Gene Signature for Predicting Metastasis in Prostate Cancer Using Primary Tumor Expression Profiles

Abstract

Prostate cancer (PCa) is currently the most commonly diagnosed cancer and second leading cause of cancer-related death in men in the United States. The development of metastases is associated with a poor prognosis in PCa patients. Since current clinicopathological classification schemes are unable to accurately prognosticate the risk of metastasis for those diagnosed with localized PCa, there is a pressing need for precise and easily attainable biomarkers of metastatic risk in these patients. Primary tumor samples from 1239 individuals with PCa were divided into development (n=1000) and validation (n=239) cohorts. In the development cohort, we utilized a meta-analysis workflow on retrospective primary tumor gene expression profiles to identify a subset of genes predictive of metastasis. For each gene, we computed Hedges’ g effect size and combined their p-values using Fisher’s combined probability test. We then adjusted for multiple hypothesis testing using the Benjamini-Hochberg method. Our developed gene signature, termed Meta-Score, achieved a robust performance at predicting metastasis from primary tumor gene expression profiles, with an AUC of 0.72 in the validation cohort. In addition to its robust predictive power, Meta-Score also demonstrated a significant prognostic utility in two independent cohorts. Specifically, patients with a higher risk-score had a significantly worse metastasis-free survival and progression-free survival compared to those with lower score. Multivariate cox proportional hazards model showed that Meta-Score is significantly associated with worse survival even after adjusting for Gleason score. Our findings suggest that our primary tumor transcriptional signature, Meta-Score, could be a valuable tool to assess the risk of metastasis in PCa patients with localized disease, pending validation in large prospective studies.