Genomic biomarkers to predict outcome in Gleason Score 9-10 disease.

Abstract

44 Background: Gleason score (GS) 9-10 prostate cancer (PCa) has classically been considered the most aggressive form of clinically localized disease. However, outcomes remain heterogeneous. Whether specific biomarkers may help guide prognostication within GS 9-10 disease remains unknown. Methods: Microarray-derived gene expression data were obtained from six retrospective radical prostatectomy cohorts (n=1076) and two prospective cohorts with data from the Decipher GRID (n=7000). A total of 957 patients had GS 9-10 disease. Clinical outcomes data (i.e., distant metastasis [DM] and prostate cancer-specific mortality [PCSM]) were available for 1077 patients (201 with GS 9-10 disease). We filtered for genes with high expression levels and differential expression between GS 9-10 and GS ≤8 (via Wilcoxon test with adjustment for false discovery rate [FDR]), and then used using weighted gene co-expression network analysis [WGCNA] to identify distinct modules. Genes with both a low p-value and a high connectivity within each module were included as potential predictors in logistic regression models constructed using elastic net regularization. We also chose genes within the GS 9-10 cohort with q-value <0.01 and <0.3 after FDR correction for outcomes of DM and PCSM, respectively. We used the cross-validated AUC for quantifying the discrimination of each gene set. Results: We identified a set of 12 genes with an AUC of 0.81 for discriminating GS 9-10 vs. GS ≤8. A separate set of 7 genes had an AUC of 0.83 for predicting DM within GS 9-10 patients, compared with an AUC of 0.68 within GS ≤8 patients. A third set of 13 genes had an AUC of 0.93 for predicting PCSM within GS 9-10 patients, but an AUC of only 0.57 for predicting PCSM within GS ≤8. Conclusions: These data suggest that a genomic biomarker signature can strongly discriminate GS 9-10 from GS ≤8. Separate gene sets can also predict DM and PCSM within GS 9-10 patients with high fidelity, but are not as predictive of these outcomes within GS ≤8 patients. These data support the hypothesis that GS 9-10 disease is a biologically distinct yet heterogeneous entity, and biomarker discovery efforts to better guide upfront treatment intensification in this subset are feasible and warranted. (NCT02609269).