The DUTRENEO trial: bulk prospective biomarker-guided therapy selection and spatial transcriptomic dissection of immunotherapy response in muscle-invasive bladder cancer

Predictive biomarkers for immune checkpoint inhibitors (ICI) have largely been identified retrospectively, but their clinical utility remains unproven in prospective settings. Here, we report the results of DUTRENEO, a prospective, randomized phase 2 trial testing whether a retrospective, well-validated, 18-gene tumor inflammation signature (TIS) could guide neoadjuvant ICI therapy in muscle-invasive bladder cancer. The trial failed to meet its primary endpoint, demonstrating that bulk gene expression stratification does not sufficiently enrich for responders in this setting. To understand the biological determinants of this failure, we generated a single-cell spatial transcriptomics dataset profiling 377 genes in ~5.4 million cells across large tissue areas. This high-resolution profiling revealed that response to ICI is governed by spatial architectures invisible to bulk assays, specifically the proximity of CD8⁺ T cells to cancer cells and the localized co-expression of checkpoints within epithelial-rich neighborhoods. Furthermore, we identified fibroblast-rich communities as a key feature of immune exclusion in non-responders. We derived a quantitative framework for the design of future clinical trials incorporating spatial analyses, demonstrating that reducing the annotation panel to ≥77 genes and sampling ≥3 mm tissue diameter preserves predictive spatial signal while enabling scalable throughput. These findings elucidate the architectural constraints that limit bulk biomarkers and provide a quantitative framework to embed spatial profiling into clinical studies.