An early report on the molecular subtyping of muscle-invasive bladder cancer (MIBC) by gene expression suggested that response to neoadjuvant chemotherapy (NAC) varies by subtype. To investigate the ability of molecular subtypes to predict pathological downstaging and survival after NAC. Whole transcriptome profiling was performed on pre-NAC transurethral resection specimens from 343 patients with MIBC. Samples were classified according to four published molecular subtyping methods. We developed a single-sample genomic subtyping classifier (GSC) to predict consensus subtypes (claudin-low, basal, luminal-infiltrated and luminal) with highest clinical impact in the context of NAC. Overall survival (OS) according to subtype was analyzed and compared with OS in 476 non-NAC cases (published datasets). Gene expression analysis was used to assign subtypes. Receiver-operating characteristics were used to determine the accuracy of GSC. The effect of GSC on survival was estimated by Cox proportional hazard regression models.The models generated subtype calls in expected ratios with high concordance across subtyping methods. GSC was able to predict four consensus molecular subtypes with high accuracy (73%), and clinical significance of the predicted consensus subtypes could be validated in independent NAC and non-NAC datasets. Luminal tumors had the best OS with and without NAC. Claudin-low tumors were associated with poor OS irrespective of treatment regimen. Basal tumors showed the most improvement in OS with NAC compared with surgery alone. The main limitations of our study are its retrospective design and comparison across datasets.
