Stage-associated mucosal microbiome remodeling and host epigenetic alterations in gastric cardia precancerous lesions: a cross-sectional multi-omics study

Gastric cardia adenocarcinoma (GCA) shows marked geographic clustering, but microbial changes in its precursor lesions remain poorly defined. We characterized the mucosa-associated microbiome across normal gastric cardia mucosa, intestinal metaplasia (IM), and intraepithelial neoplasia (IEN), and examined its association with host epigenetic alterations. In this cross-sectional study, 16S rRNA gene sequencing was performed on 71 gastric cardia mucosal samples, including 17 Normal, 44 IM, and 10 IEN samples. Microbiome data were integrated with previously published DNA methylation and RNA-seq profiles from matched samples. Helicobacter pylori colonization was assessed by immunohistochemistry (IHC) in an independent expanded cohort with no overlapping participants. Alpha diversity did not differ significantly among histological groups in unadjusted analyses, whereas microbial community composition differed among histological groups (Bray–Curtis PERMANOVA R² = 0.115, P = 0.001) and remained associated with histology after adjustment for age and sex. Helicobacter was enriched in the combined IM/IEN group. In the independent IHC cohort, H. pylori positivity was highest in the IEN group (19 of 33 IEN cases). Neisseria had lower relative abundance in lesion samples, whereas the Greengenes2-assigned putative species Gemella morbillorum and Leptotrichia trevisanii were more abundant in IEN. An exploratory dysbiosis score was associated with the gene silencing score (Spearman rho = 0.61, P = 5.51 × 10⁻⁵), and the association remained nominally significant after adjustment for age, sex, and histological group. Because microbial and host features were selected and the scores were evaluated within the study datasets, these associations are susceptible to feature-selection bias. Gastric cardia precursor lesions show microbial remodeling that co-occurs with host epigenetic alterations. The score-based association is hypothesis-generating because of within-dataset feature selection and requires independent validation. Longitudinal sampling and functional studies are needed to establish temporal relationships and biological significance.