Molecular Profiling of Cohorts of Tumor Samples to Guide Clinical Development of Pembrolizumab as Monotherapy

Abstract Purpose: Molecular profiling of large databases of human tumor gene expression profiles offers novel opportunities for informing decisions in clinical development programs. Experimental Design: Gene expression profile of programmed death ligand 1 (PD-L1) was explored in a dataset of 16,000 samples, including approximately 4,000 metastatic tumors, across >25 tumor types prevalent in the United States, looking for new indications for the programmed death 1 (PD-1) inhibitor pembrolizumab. PD-L1 expression was highly concordant with several genomic signatures indicative of immune-inflamed tumor microenvironment. Prevalence of activated immune-inflamed tumors across all tumor types was explored and used to rank tumor types for potential response to pembrolizumab monotherapy. Results: The analysis yielded 3 tiers of indications in which high levels of PD-L1 and immune-inflamed signatures were found in up to 40% to 60%, 20% to 40%, and 0% to 20% of tumors. Tier 1 contained novel indications known at the time of analysis to be responsive to PD-1 checkpoint blockade in the clinic (such as melanoma and non–small cell lung cancer), as well as indications not studied in the clinic previously, including microsatellite instability–high colorectal, head and neck, bladder, and triple-negative breast cancers. Complementary analysis of an Asian/Pacific cancer dataset (gastric cancer) revealed high prevalence of immune-inflamed tumors in gastric cancer. These data contributed to prioritization of these indications for clinical development of pembrolizumab as monotherapy. Conclusions: Data highlight the value of molecular profiling in identifying populations with high unmet needs with potentially favorable response characteristics and accelerating development of novel therapies for these patients. See related commentary by Mansfield and Jen, p. 1443