Computational identification of cross-reactive TCR epitopes with ARDitox

Cellular immunotherapies, such as those utilizing T lymphocytes expressing native or engineered T cell receptors (TCRs), have demonstrated therapeutic efficacy. However, some engineered high-affinity TCRs have caused fatal off-target immunotoxicity due to targeting epitopes later found to be expressed by both tumor cells and healthy tissues. Unfortunately, TCRs can be cross-reactive to epitopes with highly distinct sequences, making prediction difficult, and the exquisite sequence specificity of TCRs means that safety studies in mice miss human-specific epitopes.

To address this issue, we developed ARDitox, a novel in silico method based on computational immunology and artificial intelligence (AI) for predicting and analyzing potential TCR off-target toxicities. We tested the performance of ARDitox on four TCRs reported to target tumor-associated antigens, two of which are known to cause clinical immunotoxicity (MAGEA3_112-120 and MAGEA3_168-176 epitopes), one of which has experimentally identified off-target antigens (AFP_158-166 epitope), and the last one for which no cross-reactive epitopes are known (NY-ESO-1_157-165).

ARDitox confirmed the previously identified immunotoxic epitopes. We then expanded our analyses to a novel TCR targeting the tumor-associated antigen NLGN4X, frequently upregulated in gliomas. For this target, ARDitox identified a cross-reactive peptide that would not have been found using mouse models, highlighting the value of our computational approach.

Our findings underscore the value of the ARDitox in silico method for the early and reliable identification of off-target epitopes for further preclinical evaluation. This platform strongly supports the development of safer TCR-mediated immunotherapies.