Abstract
The vaginal microbiota is known to affect women's health. Yet, there is a notable paucity of highresolution follow-up studies lasting several months, which would be required to interrogate the longterm dynamics and associations with demographic and behavioural covariates. Here, we present a high-resolution longitudinal cohort study of 125 women, followed for a median duration of 8.6 months, with a median of 11 samples collected per woman. Using a hierarchical Bayesian Markov model, we characterised the patterns of vaginal microbiota community persistence and transition, simultaneously estimated the impact of 16 covariates and quantified individual variability among women. We showed that "optimal" (Community State Type (CST) I, II, and V) and "sub-optimal" (CST III) communities are more stable over time than "non-optimal" (CST IV) ones. Furthermore, we found that some covariates -most notably alcohol consumption -impacted the probability of shifting from one CST to another. We performed counterfactual simulations to confirm that alterations of key covariates, such as alcohol consumption, could shape the prevalence of different microbiota communities in the population. Finally, our analyses indicated that there is a relatively canalised pathway leading to the deterioration of vaginal microbiota communities, whereas the paths to recovery can be highly individualised among women. In addition to providing one of the first insights into vaginal microbiota dynamics over a year, our study showcases a novel application of a hierarchical Bayesian Markov model to clinical cohort data with many covariates. Our findings pave the way for an improved mechanistic understanding of microbial dynamics in the vaginal environment and the development of novel preventative and therapeutic strategies to improve vaginal health.