T-cell-based immunotherapies aim to treat various malignancies by targeting intracellular protein-derived peptides processed by the proteasome and presented on MHC class I molecules. A critical step in the development of these therapies is the accurate selection of target MHC ligands. This selection is often guided by computational prediction algorithms, particularly in the context of vaccine design, rather than experimental validation of antigen presentation. However, these algorithms frequently fail to accurately reflect the complexity of natural antigen processing and presentation, potentially leading to false-positive results and misdirected therapeutic strategies. To systematically evaluate these predictions, this thesis employs a COS-7 monoallelic cell system in liason with immunopeptidomics to screen 150 combinations of individual neoantigens/tumor-associated antigens and HLA alleles. Surprisingly, only 21.6% of predicted neoepitopes were experimentally validated, revealing a 78% false-positive rate of computational prediction algorithms. Similarly, these computational tools are frequently used to establish links between public neoepitope presentation and downstream biological effects including immune evasion through HLA loss of heterozygosity (HLA-LOH). However, this link was also disrupted in 77% of HLA-LOH cases, where the predicted neoepitopes showed no evidence of presentation on the corresponding HLA alleles, thereby challenging the assumption that HLA-LOH is mainly driven by public neoantigen presentation. Importantly, we identified 24 novel public neoepitopes across 11 HLA alleles, derived from recurrent driver mutations (e.g., TP53, EGFR) and resistance-related alterations (e.g., CTNNB1, ESR1), alongside 13 novel tumor-associated epitopes (e.g., derived from MAGE-A4, PRAME, CT83, NY-ESO-1) across 4 common HLA alleles. Two epitopes, KVDELAHFL (MAGE-A4) and KVLEHVVRV (MAGE-A4), were each presented on two common HLA alleles, offering a promising strategy to broaden patient eligibility for TCR-based therapies. Furthermore, when testing sequences used for personalized mRNA neoepitope vaccines, we observed that clinical responses were independent of the presentation status of the neoepitopes, highlighting the risk of measuring false-positive responses during T-cell monitoring. Even a presumably neoepitope-specific TCR clone from a neoantigen vaccination study could not be linked to a successfully presented neoepitope. In contrast, we observed cross-reactivity of the TCR’s targets with a CMV-specific epitope as a potential explanation for the T-cell response measured during T-cell monitoring. Overall, these data challenge current assumptions about neoantigen responses, immune evasion, and the reliability of computational predictions.
