This study demonstrates that DQ2 and DQ8 have preference for binding peptides with negatively charged anchor residues, but that the 2 HLA molecules employed different criteria for selection of deamidated gluten T cell epitopes. This can result in the selection of distinct epitopes localized in different regions of a gliadin protein, but it can also result in the selection of epitopes that combine the DQ2 and DQ8 signatures and are recognized in exactly the same binding register when bound to DQ2 or DQ8. These findings further expand our understanding of the mechanisms underlying the HLA association in celiac disease. Moreover, our observation —which we believe to be novel — that DQ heterodimers, which are encoded in trans in DQ2/DQ8 heterozygotes are functional for antigen presentation and that peptides that combine the DQ2 and DQ8 binding signatures can be presented by such trans-encoded dimers to T cells have relevance for understanding the molecular basis of the HLA association in type 1 diabetes.
The superimposable reactivity patterns toward the peptide panels of the T cell lines established by stimulation of biopsies with chymotrypsin-digested whole gluten, chymotrypsin-digested recombinant gliadin, or the collection of overlapping peptides indicate that there is no bias in our detection of epitopes. In AJ133612 α-gliadin, the DQ2- and DQ8-restricted T cell responses were localized to 2 different regions of the protein (regions 1 and 2). This observation suggested that the DQ2 and DQ8 molecules select for different epitopes. The overlapping recognition patterns of M36999 γ-gliadin peptides by DQ2- and DQ8-restricted T cell lines were therefore surprising. However, further analysis with T cell clones revealed that the DQ2- and DQ8-restricted T cells recognized different features of the M36999 γ-gliadin peptides. We found DQ8-restricted T cell clones that were reactive to peptides harboring the sequence QQPQQPFPQ following TG2 treatment. This sequence is expressed in 4 of the individual M36999 peptides, and reactivity to this epitope is a major contributor to the observed DQ8-associated reactivity pattern of the polyclonal T cell lines. The QQPQQPFPQ sequence is identical to the core region of the previously characterized DQ2–γ-VII epitope. The DQ2- and DQ8-restricted T cells recognized this peptide in the same register, but with different requirements for deamidation. A similar sequence, QQPQQPYPQ, is found in another γ-gliadin (AJ416339), and this is the core sequence of the previously characterized DQ2–γ-III epitope. We found that the DQ8-restricted T cell clones also recognized TG2-treated peptides harboring this sequence. Again the requirements for deamidation were found to be different for the DQ2- and DQ8-restricted T cell clones, following the same pattern as for the QQPQQPFQ sequence. For the DQ2-restricted T cells, deamidation at position P4 was mandatory for recognition, whereas for the DQ8-restricted T cells, deamidation at positions P1 and/or P9 were important for recognition.
The criteria employed by DQ2 and DQ8 molecules for selecting epitopes were seen from alignment of the core regions of DQ2- and DQ8-restricted gliadin epitopes (Table ). Both DQ2 and DQ8 have a preference for binding of negatively charged residues. For the DQ2-restricted gluten T cell epitopes, glutamate residues formed by TG2-mediated deamidation were found in P1, P4, P6, P7, and P9, but only deamidation in P4 and P6 — and, rarely, P7 — seem to be crucial for T cell recognition (19
). For the DQ8-restricted gluten T cell epitopes, glutamate residues formed by TG2 were found in positions P1, P4, and P9, but in contrast to DQ2, only deamidations in the positions P1 and/or P9 are critical for T cell recognition. This preference for binding of negatively charged residues in the positions P1 and P9 is consistent with other studies of the DQ8 binding motif (11
). It is striking that, for both DQ2- and DQ8-restricted T cells, it is deamidation at positions with presumed orientation of the side chains toward the MHC that affects T cell recognition. Notably, the peptide binding preferences we observed for DQ8 and DQ2 are in accordance with the X-ray crystal structures of these molecules (28
Alignment of the core region of 5 DQ2-restricted and 2 DQ8-restricted gliadin epitopes
DQ2 binds gluten peptides with the proline residues localized in P1, P3, P5, P6, and P8 but not in P2, P4, P7, or P9 (29
). This pattern is similar for DQ8, which bind peptides with proline residues in P3, P6, and P8 (Table ). An important difference between DQ2 and DQ8, however, is at P1. In most MHC class II molecules, including DQ8, there is a hydrogen bond between the amide nitrogen of the P1 residue and the backbone carbonyl of residue α53 (28
). This appears not to be the case for DQ2. A deletion of the α53 residue of DQA1*05
possibly prevents the establishment of a hydrogen bond to the P1 amide, and proline residues can thereby be accommodated at P1 without penalty (29
). The majority of the characterized DQ2-restricted gluten T cell epitopes have proline residues at P1. These epitopes would likely be unavailable for binding to DQ8 in the same binding register, and the inability of the tested DQ2–α-I– or DQ2–α-II–restricted T cell clones to recognize their epitopes in the context of DQA1*03
gives support to this notion. The fact that DQ2 is better suited than DQ8 to bind the proline-rich gluten peptides that survive gastrointestinal digestion may be the reason why DQ2 is a stronger susceptibility determinant for celiac disease than DQ8.
Celiac disease and type 1 diabetes are both associated with DQ2 (DQA1*05/DQB1*02
) and DQ8 (DQA1*03/DQB1*0302
). In celiac disease, the major susceptibility factor is DQ2, whereas DQ8 adds a small risk independent of DQ2 (3
). In type 1 diabetes, DQ8 is a stronger susceptibility factor than DQ2, and the risk associated with DQ2/DQ8 heterozygosity supersedes the combined risks associated with DQ2 and DQ8 (22
). This has led to the hypothesis that trans
-encoded dimers, i.e., DQA1*05/DQB1*0302
, are more effective to present diabetogenic epitope(s) to T cells (33
). Both the DQA1*05/DQB1*0302
and the DQA1*03/DQB1*02
dimers are shown to be expressed by DR3-DQ2/DR4-DQ8 heterozygous cells (34
). Which epitopes are involved in human type 1 diabetes and what characteristics they should have are basically unknown, although there are suggestions in the literature (35
), including posttranslationally modified antigens (38
). There is no existing evidence for a role of TG2 in the pathogenesis of type 1 diabetes by deamidating antigens, although this possibility cannot be excluded either. The molecular understanding of HLA association in celiac disease has made huge advances in recent years, much of it because of the identification of disease-relevant gluten cell epitopes. A similar advance has not taken place for type 1 diabetes, and an obvious obstacle is the lack of knowledge of disease-relevant T cell epitopes; to define them is a major goal. The gluten epitopes recognized by intestinal T cells of celiac disease patients are naturally selected by DQ2 and DQ8 and they are disease relevant. Thus this model system has advantages over transgenic mouse systems, in which the T cell epitopes studied are often the result of forced immunization with the use of adjuvants.
The trans-encoded heterodimers can possibly present a unique peptide or set of peptides. Alternatively, a peptide or limited set of peptides that could be presented by both DQ2 and DQ8 could be even more effectively presented by the trans-encoded heterodimers. The findings of this study provide support for the latter model, as we demonstrated the existence of sequence-related peptides that bound to DQ2 and DQ8 in the same registers and did so by incorporating both the DQ2 and the DQ8 binding motifs. Of particular interest is our observation that a peptide with glutamate residues at P1 and P4 was presented more effectively by 3 logs in the context of DQA1*03/DQB1*02 than in the context of DQA1*05/DQB1*02, presumably because the P1 pocket of the DQA1*03/DQB1*02 molecule better accommodates the negatively charged glutamate side chain. We cannot exclude the possibility that this effect is mediated at the level of the TCR, and future work needs to corroborate this notion by peptide binding analysis. Moreover, our testing of T cell recognition of peptides in the context of the trans-encoded heterodimers was suboptimal, as we used T cell clones that were screened and selected for their ability to recognize peptides in the context of encoded DQ2 or DQ8. Screening T cell clones from DQ2/DQ8 heterozygous individuals that recognize peptides in the context of DQA1*05/DQB1*0302 or DQA1*03/DQB1*02 should facilitate the characterization of peptides that are selected for presentation by the diabetes-related trans-encoded DQ heterodimers. Whether peptides that carry a negative charge at P6 or P7 in addition to the negative charge at P1 and/or P9 can also be better presented by the DQA1*03/DQB1*02 heterodimer is one of the questions that should be addressed.
Our observations also raise the question of why there is no synergistic effect between DQ2 and DQ8 as predisposing elements in celiac disease. This may be explained by the fact that in the overall T cell response to gluten in celiac disease patients, the responses to the DQ2–γ-III/DQ2–γ-VII epitopes are minor contributors, whereas the main response is directed against the DQ2–α-I, DQ2–α-II, and DQ2–α-III epitopes (19
), which as we showed here did not overlap with DQ8-restricted epitopes and for which we have found no evidence for presentation by any of the trans
In summary, this work give details of antigen presentation by DQ2 and DQ8 molecules that further expand the knowledge of the HLA association in celiac disease and allow us to predict essential features of peptides that are involved in type 1 diabetes.