<i>Bacteroides sp. DH3716P</i> and mediation of immunotherapy response in oligometastatic nasopharyngeal carcinoma via the gut–nasopharyngeal axis.

6031 Background: Predictors of benefit from gemcitabine/cisplatin (GP) plus PD-1 blockade in oligometastatic nasopharyngeal carcinoma (OM-NPC) remain limited. We investigated whether baseline gut microbiome features are linked to response through a gut–nasopharyngeal axis. Methods: We performed metagenomic sequencing on baseline fecal samples from 69 OM-NPC patients receiving GP chemotherapy combined with PD-1 blockade. β-diversity analyses (PCA, PCoA, NMDS) characterized microbiota compositional differences. LEfSe analysis identified differentially abundant taxa between responders (R) and non-responders (NR). Functional enrichment analyses (KEGG, KO, MetaCyc) determined pathway differences. Co-occurrence network analysis assessed microbiota stability. Single-cell RNA sequencing from 7 tumor specimens was analyzed using SAHMI algorithm to determine bacterial-immune cell interactions. Machine learning models built on microbiota, clinical parameters, and serum biomarkers predicted treatment response. Results: Microbiota composition differed significantly between R and NR groups. R group showed significant enrichment of Enterococcus , Megasphaera , and Streptococcus . Functional analysis revealed R group possessed enhanced short-chain fatty acid (SCFA) synthesis, glycolysis, and lipid metabolism pathways. Network analysis demonstrated R group had denser ecological networks with superior stability and functional redundancy compared to NR. At species level, Bacteroides sp. DH3716P abundance associated with response and prolonged progression-free survival (PFS). This strain highly expressed SusD/RagB family outer membrane proteins and glycoside hydrolase activities, suggesting xylan degradation pathway activation leading to host glycolysis stimulation and SCFA-mediated immune regulation. scRNA-seq analysis revealed Bacteroides predominantly infected CD8+ cells in tumor microenvironment. Bacteroides+ CD8+ T cells demonstrated enhanced IL-15 pathway activity, activated multiple immune response pathways, and increased sensitivity to ICIs. Cell-cell communication analysis indicated Bacteroides+ CD8+ T cells recruited proliferative T cell subsets through CCL5-CCR5 axis, promoting sustained CD8+ T-mediated anti-tumor immunity. Integrated machine learning models achieved high accuracy in predicting treatment response (AUC: 0.996, 95% CI: 0.989-1.000). Conclusions: Bacteroides sp. DH3716P sustains immunotherapy response in OM-NPC through enhanced SCFA production and IL-15-CCL5/CCR5 axis-mediated CD8+ T cell activation and proliferative T cell recruitment in the tumor microenvironment. Our microbiota-derived predictive model (AUC 0.996) establishes a clinically feasible strategy for microbiome-assisted personalized immunotherapy in OM-NPC.