A microRNA-based signature predicting immune-related toxicity (iRAEs) to pembrolizumab in prostate cancer patients.

Abstract

Listen

Translate article

269 Background: The need to predict immune-related adverse events (iRAEs) for patients receiving checkpoint inhibitors is a pressing issue, with up to 25-30% Grade 2 or higher toxicity with single agent therapy, regardless of cancer type. Our group is studying a novel class of inherited mutations that disrupt the normal regulatory process of microRNAs as they function to control the genome. We have previously applied a panel of these mutations to predict iRAEs in melanoma and NSCLC patients receiving anti-PD1/anti-PDL1 therapy. We hypothesize that our mutations will predict iRAEs across cancer types. Here we have investigated their ability to predict iRAEs in a cohort of recurrent or advanced prostate cancer patients receiving Pembrolizumab (Pembro). Methods: We previously trained several statistical classifiers on a set of 86 melanoma and NSCLC patients evaluated for iRAEs. Subjects were classified as experiencing high toxicity (grade 2 or more) versus low toxicity (less than 2). We next tested the performance of the identified biomarkers on a set of 40 prospectively-treated prostate cancer patients who had received at least three cycles of Pembro, 200mg every 3 weeks. Samples were collected via cheek swab and germ-line DNA was isolated for testing. Predictive models of high-grade toxicity were based on 49 common markers between training and test samples. Classifiers were built comparing: classification trees, LASSO-regularized logistic regression, and random forests. All models were tuned at training using leave-one-out cross validation. The final extrapolation test errors were estimated on the prostate cohort as validation data not used in training. Results: Within the melanoma and NSCLC training sample we estimate toxicity with 71% accuracy (85% sensitivity, 43% specificity). We found that our biomarker panel when applied to prostate cancer patients extrapolates quite well, where our biomarkers predict toxicity with 68% accuracy (82% sensitivity, 38% specificity). Conclusions: We have identified a panel of microRNA-based, germ-line biomarkers which predict the risk of iRAEs. This panel appears to predict iRAEs across cancer types, including prostate cancer.