Comprehensive molecular classification of localized prostate adenocarcinoma reveals a tumour subtype predictive of...

Management of localized prostate cancer (PCa) is a major clinical challenge since most of these cancers would not evolve but a majority of patients will still undergo a life-changing radical surgery. Molecular studies have shown that PCa can be classified according to their genomic alterations but none of the published PCa molecular classifications could identify a subtype corresponding to non-evolutive tumours.

Multi-omics molecular profiling was carried out on post-radical prostatectomy material from a cohort of 130 patients with localized PCa. We used unsupervised classification techniques to build a comprehensive classification of prostate tumours based on three molecular levels: DNA copy number, DNA methylation, and mRNA expression. Merged data from our cohort and The Cancer Genome Atlas cohort were used to characterize the resulting tumour subtypes. We measured subtype-associated risks of biochemical relapse using Cox regression models and survival data from five cohorts including the two aforementioned.

We describe three PCa molecular subtypes associated with specific molecular characteristics and different clinical outcomes. Particularly, one subtype was strongly associated with the absence of biochemical recurrence. We validated this finding on 746 samples from 5 distinct cohorts (P = 3.41 × 10-8, N = 746 tumour samples), and showed that our subtyping approach outperformed the most popular prognostic molecular signatures to accurately identify a subset of patients with a non-evolutive disease. We provide a set of 36 transcriptomic biomarkers to robustly identify this subtype of non-evolutive cases whose prevalence was estimated to 22% of all localized PCa tumours.

At least 20% of patients with localized PCa can be accurately predicted to have a non-evolutive disease on the basis of their molecular subtype. Those patients should not undergo immediate surgery and rather be placed under active surveillance.