Characterization of the Ovine Vaginal Microbiome and Inflammation Patterns as an Improved Testing Model of Human Vaginal Irritation

Abstract

The development of therapies targeted to improve the health of women has utilized direct vaginal delivery as a more effective and less toxic method of protection from HIV and other pathogens. Vaginal applicants and delivery devices that provide sustained effects have been met with increasing acceptability, but the efficacy and toxicity outcomes have not been successfully predicted by preclinical in vitro studies and animal modeling. We have explored the utilization of sheep as a model for testing the safety of vaginal applicants and devices based on spatial and structural similarities to the human vagina. As recently noted by the FDA, an additional safety measure is an impact on the vaginal microbiome (VMB) that is known to contribute to vaginal health and influence pathogen susceptibility and drug metabolism. To advance the utility of the sheep vaginal model, we completed a thorough molecular characterization of the ovine VMB utilizing both next-generation sequencing (NGS) and PCR methods. The process also created a custom PCR array to quantify ovine VMB community profiles in an affordable, higher throughput fashion. The results from vaginal swabs (>475 samples) collected from non-pregnant crossbred Dorset and Merino ewes treated with selected vaginal applicants or collected as sham samples established 16 VMB community types (VMB CTs). To associate VMB CTs with eubiosis or dysbiosis, we also completed custom ELISAs for six cytokines identifying IL1B, IL8, TNFa, and CXCL10 as useful markers to support the characterization of ovine vaginal inflammation. The results indicated that Pasteurella, Actinobacillus, Pseudomonas, Bacteroides, Leptotrichia, and E. coli were common markers of eubiosis (low inflammatory marker expression), and that Haemophilus, Ureaplasma, and Corynebacterium were associated with dysbiosis (high cytokine levels). Utilizing the optimized workflow, we also confirmed the utility of three commonly used vaginal applicants for impact on the VMB and inflammatory state, producing a dataset that supports the recommendation for the use of sheep for testing of vaginal applicants and devices as part of preclinical pipelines.

Keywords: cytokines; dysbiosis; inflammation; intravaginal drug delivery; ovine sheep vaginal model; toxicity; vaginal microbiome; women's health.

Figure 1

Overall study design for the three phases of the study. The iterative approach, completed over 8 years, first established collection and analysis methods (phase 1) in two breeds of sheep housed under different conditions. The optimal methods were then exported to a second research team at a second site (phase 2), allowing for refinement of techniques and confirmation of sample shipment methods before completing an expanded evaluation (phase 3) of irritants selected to impact the ovine vaginal environment with the expectation of shifted vaginal microbiome (VMB) and inflammation development to identify the most common microbiota changes and their association with inflammatory markers.

Figure 2

Average log₁₀ fold change from baseline levels of sheep vaginal cell counts (shGAPDH). The impact of Sham (top left), hydroxyethylcellulose (HEC) placebo gel (top right), nonoxynol-9 (N9) gel (lower left), or benzalkonium chloride (BZK) solution (lower right) applied daily for 10 days to the sheep vaginal cavity was assessed by collection of vaginal swabs during phase 3 of the study. Each swab was processed for shGAPDH loads by quantitative polymerase chain reaction (qPCR) that were log₁₀transformed and compared to the individual day 1 (D1) pre-dose baseline for each animal. The fold change was then averaged by the group for all study days in both the treatment (main, D1–D10) and recovery (D11–D21) periods of the study, and plotted with standard deviations. The maximal difference from baseline for the main and recovery periods are shown as individual values on the right side of each graph for the three treatment groups where recovery period samples were collected. Daily and maximum (main and recovery periods) log₁₀ fold change in shGAPDH levels were tested for normality (e.g., Shapiro–Wilk test). Daily and maximum (main and recovery periods) log₁₀ fold change in shGAPDH levels were compared for treatment effects by either parametric (e.g., ANOVA) or non-parametric (e.g., Wilcoxon) methods. Pairwise comparisons between groups were adjusted (Tukey or Dunnett) for multiple comparisons. * = p < 0.05. Non-transformed average daily values are presented in Supplementary Figure 1.

Figure 3

Average log₁₀ fold change from baseline levels of vaginal bacteria genome counts(16S). Similar analyses were completed on the same swab samples to determine the impact of Sham (top left), HEC placebo gel (top right), N9 gel (lower left), or BZK solution (lower right) applied as described for shGAPDH evaluations. Each plotted point is the averaged fold change with standard deviations. The maximal difference from baseline for the main and recovery periods are again shown as individual values on the right side of each graph for the three treatment groups where recovery period samples were collected. Comparisons were as described for the shGAPDH analyses in Figure 2. Statistical differences from group baselines are shown as * = p < 0.05; ** = p < 0.01; *** = p < 0.0001 and are more fully described in the results. Pairwise differences between groups were adjusted (Tukey's or Dunnett's test) for multiple comparisons, and p < 0.05 findings are indicated with the following comparisons showing statistically significant differences: BZK vs. Sham for D2, 3, 5, 7, and 9; BZK vs. HEC for D5, 7, 9, 11, 13, 15, and 17; BZK vs. N9 for D2; N9 vs. HEC for D17. Non-transformed average daily values are presented in Supplementary Figure 1.

Figure 4

Cluster analysis of the ovine VMB in naïve and Sham-treated animal samples. Heat map clustering [unsupervised using Morpheus online (43)] illustrated dominance by individual or combinations of bacteria as well as grouping of samples into community type (CT) categories indicated by branching shown across the top of the Figure. Proportional levels of the bacteria (right vertical labels) quantified by the VMBA are shown in rows with each sample (n = 71 distinct sheep) shown as a distinct column. Red boxes represent the maximal levels detected in a row, and blue shows the lowest detection levels. Clustering helped to identify CTs that were dominated by biomarker organisms that were confirmed by statistical comparisons, leading to the categorical CT system shown in Figure 5.

Figure 5

Proportional bar charts represent the average VMB community structure for each of the identified CTs and their inflammatory association. Top: The 10 identified biomarker organisms were proportionally charted to illustrate the average profiles from the 216 vaginal swab samples evaluated from phase 3 of the study. The number of representatives is shown below each bar. Bottom: Using the assigned CT, each of the individual swab samples was grouped (one sample was not typable by this system) and designated by one of several models of inflammation as described in the results. Using enzyme-linked immunosorbent assay (ELISA) data (Figure 6), we considered binary and tertile categorization schemes derived from all the samples regardless of CT for the four commonly detected targets. Using the binary distinctions of high (H) or low (L), assignments were made when at least 67% of the samples fell in one category. A final call of eubiotic (E) or dysbiotic (D) was made when at least three of the four targets showed consistency.

Figure 6

Scatter plots of the cytokine levels for each of the CT. The ELISA results for each phase 3 sample were grouped by assigned CT and then scatter-plotted. The bar shows average levels of the group with standard deviations shown as bar and whiskers. The tertile divisions (high, medium, and low were determined from all samples in the study) are shown as horizontal red lines, and binary divisions (high and low) are shown by the blue horizontal line. IL-8 outcomes are shown in the top left panel, IL-1b top right, TNF-a lower left, and CXCL10 lower right.