Hanne Quarsten 1, Gunnar Paulsen 1,5, Bente H. Johansen 1, Christopher J. Thorpe 2, Arne Holm 3, Søren Buus 4 and Ludvig M. Sollid 1.

Transcription

1 International Immunology, Vol. 10, No. 8, pp Oxford University Press The P9 pocket of HLA-DQ2 (non-aspβ57) has no particular preference for negatively charged anchor residues found in other type 1 diabetes-predisposing non-aspβ57 MHC class II molecules Hanne Quarsten 1, Gunnar Paulsen 1,5, Bente H. Johansen 1, Christopher J. Thorpe 2, Arne Holm 3, Søren Buus 4 and Ludvig M. Sollid 1 1 Institute of Transplantation Immunology, Rikshospitalet, University of Oslo, 0027 Oslo, Norway 2 NoHo Digital, Canberra House, Regents Street, London W1R 7YB, UK 3 Chemistry Department, Royal Veterinary and Agricultural University, 1871 Frederiksberg C, Denmark 4 Institute of Medical Microbiology and Immunology, University of Copenhagen, 2200 Copenhagen, Denmark 5 Dr Gunnar Paulsen tragically passed away before work on this manuscript could be completed Keywords: P9 anchor, peptide binding Abstract Susceptibility and resistance to type 1 diabetes are associated with MHC class II alleles that carry non-asp and Asp at residue 57 of their β chain respectively. The effect of Asp or non-aspβ57 may relate to a differential ability of distinct class II molecules to bind specific immuno-pathogenic peptides. Recent studies in man and mouse have revealed that some type 1 diabetes-predisposing non-aspβ57 class II molecules (i.e. DQ8, DR4Dw15 and I-A g7 ) preferentially bind peptides with a negatively charged anchor residue at P9. It has been suggested that this is a common feature of type 1 diabetes-predisposing class II molecules. The molecular explanation for such a phenomenon could be that class II β chains with Aspβ57 form a salt bridge between Aspβ57 and a conserved Arg of the α chain, whereas in non-aspβ57 molecules the Arg is unopposed and free to interact with negatively charged P9 peptide anchor residues. We have investigated the specificity of the P9 pocket of the type 1 diabetes-associated DQ2 molecule and in particular examined for charge effects at this anchor position. Different approaches were undertaken. We analyzed binding of a high-affinity binding ligand and P9-substituted variants of this peptide, and we analyzed the binding of a set of synthetic random peptide libraries. The binding analyses were performed with wild-type DQ2 and a mutated DQ2 with Ala at β57 substituted with Asp. Our results indicate that the wild-type DQ2 (non-aspβ57) prefers large hydrophobic residues at P9 and that there is no particular preference for binding peptides with negatively charged residues at this position. The specificity of the P9 pocket in the mutated DQ molecule is altered, indicating that the β57 residue contributes to determining the specificity of the P9 pocket. Our data do not lend support to the hypothesis that all non-asp β57 class II molecules predispose to development of disease by binding peptides with negatively charged P9 anchor residues. Correspondence to: H. Quarsten Transmitting editor: J.-F. Bach Received 5 February 1998, accepted 8 May 1998

2 1230 No preference for negatively charged P9 anchor residues in HLA-DQ2 Introduction MHC class II molecules are highly polymorphic membrane glycoproteins that display peptides on the surface of antigenpresenting cells for inspection by CD4 T cells. The peptides bind to class II molecules in a cleft jointly formed by membrane distal domains of the MHC α and β chains. The binding is partly mediated by hydrogen bonding between conserved residues of MHC and the peptide backbone, and partly by interactions between side chains of so-called anchor amino acids of the peptide and polymorphic binding pockets in the MHC class II cleft. Generally, class II molecules appear to have five pockets designated P1, P4, P6, P7 and P9 (1 3). Polymorphic MHC residues mainly form the binding pockets. Thus the size and chemical nature of the pockets and hence the overall peptide binding specificity vary between MHC allotypes. Several diseases are over-represented in subjects expressing particular MHC molecules. Many of these diseases are autoimmune in nature. Susceptibility to type 1 diabetes in man and mouse is strongly associated with given DQ (DQ8 and DQ2) and I-A (I-A g7 ) molecules respectively that carry a non-asp amino acid (i.e. Val, Ser or Ala) at position β57 (4). In contrast, DQ and I-A molecules that carry Asp at position β57 confer protection to type 1 diabetes. Certain DR and I-E molecules also influence susceptibility to type 1 diabetes (5). The DR molecule that confers the highest susceptibility, DR4Dw15 (DRB1*0405), carries Ser (i.e. non-asp) at position β57, whereas the I-E and DR molecules that confer protection against type I diabetes carry Asp at β57 (6). Susceptibility and resistance to type 1 diabetes therefore can to a certain extent be attributed to presence of non-asp and Asp residues at β57. Other polymorphic residues of the α and the β chains of class II molecules also seem to influence the degree of susceptibility/resistance, but the residue at β57 seems to be particularly important (5). In the X-ray crystal structures of DR1 and DR3, the Aspβ57 residue forms a salt bridge with the positively charged Arg at DRα76. In addition, the X-ray crystal data have furthermore demonstrated that the β57 residue contributes to the shape of the P9 pocket (1,2,7). Characterization of the peptide binding motifs of I-A g7, DQ8 and DR4Dw15 have indicated that all these molecules prefer negatively charged P9 anchor residues (i.e. Glu or Asp) (8 13). The molecular explanation for this phenomenon could be that a conserved Arg of the α chain (I-Aα76, DQα79 or DRα76) in the absence of an Aspβ57 is free to interact with a negatively charged residue at the C-terminal end of the peptide. The presence of a negatively charged P9 anchor has been suggested to be a common denominator of peptides binding to the diabetes predisposing class II molecules (14,15). These peptide MHC complexes are thought to initiate T cell immune responses that cause type 1 diabetes. Recently, however, a distinct peptide binding motif for I-A g7 which shows no particular preference for binding peptides with negatively charged P9 anchors has been identified (16). The reason for the discrepancy between the two reported I-A g7 motifs is at present unclear, but may relate to the methods used for the motif description. The motifs described for I-A g7 were both determined by truncation and substitution analysis of highaffinity binding peptides. Possibly, the motifs determined by this method can be constrained by interdependence of effects between peptide residues (i.e. flavored by the sequence of the peptides tested) (17,18). It has been suggested that DQ2 as DQ8 prefers binding of peptides with negatively charged P9 anchors (19). In a previous study we obtained results which indicated that the DQ2 molecule (hereafter termed WT-DQ2) carrying non-asp at β57 prefers bulky hydrophobic residues as the P9 anchor (20). Our results were based on substitution analysis of highaffinity binding peptides having the P9 anchor at the C- terminal end of the peptide and the observed preference for the P9 pocket could possibly have been influenced by the interference of the carboxyl group. To rule out that the C-terminal carboxyl group could be a substitute for a negatively charged P9 anchor residue, we have analyzed a highaffinity binding peptide and variants of this peptide with C- terminal extension beyond P9 for binding. Moreover, in attempt to perform an analysis which is unbiased by the sequence of the binding peptide, we have assayed for binding a set of synthetic random peptide libraries with each of the natural occurring amino acids represented at the P9 anchor position. We have also evaluated the impact of Asp/non-Asp polymorphism on the specificity of the P9 pocket by testing binding of peptides to a mutant DQ2 molecule where Ala at β57 has been changed to Asp (termed Mut57-DQ2) by site-directed mutagenesis and gene transfection. Computer models of WT-DQ2 and Mut57-DQ2 were also generated in order to provide additional information. Our results corroborate the previous findings that WT-DQ2 prefers large hydrophobic residues as P9 anchors (20,21). This is at variance with the hypothesis that all non-aspβ57 class II molecules predispose to type 1 diabetes due to a common preference for binding peptides with negatively charged P9 anchors (14,15). Methods Mutagenesis and transfection Production of transfectants expressing DQA1*0501 and DQB1*0202 was accomplished as described previously (22). Briefly, cdna of the DQB1*0202 gene (kind gift of J. Lee) was ligated into a RF-M13 vector. A mutagenic primer was used to change the codon specifying Ala at position β57 to the codon specifying Asp. The wild-type or altered DQB1*0202 cdna were ligated into the retroviral vector plncl and transfected into PA 317 cells. The DQB-vector constructs were then transferred to the B-lymphoblastoid cell line (B-LCL) SWEIG007 (homozygous for DQA1*0501 and DQB1*0301) by co-cultivation with virus-producing PA 317 cells. B-LCL transfectants were grown in select medium containing Geneticin (Gibco, Paisley, UK). Finally, introduction of DQB genes was confirmed by DNA sequencing and the DQβ*02 surface expression was verified by flow cytometry analysis. Preparation of HLA-DQ molecules The DQ molecules were purified as described earlier (23). Briefly, cells were lysed for 30 min on ice at a concentra-

3 No preference for negatively charged P9 anchor residues in HLA-DQ tion of 10 8 cells/ml in PBS containing 1% NP-40, 5 mm sodium orthovanadate, 25 mm iodoacetamide and 1 mm PMSF (24). The lysates were cleared for debris by centrifugation at 4 C. Immediately, or after storage at 70 C, the DQ molecules were affinity purified (23,25) using the mab 2.12.E11 (26) coupled to Protein A Sepharose CL-4B (27). The DQ molecules were eluted at ph 11.5 with 50 mm diethylamine containing 0.15 M NaCl and 1% β-octylglucoside, and the eluates were immediately neutralized using 2 M Tris, ph 6.3. Finally, the eluates were concentrated by vacuum dialysis. The protein content in the DQ preparations was determined by BCA protein assay (Pierce, Oud-Beijerland, The Netherlands) and the purity was confirmed by SDS PAGE under reducing conditions. Peptides The HLA class I α peptide (WIEQEGPEYW) and several analogs with substitutions at the C-terminal position (P9) were purchased from Neosystem (Neosystem Laboratoire, Strasbourg, France). The amino acids selected to replace the C-terminal position of the HLA class I peptide (i.e. Phe, Leu, Val, Ala, Asn, Gln, Ser, Glu, Asp and Lys) were chosen to represent the variety of size, chemistry and charge of natural amino acids. The HLA class I peptide extended with two Ala residues on both sides of the nonamer DQ2 binding region (AAIEQEGPEYWAA) and three variants of this peptide substituted at P9 were purchased from Research Genetics (Huntsville, AL). All peptides were 80% pure. The peptides were aliquoted and lyophilized before storage at 20 C (or 70 C). Peptide libraries Synthetic random peptide libraries (XXFXXXKXXXΩ) were synthesized with one of the natural amino acids at the C-terminal position (Ω) at a time. Mixtures of 10 amino acids (Ala, Glu, Phe, Gly, Ile, Lys, Leu, Met, Gln and Ser), considered representative for the natural occurring amino acids, were used in the random positions (X) (28). All sublibraries were generated with a fixed Phe at position 3 (P1) and a fixed Lys at position 7 (P5) to avoid binding in alternative frames. In addition, one sublibrary with random residues at the Ω position was also synthesized. All peptide libraries were quantitated by BCA protein assay before usage in the binding assay. Peptide binding assay DQ molecules were incubated with radiolabeled indicator peptide together with different unlabeled peptides in a reaction mixture containing PBS with 0.05% NP-40 and protease inhibitors as described previously (23). A Mycobacterium bovis (MB) 65 kda heat shock protein-derived peptide (KPLLI- IAEDVEGEY; MB 65 kda Y) was used as an indicator peptide for WT-DQ2 (23). Binding of several radiolabeled peptides to Mut57-DQ2 were tested and an HLA-DQβ1*0301 peptide eluted from DQ7 (DVEVYRAVTPLGPPD; HLA- DQβ1* ) (29) was selected to be an indicator peptide for this molecule. After incubation for h at 37 C (ph 5.4), peptide DQ complexes and peptides were separated on Sephadex G50 superfine spin columns. The binding capacity of each unlabeled peptide was examined by measuring the ability to inhibit binding of the radiolabeled indicator peptide. The unlabeled peptides were first tested in concentrations from 16 nm to 167 µm in 10-fold steps and IC 50 values (peptide concentration giving 50% inhibition) were calculated. To obtain a more accurate determination of IC 50, two (or more) independent experiments with 4-fold dilutions around the IC 50 value were performed. The effect of P9 substitutions on the HLA class I α peptide was expressed as the IC 50 of the HLA class I peptide divided by the IC 50 of the analog and this was defined as relative binding capacity (RBC) for the HLA class I peptide variants. Similarly, RBC values for substitution variants of the 13mer HLA class I peptide were calculated with the peptide containing the native DQ2 binding region as the reference peptide. In order to compare the results of the sublibraries with the results obtained by the HLA class I peptide variants, RBC values for peptide libraries were calculated as the ratio of the IC 50 of the sublibrary with Trp at P9 to the IC 50 of each sublibrary. Binding studies were performed with DQ molecules affinity purified from transfected cell lines (see Mutagenesis and transfection), except for the binding analysis of peptide libraries and the set of 13mer HLA class I peptides to WT- DQ2. The latter analyses were performed with DQ molecules isolated from a DQA1*0501/DQB1*0201 homozygous B-LCL (STEINLIN). The β chain (DQβ1*0201) of the WT-DQ2 molecules isolated from this cell line differs from the β chain (DQβ1*0202) of the WT-DQ2 molecules obtained from the transfected cells at residue 135. It is unlikely that this residue participates in the formation of the peptide binding groove. The WT-DQ2 molecules having either DQβ1*0201 or DQβ1*0202 exhibited similar properties when binding of several peptides was tested (data not shown). Modeling of the WT-DQ2 and Mut57-DQ2 molecules All modeling and calculation were performed in the QUANTA/ CHARMm modeling package (Molecular Simulations/Biosym, San Diego, CA) using a Silicon Graphics R4600PC workstation. The DR1 crystallographic structure (1,7) was used as a template for modeling and the sequences of the DQ α and β chains were taken from the World Health Organization collection of HLA class II sequences disseminated by the Histo World Wide Web server ( The models were built using conventional homology modeling with the exception of the β bulge region of the HLA-DQ α chain and the deletion segment in the α 1 domain α helix. These regions were built using a template forcing technique (unpublished results) with the resulting geometries for these segments being optimized locally before the global optimization of the models. The local optimization was performed using a refinement regime consisting of 50 cycles of Steepest Descent optimization and a subsequent 100 cycles of Adopted-Basis Newton Raphson optimization. Three residues on either side of the edge point of the manipulation were included actively in the calculation. All other residues within the protein were included passively. Subsequent to completion of these refinements the α 1 β 1 heterodimers were refined. Initially, only the side chains of the models were permitted to move in the refinement, with the protein backbone included passively in the calculations. The refinement procedure consisted of 50 cycles of Steepest Descent optimization followed by 1000 cycles of Adopted Basis Newton Raphson optimization. Subsequent to these steps all of the constraints were removed

4 1232 No preference for negatively charged P9 anchor residues in HLA-DQ2 and the entire model was subjected to 50 cycles of Steepest Descent optimization and 100 cycles of Adopted Basis Newton Raphson optimization. In all of the calculations the CHARMm forcefield was used, with polar hydrogen representation and charges defined from the internal dictionaries. The dielectric model used consisted of a distance dependent dielectric function with a ramped cut-off of 9 15 Å and a dielectric constant (φ) of 4. This dielectric model has previously been demonstrated to produce excellent models for MHC molecules in the absence of explicit solvent molecules (30). Modeling of DQ:peptide interaction The peptide model derived through homology modeling from the crystallographic model of the influenza haemaglutinin peptide bound to HLA-DR1 (1). The completed models for DQ molecules were then docked, using superposition, with the peptide model. The DQ:peptide models were optimized using identical force fields and dielectric models, within the QUANTA/CHARMm package as are described above. A different set of constraints for the optimization was, however, used. Initially, only the atoms of the bound peptide moiety were permitted to move, and the model was optimized using 50 cycles of Steepest Descent optimization and 1000 cycles of Newton Raphson optimization. Subsequent to the completion of this step, the restraints were removed from the entire complex and 100 cycles of Newton Raphson optimization were performed. Results Binding to WT-DQ2 First we tested the HLA class I α peptide and analogs substituted at P9 for binding to WT-DQ2 (Fig. 1A). The analogs with hydrophobic residues at P9 (i.e. Trp, Phe, Leu, Val or Ala) bound better to WT-DQ2 than the other tested analogs. Analogs with bulky hydrophobic P9 residues were especially good binders, while the analog with positively charged Lys at P9 was a particularly poor binder. The HLA class I peptide variants with negatively charged P9 residues (i.e. Glu or Asp) were intermediate binders to WT-DQ2. Between the best and poorest binding HLA class I peptide variants an almost 30- fold increase in binding capacity was observed. We investigated the binding of the HLA class I peptide extended with two Ala residues on each side of the nonamer DQ2 binding region and three P9-substituted peptide variants of this peptide (i.e. Gln, Glu or Asp). Even among these peptides, the variant with Trp at P9 was the best binder (Fig. 2). This excludes the possibility that P9 ending peptides due to a possible substitutive role of the negatively charged carboxylic group, erroneously classify bulky hydrophobic and not negatively charged residues as the optimal P9 anchor. To undertake a different and more comprehensive strategy to determine the nature of favorable P9 anchors of WT-DQ2, synthetic random peptide libraries with one of the natural amino acids represented at the C-terminus at a time were tested for binding to WT-DQ2 (Fig. 1B). Large hydrophobic P9 anchors (i.e. Trp, Phe, Ile, Leu, Val, Met and Tyr) were identified as supportive for binding of peptide libraries to WT-DQ2. On the other hand, positively charged side chains Fig. 1. Binding to WT-DQ2 of (A) the HLA class I α peptide and several analogs substituted at P9, and (B) a set of synthetic random peptide libraries. The RBC of each peptide tested for binding to WT-DQ2 was calculated as the ratio of the IC 50 value of a reference peptide to the IC 50 value of the tested peptide. The HLA class I α peptide and the sublibrary with Trp (W) as the P9 anchor residue were chosen as reference peptides. The IC 50 values (µm) are given in parentheses. Results are given as mean of two and five experiments for the HLA class I peptide and the peptide libraries respectively. Fig. 2. Binding to WT-DQ2 of the HLA class I peptide extended on both ends of the nonamer DQ2 binding region and three P9 substituted variants of this peptide. The relative binding capacity for each peptide is calculated as described for Fig. 1. The IC 50 values (µm) are given in parentheses. Results are given as mean of two experiments. were disfavorable and poorly accepted at P9. Again, we observed that libraries with negatively charged P9 residues had intermediate binding capacity. There was a 23-fold increase in binding capacity of the sublibrary that bound with highest affinity to the sublibrary binding with the lowest affinity to WT-DQ2. Binding to Mut57-DQ2 The binding of the HLA class I α peptide and its analogs and peptide libraries to Mut57-DQ2 was also

5 No preference for negatively charged P9 anchor residues in HLA-DQ Fig. 3. Binding to Mut57-DQ2 of (A) the HLA class I α peptide and several analogs substituted at P9 and (B) a set of synthetic random peptide libraries. The relative binding capacity for each peptide is calculated as described for Fig. 1. The IC 50 values (µm) are given in parentheses. Results are given as mean of two and three experiments for the HLA class I peptide and the peptide libraries respectively. analyzed (Fig. 3). Binding studies with the HLA class I peptide variants indicated that the Mut57-DQ2 molecule displayed a preference for binding of analogs with relatively small side chains at P9 (i.e. Ser, Ala, Asn, Asp or Val), while all peptide variants with large and medium-sized residues (i.e. Trp, Phe, Leu, Gln, Glu or Lys) at the C-terminus were poorer binders. The peptides with small residues at P9 apparently bound well to Mut57-DQ2 independent of the chemical nature of the side chains. There was only a 10-fold increase in the binding capacity between the best and the poorest binders for Mut57- DQ2, which is in contrast to a nearly 30-fold increase observed for the WT-DQ2 molecule. A change in P9 anchor requirement was also demonstrated by the binding analysis of synthetic random peptide libraries to Mut57-DQ2. Although, the binding pattern was not completely identical to that obtained by the HLA class I peptide and analogs, it largely indicated comparable binding features. Bulky amino acids were found to be disfavorable as P9 anchor residues, while the sublibraries carrying residues with small side chains at P9 exhibited the best binding. Only a 7-fold increase in binding capacity was observed between the poor and the good binding peptide sublibraries. Computer modeling The models of WT-DQ2 and Mut57-DQ2 clearly demonstrate the effects of substitution of position β57 on the size and shape of the P9 pocket. The majority of cleft architecture is near identical for the two molecules, with the changes in structure being exclusively clustered around β57 of the WT- Fig. 4. Graphical representations of the modeled molecular surface of the models for WT-DQ2 and Mut57-DQ2 demonstrate the differences in surface topography of the two molecules. The WT-DQ2 molecule has a distinct and deep P9 pocket (A) in which large hydrophobic and primarily aromatic residues may be comfortably bound. Indeed the electrostatic nature of this pocket appears to be predominantly hydrophobic and neutral, with a lesser basicity than the P4 and P7 pockets of the molecule which are known primarily to bind acidic residues. The P9 pocket of the Mut57-DQ2 molecule (B) is significantly shallower, and is more suited to the sequestration of small hydrophobic and polar residues. The HLA class I α peptide and a variant of this peptide containing an Ala at P9 are modeled into the peptide binding cleft of the WT-DQ2 molecule and the Mut57-DQ2 molecule respectively. DQ2 molecule (Fig. 4). The WT-DQ2 model predicts that the molecule has a broad and deep invagination, largely hydrophobic in its nature with the free basic moiety of the Arg α79 residue facing out of the pocket making interactions with carbonyl groups of the peptide backbone. This observation provides a potential explanation for the low representation of acidic ligands accommodated in the P9 pocket. Indeed the P4 and P7 pockets of the WT-DQ2 molecule, which are well known to favor the binding of predominantly acidic ligands (20,21,31), are significantly more basic in nature than the P9 pocket, adding credence to the observations that hydrophobic residues and in particular large branched chain or aromatic residues are the preferred partners for the environment of this pocket. In contrast, the P9 pocket displayed in the Mut57-DQ2 molecule is shallow yet still largely hydrophobic, being more of a recessed plateau than a deep pocket for

6 1234 No preference for negatively charged P9 anchor residues in HLA-DQ2 sequestration. However, small polar residues will also be well accommodated in the environment due to the presence of several polar residues in the region, including the Argα79/ Aspβ57 salt-bridged pair which form the shelf upon which the P9 residue of the peptides bound by this molecule rests. Analyses of potential ligands, using modeled HLA class I α peptide variants containing different C-terminal amino acids, for the pockets is in complete agreement with the biological data, with the P9 pocket of the WT-DQ2 molecule displaying an environment which is most suited to large, bulky aromatic residues (Phe, Trp and Tyr) and branched chain aliphatic residues (Ile, Leu and Val). The potential ligands of the Mut57-DQ2 molecule, however, are predicted from the analysis of the model to have strikingly smaller P9 anchors, with valine being the largest residue predicted to fit in the pocket without significant rearrangement within the pocket or alteration in the geometry of the peptide backbone. Discussion Evidence suggests that the Asp/non-Asp polymorphism of β57 in MHC class II molecules is central in determining susceptibility and resistance to type 1 diabetes. This polymorphism seems to influence the binding specificity of the P9 pocket. DQ2 is over-represented among patients with type 1 diabetes (4,32) and this molecule carries an Ala (i.e. non- Asp) at β57. In order to obtain more information on the role of β57 in determining of the binding specificity of the P9 pocket in DQ2, we have examined peptide binding to wildtype DQ2 and to a point-mutated DQ2 molecule where β57 has been changed from Ala to Asp. Interestingly, the mutant molecule has recently been demonstrated to be a naturally occurring DQ2 variant (i.e. DQA1*0501/DQB1*0203) (33). An HLA class I α peptide was previously reported to be a natural ligand for DQ2 (21,31). Several analogs of this peptide substituted at P9, as well as binding of a longer variant and three P9-substituted analogs, were tested for binding to WT-DQ2 and/or Mut57-DQ2. In addition, we examined the specificity of peptide binding by testing a set of synthetic random peptide libraries each with one of the natural amino acids represented at P9 to both DQ molecules. Our results demonstrate that the Mut57-DQ2 has a P9 anchor requirement which is different from WT-DQ2, suggesting an important role of β57 in determining peptide binding specificity. However, our results do not support the notion that all non- Asp β57 class II molecules have a preference for negatively charged P9 anchor residues, a specificity which has been implicated in the pathogenesis of type 1 diabetes (14,15). Two different approaches have until now been in use for the definition of MHC specificities. One method is based on characterization of peptides that can be eluted from MHC molecules (34 36). Motifs based on natural eluted ligands will not only reflect the binding specificities, but will also mirror processing effects and involvement of molecular chaperones. The other approach is based on binding of synthetic peptides to MHC molecules (24,37 40). Motif determination by binding analysis of analogs of good binders may have the unfortunate bias that there is an unknown component of correlated effect between the individual residues of the MHC binding peptide or, in other words, the actual binding is peptide sequence dependent (17,18). Possibly this phenomenon may explain the divergence between the motifs reported for I-A g7 (8,9,16). We reasoned that the bias related to the sequence of a given test peptide can be minimized by using positional scanning combinatorial libraries (PSCPL) (41). To take advantage of the PSCPL method, we designed a set of random peptide sublibraries carrying each natural amino acid at P9, and tested the libraries for binding to WT-DQ2 and Mut57-DQ2. To prevent the peptide libraries to bind in alternative frames, they were generated with a fixed Phe and Lys in positions 3 (P1) and 7 (P5) respectively. The bulky Phe residue is favored for accommodation in the P1 pocket while the positively charged Lys was selected as a residue unlikely to be accommodated into the P4 and P6 pockets (20,21,31). The binding of all variants of the HLA class I peptide and the libraries both suggested that WT-DQ2 preferred large hydrophobic residues in the P9 position. This is in agreement with observations obtained with an OVA peptide (20) and the HLA class I peptide (21). We observed that a positively charged residue at P9 was particular deleterious for binding. This may be due to repulsion between the Argα79 residue and positively charged P9 residues of the peptide. Regardless of the sequence of the binding peptide and the positioning of the C-terminal anchor in the peptide we found that negatively charged residues were not particularly preferred as P9 anchors. The unique combination and stereochemistry of MHC side chains in the P9 pocket of DQ2 compared to other non-aspβ57 class II molecules may be the reason for the positively charged α79 residue to be unable to create favorable interactions with a negatively charged P9 residue of a peptide. Computer modeling of the WT-DQ2 in fact suggests that the basic moiety of the Argα79 is facing out of the pocket, which may give the pocket lesser basicity than P9 pockets of other type 1 diabetes predisposing class II molecules. Our observations furthermore demonstrate that the Mut57- DQ2 molecule does not discriminate as clearly as WT-DQ2 between peptides with optimal and poor P9 anchor residues. This may indicate that the P9 pocket of the Mut57-DQ2 molecules contributes less to the overall peptide binding specificity than the corresponding pocket of the WT-DQ2 molecule. A structural explanation for this could be that the P9 pocket of the Mut57-DQ2 molecule is reduced in size compared to the deep pocket of WT-DQ2 due to the formation of a salt bridge between Argα79 and Aspβ57. Our data on DQ2 do not conform with the model claiming that certain class II molecules confer susceptibility to type 1 diabetes because they preferentially bind a disease inducing peptide with a negatively charged residue at P9 (14,15). Based on our results, other mechanisms are more likely to explain the connection between DQ2 and type I diabetes. Possibly, the DQ2-related disease susceptibility effect is mediated by the sum of various polymorphic sites of the DQ2 molecule with no particular charge effect of the P9 pocket or it could be mediated by other molecules encoded by genes in strong linkage disequilibrium with the DQ2 encoding genes. In conclusion, we have demonstrated that wild-type DQ2 preferentially binds peptides with large hydrophobic anchor residues at P9. We have also demonstrated that substitution of Ala to Asp at position β57 of DQ2 changes the preference of the P9 anchor residue, most probably as the result of a

7 No preference for negatively charged P9 anchor residues in HLA-DQ reduction in the size of the P9 pocket. Apparently, the non- Aspβ57 residue affects the specificity of the P9 pocket differently in DQ2 compared to other non-aspβ57 type 1 diabetes-predisposing class II molecules so far analyzed. Acknowledgements We thank William W. Kwok for providing the retroviral vector, Janet Lee for donating the plasmid carrying cdna for the HLA-DQB1*0202 gene, Helge Viken for the kind gift of the mab 2.12.E11 and Erik Thorsby for critically reading the manuscript. This paper is dedicated to the memory of Gunnar Paulsen. Abbreviations MB B-LCL RBC PSCPL References Mycobacterium bovis B-lymphoblastoid cell line relative binding capacity positional scanning combinatorial libraries 1 Stern, L. J., Brown, J. H., Jardetzky, T. S., Gorga, J. C., Urban, R. G., Strominger, J. L. and Wiley, D. C Crystal structure of the human class II MHC protein HLA-DR1 complexed with an influenza virus peptide. Nature 368: Ghosh, P., Amaya, M., Mellins, E. and Wiley, D. C The structure of an intermediate in class II MHC maturation: CLIP bound to HLA-DR3. Nature 378: Jardetzky, T. S., Brown, J. H., Gorga, J. C., Stern, L. J., Urban, R. G., Strominger, J. L. and Wiley, D. C Crystallographic analysis of endogenous peptides associated with HLA-DR1 suggests a common, polyproline II-type conformation for bound peptides. Proc. Natl Acad. Sci. USA 93: Todd, J. A., Bell, J. I. and McDevitt, H. O HLA-DQβ gene contributes to susceptibility and resistance to insulin-dependent diabetes mellitus. Nature 329: Nepom, G. T. and Erlich, H MHC class-ii molecules and autoimmunity. Annu. Rev. Immunol. 9: Cucca, F., Lampis, R., Frau, F., Macis, D., Angius, E., Masile, P., Chessa, M., Frongia, P., Silvetti, M., Cao, A., De Virgiliis, S. and Congia, M The distribution of DR4 haplotypes in Sardinia suggests a primary association of type I diabetes with DRB1 and DQB1 loci. Hum. Immunol. 43: Brown, J. H., Jardetzky, T. S., Gorga, J. C., Stern, L. J., Urban, R. G., Strominger, J. L. and Wiley, D. C Three-dimensional structure of the human class II histocompatibility antigen HLA- DR1. Nature 364:33. 8 Reich, E-P., von Grafenstein, H., Barlow, A., Swenson, K. E., Williams, K. and Janeway, C. A., Jr Self peptides isolated from MHC glycoproteins of non-obese diabetic mice. J. Immunol. 152: Reizis, B., Eisenstein, M., Bockova, J., Könen-Waisman, S., Mor, F., Elias, D. and Cohen, I. R Molecular characterization of the diabetes-associated mouse MHC class II protein, I-A g7. Int. Immunol. 9: Kwok, W. W., Domeier, M. E., Johnson, M. L., Nepom, G. T. and Koelle, D. M HLA-DQB1 codon 57 is critical for peptide binding and recognition. J. Exp. Med. 183: Kwok, W. W., Domeier, M. E., Raymond, F. C., Byers, P. and Nepom, G. T Allele-specific motifs characterize HLA-DQ interactions with a diabetes-associated peptide derived from glutamic acid decarboxylase. J. Immunol. 156: Kinouchi, R., Kobayasi, H., Sato, K., Kimura, S. and Katagiri, M Peptide motifs of HLA-DR4/DR53 (DRB1*0405/DRB4*0101) molecules. Immunogenetics 40: Friede, T., Gnau, V., Jung, G., Keilholz, W., Stevanovic, S. and Rammensee, H.-G Natural ligand motifs of closely related HLA-DR4 molecules predict features of rheumatoid arthritis associated peptides. Biochim. Biophys. Acta 1316: Wucherpfennig, K. W. and Strominger, J. L Selective binding of self peptides to disease-associated major histocompatibility complex (MHC) molecules: a mechanism for MHC-linked susceptibility to human autoimmune diseases. J. Exp. Med. 181: Tisch, R. and McDevitt, H Insulin-dependent diabetes mellitus. Cell 85: Harrison, L. C., Honeyman, M. C., Trembleau, S., Gregori, S., Gallazzi, F., Augstein, P., Brusic, V., Hammer, J. and Adorini, L A peptide-binding motif for I-A g7, the class II major histocompatibility complex (MHC) molecule of NOD and Biozzi AB/H mice. J. Exp. Med. 185: Saito, Y., Peterson, P. A. and Matsumura, M Quantitation of peptide anchor residue contributions to class I major histocompatibility complex molecule binding. J. Biol. Chem. 268: Udaka, K., Wiesmüller, K.-H., Kienle, S., Jung, G. and Walden, P Decrypting the structure of major histocompatibility complex class I-restricted cytotoxic T lymphocyte epitopes with complex peptide libraries. J. Exp. Med. 181: Godkin, A., Friede, T., Davenport, M., Stevanovic, S., Willis, A., Jewell, D., Hill, A. and Rammensee, H. G Use of eluted peptide sequence data to identify the binding characteristics of peptides to the insulin-dependent diabetes susceptibility allele HLA-DQ8 (DQ3.2). Int. Immunol. 9: Johansen, B. H., Vartdal, F., Eriksen, J. A., Thorsby, E. and Sollid, L. M Identification of a putative motif for binding of peptides to HLA-DQ2. Int. Immunol. 8: Van de Wal, Y., Kooy, Y. M. C., Drijfhout, J. W., Amons, R. and Koning, F Peptide binding characteristics of the coeliac disease-associated DQ(α1*0501, β1*0201) molecule. Immunogenetics 44: Paulsen, G., Lundin, K. E. A., Gjertsen, H. A., Hansen, T., Sollid, L. M. and Thorsby, E HLA-DQ2 restricted T-cell recognition of gluten-derived peptides in celiac disease. Influence of amino acid substitutions in the membrane distal domain of DQβ1*0201. Hum. Immunol. 42: Johansen, B. H., Buus, S., Vartdal, F., Viken, H., Eriksen, J. A., Thorsby, E. and Sollid, L. M Binding of peptides to HLA-DQ molecules: peptide binding properties of the diseaseassociated HLA-DQα1*0501, β1*0201) molecule. Int. Immunol. 6: Buus, S., Sette, A., Colòn, S. M., Jenis, D. M. and Grey, H. M Isolation and characterization of antigen Ia complexes involved in T cell recognition. Cell 47: Gorga, J. C., Horejsi, V., Johnson, D. R., Raghupathy, R. and Strominger, J. L Purification and characterization of class II histocompatibility antigens from a homozygous human B cell line. J. Biol. Chem. 262: Viken, H. D., Paulsen, G., Sollid, L. M., Lundin, K. E. A., Tjønnfjord, G. E., Thorsby, E. and Gaudernack, G Characterization of an HLA-DQ2 specific monoclonal antibody. Influence of amino acid substitutions in DQβ1*0202. Hum. Immunol. 42: Schneider, C., Newman, R. A., Sutherland, D. R., Asser, U. and Greaves, M. F A one-step purification of membrane proteins using a high efficiency immunomatrix. Biol. Chem. 257: Wong, W. Y., Sheth, H. B., Holm, A., Irvin, R. T. and Hodges, R. S Representative combinatorial peptide libraries: An approach to reduce both synthesis and screening efforts. In Strop, P., ed., Generation and selection of random arrays of peptides, p Academic Press, New York. 29 Falk, K., Rötzschke, O., Stevanovic, S., Jung, G. and Rammensee, H.-G Pool sequencing of natural HLA-DR, DQ and DP ligands reveals detailed peptide motifs, constraints of processing, and general rules. Immunogenetics 39: Thorpe, C. J. and Moss, D. S Molecular modeling and simulations of major histocompatibility complex class I protein peptide interactions. In Goodfellow, J. M., ed., Computer modelling in molecular biology, p VCH, New York. 31 Vartdal, F., Johansen, B. H., Friede, T., Thorpe, C. J., Stevanovic, S., Eriksen, J. A., Sletten, K., Thorsby, E., Rammensee, H.-G. and Sollid, L. M The peptide binding motif of the disease

8 1236 No preference for negatively charged P9 anchor residues in HLA-DQ2 associated HLA-DQ (α1*0501, β1*0201) molecule. Eur. J. Immunol. 26: Thorsby, E. and Rønningen, K. S Particular HLA-DQ molecules play a dominant role in determining susceptibility or resistance to type 1 (insulin-dependent) diabetes mellitus. Diabetologia 36: Olerup, O., Aldener-Cannavà, A., Fogdell-Hahn, A., Getty, R. R., Wagenknecht, D. R. and McIntyre, J. A DQB1*0202 and the new DQB1*0203 allele: a fourth pair of DQB1 alleles differing only at codon 57. Tissue Antigens 49: Falk, K., Rötzschke, O., Stevanovic, S., Jung, G. and Rammensee, H.-G Allele-specific motifs revealed by sequencing of selfpeptides eluted from MHC molecules. Nature 351: Jardetzky, T. S., Lane, W. S., Robinson, R. A., Madden, D. R. and Wiley, D. C Identification of self peptides bound to purified HLA-B27. Nature 353: Hunt, D. F., Henderson, R. A., Shabanowitz, J., Sakaguchi, K., Michel, H., Sevilir, N., Cox, A. L., Appella, E. and Engelhard V. H Characterization of peptides bound to class I MHC molecule HLA-A2.1 by mass spectrometry. Science 255: Babbitt, B. P., Allen, P. M., Matsueda, G., Haber, E. and Unanue, E. R Binding of immunogenic peptides to Ia histocompatibility molecules. Nature 317: O Sullivan, D., Sidney, J., Appella, E., Walker, L., Phillips, L., Colòn, S. M., Miles, C., Chesnut, R. W. and Sette, A Characterization of the specificity of peptide binding to four DR haplotypes. J. Immunol. 145: Jardetzky, T. S., Gorga, J. C., Busch, R., Rothbard, J., Strominger, J. L. and Wiley, D. C Peptide binding to HLA-DR1: a peptide with most residues substituted to alanine retains MHC binding. EMBO J. 9: O Sullivan, D., Arrhenius, T., Sidney, J., del Guercio, M.-F., Albertson, M., Wall M., Oseroff, C., Southwood, S., Colòn, S. M., Gaeta, F. C. A. and Sette, A On the interaction of promiscuous antigenic peptides with different DR alleles. J. Immunol. 147: Stryhn, A., Pedersen, L. Ø., Romme, T., Holm, C. B., Holm, A. and Buus, S Peptide binding specificity of major histocompatibility complex class I resolved into an array of apparently independent subspecificities: quantitation by peptide libraries and improved prediction of binding. Eur. J. Immunol. 26:1911.