Abstract 571: Meta-analysis of genomic predictors of response to immune checkpoint therapy in metastatic melanoma

Abstract

Abstract Introduction: Immune checkpoint therapies benefit a subset of patients with metastatic melanoma, but ability to predict clinical outcomes is limited. This meta-analysis of genomic predictors of outcomes to aPD1 and aCTLA4 in melanoma combines 220 sequenced tumors from 3 published cohorts, aiming to validate existing hypotheses regarding response to immune checkpoint therapies and discover new relationships with greater power. Methods: Genomic data and clinical annotations from published cohorts were analyzed with standardized pipelines for somatic variant calling, mutational signature deconvolution, and neoantigen prediction. Patients were stratified into clinical benefit (CB) and no clinical benefit (NCB) as described in Van Allen et al. 2015. Analyses were repeated using two other published response metrics (CB=PFS>6 months; CB=CR or PR). Results: Nonsynonymous mutational burden was significantly higher in CB vs. NCB using all 3 response metrics, though significance was less pronounced using PFS alone (p<0.01 vs. p<0.0001; Wilcoxon rank sum), partially due to 3 patients with high mutational burden who experienced PR for <6 months, potentially representing early acquired rather than intrinsic resistance. To assess the impact of mutational processes contributing to overall mutational burden, we used a non-negative matrix factorization framework to infer mutational activity in tumors from 6 signatures previously seen in melanoma: aging (S1), T>C substitutions (S5), mismatch repair (S6), alkylating agents (S11), UV (S7), and T>G substitutions (S17). The proportion of mutations in S7 or S11 was positively correlated with mutational burden (Spearman’s rho=0.66), while S5 and S1 were anti-correlated (rho=-0.62). In a multivariate logistic model, S7 and S11 activity were independent predictors of CB adjusting for mutational load (p<0.05), with the sum of S7 and S11 activity being a strong predictor (p<0.001). Of the patients with low mutational burden (<median) with CB, a large majority (23/29) had >1/2 of mutations in S7 or S11, compared to only 36/71 of low-mutation NCB (p<0.01; Pearson’s chi-squared). Neoantigen burden was strongly correlated with mutational burden, and did not improve ability to predict CB. In examining mutations in specific genes, >500 genes were mutated more frequently in either CB or NCB (p<0.05, Fisher’s exact). Restricting analysis to genes recurrently mutated in cancer and correcting for patient mutational burden by permutation, nonsynonymous mutations in ACSL3 and MET and truncating alterations in ARID2 were significantly enriched in CB. Conclusions: In this meta-analysis of 220 patients, harmonized clinical and whole exome analysis confirmed that mutational burden correlates with CB from aPD1 and aCTLA4 therapy, while mutational signatures and alterations in specific genes potentially provide additional predictive power. Citation Format: Diana Miao, David Liu, Daniel Keliher, Sachet Shukla, Bastian Schilling, Claire Margolis, Alicia Smart, Levi Garraway, Stephen Hodi, Dirk Schadendorf, Eliezer M. Van Allen. Meta-analysis of genomic predictors of response to immune checkpoint therapy in metastatic melanoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 571. doi:10.1158/1538-7445.AM2017-571