Using autoantibody signatures to predict immunotherapy discontinuation in melanoma patients.

Abstract

3069 Background: Immune checkpoint inhibitors (ICIs), e.g., ipilimumab (IPI) and/or nivolumab (NIVO), produce durable survival benefit in a substantial proportion of melanoma patients but can also induce severe immune-related adverse events (irAEs) requiring treatment discontinuation. There is no biomarker to predict irAEs in ICI-treated melanoma patients. Given the similar clinical manifestation between irAEs and autoimmune disorders, we hypothesized that a subset of patients possess a subclinical baseline predisposition to developing irAEs that is characterized by specific autoantibodies (autoAbs). Methods: Pre-treatment melanoma patient sera from the CheckMate-238 Phase III trial of adjuvant IPI vs. NIVO were used for autoAb profiling with HuProt proteomic arrays (CDI Labs). The outcome of interest is to predict toxicity events that caused treatment discontinuation. For each treatment arm, we allocated patients to training and testing datasets in a 3:1 ratio. We calculated the area under the curve (AUC) of the receiver operating characteristic curve to select a probability threshold, which was applied to the testing dataset to assess accuracy, sensitivity, and specificity. Functional enrichment among autoAb protein targets was assessed using Metascape. Results: There were 707 irAEs among 597 patients (IPI = 423, NIVO = 174), of which 355 required treatment discontinuation (IPI = 287, NIVO = 68). In the training sets, we identified a 170 autoAbs signature consisting of 102 autoAbs for IPI treatment and 68 autoAbs for NIVO treatment. In the independent testing set, the signatures showed AUC of 0.85 (0.78, 0.92), 82% sensitivity, 78% specificity, and overall accuracy of 81% to predict IPI discontinuation, and AUC of 0.87 (0.74, 0.99), 75% sensitivity, 97% specificity, and overall accuracy of 88% to predict NIVO discontinuation. Enrichment of nuclear lumen-associated protein targets was identified among autoAb signatures that predict IPI or NIVO discontinuation. Conclusions: The identified signature within a large Phase III trial cohort highlights the potential utility of pre-treatment autoAbs for prediction of patients at high risk of developing irAEs in the adjuvant setting necessitating treatment termination. We are currently validating and refining toxicity-associated autoAb signatures with the goal of developing a Clinical Laboratory Improvement Amendments (CLIA)-certified assay to enable clinicians to optimize immunotherapy delivery and patient selection.