Circulating Microbial Metabolites Predict Tumor Relapse and Chemotherapy Efficacy in Nasopharyngeal Carcinoma

The value of microbial metabolites in prognosis and treatment response prediction in patients with nasopharyngeal carcinoma (NPC) remains unclear. Here, through the untargeted metabolomic analysis of plasma in 48 paired NPC patients with or without tumor relapse, we identified distinct circulating metabolite atlases between NPC patients with different prognoses. We used bootstrap least absolute shrinkage and selection operator (LASSO) on a penalized Cox regression model to select metabolites and constructed a metabolite risk model comprising four microbial metabolites in a training cohort (n = 202), and validated it in an independent test cohort (n = 201) and an external validation cohort (n = 180). The model stratified patients into three risk groups. Patients in the low-risk group had optimal DFS, DMFS, and OS, compared with those in the intermediate-risk group. High-risk patients had poor survival across all clinical endpoints. Furthermore, patients in the intermediate-risk group could benefit from induction chemotherapy. In addition, we generated a nomogram integrating the risk model, N stage, and plasma EBV-DNA load, which further enhanced the predictive accuracy of the metabolite risk model. Collectively, we developed and validated a robust predictive model based on serum metabolites, promoting risk stratification and enhancing treatment outcomes in patients with NPC. We identified distinct circulating metabolite atlases between NPC patients with different prognoses in the training cohort (n = 202). A risk model, comprising four microbial metabolites, was developed to stratify patients into three risk groups. Patients in the low-risk group had optimal DFS, DMFS, and OS, compared with those in the intermediate-risk group. High-risk patients had poor survival across all clinical endpoints. Findings were validated using an independent test cohort (n = 201) and an external validation cohort (n = 180). Specifically, a nomogram integrating the risk model, N stage, and plasma EBV-DNA load enhanced predictive accuracy. Moreover, patients benefited from induction chemotherapy with improved survival in the intermediate-risk group, but not in the low-risk and high-risk groups.