MHC-I binding affinity derived metrics fail to predict tumor specific neoantigen immunogenicity

Abstract

Tumor-specific antigens emerging through somatic genomic rearrangements, known as neoantigens, play a critical role in current anticancer immunotherapy. They may or may not elicit an immune response when presented on the tumor cell surface bound to the MHC-I molecule, whose strength has been assumed as an indicator of immunogenicity. Several in silico peptide-MHC-I binding affinity predictors are used to prioritize putative immunogenic neoantigens to be experimentally and clinically explored either as biomarkers or targets for anticancer vaccines. This claims for a fair evaluation of such predictors, making essential the development of appropriate databases with experimentally validated, immunogenic/non-immunogenic neoantigens. Thus far, such a database is lacking. We herein present ITSNdb, a new and curated immunogenic neoantigen database and use it to benchmark current neoantigen immunogenicity predictors. Benchmark results failed to support the application of the predicted peptide- MHC-I binding affinity or its derived metrics as a tool to estimate neoantigen immunogenicity and the tumor neoantigen burden as an immunotherapy response biomarker. Moreover, binding affinity based immunogenicity definition leads to identifying wild-type peptide counterparts as predictors of immunotherapy response. We demonstrate that MHC-I binding affinity is insufficient to define neoantigen immunogenicity, despite being necessary for neoantigen tumor cell presentation suggesting that a paradigm shift for the emergence of new rules to identify immunogenic neoantigens is required.