Metabolic classification of bladder cancer based on multi-omics integrated analysis to predict patient prognosis and treatment response

BACKGROUND: Currently, no molecular classification is established for bladder cancer based on metabolic characteristics. Therefore, we conducted a comprehensive analysis of bladder cancer metabolism-related genes using multiple publicly available datasets and aimed to identify subtypes according to distinctive metabolic characteristics. METHODS: RNA-sequencing data of The Cancer Genome Atlas were subjected to non-negative matrix fractionation to classify bladder cancer according to metabolism-related gene expression; Gene Expression Omnibus and ArrayExpress datasets were used as validation cohorts. The sensitivity of metabolic types to predicted immunotherapy and chemotherapy was assessed. Kaplan–Meier curves were plotted to assess patient survival. Differentially expressed genes between subtypes were identified using edgeR. The differences among identified subtypes were compared using the Kruskal–Wallis non-parametric test. To better clarify the subtypes of bladder cancer, their relationship with clinical characteristics was examined using the Fisher’s test. We also constructed a risk prediction model using the random survival forest method to analyze right-censored survival data based on key metabolic genes. To identify genes of prognostic significance, univariate Cox regression, lasso analysis, and multivariate regression were performed sequentially. RESULTS: Three bladder cancer subtypes were identified according to the expression of metabolism-related genes. The M1 subtype was characterized by high metabolic activity, low immunogenicity, and better prognosis. M2 exhibited moderate metabolic activity, high immunogenicity, and the worst prognosis. M3 was associated with low metabolic activity, low immunogenicity, and poor prognosis. M1 showed the best predicted response to immunotherapy, whereas patients with M1 were predicted to be the least sensitive to cisplatin. By contrast, M2 showed the worst predicted response to immunotherapy but was predicted to be more sensitive to cisplatin, doxorubicin, and other first-line anticancer drugs. M3 was the most sensitive to gemcitabine. The risk model based on metabolic genes effectively predicted the prognosis of bladder cancer patients. CONCLUSIONS: Metabolic classification of bladder cancer has potential clinical value and therapeutic feasibility by inhibiting the associated pathways. This classification can provide valuable insights for developing precise bladder cancer treatment. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12967-021-02865-8.