Abstract

T1 high-grade (T1HG) bladder cancer represents an aggressive subset of non–muscle-invasive bladder cancer (NMIBC) with frequent Bacillus Calmette–Guérin (BCG) failure and a high risk of progression, yet current models inadequately guide treatment selection between early cystectomy and bladder preservation. Integrative multi-omics profiling of 147 tumors identifies two clinically distinct subtypes. A high-risk subtype (T1HG1) is defined by coupled anoikis resistance and immune evasion, exhibiting markedly increased progression rates (>80% vs. <20%), poor BCG responsiveness, and a higher likelihood of cystectomy. NAD(P)H:quinone oxidoreductase 1 (NQO1) is identified as a central regulator linking tumor-intrinsic survival to suppression of macrophage–T cell crosstalk. Elevated NQO1 promotes anoikis resistance and reprograms macrophages toward an immunosuppressive phenotype, limiting CXCL9-mediated T cell recruitment and facilitating immune escape. Pharmacologic inhibition of NQO1 using skullcapflavone II restores apoptotic sensitivity and enhances cisplatin efficacy, resulting in significant tumor suppression with favorable tolerability in preclinical models. A multi-omic machine learning framework for T1HG UCB, termed T1HG-UCBguider, integrating clinical, transcriptomic, and methylation features, enables individualized risk stratification and treatment guidance. Validation across seven independent cohorts, demonstrates robust performance in identifying patients at risk of progression and BCG failure. These findings establish a biologically grounded framework for precision management of T1HG bladder cancer.