Predictive Biomarkers and Personalized Medicine

Purpose: Immunomodulatory drugs differ in mechanism-of-action from directly cytotoxic cancer therapies. Identifying factors predicting clinical response could guide patient selection and therapeutic optimization. Experimental Design: Patients (N 1⁄4 41) with melanoma, non–small cell lung carcinoma (NSCLC), renal cell carcinoma (RCC), colorectal carcinoma, or castration-resistant prostate cancer were treated on an early-phase trial of anti–PD-1 (nivolumab) at one institution and had evaluable pretreatment tumor specimens. Immunoarchitectural features, including PD-1, PD-L1, and PD-L2 expression, patterns of immune cell infiltration, and lymphocyte subpopulations, were assessed for interrelationships and potential correlations with clinical outcomes. Results: Membranous (cell surface) PD-L1 expression by tumor cells and immune infiltrates varied significantly by tumor type and was most abundant in melanoma, NSCLC, and RCC. In the overall cohort, PD-L1 expression was geographically associated with infiltrating immune cells (P < 0.001), although lymphocyte-rich regions were not always associated with PD-L1 expression. Expression of PD-L1 by tumor cells and immune infiltrates was significantly associated with expression of PD-1 on lymphocytes. PD-L2, the second ligand for PD-1, was associated with PD-L1 expression. Tumor cell PD-L1 expression correlated with objective response to anti–PD-1 therapy, when analyzing either the specimen obtained closest to therapy or the highest scoring sample amongmultiple biopsies from individual patients. These correlations were stronger than borderline associations of PD-1 expression or the presence of intratumoral immune cell infiltrates with response. Conclusions: Tumor PD-L1 expression reflects an immune-active microenvironment and, while associated other immunosuppressive molecules, including PD-1 and PD-L2, is the single factor most closely correlated with response to anti–PD-1 blockade. Clin Cancer Res; 1–11. 2014 AACR.