The high intra-patient heterogeneity in multifocal primary prostate cancer (PCa) has curtailed the efficacy of current treatment options. By employing twin biopsies from multiple lesions with matched patient-derived organoids (PDO) models, the PCa molecular heterogeneity was investigated. We utilized genomics, transcriptomics and machine learning (ML) approaches to elucidate and predict the underlying mechanisms of pharmacological heterogeneity. Our data indicate a vulnerability of primary PCa organoids for small molecule inhibitors targeting receptor tyrosine kinases (MET, ALK, SRC). By exploring gene expression data from matched parental tissue in an unsupervised manner, we identified two distinct clusters of samples. Interestingly, the PDO drug responses were significantly different between the two clusters for 4/11 compounds tested. We developed a transcriptomics-based, cluster prediction model, which can accurately stratify samples into the two clusters. Notably, our prediction model is based on tissue profiles, therefore, it can be utilized to rapidly evaluate new cases and suggest promising drug candidates, even when PDO derivation is not feasible. Taken together, we propose a novel flexible stratified oncology approach that can swiftly and accurately highlight promising drug vulnerabilities of PCa patients.
