Whole-blood RNA transcript-based models can predict clinical response in two large independent clinical studies of patients with advanced melanoma treated with the checkpoint inhibitor, tremelimumab

BACKGROUND Tremelimumab is an antibody that blocks CTLA-4 and demonstrates clinical efficacy in a subset of advanced melanoma patients. An unmet clinical need exists for blood-based response-predictive gene signatures to facilitate clinically effective and cost-efficient use of such immunotherapeutic interventions. METHODS Peripheral blood samples were collected in PAXgene® tubes from 210 treatment-naïve melanoma patients receiving tremelimumab in a worldwide, multicenter phase III study (discovery dataset). A central panel of radiologists determined objective response using RECIST criteria. Gene expression for 169 mRNA transcripts was measured using quantitative PCR. A 15-gene pre-treatment response-predictive classifier model was identified. An independent population (N = 150) of refractory melanoma patients receiving tremelimumab after chemotherapy enrolled in a worldwide phase II study (validation dataset). The classifier model, using the same genes, coefficients and constants for objective response and one-year survival after treatment, was applied to the validation dataset. RESULTS A 15-gene pre-treatment classifier model (containing ADAM17, CDK2, CDKN2A, DPP4, ERBB2, HLA-DRA, ICOS, ITGA4, LARGE, MYC, NAB2, NRAS, RHOC, TGFB1, and TIMP1) achieved an area under the curve (AUC) of 0.86 (95% confidence interval 0.81 to 0.91, p < 0.0001) for objective response and 0.6 (95% confidence interval 0.54 to 0.67, p = 0.0066) for one-year survival in the discovery set. This model was validated in the validation set with AUCs of 0.62 (95% confidence interval 0.54 to 0.70 p = 0.0455) for objective response and 0.68 for one-year survival (95% confidence interval 0.59 to 0.75 p = 0.0002). CONCLUSIONS To our knowledge, this is the largest blood-based biomarker study of a checkpoint inhibitor, tremelimumab, which demonstrates a validated pre-treatment mRNA classifier model that predicts clinical response. The data suggest that the model captures a biological signature representative of genes needed for a robust anti-cancer immune response. It also identifies non-responders to tremelimumab at baseline prior to treatment.