High-throughput identification of MHC class I binding peptides using an ultradense peptide array

Abstract

16Rational vaccine development and evaluation requires identifying and measuring the magnitude 17 of epitope-specific CD8 T cell responses. However, conventional CD8 T cell epitope discovery 18 methods are labor-intensive and do not scale well. Here, we accelerate this process by using an 19 ultradense peptide array as a high-throughput tool for screening peptides to identify putative novel 20 epitopes. In a single experiment, we directly assess the binding of four common Indian rhesus 21 macaque MHC class I molecules -Mamu-A1*001, -A1*002, -B*008, and -B*017 -to 22 approximately 61,000 8-mer, 9-mer, and 10-mer peptides derived from the full proteomes of 82 23 simian immunodeficiency virus (SIV) and simian-human immunodeficiency virus (SHIV) isolates. 24Many epitope-specific CD8 T cell responses restricted by these four MHC molecules have already 25 been identified in SIVmac239, providing an ideal dataset for validating the array; up to 64% of 26 these known epitopes are found in the top 192 SIVmac239 peptides with the most intense MHC 27 binding signals in our experiment. To assess whether the peptide array identified putative novel 28 CD8 T cell epitopes, we validated the method by IFN-γ ELISPOT assay and found three novel 29 peptides that induced CD8 T cell responses in at least two Mamu-A1*001-positive animals; two 30 of these were validated by ex vivo tetramer staining. This high-throughput identification of 31 peptides that bind class I MHC will enable more efficient CD8 T cell response profiling for vaccine 32 development, particularly for pathogens with complex proteomes where few epitope-specific 33 responses have been defined. 34 35 3 Introduction 36 CD8 T cell receptors typically recognize epitopes -linear peptides of 8-11 amino acids -37 presented by major histocompatibility complex (MHC) class I molecules. These peptides are 38 usually derived from self proteins that under normal circumstances do not induce an immune 39 response. Cancerous cells, transplanted cells, and infected cells, however, present non-self 40 peptides that are recognized as foreign by CD8 T cells (1). CD8 T cell recognition of foreign 41 epitopes, including those derived from viral proteins, along with concurrent recognition of 42 costimulatory molecules on the antigen-presenting cell, leads to T cell activation and production 43 of cytotoxic compounds that trigger apoptosis in the affected cell (2). Polymorphisms in the 44 peptide binding cleft of the MHC glycoprotein determine the variety of peptides each molecule is 45 able to present, and specificity is determined in part by anchor residues in the peptide binding to 46 specific pockets in the binding cleft (3). Identifying which peptides are presented by particular 47 MHC molecules and are likely to elicit epitope-specific CD8 T cell responses is a critical task in 48 designing prophylactic and therapeutic vaccines against infectious diseases and cancers (4, 5). 50Characterizing MHC-peptide binding can be a time-and resource-intensive task. Competitive 51 binding assays, whi...

Positional SHAP (PoSHAP) for Interpretation of Machine Learning Models Trained from Biological Sequences

“…Further work is needed to interpret predictive models to understand general patterns of amino acids that will bind a given MHC I allele. Here we demonstrate that LSTM models can easily learn to perform regression directly from a peptide sequence to that peptide’s affinity to various MHC I alleles (measured by Haj et al 2020) 27 . Our main contribution is a strategy to interpret such models that we term “positional SHAP”, which reveals how each amino acid contributes to binding specifically and generally across all peptide predictions.…”

“…Data used for training and testing the model was obtained from Haj et al 2020 27 , where all possible 8-, 9-, and 10-mer peptides from 82 simian immunodeficiency virus (SIV) and simian-human immunodeficiency virus (SHIV) strains were measured by fluorescent peptide array. The data provided consists of 61,066 entries containing the peptide sequence, the peptide length, and five intensity values from the fluorescence assay for each of the five Mamu alleles tested (A001, A002, A008, B008, and B017).…”

“…These intensity values are a proxy for binding affinity between a peptide and the MHC I allele. From the methods of Haj et al, each intensity is the base 2 logarithm of the median intensity value of five replicates reported for each peptide as measured by an MS200 Scanner at a resolution of 2μm and a wavelength of 635nm 27 . For training and testing of the model, the dataset was randomly split into three categories ( Supplemental Figure 1 ).…”

New analysis pipeline for high-throughput domain-peptide affinity experiments improves SH2 interaction data

“…High-throughput studies have great value, and provide a vast quantity of often never before measured data. These methods have been useful to a wide variety of domain-motif interactions, for example SH3-polyproline interactions (44,45), PDZ domains interacting with C-terminal tails (46)(47)(48), and major histocompatibility complete (MHC) interactions with peptides (49,50). However, just as quickly, errors in these studies propagate rapidly and thereby into research results of other investigators.…”

Scoping review of the applications of peptide microarrays on the fight against human infections

“…Conventional T-cell epitope discovery methods are labour intensive and do not scale well (62). In the current review, two studies (13,62)…”