XTEP: A Structure-Informed Algorithm to Predict T Cell Cross-Reactivity 2267697

Abstract

Abstract Introduction Each T cell is able to recognize multiple unique peptides presented on MHC molecules in a phenomenon referred to as Cross-Reactivity. Cross-reactivity is important in heterologous immunity and autoimmunity and can be both beneficial and deleterious in both contexts. For example, infection can prime T cells to recognize different pathogens that have similar proteins, which can provide protection against disease, or to recognize similar self-proteins, which can induce autoimmune disease. To elicit cross-reactivity from the same T cell, two peptides need to be chemically similar enough to both bind the same MHC and be recognized by the same T cell receptor; however, in certain instances peptides that appear chemically different can elicit a cross-reactive response. Methods We sought out to develop an approach that can predict cross-reactivity using only sequence information about the peptides involved. To develop this approach, we leverage structural information to create an algorithm that emphasizes key epitope residues for homology comparison. Results We show that combining TCR-pMHC structural information with machine learning MHC binding predictions provides a reliable method of predicting cross-reactivity. We tested our algorithm using datasets measuring T cell cross-reactivity between viral peptides in cases both where the presenting MHC is known and in cases where it is not known. Our prediction approach, Cross-reactive T cell Epitope Predictor (XTEP), is TCR-sequence independent and MHC agnostic. Conclusion Our approach can be used to predict if a T cell epitope will be cross-reactive, which can be important in contexts such as vaccine design. Additionally, XTEP can predict epitopes that can be used to show cross-reactivity is occurring —for example when evaluating similarity between self-epitopes and viral peptides in autoimmunity. Our findings imply this approach is more predictive of cross-reactivity than predicting MHC binding alone or comparing full peptide similarity. Funding Source NIH T32 AI132152 Topic Categories Computational and Systems Immunology (COMP)