Quantitative profiling of the vaginal microbiota improves resolution of the microbiota-immune axis

Abstract

Background: The composition of the vaginal microbiota is closely linked to adverse sexual and reproductive health outcomes, due in part to effects on genital immunology. Compositional approaches such as metagenomic sequencing provide a snapshot of all bacteria in a sample and have become the standard for characterizing the vaginal microbiota, but only provide microbial relative abundances. We hypothesized that the addition of absolute abundance data would provide a more complete picture of host-microbe interactions in the female genital tract.

Results: We analyzed cervicovaginal secretions from 196 female sex workers in Kenya and found that bacterial load was elevated among women with diverse, bacterial vaginosis (BV)-type microbiota and lower among women with Lactobacillus predominance. Bacterial load was also positively associated with proinflammatory cytokines, such as IL-1α, and negatively associated with chemokines, such as IP-10. The associations between bacterial load and immune factors differed across bacterial community states, but L. crispatus predominance was the only microbial community where higher bacterial load was not associated with higher proinflammatory cytokines. Total vaginal bacterial load was also a stronger predictor of the genital immune environment than BV by Nugent score, the current clinical standard, in the Kenya-based cohort and in a Uganda-based confirmatory cohort.

Conclusions: Our results suggest that total vaginal bacterial load is at least as strong a predictor of the genital immune milieu as current BV clinical diagnostic tools, supporting exploration of the vaginal bacterial load as a predictor of adverse reproductive and sexual health outcomes. Video Abstract.

Fig. 1

High variability of vaginal bacterial load across community state types. A Relative abundance of the top 25 most abundant organisms identified by 16S rRNA gene sequencing, sorted by CST. B Absolute abundance of the top 25 most abundant organisms identified by 16S rRNA gene sequencing, sorted by CST. C Comparison of total vaginal bacterial load between major CSTs. p values were obtained with Tukey post-hoc tests following one-way ANOVA

Fig. 2

Vaginal soluble immune factors differ based on vaginal microbiota composition. Comparison of vaginal levels of A sE-cad, B IL-1α, C IL-1β, D IL-6, E IL-8, F IP-10, G MCP-1, H MIG, I MIP-1α, J MIP-1β, K MIP-3α, L TNF-α, and M MMP-9 between CST-I/II/V, CST-III, and CST-IV. p values obtained with Tukey post-hoc tests following one-way ANOVA

Fig. 3

Vaginal bacterial load is associated with vaginal immune factors. Association between total bacterial load and A sE-cad, B IL-1α, C IL-1β, D IL-6, E IL-8, F IP-10, G MCP-1, H MIG, I MIP-1α, J MIP-1β, K MIP-3α, L TNF-α, and M MMP-9. Data points are colored by CST. p values and test statistics were obtained with linear regression

Fig. 4

Relationship between vaginal bacterial load and genital immunology varies by vaginal microbiota composition. Heatmap displaying the association between A total bacterial load and vaginal immune factors, stratified by CST, and B estimated absolute abundance of key bacterial taxa and vaginal immune factors. Cells are shaded according to linear regression coefficients and annotated based on p values. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

Fig. 5

Vaginal bacterial load is a predictor of non-optimal vaginal microbiota states. Comparison of total vaginal bacterial load based on A Nugent BV and B molecular BV status. Receiver operator characteristic (ROC) curves for the prediction of C Nugent BV and D molecular BV with total bacterial load. p values were obtained with the Mann–Whitney U test or logistic regression

Fig. 6

Bacterial load is a strong predictor of BV in an independent, Uganda-based cohort. Comparison of total bacterial load based on A Nugent BV and B ROC curves for the prediction of B Nugent BV with bacterial load. p values obtained with the Mann–Whitney U test or logistic regression