Altered gut microbiota composition and function potential in polycystic ovary syndrome: a multicohort metagenome-assembled genomes study.

Background

Polycystic ovary syndrome (PCOS) has been associated with gut microbial dysbiosis, but cross-cohort reproducibility and genome-resolved alterations remain insufficiently characterized.

Methods

We integrated fecal shotgun metagenomic data from three independent PCOS cohorts comprising 169 participants. Taxonomic and functional profiles were generated using MetaPhlAn4 and HUMAnN, while de novo assembly and binning were used to reconstruct species-level genome bins (SGBs). Linear mixed models accounted for cohort heterogeneity. Phylogenetic, co-occurrence network, KEGG-based functional, and prevalence-bias analyses were performed. PCOS classification was evaluated across genus, species, and SGB layers using leakage-safe nested cross-validation and leave-one-cohort-out testing.

Results

PCOS was associated with reduced species richness, contraction of the core microbiota, altered community structure, and lower microbial network connectivity in the two larger cohorts. Eleven species showed differential abundance, including enrichment of Phocaeicola vulgatus and Bacteroides uniformis in PCOS. Genome-resolved analysis recovered 540 dereplicated SGBs, including 32 putative novel species. SGB prevalence bias was phylogenetically structured and remained associated with distinct KEGG Orthology profiles after adjustment for genome length and completeness. PCOS-biased SGBs were enriched in phosphotransferase-system and carbohydrate-utilization functions, whereas Healthy-biased SGBs were enriched in carbon and amino-acid metabolic pathways. Species-level XGBoost achieved the highest internal discrimination (AUC 0.757), but leave-one-cohort-out performance declined substantially, indicating limited cross-cohort generalizability.

Conclusions

PCOS is associated with multi-resolution gut microbiome alterations, with several community-level features showing cross-cohort reproducibility. These findings extend from community structure to genome-resolved functional programs, refine candidate microbial biomarkers, and highlight the need for external validation before clinical translation.