Bacterial vaginosis and health-associated bacteria modulate the immunometabolic landscape in 3D model of human cervix

Abstract

Bacterial vaginosis (BV) is an enigmatic polymicrobial condition characterized by a depletion of health-associated Lactobacillus and an overgrowth of anaerobes. Importantly, BV is linked to adverse gynecologic and obstetric outcomes: an increased risk of sexually transmitted infections, preterm birth, and cancer. We hypothesized that members of the cervicovaginal microbiota distinctly contribute to immunometabolic changes in the human cervix, leading to these sequelae. Our 3D epithelial cell model that recapitulates the human cervical epithelium was infected with clinical isolates of cervicovaginal bacteria, alone or as a polymicrobial community. We used Lactobacillus crispatus as a representative health-associated commensal and four common BV-associated species: Gardnerella vaginalis, Prevotella bivia, Atopobium vaginae, and Sneathia amnii. The immunometabolic profiles of these microenvironments were analyzed using multiplex immunoassays and untargeted global metabolomics. A. vaginae and S. amnii exhibited the highest proinflammatory potential through induction of cytokines, iNOS, and oxidative stress-associated compounds. G. vaginalis, P. bivia, and S. amnii distinctly altered physicochemical barrier-related proteins and metabolites (mucins, sialic acid, polyamines), whereas L. crispatus produced an antimicrobial compound, phenyllactic acid. Alterations to the immunometabolic landscape correlate with symptoms and hallmarks of BV and connected BV with adverse women's health outcomes. Overall, this study demonstrated that 3D cervical epithelial cell colonized with cervicovaginal microbiota faithfully reproduce the immunometabolic microenvironment previously observed in clinical studies and can successfully be used as a robust tool to evaluate host responses to commensal and pathogenic bacteria in the female reproductive tract.

Conflict of interest statement

The authors declare no competing interests.

© 2021. The Author(s).

Figures

Fig. 1. Cervicovaginal Lactobacillus and BVAB effectively colonize the surface and crevices of human 3D cervical epithelial cell model.

Scanning electron micrographs of human 3D cervical models infected with L. crispatus JV-V01 (a), A. vaginae CCUG 38953 (b), S. amnii Sn35 (c), G. vaginalis JCP8151B (d), P. bivia VPI 6822 (e), or a polymicrobial community (consisting of equal parts of four BVAB: G. vaginalis, P. bivia, S. amnii, and A. vaginae) (orange) (f). Bacterial cells had expected shapes (i.e., rod-shaped bacilli for L. crispatus or small coccobacilli for BVAB species). Bacterial cells attached to surfaces of epithelial cells and formed clusters, which interacted with multiple epithelial cells. Scale bars on micrographs show 5 μm in distance. Bacteria were pseudocolored using the Affinity Designer software: cyan blue for L. crispatus, red for A. vaginae, yellow for S. amnii, green for G. vaginalis, purple for P. bivia, and orange for the polymicrobial cocktail. These colors are also used in the following figures to indicate particular bacterial infections.

Fig. 2. BVAB species distinctively modulate host defense responses in human 3D cervical models relative to Lactobacillus and PBS controls.

A. vaginae, S. amnii, and polymicrobial community exerted the greatest proinflammatory potentials, whereas G. vaginalis and P. bivia mostly altered the epithelial barrier targets. S. amnii also induced proteins related to cellular stress and angiogenesis. Human 3D cervical models were colonized with L. crispatus type strain (1) and L. crispatus vaginal isolate (2) or infected with single BVAB species and a polymicrobial community (consisting of four tested BVAB) for 24 h. The heatmaps depict relative levels of cytokines/chemokines (a), protein targets related to physicochemical barrier (b), and growth factors and cellular stress-related proteins (c) evaluated in the cell culture supernatants. Bacterial infections were grouped using hierarchical clustering. The data were mean centered and scaled and variance was scaled for each target. Clustering was based on Euclidean distance and average linkage. Black dots indicate significant changes (p < 0.05) in protein levels when compared to both Lactobacillus and uninfected (PBS) controls. Gray dots indicate targets significantly different from PBS only, whereas white dots indicate targets significantly different from Lactobacillus only. Statistical differences between the mean levels of protein targets among the groups were determined using ANOVA with Tukey’s adjustment for multiple comparisons.

Fig. 3. Human 3D cervical models infected with A. vaginae, S. amnii, and a polymicrobial community exert more similar metabolic profiles and cluster separately from G. vaginalis, P. bivia, and L. crispatus-colonized models or uninfected control.

Human 3D cervical models were colonized with L. crispatus type strain (1) and L. crispatus vaginal isolate (2) or infected with single BVAB species and a polymicrobial community (consisting of four tested BVAB) for 24 h. Global metabolomic analysis was performed on collected cell culture supernatants. a Total number of significantly (p < 0.05) altered metabolites following bacterial infections when compared to uninfected controls. The superpathways are indicated by colored bars. The polymicrobial infection altered the greatest number of metabolites, which mostly belonged to amino acid and nucleotide superpathways. b Number of unique and shared metabolites among the monomicrobial and polymicrobial BVAB infections visualized by the Venn diagram. Only five metabolites were shared by all the tested BVAB, indicating unique metabolic profiles of BVAB species. c Hierarchical clustering of all detected metabolites based on Euclidean distance and average linkage. The analysis revealed three separate clusters. Cell culture medium control clustered separately from all other groups. L. crispatus, G. vaginalis, and P. bivia clustered together with uninfected controls, whereas A. vaginae and S. amnii clustered with the polymicrobial infection. Statistical differences were determined using Welch’s two-sample t test.

Fig. 4. Metabolites in 3D cell culture supernatants highly predict monomicrobial and polymicrobial infection with BVAB or Lactobacillus.

Metabolome data were used to predict bacterial infections using Random Forest. The analysis revealed that the most predictive metabolites belong to amino acid and nucleotide superpathways, which were altered in species-specific manner. The analysis highlighted excellent predictive accuracy (93.75%) of infection, compared to random 12.5%. a Twenty most predictive metabolites are depicted and ranked by relative importance score. The superpathways are indicated by colored dots. b The heatmap shows fold changes of the most predictive metabolites among bacterial infections compared to uninfected controls. Only the fold changes that were significant (p < 0.05; q < 0.01) are displayed. Gold- and purple-shaded squares indicate metabolite enrichment or depletion, respectively. Statistical differences were determined using Welch’s two-sample t test with FDR correction. c The confusion matrix illustrates the proportion of times each sample receives the correct classification. Among the 36 tested samples, only 2 were confused (1 from L. crispatus-colonized models and 1 from P. bivia-infected models).

Fig. 5. Cervicovaginal Lactobacillus crispatus specifically alters carbohydrate and amino acids pathways in a human 3D cervical model.

L. crispatus depletes glucose and alters energy metabolism, induce phenyllactate and N-acetylated amino acid production, and histidine degradation when compared to uninfected control and monomicrobial and polymicrobial BVAB infections. a The heatmap shows fold changes of the metabolites among bacterial infections compared to uninfected controls. Only the fold changes that were significant (p < 0.05) are displayed. Gold- and purple-shaded squares indicate metabolite enrichment or depletion, respectively. b–e Key metabolites altered by L. crispatus in human 3D cervical cell model relate to energy metabolism (b), synthesis of phenyllactate synthesis (c), N-acetylated amino acids (d), and histidine degradation pathway (e). The relative intensities of metabolites are shown as floating bar graphs. The boxes represent the median and interquartile range and whiskers range from minimum to maximum values. A plus sign (+) on plots indicates mean values. Schematics of glucose catabolism (b) and phenyllactate synthesis pathways (c) are also depicted. Colored circles indicate enriched or depleted metabolites within the pathway. Significant differences between mean intensities of metabolites among infections were determined using Welch’s two-sample t test with FDR correction. p and q values are indicated by ^(p < 0.05, q > 0.01), *(p < 0.05, q < 0.01), **(p < 0.01, q < 0.01), ***(p < 0.001, q < 0.01), ****(p < 0.0001, q < 0.01).

Fig. 6. BVAB induce production of polyamines in a species-specific manner in a human 3D cervical model.

P. bivia and S. amnii contribute to agmatine, spermidine, and N(1)-acetylspermine production, whereas A. vaginae does not induce production of bioamines similarly to L. crispatus. a The heatmap shows fold changes of the metabolites related to the polyamine pathway among bacterial infections compared to uninfected controls. Only the fold changes that were significant (p < 0.05) are displayed. Gold- and purple-shaded squares indicate metabolite enrichment or depletion, respectively. b A schematic of polyamine metabolism pathways. Colored circles indicate enriched or depleted metabolites following monoinfection with specific BVAB or infection with the polymicrobial cocktail. c Floating bar graphs shows relative intensity of polyamines in 3D cell culture supernatants following bacterial colonization. The relative intensities of polyamines and related metabolites are shown as floating bar graphs. The boxes represent the median and interquartile range and whiskers range from minimum to maximum values. A plus sign (+) on plots indicates mean values. Significant differences between mean intensities of metabolites among infections were determined using Welch’s two-sample t test with FDR correction. p and q values are indicated by ^(p < 0.05, q > 0.01), *(p < 0.05, q < 0.01), **(p < 0.01, q < 0.01), ***(p < 0.001, q < 0.01).

Fig. 7. BVAB species distinctively contribute to inflammation and physicochemical attributes of epithelial barrier in a human 3D cervical model.

A. vaginae and S. amnii impact arginine and citrulline metabolism, which leads proinflammatory signaling via nitric oxide production (a). S. amnii also induce production of oxidative stress-related metabolites (b). S. amnii induce pipecolate associated with the presence of “clue cells” (c). P. bivia can contribute to elevated cervicovaginal pH levels via asparagine degradation and ammonia production (d). BVAB species also can impact physicochemical attributes of epithelial barrier (mucin degradation and collagen remodeling) (e, f). a A schematic of arginine/citrulline metabolism pathways. Colored circles indicate enriched or depleted metabolites following monoinfection with specific BVAB or infection with the polymicrobial cocktail. a–f Floating bar graphs shows relative intensity of polyamines in 3D cell culture supernatants following bacterial colonization. The relative intensities of metabolites are shown as floating bar graphs. The boxes represent the median and interquartile range and whiskers range from minimum to maximum values. A plus sign (+) on plots indicates mean values. Significant differences between mean intensities of metabolites among infections were corrected for multiple comparisons. p and q values are indicated by ^(p < 0.05, q > 0.01), *(p < 0.05, q < 0.01), **(p < 0.01, q < 0.01), ***(p < 0.001, q < 0.01), ****(p < 0.0001, q < 0.01).

Fig. 8. Summary of immunometabolic contributions of Lactobacillus and BVAB species in the cervical microenvironment.

L. crispatus reinforces protective microenvironment via antimicrobial metabolites, including PLA. G. vaginalis and P. bivia alter physiochemical barrier-related proteins and metabolites (mucins, sialic acid, polyamines). A. vaginae and S. amnii induce robust proinflammatory and pro-oncogenic responses through induction of cytokines, iNOS, and oxidative stress. Polymicrobial infection leads to the most robust cervical immunometabolic activity. These mechanistic insights on immunometabolic landscape provide better understanding of BV pathogenesis and connect BV with adverse women’s health outcomes.