Identification and characterisation of peptide binding motifs of six autoimmune disease-associated human leukocyte antigen-class I molecules including HLA-B*39:06

Structural and biochemical analysis of highly similar HLA-B allotypes differentially associated with type 1 diabetes

Type 1 diabetes (T1D) is an autoimmune disease involving T cell-mediated destruction of the insulin-producing beta cells in the pancreatic islets of Langerhans. CD8+ T cells, responding to beta cell peptides presented by class I major histocompatibility complex (MHC) molecules, are important effectors leading to beta cell elimination. Human leukocyte antigen (HLA) B∗39:06, B∗39:01, and B∗38:01 are closely related class I MHC allotypes that nonetheless show differential association with T1D. HLA-B∗39:06 is the most predisposing of all HLA class I molecules and is associated with early age at disease onset. B∗39:01 is also associated with susceptibility to T1D, but to a lesser extent, though differing from B∗39:06 by only two amino acids. HLA-B∗38:01, in contrast, is associated with protection from the disease. Upon identifying a peptide that binds to both HLA-B∗39:06 and B∗39:01, we determined the respective X-ray structures of the two allotypes presenting this peptide to 1.7 Å resolution. The peptide residues available for T cell receptor contact and those serving as anchors were identified. Analysis of the F pocket of HLA-B∗39:06 and B∗39:01 provided an explanation for the distinct peptide C terminus preferences of the two allotypes. Structure-based modeling of the protective HLA-B∗38:01 suggested a potential reason for its peptide preferences and its reduced propensity to present 8-mer peptides compared to B∗39:06. Notably, the three allotypes showed differential binding to peptides derived from beta cell autoantigens. Taken together, our findings should facilitate identification of disease-relevant candidate T cell epitopes and structure-guided therapeutics to interfere with peptide binding.

APEX-pHLA: A novel method for accurate prediction of the binding between exogenous short peptides and HLA class I molecules

Human leukocyte antigen (HLA) molecules play critically significant role within the realm of immunotherapy due to their capacities to recognize and bind exogenous antigens such as peptides, subsequently delivering them to immune cells. Predicting the binding between peptides and HLA molecules (pHLA) can expedite the screening of immunogenic peptides and facilitate vaccine design. However, traditional experimental methods are time-consuming and inefficient. In this study, an efficient method based on deep learning was developed for predicting peptide-HLA binding, which treated peptide sequences as linguistic entities. It combined the architectures of textCNN and BiLSTM to create a deep neural network model called APEX-pHLA. This model operated without limitations related to HLA class I allele variants and peptide segment lengths, enabling efficient encoding of sequence features for both HLA and peptide segments. On the independent test set, the model achieved Accuracy, ROC_AUC, F1, and MCC is 0.9449, 0.9850, 0.9453, and 0.8899, respectively. Similarly, on an external test set, the results were 0.9803, 0.9574, 0.8835, and 0.7863, respectively. These findings outperformed fifteen methods previously reported in the literature. The accurate prediction capability of the APEX-pHLA model in peptide-HLA binding might provide valuable insights for future HLA vaccine design.

Polymorphism in F pocket affects peptide selection and stability of type 1 diabetes-associated HLA-B39 allotypes

HLA-B*39:06, HLA-B*39:01, and HLA-B*38:01 are closely related HLA allotypes differentially associated with type 1 diabetes (T1D) risk and progression. B*39:06 is highly predisposing, while B*39:01 and B*38:01 are weakly predisposing and protective allotypes, respectively. Here, we aimed to decipher molecular mechanisms underlying the differential association of these allotypes with T1D pathogenesis. We addressed peptide binding and conformational stability of HLA-B allotypes using computational and experimental approaches. Computationally, we found that B*39:06 and B*39:01 allotypes had more rigid peptide-binding grooves and were more promiscuous in binding peptides than B*38:01. Peptidomes of B*39:06 and B*39:01 contained fewer strong binders and were of lower affinity than that of B*38:01. Experimentally, we demonstrated that B*39:06 and B*39:01 had a higher capacity to bind peptides and exit to the cell surface but lower surface levels and were degraded faster than B*38:01. In summary, we propose that promiscuous B*39:06 and B*39:01 may bind suboptimal peptides and transport them the cell surface, where such unstable complexes may contribute to the pathogenesis of T1D.

Differences in F pocket impact on HLA I genetic associations with autoimmune diabetes

Introduction

Human leukocyte antigen (HLA) I molecules present antigenic peptides to activate CD8+ T cells. Type 1 Diabetes (T1D) is an auto-immune disease caused by aberrant activation of the CD8+ T cells that destroy insulin-producing pancreatic β cells. Some HLA I alleles were shown to increase the risk of T1D (T1D-predisposing alleles), while some reduce this risk (T1D-protective alleles).

Methods

Here, we compared the T1D-predisposing and T1D-protective allotypes concerning peptide binding, maturation, localization and surface expression and correlated it with their sequences and energetic profiles using experimental and computational methods.

Results

T1D-predisposing allotypes had more peptide-bound forms and higher plasma membrane levels than T1D-protective allotypes. This was related to the fact that position 116 within the F pocket was more conserved and made more optimal contacts with the neighboring residues in T1D-predisposing allotypes than in protective allotypes.

Conclusion

Our work uncovers that specific polymorphisms in HLA I molecules potentially influence their susceptibility to T1D.

Elucidation of binding mechanism, affinity, and complex structure between mWT1 tumor-associated antigen peptide and HLA-A*24:02

We have applied our advanced computational and experimental methodologies to investigate the complex structure and binding mechanism of a modified Wilms' Tumor 1 (mWT1) protein epitope to the understudied Asian-dominant allele HLA-A*24:02 (HLA-A24) in aqueous solution. We have applied our developed multicanonical molecular dynamics (McMD)-based dynamic docking method to analyze the binding pathway and mechanism, which we verified by comparing the highest probability structures from simulation with our experimentally solved x-ray crystal structure. Subsequent path sampling MD simulations elucidated the atomic details of the binding process and indicated that first an encounter complex is formed between the N-terminal's positive charge of the 9-residue mWT1 fragment peptide and a cluster of negative residues on the surface of HLA-A24, with the major histocompatibility complex (MHC) molecule preferring a predominantly closed conformation. The peptide first binds to this closed MHC conformation, forming an encounter complex, after which the binding site opens due to increased entropy of the binding site, allowing the peptide to bind to form the native complex structure. Further sequence and structure analyses also suggest that although the peptide loading complex would help with stabilizing the MHC molecule, the binding depends in a large part on the intrinsic affinity between the MHC molecule and the antigen peptide. Finally, our computational tools and analyses can be of great benefit to study the binding mechanism of different MHC types to their antigens, where it could also be useful in the development of higher affinity variant peptides and for personalized medicine.

Database search engines and target database features impinge upon the identification of post-translationally cis-spliced peptides in HLA class I immunopeptidomes

Unconventional epitopes presented by HLA class I complexes are emerging targets for T cell targeted immunotherapies. Their identification by mass spectrometry (MS) required development of novel methods to cope with the large number of theoretical candidates. Methods to identify post-translationally spliced peptides led to a broad range of outcomes. We here investigated the impact of three common database search engines - that is, Mascot, Mascot+Percolator, and PEAKS DB - as final identification step, as well as the features of target database on the ability to correctly identify non-spliced and cis-spliced peptides. We used ground truth datasets measured by MS to benchmark methods' performance and extended the analysis to HLA class I immunopeptidomes. PEAKS DB showed better precision and recall of cis-spliced peptides and larger number of identified peptides in HLA class I immunopeptidomes than the other search engine strategies. The better performance of PEAKS DB appears to result from better discrimination between target and decoy hits and hence a more robust FDR estimation, and seems independent to peptide and spectrum features here investigated.

Minimal Crossover between Mutations Associated with Omicron Variant of SARS-CoV-2 and CD8+ T-Cell Epitopes Identified in COVID-19 Convalescent Individuals

There is a growing concern that ongoing evolution of SARS-CoV-2 could lead to variants of concern (VOC) that are capable of avoiding some or all of the multifaceted immune response generated by both prior infection or vaccination, with the recently described B.1.1.529 (Omicron) VOC being of particular interest. Peripheral blood mononuclear cell samples from PCR-confirmed, recovered COVID-19 convalescent individuals (n = 30) infected with SARS-CoV-2 in the United States collected in April and May 2020 who possessed at least one or more of six different HLA haplotypes were selected for examination of their anti-SARS-CoV-2 CD8⁺ T-cell responses using a multiplexed peptide-major histocompatibility complex tetramer staining approach. This analysis examined if the previously identified viral epitopes targeted by CD8⁺ T cells in these individuals (n = 52 distinct epitopes) are mutated in the newly described Omicron VOC (n = 50 mutations). Within this population, only one low-prevalence epitope from the Spike protein, restricted to two HLA alleles and found in 2/30 (7%) individuals, contained a single amino acid change associated with the Omicron VOC. These data suggest that virtually all individuals with existing anti-SARS-CoV-2 CD8⁺ T-cell responses should recognize the Omicron VOC and that SARS-CoV-2 has not evolved extensive T-cell escape mutations at this time.

N-Terminal-Driven Binding Mechanism of an Antigen Peptide to Human Leukocyte Antigen-A*2402 Elucidated by Multicanonical Molecular Dynamic-Based Dynamic Docking and Path Sampling Simulations

We have applied our advanced multicanonical molecular dynamics (McMD)-based dynamic docking methodology to investigate the binding mechanism of an HIV-1 Nef protein epitope to the Asian-dominant allele human leukocyte antigen (HLA)-A*2402. Even though pMHC complex formation [between a Major histocompatibility complex (MHC) class I molecule, which is encoded by an HLA allele, and an antigen peptide] is one of the fundamental processes of the adaptive human immune response, its binding mechanism has not yet been well studied, partially due to the high allelic variation of HLAs in the population. We have used our developed McMD-based dynamic docking method and have successfully reproduced the native complex structure, which is located near the free energy global minimum. Subsequent path sampling MD simulations elucidated the atomic details of the binding process and indicated that the peptide binding is initially driven by the highly positively charged N-terminus of the peptide that is attracted to the various negatively charged residues on the MHC molecule's surface. Upon nearing the pocket, the second tyrosine residue of the peptide anchors the peptide by strongly binding to the B-site of the MHC molecule via hydrophobic driven interactions, resulting in a very strong bound complex structure. Our methodology can be effectively used to predict the bound complex structures between MHC molecules and their antigens to study their binding mechanism in close detail, which would help with the development of new vaccines against cancers, as well as viral infections such as HIV and COVID-19.

Association of Protective HLA-A With HLA-B∗27 Positive Ankylosing Spondylitis

Objectives

To further elucidate the role of the MHC in ankylosing spondylitis by typing 17 genes, searching for HLA-B^∗27 independent associations and assessing the impact of sex on this male biased disease.

Methods

High-confidence two-field resolution genotyping was performed on 310 cases and 2196 controls using an n-1 concordance method. Protein-coding variants were called from next-generation sequencing reads using up to four software programs and the consensus result recorded. Logistic regression tests were applied to the dataset as a whole, and also in stratified sets based on sex or HLA-B^∗27 status. The amino acids driving association were also examined.

Results

Twenty-five HLA protein-coding variants were significantly associated to disease in the population. Three novel protective associations were found in a HLA-B^∗27 positive population, HLA-A^∗24:02 (OR = 0.4, CI = 0.2–0.7), and HLA-A amino acids Leu95 and Gln156. We identified a key set of seven loci that were common to both sexes, and robust to change in sample size. Stratifying by sex uncovered three novel risk variants restricted to the female population (HLA-DQA1^∗04.01, -DQB1^∗04:02, -DRB1^∗08:01; OR = 2.4–3.1). We also uncovered a set of neutral variants in the female population, which in turn conferred strong effects in the male set, highlighting how population composition can lead to the masking of true associations.

Conclusion

Population stratification allowed for a nuanced investigation into the tightly linked MHC region, revealing novel HLA-B^∗27 signals as well as replicating previous HLA-B^∗27 dependent results. This dissection of signals may help to elucidate sex biased disease predisposition and clinical progression.

An effective CTL peptide vaccine for Ebola Zaire Based on Survivors' CD8+ targeting of a particular nucleocapsid protein epitope with potential implications for COVID-19 vaccine design

The 2013-2016 West Africa EBOV epidemic was the biggest EBOV outbreak to date. An analysis of virus-specific CD8+ T-cell immunity in 30 survivors showed that 26 of those individuals had a CD8+ response to at least one EBOV protein. The dominant response (25/26 subjects) was specific to the EBOV nucleocapsid protein (NP). It has been suggested that epitopes on the EBOV NP could form an important part of an effective T-cell vaccine for Ebola Zaire. We show that a 9-amino-acid peptide NP44-52 (YQVNNLEEI) located in a conserved region of EBOV NP provides protection against morbidity and mortality after mouse adapted EBOV challenge. A single vaccination in a C57BL/6 mouse using an adjuvanted microsphere peptide vaccine formulation containing NP44-52 is enough to confer immunity in mice. Our work suggests that a peptide vaccine based on CD8+ T-cell immunity in EBOV survivors is conceptually sound and feasible. Nucleocapsid proteins within SARS-CoV-2 contain multiple Class I epitopes with predicted HLA restrictions consistent with broad population coverage. A similar approach to a CTL vaccine design may be possible for that virus.

Circulating β cell-specific CD8+ T cells restricted by high-risk HLA class I molecules show antigen experience in children with and at risk of type 1 diabetes

In type 1 diabetes (T1D), autoreactive cytotoxic CD8+ T cells are implicated in the destruction of insulin-producing β cells. The HLA-B*3906 and HLA-A*2402 class I genes confer increased risk and promote early disease onset, suggesting that CD8+ T cells that recognize peptides presented by these class I molecules on pancreatic β cells play a pivotal role in the autoimmune response. We examined the frequency and phenotype of circulating preproinsulin (PPI)-specific and insulin B (InsB)-specific CD8+ T cells in HLA-B*3906+ children newly diagnosed with T1D and in high-risk HLA-A*2402+ children before the appearance of disease-specific autoantibodies and before diagnosis of T1D. Antigen-specific CD8+ T cells were detected using human leucocyte antigen (HLA) class I tetramers and flow cytometry was used to assess memory status. In HLA-B*3906+ children with T1D, we observed an increase in PPI5-12 -specific transitional memory CD8+ T cells compared to non-diabetic, age- and HLA-matched subjects. Furthermore, PPI5-12 -specific CD8+ T cells in HLA-B*3906+ children with T1D showed a significantly more antigen-experienced phenotype compared to polyclonal CD8+ T cells. In longitudinal samples from high-risk HLA-A*2402+ children, the percentage of terminal effector cells within the InsB15-24 -specific CD8+ T cells was increased before diagnosis relative to samples taken before the appearance of autoantibodies. This is the first study, to our knowledge, to report HLA-B*3906-restricted autoreactive CD8+ T cells in T1D. Collectively, our results provide evidence that β cell-reactive CD8+ T cells restricted by disease-associated HLA class I molecules display an antigen-experienced phenotype and acquire enhanced effector function during the period leading to clinical diagnosis, implicating these cells in driving disease.

Keratin 8 is a potential self-antigen in the coronary artery disease immunopeptidome: A translational approach

Background

Inflammation is an important risk factor in atherosclerosis, the underlying cause of coronary artery disease (CAD). Unresolved inflammation may result in maladaptive immune responses and lead to immune reactivity to self-antigens. We hypothesized that inflammation in CAD patients would manifest in immune reactivity to self-antigens detectable in soluble HLA-I/peptide complexes in the plasma.

Methods

Soluble HLA-I/peptide complexes were immuno-precipitated from plasma of male acute coronary syndrome (ACS) patients or age-matched controls and eluted peptides were subjected to mass spectrometry to generate the immunopeptidome. Self-peptides were ranked according to frequency and signal intensity, then mouse homologs of selected peptides were used to test immunologic recall in spleens of male apoE-/- mice fed either normal chow or high fat diet. The peptide detected with highest frequency in patient plasma samples and provoked T cell responses in mouse studies was selected for use as a self-antigen to stimulate CAD patient peripheral blood mononuclear cells (PBMCs).

Results

The immunopeptidome profile identified self-peptides unique to the CAD patients. The mouse homologs tested showed immune responses in apoE-/- mice. Keratin 8 was selected for further study in patient PBMCs which elicited T Effector cell responses in CAD patients compared to controls, associated with reduced PD-1 mRNA expression.

Conclusion

An immunopeptidomic strategy to search for self-antigens potentially involved in CAD identified Keratin 8. Self-reactive immune response to Keratin 8 may be an important factor in the inflammatory response in CAD.

Neoantigen characteristics in the context of the complete predicted MHC class I self-immunopeptidome

The self-immunopeptidome is the repertoire of all self-peptides that can be presented by the combination of MHC variants carried by an individual, defined by their HLA genotype. Each MHC variant presents a distinct set of self-peptides, and the number of peptides in a set is variable. Subjects carrying MHC variants that present fewer self-peptides should also present fewer mutated peptides, resulting in decreased immune pressure on tumor cells. To explore this, we predicted peptide-MHC binding values using all unique 8-11mer human peptides in the human proteome and all available HLA class I allelic variants, for a total of 134 billion unique peptide--MHC binding predictions. From these predictions, we observe that most peptides are able to be presented by relatively few (< 250) MHC, while some can be presented by upwards of 1,500 different MHC. There is substantial overlap among the repertoires of peptides presented by different MHC and no relationship between the number of peptides presented and HLA population frequency. Nearly 30% of self-peptides are presentable by at least one MHC, leaving 70% of the human peptidome unsurveyed by T cells. We observed similar distributions of predicted self-immunopeptidome sizes in cancer subjects compared to controls, and within the pan-cancer population, predicted self-immunopeptidome size combined with mutational load to predict survival. Self-immunopeptidome analysis revealed evidence for tumor immunoediting and identified specific peptide positions that most influence immunogenicity. Because self-immunopeptidome size is defined by HLA genotypes and approximates neoantigen load, HLA genotyping could offer a rapid predictive biomarker for response to immunotherapy.

Systematic identification of cancer-specific MHC-binding peptides with RAVEN

Immunotherapy can revolutionize anti-cancer therapy if specific targets are available. Immunogenic peptides encoded by cancer-specific genes (CSGs) may enable targeted immunotherapy, even of oligo-mutated cancers, which lack neo-antigens generated by protein-coding missense mutations. Here, we describe an algorithm and user-friendly software named RAVEN (Rich Analysis of Variable gene Expressions in Numerous tissues) that automatizes the systematic and fast identification of CSG-encoded peptides highly affine to Major Histocompatibility Complexes (MHC) starting from transcriptome data. We applied RAVEN to a dataset assembled from 2,678 simultaneously normalized gene expression microarrays comprising 50 tumor entities, with a focus on oligo-mutated pediatric cancers, and 71 normal tissue types. RAVEN performed a transcriptome-wide scan in each cancer entity for gender-specific CSGs, and identified several established CSGs, but also many novel candidates potentially suitable for targeting multiple cancer types. The specific expression of the most promising CSGs was validated in cancer cell lines and in a comprehensive tissue-microarray. Subsequently, RAVEN identified likely immunogenic CSG-encoded peptides by predicting their affinity to MHCs and excluded sequence identity to abundantly expressed proteins by interrogating the UniProt protein-database. The predicted affinity of selected peptides was validated in T2-cell peptide-binding assays in which many showed binding-kinetics like a very immunogenic influenza control peptide. Collectively, we provide an exquisitely curated catalogue of cancer-specific and highly MHC-affine peptides across 50 cancer types, and a freely available software (https://github.com/JSGerke/RAVENsoftware) to easily apply our algorithm to any gene expression dataset. We anticipate that our peptide libraries and software constitute a rich resource to advance anti-cancer immunotherapy.

HLA-B*39:06 Efficiently Mediates Type 1 Diabetes in a Mouse Model Incorporating Reduced Thymic Insulin Expression

Type 1 diabetes (T1D) is characterized by T cell-mediated destruction of the insulin-producing β cells of the pancreatic islets. Among the loci associated with T1D risk, those most predisposing are found in the MHC region. HLA-B*39:06 is the most predisposing class I MHC allele and is associated with an early age of onset. To establish an NOD mouse model for the study of HLA-B*39:06, we expressed it in the absence of murine class I MHC. HLA-B*39:06 was able to mediate the development of CD8 T cells, support lymphocytic infiltration of the islets, and confer T1D susceptibility. Because reduced thymic insulin expression is associated with impaired immunological tolerance to insulin and increased T1D risk in patients, we incorporated this in our model as well, finding that HLA-B*39:06-transgenic NOD mice with reduced thymic insulin expression have an earlier age of disease onset and a higher overall prevalence as compared with littermates with typical thymic insulin expression. This was despite virtually indistinguishable blood insulin levels, T cell subset percentages, and TCR Vβ family usage, confirming that reduced thymic insulin expression does not impact T cell development on a global scale. Rather, it will facilitate the thymic escape of insulin-reactive HLA-B*39:06-restricted T cells, which participate in β cell destruction. We also found that in mice expressing either HLA-B*39:06 or HLA-A*02:01 in the absence of murine class I MHC, HLA transgene identity alters TCR Vβ usage by CD8 T cells, demonstrating that some TCR Vβ families have a preference for particular class I MHC alleles.

Molecular Pathways for Immune Recognition of Preproinsulin Signal Peptide in Type 1 Diabetes

The signal peptide region of preproinsulin (PPI) contains epitopes targeted by HLA-A-restricted (HLA-A0201, A2402) cytotoxic T cells as part of the pathogenesis of β-cell destruction in type 1 diabetes. We extended the discovery of the PPI epitope to disease-associated HLA-B*1801 and HLA-B*3906 (risk) and HLA-A*1101 and HLA-B*3801 (protective) alleles, revealing that four of six alleles present epitopes derived from the signal peptide region. During cotranslational translocation of PPI, its signal peptide is cleaved and retained within the endoplasmic reticulum (ER) membrane, implying it is processed for immune recognition outside of the canonical proteasome-directed pathway. Using in vitro translocation assays with specific inhibitors and gene knockout in PPI-expressing target cells, we show that PPI signal peptide antigen processing requires signal peptide peptidase (SPP). The intramembrane protease SPP generates cytoplasm-proximal epitopes, which are transporter associated with antigen processing (TAP), ER-luminal epitopes, which are TAP independent, each presented by different HLA class I molecules and N-terminal trimmed by ER aminopeptidase 1 for optimal presentation. In vivo, TAP expression is significantly upregulated and correlated with HLA class I hyperexpression in insulin-containing islets of patients with type 1 diabetes. Thus, PPI signal peptide epitopes are processed by SPP and loaded for HLA-guided immune recognition via pathways that are enhanced during disease pathogenesis.

Perpetual complexity: predicting human CD8+ T-cell responses to pathogenic peptides

The accurate prediction of human CD8⁺ T-cell epitopes has great potential clinical and translational implications in the context of infection, cancer and autoimmunity. Prediction algorithms have traditionally focused on calculated peptide affinity for the binding groove of MHC-I. However, over the years it has become increasingly clear that the ultimate T-cell recognition of MHC-I-bound peptides is governed by many contributing factors within the complex antigen presentation pathway. Recent advances in next-generation sequencing and immunnopeptidomics have increased the precision of HLA-I sub-allele classification, and have led to the discovery of peptide processing events and individual allele-specific binding preferences. Here, we review some of the discoveries that initiated the development of peptide prediction algorithms, and outline some of the current available online tools for CD8⁺ T-cell epitope prediction.

Mass Spectrometry Profiling of HLA-Associated Peptidomes in Mono-allelic Cells Enables More Accurate Epitope Prediction

Identification of human leukocyte antigen (HLA)-bound peptides by liquid chromatography-tandem mass spectrometry (LC-MS/MS) is poised to provide a deep understanding of rules underlying antigen presentation. However, a key obstacle is the ambiguity that arises from the co-expression of multiple HLA alleles. Here, we have implemented a scalable mono-allelic strategy for profiling the HLA peptidome. By using cell lines expressing a single HLA allele, optimizing immunopurifications, and developing an application-specific spectral search algorithm, we identified thousands of peptides bound to 16 different HLA class I alleles. These data enabled the discovery of subdominant binding motifs and an integrative analysis quantifying the contribution of factors critical to epitope presentation, such as protein cleavage and gene expression. We trained neural-network prediction algorithms with our large dataset (>24,000 peptides) and outperformed algorithms trained on datasets of peptides with measured affinities. We thus demonstrate a strategy for systematically learning the rules of endogenous antigen presentation.

NetMHCpan-3.0; improved prediction of binding to MHC class I molecules integrating information from multiple receptor and peptide length datasets

Background

Binding of peptides to MHC class I molecules (MHC-I) is essential for antigen presentation to cytotoxic T-cells.

Results

Here, we demonstrate how a simple alignment step allowing insertions and deletions in a pan-specific MHC-I binding machine-learning model enables combining information across both multiple MHC molecules and peptide lengths. This pan-allele/pan-length algorithm significantly outperforms state-of-the-art methods, and captures differences in the length profile of binders to different MHC molecules leading to increased accuracy for ligand identification. Using this model, we demonstrate that percentile ranks in contrast to affinity-based thresholds are optimal for ligand identification due to uniform sampling of the MHC space.

Conclusions

We have developed a neural network-based machine-learning algorithm leveraging information across multiple receptor specificities and ligand length scales, and demonstrated how this approach significantly improves the accuracy for prediction of peptide binding and identification of MHC ligands. The method is available at www.cbs.dtu.dk/services/NetMHCpan-3.0.

Electronic supplementary material

The online version of this article (doi:10.1186/s13073-016-0288-x) contains supplementary material, which is available to authorized users.

Characterization of the peptide binding specificity of the HLA class I alleles B*38:01 and B*39:06

B*38:01 and B*39:06 are present with phenotypic frequencies <2% in the general population, but are of interest as B*39:06 is the B allele most associated with type 1 diabetes susceptibility and 38:01 is most protective. A previous study derived putative main anchor motifs for both alleles based on peptide elution data. The present study has utilized panels of single amino acid substitution peptide libraries to derive detailed quantitative motifs accounting for both primary and secondary influences on peptide binding. From these analyses, both alleles were confirmed to utilize the canonical position 2/C-terminus main anchor spacing. B*38:01 preferentially bound peptides with the positively charged or polar residues H, R, and Q in position 2 and the large hydrophobic residues I, F, L, W, and M at the C-terminus. B*39:06 had a similar preference for R in position 2, but also well-tolerated M, Q, and K. A more dramatic contrast between the two alleles was noted at the C-terminus, where the specificity of B*39:06 was clearly for small residues, with A as most preferred, followed by G, V, S, T, and I. Detailed position-by-position and residue-by-residue coefficient values were generated from the panels to provide detailed quantitative B*38:01 and B*39:06 motifs. It is hoped that these detailed motifs will facilitate the identification of T cell epitopes recognized in the context of two class I alleles associated with dramatically different dispositions towards type 1 diabetes, offering potential avenues for the investigation of the role of CD8 T cells in this disease.

Gapped sequence alignment using artificial neural networks: application to the MHC class I system

MOTIVATION

Many biological processes are guided by receptor interactions with linear ligands of variable length. One such receptor is the MHC class I molecule. The length preferences vary depending on the MHC allele, but are generally limited to peptides of length 8-11 amino acids. On this relatively simple system, we developed a sequence alignment method based on artificial neural networks that allows insertions and deletions in the alignment.

RESULTS

We show that prediction methods based on alignments that include insertions and deletions have significantly higher performance than methods trained on peptides of single lengths. Also, we illustrate how the location of deletions can aid the interpretation of the modes of binding of the peptide-MHC, as in the case of long peptides bulging out of the MHC groove or protruding at either terminus. Finally, we demonstrate that the method can learn the length profile of different MHC molecules, and quantified the reduction of the experimental effort required to identify potential epitopes using our prediction algorithm.

AVAILABILITY AND IMPLEMENTATION

The NetMHC-4.0 method for the prediction of peptide-MHC class I binding affinity using gapped sequence alignment is publicly available at: http://www.cbs.dtu.dk/services/NetMHC-4.0.