Tumor Gene Expression Profiling in Prostate Cancer (PCa) Identifies Molecular Taxonomies Associated with Node Positivity and Quantitative Imaging Features on Multiparametric MRI (mpMRI)

Abstract

Genomic classifiers based on tumor gene expression add to the predictive power of clinicopathologic features in contemporary PCa risk stratification. Increasingly, mpMRI is utilized to identify occult high grade disease as well as to aid in pre-treatment staging of localized disease. We undertook a study to examine the association of tumor gene expression profiles following radical prostatectomy (RP) with radiographic features on pre-prostatectomy mpMRI. We queried an institutional review board-approved research registry of PCa patients who had preoperative 3T mpMRI followed by post-prostatectomy tumor molecular testing performed in the course of clinical care. Genomic testing was not geographically matched to mpMRI detected lesions. Our analysis focused on correlative findings between clinical information, a genomic classifier score, and MRI features. Statistical software was used for statistical modeling of associations between gene signatures, clinicopathologic variables and imaging characteristics. Forty-eight patients met criteria for this analysis. Median age was 67 years (range 45-78) and average pre-op PSA was 11.72. Pathologic findings after RP were: 28 (58%) Gleason Score (GS) 7 (17 were 3+4, 11 were 4+3), 5 (10%) GS 8, 15 (31%) GS 9, 42 (87.5%) had EPE, 15 (31.3%) had SVI, 30 (62.5%) had positive surgical margins (SM), and 13 (27%) were node positive. A majority of patients had pre-op PI-RADS 5 lesions (n=38) detected on mpMRI. No patients had clinically or radiographically suspicious lymph nodes prior to RP. The number of patients with Decipher low, intermediate, and high risk scores were 11 (23%), 11 (23%), and 26 (54%), respectively. Non-parametric testing demonstrated a significant association between Decipher score and PI-RADS score on mpMRI (Kruskal-Wallis Test, p<0.050). Similarly, mean apparent diffusion coefficients (ADC) were directly correlated with Decipher score (Kruskal-Wallis Test, p<0.05). mpMRI features alone did not correlate with positive SM and accounting for Decipher score in univariate or multivariate models did not improve the ability of mpMRI features to predict SM status. Higher post-operative decipher score was, however, associated with likelihood of pathologic node positivity (ANOVA p=0.009). A significant correlation was observed between Decipher scores and PI-RADS classification as well as mean tumor ADC values on mpMRI. A high risk Decipher score was associated with higher rates of pathologic LN positivity otherwise undetected on mpMRI. These data support further validation of genomic classifiers in combination with mpMRI for risk stratifying PCa patients.