Integrated Radiogenomic Subtyping and Treatment Response of Intermediate and High Risk Prostate Cancer

Abstract

The capacity to prospectively predict poor treatment response in patients aggressive localized prostate cancer could allow for selective treatment intensification. We integrated clinical features, genomics, gene expression, and multiparametric MRI (mpMRI) features in intermediate and high risk prostate cancer patients receiving external beam radiation therapy (EBRT) and androgen deprivation therapy (ADT) and assessed for correlations with treatment response. Patients with intermediate or high-risk prostate cancer receiving definitive EBRT + ADT were enrolled on a prospective, IRB approved trial. mpMRI was performed pre-treatment and 6 mos post-EBRT. DNA from pre-treatment mpMRI targeted biopsies and matched germline DNA were evaluated by whole exome sequencing (mean tumor target coverage 118x). Gene expression was profiled by cDNA microarray, and analyzed using signatures available in the tumor expression software package. Multivariate logistic regression modeling included the following parameters: clinical (Gleason Grade, pre-treatment PSA, risk group), mpMRI (pretreatment tumor volume, ADC), genomic (single nucleotide variants (SNV), tumor mutational burden (TMB), % genome altered), and gene expression signatures to correlate to response (% vol reduction at 6 mos., most recent PSA, documented failure). For significant correlations receiver operating curve (ROC) and T-test analysis was performed. Analyses were performed in 57 biopsy targets/mpMRI lesions from 29 patients (median follow-up 3.8 yrs). Six patients (21%) had NCCN intermediate and 23 patients (79%) had high risk disease. On 6 mos mpMRI, % tumor response ranged from -6 to 100% (median 89.62%). %response correlated to baseline vol (p=0.0001). At last follow-up, 3 patients recurred (Phoenix criteria or biopsy proven local recurrence). MVA of genomic and clinical variables an association of non-synonymous SNV (ns-SNV) frequency with mpMRI response (OR 2.18, p<0.05), but no association of tumor response to TMB or % genome altered. On ROC analysis, # of ns-SNV (AUC=0.63) was associated with greater mpMRI response (threshold 27.5: p=1x10-9). MVA demonstrated positive correlation of the Decipher signature score with pre-treatment PSA and Gleason. Both the PORTOS and Decipher scores highly correlated with mpMRI %response (p<0.001), whereas PORTOS score also showed an associated with post-treatment PSA (p<0.05). T-test analysis confirmed the significant association (p < 0.001) of DECIPHER and PORTOS on improved PSA response. Integration of mpMRI, clinicopathologic, and molecular features may allow additional stratification of patients with intermediate and high-risk prostate cancer based on predicted response. Validation in larger cohorts will provide support for this approach.