Identification of a brand intratumor microbiome signature for predicting prognosis of hepatocellular carcinoma.

PURPOSE: Given that prognosis of hepatocellular carcinoma (HCC) differs dramatically, it is imperative to uncover effective and available prognostic biomarker(s). The intratumor microbiome plays a significant role in the response to tumor microenvironment, we aimed to identify an intratumor microbiome signature for predicting the prognosis of HCC patients accurately and investigate its possible mechanisms subsequently. METHODS: The TCGA HCC microbiome data (TCGA-LIHC-microbiome) was downloaded from cBioPortal. To create an intratumor microbiome-related prognostic signature, univariate and multivariate Cox regression analyses were used to quantify the association of microbial abundance and patients' overall survival (OS), as well as their diseases specific survival (DSS). The performance of the scoring model was evaluated by the area under the ROC curve (AUC). Based on the microbiome-related signature, clinical factors, and multi-omics molecular subtypes on the basis of "icluster" algorithm, nomograms were established to predict OS and DSS. Patients were further clustered into three subtypes based on their microbiome-related characteristics by consensus clustering. Moreover, deconvolution algorithm, weighted correlation network analysis (WGCNA) and gene set variation analysis (GSVA) were used to investigate the potential mechanisms. RESULTS: In TCGA LIHC microbiome data, the abundances of 166 genera among the total 1406 genera were considerably associated with HCC patients' OS. From that filtered dataset we identified a 27-microbe prognostic signature and developed a microbiome-related score (MRS) model. Compared with those in the relatively low-risk group, patients in higher-risk group own a much worse OS (P < 0.0001). Besides, the time-dependent ROC curves with MRS showed excellent predictive efficacy both in OS and DSS. Moreover, MRS is an independent prognostic factor for OS and DSS over clinical factors and multi-omics-based molecular subtypes. The integration of MRS into nomograms significantly improved the efficacy of prognosis prediction (1-year AUC:0.849, 3-year AUC: 0.825, 5-year AUC: 0.822). The analysis of microbiome-based subtypes on their immune characteristics and specific gene modules inferred that the intratumor microbiome may affect the HCC patients' prognosis via modulating the cancer stemness and immune response. CONCLUSION: MRS, a 27 intratumor microbiome-related prognostic model, was successfully established to predict HCC patients overall survive independently. And the possible underlying mechanisms were also investigated to provide a potential intervention strategy.