuracy for speed. We chose a default of β = 10−7 as a reasonable value that obtains a modest speedup with minimal loss of accuracy.
How well does QDB accelerate CM searches? Figure 4 shows the time required for searching the 1-Mb benchmark target sequence with each of the 51 models, as a function of the average query RNA length. QDB reduces the average-case running time complexity of the CM search algorithm from LN2.36 to LN1.32. Observed accelerations relative to the standard algorithm range from 1.4-fold (for the IRE, iron response element) to 12.7-fold (for the 5′ domain of SSU rRNA), with an average speedup per family of 4.2-fold. In total search time for the benchmark (sum of all 51 searches), the acceleration is 6-fold, because large queries have disproportionate effect on the total time.
How much does QDB impact sensitivity and specificity? Optimal alignments are not guaranteed to lie within QDB's high-probability bands. This is expected to compromise sensitivity. The hope is that QDB's bands are sufficiently wide and accurate that the loss is negligible. Figure 5 shows ROC plots (sensitivity versus false-positive rate) on the benchmark for the new version of Infernal (version 0.72) in standard versus QDB mode. These plots are nearly superposed, showing that the loss in accuracy is small at the default QDB setting of β = 10−7.
How much do our changes in parameterization (the addition of informative Dirichlet priors and entropy weighting) improve sensitivity and specificity? Figure 5 shows that the new Infernal 0.72 is a large improvement over the previous Infernal version 0.55, independent of QDB. (On average, in this benchmark, Infernal 0.55 is no better than a family-pairwise search with BlastN.) Table 6 breaks this result down in more detail, showing summary and family-specific MER scores for a variety of combinations of prior, entropy weighting, and QDB. These results show that both informative priors and entropy weighting individually contributed large improvements in sensitivity and specificity.
CM searches take a long time, and this is the most limiting factor in using the Infernal software to identify RNA similarities. Prior to this work, Infernal 0.55 required 508 CPU-hours to search 51 models against just 1 Mb of sequence in our benchmarks (Table 5). Using QDB with β banding cutoffs that do not appreciably compromise sensitivity and specificity, Infernal 0.72 offers a 6-fold speedup, performing the benchmark in 85 hours. Our eventual goal is to enable routine genome annotation of structural RNAs: to be able to search thousands of RNA models against complete genome sequences. A search of all 503 Rfam 7.0 models against the 3-GB human genome with Infernal 0.72 in QDB mode would take on the order of 300 CPU-years (down from 1,800 CPU years with Infernal 0.55). We need to be able to do it in, at the most, a few days, so we still need to increase CM search speed by five orders of magnitude. Thus, the QDB algorithm is a partial but certainly not complete solution to the problem. However, QDB combines synergistically with other acceleration techniques. Parallelization, on large clusters (although prohibitively expensive for all but a few centers), could give us further acceleration of three orders of magnitude. Software improvement (code optimization) will also contribute but probably only about 2-fold. Hardware improvements will contribute about 2-fold per year or so as long as Moore's law continues. Finally, QDB is complementary to the filtering methods recently described by Weinberg and Ruzzo [20−22]. We view QDB as part of a growing suite of approaches that we can combine to accelerate Infernal.
Is it really worth burning all this CPU time in the first place? Do CM searches identify structural RNA homologies that other methods miss? Obviously we think so, but one would like to see convincing results. For large, diverse RNA families like transfer RNA, where a CM can be trained on more than 1,000 well-aligned sequences with a well-conserved consensus secondary structure, CM approaches have been quite powerful. The state of the art in large-scale transfer RNA gene identification remains the CM-based program tRNAscan-SE [1], and CMs were also used, for example, to discover the divergent tRNA for pyrrolysine, the “22nd amino acid” [40]. But Figure 5 shows that on average, in more than 51 more or less “typical” RNA families of various sizes and alignment quality, Infernal 0.55 was actually no better than doing a family-pairwise search with BlastN. Until recently, we have spent relatively little effort on how Infernal parameterizes its models and relatively more on reducing its computational requirements [16], so previous versions of Infernal have performed best where naive parameterization works best: on very large, high-quality alignments of hundreds of sequences, which are atypical of many interesting homology search problems.
In this work, partly because the level of acceleration achieved by QDB is sensitive to transition parameterization, we have brought Infernal parameterization close to the state of the art in profile HMMs, by introducing mixture Dirichlet priors [31] and entropy weighting [36]. This resulted in a large improvement in the sensitivity and specificity of searches, as judged by our benchmark (Figure 5). The difference between Infernal and family-pairwise BlastN now appears pronounced for average-case behavior, not just best-case behavior. However, while we trust our benchmarking to tell us we have greatly improved Infernal relative to previous versions of itself, our benchmarking does not allow us to draw firm conclusions about our performance relative to other software. For that, we prefer to see independent benchmarks. Benchmarks by tool developers are notoriously biased, and however honest we may try to be, some biases are essentially unavoidable. For one thing, establishing an internal benchmark for ongoing code development creates an insidious form of training on the test set, because we accept code changes that improve benchmark performance. In particular, we set the entropy weighting target of 1.46 bits and the numbers of mixture prior components by optimizing against our benchmark. Further, our benchmark does not use a realistic model for the background sequence of the “pseudo-genome,” because we construct the background as a homogeneous independent, identically distributed (iid) sequence, and this poorly reflects the heterogeneous and repetitive nature of genomic sequence. This benchmark should be sufficient for an internal comparison of versions 0.55 and 0.72 of Infernal, because we have not altered how Infernal deals with heterogeneous compositional bias. But we cannot safely draw conclusions from our simple benchmark about the relative performance of Infernal and Blast on real searches, for example, because Blast may (and in fact does) treat sequence heterogeneity better than Infernal does. In this regard, currently we are aware of only one independent benchmark BRaliBase III [41]. BRaliBase III consists of many different query alignments of five or 20 RNA sequences, drawn from three different RNA families (U5, 5S rRNA, and transfer RNA). These authors' results broadly confirm our internal observations: while Infernal 0.55 showed mediocre performance compared with BlastN and several other tools, a recent version of Infernal stood out as a superior method for RNA similarity search.
Nonetheless, although Infernal 0.72 shows large improvements in speed, sensitivity, and specificity over previous versions, there are numerous areas where we need to improve further.
A significant gap in our current implementation is that Infernal reports only raw bit scores and does not yet report expectation values (E-values). CM local alignment scores empirically follow a Gumbel (extreme value) distribution [27], just as local sequence alignment scores do [42], so there are no technical hurdles in implementing E-values. This will be an immediate focus for the next version of Infernal. E-value calculations not only have the effect of reporting statistical significance (more meaningful to a user than a raw bit score) but also normalize each family's score distribution into a more consistent overall rank order, because different query models exhibit different null distributions (particularly in the location parameter of the Gumbel distribution). We therefore expect E-values to contribute a large increase in performance whenever a single family-independent threshold is set. Table 6 roughly illustrates the expected gain, by showing the large difference between summary MER scores and family-specific MER scores.
Parameterization of both CMs and profile HMMs remains problematic, because these methods continue to assume that training sequences are statistically independent, when in fact they are related (often strongly so) by phylogeny. Methods like sequence weighting and entropy weighting do help, but they are ad hoc hacks: unsatisfying and unlikely to be optimal. Even mixture Dirichlet priors, although they appear to be mathematically sophisticated, fundamentally assume that observed counts are drawn as independent multinomial samples, and therefore the use of Dirichlet priors is fundamentally flawed. Probabilistic phylogenetic inference methodology needs to be integrated with profile search methods. This is an area of active research [43−45] in which important challenges remain, particularly in the treatment of insertions and deletions.
Finally, QDB is not the only algorithmic acceleration method we can envision. Michael Brown described a complementary banding method to accelerate his SCFG-based RNACAD ribosomal RNA alignment software [25], in which he uses profile HMM-based sequence alignment to the target to determine bands where the more rigorous SCFG-based alignment should fall (because some regions of the alignment are well-determined based solely on sequence alignment). The gapped Blast algorithm (seed word hits, ungapped hit extension, and banded DP) can conceivably be extended from two-dimensional sequence alignment to three-dimensional CM DP lattices. Developing such algorithms, and incorporating them into a widely useful, freely available codebase, are priorities for us.
The version and options used for Blast in our benchmark are WU-BLASTN-2.0MP -kap -W = 7. For Infernal, versions 0.55 and 0.72 were used as indicated. The complete Infernal software package, including documentation and the Rfam-based benchmark described here, may be downloaded from http://infernal.janelia.org. It is developed on GNU/Linux operating systems but should be portable to any POSIX-compliant operating system, including Mac OS/X. It is freely licensed under the GNU General Public License.
The ANSI C code we used for estimating maximum likelihood mixture Dirichlet priors depends on a copyrighted and nonredistributable implementation of the conjugate gradient descent algorithm from Numerical Recipes in C [32]. Our code, less the Numerical Recipes routine, is freely available upon request. |
10.1371/journal.pbio.0050102 | High Incidence of Non-Random Template Strand Segregation and Asymmetric Fate Determination In Dividing Stem Cells and their Progeny | Decades ago, the “immortal strand hypothesis” was proposed as a means by which stem cells might limit acquiring mutations that could give rise to cancer, while continuing to proliferate for the life of an organism. Originally based on observations in embryonic cells, and later studied in terms of stem cell self-renewal, this hypothesis has remained largely unaccepted because of few additional reports, the rarity of the cells displaying template strand segregation, and alternative interpretations of experiments involving single labels or different types of labels to follow template strands. Using sequential pulses of halogenated thymidine analogs (bromodeoxyuridine [BrdU], chlorodeoxyuridine [CldU], and iododeoxyuridine [IdU]), and analyzing stem cell progeny during induced regeneration in vivo, we observed extraordinarily high frequencies of segregation of older and younger template strands during a period of proliferative expansion of muscle stem cells. Furthermore, template strand co-segregation was strongly associated with asymmetric cell divisions yielding daughters with divergent fates. Daughter cells inheriting the older templates retained the more immature phenotype, whereas daughters inheriting the newer templates acquired a more differentiated phenotype. These data provide compelling evidence of template strand co-segregation based on template age and associated with cell fate determination, suggest that template strand age is monitored during stem cell lineage progression, and raise important caveats for the interpretation of label-retaining cells.
| For each chromosome, the complementary DNA strands consist of a “younger” strand synthesized during the most recent round of DNA replication and an “older” strand synthesized during a previous cell division. When the strands separate to serve as templates for DNA synthesis during a subsequent round of replication, the two sister chromatids formed thus differ in terms of the template strand age. The “immortal strand hypothesis” predicts that a stem cell is capable of distinguishing between chromatids based on template age: when it divides, the self-renewing daughter will inherit the chromatids with the older templates, whereas the daughter destined to differentiate will inherit those with the newer templates. However, in vivo evidence in support of this hypothesis has been sparse. By labeling newly synthesized DNA in sequential divisions of stem/progenitors during muscle regeneration, we observed that almost half of the dividing cells sorted their chromatids based on template age. The more stem-like daughter inherited chromatids with older templates, and the more differentiated daughter inherited chromatids with younger templates. We propose that this phenomenon is a characteristic of asymmetrically dividing stem cells and their progeny.
| How stem cells maintain genetic and epigenetic constancy throughout repeated divisions is currently unknown. According to the “immortal strand hypothesis” [1], as the stem cell divides asymmetrically, it selectively retains those sister chromatids containing the older template DNA strands in the daughter destined to be the renewed stem cell, thus passing the younger strands (with any mutations acquired during replication), to the tissue-committed cell. This phenomenon of template strand segregation was originally based on observations in embryonic fibroblasts [2] and supported by evidence from dividing cells in the intestinal epithelium [3]. Little additional evidence in support of this hypothesis was reported until recently when the immortal DNA hypothesis was revisited, and evidence in support of this process was detected in vitro in immortalized mouse tumor cells [4] and neurosphere cultures [5], and in vivo in intestinal [6], mammary [7], and muscle [8] stem cells. However, the in vivo examples of strand segregation have been limited to at most a few percent of the cells. Thus the phenomenon has yet to be broadly accepted and is attributed to a curious, but minor, cell population.
In studies (unpublished data) of the timing of proliferation and renewal of skeletal muscle stem cells, or “satellite cells,” we used different halogenated thymidine analogs (bromodeoxyuridine [BrdU], chlorodeoxyuridine [CldU], and iododeoxyuridine [IdU]) delivered at different times during regeneration to label sequential cell divisions. To our surprise, although proliferating cells incorporated both labels when we sequentially delivered two of the analogs, approximately half of the cells that ultimately returned to quiescence contained only the second label. Theoretically, this could be explained by the ability of the self-renewing cells to selectively retain the sister chromatids with the older, unlabeled template strands, consistent with the immortal strand hypothesis [1]. We thus examined myogenic progenitors during regeneration for direct evidence of segregation of older and newer template strands.
Muscle was injured to induce regeneration, and 2 d later, pulses of CldU were administered followed by pulses of IdU approximately 12 h later. As such, cells were labeled with CldU during one replicative cycle, and with IdU during the subsequent round of DNA replication (Figure 1). Cells were then isolated, plated singly, and after allowing a short time for individual cells to complete mitosis, the cells were fixed and immunostained for CldU and IdU. The cell pairs were clearly of the myogenic lineage because this procedure yields cell pairs that are nearly all positive for Syndecan-4 and Pax7, well-established myogenic markers [9], and the pairs are clearly replicating as demonstrated by expression of Ki67 (Figure 2A). Daughter cell pairs were analyzed for the distribution of the two labels. Nearly all the pairs of cells were labeled with IdU, confirming that they had undergone DNA replication during the more recent IdU pulse. However, strikingly, we observed asymmetric inheritance of CldU, with all of the detected label in only one daughter cell (Figure 2B), indicating that during the final cell division, one daughter cell had excluded those chromatids containing the template DNA that was labeled during the earlier cell division (see Figure 1). Even more striking was the fact that this was not a rare event whatsoever; this occurred in nearly half of the pairs (Figure 2C). We also examined the inheritance of labeled DNA in an ex vivo system in which satellite cells activate and proliferate while still associated with individual muscle fibers in culture [10,11]. We again observed a very high frequency of co-segregation of labeled chromatids (Figure 2C). By contrast, when the same labeling procedure was applied to proliferating myoblasts, asymmetric segregation of label was seen in only 5% of pairs (Figure 2C). Although markedly less frequent than in satellite cells activated in vivo or ex vivo, this still reflects a mechanism that is maintained in myogenic cells throughout replicative expansion, perhaps by the presence of the few stem-like cells that are propagated in myoblast cultures [12,13]. This suggests that the mechanisms underlying template strand segregation may be most active in cells maintained in the stem cell niche.
As a further confirmation that activated satellite cells asymmetrically segregate chromatids based on template age, we used a single-label protocol. Muscle was injured and BrdU was injected early during regeneration, comparable to the timing of the CldU pulse in the previous studies. The following day, after cells that had incorporated BrdU would have divided, giving rise to two BrdU-labeled daughters, cells were isolated, plated singly, and treated with either cytochalasin D or nocodazole for several hours to arrest cytokinesis. The cells were fixed and immunostained for BrdU to test for asymmetric segregation of the BrdU label. Of the pairs showing any BrdU label, again about half clearly showed only one daughter cell inheriting the BrdU (Figure 2D).
According to the immortal strand hypothesis, the daughter cell inheriting the older template DNA is the renewing stem cell, whereas the other daughter acquires a more differentiated phenotype. However, our experimental protocol was not designed to test specifically for self-renewal, but rather focused on the proliferative expansion and myogenic lineage progression of the stem cell progeny. During myogenic lineage progression, satellite cells differentiate into fusion-competent, Desmin-expressing myoblasts [14,15]. We examined cell pairs exhibiting asymmetries in inheritance of DNA templates for the expression of Desmin to test if one daughter cell of each pair was more differentiated than the other and if specific templates segregated with specific cell fates. Pairs that showed asymmetric BrdU staining and any evidence of Desmin expression were further characterized and quantified. Strikingly, the great majority of pairs (79%) showed Desmin expression only in the daughter inheriting BrdU-labeled templates (Figure 3A and 3B), indicating a direct correlation between the inheritance of the younger template and the acquisition of a more differentiated fate, consistent with the underlying assumptions of the immortal strand hypothesis. Much smaller percentages of such pairs showed either asymmetric Desmin expression, but with the Desmin expressed in the BrdU-negative cell (18%), or symmetric Desmin expression (Figure 3B). This suggests that template strand co-segregation may not be limited to stem cell self-renewal, but may in fact occur more generally during stem cell expansion when asymmetric cell divisions or divergent daughter cell fates are determined.
Of the pairs in which BrdU-labeled templates were symmetrically inherited (about 50% of total cell pairs), nearly all were also symmetric for Desmin expression, either both positive (59%), reflecting symmetric divisions of myoblasts, or both negative (31%), reflecting symmetric divisions of early progenitors (Figure S1). Only a very small minority of pairs (9%) with symmetric BrdU showed asymmetric Desmin expression (Figure S1).
Several studies have identified Sca-1 as a marker of undifferentiated progenitors derived from skeletal muscle, demonstrating that satellite cells can give rise to progeny that express Sca-1 at least transiently [16–18]. Because of the treatments needed to detect both BrdU and Desmin immunohistochemically, we were not able to detect Sca-1 in populations that were also stained for both Desmin and BrdU. We could, however, detect clear Sca-1 immunostaining under milder conditions and compare its expression with Desmin in cell pairs. Given the strong correlation between asymmetric Desmin and asymmetric BrdU staining (Figure 3A and 3B), we used asymmetric Desmin as a surrogate marker of asymmetric inheritance of labeled template strands. Among pairs with asymmetric Desmin, the vast majority (84%) also showed asymmetric Sca-1 expression, with Desmin and Sca-1 being mutually exclusive (Figure 4A and 4B). This finding is consistent with the immortal strand hypothesis prediction that the cell inheriting the older template (in this case, the Desmin− cell) is the more undifferentiated cell as reflected by the expression of Sca-1. Only very rarely were pairs detected in which Desmin was expressed asymmetrically and Sca-1 was expressed (whether asymmetrically or symmetrically) in the Desmin+ cell (Figure 4B). Virtually all pairs expressing Desmin symmetrically did not express Sca-1 (Figure S2), consistent with high Desmin expression specifying a more differentiated myoblast and Sca-1 expression reflecting a more immature progenitor.
Using these paired cell assays, our data are thus supportive of the immortal strand hypothesis and suggest that template strand segregation is occurring in a large percentage of satellite cell progeny coincident with cell fate decisions. The more immature, Sca-1+ cells undergo asymmetric divisions in which the oldest (unlabeled, in our studies) templates segregate to the daughter that retains the less differentiated phenotype as reflected by Sca-1 expression. The other daughter, by contract, acquires the newer templates (labeled, in our studies, with CldU in the double-label experiments [Figure 2] or BrdU in the single-label experiments [Figure3]) and adopts a more differentiated phenotype as reflected by Desmin expression. These results would predict that the percentage of Sca-1+ cells that are also CldU+ (using the double-label protocol) would decrease through a round of cell division, whereas the percentage of Sca-1− cells that are also CldU+ would increase. To test this directly, we performed experiments as in Figure 2, but at the time of isolation, half of the cells were fixed immediately as a “before division” snapshot of the population. The other half was cultured for an additional 12 h before harvest and fixation as the “after division” population. Cells were then immunostained for Sca-1, CldU (the presumed younger template), and IdU (incorporated into the complimentary DNA strands of divided cells), and analyzed by fluorescence-activated cell sorting (FACS). The proportion of cells expressing Sca-1 and labeled with CldU decreased substantially during this time, whereas the Sca-1−, CldU+ proportion increased by a corresponding percentage (Figure 5). This is consistent with parental Sca-1+ cells segregating older (unlabeled) and younger (CldU-labeled) templates into two daughters that acquire different fates. These results are also consistent with the proportions of cells asymmetrically inheriting template strands and expressing Sca-1 that we observed in the paired cell assays above.
We propose a model of muscle stem cell proliferation in which muscle progenitor cells divide asymmetrically to generate both myoblasts and immature, undifferentiated cells (some of which are likely destined to return to quiescence as replacement satellite cells in vivo), and symmetrically in order to expand either the pool of progenitors or fusion-competent myoblasts necessary to promote effective muscle repair. The finding of asymmetric inheritance of template strands in the case of the asymmetric divisions and the association of the older templates and the more undifferentiated phenotype is compelling evidence in support of the immortal strand hypothesis, but extends the association of template strand co-segregation to a much broader range of stem cell lineage decisions than just self-renewal. Mechanistically, our data suggest that there must be an ongoing monitoring of template strand age and a process to segregate those strands according to age in a sequential manner, not merely the existence of one immortal strand.
The extraordinarily high frequency of muscle progenitor cells exhibiting template strand segregation during muscle regeneration, as opposed to the low frequencies observed in other in vivo systems [6,7], promises to make this system valuable to study mechanisms of asymmetric inheritance of DNA template strands. The high frequency observed in our in vivo studies may relate to the fact that we analyzed cells for asymmetric inheritance of template strands during the process of tissue repair, whereas other in vivo studies have sought evidence of this process during normal homeostatic turnover of tissues [6,7].
In addition to providing strong support for the immortal strand hypothesis and expanding the scope of that hypothesis, the findings presented here have additional important implications. First, the assessment of the proliferation kinetics of stem and progenitor cells has been carried out in many tissues by analyzing the dilution of label incorporated into DNA [19,20]. The ability of stem or progenitor cells to segregate all label to only one daughter would clearly confound the interpretation of all studies that have heretofore assumed equivalent distribution of label to daughter cells and a simple geometric relationship between label dilution and replicative history. Second, our data require a careful analysis of the use of “label retention” to identify stem cells in tissues, based on the assumption that label retention is equated with very long cell cycle times or quiescence [21–24]. Rather, our data suggest an alternative process by which a cell, even a rapidly dividing cell, could take up label and generate label-retaining progeny. If labeled chromatids continue to co-segregate through repeated rounds of DNA replication and cell division (see Figure 1), then label-retaining cells can, theoretically, be maintained indefinitely. Accordingly, such a cell would have an indeterminate replicative history since the time the label was administered. The other implication of this caveat is that a label-retaining cell could represent any stage along the lineage from the most undifferentiated stem cell to the most differentiated progeny. In our studies, the label-retaining cell was, in fact, the more committed of the two daughters, and the label-excluding cell was the more undifferentiated of the two. Clearly, the mechanisms that result in label-retaining cells in any tissue may be more complex than simply long cell cycle times, and the relationship between label retention and stage of differentiation may likewise vary from tissue to tissue and under different biological contexts.
Mouse antibody clone B44 recognizing IdU (and also BrdU) was obtained from BD Biosciences (San Diego, California, United States); rat antibody clone BU1/75 (ICR1) recognizing CldU (and also BrdU) and rat anti-Sca-1 were from Novus Biologicals (Littleton, Colorado, United States). Mouse and rabbit anti-Desmin antibodies were purchased from Sigma (St. Louis, Missouri, United States) and used at 1:200. Mouse hybridoma supernatant anti-Pax7 was from the Developmental Studies Hybridoma Bank (http://www.uiowa.edu/~dshbwww/) and used at a dilution of 1:5. Chicken IgY anti-Syndecan-4 was a generous gift from Dr. Brad Olwin (University of Colorado) and was used at 1:3,000. Rat anti Ki67 was from DakoCytomation (Glostrup, Denmark) and was used at 1:50. Isotype-matched antibodies were used as controls. Antibody staining was performed as previously described [15]. Unless otherwise indicated, primary antibodies were used at 0.5–1 μg/ml. Higher concentrations resulted in detectable cross-reactivity for the antibodies against the halogenated thymidine analogs. Secondary antibodies were Alexa 488- or 546-coupled anti-rat, anti-rabbit, anti-chicken, or anti-mouse antibodies (Invitrogen/Molecular Probes, Carlsbad, California, United States) used at 1:2,000 for immunofluorescence microscopy.
For detection of labeled DNA, cells were fixed in 70% ethanol, washed in PBS, denatured in 2.5 M HCl for 30 min, and permeabilized in 0.25% Triton-X-100 for 5 min before incubation with primary antibodies overnight in PBS/5% fetal bovine serum. For Pax7 detection, cells were fixed in paraformaldehyde and incubated as above with Triton. For Sca-1 and Syndecan-4 labeling, cells were fixed in 4% paraformaldehyde, washed, and incubated with primary antibody (in PBS/5% fetal bovine serum for Sca-1; and in 10% BlockHen [Aves Labs, Tigard, Oregon, United States] for Syndecan-4) overnight without any detergent permeabilization [17].
Muscle injury was induced by the injection of 1–2 μl of cardiotoxin I (100 μg/ml; Sigma) into 24 sites in muscles of the limb. This produces a diffuse necrotic injury and results in the activation of satellite cells throughout the muscles.
BrdU, IdU, and CldU were purchased from Sigma and used at a dose of 30 mg/kg (subcutaneously). For CldU/IdU double-labeling experiments, two doses of CldU were administered 4 h apart, with the first dose administered 48 h after the muscle injury. Approximately 8 h after the second dose of CldU, IdU was administered also by two sequential injections, the second one 4 h after the first. For in vivo BrdU-labeling experiments, two doses were administered 4 h apart, with the first dose administered 48 h after the muscle injury. For in vitro experiments, thymidine analogs were used at a final concentration of 5 μM in the media. For CldU/IdU experiments, either in vivo or in vitro, similar results were observed when the two labels were reversed.
As previously described [25], muscle was dissected, digested in 0.25 U/ml collagenase type II (Sigma) in HEPES buffered media, and dissociated by trituration into fiber fragments. Fiber-associated cells were liberated either by further digestion in 0.5 U/ml dispase and 80 U/ml collagenase in media and then filtration and subsequent washing of cells in PBS, or by trituration in media through a 20-gauge needle [26]. Both methods gave similar results. Cells obtained by this methods are more than 95% positive for the myogenic cell markers CD34 and M-cadherin and less than 2% positive for the endothelial cell marker PECAM [18,25].
Cells labeled in vivo and prepared as above were plated at a very low density (~10 cells/mm2) onto 4- or 8-well chamber slides coated with ECM gel (Sigma) diluted to 1:100. Satellite cells activated ex vivo in bulk myofiber explant cultures were labeled with a pulse for 8 h on day 2 after explantation, maintained in growth medium for an additional 12 h, and then liberated from the fibers and plated singly, as above, early on day 3. Established myoblast cultures (passage 20–30 after isolation as bulk cultures [25]) were labeled, plated singly, and analyzed identically. Direct microscopic examination revealed that sparsely plated cells adhered within about 1 h and that negligible cell migration occurred during the subsequent period before analysis of cell pairs. After cells were attached, cytochalasin D (2 μM final concentration; Sigma) or nocodazole (1 μM final concentration; Sigma) was added to block cytokinesis. Cells were fixed 2–4 h after plating, immunostained, and scored as a “divided pair” if they were within one cell diameter of each other, and more than 50 cell diameters away from other cells in the 20× field of view. Between 100 and 200 cell pairs were scored per experiment, and the number of replicate experiments is described in the figure legends.
For co-staining of Sca-1, CldU, and IdU for FACS analysis, Sca-1 was detected as above using Alexa 647 as the secondary antibody. The samples were then re-fixed, permeablized with 0.25% Triton-X-100, digested with DNAse1 in F-10 medium [8], and immunostained for CldU and IdU as above, using Alexa 488 or R-phycoerythrin anti-rat secondary antibodies, and PE- or FITC-conjugated mouse anti-BrdU (IdU) clone B44. Isotype-matched antibodies were used as negative controls and for gating. FACS acquisition was performed on a FacsCaliber model (BD Biosciences), and analysis was performed using WinMDI 2.8 software (Joseph Trotter, http://facs.scripps.edu).
|
10.1371/journal.ppat.1003848 | DAMP Molecule S100A9 Acts as a Molecular Pattern to Enhance Inflammation during Influenza A Virus Infection: Role of DDX21-TRIF-TLR4-MyD88 Pathway | Pathogen-associated molecular patterns (PAMPs) trigger host immune response by activating pattern recognition receptors like toll-like receptors (TLRs). However, the mechanism whereby several pathogens, including viruses, activate TLRs via a non-PAMP mechanism is unclear. Endogenous “inflammatory mediators” called damage-associated molecular patterns (DAMPs) have been implicated in regulating immune response and inflammation. However, the role of DAMPs in inflammation/immunity during virus infection has not been studied. We have identified a DAMP molecule, S100A9 (also known as Calgranulin B or MRP-14), as an endogenous non-PAMP activator of TLR signaling during influenza A virus (IAV) infection. S100A9 was released from undamaged IAV-infected cells and extracellular S100A9 acted as a critical host-derived molecular pattern to regulate inflammatory response outcome and disease during infection by exaggerating pro-inflammatory response, cell-death and virus pathogenesis. Genetic studies showed that the DDX21-TRIF signaling pathway is required for S100A9 gene expression/production during infection. Furthermore, the inflammatory activity of extracellular S100A9 was mediated by activation of the TLR4-MyD88 pathway. Our studies have thus, underscored the role of a DAMP molecule (i.e. extracellular S100A9) in regulating virus-associated inflammation and uncovered a previously unknown function of the DDX21-TRIF-S100A9-TLR4-MyD88 signaling network in regulating inflammation during infection.
| The lung disease severity following influenza A virus (IAV) infection is dependent on the extent of inflammation in the respiratory tract. Severe inflammation in the lung manifests in development of pneumonia. Therefore, it is very critical to identify cellular factors and dissect the molecular/cellular mechanism controlling inflammation in the respiratory tract during IAV infection. Knowledge derived from these studies will be instrumental in development of therapeutics to combat the lung disease associated with IAV infection. Towards that end, in the current study we have identified a cellular factor S100A9 which is responsible for enhanced inflammation during IAV infection. In addition, we have characterized a signal transduction pathway involving various cellular receptors and signaling adaptors that are involved in mediating S100A9-dependent inflammatory response. Thus, our studies have illuminated a cellular/molecular mechanism that can be intervened by therapeutics to reduce and control IAV-associated lung inflammatory disease like pneumonia.
| Pathogen-associated molecular patterns (PAMPs) are molecular signatures of pathogens which facilitate induction of the host immune response [1], [2]. PAMPs activate cellular pattern-recognition-receptors (PRRs) such as toll-like receptors (TLRs) to induce immunity [1], [2]. Wide arrays of pathogens activate PRRs in the absence of PRR-specific PAMPs. It is thought that during infection cellular factors can activate PRRs and thus indirectly fulfill the function of PAMPs. The mechanism regulating the activity and function of non-PAMP dependent immune response during virus infection is still an enigma. Damage-associated molecular patterns (DAMPs), which are molecules produced from damaged or dead cells induce an inflammatory response in paracrine fashion via TLR activation [3]. However, whether DAMPs can function as a host-derived molecular pattern during virus infection is not known. In this study, we determined that during influenza A (IAV) virus infection, S100A9 protein (also known as Calgranulin B or MRP-14), which is classified as a DAMP, is released from undamaged infected cells to activate the TLR4/MyD88 pathway for induction of innate and inflammatory responses against IAV. Thus, we have identified extracellular S100A9 as a critical host-derived molecular pattern during IAV infection. This protein has an essential role in enhancing the inflammatory response, which culminates in exacerbated IAV pathogenesis and lung disease.
Influenza A virus (IAV) is a negative-sense, single-stranded RNA virus that causes severe respiratory tract infection [4]–[6]. Infection among high-risk people such as elderly and immuno-compromised individuals manifests in massive airway inflammation, which leads to the development of pneumonia [4]–[6]. Furthermore, there is a constant threat from naturally evolving IAV strains in avian and animal reservoirs that can lead to an epidemic or pandemic. Death of more than 200,000 individuals due to swine IAV (2009 H1N1 IAV) associated infection [7] is an example of the catastrophic nature of IAV infection.
Innate immunity, comprised of antiviral activity (via type-I interferons, IFN-α/β) and a controlled inflammatory response, is critical host defense machinery for virus clearance and the resolution of virus-induced disease [8]–[16]. PRRs recognize PAMPs to induce innate immunity in response to pathogen invasion. During IAV infection, both membrane-bound (e.g., TLRs) and cytosolic (e.g., RIG-like receptors such as RIG-I and Nod-like receptors such as NLRP3 and Nod2) PRRs are required to launch an effective innate response [13], [17]–[31].
Activation of PRRs could serve as a double-edged sword: While operating as host defense factors, activated PRRs can also contribute to the progression of virus-induced disease. For example, although TLR4 is activated during IAV infection, studies with TLR4 KO mice have shown that TLR4 contributes to exacerbated lung disease and mortality in IAV-infected animals [22], [23]. Since pneumonia is an inflammatory disease [6], [32], it is imperative to characterize the molecular mechanisms and cellular factors responsible for uncontrolled inflammation mediated by TLR4 during IAV infection [23]. Although activated TLR4 is a key contributor to exacerbation of disease, the mechanism by which TLR4 is activated in IAV-infected cells is unknown, especially since IAV does not have TLR4-specific PAMP ligand lipopolysaccaride (LPS). Therefore, it is crucial to identify and characterize “non-PAMP” host-derived molecular pattern, which can activate PRRs during virus infection. We expect that this line of investigation will illuminate the role of host-factors in contributing, either positively or negatively, to IAV-associated disease and pathogenesis. These studies will be a stepping stone for development of therapeutics to combat IAV-associated lung disease.
We are interested in understanding the role of secreted soluble factors (e.g. defensins, interferon-alpha induced soluble factor) in viral innate immunity [15], [21], [33], [34]. During our studies to further understand how TLR response modulates expression/production of soluble secreted factors during infection, we found that cells lacking TLR adaptor TRIF failed to release S100A9 following IAV infection. We specifically focused on S100 proteins, since expression of both defensins and S100 proteins are concurrently enhanced during various cellular stimuli [35]–[37] and S100 proteins (S100A9 and S100A8) has implicated in activation of TLR4 pathway during LPS stimulation [38].
In the current study, we have identified extracellular S100A9 protein as a host-derived molecular pattern regulating the pro-inflammatory response, cell death, and pathogenesis during IAV infection. We also show that DDX21/TRIF and TLR4/MyD88 pathways are respectively required for S100A9 gene expression and activity. In addition, we have uncovered DDX21-TRIF-S100A9-TLR4-MyD88 signaling network as a critical regulator of inflammation. This network may also contribute to inflammation and disease during both infection-associated and noninfectious inflammatory diseases and disorders.
Macrophages are essential immune cells that modulate host defense, inflammation, and disease pathogenesis during IAV infection. Macrophages are also the major cytokine- and chemokine-producing cells during IAV infection and thus contribute to lung tissue damage [39]–[42]. To investigate whether IAV infection stimulates S100A9 secretion, we infected macrophages with IAV for 4–16 h. After infection, medium supernatant was collected to assess S100A9 protein levels by ELISA. We found that following IAV infection both human (U937 cells) (Figure 1A) and mouse [J774A.1 macrophage cell-line, primary alveolar macrophages and primary bone marrow-derived macrophages (BMDMs)] macrophages (Figure 1B–D) secreted high levels of S100A9. The physiological significance is evident from the ability of primary macrophages (i.e. alveolar macrophages and BMDMs) (Figure 1C and D) to secrete S100A9 upon IAV infection. Interestingly, S100A9 secretion was detected as early as 4–8 h postinfection. Release of S100A9 is not due to cell cytotoxicity or damage, since an LDH release cytotoxicity assay showed minimal cytotoxicity in macrophages at 8 and 16 h postinfection (Figure S1A). Similarly, no cell death (apoptosis or necrosis) was observed during the 8–16 h postinfection period (not shown). These results demonstrated that following IAV infection, S100A9 is released to the extracellular environment from undamaged macrophages.
There have been no studies to date on the signaling mechanism that regulates gene expression of S100 family of proteins. We examined the signaling mechanism involved in S100A9 expression. We infected BMDMs derived from wild-type (WT), TLR2 knockout (KO), TLR4 KO and TRIF KO mice with IAV. At 24 h postinfection, we evaluated S100A9 levels in the medium. TLR2 and TLR4 were not involved, since comparable S100A9 secretion was noted in WT and TLR KO BMDMs (Figure 2A). A similar result was obtained with TRAM KO and TIRAP KO cells (not shown). In contrast, significant reduction in S100A9 secretion was observed in IAV-infected TRIF KO BMDMs (Figure 2A). RT-PCR analysis showed that loss of S100A9 secretion was caused by the absence of S100A9 mRNA in infected TRIF KO cells (Figure 2B). Apart from TLR4, which uses TRIF for MyD88-independent signaling, TLR3 also recruits TRIF for TLR3-mediated signal transduction. However, TLR3 is not involved in this process, as shown by the fact that S100A9 secretion was not reduced in TLR3 KO BMDMs (Figure 2C). These results demonstrated that S100A9 gene induction occurs via the TLR-independent TRIF-dependent pathway.
Recently, DEAD box proteins (also known as DDX protein) having RNA helicase activity has been implicated in innate immunity [43]. DDX proteins (e.g. DDX21) can function as cytosolic PRR in mouse dendritic cells (mDCs) to induce type-I interferon during infection [43]. Interestingly, DDX signaling was TRIF-dependent and DDX21 interacted with TRIF during signaling [43]. Therefore, we examined whether DDX21 has a role in S100A9 expression during IAV infection of macrophages. Since KO animals lacking DDX proteins do not exist, we used siRNA technology to silence DDX21 expression in macrophages.
Mouse alveolar macrophages (MH-S cell line) were transfected with DDX21-specific siRNA or control scrambled siRNA, after which these cells were infected with IAV. DDX21 expression was monitored by RT-PCR. We observed induction of DDX21 expression following IAV infection (Figure 2D). The silencing efficiency was evident from the loss of DDX21 expression in IAV-infected cells transfected with DDX21-specific siRNA (Figure 2D). We used the silenced cells to deduce the role of DDX21 in S100A9 gene expression following IAV infection. DDX21 is critical for S100A9 gene expression, since drastic loss of S100A9 mRNA was observed in IAV-infected DDX21 silenced cells (Figure S1B). Accordingly, reduced S100A9 mRNA expression in DDX21 silenced cells led to diminished S100A9 secretion following IAV infection of these cells (Figure 2E). The DDX/TRIF dependent S100A9 expression was independent of virus replication, since IAV hemagglutinin (HA) mRNA levels were similar in DDX21 silenced and TRIF KO cells (Figure S2A and S2B). Moreover, S100A9 expression (not shown) and production (Figure S2C and S2D) was not significantly altered in IAV infected MyD88 KO and MAVS KO cells, which implicated MyD88 aααnd MAVS independent expression/production of S100A9 during IAV infection. In addition, we failed to observe significant difference in S100A9 expression/production from IAV infected WT vs. TLR7 KO cells (Figure 2F). It is interesting to note that DDX21 expression was undetected at 48 h postinfection (Figure 2D), which co-relates with loss of S100A9 production during that time frame (not shown). This suggests that to maintain homeostasis and to avoid hyper-inflammation cells may negatively regulate DDX21 expression to reduce S100A9 production. These results demonstrated that the DDX21/TRIF pathway is required for S100A9 gene induction and the resulting S100A9 secretion following IAV infection.
In the preceding studies, the high levels of S100A9 secretion during infection suggested that secreted extracellular S100A9 may have some role during IAV infection. Therefore, we focused on the role and function of extracellular S100A9 during IAV infection. Earlier studies have shown that the S100A9/S100A8 complex is required for optimal LPS-induced TLR4-dependent TNF-α (TNF) production in bone marrow cells (comprised of undifferentiated monocytes and DCs) [38]. However, few studies have shown the pro-inflammatory activity of S100A9 in the absence of S100A8 and LPS. Moreover, it is not known whether S100A9 can launch a pro-inflammatory response in macrophages. Since our studies are focused on the innate responses of IAV-infected macrophages, we investigated whether extracellular addition of purified S100A9 protein (to mimic secreted S100A9) promotes the release of pro-inflammatory cytokines IL-6 and TNF-α (TNF). These pro-inflammatory cytokines are produced early during IAV infection, a period that corresponds with S100A9 secretion kinetics.
Mouse (J774A.1) and human (U937 cells) macrophages were incubated with purified mouse or human S100A9 proteins, respectively for 6–12 h (Figure 3). After treatment, medium supernatant was collected to analyze TNF and IL-6 levels by ELISA. S100A9 alone stimulates a pro-inflammatory response in macrophages, as is evident from high levels of TNF (Figure 3A and C) and IL-6 (Figure 3B and D) production by macrophages treated with purified S100A9 protein. Both human (Figure 3A and B) and mouse (Figure 3C and D) macrophages produced pro-inflammatory cytokines upon incubation with human and mouse S100A9 protein. Interestingly, the response was rapid, since substantial TNF and IL-6 production occurred within 6 h of treatment with S100A9 protein. RT-PCR analysis showed that production of TNF and IL-6 by S100A9 was due to activation of their corresponding genes (Figure S3). Since the pro-inflammatory activity of purified S100A9 protein could be inhibited by anti-S100A9 blocking (neutralizing) antibody (not shown), the observed response was due to S100A9 protein. Moreover, the effect observed with purified S100A9 protein was not due to LPS contamination (Figure S4A and S4B). These studies demonstrated that S100A9 functions as an extracellular host factor to launch a pro-inflammatory response in macrophages.
We next examined the role of secreted S100A9 in eliciting a pro-inflammatory response during IAV infection. We used blocking antibody against S100A9 to neutralize the activity of extracellular (secreted) S100A9. Previously, it was shown that this blocking antibody specifically inhibited the activity of the secreted extracellular form of S100A9 both in vitro and in vivo [44]–[50]. J774A.1 cells were infected with IAV in the presence of either control antibody (control IgG) or S100A9-specific blocking antibody. At various postinfection time points, IL-6 and TNF levels were examined by ELISA. Extracellular S100A9 plays a key role in inducing the pro-inflammatory response during IAV infection, since significant reduction in IL-6 (Figure 4A) and TNF (Figure S4C) levels were observed in infected cells treated with S100A9 blocking antibody. RT-PCR showed that loss of IL-6 and TNF production was due to diminished gene expression (not shown). Similar results were obtained following treatment of IAV-infected primary macrophages (BMDM) with S100A9 blocking antibody (Figure S4D and S4E). Diminished IL-6 (Figure S4D and S4E) and TNF (not shown) production (Figure S4D) and expression (Figure S4E) was observed in infected BMDM treated with S100A9 blocking antibody. The loss of pro-inflammatory response was not due to reduced IAV replication, since IAV HA expression was similar in control antibody and S100A9 blocking antibody treated J774A.1 cells (Figure S5A) and BMDMs (Figure S5B). Thus, extracellular S100A9 modulates pro-inflammatory response independent of IAV replication.
Our finding that S100A9 contributes to the pro-inflammatory response during IAV infection was validated by using BMDMs derived from S100A9 KO mice. WT and S100A9 KO BMDMs were infected with IAV, after which TNF and IL-6 levels in the medium supernatant were measured by ELISA. As compared to WT cells, there were significant reductions in IL-6 (Figure 4B) and TNF (Figure 4C) production from infected S100A9 KO cells. Once again, this was a consequence of the loss of pro-inflammatory gene expression in IAV-infected S100A9 KO BMDMs (Figure S5C). The critical function of secreted (extracellular-form) S100A9 during this response was apparent from the observation that addition of purified mouse S100A9 protein to S100A9 KO BMDMs restored the pro-inflammatory response in IAV-infected S100A9 KO BMDMs (Figure 4D). This result also suggested that intracellular S100A9 does not play a role in inducing a pro-inflammatory response. Treatment of WT or S100A9 BMDMs with S100A9 protein did not alter IAV replication status in the corresponding cells (not shown). We also observed production of S100A9 following treatment of BMDMs with synthetic dsRNA (poly-IC) (Figure S6A). The pro-inflammatory activity of S100A9 was specific for IAV and dsRNA (which serves as a replicative intermediate during IAV infection and induces DDX21/TRIF pathway) since dsRNA (poly-IC) dependent TNF and IL-6 release was significantly diminished in S100A9 KO cells (Figure S6B and S6C), and treatment of KO cells with purified S100A9 protein restored the pro-inflammatory response in poly-IC treated S100A9 KO cells (Figure S6D and S6E). In contrast, TNF and IL-6 release from S100A9 KO BMDMs was not affected following imiquimod (which activates TLR7 dependent pro-inflammatory response) treatment (Figure S6F and S6G). In addition, treatment of WT and S100A9 KO BMDMs with TNF (to induce NF-κB dependent inflammatory response via TNF receptor) revealed similar levels of IL-6 production from both WT and KO cells (Fig. S6H).
During these studies we observed that IAV replication (as deduced from IAV HA mRNA expression) was significantly reduced in S100A9 KO BMDMs compared to WT cells (Figure S7A). This result suggested that although extracellular S100A9 plays a critical role in modulating pro-inflammatory response (Figure 4D), intracellular S100A9 may be involved in negatively regulating antiviral factor expression/production or it is required for efficient virus infection/replication. This is not surprising in light of previous reports illustrating differential function of extracellular vs. intracellular S100 proteins.
It is important to mention that we observed S100A9 production from IAV-infected BMDMs at 4 h postinfection (Figure 1C) and that TNF and IL-6 are produced from IAV-infected BMDMs at 8–12 h postinfection (not shown); these cytokines are undetectable at 4 h postinfection (not shown). Thus, S100A9 secretion and production of early pro-inflammatory mediators (e.g. TNF, IL-6) are temporally regulated during IAV-infection. Therefore, extracellular S100A9 is a key regulator of the pro-inflammatory response during IAV infection.
Macrophages undergo apoptosis during IAV infection [41], [42]. Several studies have demonstrated that S100A9 has a pro-apoptotic function in epithelial cells, muscle cells, and neutrophils [51]–[55], but no apoptosis-inducing activity of S100A9 (or any other S100 proteins) in macrophages has been reported. Since IAV infection resulted in high levels of S100A9 secretion, we examined the ability of extracellular S100A9 to induce apoptosis in macrophages and the role of secreted S100A9 in apoptotic induction during IAV infection.
We treated J774A.1 and MH-S macrophages with purified S100A9 protein for 48 and 72 h, then examined the apoptotic status of cells by monitoring annexin V and PI staining. The apoptosis rate was calculated based on the number of annexin V positive/PI negative cells (denoting early apoptosis)+number of annexin V positive/PI-positive cells (denoting late apoptosis) per total number of cells. We noted apoptosis in S100A9 protein-treated mouse macrophages (Figure 5A and B). The result obtained with annexin V and PI staining was confirmed by performing TUNEL analysis (Figure S7B and S7C). Similar results were obtained following treatment of human U937 macrophages (not shown).
Since IAV infection triggers S100A9 secretion, we next examined whether extracellular S100A9 has a role in apoptosis of IAV-infected macrophages. J774A.1 macrophages were infected with IAV for 48 h in the presence of either control antibody (control IgG) or S100A9 blocking antibody. Significantly diminished apoptosis (reduction of 27%) occurred in macrophages treated with S100A9 antibody (Figure 5C). These results were further confirmed by TUNEL analysis (Figure S7D). Thus, extracellular S100A9 has a critical function in regulating apoptosis of IAV-infected macrophages.
Previous studies have found that optimal TLR4 activation by LPS in bone-marrow cells required the activity of extracellular S100A9/S100A8 complex [38]. However, it is not known whether S100A9 alone activates TLR4, especially in macrophages. In addition, there has been no report of DAMP proteins like S100A9 activating PRR signaling during virus infection. Therefore, we investigated the role of the TLR4/MyD88 pathway in the macrophage pro-inflammatory response by S100A9 alone (in the absence of S100A8), and the function of the S100A9/TLR4/MyD88 pathway in regulating the pro-inflammatory response in IAV-infected macrophages. We incubated WT and TLR4 KO BMDMs with purified S100A9 protein, and then measured IL-6 (Figure 6A) and TNF (Figure 6B) levels by ELISA. Drastic loss of IL-6 and TNF production was detected in S100A9 protein-treated TLR4 KO BMDMs (Figure 6A and B), indicating that TLR4 is absolutely required for the S100A9-mediated response. Drastic reductions in IL-6 (not shown) and TNF (Figure S8A) transcripts occurred in S100A9 protein treated TLR4 KO cells.
Since MyD88 is one of the critical adaptors for activated TLR4, we next investigated the role of MyD88 by using MyD88 KO BMDMs. Incubation of WT and MyD88 KO BMDMs with purified S100A9 protein significantly reduced production of IL-6 (Figure 6C) and TNF (not shown) from MyD88 KO cells. The loss of cytokine protein production was due to reduced TNF (Figure S8B) and IL-6 (not shown) gene expression in MyD88 KO BMDMs, thus, demonstrating that the TLR4/MyD88 pathway is required for the S100A9-mediated pro-inflammatory response.
We next studied the role of TLR4/MyD88 in stimulating the pro-inflammatory response following IAV infection. After WT, MyD88 KO, and TLR4 KO BMDMs were infected with IAV, IL-6 levels were assessed by ELISA. Our study revealed that TLR4/MyD88 is an essential regulator of pro-inflammatory response during IAV infection, since significant reduction in IL-6 (Figure 6D) and TNF (Figure 6E) production was noted in IAV infected TLR4 KO (Figure 6D and E) and MyD88 KO (Figure 6D) BMDMs. RT-PCR analysis demonstrated diminished expression of IL-6 mRNA in TLR4 KO (Figure S8C) and MyD88 KO (not shown) BMDMs. Similarly, we noted significant reduction in TNF production from IAV-infected TLR4 KO (Figure 6E) and MyD88 KO (not shown) BMDMs. The observed effect was independent of virus replication since compared to WT cells, no change in HA mRNA expression was noted in TLR4 KO (Figure S8D) and MyD88 KO (not shown) cells. Thus, TLR4/MyD88 activation is a key step for inducing the S100A9-mediated pro-inflammatory response. Also, the S100A9/TLR4/MyD88 pathway is a crucial regulator of the pro-inflammatory response during IAV infection.
Our study showed that extracellular S100A9 uses TLR4/MyD88 signaling for the pro-inflammatory response during IAV infection. TLR4 activation has been associated with apoptosis induction via various mechanisms, including activation of the pro-apoptotic function of NF-κB, modulation of tumor-suppresser expression or function etc [56]–[62]. To assess the role of TLR4 in S100A9-mediated apoptosis, we treated WT and TLR4 KO BMDMs with purified S100A9 protein for 72 h. Treatment of WT BMDMs with S100A9 protein induced apoptosis (Figure 7A), which was consistent with our previous findings. However, significant loss of apoptosis was observed in S100A9 protein-treated TLR4 KO BMDMs (Figure 7A).
We next examined the role of TLR4 and MyD88 in apoptosis induction during IAV infection. We infected WT and TLR4 KO BMDMs with IAV and evaluated apoptosis 48 h later. Apoptosis analysis by annexin V staining (Figure 7B) and TUNEL (Figure 7C) revealed that while IAV infection resulted in apoptosis of WT macrophages, a significant reduction in apoptosis was detected in TLR4 KO cells. Diminished apoptosis was also observed in infected MyD88 KO BMDMs (Figure 7D), indicating that MyD88 is also required during this event. Thus, the S100A9/TLR4/MyD88 pathway constitutes one of the mechanisms that modulate apoptosis of IAV-infected cells.
To establish the in vivo role of S100A9 in regulating innate response during IAV infection of the airway, we next evaluated S100A9 expression and its secretion in the IAV-infected mouse respiratory tract. Mice were intratracheally inoculated with IAV and, at 1–6 days postinfection, lungs were harvested. S100A9 mRNA expression in the lungs were analyzed by RT-PCR. S100A9 transcripts were observed in infected lungs but not in lungs from uninfected animals (Figure 8A), indicating that IAV infection led to robust induction of S100A9 gene expression. We also detected high levels of S100A9 protein in the lungs of IAV-infected mice (Figure 8B). Immunohistochemical analysis of lung sections confirmed the presence of S100A9 protein in IAV-infected animals (Figure 8C), while S100A9 was nearly undetectable in mock-infected lungs. Analysis of bronchoalveolar lavage fluid (BALF) by ELISA confirmed the presence of S100A9 protein in the airway of IAV-infected animals (Figure 8D). Thus, IAV infection of the respiratory tract results in induction of S100A9 gene expression and secretion of S100A9 protein in the airway.
Macrophages play a vital role in the innate response to IAV infection by producing pro-inflammatory mediators that determine the inflammation status in the lung [39]–[42]. Moreover, debris from dead cells, originating from apoptosis of immune cells, contributes to airway inflammation [40]–[42], [63]–[68]. Since extracellular S100A9 acted as a positive regulator of pro-inflammatory response and induced apoptosis, we hypothesized that extracellular S100A9 exacerbates IAV-associated lung disease. To test this, we used anti-S100A9 blocking antibody, which neutralizes extracellular (secreted) S100A9 protein.
We used anti-S100A9 antibody instead of doing our in vivo studies with S100A9 KO mice because S100 proteins have both intracellular functions (such as cytoskeletal rearrangement, cell metabolism, intracellular calcium response etc) and extracellular functions [69], [70]. Since we have elucidated a role of extracellular (secreted form) S100A9, results from KO mice might not distinguish whether the observed effects are due to activity of extracellular S100A9 or intracellular S100A9 function. Most importantly our studies demonstrated that intracellular S100A9 could be involved in negatively regulating antiviral response or it is required for IAV infection/replication, since reduced virus replication was noted in S100A9 KO BMDM compared to WT cells (Figure S7A). In that scenario, S100A9 KO mice may not serve as an appropriate model to study IAV-induced pro-inflammatory (and apoptotic) response in vivo, since viral burden in the lung is directly proportional to the degree of pro-inflammatory (and apoptotic) response (i.e. if there is less viral burden then concomitantly reduced pro-inflammatory response and apoptosis will occur). However, neutralizing the activity of extracellular S100A9 with S100A9 blocking antibody did not alter IAV replication in vitro (Figure S5A and S5B) and in vivo (please see below). We therefore used anti-S100A9 blocking antibody to specifically inhibit the activity of extracellular S100A9 in mice. We have previously shown that anti-S100A9 blocking antibody has extracellular S100A9 blocking activity [44]–[50]. Specifically, intraperitoneal (i.p) injection of S100A9 blocking antibody inhibited the activity of mouse S100A9 during S. pneumoniae infection [49]. Thus, this antibody [44], [45], [48]–[50] is useful to assess the functional role of extracellular (secreted form) S100A9. Furthermore, similar levels of S100A9 protein were detected in the BALF of control IgG-treated and S100A9-antibody treated mice (Figure S9A). Thus, i.p.-injected anti-S100A9 antibody did not significantly affect S100A9 protein production in the airway-lumen following IAV infection. As in previous reports, we detected anti-S100A9 antibody (administered i.p.) in lung homogenate (Figure S9B). Thus, anti-S100A9 antibody could effectively block lung-localized S100A9 during IAV infection [49], [71], [72]. The clinical significance of utilizing neutralizing antibody is obvious from possible passive immunization with S100A9 antibody as a new therapeutic strategy to control lung inflammation and associated lung disease during IAV infection.
Initially, we investigated the role of secreted S100A9 in regulating IAV susceptibility. For these studies, mice were i.p. injected with either control IgG or anti-S100A9 blocking antibody. One day later, mice were infected with IAV via intra-tracheal (I.T) inoculation. Survival of IAV-infected mice was monitored until 8 days postinfection. Blocking S100A9 activity significantly reduced the mortality of IAV-infected mice (Figure 9A), demonstrating that extracellular S100A9 is a key regulator of IAV susceptibility. Extracellular S100A9 also contributes to morbidity since mice treated with S100A9 blocking antibody exhibited reduced weight loss upon IAV infection (Figure S9C). Thus, extracellular S100A9 contributes to both IAV-induced mortality and morbidity.
In addition, inflammation was reduced following inhibition of extracellular S100A9 activity (Figure 9B and S10A). These results demonstrated that extracellular S100A9 contributes to the severity of IAV-associated lung inflammation and serves as a critical host factor for heightened IAV susceptibility and IAV-induced death. The clinical significance of our result is borne out by the possibility of passive immunization with anti-S100A9 antibody to reduce the severity of respiratory disease associated with IAV infection.
We have identified extracellular (secreted) S100A9 as a critical regulator of the pro-inflammatory response following IAV infection of macrophages. To examine the physiological role of secreted S100A9 in lung inflammation, we tested whether intratracheal (I.T.) administration of purified S100A9 protein would trigger a pro-inflammatory response in the lungs. Indeed, this led to production of TNF (Figure 9C) and IL-6 (Figure S10B) in the respiratory tract due to S100A9-mediated upregulation of TNF and IL-6 gene expression in the lung (not shown). The ability of S100A9 protein to trigger pro-inflammatory mediators in the lung is further reflected by observing airway inflammation in S100A9 protein administered (via I.T) mice (Figure S10C).
Based on this observation, we next examined the role of extracellular S100A9 in airway pro-inflammatory response following IAV infection. Mice were given i.p. injections of control IgG antibody or anti-S100A9 blocking antibody. At 1 d post-antibody treatment, mice were infected with IAV via I.T route. Levels of IL-6 and TNF in the lung were measured by ELISA. Extracellular S100A9 contributes to production of pro-inflammatory mediators during infection as evident from reduced TNF (Figure 9D) and IL-6 (Figure S11A) levels in the lung of S100A9 antibody treated mice. Reduced pro-inflammatory cytokine production was caused by loss of TNF (Figure S11B) and IL-6 (data not shown) mRNAs in the lungs of IAV-infected mice treated with S100A9 blocking antibody. Diminished pro-inflammatory response is not due to reduced IAV infection, since both control antibody and S100A9 antibody treated mice exhibited similar IAV infection status (i.e. viral burden) (Figure S12A). Interestingly, S100A9 antibody could also be utilized as therapeutics to control IAV-associated disease, since administration of S100A9 blocking antibody after IAV infection significantly reduced pro-inflammatory response and lung inflammation (Figure S12B and S12C).
In order to provide evidence for direct neutralization of S100A9 activity in the airway following i.p. administration of S100A9 antibody, we administered S100A9 antibody (via i.p.) to mice and after one day (to exactly mimic IAV infection studies) mice were inoculated with S100A9 protein via I.T route. Significant inhibition in pro-inflammatory activity was noted in the presence of S100A9 antibody (Figure S13A), which shows that i.p. administered blocking antibody can neutralize S100A9 protein in the airway.
The role of extracellular S100A9 was further validated by conducting ex vivo experiment with BALF-associated cells derived from IAV infected mice administered (via i.p) with either control antibody or S100A9 blocking antibody. Significant reduction in IL-6 and TNF production from BALF cells was observed in S100A9 blocking antibody treated mice (Figure 9E). This result once again validates blocking of S100A9 activity in the alveolar space localized (i.e. present in the BALF) cells. These studies illustrate the importance of secreted S100A9 in regulating pro-inflammatory cytokine gene expression and production during IAV infection of the airway.
Our studies with macrophages have illuminated a vital role of extracellular S100A9 in inducing apoptosis of IAV-infected cells. We have extended those observations in mice to establish the in vivo physiological relevance of extracellular S100A9 as a regulator of apoptosis. Further, it is known that apoptosis significantly contributes to IAV infection severity and associated lung disease [40], [63]–[68]. Therefore, reduced apoptosis in IAV-infected S100A9-blocked mice may contribute to reduced susceptibility and diminished airway disease (as shown in Figure 9A and 9B and S9C). To examine this possibility, mice treated with control IgG and S100A9 blocking antibody were inoculated with IAV via the I.T route. On the third day post-infection, we performed in situ TUNEL assay with lung sections to determine the apoptotic status of the IAV-infected respiratory tract. We found significantly less apoptosis in the lungs of mice given S100A9 blocking antibody than in the lungs of control mice (Figure 10A and B). These results demonstrated that secreted S100A9 is a pivotal regulator of lung apoptosis following IAV infection.
The role of DAMPs as a host-derived molecular pattern during virus infection is not known. In the current study we have demonstrated that extracellular S100A9 protein functions as a host-derived molecular pattern during infection. Although S100A9 is classified as a DAMP, using clinically important influenza A virus (IAV) infection model, we show release of S100A9 from “undamaged” cells during IAV infection; which triggered PRR (i.e., TLR) signaling. Surprisingly, we observed that extracellular (secreted) S100A9 regulates two key mechanisms that contribute to inflammation during IAV infection. These are pro-inflammatory cytokine production during early infection and induction of apoptosis. We also found that S100A9-mediated activation of the TLR4/MyD88 pathway resulted in increased inflammation, which culminated in exacerbated IAV pathogenesis. Thus, our study shows a role of “non-PAMP” PRR-activating DAMPs in modulating immunity and inflammation during virus infection. We also have identified DDX21-TRIF-S100A9-TLR4-MyD88 as a novel signaling “network” that regulates inflammation. It is possible that a similar pathway is used to promote disease during infection with other viruses including highly pathogenic RNA viruses like SARS, Ebola virus, Marburg virus.
Based on our results, we propose a model (Figure 10C) whereby the S100A9 gene is activated by the DDX21-TRIF pathway and the resulting S100A9 protein is secreted during IAV infection. Extracellular S100A9 activates the TLR4/MyD88 pathway via an autocrine or paracrine mechanism. As a consequence, S100A9/TLR4 activity exacerbates lung disease by promoting a pro-inflammatory response and inducing cell-death. Our studies are the first to highlight the role of S100A9 protein and DDX21-TRIF-S100A9-TLR4-MyD88 signaling network in modulating inflammation during virus infection. The clinical significance of our study is borne out by detection of S100A9 protein in mucosal secretions from IAV-infected individuals [73]. The clinical significance of utilizing neutralizing antibody is obvious from possible passive immunization with S100A9 antibody (as shown in Figure 9A and B and S9C) as a new therapeutic strategy to control lung inflammation and associated lung disease during IAV infection.
Mortality among IAV-infected individuals is associated with pneumonia, a disease characterized by massive lung inflammation leading to tissue damage and endothelial barrier disruption, resulting in fluid leakage in the airway and the development of edema. Highly pathogenic IAV strains have greater propensity to launch a hyper-inflammatory response in the respiratory tract upon infection, culminating in the development of pneumonia. Among the cellular factors that regulate IAV-induced lung disease, TLR4 is a major contributor to susceptibility and exacerbated pathophysiology associated with IAV infection [22], [23]. TLR4 is activated during IAV infection and reduced mortality and diminished lung disease (and inflammation) was observed in IAV infected TLR4 KO mice [22], [23]. The mechanism of TLR4 activation by IAV is unknown, especially since IAV doses not posses TLR4 ligand LPS. In that regard, our current study has elucidated a role of extracellular S100A9 in activation of TLR4/MyD88 pathway during IAV infection.
Previous studies reported that S100A9-S100A8 complex optimizes LPS-mediated TLR4 activation [38]. Bone marrow cells, including undifferentiated monocytes and DCs, and mice were treated/infected with LPS and LPS- expressing bacteria (E. coli 018:K1) to demonstrate augmentation of LPS activity by extracellular S100A9-S100A8 complex [38]. However, we show for the first time that S100A9 alone can directly activate TLR4 (in the absence of LPS) and contribute to the regulation of inflammation during infection with IAV, a non-LPS-expressing pathogen. Thus, extracellular S100A9 can directly modulate immune response via TLR4 activation. We also demonstrated that S100A9 alone (independent of S100A8) is a critical regulator of pro-inflammatory response in vitro and in vivo. Although few studies have been done on the mechanism of S100 gene induction during biological responses, we have shown that the S100A9 gene is induced by the DDX21-TRIF pathway. These studies also demonstrated the critical function of the DDX family PRRs in regulating expression of a host factor (S100A9) that is not a cytokine (i.e. IFNs).
We have determined that the DDX21-TRIF pathway is required for S100A9 gene expression. A recent study has shown that the DDX1-DDX21-DHX36/TRIF pathway triggers a type-I interferon response in myeloid dendritic cells (mDCs) during virus infection and treatment of cells with dsRNA (poly-IC) [43]. In these studies, direct interaction of viral dsRNA with DDX proteins was not shown. Thus, during virus infection viral dsRNA can directly or indirectly activate DDX proteins for signaling. In that context, S100A9 related S100A8 gene expression was induced by dsRNA via MAPK pathway [74].
In accord with the previous studies [43], [74], we noted that IAV replication, which will generate viral dsRNA, is required for S100A9 production, since UV inactivated IAV failed to secrete S100A9 from macrophages (Figure S13B). Surprisingly, our study demonstrated that apart from the reported DDX/TRIF-dependent IFN production in mDCs [43], the DDX/TRIF pathway is also important in inducing a pro-inflammatory response during IAV infection of macrophages, which is mediated by DDX21-TRIF dependent activation of S100A9 gene expression and resulting autocrine/paracrine action (via TLR4/MyD88 pathway) of secreted S100A9 protein. Thus, our studies have illustrated a role of two PRRs in modulating inflammation during IAV infection – a cytosolic PRR (i.e. DDX21) regulating S100A9 gene expression and membrane-localized PRR (i.e. TLR4) transducing the biological activity (i.e. pro-inflammatory response and cell death) of S100A9. This shows how concerted activity of two PRRs is used to control inflammation, since the inflammatory response has to be “regulated” at several levels due to the detrimental effect of uncontrolled inflammation on promoting cell and tissue damage.
We have also identified extracellular S100A9 as one of the host factors that regulate apoptosis during IAV infection. Apoptosis is a key contributor to pathogenesis and the pathology associated with IAV infection [40], [63]–[68], [75]. Cell death intensifies inflammation in the respiratory tract, culminating in exacerbated lung disease. Previous studies have shown the ability of S100 proteins to induce cell death by various mechanisms [51]–[55]. Similar mechanisms may contribute to S100A9-mediated cell-death following IAV infection.
Two arms of viral innate immunity consist of antiviral and inflammatory responses. Our studies have indicated that extracellular and intracellular S100A9 may function differently in terms of innate immune response during IAV infection, whereby extracellular S100A9 modulates pro-inflammatory response (independent of viral replication) and intracellular S100A9 is involved in orchestrating antiviral response to reduce viral burden. The differential activity of extracellular vs. intracellular S100A9 protein has been noted previously [69], [70]. In that regard, two different pools (i.e. extracellular and intracellular) of S100A9 exist in IAV infected macrophages (Table S1). During infection, while 10%–25% of S100A9 protein is released, the rest is localized inside the cell (Table S1). In the current study we demonstrated that extracellular S100A9 (secreted form) triggers pro-inflammatory response and apoptosis. In contrast, we speculate that intracellular S100A9 may be involved in negatively regulating antiviral response or it is required for efficient IAV infection/replication. This conclusion was based on the observation that - a) treatment of macrophages (and mice) with S100A9 blocking antibody diminished inflammatory response, while virus replication/infection was unchanged (Figure S5A, S5B and S12A), b) virus replication in S100A9 KO macrophages is reduced compared to WT cells (Figure S7A), and c) addition of S100A9 protein (to mimic extracellular S100A9 protein) to IAV infected S100A9 deficient (KO) cells led to a pro-inflammatory response even in the absence of intracellular S100A9 (i.e. in S100A9 KO BMDMs) (Fig. 4D). Thus, extracellular S100A9 regulates inflammatory response independent of virus replication, while intracellular S100A9 may negatively regulate expression/production of antiviral factor(s) or it functions as a host factor required for efficient IAV infection/replication. In the future we will further dissect the exact mechanism(s) by which intracellular S100A9 modulates IAV infection/replication.
S100A9 confers protective immunity during Klebsiella pneumoniae infection, since enhanced bacterial dissemination, lung damage, and susceptibility was observed in mice deficient in S100A9 expression [76]. In these studies, no distinction was made in terms of unique and differential function of extracellular vs. intracellular S100A9. However, our studies have shown that extracellular S100A9 is one of the factors that dictate detrimental host (inflammatory and apoptotic response) response during IAV infection, and this response is independent t of virus replication. In contrast, we show that intracellular S100A9 (which constitutes majority of S100A9 protein in IAV infected cells) is required for efficient IAV infection/replication. Therefore, our studies have illustrated a distinct role of extracellular vs. intracellular S100A9 during IAV infection. Thus, in accord with the previous study with Klebsiella we speculate that IAV infection of S100A9 KO mice will result in diminished IAV replication and as a consequence, these mice will exhibit reduced inflammation, susceptibility and pathogenesis. We will conduct these studies in the future to elucidate the exact role (and underlying mechanism) of intracellular S100A9 in controlling IAV infection/replication.
Apart from macrophages, epithelial cells (primary mouse lung epithelial cells) also produced S100A9 upon IAV infection (Figure S14). Interestingly, lower levels of S100A9 was released from primary lung epithelial cells compared to primary alveolar macrophages [please compare Figure S14 (lung epithelial cells) vs. Figure 1D (alveolar macrophages)]. In the future we plan to perform in-depth study to investigate the role of macrophages and lung epithelial cells (and their cross-talk) during S100A9 mediated inflammatory response following IAV infection.
In summary, we have identified extracellular S100A9 as a host-derived molecular pattern that regulates inflammation during virus infection. In addition, we have uncovered DDX21-TRIF-S100A9-TLR4-MyD88 as a novel signaling “network” that regulates inflammation. Future studies dealing with identification and characterization of host-derived molecular patterns (e.g. DAMPs) during virus invasion may lead to the development of measures to combat infection-associated inflammatory diseases.
Animal studies were performed according to housing and care of laboratory animals guidelines established by National Institutes for Health. All animal experiments were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of University of Texas Health Science Center at San Antonio. The Animal Welfare Assurance # is A3345-01.
Influenza A [A/PR/8/34 (H1N1)] virus was grown in the allantoic cavities of 10-day-old embryonated eggs [21], [75]. Virus was purified by centrifuging two times on discontinuous sucrose gradients [21], [75], [77]. J774A.1 cells were maintained in DMEM supplemented with 10% fetal bovine serum (FBS), penicillin, streptomycin, and glutamine. U937 cells were maintained in RPMI 1640 medium supplemented with 10% FBS, 100 IU/mL penicillin, 100 µg/mL streptomycin, 1 mM sodium pyruvate, and 100 nM HEPES. MH-S cells were maintained in RPMI 1640 medium supplemented with 10% FBS, 100 IU/mL penicillin, and 100 µg/mL streptomycin. Bone-marrow-derived macrophages (BMDMs) were obtained from femurs and tibias of wild-type (WT) and knock-out mice and were cultured for 6–8 days as described earlier [21], [75]. Cells were plated on 12-well plates containing RPMI, 10% FBS, 100 IU/mL penicillin, 100 µg/mL streptomycin, and 20 ng/ml GM-CSF. Alveolar macrophages were obtained from the broncho-alveolar lavage fluid (BALF) of wild-type C57BL/6 mice. The IAV titer was monitored by plaque assay analysis with MDCK cells.
S100A9 KO mice were generated at University of Laval, Quebec, Canada. Other KO mice (TLR4, TLR2, TRAM, TRIF, TIRAP) were originally provided by Dr. Doug Golenbock (University of Massachusetts Medical School, Worcester, MA) under a Materials Transfer Agreement with Dr. Shizuo Akira (Osaka University, Osaka, Japan). TLR3 KO, TLR7 KO and MyD88 KO mice were obtained from Jackson Laboratory, Bar Harbor, ME.
Murine S100A9 neutralizing antibody purified IgG from the serum of S100A9 immunized rabbits was generated as described previously [44]–[45]. This antibody has been successfully used to block the activity of extracellular mouse S100A9 [44]–[50]. Human S100A9 antibody was acquired from AbCam, Cambridge, MA (goat anti-human S100A9 antibody) and R&D Systems (mouse anti-human antibody). Recombinant human and mouse S100A9 proteins were generated as previously described [44]–[50]. Briefly, full length human S100A9 cDNA was cloned into pET28 expression vector (Novagen, Madison, WI). S100A9 protein expression was induced with 1 mM isopropyl β-D-thiogalactoside (IPTG) in E. coli HMS174 (Boehringer Mannheim, Mannheim, Germany) for 16 h at 16°C. After IPTG treatment, the bacteria were centrifuged at 5000×g for 10 min and the pellet was re-suspended in PBS [(containing NaCl (0.5 M) and imidazole (1 mM)] and lysed by sonication. Upon centrifuging the lysate at 55,000×g for 30 min at 4°C, the supernatant was collected. Recombinant His-Tag S100A9 was purified by using a nickel column. S100A9 bound to the column was incubated with 10 U of biotinylated thrombin (Novagen) (for 20 h at room temperature) to free S100A9 from its His-Tag. Recombinant S100A9 was then eluted with PBS. The digestion and elution processes were repeated one more time to cleave the remaining undigested recombinant proteins, and streptavidin-agarose (Novagen) was added to remove contaminating thrombin. Finally, the protein preparation was passed through a polymyxin B-agarose column (Pierce, Rockford, IL) to remove endotoxins.
Recombinant proteins were prepared in Hank's buffered salt solution (HBSS) buffer. The absence of endotoxin contamination in antibody and protein preparations was confirmed using the limulus amebocyte assay (Cambrex).
Total RNA was extracted using Tri Reagent (Invitrogen). cDNA was synthesized using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). PCR was done using 0.25 units of Taq polymerase, 10 pmol of each oligonucleotide primer, 1 mM MgCl2, and 100 µM deoxynucleotide triphosphates in a final reaction volume of 25 µl. Following amplification, the PCR products were analyzed on 1.5% agarose gel. Equal loading in each well was confirmed by analyzing expression of the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH). The primers used to detect the indicated genes by RT-PCR were:
GAPDH forward, 5′-GTCAGTGGTGGACCTGACCT, GAPDH reverse, 5′-AGGGGTCTACATGGCAACTG,
Mouse GAPDH forward, 5′-GCCAAGGTCATCCATGACAACTTTGG, Mouse GAPDH reverse, 5′-GCCTGCTTCACCACCTTCTTGATGTC
Mouse S100A9 forward, 5′-GTCCTGGTTTGTGTCCAGGT, Mouse S100A9 reverse, 5′-TCATCGACACCTTCCATCAA
Mouse DDX21 forward, 5′-GATCCCCCTAAATCCAGGAA, Mouse DDX21reverse, 5′-TTCGGAAGGCTCCTCTGTTA
Mouse TNF-α forward, 5′-CCTGTAGCCCACGTCGTAGC, Mouse TNF-α reverse, 5′-TTGACCTCAGCGCTGAGTTG
Mouse IL-6 forward, 5′-TTGCCTTCTTGGGACTGATGCT, Mouse IL-6 reverse, 5′-GTATCTCTCTGAAGGACTCTGG
IAV HA forward, 5′- CCCAAGGAAAGTTCATGG, IAV HA reverse, 5′-GAACACCCCATAGTACAAGG
U937 cells, alvelolar macrophages, BMDM, MH-S, and J774A.1 were infected with purified IAV [1 multiplicity of infection (MOI)−2 MOI as indicated] in serum-free, antibiotic-free OPTI-MEM medium (Gibco). Virus adsorption was done for 1.5 h at 37°C, after which cells were washed twice with PBS. Infection was continued in the presence of serum containing DMEM or RPMI medium for the specified time points.
In some experiments, cells were infected in the presence of 2 ng–10 ng/ml control IgG (purified rabbit IgG, Innovative Research, Novi, MI) or 2 ng–10 ng/ml anti-S100A9 blocking antibody. Following virus adsorption, antibodies were added to the cells and the infection was carried out in the presence of the antibodies. In addition, in some experiments infection was done in the presence of purified S100A9 protein or HBSS buffer (vehicle control). Purified S100A9 protein (5 µg/ml) was added to S100A9 KO BMDMs following virus adsorption. Purified protein was present during infection.
Control siRNA and mouse DDX21 siRNA were purchased from Santa Cruz Biotechnology. MH-S cells were transfected with 40 pmol of siRNAs using Lipofectamine 2000 (Invitrogen). At 48 h posttransfection, the cells were infected with IAV.
Medium supernatant and mouse lung homogenate were analyzed for TNF and IL-6 levels by using a TNF and IL-6 specific ELISA kit (eBioscience, San Diego, CA). For S100A9 ELISA, Costar High-Binding 96-well plates (Corning, NY) were coated overnight at 4°C with 800 ng/well of purified rabbit IgG against mouse S100A9 or 100 ng/well of goat polyclonal human S100A9 antibody (Abcam) diluted in 0.1 M carbonate buffer, pH 9.6. The wells were blocked with PBST+1% BSA for 1 h at room temperature. The samples were added and incubated overnight at 4°C. The plates were washed three times with PBST and incubated with either goat anti-mouse IgG (300 ng/well) (R&D) (for mouse S100A9) or mouse anti-human IgG (50 ng/well) (R&D) (for human S100A9) in PBST+0.1% BSA for 2 h at room temperature. The plates were then washed three times in PBST. To detect mouse S100A9, rabbit anti-goat HRP (Bio-Rad) was added to the plates. To detect human S100A9, goat anti-mouse HRP (Bio-Rad) was added. After 1 h incubation at room temperature, the plates were washed three times with PBST. TMB-S substrate (100 µl/well) (Sigma-Aldrich) was added to the plates according to the manufacturer's instructions. The ODs were detected at 450 nm, using a Modulas micro-plate reader.
To detect i.p.-injected S100A9 antibody in the lung homogenate, Costar High-Binding 96-well plates were coated overnight at 4°C with mouse S100A9 protein diluted in 0.1 M carbonate buffer, pH 9.6. The wells were blocked with PBST+1% BSA for 1 h at room temperature. The lung homogenate was added and incubated overnight at 4°C. The plates were washed three times with PBST and goat anti-rabbit HRP (Bio-Rad) was added. After 1 h of incubation at room temperature, the plates were washed three times with PBST. TMB-S substrate (100 µl/well) (Sigma-Aldrich) was added to the plates according to the manufacturer's instructions. ODs were detected at 450 nm by using a Modulas micro-plate reader.
For survival experiments, 6–8-week old pathogen-free WT C57BL/6 mice (Jackson Laboratory) were injected i.p. with 2 mg/mouse of either control IgG or anti-S100A9 antibody. One day later, mice were anesthetized and inoculated via the intratracheal or I.T route with IAV (1×105 pfu/mouse) in 100 µl of PBS (Invitrogen). Control mice were sham-inoculated with 100 µl of PBS. Survival was monitored until 8 days postinfection. For pathogenesis assay, mice were inoculated with IAV (2×104 pfu/mouse via the I.T route) at 1 day after antibody treatment. At 3 days after infection, lungs and BALF were collected. Lung tissue sections were used for H&E analysis and in-situ TUNEL analysis. Lung homogenate was used for ELISA analysis (for TNF and IL-6). RT-PCR analysis for TNF and IL-6 expression was done with RNA isolated from mouse lungs. BALF was used for Western blotting with S100A9 antibody and S100A9 ELISA analysis.
In some experiments, purified mouse S100A9 protein (15 µg/mouse) diluted in PBS or HBSS buffer diluted in PBS (vehicle control) was administered to mice via the I.T route. At 8 h posttreatment, TNF and IL-6 expression and production in the lung was monitored by RT-PCR and ELISA.
Lung sections from mock- or IAV-infected mice were stained with goat anti-mouse S100A9 antibody (1∶100 dilution) (R&D) for 2 h at room temperature. After washing five times with PBS, lung sections were incubated with anti-goat Texas Red (1∶50 dilution) (Vector Labs) for 1 h at room temperature. After washing three times with PBS, sections were mounted with DAPI containing mounting solution (Invitrogen). Sections were visualized by fluorescence microscopy.
To study apoptosis in the respiratory tract, TUNEL assays were done. Formalin-fixed lungs from IAV-infected mice were used. The TUNEL assay was done using an ApopTag Peroxidase In Situ Apoptosis Detection Kit (Milipore, MA). Digital images of TUNEL-stained lung sections were examined by light microscopy. Digital images were used to count the number of TUNEL-positive cells, using Image J software from NIH (http://rsbweb.nih.gov/ij/) as described previously by us [75]. For each analysis, an area of 5.39×102 µm×4.09×102 µm of TUNEL-stained lung section was scanned by Image J software. Gross apoptotic area was expressed as pixels per micron. This value was used to calculate the percentage of the apoptotic area in each analysis. Three IAV- infected mice treated with control IgG and three IAV-infected mice treated with S100A9 antibody were used. Data were collected from 9 areas per mouse from each experimental group. The values obtained from the 27 lung section areas of each experimental group were used for statistical analysis.
Hematoxylin and eosin (H&E) staining was performed on paraffin-embedded mouse lung sections. Briefly, slices of lung were sequentially rehydrated in 100% and 95% ethanol followed by xylene deparaffinization. After rinsing with distilled water, sections were stained with hematoxylin for 8 min and counterstained in eosin for 1 min followed by serial dehydration with 95% and 100% ethanol. Sections were then mounted on coverslips.
IAV-infected and S100A9 protein-treated cells were examined for apoptosis by annexin V labeling, using an annexin V/propidium iodide (PI) apoptosis detection kit (BioVision, CA) [75], [78], [79]. For TUNEL assay cells were grown in cover slips (12 mm diameter) (Ted-Pella, CA). TUNEL assay with macrophages was performed by using DeadEnd Colorimetric TUNEL System (Promega, WI). Digital images of TUNEL-stained macrophages were examined by light microscopy. Digital images were used to count the number of TUNEL-positive cells using Image J software (please see above). At least eight different fields were counted for each cover slip and two cover slips (duplicate) were examined for each experiment. Furthermore, each experiment was repeated independently three times.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.