Exploring the Potential of Structure-Based Deep Learning Approaches for T cell Receptor Design.

Deep learning methods, trained on the increasing set of available protein 3D structures and sequences, have substantially impacted the protein modeling and design field. These advancements have facilitated the creation of novel proteins, or the optimization of existing ones designed for specific functions, such as binding a target protein. Despite the demonstrated potential of such approaches in designing general protein binders, their application in designing immunotherapeutics remains relatively unexplored. A relevant application is the design of T cell receptors (TCRs). Given the crucial role of T cells in mediating immune responses, redirecting these cells to tumor or infected target cells through the engineering of TCRs has shown promising results in treating diseases, especially cancer. However, the computational design of TCR interactions presents challenges for current physics-based methods, particularly due to the unique natural characteristics of these interfaces, such as low affinity and cross-reactivity. For this reason, in this study, we explored the potential of two structure-based deep learning protein design methods, ProteinMPNN and ESM-IF, in designing fixed-backbone TCRs for binding target antigenic peptides presented by the MHC through different design scenarios. To evaluate TCR designs, we employed a comprehensive set of sequence- and structure-based metrics, highlighting the benefits of these methods in comparison to classical physics-based design methods and identifying deficiencies for improvement.

Combinatorially restricted computational design of protein-protein interfaces to produce IgG heterodimers.

Redesigning protein-protein interfaces is an important tool for developing therapeutic strategies. Interfaces can be redesigned by in silico screening, which allows for efficient sampling of a large protein space before experimental validation. However, computational costs limit the number of combinations that can be reasonably sampled. Here, we present combinatorial tyrosine (Y)/serine (S) selection (combYSelect), a computational approach combining in silico determination of the change in binding free energy (ΔΔG) of an interface with a highly restricted library composed of just two amino acids, tyrosine and serine. We used combYSelect to design two immunoglobulin G (IgG) heterodimers-combYSelect1 (L368S/D399Y-K409S/T411Y) and combYSelect2 (D399Y/K447S-K409S/T411Y)-that exhibit near-optimal heterodimerization, without affecting IgG stability or function. We solved the crystal structures of these heterodimers and found that dynamic π-stacking interactions and polar contacts drive preferential heterodimeric interactions. Finally, we demonstrated the utility of our combYSelect heterodimers by engineering both a bispecific antibody and a cytokine trap for two unique therapeutic applications.

Evaluation of AlphaFold antibody-antigen modeling with implications for improving predictive accuracy.

High resolution antibody-antigen structures provide critical insights into immune recognition and can inform therapeutic design. The challenges of experimental structural determination and the diversity of the immune repertoire underscore the necessity of accurate computational tools for modeling antibody-antigen complexes. Initial benchmarking showed that despite overall success in modeling protein-protein complexes, AlphaFold and AlphaFold-Multimer have limited success in modeling antibody-antigen interactions. In this study, we performed a thorough analysis of AlphaFold's antibody-antigen modeling performance on 427 nonredundant antibody-antigen complex structures, identifying useful confidence metrics for predicting model quality, and features of complexes associated with improved modeling success. Notably, we found that the latest version of AlphaFold improves near-native modeling success to over 30%, versus approximately 20% for a previous version, while increased AlphaFold sampling gives approximately 50% success. With this improved success, AlphaFold can generate accurate antibody-antigen models in many cases, while additional training or other optimization may further improve performance.

Structural basis for T cell recognition of cancer neoantigens and implications for predicting neoepitope immunogenicity.

Adoptive cell therapy (ACT) with tumor-specific T cells has been shown to mediate durable cancer regression. Tumor-specific T cells are also the basis of other therapies, notably cancer vaccines. The main target of tumor-specific T cells are neoantigens resulting from mutations in self-antigens over the course of malignant transformation. The detection of neoantigens presents a major challenge to T cells because of their high structural similarity to self-antigens, and the need to avoid autoimmunity. How different a neoantigen must be from its wild-type parent for it to induce a T cell response is poorly understood. Here we review recent structural and biophysical studies of T cell receptor (TCR) recognition of shared cancer neoantigens derived from oncogenes, including p53R175H, KRASG12D, KRASG12V, HHATp8F, and PIK3CAH1047L. These studies have revealed that, in some cases, the oncogenic mutation improves antigen presentation by strengthening peptide-MHC binding. In other cases, the mutation is detected by direct interactions with TCR, or by energetically driven or other indirect strategies not requiring direct TCR contacts with the mutation. We also review antibodies designed to recognize peptide-MHC on cell surfaces (TCR-mimic antibodies) as an alternative to TCRs for targeting cancer neoantigens. Finally, we review recent computational advances in this area, including efforts to predict neoepitope immunogenicity and how these efforts may be advanced by structural information on peptide-MHC binding and peptide-MHC recognition by TCRs.

A single C-terminal residue controls SARS-CoV-2 spike trafficking and incorporation into VLPs.

The spike (S) protein of SARS-CoV-2 is delivered to the virion assembly site in the ER-Golgi Intermediate Compartment (ERGIC) from both the ER and cis-Golgi in infected cells. However, the relevance and modulatory mechanism of this bidirectional trafficking are unclear. Here, using structure-function analyses, we show that S incorporation into virus-like particles (VLP) and VLP fusogenicity are determined by coatomer-dependent S delivery from the cis-Golgi and restricted by S-coatomer dissociation. Although S mimicry of the host coatomer-binding dibasic motif ensures retrograde trafficking to the ERGIC, avoidance of the host-like C-terminal acidic residue is critical for S-coatomer dissociation and therefore incorporation into virions or export for cell-cell fusion. Because this C-terminal residue is the key determinant of SARS-CoV-2 assembly and fusogenicity, our work provides a framework for the export of S protein encoded in genetic vaccines for surface display and immune activation.

SARS-CoV-2 infection establishes a stable and age-independent CD8+ T cell response against a dominant nucleocapsid epitope using restricted T cell receptors.

The resolution of SARS-CoV-2 replication hinges on cell-mediated immunity, wherein CD8+ T cells play a vital role. Nonetheless, the characterization of the specificity and TCR composition of CD8+ T cells targeting non-spike protein of SARS-CoV-2 before and after infection remains incomplete. Here, we analyzed CD8+ T cells recognizing six epitopes from the SARS-CoV-2 nucleocapsid (N) protein and found that SARS-CoV-2 infection slightly increased the frequencies of N-recognizing CD8+ T cells but significantly enhanced activation-induced proliferation compared to that of the uninfected donors. The frequencies of N-specific CD8+ T cells and their proliferative response to stimulation did not decrease over one year. We identified the N222-230 peptide (LLLDRLNQL, referred to as LLL thereafter) as a dominant epitope that elicited the greatest proliferative response from both convalescent and uninfected donors. Single-cell sequencing of T cell receptors (TCR) from LLL-specific CD8+ T cells revealed highly restricted Vα gene usage (TRAV12-2) with limited CDR3α motifs, supported by structural characterization of the TCR-LLL-HLA-A2 complex. Lastly, transcriptome analysis of LLL-specific CD8+ T cells from donors who had expansion (expanders) or no expansion (non-expanders) after in vitro stimulation identified increased chromatin modification and innate immune functions of CD8+ T cells in non-expanders. These results suggests that SARS-CoV-2 infection induces LLL-specific CD8+ T cell responses with a restricted TCR repertoire.

Structure of engineered hepatitis C virus E1E2 ectodomain in complex with neutralizing antibodies.

Hepatitis C virus (HCV) is a major global health burden as the leading causative agent of chronic liver disease and hepatocellular carcinoma. While the main antigenic target for HCV-neutralizing antibodies is the membrane-associated E1E2 surface glycoprotein, the development of effective vaccines has been hindered by complications in the biochemical preparation of soluble E1E2 ectodomains. Here, we present a cryo-EM structure of an engineered, secreted E1E2 ectodomain of genotype 1b in complex with neutralizing antibodies AR4A, HEPC74, and IGH520. Structural characterization of the E1 subunit and C-terminal regions of E2 reveal an overall architecture of E1E2 that concurs with that observed for non-engineered full-length E1E2. Analysis of the AR4A epitope within a region of E2 that bridges between the E2 core and E1 defines the structural basis for its broad neutralization. Our study presents the structure of an E1E2 complex liberated from membrane via a designed scaffold, one that maintains all essential structural features of native E1E2. The study advances the understanding of the E1E2 heterodimer structure, crucial for the rational design of secreted E1E2 antigens in vaccine development.

Evaluation of AlphaFold Antibody-Antigen Modeling with Implications for Improving Predictive Accuracy.

High resolution antibody-antigen structures provide critical insights into immune recognition and can inform therapeutic design. The challenges of experimental structural determination and the diversity of the immune repertoire underscore the necessity of accurate computational tools for modeling antibody-antigen complexes. Initial benchmarking showed that despite overall success in modeling protein-protein complexes, AlphaFold and AlphaFold-Multimer have limited success in modeling antibody-antigen interactions. In this study, we performed a thorough analysis of AlphaFold's antibody-antigen modeling performance on 429 nonredundant antibody-antigen complex structures, identifying useful confidence metrics for predicting model quality, and features of complexes associated with improved modeling success. We show the importance of bound-like component modeling in complex assembly accuracy, and that the current version of AlphaFold improves near-native modeling success to over 30%, versus approximately 20% for a previous version. With this improved success, AlphaFold can generate accurate antibody-antigen models in many cases, while additional training may further improve its performance.

TCRmodel2: high-resolution modeling of T cell receptor recognition using deep learning.

The cellular immune system, which is a critical component of human immunity, uses T cell receptors (TCRs) to recognize antigenic proteins in the form of peptides presented by major histocompatibility complex (MHC) proteins. Accurate definition of the structural basis of TCRs and their engagement of peptide-MHCs can provide major insights into normal and aberrant immunity, and can help guide the design of vaccines and immunotherapeutics. Given the limited amount of experimentally determined TCR-peptide-MHC structures and the vast amount of TCRs within each individual as well as antigenic targets, accurate computational modeling approaches are needed. Here, we report a major update to our web server, TCRmodel, which was originally developed to model unbound TCRs from sequence, to now model TCR-peptide-MHC complexes from sequence, utilizing several adaptations of AlphaFold. This method, named TCRmodel2, allows users to submit sequences through an easy-to-use interface and shows similar or greater accuracy than AlphaFold and other methods to model TCR-peptide-MHC complexes based on benchmarking. It can generate models of complexes in 15 minutes, and output models are provided with confidence scores and an integrated molecular viewer. TCRmodel2 is available at https://tcrmodel.ibbr.umd.edu.

Structural insights into protection against a SARS-CoV-2 spike variant by T cell receptor diversity.

T cells play a crucial role in combatting SARS-CoV-2 and forming long-term memory responses to this coronavirus. The emergence of SARS-CoV-2 variants that can evade T cell immunity has raised concerns about vaccine efficacy and the risk of reinfection. Some SARS-CoV-2 T cell epitopes elicit clonally restricted CD8+ T cell responses characterized by T cell receptors (TCRs) that lack structural diversity. Mutations in such epitopes can lead to loss of recognition by most T cells specific for that epitope, facilitating viral escape. Here, we studied an HLA-A2-restricted spike protein epitope (RLQ) that elicits CD8+ T cell responses in COVID-19 convalescent patients characterized by highly diverse TCRs. We previously reported the structure of an RLQ-specific TCR (RLQ3) with greatly reduced recognition of the most common natural variant of the RLQ epitope (T1006I). Opposite to RLQ3, TCR RLQ7 recognizes T1006I with even higher functional avidity than the WT epitope. To explain the ability of RLQ7, but not RLQ3, to tolerate the T1006I mutation, we determined structures of RLQ7 bound to RLQ-HLA-A2 and T1006I-HLA-A2. These complexes show that there are multiple structural solutions to recognizing RLQ and thereby generating a clonally diverse T cell response to this epitope that assures protection against viral escape and T cell clonal loss.

Mucosal nanobody IgA as inhalable and affordable prophylactic and therapeutic treatment against SARS-CoV-2 and emerging variants.

Anti-COVID antibody therapeutics have been developed but not widely used due to their high cost and escape of neutralization from the emerging variants. Here, we describe the development of VHH-IgA1.1, a nanobody IgA fusion molecule as an inhalable, affordable and less invasive prophylactic and therapeutic treatment against SARS-CoV-2 Omicron variants. VHH-IgA1.1 recognizes a conserved epitope of SARS-CoV-2 spike protein Receptor Binding Domain (RBD) and potently neutralizes major global SARS-CoV-2 variants of concern (VOC) including the Omicron variant and its sub lineages BA.1.1, BA.2 and BA.2.12.1. VHH-IgA1.1 is also much more potent against Omicron variants as compared to an IgG Fc fusion construct, demonstrating the importance of IgA mediated mucosal protection for Omicron infection. Intranasal administration of VHH-IgA1.1 prior to or after challenge conferred significant protection from severe respiratory disease in K18-ACE2 transgenic mice infected with SARS-CoV-2 VOC. More importantly, for cost-effective production, VHH-IgA1.1 produced in Pichia pastoris had comparable potency to mammalian produced antibodies. Our study demonstrates that intranasal administration of affordably produced VHH-IgA fusion protein provides effective mucosal immunity against infection of SARS-CoV-2 including emerging variants.

Benchmarking AlphaFold for protein complex modeling reveals accuracy determinants.

High-resolution experimental structural determination of protein-protein interactions has led to valuable mechanistic insights, yet due to the massive number of interactions and experimental limitations there is a need for computational methods that can accurately model their structures. Here we explore the use of the recently developed deep learning method, AlphaFold, to predict structures of protein complexes from sequence. With a benchmark of 152 diverse heterodimeric protein complexes, multiple implementations and parameters of AlphaFold were tested for accuracy. Remarkably, many cases (43%) had near-native models (medium or high critical assessment of predicted interactions accuracy) generated as top-ranked predictions by AlphaFold, greatly surpassing the performance of unbound protein-protein docking (9% success rate for near-native top-ranked models), however AlphaFold modeling of antibody-antigen complexes within our set was unsuccessful. We identified sequence and structural features associated with lack of AlphaFold success, and we also investigated the impact of multiple sequence alignment input. Benchmarking of a multimer-optimized version of AlphaFold (AlphaFold-Multimer) with a set of recently released antibody-antigen structures confirmed a low rate of success for antibody-antigen complexes (11% success), and we found that T cell receptor-antigen complexes are likewise not accurately modeled by that algorithm, showing that adaptive immune recognition poses a challenge for the current AlphaFold algorithm and model. Overall, our study demonstrates that end-to-end deep learning can accurately model many transient protein complexes, and highlights areas of improvement for future developments to reliably model any protein-protein interaction of interest.

Structural Features of Antibody-Peptide Recognition.

Antibody recognition of antigens is a critical element of adaptive immunity. One key class of antibody-antigen complexes is comprised of antibodies targeting linear epitopes of proteins, which in some cases are conserved elements of viruses and pathogens of relevance for vaccine design and immunotherapy. Here we report a detailed analysis of the structural and interface features of this class of complexes, based on a set of nearly 200 nonredundant high resolution antibody-peptide complex structures that were assembled from the Protein Data Bank. We found that antibody-bound peptides adopt a broad range of conformations, often displaying limited secondary structure, and that the same peptide sequence bound by different antibodies can in many cases exhibit varying conformations. Propensities of contacts with antibody loops and extent of antibody binding conformational changes were found to be broadly similar to those for antibodies in complex with larger protein antigens. However, antibody-peptide interfaces showed lower buried surface areas and fewer hydrogen bonds than antibody-protein antigen complexes, while calculated binding energy per buried interface area was found to be higher on average for antibody-peptide interfaces, likely due in part to a greater proportion of buried hydrophobic residues and higher shape complementarity. This dataset and these observations can be of use for future studies focused on this class of interactions, including predictive computational modeling efforts and the design of antibodies or epitope-based vaccine immunogens.

Induction of broadly neutralizing antibodies using a secreted form of the hepatitis C virus E1E2 heterodimer as a vaccine candidate.

SignificanceHepatitis C virus chronically infects approximately 1% of the world's population, making an effective vaccine for hepatitis C virus a major unmet public health need. The membrane-associated E1E2 envelope glycoprotein has been used in clinical studies as a vaccine candidate. However, limited neutralization breadth and difficulty in producing large amounts of homogeneous membrane-associated E1E2 have hampered efforts to develop an E1E2-based vaccine. Our previous work described the design and biochemical validation of a native-like soluble secreted form of E1E2 (sE1E2). Here, we describe the immunogenic characterization of the sE1E2 complex. sE1E2 elicited broadly neutralizing antibodies in immunized mice, with increased neutralization breadth relative to the membrane-associated E1E2, thereby validating this platform as a promising model system for vaccine development.

An extended motif in the SARS-CoV-2 spike modulates binding and release of host coatomer in retrograde trafficking.

ß-Coronaviruses such as SARS-CoV-2 hijack coatomer protein-I (COPI) for spike protein retrograde trafficking to the progeny assembly site in endoplasmic reticulum-Golgi intermediate compartment (ERGIC). However, limited residue-level details are available into how the spike interacts with COPI. Here we identify an extended COPI binding motif in the spike that encompasses the canonical K-x-H dibasic sequence. This motif demonstrates selectivity for αCOPI subunit. Guided by an in silico analysis of dibasic motifs in the human proteome, we employ mutagenesis and binding assays to show that the spike motif terminal residues are critical modulators of complex dissociation, which is essential for spike release in ERGIC. αCOPI residues critical for spike motif binding are elucidated by mutagenesis and crystallography and found to be conserved in the zoonotic reservoirs, bats, pangolins, camels, and in humans. Collectively, our investigation on the spike motif identifies key COPI binding determinants with implications for retrograde trafficking.

T cell receptors employ diverse strategies to target a p53 cancer neoantigen.

Adoptive cell therapy with tumor-specific T cells can mediate durable cancer regression. The prime target of tumor-specific T cells are neoantigens arising from mutations in self-proteins during malignant transformation. To understand T cell recognition of cancer neoantigens at the atomic level, we studied oligoclonal T cell receptors (TCRs) that recognize a neoepitope arising from a driver mutation in the p53 oncogene (p53R175H) presented by the major histocompatibility complex class I molecule HLA-A2. We previously reported the structures of three p53R175H-specific TCRs (38-10, 12-6, and 1a2) bound to p53R175H and HLA-A2. The structures showed that these TCRs discriminate between WT and mutant p53 by forming extensive interactions with the R175H mutation. Here, we report the structure of a fourth p53R175H-specific TCR (6-11) in complex with p53R175H and HLA-A2. In contrast to 38-10, 12-6, and 1a2, TCR 6-11 makes no direct contacts with the R175H mutation, yet is still able to distinguish mutant from WT p53. Structure-based in silico mutagenesis revealed that the 60-fold loss in 6-11 binding affinity for WT p53 compared to p53R175H is mainly due to the higher energetic cost of desolvating R175 in the WT p53 peptide during complex formation than H175 in the mutant. This indirect strategy for preferential neoantigen recognition by 6-11 is fundamentally different from the direct strategies employed by other TCRs and highlights the multiplicity of solutions to recognizing p53R175H with sufficient selectivity to mediate T cell killing of tumor but not normal cells.

Structural assessment of HLA-A2-restricted SARS-CoV-2 spike epitopes recognized by public and private T-cell receptors.

T cells play a vital role in combatting SARS-CoV-2 and forming long-term memory responses. Whereas extensive structural information is available on neutralizing antibodies against SARS-CoV-2, such information on SARS-CoV-2-specific T-cell receptors (TCRs) bound to their peptide-MHC targets is lacking. Here we determine the structures of a public and a private TCR from COVID-19 convalescent patients in complex with HLA-A2 and two SARS-CoV-2 spike protein epitopes (YLQ and RLQ). The structures reveal the basis for selection of particular TRAV and TRBV germline genes by the public but not the private TCR, and for the ability of the TCRs to recognize natural variants of RLQ but not YLQ. Neither TCR recognizes homologous epitopes from human seasonal coronaviruses. By elucidating the mechanism for TCR recognition of an immunodominant yet variable epitope (YLQ) and a conserved but less commonly targeted epitope (RLQ), this study can inform prospective efforts to design vaccines to elicit pan-coronavirus immunity.

An Antigenically Diverse, Representative Panel of Envelope Glycoproteins for Hepatitis C Virus Vaccine Development.

BACKGROUND & AIMS: Development of a prophylactic hepatitis C virus (HCV) vaccine will require accurate and reproducible measurement of neutralizing breadth of vaccine-induced antibodies. Currently available HCV panels may not adequately represent the genetic and antigenic diversity of circulating HCV strains, and the lack of standardization of these panels makes it difficult to compare neutralization results obtained in different studies. Here, we describe the selection and validation of a genetically and antigenically diverse reference panel of 15 HCV pseudoparticles (HCVpps) for neutralization assays. METHODS: We chose 75 envelope (E1E2) clones to maximize representation of natural polymorphisms observed in circulating HCV isolates, and 65 of these clones generated functional HCVpps. Neutralization sensitivity of these HCVpps varied widely. HCVpps clustered into 15 distinct groups based on patterns of relative sensitivity to 7 broadly neutralizing monoclonal antibodies. We used these data to select a final panel of 15 antigenically representative HCVpps. RESULTS: Both the 65 and 15 HCVpp panels span 4 tiers of neutralization sensitivity, and neutralizing breadth measurements for 7 broadly neutralizing monoclonal antibodies were nearly equivalent using either panel. Differences in neutralization sensitivity between HCVpps were independent of genetic distances between E1E2 clones. CONCLUSIONS: Neutralizing breadth of HCV antibodies should be defined using viruses spanning multiple tiers of neutralization sensitivity rather than panels selected solely for genetic diversity. We propose that this multitier reference panel could be adopted as a standard for the measurement of neutralizing antibody potency and breadth, facilitating meaningful comparisons of neutralization results from vaccine studies in different laboratories.

Molecular Determinants of Filament Capping Proteins Required for the Formation of Functional Flagella in Gram-Negative Bacteria.

Bacterial flagella are cell surface protein appendages that are critical for motility and pathogenesis. Flagellar filaments are tubular structures constructed from thousands of copies of the protein flagellin, or FliC, arranged in helical fashion. Individual unfolded FliC subunits traverse the filament pore and are folded and sorted into place with the assistance of the flagellar capping protein complex, an oligomer of the FliD protein. The FliD filament cap is a stool-like structure, with its D2 and D3 domains forming a flat head region, and its D1 domain leg-like structures extending perpendicularly from the head towards the inner core of the filament. Here, using an approach combining bacterial genetics, motility assays, electron microscopy and molecular modeling, we define, in numerous Gram-negative bacteria, which regions of FliD are critical for interaction with FliC subunits and result in the formation of functional flagella. Our data indicate that the D1 domain of FliD is its sole functionally important domain, and that its flexible coiled coil region comprised of helices at its extreme N- and C-termini controls compatibility with the FliC filament. FliD sequences from different bacterial species in the head region are well tolerated. Additionally, head domains can be replaced by small peptides and larger head domains from different species and still produce functional flagella.

Structural and energetic profiling of SARS-CoV-2 receptor binding domain antibody recognition and the impact of circulating variants.

The SARS-CoV-2 pandemic highlights the need for a detailed molecular understanding of protective antibody responses. This is underscored by the emergence and spread of SARS-CoV-2 variants, including Alpha (B.1.1.7) and Delta (B.1.617.2), some of which appear to be less effectively targeted by current monoclonal antibodies and vaccines. Here we report a high resolution and comprehensive map of antibody recognition of the SARS-CoV-2 spike receptor binding domain (RBD), which is the target of most neutralizing antibodies, using computational structural analysis. With a dataset of nonredundant experimentally determined antibody-RBD structures, we classified antibodies by RBD residue binding determinants using unsupervised clustering. We also identified the energetic and conservation features of epitope residues and assessed the capacity of viral variant mutations to disrupt antibody recognition, revealing sets of antibodies predicted to effectively target recently described viral variants. This detailed structure-based reference of antibody RBD recognition signatures can inform therapeutic and vaccine design strategies.