Structural and Functional Studies of trans-Encoded HLA-DQ2.3 (DQA1*03:01/DQB1*02:01) Protein Molecule*

Background: trans-Encoded HLA-DQ molecules are biologically interesting, but no structures of such molecules exist. Results: X-ray crystal structure of the trans-encoded DQ2.3 (DQA1*03:01/DQB1*02:01) was determined. Structural data are presented together with functional T-cell data. Conclusion: DQ2.3 has preference for negative charged anchors at P1 and P4. Significance: This work helps to understand why DQ2.3 is associated with a particular risk for type 1 diabetes. MHC class II molecules are composed of one α-chain and one β-chain whose membrane distal interface forms the peptide binding groove. Most of the existing knowledge on MHC class II molecules comes from the cis-encoded variants where the α- and β-chain are encoded on the same chromosome. However, trans-encoded class II MHC molecules, where the α- and β-chain are encoded on opposite chromosomes, can also be expressed. We have studied the trans-encoded class II HLA molecule DQ2.3 (DQA1*03:01/DQB1*02:01) that has received particular attention as it may explain the increased risk of certain individuals to type 1 diabetes. We report the x-ray crystal structure of this HLA molecule complexed with a gluten epitope at 3.05 Å resolution. The gluten epitope, which is the only known HLA-DQ2.3-restricted epitope, is preferentially recognized in the context of the DQ2.3 molecule by T-cell clones of a DQ8/DQ2.5 heterozygous celiac disease patient. This preferential recognition can be explained by improved HLA binding as the epitope combines the peptide-binding motif of DQ2.5 (negative charge at P4) and DQ8 (negative charge at P1). The analysis of the structure of DQ2.3 together with all other available DQ crystal structures and sequences led us to categorize DQA1 and DQB1 genes into two groups where any α-chain and β-chain belonging to the same group are expected to form a stable heterodimer.

MHC class II molecules are composed of one ␣-chain and one ␤-chain whose membrane distal interface forms the peptide binding groove. Most of the existing knowledge on MHC class II molecules comes from the cis-encoded variants where the ␣and ␤-chain are encoded on the same chromosome. However, trans-encoded class II MHC molecules, where the ␣and ␤-chain are encoded on opposite chromosomes, can also be expressed. We have studied the trans-encoded class II HLA molecule DQ2.3 (DQA1*03:01/DQB1*02:01) that has received particular attention as it may explain the increased risk of certain individuals to type 1 diabetes. We report the x-ray crystal structure of this HLA molecule complexed with a gluten epitope at 3.05 Å resolution. The gluten epitope, which is the only known HLA-DQ2.3-restricted epitope, is preferentially recognized in the context of the DQ2.3 molecule by T-cell clones of a DQ8/ DQ2.5 heterozygous celiac disease patient. This preferential recognition can be explained by improved HLA binding as the epitope combines the peptide-binding motif of DQ2.5 (negative charge at P4) and DQ8 (negative charge at P1). The analysis of the structure of DQ2.3 together with all other available DQ crystal structures and sequences led us to categorize DQA1 and DQB1 genes into two groups where any ␣-chain and ␤-chain belonging to the same group are expected to form a stable heterodimer.
Major histocompatibility complex (MHC) proteins play a critical role in immune recognition by displaying antigens to T-cell receptors (TCRs) 5 within the context of MHC-peptide complex to elicit a T-cell-mediated immune response. In humans, MHC proteins are encoded by the human leukocyte antigen (HLA) genes found on chromosome 6. HLA is the single most important genetic factor that predisposes to most autoimmune diseases (1) and contributes 35-50% of the genetic disease association in type 1 diabetes and celiac disease (2,3). CD4 ϩ T-cells recognize antigens in the context of MHC class II molecules that are heterodimers of ␣and ␤-chains. In humans, there are three isotypes of MHC class II molecules: HLA-DR, HLA-DQ, and HLA-DP. In HLA-DR, the polymorphic variation is provided by the ␤-chain alone as the ␣-chain is monomorphic. However, in DQ and DP, both the ␣-chains and the ␤-chains are polymorphic. As a result, unique DQ and DP molecules can be formed with ␣and ␤-chains encoded on the same chromosome (i.e. encoded in cis) or on opposite chromosomes (i.e. encoded in trans). The occurrence of trans-encoded HLA class II molecules is well documented in the literature (4). However, evidence suggests that not every ␣and ␤-chain pairing will form a stable heterodimer (5,6). Hence, it is generally considered that alleles of DQ␣and DQ␤-chains pair up predominantly in cis rather than in trans (5,7). Nevertheless, studies on type 1 diabetes indicate that trans-encoded HLA molecules may play a role in pathogenesis (8). It has been observed that individuals who are heterozygous for DQ2.5 (DQA1*05: 01/DQB1*02:01) and DQ8 (DQA1*03:01/DQB1*03:02) are susceptible to type 1 diabetes with an almost 5-fold higher risk than those who are homozygous for either of the DQ variants (1,9,10). This phenomenon can be explained by the formation of trans-encoded molecules DQ8.5 (DQA1*05:01/DQB1*03:02) and DQ2.3 (DQA1*03:01/DQB1*02:01), which could present one or a few specific diabetogenic epitopes to CD4 ϩ T-cells, possibly inducing an immune response that leads to destruc-tion of insulin-producing pancreatic ␤-islet cells (8). A strong argument for involvement of the DQ2.3 heterodimer in type 1 diabetes comes from trans-racial gene mapping studies that have found that this heterodimer, which is typically found in the trans-configuration among Caucasian subjects, exists and is overrepresented in the cis-configuration among type 1 diabetes patients of African origin (11,12). The increased diabetes risk of the African DQ2.3 (DQA1*03:01/DQB1*02) carrying DR7 haplotype is contrasted by a protecting effect of the DQ2.2 (DQA1*03:01/DQB1*02) carrying DR7 haplotype of European origin, speaking to the functional importance of ␣-chain in the DQ2.3 molecule (12).
Celiac disease patients mount T-cell responses to gluten (consisting of the ␣-, ␤-, and ␥-gliadins as well as glutenin components) in the context of the celiac disease-associated DQ2.5 and DQ8 molecules (13). This human disease with in vivo expansion of T-cell clones specific for naturally selected epitopes offers a unique system to study the structure-function relationship of peptide-MHC complexes. Here, we have used this model system, offering a natural T-cell epitope restricted by HLA-DQ2.3, to study the x-ray crystal structure and function of a trans-encoded HLA-DQ molecule.
Production of DQ2.3 Heterodimer-Constructs for production of the HLA ␣-chain (DQA1*03:01 made from cDNA of the DQA1*03:01/DQB1*03:02 homozygous lymphoblastoid B-cell IHW9092) and ␤-chain (DQB1*02:01 modified from pAcAB3-DQB1*02:01) (20) were introduced into the expression plasmid pRmHa3. The sequence coding for the transmembrane region of each chain was excluded. A nucleotide segment was introduced so that the peptide RDSGPQPEQPEQPFPQPQ (underlined sequence representing residues 68 -81 of a ␥-gliadin; National Center for Biotechnology Information (NCBI) accession number AAK84776) with Gln-to-Glu substitutions at two positions (P1 and P4) followed by a 15-residue thrombin-cleavable linker GAGSLVPR2GSGGGGS (arrow indicates cleavage site) was linked to the N-terminal part of the DQ␤-chain (21). The C-terminal part of the ␣-chain contained a Cys followed by a human rhinovirus 3C protease cleavage site, an acidic leucine zipper, and a His 6 tag. The C-terminal part of the ␤-chain also contained a Cys residue, an HRV 3C protease cleavage site, a basic leucine zipper, a FLAG tag, and a BirA biotinylation site. An interchain disulfide bond was formed by the C-terminally introduced Cys residues and the leucine zipper pair promoted MHC dimer formation, whereas the His 6 and FLAG tags facilitated protein purification. Drosophila Schneider 2 (S2; Invitrogen) cells were co-transfected with a pair of DNA plasmids encoding the ␣and ␤-chain separately as well as with a blasticidin resistance gene (pCoBlast, Invitrogen) that ensured survival of the transfected cells. Expression of DQ2.3 was induced with CuSO 4 . The recombinant protein was isolated from culture supernatant by anti-FLAG tag affinity chromatography (anti-FLAG M2 affinity gel, Sigma) followed by gel filtration chromatography (Superdex 200, GE Healthcare). Gel filtration revealed three peaks. The first peak was a high molecular weight material, i.e. aggregated DQ2.3 molecules. MS analysis showed that the second peak contained the DQ2.3 molecule and a protein identified as Drosophila super coiling factor (Fly-Base database; FlyBaseID: FBgn0025682) (22), likely competing with the linked gluten peptide for binding to the peptide groove. The third peak contained the DQ2.3 molecule and was further purified before crystallization.
Functional Testing of Water-soluble DQ2.3 Molecules-Purified molecules were site-specifically biotinylated using the biotin-protein ligase BirA (Avidity). Biotinylated molecules were bound to streptavidin-conjugated plates (Roche Applied Science) overnight. T-cell clones were added to the plate, and recognition of HLA molecules was measured as [ 3 H]thymidine incorporation in a proliferative assay. Biotinylated DQ2.3 molecules were conjugated to multimers with streptavidin-phycoerythrin (Invitrogen). Staining of DQ2.3 reactive T-cell clones was analyzed on a FACSCalibur flow cytometer (BD Biosciences).
X-ray Crystallography-For crystallization, leucine zipper portion of the DQ2.3 heterodimer was removed by human rhinovirus 3C protease digestion for 16 h at 4°C. The resulting mixture was purified by anion-exchange chromatography using a Resource Q column (GE Healthcare) followed by size exclusion chromatography using a Shodex KW-803 column (Showa Denko K.K.). Crystals were grown using the hanging drop vapor diffusion method at 18°C. Protein concentration was 4 mg/ml, and the crystallization solution was 0.2 M Li 2 SO 4 , 0.1 M Tris, pH 8.5, 30% (w/v) PEG4000, 8% glycerol. Crystals grew to full size in 1 week. Crystals were dehydrated by the addition of 5 l 0.2 M Li 2 SO 4 , 0.1 M Tris, pH 8.5, 40% (w/v) PEG4000, 8% glycerol to the hanging drop and 1 ml of the same buffer to the well and incubating at 18°C for 12 h. The DQ2.3 crystal belonged to the space group C2 with cell dimensions a ϭ 74.9 Å, b ϭ 114.9 Å, c ϭ 138.0 Å, and ␤ ϭ 103.4°. The initial diffraction data set was collected at beam line BL13B1 of the Taiwan National Synchrotron Radiation Research Center, and the final data set was collected at beam line 9-3 of the Stanford Synchrotron Radiation Laboratory. X-ray diffraction data were indexed and integrated using HKL2000 (23). Structure was solved by molecular replacement using Phaser (CCP4) (24). The ␣-chain of DQ8 (1JK8) and ␤-chain of DQ2.5 (1S9V) were used as search models. Structure was refined by Refmac (CCP4) (25), Coot (26), Buster (27), and PHENIX (28). The final R work and R free values are 0.210 and 0.283, respectively. The quality of the final model was verified by PROCHECK (29). Crystallographic parameters, data collection, and refinement statistics are given in supplemental Table 1.

T-cell Recognition of DQ2.5-Glia-␥-4c/DQ8-Glia-␥-1a Gluten Epitope Presented via cis-or trans-Encoded Heterodimers-
We describe three T-cell clones (TCC548.3.5.6, TCC548.3.5.3, and TCC548.1.8.5) derived from a DQ8/DQ2.5 heterozygous celiac disease patient that are specific for variants of the DQ2.5-glia-␥-4c epitope that harbor deamidated residues (PQPEQPEQPFPQPQ or PQTEQPEQPFPQPQ). These T-cell clones, in contrast to similar T-cell clones described earlier (14), have undergone maturation in thymus in the presence of the DQ2.3 molecule. Also of note, the epitope studied is recognized in the same register by DQ8-restricted T-cells (then named DQ8-glia-␥-1a). The preference for the positioning of the Glu residues introduced by deamidation is different for the DQ2.5-and DQ8-restricted T-cells, DQ2.5restricted T-cells being sensitive to deamidation at position P4 and DQ8-restricted T-cells being sensitive to deamidations at P1 (14). The three T-cell clones carried different TCRs, but they showed similar reactivity patterns when tested against variants of the DQ2.5-glia-␥-4c epitope in the context of different antigen-presenting cells (APCs). Importantly, we found that a peptide with deamidations at both P1 The T-cell clone TCC548.3.5.6 was derived from a DQ8/DQ2.5 heterozygote celiac disease patient and tested against peptides with Gln to Glu substitutions (indicated by single-letter codes) at position P1 and/or P1 and P4 with four different APCs: DQ2.5 (CD114), DQ2.3 (IHW9102), DQ8.5 bare lymphocyte syndrome (BLS WK), and DQ8 (IHW9092). An increased response is seen when the peptide with Gln-to-Glu substitutions at both P1 and P4 is presented in the context of trans-encoded DQ2.3 to the T-cell clone from the heterozygote DQ8/DQ2.5 patient. and P4 was recognized much better when presented by the trans-encoded DQ2.3 molecule than the cis-encoded DQ2.5 molecule (Fig. 1). The same three T-cell clones were tested for their ability to recognize peptides with Gln-to-Glu substitutions at P1 and/or P4 in the context of various cis-or trans-encoded HLA molecules, and they gave similar response patterns. Substitution at P1 had an effect in the context of DQ2.3 and DQ8, whereas substitution at P4 had an effect in the context of both DQ2.3 and DQ2.5 (Fig. 1).
Peptide Binding to cis-or trans-Encoded Heterodimers-Peptide binding assays showed that the peptide variant with Glu at both P1 and P4 bound stronger to the DQ2.3 molecule (IC 50 ϭ 7.50 M) when compared with peptide variants with Gln-to-Glu substitutions at either P1 or at P4 alone (both IC 50 Ն 200  M) (Fig. 2A). This correlated with the T-cell response pattern described above, suggesting that the improved T-cell recognition (Fig. 1) could be explained by improved HLA binding. Similar peptide binding analysis for DQ2.5 showed that Gln-to-Glu substitution at P4 exerted the biggest effect as the variants with Gln-to-Glu substitutions at both P1 and P4 or only at P4 bound better than the variants with no substitution or substitution only at P1 (Fig. 2B). Thus, enhanced HLA binding could also explain the improved T-cell recognition of the P4 Glu-substituted peptide in the context of DQ2.5 (Fig. 1).
Determining T-cell Recognition by Ala and Lys Scans-Single position Ala-or Lys-substituted variants of PQPEQPEQPF-PQPQ were next tested in proliferative T-cell assays with the T-cell clones TCC387.19 and TCC548.3.5.6 using APCs expressing DQ2.5 or DQ2.3 (supplemental Fig. 2). The most interesting observation in this experiment related to recognition of the peptide with Ala substitution at P1. Both TCC548.3.5.6 and TCC387.19 showed impaired recognition of the P1 Ala peptide when compared with the P1 Glu peptide in the context of DQ2.3, whereas in the context of DQ2.5, both T-cell clones recognized the P1 Ala peptide best (Fig. 3). As this effect was seen with both T-cell clones, it may suggest that the P1 pocket of DQ2.5 is better at accommodating Ala than the P1 pocket of DQ2.3.
Functional Testing of Water-soluble trans-Encoded DQ2.3 Molecules-We produced water-soluble DQ2.3 molecules in S2 cells displaying the PQPEQPEQPFPQPQ peptide. Purified monomeric DQ2.3 molecules were biotinylated and added to streptavidin-coated plastic wells to test whether the recombinant molecules were recognized by antigen-specific T-cell clones. Both TCC387.19 and TCC548.3.5.6 recognized the recombinant molecules with the tethered peptide (Fig. 4). Moreover the clones were specifically stained with multimerized HLA/peptide molecules obtained by coupling biotinylated DQ2.3 HLA molecules to phycoerythrin-labeled streptavidin. As evident in Fig. 4, both TCC387.19 and TCC548.3.5.6 were effectively stained, indicating that the water-soluble HLA molecules were indeed correctly folded and were properly recognized by the T-cells. Once the integrity of the transencoded DQ2.3 molecule was verified, we went on to determine the x-ray crystal structure of DQ2.3 to gain structural insight into trans-encoded heterodimer formation and epitope recognition.

Structure of trans-Encoded HLA-DQ Molecule
tics for data collection and structure refinement are given in supplemental Table 1, and a representative electron density map is shown in supplemental Fig. 3. Asn-␤33 is the only residue that is in the disallowed region of the Ramachandran plot. The loop structure at this particular position adopts an unusual type II ␤-turn (30), which places the and angles of the residue at the iϩ1 position of the turn in otherwise disallowed configurations. The overall structure of DQ2.3 is highly similar to previously reported DQ crystal structures: DQ2.5, DQ8, and DQ6.2.
Two unusual structural features are present in DQ2.3 that are not seen in other MHC class II molecules. The first unique structure is the "gate" formed by Arg-␣52 and Arg-␤88 (Fig.  5C), which is located at the P1 end of the peptide-binding groove. The guanidinium group of Arg-␣52 and Arg-␤88 is stacked in a face-to-face manner with a C to C separation distance of 3.6 Å. Although Arg-Arg pairing is not common in protein structures, a survey of the Protein Data Bank by Magalhaes et al. (32) found 41 analogous Arg-Arg interactions. This gate creates a shallow barrier at the P1 end of the peptidebinding groove. Related to this, the P-3 to P-1 segment of the peptide is protruding out at a roughly 90°angle relative to the rest of the peptide sitting in the MHC-binding groove. The second unique structure is the "roof" formed by Phe-␣58 and Arg-␤77, which lies over the P3 and P4 section of the bound peptide (Fig. 5D). Arg-␤77 side chain forms a hydrogen bond with the P2 Gln side chain and the P3 main chain carbonyl oxygen. The occurrence of hydrogen-bonding interaction between Arg-␤77 and the nonanchoring P2 side chain and P3 backbone is unusual. Extensive shielding of the bound peptide by MHC side chains is not seen in any of the previous class II MHC crystal structures. Incidentally, ␣58 and ␤77 are involved in TCR recognition, particularly with the CDR3 loops, as observed in different class II MHC-TCR complex structures determined to date (33).

DISCUSSION
Our results demonstrate that the DQ2.3 molecule combines the peptide binding signatures of the DQ2.5 and DQ8 molecules. This results in a binding motif with preference for negatively charged anchor residues at both the P1 and the P4 positions. In this way, some epitopes can be presented even more effectively in the context of the trans-encoded DQ2.3 molecule. This has relevance for understanding how the trans-encoded DQ2.3 molecule is predisposing to type 1 diabetes.
We found that the responses of the T-cell clones in the context of the DQ2.3 molecule were improved by Gln-to-Glu substitutions in the peptide at the P1 and P4 positions (Fig. 1). Results from peptide binding assays suggest that the enhanced T-cell recognition of the deamidated peptides, at least in part, is explained by improved HLA binding (34). There are multiple positively charged residues near the P1 pocket (Arg-␣52 and His-␣24) and the P4 pocket (Arg-␤77, Arg-␤70, Lys-␤71) that may establish long range electrostatic interaction with the P1 or P4 Glu of the gliadin peptide or establish hydrogen bonds with the peptide through water molecules, which in the current crystal structure are not visible due to insufficient resolution.
Differential recognition of the P1 Ala-substituted PQPEQPEQPFPQPQ peptide by a T-cell clone in the context of DQ2.5 versus DQ2.3 points to structural differences between the two HLA molecules around the P1 pocket (Fig. 3). Although the overall shape and size of the P1 pocket in DQ2.3 and DQ2.5 are highly similar, the polarity of the residues found at the surface of the respective P1 pockets is dissimilar (Fig. 6). In DQ2.5, Gln-␣31 and Glu-␤86 are found at the bottom of the P1 pocket and form a bidentate hydrogen bond to one another (donoracceptor distances are 2.8 and 3.0 Å). In contrast, Glu-␣31 and Glu-␤86 are found at the bottom of the P1 pocket in DQ2.3 (carboxylate oxygen to carboxylate oxygen distances are 2.9 and 3.4 Å). Furthermore, DQ2.3 has Arg-␣51 on the side of the P1 pocket, whereas DQ2.5 has Phe-␣51. Binding of the P1 Alasubstituted peptide to either DQ2.3 or DQ2.5 will result in a mostly vacant P1 pocket, which may lead to conformational shift of the above mentioned residues as well as introduction of solvent molecules inside the P1 pocket. Although the details of such rearrangement need to be assessed in a future study, we predict that the DQ2.5 will form a more stable complex with an aliphatic amino acid such as Ala in P1 in contrast to the DQ2.3 molecule that prefers a negatively charged amino acid in this position.
It is generally considered that alleles encoded in cis (i.e. on a single chromosome) have been evolutionarily selected to form a stable heterodimer, whereas trans-encoded heterodimers may not necessarily be able to form a stable or functional MHC (35). Regardless, a number of studies have shown that ␣␤ heterodimers can form among a subset of different MHC class II molecules, including pairing of mixed haplotypes (e.g. mouse MHC class II A␣ b A␤ k (35)), mixed isotypes (e.g. human MHC class II DR␣DQ␤ (36) or mouse MHC class II E␣ d A␤ d (37)), and mixed species (e.g. human and mouse MHC class II DR␣A␤ d (38)). Some structural explanations for the bias for cis-encoded heterodimer formation have been given for mouse MHC class II I-A molecules, where a subset of specific residues has been identified that can dictate which of the ␣and ␤-chain alleles can form a stable heterodimer (39). However, limited knowledge exists as to the rules that determine formation of MHC class II heterodimers.
In DQ molecules, polymorphic residues at the ␣␤ dimerization interface are concentrated at the two ends of the peptidebinding groove. On the P1 side, ␣44 -␣54 and ␤84 -␤90, two clusters in the ␣ and ␤ chains where polymorphism occurs most extensively, come together to form the ␣/␤ interface. On the P9 side, polymorphic residues ␣73/␣77 and ␤53 contact each other at the ␣/␤ interface. Sequence distribution at these polymorphic positions reveals that DQ alleles can be divided into two major groups. Group I consists of DQA1*02 through DQA1*06 and DQB1*02 through DQB1*04, whereas group II is composed of DQA1*01 and DQB1*05 and DQB1*06 (Fig. 7A). Interestingly, according to the dbMHC database (40), only 10 out of 4,233 reported HLA-DQ haplotypes carry a group I-group II mixed pair, indicating potential evolutionary selection pressure against the cross-group pairing. A survey of available DQ crystal structures revealed some key structural features on the ␣␤ dimerization interface that may be dictating the observed allele-pairing bias.
On the P1 side of the molecule in DQ2.5 and DQ8 (group I), absolutely conserved Tyr-9␣ and Glu-86␤ side chains form an H-bond to each other (Fig. 7B). In DQ6 (group II), however, ␤87 is found at the position where ␤86 occurs in DQ2.5 and in DQ8 (both group I) (Fig. 7C). Cys-9␣ is absolutely conserved, and ␤87 carries either a Tyr or a Phe residue, allowing the pair to have a good steric fit. However, mixing of group I and group II alleles will result either in the loss of an H-bond or the introduction of a steric clash. Furthermore, although the ␣44 -␣54 and ␤84 -␤90 polymorphic clusters both assume a helical structure, their local conformations are different in groups I and II. In DQ6 (group II), the backbone conformation near Phe-51␣ draws the bulky side chain close to the ␤-chain at Ala-86␤ and Gly-89␤. In group I, ␤86 and ␤89 are conserved for Glu and Thr, respectively. Simple overlay of DQ6 and DQ8 structures shows steric incompatibility between Phe-51␣ and Thr-89␤ side chains. Similarly, in DQ8 (group I), the difference in the helical structure shifts the ␤-chain-contacting residue from Phe-51␣ to Arg-52␣, where the Arg-52␣ side chain is buried deep inside the ␣␤ interface and forms an H-bond with the previously mentioned Glu-86␤. Again, overlay of DQ6 and DQ8 structures indicates that the Arg-52␣ will clash against Ala-86␤. At the other end of the peptide-binding groove, polymorphic ␣77 and ␤53 residues contact each other. In group I, ␣77 and ␤53 are conserved for Ser and Leu, respectively (Fig.   7D). In group II, ␣77 and ␤53 are conserved for Tyr and Gln (Fig. 7E). Structural overlay of DQ2.5, DQ8, and DQ6 shows that pairing of the group II ␣-chain and the group I ␤-chain will likely suffer from steric crash between the Tyr-77␣ and Leu-53␤ side chains. Taken together, mixed chain pairing (group I ␣-chain paired with group II ␤-chain or group II ␣-chain paired with group I ␤-chain) will lead to loss of interchain hydrogenbond interaction or result in suboptimal packing at the heterodimer interface.
Our crystal structure shows that the P1 pocket in DQ2.3 is significantly different from that of DQ2.5 due to the polymorphic MHC residues found in this region. Additionally, we have demonstrated that DQ2.3 presents a gluten epitope to T-cells much more efficiently than DQ2.5. Therefore, trans-encoded MHCs have the potential to drastically alter an individual's immune response toward a particular antigen. Sequence analysis of the HLA-DQ gene products has revealed that certain haplotype pairings are more likely to produce a structurally stable trans-encoded ␣␤ heterodimer than others, which is also reflected in the high frequency of compatible pairings seen in the NCBI dbMHC database.