Residue 81 confers a restricted C-terminal peptide binding motif in HLA-B*44:09

ACME: pan-specific peptide–MHC class I binding prediction through attention-based deep neural networks

Motivation

Prediction of peptide binding to the major histocompatibility complex (MHC) plays a vital role in the development of therapeutic vaccines for the treatment of cancer. Algorithms with improved correlations between predicted and actual binding affinities are needed to increase precision and reduce the number of false positive predictions.

Results

We present ACME (Attention-based Convolutional neural networks for MHC Epitope binding prediction), a new pan-specific algorithm to accurately predict the binding affinities between peptides and MHC class I molecules, even for those new alleles that are not seen in the training data. Extensive tests have demonstrated that ACME can significantly outperform other state-of-the-art prediction methods with an increase of the Pearson correlation coefficient between predicted and measured binding affinities by up to 23 percentage points. In addition, its ability to identify strong-binding peptides has been experimentally validated. Moreover, by integrating the convolutional neural network with attention mechanism, ACME is able to extract interpretable patterns that can provide useful and detailed insights into the binding preferences between peptides and their MHC partners. All these results have demonstrated that ACME can provide a powerful and practically useful tool for the studies of peptide–MHC class I interactions.

Availability and implementation

ACME is available as an open source software at https://github.com/HYsxe/ACME.

Supplementary information

Supplementary data are available at Bioinformatics online.

The diversity of the HLA-E-restricted peptide repertoire explains the immunological impact of the Arg107Gly mismatch

Human leukocyte antigen (HLA)-E molecules are potent inhibitors of NK cell-mediated killing. Low in polymorphisms, two alleles are widely expressed among diverse populations: HLA-E*01:01 and HLA-E*01:03. Both alleles are distinguished by one SNP resulting in the substitution Arg107Gly. Both alleles present a limited set of peptides derived from class I leader sequences physiologically; however, HLA-E*01:01 presents non-canonical peptides in the absence of HLA class I molecules. To further assess the functional differences between both alleles, we analyzed the peptide repertoire of HLA-E*01:03 by applying soluble HLA technology followed by mass-spectrometric peptide sequencing. HLA-E*01:03 restricted peptides showed a length of 9-17 amino acids and differed in their biophysical properties, no overlap in the peptide repertoire of both allelic variants could be observed; however, both alleles shared marginal peptides from the same proteomic content. Artificial APCs expressing empty HLA-E*01:01 or E*01:03 molecules were generated and stabilized using cognate HLA class I-derived peptide ligands to analyze the impact of residue 107 within the HLA-E heavy chain on the NKG2/CD94 receptor engagement. Differences in peptide stabilization could be translated to the density and half-life time of peptide-HLA-E molecules on the cell surface that subsequently impacted NK cell inhibition as verified by cytotoxicity assays. Taken together, these data illustrate functional differences of HLA-E allelic variants induced by a single amino acid. Furthermore, the function of HLA-E in pathophysiologic situations when the HLA processing machinery is interrupted seems to be more emphasized than previously described, implying a crucial role for HLA-E in tumor or viral immune episodes.

HLA-E: Presentation of a Broader Peptide Repertoire Impacts the Cellular Immune Response-Implications on HSCT Outcome

The HLA-E locus encodes a nonclassical class Ib molecule that serves many immune functions from inhibiting NK cells to activating CTLs. Structural analysis of HLA-E/NKG2A complexes visualized fine-tuning of protective immune responses through AA interactions between HLA-E, the bound peptide, and NKG2A/CD94. A loss of cellular protection through abrogation of the HLA-E/NKG2A engagement is dependent on the HLA-E bound peptide. The role of HLA-E in posttransplant outcomes is not well understood but might be attributed to its peptide repertoire. To investigate the self-peptide repertoire of HLA-E^∗01:01 in the absence of protective HLA class I signal peptides, we utilized soluble HLA technology in class I negative LCL cells in order to characterize HLA-E^∗01:01-bound ligands by mass-spectrometry. To understand the immunological impact of these analyzed ligands on NK cell reactivity, we performed cellular assays. Synthesized peptides were loaded onto recombinant T2 cells expressing HLA-E^∗01:01 molecules and applied in cytotoxicity assays using the leukemia derived NK cell line (NKL) as effector. HLA-E in complex with the self-peptides demonstrated a shift towards cytotoxicity and a loss of cell protection. Our data highlights the fact that the HLA-E-peptidome is not as restricted as previously thought and support the suggestion of a posttransplant role for HLA-E.

A Micropolymorphism Altering the Residue Triad 97/114/156 Determines the Relative Levels of Tapasin Independence and Distinct Peptide Profiles for HLA-A(*)24 Allotypes

While many HLA class I molecules interact directly with the peptide loading complex (PLC) for conventional loading of peptides certain class I molecules are able to present peptides in a way that circumvents the PLC components. We investigated micropolymorphisms at position 156 of HLA-A(*)24 allotypes and their effects on PLC dependence for assembly and peptide binding specificities. HLA-A(*)24:06(156Trp) and HLA-A(*)24:13(156Leu) showed high levels of cell surface expression while HLA-A(*)24:02(156Gln) was expressed at low levels in tapasin deficient cells. Peptides presented by these allelic variants showed distinct differences in features and repertoire. Immunoprecipitation experiments demonstrated all the HLA-A(*)24/156 variants to associate at similar levels with tapasin when present. Structurally, HLA-A(*)24:02 contains the residue triad Met97/His114/Gln156 and a Trp156 or Leu156 polymorphism provides tapasin independence by stabilizing these triad residues, thus generating an energetically stable and a more peptide receptive environment. Micropolymorphisms at position 156 can influence the generic peptide loading pathway for HLA-A(*)24 by altering their tapasin dependence for peptide selection. The trade-off for this tapasin independence could be the presentation of unusual ligands by these alleles, imposing significant risk following hematopoietic stem cell transplantation (HSCT).

HLA-E: a novel player for histocompatibility

The classical class I human leukocyte antigens (HLA-A, -B, and -C) present allele-specific self- or pathogenic peptides originated by intracellular processing to CD8(+) immune effector cells. Even a single mismatch in the heavy chain (hc) of an HLA class I molecule can impact on the peptide binding profile. Since HLA class I molecules are highly polymorphic and most of their polymorphisms affect the peptide binding region (PBR), it becomes obvious that systematic HLA matching is crucial in determining the outcome of transplantation. The opposite holds true for the nonclassical HLA class I molecule HLA-E. HLA-E polymorphism is restricted to two functional versions and is thought to present a limited set of highly conserved peptides derived from class I leader sequences. However, HLA-E appears to be a ligand for the innate and adaptive immune system, where the immunological response to peptide-HLA-E complexes is dictated through the sequence of the bound peptide. Structural investigations clearly demonstrate how subtle amino acid differences impact the strength and response of the cognate CD94/NKG2 or T cell receptor.

Soluble HLA technology as a strategy to evaluate the impact of HLA mismatches

HLA class I incompatibilities still remain one of the main barriers for unrelated bone marrow transplantation (BMT); hence the molecular understanding of how to mismatch patients and donors and still have successful clinical outcomes will guide towards the future of unrelated BMT. One way to estimate the magnitude of polymorphisms within the PBR is to determine which peptides can be selected by individual HLA alleles and subsequently presented for recognition by T cells. The features (structure, length, and sequence) of different peptides each confer an individual pHLA landscape and thus directly shape the individual immune response. The elution and sequencing of peptides by mass spectrometric analysis enable determining the bona fide repertoire of presented peptides for a given allele. This is an effective and simple way to compare the functions of allelic variants and make a first assessment of their degree of permissivity. We describe the methodology used for peptide sequencing and the limitations of peptide prediction tools compared to experimental methods. We highlight the altered peptide features that are observed between allelic variants and the need to discover the altered peptide repertoire in situations of "artificial" graft versus host disease (GvHD) that occur in HLA-specific hypersensitive immune responses to drugs.

Dysregulation of cell cycle control caused by overexpression of the oncogene pp32r1 (ANP32C) and the Tyr>His mutant pp32r1Y140H

The pp32 (ANP32A) gene acts as a tumor suppressor while its closely related homologue pp32r1 (ANP32C) is oncogenic and is overexpressed in breast, prostate and pancreatic tumors. The transduction of p53wt cell lines (ACHN and HeLa) with pp32r1 or pp32r1Y140H lentivirus increased the proliferation of p53wt cell lines compared to the untransduced control cells while transduction of the p53(R248W) MiaPaCa2 cell line had no effect. Cell cycle analysis of transduced ACHN cells by PI staining and BrdU incorporation illustrated a pronounced shift toward the S-phase of the cell cycle in cells overexpressing the pp32r1 and pp32r1Y140H proteins. Confocal microscopy and western blotting demonstrated that pp32r1 and the pp32r1Y140H mutant protein reside predominantly in the cytoplasm in constrast to pp32 which is a nuclear/cytoplasmic shuttling protein. To determine the effects of pp32r1 or pp32r1Y140H overexpression at the proteomic level we performed a comprehensive proteome analysis on ACHN, ACHN-pp32r1 and ACHN-pp32r1Y140H cell lysates using the isotope-coded protein label (ICPL) method. Among those proteins with >40% regulation were Macrophage Capping protein (CAPG) and Chromodomain Helicase DNA binding protein 4 (CHD4) proteins which were significantly upregulated by pp32r1 and pp32r1Y140H overexpression. This increase in CHD4 also appears to influence a number of cell cycle regulator genes including; p53, p21 and cyclinD1 as judged by western blotting. Silencing of CHD4 in ACHN-pp32r1Y140H cells using specific shRNA reverted the cell cycle dysregulation caused by pp32r1Y140H expression to that of the untransduced ACHN cell line, suggesting that CHD4 is the prominent effector of the pp32r1/pp32r1Y140H phenotype.