An In Silico Framework for the Rational Design of Vaginal Probiotic Therapy

Introduction: The vaginal microbiome (VMB) is a crucial determinant to women’s reproductive health. A community predominantly composed of Lactobacillus species is widely acknowledged as a healthy state. However, a transition to a more diverse composition, including anaerobic bacteria such as Gardnerella vaginalis often characterizes bacterial vaginosis (BV), a prevalent condition affecting approximately 30% of reproductive-age women globally. BV is linked to adverse health consequences, including heightened susceptibility to sexually transmitted infections, pelvic inflammatory disease, and preterm birth. Standard antibiotic regimens such as metronidazole stand as a common first-line treatment offering temporary relief; however, its long-term effectiveness is constrained by its inability to restore the dominance of Lactobacillus species. An alternative treatment approach gaining popularity involves the utilization of probiotics, which exhibit potential in reinstating the optimal composition of the vaginal microbiome, however, the efficacy varies depending on the strain, route of administration, and dosing strategies. In this study, we developed a model that can systematically test probiotic characteristics and dosing strategies against a variety of in silico BV communities. Lastly, the modeling framework was evaluated in the context of regimen reported for Lactin-V in a phase 2b clinical trial.

Materials and

Methods: A generalized Lotka-Volterra model (gLVM) with four equations was used to predict the population changes in non-optimal anaerobic bacteria (nAB), Lactobacillus inners (Li), optimal lactobacillus (oLB), and probiotics (P) species. This model included parameters such as growth rate, self-interaction, and inter-species interaction terms. To test the impact of probiotic strains on heterogenous populations, a virtual patient population was generated using Latin Hypercube Sampling (LHS) of a physiologically relevant parameter space, and probiotic strains were tested on 2000 virtual BV+ patients. A sensitivity analysis was used to determine probiotic strain characteristics that promoted the return to an optimal VMB state. Two main probiotic regimens were evaluated: a short-term probiotic therapy with and without antibiotic pre-treatment, and a long-term therapy with antibiotic treatment modeled similar to a regimen used in a previous study of Lactin-V. The impact of different probiotic strains was evaluated at multiple time points by classifying the community state types (CSTs) after treatment using a nearest centroid classifier. Moreover, for both short-term and Lactin-V regimens, the association between probiotic strain characteristics and the response classification was assessed using a PLS-DA model. All model simulations and statistical analyses including Chi-square tests to compare frequencies of response types, and Wilcoxon rank sum tests to compare numerical data, were completed in MATLAB.

Results, Conclusions, and Discussions: The results of sensitivity analysis suggested that the effects of nAB species on the growth of probiotic species (αnAB->P) was a highly sensitive parameter for probiotic efficacy. The model suggested that this parameter was more sensitive than interactions of probiotics on nAB (αP->nAB), which is traditionally used as probiotic selection criteria. The sensitivity analysis also revealed that while probiotic interactions with endogenous Lactobacillus species (αP->Li, αP->oLB, αLi->P, αoLB->P) did not influence efficacy, they did determine which type of Lactobacillus species dominated in the community assembled post-treatment. To illustrate these ideas more concretely, we compared the efficacy of a traditional probiotic (Figure 1; “Traditional”; αP->nAB = -0.01 density-1d-1) to a probiotic that was positively influenced by nAB species (Figure 1; “1p: Best”; αnAB->P = +0.01 densitiy-1d-1) and different combinations of the two on efficacy in a heterogeneous in silico BV+ population (Figure 1). Results showed that the traditional strain had a higher rate of treatment failure than a strain with a positive αnAB->P (Figure 1; 31.9% vs. 14.4%; p< 1x10-6). Additional simulations illustrated how treatment failure could be further reduced by combinatorial changes of multiple parameters (Figure 1; “2p” & “3p”). Model results were also able to replicate results from a recent phase 2b clinical trial for Lactin-V, illustrating how Lactin-V requires adjunctive antibiotic therapy to be effective, and that increasing the dosing frequency of the probiotic could have a moderate impact on BV recurrence at 12 and 24 weeks post-treatment.

Overall, this work provides a new framework to characterize and test different probiotic strains and regimen strategies across heterogeneous VMB communities in different women. It will aid in the future rational selection of probiotic strains and intelligent design of dosing regimens for BV treatment.

Acknowledgements (Optional):