Background Dysbiosis of vaginal bacterial communities have been associated with increased risk for sexually transmitted infections and bacterial vaginosis. This is the first observational study to model temporal dynamics of vaginal microbiota using frequently collected samples, behavioural data and culture-independent methods.

Methods Thirty-three asymptomatic, reproductive-age women self-collected mid-vaginal swabs every 3rd day for 16 weeks (998 samples). Participants reported behaviours and menstrual data on daily diaries. Bacterial communities were determined by pyrosequencing of barcoded 16S rRNA genes (V1–V2 region). Participants were clustered into five community classes based on temporal patterns of vaginal bacterial community composition using transition probabilities. A linear mixed effect model for the log of Jensen-Shannon rate of community change was utilised. The model accounted for correlations between samples from the same participant and was adjusted for time-varying confounders and normalised menstrual cycle time.

Results Three of the community classes were most often dominated by Lactobacillus iners, L crispatus, or L gasseri, respectively, while two lacked significant numbers of Lactobacillus spp. The latter classes were split into subtype A typified by Corynebacterium, Anaerococcus, Peptinophilus, Prevotella, and Finegoldia, while those of subtype B showed a higher abundance of the genus Atopobium. The rank abundance and species composition of bacterial communities in some women changed markedly over short periods of time while others were relatively stable. Classes dominated by L crispatus and L gasseri experienced the fewest fluctuations in community composition, and communities that lacked significant number of Lactobacillus spp. also demonstrated some stability. Vaginal communities dominated by L iners demonstrated either a lack of constancy or notable stability. The menstrual cycle was associated with temporal dynamics, but these effects were influenced by bacterial community class. Sexual activity the day prior to sampling was of borderline statistical significance (p=0.065) and is a variable of interest in supplementary modelling.

Conclusions Vaginal microbiota can fluctuate rapidly. Future studies should investigate the role of temporal changes in vaginal microbiota on sexually transmitted infection risk. Longitudinal studies of the vaginal microbiome may allow for the future development of targeted individualised therapeutic approaches.