Oral microbiome signatures of post-stroke cognitive impairment

To investigate the characteristics of oral microbiota in patients with post-stroke cognitive impairment (PSCI) and to evaluate its potential as a noninvasive candidate microbial signature associated with PSCI.

This cross-sectional study enrolled 108 participants, including 40 PSCI patients, 40 post-stroke patients with normal cognition (PSNC), and 28 healthy controls (HC). Saliva samples were collected for 16S rRNA sequencing. Diversity analyses, differential taxa identification, and correlation analyses were performed. Candidate microbial signatures were screened using the least absolute shrinkage and selection operator (LASSO) regression and Random Forest (RF), followed by the construction of multiple machine learning models.

Compared with PSNC and HC, patients with PSCI exhibited significantly reduced richness and diversity of the oral microbiota, and beta-diversity analyses suggested group-level differences in community composition. The relative abundances of gram-negative taxa, such as Proteobacteria, Campylobacterota, Gammaproteobacteria, Pseudomonadales, and Alloprevotella were increased in PSCI samples. In contrast, commensal taxa Leptotrichia, and Veillonella were markedly decreased. Significant associations were observed between differential taxa and cognitive scores (MMSE and MoCA). Three key microbial features were ultimately identified. Models based on these features showed favorable exploratory internal discriminatory performance between PSCI and PSNC within this dataset. In the test set, the RF model achieved an AUC of 0.979 and an Average Precision of 0.983. However, these findings should be interpreted cautiously and require validation in larger independent cohorts.

Patients with PSCI present with notable oral microbiota dysbiosis, characterized by depletion of commensal taxa and enrichment of Gram-negative bacteria. Oral microbiota-based models showed favorable exploratory internal discriminatory performance within this dataset, but these findings remain hypothesis-generating and require external validation in larger independent cohorts.