Vaginal epithelial dysfunction is mediated by the microbiome, metabolome, and mTOR signaling

Abstract

Bacterial vaginosis (BV) is characterized by depletion of Lactobacillus and overgrowth of anaerobic and facultative bacteria, leading to increased mucosal inflammation, epithelial disruption, and poor reproductive health outcomes. However, the molecular mediators contributing to vaginal epithelial dysfunction are poorly understood. Here we utilize proteomic, transcriptomic, and metabolomic analyses to characterize biological features underlying BV in 405 African women and explore functional mechanisms in vitro. We identify five major vaginal microbiome groups: L. crispatus (21%), L. iners (18%), Lactobacillus (9%), Gardnerella (30%), and polymicrobial (22%). Using multi-omics we show that BV-associated epithelial disruption and mucosal inflammation link to the mammalian target of rapamycin (mTOR) pathway and associate with Gardnerella, M. mulieris, and specific metabolites including imidazole propionate. Experiments in vitro confirm that type strain G. vaginalis and M. mulieris supernatants and imidazole propionate directly affect epithelial barrier function and activation of mTOR pathways. These results find that the microbiome-mTOR axis is a central feature of epithelial dysfunction in BV.

Keywords: BV; CP: Microbiology; bacterial vaginosis; epithelial barrier; imidazolepropionic acid; in vitro models; inflammation; mTOR signaling; metabolomics; proteomics; transcriptomics; vaginal dysbiosis; vaginal microbiome.

Graphical abstract

Figure 1

Vaginal microbiome composition Vaginal swabs collected from 315 women were analyzed by mass spectrometry-based metaproteomics and passed quality control criteria (dark-blue participant group). 3,280 bacterial proteins from 20 unique genera and 142 species were identified in this group. Vaginal swab and vaginal tissues collected from another 90 HIV-negative women were analyzed by mass spectrometry and passed quality control criteria (light-blue participant group). 522 bacterial proteins from 15 unique genera were identified in this second group. All 405 samples were clustered and designated microbiome profile groups 0–4. Bacterial functions were also annotated using the KEGG database and are shown in the bottom panel in the same order as in the microbiome grouping. See also Table S2.

Figure 2

Proteomic and transcriptomic differences in cervicovaginal mucosal and tissue samples between vaginal microbiome groups (A) Heatmap showing differentially abundant proteins (353, 5% FDR) between vaginal microbiome groups, with functional annotation of biological pathways (p < 0.0001) shown on the right, for participant group 1. See Figure S1A for proteomics on participant group 2. (B) Gene set enrichment analysis (GSEA) of vaginal mucosal proteome differences between microbiome groups for participant group 1. See Figure S1B for pathway analysis of participant group 2. (C) Heatmap of differentially expressed genes by transcriptomic analysis of vaginal biopsy samples from a subset of 80 women (1,435 genes [<5% FDR]) between vaginal microbiome groups. Functional annotated pathways are indicated with major gene clusters. (D) GSEA of differentially expressed genes in vaginal tissue between microbiome groups. (E) Heatmap of vaginal mucosal metabolites significantly different between vaginal groups (27, 5% family-wise error rate). See Figure S5 for a summary of these interactions. See also Table S3. Protein expression from mucosal fluid taken by a vaginal swab in women from the first participant group, related to Figure 2, Table S4. Protein expression from mucosal fluid taken by a vaginal swab in women from the second participant group, related to Figures 2 and S1, Table S5. Gene (transcriptomics) expression from vaginal biopsies in women from the second participant group, related to Figure 2 for protein and gene expression.

Figure 3

Relationship between the mTOR activity signature in the vaginal microenvironment (A) Heatmap of the leading mTOR edge proteins identified by GSEA and relationship to vaginal microbiome. (B) Core mTOR proteins identified using the Pathway Interaction Database were used to identify mTOR expression. An mTOR activity score was determined using partial least-squares discriminant analysis (PLS-DA) inference of the latent variable combining pathway proteins or transcripts to describe mTOR pathway activity relative to LD status (“mTOR Score”). This numeric score was then used to determine both host and bacterial factors that associate with a high mTOR score. See Figure S2A for mTOR score analysis performed using Nugent score. (C) Vaginal tissue gene expression for immune response pathways (inflammation, TNFα, IFNγ) overlapped with genes associated with mTOR and LPS-stimulated genes identified using significant pathways identified GSEA when comparing Lactobacillus-dominant (LD) with non-dominant (nLD) groups. (D) mTOR score is significantly correlated with measured epithelial barrier protein expression in the mucosal fluid (Spearman’s correlations, p < 0.0001) showing key distinctions in barrier function and structural changes that were clearly separated by cluster analysis.

Figure 4

Host mTOR activity associates with bacterial variables and host immune and barrier functions in the mucosal environment Construction of multi-omics meta-model for microbiome-to-host regulation of mTOR activity. (A) Latent variables were computed for each data stream (microbiome composition, bacterial proteins, and metabolites) using partial least-squares regression (PLSR) to predict mTOR score before integration via a generalized linear model with interaction effects. (B–E) The mTOR activity score was then regressed against (B) bacterial taxa, (C) bacterial proteins, (D) bacterial GO functions, and (E) metabolites measured in the vaginal fluid. (F) Multi-omics integration via latent variable interaction effects linear modeling identified top variables that associate with an activated mTOR pathway. The significant interactions between mTOR pathway and latent variables, bacterial composition LV1, and metabolome LV2 are shown here. Bacterial taxa positively associated with an activated mTOR score include Gardnerella, Prevotella amnii, and Mobiluncus mulieris. Vaginal metabolites associated with activated mTOR expression include 5-aminovalerate, N-acetylputrescine, imidazolepropionic acid, xanthine, and tyramine. See also Figures S2B–S2D and S3; Table S6.

Figure 5

BV-associated bacterial supernatants affect vaginal epithelial barrier function in vitro and affect pathways involved in mTOR activity and epithelial function (A) Heatmap showing differentially abundant proteins in VK2 cells treated with different bacterial culture supernatants for 24 h, as observed by mass spectrometry. Proteins that activate mTOR cluster in the group that is upregulated in BV-associated bacterial treated cells (red box). Numerous structural and signaling adherens proteins were also differentially regulated by BV-associated bacteria (highlighted in green). (B) Pathway analysis showing top pathways including wound healing, tissue remodeling, and mTOR activation. (C) G. vaginalis inhibited VK2 cell differentiation and thickness compared with L. crispatus (n = 9). (D) G. vaginalis inhibited wound-healing ability as measured by scratch assay in Hec1a cells (n = 9). (E) M. mulieris decreased barrier integrity as measured by electrical resistance (TEER) on transwell membranes of Hec1a cells (n = 12). (F) M. mulieris induced a leaky barrier with protein-sized particles (dextran-FITC, 70 kDa) able to translocate across the membrane, but not larger-sized particles (n = 6). All error bars represent standard deviation from the mean, p values represent unpaired t test results. See also Figure S4.

Figure 6

Bacterial and mTOR-associated metabolite imidazolepropionic acid leads to epithelial barrier disruption (A–C) (A) One of the metabolites identified in our multi-omics model (imidazolepropionic acid) to be significantly correlated with mTOR activation and BV-associated bacteria was found to be produced by G. vaginalis in supernatants of pure cultures measured against negative control bacterial media samples. Imidazolepropionic acid purchased from Sigma was used to treat Hec1a cells and proteomics, analysis of which showed both (B) protein expression and (C) pathway-level disruption of mTOR activity. (D) We next showed that imidazolepropionic acid disrupted barrier integrity in a dose-dependent manner using transepithelial electrical resistance of Hec1a cells on transwell membranes. The error bar represents the standard deviation from the mean; n = 12, p value represents unpaired t test results.