Development and validation of a ferroptosis and metabolic reprogramming gene signature for risk stratification in bladder urothelial carcinoma

Bladder urothelial carcinoma (BLCA) is clinically and molecularly heterogeneous, and conventional clinicopathologic factors do not fully reflect the biological variation associated with survival. We aimed to develop and externally evaluate a prognostic signature based on genes related to both ferroptosis and metabolic reprogramming and to examine its clinical and biological associations. The Cancer Genome Atlas (TCGA)-BLCA transcriptomic and clinical data were used for candidate-gene identification and model development. A 10-gene least absolute shrinkage and selection operator Cox signature was evaluated by complete-pipeline bootstrap validation. After platform harmonization, the same coefficients and TCGA-derived cutoff were applied to GSE13507 and GSE48075. Clinical adjustment, stage-specific analyses, immune analyses accounting for the IFNG score contribution, tumor purity, and molecular subtype, exploratory drug-response prediction, and expression analysis in 15 paired institutional tumor and adjacent tissues were performed. The signature comprised JUN, IFNG, CAV1, SREBF1, G6PD, TFRC, SCD, ZEB1, SLC7A11, and WWTR1. In TCGA-BLCA (401 patients; 176 deaths), the high-risk group had poorer overall survival (hazard ratio [HR] = 2.313; 95% confidence interval = 1.697-3.154; P < .001). The fixed TCGA cutoff also identified high-risk groups in GSE13507 (165 patients; 69 deaths; HR = 2.202; P = .0010) and GSE48075 (73 patients; 45 deaths; HR = 2.476; P = .0082). The 10-gene RiskScore yielded 1-, 3-, and 5-year areas under the curve of 0.644, 0.657, and 0.668 in TCGA-BLCA; the corresponding values ranged from 0.583 to 0.696 in the external cohorts. RiskScore remained associated with survival after adjustment for age, sex, and stage in TCGA (HR per standard-deviation increase, 1.573; P < .001). In the paired institutional cohort, 5 genes differed between tumor and adjacent tissues at a false discovery rate below 0.05, and 9 of 10 genes showed expression directions concordant with TCGA. Enrichment analyses highlighted oxidative-stress and metabolic pathways; several immune and checkpoint associations remained after sensitivity analyses, and exploratory modeling identified risk-group differences in predicted drug sensitivity. The 10-gene signature consistently stratified overall survival across TCGA-BLCA and 2 independent Gene Expression Omnibus cohorts. Its associations with clinical outcome, paired-tissue expression, and immune features support further investigation of ferroptosis-related metabolic heterogeneity in BLCA.