The vaginal microbiome has long been recognized as a player in reproductive health. A community dominated by Lactobacillus species, particularly Lactobacillus crispatus, is generally associated with better outcomes, while overgrowth of diverse anaerobic bacteria marks a state called bacterial vaginosis that raises risks of preterm birth and infection. But during assisted reproductive treatment, the picture becomes more complicated. Hormonal stimulation, antibiotics, and the physical interventions of egg retrieval and embryo transfer all reshape the microbial landscape in ways that standard sequencing cannot fully capture. Most studies to date have relied on species-level resolution, which treats all members of a species as functionally equivalent and leaves the within-species differences unexplored.

To push past that limitation, Peigen Chen, Hao Shi, and their colleagues conducted longitudinal shotgun metagenomic sequencing of vaginal samples collected from 41 women undergoing IVF or intracytoplasmic sperm injection. The samples spanned six timepoints, from a pre-stimulation baseline through ovarian stimulation, egg retrieval, embryo transfer, and finally pregnancy testing, yielding 186 samples in total. Rather than simply cataloging which species were present, the team used a strain-level analysis tool called inStrain to reconstruct the genomic sequences of individual bacterial strains within each sample, tracking how those strains changed over the course of treatment.

What emerged was a striking reciprocal pattern involving two Lactobacillus species that often dominate the vaginal environment. Lactobacillus crispatus, generally considered the more favorable species, showed a pattern of genomic stability in women who achieved pregnancy: the strains persisted with relatively few mutations accumulating over time, suggesting a robust and resilient presence. Lactobacillus iners, a more enigmatic species that has been associated with both health and dysbiosis, displayed a different behavior in successful pregnancies. Rather than remaining stable, L. iners strains showed adaptive flexibility, accumulating mutations in ways that suggested the bacteria were actively responding to the shifting conditions of the reproductive tract under hormonal stress.

The team then turned to functional protein analysis to identify which bacterial proteins might connect these strain-level dynamics to pregnancy outcomes. Two targets stood out. The first was Inerolysin, a cholesterol-dependent cytolysin produced by L. iners. Cholesterol-dependent cytolysins are a family of bacterial toxins that bind cholesterol in host cell membranes and punch holes in them, and Inerolysin’s role in the vaginal environment has been debated. The second was L-lactate dehydrogenase, or L-LDH, an enzyme from L. crispatus that catalyzes the production of lactate, a molecule thought to help maintain the acidic environment favorable to reproductive health.

For each protein, the team identified specific mutations that distinguished women who achieved pregnancy from those who did not. In Inerolysin, a Success-specific mutation at position 429 changed the amino acid serine to proline. Structural modeling suggested that this S429P mutation is predicted to disrupt the protein’s cholesterol-binding capacity, potentially attenuating its cytolytic activity. In other words, women whose L. iners strains carried this mutation appeared to harbor a less damaging version of the toxin. In L-LDH, a mutation at position 169 changed proline to leucine, and computational predictions suggested this P169L change may enhance the enzyme’s catalytic efficiency, potentially boosting lactate production.

To test whether these structural interpretations were robust, the team applied four orthogonal computational methods. FoldX thermodynamic calculations estimated the impact of each mutation on protein stability, ESM-1v evolutionary scoring assessed how likely each variant was to be tolerated based on patterns across related proteins, ConSurf conservation analysis determined whether the mutated positions sat in functionally constrained regions, and 10-nanosecond molecular dynamics simulations tracked how the mutated proteins behaved in silico. All four methods provided convergent evidence supporting the structural interpretations, lending weight to the idea that these mutations genuinely alter protein function rather than representing neutral genetic drift.

Perhaps the most consequential finding concerned timing. The protective variants were enriched at baseline, before any hormonal stimulation had begun, rather than arising during treatment. This indicates that pre-existing strain genetics predispose women to differential responses under the hormonal stress of IVF, rather than the microbiome adapting favorably in real time. In practical terms, the bacterial strains a woman carries before treatment starts may already determine how her vaginal environment responds to the interventions that follow.

Building on this insight, the team constructed a baseline prediction panel using 18 single-nucleotide variants. When tested with internal cross-validation, the panel achieved an area under the curve of 0.945, indicating strong discriminatory power. In an external validation cohort of 69 women sequenced on a multiplex amplicon PCR platform, the panel achieved an AUC of 0.784, a more modest but still meaningful performance that suggests the approach could generalize beyond the original study population. Such a panel could eventually serve as a biomarker for personalized risk assessment before treatment begins.

The findings open several avenues for future research. The reciprocal dynamics of L. crispatus stability and L. iners adaptive flexibility, together with the dual-protein mutation pattern, suggest targets for strain-specific probiotic interventions tailored to individual microbiome profiles. Rather than prescribing generic Lactobacillus supplements, clinicians might one day select specific strains engineered or selected for favorable variants of proteins like Inerolysin and L-LDH. The team emphasizes that all computational predictions require experimental validation through biochemical and functional assays before such interventions become reality, but the study provides a detailed mechanistic framework for understanding how the vaginal microbiome shapes reproductive success at a resolution far beyond what species-level analysis can offer.

Subject of Research: Longitudinal strain-level metagenomics reveals reciprocal Lactobacillus dynamics and outcome-associated protein variants during IVF treatment

Article Title: Longitudinal strain-level metagenomics reveals reciprocal Lactobacillus dynamics and outcome-associated protein variants during IVF treatment

Article References: Chen, P., Shi, H., Liu, Z., Zhao, X., Zhu, Y., Chen, W., Liang, X., & Yang, X. (2026). Longitudinal strain-level metagenomics reveals reciprocal Lactobacillus dynamics and outcome-associated protein variants during IVF treatment. Microbiome. https://doi.org/10.1186/s40168-026-02560-0

Image Credits: AI Generated

DOI: 10.1186/s40168-026-02560-0

Keywords: Longitudinal, strain-level, metagenomics, reveals, reciprocal, Lactobacillus, dynamics, outcome-associated, protein, variants, during, treatment