Integrating epidemiologic and molecular methods to improve vaginal health

Challenges measuring the microbiome and limited incorporation of epidemiologic methods in microbiome science have hindered the progress of vaginal microbiome research, and major questions in the field remain unanswered despite decades of work. This dissertation seeks to address these gaps by describing measurement error and resulting bias in vaginal microbiota research; evaluating the state of epidemiologic evidence regarding the role of Lactobacillus iners, a controversial vaginal bacterial species, in various sexual health outcomes; and investigating bacterial drivers of BV symptomatology. The most popular method for characterizing the vaginal microbiota is 16S rRNA gene amplicon sequencing, which provides information on the taxonomic composition of a bacterial community. Shotgun metagenome sequencing characterizes the bacterial genes in a sample and provides information on the community’s functional potential, which is more relevant to understanding mechanisms and causal relationships between the microbiome and health outcomes than taxonomic composition. Metagenome inference methods attempt to bridge the gap between 16S rRNA gene amplicon sequencing and shotgun metagenomics by predicting a bacterial community’s metagenome based on its taxonomic composition and annotated whole genome sequences of its members. Several studies have compared metagenome inference performance in different human body sites; however, none specifically reported on the vaginal microbiome. In Chapter 2, we compared the performance of two popular metagenome inference methods, PICRUSt2 and Tax4Fun2, using paired 16S rRNA gene amplicon sequencing and shotgun metagenome sequencing data from vaginal samples from 72 pregnant individuals enrolled in the Pregnancy, Infection, and Nutrition cohort. PICRUSt2 and Tax4Fun2 performed modestly overall (median Spearman correlations between observed and predicted KEGG ortholog [KO] relative abundances 0.20 and 0.22, respectively). Both methods performed best among Lactobacillus crispatus-dominated vaginal microbiotas (median Spearman correlations 0.24 and 0.25, respectively) and worst among L. iners-dominated microbiotas (median Spearman correlations 0.06 and 0.11, respectively). Differential metagenome inference performance across vaginal microbiota community types can be considered differential measurement error, which often results in differential misclassification. As such, metagenome inference will introduce hard-to-predict bias in vaginal microbiome research. These findings demonstrate the importance of considering common epidemiologic biases in designing and evaluating novel microbiome analytic methods. The vaginal microbiota of reproductive-age individuals is often dominated by a single Lactobacillus species, most commonly L. iners or L. crispatus. While L. crispatus-dominated vaginal microbiotas are widely thought to protect against adverse sexual health outcomes, the role of L. iners-dominated microbiotas is less clear and has been debated. To better understand the role of L. iners, in Chapter 3 we conducted systematic reviews of the associations between L. iners compared to L. crispatus and the following outcomes: bacterial vaginosis (BV); Chlamydia trachomatis, Neisseria gonorrhoeae, Trichomonas vaginalis, human papillomavirus (HPV), HIV, and genital herpes simplex virus type-2 (HSV-2) infections; and cervical dysplasia. We searched four databases for epidemiologic studies of reproductive-age, nonpregnant, cisgender women that used marker gene sequencing to characterize vaginal microbiota composition and presented an effect estimate for the association between L. iners compared to L. crispatus and outcomes of interest. For outcomes with >3 eligible studies presenting the same form of effect estimate, we conducted random-effects meta-analysis. Three BV studies were included in meta-analysis, which indicated L. iners-dominated microbiotas were associated with 2.1-fold higher prevalence of BV compared to L. crispatus-dominated microbiotas (95% CI 0.9-4.9). Six C. trachomatis studies were included in meta-analysis, which showed L. iners-dominated microbiotas were associated with 3.4-fold higher odds of chlamydia compared to L. crispatus-dominated microbiotas (95% CI 2.1-5.4). L. iners-dominated vaginal microbiotas may be suboptimal compared to L. crispatus-dominated microbiotas for BV and chlamydia, which is consistent with prior in vitro, in silico, and genomic work. Evidence was sparse for other outcomes. Nearly all included studies assessed microbiota composition and outcome status cross-sectionally and were at serious risk of bias, critically limiting the quality of evidence reviewed. In contrast to Lactobacillus-dominated microbiotas, BV is a polymicrobial condition characterized by a diverse community of anaerobic and facultative bacteria. It is the most common cause of vaginal discharge worldwide, affecting approximately one quarter of reproductive-age individuals. In high-resource settings, clinical BV diagnosis is typically based on the presence of at least three of four signs and symptoms termed Amsel criteria: amine odor on addition of potassium hydroxide to vaginal fluid; clue cells on vaginal wet prep; thin, gray, homogeneous vaginal discharge; and elevated vaginal pH. Because bacterial colonization and associations with these signs and symptoms may vary between populations, in Chapter 4 we assessed relationships between vaginal bacteria and Amsel criteria among two distinct populations. We included Kenyan participants from the placebo arm of the Preventing Vaginal Infections (PVI) trial and participants from a Seattle-based cross-sectional BV study in this analysis. Amsel criteria were recorded at study visits, and the vaginal microbiota was characterized using 16S rRNA gene amplicon sequencing. We fit logistic regression models to evaluate associations between bacterial relative abundance and each Amsel criterion. BV-associated bacterium 1 (BVAB1) was positively associated with all Amsel criteria in both populations. Eggerthella type 1, Fannyhessea (Atopobium) vaginae, Gardnerella spp., Sneathia amnii, and Sneathia sanguinegens were positively associated with all Amsel criteria in the Seattle study, and all but discharge in the PVI trial. This core group of vaginal bacteria may play a key role in the manifestation of BV signs and symptoms across diverse populations, and these findings are consistent with growing evidence regarding the role of biogenic amines and extracellular enzymes in BV etiology and symptom manifestation. Finally, in Chapter 5 I discuss the implications of this dissertation work for microbiome science, medicine, and public health. I recommend avenues by which investigators can better incorporate epidemiologic methods and principles into their work, I provide a novel characterization of L. iners, and I weigh various strategies to improve BV diagnostics and treatment in high- and low-resource settings.