ABSTRACT
HIV-1 has been shown to evolve independently in different anatomical compartments, but studies in the female genital tract have been inconclusive. Here, we examined evidence of compartmentalization using HIV-1 subtype C envelope (Env) glycoprotein genes (gp160) obtained from matched cervicovaginal lavage (CVL) and plasma samples over 2 to 3 years of infection. HIV-1 gp160 amplification from CVL was achieved for only 4 of 18 acutely infected women, and this was associated with the presence of proinflammatory cytokines and/or measurable viremia in the CVL. Maximum likelihood trees and divergence analyses showed that all four individuals had monophyletic compartment-specific clusters of CVL- and/or plasma-derived gp160 sequences at all or some time points. However, two participants (CAP177 and CAP217) had CVL gp160 diversity patterns that differed from those in plasma and showed restricted viral flow from the CVL. Statistical tests of compartmentalization revealed evidence of persistent compartment-specific gp160 evolution in CAP177, while in CAP217 this was intermittent. Lastly, we identified several Env sites that distinguished viruses in these two compartments; for CAP177, amino acid differences arose largely through positive selection, while insertions/deletions were more common in CAP217. In both cases these differences contributed to substantial charge changes spread across the Env. Our data indicate that, in some women, HIV-1 populations within the genital tract can have Env genetic features that differ from those of viruses in plasma, which could impact the sensitivity of viruses in the genital tract to vaginal microbicides and vaccine-elicited antibodies.
IMPORTANCE Most HIV-1 infections in sub-Saharan Africa are acquired heterosexually through the genital mucosa. Understanding the properties of viruses replicating in the female genital tract, and whether these properties differ from those of more commonly studied viruses replicating in the blood, is therefore important. Using longitudinal CVL and plasma-derived sequences from four HIV-1 subtype C-infected women, we found fewer viral migrations from the genital tract to plasma than in the opposite direction, suggesting a mucosal sieve effect from the genital tract to the blood compartment. Evidence for both persistent and intermittent compartmentalization between the genital tract and plasma viruses during chronic infection was detected in two of four individuals, perhaps explaining previously conflicting findings. In cases where compartmentalization occurred, comparison of CVL- and plasma-derived HIV sequences indicated that distinct features of viral populations in the CVL may affect the efficacy of microbicides and vaccines designed to provide mucosal immunity.
INTRODUCTION
Globally, there were almost two million new HIV-1 infections during 2017, with the majority occurring among women in sub-Saharan Africa via vaginal intercourse (1). HIV-1 subtype C is the predominant circulating genetic subtype in southern Africa, where the epidemic is mostly concentrated (1). It is not well understood whether viral populations in the female genital tract are genetically distinct from those in the blood compartment and, if so, how these populations emerge and evolve over time. Such studies on the properties of HIV populations in the female genital tract are important for the design of effective microbicides and vaccines that have the potential to reduce vaginal HIV transmissions in this region (2).
Compartmentalization is well described in HIV infection mostly using envelope (Env) glycoprotein sequences (gp160), which encompass the most diverse regions of HIV-1 genomes (3, 4). A number of factors may favor compartmentalization, including barriers between tissues, differential host cell availability, tissue-specific immune selection pressures, varying viral replication rates in compartments, and random genetic drift (3, 5). As a result, compartmentalized viral populations often possess tissue-specific phenotypic characteristics distinct from those of plasma viruses, including differences in cell tropism, the extent of glycosylation, and drug resistance (3). Compartmentalization is well described in the cerebrospinal fluid and the central nervous system, where it has been reported to affect responses to treatment and is associated with various brain complications (6–11). Independent viral populations in different compartments may produce variants with advantageous phenotypes for viral proliferation compared to parental strains (12). There is also convincing evidence that the semen constitutes a distinct compartment where locally produced HIV lineages evolve under selection pressures different from those in the blood (13–17). Relative to blood-derived viral populations, semen-derived populations show less diversity, decreased levels of positive selection, decreased CXCR4 coreceptor utilization, and altered glycosylation patterns (18). These differences may be at least in part attributable to an enrichment of cytokines and chemokines in the seminal tracts of HIV-infected men that promote T cell activation and viral replication (19).
Studies of HIV compartmentalization in the female genital tract have shown evidence of tissue-specific differences in some, but not all, studied women (20–30). In certain individuals, compartmentalization was shown to be associated with distinct viral features, including coreceptor usage, numbers of N-linked glycosylation sites, levels of neutralization resistance, and diversity (22, 25, 28, 29, 31). The temporal scope of most of these studies has largely been limited by the use of cross-sectional sampling. However, investigations that used more advanced computational and statistical analyses failed to find evidence of compartmentalization and have suggested that such studies can be biased by monotypic and low-diversity sequences (32, 33). A more recent study showed that there was higher HIV-1 Env sequence diversity in the vaginal tract early in infection than there was in the blood (34). This suggested the possibility of either restrictions in the movement of viruses from the genital tract to the systematic blood compartment or differences in HIV evolution within these respective compartments (34). However, this study utilized only the C2-V3-C3 region of Env and might have missed other tissue-specific signatures elsewhere in Env. The only detailed study of HIV compartmentalization in the female genital tract that examined complete gp160 sequences analyzed only a small number of sequences and did not include longitudinal samples from the same individuals (27).
In this study, we amplified longitudinally sampled complete HIV-1 subtype C gp160 envelope sequences from cervicovaginal lavage (CVL) and plasma samples and analyzed these for evidence of compartmentalization using various statistical methods. We present evidence for the existence of separate populations that are maintained over time in the genital tract and plasma compartments of some study participants. These participants showed an asymmetric pattern of viral migrations between the genital tract and plasma with fewer movements to the plasma compartment, suggesting a mucosal sieve. We conclude that compartmentalization may occur in some individuals, yielding viruses in the genital tract that are genetically distinct from those in the plasma.
RESULTS
HIV-1 gp160 amplification from the female genital tract over time.CVL samples were obtained from 18 women enrolled in the CAPRISA acute infection study in Durban, South Africa, within 2 to 15 weeks of HIV-1 infection (Table 1). Women were not menstruating at the time of sample collection, and there was no evidence of macroscopic blood contamination in CVL samples. Most samples were found to have low numbers of erythrocytes, typical of normal mucosal vascularization (35). All participants had detectable plasma viral loads (VL), but only two (CAP177 and CAP217) had detectable CVL viral loads (Table 1). Statistical analyses showed that the difference in CVL VL between those with amplifiable gp160 and those without approached significance (P = 0.0659). Full-length gp160 Env sequences were successfully amplified from the CVL samples from these two participants plus two others (CAP261 and CAP270) who had CVL viral loads of less than 50 copies/ml.
List of 18 CAPRISA individuals who donated early-infection CVL samples used in this studya
Longitudinal CVL samples from 2.5 to 3.5 years following acute infection were used to isolate additional sequences from these four participants. CVL viral loads were detectable in at least one time point in all cases (Table 2). Amplicons were obtained from the CVL supernatants at all seven sampling time points for CAP270, while fewer time points yielded amplicons for CAP177, CAP217, and CAP261 (5/10, 4/9, and 3/6 time points, respectively). The total number of sequences from CVL varied from 1 to 19 per time point and totaled 42 to 62 over time for each participant (Table 2). Similar numbers of HIV-1 gp160 Env sequences from plasma were obtained for all participants at matching time points.
Longitudinal CVL samples from four CAPRISA donors where gp160 was successfully amplified at acute infection
Association between CVL cytokine levels and gp160 amplification during acute infection.We investigated whether there was a correlation between Env amplification from CVL and local immune activation in the genital tract by measuring cytokine levels (36, 37). CVL samples from acute infection were available from 14 of the 18 participants for this analysis. Participants were grouped according to the concentrations of 20 cytokines measured in CVL using unsupervised hierarchical clustering. Three of the four participants with amplifiable gp160 clustered together on the heatmap, suggesting an association with higher concentrations of cytokines (Fig. 1A). Statistical analysis revealed that the proinflammatory cytokine cluster was significantly associated with CVL gp160 amplification (P = 0.024 by Mann-Whitney U test) (Fig. 1B).
Association between cytokine levels, STIs, and gp160 amplification in CVL at acute infection. (A) Heat map showing CVL cytokine levels, HIV gp160 PCR results, sexually transmitted infections (STIs), and bacterial vaginosis (BV). Participants with amplifiable CVL gp160 are shown in boldface. (B) Confirmatory factor analysis was used to group cytokines according to biological functions and generate factor scores for each cytokine group for each participant. Mann-Whitney U test was used to evaluate differences in factor scores between women with amplifiable gp160 (Pos) and those without (Neg). Lines indicate medians. P values of <0.05 were considered statistically significant and are shown in boldface.
Sexually transmitted infections (STIs) and bacterial vaginosis (BV) frequently result in genital tract inflammation and also increase HIV replication rates in the female genital tract via proinflammatory signaling pathways (38–40). We therefore investigated if the ability to amplify gp160 in the genital tract was related to the presence of STIs and BV. Eight of the 14 participants had evidence of an active STI (Fig. 1A), including the three participants from whom CVL gp160 was amplified. This included CAP177 (Neisseria gonorrhoeae positive), CAP261 (herpes simplex virus 2 PCR positive), and CAP217 (Mycoplasma genitalium positive), who all had infections known to be associated with increased HIV shedding (41–44). However, the presence of an STI was not associated with gp160 amplification (P = 0.5804 by Fisher's exact test). Although all four participants with amplifiable gp160 had BV, 6/10 participants from whom gp160 was not amplifiable also had BV, and the difference was not statistically significant (P = 0.2507 by Fisher's exact test).
We next performed logistic regression to further evaluate the relationship between inflammation and amplification. STIs, BV, and CVL VL were not significantly associated with proinflammatory cytokines (P = 0.296, P = 0.212, and P = 0.141, respectively). However, despite the small sample size, proinflammatory cytokines remained significantly associated with amplification after adjusting for STIs and CVL VL (P = 0.0357 and P = 0.0303, respectively) and approached significance after adjusting for BV (P = 0.0507). Thus, although proinflammatory cytokine production may be influenced by STIs, BV, and CVL VL, this analysis suggests that inflammation was associated with viral amplification regardless of the cause.
Analysis of compartmentalization in plasma and CVL samples.To determine if sequences in the CVL differed from those in plasma, we constructed maximum likelihood trees to identify monophyletic clades comprised of sequences sharing common phenotypic characteristics and ancestry. For all four participants the plasma and CVL sequences were largely intermingled on the trees, although there was some evidence of sequences grouping together by time point and compartment (Fig. 2). This raised the possibility of compartmentalization in these individuals.
Maximum likelihood trees of CVL and plasma Env sequences from four donors constructed in PhyML. Plasma and CVL sequences are indicated by filled and open circles, respectively, and the circle color represents the different time points. Node support greater than 0.348 is represented by black diamonds. The number of sequences included is shown in parentheses.
We next performed statistical tests for evidence of compartmentalization using Bayesian tip-association significance testing (BaTS) (45). We first analyzed all sequences from all time points (range, 40 to 82 sequences per compartment) and then repeated the analysis with identical sequences removed (range, 22 to 49 sequences per compartment) to rule out the possibility of inaccurate inferences resulting from localized replication (Table 3). There was strong evidence of compartmentalization in CAP177 and CAP217 regardless of whether identical sequences were included or not (Table 3 and Fig. 3A). The monophyletic clade size statistic for CVL was significantly higher than that for plasma for both CAP177 and CAP217, providing more support for genital tract compartmentalization. This was consistent with the hypothesis that viruses move less frequently from the genital tract to the plasma compartment than they do in the opposite direction. There was no detectable compartmentalization in CAP261, with or without identical sequences included in the analysis, while for CAP270 the evidence was lost when identical sequences were removed, suggesting that bursts of local replication can account for the observed compartmentation in this participant (Table 3 and Fig. 3A).
Compartmentalization tests using data sets with monotypic sequences included and excludeda
Statistical analysis of compartmentalization using BaTS. (A) Analysis of all sequences for each of the 4 participants with or without identical sequences removed. (B) Longitudinal analysis of sequences from 2 participants with evidence of compartmentalization. Each dot represents an independent statistical test for compartmentalization. Circles indicate cases where at least 2 tests were significant (P < 0.05) for that compartment. Blue and red represent CVL and plasma, respectively. The time points and numbers of sequences used in this analysis are shown in Tables 3 and 4.
Having confirmed evidence of compartmentalization in CAP177 and CAP217, we next determined whether this changed during the course of infection by analyzing the data at the four sampling time points per participant that yielded the highest numbers of sequences (range, 11 to 40 sequences per time point) (Table 4). We identified compartment-specific clusters at all time points in CAP177 and at the first and last time points in CAP217 (9 and 190 weeks) (Fig. 3B). We also observed significantly higher monophyletic clade size statistics for CVL than for plasma, providing more support for genital tract compartmentalization than for plasma compartmentalization. In summary, the Bayesian statistical analyses suggested compartmentalization in two of the four participants: in CAP177, where it was persistent, and in CAP217, where it was intermittent.
Longitudinal compartmentalization tests for CAP177 and CAP217 using BaTSa
HIV migration between genital tract and plasma compartments.We next compared viral migration between the plasma and CVL samples of the four individuals to assess the role of restricted viral migration as a cause of compartmentalization. To do this, we estimated the numbers of distinct migration events during the histories of the sampled viruses using Markov jumps in Bayesian evolutionary analyses by sampling trees (BEAST) (46, 47). The sequence phylogenies from all four individuals yielded evidence of more viral migration events from the plasma to the genital tract compartment (Markov counts from 16 to 23) than from the genital tract to the plasma compartment (Fig. 4). This was particularly evident for the CAP177 and CAP217 sequence phylogenies, both of which displayed evidence of only two instances of viruses moving from the genital tract to plasma (i.e., two Markov counts) during the evolutionary histories of all the sampled viruses. In contrast, the CAP261 and CAP270 phylogenies displayed evidence of eight and nine genital tract to plasma migration events, respectively. There was good statistical support for the occurrence of at least some movements from the genital tract to plasma, as indicated by the Bayes factors (BFs) (Fig. 4). Specifically, in all of the participants other than CAP270, the BFs for CVL to plasma movements was >5, indicating that there was approximately greater than 5 times more support for the occurrence of these movements than for their absence. These data suggest a mucosal sieve effect of viral movements from the genital tract to plasma, which is more severe in individuals with evidence of compartmentalization.
Viral migration events between CVL and plasma compartments. Migration events estimated as Markov counts in BEAST and the Bayes factor (BF) support for the movements between anatomical compartments are shown. Bayes factor values of >5 indicate statistical support for the movements.
Comparison of evolutionary differences between sequences in plasma and CVL.As compartmentalization may arise due to differential viral population diversification in tissues, we estimated and compared the average pairwise genetic distances between the sequences sampled at each of the time points in the separate compartments. We observed differences in diversity patterns between plasma- and CVL-derived sequences for CAP177 and CAP217 over the course of infection (Fig. 5A). This suggested that these two compartments exerted distinct pressures resulting in differential viral population diversification. CAP261 displayed higher viral sequence diversity in CVL than plasma at all time points (P = 0.0422 by two-way analysis of variance [ANOVA] test), but this did not result in detectable viral compartmentalization. CAP270 showed similar diversity patterns and levels in both compartments.
Comparison of viral evolution in plasma and CVL compartments of four individuals. (A) Longitudinal Env sequence diversity analyses in plasma and CVL (shown in red and blue, respectively) of four individuals. The average nucleotide pairwise distances were estimated in MEGA6. (B) Evolutionary rates and time of the most recent common ancestor (tMRCA) for plasma and CVL sequences estimated in BEAST using sequences from all time points. CAP177 showed clear differences between CVL- and plasma-derived sequences and is highlighted with an asterisk.
We next assessed whether compartmentalization was a result of differences in evolutionary rates and the ancestry of the viral lineages in the separate anatomical compartments using BEAST (48). We used all sequences obtained at all available time points for these analyses. In three of the four participants, no differences in either estimated nucleotide substitution rates or estimated times to most recent common ancestors (tMRCAs) were noted between the CVL and plasma sequences (Fig. 5B). However, for CAP177, the estimated nucleotide substitution rates and the tMRCA of plasma-derived sequences were significantly different from those of the CVL-derived sequences, as indicated by the 95% high posterior density (HPD) intervals that did not overlap (Fig. 5B, asterisk). In addition, for both CAP177 and CAP261 the estimated tMRCA for all sequences was greater than the actual time of infection, confirming that these individuals were likely to have been initially infected by multiple variants (49, 50, and unpublished data). In summary, these findings show that differences in nucleotide substitution rates and tMRCA of plasma- and CVL-derived gp160 sequences were only observed in viruses infecting CAP177, the participant displaying the strongest evidence of compartmentalization.
Comparison of CVL and plasma amino acid sequences in participants with evidence of compartmentalization.We next compared amino acids in the Env sequences at all sites in the nonoverlapping regions to identify compartment-specific differences between the plasma- and CVL-derived viruses. Using the highlighter tool on the HIV LANL website (51), we identified a total of 43 and 31 amino acid residues in CAP177 and CAP217 Env sequences that were either more or less common in CVL-derived viruses than in plasma-derived viruses, by a margin of 40% to 100%. We performed the comparisons using plasma and CVL sequences at time points that showed evidence of compartmentalization. The relative frequencies of synonymous and nonsynonymous substitutions within the codons encoding these amino acids estimated using the FUBAR (52) and MEME (53) natural selection detection methods suggested that these amino acid differences were a consequence of positive selection in CAP177 and both positive selection and insertions/deletions in CAP217 (Fig. 6). These amino acid differences resulted in charge differences at similar positions in the Envs of the plasma- and CVL-derived sequences, suggesting that different amino acids are preferred at these sites in the different compartments. In CAP217, insertions were more common in the V1V2 regions of CVL-derived sequences than they were in plasma-derived sequences (data not shown). These amino acid differences also increased with time in both participants (Fig. 6). When we compared the distribution of these sites on the Env of both participants, they were mostly in the V4 and C4 regions of CAP177 Env sequences and in the V1V2 regions of CAP217 Env sequences. In summary, these results suggest that the compartmentalization observed in CAP177 and CAP217 was associated with distinct, possibly compartment-specific, selection acting at gp160 codon sites and/or the differential accumulation in gp160 of insertions and deletions in CVL-derived and plasma-derived viruses.
Identification of HIV-1 Env glycoprotein sites that distinguish CVL and plasma sequences in individuals with evidence of compartmentalization. (A) Number of sites that are different between plasma and CVL sequences are shown in gray (P value of ≤0.05), positive selection on those sites estimated in FUBAR and MEME are in black, insertions/deletions (INDELS) on the sites are shown in green, and the amino acid differences that resulted in charge changes are shown in purple. (B and C) Distribution of the amino acid differences on the glycoprotein Envs over time. The bars are colored according to time point.
DISCUSSION
Our study provides evidence of compartmentalization of HIV-1 subtype C in the genital tract in two of four women studied. Using multiple analytical approaches, we showed that compartmentalization could be persistent or intermittent. Furthermore, we identified Env features in viruses from the genital tract that differentiated them from those found in plasma.There was evidence of restricted migrations of viruses from the genital tract to plasma in all four cases, which was more pronounced in individuals with compartmentalization. These data suggest that the female genital tract and the plasma can, in some cases, act as separate compartments, each with a virus population that evolves with a degree of independence from the other compartment. This highlights the fact that compartmentalization can occur in individuals to different degrees, which could impact the evolutionary dynamics of HIV-1, which vary from person to person (54).
Studies focusing on the detection and characterization of viral populations in the female genital tract have proven difficult due to challenges in collecting genital tract samples and the generally low viral loads in this compartment (34, 55–57). We made use of CVL samples that were collected at multiple time points from women in the CAPRISA cohort enrolled during acute HIV-1 infection and monitored through to chronic infection (36). Although sampling by means of CVL yields a good representation of a large genital tract area, the high degree of genital fluid dilution likely contributed to our failure to amplify Env from most of the CVL samples. Other studies that have had greater success amplifying HIV sequences from the female genital tract have made use of swabs or endocervical cytobrushes (56, 58–61). However, these methods capture only the viral population at the specific site in the cervix from which samples were taken and cannot be used to study cytokine levels. The use of CVL samples allowed us to discover that amplification of gp160 was significantly associated with levels of proinflammatory cytokines in the female genital tract. Furthermore, all four women from which viral sequences in CVL samples were amplifiable at the acute infection stage had clinical or laboratory evidence of an STI and/or detectable CVL viral loads. Local immune activation associated with STIs often results in an influx of leukocytes, elevated cytokine concentrations, and increased HIV shedding that may have favored Env amplification (61, 62).
We used a combination of genetic distance-based methods, Bayesian methods, and natural selection analyses of sequences sampled from plasma and the genital tract to detect and characterize compartmentalization of viral populations at these sites. Previous studies have used approaches such as genetic diversity and phylogenetic analyses to infer compartmentalization (23, 24, 27, 30), but as shown here, it is difficult to achieve robust conclusions based on the results from a single analytical approach. Until recently, previous compartmentalization studies did not account for low diversity and monotypic sequences which could occur because of local viral replication (32, 33). We accounted for these potentially confounding effects by repeating analyses with identical sequences removed and ensuring that almost equal numbers of single-genome amplification (SGA)-derived gp160 sequences were analyzed from CVL and plasma at matched time points. The numbers of sequences examined impacted the analyses in one participant (CAP270): time points with the most sequences (19 to 20 sequences at 9 and 190 weeks) showed strong statistical support compared to the time points with the lowest number of sequences (6 in plasma and 5 in CVL at 60 and 110 weeks), a finding that is probably attributable to the statistical power to detect compartmentalization increasing with increasing numbers of sampled sequences. In CAP177, however, compartmentalization was detectable at all time points regardless of the number of sequences examined. We confirmed evidence of compartmentalization with a combination of approaches that examined different aspects of viral evolution and identified individual-specific residues in Env that differentiated CVL-derived viruses from plasma-derived viruses.
Evidence of HIV compartmentalization between the female genital tract and plasma was strongly supported in two of the four participants. In CAP177, compartmentalization was persistent, with evidence for restricted viral movements from the genital tract to the plasma found at all four time points up to 186 weeks postinfection. In CAP217, compartmentalization was only detected at the first time point and at 190 weeks postinfection and not at the other two time points. Intermittent compartmentalization of HIV populations between the plasma and female genital tract has not been observed previously but has been reported between the male genital tract and the blood (16). CAP177 and CAP217 were the only two individuals who had detectable levels of virus in CVL at acute infection, suggesting that local viral replication and diversification favored subsequent compartmentalization of the genital tract HIV population from that in the blood.
Two of the study participants, CAP261 and CAP270, displayed no convincing evidence of compartmentalization, in that both CVL and plasma populations displayed similar patterns of diversity, indistinguishable evolutionary rates, and tMRCA estimates. There were also similar numbers of migration events detected in both directions between the genital tract and plasma. Other studies have similarly reported the lack of compartmentalization between the female genital tract and blood (20, 22, 27). By studying longitudinal samples, we have further shown that compartmentalization can occur sporadically in some individuals.
Regardless of whether there was strong evidence of compartmentalization, all four participants consistently showed higher numbers of virus migration events from the plasma to the genital tract compartment. Low levels of virus in the CVL may have limited our ability to detect migration events from the CVL compartment. However, diversity rather than viral population size has been shown to be a key factor in detecting compartmentalization (18, 25, 63). Here, we show that the levels of diversity were similar in both compartments despite the higher viral levels in plasma, which enabled us to quantify migration events in both directions.
The Env amino acid composition of the two individuals with evidence of compartmentalization showed distinctive differences that may have impacted properties of the protein, such as its charge distribution. Such differences have been reported to significantly alter key virologic properties, including cellular tropism and transmission (64). We speculate that the transmitted/founder (T/F) viruses in these individuals had properties that favored the genital tract over the plasma compartment. Individuals with no compartmentalization may have had T/F viruses with amino acid properties that were equally well adapted to the plasma and genital tract compartments. Persistent compartmentalization was associated with distinct evolutionary rates and tMRCA differences between the plasma and CVL-derived sequences, whereas intermittent compartmentalization was associated with more subtle evolutionary differences. A recent study showing that early HIV-1 viruses from the endocervix had higher degrees of genetic diversity than those from blood is consistent with the independent evolution of viruses in the female genital tract and/or the presence of a mucosal sieve effect between these two compartments (34).
In conclusion, compartmentalization of HIV populations between the female genital tract and plasma likely does occur in some individuals despite frequent movements of viruses from the blood to the female genital tract. Our data indicate that in cases where compartmentalization occurs, distinct genetic features of viral Env populations in the genital tract may impact the efficacy of microbicides and vaccines designed to provide mucosal immunity.
MATERIALS AND METHODS
Participants.This study made use of participants from the CAPRISA 002 Acute Infection study, a cohort of 245 uninfected high-risk women, which was established in KwaZulu-Natal in 2004 (65). The parent study was reviewed and approved by the research ethics committees of the University of KwaZulu-Natal (E013/04), the University of Cape Town (025/2004), and the University of the Witwatersrand (MM040202). All participants provided written informed consent for sample storage for research purposes. Ethics approval for this substudy was obtained from the University of the Witwatersrand (M160340).
Stored plasma and CVL samples from 18 women (ages 18 to 59 years) collected during acute HIV infection were selected for this study. The time of infection was defined as the midpoint between the last HIV-1 antibody-negative test and the first HIV-1 antibody-positive test or estimated to be 14 days prior to a positive RNA PCR assay result when the HIV-1 enzyme immunoassay (EIA) was negative. Plasma and CVL samples were collected at multiple matched time points from acute to chronic HIV infection (2 to 190 weeks postinfection).
Collection of CVL samples.The cervix was bathed with 10 ml of phosphate-buffered saline before aspirating the fluid from the posterior fornix. This was repeated with the same fluid 2 to 3 times. The collected fluid was centrifuged at 400 × g for 10 min to fractionate the cellular component from the supernatant. The supernatant was aliquoted and stored at −70°C. CVL samples were not collected during menstruation, and supernatants with macroscopic blood contamination were not used. The presence of microscopic blood was detected with Roche Cobas combur dipsticks, which detect 5 to 10 erythrocytes/µl (1+ score) to 250 erythrocytes/µl (4+ score). Viral load was measured in the 1-ml CVL supernatants with NucliSENS EasyQ HIV1, version 1.2, which has a limit of detection of 50 copies per ml.
Isolation of HIV RNA from CVL supernatant and plasma.Two different extraction procedures were tested to isolate viral RNA from CVL, either VitalVirus followed by QIAamp viral RNA Minikit (VitalVirus+QIAamp) or QIAamp viral RNA Minikit on its own. The µMACS VitalVirus HIV isolation kit (Miltenyi Biotec, Bergisch Gladbach, Germany) captures virus particles with magnetic microbeads that bind to host-derived CD44 in the viral envelope. Viral RNA was then isolated from the captured concentrated, intact virions with the QIAamp viral RNA Minikit (Qiagen, Hilden, Germany).
Using the QIAamp viral RNA Minikit, RNA was isolated from 0.2 to 1.75 ml CVL supernatant or from 140 µl plasma. When sample volumes exceeded 0.6 ml, the supernatant was centrifuged at 23,000 × g for 1 h at 4°C and the supernatant was discarded, except for 140 µl in the base of the tube. In order to ensure that all PCR products had been amplified from cDNA and not DNA, during RNA purification an on-column DNase digestion step (RNase-free DNase set; Qiagen) was added to the QIAamp procedure. The QIAamp extraction procedure on its own yielded more amplicons from CVL than VitalVirus+QIAamp, so the former was used for all subsequent amplifications.
SGA and gp160 sequencing.Viral RNA was reverse transcribed to cDNA using SuperScript III reverse transcriptase (Invitrogen, Carlsbad, CA) and the env primer OFM19 (66). The complete env was amplified in a nested PCR. In order to ensure that env genes were amplified from single cDNA copies, cDNA prepared from plasma was diluted before PCR until fewer than 30% of reactions were positive. In most cases, it was not necessary to dilute the CVL cDNA, as few reactions were positive. Amplicons were directly sequenced using the ABI PRISM BigDye Terminator cycle sequencing ready reaction kit (Applied Biosystems, Foster City, CA) and resolved on an ABI 3100 automated genetic analyzer. The full-length env sequences were assembled and edited using Sequencher v.4.5 software (Gene Codes, Ann Arbor, MI). Amplicons with double peaks or interrupted reading frames were not included in the sequence analysis.
Measurement of cytokines.The concentrations of 20 cytokines were measured in CVL samples from acute infection using Luminex multiplex flow-cytometric assays. After thawing, CVL samples were prefiltered by centrifugation using 0.2-μm cellulose acetate filters (Sigma, USA). Concentrations of eotaxin/CCL11, fractalkine/CX3CL1, granulocyte colony-stimulating factor (G-CSF), interleukin-1α (IL-1α), IL-8/CXCL8, IL-12p40, IL-15, monocyte chemotactic protein 1 (MCP-1)/CCL2, macrophage inflammatory protein 1α (MIP-1α)/CCL3, MIP-1β/CCL4, RANTES/CCL5, and soluble CD40 ligand (sCD40L) were measured using human cytokine LINCOplex kits (LINCO Research, MO, USA) according to the manufacturer’s protocol. The sensitivity of these kits ranged between 0.3 and 18.3 pg/ml for each of the 12 cytokines measured. Concentrations of IL-1β, IL-2, IL-6, IL-7, IL-10, IL-12p70, granulocyte macrophage (GM)-CSF, and tumor necrosis factor alpha (TNF-α) were measured in CVL using high-sensitivity human cytokine LINCOplex kits. The sensitivity ranged between 0.01 and 0.48 pg/ml for each of the 8 cytokines measured with high sensitivity. Data were collected using the Bio-Plex suspension array reader (Bio-Rad Laboratories Inc.), and a 5 PL regression formula was used to calculate sample concentrations from the standard curves. Data were analyzed using Bio-Plex manager software (version 4). Cytokine concentrations that were below the lower limit of detection of the assay were reported as the midpoint between the lowest concentration measured for each cytokine and zero.
Multivariate statistical analyses were performed using STATA (StataCorp, Texas, USA), and a hierarchical clustering analysis was performed using R. Confirmatory factor analysis was used to group cytokines according to biological functions and to generate factor scores for each participant. Factor scores are linear combinations of the concentrations of each cytokine in a-factor, weighted according to their factor loadings. Cytokines were grouped as proinflammatory (IL-1α, IL-1β, IL-6, IL-12p40, IL-12p70, and TNF-α), hematopoietic (G-CSF and GM-CSF), chemokines (IL-8, fractalkine, eotaxin, MCP-1, MIP-1α, MIP-1β, and RANTES), anti-inflammatory (IL-10), and adaptive immune mediators (IL-2, IL-7, IL-15, and sCD40L). Factor scores were compared using Mann-Whitney U test, and P values of <0.05 were considered statistically significant.
Sequence alignments.Envelope (env) nucleotide sequences from each participant were initially aligned with ClustalO (67) using HXB2 as the reference sequence. Sequence alignments were then codon aligned using the Se-Al v2.0a11 (http://tree.bio.ed.ac.uk/software/seal/) sequence alignment editor to avoid the inclusion of DNA encoding frame shifts and mistranslated sequences that would affect downstream analyses.
Phylogenetic analyses.Pairwise genetic distances between sequences were calculated using MEGA, v6 (68), and the statistical tests for significance (P < 0.05) were performed using a two-way ANOVA method with a Sidak’s multiple-comparison test. Maximum likelihood trees were constructed using PHYML (69) implemented in Recombination Detection Program (70). All sequences were analyzed in Data Analysis in Molecular Biology and Evolution (DAMBE), v7, software, and no evidence of nucleotide substitution saturation was found (71). Maximum clade credibility (MCC) trees were produced for CVL and plasma sequences in BEAST, v1.10.4, using best-fit demographic and clock models selected by path sampling and stepping-stone methods (47, 48, 72). The phylogenetic trees were viewed in FigTree (http://tree.bio.ed.ac.uk/software/figtree).
Bayesian analysis of viral compartmentalization.Posterior distributions containing 1,000 trees for both plasma and CVL sequences were constructed in BEAST for all time points and used as input for the BaTS analysis (45). Time points that only had sequences from one compartment type and identical sequences or low-diversity sequences that could inflate the appearance of structure were excluded in subsequent BaTS analyses. The BaTS analysis performs statistical analyses for evidence of lineage movements between locations (the female genital tract and blood plasma) indicative of compartmentalization using the association index (AI), Fitch parsimony score (PS) (also known as the Slatkin Maddison test), and the maximum exclusive single-state clade (MC) tests (45). Evidence of compartmentalization was inferred when significant results for at least two of the three statistical methods were obtained.
Analysis of HIV migration between CVL and plasma compartments.Phylogeography analyses of viral movements between CVL and plasma were performed in BEAST, v1.10.4 (47). The direction of viral flow was determined by applying a discrete diffusion asymmetric model with a Bayesian stochastic search variable selection (BSSVS) and Bayes factor tests to identify the most statistically supported viral movements between the two compartments (73, 74). The average number of migration events between the genital tract and plasma compartments was estimated by reconstructing the state change counts using Markov jumps (46). Bayes factor support of >5 was considered significant statistical evidence for viral movement between compartments (73).
Identifying amino acid differences in viruses from different compartments.Amino acid sites in CVL-derived sequences that were different from those in plasma-derived sequences were identified using the Highlighter tool (http://www.hiv.lanl.gov/). The consensuses of plasma-derived sequences were used to identify amino acid sites that were different from those found in the CVL-derived sequences. Only codons in the nonoverlapping regions of the Env were analyzed. The frequencies of any differences per site and the percent differences were then calculated. Sites that were evolving under positive selection were detected using the Fast Unbiased Bayesian Approximation (FUBAR) and Mixed Effects Model of Evolution (MEME) methods implemented in HyPhy (52, 53, 75).
ACKNOWLEDGMENTS
We thank participants in the CAPRISA cohort for their commitment to this study and the clinic and laboratory staff at CAPRISA for sample collection. We thank staff and students at NICD and UCT for generating the SGAs. We are grateful for funding from the Department of Science and Technology (DST) of South Africa, the South African Medical Research Council, the Poliomyelitis Research Foundation (PRF), and the CAPRISA Centre of Excellence program. B.M. and P.L.M. are supported by the South African Research Chairs Initiative (SARChI) of the DST and the National Research Foundation (NRF) of South Africa (grant no. 98341).
FOOTNOTES
- Received 22 February 2019.
- Accepted 27 February 2019.
- Accepted manuscript posted online 6 March 2019.
- Copyright © 2019 American Society for Microbiology.
