The role of HLA-DQ8 β57 polymorphism in the anti-gluten T-cell response in coeliac disease

A specific assessment of the role of MHC class II alleles in antigen-specific T-cell responses is possible in coeliac disease, because it is caused by a single and well-defined antigen, gluten, which is present in wheat, rye and barley (reviewed in ref. 12). Gluten is rich in glutamine and proline, and the high content of proline renders gluten resistant to intestinal enzymatic digestion, explaining the persistence of intact peptides for immune recognition. Notably, despite the presence of few negatively charged amino acid residues in gluten and the expression of other MHC alleles potentially capable of presenting gluten peptides, all gluten-specific CD4⁺ T cells isolated from the intestinal mucosa of adult patients are restricted by HLA-DQ2 or HLA-DQ8 (refs 10, 13 and 14), and most recognize a dominant peptide with a deamidated Gln—that is, a negatively charged Glu residue^5,6.

To study the T-cell response to native and deamidated gluten peptides (Fig. 1a), we immunized humanized HLA-DQ8 mice with the native 24-amino-acid gluten peptide 219–242 from the α2 (AJ133612) gliadin comprising the commonly recognized DQ8-α–I epitope (QGSFQPSQQ)¹⁰, or with a peptide deamidated at positions 229 and 237, the known targets of TG2 (ref. 6). As previously reported¹⁵, the deamidated peptide failed to induce superior effector or recall proliferative (Fig. 1b, left panel and Supplementary Fig. 1a, left panel) or cytokine (Supplementary Fig. 1b, left panels) CD4⁺ T-cell responses. Similar responses were also observed for lower, limiting doses of immunizing peptide (data not shown). However, under conditions of competition, when the two peptides were administered simultaneously, we found that the response against the deamidated peptide largely predominated over the native peptide (Fig. 1b, right panel and Supplementary Fig. 1, right panels). When peptides with single Gln to Glu substitutions at position 229 or 237 (Fig. 1a) were compared in the same competition assay against the native peptide, the E237 peptide behaved like the deamidated peptide, whereas the E229 peptide resembled the native peptide (Fig. 1b middle and lower panels). In accordance with previous studies concluding that the Glu residue at position 229 binds to pocket 1 (P1) of HLA-DQ8, whereas Glu at position 237 binds to pocket 9 (P9)^6,9,16, these results indicate that an acidic residue at position P9 confers a functional advantage with respect to immunization.

Using a panel of T-cell hybridomas derived from immunized lymph nodes, we compared the specificity and T-cell receptor (TCR) repertoire of CD4⁺ T cells elicited by the native and the deamidated peptides. Immunization with native and deamidated peptides induced crossreactive T cells, in addition to T cells recognizing specifically one or the other form of the peptide. Notably, nearly a quarter of the hybridomas generated against the native peptide showed a ‘heteroclitic’ response¹⁷ against the deamidated peptide, in that they responded much better to the deamidated than to the native peptide, with a shift in the dose response curve of up to 100-fold (Fig. 1c and Supplementary Fig. 2). Immunization with the native peptide, however, did not induce T cells that recognized the deamidated peptide exclusively, confirming the absence of in vivo deamidation in this experimental setting. Importantly, a heteroclitic response was never observed after immunization with the deamidated peptide.

Analysis of the TCR repertoire (Fig. 1d) revealed that most T cells recognizing the native peptide—whether elicited by immunization with the native or the deamidated peptides—preferentially used a limited set of Vβ chains including Vβ11, Vβ5, Vβ14 and Vβ2 T cells. T cells that exclusively recognized the native peptide predominantly expressed Vβ11 (45%) and to a lesser degree Vβ14 (15%), whereas T cells that equivalently recognized the two peptides mostly expressed Vβ5 (75%). In marked contrast, T cells specific for the deamidated peptide showed a broad TCR repertoire with no preferential Vβ (Fig. 1d) or Vα chain use (data not shown).

By studying the response of T-cell hybridomas to amino- and carboxy-terminal deletion mutant peptides, we identified five different patterns. Most (82%) of the T cells specific for the native peptide responded to the minimal ²²⁹QGSFQPSQQ²³⁷ 9-amino-acid determinant (pattern 1), whereas the remaining clones responded to the minimal 9-amino-acid ²²⁸GQGSFQPSQ²³⁶ determinant (pattern 2; Supplementary Table 1, lower panel). Hybridomas responding equally well to native and deamidated peptides also showed a relatively restricted response pattern; that is, pattern 1 and 4 accounted for 78% of these hybridomas (Supplementary Table 1, lower panel). In contrast, T cells specific for the deamidated peptide showed at least four different response patterns, none of which was predominant (Supplementary Table 1, lower panel). Notably, response pattern 2, characterized by a response to ²²⁸GEGSFQPSQ²³⁶, was never observed for T cells specific for the deamidated peptide (Supplementary Table 1, lower panel), indicating a requirement for the acidic residue at position 237. The diversity of response patterns observed with T cells specific for the deamidated peptide might be due to shifts in the peptide register, as the structure of MHC class II molecules allows peptides to frameshift within the MHC groove¹⁸. Alternatively, a single register might be occupied by the deamidated peptide, but the diverse TCRs could contact different residues. Altogether, these results indicate that recognition of the native peptide imposes strict requirements leading to limited TCR use, whereas by displaying several epitopes by a yet undetermined mechanism, the deamidated peptide recruits a far broader TCR repertoire.

Examination of TCR sequences revealed the notable conservation of a negative charge at position 3 of CDR3β in most T cells recognizing the native peptide, including 92% of those specific for the native peptide, and 60–92% of those that responded to both the native and the deamidated peptides (Table 1). Importantly, 91% of the negative charges were non-germline-encoded, indicating an antigen-driven selection process (Supplementary Table 2). During TCR V(D)J recombination, further diversity is generated by the stochastic addition and deletion of palindromic and non-germline nucleotides at the coding joints between the gene segments. Specifically, here more than 35% of the non-germline-encoded negatively charged residues resulted from one non-germline nucleotide addition, 19% from three palindromic nucleotide additions and the rest from two non-germline, three non-germline, or one palindromic nucleotide additions and from the combination of two palindromic and one non-germline nucleotide additions (Supplementary Table 2). Interestingly, published TCR sequences from early islet infiltrates in NOD mice are also characterized by a conserved negative charge at position 3 of the CDR3β loop^19,20. In contrast, only 29% of T cells specific for the deamidated gluten peptide had a negative charge at that position (Table 1). A possible model accounting for these findings is that HLA-DQ8- and IA^g7-restricted TCRs against peptides lacking acidic residues must use a negative charge in their CDR3β to stabilize peptide–MHC (pMHC) complexes through charge–charge interactions.

To test whether the negative charge at position 3 of CDR3β is required for recognition of the native peptide, we generated a TCR mutant lacking the conserved negative charge (Fig. 2a), as well as a negatively charged HLA-DQ8 mutant in which an Asp was introduced at position β57 (Fig. 2b). The TCR was obtained from a representative Vβ5.1/Vα17 hybridoma that responded equivalently to the native and deamidated peptides and was retrovirally transferred into 58α⁻β⁻ T cells. Notably, mutation of the conserved negative charge at position 3 of CDR3β (Vβ5.1 (E→Q)) resulted in complete loss of recognition of the native peptide, whereas the response to the deamidated peptide was preserved (Fig. 2a, middle panel). As a control, mutation of a non-conserved negative charge at position 8 of CDR3β (Vβ5.1 (D→N)) did not alter the response to either native or deamidated peptides (Fig. 2a, right panel). We next performed the reciprocal experiment in which HLA-DQ8 was mutated to introduce an Asp at position β57. Using T-cell hybridomas responding equally well to native and deamidated peptides, we observed a marked loss of response to the native peptide (Fig. 2b, left panel). Loss of recognition of the native peptide with the Asp 57 DQ8 mutant was also confirmed with three hybridomas recognizing selectively native peptides (data not shown). In contrast, recognition of the deamidated peptide was only moderately affected (Fig. 2b, right panel), consistent with previous studies showing that MHC alleles carrying an Asp at position β57 can bind peptides harbouring acidic residues. These results support the hypothesis that HLA-DQ8 β57 polymorphism can elicit strong responses against non-negatively charged peptides by recruiting TCRs that express a negative charge at position 3 of CDR3β.

The amplitude of the T-cell response results from the combined strength of the binding of peptide to MHC and the binding of TCR to pMHC complexes. The strong response elicited by the native peptide was surprising given the lack of a negative charge to match the positive P9 pocket in HLA-DQ8. Indeed, whereas the deamidated peptide readily competed with the binding of indicator peptide to HLA-DQ8, binding of the native peptide was below detection (Supplementary Fig. 3a). When indirectly measured by hybridoma stimulation using a clone responding to both peptides, the binding of the native peptide exhibited similar on-rate but increased off-rate (Supplementary Fig. 3b). Overall these results are in accordance with previous observations showing strong in vivo and in vitro T-cell responses against peptides for which no binding to MHC could be detected by direct binding assays^21,22. Furthermore, our data suggest that the instability of complexes between native peptide and HLA-DQ8 is compensated by the presence of a negatively charged residue in the CDR3β loop. In agreement with these findings, the presence of T cells during the short incubation time (1 h) of antigen-presenting cells with gluten peptides influenced the magnitude of their response against the native but not the deamidated peptide (Supplementary Fig. 3c).

Because the native gluten peptides recruit a T-cell population that is not only largely crossreactive but also substantially heteroclitic against the deamidated peptides, these cells could be recruited and continuously activated once inflammation is triggered and TG2 is activated to produce deamidated peptide. Consistent with this model, quantification of the CD4⁺T-cell response by enzyme-linked immunospot (ELISPOT) showed that the frequency of T cells recognizing the deamidated peptide was consistently highly increased after immunization with a mixture of native and deamidated peptides (Fig. 3).

To assess the pathogenic relevance of our findings further, we examined the TCR structure of human T-cell clones specific for the native DQ8-α–I epitope (QGSFQPSQQ). Notably, six out of six human T-cell clones, obtained from two independent research centres in Leiden and Oslo, expressed a TCR with a non-germline-encoded acidic residue precisely at position 3 of their CDR3β (Table 2). All these negatively charged residues were non-germline encoded and resulted from one non-germline (16.7%), two non-germline (33.3%), two palindromic (33.3%) and three non-germline (16.7%) nucleotide additions (Supplementary Table 3). In contrast, none of the three non gluten-specific T-cell clones derived from the same patients had this signature charge in their CDR3β (Table 3). To integrate these results in a more general context, we analysed 125 published TCR sequences obtained from HLA-DQ8 individuals with type I diabetes (n = 116), Birch (n = 5) and nickel (n = 1) allergy, or HIV infection (n = 3) and found that only 17 had a negatively charged residue at position 3 of the CDR3β loop. Interestingly, all of these 17 TCRs originated from the type I diabetes patients (data not shown).

This study reveals a new mechanism of peptide recognition in the context of HLA-DQ8, the MHC class II allele that predisposes to both type I diabetes and coeliac disease. The P9 pocket of HLA-DQ8, lacking the canonical Asp 57 found in other alleles, has the potential to interact not only with peptides harbouring an acidic residue at the P9 position, but also with TCRs harbouring a negative charge in their CDR3β loop. The potent response generated in vivo against native peptides with extremely low affinity for HLA-DQ8 may be explained by the ability of these TCRs to stabilize pMHC complexes. A large proportion of the TCRs recruited by native peptides display a cross-reactive heteroclitic response to the deamidated peptide, which binds strongly to HLA-DQ8. Furthermore, because deamidated peptides do not impose charge constraints on the TCR, they can recruit an additional TCR repertoire. Consequently, the magnitude of the response against the deamidated peptide is increased under conditions in which native and deamidated peptides are both present. This may be particularly relevant in the context of coeliac disease, because TG2 is probably unevenly activated once inflammation is initiated. Furthermore, the magnitude of the T-cell response is viewed as a limiting factor for disease development^23,24. In the absence of a disease model of coeliac disease, it is at present not possible to assess whether this process is important for disease onset, disease progression or other aspects of the disease. It also remains to be determined whether DQ β57 polymorphism also contributes to the strong association of coeliac disease with HLA-DQ2.

It is tempting to speculate that similar mechanisms may explain the association of type I diabetes with HLA-DQ8. The TCR repertoire of early pancreatic islet infiltrate is also characterized by the frequent presence of a negative charge at position 3 of CDR3β^19,20,25. Furthermore, our preliminary studies suggest that TG2 activity is significantly increased in the pancreas, but not in the draining pancreatic lymph node and the spleen of 12-week-old NOD mice that have ongoing pancreatic inflammation (B.J. and L.T., unpublished observations). The ability to present native and deamidated peptides and recruit crossreactive TCR repertoires directly linked to the polymorphism at position β57 may underlie the remarkable association of HLA-DQ8 with severe immunopathology in mouse and human.