Abstract
BACKGROUND: Significant gaps exist in understanding the gastrointestinal microbiota in patients with pancreatic cancer (PCA) versus benign or low-grade malignant pancreatic tumors (NPCA). This study aimed to analyze these microbiota characteristics and explore their potential use in distinguishing malignant pancreatic lesions. METHODS: Between September 2020 and May 2024, fecal and oral samples were collected from 121 patients undergoing surgical resection or diagnostic biopsy of pancreatic lesions, including 75 patients with PCA and 46 patients with NPCA, and 16s rRNA sequencing was performed. Random forest models based using fecal and oral microbiota data were developed to diagnose PCA and NPCA, with performance assessed using the leave-one-out cross validation method. RESULTS: The Shannon index and PCoA analysis revealed significant differences in oral microbiota composition between PCA and NPCA (p < 0.001 and p = 0.001, respectively). Fecal microbiome richness differed significantly (p = 0.02), though composition similarity was noted (p = 0.238). LEfSe identified 16 and 23 genera with significant differences in fecal and oral microbiomes, respectively. Random forest classifiers based on fecal and oral microbiota achieved areas under the curves (AUCs) of 89.4% and 96.3%, respectively, for distinguishing PCA and NPCA. In the mucinous tumor cohort, oral and fecal microbiome classifiers outperformed CA19-9, yielding AUCs of 83.0% and 85.2%, respectively. CONCLUSION: Fecal and oral microbiota compositions were significantly different between PCA and NPCA patients. Random forest classifiers utilizing fecal and oral microbiota data effectively distinguish between benign or low-grade malignant and malignant pancreatic lesions.