Prediction of pan-solid tumor pembrolizumab benefit by integrating tumor mutation and gene expression profiling

Abstract

Pembrolizumab is approved in many advanced solid tumor types, however predictive biomarkers and the proportion of pembrolizumab-benefiting patients vary. Biomarkers beyond PD-L1 immunohistochemistry, microsatellite instability (MSI) status, and tumor mutation burden (TMB) may improve benefit prediction. Here, leveraging treatment data (time to next treatment [TTNT]) and comprehensive genomic and quantitative transcriptomic profiling on routine tumor tissue from 708 patients (24 tumor types) collected in an ongoing observational trial (NCT03061305), we report a multivariate, integrative predictor of pan-solid tumor pembrolizumab benefit. The Immune Response Score (IRS) model, which includes TMB and quantitative PD-1, PD-L2, ADAM12 and CD4 RNA expression, was confirmed as predictive through comparison of pembrolizumab TTNT with previous chemotherapy TTNT in a subset of 166 patients treated with both. Applying IRS to the entire NCT03061305 cohort (n=25,770 patients), 13.2-30.7% of patients (2.2-9.6% of patients outside of pembrolizumab approved tumor types [including TMB-High and MSI-High]) are predicted to benefit substantially from pembrolizumab. Hence, if prospectively validated, the IRS model may improve pembrolizumab benefit prediction across approved tumor types including patients outside of currently approved indications.