Accurate structure prediction of peptide-MHC complexes for identifying highly immunogenic antigens

Peptide-binding specificity prediction using fine-tuned protein structure prediction networks

Peptide-binding proteins play key roles in biology, and predicting their binding specificity is a long-standing challenge. While considerable protein structural information is available, the most successful current methods use sequence information alone, in part because it has been a challenge to model the subtle structural changes accompanying sequence substitutions. Protein structure prediction networks such as AlphaFold model sequence-structure relationships very accurately, and we reasoned that if it were possible to specifically train such networks on binding data, more generalizable models could be created. We show that placing a classifier on top of the AlphaFold network and fine-tuning the combined network parameters for both classification and structure prediction accuracy leads to a model with strong generalizable performance on a wide range of Class I and Class II peptide-MHC interactions that approaches the overall performance of the state-of-the-art NetMHCpan sequence-based method. The peptide-MHC optimized model shows excellent performance in distinguishing binding and non-binding peptides to SH3 and PDZ domains. This ability to generalize well beyond the training set far exceeds that of sequence-only models and should be particularly powerful for systems where less experimental data are available.

Physicochemical Heuristics for Identifying High Fidelity, Near-Native Structural Models of Peptide/MHC Complexes

There is long-standing interest in accurately modeling the structural features of peptides bound and presented by class I MHC proteins. This interest has grown with the advent of rapid genome sequencing and the prospect of personalized, peptide-based cancer vaccines, as well as the development of molecular and cellular therapeutics based on T cell receptor recognition of peptide-MHC. However, while the speed and accessibility of peptide-MHC modeling has improved substantially over the years, improvements in accuracy have been modest. Accuracy is crucial in peptide-MHC modeling, as T cell receptors are highly sensitive to peptide conformation and capturing fine details is therefore necessary for useful models. Studying nonameric peptides presented by the common class I MHC protein HLA-A*02:01, here we addressed a key question common to modern modeling efforts: from a set of models (or decoys) generated through conformational sampling, which is best? We found that the common strategy of decoy selection by lowest energy can lead to substantial errors in predicted structures. We therefore adopted a data-driven approach and trained functions capable of predicting near native decoys with exceptionally high accuracy. Although our implementation is limited to nonamer/HLA-A*02:01 complexes, our results serve as an important proof of concept from which improvements can be made and, given the significance of HLA-A*02:01 and its preference for nonameric peptides, should have immediate utility in select immunotherapeutic and other efforts for which structural information would be advantageous.

Structural Prediction of Peptide-MHC Binding Modes

The immune system is constantly protecting its host from the invasion of pathogens and the development of cancer cells. The specific CD8⁺ T-cell immune response against virus-infected cells and tumor cells is based on the T-cell receptor recognition of antigenic peptides bound to class I major histocompatibility complexes (MHC) at the surface of antigen presenting cells. Consequently, the peptide binding specificities of the highly polymorphic MHC have important implications for the design of vaccines, for the treatment of autoimmune diseases, and for personalized cancer immunotherapy. Evidence-based machine-learning approaches have been successfully used for the prediction of peptide binders and are currently being developed for the prediction of peptide immunogenicity. However, understanding and modeling the structural details of peptide/MHC binding is crucial for a better understanding of the molecular mechanisms triggering the immunological processes, estimating peptide/MHC affinity using universal physics-based approaches, and driving the design of novel peptide ligands. Unfortunately, due to the large diversity of MHC allotypes and possible peptides, the growing number of 3D structures of peptide/MHC (pMHC) complexes in the Protein Data Bank only covers a small fraction of the possibilities. Consequently, there is a growing need for rapid and efficient approaches to predict 3D structures of pMHC complexes. Here, we review the key characteristics of the 3D structure of pMHC complexes before listing databases and other sources of information on pMHC structures and MHC specificities. Finally, we discuss some of the most prominent pMHC docking software.

Predicting Cross-Reactivity and Antigen Specificity of T Cell Receptors

Adaptive immune recognition is mediated by specific interactions between heterodimeric T cell receptors (TCRs) and their cognate peptide-MHC (pMHC) ligands, and the methods to accurately predict TCR:pMHC interaction would have profound clinical, therapeutic and pharmaceutical applications. Herein, we review recent developments in predicting cross-reactivity and antigen specificity of TCR recognition. We discuss current experimental and computational approaches to investigate cross-reactivity and antigen-specificity of TCRs and highlight how integrating kinetic, biophysical and structural features may offer valuable insights in modeling immunogenicity. We further underscore the close inter-relationship of these two interconnected notions and the need to investigate each in the light of the other for a better understanding of T cell responsiveness for the effective clinical applications.

Combining Three-Dimensional Modeling with Artificial Intelligence to Increase Specificity and Precision in Peptide-MHC Binding Predictions

The reliable prediction of the affinity of candidate peptides for the MHC is important for predicting their potential antigenicity and thus influences medical applications, such as decisions on their inclusion in T cell-based vaccines. In this study, we present a rapid, predictive computational approach that combines a popular, sequence-based artificial neural network method, NetMHCpan 4.0, with three-dimensional structural modeling. We find that the ensembles of bound peptide conformations generated by the programs MODELLER and Rosetta FlexPepDock are less variable in geometry for strong binders than for low-affinity peptides. In tests on 1271 peptide sequences for which the experimental dissociation constants of binding to the well-characterized murine MHC allele H-2Db are known, by applying thresholds for geometric fluctuations the structure-based approach in a standalone manner drastically improves the statistical specificity, reducing the number of false positives. Furthermore, filtering candidates generated with NetMHCpan 4.0 with the structure-based predictor led to an increase in the positive predictive value (PPV) of the peptides correctly predicted to bind very strongly (i.e., K d < 100 nM) from 40 to 52% (p = 0.027). The combined method also significantly improved the PPV when tested on five human alleles, including some with limited data for training. Overall, an average increase of 10% in the PPV was found over the standalone sequence-based method. The combined method should be useful in the rapid design of effective T cell-based vaccines.

Machine Learning for Cancer Immunotherapies Based on Epitope Recognition by T Cell Receptors

In the last years, immunotherapies have shown tremendous success as treatments for multiple types of cancer. However, there are still many obstacles to overcome in order to increase response rates and identify effective therapies for every individual patient. Since there are many possibilities to boost a patient's immune response against a tumor and not all can be covered, this review is focused on T cell receptor-mediated therapies. CD8+ T cells can detect and destroy malignant cells by binding to peptides presented on cell surfaces by MHC (major histocompatibility complex) class I molecules. CD4+ T cells can also mediate powerful immune responses but their peptide recognition by MHC class II molecules is more complex, which is why the attention has been focused on CD8+ T cells. Therapies based on the power of T cells can, on the one hand, enhance T cell recognition by introducing TCRs that preferentially direct T cells to tumor sites (so called TCR-T therapy) or through vaccination to induce T cells in vivo. On the other hand, T cell activity can be improved by immune checkpoint inhibition or other means that help create a microenvironment favorable for cytotoxic T cell activity. The manifold ways in which the immune system and cancer interact with each other require not only the use of large omics datasets from gene, to transcript, to protein, and to peptide but also make the application of machine learning methods inevitable. Currently, discovering and selecting suitable TCRs is a very costly and work intensive in vitro process. To facilitate this process and to additionally allow for highly personalized therapies that can simultaneously target multiple patient-specific antigens, especially neoepitopes, breakthrough computational methods for predicting antigen presentation and TCR binding are urgently required. Particularly, potential cross-reactivity is a major consideration since off-target toxicity can pose a major threat to patient safety. The current speed at which not only datasets grow and are made available to the public, but also at which new machine learning methods evolve, is assuring that computational approaches will be able to help to solve problems that immunotherapies are still facing.

Structure Based Prediction of Neoantigen Immunogenicity

The development of immunological therapies that incorporate peptide antigens presented to T cells by MHC proteins is a long sought-after goal, particularly for cancer, where mutated neoantigens are being explored as personalized cancer vaccines. Although neoantigens can be identified through sequencing, bioinformatics and mass spectrometry, identifying those which are immunogenic and able to promote tumor rejection remains a significant challenge. Here we examined the potential of high-resolution structural modeling followed by energetic scoring of structural features for predicting neoantigen immunogenicity. After developing a strategy to rapidly and accurately model nonameric peptides bound to the common class I MHC protein HLA-A2, we trained a neural network on structural features that influence T cell receptor (TCR) and peptide binding energies. The resulting structurally-parameterized neural network outperformed methods that do not incorporate explicit structural or energetic properties in predicting CD8+ T cell responses of HLA-A2 presented nonameric peptides, while also providing insight into the underlying structural and biophysical mechanisms governing immunogenicity. Our proof-of-concept study demonstrates the potential for structure-based immunogenicity predictions in the development of personalized peptide-based vaccines.

Broad cross-reactivity of the T-cell repertoire achieves specific and sufficiently rapid target searching

The molecular recognition of T-cell receptors is the hallmark of the adaptive immunity. Given the finiteness of the T-cell repertoire, individual T-cell receptors are necessary to be cross-reactive to multiple antigenic peptides. In this study, we quantify the variability of the cross-reactivity by using a string model that estimates the binding affinity between two sequences of amino acids. We examine sequences of 10,000 human T-cell receptors and 10,000 antigenic peptides, and obtain a full spectrum of cross-reactivity of the receptor-peptide binding. Then, we find that the cross-reactivity spectrum is broad. Some T-cells are reactive to 1000 peptides, but some T-cells are reactive to only one or two peptides. Since the degree of cross-reactivity has a correlation with the (un)binding affinity of receptors, we further investigate how the broad cross-reactivity affects the target searching of T-cells. High cross-reactive T-cells may not require many trials for searching correct targets, but they may spend long time to unbind from incorrect targets. In contrast, low cross-reactive T-cells may not spend long time to ignore incorrect targets, but they require many trials for screening correct targets. We evaluate this hypothesis, and show that the broad cross-reactivity of the natural T-cell repertoire can balance the trade-off between the rapid screening and unbinding penalty.

GradDock: rapid simulation and tailored ranking functions for peptide-MHC Class I docking

Motivation

The identification of T-cell epitopes has many profound translational applications in the areas of transplantation, disease diagnosis, vaccine/therapeutic protein development and personalized immunotherapy. While data-driven methods have been widely used for the prediction of peptide binders with notable successes, the structural modeling of peptide binding to MHC molecules is crucial for understanding the underlying molecular mechanism of the immunological processes.

Results

We developed GradDock, a structure-based method for the rapid and accurate modeling of peptide binding to MHC Class I (pMHC-I). GradDock explicitly models diverse unbound peptides in vacuo and inserts them into the MHC-I groove through a steered gradient descent with a topological correction process. The simulation process yields diverse structural conformations including native-like peptides. We completely revised the Rosetta score terms and developed a new ranking function specifically for pMHC-I. Using the diverse peptides, a linear programming approach is applied to find the optimal weights for the individual Rosetta score terms. Our examination revealed that a refinement of the dihedral angles and a modification of the repulsion can dramatically improve the modeling quality. GradDock is five-times faster than a Rosetta-based docking approach for pMHC-I. We also demonstrate that the predictive capability of GradDock with the re-weighted Rosetta ranking function is consistently more accurate than the Rosetta-based method with the standard Rosetta score (approximately three-times better for a cross-docking set).

Availability and implementation

GradDock is freely available for academic purposes. The program and the ranking score weights for Rosetta are available at http://bel.kaist.ac.kr/research/GradDock.

Contact

hskim76@kaist.ac.kr.

Supplementary information

Supplementary data are available at Bioinformatics online.

Analysis of the affinity of influenza A virus protein epitopes for swine MHC I by a modified in vitro refolding method indicated cross-reactivity between swine and human MHC I specificities

In vitro refolding assays can be used to investigate the affinity and stability of the binding of epitope peptides to major histocompatibility complex (MHC) class I molecules, which are key factors in the presentation of peptides to cytotoxic T lymphocytes (CTLs). The recognition of peptide epitopes by CTLs is crucial for protection against influenza A virus (IAV) infection. The peptide-binding motif of the swine SLA-3*hs0202 molecule has been previously reported and partly overlaps with the binding motif of the most abundant human MHC allele, HLA-A*0201. In this study, we screened all the protein sequences of the swine-origin epidemic IAV strain A/Beijing/01/2009 (H1N1), and a total of 73 9-mer epitope peptides were predicted to fit the consensus motif of the swine SLA-3*hs0202 or HLA-A*0201 molecule. Then, 14 peptides were selected, and their affinities to SLA-3*hs0202 were tested by a modified in vitro refolding assay. Our results show that ten epitopes could tolerate gel filtration, indicating that these epitopes formed stable or partly stable complexes with SLA-3*hs0202. Eight out of the ten epitopes have been previously reported as HLA-A2-restricted epitopes, which implied cross-reactivity between swine and human MHC I specificities. Furthermore, the modified mini-system refolding method could be applied for the screening of peptides because the refolding efficiency remained almost unchanged with the positive peptide (HA-KMN9) subjected to size-exclusion chromatography and Resource Q anion-exchange chromatography. The results presented here provide new insight into the development of epitope-based vaccines to control IAV and increase our understanding of swine molecular immunology.

Platform technologies for modern vaccine manufacturing

Improved understanding of antigenic components and their interaction with the immune system, as supported by computational tools, permits a sophisticated approach to modern vaccine design. Vaccine platforms provide an effective tool by which strategically designed peptide and protein antigens are modularized to enhance their immunogenicity. These modular vaccine platforms can overcome issues faced by traditional vaccine manufacturing and have the potential to generate safe vaccines, rapidly and at a low cost. This review introduces two promising platforms based on virus-like particle and liposome, and discusses the methodologies and challenges.

Enhancement of viral escape in HIV-1 Nef by STEP vaccination

OBJECTIVE

Properly priming cytotoxic T-lymphocyte (CTL) responses is an important task in HIV-1 vaccination. However, the STEP trial showed no efficacy even though the vaccine elicited HIV-specific CTL responses. Our study is to investigate whether or not the STEP vaccine enhanced viral escape in infected volunteers.

METHODS

The signature of viral escape, the presence of multiple escape variants, could be falsely represented by the existence of multiple founder viruses. Therefore, we use a mathematical model to designate STEP study patients with infections from a single founder virus. We then conduct permutation tests on each of 9988 Gag, Pol, and Nef overlapping peptides to identify epitopes with significant differences in diversity between the vaccine and placebo groups using previously published STEP trial sequence data.

RESULTS

We identify signatures of vaccine-enhanced viral escape within HIV-1 Nef from the STEP trial. Vaccine-treated patients showed a greater level of epitope diversity in one of the immunodomiant epitopes, EVGFPVRPQVPL (Nef65-76), compared with placebo-treated patients (P = 0.0038). In the other three Nef epitopes, there is a marginally significant difference in the epitope diversity between the vaccine and placebo group (P < 0.1). This greater epitope diversity was neither due to any difference in infection duration nor overall nef gene diversity between the two groups, suggesting that the increase in viral escape was likely mediated by vaccine-induced T-cell responses.

CONCLUSION

Viral escape in Nef is elevated preferentially in STEP vaccine-treated individuals, suggesting that vaccination primarily modulated initial CTL responses. Our observations provide important insights into improving vaccine-primed first immune control.

CD8 T-cell responses against the immunodominant Theileria parva peptide Tp249-59 are composed of two distinct populations specific for overlapping 11-mer and 10-mer epitopes

Immunity against Theileria parva is associated with CD8 T-cell responses that exhibit immunodominance, focusing the response against limited numbers of epitopes. As candidates for inclusion in vaccines, characterization of responses against immunodominant epitopes is a key component in novel vaccine development. We have previously demonstrated that the Tp249-59 and Tp1214-224 epitopes dominate CD8 T-cell responses in BoLA-A10 and BoLA-18 MHC I homozygous animals, respectively. In this study, peptide-MHC I tetramers for these epitopes, and a subdominant BoLA-A10-restricted epitope (Tp298-106 ), were generated to facilitate accurate and rapid enumeration of epitope-specific CD8 T cells. During validation of these tetramers a substantial proportion of Tp249-59 -reactive T cells failed to bind the tetramer, suggesting that this population was heterogeneous with respect to the recognized epitope. We demonstrate that Tp250-59 represents a distinct epitope and that tetramers produced with Tp50-59 and Tp49-59 show no cross-reactivity. The Tp249-59 and Tp250-59 epitopes use different serine residues as the N-terminal anchor for binding to the presenting MHC I molecule. Molecular dynamic modelling predicts that the two peptide-MHC I complexes adopt structurally different conformations and Tcell receptor β sequence analysis showed that Tp249-59 and Tp250-59 are recognized by non-overlapping T-cell receptor repertoires. Together these data demonstrate that although differing by only a single residue, Tp249-59 and Tp250-59 epitopes form distinct ligands for T-cell receptor recognition. Tetramer analysis of T. parva-specific CD8 T-cell lines confirmed the immunodominance of Tp1214-224 in BoLA-A18 animals and showed in BoLA-A10 animals that the Tp249-59 epitope response was generally more dominant than the Tp250-59 response and confirmed that the Tp298-106 response was subdominant.

Computational Modeling of T Cell Receptor Complexes

T-cell receptor (TCR) binding to peptide/MHC determines specificity and initiates signaling in antigen-specific cellular immune responses. Structures of TCR-pMHC complexes have provided enormous insight to cellular immune functions, permitted a rational understanding of processes such as pathogen escape, and led to the development of novel approaches for the design of vaccines and other therapeutics. As production, crystallization, and structure determination of TCR-pMHC complexes can be challenging, there is considerable interest in modeling new complexes. Here we describe a rapid approach to TCR-pMHC modeling that takes advantage of structural features conserved in known complexes, such as the restricted TCR binding site and the generally conserved diagonal docking mode. The approach relies on the powerful Rosetta suite and is implemented using the PyRosetta scripting environment. We show how the approach can recapitulate changes in TCR binding angles and other structural details, and highlight areas where careful evaluation of parameters is needed and alternative choices might be made. As TCRs are highly sensitive to subtle structural perturbations, there is room for improvement. Our method nonetheless generates high-quality models that can be foundational for structure-based hypotheses regarding TCR recognition.

Genomic and bioinformatic profiling of mutational neoepitopes reveals new rules to predict anticancer immunogenicity

The mutational repertoire of cancers creates the neoepitopes that make cancers immunogenic. Here, we introduce two novel tools that identify, with relatively high accuracy, the small proportion of neoepitopes (among the hundreds of potential neoepitopes) that protect the host through an antitumor T cell response. The two tools consist of (a) the numerical difference in NetMHC scores between the mutated sequences and their unmutated counterparts, termed the differential agretopic index, and (b) the conformational stability of the MHC I-peptide interaction. Mechanistically, these tools identify neoepitopes that are mutated to create new anchor residues for MHC binding, and render the overall peptide more rigid. Surprisingly, the protective neoepitopes identified here elicit CD8-dependent immunity, even though their affinity for K(d) is orders of magnitude lower than the 500-nM threshold considered reasonable for such interactions. These results greatly expand the universe of target cancer antigens and identify new tools for human cancer immunotherapy.