Muscle-invasive bladder carcinoma (MIBC) constitutes a heterogeneous group of tumors with a poor outcome. Molecular stratification of MIBC may identify clinically relevant tumor subgroups and help to provide effective targeted therapies. From seven series of large-scale transcriptomic data (383 tumors), we identified an MIBC subgroup accounting for 23.5% of MIBC, associated with shorter survival and displaying a basal-like phenotype, as shown by the expression of epithelial basal cell markers. Basal-like tumors presented an activation of the epidermal growth factor receptor (EGFR) pathway linked to frequent EGFR gains and activation of an EGFR autocrine loop. We used a 40-gene expression classifier derived from human tumors to identify human bladder cancer cell lines and a chemically induced mouse model of bladder cancer corresponding to human basal-like bladder cancer. We showed, in both models, that tumor cells were sensitive to anti-EGFR therapy. Our findings provide preclinical proof of concept that anti-EGFR therapy can be used to target a subset of particularly aggressive MIBC tumors expressing basal cell markers and provide diagnostic tools for identifying these tumors.

Deep learning methods for digital pathology analysis are an effective way to address multiple clinical questions, from diagnosis to prediction of treatment outcomes. These methods have also been used to predict gene mutations from pathology images, but no comprehensive evaluation of their potential for extracting molecular features from histology slides has yet been performed. We show that HE2RNA, a model based on the integration of multiple data modes, can be trained to systematically predict RNA-Seq profiles from whole-slide images alone, without expert annotation. Through its interpretable design, HE2RNA provides virtual spatialization of gene expression, as validated by CD3-and CD20-staining on an independent dataset. The transcriptomic representation learned by HE2RNA can also be transferred on other datasets, even of small size, to increase prediction performance for specific molecular phenotypes. We illustrate the use of this approach in clinical diagnosis purposes such as the identification of tumors with microsatellite instability.