Repertoire-scale determination of class II MHC peptide binding via yeast display improves antigen prediction

Abstract

CD4 + helper T cells contribute important functions to the immune response during pathogen infection and tumor formation by recognizing antigenic peptides presented by class II major histocompatibility complexes (MHC-II). While many computational algorithms for predicting peptide binding to MHC-II proteins have been reported, their performance varies greatly. Here we present a yeast-display-based platform that allows the identification of over an order of magnitude more unique MHC-II binders than comparable approaches. These peptides contain previously identified motifs, but also reveal new motifs that are validated by in vitro binding assays. Training of prediction algorithms with yeast-display library data improves the prediction of peptide-binding affinity and the identification of pathogen-associated and tumor-associated peptides. In summary, our yeast-display-based platform yields high-quality MHC-II-binding peptide datasets that can be used to improve the accuracy of MHC-II binding prediction algorithms, and potentially enhance our understanding of CD4 + T cell recognition.

Machine learning optimization of peptides for presentation by class II MHCs

“…Prediction algorithms have been utilized to predict peptide-MHC binding. Recent strides in algorithmic performance have been enabled by advances in computational methods ( Chen et al , 2019 ; O’Donnell et al , 2020 ; Racle et al , 2019 ; Reynisson et al , 2020 ; Zeng and Gifford, 2019 ) and the development of new methodologies for generating training data, such as mono-allelic mass spectrometry ( Abelin et al , 2019 , 2017 ; Sarkizova et al , 2020 ) and yeast display ( Rappazzo et al , 2020 ). With the help of these tools, peptide vaccines with constituent peptides computationally selected for the ability to be displayed by MHCs have been utilized to amplify T cell responses and proven clinically successful for patients with cancer after eliciting CD8+ and CD4+ T cell responses ( Abelin et al , 2017 ; Hu et al , 2018 ; Ott et al , 2017 ).…”

“…For our objective function, we use predictions from the PUFFIN peptide-MHC binding model ( Zeng and Gifford, 2019 ) trained on peptide-binding data from a MHCII yeast display platform ( Rappazzo et al , 2020 ). PUFFIN uses an ensemble of deep residual networks to quantify its uncertainty about its predictions, while achieving state-of-the-art performance on MHCII-binding prediction tasks ( Zeng and Gifford, 2019 ).…”

High resolution profiling of MHC-II peptide presentation capacity, by Mammalian Epitope Display, reveals SARS-CoV-2 targets for CD4 T cells and mechanisms of immune-escape

“…This is exemplified in this study by the very immunogenic membrane protein peptide M 146-165 , recognized by TCR091 in the context of DRB1*11:01 and shown by MEDi to be also presentable by several other HLAs, also not predicted by netMHCIIpan. However, the information gained from MEDi can support further training of predictive models similar to Rappazzo et al 13 .…”

“…In this respect, the recently improved NetMHCIIpan4 shows better performance than conventional binding prediction algorithms 12 , but is accurate only for a limited number of alleles, owing to the lack of suitable peptide datasets for training. To circumvent this, a recently published study improved algorithm performance using yeast-display peptide libraries 13 . Still, there is a big gap from the several HLAs with high-quality in-silico prediction scores and the thousands of unique HLA alleles present in the human population.…”

MS-Based HLA-II Peptidomics Combined With Multiomics Will Aid the Development of Future Immunotherapies

“…HLA-II peptidome profiling is accomplished through LC-MS/MS sequencing of peptides obtained from either multi-or mono-allelic systems. Other methods for HLA-II peptidome profiling include yeast display and peptide microarray systems 26,27 . Mono-allelic systems have only one HLA heterodimer J o u r n a l P r e -p r o o f expressed or purified to ensure that all peptides can be assigned to a single heterodimer.…”