In silico immunogenicity assessment for sequences containing unnatural amino acids: A method using existing in silico algorithm infrastructure and a vision for future enhancements

Abstract

Thein silicoprediction of T cell epitopes within any peptide or biologic drug candidate serves as an important first step for assessing immunogenicity. T cell epitopes bind human leukocyte antigen (HLA) by a well-characterized interaction of amino acid side chains and pockets in the HLA molecule binding groove. Immunoinformatics tools, such as the EpiMatrix algorithm, have been developed to screen natural amino acid sequences for peptides that will bind HLA. In addition to commonly occurring in synthetic peptide impurities, unnatural amino acids (UAA) are also often incorporated into novel peptide therapeutics to improve properties of the drug product. To date, the HLA binding properties of peptides containing UAA are not accurately estimated by most algorithms. Both scenarios warrant the need for enhanced predictive tools. The authors developed anin silicomethod for modeling the impact of a given UAA on a peptide’s likelihood of binding to HLA and, by extension, its immunogenic potential.In silicoassessment of immunogenic potential allows for risk-based selection of best candidate peptides in further confirmatoryin vitro, ex vivo,andin vivoassays, thereby reducing the overall cost of immunogenicity evaluation. Examples demonstratingin silicoimmunogenicity prediction for product impurities that are commonly found in formulations of the generic peptides teriparatide and semaglutide are provided. Next, this article discusses how HLA binding studies can be used to estimate the binding potentials of commonly encountered UAA and “correct”in silicoestimates of binding based on their naturally occurring counterparts. As demonstrated here, thesein vitrobinding studies are usually performed with known ligands which have been modified to contain UAA in HLA anchor positions. An example using D-amino acids in relative binding position 1 (P1) of the PADRE peptide is presented. As more HLA binding data become available, new predictive models allowing for the direct estimation of HLA binding for peptides containing UAA can be established.