20 Background: Racial disparities in prostate cancer (PCa) incidence and mortality are well known. PCa is known to be more aggressive in African American men (AAM) in terms of higher incidence and mortality rates. Here we validate a tumor gene expression pan-cancer race model in men with PCa and further characterize genomic differences that may contribute to disparate clinical outcomes Methods: We obtained de-identified genome-wide expression profiles from clinical use of the Decipher RP test in 9,953 men from the GRID registry database. A subset of men (n = 313) had known race status. A pan-cancer race model, developed to predict patient AAM race from analysis of gene expression patterns in 4,162 tumors from retrospective cohorts with known race status was applied to the prospective cohort for race prediction. Gene expression data was used to define genomic differences. Results: The race model has an AUC of 0.98 discriminating EAM from AAM in independent PCa cohort. The model was then applied to the 9,640 GRID patients with unknown race status and classified 6,831 as EAM, 1,058 as AAM with 1,751 as having indeterminate race. Characterizing the molecular subtypes, we found known and predicted AAM to be enriched with SPINK1+ tumors (21% and 24%, respectively) compared to predicted EAM (8%). In contrast, while ERG+ was found 22% and 19% in known and predicted AAM, respectively compared to 46% in predicted EAM. Based on PAM50 prostate cancer classifier, 61% of AAM were classified as basal-like tumors, whereas 41% were basal-like in EAM. Similarly, 28% of AAM had low AR-A while only 11% of EAM had low AR-A. AAM tumors had higher levels of immune infiltration signatures as well as higher scores for inflammatory and interferon gamma responses, and Interleukin 6 (IL6) signaling activity scores. AAM had lower DNA repair and glycolysis pathway activity compared to EAM Conclusions: Known and predicted AAM, were enriched with SPINK1+ tumors, higher immune infiltration and activation but lower ERG+, DNA repair and AR activity tumors. Using such large GRID data with known race, we will further understand the underlying causes associated with prostate cancer racial disparities which could lead to personalized diagnosis and treatment.
Copy number and gene expression differences between African American and Caucasian American prostate cancer
