Integrated multicenter single-cell atlas of bladder cancer reveals tumor heterogeneity associated with immunotherapy

Background: Molecular subtyping of bladder cancer (BCa) using bulk transcriptomic data is hindered by tumor microenvironment (TME) contamination and intra-tumoral heterogeneity. These factors also contribute to variable responses to immune checkpoint inhibitor (ICI) therapy. Single-cell RNA sequencing offers a high-resolution approach to overcome these limitations by enabling tumor-specific analysis and improving therapeutic target identification. Materials and methods: We constructed a large multicenter single-cell atlas of BCa by integrating nine cohorts comprising 123 samples and 842 127 cells. Tumor epithelial cells ( n = 265 962) were identified via copy number variation (CNV) analysis using inferCNV. Pseudobulk RNA profiles were generated to simulate conventional bulk samples (Pseudobulk all ) and purified tumor profiles (Pseudobulk tumor ), enabling systematic evaluation of bulk-based classifiers. Inter-tumoral and intra-tumoral heterogeneity programs were defined based on transcriptional profiles and validated through in vitro and in vivo experiments. Additionally, we cross-analyzed five immunotherapy datasets encompassing 13 immune response phenotypes to identify molecular features associated with therapeutic resistance and sensitivity. Results: Unsupervised clustering identified 11 major cell types and 55 non-epithelial subtypes, with TME heterogeneity driven by tissue origin and malignancy status. Existing molecular classifiers UNC (University of North Carolina at Chapel Hill) showed compromised concordance between Pseudobulk all and Pseudobulk tumor , particularly in stroma-rich samples. We defined three inter-tumoral (GE inter 2–4) subtypes with distinct CNV and transcription factor signatures that remained stable even at low tumor purity. Seven intra-tumoral (GE intra ) programs captured key biological states, including an EMT/Stem-like module associated with immunosuppression and ICI resistance. Mechanistically, TGF-β1 induced expression of EMT/Stem-like genes (e.g., PTHLH ), while Pthlh knockdown synergized with anti-PD-L1 therapy in murine models by enhancing T cell cytotoxicity without altering T cell infiltration. Conclusions: This study provides a robust single-cell framework for resolving inter- and intra-tumoral heterogeneity in BCa, overcoming limitations of bulk-based subtyping. The TGF-β1-PTHLH axis contributes to ICI resistance, positioning PTHLH as a promising therapeutic target. These findings offer new insights for improving immunotherapeutic strategies in BCa.