Abstract
<title>Abstract</title> <p> Background Gastric cancer (GC) remains a leading cause of cancer-related morbidity and mortality globally. Emerging evidence suggests that alterations in the oral microbiota are closely associated with GC progression via mechanisms such as ectopic colonization, metabolic disruption, and localized inflammation. However, a systematic evaluation of oral microbiota signatures across distinct stages of GC and their diagnostic utility is lacking. This study aimed to characterize the oral microbiota across different stages of GC progression and to evaluate its potential as a non-invasive tool for early screening and risk stratification. Methods We prospectively enrolled 129 participants, including patients with non-GC chronic gastritis (normal controls, NC), precancerous lesions of gastric cancer (PLGC), early gastric cancer (EGC), and advanced gastric cancer (AGC), confirmed by endoscopy and histopathology. Saliva samples were analyzed using 16S rRNA gene sequencing to assess microbial composition, diversity, and co-occurrence networks. LEfSe and MetaCyc were utilized to identify differential taxa and predict functional pathways. Furthermore, diagnostic models based on Least Absolute Shrinkage and Selection Operator regression were developed using strictly filtered microbial features, with predictive performance evaluated via Leave-One-Out Cross-Validation. Results Microbial profiling revealed <italic>Bacteroidota</italic> , <italic>Firmicutes</italic> , <italic>Proteobacteria</italic> , and <italic>Fusobacteriota</italic> as the predominant phyla across all groups. Stage-specific variations were observed: <italic>Streptococcus</italic> and its higher taxonomic levels were significantly enriched in AGC; <italic>Oribacterium sinus</italic> characterized EGC; unclassified <italic>Prevotella sp.</italic> were elevated in PLGC; and commensals such as <italic>Alloprevotella</italic> and <italic>Rothia mucilaginosa</italic> were enriched in NC. The LASSO regression models demonstrated robust discriminatory performance, achieving Area Under the Curve (AUC) values of 0.911 (95% CI: 0.837–0.985), 0.888 (95% CI: 0.813–0.962), and 0.902 (95% CI: 0.826–0.979) for identifying PLGC, EGC, and AGC, respectively. Conclusion The structural composition and functional pathways of the oral microbiota vary significantly across different stages of gastric carcinogenesis. The microbiota-based LASSO models exhibited high diagnostic efficacy, suggesting that oral microbiome profiling may serve as a potential non-invasive adjunct for early GC screening. These findings provide a theoretical basis for developing microbial diagnostic tools, though multi-center prospective cohorts are warranted to validate their clinical generalizability. Trial registration Clinical trial number: not applicable. </p>