Genome-scale metabolic network reconstruction analysis identifies bacterial vaginosis-associated metabolic interactions

Abstract

Bacterial vaginosis (BV) is the most prevalent vaginal condition among reproductive-age women presenting with vaginal complaints. Despite its significant impact on women's health, limited knowledge exists regarding the microbial community composition and metabolic interactions associated with BV. In this study, we analyze metagenomic data obtained from human vaginal swabs to generate in silico predictions of BV-associated bacterial metabolic interactions via genome-scale metabolic network reconstructions (GENREs). While most efforts to characterize symptomatic BV (and thus guide therapeutic intervention by identifying responders and non-responders to treatment) are based on genomic profiling, our in silico simulations reveal functional metabolic relatedness between species as quite distinct from genetic relatedness. We grow several of the most common co-occurring bacteria (Prevotella amnii, Prevotella buccalis, Hoylesella timonensis, Lactobacillus iners, Fannyhessea vaginae, and Aerrococcus christenssii) on the spent media of Gardnerella species and perform metabolomics to identify potential mechanisms of metabolic interaction. Through these analyses, we identify BV-associated bacteria that produce caffeate, a compound implicated in estrogen receptor binding, when grown in the spent media of other BV-associated bacteria. These findings underscore the complex and diverse nature of BV-associated bacterial community structures and several of these mechanisms are of potential significance in understanding host-microbiome relationships.

Fig. 1. Differential community composition of symptomatic vs asymptomatic BV.

A heatmap illustrating relative frequency of Gardnerella vaginalis co-occurrence with non-Gardnerella species in BV symptomatic, BV asymptomatic, and healthy samples.

Fig. 2. Pairwise in silico bacterial interactions.

Heatmap illustrating primary bacteria biomass flux change in the presence of co-occurring bacteria when simulating (A) mutualism, (B) competition, (C) net interactions; t-SNE plots of primary bacteria biomass flux change reduced across all co-occurring bacteria when simulating (D) mutualism, (E) competition, (F) net interactions.

Fig. 3. Relative competition versus mutualism and the metabolites that underly these interactions.

A average biomass increase due to mutualism benefit vs. average biomass decrease due to competition cost for each bacteria (B) dendrogram of Gardnerella species, with light stars denoting top four most mutualistic bacteria and dark stars indicating top four most competitive bacteria across all interactions simulations. C Most highly competed for (competition) and/or shared (mutualism) metabolites across all interaction simulations. D G. piotii growth curves in NYC III enriched media (gray) and NYC III enriched media supplemented with 0.1% L-histidine (black), with mean and ½ standard deviation spread (one-tailed t-test, t = 3.487, p-value < 0.01).

Fig. 4. Pairwise in vitro bacterial growth interactions.

A–H growth curves in NYC III enriched media (black), spend G. piotii media (light pink), and spent G. vaginalis media (dark pink/red), with mean and two standard deviation spread. G. piotii and A. christensenii have highly flocculent phenotypes.

Fig. 5. Pairwise in vitro bacteria metabolic interactions.

A–E Volcano plots of differential metabolites (blue: consumed; green: produced; gray: not significant (two sided t-test, Bonferroni multiple hypothesis correction) from co-occurring species grown in G. vaginalis spent media (F) Caffeic acid production of F. vaginae grown in G. piotii spent media versus NYC III (n = 6 biological replicates, p-value = 0.07 (one-sided t-test) G. piotii Spent Media boxplot: minimum = 0.022, maximum = 0.539, median = 0.104, 25th percentile = 0.024, 75th percentile = 0.184, upper whisker = 0.184, lower whisker = 0.022, interquartile range = 0.16. NYC III Media boxplot: minimum = 0.016, maximum = 0.048, median = 0.0185, 25th percentile = 0.018, 75th percentile = 0.027, upper whisker = 0.027, lower whisker = 0.027, interquartile range = 0.009.