Human T cells recognize HLA-DP–bound peptides in two orientations

Abstract

Human leukocyte antigen (HLA) molecules present small peptide antigens to T cells, thereby allowing them to recognize pathogen-infected and cancer cells. A central dogma over the last 50+ y is that peptide binding to HLA molecules is mediated by the docking of side chains of particular amino acids in the peptide into pockets in the HLA molecules in a conserved N- to C-terminal orientation. Whether peptides can be presented in a reversed C- to N-terminal orientation remains unclear. Here, we performed large-scale identification of peptides bound to HLA-DP molecules and observed that in addition to peptide binding in an N- to C-terminal orientation, in 9 out of 14 HLA-DP allotypes, reverse motifs are found, compatible with C- to N-terminal peptide binding. Moreover, we isolated high-avidity human cytomegalovirus (CMV)-specific HLA-DP–restricted CD4 + T cells from the memory repertoire of healthy donors and demonstrate that such T cells recognized CMV-derived peptides bound to HLA-DPB1*01:01 or *05:01 in a reverse C- to N-terminal manner. Finally, we obtained a high-resolution HLA-DPB1*01:01-CMVpp65 (142–158) peptide crystal structure, which is the molecular basis for C- to N-terminal peptide binding to HLA-DP. Our results point to unique features of HLA-DP molecules that substantially broaden the HLA class II bound peptide repertoire to combat pathogens and eliminate cancer cells.

Accurate prediction of HLA class II antigen presentation across all loci using tailored data acquisition and refined machine learning

“…This suggests that the HLA-DP α chain, although bearing a limited specificity-determining role (see later), is the major determinant for the acceptance of the inverted peptide binding mode. These observations are in alignment with recent studies ( 28 , 33 ), which have shown that widespread inversion of peptide binders is only observed for DP and only for DP molecules with certain α chains (namely, DPA1*02:01 and DPA1*02:02). Although molecules with DPB1*03:01 have not been observed in previous studies to have inverted binders ( 32 ), our method predicts a sizable percentage (~8.1%, 83 of 1029 peptides) of inversions for DPA1*02:01-DPB1*03:01.…”

“…On the basis of this, they hypothesized that for this molecule, peptides can bind both in a canonical (N- to C-terminal) and inverted (C- to N-terminal) orientation. This inverted binding mode was later confirmed experimentally in independent studies ( 28 , 33 ). In the context of T cell epitope prediction, Racle et al.…”

“…On the basis of this, they hypothesized that for this molecule, peptides can bind both in a canonical (N-to C-terminal) and inverted (C-to N-terminal) orientation. This inverted binding mode was later confirmed experimentally in independent studies (28,33). In the context of T cell epitope prediction, Racle et al (28) showed in a recent publication that by incorporating inverted binding prediction into their MixMHC2pred method, they were able to identify several epitopes bound inverted to DPA1*02:01-DPB1*01:01, which elicited a CD4 + immune response.…”

“…Before model training, the MS data were processed and enriched with artificial random natural negative peptides (as described in Materials and Methods). To accommodate the inverted binding mode recently observed for some DP molecules ( 28 , 33 ), we used a modified version of the NNAlign_MA machine learning framework (see fig. S1), which includes an option to simultaneously predict both the binding core offset and the orientation (forward or inverted) of the peptide ligands (for details on the implementation, refer to Materials and Methods).…”

“…In this study, we set out to complete the journey of characterizing the rules of and developing prediction methods for HLA class II antigen presentation. To achieve this, we first compiled a comprehensive immunopeptidomics dataset, integrating the training data for the NetMHCIIpan-4.2 method including DQ-specific immunopeptidomics data covering 14 HLA-DQ molecules (25) with DPspecific data from van Balen et al and related studies covering 19 HLA-DP molecules (32)(33)(34), along with additional data for HLA-DR (24) and BoLA-DR (see table S1 for more information on all the included datasets) (35). Figure 1A gives an overview of this combined dataset in terms of the SA and MA data categories, illustrating that the majority of the training data are derived from MA datasets.…”