Background: Urothelial carcinoma (UC) is a biologically heterogeneous disease, and current molecular classifications have limited integration into clinical decision-making. To further pursue precision oncology efforts in UC, we developed a molecular classification framework applicable to transcriptomic and proteomic data from non-muscle-invasive bladder cancer (NMIBC), muscle-invasive bladder cancer (MIBC) and urothelial cancer cell lines.
Methods: Using a whole-transcriptome self-organised map and regularised semi-supervised clustering of 4439 bulk NMIBC and MIBC transcriptomes and proteomes, and 33 UC cell lines, we identified three molecular UC clusters. Making use of both in silico and in vitro approaches, we selected promising treatment approaches for each cluster.
Results: The three developed clusters displayed distinct signatures of mRNA, proteins, biological processes, metabolism and essential driver genes. They also differed in prognosis and machine learning-predicted treatment vulnerabilities and resistance. High-risk, stroma-rich Cluster #1 cancers were predicted to respond to selected cytotoxic drugs, ferroptosis inducers and PARP inhibitors. For the aggressive, fast-proliferating, immune-infiltrated Cluster #2 tumours with basal/squamous differentiation, cytotoxic agents and EGFR/ERBB- and MEK/ERK-targeting therapies were proposed. Cluster #3 cancers of predominantly luminal papillary phenotype with scarce stroma and immune infiltration were enriched with NMIBC and low-risk malignancies. For patients with Cluster #3 tumours, selected epigenetic drugs or EGFR/FGFR inhibitors may represent attractive treatment options.
Conclusions: Our novel molecular taxonomy holds promise as a practical framework for patient risk stratification and clinical trials in UC. Our molecular classification scheme may facilitate personalised transcriptome- and proteome-based risk assessment and clinical trial design for the development of various therapeutics.
Key points: We developed three UC clusters, applicable for MIBC and NMIBC, which were validated using transcriptomic- and proteomic datasets. Publically available UC cell lines were assigned to the clusters, to have in vitro models representing each cluster. The clusters differ in molecular and biological signatures, with distinct prognostic and therapeutic characteristics.
Keywords: bladder cancer; drug response; molecular classification; proteomics; transcriptomics.
© 2026 The Author(s). Clinical and Translational Medicine published by John Wiley & Sons Australia, Ltd on behalf of Shanghai Institute of Clinical Bioinformatics.