A multi-platform metabolomics approach identifies highly specific biomarkers of bacterial diversity in the vagina of pregnant and non-pregnant women

Bacterial vaginosis (BV) increases transmission of HIV, enhances the risk of preterm labour, and is associated with malodour. Clinical diagnosis often relies on microscopy, which may not reflect the microbiota composition accurately. We use an untargeted metabolomics approach, whereby we normalize the weight of samples prior to analysis, to obtained precise measurements of metabolites in vaginal fluid. We identify biomarkers for BV with high sensitivity and specificity (AUC = 0.99) in a cohort of 131 pregnant and non-pregnant Rwandan women, and demonstrate that the vaginal metabolome is strongly associated with bacterial diversity. Metabolites associated with high diversity and clinical BV include 2-hydroxyisovalerate and γ-hydroxybutyrate (GHB), but not succinate, which is produced by both Lactobacillus crispatus and BV-associated anaerobes in vitro. Biomarkers associated with high diversity and clinical BV are independent of pregnancy status, and were validated in a blinded replication cohort from Tanzania (n = 45), where we predicted clinical BV with 91% accuracy. Correlations between the metabolome and microbiota identified Gardnerella vaginalis as a putative producer of GHB, and we demonstrate production by this species in vitro. This work illustrates how changes in community structure alter the chemical composition of the vagina, and identifies highly specific biomarkers for a common condition.