Participant characteristics.
This study included 200 female sex workers (FSW) from Nairobi, Kenya (see Methods section for exclusion criteria) with deliberate enrichment of participants with BV by Nugent score. Full demographic data are shown in Table 1. Participants had a median age of 30 years. The participants had a median of 5 years in sex work, with a median of 3 clients in the past week. All participants had abstained from vaginal sex for at least 24 hours. More than half (52.5%) of the participants reported vaginal washing within the last 30 days. Three participants (1.5%) were seropositive for syphilis based on RPR (rapid plasma reagin) testing.
Table 1
Table of participant characteristics (n = 200). Values are reported as either count (percentage) or as median [IQR].
Characteristic | Total (n = 200) |
Age | 30 [24–37] |
Contraceptive Methods Oral contraceptives Copper IUD Injectables Implant | 25 (12.5%) 5 (2.5%) 34 (17%) 34 (17%) |
Years in sex work | 5 [3–9] |
Days since last vaginal sex | 3 [2–3] |
Number of clients in the last 7 days | 3 [0-5.5] |
Vaginal washing | 105 (52.5%) |
HSV-2 seropositivity | 107 (53.5%) |
Positive syphilis rapid plasma reagin | 3 (1.5%) |
Bacterial Vaginosis (by Nugent) BV Intermediate Flora Normal | 90 (45%) 45 (22.5%) 66 (33%) |
Anal sex practice in the last 6 months | 3 (1.5%) |
Total cervicovaginal IgA and IgG levels.
First, we measured the concentration of total IgA and IgG in cervicovaginal secretions from all participants. One participant (0.5%) had levels of IgA below the limits of quantification, and 15 (7.5%) participants had levels of IgG below the limits of quantification, these participants were assigned the value of the lower limit of quantification. The mean concentration of IgA was log10 4.944 ± 0.598 pg/mL and the mean concentration of IgG was log10 4.736 ± 0.994 pg/mL, and within participants, the total cervicovaginal concentration of IgA was higher than IgG (log10 difference: +0.209, paired t-test p = 0.004; Fig. 1).
IgA and IgG concentrations were not significantly associated with age, age of first vaginal sex, age when the participant started selling sex, the number of years selling sex, days since last vaginal sex, clients in the last 7 days, condom use, contraceptive method, HSV status or vaginal douching. Total concentrations of IgA and IgG were substantially lower among women with Nugent BV than among women with either intermediate flora or normal flora (IgA: mean difference of BV vs. intermediate flora: log10-0.581 ± 0.102 ng/mL p < 0.0001, mean difference of BV vs. normal flora: log10-0.354 ± 0.089 ng/mL p = 0.0003; IgG: mean difference of BV vs. intermediate flora: log10-1.083 ± 0.155 ng/mL p < 0.0001, mean difference of BV vs. normal flora: log10-1.054 ± 0.138 ng/mL p < 0.0001).
Microbe-binding IgA and IgG in cervicovaginal secretions.
Having demonstrated that the vaginal microbiota has an important effect on total levels of IgA and IgG, we next quantified the binding of cervicovaginal IgA and IgG to four key vaginal bacteria: Lactobacillus crispatus, Lactobacillus iners, Gardnerella vaginalis, and Prevotella bivia using a flow cytometry-based method (SFigure 3A; Materials and Methods). We incubated four serial dilutions (1/40, 1/80, 1/160, 1/320) of cervicovaginal secretions with each cultured bacterium and then stained with fluorophore-labelled secondary antibodies specific for either human IgA or IgG and analyzed by flow cytometry. MFI values were recorded for each serial dilution (Fig. 2A for representative plots, and SFigure2 for plots of all participants).
Microbe-binding antibodies were common amongst the cohort, with most participants having both IgA and IgG that recognize all four bacteria (Fig. 2B). However, the distribution of the MFI responses did not indicate a clear-cut threshold to define a response. Notably, among participants with similar ranges of either total IgA or IgG, we observe at least a one-log variance in the microbe-binding IgA and IgG MFI responses to any of the four bacterial species (SFigure 4). Participants with total vaginal IgG concentrations below the limits of quantification generally exhibited low microbe-binding responses against all four taxa (most within the lower 10% percentile of all MFI values), except one participant who demonstrated moderate responses against L. iners and L. crispatus (MFI: 1774,7 and 647.6 respectively). The one participant with IgA below the limits of quantification, suspected to have IgA deficiency, had very low microbe-binding responses against all taxa (all MFI values < 3). Additionally, we observed differences in the antibody-bound proportion of each species. G. vaginalis, P. bivia, and L. crispatus were bound by IgA at a higher proportion than IgG (median difference of IgA + and IgG + bacteria: 47.48%, 41.15%, 20.75% respectively), whereas L. iners was bound by a slightly greater proportion of IgG than IgA (median difference of IgA + and IgG + bacteria: -0.233%; Fig. 2C). BV-associated bacteria, P. bivia and G. vaginalis were more commonly bound by only IgA antibodies, whereas L. iners and L. crispatus were more commonly bound by both IgA and IgG (Fig. 2D).
We assessed the degree of binding using the AUC value generated by the series of MFI values measured for each sample. The AUC values are strongly correlated with MFI values (all Spearman’s rho > 0.895 and p < 0.0001) and highly reproducible across experiments (coefficient of variation < 10% for MFI values above 50). The mean IgA AUC response to G. vaginalis, P. bivia, L. iners and L. crispatus were 2.183 ± 0.514, 2.212 ± 0.521, 2.867 ± 0.571 and 2.682 ± 0.472, respectively, while the mean IgG AUC response was 1.600 ± 0.621, 1.586 ± 0.793, 2.854 ± 0.496, and 2.281 ± 0.829 (Table 2).
Overall, microbe-binding IgA and IgG AUCs were strongly correlated with total IgA and IgG concentrations (all IgA r > 0.631, p > 0.0001; all IgG r > 0.770, p < 0.0001, Fig. 2D-E). However, using hierarchical clustering, we clustered the microbe-binding IgA and IgG AUC responses within the cohort (Fig. 2F-G) and observed a spectrum of responses ranging from strong responses against all four bacteria to low responses against all four bacteria, with some participants demonstrating heterogeneity in bacterial responses, suggesting that these responses are not purely dependent on total antibody concentrations. In summary, cervicovaginal IgA and IgG frequently and heterogeneously bound key vaginal bacteria species G. vaginalis, P. bivia, L. iners, and L. crispatus.
Table 2
Microbe-binding antibody responses across participants
Species | MFI Median (IQR) | AUC Mean ± SD |
IgA MFI | IgG MFI | IgA AUC | IgG AUC |
G. vaginalis | 642.98 (315.65–1334.64) | 173.80 (59.75 − 367.5) | 2.183 ± 0.514 | 1.600 ± 0.621 |
P. bivia | 807.26 (331.16–1522.16) | 223.15 (80.75–590.60) | 2.212 ± 0.521 | 1.586 ± 0.793 |
L. iners | 4119.50 (1870–8347.5) | 3036.85 (1924.80–4353.40) | 2.867 ± 0.571 | 2.854 ± 0.496 |
L. crispatus | 2473.35 (1072.15–5002.10) | 1040.15 (357.80–2744.25) | 2.682 ± 0.472 | 2.281 ± 0.829 |
Microbial antibody responses amongst women with and without BV.
We then evaluated whether BV was associated with differences in microbe-binding antibody responses as well as total antibody concentration. Specifically, we compared total antibody and microbe-binding antibody responses associated with molecular BV (molBV) using 16S rRNA sequencing of cervicovaginal secretion samples and the Valencia classification program: 123 participants were identified as having molBV, while 75 participants were identified as having normal flora. Two participants had an insufficient number of reads (< 1000 reads) to ascertain molecular BV status using Valencia and were subsequently excluded from all BV-specific analyses.
First, we compared the concentration of total IgA and IgG between women with and without molBV. Women with molBV had approximately a 33% reduction in total IgA (Fig. 3A; BV positive: log10 4.861 ± 0.624 pg/mL, BV negative: log10 5.066 ± 0.532 pg/mL, p = 0.0148) and an 80% reduction total IgG (Fig. 3A; molBV positive: log10 4.465 ± 1.138 pg/mL, molBV negative: log10 5.149 ± 0.452 pg/mL, p < 0.0001) when compared to women without molBV. Of note, all participants with IgG below the limit of quantification had molBV, and those with quantifiable IgG levels were less likely to have molBV (OR: 0 [CI: 0-0.369], p < 0.0006). Even with the exclusion of undetectable IgG values, total IgG levels were lower amongst participants with molBV (p = 0.0001) than participants without molBV. There was relative enrichment of IgA in women with molBV (Fig. 3B), with an IgA to IgG ratio of 1.71 [IQR: 0.56–8.29] vs. 0.77 [IQR: 0.51–1.50], p < 0.0001) in those who were molBV negative. Women with molBV had higher levels of IgA than IgG (comparison to 1.0 ratio, p < 0.0001), and women without molBV had comparable levels of IgA and IgG (p = 0.5199).
Next, we compared IgA and IgG AUC responses to the four key vaginal bacteria analyzed in our flow cytometry assay: G. vaginalis, P. bivia, L. crispatus and L. iners. In keeping with the reduced overall antibody concentrations, women with molBV also had lower levels of IgA binding to G. vaginalis and L. crispatus, a trend towards lower levels of IgA binding to P. bivia (p = 0.0754), and decreased IgG binding to all four taxa measured (all p < 0.0001, Fig. 3C). When binding was normalized for the reduced total IgA and IgG concentrations, women with molBV demonstrated higher microbe-binding IgA AUCs for all four bacterial species, as well as higher IgG responses against G. vaginalis, L. crispatus and L. iners (Fig. 3D).
Microbe-binding IgA and IgG and the detectability of key vaginal bacteria.
We then assessed whether detectable levels of the four key vaginal bacteria in the vaginal microbiome were associated with microbe-binding IgA and IgG. Here, the four key vaginal bacteria were quantified using qPCR and normalized based on cervicovaginal secretion volume. Two participants were excluded from qPCR analyses since their CVS volumes were not recorded. P. bivia was detectable by qPCR above the limit of quantification (LLOQ) in 38.9% (77/198) of participants, L. crispatus in 22.7% (45/198) of participants, L. iners in 80% (156/198) of participants, and G. vaginalis in 93.4% (185/198) of participants (Supplementary Table 2). Detectable levels of most key bacteria above the LLOQ (ie: presence/absence) were not associated with elevated levels of their respective microbe-binding IgA and IgG (Table 3); the detection of L. crispatus was associated with elevated levels of L. crispatus binding IgG, but also elevated levels of microbe-binding IgG against other key bacteria. G. vaginalis detection was associated with decreased G. vaginalis binding IgG, and P. bivia detection was associated with decreased P. bivia binding IgA and IgG; similarly, each was also associated with decreased microbe-binding antibodies against other key bacteria.
Subsequently, we assessed whether a higher relative abundance of the key vaginal bacteria was associated with elevated microbe-binding response, which may suggest dominance is required to induce microbe-response. G. vaginalis was the dominant microbe (> 50% relative abundance) in 10.6% (21/198) participants, L. iners in 22.7% (45/198) participants, and L. crispatus in 10.1% (20/198) participants. P. bivia was not a dominant microbe in any vaginal microbiome observed. Dominance by either L. crispatus or L. iners was associated with elevated microbe-binding IgG AUC responses against their respective species, but also with elevated microbe-binding antibody responses against other key vaginal bacteria. Dominance by G. vaginalis was not associated with elevated microbe-binding antibodies (Supplementary Table 3). Overall, microbe-binding IgA and IgG do not seem to be associated with the detectability of their respective vaginal bacteria.
Table 3
Microbe-binding AUC in participants with and without key vaginal bacteria.
| Difference of G. vaginalis IgA AUC | Difference of G. vaginalis IgG AUC | Difference of P. bivia IgA AUC | Difference of P. bivia IgG AUC | Difference of L. iners IgA AUC | Difference of L. iners IgG AUC | Difference of L. crispatus IgA AUC | Difference of L. crispatus IgG AUC |
G. vaginalis detectable vs. undetectable | -0.230 | -0.534* | 0.042 | -0.415** | -0.126 | -0.392**** | -0.054 | -0.526**** |
P. bivia detectable vs. undetectable | -0.034 | -0.059 | 0.022 | -0.074 | 0.056 | -0.001 | 0.018 | -0.069 |
L. iners detectable vs. undetectable | -0.164 | -0.282* | -0.151 | -0.338* | -0.353** | -0.225* | -0.1193 | -0.368** |
L. crispatus detectable vs. undetectable | -0.032 | 0.212* | 0.032 | 0.388*** | 0.096 | 0.306**** | 0.105 | 0.522**** |
Differences of means by Welch’s T-test. Significant values are bolded and p-values are as indicated: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.
Antibody associations with individual bacterial species
Having assessed relationships with relative abundance, we next evaluated the relationship between microbe-binding IgA and IgG and the absolute abundance of key vaginal bacteria. We quantified the total bacterial abundance based on 16S copies measured by qPCR. In keeping with our finding that molBV was associated with reduced total IgA and IgG levels, total levels of IgA and IgG were strongly negatively correlated with total bacterial abundance (Fig. 4; IgA: r= -0.438, p < 0.0001; IgG: -0.539, p < 0.0001). The relationship between total immunoglobulin levels and bacterial abundance was primarily driven by participants with molBV (n = 123; IgA: r= -0.596, IgG: r= -0.531, both p < 0.0001). In those without molBV, there was no significant correlation between total IgA/IgG and overall bacterial abundance (n = 75; IgA: r = 0.069; IgG: r= -0.159). Both total and G. vaginalis-binding IgA and IgG correlated with a lower absolute abundance of total bacteria and G. vaginalis, and both total and L. crispatus-binding IgG correlated with a higher total abundance of L. crispatus (Table 4). However, only L. iners-binding IgA and IgG (and not total overall IgA and IgG) correlated with lower levels of L. iners, suggesting a possible species-specific effect. Of note, most vaginal bacteria strongly co-associate with each other and total bacteria abundance and therefore can have similar associations with antibody responses, except L. iners does not correlate with total bacteria or BV-associated bacteria. (Supplementary Table 4). Overall, total and microbe-binding IgA and IgG were strongly correlated with reduced levels of bacteria abundance in the FGT, in particularly in the context of molBV.
Table 4
Associations of total and microbe-binding IgA and IgG with vaginal bacteria abundance
Antibody | Total bacterial abundance (log ng/mL of CVS) | G. vaginalis (log ng/mL of CVS) | P. bivia (log ng/mL of CVS) | L. iners (log ng/mL of CVS) | L. crispatus (log ng/mL of CVS) |
G. vaginalis binding IgA | -0.369**** | -0.264*** | -0.076 | -0.015 | 0.011 |
G. vaginalis binding IgG | -0.526**** | -0.444**** | -0.070 | -0.009 | 0.152* |
P. bivia binding IgA | -0.318**** | -0.210** | -0.003 | -0.055 | 0.069 |
P. bivia binding IgG | -0.483**** | -0.370**** | 0.031 | -0.039 | 0.170* |
L. iners binding IgA | -0.277**** | -0.235*** | -0.043 | -0.273*** | 0.119 |
L. iners binding IgG | -0.574**** | -0.503**** | -0.074 | -0.160* | 0.296**** |
L. crispatus binding IgA | -0.404**** | -0.342**** | -0.056 | 0.013 | 0.133 |
L. crispatus binding IgA | -0.454**** | -0.391**** | -0.044 | -0.050 | 0.263*** |
Total IgA (log ng/mL) | -0.448**** | -0.316**** | -0.018 | -0.087 | 0.123 |
Total IgG (log ng/mL) | -0.589**** | -0.469**** | -0.006 | -0.050 | 0.221** |
Spearman’s correlation. Significant values are bolded and p-values are as indicated: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.
Associations of soluble immune factors, microbe-binding antibodies, and the vaginal microbiome
Molecular BV was strongly associated with elevated levels of the pro-inflammatory cytokines IL-1α, IL-1β and TNF, as well as elevated levels of epithelial disruption markers sE-cad and MMP-9, and with decreased levels of the chemokines IP-10, MCP-1, MIG, and MIP-3α (SFigure 5A). High bacterial abundance is a hallmark of molBV and correlated with similar soluble immune factors. Specifically, total bacterial abundance correlated with elevated vaginal levels of sE-cad, IL-1α and IL-1β, and with reduced IL-6, IP-10, MCP-1, MIG, MIP-1α, MIP-1β and MIP-3α (SFigure 5B). Therefore, subsequent analyses of microbiome-immune associations controlled for total bacterial abundance.
We hypothesized that microbe-binding IgA and IgG would have anti-inflammatory properties; therefore, we next assessed their correlation with both pro-inflammatory cytokines and chemokines, using partial correlations to control for total bacterial density. Higher concentrations of total IgA and IgG were correlated with elevated levels of the pro-inflammatory cytokines (IL-6, TNF) and chemokines (IL-8, IP-10, MCP-1, MIG, MIP-1α, MIP-1β, MIP-3α; Fig. 5A). Total IgA correlated with elevated IL-1β and markers of epithelial disruption (sE-cadherin and MMP-9), while total IgG was inversely correlated with vaginal levels of IL-1α. Next, we assessed if there were taxonomically specific immune effects of microbe-binding IgA and IgG. Once again, controlling for total bacterial abundance, we generally observed a positive correlation with inflammatory cytokines (IL-1β, IL-6, TNF) and chemokines (IL-8, IP-10, MCP-1, MIG, MIP-1α, MIP-1β, MIP-3α), and inversely correlated with IL-1α (Fig. 5B). Of note, microbe-binding IgA and IgG responses did not associate with sE-cadherin, and L. crispatus-binding IgA and IgG and L. iners-binding IgG had the strongest inverse associations with IL-1α (all r > 0.215, p > 0.01). In summary, total and microbe-binding IgA and IgG correlate with elevated levels of pro-inflammatory cytokines and chemokines.