A recurrent cancer-associated fibroblast/TGF-beta/endothelial barrier direction is associated with immune exclusion...

Immune checkpoint inhibitors benefit only a subset of patients with urothelial carcinoma, and bulk biomarkers may not capture spatially organized stromal and vascular programs that restrict immune access. We performed a retrospective public multi-cohort transcriptomic study to examine whether a cancer-associated fibroblast (CAF)/TGF-beta/endothelial barrier direction is associated with immune exclusion and checkpoint resistance in bladder cancer. GSE171351 was used for spatial discovery, GSE319536 for external spatial recurrence testing, TCGA-BLCA for prognosis-oriented support, and IMvigor210, GSE176307 and GSE328930/DUTRENEO for treatment-response projection. In the four-section discovery cohort, rank-5 non-negative matrix factorization separated an E2 barrier-enriched program from an E4 hypoxic epithelial program, although component assignment was strongly associated with section identity. The external cohort reproduced the direct CAF/TGF-beta/endothelial barrier direction in 22 of 22 sections but did not reproduce E2/E4 mutual exclusivity: projected E2 and E4 were positively correlated (rho = 0.928). In IMvigor210, top-quartile E2-high tumors showed suggestive, but not statistically conclusive, evidence of anti-PD-L1 non-response after full biologic covariate adjustment (OR=2.57, 95% CI 1.00-6.64, p = 0.051; delta AUC = 0.009). A prespecified heuristic barrier-exclusion score was associated with non-response in exploratory fixed- and random-effects summaries (OR=1.52, 95% CI 1.20-1.92), including a sensitivity analysis excluding GSE328930. These results identify a recurrent barrier direction rather than a universal two-niche architecture and support a hypothesis-generating mechanism-to-translation framework that requires prospective spatial and pathology-level validation.