Multi-omics and pathological landscape of a senescence-associated prognostic model in bladder urothelial carcinoma

Background: Cellular senescence contributes to tumor progression and shapes the tumor microenvironment (TME), yet its prognostic and biological significance in bladder urothelial carcinoma (BLCA) remains incompletely understood. This study aimed to identify senescence-associated prognostic signals through pan-cancer screening and to develop an integrated risk stratification framework for BLCA that bridges molecular characterization with translational relevance.

Methods: In BLCA, weighted gene co-expression network analysis (WGCNA) and least absolute shrinkage and selection operator (LASSO)-Cox regression were used to construct a 15-gene prognostic model. Molecular subtype distribution, immunohistochemical validation, immune microenvironment, predicted immunotherapy response, whole-slide image (WSI)-based pathology prediction, single-cell transcriptomic features, and spatial transcriptomic characteristics were evaluated. Machine-learning and SHapley Additive exPlanations (SHAP) analyses were used to prioritize key predictive genes, followed by somatic single-nucleotide variant (SNV), copy number variation (CNV), and exploratory cis-expression quantitative trait locus (cis-eQTL) annotation.

Results: Pan-cancer analysis covering solid tumors including kidney renal clear cell carcinoma (KIRC), liver hepatocellular carcinoma (LIHC), and lung adenocarcinoma (LUAD) revealed cancer-type-specific prognostic effects of senescence signatures, with significant survival associations in BLCA, LIHC, and glioma. The 15-gene LASSO risk model effectively stratified BLCA prognosis across multiple cohorts and was associated with aggressive molecular subtypes. Immunohistochemical validation supported distinct epithelial differentiation and proliferative features between LASSO risk groups. High-risk tumors showed an immune-cold microenvironment and poorer predicted response to immune checkpoint blockade. WSI-based deep learning predicted LASSO risk groups with favorable performance. Single-cell and spatial analyses showed that the risk signature was enriched in high-CNV malignant urothelial carcinoma cells, accompanied by suppressed programmed cell death pathways and tumor-core enrichment. Machine-learning interpretation narrowed the 15-gene signature to 10 key predictive genes. These genes showed limited somatic SNV alterations, heterogeneous CNV patterns, and exploratory cis-eQTL signals, providing complementary genomic and regulatory context.

Conclusions: This study establishes a senescence-associated multi-dimensional risk stratification framework for BLCA. The identified signature links prognosis with molecular subtype, immune status, histopathological features, malignant cell states, spatial heterogeneity, and selected genomic/regulatory annotations, supporting its potential value for understanding BLCA tumor ecosystem remodeling.