Integrated-omics of MRI-visible and -invisible prostate cancer identifies molecular correlations with clinical outcome.

Abstract

e17506 Background: The inclusion of multiparametric MRI (mpMRI) to prostate cancer (PCa) diagnostics has increased the detection rate and has improved the prediction rate of clinically significant PCa. Yet, mpMRI has a false negative rate of 12.6%, missing 6-13% of clinically significant PCa. The mechanisms underlying MRI visibility are poorly understood; therefore, to probe the molecular and cellular underpinnings of PCa MRI visibility, we profiled tissue from Gleason score and clinically matched patients with MRI-visible and MRI-invisible PCa who underwent radical prostatectomy. Methods: MRI-invisible (n = 7) and MRI-visible (n = 8) PCa tumors were evaluated with multiplex immunofluorescence (MxIF; 14 markers) and were subjected to gene expression profiling using the HTG EdgeSeq Oncology Biomarker Panel (2,549 genes). Gene expression analysis was also performed using The Cancer Genome Atlas (TCGA), including normal prostate (n = 52) and PCa (n = 387) tissue. Analyses were performed using the BostonGene automated pipeline. Results: MpMRI-visible PCa tumors (62.5%) displayed compact epithelial tumor architecture compared with mpMRI-invisible PCa tumors (28.5%). mpMRI-visible PCa had higher malignant cell density (p = 0.04) and increased neighboring malignant cells (p = 0.07), correlating with MRI-visible PCa complex tumor architecture (r = 0.49 for neighboring malignant cells vs tumor cell density). Tumor stromal organization differences were determined by measuring Wasserstein distances between distributions, and mpMRI-invisible PCa stroma appeared more similar to normal tissue. The visible group exhibited lower expression of stroma-enriched genes such as filamin C (FLNC) (FDR < 0.1) and cellular adhesion-related genes (FDR < 0.4), with gene expression signatures markedly different compared to normal prostate tissue. Higher malignant cell density, neighboring malignant cells, and Wasserstein distances, and low FLNC expression – all mpMRI visibility characteristics – were associated with patient relapse (p = 0.02). Low stroma signature expression in the TCGA cohort correlated with inferior PCa PFS (p = 0.005). Conclusions: This is the first integrated multi-omics analysis of clinically matched mpMRI-visible and -invisible PCa. mpMRI-invisible tumors exhibited molecular, cellular, and structural characteristics more akin to normal prostate tissue, which may render them undetectable by imaging. A stroma-associated gene signature, a mpMRI-invisible tumor feature, correlated with better PCa clinical outcomes.