Correlating HLA variants and the emergence of immune-related adverse events (irAEs) from ipilimumab/nivolumab (I/N) in patients (pts) treated on CheckMate 214

659 Background: Combination therapy with I/N can achieve long-term benefit in a sizable proportion of RCC pts, but broad application has been challenged by a high incidence of irAEs. Host features affecting antigen presentation are candidate determinants of irAE pathogenesis, and isolated reports have linked Human Leukocyte Antigen (HLA) variations to organ- specific irAEs. We analyzed data for pts treated on a phase 3 trial to investigate association of class 1 HLA variants with development of irAEs from I/N and develop a tool to predict risk for toxicity. Methods: We analyzed germline HLA + clinical data for 472 pts receiving I/N vs sunitinib (SUT). Based on allelic variants for HLA-A and B pts were categorized into 12 established HLA "Super-Types" (ST), sets of HLA variants with largely overlapping peptide binding specificity. We conducted uni- and multivariate (UV; MV) Cox proportional hazards regression to correlate HLA-ST variables with time to treatment-related AEs (TTrAE, grade 2+) in I/N pts using the Kaplan Meier method. Several MV HLA-ST models were developed on a discovery set (DISC; 2/3 random sample of I/N pts without replacement ) and compared using model concordance (c-index) on a validation set (VAL; remaining 1/3). Model coefficients were used to create a “HLA-ST score” that could be computed for individual pts. Results: 235 pts and 237 pts received I/N vs SUT and had available HLA data, respectively. On UV analysis for I/N, HLA-B07 ST had protective association TTrAE (HR= 0.65, 95% CI: 0.46,0.90; p= 0.010), while B62 associated adversely (HR=1.64, 95% CI: 1.12, 2.40; p=0.014). Relevant to the development of our model we identified interactions between several pairs of HLA ST. The HLA ST model with best performance in DISC (c-index= 0.606) and VAL (c-index=0.595) integrates B07, B62, A01, B08, and two interactions: B07-B08 and B07-A01. The model-generated HLA-ST score was significantly associated with TTrAE after adjusting for race, BMI, region, PDL1 status, and MSKCC risk score (DISC p= vs < median had significantly different TTrAE both in DISC (p=0.004) and VAL (p=0.009); again, no difference was observed in the SUT arm (p=0.5). The weighted HLA-ST score had no association with PFS or OS for I/N pts. Conclusions: In this large sample of I/N-treated patients, class I HLA variants were associated with the risk of developing irAEs. We developed a HLA ST based score that correlated with treatment toxicity independent of relevant clinical and demographic features. No association with TTrAE was seen in SUT-treated pts. These results highlight the potential of characterizing germline features to predict immune related toxicity upfront and deserves further study. Clinical trial information: NCT02231749 .